Dölling Etc., Language, Context, and Cognition

Event Structures in Linguistic Form and Interpretation
Language, Context, and Cognition

Edited by
Anita Steube
Volume 5
w
DE
G
Walter de Gruyter Berlin New York
Event Structures
in Linguistic Form
and Interpretation
Edited by
Johannes Dlling
Tatjana Heyde-Zybatow
Martin Schfer
w
DE
G
Walter de Gruyter Berlin New York
Printed on acid-free paper which falls within the guidelines

of the ANSI to ensure permanence and durability.
Library of Congress Cataloging-in-Publication
Data
Event structures in linguistic form and interpretation / edited by

Johannes Dlling, Tatjana Heyde-Zybatow, Martin Schfer,
p. cm. (Language, context, and cognition)
Includes bibliographical references and index.
ISBN 978-3-11-019066-3 (alk. paper)
1. Grammar, Comparative and general Syntax.
2. Grammar,
Comparative and general Verb phrase.
3. Semantics. I. Dlling,
Johannes.
II. Heyde-Zybatow, Tatjana, 1973
III. Schfer, Martin, 1975P295.E96 2007
415dc22
2007005990
ISBN 978-3-11-019066-3
Bibliographic information published by the Deutsche
Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the Deutsche

Nationalbibliografie; detailed bibliographic data are available in the Internet
at http://dnb.d-nb.de.
Copyright 2008 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin

All rights reserved, including those of translation into foreign languages. No part of this
book may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopy, recording or any information storage and retrieval
system, without permission in writing from the publisher.
Printed in Germany
Cover design: Christopher Schneider, Berlin
Printing and binding: Hubert & Co. G m b H & Co. KG, Gttingen
Contents
Introduction
IX
Section I: Event Structure and Syntactic Construction

Patients in Igbo and Mandarin
Alexander Williams
Event decomposition and the syntax and semantics

of durative phrases in Chinese
Jo-wang Lin
31
Syntactic decomposition of events in Korean and Standard Indonesian
55
Minjeong Son and Peter Cole
Section II: Event Structure and Modification

Unifying illegally
Kyle Rawlins
81
Adverbial modification of adjectives: Evaluatives and a little beyond

Marcin Morzycki
103
The structure of criterion predicates

Kjell Johan Sceb0
127
Reference to embedded eventualities
149
Markus Egg
Section III: Event Structure and Situation Aspect

Two puzzles for a theory of lexical aspect: Semelfactives
and degree achievements
Susan Rothstein
The notion of 'path' in aspectual composition: Evidence from Japanese
Eri Tanaka
175
199
VI
Contents
Reflexive intransitives in Spanish and event semantics

Eric McCready and Chiyo Nishida
223
Scalar complexity and the structure of events
245
John Beavers
Section IV: Event Structure and Plurality

On the plurality of verbs
Angelika Kratzer
269
Event quantification and distributivity

Kimiko Nakanishi
301
The event structure of irreducibly symmetric reciprocals

Alexis Dimitriadis
327
Existential readings for bare plurals in object position
355
Sheila Glasbey
Section V: Event Structure and Temporal Location

Tense and adverbial quantification
Cornelia Endriss and Stefan Hinterwimmer
389
Phase structure and quantification

Marko Malink
413
Cohesion in temporal context: The role of aspectual adverbs

Alice G.B. ter Meulen
435
Mandarin sentential -le, perfect and English already
447
Hooi Ling Soh and Meijia Gao
Section VI: Event Structure and Natural Language Ontology

The lower part of event ontology
Regine Eckardt
477
Verbs of creation
Christopher Pin
493
Contents
VII
Portraits of the Authors
523
Index
527
Introduction
This volume comprises a selection of papers presented at the workshop "Event
Structures in Linguistic Form and Interpretation", which took place at the University of Leipzig in March 17-19, 2004. The workshop was hosted by the
research project "Event Structures: Grammatical and Conceptual Components
of Utterance Interpretation" at the Department of Linguistics of the University
of Leipzig. The central topic to be addressed was how conceptual information
on event structure is encoded in linguistic expressions and how such information can be reconstructed from utterances. Answers to these questions essentially contribute to a better understanding of the relationship between lexical
semantics, syntactic structure, pragmatic inference, and world knowledge in a
broader cognitive perspective.
Among the many collections on event-based semantics and syntax appearing over the last ten years (e.g. Rothstein, 1998; Tenny and Pustejovsky, 2000;
Higginbotham, Pianesi and Varzi, 2000; Lang, Maienborn and FabriciusHansen, 2003; Austin, Engelberg and Rauh, 2004; Maienborn and Wllstein,
2005; and Verkuyl, de Swart and van Hout, 2005), this volume adopts a decidedly applied attitude in that the existence of Davidsonian event arguments is
taken as given and that problems of the fundamental methodology are of minor
concern. Instead, it demonstrates how the idea of event structure can be successfully applied to a wide range of empirical problems in an increasing number of languages. Thus, the topic is discussed not only on the basis of English
and German but, among others, of Mandarin Chinese, Japanese, Korean, Indonesian, and Igbo as well.
The contributed papers fall into several broad classes - accounting for event
structure in connection with syntactic construction, modification, situation
aspect, plurality, temporal location, and natural language ontology. Accordingly, the volume is organized into six sections.
Section I: Event Structure and Syntactic Construction

The assumption of an event-related framework in research on verb meaning
raises important questions for the analysis of the syntax/semantics interface.
While it is uncontroversial that event structure is to some degree reflected in
syntactic structure, there is an ongoing discussion about the sort of verbal information that should play a role in the syntactic derivation. Two different
strategies of meaning decomposition, which can be traced back to the Genera-
Introduction
tive Semantics tradition, dominate. One line (cf. e.g. Dowty, 1979; Jackendoff,
1990; Pustejovsky, 1991; Bierwisch, 1997; Rappaport and Levin, 1998) assumes that verbs have to be decomposed in the lexicon, independently of syntax, into more primitive predicates representing the structure of events designated. The fundamental assumption behind such lexicalist accounts is that the
lexical-semantic entry of a verb determines its syntactic behaviour. The second, more recent line (cf. e.g. Hale and Keyser, 1993; von Stechow, 1996;
Ritter and Rosen, 1998; Travis, 2000; Ramchand, 2003; Borer, 2005) follows
the idea that event structure is explicitly encoded in syntax. Accordingly, the
meaning of verbs is viewed as being compositionally constructed from primitives linked immediately to functional heads and abstract verbal roots. In this
connection, one essential problem is whether thematic role information is projected from the verb's lexical representation, or imposed by the structural context in which the verb occurs. In particular, Kratzer (1996, 2003) has given
strong evidence for the claim that external arguments are attached in the syntactic derivation through the mediation of Davidsonian event arguments.
Building on the proposal by Kratzer, Alexander Williams additionally
questions the common assumption that patients have to be treated as internal
arguments of the verb. He lays out data from resultative constructions (RCs) in
Igbo and Mandarin to show that even if a verb in a simple clause requires a
direct object - interpreted as patient - the same verb has no such requirement
if it serves as a means predicate in a RC. These data can be explained if the
patient relation is introduced by VP structure, and not by the verb. Williams
argues that languages differ with respect to the number of lexically encoded
arguments within the range of possible participants of the described event.
This makes it possible that the lexical entries of verbs in English contain the
patient argument position but the entries of verbs in Igbo and Mandarin do not.
A further important general consequence of his analysis is that one has to recognize patient as a basic thematic predicate.
The paper by Jo-wang Lin is concerned with the syntactic distribution and
the semantics of durative phrases in Mandarin. He argues that they can be adjoined to every maximal projection, provided that we can interpret them there
without violating the homogeneity requirement. According to Lin, the event
structure of verbs should be decomposed in overt syntactic structures, similar
to von Stechow's treatment of the scope ambiguity of wieder ('again'). According to the author, this decompositional analysis explains why result-related
durative phrases are syntactically more restricted - they can only occur after
the direct object of the sentence - than process-related durative phrases. Finally, Lin discusses the implication of his structural account for data containing durative phrases in conjunction with incremental theme verbs, arguing that
they have to be analyzed as inherently telic.
Minjeong Son and Peter Cole also present empirical support for the view
that the decomposition of verb meaning is explicitly reflected in (morpho-)
Introduction
XI
syntax. According to them, such an approach provides satisfactory explanation

not only for the scope ambiguity of tasi ('again') associated with morphological as well as lexical causatives in Korean, but also for the apparent lexical
ambiguity of the verbal suffix -kan in Standard Indonesian. For Korean, the
authors assume that an abstract CAUSE can be overtly realized by the suffix i-, while the Indonesian suffix -kan is argued to be the overt instantiation of a
RESULT predicate. Furthermore, Son and Cole suggest that their account unifies the different uses of the suffix in causative and benefactive constructions.
A broader question raised by this analysis is in what extent languages differ in
the way conceptual components of event structure are mapped into the syntax.
Section II: Event Structure and Modification

Sentences containing adverbial modification were one of the main reasons for
Davidson (1967) to introduce event variables into semantic representations.
According to him, modifiers of verbal expressions can be analyzed as first
order predicates that add information about the events introduced by the respective expressions. One advantage of this approach is that it allows one to
draw inferences that relate to adverbial modifiers by virtue of conjunction reduction. While Davidson's account was restricted to instrumental, local and
temporal adverbials, Parsons (1990) argued that the event variable can be used
in the formalization of all VP modifiers that is, including, e.g., manner adverbs
such as slowly, gently or quietly. In the wake of Parsons, the event-based
framework has been applied to a far broader range of modification phenomena
and has yielded many fruitful results. However, a number of areas have shown
themselves to be somehow intricate. Thus, it is well-known that many adverbials can receive different interpretations depending on their syntactic position
(cf. e.g. Ernst, 2002). Besides a few proposals covering specific types of modification (cf. e.g. Maienborn, 2001), till now there is no general approach to the
systematic effects that syntactic variation bears on the resulting event structure.
Moreover, an open question is to what extent adverbial modification can involve reinterpretations which are triggered by semantic mismatches and executed by shifting of ontological event type (cf. e.g. de Swart, 1998; Geuder,
2000; Dolling, 2003, 2005).
The paper by Kyle Rawlins starts from the afore-mentioned observation
that the meaning of modifying adverbs may vary depending on their syntactic
position. In particular, the author is concerned with several modifier uses of
adverbs like illegally: (a) a clausal use (= high position), (b) a manner use (=
low position) and (c) two pre-adjectival uses. Rawlins argues that the meaning
differences induced by placing such an adverb in different positions result
purely from scope, especially from the relative scope with respect to the existential quantification over events and with respect to tense. After introducing a
semantics for all four uses of illegally separately, he proposes one lexical entry
XII
Introduction
that expresses the core meaning of this adverb, which is identical with its interpretation in the high position. To enable adverbs to compose in a variety of
positions, Rawlins assumes a family of type-shift operators that coerce sentence modifiers into modifiers of the respective types.
Adverbs in a pre-adjectival position stand in the centre of the paper by
Marcin Morzycki. It examines adverbs like remarkably and surprisingly
which modify adjectival phrases and give rise to judgments about having a
property to a particular degree, although they are not degree words. On the
approach proposed, the evaluative adverbs have the effect of domain widening,
similar to effects observed for embedded exclamatives, and are interpreted as
arguments of an unrealized degree morpheme in much the same way as nominal measure phrases have been proposed to be. At the same time, Morzyski
shows that they themselves have the same denotation as the corresponding
adjectives and their adverbial counterparts in the clause-modifying position.
Finally, the author suggests how the analysis can be extended to ad-adjectival
subject-oriented adverbs.
Kjell Johan Sseb0 is interested in the question how the modification of abstract predicates by instrumental fey-phrases can be accounted for in an eventbased framework. Criterion predicates like obey doctor's order and do me a
favour form one type of these predicates and have remained ill-understood
until now. Also, there is no consensus on the proper analysis of the second
major type of abstract predicates, the manner-neutral causative predicates like
create a fiction and ruin my reputation. As is argued by the author, criterion
predicates as well as manner-neutral causatives are characterized by a certain
degree of indeterminacy. Whereas the former do not specify the physical criteria which an action must meet, the latter are unspecific about the way in which
the change of state is brought about. Saeb0 proposes an analysis where both
types of predicates involve an indeterminate event predicate and where the
function of the instrumental fry-adjunct is to fill it with content via unification.
Another aspect of modification of event expressions is highlighted by Markus Egg. He offers a unified approach to cases in which modifiers or affixes
refer to embedded eventuality arguments in the semantics of the modified expression or base, respectively. Addressing modification of deverbal nouns by
an adjective like beautiful in beautiful dancer, restitutive readings of againsentences, and the effect of prefixation in German verb nominalisations like
Losgerenne, the author argues that all three cases can be analyzed as interface
phenomena in which a modifier or affix may semantically apply to only a part
of the semantic contribution of its modified expression or base. Within this
part, the eventualities that are bound in the semantics of the modified expression or base as a whole emerge as open arguments. The analysis is modelled in
a framework based on underspecification and makes use of potential scope
ambiguities in the semantics of the expressions discussed.
Introduction
XIII
Section III: Event Structure and Situation Aspect

Situation aspect (cf. Smith, 1991), also called lexical aspect or Aktionsart,
which is concerned with the internal temporal constitution of events and its
linguistic reflection, is perhaps the most prominent semantic field where the
Davidsonian approach has achieved remarkable advances. Although the literature on this phenomenon is vast, there remain many questions that are not convincingly solved yet. Since Vendler's (1967) classification of verbal predicates
into four classes - activities, states, accomplishments and achievements - researchers aim at the clarification of the determining factors of these classes.
Features like dynamicity, durativity, telicity and gradability play a central role
in the discussion. Do such features characterize the underlying events or are
they part of the verbal meaning or both of them? If they are considered to be
features of the events, then one has to distinguish between several ontological
types (or sorts) of events, i.e. between events of change, processes, and states
(cf. e.g. Bach, 1986a,b; Parsons, 1990; Pin, 1995; Dlling, 2005). Another
possibility that has proven itself to be of great benefit is the approach proposed
by Krifka (1989, 1992, 1998), which identifies quantization (telicity) and cumulativity (atelicity) as properties of verbal predicates. Here, the tight interaction between the referential information of a verb and that of its arguments (or
adjuncts) is crucial for the explanation of aspectual composition (cf. e.g. Filip,
1999; Engelberg, 2000; Rothstein, 2004).
Susan Rothstein argues that the semantics behind the Vendlerian verb
classes is best captured with the help of two sets of features: whether or not the
event in its denotation is durative, and whether or not it denotes an event of
change. But the verbal class of semelfactives (e.g. knock) and that of degree
achievements (e.g. cool) pose notorious problems for the Vendlerian classification. Rothstein proposes an operation of S-summing in order to explain why
degree achievements, although verbs of change, also have atelic interpretations. This operation forms singular events out of sums of temporally adjacent
events. While S-summing normally does not apply to verbs of change, because
two events of change cannot be immediately adjacent to each other, it can apply to degree predicates, because they are characterized as changes in values
on a scale. Semelfactives differ from activities in that they come with natural
beginnings and endpoints and therefore can be perceived as atoms. If that is
the case, the interpretation of a semelfactive verb is telic.
Eri Tanaka's paper aims at an explanation for why Japanese lacks socalled strong adjectival resultative constructions. Its starting point is the assumption that one should differentiate between incremental theme verbs and
motion verbs on the one hand, and change of location/state verbs on the other
hand. Only the latter should be represented by the BECOME-operator that
serves as a source for their telic interpretation. Verbs of the first group are atelic in Japanese and are interpreted as referring to a path or scale. Telic inter-
XIV
Introduction
pretations with verbs of this group are only possible if the path/scale is
bounded by the object-NP or a postposition (e.g. -made 'up to'). Tanaka takes
the complementary distribution of the two postpositions -ni ('in/at/to') and made as further evidence for the distinction between the two verb groups. She
suggests that weak resultatives are formed on the basis of change of location/state verbs - hence BECOME-verbs - and that the adjectives modify the
result state. Strong adjectival resultatives are based on verbs of the first group,
but Japanese adjectives lack the possibility to bind/to limit the path/scale and
therefore Japanese lacks this sort of construction.
The paper by Eric McCready and Chiyo Nishida is concerned with Spanish reflexive intransitives (Ris), i.e. constructions containing the reflexive clitic
se in conjunction with a non-transitive verb. After establishing the differences
between RIs and their transitive counterparts, the authors note that (dynamic)
RIs have three distinctive properties. They require a quantized subject NP, they
license a dative argument which stands in some relation to the event described
by the sentence, and they coerce the verbs they appear with to achievements
denoting a transition, or the onset or the final end of a process or a state.
McCready and Nishida model these facts by introducing a presupposition of
telicity on the VP associated with reflexive se, allowing RIs to be associated
with an additional, possibly implicit, argument, and, finally, placing meaning
postulates on verb classes associating the temporal interval of the event with
the initial or final point of a path.
John Beavers focuses primarily on the factors governing durativity in dynamic predicates to build a broader picture of the aspectual behaviour of descriptions of events of change. His main objective is to show that there is a
general correlation between the durativity of an event and the gradability of the
scale of change of a participant. He argues that Rrifka's homomorphism model
designed originally to explain the nature of telicity of incremental theme verbs
can be generalized to cover a wide range of dynamic predicates following a
scalar approach to telicity and, in consequence, also explains the durativity/gradability correlation. Essential for the analyses by Beavers is a distinction
between two types of mereological complexity: structures with two sub-parts
(begin, end) and structures with at least three sub-parts (begin, middle and
end). In addition, the paper outlines relevant lexical, pragmatic and contextual
constraints on durativity and gradability and discusses their possible origins.
Section IV: Event Structure and Plurality

Quite a number of recent event-based studies deal with the question of how
distributive, collective and cumulative readings arise. Traditional analyses attribute this difference between the addressed readings to the quantificational
make-up of the NPs occurring in the respective sentences. But since the work
of Schein (1993) and Lasersohn (1995), it has been recognized that event
Introduction
XV
structure must play a major role in accounts of plurality, including distributive,

collective and cumulative interpretations (cf. e.g. Landman, 2000; Kratzer,
2003). In this connection, there is an ongoing discussion about the question of
whether verbal predicates are lexically already marked for plurality. Moreover,
the plurality question points towards a broader problem: although there is a
huge amount of literature on quantification, the formal approaches to these
phenomena rarely make use of events. With the growing acceptance of event
semantics it became apparent that a combination of event-based theories and
theories of quantification is necessary. In this context, the influence of the
event type on the existential force of the subjects and objects of verbs presents
another challenge.
Angelika Kratzer pursues some of the consequences of the idea that there
are at least two sources for distributive/cumulative alternation of readings in
English. One source is lexical pluralization. Following a proposal by Krifka
and Landman, she assumes that all verb stems are born as plurals independent
of the particular language and the particular nature of its NPs. The second
source of cumulative/distributive interpretation in English is directly provided
by plural NPs. Kratzer proposes that such phrases with plural agreement features are able to pluralize adjacent verbal projections. The difference between
the two possibilities lies in the fact that distributive/cumulative interpretations
on the basis of lexical pluralization allow the occurrence of singular NPs
within the sentences, which is impossible with the second type. She shows that
the phenomena discussed in her paper all pose conceptual problems for analyses not based on events, but can be given elegant accounts within a version of
the Davidsonian event semantics.
The objective of the contribution by Kimiko Nakanishi is closely connected to the last one: It examines empirical data on collective and distributive
interpretations of constructions in Japanese and German, where a quantificational expression appears separated from its host NP. The author assumes that
such split quantifiers measure the events in the verbal domain, but only if the
relevant VP denotes a part-whole structure, i.e. a lattice of events. Furthermore, she assumes that, with the help of the mapping based on the homomorphism from events to objects, the measure function applies indirectly to the
events by measuring the objects denoted by the host NP. According to her, this
mechanism of event measurement satisfactorily accounts for why split quantifier constructions usually disallow collective readings. In addition, Nakanishi
discusses some examples where, by contrast, a collective reading is available
in such a construction and argues that the proposed analysis is capable of handling these cases, too.
A specific type of multiple participants in a single event is focussed by
Alexis Dimitriadis' paper. The author presents a wide range of data on reciprocals, especially on irreducibly symmetric and discontinuous reciprocals,
found in numerous languages. It is essential for an irreducibly symmetric
XVI
Introduction
predicate like meet that it expresses a relationship between two arguments

where, in contrast to ordinary reciprocals, both arguments necessarily have
thematically identical participation in any event described by means of it. As it
is shown by Dimitriadis, irreducible symmetry plays a prominent role in many
discontinuous constructions, in which the logical subject of a reciprocal verb
appears to be split between the syntactic subject and a comitative vv/i/i-phrasc.
Taking into account the well-known problems with assigning the same thematic role to different participants, he proposes an analysis where the symmetric events are decomposed into sub-minimal events specifying the distinct relations of each participant to the complex event.
Sheila Glasbey investigates which verbal predicates disallow existential
readings for their bare plural objects and, in addition, which adjectival predicates disallow existential readings for their bare plural subjects. One of her
observations is that, contrary to the usual picture, many verbs and adjectives
which might well be classified as individual level predicates give not only generic but also existential interpretations. Using a situation-theoretic framework, the author argues that, generally, such a reading is made possible by the
presence of a localising situation, which may be provided by the event argument of the verb or by an appropriate context. According to her, psych-verbs
with experiencer subjects like hate as well as adjectival predicates lack event
arguments and, therefore, may allow an existential reading for their bare plurals only with the help of a specific context. The proposal is that, in both cases,
the reading comes from a so-called existential inference which is licensed by
the respective situation.
Section V: Event Structure and Temporal Location

The significance of temporal location for the determination of event structure
should be obvious (cf. e.g. Parsons, 1990; Kamp and Reyle, 1993). According
to common understanding, the tense of the verb serves primarily to localize the
respective event in time - notably before, around, or after the time of utterance
(or speech). A more recent approach takes another stance (cf. Klein, 1994):
Tense does not directly specify the time of the event; rather, it locates the time
interval about which the utterance asserts something with respect to the utterance time. Viewpoint (or grammatical) aspect (cf. Smith, 1991), i.e. the device
for making distinctions between, e.g., perfective and imperfective, on the other
hand, concerns the relationship of event time to tense time. Thus, not only
tense but also viewpoint aspect is defined in terms of temporal relations. While
there is a huge amount of research on the two grammatical categories and on
the interaction between them, there is much less work on how they interact
with quantificational NPs or with quantificational adverbials in terms of
scopes, as well as with particles like already or still. In addition, specific questions arise from languages which lack at least one of the categories to express
Introduction
XVII
the temporal location of events. For example, Mandarin has no tense marker
and, hence, no grammaticalized means to impose a constraint on the time
about which the assertion is made. An important problem which has to be investigated is how the temporal location system fills such a 'gap'.
In their paper, Cornelia Endriss and Stefan Hinterwimmer present novel
evidence for an analysis of Quantificational Variability Effects (QVEs) as byproducts of a quantification over events/situations. They compare adverbially
quantified sentences containing indefinites modified by relative clauses with
sentences containing corresponding quantificational NPs modified by relative
clauses, showing that the former have to obey a constraint in order for QVEs to
arise that does not hold for the latter: the tense of the relative clause verb has to
agree with the tense of the matrix verb. While this is completely unexpected
under the assumption that QVEs come about via direct quantification over
individuals, the authors show that a natural explanation is possible under the
assumption that the events quantified over have to be located in a salient time
interval, and that this interval is determined on the basis of a pragmatic strategy dubbed Interval Resolution Strategy that favours local information.
Marko Malink deals with the interplay of negation, quantifiers and phase
particles schon ('already'), noch ('still'), noch nicht ('not yet') and nicht mehr
('not anymore'). In particular, he provides a scope analysis of German sentences such as Einige sind noch nicht da ('Some people are not there yet') or
Niemand ist mehr da ('Nobody is there anymore'). In order to correctly account for the intuitive meaning of such sentences, he proposes to split negative
phase particles such as noch nicht and negative quantifiers such as niemand
into a negation and a purely positive part. He then shows that the negation has
a different scope position than the positive part of the phase particle or quantifier to which it belongs. He goes on to specify a compositional formal account
of quantified phase structures within a generalized quantifier framework. Malink argues that in the sentences under consideration, phase particles have a
bridging function connecting the VP and the subject-quantifier.
A different view on properties of phase particles or, in other words, aspectual adverbs is taken by Alice G. B. ter Meulen. With the overall aim of a
better understanding of temporal coherence of information states, she examines the relation between the objective content of English aspectual adverbs
and the subjective information which can be also conveyed by them. According to the author, the four basic aspectual adverbs - not yet, already, still and
not anymore - constitute a logical polarity square in the temporal domain of
events, showing the fundamental relationship between current, past and future
reference times. In addition, they have usages associated with marked high
pitch prosody, where, besides the factual information, the speaker's attitude
regarding the flow of events or its perceived speed is expressed. To capture
them, ter Meulen introduces a modal operator which quantifies over alterna-
XVIII
Introduction
tives to the current course of events, dependent upon the speaker's epistemic
state.
Hooi Ling Soh and Meijia Gao's paper investigates the meaning of Mandarin sentential -le with specific focus on its relation to English perfect tense
and to the particle already. They propose that the transition marker -le encodes
the assertive meaning that the situation in question is realized prior to a reference time (in a Reichenbachian sense), along with the presupposition that a
situation opposite to the one described by the sentence exists immediately before the point of realization. The reference time is either the utterance time, or,
when the particle jiu is used, some specified time in the past or the future. The
authors discuss the similarities that have been previously noted between sentential -le and English perfect in terms of the presence of a result state and a
continuative reading and argue that the relevant readings are not entailed by le. On their analysis, the transition marker shares its assertive meaning with
perfect tense and its presupposition with already.
Section VI: Event Structure and Natural Language Ontology

There is perhaps no topic in semantics in which questions of meaning are so
intimately interwoven as with ontological questions such as the concept of
event. In order to analyse what information on event structure is encoded in
linguistic expressions one needs an account of the ontology underlying natural
language. Such a theory of the world, i.e. of what basic sorts of entities there
are, what fundamental properties the entities have and how they are related,
does not necessarily correspond to the categorial commitments of current natural science, especially of micro physics. Instead, natural language ontology (cf.
Bach, 1986a) has to accommodate every kind of entity which can be the object
of ordinary talking and thinking. More specifically, the ontological analysis is
not primarily concerned with the way the world really is but rather with the
way we conceptualize it for the purposes of every-day life. Thus, it has to do
with entities which result from projecting our cognitive framework onto environmental input and, for this reason, are also dependant on the concrete point
of view we take. Over the past years, an increasing number of domains of such
entities have been distinguished in investigating the common sense world. Following the crucial innovation by Link (1983) to assume specific lattice structures on the domains of pluralities and of quantities of matter, Bach (1986b)
and, in particular, Krifka (1989, 1992, 1998) have extended this algebraic approach to events. Many others, of which only some could be referred to in this
introduction, have contributed to this far-reaching field of research. As a result,
semantics is provided now with a system of assumptions that precisely characterize the structure of the domain of events and its relationship to the domain
of objects.
Introduction
XIX
Regine Eckardt addresses an apparent conflict between two applications of

event ontology for semantics. Thus, scholars aiming at modelling the difference between telic and atelic predicates commonly assume that certain properties - the properties of homogenous reference - of events are inherited by all
their parts, no matter how small they are. However, proposals modelling negative polarity items are forced to assume a level where parts of events are so
small that they can no longer inherit properties denoted by a natural language
predicate. After summarising the two positions, Eckardt proposes three possible solutions for this dilemma, and discusses two of them, the construction of
infinitesimal events and a solution capitalizing on the Sorites paradox. Both of
the treatments appear to capture some of the essence of how in every-day reasoning very small events are thought about. She concludes that till now there
are no conclusive arguments in favour of one or the other of these options.
Another subject which is immediately connected with basic problems of ontology is investigated by Christopher Pin. His paper gives an account of
verbs of creation (e.g., build, compile, draw, write) which are known to take an
internal argument denoting a physical object that is effected or brought into
being as a result of the event named by the verb. The author argues that such
verbs are actually ambiguous with respect to the sortal character of their internal argument. In particular, the internal argument of a verb of creation may
also denote what he calls a template, i.e. an abstract individual that is physically instantiated in the course of the event described by the verb. This idea
allows to analyze performance verbs of creation (as sing in sing a song) and
those denoting the creation of templates (as compose in compose a symphony).
Moreover, Pin can also explain data such as Sarah built the house that Rebecca designed where the house that Rebecca designed designates a house
template (or design) which Sarah builds an instantiation of. To this end, he
proposes a set of sort-shifters that serve to capture systematic ambiguity
among verbs of creation like build.
We like to thank the reviewers of the workshop for their great help and patience: Markus Egg, Veronika Ehrich, Stefan Engelberg, Hana Filip, Fritz
Hamm, Angelika Kratzer, Manfred Krifka, Claudia Maienborn, Susan Rothstein, Arnim von Stechow, and Henriette de Swart. Furthermore, we want to
thank the authors of the contributed papers for their cooperation all the way
through.
The conference received financial support from the Centre for Higher Studies at the University of Leipzig, which is gratefully acknowledged. The research project "Event Structures: Grammatical and Conceptual Components of
Utterance Interpretation" at the Department of Linguistics of the University of
Leipzig as well as the preparation of manuscript for publication were funded
by the German Research Foundation (DFG).
XX
Introduction
Sebastian Hellmanii and Stefan Keine deserve special thanks for their accurate formatting of the manuscript. Thanks also to David Dichelle and Ryan
Young for checking the English of non-native speakers involved in the volume.
Leipzig, January 2007
Johannes Dlling, Tatjana Heyde-Zybatow and Martin Schfer
References
Austin, Jennifer, Stefan Engelberg and Gisa Rauh (eds.) (2004): Adverbials. The Interplay between Meaning, Context, and Syntactic Structure. Amsterdam, Philadelphia:
Benjamin.
Bach, Emmon (1986a): 'Natural Language Metaphysics'. In: R. Barcan Marcus, G.
Dorn and P. Weingartner (eds.): Logic, Methodology and Philosophy of Science.
Vol. VII., Amsterdam: Elsevier, 573-595.
Bach, Emmon (1986b): 'The Algebra of Events'. Linguistics and Philosophy 9: 5-16.
Bierwisch, Manfred (1997): Lexical Information from a Minimalist Point of View. In:
Ch. Wilder, H.-M. Grtner and M. Bierwisch (eds.): The Role of Economy Principles in Linguistic Theory. Berlin: Akademie Verlag, 227-266.
Borer, Hagit (2005): Structuring Sense. An Exo-Skeletal Trilogy. Vols. 1 and 2. Oxford: Oxford University Press.
Davidson, Donald (1967): 'The Logical Form of Action Sentences.' In: N. Resher (ed.):
The Logic of Decision and Action. 81-95. Pittsburgh: University of Pittsburgh
Press.
Dlling, Johannes (2003): 'Flexibility in Adverbal Modification: Reinterpretation as
Contextual Enrichment'. In: E. Lang and C. Maienborn and C. Fabricius-Hansen
(eds.): Modifying Adjuncts. Berlin, New York: Mouton de Gruyter, 511-552.
Dlling, Johannes (2005): Semantische Form und pragmatische Anreicherung: Situationsausdrcke in der uerungsinterpretation. Zeitschrift fr Sprachwissenschaft
24 : 159-225.
Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Kluwer.
Engelberg, Stefan (2000): Verben, Ereignisse und das Lexikon. Tbingen: Niemeyer.
Ernst, Thomas (2002): The Syntax of Adjuncts. Cambridge: Cambridge University
Press.
Filip, Hana (1999): Aspect, Eventuality Types and Nominal Reference. New York:
Garland Publishing.
Geuder, Wilhelm (2000): Oriented Adverbs. Issues in the Lexical Semantics of Event
Adverbs. Dissertation, Universitt Tbingen.
Introduction
XXI
Hale, Kenneth and Keyser, Jay (1993): 'On argument structure and the lexical expression of syntactic relations'. In: K. Hale and J. Keyser (eds.): The view from building 20: A festschrift for Sylvain Bromberger. Cambridge, MA: MIT Press, 53-108.
Higginbotham, James and Pianesi, Fabio and Varzi, Achille C. (eds.) (2000): Speaking
of Events. Oxford: Oxford University Press.
Jackendoff, Ray (1990): Semantic Structures. Cambridge, MA: MIT Press.
Kamp, Hans and Reyle, Uwe (1993): From Discourse to Logic. Introduction to
Modeltheoretic Semantic of Natural Language. Dordrecht: Kluwer.
Kratzer, Angelika (1996): 'Severing the external argument from its verb'. In: J. Rooryck and L. Zaring (eds.): Phrase Structure and the Lexicon. Dordrecht: Kluwer,
109-137.
Kratzer, Angelika (2003): The Event Argument and the Semantics of Verbs. Ms. University of Massachusetts, Amherst.
Krifka, Manfred (1989): Nominalreferenz und Zeitkonstitution. Zur Semantik von Massentermen, Pluraltermen und Aspektklassen. Mnchen: Fink.
Krifka, Manfred (1992): 'Thematic relations as links between nominal reference and
temporal constitution'. In: I. A. Sag and A, Szabolcsi (eds.): Lexical Matters. Stanford, CA: CSLI Publications, 29-53.
Krifka, Manfred (1998): 'The origins of telicity'. In: S. Rothstein (ed.): Events and
Grammar. Dordrecht: Kluwer, 197-235.
Klein, Wolfgang (1994): Time in Language. London and New York: Routledge.
Landman, Fred (2000): Events and Plurality. The Jerusalem Lectures. Dordrecht: Kluwer.
Lang, Ewald and Maienborn, Claudia and Fabricius-Hansen, Claudia (eds.) (2003):
Modifying Adjuncts. Berlin, New York: Mouton de Gruyter.
Lasersohn, Peter (1995): Plurality, Conjunction and Events. Dordrecht: Kluwer.
Link, Godehard (1983): 'The Logical Analysis of Plurals and Mass Terms: A LatticeTheoretical Approach'. In: R. Buerle,Ch. Schwarze and A. von Stechow (eds.):
Meaning, Use, and Interpretation of Language. Berlin, New York: de Gruyter, 303323.
Maienborn, Claudia (2001): 'On the Position and Interpretation of Locative Modifiers'.
Natural Language Semantics 9: 191-240.
Maienborn, Claudia and Wllstein, Angelika (eds.) (2005): Event Arguments: Foundations and Applications. Tbingen: Niemeyer.
Parsons, Terence (1990): Events in the Semantics of English. A Study in Subatomic
Semantics. Cambridge, MA: MIT Press.
Pin, Christopher (1995): An Ontology for Event Semantics. PhD Thesis. Stanford
University.
Pustejovsky, James (1991): 'The syntax of event structure'. Cognition 41: 47-81.
XXII
Introduction
Swart, Henriette de (1998): 'Aspect shift and coercion'. Natural Language and Linguistic Theory 16: 347-385.
Ramchand, Gillian (2003): First Phase Syntax. Ms. Oxford University.
Rappaport Hovav, Malka and Levin, Beth (1998): 'Building Verb Meanings'. In: M.
Butt and W. Geuder (eds.): The Projection of Arguments: Lexical and Compositional Factors. Stanford: CSLI Publications, 97-134.
Ritter, Elizabeth and Rosen, Sara Thomas (1998): 'Delimiting Events in Syntax'. In: M.
Butt and W. Geuder (eds.): The Projection of Arguments: Lexical and Compositional Factors. Stanford, CA: CSLI Publications, 135-164.
Rothstein, Susan (ed.) (1998): Events and Grammar. Dordrecht: Kluwer Academic
Publishers.
Rothstein, Susan (2004): Structuring Events. A Study in the Semantics of Lexical Aspect. Oxford: Blackwell.
Schein, Barry (1993): Plurals and Events. Cambridge, MA: MIT Press.
Smith, Carlotta (1991): The Parameter of Aspect. Dordrecht: Kluwer.
Stechow, Arnim von (1996): 'The different readings of wieder "again": A structural
account'. Journal of Semantics 13: 87-138.
Tenny, Carol and James Pustejovsky (eds.) (2000): Events as Grammatical Objects.
Stanford, CA: CSLI Publications.
Travis, Lisa (2000): 'Event structure in syntax'. In: C. Tenny and J. Pustejovsky (eds.):
Events as Grammatical Objects. Stanford, CA: CSLI Publications,, 145-185.
Vendler, Zeno (1967): Linguistics in Philosophy. Ithaca, NY: Cornell University Press.
Verkuyl, Henk J. and de Swart, Henriette and van Hout, Angeliek (eds.) (2005): Perspectives on Aspect. Dordrecht: Springer.
SECTION I
Event Structure and Syntactic Construction
Alexander Williams (Maryland)

1 Introduction
The development of event semantics has facilitated discussion of a basic question in verbal grammar. For a given thematic relation between (the meanings
of) a verb and a phrase local to it, is that relation introduced by the lexical representation of the verb, or by the structural context in which the verb occurs?
More briefly, is the phrase an argument of the verb or not? By supplying a parameter of which thematic relations can be conjunctively predicated - the event
variable - event-based representations have allowed questions and answers on
this topic to be formulated in a clear and useful way (e.g. Carlson 1984, Dowty
1989, Kratzer 1996, Marantz 1997, Rothstein 2001).
Yet the two ways of modeling a thematic relation, as projecting from the verb
or as imposed by its context, are often hard to distinguish empirically. In this
paper I discuss a case where the facts seem decisive: resultative constructions in
Mandami and Igbo, like (1) and (2) respectively.1
(1)
(2)
tati
duii -le ntio miibn.
3s kick snap -PFV that plank
'He made that plank snap by kicking.'
O ku
wa -ra
ba ahu.
3sS strike split - F A C T gourd that
'He made that gourd split by striking.'
(ex. Hale, Ihionu, and Manfredi 1995, trans. AW)
The grammar of these constructions demonstrates that, characteristically in Igbo

and Mandarin, neither agents nor patients are arguments of the verb. Basic the1
Mandarin is a Sinitic language and the national language of China. In glosses of Mandarin,
PFV means 'perfective,' and PRT means 'sentence final particle'. Igbo ([io]) is a Benue-Congo
(or Eastern Kwa) language spoken mainly in Nigeria (see Swift et al. 1962, Green and Igwe
1963, Emenanjo 1978, and Igwe 1999). Glosses of Igbo use the following abbreviations. FACT
means 'factative'; roughly, a predicate in the factative has past time reference when eventive and
nonpast time reference when stative. VC means 'bound verb cognate' ( see Nwachukwu 1987 and
Emenanjo 1978). The BVC is a nominalization of the verb group; in all the data presented here, it
is used solely to satisfy the requirement that a verb group in the factative has not be clause-final
(Nwachukwu 1987: 19-21). PROG means 'progressive,' SBRD means 'subordinate verb prefix,'
and means 'all-purpose preposition.'
Alexander Williams
matic relations are instead introduced by the structure in which the verb occurs.
I have defended this conclusion for agents elsewhere (Williams 2004, 2005).
In this paper I make the case for patients, a more surprising hypothesis. Almost
always, patients are treated as arguments of the verb, and strong conceptual support has been given for this decision, notably in Kratzer (2003). Here I counter
with a grammatical argument. For Igbo and Mandarin, the relative distribution
of verbs and patients is best explained if patients are introduced by verb pirrase
structure, and not by the verb. Based on this I abstract a broader conclusion. If
the event of a verb necessarily has a patient as a participant, we still cannot conclude that the verb has a patient as an argument. Lexical meaning is not lexical
valence.
The program is as follows. I begin in section 2 by sketching what it means for
an argument to be introduced by the verb or by its context. Sections 3 and 4 then
define what resultative constructions are, and how they can be used to test for the
valence of verbs within them. The lessons of English resultatives are discussed
in section 5, before I lay out the target data from Igbo and Mandarin in section
6. Section 7 is the core of the paper, setting out the claim that the facts of section
6 are best explained if patients are not arguments of the verb. Sections 8 and 9
consider and dismiss various alternatives. I then mention a semantic objection
to the theory in section 10 before concluding.
2 Projectionist and nonprojectionist models

Suppose that the direct object in a simple clause names the patient of the verb's
event, as in (3). There are two ways to model this, differing in how the relation
is introduced in deriving the semantic representation of the clause (see Carlson
1984, Dowty 1989, Kratzer 1996, Borer 2003).
(3)
\A1 pounded the cutlet\ = 3e.pound{e) PAT(e) = c AG(e) = a
We might say that the patient relation is introduced by the lexical representation
of the verb, as in (4) for example. In this case the patient relation projects from
the verb, and the patient is a lexical argument of the verb.
(4)
poimdj = Xy(Xx)Xe.pound(e)
PAT(e) = y ( AG(e) = )
Or we might say that the patient relation is not introduced by the verb, but by
the structure in which it occurs, perhaps as in (5).2 Then the patient, while it is
identified by an argument category in the clause, is not a lexical argument of the
verb.
In writing "PAT(e ) = [DP]]," I am presuming as a convenience that D P here denotes in the set
of individuals.
(5)
(a)
\pound\ = \e.pound {e)
(b)
[ [VP V D P ] ] = Ae. [ V ] (e) P A T ( e ) =
[DP]
I will call models of the first type projectionist, and models of the second type
nonprojectionist.
The two models differ in how many lexical arguments they assign the verb.
But they need not differ in how many participants they assign the event-type
that the verb describes. If a verb has an argument to which it assigns a certain
thematic relation, then it describes an event involving a participant who bears
that relation. But the converse is not necessarily true. Compare (4) and (5a) for
example. Given existential closure of y (and x) in the fonner, each defines a
predicate of events. These two predicates do not necessarily differ in extension.
Any event that verifies the predicate from (4) must have a patient; this is stated
explicitly in the formula. But the same might be true of (5a), albeit implicitly, if
the metalanguage predicate pound is defined to have (6) as a consequence (see
Dowty 1989: 85). In that case (4) and (5a) will describe exactly the same set of
events.
(6)
Ve [pound (e) s 3j/.PAT(e) = y]
Thus the choice between projectionist and nonprojectionist models is primarily

not semantic, in the strict sense, but grammatical. Given two theories of either
sort which assign the verb the same type of event, which one yields the simpler
and more explanatory grammar?
It is sometimes suggested that, pursuant to some universal principle, a verb
has a lexical argument for each participant in its event (see e.g. Lidz, Gleitman
and Gleitman 2003, Kratzer 2003: Chapter 1, page 18). But this is an empirical
hypothesis, not an a priori principle. And if my understanding of Igbo and Mandarin is correct, it is a hypothesis challenged by the facts of these languages (see
also Davis and Demirdache 2000, Bhatt and Embick 2004).
3 Resultative constructions
Resultative constructions, henceforth RCs, are single clause constructions comprising two predicates, a means predicate (M) and a result predicate (R). Neither
M nor R is introduced by a conjunction, adposition, or complementizer. (7) is
an English example, where M is pound and R is flat.
(7)
Al pounded the cutlet flat.
Semantically, RCs express a relation of causation between the eventualities described by M and R, without this relation being indicated by any overt morpheme
(Dowty 1979, a.o.): (7) says that pounding caused flatness. One aspect of this
Alexander Williams
meaning is that some object changes state, entering the result condition defined
by R. The pirrase that names this object, I will say, controls R. In (7) the cutlet
controls flat, since (7) entails that the cutlet became flat.
(1) and (2), repeated here, are RCs from Mandarin and Igbo. In (1) M is ti
'kick,' R is dun 'snap,' and R is controlled by ntio miban 'that plank.' The
sentence says that kicking caused snapping, and what wound up snapped was
the plank. (2) says that striking caused splitting, and what wound up split was
the gourd. M here is ku 'strike' and R is wa 'split.'
(1)
(2)
tati
dun-le
ntio mbn.
3s kick snap -PFV that
plank
' H e made that plank snap by kicking.'
ku
wa -ra
ba
ahu.
3sS strike split -FACT gourd that
' H e made that gourd split by striking.'
My glosses will follow a fixed fonnat: 'subject made object R by M'ing.' I will
discuss the syntax of Mandarin and Igbo RCs briefly in section 6.
What will filterest me primarily are the understood thematic relations of subject and object to the event of M, the means event. In (7) Al names the agent of
pounding and the cutlet names its patient. In (1) and (2) as well, the subject is
the agent of the M event, and the object is the patient. But we will see in section 6 that Mandarin and Igbo differ from English in not requiring this particular
pattern of relations. And this will form the basis of my central conclusion, that
verbs are typically without arguments in Igbo and Mandarin.
One last distinction sfiould be made, between what I will call transitive and
intransitive RCs. In English the distinction is readily made in tenns of surface
syntax. Transitive RCs have a subject and an object, (8), while intransitive have
only a (surface) subject, (9).
(8)
(9)
(a)
(b)
(a)
(b)
Al pounded the cutlet flat,

Al yelled Iiis throat hoarse.
The lake froze solid,
The door swung shut.
But the criterion of the distinction, as I intend it, is semantic. Transitive but
not intransitive RCs include an argument that is construed as the agent of causation3 (i.e. the causer); as it happens, this is always the subject. Concomitantly
there is a difference in control of R. Control is by the object in transitives and
the (surface) subject in intransitives (Simpson 1983, Y. Li 1995). Classing RCs
along these lines assists in cross-linguistic comparison, allowing generalizations
3
I assume a very broad understanding of the agent relation, similar to what Van Valin and Wilkins
1996 have for their "Effector" relation. I do not assume that agents must be animate or volitional
actors.
to emerge which are otherwise obscured by independent differences in syntax

(Williams 2005). In this paper I rely on it only to limit the scope of discussion:
I will discuss only transitive RCs.
4 The relevance of resultative constructions

RCs are sometimes analyzed as complex predicates (e.g. Dowty 1979, Larson
1991). It is assumed, that is, that M contains no argument positions. Instead the
means verb combines directly with R to the exclusion of the object, (10). 4
(10) [ Object [ Vmeans R ] ]
(linear order irrelevant)
Insofar as this analysis is correct for a given RC, the construction will provide
a diagnostic environment. By putting a verb in M, we stand to learn something
about its lexical argument structure. Let us see why.
Suppose we are choosing between two denotations for pound, (11) and (12).
The two options make different predictions when pound occurs in M, if the RC
is a complex predicate.
(11) [pound ] = XyXe.pound(e) PAT(e) = y
(12) [pound J = \e.pound{e)
Given (11), the verb, when it occurs in M, will have an argument that is not
immediately sahnated, since there the verb's sister will be R, an expression that
cannot provide a patient. We consequently expect that the complex predicate
will inherit this unsaturated argument from M. We expect, for example, that the
denotation ofpoundflat will have the outlines in ( 13 ).
(13) [poundflat ] = \y . . . pound(e) PAT(e) = y . ..
So by assigning the argument to the verb lexically, we encode an expectation
that the verb will be subject to the same requirement in an RC as in a simple
clause. In both contexts it will cooccur with a phrase understood as its patient.
Any deviations from this expectation will count as special cases, in need of
explanation. For example, we might need to posit a covert operator that binds
the verb's unsaturated argument.
We have no such expectation, however, if the patient is not an argument of
the verb, (12). Then there will be no argument left unsaturated when M and R
combine, and no argument to pass along to the complex predicate. So there will
be no assumption, based solely on the verb's lexical representation, that it will
4
Some of the points I make about the complex predicate analysis can also be made about the
Small Clause analysis (Kayne 1985, Hoekstra 1988), according to which RCs have the structure:
[ Vmeans [ Object R ] ]. Yet I will not discuss the Small Clause analysis here, as I find it
unattractive for the languages under consideration (cf. Sybesma 1999).
Alexander Williams
enter the same thematic relation in RCs as it does in simple clauses. It will not
come as a surprise if no noun pirrase in an RC is interpreted as the patient of the
means event.
If they are complex predicates, therefore, RCs can provide evidence for whether or not a given thematic relation projects from the verb. If the relation obtains
wherever the verb occurs, equally in RCs and simple clauses, then it is likely
introduced by the verb itself, lexically. But if it should matter where the verb
occurs - with the relation required in simple clauses, but not in RCs - then
perhaps it is introduced not by the verb, but by its context.
5 English resultative constructions and verbal valence

Before heading into Igbo and Mandarin, it will be useful to consider English as
a counterpoint. We will see that, in English, the grammar of RCs can be used to
argue that (at least) patients are arguments of the verb.
In English, a verb is typically subject to the same requirements in an RC as
m a simple clause. A verb will require a patient (or theme) when in M, for
example, to the same extent, and under the same conditions, that it requires one
m a simple clause (Dowty 1979: 222, Carrier and Randall 1992: 187, Levin and
Rappaport Hovav 1995: 39, but cf. Boas 2003: 113). The verb yell, for example,
does not require identification of its theme (i.e. that which is yelled) in simple
clauses, (14), and the same is true in RCs, (15).
(14) Al yelled.
(15) Al yelled his throat hoarse.
The verb hammer generally does occur with an object naming the patient of
hammering. But sometimes, particularly when the hammering is repetitive, the
patient may go unexpressed, (16). Again, this is true in RCs as in simple clauses;
(17) does not tells us what was hammered.
(16) Al hammered ? (nails).
(17)
7Al hammered his wrist sore.
Finally, verbs like cut and cany do not tolerate drop of their patients in simple
clauses (18), and the same intolerance is shown in RCs (19). Carrier and Randall
(1992: 187) illustrate the same point for the verb frighten, (20).
(18) (a)
(b)
Al cut *(the frozen meat).

Al carried *(the luggage).
(19) (a)
(b)
(20) (a)
*Al cut the knife dull.

*Al carried his neck sore.
The bears frightened *(the campers).
(b)
*The bears frightened the campground
empty.
Thus in each case the behavior of the verb in RCs corresponds to its behavior in
simple clauses.5
The same pattern governs grammatical relations. A verb in M will find its
thematic relata bearing the same grammatical relations in the RC clause that they
would have in a simple clause. In the simple clauses (21a) and (22a), yell and
pound find their agent in the subject and their (theme or) patient in the object;
the opposite arrangement is impossible, (21b), (22b).
(21) (a)
(b)
(22) (a)
(b)
A! yelled slogans.
*The slogans yelled Al.
Rocky's fists pounded the frozen meat.
* The frozen meat pounded Rocky 's fists.
Just so, neither verb can occur in a RC where the object names its agent and the
subject names its theme, (23), (24). Notice that the intended meanings here are
entirely plausible.
(23) *The slogans yelled Al hoarse.
Intended: 'The slogans made Al hoarse by Iiis yelling them.'
(24) * The frozen meat pounded Rocky 's fists bloody.
Intended: 'The meat made the fists bloody by their pounding it.'
When a verb is subject to the same argument requirements in both simple clauses
and RCs, I will say that it shows uniform projection. And when it is characteristic
of a language that its verbs show uniform projection, I will say that the language
has the uniform projection property, or UPP. Thus English has the UPP. Knowledge of this is revealed by our reaction to a quote attributed to Mormon pioneer
Brigham Young, (25).
(25) '"God almighty will give the United States a pill that will puke them to death,'
Young said during tensions in the late 1850's."
(T. Egan, New York Times, 3 February 2002)
From this unusual sentence we deduce immediately that Young's grammar must
have allowed sentences like (26). Were the UPP not a characteristic of English,
the strength of this inference would be surprising.
(26) This bitter pill will surely puke you.
Uniform projection is explained if argument requirements are stated as lexical

properties of the verb, since they will then be expressed wherever the verb occurs. English RCs have therefore been take to support a projectionist model
5
Essentially the same pattern is found with respect to agent arguments. A verb will require or
refuse an agent to the same degree in M as in simple clauses.
10
Alexander Williams
of argument relations, most emphatically in Levin and Rappaport Hovav 1995

(chapter 2).6
6 Arguments in Igbo and Mandarin

In this section we will see that Igbo and Mandarin do not have the UPP. Systematically, verbs that must cooccur with a patient in simple clause have no such
requirement in RCs.
First let us get a clearer idea of the basic structure of RCs in these languages,
which differs somewhat from that of the RC in English. In English, M is a verb
but R never is. In Igbo and Mandarin, however, M and R are both verbs, roots
that can serve as the sole predicate of a clause without auxiliary support. Thus
the R predicates in (27) and (29) can head clauses on their own, (28) and (30).
Notice that the R verb is moreover not constrained to be stative; in both these
examples, it is eventive.
(27) t ti
duii -le ntio mbii.
3s kick snap -PFV that plank
'He made that plank snap by kicking it.'
(28) ntio nibn dnn -le.
that plank snap -PFV
'That plank snapped.'
(29) O ku
wa -ra
oba ahu.
3sS strike split -FACT gourd that
'He made that gourd split by striking it.'

(30) Oba ahu wa -ra
awa.
gourd that split -FACT BVC
'That gourd split.'
In English, R is phrasal, in that it may contain modifiers in addition to its head,

(31). But the head of R cannot be modified in Igbo and Mandarin, as shown for
Mandami in (32). Thus R is a verbal head simply, and not a pirrase.
(31) AI pounded the cutlet very flat.
(32) t z
(*hn) ping -le nkni ru.
3s pound (*very) flat -PFV that
meat
'He pounded that meat (*very) flat.'
Tense and aspect suffixes follow both verbs in Igbo and Mandami, and do not
attach to M independently. The direct object likewise follows both M and R, and
6
We should appreciate that Levin and Rappaport Hovav's conclusion is persuasive only if English
RCs are complex predicates. If instead the means verb combines immediately with the object
NP, as argued by Carrier and Randall ( 1992), then the local syntactic context of the verb will be
the equivalent in RCs and in simple clauses. A n d in that case we would expect the U P P pattern
whether the arguments project f r o m the verb or not.
11
cannot occur between them, (33), (34).

(33)
*t ti
(-le) ntio rniibn diin (-le).
3s kick ( - P F V ) that plank snap ( - P F V )
Intended: 'He made that plank snap by kicking.'
(34) *0 ku
(-ru)
oba ahu wa (-ra).
3sS strike ( - F A C T ) gourd that split ( - F A C T )
Intended: 'He made that gourd split by striking.'
Mainly because M and R are in this way inseparable, it is widely agreed that
Mandami and Igbo RCs are complex predicates. The means verb combines directly with R, and does not combine first with any noun phrase arguments (see
e.g. Thompson 1973, Y. Li 1990, and Huang 1992 for Mandami; Lord 1975 and
Hale, Ihiono, and Manfredi 1995 for Igbo). I will assmne that, more specifically,
the first node dominating both M and R is a as in (35a), with an intermediate
head introducing the relation of causation between events.7 Details of the semantic derivation will be proposed in section 7; but in outline it will proceed as
in (35b).
(35) (a)
CAUSE
(b)
[Q'] =
V R
[CAUSE]([V
])([VMJ)
6.1 Unrealized patients

In both Mandami and Igbo, a verb that must cooccur with a patient in simple
clauses need not do so when serving as M in an RC.
For Mandarin this observation is commonplace (L.Li 1980, Lii 1986, Ma
1987, Tan 1991, among others). Take the verb qi 'cut,' for example. In simple
clauses, (36), it requires an object naming the patient of cutting. Thus sentences
like (36b) or (36c) can only be analyzed as including a silent object pronomi,
referring to some individual salient in the discourse. They cannot mean simply
that there was an event of Lao Wei cutting something, or that there is such an
event ongoing.
(36) (a)
Lo Wi qi -le zhsn.
L.W.
cut -PFV bamboo shoot
'Lao Wei cut bamboo shoots.'
I have no strong objection to an alternative structure where the node a contains just the two
verbs, and the relation of causation is introduced by semantic rule. But there is some slight
morphological evidence for the presence of the head in Mandarin.
12
Alexander Williams
(b)
(c)
*Lo Wi qi -le.
L.W.
cut - P F V
Intended: cut' (Can mean: 'He cut it.')
*Lo Wi zi
qi.
L.W.
PROG cut
Intended: 'Lao Wei is cutting.'

(Can mean: 'Lao Wei is cutting it.')
Wlien qi 'cut' is the means verb of an RC, however, no such requirement holds.
The RC in (37), for example, can mean just that the subject made the knife dull
by cutting something. No nomi phrase names what is cut.
(37) t hi qi dim -le nde cido.
3s also cut dull -LE your food.knife
'He also made your cleaver dull by cutting.'
(Adapted from Ma 1987: 428)
(37) does not contain a silent object pronomi, referring to the patient. Syntactically the sentence has no space for a second object, (38).
(38)
*ta hi qi dim -le (zhsn) nde cido

(zhsn).
3s also cut dull -LE (bamboo) your food.knife (bamboo)
Intended: 'He also made your cleaver dull by cutting bamboo.'
Pragmatically, moreover, (37) is not constrained to occur only in a context that

would license silent pronominal reference to the patient of cutting. The context
of (39a), for instance, does not license pronominal reference to anything but the
cleaver, yet (39b) is felicitous nonetheless.
(39) (a)
(b)
cido znme hishi

a?
cleaver how happened RT
'What happened with the cleaver?'
Lo Wi qi dim -le [pro].
L.W.
cut dull -PFV it
'Lao Wei made it dull by cutting.'
Should the speaker of (37) want to identify what was cut, this can be done
(among other ways) by adjoining an adverbial verb phrase, as in (40). Yet regardless of whether this addition is required by the conversation, it is not required by
the syntax.8
(40) Lo Wi qi zhsn,
qi dim -le cido.
L.W.
cut bamboo shoots, cut dull -PFV cleaver
'Cutting bamboo shoots, Wei made the cleaver dull by cutting.'
13
Filially we cannot say that the direct object in (37), cido 'cleaver,' is itself an
argument of the means verb. The cleaver is indeed the instrument of the means
event; but in simple clauses qi 'cut' cannot take an instrument as its direct
object, (41).
(41)
*t qi -le ni de cido.
3s eut -LE your cleaver
Intended: 'He cut with your cleaver.'
This pattern is systematic. With few exceptions, any verb in M can occur without
the patient argument required in simple clauses. (42)-(44) give further examples.
(42) w c zng -le lingkui mb.
Is wipe dirty-PFV two
towels
made two towels dirty by wiping.'
(ex. Wang 1995: 148, trans. AW)
(43) t pai
tng -le shu.
3s smack hurt -PFV hand
Can mean: 'He made Iiis hand hurt by smacking [something else].'
(Adapted from L. Li 1980: 98, trans. AW)
(44) t mi -kng -le qanbao.
3s buy -empty -PFV wallet
'He bought (so much that) his wallet (got) empty.'
(ex. and trans. Tan 1991: 100)
It can be shown, just as it was for (37), that none of these RCs includes a nomi
phrase naming the patient (or theme) of the means event; yet in each case M is
a verb that must cooccur with a patient (or theme) argument hi simple clauses,
and cannot take an instrument as its direct object.
Igbo displays the same pattern as Mandami, just as systematically. A verb
required to cooccur with a patient in simple clauses is subject to no such requirement when in M. Take the Igbo verbs bi 'cut' and gwu 'dig out', for example. 9
In simple clauses like (45) and (46), these verbs must cooccur with an argument
nomi phrase that identifies what was cut or what was dug out.
(45) O bi -ri
osisi.
3sS cut -FACT wood
'He cut wood.'
Such VPs are considered adjuncts not only because they can be dropped, but also because they
cannot include aspectual suffixes or modal verbs.
The Igbo data I present here come mainly from primary research I conducted with native speakers
from Nigeria, now living in the Philadelphia area. More information on Igbo RCs can be found in
Lord 1975; Nwachukwu 1987; Uwalaka 1988; Dchane 1993; Hale, Ihionu, and Manfredi 1995;
and Igwe 1999.
14
Alexander Williams
(46) O
gwu -ru
3sS dig
ji.
-FACT y a m
'He dug up yams.'
Unlike Mandarin, Igbo has no silent object pronouns; so (47) and (48) have no
grammatical analysis at all.
(47) (a)
*0 bi -ri
(ebi).
3 s S c u t -FACT ( B V C )
(b)
Intended: 'He cut [stuff].'

*0 na
ebi (ebi).
3 s S PROG SBRD- c u t ( B V C )
(48) (a)
Intended: 'He is cutting [stuff]'

*0 gwu -ru
(egwu).
3sS dig
(b)
-FACT ( B V C )
Intended: 'He dug up [stuff].'

*0 na
egwu (egwu).
3 s S PROG S B R D - d i g
(BVC)
Intended: 'He is digging up [stuff].'
Yet when bi 'cut' and gwu 'dig out' appear in M, there is no need for a patient. (49) and (50) are perfectly natural, despite the absence of any nomi pirrase
identifying what was cut or what was dug out.
(49) O
3sS
'He
(50) O
3sS
'He
bi -kpu -ru
mma.
cut -blunt -FACT knife
made his knife blunt by cutting [stuff].'
gwu -ji
-ri
ogu.
dig.up -snap -FACT hoe
made the hoe snap by digging up [stuff].'
Again, these are not cases of silent anaphora, since Igbo has no silent object
pronouns. Nor do they express alternative argument structures for bi 'cut' and
gwu 'dig up', alternatives which select an instrument rather than a patient as
object. In simple clauses an instrumental object is impossible, (51), (52).
(51)
*0 bi -ri
mma (n' osisi).
3sS cut -FACT knife (P wood)
Intended: 'He cut with a knife (at wood).'
(52) *0 gwu -ru ogu (na ji).
3sS dig.up -rV hoe (P yam)
Intended: 'He dug with Iiis hoe (at yams).'
We can only conclude that the requirement associated with these verbs hi simple
clauses is absent hi RCs. Should the speaker want to identify the patient of the
means event, this can be done by means of an adjunct PP, as in (53) and (54).
(53) O
3sS
'He
(54) O
3sS
'He
15
bi -kpu -ru
mma n' osisi.
cut -blunt -FACT knife wood
made his knife blunt cutting wood.'
gwu -ji
-ri
ogunaji.
dig.up -snap -FACT hoe yarn
made the hoe snap digging up yams.'
But the addition of this information is not syntactically required.

The behavior of bi 'cut' and gwu 'dig out' is in no way exceptional. Verbs
that require a patient in simple clauses do not when in M, quite generally. Three
more examples are given in (55)-(57).
(55) O so -ja
-ra
osisi.
3sS poke -splayed -FACT wood
'He splayed the stick by poking [with it].'
(Can also mean: 'He splayed the stick by poking it.')
(56) O de
-ji
-ri
pensili.
3sS write -snap -FACT pencil
'She made the pencil (nib) snap by writing.'
(57) O bu -lio -ro
olu (' ibu).
3sS carry -sore -FACT neck ( load)
'She made her neck sore by carrying (a load).'
In none of these RCs is there a noun pirrase identifying the patient of the means
event. Yet so 'poke', de 'write', and bu 'carry (on the head)' are all verbs that
require a patient in simple clauses, and cannot take an instrument as object.
6.2 Patients in unexpected places
The suspension of simple-clause requirements is also evident in the correspondence between thematic and grammatical relations. A verb constrained to find
its patient in the direct object of a simple clause may seem to find a patient in
the subject of an RC.
For Mandarin this has been observed in L. Li (1980), Lli (1986), Ma (1987),
Tan (1991), and elsewhere; the most widely known discussions are in Y. Li ( 1990
and 1995). Consider (58) and (59) for example.
(58) (a)
(b)
jiji
x
-le ylf.
elder sister wash -PFV clothes
'Big sister washed (the) clothes.'
*ylfu x
-le jiji.
clothes wash -PFV elder.sister
Intended: 'Big sister washed the clothes.'
16
Alexander Williams
(59) ylfu
li -le jiji.
clothes wash tired -PFV elder.sister
'The clothes made big sister tired from [her] washing [them].'
(ex. Ren 2001: 326, trans. AW)
In simple clauses, (58), the verb xi 'wash' is constrained to find its patient in the
object and its agent in the subject. Yet in the RC (59), the subject is understood
as naming the patient of the means event, and the object, its agent: big sister
washes the clothes. So constraints on the correspondence between grammatical
and thematic relations in simple clauses are apparently voided when the verb is
in M.
Tan ( 1991 ) suggests that sentences like these reflect the possibility of the verb
in M occurring intransitively and nonagentively, as in (60).111
(60)
ylfu
xi
-le.
c l o t h e s w a s h -PFV
'The clothes are washed.'
She then proposes that, in RCs like (59), the means verb occurs in its intransitive
guise, and consequently assigns its patient role to the subject. But this cannot
be correct. Construed as nonagentive intransitives, sentences like (60) have a
result-state interpretation. (60) means that the clothes are in the state that results
from washing, for example. Yet this meaning is no part of (59). (59) does not
mean: 'The clothes being in a washed state made big sister tired.' It means
rather that washing the clothes made her tired. The contribution of the means
verb here is eventive, and not (result-) stative. Thus we should assmne that the
verb in (59) is the eventive transitive of (58) and not the result-state intransitive
of (60).
Now let us turn to Igbo. My interviews with Igbo speakers have hinted that
sentences like Mandami (59) are possible in Igbo as well: transitive RCs where
the subject is the patient of the means event while also being the agent of causation. Of the four speakers I consulted with most regularly, two accepted (61)
and two rejected it.
(61) % Ji alili gwu -ji
-ri
oguya.
yam that dig.out -snap -FACT hoe 3sPOSS
'That yam snapped Iiis hoe by digging [it] out'
For those who accept this sentence, the subject, ji ahu 'that yam,' is understood
as the patient of the means event: the yam is what was dug out. In simple clauses,
however, gwu 'dig out' must find its patient in the object, (62). 11
10
Tan demonstrates that sentences like (60) do indeed have an intransitive analysis, under which
there is no silent pronoun referring to an agent, and the patient NP is the (surface) subject. This
conclusion accords with the consensus in the Chine se-language literature (e.g. Gong 1980), and
with the perspective in Li and Thompson ( 1994); but see LaPolla ( 1988) for disagreement.
(62) (a)
gwu
-rii
17
ji.
3 s S d i g . o u t -FACT y a m
(b)
'He dug out yams.'

*Ji aliu gwu
-ru
ya.
y a m t h a t d i g . o u t -FACT 3s
Intended: 'He dug out that yam.'
So for some speakers of Igbo, constraints on the correspondence between thematic and grammatical relations are relaxed when a verb appears in an RC. I do
not know what to make of the disagreement among speakers. But it is interesting
that (61) was sensible to any speakers at all. Contrast the English caique in (63),
which provokes only bafflement.
(63) *Thatyam
dug his hoe apart.
6.3 Summary of the data

The thematic relations an Igbo or Mandarin verb must enter in simple clauses, it
need not enter when in the means predicate of an RC.12 Correspondingly, while
the interpretation of subject and object is fixed with respect to the verb in a
simple clause, it is largely free with respect the means verb in an RC.
At the same time, two correlated aspects of interpretation remain fixed. The
subject names the agent of the event of causation, and the object names the thing
caused to enter the result state defined by R. Unlike thematic relations to the
means event, these relations to the event of causation are never reversed (Y. Li
1995). The understood causer, for example, is never named by the object. And
while it may happen that the sentence has no noun phrase naming the patient
11
One of the speakers who accepted (61 ) also accepted (i). This sentence could not be tested with
m y other consultants, however, as their dialects do not include the verb no 'tired, sore' (Green and
Igwe 1963: 232, Igwe 1999: 559).
(i)
Ibu bu
-no -ro
ya olu.
load carry -sore -FACT 3s neck
'The load made his neck sore f r o m carrying.'
Here M is bu 'to carry on the head', and the subject names what is carried. But this is impossible
when bu is on its own, ( ii ).
( ii )
*Ibu bu -ru ya.

load carry -FACT 3s
Intended: 'He carried the load.'
The speaker w h o accepted ( ii )grew up in the Isu-ikwu-ato region of an area now known as A m biya, formerly a part of Imo State. The dialect studied in Green and Igwe ( 1963) was spoken
"near Umuahia by the people known as O h u h u " ( 1963: xiii).
12
It is also true that certain HOH-semantic requirements shown in simple clauses m a y be suspended
when a verb is in M. Verbs that are required to cooccur with a certain semantically vacuous object
noun phrase in simple clauses are under no such obligation when in M. This is a major topic in
the two seminal papers f r o m which the present work derives, Thompson ( 1973 ) and Lord ( 1975 ).
Yet I lack the space to discuss it here.
18
Alexander Williams
of the means event, the 'causee' in the event of causation is always identified,
namely by the phrase that controls R.
A theory of Igbo and Mandarin must therefore answer three questions. Why
does the observed degree of freedom in interpretation obtain only in RCs? W i y
is interpretation in RCs free only with respect to the means event? And how are
Igbo and Mandami are different from English? I believe the only explanatory
answers to these questions are provided by the theory I will now describe, the
No Argument Theoiy for Igbo and Mandami, or NAT.
7 The No Argument Theoiy

The facts of section 6 follow directly if we assmne that patients, as well as
agents, are not arguments of the verb in Igbo and Mandami. 13 The typical transitive verb in these languages characteristically has no arguments lexically; it
simply denotes a sortal on events, as illustrated in (64) for verbs meaning 'cut.'
(64)
Mandarin 'cut':
Igbo 'cut':
\qie\
[6]
=
=
Xe.cut(e)
e. cut. (e)
Correspondingly, thematic relations are introduced by the environment the verb

occurs in. Kratzer's (1996) proposal for introducing agents structurally is familiar and I will adopt it here. (65) sketches the proposal by pairing syntactic nodes
with their interpretation. head denoting the agent relation combines with
VP, and the resulting is interpreted by a rule known as "event identification."
(65)
* . 3 e [ [ V P J ( e ) ( ) (e)]
VP
VAG
yei.AG(ei) = y
I add that, in Igbo and Mandarin, patient relations are also introduced structurally. This is done, I will assmne, by means of a semantic rule that applies at
VP, as in (66). 14
(66)
VP
e.[V](e)APAT(e) = [DP]
DP
13
14
Lin (2001 ) arrives at similar conclusions for Mandarin, but by a very different route.
In (66) I have the direct object preceding the verb, on the assumption that verb raising will derive
the correct surface order. As nothing here depends on this, however, the reader is free to reject my
assumption, and to presume instead that V precedes DP underlyingly.
19
Others might prefer to posit a head that denotes the patient relation, combining
this with the verb by event identification, just as in the case of agents.
For a simple clause whose verb is Mandarin qi 'cut' or Igbo bi 'cut,' (66)
yields (67a) as the denotation for VP. Plugging this into (65) then yields (67b)
for .
(67) (a)
[ [VP D P qi ] ] = [ [ V P D P bi ] ] =
e. cut (e)
(b)
PAT(e) =
[ [' *>AG [ V '
\x\e.cut{e)
D P qi]]
[DP]
] = [ [, I'AG [ V ' D P bi ]] ] =
PAT(e) = [DP] AG(e)
=x
(67b) states directly that the object DP is the patient of pounding, and the subject,
when it comes along, will be its agent. The grammar thus predicts correctly that
the interpretation of subject and object in simple clauses will be fixed.
But crucially, the same grammar yields a vague interpretation for subject and
object in a RC, given two ordinary assumptions: the M verb and R constitute a
complex predicate, and this predicate has the distribution of a simple verb.
If the M verb fonns a complex predicate with R, it does not combine first
with an object. In the present context, this means it does not first enter any
structure that introduces a patient. Moreover, the minimal assumption about the
semantics of combining M and R is that it introduces no content beyond the
relation of causation. The smallest constituent containing both M and R, then,
has the interpretation in (68). 15 I assiune, recall, that the CAUSE' relation here
is introduced by a silent head, CAUSE, located between the two verbs; this head
can be taken to denote as in (69).
(68)
[ M R ] = Ae.3ei3e2[CAUSE'(e,ei,e2) [M](ei) [R](e2)]
(69)
[CAUSE] = ATCA_MAe.3ei3e2[CAUSE/(e,ei,e2)
M(e)
TZ(e2)]
Given the lexical denotations in (70a), therefore, the complex predicates qi dim 'cut
dull' and bi kpu 'cut dull' will denote as in (70b).
(70) (a)
(b)
[ dim ] = [ kpu ] = Ae.dull(e)

[ qi dim ] = [ bi kpu ] = A e . 3 e i 3 e 2 [ C A U S E ' ( e , e i , e 2 )
A cut(e 1) dull fa)]
15
(68) is similar to Rothstein's rule of "resultative conjunction," (i).

(i)
Resultative conjunction (Rothstein 2001: 158)
A + RB = A ' j / A e . 3 e i 3 e 2 [ ( e = e i U e 2 ) ( C U L (e ) . e 2 ) A f a , Y) B f a , y ) \
(i) presupposes an analysis of what (68) has as "CAUSE'" into a sequence of two relations, namely
the first two conjuncts in the body of the formula. But we are free to import this analysis into ( 68 ).
The only real difference between ( 68 ) and ( i ) is that ( i ) identifies the presumed internal arguments
of M and R by lambda-abstraction, while (68) includes no such operation, since it combines verbs
that have no internal arguments. If this difference is factored out, (68) and (i) can be seen as
equivalent.
20
Alexander Williams
The RC predicate thus denotes a predicate true of events e wherein one event ei
causes another e 2 - but it specifies no thematic relations to the means or result
events individually.
Now let us assmne that the minimal RC predicate has the same syntactic
distribution as a simple verb. This assumption is common in the literature, where
Igbo and Mandami RCs are often described as compound verbs. Here it means
that complex predicates like qi dim 'cut dull' and bi kpu 'cut dull' will occur
in the V slot of the VP structure in (66), yielding (71). Plugging this into the
structure of (65) yields (72) in turn.
(71) [ [VP DP [V qi dim ]] ] = [ [Vp DP [ v bi kpu ]] ] =
Ae.3ei3e2[CAUSE'(e, ei, e 2 ) cut (e ) dull fa)
PAT(e) = [DP] ]
(72) [ [ T/ [VP DP [V qi dim ]]]] = [ [, [VP DP [ v bi kpu ]]] ] =
AiAe.CAUSE'(e, ei, e 2 ) cut (e ) dull fa)
PAT(e) = [DP] AG(e) =
The VP and structures introduce thematic relations. But as a matter of locality,
these relations predicate of the mam event of causation, and not of its subevents
of means and result. The semantics thus tells us that the subject is the agent of
causation and the object is its patient, but says nothing explicit about their relations to either the means or the result events. Interpretation with respect to these
events is consequently free - except insofar as it is constrained, semantically and
pragmatically, by being the agent and patient of a certain event of causation.
This predicted degree of vagueness is exactly what the Mandarin and Igbo
data show, I suggest. 16 The subject and object of a RC may be construed as
bearing any plausible thematic relation to the means event, or no relation at all,
because the semantic representation insists on none in particular.
Construal with respect to the result event, on the other hand, is limited by the
one semantic constraint that seems natural. Any definition of the basic predicates
PAT and CAUSE' should have (73) as a theorem.
(73) If is the patient of e m causing e r , then is the patient of e r . 17
So if a plank is the patient of kicking causing snapping, then the plank is the
patient of snapping, and hence winds up snapped. This is simply what it means
to be the patient of an event of causation. Parsons makes essentially the same
claim for his "Themes" of "BECOME" events - which, after all, can be regarded
16
17
Sybesma ( 1999) has similarly suggested that vagueness is what is behind the facts of Mandarin,
though his analysis of the RC is otherwise different.
In case e r is a state, rather than an event of state change, we will have to consider the patient of a
state as its holder. If this is unacceptable, we can simply restate ( 73 ) less gracefully as (i ).
(i)
If is the patient of e m causing e r , then is the holder of the result state in e r .
21
as events of causation with no means event or agent specified (Dowty 1979,

Parsons 1990): "The Theme of [BECOME's] event is the same as the Theme of
its Target state: BECOME(e,s) -* [Theme(e,z) Theme(s,x)]" (Parsons 1990:
119).
Given this semantics, it follows definitionally that the direct object in an RC,
because it necessarily names the patient of the CAUSE' event, also controls the
result predicate R. Take (74), for example.
(74) (a)
(b)
Lo Wng ti duii -le ntio mbii.

L.W.
kick snap -PFV that plank
'Lao Wang made that plank snap by kicking.'
[(74a)] = x e . C A U S E ( e , e i , e 2 ) kick (e )
PAT(e) = plank AG(e) = laowang
snap(e2)
Here the object controls R, but not because the denotation in (74b) states any
relation between the plank and the snapping. Rather, it establishes a patient relation between the plank and the event of kicking causing snapping. The relation
to the snapping event, in virtue of which we say that the direct object controls R,
is a definitional consequence.
Evidently the meaning of CAUSE' does not entail identity between the agent
or patient of causation and any particular participant in the means event. But
there do seem to be default inferences; strongest among them, the inference that
the agent of causation is in general the agent of the means event. Apparent variation in the strength of this inference cross-linguistically is discussed in Williams
(2005).
8 Attractions and alternatives

Two aspects of the NAT are attractive. First, it relies on no special valencereducing operations, posited ad hoc in the RC context, without morphological
motivation. The account derives just from defining the lexical primitives, and
observing that RCs are complex predicates, at least in Mandami and Igbo.
Second, it implies a natural point of cross-linguistic difference. We can assume that Igbo and Mandarin differ from English just in the lexical valence of
verbs which describe the same sort of event, (75). In English the patient is an
argument of the verb, and consequently English shows uniform projection (see
sections 4 and 5).
(75) (a)
(b)
(c)
Mandarin'cut': \_qie\ =
Xe.cut.(e)
Igbo 'cut': [fa'] = e. cut (e)

English 'cut': [c*rf] = Xx . . . Xe.[cut(e) PAT(e) = . . .}
That verbs with similar meaning may differ in apparent valence is a familiar
observation. Discuss and argue describe very similar activities, but only discuss
22
Alexander Williams
requires a direct object to identify the topic of conversation. Wliat the NAT asks
us to assume is just that languages may exhibit characteristic differences in how
many arguments they assign to a verb lexically, within the range allowed by the
number of participants in its event. 18 This seems a plausible assumption.
Now let us consider alternative accounts. How might one model the Mandarin and Igbo data while assuming, contra the NAT, that (at least) patients are
arguments of the verb? I see three clear possibilities, but I think they all fail as
explanations.
First, we could say that each verb has multiple lexical argument structures,
but most are permitted only in the M context. Perhaps xf 'wash' has several
lexical entries, for example, corresponding to the several denotations in (76),
but only the entry with denotation (a) occurs freely. The others are constrained
to occur only in M.
(76) [ * i ] =
(a)
XyXxXe.wash{e) A PAT(e) = y A AG(e) =
(b)
(c)
\x\e.wash(e)
\e.wash(e)
(d)
\x\y\e,wash(e)
A AG(e) =
A PAT(e) = y A AG(e) =
Second, we might keep lexical verbs unambiguous, granting them only those
argument structures that are manifested in simple clauses, and locate ambiguity
in the complex predicate instead. The same pair of unambiguous verbs in M
and R, that is, might yield a complex predicate with several distinct argument
structures, (77). These differ in the thematic relations they establish between the
means event and the subject or object referents.
(77) [ xli ] =
(a)
Aj/AiAe.CAUSE'(e,ei,e 2 ) (wash(ei)
A PAT(ei) = y
A AG(ei) = x) A (tired{e 2 ) A PAT(e 2 ) = y)

(b)
A j / A i A e . C A U S E ' ( e , e i , e 2 ) (wash(ei)
A AG(ei) =
x)
A (tired(e2) A PAT(e 2 ) = y)
(c)
Aj/AiAe.CAUSE'(e,ei,e 2 ) (wash(ei)
A PAT(ei) =
A AG(ei) = y) A (tired(e2) A PAT(e 2 ) = y)

The operation of resultative predicate formation would then not define a function. More specifically, it would have the effect of arbitrarily permuting or deleting the lexical arguments of the means verb. A version of this solution was
developed by Y.Li (1990, 1995).
Finally, formation of an RC predicate might suppress the lexical arguments of
the verb in M, through deletion or existential binding. The scheme for interpret18
The NAT itself says nothing about whether languages differ in how many participants in the verb's
event must be identified in a simple clause. It says only that languages may differ in how many
arguments in a simple clause are lexical arguments of the verb.
23
ing RC predicates might be as in (78), for example, where 3z binds a presumed

lexical argument of M.
(78)
[MR] =
Aj/Ae.3;3ei3e2[CAUSE'(e,ei,e2) [M]()(ei) [R](j/)(e2) ]
We would then be free to assume that xi 'wash' does have a patient argument
lexically, since suppression of this argument under complex predicate formation
would ensure that it is assigned to no pirrase in the RC clause. Any understood
thematic relation to the means event would be regarded as the result of inference,
just as proposed within the NAT (see also Sybesma 1999).
This last alternative is the most attractive. It neither multiplies dubious lexical
entries nor introduces a nonfunctional operation into the grammar. But it shares
with the other alternatives one basic problem.
Each proposes that the M context is somehow special. It licenses argument
structures not otherwise licit; it allows the verb's lexical arguments to be permuted; or it supresses them altogether. But why should the M context have these
effects? More pressingly, why should it have these effects in Igbo and Mandarin
but not English? If M's arguments are existentially bound in Mandarin and Igbo,
for example, why shouldn't the same be true in English? Unless these questions
find a good answer, the descriptive postulates of all three alternatives will seem
ad hoc.
I believe there is no good answer, no independent feature of the M context,
just m Mandarin and Igbo, that should have any special effect on the argument
structure of its occupant. Sometimes changes in valence are linked to changes
m aspectuality (e.g. eventive versus stative), or in what event the verb describes
(e.g. a spontaneous change versus one wrought by an agent). But it is clear
that no such change affects the means verb in Igbo and Mandarin. Or changes
m valence accompany changes in lexical category; Dowty (1989), for example,
suggests that verbs lose their arguments under lexical nominalization. But there
is no evidence that the lexical category of a root is different in M than in simple
clauses. And finally, there is no fonnal indication that passive or antipassive
operations apply to the means verb in Mandarin or Igbo. Only one aspect of
RCs m these languages has any allure as an explanatory factor: they, unlike the
RCs of English, involve a compound of two verbal heads. Yet I will show in
section 9 that this prospect too is a dead end.
The alternatives are therefore empirical failures. So long as we presmne that,
m Igbo and Mandami as in English, patients are arguments of the verb, the RC
data cannnot be explained. Yet once this presumption is removed, an explanation follows, just from the agreed fact that RCs are complex predicates with the
distribution of simple verbs.
24
Alexander Williams
9 Size does not matter

English allows R to be phrasal, but Igbo and Mandarin do not. For this reason
Igbo and Mandarin RCs are often described as compounds, and sometimes as
compounds fonned 'in the lexicon.' If we needed to claim that the M context of
Mandarin and Igbo has special effects on argument structure, this difference in
syntax might seem to promise an explanation of why. But in fact it could provide
no explanation, for three reasons. 19
First, there is no a priori reason that combining two lexical heads, whether
m the lexicon or in the syntax, should cause the argument structures of either
one to be modified or suppressed. Any such effect would have to be stipulated
specially, and on no clear basis.
Second, the required stipulation would conflict with many studies of compounds, which have found it useful to assiune that, if a verb has argument structure, it is preserved under compounding. The interpretation of compounds like
English god-fearing, for instance, has often been explained by assuming that the
root verb (here,fear) maintains its argument structure, and assigns the noun its
internal thematic role (see Grimshaw 1990).
Third and most importantly, there is direct evidence from Mandarin that the
size of R is not what matters. What accounts for the lack of uniform projection
is the formation of a complex predicate, regardless of whether its secondary
predicate is a head or a phrase. The evidence comes from another complex
predicate type in Mandarin, called the V-de construction.
The V-de construction consists of a verb, transitive or intransitive, bearing the
enclitic -de, followed possibly by a noun phrase (NP 0 ) and necessarily by a verb
phrase (VP2), (79). V P 2 is controlled by N P 0 when present, and otherwise by
the nearest NP outside V P i .
(79)
V-ife c o n s t r u c t i o n : [VPlV-de
(NP0) VP2 ]
There are two or three subtypes of V-de construction, differing in what semantic
relation holds between the meanings o f V and VP 2 (Huang 1988, Lamarre 2001,
Yue 2001). In one the meaning is roughly causative, and here glossing -de as
'such that' yields an appropriate paraphrase.
(80)
t h n
-d w o m e n d u l o x i -le
3s s c r e a m -DE w e
all
fall
ynli.
-PFV t e a r
' H e s c r e a m e d s u c h t h a t w e all s h e d tears.'

(ex. L . Li 1963: 4 0 5 , trans. A W )
Two major studies of the V-de construction are L. Li (1963) and Huang (1992).
19
Most likely the size of R does explain a regular difference in word order. Among SVO languages
with RCs, those that constrain R to be a head place the direct object after both M and R ( Sbj M R
Obj), while the others have the object interceding (Sbj M Obj R); see Williams (2005).
25
Both conclude that V and VP 2 form a complex predicate to the exclusion N P 0 ; V

combines first with VP2 and then the result combines with NP 0 , 2 " Underlyingly,
therefore, the V-de construction is isomorphic to the verb-verb RC. Both are
complex predicates; they differ just in the size of their secondary predicates.
The NAT therefore predicts that, in V-de constructions as in an RCs, a verb
will not be subject to the cooccurrence requirements it is subject to in simple
clauses. And this is correct. Compare (81) and (82). In the simple clause (81),
kii 'praise' must cooccur with a patient, but in the V-de construction (82), it
need not.
(81)
*w kii
-le.
Is
praise -PFV
Intended: praised.' (Can mean: praised him/her.')
(82) w pi
Lo Wi -de mpi,
kn
-d lin ta titi y
Is smack Lao Wei's
bhoyis
horse-rump, praise -DE even his wife also embarrassed
le.
PRT
'Flattering Lao Wei, I praised [him] such that even his wife got embarrassed.'
(82) cannot be analyzed as containing a silent pronomi, serving as the object of

kii and referring to the understood recipient of praise. Syntactically there is no
space for such a pronomi, either before the verb or after, (83).
(83)
*w pi
Lo Wi -de mpi,
(t) ku
(t) -d (t) lin t titi
Is
smack L.W.'s
horse rump, (him) praise (him) -DE (him) even his wife
y bhoysi
le.
also embarrassed PRT
Intended: Same meaning as (82).
We also find that, again, notional thematic relata may be found hi unusual syntactic positions, as hi (84).
(84) wandu chi -d rn
ti fa run.
peas
eat -DE people legs go soft
'Peas make people go weak in the knees from eating them.'
(L.Li 1963:405, quoting Liu Ke)
Here the understood patient of eating, wandu 'peas,' is the subject of the clause,
and the understood agent is the object, rn 'people'. This arrangement is not
possible hi simple clauses, (85).
20
For Huang (1992) the surface discontinuity of the predicate is an effect of verb-raising, which
here applies to V alone. Huang also regards what I label VP2 as a clause whose subject is a silent
anaphor, controlled by the nearest noun phrase.
26
(85)
Alexander Williams
*wandu chi rn.

peas
eat people
Intended: 'People eat peas.'
One can plausibly object that wanddu 'peas' in (84) is a topic, whose thematic
relation to chT 'eat' is only inferred and not assigned grammatically. But in that
case chTenters no patient relation in (84), and this is itself significant, since in
simple clauses the patient relation is required, (86).
(86)
*Lo Wi chi -le.

L.W.
eat -PFV
Intended: 'Lao Wei ate.' (Can mean: 'Lao Wei ate it.')
The fact that Mandami verbs seem to lose their arguments in verb-verb RCs is
thus part of a larger pattern. Their arguments seem to get lost in any complex
predicate, whether its secondary predicate is a single verb or a pirrase. One
cannot use the size of R to explain the lack of uniform projection in Mandarin,
therefore, without missing a major generalization.
Given this, I will assmne that the size of R cannot explain why Igbo verbs do
not show uniform projection either, or why English verbs do. This seems to be
the null hypothesis.
10 Might patients be a vice?

There is much doubt about the semantic legitimacy of a generalized patient
(or theme) relation. More so than agents, the presmned patients of distinct
event-types share few distinguishing properties (Parsons 1990, Dowty 1991).
As Kratzer (2003) observes, we generally cannot recognize the patient of an
event except under a particular description of the event imposed by the verb.
Thus it is unclear whether a highly general patient predicate could be given truth
conditions of any substance. Correspondingly, if there is such a predicate, it will
surely require that events are individuated to a very fine grain - fine enough that
each event will, so to speak, wear its patient on its sleeve (cp. Parsons 1990 and
Landman 2000 on agents). Yet such finegrainedness tends to undermine a motivating ambition in semantic theory, that of relating language to a denotational
domain of significantly independent structure. For this reason, the idea of a basic
predicate that denotes a generalized patient relation is often considered suspect.
With this paper I mean only to cast an opposing doubt, based on the distributional facts of Igbo and Mandarin. In these languages patient arguments seem
to be introduced syntactically, and not only in a small group of special cases.
If introducing an argument syntactically means introducing a thematic predicate
mto the semantic derivation (as is commonly but not always assumed 21 ) then the
27
grammars of Igbo and Mandarin do include patient as a basic predicate. And if

this is correct, then perhaps our semantic ambition should be moderated.22
11 Conclusion
The grammar of resultatives in Igbo and Mandami is explained directly if patients in these languages are not lexical arguments of the verb. The explanation
is attractive because it does not require any special operations on argument structure that apply in Mandami and Igbo only. There is no independent indication
that such operations do apply, and if they were to be postulated, it would be hard
to say why they don't apply in English resultatives as well. By adopting the
No Argument Theory, therefore, we afford ourselves an account of resultative
structure that is cross-linguistically more uniform. The source of the observed
variation is relocated to the lexicon; or, more precisely, to differences in whether
a certain argument type is introduced by the verb or by the structure it occurs in.
These conclusions imply two claims of general relevance. First, we need to
distinguish between what sort of event a verb describes, and what combinatory
requirements are associated with the verb lexically. A verb need not have as
many lexical arguments as its event has thematic participants. The idea that it
should has guided much research, both grammatical and psycholinguistic. But
if I am right, there is empirical evidence against it. Second, we need to include
patient in the inventory of basic thematic predicates, despite semantic arguments
to the contrary.
References
Bhatt, Rajesh, and David Embick (2004): Causative derivations in Hindi. Unpublished
manuscript. University of Texas at Austin and University of Pennsylvania.
Boas, Hans (2003): A Constructional Approach to Resultatives. Stanford: CSLI.
Borer, Hagit (2003): Exo-skeletal and endo-skeletal explanation: Syntactic projections
and the lexicon. In The NaUire of Explanation in Linguistic Theory, John Moore and
Maria Polinsky (eds.). Chapter 3. Stanford: CSLI.
Carlson, Greg (1984): Thematic roles and their role in semantic interpretation. Linguistics 22: 259-279.
Carrier, Jill, and Janet H. Randall (1992): The argument structure and syntactic structure
of resultatives. Linguistic Inquiry 23: 173-233.
21
22
For Rothstein (2001 ), syntactic introduction of an argument does of mean introducing a thematic
relation. It just means abstracting over a designated variable in a structured denotation. If this
view were preferred - and if the NAT could be implemented in its terms without formal difficulty
(which is not perfectly clear) - then it might be possible to state the NAT without making any
reference to a patient relation. I lack the space to discuss this possibility here.
With Landman (2000), we might accept that a denotational domain of fine-grained events is linguistically necessary, but proceed to relate this domain to one of coarser-grained "situations," by
construing events as properties of situations.
28
Alexander Williams
Davis, Henry, and Haniida Demirdache (2000): On lexical meanings: Evidence from
Salish. In Events as Grammatical Objects, Carol Tenny and James Pustejovsky (eds.),
97-142. Stanford: CSLI.
Dchane, Rose-Marie (1993): Predicates across Categories: towards a Category Neutral
Syntax. Amherst [Mass.]. Doctoral diss.
Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Reidel.
Dowty, David (1989): On the semantic content of the notion of 'thematic role.' In Properties, types and meanings, II, Gennaro Chierchia, Barbara H. Partee, and Raymond
Turner (eds.), 69-130. Dordrecht: Kluwer.
Dowty, David (1991): Thematic proto-roles and argument selection. Language 67 (3):
547-619.
Emenanjo, Nolue (1978): Elements of M o d e m Igbo Grammar. Ibadan: University Press
Limited.
Goldberg, Adele (1995): Constructions. Chicago: University of Chicago Press.
Gong, Qianyan (1990): Xiandai Hanyuli-de shoushi zhuyu j u [Patient-subject clauses in
M o d e m Chinese]. Zhongguo Yuwen 1980 (5): 335-344.
Green, M. M., a n d G . E. Igwe (1963): A Descriptive Grammar of Igbo. Berlin: Akademie
Verlag.
Grimshaw, Jane (1990): Argument Structure. Cambridge, MA: The MIT Press.
Hale, Ken, U. Peter Ihionu, and Victor Manfredi (1995): Igbo bipositional verbs i n a syntactic theory of argument structure. In Theoretical Approaches to African Linguistics,
Akinbiyi Akinlabi (ed.), 83-107. Trenton: Africa World Press.
Hoekstra, Temi (1988): Small Clause results. Lingua 74: 101-139.
Huang, C.-T. James. (1988): Wo pao de kuai and Chinese phrase structure. Language 64:
274-311.
Huang, C.-T. James. (1992): Complex predicates in Control In Control and Grammar,
Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham (eds.),
109-147. Dordrecht: Kluwer.
Igwe, G. Egemba (1999): Igbo-English Dictionary. Ibadan: University Press PLC.
Kayne, Richard (1985): Principles of particle constructions. In Grammatical Representation, Jean Mark Gawron, Hans G. Obenauer, and Jean-Yves Pollock (eds.) 101-140.
Dordrecht: Foris.
Kratzer, Angelika (1996): Severing the external argument from its verb. In Phrase Structure and the Lexicon, Johan Rooryck and Laurie Zaring (eds.), 109-137. Dordrecht:
Kluwer.
Kratzer, Angelika (2003): The Event Argument. Manuscript from
www.semanticsarcliive.net.
Lamarre, Christine (2001): Verb complment constructions in Chinese dialects: Types
and markers. In Sinitic Grammar, Hillary Chappell (ed.), 85-120. Oxford: Oxford
University Press.
Landman, Fred (2000): Events and Plurality. Dordrecht: Kluwer.
LaPolla, Randy J. (1988): Topicalization and the question of the lexical passive in Chinese. In Proceedings of the Third Annual Conference on Chinese Linguistics, Mar-
29
jorie Chan and Thomas Emst (eds.), 170-188. Bloomington: Indiana University
Linguistics Club.
Larson, Richard (1991): Some issues in verb serialization. In Serial Verbs, Clair Lefebvre
(ed.), 185-210. Philadelphia: John Benjamins.
Levin, Beth, and Malka Rappaport Hovav (1995): Unaccusativity. Cambridge, MA: The
MIT Press.
Li, Charles N., and Sandra A. Thompson (1994): On 'middle voice' verbs in Mandarin.
In Voice: Forni and Function, Barbara Fox and Paul J. Hopper (eds.), 231-246.
Philadelphia: John Benjamins.
Li, Linding (1963): Dai de zi de buyu ju [Sentences with complements including the
morpheme de]. Zhongguo Yuwen 1963 (5): 396-410.
Li, Linding (1980): Dongbu ge jushi [Sentences with the verb-complement construction].
Zhongguo Yuwen 1980 (2): 93-103.
Li, Yafei (1990): On V-V compounds in Chinese. Natural Language and Linguistic Theory 8: 177-207.
Li, Yafei (1995): The thematic hierarchy and causativity. Natural Language and Linguistic Theory 13: 255-282.
Lidz, Jeffrey, Henry Gleitman, and Lila Gleitman (2003): Understanding how input matters: The footprint of universal grammar on verb learning. Cognition 87: 151-178.
Lin, Tzong-hong (2001): Light verb syntax and the theory of phrase structure. Ph.D.
diss., Department of Linguistics, University of California at Irvine.
Lord, Carol (1975): Igbo verb compounds and the lexicon. Studies in African Linguistics
6 (1): 23^18.
Lii, Shuxiang (1986): Hanyu jufa de linghuoxing [The flexibility of Chinese sentence
grammar]. Zhongguo Yuwen 1986 (1): 1-9.
Ma, Xiwen (1987): Yudonjieshi donci youguan de mouxie juzhi [Some sentence patterns
relevant to verbs in the verb-result construction]. Zhongguo Yuwen 1987 (6): 424441.
Marantz, Alec (1997): No escape from syntax: Don't try morphological analysis in the
privacy of your own lexicon. U. Penn Working Papers in Linguistics 4 (2): 201-226.
Nwachkwu, P. Akujuobi (1987): The argument structure of Igbo verbs. Technical Report
18, MIT Lexicon Project Working Papers.
Parsons, Terence (1990): Events in the Semantics of English. Cambridge, MA: The MIT
Press.
Ren, Ying (2001): Zhubin kehuanwei dongjieshi shuyujiegou fenxi [An analysis of verbresult complement constructions where the subject and object can switch places].
Zhonguo Yuwen 2001 (4): 320-328.
Rothstein, Susan (2001): Predicates and Their Subjects. Dordrecht: Kluwer.
Simpson, Jane (1983): Resultatives. In Papers in Lexical Functional Grammar, Malka
Rappaport, Lori Levin, and Annie Zaenen (eds.), 143-157. Bloomington: Indiana
University Linguistics Club.
Swift, L.B., A. Ahaghotu, and E. Ugorji (1962): Igbo Basic Course. Washington, D.C.:
Foreign Service Institute.
Sybesma, Rint (1999): The Mandarin VP. Dordrecht: Kluwer.
30
Alexander Williams
Tail, Fu (1991): Notion of subject in Chinese. Ph.D. diss.. Department of Linguistics,

Stanford University.
Thompson, Sandra A. (1973): Resultative verb compounds in Mandarin Chinese: A case
for lexical rules. Language 42 (2): 361-344.
Uwalaka, Mary Angela A.N. (1988): The Igbo Verb: A Semantico-Syntactic Analysis.
(Beitrage zur Afrikanistik, Band 35.) Vienna: Institute fr Afrikanistik mid Aegyptologie der Universitt Wien.
Van Valili, Robert D., and David P. Wilkins (1996): The case for 'Effector': Case roles,
agents and agency revisited. In Grammatical Constructions: Their Forni and Meaning, Masayoshi Shibatani and Sandra A. Thompson (eds.), 289-322. Oxford: Clarendon.
Wang, Hongqi (1995): Dongjieshi shubu jiegou peijia yanjiu [Studies on the valence of
resultative complement constructions]. In Xiandai Hanyu Peiji Yufa Yanjiu, Yang
Shen and Ding'ou Zheng (eds.), 144-167. Beijing: Beijing University Press.
Williams, Alexander (2004): Intransitive resultatives and Igbo. Paper presented at the
74th Annual Meeting of the Linguistic Society of America.
Williams, Alexander (2005): Complex causatives and verbal valence. Ph.D. diss., Department of Linguistics, University of Pennsylvania.
Yue, Anne (2001): The verb complement construction in historical perspective with special reference to Cantonese. In Sinitic Grammar, Hillary Chappell (ed.), 232-265.
Oxford: Oxford University Press.
Jo-wang Lin (Taiwan)
Event decomposition and the syntax and semantics of

durative phrases in Chinese
1
Introduction
This paper deals mainly with the syntactic distribution and semantics of durative
phrases in Chinese. It has been pointed out that durative phrases in Chinese may
be interpreted differently depending upon the situation type of the sentence that
they modify. According to Ernst (1987) and Li (1987), one interpretation of
durative phrases is that they measure the duration of an event such as (1). This
interpretation obtains when the situation described involves no change of state
or non-completion of the event. By contrast, when the situation described involves a change of state or the completion of an event, they claim that durative
phrases are interpreted as describing the time elapsed since completion of the
event as the translation in (2) indicates.
(1)
Women
we
zou-le
liang-ge
walk-Asp two-CL
xiaoshi
hour
'We walked for two hours.'

(2)
Ta yijing jiehun
san
nian le
he already get-married three year PAR

'It has been three years since he got married.'
With regard to the interpretations of durative phrases in Chinese, two remarks

are in order here. First, unlike English durative /or-phrases which are generally
incompatible with change of state verbs unless the sentence is coerced into a
repetitive or iterative interpretation, it is very natural for Chinese durative
phrases to occur with such verbs without meaning coercion. Second, although
An earlier version of this paper was presented at the second Workshop on Formal Syntax and
Semantics in Academia Sinica in Taipei on September 27-28, 2003 and at the Workshop on
Event Structures in Linguistic Form and Interpretation at the University of Leipzig on March
17-19, 2004. I would like to thank the participants for their questions and comments. I am also
grateful to Waltraud Paul for her written comments, though I am not able to incorporate most of
them into the text due to constraint of space. This work was supported by National Science
Council of Taiwan, grant No. 93-241 l-H-009-011.
32
Jo-wang Lin
Emst (1987) and Li (1987) have described the second reading of durative
phrases as "the duration since completion of an event" (SCE reading), it is more
correct to say that this reading involves measuring the duration of the consequent state resulting from an event. 1 In fact, one of the goals of this article is to
clarify the interpretation of durative phrases which occur with change of state
verbs. I will refer to durative phrases associated with a change of state verb as
R-related (result-related) duratives, pending my arguments in a later section. In
turn, I will refer to the first reading of durative phrases as the P-related (process-related) interpretation.
P-related and R-related durative phrases raise many interesting semantic and
syntactic questions. For example, semantically speaking, do the two interpretations represent lexical ambiguity of durative phrases, as Ernst's (1987) and Li's
(1987) description of the two readings as the 'since completion' and 'event
duration' readings might imply? It is the goal of this paper to show that durative
phrases in Chinese are not lexically ambiguous.
Syntactically, R-related duratives differ from P-related duratives in that the
latter may precede or follow the direct object, whereas the former can only
follow the direct object. This is illustrated by the examples in (3) and (4).
(3)
(a)
Wo
(yijing) kai
jichengche ershi
nian
le
already
taxi
year
PAR
drive
twenty
have already driven a taxi for twenty years.'

(b)
Wo
yijing
kai
ershi
already
drive
twenty year
nian jichengche
taxi
le
PAR
have driven a taxi for twenty years.'

(4)
(a)
Women yijing
dida
zhongdian
shi
fengzhong
le
we
reach
destination
ten
minute
PAR
already
'It's already been ten minutes since we reached the destination.'

(b)
*Women
we
yijing
dida
shi
fengzhong
zhongdian
le
already
reach
ten
minute
destination
PAR
'It's already been ten minutes since we reached the destination.'
Another goal of this article is to explain why R-related duratives behave differently from P-related duratives with respect to the word order problem.
In addition to the above syntactic and semantic issues, durative phrases are
worth studying because of their interaction with Incremental Theme verbs. It is
well-known that the combination of an Incremental Theme verb with a quantized object NP produces a quantized, i.e., telic, predicate (Krifka 1989, 1992,
Although I do not agree with Ernst (1987) and Li (1987) about the "since completion" reading, I
will still use the smce-construction to translate the Chinese sentences in many cases, as this is the
most idiomatic way to translate it. However, this has no implication for the analysis of these
sentences. Also see Paul (1988) for a different description of the two interpretations.
Event decomposition
33
1998; Filip 1999, among many others). This then predicts that durative phrases
in an Incremental Theme sentence should elicit an R-related reading, and this
prediction is correct. To our surprise, however, such sentences also allow durative phrases to receive the P-related reading, as is evidenced by the ambiguity of
(5).
(5)
Zhe-dong
fangzi wo
gai-le
san
nian
le
this-CL
house
build
three
year
PAR
(i)
have been building this house for three years'
(ii)
'It's three years since I built the house./I built this house and the result state
has existed for three years.'
The final goal of this article is to explore how the ambiguity of the durative
phrase in (5) arises and discuss the implications of such constructions for the
semantic representation of Incremental Theme verbs.
2 Previous analyses of durative phrases

Several syntactic analyses of Chinese durative phrases have been proposed in
the literature (Teng 1975, Huang 1982, 1991, 1997, Li 1987, Ernst 1987, Tang
1990,1994, Sybesma 1999). However, none of them adequately accounts for all
of the relevant data discussed in the last section. I will discuss these analyses in
turn.
To begin with, let us consider Tang's (1990, 1994) analysis. Following
Bowers (1988) and Larson (1988), she has assumed that direct objects are projected in the specifier position of V P and the verb is moved to a higher functional head. Moreover, she proposes that durative (and frequency) phrases can
be adjoined to VP, as is shown in (6).
(6)
[FP F
[vp Duration [ w
Object
[ V
Duration]]]
In (6), when the verb is raised to the functional head F, the durative-object order
is derived. To derive the object-durative order, on the other hand, she proposes,
following Larson's (1988) idea that oblique expressions may be base-generated
as the complement of V, that durative phrases in Chinese can also be projected
under the minimal V ' as the complement of V.
Sybesma (1999) proposes that durative phrases function as massifiers
(mass-classifiers) just as a numeral-classifier does in a nominal structure such as
san ping jiu 'three bottle wine'. Therefore, the durative phrase may precede the
object NP. As for the object-durative order, he has assimilated the structure to
34
Jo-wang Lin
locative resultatives with zai and dative structures. The exact details of Sybesma's proposal are beyond the scope of this article.
The major empirical problem with Tang's and Sybesma's solutions to the
distribution of duratives is that both analyses fail to distinguish P-related duratives from R-related duratives. As noted, while P-related duratives may precede
or follow the direct object, R-related duratives only follow the object. As far as I
can see, their solutions would apply blindly to any type of verb, thus incorrectly
predicting that R-related duratives should be able to precede the object NP just
like P-related duratives.
Huang (1991, 1997) does not directly discuss the word order problem of
durative phrases. He does, however, make a very specific proposal about their
syntactic position. According to Huang, a sentence like Ta kan-le san tian shu
'he read-Asp three days books' is analyzed as having a gerundive IP embedded
to a light verb DO. The durative phrase is adjoined to I' or is the specifier of the
IP. After the verb moves to DO, the durative-object word order is obtained. The
problem with this approach is that it is not clear how the object-durative word
order is derived. Nor is it clear how his analysis prevents R-related duratives
from appearing before the object NP.
Teng (1975), Ernst (1987) and Li (1987) do not discuss word order variation
in detail but they do provide explanations for why the R-related duratives do not
appear before the object NP. According to Teng, an R-related durative must
occur after the object NP, because it is not a constituent within the VP but the
main predicate of the sentence. 2 Li (1987) does not argue against Teng's main
predicate hypothesis of R-related durative phrases. She does, however, convincingly demonstrate that in addition to functioning as the main predicate of a
sentence, R-related duratives (and P-related duratives) can also be VP constituents. This manifests itself in the light of the distribution of adverbs such as
yijing 'already', which may occur either before the verb or before the durative
phrase, and the scope interaction between negation markers and durative
phrases. Since R-related duratives can remain within the projections of a verb,
the question still needs to be answered as to why they cannot occur before the
object NP. Li's answer to this question is that, when a durative phrase occurs
before the object NP, it forms a constituent with the object NP and is the
specifier of the NP. She argues that this analysis is supported by the fact that de,
a general modification marker, can be inserted between the durative phrase and
2
That a durative phrase can be (part of) the main predicate is supported by the fact that the verb
you 'have' can be placed before the durative as is shown in (i) (Teng 1975, Ernst 1987).
(i)
Ta lai
meiguo
you liang nian le
he come America have two
year PAR
'It has been two years since he came to America.'
In this article I will not discuss the status of you. It can be analyzed as part of the main predicate,
as in Teng (1975) and Ernst (1987). The major concern in this article will be the occurrences of
the duratives that are not the main predicate of the sentence.
Event decomposition
35
the object NP. Moreover, she proposes that in such a structure the durative
phrase is interpreted as quantifying over the verb, i.e., "the quantity or extent of
certain activity". According to Li, this is why an achievement sentence like *Ta
lai san-ge yue (de) meiguo 'It's been three months since he came to America' is
ill-formed. The problem with Li's approach is that it is not possible that a
durative phrase is always the specifier of an NP. Notice that /-insertion is
possible only when the object NP is a bare noun. If the object NP is a full NP
such as zhe-ben shu 'this-Cl book', de-insertion is impossible (cf. * san-ge
xiaoshi de zhe-ben shu 'three hours of this book'). Li's analysis seems to have
nothing to say about this. Her explanation also raises the question of why a
[duration + NP] phrase quantifies over events rather than result states. For
constructions involving an activity verb and a bare noun object, this constraint is
understandable, because atelic activity situations have no result states. For telic
situations, on the other hand, it is not clear why the restriction holds. Suppose
that there is a non-instantaneous change of state verb which denotes a property
of the result state. Then there is no a priori reason why a result state cannot be
quantified over. Therefore, unless there is a better explanation of why the
structure of [V + duration + NP] is restricted to quantification over the extent of
an event rather than the result state of that event, Li's analysis is not so much an
explanation as a stipulation.
Ernst's (1987) account for why R-related duratives may not precede object
NPs is inadequate as well. According to Ernst, this is so because "the semantic
rule interpreting this marked reading restricts it to VP-final position". This
answer, however, does not help us better understand the nature of the R-related
duratives, if no concrete proposal is offered concerning what the semantic rule is
and why this rule has the property it has.
In view of the above problems of the previous analyses of R-related durative
phrases, I would like to pursue a different analysis in the following section that
may accommodate the relevant data in a more enlightening way. I will propose a
semantics-based structural account of the distribution of durative phrases inspired by von Stechow's (1995, 1996) analysis of the German word wieder
'again'.
3 An alternative account
Before proposing my analysis of durative phrases in Chinese, it is helpful to first
discuss the German word wieder 'again', which displays the following interpretive contrast (von Stechow 1995, 1996).3
(7)
(a)
Ali Baba
Sesam
wieder ffnete
Subj
Obj
again
(restitutive/repetive)
opened
(7a) and (7b) should be understood as subordinate clauses.
36
Jo-wang Lin
(b)
AH Baba
Subj
wieder Sesam ffnete (only repetitive)

again Obj
opened
(7a) has both a restitutive reading in which wieder modifies the result state and a
repetitive reading in which wieder modifies the whole event, whereas (7b)
provides only the latter reading. To account for the above data, von Stechow
(1995, 1996) argues that the ambiguity of wieder is a reflection of syntactic
scope. 4 He has assumed a rather abstract syntax which decomposes telic verbs
into their subcomponents (CAUSE + BECOME + STATE). Following Kratzer
(1996), he suggests that the external argument of a verb is introduced by a
functional head called Voice, which identifies the CAUSE component. Above
the VoiceP is AgroP, where accusative Case is checked. Below the VoiceP is
VP, whose head is BECOME, which in turn takes a result state projection. More
concretely, his proposal can be represented as follows:
FP
(8)
REPETITIVE
e[VoiceP(e)]
MODIFIER
again
NP
::'
SUBJECT
(&)
voice
ACTIVE
V
agent(x)(e)
BECOME
RESTITUTIVE MODIFIER
again
(9)
(a)
(b)
NP
OBJECT
RESULT
3e[again(Ae[agent(Ali Baba)(e) & BECOME (opened(fcam))(e)])]

3e[agent(Ali Baba)(e) & BECOMEfagain (opened(5eram))(e)]]
In the above structure, the restitutive again is adjoined to the result XP, hence
only modifying the result state, whereas the repetitive again is adjoined to a
position higher than VP, hence modifying the whole event. Both the subject and
the object have to be raised out of VP to check their Case in [SPEC, AgrsP] and
[SPEC, AgroP], respectively. In this analysis, (7a), in which again follows both
the subject and the object, is ambiguous because again can be adjoined to either
the result XP or a recursive VP. In contrast, in (7b), in which again follows the
4
See also Dowty's (1979) treatment of again and/or-phrases in English.
Event decomposition
37
subject and precedes the object, it can only be adjoined to a position higher than
AgroP. Hence, only the repetitive reading is possible.
A very important aspect of von Stechow's analysis is its overt syntactical
expression of the sub-eventuality of a bigger event, hence making that
sub-eventuality accessible to syntactic modification and semantic composition.
In a similar vein, Foli (2002) and Ramchand (2003) have recently argued that
there are three sub-event projections for a lexical expression that entails a result
state. These three projections are vP, VP and RP. According to Ramchand, "vP
introduces the causation event and licenses different types of external argument", "VP specifies the nature of the change or process and licenses the entity
undergoing change or process" and "RP gives the 'telos' or 'result state' of the
event and licenses the entity that comes to hold the result state". In this paper, I
share the above authors' view of the syntax of event structure and will propose
an account for Chinese durative phrases based on such event decomposition in
overt syntax. Following Ramchand (2003), I assume that each verb carries some
semantic features that need to be checked by a head in the syntax. In particular, I
assume that a result predicate in the representation of a verb meaning has to
check the result feature of the head R of RP, the event predicate has to check the
process feature of the head V of VP, and the Agent predicate has to check the
agent feature of VoiceP.
The last point about my syntactic assumption of the phrase structure is the
position of AgroP. As noted, von Stechow (1996) has assumed that, in German,
AgroP is located above VP. In Kratzer's (1996) original proposal, however,
accusative Case is actually assigned below VoiceP in the specifier position of
VP via government. Kratzer's analysis indicates that AgroP can, in principle, be
located below VoiceP if her framework is translated into Chomsky's minimalist
program. Indeed, in this paper, I will assume that, in Chinese, AgroP is located
below VoiceP and above VP. Similar assumptions for this kind of structure can
be found in Basilico (1998) and Sanz (2000). The different positions of AgroP
can be thought of as choices of different parameters.
In addition to the above syntactic assumptions, there is a semantic requirement of durative phrases that needs clarifying before I am able to explain their
distribution. It is well-known that expressions like for two days/in two days are
used to distinguish between accomplishments and achievements, and activities
(processes) and states. While expressions like for two days occur more naturally
with processes and states, expressions like in two days generally occur with
accomplishments and achievements. This suggests that durative phrases impose
an aspectual homogeneity requirement on the constituent that they modify
(Moltmann 1991, Dini and Bertinetto 1995). I assume that durative phrases in
Chinese have this homogeneity requirement as part of their selectional restriction.
Now let us return to the distribution of durative phrases. As we saw earlier in
(3), when a durative phrase occurs with a process/activity, the durative phrase
38
Jo-wang Lin
may precede or follow the direct object. The permutation of word order can be
explained by adjoining the durative phrase to AgroP or VP as shown below.
( 1 0 ) UgrsP
W0 k [voiteP t k [voice'kaij [ Ag roP e r s h i niai) [AgroP j i c h e n g c h e i [ A g r o' AgrOj [ v p e r s h i n i a n [ v p t [ v V]]]]]]11]
rive
twenty year
i ^
ta|i
twenty year
In (10), just as the subject NP has to move to the specifier position of AgrsP to
check its Nominative Case, the object NP jichengche 'taxi' must move to the
specifier position of AgroP to check its Accusative Case. The motivation of
such movement has been discussed thoroughly in the syntax literature. As for
the movement of the verb, we can assume Ramchand's (2003) feature checking
system, in which the verb has to check all relevant eventuality features. When
the durative phrase is adjoined to VP, the object-durative order is derived; when
the durative phrase is adjoined to AgroP, the durative-object order is derived.
It is important to note here that in (10) the durative phrase can be adjoined to
VP, AgroP or VoiceP, because they are all process projections which meet the
homogeneity requirement. However, durative phrases are rarely adjoined to
VoiceP or higher projections unless they are contrastive as is shown in (11)
below.
(11)
Ta
san
he
three day
tian
han
baba
with
father sleep
shui,
tian
han
mama
shui
four day
si
with
mother
sleep
'He sleeps with his father for three days and sleeps with mother for four days.'
In (11), the durative phrases can also occur after the verb shui 'sleep' without
affecting the meaning. Although I do not know exactly why P-related durative
phrases have to be contrastive in order to occur in a preverbal position, I would
like to speculate the following: Indefinite NPs (in Chinese and in many other
languages) occurring in a preverbal position must receive a specific interpretation. Since durative phrases are mainly used to assert cardinality rather than
existence of referents, special contexts are required to license their presence in a
preverbal position. A contrastive context is one context that may remove the
specificity requirement. Whether or not this explanation is correct, the fact that
P-related duratives may appear in a preverbal position is predicted by the homogeneity requirement of durative phrases.
Similarly, the homogeneity account predicts that durative phrases may appear in the sentence-initial position, but this prediction is not borne out. Again,
an account like what the above suggested may apply to this case. The sentence-initial position is the topic position, which requires an NP to be definite or
specific. Because durative phrases are not definite or specific NPs, they are
excluded from the sentence-initial position. The situation here is similar to other
39
Event decomposition
indefinite object NPs as the ungrammatically of the sentence *Liang-ben shu,

wo kan-le 'Two books I read-Asp' shows.
Next, let us consider cases when the main verb is inherently telic. In such
cases, the phrase structure contains a result state projection. Take (4a) for instance. It is represented as follows: 5
(12)
[AgrsP
women k
[Volccp
tk
[voice 1
didaj [ AgroP *shi fenzhong [ AgroP zhongdian, [Agr Agro;

reach
ten minute
destination
[vp *shi fenzhong [ w t, [ v Vj [RP shi fenzhongfap ek [R. t t] ]]]]]]]]]]]]

ten minute
ten minute
In (12), among the five projections, RP, VP, AgroP, VoiceP and AgrsP, only the
most deeply embedded projection RP represents a homogeneous eventuality.
Neither a process plus a result state nor a causing event plus a process plus a
result state can be homogeneous. Therefore, it follows from the homogeneity
requirement of durative phrases that they can only attach to the result projection
RP in (12). Since the verb has to be raised to the Voice head and the object NP
must move to AgroP, it follows that the durative NP must follow the direct
object. This not only explains why a durative phrase always has an R-related
interpretation when it occurs with a verb that lexically entails a result state, but
also accounts for why R-related duratives may not appear before the direct
object NP.
4 The semantics of durative phrases

In this section, I turn to the semantics of durative phrases. I will show that the
very same meaning of durative phrases may apply to both P-related and
R-related durative phrases, thus arguing against an ambiguity approach to the
interpretations of Chinese durative phrases.
I define the meaning of a durative NP as in (13):
(13)
[[Durative NP]] =: R Xe/s[R (e/s) & MU(x(e/s)) = Num Hom(e/s)]
where e is a variable ranging over events, s is a variable ranging over states, is

a function which, when applied to an eventuality, gives its temporal trace, MU is
a measure unit function which yields a number when applied to a temporal trace
and Horn is a predicate representing homogeneity, 'e/s' means either a variable
of e or variable s.6 As for the notation 'R ', it is intended to stand for a variable
with a flexible semantic type involving a sequence of one or more eventuality
'e' in (12) represents an empty category that is co-referential with the subject NP.
We can use a single eventuality variable here. However, in order to make the meaning clearer, I
distinguish e from j so that no confusion will occur.
40
Jo-wang Lin
arguments. That is, ' R ' can be of type <s,t>, <s,<,s,t>>, < s c , s < s , t > , etc. (See
Pin 1999 for the use of such a notation.) The motivation for this notation will
be explained later. (13) says that a durative phrase takes an expression of type
R as an argument, and that the duration of the first eventuality argument of
R in terms of the unit M U is the number Num as specified by the durative NP,
and finally that this eventuality is homogeneous. Now let us see how this
meaning of durative phrases apply to the examples discussed in the last section.
Consider the representation in (10) first, in which the durative phrase elicits a
P-related interpretation. Assume that when a verb is raised, it leaves a variable
of the same semantic type, and that object NPs leave a trace of type e. The step
by step computation of (10) is as follows:
(14)
When the durative phrase is adjoined to VP

[[Vj]] = R<e< s ,, 7
[[VP]] = R(x)
[[ershi-nian VP]] = R [ (e) & Year(x(e)) = 20 Hom(e)](R(x))
= e[R(x;)(e) & Year(x(e)) = 20 Hom(e)]
[[AgroP]] = Xje[R(x,)(e) & Year(x(e)) = 20 Hom(e)](taxi)8
= ^ ^ ) Year(x(e)) = 20 Hom(e)]
[[VoiceP]] = xke[Agent(xk)(e) drive(taxi)(e) Year(x(e)) = 20 Hom(e)]9
[[AgrsP]] = ^ ) drive(taxi)(e) Year(T(e)) = 20 Hom(e)]]
After the existential closure of the event variable, (14) is equivalent to saying
that there is an event of driving a taxi whose agent is I and the duration, i.e., the
temporal trace, of that event is 20 years and that event is homogeneous.
When the durative phrase is adjoined to AgroP rather than VP, the computation steps are similar. I leave this as an exercise to the reader.
Next, let us consider change of state verbs such as achievement verbs and
resultative verb compounds. I assume that a change of state verb lexically entails a result state. Following Pin (1999), I define a result state as an eventuality which immediately follows the event that brings it about, and which has a
theme participant identical to the theme participant of the event. That is, a result
is a four-place relation, as Pin (1999) has put it in (15a). I will also join him in
assuming that "if an object is in a result state of a given type at the end of an
event, then it is not in a result state of that type before the event ends" as (15b)
indicates. This notion will reappear in my discussion of Incremental Theme
sentences later in this paper.
Here I assume Kratzer's (1996) analysis where the external argument is introduced by Voice. So
a transitive verb only has one internal individual argument.
I assume with Heim and Kratzer (1998) that the index of a raised constituent functions as a
lambda-abstractor.
Here I assume Kratzer's (1996) semantics of the agent predicate and the event identification rule.
See Kratzer (1996) for details.
41
Event decomposition
(15)
(a)
Result(e, , s, P) := Theme(e, ) e
s P(s) Theme(s, x)
('event e with theme has result state 5 of type with theme x')
(b)
Result(e, x, s, P)
Ve' [e' c e
-,3s'[Result(e', s', , )]
('result states do not begin earlier than they do')
Pin (1999, 10)
With the above definitions in mind, I propose that a resultative verb such as dida
'reach' in (4a) denotes something like (16):
(16)
[[dida]] = yxse[reach(y)(S)(E) Result(e, x, s, be-at-the-place-of-y)]
That is, resultative verbs express relations between ordinary objects, result
states and events. Notice that the third and fourth arguments of the verb dida
'reach' are eventuality arguments. So after the first two individual arguments
are satisfied, the semantic type is <s,<s,t>> rather than <s,t>. This is, in other
words, a sequence of type s argument. This is why I use the notation ' R ' in
(13). I want the durative phrase to be able to modify not only processes of type
<s,t> but change of state situations of type <s,<s,t>> as well. I will also join
Pin (1999: 11) in assuming that a default mechanism will existentially close
the result state argument of a verb or verbal complex if nothing else does. This is
done through the application of the following existential closure rule:
(17)
RXeps[R(s)(e)]]
Applying (16) to the RP in (12), we obtain the property of the result state of the
reaching event. The meaning of the durative can then apply, measuring the
duration of the result state as is shown in (18).
(18)
[[RP]] = Rj(x,)
[[shi fenzhong RP]] = Xs[Rj(Xi)(yk)(s) & Minute(T(s)) = 10 Hom(s)]
[[VP]] = [[RP]]
[[AgroP]] = Xs[R,(the-destination)(y k )(s) & Minute(x(s)) = 10 Hom(s)]
[[VoiceP]] = e3s[Agent(y K )(E) reach(the-destination)(e) Result(e, yk, s,
Be-at-the-place-of-the-destination) & Minute(t(s)) = 10 Hom(s)]
(By Default existential closure (17) and Kratzer's Event identification in footnote 9)
[[AgrsP]] = / x 3 s [ Agenti we )(e) reach(the-destination)(e) Result(e, we, s,
Be-at-the-place-of-the-destination) & Minute(t(s)) = 10 Hom(s)]
Once the text level existential closure applies, the denotation of AgrsP amounts
to saying that there is an event of reaching the destination whose agent is "we",
and this event leads to the result state of our being at the destination and the
duration of the result state being 10 minutes. This seems to be the correct interpretation of the sentence.
Having shown how the P-related and R-related interpretations of durative
phrases are compositionally derived, I would like to emphasize that both inter-
42
Jo-wang Lin
pretations utilize the same meaning of durative phrases as given in (13), the only
difference being that P-related duratives are adjoined to VP or AgroP, but
R-related duratives are adjoined to the most deeply embedded RP. This purely
structural account of durative phrases indicates that it is not necessary to resort
to an unmotivated multiplication of interpretive rules or multiplication of lexical
entries for the semantics of Chinese durative phrases. If this account of Chinese
durative phrases is correct, it will constitute an argument for a structured semantic representation. This account also contrasts with Dowty's (1979) treatment of durative phrases in English which assumes the lexical ambiguity of the
preposition for.
5 Against the "since-completion" reading of

R-related duratives
Recall that instead of assuming that R-related durative phrases measure the
duration of a result state, Ernst (1987) and Li (1987) have proposed that they
measure "the duration since completion of an event". These two different
characterizations are actually not equivalent. In what follows, I will first show
that the "since completion" reading can be derived as an inference to the
R-related reading, and I will then provide cases showing that Ernst-Li's "since
completion" reading does not offer a correct interpretation.
Li-Ernst's "since completion" reading can be derived as an inference of the
proposed R-related interpretation, though the latter says nothing about the
"since" interpretation. Take (5) for example. If you say that you have built a
house and that the duration of the existence of the house is 3 years, one can
naturally infer from this that you built the house three years ago. Therefore, the
R-related reading does not need to resort to "since" to have the "since completion" reading.
In fact, both Ernst and Li do not explicitly explain the origin of the meaning
of "since" in their approach. Since durative phrases in Chinese are not accompanied by a preposition, the meaning of "since" presumably comes from numerals. This then implies that numerals in Chinese are lexically ambiguous. But
if the "since" reading derives from lexical entries, this predicts that any type of
completed events, including atelic activities, should be able to elicit the "since
completion" reading. This, however, does not seem to be the case. Ernst's "since
completion" reading never arises when the sentence denotes an atelic event such
as Ta pao-le san-ge xiaoshi 'He ran for three hours'. The above sentence only
obtains a reading in which the event of running lasted for three hours. It does not
elicit the reading 'It's been three hours since I finished running'. It would be
unreasonable to say that atelic events cannot be completed. When an atelic event
is terminated at any arbitrary time, that terminated event is also a completed
event. Clearly, to solve this problem, Ernst and Li need to resort to some
stipulative rule so that it only applies to telic situations.
43
Event decomposition
Another argument in favor of the R-related interpretation, as opposed to the

"since completion" reading, derives from the consideration of the following
sentence.
(19)
Yuehan
ba
men
guan-le
John
BA
door
close-Asp three-CL
san-ge
xiaoshi
le
hour
PAR
(Lit.) 'John has closed the door for three hours.'
(19) is true in situations in which the consequent state of the door being closed
lasts for three hours. Now consider another situation. Two hours after John had
closed the door, he reopened it and one hour has elapsed since the reopening. In
this scenario, (19) is false. However, according to the "since completion"
reading, (19) should be true, because the duration since completion of the
closing event is indeed three hours. This example clearly shows that in (19), the
duration of the time as specified by the durative phrase is related only to the
result state. We are not just measuring the time elapsed since an event is completed.
Finally, the "since completion" reading requires that the left boundary of the
interval specified by a R-related durative phrase be the point at which the event
is completed, and that the right boundary be the speech time. Let us refer to this
interval as the 'since interval'. In normal cases, this "since interval" is identical
to the duration of the result state when the right boundary is the speech time.
However, when the time elapsed since the attainment of the result state and the
duration of the result state are not the same, the predictions of the 'since completion' reading and the R-related reading are different. Consider the following
examples. 10
(20)
(a)
Ta
zuotian
(zhi)
likai-le
jiaoshi
shi
fenzhong
he
yesterday
only
leave-ASP
classroom
ten
minute
'Yesterday he (only) left the classroom for ten minutes.'
10
Sometimes such examples sound a little odd as (i) below shows.

(i)
*Zuotian
ta
yesterday he
dida
(le)
zhongdian
shi
fenzhong
reach
ASP destination
ten
minutes
'Yesterday he reached the destination and stayed there for ten minutes.'
I do not know why (i) is not good. However, even such examples can be made fully acceptable
when a richer context is given, as is shown in (ii).
(ii)
Zuotian
ta
dida
zhongdian liang fenzhong hou, jiu
tili
yesterday he
reach destination two
minute
after then energy
bu
zhi
fundao le
not
sufficient faint
ASP
'After he reached the destination for two minutes yesterday, he fainted because
of lack of energy. '
44
Jo-wang Lin
(b)
Ta
zuotian
kai-hui
deshihou
turan
hunmi-le
he
yesterday
open-meeting
when
suddenly
faint-ASP ten
shi
fenzhong
minute
'Yesterday, h e suddenly fainted for ten minutes during the meeting. '
(20a) obtains the interpretation "there was an event of leaving the classroom,
which took place yesterday, whose agent was "he", and this event led to a result
state of his not being in the classroom, and the duration of this state was 10
minutes". This is exactly the interpretation assigned by the proposed analysis of
R-related duratives. However, Ernst-Li's 'since completion' analysis assigns
the sentence a false interpretation. In their analysis, there must be some time
point that serves as the right boundary of the "since interval". However, the
speech time cannot be the right boundary, because there are far more than ten
minutes between the time point at which the event is completed and the speech
time. The interval denoted by the temporal adverb zuotian 'yesterday' cannot be
the right boundary of the since-interval either, since this would give rise to an
interpretation that (20a) does not have. I conclude that durative phrases in
Chinese do not have the "since completion" reading that Ernst and Li have
claimed. On the contrary, that which they refer to as the "since completion"
reading is actually the R-related interpretation defended in this article.
6 Incremental theme sentences

In the above discussion, I have shown that when a verb lexically entails telicity,
the durative phrase occurring with it must be interpreted as an R-related durative. At the beginning of this article, however, I mentioned that when the verb of
the sentence is an Incremental Theme verb such as kan 'read', xie 'write', and
gai 'build', a durative phrase is ambiguous between the P-related and the
R-related reading, and thus constitutes a challenge to aspectual composition as
proposed in Kriflka (1992,1998). Here are some examples to refresh the reader's
memory:
(21)
(a)
Ta
(yijing)
xie-le
he
already
write-ASP that-CL
na-feng
xin
san
tian
le
letter
three
day
PAR
(i) ' H e has been writing that letter for three days.'
(ii) 'It has been three days since he finished writing that letter.'
(b)
Ta
xie-le
san
tian
na-feng
xin
le
he
write-ASP
three
day
that-CL
letter
PAR
'He has already been writing that letter for three days.'
With the durative phrase placed at the end of the sentences, (21a) is ambiguous
between the P-related and R-related reading. The ambiguity is even more obvious when the object NP is preposed to the beginning of the sentence as in (22).
45
Event decomposition
(22)
(a)
Na-feng
xin,
wo
xie-le
san
tian le,
keshi
hai
mei
that-CL
letter
write-ASP
three
day
but
yet
not
PAR
ji-chuqu
send-out
'It's been three days since I wrote that letter, but I haven't sent it out.'
(b)
Na-feng
xin,
wo
xie-le
san
that-CL
letter
write-ASP
three day
tian,
jieguo
mei
xie-wan
result
not
write-finish
wrote that letter for three days, but didn't finish it.'
(c)
Na-feng
xin,
wo
xie-le
san
that-CL
letter
write-ASP
three day
tian,
cai
xie-wan
EMP
write-finish
'It took me three days to write that letter.'
The durative phrase in (22a) obtains the R-related interpretation. The P-related
durative receives two possible interpretations. In one reading, the durative
phrase describes the duration of an incomplete event. This is the case in (22b).
The other reading tells us how much time the whole event takes. In other words,
a durative phrase in an Incremental Theme sentence may in fact be three-way
ambiguous.
However, it is important to note that if the durative phrase precedes the
postverbal object NP, the R-related reading disappears. This is illustrated by
(21b). In other words, the P-related and R-related duratives in Incremental
Theme sentences behave exactly the same way as they do in non-Incremental
Theme sentences. This suggests that the proposed analysis of durative phrases
should apply to their occurrences in Incremental Theme sentences as well.
Another important observation of Chinese Incremental Theme sentences is
that there is a contrast between numeral objects and definite objects according to
whether the events culminate. According to Zhang (1999), Soh and Kuo (2001)
and Liu (2003), while there is no implication of culmination for a definite object
as discussed above, a numeral object always implies the completion of the
event. This is illustrated by Soh and Kuo's examples below. 11
(23)
(a)
Ta
chi-le
he
eat-ASP
#liang-ge
two-CL
dangao/na-ge dangao, keshi
mei
chi-wan
cake/that-CL
not
eat-finish
cake
but
' H e ate two cakes/that cake, but he did not finish eating them/it.'
11
Soh and Kuo (2001) have cited (i) below as a fully acceptable sentence.
(i)
W o zuotian
kan-le
yi-ben
shu,
keshi mei kan-wan
I
yesterday read-Asp one-CL book but
not read-finish
read a book yesterday, but I didn't finish reading it.
However, the same sentence has been cited by Zhang ( 1999) as a contradiction. I agree with Soh
and Kuo (2001) that the variation here has to do with how one interprets the numeral yi 'one'.
W h e n yi is construed as specific, hence referential, (i) is more acceptable. When yi 'one' is read
as asserting a quantity, (i) is less acceptable.
46
Jo-wang Lin
(b)
Ta
kan-le
he
read-ASP two-CL
#liang-ben
shu/na-ben
shu,
keshi
mei
kan-wan
book/that-CL
book
but
not
read-finish
'He read two books/that book, but he did not finish reading them/it.'
The ambiguity of sentences such as (21a) and the contrast between definite and
object NPs in (23) raise very interesting questions involving the semantics of
Incremental Theme verbs and durative phrases as well as their interaction.
These issues will be dealt with in this section.
First, I will discuss the R-related reading of durative phrases in Incremental
Theme sentences. This reading entails culmination of an incremental theme
event. If we adopt the traditional assumption, as we did for achievement verbs,
that accomplishment verbs incorporate telicity as part of their inherent meaning,
then the semantic interpretation and distribution of durative phrases in incremental theme sentences will parallel those in achievement sentences. Nothing
new needs to be added to the theory of grammar. What need to be explained is
how the P-related reading of durative phrases arises in Incremental Theme
sentences. I will now turn to this matter.
To begin, let me first discuss the interaction between noun phrases and verbal
predicates with respect to aspectual composition. It has now been well established that the quantization properties of nominal arguments may influence the
quantization properties of verbal predicates, i.e, lexical aspect, when the verb is
an Incremental Theme Verb (See Krifka 1998, Filip 1999 and the references
cited therein). It has also been observed that grammatical aspect, i.e., the perfective vs. imperfective distinction, may constrain the quantificational and
(in)definiteness properties of Incremental Theme Noun Phrases in Slavic languages (Krifka 1992, Pin 1995). Filip (2001) argues that to account for the
phenomenon of mutual constraining between nominal domain and verbal domain, Krifka's (1992) rule of "aspectual composition" is inadequate. 12 In particular, she has shown that there are cases for which the nominal argument has a
quantized interpretation but the grammatical aspect is imperfective. (24) below,
from Filip (2001, 466), is such an example.
(24)
Ivan jel 1
jbloko
pjat' mint/??za
pjat' mint
Ivan
apple.SG.ACC
five minutes/??in
five minutes
eat.PAST
Russian
(i) 'Ivan was eating an/some/the apple ??for five minutes/in five minutes"
(ii) 'Ivan ate an/some/the apple for five minutes/??in five minutes.'
Filip argues that the noun phrase jbloko 'apple' receives a count rather than a
mass interpretation, i.e., apple-stuff or apple sauce, despite the fact that the
predicate is cumulative as evidenced by the use of the durative adverbial. That
12
Roughly, Krifka's aspectual composition says that the combination of a quantized Incremental
Theme noun phrase with a verb yields a quantized predicate, whereas the combination of a
cumulative Incremental Theme noun phrase yields a cumulative predicate.
47
Event decomposition
jbloko 'apple' obtains a count interpretation is also supported by the fact that a
pronoun can refer back to it, which is impossible if jbloko 'apple' has a mass
interpretation. This is shown by the Russian sentence (25).
(25)
Ivan jel1
jblokoi,
a
Boris eg, jel1
Ivan eat.PAST
apple.SG.Acq and Boris it
eat.PAST
(i) 'Ivan ate an/the/some apple and Boris ate it, too.'
(ii) 'Ivan ate an/the/some apple, and Boris ate iti too.'
tze.
too
Such examples indicate that a quantized nominal argument can retain its count
interpretation even in the scope of the imperfective aspect, and therefore fail the
principle of aspectual composition. In view of this, Filip suggests that the contribution of Incremental Theme noun phrases (quantization-cumulativity) to the
lexical aspect should be separated from the contribution of imperfective aspect
(partitivity-totality). While the Incremental Theme argument encodes the information relevant to quantization-cumulativity, the verb encodes the relation of
'partitivity' (imperfective) and 'totality' (perfective). According to Filip, this
treatment avoids the contradiction of combining a quantized noun phrase with
an imperfective verb. She takes the perfective and imperfective aspects as
eventuality-type shifters and represents perfective and imperfective verbs as
below:
(26)
Perfective Verb:
Imperfective Verb:
[() & TOT(P)]

[() & PART(P)]
PART = '3[() e'<e]
(Fililp 2001,474)
(Fililp 2001, 475)
According to the above analysis, a perfective verb represents a given state of

affairs in its totality and is therefore quantized. By contrast, the imperfective
operator takes predicates of states, processes or events and maps them onto
partial states, processes and events. Take (24) for example. According to
Krifka's aspectual composition, the verbal predicate is quantized because the
noun argument is quantized. On the other hand, the imperfective operator enables the partitivity condition. Therefore (24) asserts that a part of the apple-eating event took place. From this it follows that a part of the apple was
eaten, if events and objects have a one-to-one mapping as Krifka's (1998)
Mapping to Object principle requires. It is worth mentioning here that Filip has
used the part relation '<' (proper part or improper part) in (26), because imperfective verb forms in Russian can not only be used to denote incomplete
(partial) events but also completed events, given an appropriate context. I refer
the reader to her article for more details.
Like Filip (2001), Tatevosov (2002) has reported a similar phenomenon in
Turkic languages such as Chuvash and Karachay-Balkar, which are spoken in
Russia. He also resorts to an empty PART operator to account for the data,
48
Jo-wang Lin
although his PART operator applies to nominal predicates rather than verbal
predicates.
Returning to Chinese, it is easy to see that the ambiguity of (21a) is very
similar to the Russian data discussed by Fililp, and the Chuvash and Karachay-Balkar data reported by Tatevosov. In Chinese, a quantized Incremental
Theme does not necessarily result in a quantized predicate either. This suggests
that we may extend Filip-Tatevosov's analysis of Incremental Theme sentences
to Chinese. I will indeed adapt Filip's semantics of the PART operator to account for the Chinese data. An advantage of associating the PART operator with
the verb is that it can be constrained via a function-argument selectional restriction, to apply to only Incremental Theme verbs. In this approach, the partitive effect on the theme argument can be derived from Krifka's Mapping to
Object principle, i.e., every subpart of an event corresponds to a subpart of the
Incremetnal Theme. In what follows, I will extend, with a slight modification,
Filip's and Tatevosov's ideas of partitivity to explain Chinese Incremental
Theme Sentences.
Let us take the verb gai 'build' as an example. As noted, I assume Pin's
(1999) definition of result states. Given this, the verb gai 'build' can be defined
as in (27) with the result state as part of its inherent meaning. If a durative phrase
occurs with such a verb, it obtains the R-related interpretation just like the case
of achievement verbs. On the other hand, the P-related reading arises as a consequence of the operation of the partitive operator whose definition is given in
(28). Recall that Pin (1999) has proposed that a result state does not begin
earlier than the events do, i.e., part of an event does not have a result state. (See
the definition in (15).) Since the function of the PART operator is to focus on the
parts of an event, applying the PART operator entails the non-existence of the
result state. I incorporate this into the meaning of the PART operator. This is
why we have ' - , 3 s ' in (28). Applying (28) to (27) yields the result in (29). It
follows from this that a partitive Incremental Theme verb applies only to the
process of an event, but not the result state. As there is no result state in the verb
meaning, the RP projection will not be licensed (or projected) in overt syntax,
i.e., feature checking will fail. Therefore the combination of a partitive Incremental Theme verb with a durative phrase will only produce a P-related interpretation.
(27) [[gai]] = XyXsXe[Build(y)(e) Result(e, y, s, Be-built)]
(28) [[PART]] = P<e,<5,<s,t>yXe'Be-I3S [P(y)(s)(e') e'<e]
(29) [[gaipART]] = ye'5e -,Bs[Build(y)(e') Result(e', y, s, Be-built) e'<e]
It is interesting to note that the analysis in (28) predicts, as in the
Filip-Tatevosov analysis, that a P-related durative is compatible with a completed or incomplete event. This prediction is correct, because a Chinese sentence like Wo du-le na-ben shu liang-ge xiaoshi may mean either that reading
Event decomposition
49
part of that book took two hours or that reading the whole book took two hours
(cf. (22b), (22c)). For either reading, the durative phrase modifies the process of
the event. The former reading arises for 'e'<e' and the latter reading arises for
'e'=e\ At this point, the alert reader might think that a contradiction exists
between precluding a result state and the 'e'=e' reading. This is a legitimate
question but can be solved technically. Caudal (1999) has proposed that events
canonically break down into Inner stages (event's development) and Result
stages. Moreover, telic events have binary result stages consisting of a primary
result stage, related to the development of events, and a secondary result stage,
related to the state of affairs arising from the final completion of the event.
According to Caudal, "the secondary RS should be the complementary of the
primary RS, so as to cause a definite CoS [change of state]" and "the secondary
RSs should be viewed as a transition function". The event structure of the verb
drink proposed by Caudal (1999: 235) is the following:
(30)
drink (e, x, y)
drink_IStage (ei,x,y)
drink_RStage fe.y)
drink_P_RS (e3, y)
drink_S_RS (e4, y)
Returning to our "e'=e' reading, I suggest that the ' e " here is an event which
consists of the inner stage and the primary result stage but to which the 'transition function' has not applied. There is therefore no definite change of state yet.
This explains why it is still possible to claim the non-existence of the result state
when 'e'=e', just as in the definition of the PART operator in (28). Presumably,
this should also be the reason why the Russian data, as discussed in Filip (2001),
are still compatible with the Russian durative /or-phrase.
Although the use of the PART operator is quite successful in accounting for
the partitive reading of a definite Incremental Theme, we still have to explain
why indefinite numeral incremental theme objects always imply the attainment
of a goal as the examples in (23) indicate. With the introduction of the PART
operator, it seems that we would wrongly predict that a partitive reading is also
possible with a numeral Incremental Theme object. I would like to suggest that
partitive readings are excluded for numeral NPs for an independent reason.
It has been pointed out repeatedly that partitive constructions (in English)
are subject to a constraint, to the effect that the embedded NP must be definite
(Selkirk 1977, Jackendoff 1977, Barwise and Cooper 1981, de Hoop 1998,
among many others). This is illustrated by the examples in (31) taken from de
Hoop (1998, p.151).
(31)
(a)
one of these/the/my cat
(b)
*one of some/three/no cats
50
Jo-wang Lin
Whatever the explanation for the partitive constraint may be, I suggest that
something similar might work for the proposed PART operator. As we have
seen, though the PART operator takes an incremental theme verb as its argument, there is a partitivity effect on the incremental theme argument through
Krifka's (1998) Mapping to Object principle. It is therefore reasonable to suggest that, like the most deeply embedded NP in a partitive construction, the
incremental theme argument is subject to the Partitive Constraint.13 If something along these lines is correct, partitive readings can be excluded for numeral
incremental theme arguments. There is in fact evidence demonstrating that this
line of thinking is correct. In Chinese, expressions like bufen (neirong) 'part
(content)' may modify a definite NP but not a numeral NP. This is shown by the
contrast between (32) and (33).
(32)
(a)
(b)
(33)
(a)
(b)
Wo yijing
I
already
already read
Wo yijing
I
already
already read
*Wo yijing
du-le
read-ASP
part of that
du-le
read-ASP
part of that
du-le
nei-ben shu
that-CL book
book'.
bufen nei-ben
part
that-CL
book.'
yi/liang-ben
de
Poss
shu
book
shu
bufen
part
neirong
content
le
PAR
le
PAR
de
bufen
I
already read-ASP one/two-CL book POSS part
already read part of one/two book(s)'.
*Wo yijing
du-le
bufen yi/liang-ben
shu le
I
already read-ASP part
one/two-CL
book PAR
already read part of one/two book(s).'
neirong le
content PAR
Summarizing this section, I have shown that although durative phrases are
ambiguous when they occur in an Incremental Theme sentence, their syntactic
behavior is not different to their behavior in other types of sentences. That is,
when a durative phrase receives a P-related reading, it may precede or follow the
object NP; when it has an R-related reading, it can only follow the object NP. To
account for this fact, I propose that telicity is built into the semantic representations of Incremental Theme verbs. Thus, an Incremental Theme verb may
license RP projection in syntax, to which the R-related durative is adjoined.
Since RP is in the lowest position and the object NP is raised to AgroP, the
R-related durative must follow the object NP. On the other hand, the P-related
reading arises as the result of a covert PART operator canceling the result state
of the Incremental Theme verb, turning it into an atelic process. As there is no
result state in the semantic representation of partitive Incremental Theme verbs,
13
This is even more apparent if we state the meaning of PART in a manner more like that of
Tatevosov's proposal, (i) below is such a possibility,
(i)
[[PART]] = Pxe3y-s [P(y)(s)(e) y < ]
Event decomposition
51
RP projection is not licensed. It follows that the P-related durative phrase may
freely adjoin to VP or AgroP, hence producing two different orders.
7 Conclusions and theoretical implications

In this paper I have proposed a syntactic-semantic analysis of Chinese durative
phrases, arguing that they can be adjoined to every maximal projection, provided that we can interpret them in such a position without violating the homogeneity requirement. I have shown that all superficially different interpretations of durative phrases actually reflect one and the same meaning. In addition,
I have argued for a decompositional analysis of verb meanings in overt syntax.
This decompositional analysis predicts that a durative phrase modifying a result
state must be adjoined to the most deeply embedded result XP projection.
Consequently, such durative phrases can appear only after the object NP, in
contrast to durative phrases modifying a process, which allow both the durati ve-object and object-durative orders. I also have provided an alternative
analysis of incremental theme verbs, in which they are analyzed as being inherently telic. These results have several interesting theoretical implications.
First, if the proposed account of Chinese durative phrases is correct, it constitutes an argument for a structured semantic representation, for example, von
Stechow's event decomposition in overt syntax, in which a result state can be
modified. Second, the analysis stipulating that durative phrases are free to adjoin to any (maximal) projection as long as we can interpret them in such a
position provides further support to Haider's (2000) and Ernst's (2002) claims
that there is no specific adjunction site for adverbials as long as the adjunction
site meets the interpretational potential or the lexico-semantic property of the
adverbials. Finally, if the proposed analysis of Incremental Theme verbs is
correct, it implies that Chinese accomplishment verbs include a result state as
part of the verbal semantics, as is reflected by the syntactic distribution of
Chinese durative phrases. This contrasts with the recent proposal made by
Kratzer (2004) that accomplishment verbs in some languages are inherently
atelic.
References
Barwise, J. and Cooper, R. (1981): Generalized Quantifiers and Natural Language.
Linguistics and Philosophy 4, 159-219.
Basilico, D. (1998): Object Position and Predication Forms. Natural Language and
Linguistic Theory 16.3: 541-595.
Bowers, J. (1988): A Structural Theory of Predication, ms. Cornell University.
Jo-wang Lin
52
Caudal, P. (1999): Result Stages and the Lexicon: The Proper Treatment of Event
Structure. In: Proceedings of the Ninth Meeting of the European Chapter of the
Association for Computational Linguistics (EACL'99). University of Bergen, Bergen, Norway.
Dini, L. and Bertinetto, P. M. (1995): Punctual Verbs and the Ontology of Events. Quad.
Lab. Ling. SNS 9/1995.
Dowty, D. (1979): Word Meaning and Montague Grammar. Reidel: Dordrecht.
Ernst, T. (1987): Durational Adverbials and Chinese Phrase Structure. Journal of Chinese Language Teachers Association 22.2: 1-11.
Ernst, T. (2002): The Syntax of Adjuncts. Cambridge: Cambridge University Press.
Filip, H. (1999): Aspect, Eventuality Types and Noun Phrase Semantics. New York:
Garland Publishing.
Filip, H. (2001): Nominal and Verbal Semantic Structure: Analogies and Interactions.
Language Sciences 23: 453-501.
Foli, R. (2002): Resultatives: Small Clauses or Complex VPs?. In: Beyssade, Claire,
Bok-Bennema, Reineke, Drijkoningen, Frank and Monachesi, Paola (eds.) Romance
Languages and Linguistic Theory 2000. Amsterdam/Philadelphia: John Benjamins
Publishing Company, pp. 153-170.
Haider, H (2000): Adverb Placement - Convergence of Structure and Licensing. In:
Alexiadou, A. and Svenonius, P. (eds.). Linguistics in Potsdam, vol. 6, Potsdam:
University of Potsdam, pp. 50-77.
Heim, I. and Kratzer, A. (1998): Semantics in Generative Grammar. Maiden: Blackwell
Publishers.
de Hoop, H. (1998): A Semantic Reanalysis of the Partitive Constraint. Lingua
103:151-174.
Huang, C.-T. James (1982): Logical Relations in Chinese and the Theory of Grammar.
Doctoral dissertation. MIT.
Huang, C.-T. James (1991): On Verb Movement and Some Syntax-Semantics Mismatches in Chinese. In: Proceedings of the 2 nd International Symposium of Chinese
Languages and Linguistics. Taipei: Academia Sinica.
Huang, C.-T. James Huang (1997): On Lexical Structure and Syntactic Projection. In:
Proceedings of the 3rd International Symposium of Chinese Languages an Linguistics. Taipei: Academia Sinica.
Jackendoff, R. (1977): X-bar Syntax: A Study of Phrase Structure. Camridge, Mass: MIT
Press.
Kratzer, A. (1996): Severing the External Argument from its Verb. In: J. Rooryck and
Zaring, L. (eds.). Phrase Structure and the Lexicon. Dordrecht: Kluwer Academic
Publishers.
Event decomposition
53
Kratzer, . (2004): Telicity and the Meaning of Objective Case. In Guron, Jacqueline
and Lecarme Jacqueline (eds.) The Syntax of Time. Cambridge: The MIT Press, pp.
389-424.
Kratzer, A. (forthcoming): The Event Argument and the Semantics of Verbs. Manuscript. University of Massachusetts, Amherst.
Krifka, M. (1989): Nominal Reference, Temporal Constitution and Quantification in
Event Semantics. In: Renate Bartsch, Johan van Bentham, and Peter van Emde Boas
(eds.) Semantics and Contextual Expressions. Dordrecht: Foris, pp. 197-235.
Rrika, M. (1992): Thematic Relations as Links Between Nominal Reference and Temporal Constitution. In: Ivan Sag and Anna Szabolcsi (eds.). Lexical Matters. Stanford: CSLI, pp. 29-53.
Krifka, M. (1998): The Origins of Telicity. In Susan Rothstein (ed.) Events and Grammar. Dordrecht: Kluwer Academic Publishers, pp. 97-134.
Larson, R. (1988): On double Object Constructions. Linguistic Inquiry 19: 335-391.
Li, Y.-H. Audrey (1987): Duration Phrases: Distributions and Interpretations. Journal of
Chinese Language Teachers Association 22.3: 27-65.
Liu, F. (2003): Definite NPs and Telicity in Chinese. Snippets, July issue.
Moltmann, F. (1991): Measure Adverbials. Lingusitics and Philosophy 14.4: 629-660.
Paul, W. (1988): The Syntax of Verb-Object Phrases in Chinese: Constraints and Reanalysis. Paris: Languages Croiss.
Pin, C. (1995): An Ontology for Event Semantics. Doctoral dissertation. Stanford
University.
Pifin, C. (1999): Durative Adverbials for Result States. Paper presented at WCCFL18,
University of Arizona, Tucson.
Ramchand, G. (2003): First Phase Syntax. Manuscript. University of Oxford.
Sanz, M. (2000): Events and Predication: A New Approach to Syntactic Processing in
English and Spanish. Amsterdam/Philadlphia: John Benjamin.
Selkirk, E. (1977): Some Remarks on Noun Phrase Structure. In: Culicover, P., Watsow,
T. and Akmajian, A. (eds.) Formal Syntax. New York: Academic Press, pp. 285-316.
Soh Hooi Ling and Kuo, Y. Jenny (2001): Perfective Aspect And Accomplishment
Situations in Mandarin Chinese. In: Proceedings of the Conference Perspective on
Aspect. December 12-14. Utrecht Institute of Linguisitcs.
Sybesma, R. (1999): The Mandarin VP. Dordrecht: Kluwer Academic Publishers.
Tang, C.-C. Jane (1990): Chinese Phrase Structure and the Extended X'-Theory. Doctoral dissertation. Cornell University.
Tang, C.-C. Jane (1994): Conditions on the Distribution of Postverbal Duration and
Frequency Phrases in Chinese Revisited. Chinese Language and Linguistics 2:
641-669.
54
Jo-wang Lin
Tatevosov, S. (2002): Quantized Nominal Arguments Yielding Cululative Verbal

Predicates: Aspectual Composition and NP Semantics. Paper presented in Sinn Und
Bedeutung VII. University of Konstanz, Germany.
Teng, S. (1975): Predicate Movements in Chinese. Journal of Chinese Linguistics 3(1):
60-75.
von Stechow, A. (1995): Lexical Decomposition in Syntax. In: Egli, U., Pause, P. E.,
Schwarze, Ch., Stechow, A. von, Wienold, G (eds.). The Lexicon in the Organization
of Language. Amsterdam/Philadephia: John Benjamin Publishing Company, pp.
81-118.
von Stechow, A. (1996): The Different Readings of Wieder "Again": A Structural Account. Journal of Semantics 13: 87-138.
Zhang, Z. (1999): Aspect and Discourse Properties of Definite and Indefinite NPs. In:
He, Baozhang and Wenze Hu (eds.) The Proceedings of 11th North American
Conference on Chinese Linguisitcs. East Asian Language Programs. Harvard University.
Minjeong Son and Peter Cole (Delaware)
Syntactic decomposition of events in Korean and

Standard Indonesian*
1 Introduction
This paper provides empirical evidence in favor of lexical decomposition of
events in the syntax. We demonstrate that a direct mapping between lexical
semantics associated with event structure and syntax can provide satisfactory
explanations not only for the scope ambiguity of tasi 'again' associated with
morphological causatives in Korean but also for the apparent lexical ambiguity
of the verbal suffix -kan in Standard Indonesian.
The idea of representing lexical semantics associated with event structure in
the syntax goes back to McCawley (1968) in the Generative Semantic framework. Since then, there have been a number of different approaches to event
structure in lexical semantic representations employing different mechanisms
and rules (e.g. Carter 1976, Dowty 1979, Levin and Rapoport 1988, Jackendoff 1990). Despite the different principles and issues a now widely held position in the literature on verbal meaning is that the lexical semantic representation of eventive verbs like 'break' or 'kill' involves complex event structure
with semantic components like CAUSE and BECOME (e.g. Dowty 1979);
events are not unanalyzable atomic units but are decomposable into subparts of
events such as causation and change of state. The logical form of 'break,' for
instance, can be represented along the lines of (1), which can be roughly paraphrased as 'x causes y to change into a state of being (or becoming) broken.'
(1)
(a)
[ x CAUSE
(b)
[x CAUSE [y BECOME (AT) broke BY [x 'wipe' floor]]]
[ y BE BROKE ] CHANGE]]
(c)
[[x DO something] CAUSE [BECOME [the cup broke]]]
Carter (1976)
Levin and Rapoport (1988)

Dowty (1979)
We would like to thank the audience at the Workshop on Event Structures in Linguistic Form
and Interpretation for helpful comments and questions. We have also benefited greatly from the
insightful comments and suggestions of Satoshi Tomioka and Benjamin Bruening on earlier
versions of this work. Special thanks go to Yassir Tjung for the data on Indonesian.
56
A growing body of literature (e.g., Hale and Keyser 1993; Travis 2000) has
shown that the lexical semantic representations of verbs associated with event
structure are directly reflected in the syntactic structure, where the semantic
components like CAUSE are present in the syntax as abstract morphemes. This
is illustrated in (2), taken from Hale and Keyser (1993), the structure of which
is argued to be operative in the lexicon, i.e., lexical syntax.
(2)
Hale and Keyser (1993)

VP
^ ^
Causing Event
NP
V'
John
^ , ^ " - ^ ' '
C A U S E (=v) ^ ' ' VP

NP
the
Caused Event(uality)
V'
cup
AP
break
The postulation of CAUSE in (2) has received considerable empirical support,

since in many languages the CAUSE predicate associated with an external
argument (e.g. John) is overtly realized in the morphology. In Salish and Tagalog, for instance, causative verbs are derived by an overt causative morpheme
that introduces an external causer argument. Examples of such morphemes are
illustrated in (3) and (4) from Salish and Tagalog, respectively.
Salish (Davis and Demirdache 2000)
(3)
Unaccusative
Causatives
(a)
kwis
'to fall'
Vkwis-ts
'to drop something'
(b)
t'iq
'to arrive/to get here'
Vt'iq-s
'to bring something'
(c)
us
'to get thrown out'
Vus-W
'to throw out something'
Tagalog (Travis 2000)

(4)
Unaccusative
Causatives
(a)
t-um-umba
'x fall d o w n '
m-pag-tumba
'y knock d o w n '
(b)
s-um-abog
'x explode'
m-/>ag-sabog
'y scatter x '
(c)
1-um-uwas
'x go into the city'
m-/>ag-luwas
'y take x '
The postulation of a result-state-denoting constituent, which roughly corresponds to the lower VP in (2), has also been shown to be grammatically substantial from the semantic and syntactic facts with respect to adverbial modifi-
We use Pesetsky's (1995) notation of V to express the root morpheme.
57
cation (e.g., again, almost, etc.). For instance, it has been argued that different
readings of 'again' in German and English provide a test for the syntactic and
semantic decomposition of predicates and for a syntactic constituent denoting
a result state encoded in the verb meaning (see von Stechow 1996 and Beck
and Johnson 2004 for examples in German and English).
In the two sections which follow, we shall show that Korean and Standard
Indonesian provide additional empirical support for the view that the lexical
decomposition of events is explicitly reflected in the (morpho-)syntax. We
argue that Korean not only displays an overt instantiation of CAUSE associated with an external argument but also reveals the existence of a result-statedenoting constituent by replicating the 'again' test used for German and English. Indonesian is shown to exhibit an overt morphological operation sensitive
to a result-state-denoting constituent.
The analysis presented in this paper relies on the notion of "Late-Insertion"
as proposed by Halle and Marantz (1993), according to which lexical realization occurs en route to Phonological Form. In other words, syntactic categories
are purely abstract, having no phonological content. The phonological expressions of lexical items are inserted in a process called Spell-Out only after syntactic operations take place. Based on this approach, it is not necessary to make
a distinction between lexical syntax (Hale and Keyser 1993) and regular
clausal syntax. We consider both to be one module, that is, syntax. The semantic analysis is rendered in neo-Davidsonian fashion using event variables,
which are always bound by an existential quantifier since event arguments
have no overt correlates in the syntax, unlike other thematic arguments of
verbs (e.g., agents, themes, etc.). In this account, verbs are taken as one-place
predicates of events to individuals. These are the individuals who participate in
the events, and the semantic and lexical content of the thematic role indicates
the nature of the participation.
2 Overt realization of CAUSE: Korean

2.1 Ambiguity of tasi 'again' and morphological causatives
In this section, we examine morphological causatives in Korean, the predicates
of which are formed by attaching the suffix -i- to verbal roots. By replicating
the 'again' test, used for German by von Stechow (1996), we demonstrate that
the combination of a verb plus -i- does not form a semantic and syntactic unit,
but is decomposable into two separate grammatical components. The suffix
-i-, then, is argued to be an overt realization of CAUSE associated with an
58
external argument. Sentences in (5) and (6) exemplify morphological causatives (hereafter MCs) with unaccusative verb bases. 2
(5)
(a)
sakwacup-i
nok-ess-ta.
apple juice-NOM
melt-PST-DEC
'The apple juice melted.'

(b)
Chelswu-ka
sakwacup-ul
nok-i-ess-ta.
Chelswu-NOM
apple juice-ACC
melt-CAUSE-PST-DEC
'Chelswu melted the apple juice.'

(6)
(a)
aiskrim-i
el-ess-ta.
ice cream-NOM
freeze-CAUSE-PST-DEC
'The ice cream froze.'

(b)
Yenghi-ka
aiskrim-lul
el-//-ess-ta.
Yeonghi-NOM
ice cream-ACC
'Yeonghi froze the ice cream.'
As seen above, the causative verbs in the (b) sentences are formed by attaching
the suffix -i- to the unaccusative predicates, Vnok 'melt' and Ve/ 'be(come)
frozen'. In addition, the causative sentences in (b) involve an additional argument that is interpreted as a causer of the event described by the base verb; the
causative morpheme introduces an external argument.
Despite the visibly realized morpheme -i- associated with a causative interpretation, it has often been argued (e.g., Kim 1998) that the causative verbs in
(5) and (6) with a non-agentive causee (hereafter non-agentive MCs) are not
decomposable but enter into syntax as atomic units; non-agentive MCs are
formed in the lexicon (see Kim 1998 and Um 1995 for the lexical approach to
non-agentive MCs). 3 However, a consideration of further data associated with
eventive adverbs like 'again' suggests that non-agentive MCs are semantically
decomposable into a causing event and a result state, as also argued by Son
(2004) with transitive verb bases (e.g., verbs of the 'put-on' class). Consider
(7):
(7)
(a)
Chelswu-ka
sakwacup-ul
tasi
nok-/-ess-ta.
Chelswu-NOM
apple juice-ACC
again
melt-CAUSE-PST-DEC
(i)
'Chelswu caused [the apple juice to melt again].'
(ii) 'Chelswu again caused the apple juice to melt.'
The following abbreviations are used in this paper.

NOM: Nominative ACC: Accusative DAT: Dative CAUSE: Causative Morpheme PST: Past
Tense DEC: Declarative Marker
It has been argued that morphological causatives in Korean are divided into at least two types:
agentive MCs, in which the causee is interpreted as an agent, and non-agentive MCs, in which
the causee is interpreted as a non-agent (e.g., theme) (see Son 2004). Since there is no controversy concerning syntactic and semantic decomposability of causative predicates associated
with agentive MCs, our discussion in this paper is restricted to non-agentive MCs.
(b)
Yenghi-ka
aiskrim-lul
tasi
el-//-ess-ta.
Yenghi-NOM
ice cream-ACC
again
(i)
59
'Yenghi caused [the ice cream to freeze again].'
(ii) 'Yenghi again caused the ice cream to freeze.'
As shown in (7), tasi 'again' creates a scope ambiguity; both sentences in (7)
are ambiguous between two readings, a restitutive and a repetitive reading. The
restitutive reading in (7a-i), for example, presupposes that there was an event
during which the apple juice melted before, but it is not required that Chelswu
himself caused that event before; in fact, it is not necessary that anyone caused
that event. In contrast, the repetitive reading in (7a-ii) presupposes that
Chelswu himself had made the apple juice melt before; Chelswu melted the
apple juice twice. Therefore, tasi can modify either the result state, as in (7a-i),
or the causing event, as in (7a-ii).
Adopting the structural theory of 'again' proposed by von Stechow (1996),
we suggest that the ambiguity with respect to the scope of tasi is sufficient
evidence that non-agentive MCs involve two events, and thus have a complex
underlying structure. The scope ambiguity of tasi in (7) can be readily accounted for if we decompose the verb nok-i- into 'melt + CAUSE,' and this
decomposition is reflected in the syntax. For instance, the ambiguity of (7) can
be explained by assuming that tasi has narrow scope with respect to CAUSE if
it is generated in the lower position, as in (8a), hence the restitutive reading. If
tasi is generated in the higher position, it has wide scope with respect to
CAUSE, as shown in (8b), and we get the repetitive reading.
(8)
(a)
[[vp again [Vp put-on]] CAUSE]
Restitutive reading
(b)
[again [Vp [vp put-on] CAUSE]]
Repetitive reading
With this decompositional approach in mind, in the subsequent sections, we

shall provide an explicit semantic and syntactic representation of non-agentive
MCs and that of the scope ambiguity of tasi 'again'.
2.2 Analysis
Before turning to the main analysis, the following subsections will review the
position of external arguments and the representation of CAUSE, both of
which are presupposed in the present analysis. The core ideas are based on
Kratzer (1996) and Pylkknen (2002), respectively.
2.2.1 The position of external arguments and Voice
Kratzer (1996) proposes that external arguments are not arguments of verbs,
but rather arguments of a separate functional element, namely VoiceP, equiva-
60
lent to vP in Chomsky (1995), to Predicate Phrase in Bowers (1993), or to a

neo-Davidsonian predicate Agent (x, e); the external argument is introduced by
the functional head Voice in the projection immediately above VP. Kratzer
notes that external arguments seem to have a special status because they are
rarely able to trigger a special interpretation of the verb (e.g., idiomatic readings), unlike internal arguments. Therefore, she suggests that there is a distinction between the internal arguments of verbs, which are part of the lexical
entry and appear in the lexical semantic representation as arguments of the
main predicate, and external arguments, which are introduced by a separate
functional head Voice (=v). She further proposes that the external-argumentintroducing head combines with the VP by a rule of semantic composition
which she dubs "Event Identification". This compositional rule is a special
kind of conjunction operation which chains together various conditions for the
event described by a sentence. In particular, the conjunction operation combines the external argument with the event described by the VP, as stated and
illustrated in (9) and (10).
(9)
Event Identification (Kratzer 1996):

/
<e <s,t>
<s,t>
. Ag(e,x) . wash (e, clothes)

(10)
J
Ag
h
<e, <s,t>>
^ by Event
. Ag(e,x) & wash (e, clothes)

Identification
h < e < s , t . Ag(e,x) & wash (e) & Th(e, clothes)

Voice
VP g <s,t> ^35() & Th(e, clothes)
/ <e<s,t
XxXc.Ag(e.x) wash
clothes
. wash(e) & Th(e,y)
The external arguments (e.g., agents) compose with a separate light verb (i.e.,
Voice) and then are conjoined with the lower predicate (e.g., VP) via Event
Identification. Therefore, in (9) and (10), the verb combines with its object to
produce a function from an event argument to a truth value, g <s,t>. The voice
head / of type <e<s,t>> is a function taking an agent and an event as arguments. Event identification combines these functions by unifying their event
arguments. We will adopt Kratzer's external-argument-introducing Voice for
the position of external arguments in underlying representations of both Korean and Indonesian sentences.
2.2.2 Representation of CAUSE (Pylkknen 2002)
For the representation of CAUSE, we adopt Pylkknen's (2002) proposal that
CAUSE and the external thematic relation form a syntactic unit in the formation of causatives in Korean; a causal relation between two events and the
61
external theta-role, (e.g., causer, agent, source, etc.), are bundled into one
causative morpheme, as illustrated in (11).4
(11)
The causative morpheme : [CAUSE, ], where

CAUSE : Xf<s,t> . 3e' [f (e') & CAUSE (e, e')] and
: . (e,)
The type 's' refers to eventualities, 'e' is the type of individuals, a n d ' t ' is the
type of truth-values. Thus, CAUSE is of type s , t , > < s , t and 6 E xts of type
<e,<s,t>>. CAUSE is defined as 'for all eventualities e, e\ CAUSE (e, e')=l
iff e ' is a caused eventuality of e.5 This analysis is in line with the Distributed
Morphology assumption that the current analysis is based on, which hypothesizes that morphemes are bundles of features and that it is these feature bundles that occur in the terminal nodes of syntax (Halle and Marantz 1993,
among others). Therefore (11) maps two interpretable features into a single
syntactic head and yields the structure and the interpretation as (12) and (13),
respectively.
(12)
CAUSE-Voice
J o / m ^ ^ C A U S E . Voice'
[CAUSE, ]
(13)
VP
CAUSE-Voice
Mary
STEP2: ( (CAUSE VP)): . 3e' [ (e, ) & melt(e')

& Th(e', ice) & CAUSE (e, ')]
STEP 1: (CAUSE (VP)): . 3' [melt(e') & Th (e', ice) & CAUSE (e, e')]
[CAUSE,
VP (melticc)
Adopting Pylkknen's (2002) proposal concerning the interpretation of the

CAUSE-Voice head in (12), we assume that the contents of a CAUSE-Voice
head in which two features are bundled together is semantically complex, as
shown in (13). In other words, CAUSE and are a unit only syntactically,
but not semantically. The interpretation of the CAUSE-Voice head thus is
4
Pylkknen (2002) s8parat8s CAUSE from an 8xternal-argument-introducing Voice based on a

cross-linguistic observation that in some languages a causative meaning is possible without an
external argument, e.g., unaccusative causatives (e.g., Finnish desiderative causatives). She further proposes that this s8paration may be universal and that CAUSE itself would never introduce an external argument. Since we observe no empirical evidence for the 8xist8nce of unaccusative causatives in Korean, we assume that two features responsible for the causative meaning and the introduction of an external argument, CAUSE and , form a syntactic unit that
receives an overt realization as -/-.
The term 'eventuality' is a neutral term for events, proposed by Bach (1981) in event semantics, given that sometimes the term event excludes states (8.g., Kratzer 1996; Harley 1995).
62
carried out in two steps, in which CAUSE applies before (see Pylkknen
2002 for justification of this approach).
We assume that causation associated with causative predicates in Korean is
expressed by the same syntactic head that introduces an external argument
(henceforth, CAUSE-Voice). The semantic and syntactic representation of
CAUSE-Voice presented in (13) is assumed to be the same in the underlying
functional architecture of all types of causatives in Korean.
2.2.3 Semantic and syntactic decomposition of causative predicates
Based on the assumptions regarding the position of an external argument and
the representation of CAUSE outlined above, let us now consider how morphological causatives are represented syntactically and semantically in Korean.
Example (5b) is repeated as (14).
(14)
Chelswu-ka
sakwacup-ul
nok-i-ess-ta.
Chelswu-NOM
apple juice-ACC
melt-CAUSE-PST-DEC
'Chelswu melted the apple juice.'
(14) receives a semantic and syntactic representation as (15).6

(15)
CAUSE-VoiceP . 3e'[Ag(e, Chelswu) & melt(e') & Th(e', apple juice)

& CAUSE (e, e')]
Chelswu
STEP2: . 3e'[Ag(e, y) & melt(e')

& Th(e', apple juice) & CAUSE (e, e')]
STEP1: . 3e'[melt(e') & Th(e', apple juice) & CAUSE(e, e')]
VP
apple juice
CAUSE-Voice
Xf<s,t> . 3e'[f (e') & CAUSE (e, ')]
Vmelt . ) & Th(e, )
The interpretation based on the semantic composition in (15) can be paraphrased as 'there exists e ' such that the apple juice melts in e ' and e ' is a result
state of e such that Chelswu is an agent in e.'
By having a VP as a complement of CAUSE under the decompositional approach, an explanation for the scope ambiguity associated with tasi 'again'
For the und8rlying representation of the inchoative counterpart nok- melt (intran.)', we a s s u r e
that VP merges with BECOME-Voice, which does not introduce an external argument in its
specifier position. This assumption is in line with Harley (1995) and Folli and Harley (2002),
who argue that eventiveness of predicates is determined by the presence of an Event Phrase
(equivalent to VoiceP in our analysis) and that there are different kinds of Voice heads (e.g.,
CAUSE-Voice, BECOME-Voice, etc.). See Son (2004) for a discussion on how the argument
structure of predicates in Korean is determined based on this assumption.
63
falls out naturally; tasi can be adjoined either to the VP, where it takes scope
over the caused event, or to the CAUSE-VoiceP where it takes scope over
CAUSE (i.e., a causing event). Different syntactic positions of tasi, therefore,
trigger two different interpretations, the restitutive reading, as in (16b), and the
repetitive reading, as in (16a).
(16)
(a)
CAUSE-VoiceP again ([. 3e'[Ag(e, Q & melt(e') & Th(e\ apple juice)
& CAUSE (e, e')]](e")
tasi
CAUSE-VoiceP Xe.3e'[Ag(e, Q & melt(e') & Th(e\ apple juice)
repetative^^
Chelswu
'
& CAUSE (e, e')]

STEP2: ye. 3e'[Ag(e, y) & melt(e')
& Th(e', apple juice) & Cause(e, e')]
STEP1: . 3e'[melt(e') & Th(e', apple juice) & CAUSE(e, e')]
(b)
VP
tasi
CAUSE-Voice ^, . 3e'[f (e') & CAUSE(e, e')]

VP
restitutive^-^'
apple juice
again (['. melt(e') & Th(e', apple juice)])(e")
_
Vmelt . melt(e) & Th(e, x)
For the meaning of 'again,' we adopt the semantics proposed by von Stechow
(1996); 'again' is of type s , t > < s , t , a function taking two event arguments.
Its meaning is defined as (17) and (18).7
(17)
(18)
Let be a property of eventualities and let e be an eventuality.

again(P)(e) is defined only if 3e' [ [MAX] (P)(e') = 1 & e' < e ]
Where defined, [again] (P)(e) = 1 iff P(e) = 1
MAX is a symbol of type s , t > , < s , t . [MAX] (P)(e) = 1 iff P(e) and there is no e' such
that e is a proper part of e' and P(e') = 1
The semantics of again presupposes that the same kind of event had occurred
previously. Therefore, the denotation of the higher VP in (16b), the restitutive
reading, can be read as 'e ' is an event such that the apple juice melted in e ' and
there has been a maximal event of the same kind before'. The repetitive reading is derived by adjoining tasi to the CAUSE-VoiceP. The denotation of the
higher CAUSE-VoiceP in (16a), therefore, can be read as 'there exists e' such
that the apple juice melted in e ' and e ' is a caused event of e in which Chelswu
is an agent in e, and there has been a maximal event of the same kind as e
before'.
The definition is read as 'P is a property of eventualities, and '<' is the relation of temporal
precedence'. It is true of any two eventualities if the first is temporally located entirely before
the second. MAX(P)(e') means that e' is a maximal P-event, where MAX is defined as in (18).
64
2.3 Lexical causatives and ambiguity of tasi 'again'

It should be noted that the aforementioned again test shows ambiguity for
lexical causatives as well, the causative meaning of which is embedded in a
single lexical verb without an overt causative morpheme. Consider examples
in (19) and (20).
(19)
(20)
Chelswu-ka
mwun-ul
tasi
yel-ess-ta.
Chelswu-NOM
door-ACC
again
open-PST-DEC
(a)
'Chelswu [open the door again],'
(b)
'Chelswu again [open the door].' [Repetitive]
[Restitutive]
Sensayngnim-i
Yenghi-eykey
(kyosil) chengso-lul
tasi
siki-ess-ta.
teacher-NOM
Yeonghi-DAT
(classroom)cleaning-ACC
again
cause. do-PST-DEC
(a)
'The teacher made [Yeonghi do the cleaning again].'
[Restitutive]
(b)
'The teacher again made [Yeonghi do the cleaning].'
[Repetitive]
Both sentences in (19) and (20) are ambiguous between two readings, a restitutive and a repetitive reading. The restitutive reading in (20a), for example,
presupposes that there had been an event during which Yeonghi cleaned the
classroom (did the cleaning) sometime before, but it is not required that the
teacher himself/herself caused that event. On the other hand, the repetitive
reading presupposes that the teacher himself/herself had made Yeonghi clean
the classroom before. The fact that tasi creates scope ambiguity when it occurs
with lexical causatives indicates that they also involve more complex event
structures than was previously recognized. We argue that there exists an abstract CAUSE morpheme present in the underlying representation of (19) and
(20), which introduces an external argument, as shown in (21).
Therefore, tasi 'again' can be adjoined either to the VP having scope over a
result state, which derives a restitutive reading, or to the CAUSE-VoiceP having scope over the causing event, which yields a repetitive reading. The scope
ambiguity of tasi associated with lexical causatives then suggests that both
lexical and morphological causatives contain an abstract CAUSE head in their
underlying representations, regardless of whether CAUSE is visibly realized in
the morphology.
By demonstrating the scope ambiguity of tasi 'again' in connection with
non-agentive MCs and lexical causatives, we have seen thus far that Korean
provides empirical evidence for the grammatical realization of complex event
65
structure encoded in the verb meaning. It has been shown that the semantic
primitive CAUSE associated with an external argument is present as an abstract morpheme in the syntax of causatives, which is overtly realized as the
suffix -i- in morphological causatives.
In the following section, we shall show that Indonesian also provides evidence in favor of semantic and syntactic decomposition of events. Despite
patterns similar to Korean in the formation of causative predicates, Indonesian
is shown to display an overt instantiation of a different semantic component
than CAUSE, i.e., RESULT. This argument draws from the facts particular to
Indonesian, which makes use of the same morpheme to derive, among other
constructions, causative and benefactive constructions.
3 Overt instantiation of RESULT: Indonesian

In this section, we show that Indonesian manifests an overt instantiation of a
RESULT head, rather than a CAUSE head. This argument crucially relies on
the fact that the verbal suffix -kan, which is primarily used to derive a causative verb, is also found in other constructions (e.g., benefactive constructions).
In order to accommodate the fact that -kan is found in seemingly unrelated
constructions, a slightly modified version of event structure is proposed, which
contains a meaning postulate RESULT. The RESULT is analogous to CAUSE
claimed for Korean, but differs from it in terms of its reflection in the syntactic
structure. While CAUSE is generated above the verbal domain associated with
an external argument, RESULT is argued to be deeply embedded in the VP
structure.
3.1 Distribution of-
The suffix -kan is lexically ambiguous, in that it occurs in a number of different constructions. The constructions with which we are primarily concerned in
this paper are causative and benefactive constructions.8
Let us first consider the use of -kan as a causative morpheme, illustrated in
(22) and (23).
(22)
(a)
Cangkir-nya
cup-3
pecah.
break
'The cup broke.'
(b) Janet memecah-kan

Janet meN-break-KAN
cangkir-nya.
cup-3
'Janet broke her cup.'
See Son and Cole (2004) for other -kan constructions and for how the analysis presented here
extends to those constructions. See also Cole and Son (2004) for a different approach to the
syntax of -kan.
66
(23)
(a)
Balon
itu
(b) Mereka
terbang.
Balloon that
3PL
fly
menerbang-kan
balon
meN-fly-KAN
balloon that
itu.
'They caused the balloon to fly.'
'The balloon is flying.'
As seen above, the causative verbs in (b) are formed by attaching the suffix
-kan to the unaccusative predicates (e.g., pecah 'break'), in a similar fashion
to the formation of morphological causatives in Korean. In addition, the causativized sentences in (b) involve an obligatory presence of a morpheme associated with Voice, which is instantiated by meN- in the active, as shown above,
and by diin the passive (see examples in (26) and (28)). On the basis of
causative constructions alone, one could argue that -kan is another instance of
an overt realization of CAUSE, as in Korean and many other languages which
employ an overt causative morpheme. However, a number of facts associated
with the presence of -kan in benefactive constructions suggest that -kan may
not be a realization of CAUSE, but rather something else.
Let us now consider the occurrence of -kan in benefactive constructions.
This is exemplified in (24) and (25).
(24)
(a)
Tika
memanggang
roti
itu
(untuk
Tika
meN-bake
bread
the
for
Eric).
Eric
'Tika baked the bread for Eric.'

(b)
Tika memanggang-*(kan) Eric
roti
itu.
Tika
bread
the
meN-bake-KAN
Eric
'Tika baked Eric the bread.'

(25)
(a)
Eric
membuat
rumah-rumahan
itu (untuk
anak-nya).
Eric
meN-make
RED-house-AN
the
child-3
for
'Eric made the toy house for his child.'

(b)
Eric
membuat-*(kan)
anak-nya
rumah-rumahan
itu.
Eric
meN-make-KAN
child-3
RED-house-AN
the
'Eric made his child the toy house.'
The examples above show the use of -kan as an applicative suffix associated
with a benefactive interpretation. As shown in (b), when -kan is attached to a
transitive verb base, the NP object of an optional prepositional phrase in (a)
(e.g., Eric and anak-nya 'his child') occurs as a bare NP adjacent to the derived
verb (hereafter an NP+NP frame). When the benefactive argument occurs as a
bare NP, it functions like the primary object of the derived verb on the basis of
passivization. This is shown in (26).
(26)
(a)
Eric
dipanggang-kan
Eric
DI-bake-KAN
roti
itu
(oleh Tika).
bread
the
(by
Tika)
'Eric was baked that bread (by Tika).'

(b)
*Roti
itu
bread
the DI-bake-KAN
dipanggang-kan
Eric (oleh Tika).

Eric (by
'That bread was baked for Eric (by Tika).'
Tika)
67
On the basis of the examples from (24) through (26), -kan would appear to be
an applicative suffix with a distribution similar to that of prototypical applicatives found in many Bantu languages (see Baker 1988 and Marantz 1993,
among others). However, the consideration of a full range of examples suggests that Indonesian benefactives with -kan differ from typical applicatives
with respect to the effect of the affix on the argument structure. In prototypical
applicatives the object of a preposition in the base sentence must appear as the
primary object in the applicative construction. In contrast, in Indonesian, the
benefactive NP may also occur in a prepositional phrase (hereafter an NP+PP
frame). This is despite the presence of -kan on the verb, as shown in (27).
(27)
(a)
Tika memanggang-kan
roti
itu * (untuk
Tika meN-bake-KAN
bread
the
for
Eric).
Eric

(b)
Eric membuat-kan
rumah-rumahan
itu *(untuk
Eric meN-make-KAN
RED-house-AN
the
for
anak-nya).
child-3
'Eric made the toy house for his child.'
Furthermore, on the assumption that it is the primary object that is made subject by passivization, the theme, not the benefactive, is the primary object in
(27), as shown in (28).
(28)
(a)
Roti
itu
dipanggang-kan
bread the DI-bake-KAN
untuk
Eric.
for
Eric
'That bread was baked for Eric.'

(b)
Rumah-rumahan
itu
dibuat-kan
RED-house-an
the
DI-make-KAN
untuk
for
anak-nya.
child-3
'The toy house was made for his child.'
While the examples in (27) seem to have the same argument structure as those
sentences with a 'for' beneficiary without -kan (e.g., 24a-25a), syntactic and
semantic differences between them suggest that the two structures are not
identical. For instance, when -kan is present, the benefactive PP is an obligatory oblique complement (i.e., a subcategorized constituent). This is shown by
the fact that when the benefactive PP is omitted in the passive shown in (28),
the sentence is interpreted as having a null benefactive interpretation, as illustrated in (29).
(29)
(a)
Roti
itu
dipanggang-kan.
bread
the
DI-bake-KAN
'That bread was baked for someone. '

(b)
Rumah-rumahan itu
RED-house-an
the
dibuat-kan.
DI-make-KAN
'The toy house was made for
someone.'
68
Such an interpretation with an implicit beneficiary is not observed when -kan

is absent, as shown in (30).
(30)
(a)
Roti
itu
bread the
dipanggang.
DI-bake
'That bread was baked.'

(b)
Rumah-rumahan
itu
dibuat.
RED-house-an
the
DI-make
'The toy house was made.'
Another difference between the sentences with -kan and those without -kan
comes from their interpretation; the NP+PP frame with -kan is synonymous
with the NP+NP frame shown in the (b) sentences of (24) and (25). In the
NP+NP frame, the benefactive NP adjacent to the derived verb is interpreted as
a prospective possessor of the theme argument, similar to English double object constructions. For instance, sentence (24b), repeated as (31a), carries with
it a strong implication that Eric is in possession of the bread that Tika baked at
the end of the event, and only this interpretation is possible. The corresponding
NP+PP frame, given in (31b), has the same interpretation, in which Eric is
expected to get the bread as a result of Tika's baking the bread.
(31)
(a)
Tika
memanggang-kan
Eric
roti
itu.
Tika
meN-bake-KAN
Eric
bread
the
'Tika baked Eric the bread.'

(b)
Tika
memanggang-kan
roti
itu
untuk Eric.
Tika
meN-bake-KAN
bread
the
for
Eric
However, as shown in (32), the NP+PP structure without -kan is ambiguous

between a possessive and a purely beneficiary reading; the sentence can be
interpreted either as 'Tika baked the bread to give it to Eric' or 'Tika baked the
bread in place of Eric or for the benefit of Eric.'
(32)
Tika
memanggang
roti
itu
untuk
Eric.
Tika
meN-bake
bread
the
for
Eric
(a)
'Tika baked the bread to give it to Eric.'
(b)
'Tika baked the bread in place of Eric (since Eric was busy).'
The same kind of semantic contrast has been observed between the double
object and the oblique complement constructions in English, as noted, for
example, by Harley (2002) and Beck and Johnson (2004). Consider (33).
(33)
(a)
Sally knitted John a sweater.

(=John has the sweater as a result of Sally's kitting.)
(Only possession reading)
69
(b)
Sally knitted a sweater for John.

(i)
(Ambiguous)
Sally knitted a sweater to give it to John.
(ii) Sally knitted a sweater in place of John.
The sentences in (33) show an important semantic contrast; the referent of the
goal in (33a) must be a prospective possessor of the referent of the theme,
hence the possessor account (Pesetsky 1995; Harley 2002, inter alia). Due to
the possessor implication associated with the goal in (33a), double object constructions are known to have an animacy restriction on the first NP of double
NP objects, as shown in (34).
(34)
*Sally knitted Christmas a sweater.
In the oblique complement construction, such a restriction does not apply;

(33b) is ambiguous between a possession and a purely beneficiary reading
without a necessary possessive implication. John may be intended to be a
possessor, but it is also possible to interpret (33b) to describe a situation in
which Sally knitted the sweater for the sake of John or in place of John.
Therefore, as illustrated in (35), an inanimate object is allowed as a complement of the preposition, contrary to (34).
(35)
Sally knitted a sweater for Christmas.
The 'possessor' account proposed for English double object constructions

seems to extend to Indonesian -kan benefactives as well, regardless of whether
the benefactive argument is realized as an NP or a PP. This is evidenced by the
fact that both an NP+NP and NP+PP structure do not allow an inanimate NP to
appear as an applied benefactive argument, as shown in (36) and (37).
(36)
(a)
*Saya
men-(p)anggang-Ara/i
perayaan
1SG
meN-bake-KAN
celebration birthday
ulangtahun
Eric
biskuit
itu
Eric
biscuit
the
baked a biscuit for Eric's birthday.'

(b)
*Dia
mem-buat-o
perayaan
Halloween
rumah-rumahan
itu
3SG
meN-build-KAN
celebration
Halloween
RED-house-AN
the
'He built a toy house for Halloween.'

(37)
(a)
Saya men-(p)anggang-ftan biscuit itu untuk perayaan

1SG
meN-bake-KAN
biscuit the for
ulangtahun Eric
celebration birthday
Eric
baked someone the biscuit for Eric's birthday.'

(b)
Dia
mem-buat-
3SG
meN-build-KAN
rumah-rumahan
itu
RED-house-AN the
untuk perayaan
Halloween
for
Halloween
celebration
'He built someone the toy house for Halloween.'
In (37), the animacy restriction on the benefactive argument may be concealed

by the fact that Indonesian allows phonologically null arguments. However,
70
the fact that the sentences in (37) are interpreted as having an implicit goal
argument as a potential beneficiary suggests that the animacy restriction also
applies to the NP+PP frame when -kan is present.
Examples in (38) and (39) illustrate the same point: the benefactive argument in sentences with -kan is understood to be a prospective possessor which
must be animate.9
(38)
(a)
Saya
menyulam-
bayi
kita
baju hangat
ini.
1SG
meN-knit-KAN
baby
1PL
shirt
this
warm
knitted our baby this sweater.'

(b)
Saya
menyulam-Ae
baju
hangat
ini
untuk
bayi
kita
1SG
meN-knit-KAN
shirt
warm
this
for
baby
1PL
knitted this sweater for our baby (with the baby being a prospective possessor).'
(39)
Saya
menyulam
1SG
meN-knit
baju
shirt
hangat
warm
ini
untuk
bayi
kita.
this
for
baby
1PL
knitted this sweater for our baby.'
In (38), both frames of the -kan benefactive have the implication that the baby
exists. If the baby must bear a possessor role in (38) by virtue of occurring in
the -kan construction, it must be animate (i.e., alive), and hence has already
been born. In (39), in contrast, when -kan is absent, the baby may or may not
exist in a real world; the female speaker may simply be pregnant.
We have seen thus far that unlike prototypical applicative suffixes, -kan
does not necessarily make a benefactive argument the primary object of a
derived verb in applicative constructions; -kan has the effect of making either
an NP or a PP benefactive internal to the argument structure of the derived
verb, and hence allows both NP+NP and NP+PP structures. We have also seen
that the two variants of the -kan benefactive differ only in their surface structure but have the same semantic interpretation, which involves the implication
of a possession relation between the goal and the theme.
Simply inspecting causatives and benefactives, previous accounts of -kan
have assumed that the primary function of -kan is to increase the valence of a
VP by introducing an extra argument in the argument structure; -kan is a transitivizer (e.g., Sie 1988; Arka 1992; Postman 2002). In the following section,
we briefly review one of the most recent approaches to -kan that syntactically
formalizes the idea of treating it as a transitivizer. We then point out the inadequacy of the account of -kan as a realization of some functional head situated
above the VP domain, given that it fails to account for the NP+PP frame of
-kan benefactives (e.g., (27)).
These examples are inspired by the corresponding English examples provided by Harley
(2002).
71
3.2 Syntactic account of -kan as a transitivizer: Postman (2002)

Postman (2002) formalizes the idea of analyzing -kan as a transitivizer (e.g.,
Sie 1988) in syntactic structure; she argues that -kan is an overt realization of
the functional head Tr based on the Transitive Phrase Hypothesis (TPH) proposed by Bowers (2001). According to TPH, transitivity is structurally represented by the existence of a Transitive Phrase (TrP); transitivity of verbs is
determined by the number of TrPs in syntactic structure. Adopting Bower's
hypothesis, Postman (2002) claims that -kan projects an additional TrP above
the VP domain and bears the [+dative] case feature in the case of benefactive
constructions, as illustrated in (40a).
Postman assumes that the theme argument moves to the lower [Spec,TrP] to
check the ACC case feature, and then the verb moves to the lower transitive
head. This head movement makes the two NPs equidistant from the higher Tr
head. Therefore, the benefactive NP can move to the higher [Spec, TrP] to
satisfy the case feature of Tr, the head of which is realized by -kan. She further argues that since -kan attracts the benefactive NP, this syntactic licensing
of the benefactive NP by -kan explains a natural relationship between the
suffix and the benefactive NP. However, in this analysis (as in most of the
previous analyses), the NP+PP structure compatible with -kan is not mentioned. Although the dependency between the suffix -kan and the benefactive
NP is accounted for by means of case assignment, the dependency between
-kan and the benefactive PP in an NP+PP structure cannot be explained, since
as seen in (40b), the NP object of untuk 'for' cannot move to the upper [Spec,
Tr0]. If it moves, an incorrect surface word order is derived, stranding the
preposition. Furthermore, the NP movement out of PP violates the Minimal
Link Condition in the sense of Chomsky (1995). If the NP stays, however, the
case feature of Tr {-kan) remains unchecked, and the derivation crashes. The
purely syntactic account of -kan as a transitivizer thus fails to explain the
occurrence of -kan in benefactive constructions.
72
(40)
Benefactive Constructions
(a) NP+NP structure
TrP
/ \
Spec
(b) *NP+PP structure

TrP
Tr'
/ \
Spec
[+DAT,
The questions that naturally arise then are :1) how do we effectively account
for the compatibility of both structures with -kan and for the fact that -kan has
the effect of allowing both types of arguments to be internal arguments? 2)
what is the correct characterization of -kan that gives rise to the observed occurrence of the suffix in the two seemingly unrelated constructions? In order to
successfully answer these two questions, we propose a radically different approach to the function of -kan, advocating event decomposition of predicates
in the syntax and semantics.
3.3 Event-based account of -kan
An investigation of causative and benefactive constructions from the perspective of event decomposition of predicates suggests that they share the same
internal aspectual properties associated with the meaning of causation. Therefore, benefactives are treated as being parallel to causatives in their underlying
event structure, as also argued for English double object constructions by a
number of researchers (e.g., Pesetsky 1995; Harley 2002; Ramchand 2003;
Beck and Johnson 2004). We suggest that despite their superficial differences
the internal structures of the two constructions involve the same event components of a causing event and a caused eventuality, which correspond to separate verbal projections in the phrase structure. In order to correctly identify the
function of -kan in these two constructions, we propose a slightly modified
version of event structure with a meaning postulate RESULT(R), analogous to
RESULT in Ramchand (2003). By adopting the theory of event structure propounded by Ramchand (2003), we hypothesize that there is a semantic primi-
73
tive RESULT involved in -kan constructions, which projects a Result Phrase

(RP) that is deeply embedded in the VP structure, -kan is argued to be a morphological reflex of the Result head, rather than the head of some functional
projection above VP (as proposed, for example, by Postman 2002). We further
argue that it is the semantics of R that gives rise to a causative interpretation;
RESULT introduces an implicit event argument (i.e., a causing event) and
establishes a relation between a causing event and a caused eventuality (i.e., a
result state). The proposed semantics of RESULT is formalized as (41).
(41 )
Semantics of RESULT
- Causatives: Xf <e<s,t>> . Xg <s,t> . Be' [Result (e',e) & f(e',x) & g(e)]
- Benefactives: < e < s , t . Xg <e<s,t> . 3e' [Result (e',e) & f(e',x) & g (e,x)]
RESULT is defined as 'for all eventualities e, e\ RESULT ( e \ e)=l iff e ' is a

result state of e.' Given the proposed semantics of RESULT, we argue that the
postulation of an abstract CAUSE predicate in the semantic-syntactic representation of causative and benefactive constructions is not necessary, since R
entails the relevant semantic function provided by CAUSE. 10
Let us then see how the proposed analysis provides a straightforward explanation for the presence of -kan in causative and benefactive constructions.
3.3.1 Representing causatives
Based on the proposed semantics and syntax of RESULT, the meaning of the
causative sentences in (22b), repeated as (42b), can be represented along the
lines of (43).
(42)
(a)
Cangkirnya
pecah
cup-3
break
'The cup broke.'

(43)
(b)
Janet memecah-kan
cangkirnya
Janet meN-break-KAN cup-3

'Janet broke her cup.'
(a)
(b)
Paraphrase: the event such that Janet is the agent of doing something brings about a
VOICEP
Janet VOICE [ VP V-do (something) [ RP RESULT
[AP
the cup broken] ]]]
caused eventuality e ' such that the cup is broken.
The lexical semantic representation in (43a) is reflected in the syntax as (44).
10
Given that RESULT bears similarity to CAUSE head in its semantics, one could still argue that
-kan is a realization of CAUSE situated above VP. However, we reject this idea due to various
problems that arise otherwise. See Son and Cole (2004) for an argument against the treatment
of -kan as CAUSE generated above VP.
74
(44)
VoiceP
. 3e'[Result(e', e) & broken(e') & Th(e', cup) & Ag(e, Janet) & do(e)]
Ext
Janet
. 3e'[Result(e', e) & broken(e') & Th(e', cup) & Ag(e, z) & do(e)]
^ ^
"
by Event
Identification
Voice
VP . 3e'[Result(e', e) & broken(e') & Th(e',cup) & do(e)]
. Ext(e, z)
V
RP
. do-something(e) ^ ^ ^ ^
cangkirnfa
'cup'
-a/i
. 3e'[Result(e', e) & broken(e') &

Th(e', cup) & g(e)]
R1 xg<s,t> . 3e'[Result(e', e) & broken(e') &
Th(e', x) & g(e)]
AP < e , < s , t
tewas
". broken(e") & Th(e", y)
'break'
< 8 , < s , t xg <s,t> . 3e'[Result(e',e) & f(e',x) & g(e)]
Based on the proposed semantics of R in (41), the R head takes the function /
as its first argument, which corresponds to the result state that is expressed by
the Adjective Phrase in (44). The R further takes the function g denoted by the
verb as a solution for deriving the correct semantic interpretation.11 The compositional interpretation of (44) then proceeds as indicated in the proposed
structure. At the end of the composition, we get the intended interpretation as
'a set of eventualities e such that Janet is the agent of doing something in e and
there is e ' such that the cup is broken in e ', and e ' is the result of e. '
3.3.2 Representing benefactives
By adopting the idea of treating benefactive verbs as parallel to explicit causatives in Indonesian, the structural configuration and semantics proposed for
causatives can naturally extend to the -kan benefactives. For example, the
lexical semantics of the benefactive 'make' can be represented along the lines
of (45), in perfect analogy to the representation of apparent causatives.
(45)
Eterwfactivc 'mak8':
[voiceP Eric VOICE [vp making (toy house) [RP RESULT the child has the toy house.]]]
Given that the two variants of the -kan benefactive involve an implication of
possession between the goal and the theme, we assumed that the logical representation of the benefactive 'make' in (45) implies that the verb denotes causation of change of possession as part of its lexical meaning. (45) can be para-
"
V(erb) in (44) is not semantically vacuous but is assigned the meaning 'do something'. We
make this claim because the causation expressed in (43b) need not involve a direct action imposed on the theme that undergoes a change of state; (43b) can be interpreted as either 'Janet
deliberately broke the cup (by throwing it away or dropping it on the floor), or 'Janet broke the
cup by shaking the table on which the cup was placed'. We assume that such an unspecified
action is represented in the logical form of V, which tells us that Janet did the breaking and the
cup broke, but it does not specify exactly what Janet did to break the cup.
75
phrased as 'the event of Eric's making (the toy house) brings about a result
state such that the child is a possessor of the toy house'.
On the basis of different syntactic behaviors between two variants of the
kan benefactive (e.g., passivization), we take a base-generation approach to the
two different structures compatible with -kan. As shown in (46), we assume
that the benefactive argument can be realized either in [Spec, RP] as a subject
of the result state, or as a complement of PP selected by the R head. In both
cases, a dependency between -kan and the benefactive argument is ensured; kan is the locus of projecting the benefactive argument, regardless of whether
it is realized as an NP or a PP.
(46)
(a)
N P + N P Frame
(b)
RP
NP
N P + P P Frame
RP
Theme N P
Benefactive PP
A fully specified syntactic and semantic representation of (46b), for example,

is given in (47).12
(47)
VoiceP
. 3e'[Result(e', e) & in p o s s e s s i o n ^ ' , toy house, his child) & making(e)

& Th(e, toy house) & Ag(e, Eric)]
. 3e'[Result(e', e) & in possession (e1, toy house, his child) & making(e) &
Ext
Eric
'
-Voice
r-
by Event
Identification
. 3e'[Result(e', e) & in possession (e', toy house, his child)

& making(e) & Th(e, toy house)]
. Ext (e,x)
.
Th(e, toy house) & Ag(e, x)]

VP
RP X g < e < s , t X e . 3e'[Result(e', e) & in possession (e', toy
buat ' m a k e ' ^ ^ ^ ^ ^
ru
rumah-rumahan
'the toy h o u s e '
'the
XyXe. making(e) &
-kan
itu
house, his child & g(e, toy house)]

R ' z g < e < s , t e . 3e'[Result(e', e) &
in p o s s e s s i o n ^ ' , z, his child) & g(e,z)]
PP " . is in possession of his child in e"
Th(e,y)
untuk ' f o r '
DP
anak-nya 'his child'
" ' is in possession of y in e" ( = in possession(e", x, y))

< e < s , t <e<s,t> . 3e'[Result(e', e) & f(e', ) & g(e, )]
In benefactives, the result state is expressed by a PP, which involves a necessary possession relation between the goal and the theme. This denotation is
reflected in the semantics of P, which we argue is overtly realized as 'for' for
case reasons. The interpretation based on the semantic composition given in
(47) can be expressed as 'a set of eventualities e such that e is making the toy
12
See Son and Cole (2004) for the fully specified semantic and syntactic representation of
N P + N P structure.
an
76
house and Eric is the agent of e and there is an e ' such that his child is in possession of the toy house in e ' and e ' is the result eventuality of e. '
As we have seen thus far, the presence of -kan in causative and benefactive
constructions receives a natural explanation in our event-based account in
which the lexical semantic representation of verbs associated with events is
directly mapped into the syntax. By treating -kan as an overt instantiation of
the RESULT head, our analysis of -kan not only provides important empirical
support for the semantic and syntactic decomposition of predicates but also
leads to a unified account of -kan, which straightforwardly captures the occurrence of the suffix in seemingly unrelated constructions. Furthermore, by demonstrating an overt instantiation of a result-state-denoting constituent in the
syntax of a benefactive construction, our analysis supports the idea of treating
it as parallel to causatives in their structure and meaning in English (e.g., Pesetsky 1995; Harley 2002; Beck and Johnson 2004).
4 Concluding remarks
We have shown so far that Korean and Indonesian provide empirical support
for the predicate decomposition of events in the (morpho-)syntax. It has been
argued that the syntactic and semantic decomposition of predicates associated
with event structure provides straightforward explanations for a number of
linguistic phenomena in Korean and Indonesian. We showed that the scope
ambiguity of the eventive adverb tasi 'again' associated with lexical and morphological causatives in Korean can be easily accounted for if we adopt the
decompositional model of predicate formation. It was also shown that an abstract CAUSE present in the underlying representation of causatives is overtly
realized by the suffix -/-, which shows a close correlation between the semantic decomposition of events and the morpho-syntax. Indonesian has also been
argued to provide empirical support for the lexical decomposition of events in
the semantics and syntax. We have argued that Indonesian manifests an overt
instantiation of a RESULT predicate that we postulated in order to accommodate the facts peculiar to Indonesian; the verbal suffix -kan is lexically ambiguous occurring in seemingly unrelated constructions, causative and benefactive constructions. In order to provide a correct characterization of the function of -kan, we have argued that -kan is an overt realization of RESULT
situated inside the verbal domain, rather than CAUSE generated above VP.
Due to the semantics of the RESULT that gives rise to a causative interpretation, the postulation of an abstract CAUSE morpheme in the underlying representation of causatives and benefactives was not necessary.
The analysis propounded in this paper is of both empirical and theoretical
interest because it raises questions concerning whether languages differ in the
way semantic components of events are mapped into the syntax, i.e., whether
or not different languages grammaticalize the same semantic components of
77
events. Although it is rather premature to make a broader typological claim,

investigation of Korean and Indonesian indicates that although both languages
form causative verbs by means of an overt causative suffix, they nevertheless
differ in terms of which semantic component of causation is reflected in the
syntax. In Korean, it is the CAUSE predicate associated with an external argument that is overtly realized in the (morpho-)syntax. In Indonesian, however, it is the RESULT predicate that receives an overt realization, and the
RESULT is irrelevant to introducing an external argument, unlike CAUSE in
Korean.
References
Arka, I.W. (1992): The -kan
causative in Indonesian. M.Phil, thesis, University of
Sydney, Australia.
Bach, E. (1981): On time, tense, and aspect: an essay in English metaphysics. In: Peter
Cole (ed.), Radical Pragmatic, Academic Press, New York.
Baker, M. (1988): Theta theory and the syntax of applicatives in Chichew-La. Natural
Language and Linguistic Theory 6:353-389.
Beck, S. and Johnson, K. (2004): Double object again. Linguistic Inquiry, Vol. 35.
Issue 1, 97-123. MIT Press.
Bowers, J. (1993): The syntax of predication. Linguistic Inquiry 24, 591-656. MIT
Press.
Bowers, J. (2001): Transitivity. Linguistic Inquiry 33, 183-224.
Carter, R. J. (1976): Some linking regularities. In On linking: Papers by Richard Carter,
ed. . Levin and C. Tenny, 1-92. Cambridge, MA: Lexicon Project, MIT Center
for Cognitive Science.
Chomsky, N. (1995): The minimalist program. Cambridge, MA: MIT Press.
Cole, P. and Son, M.J. (2004): The argument structure of verbs with the suffix - k a n in
Indonesian. To appear in Oceanic Linguistics, Volume 43, no. 2 (December 2004),
University of Hawai'i Press.
Davis, H., and Demirdache, H. (2000): On lexical verb meaning: evidence from Salish.
In: C. Tenny and J. Pustejovsky, (eds.), Events as grammatical objects, 97-141.
Stanford: CSLI.
Dowty, D. (1979): Word meaning and Montague grammar. Dordrecht: Reidel.
Folli, R. and Harley, H. (2002): Consuming results in Italian and English:vors of v.
Paper presented at the NSF Workshop on Aspect, University of Iowa. To appear in
a Kluwer volume edited by Paula Kempchinsky and Roumyana Slabakova.
Hale, K. and Keyser, J. (1993): On argument structure and the lexical expression of
syntactic relations. In The view from building 20: A festschrift for Sylvain Bromberger, ed. Ken Hale and Jay Keyser, 53-108. Cambridge, MA: MIT Press.
78
Halle, M. an Marantz, A. (1993): Distributed Morphology, in Hale and Keyser eds. The
View from Building 20, MIT Press: Cambridge, Massachusetts.
Harley, H. (1995): Subjects, Events, and Licensing, Ph.D. Thesis, MIT.
Harley, H. (2002): Possession and the double object construction. To appear in Linguistic variation yearbook vol. 2, ed. Pierre Pica and Johan Rooryck.
Jackendoff, R. (1990): Semantic structures. Cambridge, MA: MIT Press.
Kim, A. R. (1998): VP complement of Hi-Causative. In: David James Silva eds.
Japanese/Korean Linguistics 8,445-458. CSLI
Kratzer, A. (1996): Severing the external argument from its verb. In: Phrase structure
and lexicon, ed. Johan Rooryck and Laurie Zaring, 109-137. Dordrecht: Kluwer.
Levin, B. and Rapoport, T. (1988): Lexical subordination. In: Proceedings of the Chicago Linguistics Society, 275-289.
McCawley, J. (1968): Lexical insertion in a transformational grammar without deep
structure. Proceedings of the Chicago Linguistic Society 4, 71-80.
Marantz, A. (1993): Implications of asymmetries in double object constructions. In:
Theoretical aspects of Bantu grammar 1, ed. Sam A. Mchombo, 113-151. Stanford,
CA: CSLI Publications.
Pesetsky, D. (1995): Zero syntax. Cambridge, MA: MIT Press.
Postman, W. A. (2002): Thematic role assignment in Indonesian: A case study of
agrammatic aphasia. Doctoral dissertation, Cornell University, Ithaca, NY.
Pylkknen, L. (2002): Introducing arguments. Doctoral dissertation. MIT.
Ramchand, G. (2003): First phase syntax. Ms., Oxford University.
Sie, I.D. (1988): The syntactic passive in Bahasa Indonesia: a study in governmentbinding theory. Doctoral Dissertation, Universiteit van Amsterdam.
Son, M. J. (2004): A unified syntactic account of morphological causatives. In: Korean.
Proceedings of the 13th Japanese/Korean Linguistics. CSLI
Son, M. J. and Cole, P. (2004): An Event-Based Account of -kan Constructions in
Indonesian. Ms. University of Delaware.
Travis, L. (2000): Event structure in syntax. In: Events as grammatical objects, ed. C.
Tenny and J. Pustejovsky, 145-185. Stanford, CT: CSLI.
Um, H. (1995): Argument Structure of Korean Causatives. In: Harvard Studies of Korean Linguistics 6, 413-428. Cambridge, MA: Harvard University Press.
von Stechow, A. (1996): The different readings of wieder "again": A structural account.
Journal of Semantics 13:87-138.
SECTION II
Event Structure and Modification
Kyle Rawlins (Santa Cruz)
Unifying illegally*
1 Introduction
A puzzling observation, first noted in Austin ( 1956), is that the syntactic position
of an adverb can affect the meaning of a sentence.
(1)
(2)
(3)
(4)
(a)
(b)
(a)
(b)
(a)
(b)
(a)
(b)
Clumsily he trod on the snail.

He trod on the snail clumsily.
Cleverly, John dropped his cup of coffee.
John dropped Iiis cup of coffee cleverly.
Louisa rudely departed.
Louisa departed rudely.
Appropriately, Kim kissed Sandy.
Kim kissed Sandy appropriately.
(Austin 1956)
(Jackendoff 1972)
(McConnell-Ginet 1982)
(Wyner 1994)
In (2a), John was clever to drop his coffee when he did. In (2b) however, the
way in which he dropped it was clever. Similarly, in (3a), it was rude of Louisa
to depart when she did. In (3b), her timing may have been fine, but some aspect
of her departure was rude.
There are two ways that this pattern has been accounted for. The first, proposed in different ways by Thomason and Stalnaker (1973) and Cresswell (1977),
explains the meaning differences in tenns of scope. Adverbs in the sentenceinitial position (which I will call high adverbs) compose with something other
than what sentence-final (low) adverbs compose with. I will call this the scope
approach.
The second kind of approach proposes that adverbs have multiple lexical entries, related by a lexical rule. According to McConnell-Ginet (1982) the two
uses of adverbs like rudely (3a)-(3b) involve different words of different syntactic category. Their meaning is connected by a lexical rule. I will call this the
lexical approach.
Recent work has tended to offer mixed approaches. Wyner (1994) suggests
that rudely is ambiguous between a "whole-event" and a manner reading, where
Thanks to Donka Farkas, Angelika Kratzer, Marcin Morzycki, and Barbara Partee for much interesting discussion of this topic, and to James Isaacs, Ruth Kramer, and Martin Schfer for
comments on drafts of this paper. Thanks also to audiences at UMass Amherst, UC Santa Cruz,
and the Event Structures workshop in Leipzig for questions and comments.
82
Kyle Rawlins
rudeness is predicated of maximal events and subevents respectively. Geuder

(2000) puts the brunt of the work into the lexicon. Emst (2002) presents a mixed
approach, appealing both to scope and lexical rules, that derives maimer adverbs
from sentence adverbs.
The lexical and mixed approaches, while quite effective, are counter-intuitive
- the different lexical entries behind a single surface fonn appear in complementary distribution. They are also too powerful - lexical rules of this kind can
describe many meaning changes in adverbs that are never seen. This paper argues for the scope approach, and suggests that the meaning differences between
the maimer and clausal uses can be analyzed without resorting to a lexical rule or
ambiguity. I present this argument through a case study of the adverb illegally.
My proposal extends to a class of adverbs that have been traditionally analyzed
as agent or subject oriented, including rudely, politely, legally, cleverly.1 I also
account, in somewhat less detail, for the pre-adjectival fonn of illegally. Due to
space limitations, I will not be able to discuss the adjectival illegal.
This analysis assmnes that what combines with the sister of an adverb is not
necessarily the lexical fonn of the adverb, but can be a type-shifted version of
this adverb. This idea originates with Morzycki (2002), which says that "Each
position [for a modifier]... is not so much the natural home for a particular kind
of modifier as for a particular kind of interpretation". On the view here, the kind
of interpretation of a use of an adverb is determined entirely by the compositional surroundings. The range of surroundings that I will consider is given in
(5)-(7).
(5)
(6)
(7)
Illegally, Alfonso moved a pawn.

Alfonso moved a pawn illegally.
"It's like having a radio with an illegally tall antenna," he says. (Google)
The remainder of 1 discusses background assumptions and terminology. In 24 I give independent semantics for the structurally high, low, and pre-adjectival
uses respectively. The "unifying" happens in 5, where I factor out what each
use has in common. This common part fonns the core lexical meaning of the
modifier, and the remainder is treated as a family of type-shift operators that
coerce sentence modifiers into modifiers of unsaturated types. These type-shifts
apply generally across a class of adverbs that are in their basic fonn, sentenceoperators. Finally, in 6 I demonstrate how to extend the analysis to other adverbs, and argue that agent-orientation, comparison class sensitivity, and gradability fall out from the modal force of the adverbs.
This class excludes "passive sensitive" or Thematically-Dependant adverbs, such as reluctantly

and appropriately. See Wyner ( 1994) inter alia.
Unifying illegally
83
1.1 Background
1.1.1 Why illegally?
Illegally is interesting for several reasons. It shows a meaning difference depending on syntactic position. However, factors that have been associated with
this difference such as agent-orientation (Jackendoff 1972, Wyner 1994, Emst
2002), and comparison-class sensitivity (Emst 2002) are absent. There is also
very little effect of gradability. Considering adverbs in the absence of these factors leads to a view of the factors as derived or secondary, a view I pursue in
6.
Another reason for focusing on illegally is that judgments about legality in

situations involving games with small sets of rules such as chess can be made
very clear; in fact much more clear than examples involving, say, rudeness or
cleverness.
1.1.2 Terminology
I will assume that, with respect to truth-conditional meaning, there are only two
kinds of positions/uses for adverbs like illegally2 These are the positions sometimes called "clausal" and "manner". I will refer to these positions by the neutral
structural tenns high and low respectively. The use of an adverb in these positions will be referred to as the high use and low use. The analysis starts by
treating each use separately, so I will subscript an adverb to indicate which use
is under discussion. Thus, illegally H will be illegally in its high position, and
illegallyL will be illegally in its low position.
The notation here will be as close to Heim and Kratzer (1998) as possible, for
expository clarity. This means that I will use then notational variant of a typed
lambda calculus, where variable sort is indicated as part of the lambda tenn, and
assiune familiarity with their implementation of Function Application (FA) and
Predicate Modification (PM).
I will use the following definition of Intensional Function Application (IFA).
Ds is the domain of possible worlds.
(8)
(9)
For any constituent , the intension of , notated [ a ] i s Xw' Ds . [a] 'c

Intensional Function Application If a and are daughters of 7, [a]'c is of any
type T, and [/3] 'c is a function with a domain of type (sT), then
br
[/3rc([or'c0)
See Cinque ( 1999) for arguments for more distinctions. For instance, manner positions are distinguished under e.g. the middle alternation. To the best of my knowledge, all further distinctions
concern grammaticality and not differences in truth-conditions.
84
Kyle Rawlins
2 The high position

Consider the following sentences, uttered by someone describing a chess game.
(10) Illegally, White moved.
(11) Illegally, White moved a pawn.
(12) Illegally, White moved a pawn diagonally.
(13) Illegally, White moved a pawn backwards.
Sentence (10) says that it was illegal for White to move at all. White probably
moved out of turn (or after the end of the game, or when the referee was not
looking). Similarly, (11) says that it was illegal for White to move the pawn in
question (at the relevant time). This would be true in a situation where the pawn
could not move forward (it was blocked; pawns in chess cannot capture moving
forward), and there were no possible captures (pawns only captine when moving
diagonally to the front, and otherwise cannot move diagonally). Sentence (12)
is true only when the pawn moved diagonally, and there was nothing to captare
there. Sentence ( 13 ) is true in any situation where a pawn was moved backwards;
pawns can never do this in chess.
In all of these cases, though there was a particular move made, it is not just
this move that is declared illegal, hi ( 10) it is any move by White, in ( 11 ) it is any
move of the pawn (at that time), in (12) it is any diagonal move of the pawn (at
that time), and in (13) it is any backward move of the pawn (at that time). These
sentences mie out a class of related possible moves as illegal. No possible move
at the time of the actual move, with the properties of the actual move mentioned
in the sentence, would have been legal. This is why we have the intuition, in
(10), that White moved out of turn; no White move at the time was legal, and
this situation aptly describes moving out of turn.
Emst (2002) proposes that adverbs in the high position can be analyzed as
modifying an eventuality. The adverb would modify an event picked out by the
verb, saying that the event is illegal. However, given the intuitions discussed
above, the adverb would need to modify not just the particular eventuality that
is asserted to have occurred, but an entire class of eventualities that might have
occurred. Ernst's proposal also uses a particular comparison class, saying that
the event was Adv in relation to other events that might have taken place. What
we want to say, however, is that all events (at that time) that meet the description
of the sentence were Adv (as opposed to other events that might have happened).
Because of this problem, I do not adopt Ernst's analysis, though I return to the
issue of comparison classes in 6.2. For similar reasons, I do not adopt Wyner
(1994)'s analysis, where the high position involves modification of a maximal
event.
Instead, I treat illegallyn as a sentence operator - a function from propositions
to propositions. In McComiell-Ginet (1982)'s tenns, this makes illegallyn an
ad-sentence. Illegally has a deontic flavor, declaring something to be disallowed
Unifying illegally
85
according to some rules or law. The most straightforward way to capture this in
the high use would be to treat illegallyn as a deontic modal operator, following
Jackendoff (1972). Illegally H differs from a typical modal in that it is factive sentences containing illegallyn entail the truth of the adverb's complement. On
top of the modal character, an additional provision must be made to account for
factivity.
Illegally is also context-dependent. When used in some context, it seems to
make reference to a salient code of laws in that context. In the above examples,
this has been the laws of chess, but it might equally well be the laws of the state,
of the country, etc. To model this, I will use the semantics of modals developed
in Kratzer (1977, 1981, 1991). Kratzer-style modals are interpreted relative to
conversational backgrounds (c.b.s). A conversational background maps sets of
propositions onto worlds, and serves in roughly the same capacity as accessibility functions (see e.g. Hughes and Cresswell 1996, Gamut 1990). C.b.s encapsulate the circumstantial, epistemic, and deontic background knowledge needed
for interpreting modals. I treat this as the same kind of background knowledge
involved in interpreting adverbs.
For example, recall that in (10), we want to say that no White move is legal.
If a move is not legal, it is not sanctioned by the law (of chess). If we think of
the law as a set of propositions (in this case saying things like "Chess occurs
in turns. White and Black alternate turns. Each can only move in their turn."
and so on), White moving at all cannot follow from these propositions. A set of
propositions is a set of sets of worlds, so that White cannot have moved in any
worlds in the union of these sets. That is, in no world that accords with the law
is it true that White moved. A conversational background is the object furnished
by the context which picks out, in each world, the set of propositions that make
up a body of law.
For the sake of exposition, I will use a singly relative modal operator for the
modal component of illegallyn This means that I will not formally represent
an ordering semantics or theory of inertial worlds in formulas. Kratzer's treatment is doubly-relative, interpreting modals w.r.t. to an ordering source and a
modal base. The ordering source orders accessible worlds relative to some set
of propositions. I do assume informally that an ordering semantics is operating
in the background, and make several assumptions about how it is behaving. 3
The definitions for modal operators and illegally H are given in (14)-(17).
(14) A conversational background (c.b.) b is legal iff it maps to each world a set of
propositions given by a single legal system in those worlds. 4
In particular, I assume that the lexical content of the adverb would pick out the content of the
ordering source, making it a legal ordering source for illegally. The modal base would be a
circumstantial conversational background. The singly relative treatment uses only the legal c.b.
All legal conversational backgrounds are deontic, but there are deontic backgrounds which are
not legal. Consider codes of behavior, politeness, or morality, for instance.
86
Kyle Rawlins
(15) Singly-relative necessity. A proposition is a necessity in a world w in view of a

conversational background b iff [Vu>' : w Ds A w' f~] b(w)] (w' p)
(16) Singly-relative possibility. A proposition is a possibility in a world w in view of
a conversational background b iff - is not a necessity in w in view of b.
(17) [illegallyg]' c is defined iff c picks out exactly one conversational background
that is legal. 5 In this case, [illegally H ] c =
-D(st) ^
'-*{p is a possibility in w in view of bc)
where bc is the legal conversational background picked out by c.
(17) straightforwardly casts the adverb as a negated (due to "il-") possibility

operator, with two modifications.
Unlike a normal modal, illegallyu is "factive". While it rules out the truth of a
proposition in deontically ideal worlds, it still asserts the truth of the proposition
in our world. I have cast this as straightforward entailment, but it might be
considered to be a presupposition.6 This also results in the entailment that the
proposition is illegal hi our world.
Secondly, the conversational background is lexically restricted hi a stronger
fashion than is typical for a modal. That is, modals impose only large-scale
restrictions hi the kind of c.b. they use, such as deontic vs. epistemic. To the
best of my knowledge, no true modals force the background to be something as
specific as sets of rules or laws.
Now for an example. I will show the computation of (10) to demonstrate how
the high use works. Later I'll give a denotation of this sentence with a more
complex event semantics, but for this example I'll assmne that move denotes a
predicate of individuals:
(18)
(19)
[move]' c = \x De . moved in w
[illegallyg, white moved] " ' 1
iff [illegallya] c ( [white moved]
= 1
iff [illegallyJJ]
(Aw/ Ds . white moved i W ) = 1
White moved in w
\
-.((w' Ds . White moved in w')

is a possibility in w in view of bc)
The sentence asserts two things: (i) White moved in the evaluation world, and
(ii) In all legally ideal worlds (w.r.t. the body of law picked out, which is the
laws of chess) it's false that White moved.
5
I will not be specific about how a context picks out a c.b.; there are several ways to go about it.
One simple way would be to assign the adverb an index and treat backgrounds as references in
the assignment function.
Wyner (1997) has proposed that adverbs like "wisely" in high position are predicates of facts.
Under this kind of analysis, the use of such an adverb might presuppose the existence of a fact
verifying the main sentence, effectively presupposing that the main sentence is true. Here I will
continue to treat the apparent factivity as entailment, for reasons that will be apparent later.
Unifying illegally
87
3 Low position
The predominant trend in analyzing low/manner adverbs in recent work has
been to take them to be predicates of events, following Davidson (1967)'s treatment of other adverbials. I follow this trend, and in particular I assume the
neo-Davidsonian analysis (Parsons 1990, Wyner 1994, Eckardt 1998, Landman
2001, among others)7.
Consider the sequence of claims, again about a chess game, in (20)-(23).
(20)
(21 )
(22)
(23)
White moved illegally.

White moved a pawn illegally.
White moved a pawn diagonally illegally.
White moved a pawn backwards illegally.
A sentence like (20) is true if some aspect of White's move constituted a violation of
the law. We are not told which aspect, and there may have been other aspects which did
not violate the law. I take this to be the most important intuition about the low position
sentences.
Unlike the parallel high-position sentence (10), (20) says nothing about moves that
White did not make. The sentence declares the particular move White made to have been
illegal, and nothing more. All the other sentences are the same, except that they provide
more information about the move. In fact, they could all be true in the same context, a
property which the high-position sentences in ( 10)-( 13) do not have.
I will make two basic assumptions about eventualities. First, I will take them to be
particulars, in their own domain Dv. Second I will assume Kratzer (1996)'s "severing"
of the agent argument from the verb. On this view, the verb does not lexically have an
external argument, but this argument is added in by a Voice head in the Infi range.
(24)
[Voice]' c = XP e D{vt)
. De . \e Dv . P ( e ) Agent (e, )
In most neo-Davidsonian systems, the event argument is saturated by an existential quantifier at some point in the derivation. It is not clear where this quantifier
is introduced compositionally. Kratzer (1998) makes it part of the meaning of
Aspect morphemes, but other authors (e.g. Landman 2001, Chung and Ladusaw
2004) assmne a type-shift or compositional operation called Existential Closure
(EC), which gets applied when we have an unsaturated predicate and need a saturated one. I will assmne that EC is a type-shift applying as a last resort, and that
composition of a sentence is not complete unless the top node is a truth value.
(25) Existential Closure
If a constituent denotes a function of type (vX) for any type X, then
E C ( [ a < r - c ) = 3e Dv . [ "
The analysis here should be broadly compatible with non-Davidsonian theories of adverbial modification. See Wyner ( 1989), Landman (2001 ) 3.4.2. It is not compatible with McConnell-Ginet
( 1982 )'s analysis of manner modification.
88
Kyle Rawlins
Verb denotations take the following fonn:

(26)
[ m o v e ] ' c = \e
E Dv
. e is a m o v i n g i n w
A verb binds the event to the evaluation index, as well as picking out its type.
I am informally assuming a theory of counterpart relations that applies to the
domain of eventualities as well as individuals (Lewis 1968,1986, among others).
An event that has no counterpart in w would make a verbal predicate false at w
- this is the main effect of the "in w".
Given these assumptions, we are now in a position to define illegally^ As
with the high use, I will use deontic modality, but the immediate application of
it is slightly more complicated than for the high case. If an event is illegal, it has
some aspect that is not allowed in deontically ideal worlds. We can model this
by ruling out occurrences of the actual event (that are sufficiently similar to it)
in the ideal worlds.
I will assmne that the ordering semantics or theory of inertia worlds will
cause us to consider only worlds that are close enough so that if an event has an
occurrence (i.e. a counterpart), that occurrence shares all properties of the actual
event. No event will have a counterpart in an ideal world that is not identical to
it. Thus, a predicate like [move]"'' 0 will be true in w of an event that occurs in
w' only if that event has an identical counterpart in w. 8
By abstracting over the evaluation world and applying the event argument,
we can fonn a proposition of the right kind.
(27)
[ i l l e g a l l y L ] 'c is d e f i n e d iff c p i c k s out e x a c t l y o n e c o n v e r s a t i o n a l b a c k g r o u n d that

is legal. I n this case, [ i l l e g a l l y L ] w'c = \P
(
l
^ t
Ae Dv
P{e)
-i((Aii)'
\
Ds
. P(w')
( e ) ) is a p o s s i b i l i t y i n w i n v i e w of bc)
w h e r e bc is t h e legal c o n v e r s a t i o n a l b a c k g r o u n d p i c k e d out b y c.
This denotation has two mam components. Using the letter V to stand hi for
the constituent illegallyL modifies, a V-ing illegallyl is also a V-ing in the
evaluation world. This is the low position correlate of factivity. Additionally, in
legally ideal worlds, it is not possible for such an event to take place. By ruling
out identical counterparts of the event, we entail that some property of the event
8
Pete Alrenga (p.c.) has suggested to me that this is effectively the same assumption we make
about sentences like "John might fail the exam" when we hold the facts about John, the exam,
and what constitutes failing constant in ideal worlds (even if a listener is not necessarily aware of
all of them). In Kratzer's ordering semantics, this would be modeled by using a circumstantial
modal base that fixes the relevant facts, and a deontic ordering source that picks out the closest
ideal worlds satisfying the circumstances. Note also, as pointed out to me by Donka Farkas (p.c.)
that this is not a problem about counterpart theory, as the same fixing of the facts is necessary in
other theories of transworld identity (e.g. Kripke ( 1980)). One alternative would be to quantify
explicitly over properties of events, effectively moving the work into the semantics proper. This
would be similar to Wyner ( 1994 )'s treatment, which quantifies over subparts of the event.
89
Unifying illegally
has caused it to be disallowed from occurring in worlds that are legally ideal.
Let us consider the example computation of (20), assuming the structure in
(20)' and that White is a proper name,
(20') [ip White [Voice [ y p moved illegally] ] ]
In the following formulas, bc always refers to the unique legal background picked
out by the context. I will start from the bottom and work upwards. IllegallyL
and moved need to combine via Intensional Function Application, as the denotation of illegallyL takes the intension of a predicate of events for its first
argument. Combining the two (and assuming the context will provide a suitable
conversational background) gives us:
[illegallyL] ' c ( [moved] w ' c ) =
Ae Dv .
e is a moving in w
-i((Au>' Ds . e is a moving in w')

Next, the Voice head introduces an argument place in the composition for the
external argument.
[Voice] ' c ( [illegallyL] 'c ( [moved] ) ) =
Xx De . Xe Dv .
/
Agent (e, )
e is a moving in w
-i((Aii)' Da . e is a moving in w')
Next, the external argument composes straightforwardly by Function Application, and Existential Closure saturates the event argument to create a truth-value.
Substituting the definition of possibility, we get the final result:
(28)
[White Voice moved i l l e g a l l y L ] p c = 1 iff [3e : e e Dv]

/
Agent (e, White)
\
e is a moving in w
[Vu/ : w' Da Aw' )1(

-i(e is a moving in w'))
The previous example demonstrating high use did not make use of an event
semantics. For the purposes of comparison, it is useful to compute the denotation
of that example (10) with an event semantics. The result is:
90
Kyle Rawlins
(19') [illegallyH, white moved]" '
r_
A
\
, /
Be : e DJ
L
V a
[Vw' : w' Ds Aw'
1 iff
e is a moving in w
7, . ,
A, ent e , whlte
('
)
J
f]bc(w)}(
-i([3e : e Dv] (
\
Agent(e, White)
e is a moving in w J /
Comparing (19') and (28) reveals that the difference between the two sentences
derives from the relative scope of the universal quantifier over worlds (i.e. the
possibility operator due to the adverb) and the existential quantifier introducing
the event variable. In the low use, the existential quantifier over events scopes
above the universal quantifier over worlds. The high use reverses this, and the
universal takes scope over the existential.
4 Pre-adjectival uses
The picture becomes more complex when we consider adverbial modification
of adjectives. Examples of this are most natural when the adjective is directly
deverbal, but are not limited to this case. Here are some naturally occurring examples involving a variety of adverbs, and some constructed examples involving
illegally.
(29) ...with zoomy homage to the age of the camcorder and a clumsily realistic spontaneity among its performers. (Google)
(30) It's called The Score, an appropriately generic title for a droning, high-toned little
heist picture with no dash and no raison d'tre. (Google)
(31) ...the consequences of operating an illegally uninsured business could bring significant criminal and civil consequences. (Google)
(32) When they find an illegally colored house, they'll kick down the door and drag the
homeowner off to jail. (Google)
(33) Alfonso noticed an illegally red house.
(34) Alfonso noticed an illegally built house.
Here I will not try to account for the conditions under which illegally can modify
an adjective, but will provide an analysis that gives the correct truth-conditions
for sentences like (31)-(34).
Sentences involving illegally in this use seem to have two readings. In (33),
for example, it could be that zoning laws in the town forbid red houses entirely.
In this case, the sentence says that it was illegal for the house to be red at all.
This meaning, like the high use, gives rise to the entailment that no other shade
of red would have been acceptable; the law bans any red houses. I will call this
the whole-predicate use, and write this version of illegally as illegallyW P . This
use is directly analogous to the high use.
Unifying illegally
91
The second reading can be paraphrased by saying that the way/manner in

which the house is red is not allowed. The law might allow some shades of red,
but ban others. It is only the particular shade of red that matches the house that
is called illegal. I'll call this second use the sub-predicate use, and write this
version of illegally as illegally sp. This use is directly analogous to the low use.
With a deverbal case like (34), this second reading comes out more clearly,
although the first is still present. It may be that it was illegal to build the house
in the first place (e.g. if the land is zoned for factories), or that some part of
the house violates building codes (e.g. the foundation does not meet earthquake
code, and the house is in California).
4.1 The whole-predicate use
Intuitively, we want illegallywp to declare that any instantiations of the adjectival predicate combined with its nomi are declared illegal. If a house is
illegally WP red, we want no legal ideals to include houses that are red at the
evaluation index. This can be accomplished in a similar way to the low and
high cases, where we ruled out eventualities satisfying the verbal predicate, and
propositions satisfying the sentence, respectively. Here we mie out individuals satisfying the adjectival predicate. The following denotation works only for
attributive adjectives that normally combine with a nommai predicate via Predicate Modification. 9
where bc is the legal conversational background picked out by c.
This denotation is quite similar to the denotation for the low use, except that the
predicates involved are predicates of individuals, not events.1" The correlate of
factivity in this domain is to assert that the property holds of the entity argument
at the evaluation world.
Here is the denotation of "illegally WP red house" assuming standard denotations for the other items.
10
We could also create a higher-typed version of illegallywp if some adjectives that allow adverbs
need to have higher types such as ((s(et))(et)).
This denotation is also a candidate for the meaning of illegal, since the types involved are the
same. However, with the adjective, we run into issues of stage vs. individual level illegality that
there is not space to discuss here. Note also that this coincides with Geuder ( 2000 )'s suggestion
that some adjectives are not more basic than their adverbial counterpart.
92
Kyle Rawlins
[illegally red house] w * c = (IFA, PM)
/
Xx De
is red in w
is a house in w
-i((Aii)' Ds . is red in w')
\
is a possibility in w in view of bc) /
where bc is the legal conversational backgroimd picked out by c.
In English, it is not possible for the house to have counterparts in ideal worlds
and still be red. The house in the evaluation world, however, is red.
4.2 The Sub-predicate use
I will analyze this reading by assuming that illegallysp modifies a predicate of
states, just as illegallyL modifies an event-predicate. This requires assuming
that at least some adjectives have a state argument that is saturated by existential
closure. This has been independently proposed by Parsons (1990) 10.4. We
also must assume that there is such a thing as a particular state. An illegally red
house on this reading is a house where there is a state of redness, the house is in
that state, and the particular state of redness that the house is in does not occur
m legally ideal worlds.
If this analysis is right, the denotation of illegally SP would be identical to the
denotation for the low use, assuming that Dv contains states as well as eventualities of other kinds.
5 Unification
In the previous sections, I have accounted independently for the high (clausal)
and low (maimer) uses of illegally, as well as two pre-adjectival meanings. This
section unifies these uses, extracting a single core lexical meaning, and a family of type-shifts which apply to allow the adverb to compose in a variety of
positions.
For the sake of comparison, here are the denotations of each use laid out next
to each other. Each denotation, as before, is defined only when the context picks
out exactly one legal backgroimd, and bc in each denotation refers to this backgroimd.
(17)
[illegallyHr-c= D<st) .
p(w)
- ( is a possibility in w in view of bc)
(27) [illegally^ 'c = XP D(s{vt})
. Xe Dv .
P{e)
-i((Aii)' Ds . P(w')(e)) is a possibility in w in view of bc
(35) [illegally W P r- c = XP D{a{et)) . Xx De .
(
l
PH(x)
-i((Xw' Ds . P(w')(x))
is a possibility in w in view of bc
Unifying illegally
(36)
[illegallysp] -
= [illegally^
93
Anything that is common to these denotations should be part of the core meaning of the adverb. The following things are shared: (i) the "factivity", (ii) the
modal force, and (iii) the restriction on conversational backgrounds. The differences boil down to the mechanisms used to convert the modified type into a
proposition. Since illegally H needs no mechanisms, we can take it to be basic,
and factor out the differences as follows:
[illegally]' c = \p D{st) . p(w) -( is a possibility in w in view of bc) where
bc is the legal conversational background picked out by c.
(38) [illegallyg] ' c = [illegally] w ' c
(39) [ L r c = [ s p r c =
(37)
\A D((st)t) -R D(s(vt Ae <= Dv . A{\w' Ds . R(w')(e))

(40) [ W P r c = A-4 D{{st)t) . XP D{s{et)) ,\xDe.
A{Xw' Da . P(w')(x))
Each of these operations can be thought of as a type-shift from a modifier of a
saturated type to a modifier of an unsaturated type. The core denotation of this
kind of adverb is taken to be a property of propositions. Type-mismatches occur
in the stmcturally low position, because in order to combine with the verb in its
basic fonn, illegally would have to first trigger existential dosine. The rest of
the sentence would not be able to compose after this. The type-shift given by
L resolves this type mismatch. Before an adjective, a type-mismatch occurs as
well, but there are two sequences of type-shifts which can be used to resolve
it. The first sequence is existential closure (over a state argument) followed by
wp, resulting in the whole-predicate reading. The second is L (i.e. SP) followed
by existential closure, and this results in the sub-predicate reading. As with
the better-known nominal type-shifts (Partee 1986), these apply as a last resort,
when composition would otherwise fail due to a type-mismatch.
The core lexical content contains elements shared across the class of adverbs
under discussion. These elements might be factored out as well, into a (lexical)
property of this class. Left behind as the core lexical content of a particular
adverb would be only a restriction on the background information present in the
context at the time of utterance.
Theoretically, these type-shifts have two mam functions. They allow lexical
meanings of adverbs to be imderspecified for type, and this is why I proposed
them. They are also independent of particular adverbs, and therefore allow a
straightforward scope-based explanation of the low-high alternation. Ultimately,
a theory-independent justification of the existence of these type-shifts (as opposed to any other possible ones) is desirable, and I will leave this to future
work.
Geuder (2000) suggests (2.2.5f, 3.6.3) that some sortal mismatches might
be dealt with by employing Nimberg (1995)'s predicate transfer in the lexicon.
While Geuder's goals are slightly different (explanation of the relationship be-
94
Kyle Rawlins
tween adjectival and adverbial fonns when the adjectival is taken to be basic),
the idea is similar to that of type-shifting. Predicate transfer acts as a very restricted fonn of type-coercion, subject to lexical idiosyncrasies. Geuder argues
that a fully productive (e.g. non-lexical) approach (such as mine) would be too
powerful, and that a lexical approach would allow us to specify that certain readings were blocked. However, in the class of adverbs under consideration here, I
do not know of any cases of the kind of idiosyncrasy that Geuder is concerned
about. The relationship between high, low, and pre-adjectival uses in this class
seems entirely productive, and subject at best to pragmatic restrictions. I do not
address the relationship of illegally-class adverbs to their adjectival fonn here,
and this may be where lexical idiosyncrasies start to appear.
I have described one part of a family of typeshifts. They take two fonns,
turning a sentence operator into an event-modifier, and into a predicate-modifier.
That is, we typeshift from ((st)t) into ((s(et)}(et)}, and into ((s(vt))(vt)). There
are other potential members of this family that have not yet been needed. For
instance, we could attempt the reverse shifts, from unsaturated modifiers to saturated modifiers. The reverse type-shifts of the ones I've given are difficult to
formulate without some use in mind. It does not seem to be possible to provide
any type-shift that would allow an acfiial reversal of the shifts I've given. The
technical problem is how to dispose of the unsaturated argument place.
Interestingly, the most obvious candidates for such a type-shift are "pure manner" adverbs such as loudly (Schfer 2001, 2002) which seem to be more basically event predicates. At least superficially, these adverbs do not show scope
effects (they perfonn maimer modification in a high position), and the assumption that they are in their basic fonn event modifiers, combined with the lack of
a type-shift, might explain part of their scopeless behavior.
Assuming the existence of these type-shifts, we must consider how they fit in
with existing theories of the syntax of adverbs.
5.1 An alternative conception of the type-shifts
The analysis so far, relying on type-shifts, is roughly in line with the analysis
of the syntax of adverbs in Emst (2002): adverbs are adjoined where they can
compose. Emst assmnes various type-shifts which are not spelled out, and the
details differ greatly from here, but the type-shifts above can be thought of as
making explicit some parts of Ernst's semantics for adverbs.
In Cinque (1999), the main competing approach to the syntax of adverbs,
adverbs appear in the specifiers of functional projections. A functional head is
associated with a class of adverbs, and only that class can appear in the specifier
of that head. The type-shifting components presented above could be thought of
as fixed meanings of particular functional heads. An idea of this sort has been
suggested by Morzycki (2002). On this version of my analysis, the functional
heads would more or less prepare their complement for composition with an adverb. While the order of composition is different, the meanings above can easily
Unifying illegally
95
be refactored to compensate. For instance, the "maimer" morpheme would look

as follows:
(41) t r
XR
e D(s(vt}) A-4 G D{{at)t)
. \v Dv . A(\w' Ds . R(w')(e))
On this view, the meaning pieces introduced above would not be type-shifts, but
real (though in English, invisible) morphemes in the syntax.
This demonstrates that the broad analysis here is compatible with Cinque's
syntax. However, adopting the functional-head based version leaves some difficult questions.
First, I have used the same type-shift in more than one place. Both the subpredicate and low use involve the same operation, and as suggested briefly in 4,
the adjectival type-shift looks quite like the whole-predicate type-shift. This is
to be expected if there are a limited number of type-shifting operations that are
used as a last resort; if one can apply somewhere, it will. On a functional-head
approach, however, there is no obvious reason why the same meaning would
be used in fairly disparate points in the syntactic structure. On a related note,
we would have to postulate an ambiguity for the morpheme introducing preadjectival adjectives. This ambiguity falls out naturally from the family of typeshifts I have proposed.
Second, when there is no adverb in a specifier of some functional head, we
would not want the meaning pieces like (41) to appear in the functional head,
because this would produce a type-mismatch in the absence of an adverb. We
could solve this by putting syntactically null adverbs that defuse the functional
head's type-changing component in some vacuous way, but this solution does
not seem ideal. In many cases, Cinque makes use of functional heads that have
been used independently for semantic purposes. In the case of functional heads
that do something else (for instance if Voice on Kratzer (1996)'s analysis also
served to introduce some class of adjectives), we would need functional heads
to be systematically ambiguous. This also does not seem ideal.
Third, there is the question of why the functional heads mean the things they
do. In hindsight at least, the meanings involved are fairly simple ways of shifting modifier types from saturated to unsaturated. They affect the meanings of
then arguments in very minimal ways. We might expect the meanings of such
functional heads to be more arbitrary than this.
These questions at their heart ask whether the functional-head based approach
to adverbial syntax is, from the semantics side, too powerful. The questions do
seem to lead to the conclusion that both approaches, at least with respect to the
semantics, are attempting to model the same thing.
96
Kyle Rawlins
6 Other adverbs
In this section I will broaden the class of adverbs under consideration to include
rudely, politely, and legally. What I say here applies to any adverb which can
plausibly be taken to have its core lexical meaning based on a set of facts that are
part of the background knowledge of a speaker. I take this to include cleverly,
stupidly, wisely, foolishly, tactfully, craftily, ostentatiously, graciously, eagerly,
absent-mindedly, and others which have been traditionally categorized as agent
or subject oriented. Additional extensions may be possible to speaker-oriented
adverbs {frankly; ideally) and domain adverbs (mathematically, semantically).
First I'll consider legally. Care must be taken to ignore the extra readings that
this adverb has. We are only interested in one where there is permission given
by some body of law, as in (42). The additional readings that legally has seem
to be domain readings (Bellert 1977, Ernst 2004), illustrated in (43), and these
are not under consideration here.
(42) Alfonso left the country legally.
(43) (a) Alfonso is legally blind.
(b) Legally, Alfonso is skilled.
(c) Legally(/in view of the law), you must cross the street at a crosswalk.
The only difference between permissive legally and illegally on my analysis is

that there is no negation; something is legal if it's possible in view of the law.
The type-shifts still apply as described above.
(44) [legally]' c = \p -D<et) . p{w) (p is a possibility in w in view of bc) where bc
is the legal conversational background picked out by c.
Consider now rudely.

(45) Rudely, Alfonso departed.
(46) Alfonso departed rudely.
Rudely also has a modal character in that it involves laws of etiquette. If conversational backgrounds can be restricted to codes of permissible behavior, rudely
can be given the obvious analysis in (47):
(47) [ r u d e l y ] ' ^ e D(st) p{w) -(p is a possibility in w in view of bc) where
bc is the politeness-based conversational background picked out by c.
The main difference from illegally is the restriction on what kind of conversational backgrounds are involved. A conversational background about politeness
consists of a series of statements about what is considered polite, in some system of politeness. The ideal picked out by such a background consists of worlds
where only polite things happen. We would expect all the type-shifts introduced
previously to apply here.
Unifying illegally
97
However, there are several ways in which sentences with rudely differ from
those with illegally, and I will talk about these in the following sections. My
analysis makes some predictions about entailments which seem acceptable for
illegally, but are not for other adverbs. I discuss this in 6.1. Ernst (2002) has
analyzed the rudely-class as being sensitive to a comparison class, and I describe
how to derive this in 6.2. In 6.3, I discuss the fact that rudeness has been
described as a scalar or gradable notion, with a context-dependence threshold
set somewhere on a scale.
I argue that none of these are problems with the proposed family of typeshifts. Each can be addressed by either making further assumptions about operators that scopally intervene between adverb positions, or making further assumptions about the structure of information picked out by various kinds of conversational backgrounds.
An important issue that I do not have space to discuss is that rudely and related
adverbs have often been described as "subject/agent-oriented" (Jackendoff 1972,
among others). Rudely (in some sense) predicates rudeness of the agent of the
sentence, cleverly predicates cleverness, and so on.
6.1 Incorrect entailments
The analysis so far predicts the entailment patterns in (48) and (49) to be acceptable. Some speakers judge the entailment (48) to be valid, but (49) is clearly
incorrect.
(48) ? Illegally, White moved. f= White moved illegally.
(49) * Rudely, Alfonso departed. |= Alfonso departed rudely.
For high rudely, there is not necessarily anything rude about how Alfonso departed; it is really just the timing of a departure, or the fact of a departure at all.
We could copy the actual departing event in its entirety to another time, and it
might be polite. In the low case, the timing does not come into play - it is something essential to the particular departing that was rude. This departing would
be rude in many contexts or at many times.
These patterns are predicted because ruling out counterparts of a particular
event of moving/departing is a special case of ruling out all events of the agent
moving/departing. The quantifier scope for the high use is Vx-i3j/0, which entails 3y~<Vxip, the scope for the low use, if entails V'11
Until now I have been treating tense informally. The representation of tense
is the missing factor, and once we take it into account, the entailment patterns
in (48) and (49) are not predicted. We also make the truth-conditions more accurate. Specifically, tense intervenes scopally between the two adverb positions,
and is inside the scope of the modal force only in the high position.
11
This was pointed out to me by Marcin Morzycki (p.c.).
98
Kyle Rawlins
Here I will consider only cases of past tense, which I will take to be referential, following Partee (1973), Kratzer (1998), Stone (1997), Schlenker (1999,
2004). A past tense morpheme acts like a pronomi in that it picks out an interval,
and it says that its event argument culminates (indicated with the predicate Cui,
from Parsons (1990)) at that interval.
(50) [ - e d ] 1 ^ ^ XP D{et) . \e Dv . P ( e ) Ci(e, t*, w) where t* is the interval
picked out by c such that t* < t.
Composition of examples (45) and (46) now results in the following formulas:
(51) [-ed Alfonso Voice depart rudelyL] 1 c = 1 iff [3c : e e l>, \
/
Agent (e, White)
\
Cul(e,f*,u)
e is a departing in w
A [Vu' : w' Ds Aw' ) ] (
y
-i(e is a departing in w')) )
where bc is the polite conversational background picked out by c, and t* is the time
picked out by c s.t. t* < t.
(52) [rudely H , -ed Alfonso Voice depart] 1 c = 1 iff
/
/
e is a departing in w \
\
[3e : e Dv] ( Cul(e,f*,u>)
I
\ Agent (e, Alfonso)
/
[Vu/ : w' G Ds Aw' G ) ] (
Agent(e, Alfonso)
\
Cul(e,f*,u')
e is a departing in w' / /
where bc is the polite conversational background picked out by c, and t* is the time
picked out by c s.t. t* < t.
The crucial difference between the two is that in the high position, the past tense
operator is under the scope of the adverb, whereas in the low position, it is not.
The high use says that no event of Alfonso departing at that time occurs in ideal
worlds, and the low use says that the particular event of departing that did happen
in the evaluation world has no counterparts in ideal worlds. On the low use, the
time of the event is fixed hi the evaluation world, but its counterparts in ideal
worlds do not have fixed times. On the high use, the time is fixed across ideal
worlds.
We are left with the question of why (48) seems intuitively valid for some
speakers. I do not have a good answer, except to point out that both (53) and
(54) seem substantially worse than (48).
(53) *? Illegally, White moved a pawn. |= White moved a pawn illegally.
(54) * Illegally, Alfonso left the country. |= Alfonso left the country illegally.
Unifying illegally
99
Entailment patterns of this kind do not seem as straightforward as one might

expect, and more work is needed.
6.2 Comparison class sensitivity
Emst (2002) describes rudely as being sensitive to comparison classes, in a way
that differs between the different uses. In particular, in the high use, the actual
occurrence is rude in comparison to other events that could have happened at the
same time. In the low use, the actual occurrence is rude in comparison to other
events of Ving, substituting in the verb for V.
My analysis suggests that the apparent comparison class sensitivity is a secondary effect, derivable from the interaction of modality and tense. To find what
the comparison class consists of, we see what can potentially take the place of
the actual occurrence in ideal worlds.
The high use in (45) effectively calls any event of departing at the referenced
time rude. We can infer that in the ideal worlds, some other events with Alfonso
as the agent took place in lieu of the departing at roughly the same time. The
actual event is rude compared to these.
The low use calls the particular event that occurred to be rude. Any identical
copies are ruled out, but in its place are events of departing. Since they are not
counterparts of the actual departing, they may be non-identical in more ways.
Thus the event seems rude in comparison to other departings that Alfonso could
have been the agent of.
6.3 Gradability
Rudely is gradable; something can be done more or less rudely. Even without degree modifiers, rudely is scalar and sensitive to some contextual standards. That
is, in one context, on a degree analysis, we might have the threshold for rudeness set different than in another context. The nature of my analysis forces this
to be derived indirectly; we cannot simply place events on a scale of rudeness.
I suggest that with respect to both degree modifiers, and apparent sensitivity to
the position of a standard on a scale, there is only the appearance of a true linear
scale. Facts about how codes of politeness relate to each other make the rudeness
scale look more linear than the legality scale.
Codes of politeness are generally related. When we move from fonnal contexts to less fonnal contexts, more things become allowed, and when we move
the other direction, less things become allowed. If something is not allowed in
a less fonnal setting, it very likely is not allowed in any more fonnal setting.
Thus, between different codes of politeness, there is the appearance of a rough
subset-superset relation, and this can lead to a rough mapping on a linear scale of
politeness. The most fonnal settings are at the top of the scale, and the least are
lower down. Below those are codes of politeness you would get by subtracting
100
Kyle Rawlins
restrictions from the least formal settings, though these may not be actually in
force anywhere.
Laws tend to behave in a very different way. The laws of chess are completely non-overlapping with the laws of the US. Even federal laws in the US do
not share much with the town laws. Though there may often be some relation
between codes of law as we move from e.g. town to town, it does not look like a
subset-superset relation at all. Thus there is no appearance of a linear scale, and
illegally seems to be non-gradable.
An action is more polite than another if it is possible under more restricted
(or even simply more) codes of politeness than the other. Similarly, things which
are very illegal are things which would be illegal under a great many codes of
law. If someone were to call murder very illegal (which it is), this would be what
they mean.
7 Conclusions
In this paper I have defended two main claims: (i) The meaning differences
induced by placing an adverb in different positions result purely from scope,
and (ii) This position is consistent with an adverb having only one lexical entry.
The positional meaning differences follow from the relative scope of the adverb with respect to the existential quantifier over events and with respect to
tense. While I have confined the investigation here to sentences with simple past
tense, an interesting future topic would be whether aspect plays a role, since it
would also intervene scopally.
In order to give adverbs only one lexical entry, I have proposed a family of
type-shifts between modifier types. As with type-shifts in the nominal domain,
these apply as a last resort when composition would fail. They shift between
modifiers of safiuated types (sentence operators) to modifiers of unsaturated
types. I have proposed that the basic lexical entries for illegally-type adverbs
are factive modal operators, with relatively free composition mediated by typeshifts.
Much work still remains. I have only touched on the range of adverbs that
show positional effects - candidates for the modal type-shifting analysis include
domain adverbs (e.g. semantically) and speaker-oriented adverbs (frankly). I
have argued that comparison class sensitivity and gradability can be derived from
the modal force of adverbs like rudely, but I leave an in-depth treatment of these
effects for the future.
References
Austin, J. L. (1956): A plea for excuses. In: Proceedings of the Aristotelian Society.
Reprinted in Austin (1961).
Austin, J. L. (1961): Philosophical papers. Oxford University Press.
Unifying illegally
101
Bellert, I. (1977): On semantic and distributional properties of sentential adverbs. Linguistic Inquiry. 8:337-351.
Chung, S., and W. Ladusaw (2004): Restriction and saturation. MIT Press.
Cinque, G. (1999): Adverbs and functional heads: A cross-linguistic perspective. Oxford
University Press.
Cresswell, M. (1977): Adverbs of space and time. In Formal semantics and pragmatics
for natural languages., ed. F. Guenthnerand S. J. Schmidt, 171-199. Reidel. Reprinted
in Cresswell (1985).
Cresswell, M. (1985): Adverbial modification. D. Reidel Publishing Company.
Davidson, D. (1967): The logical forni of action sentences. In The logic of decision and
action. University of Pittsburgh Press.
Eckardt, R. (1998): Events, adverbs, and other things. Max Niemeyer Verlag.
Emst, T. (2002): The syntax of adjuncts. Cambridge University Press.
Emst, T. (2004): Domain adverbs and the syntax of adjuncts. In Adverbials: the interplay between meaning, context, and syntactic structure., ed. Jennifer Austin, Stefan
Engelberg, and Gisa Rauli. John Benjamins.
Gamut, L.T.F. (1990): Logic, language, and meaning volume 2: Intensional logic and
logical grammar. University of Chicago Press.
Geuder, W. (2000): Oriented adverbs: Issues in the lexical semantics of event adverbs.
PhD dissertation, University of Tbingen.
Heim, I., and . Kratzer. (1998): Semantics in generative grammar. Blackwell.
Hughes, G.E., and M.J. Cresswell. (1996): Anew introduction to modal logic. Routledge.
Jackendoff, R. S. (1972): Semantic interpretation in generative grammar. The MIT Press.
Kratzer, A. (1977): What must and can must and can mean. Linguistics and Philosophy.
1:337-355.
Kratzer, A. (1981): The notional category of modality. In Words, worlds, and contexts:
New approaches in world semantics., ed. Hans-Jiirgen Eikmeyer and Hannes Rieser,
38-74. Berlin: Walter de Gruyter.
Kratzer, . (1991): Modality. In Semantics: an international handbook of contemporary
research., ed. A. von Stechow and D. Wunderlich, 639-650. Berlin and New York:
Mouton de Gruyter.
Kratzer, . (1996): Severing the external argument from its verb. In Phrase structure and
the lexicon., ed. J. Rooryck and L. Zaring. Kluwer Academic Publishers.
Kratzer, A. (1998): More structural analogies between pronouns and tenses. In Proceedings of Semantics and Linguistic Theory Vili., ed. Devon Strolovitch and Aaron Lawson, 92-110. CLC publications.
Kripke, S. (1980): Naming and necessity. Harvard University Press.
Landman, F. (2001): Events and plurality. Kluwer Academic Publishers.
Lewis, D. (1968): Counterpart theory and quantified modal logic. Journal of Philosophy.
65:113-126.
Lewis, D. (1986): On the plurality of worlds. Blackwell.
McConnell-Ginet, S. (1982): Adverbs and logical form: A linguistically realistic theory.
Language. 58:144-184.
Morzycki, M. (2002): Mediated modification. PhD Defense Handout.
Nimberg, G. (1995): Transfers of meaning. Journal of Semantics. 12:109-132.
Parsons, T. (1990): Events in the semantics of english. Cambridge, MA: MIT Press.
Partee, . (1973): Some structural analogies between tenses and pronouas in english.
102
Kyle Rawlins
Journal of Philosophy. 70:601-609.

Partee, . (1986): Nomi pirrase interpretation and type-shifting principles. In Studies in
discourse representation theory and the theory of generalized quantifiers., ed. J. Groenendijk, D. de Jongh, and M. Stokhof, 115-143. Foris Publications.
Schfer, M. (2001 ): Pure manner adverbs. In Ereignisstrukturen., ed. J. Dlling and T. Zybatow, 251-272. Linguistische Arbeitsberichte 76, Universitt Leipzig.
Schfer, M. (2002): Pure manner adverbs revisited. In Sinn und Bedeutimg VI, Osnabrck 2001., ed. G. Katz, S. Reinhard, and P. Reuter, 311-323. Publications of the
Institute of Cognitive Science, University of Osnabrck.
Schlenker, P. (1999): Propositional attitudes and indexicality (a cross-categorial approach). Phd dissertation, MIT.
Schlenker, P. (2004): Sequence phenomena and double access readings generalized. In
The syntax of time., ed. Lecarme and Gueron. MIT Press.
Stone, M. (1997): The anaphoric parallel between modality and tense. IRCS Report 9706.
Thomason, R., and R. Stalnaker (1973): A semantic theory of adverbs. Linguistic Inquiry.
4:195-220.
Wyner, A. (1989): Adverbs and argument structure. In Proceedings of ESCOL 89., ed.
. de Jong and Y. No. Ohio State University.
Wyner, A. (1994): Boolean event lattices and thematic roles in the syntax and semantics
of adverbial modification. PhD dissertation, Cornell University.
Wyner, A. (1997): On factive adverbials. In Sinn und Bedeutimg 97.
Marcin Morzycki (Michigan)
Adverbial modification of adjectives:

Evaluatives and a little beyond*
1 Introduction
One of the principal analytical challenges of adverbial modification is how to account for the intricate and often subtle correlation between an adverb's syntactic
position and its interpretation. Why, to consider one familiar class of examples,
should subject-oriented readings be associated with an intermediate position in
the clause? Wiy should maimer readings be associated with relatively lower positions, and speaker-oriented readings be associated with higher ones? Attempts
to grapple with these kinds of issues, from Jackendoff (1972) and McConnellGinet (1982) to Cinque (1999) and Emst (2002), have focused on adverbial modification in the verbal and sentential domain. But adverbial modification can be
found elsewhere as well - in English and many other languages, adverbs can also
occur in the extended AP. Importantly, the interpretation adverbs receive in these
less understood 'ad-adjectival' positions varies predictably from the one they receive elsewhere. And, strikingly, the position of adverbs M 'ithin the extended
AP varies in a similar way. For example, subject-oriented(-like) interpretations,
such as the one defiantly receives, are possible only right of degree words:
(1 )
(a)
(b)
He seemed enormously more defiantly sedentary than Greta,

*He seemed defiantly more enormously sedentary than Greta.
Similarly, evaluative readings are not possible to the right of an adjective, though
domain-adverb readings1 are:
(2)
(a)
(b)
(c)
(d)
George seems intellectually inadequate.

George seems inadequate intellectually.
George seems shockingly inadequate.
*George seems inadequate shockingly.
Thanks to Angelika Kratzer, Amia Maria Di Sciullo, Anne-Michelle Tessier, Barbara Partee, Chris
Kennedy, Klaus Abels, Kyle Johnson, Kyle Rawlins, Lisa Matthewson, Meredith Landman, Stefan Engelberg, Susan Rothstein, Tom Ernst. This research was supported by grants to Anna Maria
di Sciullo from the Social Sciences and Humanities Research Council of Canada.
The term here is intended in the sense of Ernst ( 2002 ); see also Rawlins ( 2003 ), and under different
names, Bartsch ( 1976), Moltmann ( 1997).
104
Marcin Morzycki
Because of these properties, then, adverbial modification in the extended adjectival projection may offer a fresh perspective on the larger problem.
This paper examines one large natural class of such AP-modifying adverbs,
which have an evaluative interpretation and include remarkably, surprisingly,
and breathtakingly, among many others, and considers how the account proposed might extend to other varieties of ad-adjectival adverbs. The core proposal will be that these adverbs widen the domain of salient degrees, and are
interpreted as arguments of unrealized degree morphology in much the same
way as measure pirrases have been proposed to be. This approach turns out to
extend naturally to uses of these adverbs in other positions.
Section 2 identifies the class of adverbs of filterest here and explores its distinguishing characteristics. Section 3 develops an analysis of the semantics of
sentences containing remarkably adverbs based in part on a notion of domain
widening in tfie degree domain, assimilating tfiem to certain exclamatives. Section 4 confronts problems of compositionality tfiese adverbs pose, and arrives at
a kind of decomposition in wfiicfi part of tfie interpretation of a remarkably adverb is contributed by its lexical semantics and part is contributed directly by its
place in tfie architecture of the extended adjectival projection. Section 5 sketches
how these syntactic and semantic assumptions can be the foundation of a more
general theory of how the meaning of these adverbs is related to the meaning
tfiey fiave in otfier structural positions. Section 6 applies tfie approach that has
been developed more broadly, examining ad-adjectival uses of subject-oriented
adverbs. Section 7 concludes.
2 Remarkably adverbs
2.1 Tfie cast of characters
Very roughly, the adverbs of interest here, henceforth -emarkably adverbs', give
rise to a judgment about having a property to a particular degree - that it is, say,
remarkable or surprising or fiorrible:2
(3)
(a)
(b)
(c)
(d)
Clyde is remarkably tall.

Floyd is surprisingly ugly.
Tranquility is heartbreakingly difficult to attain.
Self-referential example sentences are disappointingly
distracting.
Tfiis class of adverbs is quite large - indeed, it is an open class. Among its many
otfier members are amazingly, astoundingly, amusingly, calmingly, disappointingly, earth-shatteringly, extraordinarily, frighteningly, grotesquely, heartbreakingly, impressively, inconceivably, infuriatingly, interestingly, mind-numbingly,
nauseatingly, provocatively, revoltingly, shockingly, terrifyingly, imnen'ingly,
2
If the adverb receives parenthetical intonation, it can receive the same reading it receives in clausemodifying positions.
Abverbial modification of adjectives
105
(un)pleasantly, (un)remarkably, and wonderfully. New adverbs of this class can

be coined quite easily (I might describe shoes as cringe-indncingly uncomfortable, for example). Importantly, in all these cases, there is a predictable semantic
relation between the adverb and the corresponding adjective.
2.2 Contrast with clause-modifying uses
These adverbs can occur high in a clause-modifying position as well, where they
receive a different reading entirely:
(4)
(a)
(b)
(c)
(d)
Remarkably, Clyde is tall.

Surprisingly, Floyd is ugly.
Heartbreakingly, tranquility is difficult to attain.
Disappointingly, self-referential example sentences are distracting.
Here, no judgment is being rendered specifically about having a property to any

particular degree. Rather, very crudely, the judgment in these sentences is about
the proposition expressed by the sentence as a whole. These readings are truthconditionally distinct - if, for example, Clyde is a professional basketball player
and therefore expected to be very tall, Clyde is remarkably tall could be true
while (4a) could be false. Indeed it is not possible to construe any of the sentences m (4) as having the interpretations of their counterparts in (3). Just as the
meaning of a remarkably adverb is predictable from its adjective counterpart,
the meaning of sentences like those in (3) is predictable from their counterparts
in (4). These facts, then, seem to reveal robust, apparently exceptionless grammatical regularities.
2.3 Not degree words
One natural analytical impulse is to suppose that remarkably adverbs are in fact
a species of degree words (that is, of Degree head; I will use these interchangeably), like too, very, pretty, or comparative morphology. But that does not seem
to be the right approach, for several reasons.
Perhaps the clearest of these is that, unlike degree words, remarkably adverbs
support degree words of their own: 3
(5)
(6)
(a)
Clyde is [more remarkably] tall,
(b)
*Clyde is [more quite] tall.
(a)
Floyd is [quite surprisingly] ugly,
(b)
*Floyd is [quite too] ugly.
One might object at this point that there is a conceivable alternative parse of the
3
Many of the sentences starred here are possible as metalinguistic comparatives (like e.g. ? Floyd
is less surprisingly ugly than he is a minor annoyance ).
106
Marcin Morzycki
sentences in (5)-(6) in which the degree word is associated with the adjective
rather than the adverb, as indicated in (7):
(7)
Clyde is [more [remarkably tall]].
(parse to be rejected)
If this were the structure of (7), a puzzle would arise immediately - more tall is
not the comparative fonn of tall, taller is. Yet what we find in (7) on this structure is comparative morphology applying to an AP headed by tall, so we would
expect taller to occur here. Assuming that the way comparative morphology and
adjectives combine morphologically is by head movement of the adjective to a
higher position where it finds the comparative morpheme, we would expect that
the adjective would move over remarkably, as in (8):
(8)
*Clyde is [tall-er [remarkably t\\
This, as indicated, results in an imgrammatical sentence. Nor is there evidence

for a structure like (7) from interpretation. It is probably true that if Clyde is said
to be very remarkably tall, he must also be very tall. But this is not evidence for
construing vei-y as applying to remarkably tall, because of the way being tall
is related to being remarkably tall. The only way Clyde's height can be more
remarkable (in the way relevant to remarkably adverbs) is to be greater; the only
way for it to be less remarkable is for it to be smaller. Consequently, increasing
or decreasing the extent to which Clyde's height is remarkable also increases or
decreases his height correspondingly. The effect of a degree word, then, will be
in this respect the same irrespective of which structure is adopted.
There are broader considerations that militate against treating remarkably adverbs as degree words. Remarkably adverbs constitute an open class, and can
be comed essentially on-the-fly; there is no comparably productive, readilyaccommodated means of coining new degree words (though to be sine, doing
so is not impossible). No doubt related to this is the relative scarcity of degree
words - it does not seem at all out of the question that one might be able to compile an exhaustive list. Compiling an exhaustive list of remarkably adverbs, on
the other hand, would be an enormous undertaking at best, and nearly as futile
as attempting to compile an exhaustive list of nouns might be. Another signature
characteristic of remarkably adverbs is their systematic relationship to their adjective counterparts, and to their corresponding uses in clausal positions. Degree
words manifest neither of these characteristics.
107
3 Developing an interpretation
3.1 Some paraphrases
Given the systematic relationship between remarkably adverbs and adjectives, it
seems appropriate to construct the denotations of remarkably adverbs in tenns
of their adjective counterparts, taking paraphrases like those in (9)(10) as a
starting point:
(9)

(a)
(b)
(c)
It is remarkable that Clyde is as tall as he is.

It is remarkable to be as tall as Clyde is.
It is remarkable how tall Clyde is.
(10) Floyd is surprisingly ugly.

(a)
(b)
(c)
It is surprising that Floyd is as ugly as he is.

It is surprising to be as ugly as Floyd is.
It is surprising how ugly Floyd is.
Not all of these paraphrases are equally good. The (a) and (b) paraphrases all
suffer from a problem of ambiguity. For (9a), there is a reading in which what
is remarkable is the fact that Clyde is as tall as Clyde. Similarly, in (10a), what
is surprising could be the fact that Floyd is as ugly as Floyd. 4 The remarkably
adverb sentences do not have this reading. But this problem could be avoided one could imagine pursuing paraphrases of the fonn Floyd is tall to some degree,
and it's remarkable that he's that tall, or, in linguist quasi-English, Floyd is dtall and it's remarkable to be d-tall. There is, however, a deeper problem.
An inkling of this problem is reflected in (9a) and (9b). If what is remarkable
about Clyde's height is that he is very short, both of these paraphrases would
be true; but of course, the remarkably adverb sentence cannot mean this. This
is still only an inkling of the problem, in that it too could be solved relatively
straight-forwardly, in this case by adding to the denotation a requirement that, in
this instance, Clyde be tall.
The full measure of the problem emerges more clearly in a situation in which
we know Clyde to be the victim of a creepy nmnerological accident. We know
that he was bom at precisely 5:09 in the morning, on the fifth day of the ninth
month of 1959. We further know that he currently lives at 59 Fifty-ninth Street.
Discussing this strange happenstance, I might inform you that Clyde's height is
precisely five feet and nine inches. So Clyde is not very tall, but he is not very
short either. It would be quite natural for you to say, upon having heard this
news, that it is remarkable that Clyde is five feet nine inches tall, or to utter (9a).
But it would not be natural at all to say that Clyde is remarkably tall - indeed,
4
This is essentially the same ambiguity as in Russell ( 1905 )'s Your yacht is larger than I thought it
108
Marcin Morzycki
given typical contemporary expectations about adult male height, it would be

false.
In this situation, the problem cannot simply be that Clyde is not tall. If we
increment all the numbers that seem to haunt Clyde to the point that he might
qualify as just barely tall but not very tall, the result stays the same - it is still
remarkable that he is as tall as he is, in light of the numeric coincidences in his
life, but he is not remarkably tall.
What this demonstrates is that to qualify Clyde as remarkably tall, it is not
sufficient that he be tall and that there be something remarkable about his height.
It must also be the case that what is remarkable about Ms height is how great it
is. Similar facts hold for other remarkably adverbs- for (10), for example, what
is surprising must be how great Floyd's ugliness is, not simply that he is ugly.
So there is something fundamentally inadequate about the (a) and (b) paraphrases above, and more generally about paraphrases that involve predicating
an adjective of a proposition in this way. But all this also strongly suggests
that the (c) paraphrases above, which involve embedding wA-clauses, are on the
right track. They face none of these difficulties. They do not give rise to the
undesirable ambiguity discussed above - they have only the interpretation that
remarkably adverbs have. Nor do they fail to reflect that remarkably adverbs always seem to require that the degree in question be high, and that it must be the
highness of the degree that leads to the judgment expressed by the remarkably
adverb.
In light of the close parallel between these paraphrases and remarkably adverbs, then, taking them as a guide seems to be an approach with some empirical
support - these really are very close paraphrases, close enough to suggest that
what they reflect is genuine.
3.2 Embedded exclamatives
There is, however, a complication in taking the semantics of these wA-paraplirases
as a guide: it is less than clear what the semantics of these paraphrases themselves is. The M'/i-clause in these paraphrases is not, as it might initially seem,
an indirect question. Rather, it is an embedded exclamative of the sort discussed
in Grimshaw (1979) - a less-studied construction.
Perhaps the clearest evidence for this involves veiy. As Grimshaw observed,
very is impossible with w/i-words hi questions, as hi ( 11 ), but possible hi exclamatives, as in (12):
(11) (a)
(b)
(12) (a)
(b)
*How very tall is Clyde?

*How very ugly is Floyd?
How very tall Clyde is!
How very ugly Floyd is!
109
This contrast holds under embedding as well. Embedded clauses that are relatively clearly indirect questions do not admit vety:
(13) (a)
(b)
*I wonder how very tall Clyde is.

*Someone asked how very ugly Floyd is.
But embedded exclamatives do:

(14) (a)
(b)
It is remarkable how very tall Clyde is.

It is surprising how very ugly Floyd is.
Another diagnostic is based on the observation, due to Elliott (1974) and noted
by Zanuttini & Portner (2003), that exclamatives do not seem to occur comfortably under negation in declaratives:
(15) (a)
(b)
(16) (a)
(b)
I don't (particularly) wonder how tall Clyde is.

No one asked how ugly Floyd is.
*?It isn't remarkable how very tall Clyde is.
*?It isn't surprising how ugly Floyd is.
Zanuttini & Portner observe that curiously, in questions the situation is reversed exclamatives can occur with negation, as in (17), but not without it, as in (18):
(17) (a)
(b)
(18) (a)
(b)
Isn't it remarkable how tall Clyde is?

Isn't it surprising how ugly Floyd is?
*?Is it surprising how ugly Floyd is?
*?Is it remarkable how very tall Clyde is?
So in this respect too, these paraphrases pattern with embedded exclamatives.

Building on the foundation these paraphrases provide, then, we are now led
to a semantics for remarkably adverbs framed in tenns of their corresponding
adjectives and embedded exclamatives.
3.3 The interpretation of exclamatives
The semantics of exclamatives, though, is murky (at least from a formal-semantic
perspective; informal discussions include McCawley 1973, Elliott 1974, and
Michealis and Lambrecht 1996). Still less clear is the semantics of exclamatives
under embedding. Zanuttini and Portner (2003), who develop an approach to
these issues, will serve here as a guide through this thicket of uncertainty.
Their first move is to observe that exclamatives do not have truth values, and
hence should not be analyzed as proposition-denoting. Zanuttini & Portner suggest that instead, exclamatives have denotations of the same type as questions
do - sets of propositions (following, for questions, Hamblin (1973), Karttimen
110
Marcin Morzycki
(1977), Groenendijk and Stokhof (1984), and others). This reflects quite clearly
the deep syntactic parallel between questions and exclamatives. It also sets aside
the difference between the two in illocutionary force, which can be reflected in
other ways (as they convincingly argue). Adopting the Karttimen (1977) view
that a question denotes the set of its true answers, they treat exclamatives as
likewise denoting a set that includes only true propositions. So, they suggest, an
exclamative such as (19a) will denote a set of propositions that might, under the
appropriate circumstances involving discussion of chili pepper consumption, be
as in (19b):
(19) (a)
(b)
What surprising tilings he eats!

{'he eats poblanos', 'he eats serranos', 'he eats jalapeos'}
More generally, then, (19a) will denote the set of true propositions of the fonn
'he eats x' for some (surprising) value of x:
(20)
[ What surprising things he eats! ] = {: is true and there is a surprising tiling

such that is the proposition that he eats x }
Zanuttini & Portner identify two principal ingredients in the semantics of exclamatives. One of them is /activity - exclamatives systematically presuppose the truth
of a corresponding declarative. While remarkably adverbs have a similar property, as these examples show, this will not be a central focus at the moment. The
other ingredient, which will figure prominently in the analysis of remarkably adverbs proposed here, is widening of the domain of quantification of the displaced
M'/i-expression. To illustrate how this works, consider a context in which we are
discussing what Herman eats. If I say Herman eats eveiything, the domain of
quantification of the universal is constrained by a contextual domain restriction,
so one would not conclude from my utterance that Herman eats light bulbs or his
relatives. It is very probable that what we might expect Herman to eat would be
even more constrained than this - assuming the appropriate cultural background,
we might also fail to conclude that Herman eats serrano chilies. Zanuttini &
Portner propose that exclamatives affect essentially this sort of domain restriction, widening it to include things we otherwise would not have considered.
So if what I had uttered instead was the exclamative What surprising things he
eats!, its effect would be to cause my interlocutors to entertain some possibility they previously had not - say, that Herman eats serranos. The denotation of
the exclamative, then, will because of this widening include more propositional
alternatives than it otherwise would have. As Zanuttini & Portner observe, this
bears a close family resemblance to Kadmon and Landman (1993)'s analysis of
what any does.
This idea elegantly gathers together several otherwise slippery and elusive
intuitions about what exclamatives mean. Among these are the intuition that
111
exclamatives somehow involve an 'extreme' value for something, and that exclamatives convey that something is unexpected in a particular way.
3.4 Interpreting exclamatives embedded
The next question relevant to understanding exclamative paraphrases of remarkably adverb sentences is what happens when an exclamative is embedded. This
presents one slight additional complication, but it eliminates another one. The
additional complication is that some assumptions have to be made about the
semantics of the embedding predicate - hardly a minor point here, since this
embedding predicate is what corresponds to the remarkably adverb. Here too,
Zanuttini & Portner lead the way. They suggest that amazing, which embeds
both exclamatives and finite indicatives, can be understood as having two fonns,
one for each type of complement. The garden-variety fonn applies to propositions and hence embeds finite indicatives. Its semantics is relatively straightforward - it predicates of a proposition that it is amazing: 5
(21) [ dlfl dZl fig garden-variety 1 = Ap<s>,) . amazing(p)
The other fonn of amazing applies to sets of propositions and hence embeds
exclamatives. It is interpreted as requiring that some proposition in this set be
amazing:
(22) I 77//(IZingexclamative-embeddmg ] = AE<<S ,) ,) . 3p[E(p) amaZllg(p)]
For an exclamative denotation to be amazing, then, it must include a proposition
which is amazing. So, supposing that Clyde is 6 feet 4 inches tall, one might
utter (23 a), and the exclamative will have a denotation like the one indicated
schematically in (23b):
(23) (a)
(b)
It is amazing how tall Clyde is.

3p[p{'Clyde is 6 feet 1 inch tall', . . ' C l y d e is 6 feet 2 inches tall', ...,
'Clyde is 6 feet 3 inches tall', . .., 'Clyde is 6 feet 4 inches tall'} aniazing(p)]
In light of (22), (23a) can be interpreted as requiring that one of the propositions
in the set in (23b) be amazing. If it is the case that it is amazing to be 6 foot
4, then, this will be true. More generally, we might assmne that embedded
exclamatives (at least ones embedded under the relevant sort of predicate) are
interpreted in a way that parallels (23).
While hi some respect complicating things slightly, this simplifies the situation in another respect. In light of the denotation arrived at for these sorts of
This is not precisely their formalism, but the content is (intended to be ) the same.
112
Marcin Morzycki
structures, for current purposes, it will be possible to do away with making reference in these denotations to sets of propositions, replacing them with sets of
degrees. This is so because asserting (23) amounts to claiming that it is amazing
that there is a degree (in a particular set of degrees with the relevant properties)
to which Clyde is tall:
(24) amazing( A 3d[d{6 feet 1 inch, . . . . 6 feet 2 inches, . . . . 6 feet 3 inches, . .., 6 feet
4 inches} Clyde is d-tall])
All embedded-exclamative paraphrases of remarkably adverbs involve adjectives, so hi all of them it will be possible to make this simplifying move, quantifying over degrees rather than over propositions. To capture the meaning of
embedded exclamatives, and by extension of sentences containing remarkably
adverbs, it will also be necessary to say something about what the set of degrees
being quantified over is - specifically, it will be necessary to capture the effect of
domain widening.
3.5 Brief interlude: some assumptions about adjectives
Before proceeding further, though, it may be helpful to briefly lay out some
background assumptions about the interpretation of adjectives. First, a degree
is an interval on a scale abstractly representing measurement (Kennedy 1997,
Schwarzschild and Wilkinson 2002). A scale is a dense, linearly ordered set of
points. Second, a gradable adjective denotes a relation between an individual
and a degree - a relatively standard assumption (Semen 1973, Cresswell 1976,
von Stechow 1984, Bierwisch 1989, Klein 1991, Rulhnan 1995, Kennedy and
McNally 2004). In a sentence like (25), then, tall relates Clyde to some degree
of height, here one measuring six feet:
(25) (a)
(b)
{tailJ = AxAd . tall(x)(d)

[ Clyde is six feet tall ] = 3d[tall(Clyde)(d) the measure in feet of d is 6]
If no overt measure phrase is present, the adjective will be interpreted with respect to a contextually-supplied standard degree of talhiess. In (26), for example, tall relates Clyde and the standard for tallness siB//provided by the context of
utterance:
(26) [ Clyde is tall ] = 3d[tall(Clyde)(d) d > s,fl]
What (26) requires is that Clyde be tall to some degree and that this degree meet
or exceed the standard stau
113
3.6 The interpretation of remarkably adverb sentences

Returning to the mam thread of the discussion, it is now possible to propose an
interpretation for exclamative paraphrases of remarkably adverbs in the spirit of
Zanuttini & Portner, and thereby one for the corresponding remarkably adverb
sentences as well. Given what has already been said, a sentence such as the
now-familiar (27a), along with its exclamative paraphrase (27b), might (in a
particular circumstance) receive an interpretation such as (27c):
(27) (a)
(b)
(c)

It is remarkable how tall Clyde is.
remarkable( A 3d[ d { 6 feet 1 inch, .... 6 feet 2 inches, .... 6 feet 3 inches, ...,
6 feet 4 inches} Clyde is d-tall ])
So, as before supposing that Clyde is 6 foot 4, (27a) might assert that it is remarkable that Clyde is tall to a degree in the set indicated schematically in (27c). To
spell things out a bit more precisely, a means of representing domain restrictions
will be needed. One way of doing this, though not the path taken by Zanuttini & Portner, is to make use of resource domain variables (von Fintel 1994,
Westersthl 1985). Just as a resource domain variable can be used to reflect
contextual domain restrictions on determiner and adverbial quantification, it can
also be used to reflect contextual domain restrictions on quantification inside the
extended AP. The denotation of Clyde is tall in (26) can be thus elaborated with
the addition of a resource domain variable C, which will restrict an existential
quantifier over degrees as in (28):
(28)
[ Clyde is talle ] = 3d[df=C tall(Clyde)(d) d>s, fl ]
The resource domain variable C has as its value a contextually-salient set of degrees; (28) requires that the degree quantified over be in this set. It is a fairly
significant step, and one that will be crucial here, to suppose that quantification
over degrees is contextually restricted in the way that quantification over individuals or events (or situations) is. But since domain restrictions seem to be a
general property of natural language quantification, this is a natural assumption.
With this in place, the widening effect of remarkably adverbs can now be
represented fairly straightforwardly. As a first step, without yet reflecting the
effect of widening in the denotation, we can take (29a) to have the denotation in
(29b):
(29) (a)
(b)

[ Clyde is remarkably talle ]
= remarkable( A 3d[df=C tall(Clyde)(d) d>s, fl ])
(not final)
This merely predicates remarkable-ness of the proposition expressed by Clyde

is tall, yielding a meaning that might be paraphrased 'it is remarkable that Clyde
114
Marcin Morzycki
is tall' (which is an inadequate paraphrase for reasons discussed in section 3.1).

To introduce the effect of domain widening, we might modify (29b) by existentially quantifying over a domain larger than the contextually-supplied domain
provided by the resource domain variable C:
(30)
[ Clyde is remarkably
talle ]
(not final)
= remarkable(A3d3C'[C/DC deC'A tall(Clyde)(d) d>s,fl])

This amounts to loosening the requirement that a degree of Clyde's talhiess be
among the contextually salient degrees, permitting it instead to be either among
these degrees or in some larger domain C that includes these degrees.
Still, this is not yet quite adequate, because remarkably adverbs, like exclamatives, contribute domain widening in a particular sense that (30) does not reflect.
Unlike the variety of widening that Kadmon and Landman (1993) argue any
involves, exclamatives and remarkably adverbs impose the further requirement
that the degree quantified over not be in the imwidened portion of the domain.
For Clyde to be remarkably tall, it is not sufficient that he be tall to a degree
that is among the contextually salient ones. Rather, Clyde has to be tall to some
degree that is not among the degrees already contextually salient - he must be
tall to a degree that has been added to the domain by widening, as (31 ) reflects:
(31)
[ Clyde is remarkably
talle ]
= remarkable(A3d3C'[C/DC deC'-C tall(Clyde)(d) d>s,fl])

This requires that there be a degree to which Clyde is tall which exceeds the
standard and that it is in the portion of the widened domain C that excludes the
original domain C.
This denotation seems to be an adequate representation of the meaning of
Clyde is remarkably tall. It reflects that this sentence involves a claim that something is remarkable, and that what is remarkable is not merely that Clyde is tall or
even that there is some particular degree such that it is remarkable that he is tall
to that degree. Rather, what is claimed to be remarkable is that Clyde's height
is so great that it exceeds all the heights one would otherwise have entertained.
In this way, this denotation reflects the same sort of domain-widening that an
embedded exclamative would contribute, thereby explaining the semantic correlation with the embedded exclamative paraphrase. The factivity entailment that
is also characteristic of both remarkably adverbs and exclamatives is predicted
here, too, because this denotation requires that there be a degree to which Clyde
is tall that exceeds the standard for talhiess. Maintaining this requirement of
exceeding the standard is cmcial to capturing the factivity entailment - the requirement of widening the domain on its own would not suffice, since it would
not rule out the possibility that Clyde is tall to a degree smaller than any in the
115
domain, and that what is remarkable about his height is how small it is.6 Other
remarkably adverb sentences can be given interpretations analogous to this one.
4 Assembling the pieces

The previous section arrived at a model of the interpretation of remarkably adverb sentences, but nothing has so far been said about how this interpretation
is assembled compositionality. It will emerge in this section that the familiar
means of semantically combining an adverb and an expression it modifies are
not adequate for the task that needs to be perfonned here - and that a further
examination of the syntax suggests another path to take.
4.1 The trouble with the usual options
The most basic means of interpreting a modifier is intersectively, by a rule like
Heim and Kratzer (1998)'s Predicate Modification. There is no straightforward
way of doing this for remarkably adverbs. The principal difficulty is that for two
expressions to be interpreted intersectively they must be of the same semantic
type. In order to implement an intersective interpretation for remarkably adverbs
and the adjectival projections they modify, it will thus be necessary to find a
single type for the denotations of both the remarkably adverb and its sister. But
what could this type be?
One possibility that seems initially appealing is that both the remarkably adverb and its sister denote properties of degrees. This, though, is problematic, and
at a minimum requires complicating the ontology of degrees significantly. To begin with, it would be necessary to find a way to construe the remarkably adverb
itself as a property of degrees. Given denotations like the one arrived at above, it
is at best highly unclear how this might be done. Of course, one might conclude
from this that there is something severely wrong with these denotations. It could
in principle be that remarkably adverbs are interpreted simply by predicating
them directly of degrees. This has the appeal of simplicity, but, among other difficulties, such an approach would have to be spelled out far more before it could
be made sense of. Certainly, if a degree is simply an interval on a scale as assumed here (following Kennedy 1997 and Schwarzschild and Wilkinson 2002),
predicating of this interval that it is remarkable or surprising or disappointing
or strange would at a minimum fail to make obvious predictions, and at worst
might be as irredeemably incoherent as a claim like '12 is remarkable'.
Another, perhaps less serious but non-trivial difficulty is what one might do
with the type that would result when a remarkably adverb and its sister are interpreted - if this type is itself a property of degrees, as would result from an
intersective interpretation, an account would have to be provided of how this
6
This presupposes that the standard will always be in the domain of quantification - an assumption
natural at least, and perhaps unavoidable.
116
Marcin Morzycki
can ultimately be predicated of individuals. Certainly, there are ways in which

this can be done, both by altering syntactic assumptions or semantic ones. One
interesting semantic approach toward this problem may be available if degrees
are formalized, as Faller (2000) proposes, as vectors in a Vector Space Semantics
(Zwarts 1997, Zwarts and Winter 2000, Winter 2001). In this sort of framework,
there are independently necessary type shifts that map properties of vectors ( qua
degrees) to properties of individuals. Any approach in which a remarkably adverb is predicated directly of a degree also faces the problem of explaining what
the relationship is between predicating a remarkably adverb of a degree and
predicating its adverbial or adjectival cognates of individuals and propositions
(and perhaps eventualities). One can certainly claim that remarkably simply
denotes a property of remarkable degrees, and remarkable a property of remarkable individuals - but this merely conceals the problem behind the metalanguage
predicate 'remarkable'.
When an intersective denotation for a modifier is not possible, one usually
simply adopts a higher, predicate-modifier type denotation - construing it as a
function that applies directly to the modified expression. But for remarkably adverbs, this road too has dangerous pitfalls. If remarkably adverbs were predicate
modifiers, they would presumably denote functions from AP denotations to AP
denotations - given the assumptions here, expressions of type {{e, dt), (e, dt)). This
would certainly help with the problems noted in the previous section, since the
remarkably adverb could now 'have access' to the adjectival denotation in a way
that would make it possible to build up a denotation like the one arrived at in
section 3. But this is inconsistent with the syntactic behavior of these expressions. As we have already seen (in ( 5 )-( 6 ) ), remarkably adverbs project further
structure:
(32) (a)
(b)
(c)
Clyde is [[quite remarkably] tall].

Floyd is [[rather surprisingly] ugly].
Many voters are [[pretty horribly] conservative].
In light of this, it is not the remarkably adverb itself but rather the extended AdvP
in which it occurs which must have the higher-type denotation. But to achieve
this, barring some complicated, previously unattested type shift, it would be
necessary to assume that other elements of the adverbial extended projection
- including comparative morphology, very, and all other Degs - are systematically ambiguous between their regular denotations and ones that yield this very
high AP-modifying type. This would be an exceptionally implausible and costly
assumption at best.
4.2 Building up more syntax: analogy to measure phrases
If, as the previous section argued, remarkably adverbs cannot be interpreted intersectively or as predicate modifiers, how should they be interpreted? A closer
examination of the syntax suggests an answer.
117
One especially clear aspect of the syntax of these expressions is that they
resemble nominal measure phrases - they occur in the same linear position, and
they are in complementary distribution with overt degree words modifying the
APs in which they occur:
(33) (a)
(b)
Floyd is {six feet/remarkably} tall {*six feet/*remarkably}.

Floyd is {*six feet/*remarkably} very tall.
It seems reasonable, then, to pursue a parallel syntactic analysis. I will assume

that APs with absolute adjectives and measure phrases have a structure like the
one in (34), in which the measure phrase occupies the specifier position of a
Deg(ree) head (Abney 1987, Corver 1990, Grimshaw 1991, Kennedy 1997):
(34)
DegP
DP
sixfeet
Deg'
Deg
AP
[ABS]
tall
Under other circumstances, the Deg head can be spelled out overtly as a comparative morpheme (or other degree morpheme) or as a degree word. With absolute adjectives, it cannot be overtly spelled out; in these cases, Kennedy (1997)
suggests the Deg head is instead occupied by a null degree morpheme [ABS]. In
light of the similarities, it is natural to assign remarkably adverbs a similar structure, in which their phrasal projections likewise occupy the specifier position of
DegP: 7
(35)
DegP
Proposals of roughly this form for degree adverbs in general - by which is typically meant any true adverbs in AP - have been made before. Abney (1987)
suggests a structure similar to (35), with adverbs in a specifier position, and the
structures Jackendoff (1977) has in mind would have ones like (35) among their
contemporary analogues.
7
I use DegAdvP here to distinguish the degree projection of the adjective and that of the adverb.
118
Marcin Morzycki
Kennedy's [ABS] has in (35) been replaced with a similar feature [R]. Although a stronger reason to distinguish these will emerge shortly, there are at
least two other, purely syntactic reasons this distinction may be useful. One of
these is that [ABS] licenses a DP in its specifier, so it is Case-licensing. Remarkably adverbs, on the other hand, have no need to check Case. Another
consideration here is a slight difference in distribution - measure pirrases, unlike
remarkably adverbs, are possible in comparatives:
(36) Clyde is {two feet/*remarkably/*siirprisingly} taller than Floyd.
It will be necessary, then, to distinguish the ability to license measure pirrases

and remarkably adverbs in order to reflect that certain Degs may license one
but not the other. In light of this independent necessity, there would not be
any advantage to uniting the ability to license measure pirrases and remarkably
adverbs in one Deg, [ABS].
This sort of structure has a number of syntactic advantages. It can account

for the complementary distribution of measure pirrases and remarkably adverbs,
since these both occupy the same structural position. It can account for why
remarkably adverbs are obligatorily left of the adjective. And it can account for
why they are in complementary distribution with overt Degs, since they require
a particular (null) Deg to license them. 8
4.3 Putting the syntax and semantics together
With these syntactic structures in place, it is now possible to look on the semantic
compositionality puzzle with a fresh eye. For measure-phrase structures like
(34), Kennedy suggests that the semantics is assembled as in (37):
(37)
[ Clyde is six feet [ABS] tall ] =

[ [ABS] ] ( [ tall] )([six
feet ] )([ Clyde
])
The Deg [ABS] yields a property of individuals as the denotation of the DegP.
It does the semantic work of relating the AP and the measure pirrase. Given the
parallels, it is natural to suppose that semantic composition works similarly in
(35). The [R] feature can be taken to be interpretable, and paralleling [ABS], to
be what relates the AP and the remarkably adverb semantically:
(38)
[Clyde
is remarkably
[R] tall J =
[ [R] ] ( [ tall ] )( [ remarkably
] )( [ Clyde ] )
This structure also predicts that it should not be possible to stack remarkably adverbs, but that it
should be possible to introduce them recursively. That is, while exactly one remarkably adverb
phrase can occur for each AP, a remarkably adverb phrase can itself contain a remarkably adverb
(e.g. [[surprisingly [terrifyingly]] ugly]).
119
This means of putting the pieces together, via the mediation of [R], will be the
key to solving the compositionality problem and arriving at the desired interpretation.
It is now possible to suppose that the denotation of remarkably is identical
to that of the adjective remarkable. To illustrate this, though, it will be useful
to make two simplifying assumptions purely for exposition. First, I will omit
the degree argument in the denotation of both remarkably adverb and their corresponding adjectives. Second, I will for the moment suppose that these corresponding adjectives denote properties of propositions rather than, say, ordinary
individuals. Both remarkably and remarkable can thus be taken to have the denotation in (39):
(39) [remarkable J = [remarkably ] = . remarkable(p)
This is a very simple denotation, and reflects only the barest, most minimal
lexical core of the meaning of these expressions. It is a long way from the
making the semantic contribution that was attributed to remarkably adverbs in
section 3. But the challenge of getting from one to the other can now be met
straightforwardly - the additional semantic work that needs to be done can be
attributed not to the adverb itself, but rather to the [ R] feature in Deg that licenses
it. Just as the adverb itself can now have as its denotation only the irreducible
essence of its lexical semantics, so too the [R] can now have as its denotation
only those aspects of meaning that characterize the class of remarkably adverbs
generally, independent of the particular choice of adverb:
(40) [ [R] ] = \A{e,{dj))
XR{SIJ) . R(A3d3C'[C'DC deC'-C A(x)(d) d>s A ])
This denotation reflects exactly the semantic properties identified in section 3 as

characteristic of remarkably adverbs- among the more prominent ones, domain
widening. It also serves as a kind of semantic glue, helping hold together typetheoretically the adjective and adverb denotations. These pieces fit together in a
way that yields the desired result:
(41)
[Clyde is remarkably [R] tall J

= ['[R]
j([tallj)([remarkablyj)([Clydej)
= remarkable(A3d3C'[C/DC deC'-C tall(Clyde)(d) d>s,fl])

This is exactly the denotation ultimately arrived at in section 3 in (31 ).
4.4 Problems averted
This division of labor avoids the problems raised by the alternative approaches
to introducing remarkably adverbs into semantic composition. The problems
associated with an intersective interpretation do not arise here because this ap-
120
Marcin Morzycki
proach does not impose the requirement that the adverb and its sister be of the
same type. Consequently, we are not forced into any uncomfortable further assumptions to sustain these types. In particular, there is no analytical pressine
on this view to treat remarkably adverbs as properties of degrees. Rather, the
denotation of a remarkably adverb is ultimately predicated of a proposition, as
seems most natural. The problems associated with a predicate modifier denotation are avoided as well. On the current account, the type of the remarkably
adverb and its projections remains very simple, and more important, the same
as the corresponding adjective. So it is no surprise - and indeed, expected - that
remarkably adverbs should support their own degree words and project the full
adverbial extended projection. It will not be necessary to assmne either massive
systematic ambiguity of Degs or any novel otherwise unmotivated type-shifts,
because the types of all elements of the adverbial projection will be exactly the
same as they would otherwise be.
5 Relation to clausal counterparts

What has now been introduced is a kind of decomposition - the apparent meaning of remarkably adverbs has been split into two parts, one associated with the
remarkably adverb itself and one associated with its position. Among the chief
advantages of having done things this way is that it provides a simple theory not
only of the relation to the corresponding adjectives, but also of the relation to
clause-modifying uses. Given the same denotation for remarkably proposed in
(39), the right interpretation for its clause-modifying use in (42) will follow:
(42) (a)
(b)
[ Clyde is talle ] = 3d[df=C tall(Clyde)(d) d>s,fl]

\Remarkably, Clyde is talle ] = remarkable(A3d[dC
tall(Clyde)(d) d>s,flH] )
The denotation in (42) requires only that it be remarkable that Clyde is tall,
which seems to reflect what the clause-modifying use of remarkably means.
6 Approaching ad-adjectival subject-oriented adverbs

This approach is certainly not a comprehensive theory of ad-adjectival adverbial
modification. It addresses only one class of adverbs that occur in the extended
AP - but there are others. Among them is a class, to which I now fimi, that
bears some resemblance to subject-oriented adverbs in VP and hence takes us
one step closer to the most basic broader questions about adverbial modification.
These AP-modifying adverbs include some of the canonical examples of subjectoriented adverbs, which seem to contribute roughly their usual interpretation: 9
9
'Subject-oriented' is an especially unfortunate term in this context, but I stick with it for its familiarity. Many of the examples in this section have their roots in a collection of naturally-occurring
121
(43) Clyde seemed {intentionally/deliberately/accidentally/willingly} reliant on Herman.
That these do in fact have the agentive semantics that is a signature of subjectorientation in VP is clear from the coimterpragmatic inferences they create in an
environment like (44):
(44) #When he was served to his hungry Martian overlords,
Clyde seemed {defiantly/imapologetically/rudely} raw on the inside.
This sentence leads us to suppose Clyde had some control over his being raw on
the inside, in a way that we would not if the adverb were absent. Another distinguishing feature of (VP-)subject-oriented adverbs is focus-sensitivity, a characteristic ordinary maimer adverbs do not have (Wyner 1994, Geuder 2000).
These adverbs pattern with subject-oriented VP adverbs in this respect, too, as
the non-synonymy of (45a) and (45b) reflects:
(45) (a)
(b)
Greta seemed rudely reliant on HERMAN to clean up after her.

Greta seemed rudely reliant on Herman to CLEAN UP after her.
But despite this evidence for thinking these adverbs are in a meaningful sense
subject-oriented, there is a fundamental problem here. Adjectives by then very
natine are stative; subject-orientation by its very natine requires agentive or voluntary eventualities, which states cannot in principle be.
There is another difficulty as well, a compositional one similar to the one remarkably adverb gave rise to: If these adverbs are essentially subject-oriented,
how can they compose with an (extended) AP denotation? As before, a simple
intersective interpretation does not lead very far here. It is very unclear at best
how an adequate denotation for the adverb could be framed in the appropriate
way. Subject-oriented adverbs may denote properties of events, but - even setting aside the sortal difficulty about states versus events - there is no obvious
place to plug such a thing into the structure of an AP. Moreover, an intersective
interpretation is inherently symmetrical, which makes the prediction that (46a)
should feel redundant, which it does not, and that (46b) should be a contradiction, which it is not:
(46) (a)
(b)
Clyde seems both rudely vocal and vocally rude.

Clyde didn't seem rudely vocal; rather, he seemed vocally rude.
Perhaps it might be possible to swat this sort of observation away by appeal

to pragmatics. This does not seem implausible, but it is definitely swimming
against the empirical current.
examples gathered by Tom Ernst.
122
Marcin Morzycki
Naturally, the alternative of simply assigning these adverbs predicate modifier

denotations remains. This is, in fact, how Wyner (1998) treats subject-oriented
adverbs (in VP), proposing that they denote functions from properties of events
to properties of events. This sort of approach, though, presents the same problem with respect to further AdvP structure encountered above with remarkably
adverbs. If these adverbs denoted predicate modifiers, they would not be compatible with degree word denotations. But adverbs of this sort occur with degree
words quite readily:
(47) Clyde appeared {rather rudely/quite thoughtlessly/very cleverly} indifferent to others.
The puzzle this leaves us with is an echo of the one remarkably adverb presented.
It seems only natural, then, to consider applying the same tools here.
The model of the interpretation of remarkably adverbs developed above, in
which the adverb can enter semantic composition through the mediation of a
Deg that stitches things together type-theoretically and makes its own particular
semantic contribution, can help address both of these problems. To illustrate
this fact about the semantic combinatorics, it will be necessary to sidestep the
extremely important and obviously relevant but still rather murky issue of what
exactly the semantics of subject-orientation is. I thus will adopt the toy semantics - certainly inadequate - in (48), in which deliberately and deliberate both
simply denote properties of events:1"
(48) [deliberately^ = Ae . deliberate(e)
Accepting this, it seems plausible that a sentence such as (49a) might receive an
interpretation like (49):
(49) (a)
(b)
Clyde is deliberately reliant on Herman.

[ Clyde is deliberately reliant on Herman ] =
3s3e3e'[reliant-on-Herman(Clyde)(s) deliberate(e)
agent(e)(Clyde) cause(e)(e') become(e')(Clyde)(reliant-on-Hennan)]
What (49) means, then, is that Clyde was the agent of a deliberate event that
caused an event of Clyde becoming reliant on Herman. The claim that underlies
this is that a subject-oriented adverb in the extended AP is interpreted with respect to an event that stands in a particular (causal) relation to the state associated
with the adjective.
Because these adverbs and remarkably adverbs have apparently the same distribution inside AP, the syntax from which this sort of denotation will be built
can mirror the one proposed for remarkably adverbs above:
10
Attempts to address the semantics of subject-orientation in earnest include McConnell-Ginet

( 1982), Wyner ( 1994, 1998), Geuder (2000), Ernst (2002) and Rawlins (2003).
(50)
123
DegP
deliberately
Deg
I
[AGT]
reliant on Herman
The price to be paid here is that in order to build (49), a distinct Deg [AGT]
will have to be posited. With that done, though, the composition is relatively
straightforward, and proceeds along the same lines as Kennedy's for measure
phrases and the one for remarkably adverbs above:
(51)
[ Clyde is deliberately [AGT] reliant on Herman ] =

[ [AGT] ] ( [ reliant on Herman ] )( [ deliberately ] )( [ Clyde ] )
The additional agentive meaning, and the additional events that underlie it, will
b e i n t r o d u c e d b y [AGT]:
(52)
[ [AGT] ] = AAASAxAs . 3e3e'[A(x)(s) S(e) agent(e)(x)

cause(e)(e') become(e')(x)(A)]
(53) [ [AGT] ] ( [ reliant on Herman ] )([ deliberately ])([ Clyde ]) =
As3e3e'[reliant-on-Herman(Clyde)(s) deliberate(e)
agent(e)(Clyde) cause(e)(e')
become(e')(Clyde)(reliant-on-Herman)]
So the same theoretical architecture that provided an account of remarkably adverbs above seems to provide solutions to both of the problems this section began
with. The compositional issue is solved exactly as before - the adverb is interpreted as an argument of degree morphology, which does the essential compositional work. And the problem of relating subject-orientation, which is a notion
bound up with events, and the semantics of the extended AP, which is stative, is
solved as well: the degree head contributes a semantics that makes available to
the adverb a causing event of which it can be predicated. As before, the adverb
can have a simple, first-order denotation that (plausibly) remains constant across
its uses in various positions.
Though no substantive proposal will be offered here of how ad-adjectival
uses of subject-oriented adverbs relate to subject-oriented uses in VP, it is worth
noting that elements of the account suggested here bear a surprising resemblance
to the model of subject-orientation that Wyner (1998) constructs. He argues that
a verbal functional head, a 'volitional' fonn of the passive auxiliary be, is crucial
in explaining why the interpretation of subject-oriented adverbs can be affected
by passivization. Both this element and [AGT] occupy functional heads in the
124
Marcin Morzycki
extended projection of the modified expression, and both contribute an agentivity

inference to the interpretation of subject-oriented adverbs.
7 A final word
The principal argument here has been that the syntactic and semantic architecture of at least one and perhaps two classes of AP-modifying adverbs involves
precisely the same semantics for the adverb itself as in other positions, with additional, specifically ad-adjectival meaning arising through its interaction with a
degree morpheme that introduces it. For remarkably adverbs, this additional semantics seems to involve widening the domain of degrees; for subject-oriented
ad-adjectival adverbs, it seems to involve some notion of agentivity. That this
'factoring-out' approach proved useful in both cases may suggest that it may be
fruitfully applied more widely, perhaps in some f o n n to more prototypical adverbiale as well. But one way or another, these AP-modifying adverbs offer a
novel perspective on familiar larger questions about adverbial modification.
References
Abney, S. (1987): The English Nomi Phrase in Its Sentential Aspect. Doctoral dissertation, MIT.
Bartsch, R. (1976): The Grammar of Adverbials. North-Holland, Amsterdam.
Bierwisch, M. (1989): 'The semantics of gradation'. In Bierwisch, Manfred and Ewald
Lang, eds., Dimensional Adjectives. Springer-Verlag, Berlin.
Cinque, G. (1999): Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press, New York.
Corver, N. (1990): The Syntax of Left Branch Extractions. Doctoral dissertation, Tilburg
University.
Cresswell, M. J. (1976): 'The semantics of degree'. In Partee, Barbara H., ed., Montague
Grammar. Academic Press, New York.
Elliott, D. E. (1974): 'Toward a grammar of exclamations'. Foundations of Language
11:231.
Emst, T. (2002): The Syntax of Adjuncts. Cambridge University Press, Cambridge.
Faller, M. (2000): 'Dimensional adjectives and measure phrases in vector space semantics'. In Faller, M., S. Kaufmann, and M. Pauly, eds., Formalizing the Dynamics of
Information. CSLI Publications, Stanford.
Geuder, W. (2000): Oriented Adverbs: Issues in the Lexical Semantics of Event Adverbs.
Doctoral dissertation, Universitt Tbingen.
Grimshaw, J. (1979): 'Complement selection and the lexicon'. Linguistic Inquiry
19(2):279.
Grimshaw, J. (1991): 'Extended projection'. In Lexical Specification and Lexical Insertion. Erlbaum, Hillsdale, NJ.
Groenendijk, J. and M. Stokhof (1984): 'On the semantics of questions and the pragmatics of answers'. In Landman, Fred and Frank Veltman, eds., Varieties of Formal
Semantics. Foris, Dordrecht.
125
Hamblin, C. (1973): 'Questions in Montague English'. Foundations of Language

10(1):41.
Heim, I. and A. Kratzer (1998): Semantics in Generative Grammar. Blackwell Publishers,
Oxford.
Jackendoff, R. (1972): Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, Massachusetts.
Jackendoff, R. (1977): X-Bar Syntax: A Study of Phrase Structure. No. 2 in Linguistic
Inquiry Monographs. MIT Press, Cambridge, Mass.
Kadmon, N. and F. Landman (1993): 'Any'. Linguistics and Philosophy 16(4):353.
Karttimen, L. (1977): 'Questions revisited'. Unpublished manuscript, The Rand Corporation.
Kennedy, C. (1997): Projecting the Adjective: The Syntax and Semantics of Gradability and Comparison. Doctoral dissertation, University of California, Santa Cruz. Published in 1999 by Garland, New York.
Kennedy, C. and L. McNally (2004): 'Scale structure and the semantic typology of gradable predicates'. Under review for Language.
Klein, E. (1991): 'Comparatives'. In von Stechow, Arnim and Dieter Wunderlich, eds.,
Semantik: Ein internationales handbuch der zeitgenssischen forschimg. Walter de
Gruyter, Berlin.
McCawley, N. (1973): 'Boy, is syntax easy!' In Coram, C., T. Cedric Smith-Stark, and
Ann Weiser, eds., Papers from the Ninth Regional Meeting of the Chicago Linguistic
Society. CLS.
McConnell-Ginet, S. (1982): 'Adverbs and logical forni: A linguistically realistic theory'. Language 58:144.
Michealis, L. and K. Lambrecht (1996): Conceptual Structure, Discourse, and Language.
CSLI, Stanford.
Moltmann, F. (1997): Parts and Wholes in Semantics. Oxford University Press, New
York.
Rawlins, K. (2003): study in some adverb denotations'. B.A. honors thesis, University
of Massachusetts Amherst.
Rullman, H. (1995): Maximality in the Semantics of lf7?-Constractions. Doctoral dissertation, University of Massachusetts.
Russell, B. (1905): 'On denoting'. Mind 14:479.
Schwarzschild, R. and K. Wilkinson (2002): 'Quantifiers in comparatives: A semantics
of degree based on intervals'. Natural Language Semantics 10(1 ): 1.
Seuren, P. A.M. (1973): 'The comparative'. In Generative Grammar in Europe. Reidel,
Dordrecht.
von Fintel, . (1994): Restrictions on Quantifier Domains. Doctoral dissertation, University of Massachusetts Amherst.
von Stechow, A. (1984): 'Comparing semantic theories of comparison'. Journal of Semantics 3:1.
Westersthl, D. (1985): 'Determiners and context sets'. In van Bentham, Johan and Alice
terMeulen, eds., Generalized Quantifiers in Natural Language. Foris, Dordrecht.
Winter, Y. (2001): 'Measure phrase modification in vector space semantics'. In
Megerdoomian, Karine and Leora A. Bar-el, eds., Proceedings of WCCFL XX. Cascadilla Publications, Somerville, Mass.
Wyner, A. (1994): Boolean Event Lattices and Thematic Roles in the Syntax and Seman-
126
Marcin Morzycki
tics of Adverbial Modification. Doctoral dissertation, Cornell University.

Wyner, A. (1998): 'Subject-oriented adverbs are thematically dependent'. In Rothstein,
Susan, ed., Events in Grammar. Dordrecht: Kluwer.
Zanuttini, R. and P. Portner (2003): 'Exclamative clauses: At the syntax-semantics interface'. Language 79(1 ):39.
Zwarts, J. (1997): 'Vectors as relative positions: A compositional semantics of modified
PPs'. Journal of Semantics 14:57.
Zwarts, J. and Y. Winter (2000): 'Vector space semantics: A modeltheoretic analysis of
locative prepositions'. Journal of Logic, Language and Information 9:171.
Kjell Johan Sasbo (Oslo)

1 Introduction
A number of verbal predicates have an intuitively rather abstract meaning; they
may specify some higher-order, typically modal, property of an action but they
remain tacit on what is actually going on. Here are some examples:
(1 )
(a)
(b)
obey doctor's orders, do me a favour, transgress Holy Law,

give way, respond, start the nest, badger the bureaucracy,
take revenge on the remote father
create a fiction, ruin my reputation, surprise the nation,
waste fuel, help the campaign of Senator John Kerry,
undermine the war on terrorism
The items in (lb) are causative predicates, more precisely such that do not
specify the way in which the change of state is brought about; Kearns (2003)
calls them causative upshot predicates. I call them manner-neutral causatives.
The predicates in (la) are not causative. Ryle (1949: 125-147) classifies them
as achievements; Kearns (2003: 599) refers to them as criterion predicates:
The key notion here is that there is some conventional criterion an
action must meet in order to qualify as an event of the criterionmatching kind.
While criterion predicates specify conventional (nonnative) or intentional criteria, they are imspecific about the physical criteria an action must meet. Usually,
there is a need for more information on how the action is perfonned. If you ask
me to do you a favour, I will want to know what it is. If you tell me that you
are obeying doctor's orders, you are probably alluding to a familiar action. Very
often, the context will, in various ways, fill in the picture. Much the same is true
of manner-neutral causatives.
One way of specifying more concrete criteria is to modify the VP with an
adjunct; a clause or a PP. In English, the natural choice is a by adjunct with a
present participle complement, as in (2)-(5).
(2)
(3)
The City retaliated by electing its own mayor.

Mowgli kept a promise by killing Shere Khan.
128
Kjell Johan Sbo
(4)
By sending rain, Yahweh had usurped the function of Baal.
(5)
It tries to escape by moving as fast as possible away from the predator.
This extends to manner-neutral causatives, as in (6)-(9).

(6)
(7)
(8)
(9)
By calling and dancing, he entices a female to Iiis bower.

Yahweh saved the Israelites by opening the Sea of Reeds.
Yahweh made Adam by scooping up some clay and breathing on it.
In Germany they portrayed the Plague as a maid travelling through the air like a
blue flame, killing her victims by raising an arm.
(I will subsume criterion predicates and manner-neutral causatives under the

tenn abstract predicate.) There is a strong intuition that in each case, the merge
of the by plnase and the plnase it modifies denotes one set of events, and that
somehow, the by plnase predicate fills a slot in the abstract predicate. My intent is to account for these intuitions through a fonnal analysis of the abstract
predicate, the instrumental adjunct, and the way they are composed.
The rest of the paper is structured as follows. In Section 2,1 review the recent
work of Kearns (2003) on abstract predicates and argue that it is incomplete,
both as it stands and as a basis for an analysis of the by locution. Section 3
provides a critical survey of work on the by locution (Bennett 1994) and on by
adjuncts in connection with causative predicates (Dowty 1979). In Section 4,
I develop my own analysis based on unification in recent DRT. In Section 5, I
discuss the limits to the by locution, arguing that they can be stretched through
causative or criterial interpretations of predicates that are not strictly causative
or criterion predicates. Section 6 offers conclusions.
2 Kearns 2003 and the Anscombe thesis

According to Kearns (2003), criterion predicates do not refer to basic actions or
events but to actions or events that depend on basic actions or events, to parasite
events depending on host events (p. 600).
What is ... parasitic about criterion predicates,... , is that the eventualities described cannot simply come about, but must be realized
in the occurrence of some event which is describable in different
tenns. (Kearns 2003: 627)
In the acfiial analysis, there is only one event involved, but there is a (usually
rather vacuous) host and a parasite description of that event. The latter, the criterial component of the predicate, is an "individual-level predicate on events",
as indicated in paraplnases like those in (10) (p. 628):
(10) (a)
(b)
Jones broke the law.

'Jones did something, and what he did was illegal'
(c)
129
3 e (do(j, e) & illegal(e))
This analysis is questionable for two reasons. First, it is hardly reasonable to

ascribe properties like legality to events. Generally, it would seem that what
must meet conventional criteria are event types - predicates - and not event tokens.1 Note that cases like (11a) and (12a) turn out to be trivial - contradictory
or tautologous - on an analogous analysis, as demonstrated in (11c) and (12c).
Obviously, one and the same event e can only occur once.
(11) (a)
(b)
(c)
(12) (a)
(b)
(c)
Joan made a common mistake.

'Joan did something, and what she did was a common mistake'
3 e (do( j, e) & mistake(e) & common(e))
Joan did something noone had ever done before.
'Joan did something, and what she did noone had ever done'
3 e (do( j, e) & noone had ever done e before)
In fact, there is another possible formalisation in the style of the (c) formulae,
equally in accordance with the (b) paraphrases, avoiding these problems:
(10) (d)
(11) (d)
(12) (d)
3 e 3 (P( j, e) & illegal(P))

3 e 3 (P(j, e) & mistake(P) & common(P))
3 e 3 (P(j, e) & noone had ever done before)
This is close to what I will propose in Section 4.

Second, this analysis is questionable because it fails to fonn a sound basis for
an analysis of the by locution. Kearns does not offer an explicit analysis of the by
locution. In fact, she assmnes that criterion predicates, as opposed to causative
predicates, occur more naturally with in -ing adjuncts (p. 629). It may be that
in adjuncts of the type illustrated in (13) and (14) preferably modify criterion
predicates. But cases of such predicates with by adjuncts abound in corpora, and
Kearns herself discusses (p. 599) a case in which "a soldier obeys an order by
fixing his bayonet".
(13) In naming him, Putin ended a guessing game that had begun to overshadow a predictable presidential election two weeks from now that is seen as a sure tiling for
Putin.
(14) In naming him, Putin avoided creating an alternative center of power or a rival for
the political spotlight.
In any case, it is clear that Kearns considers in or by adjuncts to offer host descriptions, potentially specifying the profonna "do" predicate in formulae like
(10c) above. One can thus conjecture that (15a) may analyse as (15c).
There is an interesting parallel to the notion of felicity: Krifka ( 1998) argues, contra, e.g., Rothstein ( 2004 ), that telicity cannot be a property of event tokens but must be a property of event
types.
130
Kjell Johan Sbo
(15) (a)
(b)
(c)
Jones broke the law by limiting.

'Jones limited, and Iiis limiting was illegal'
3 e (hunt( j, e) & illegal(e))
This corresponds to what has been referred to in the philosophical literature as

the "Anscombe thesis" in its very simplest fonn.
The Anscombe thesis (according to Bennett 1994)
If someone 0s by 7ring, and F is the act which makes it the case that
she 0s, and is the act which makes it the case that she 7rs, then F
is P.
In other words, the modified predicate and the by adjunct describe one event
in two ways. To be sure, there is a strong intuitive basis for this assumption.
However, as spelt out in (15c) or any formula where the host and the parasite
are parallel predicates and where the latter is a first-order predicate on events, it
causes two (closely related) problems.
First, it predicts that the construction is closed under weakening, which it is
arguably not:
(16) (a)
(b)
He broke the Jungle Law by limiting at the pool in a drought

He broke the Jungle Law by limiting at the pool
He broke the Jungle Law by limiting
3 e (hmit(j, e) & illegal(e) & atthepool(e) & inadrought(e))
3 e (hmit(j, e) & illegal(e) & atthepool(e)) =>
3 e (hmit(j, e) & illegal(e))
Second, as pointed out, i.a., by Bennett (1994), the Anscombe thesis is liable to
predict a symmetry between the by adjunct and the modified predicate:
(16) (c)
(d)
He broke the Jungle Law by limiting.

? He limited by breaking the Jungle Law.
3 Second-order predicates decomposed

In this section, I discuss two approaches to the semantics of the by adjunct which
avoid the two problems noted above by treating what the by adjunct adjoins to as
a composite expression containing an existential quantification over things such
as those expressed in the by adjunct. This is a significant step forward. What
these approaches do not provide is a compositional analysis.
3.1 Bennett 1994 and the "namely" analysis
According to Bennett (1994), the asymmetry of the by construction falsifies the
Anscombe thesis that we have two descriptions of one and the same event. His
analysis differs from the one sketched in the last section in two ways:
131
1. At the relevant level of analysis, the by complement does not denote a set
of events; in fact, it denotes a (true) proposition (a fact).
2. At the relevant level of analysis, the pirrase modified by the by pirrase
denotes a second-order entity; in fact, a set of true propositions (facts).
(I will argue that 1. is inessential while 2. is essential.) Bennett paraphrases
(17a) as (17b). A formalisation in the style of the formula in (10c) or (15c) (ignoring tense) could yield (17c):
(17) (a)
(b)
(c)
Jones broke a promise - by - coming home late.

Some fact about his behavior conflicted with a promise he had made earlier namely the fact that - he came home late.
comehomelate(he) & promise( hiot(comehomelate(he)))(he)
Without violating the spirit of this analysis, one could reintroduce events to represent the verb pirrase break a promise by being late as (17d) (simplified):
(17) (d)
e ( la.te(i)(e) 3 ei ( promise(
3 ei (late(x)(e2)))(x)(ei)))
As a representation of the modified VP, this is intuitively not far off the mark,
and it seems to avoid the two problems noted above: The construction is not
predicted to be closed under weakening or to be symmetric. The reason is that
the "parasite", the criterial component, is not at the same level as the "host";
it is one level up and has an argument place for the host. In principle, this
pattern generalises to manner-neutral causative predicates, but this is yet to be a
semantic analysis - it is not clear how the meaning of the modified VP comes
from the meaning of its two daughters through the "namely" operation. Indeed,
it is not easy to develop a compositional analysis along these lines.
3.2 Dowty 1979 and the by postulate
Dowty (1979: 227-229) treats by adjuncts as modifying causative VPs. He
first considers ascribing a causative element to by, but rejects this because it
seems to result in a double causation. 2 It may be added that such a move is also
problematic in connection with criterion predicates, for which the result does
not seem to involve any element of causation. What Dowty proposes, instead of
a translation of the preposition, is a meaning postulate:
VpVPVQ Vx [ b y f ( P ) ( y [Q{y} CAUSE > ] ) ( x ) - - [P{x} CAUSE > ] ]
This ensures that if John awakened Mary by shaking her, then his shaking her
awakened her - a welcome result. Dowty did not use events, but in principle,
2
In fact, one may be tempted to such a move by considering predicates, activities or achievements,
that are neither clear causative nor clear criterion predicates; cf. Section 5.
132
Kjell Johan Sbo
the by postulate could be reformulated in terms of events. And in principle, it

generalises to criterion predicates (on an appropriate decomposition).
But of course, the postulate does not amount to a compositional analysis. It
does not specify the meaning of the by pirrase, and in particular, it does not say
what, if any, predicates the by pirrase cannot meaningfully modify. If the by
pirrase combines with a predicate not of the fonn (y [Q{y} CAUSE p]), the
meaning postulate does not apply, so it is unable to predict negative facts like
those m (18).
tied his necktie
combed his hair
b u t t o n e d his shirt
polished his nails
p u t on his t o p h a t
> by . . .
J
The scope of such negative facts may be debatable; Dowty himself (p. 229)
mentions the case in which John "hammers the metal flat by pounding it with a
pipe wrench". The boundary between abstract and concrete predicates is fuzzy
and flexible, but the cases in (18) are evidence that there are predicates that are
definitely too "concrete" to be modified by by pirrases. I will retimi to this topic
in Section 5.
One way to build a compositional analysis is to give the abstract predicate a
separate argument place for a by phrase predicate (simplified):
"AwakenMary": APAe 3ei[Cause(Become(awake(m))(ei))(P(e))]
"Keep a promise": A e [ P(e) Promise(P)( Agent(e)) ]
But this is hardly plausible considering the cases where the abstract predicate
occurs "on its own", without being modified by anything more specific. It would
seem, therefore, that one must look farther afield for a compositional analysis
preserving the ideas of Bennett and Dowty.
4 The analysis
The discussion hi the last two sections has suggested that abstract predicates
should not be described (only) as predicates of events but (also) as predicates of
predicates of events, that is, as second-order predicates of events, and that when
modified by a by phrase, they are predicated of the by phrase predicate. More
precisely, there is reason to assume the following hypothesis:
Hypothesis
If someone 0s by 7ring, then says that she does a such that
... (for instance, is something promised, or her doing - causes
something), and - is .
133
To develop this hypothesis into a viable analysis, I will first show how one can
make fonnal sense of it in a version of Discourse Representation Theory. Next,
I will show how the problem of composing the representations can be overcome
through the notion of unification used in recent DRT. I illustrate various combinations of predicates and show how negative facts can follow from a failure of
unification. On the resulting analysis,
Bennett's and Dowty's ideas are rendered in a compositional version
the Anscombe thesis is vindicated: There are two descriptions of one event
the symmetry problem is solved: There is symmetry at event token level
but asymmetry at event type level
a prepositional notion of causation is vindicated.
4.1 The desideratum
I will assmne that the result of by adjunction denotes a set of events: Ae [...],
and that the by phrase predicate is predicated of those events; if the by phrase
is by reversing, we have Ae [... reverse(e)...}.
Considering a sentence like
(19a), I will assmne that the by phrase adjoins at the level of the VP, cf.(19b),
and that the Agent relation comes into play at a later stage (cf. Kratzer 1996).
(19) (a)
(b)
(c)
Neither would give way by reversing.

give way by reversing
Ae [. .. reverse(e). . . ]
On top of this VP, three functional heads round off the sentence:
Voice: (e.g.) XxXe [Agent(x)(e)]
(a function from objects to sets of events)
Aspect: (e.g.) \P\t3e
[P(e) Perfective(e)()]
(a function from sets of events to sets of times)
Tense: (e.g.) Past( 0 )(i)
(a time)
These three functions will be disregarded in the following.
What remains in the skeletal representation of the result of by adjunction
Ae [... P(e)... ], where is the by phrase predicate, is a representation of the
modified abstract predicate that involves P , and this requires decomposition.
Wien, as in (19a-c), the modified abstract predicate is a criterion predicate, decomposition is especially difficult. Let us begin with a causative predicate, for
which we have some experience with decomposition. Consider (20a-c):
134
(20) (a)
(b)
(c)
Kjell Johan Sbo

She maddened me by dancing.
madden me by dancing
\e [. . . dance(e) . . . ]
According to an event-based notion of causation which has become customary

over the last years (cf. e.g. Pylkknen 2002), one would expect (20c) to take the
more specific fonn of (20d):
(20) (d)
\e 3 ei [Become('mad(?'))(ei) dance (e) A Ca.use(ei)(e) ]
But this is a representation of the causative predicate (madden ) which does not
involve dance, the by pirrase predicate, and it is difficult to see how the symmetry
problem can be overcome on such an analysis. There is, however, an alternative
decomposition, more in line with Dowty's work (1976, 1979), where causation
is not a relation between events but between propositions (although intensions
are notationally disregarded below):
(20) (e)
e[3ei [Become('mad())(ei) dance(e)

Cause(Become(mad())(ei))(dance(e)) ]
Here, it is clear that the abstract predicate involves the by pirrase predicate dance occurs twice in the representation.
To be sure, it is debatable whether this is the best formulation of causative
verb causation, but in any case, a coimterfactual analysis of causation (Lewis
1973) is more natural on the basis of a formulation where, as here, the causing
event type enters into the causation relation than on the basis of one where only
the causing event token enters into it.
Turning to criterion predicates with by adjuncts, a similar pattern emerges: To
the extent that a decomposition is feasible, it will involve the by adjunct. Take
the predicate give way from (19a). This seems to entail doing something another
party insists upon, although it does not follow from objective norms. Let us assume that a decomposition along these lines is theoretically possible (although
rather indeterminate). Then give way by reversing seems to be the same plus the
condition that reversing is that something. On the assumption that something
similar holds of all criterion predicates, we can focus on one case where a decomposition is not merely possible but relatively practical, keep a promise (still,
(21c) is a simplification):
(21) (a)
(b)
(c)
She kept a promise by dancing.

keep a promise by dancing
\e 3 ei [Pro'mjse(dance(pro))(Agent(e))(ei) dance(e)]
In order to prepare the ground for a compositional analysis - how to arrive at

representations like (20e) and (21c) - it is useful to cast the representations in a
Discourse Representation Structure fonnat ((20b)=(22a), (21b)=(23a)):
(22) (a)
135
madden me by dancing
ei
(b)
(23) (a)
Ae
dance (e)
Become(mad(?'))(ei)
Cause(Become ( m a d (?') ) (ei ) ) {dance (e) )
keep a promise by dancing
ei Q
(b)
Ae
dance (e)
dance C Q
Promise (Q (pro) ) (Agent (e) ) (ei )
So far, so good - but the problem is that it is far from obvious how to derive
these structures in a compositional maimer. We can go some way towards identifying the contribution of the abstract predicate and that of the adjunct through
formulations corresponding one-to-one to the above hypothesis; "if someone
by 7ring, then says that she does a - such that... and - is ":
ei
(22) (c)
Ae
P(e)
Become(mad(?'))(ei)
Cause(Become(mad(?'))(ei))(P(e))
! = dance !
ei Q
(23) (c)
Ae
P(e)
PQ
Promise
(Q (pro) ) (Agent (e) ) (ei )

' = dance '
This serves to isolate the problem: The sole contribution of the by adjunct seems
to consist in the condition = dance', but if the contribution of the modified
predicate is everything but that condition, it is difficult to see how the by phrase
can have access to the event type discourse referent - as long as we maintain
traditional wavs of comnosition.
136
Kjell Johan Sbo
4.2 Composition by unification: Stores and binding conditions

There are a variety of ways of composing, . . .
Duke Ellington
Recent work in DRT (e.g. Bende-Farkas and Kamp 2001, Kamp 2001) uses
unification rather than functional application as a method of composition. So
far, this method has mainly been used for the representation of semantic incorporation (see below); I will argue that abstract predicate modification represents
another case for which it can make a positive difference. First, it is necessary to
describe the novel features in general tenns.
A preliminary representation of a node consists of a store and a content. A
store consists of triples: A variable, constraints, and a binding condition. Here
I will assmne just pairs: A variable and a binding condition, (
,
}. A
content is a DRS:
When two nodes meet, the unification of store variables of the same type is
driven by the binding conditions, and the two content DRSs are then merged.
The binding conditions that a store variable may be subject to include:
definite (BCdef), indefinite (BC in d e f), and quantificational (BCq).
A quantificational store variable must find an indefinite store variable to bind.
Bende-Farkas and Kamp (2001) use this to account for Definiteness Effects in
semantic incorporation (cf. Bende-Farkas 1999 and Farkas and de Swart 2003);
for instance, there be in English comes with a quantificational variable, and if the
matching variable from the sister NP is quantificational or definite, unification
will fail and the merge will be incoherent.
Indefinite store variables, on the other hand, do not need to be bound, although
they easily are; if they are not, they eventually enter the content DRS as normal
(indefinite) discourse referents.
I will use three binding conditions,
for 'classical' abstraction,
indefinite, and
constant as a subsort of BCQ,
137
and I will assume that the by phrase introduces a constant predicate variable
while the abstract predicate introduces another, indefinite predicate variable.
When the two phrases meet, the fonner will bind the latter. If the by pirrase
meets a "concrete" predicate not introducing an indefinite predicate variable, the
unification fails and the composition terminates. If the abstract predicate does
not meet a by pirrase (or a similar modifier), the indefinite predicate variable
enters the content DRS as an ordinary discourse referent.
4.3 Examples
Let us first look at preliminary representations of two abstract predicates, one
causative and one criterial. (Note that these representations abstract away from
intensions, and that they represent oversimplifications in other respects as well.
Recall that I follow Kratzer 1996 in assuming the Agent relation to come into
play at a later stage; more on agentivity below.)
(24) (a)
(b)
(25) (a)
(b)
madden me
),
( , indefinite )
keep a promise
),
( , indefinite }
Next, let us see what a representation of a simple by adjunct might look like. I
will assmne that the function of the preposition is purely identificational: Essentially, it takes a predicate and returns a store-content pair introducing a constant
predicate variable with the content identifying this as the predicate:
(26) (a)
(b)
by dancing
( , constant. ) J>
138
Kjell Johan Sbo
When a by phrase like this modifies an abstract predicate such as (24a) or (25a),
the variable from the fonner binds the variable from the latter, the latter
being substituted for the fonner and entering the universe of the merged content
DRS. Below are some illustrations of this, as well as illustrations of cases in
which there is no store variable or no store variable is provided.
4.3.1 Simplex cases

Wien a manner-neutral causative predicate is modified by a by adjunct, (24c)
depicts how the by pirrase predicate identifies the maimer by unification:
(24) c. make me cry

; e , A),
I , indefinite )
unification
by calling
{, constant
make me ay by calling
ei
{ (e,A>
Pie)
Bec(cry(?'))(ei)
Cause(Bec(cry(?;))(ei))(P(e))
= Xe
collie)
ei
ie, A
call{e)
Bec(cry(?'))(ei)
Cause(Bec(cry(-))(ei))(caZZ(e))
139
Note that although the by adjunct is only ascribed an identifieational meaning

(and in particular, not a causative meaning), it is fully possible to represent what
seems to be the negation of this meaning, as in (27a); unification occurs but the
bound variable is claimed to be different from the constant predicate.
(27) (a)
sadden me (but) not by dancing (but... by singing)

ei
P(e)
(b)
Bec(sad(?'))(ei)
Cause(Bec(sad(i))(ei))(P(e))
\e
dance (e)
Note, also, that when an agent is eventually connected to the modified VP, via
the relation Agent(x) (e) (Kratzer 1996), it is the causing event, described by the
modifier, that is assigned agentivity; the caused event may well be unintentional.
This is as it should be.
It is an interesting question whether the empty grammatical subject of the
by phrase is always an external argument, essentially an agent; as it stands, the
analysis presupposes that it is. As the by phrase is not represented with a PRO
subject, a theme trace variable cannot be bound by anything. This predicts, in
particular, that there should be no passives in by adjuncts, and passives are indeed very rare; when they cannot be interpreted as covert actives, along the lines
of (28b), they seem rather marginal, cf. (29):
(28) (a) By being defeated, you have ruined everything.
(b) By letting yourselves be defeated, you have ruined everything.
(29) ? The mullah lost his honour by being lifted off the floor.
On the other hand, if desired, it would be possible, only more complicated, to

supply the representation of the by phrase with a PRO subject, controlled by an
agent DP or by a raised theme argument binding a theme trace variable in the
modified VP; hi the latter case, PRO could bind a theme trace variable hi the by
phrase. Whether this is deshable is primarily an empirical question. As long as
subjects seem to be agents, I will assmne that they are.
The composition of criterion predicates like (25a) and by adjuncts like (26a)
will parallel the composition of causative predicates like (24a) and by adjuncts
as shown in (24c) above. So will, hi principle, the composition of criterion
predicates and more complex by adjuncts, as shown hi (30) below.
Kjell Johan Sbo

4.3.2 Complex cases
keep a
promise
/ Q
; e , A),
P(e)
PQ
Promise(Q(j>To)) (Agent (e)) (/)
I , indefinite )
unification
by killing Shere
Khan
Q ei
' , constant
keep a promise
= Ae
by killing Shere
/
Khan
Qi
(pro)) (Agent (e)) (/)
Promise{Qi
(e,A>
PQQi
P(e)
Q e2
= Aei
e2
tW(ei)
Cause(t(s)(e1))(Q(e))
Q(e i )
tW(e2)
Cause(t(s)(e2))(Q(e1))
Qi
P'O7M'se(Qi(p]-o))(Agent(e))(/)
Cause(t(s)(e2))(Q(e))
tW(e2)
Ae
/
Aei
Q e2
Q(ei)
tW(e2)
Cause(t(s)(e2))(Q(e1))
Qi
141
(30) illustrates the modification of a criterion predicate through a by adjunct

whose predicate is itself complex, here a manner-neutral causative predicate
(whose indefinite predicate store variable is entered into the content DRS). The
bottom structure says that to keep a promise by killing Shere Khan is to do
something causing the death of Shere Khan such that doing something causing
the death of Shere Khan entails something that has been promised.
To be sine, the exact fonn of the representations is open to modifications.
This is particularly true of the decomposition of the concept keep a promise,
where modal and temporal parameters, while relevant, are not made explicit.
The event of keeping a promise must succeed the event of making a promise,
and the indefinite predicate referent may stand for a predicate in intension.
The essential thing is the unification of that referent, made accessible as a store
variable, and the predicate from the representation of the by phrase.
Note that when a manner-neutral causative predicate by phrase modifies another manner-neutral causative predicate, there is an asymmetry between the
two: With reference to (7), repeated here as (31a), to save the Israelites by opening the Sea is not to do something causing both the Sea to become open and the
Israelites to become safe but to do something causing the Sea to become open
such that that doing something causing the Sea to become open causes the Israelites to become safe, - as illustrated in (3 lb):
(31) (a)
(b)
Q
Yahweh saved the Israelites by opening the Sea of Reeds,

save the Israelites by opening the Sea:
ei e2
Q(e)
Bec(ja/())(ei)
Cause(Bec(o/7en(s))(e2))(Q(e))
Ae
Bec(open(s))(e2^
e2
Q(e)
Cause(Bec(ja/())(ei))
V
Bee (open (s))(e2)

Cause (Bec(op>en(s))(e2))(Q(e))
Note, finally, that the modification-by-unification mechanism is recursive; there

is no difficulty in representing the appropriate meaning of, say, (32): 3
(32) Mowgli kept a promise by killing Shere Khan by stampeding buffalo through a
ravine.
4.3.3 Concrete parasites

If composition by unification is to succeed when a predicate is merged with a by
adjunct, the predicate must provide an indefinite predicate store variable. Not all
3
That is, there is no theoretical problem; in practice, however, such a representation will easily
become very complex.
142
Kjell Johan Sbo
predicates do. I will argue later, in Section 5, that this is not a sharp, absolute
distinction and that predicates can be quite flexible in this regard; but some are
simply too concrete or manner-specific to be interpreted as providing an indefinite predicate variable playing a part in their interpretation. These predicates
supply the negative facts about by pirrase modification.
(18) ?? Fred ^
tied his necktie

combed his hair
b u t t o n e d his shirt
polished his nails
p u t on his t o p h a t
} by . . .
(33) illustrates the failure of composition by unification for the event type spew
all over a man and a woman and the by pirrase by getting blind dnmk on seven
gins and umpteen pints (inspired by Saturday night and Sunday morning by Alan
Sillitoe), a combination which would not be implausible were the by pirrase to
convey a causal relation on its own:
(33)
spew ail over a man and a woman

({
(e, ) I ,
spew...
(e)
by getting blind dnmk on seven gins and umpteen pints

', , constant )
fail.
Unification fails because the constant binding condition for , a subsort of Q for
Quantifieational, necessitates the binding of a variable with air indefinite binding condition in the store of the sister. Here there is none to be found, or even
accommodated.
Note the parallel to presupposition failure as failure of anaphora binding;
store elements with binding conditions of the Quantificational sort can be viewed
as intrasentential presuppositions-as-anaphora.
4.3.4 Lone parasites
The by adjunct requires a predicate providing an indefinite predicate variable,
but not vice versa: A causative or a criterion predicate can very well occur on
its own, without any sort of modifier, because the indefinite predicate store variable is transferred to the content DRS as a normal discourse referent if nothing
143
happens. It does not need to be bound; if it is not, it stays indefinite, as in these

examples:
(34) (a)
(35) (a)
(36) (a)
He did me a favor.
You have done a great deed.
The boy insulted me in your bar.
This is not to say that it stays indefinite in a broader context. In isolation, the
sentence may be represented with an indefinite predicate discourse referent:
e t
eC t
t < n
P(e)
Agent(e)(yoz/)
great(P)
But this can serve as a source or a target for intersentential unification, so
that the final representation of the discourse includes conditions of the fonn
= Xe ... and e = . . . , or equivalently, much as if there were a by adjunct.
This seems particularly common with criterion predicates, as shown below:
(34) (b)
(35) (b)
(36) (b)
"You want to tell me what this is about?"

"He did me a favor. I want to say thanks. That's all it is."
"It must have been quite a favor" I said. "Do you mind if I ask what he did?"
"He showed me a kindness when I was down on my luck."
You have saved the world from the evil witch. [ . . . ]
You have done a great deed.
The boy insulted me in your bar. He told me to shut up.
The establishment of such binding relations is based on pragmatic reasoning and

accompanied by discourse relations. In (35b), succeeds its unifier, and we may
speak of abstraction, while in (36b), we may speak of elaboration (Asher and
Lascarides 2003: 204-207, Behrens and Fabricius-Hansen 2002).
5 The scope of abstractness

Sometimes, a by phrase adjoins to a predicate which does not seem to be an abstract predicate, either a criterion or a manner-neutral causative predicate; typically, then, the by phrase seems to convey a causal relation on its own. Such
cases are, of course, a threat to the analysis proposed in the last section. My defense will be to argue that on closer inspection, predicates which do not appear
to be abstract really are, at least under the given circumstances; that is, predicates that may not be intrinsically causative or criterial can, under the influence
of certain factors, be interpreted as causative or criterial, one of these factors
being a merge with a by phrase. There is independent support for this, and at a
144
Kjell Johan Sbo
general level, there is reason to embrace the idea that criteriality and causativity
are not fixed and lexical but flexible and contextual categories.
Let us first consider some activity predicates modified by by pirrases.
(37)
(38)
(39)
(40)
Snakes move by throwing their bodies into backward-moving waves.

They feed by filtering food particles from the water.
It swims by flexing its body from side to side.
The majority of their people live by fanning.
It is not unreasonable to assume that the predicates move, feed, swim, live are
used here in a slightly derived, abstract sense:
to move or swim in the relevant sense is to propel oneself
(through water)
to feed in the relevant sense is to obtain food
m the genetically encoded way
to live m the relevant sense is to sustain life;
to satisfy one's "basic needs"
Thus interpreted, the apparently intransitive activity (or even stative) verbs are
m actual fact transitive and causative accomplishments (though temporally, they
remain atelic due to iterativity or to the fact that a change of state is not brought
about but prevented (cf. Dowty 1979: 124)). As such, they introduce indefinite
predicate variables for the causing activity. The verb feed features an additional
criterion that the indefinite predicate must satisfy, as suggested by the formulation "in the genetically encoded way".
Let us next consider some achievement predicates modified by by phrases.
Much the same story can be told about them:
(41 ) They find prey by detecting minute vibrations from a distance away.
(42) . . . . a project to reach India not by following the coastline of Africa ... but rather
by plunging boldly into the unknown Western ocean.
(43) He was forced to forfeit the medal he had won by cheating.
(44) He claimed that he had escaped by crossing the Congo.
In fact, a relevant story has already been told about such cases: To account for
"progressive achievements", Rothstein (2004: 45-50, 136-139) proposes that
achievement predicates have a double nature: They can be coerced, or shifted,
to activities culminating in achievements, that is, to accomplishments:
S H I F T ( V P p u n c t u a i ) : Ae.(BECOME)(e)
Ae.3eieo [e = s (eiUeo)A(DO(a)(ei)A(BECOME(VP))(eo)ACul(e) =
e2]
145
The dummy predicate DO corresponds to the indefinite predicate variable

m the representations of abstract predicates in the last section. To find prey,
to escape, to reach India, or to win the medal in the broader, accomplishment
sense is to do something culminating in finding prey, reaching India, winning
the medal, or escaping in the narrower, achievement sense.
If, following Dowty (1979: 183), we take the presence of a causal event to
be the most salient distinction between achievements and accomplishments, we
can represent, e.g., escape in the shifted sense, escape+, as:
ei
<e, >,
( , indefinite } J
P(e)
ay7/?e(Agent(e))(ei)
C ause (escape ( Agent ( e ))( e ))( ( e ) )
Rothstein's shifting operation is supposed to be triggered by progressive aspect;

by adjunction now emerges as another factor triggering accomplishment readings of apparent achievements.
It should be accentuated that assuming criteriality and causativity to be elastic
notions is in no way a costly concession. On the contrary, it is what we should
expect. It would be surprising if the class of abstract predicates were closed
and solely lexically determined. Elasticity is welcome because it reflects the
basically relative (functional, pragmatic) nature of abstractness. In this light, it
is not surprising that the limits to the by locution are fuzzy. They are, we may
say, just as fuzzy as they ought to be.
6 Conclusions
It seems, then, that the key to a better understanding of the by locution is a
better understanding of the things it modifies, namely, abstract predicates, and
vice versa. The need to overcome the "symmetry problem" forces a reassessment of criterion predicates and manner-neutral causatives as predications not
merely over events but over sets of events. Conversely, once it is appreciated
that predicates with a by adjunct involve a second, indetenninate predicate, it
becomes clearer what the contribution of the by adjunct should consist in: The
determination of that second predicate.
This does not proceed on its own, however. A lexical decomposition where
the indetenninate "second predicate" is visible remains useless as long as this
second predicate is inaccessible for determination through the by adjunct. Some
innovative method of composition is called for, and in fact available: Recent
work in DRT supplements (or supplants) reduction by unification. Constituent
representations are bipartite, and discourse referents figuring in the content section are entered as variables in the store section along with (constraints and)
so-called Binding Conditions that drive the unification. The by phrase can thus
146
Kjell Johan Sbo
be translated as a structure where the embedded predicate is represented by a

store variable with a condition ensuring its unification with the store variable for
the "secondpredicate" 4
This can be carried out in recent DRT; it can presmnably be modelled in
another framework as well. However, by pirrase or other intrasentential modification of abstract predicates is part of a larger picture encompassing intersentential fonns of unification between (in)determinate predicate referents. Here,
DRT will make a positive difference, inasmuch as even the representations of
full root sentences are in this framework only preliminary, open to linkings and
bindings driven by more or less "pragmatic" presuppositions. Ahead lies a better
understanding of a discourse relation like elaboration.
There are negative facts about by phrase modification, and they can be accounted for; but the limits to the locution are not that sharp. This reflects the
vagueness and context sensitivity of the boundary between the abstract and the
concrete. A predicate appearing concrete in one perspective may appear abstract
in another. It may be assmned that an instrumental adjunct can itself effect such
a shift in perspective.
The Anscombe thesis is vindicated: Just one action is indeed perfonned if one
signals by waving one's arm. This thesis seemed to rim afoul of the symmetry
problem as long as predicates like signai were not taken apart; once there is
asymmetry at event type level, however, symmetry at event token level ceases to
be a problem. An appealing intuition is thus proven viable.
References
Anscombe, Gertrude (1957): Intention. Oxford: Blackwell.
Asher, Nicholas and Alex Lascarides (2003): Logics of Conversation. Cambridge: Cambridge University Press.
Behrens, Bergljot and Cathrine Fabricius-Hansen (2002): Connectives in contrast: A discourse semantic sUidy of Elaboration. In Information StracUire in a Cross-Linguistic
Perspective, Hilde Hasselgrd et al. (eds.), 45-61. Amsterdam: Rodopi.
Bende-Farkas, Agnes (1999): Incorporation as Unification. Proceedings of the 12th Amsterdam Colloquium, Paul Dekker (ed.). Amsterdam: ILLC.
Bende-Farkas, Agnes and Hans Kamp (2001): Indefinites and Binding: From Specificity
to Incorporation. ESSLLI, Helsinki.
(http://www.helsinki.fi/esslli/courses/readers/K29.pdf)
Bennett, Jonathan (1994): The "Namely" Analysis of the by Locution. Linguistics and
Philosophy 17: 29-51.
It may be debatable in how strict a sense this scheme adheres to compositionality; if it is ultimately judged to transcend one's preferred compositionality notion, it is at least in good company
with recent work on incorporation and related matters arguing the need for moderately innovative
methods of composition; cf. e.g. Farkas and de Swart 2003.
147
Dowty, David (1976): Montague Grammar and the Lexical Decomposition of Causative
Verbs. In Montague Grammar, Barbara Partee (ed.), 201-245. New York: Academic
Press.
Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Reidel.
Farkas, Donka and Henriette de Swart (2003): The Semantics of Incorporation: From
Argument Structure to Discourse Transparency. Stanford: CSLI Publications.
Kamp, Hans (2001 ): The Importance of Presupposition. In Linguistic Forni and its Computation, Rohrer, Christian, Antje Rossdeutscher and Hans Kamp (eds.), 207-254.
Stanford: CSLI Publications.
Kearns, Kate (2003): Durative Achievements and Individual-Level Predicates on Events.
Linguistics and Philosophy 26: 595-635.
Kratzer, Angelika (1996): Severing the External Argument from its Verb. In Phrase Structure and the Lexicon, Rooryck, Johan and Laurie Zaring (eds.), 109-137. Dordrecht:
Kluwer.
Krifka, Manfred (1998): The Origins of Telicity. In Events and Grammar, Rothstein,
Susan (ed.), 197-235. Dordrecht: Kluwer.
Lewis, David (1973): Causation. Journal of Philosophy 70: 556-567.
Pylkknen, Liina (2002): Introducing Arguments. MIT Dissertation.
Rothstein, Susan (2004): Structuring Events: A Study in the Semantics of Aspect. Oxford: Blackwell.
Ryle, Gilbert (1949): The Concept of Mind. Harmondsworth: Peregrine Books.
Markus Egg (Groningen)

1 Introduction
In the interfaces between morphology or syntax and semantics, semantic contributions of the involved syntactic or morphological constituents are opaque, i.e.,
processes of semantic construction may only handle them as a whole, they are
blind to the inner structure of these constituents.
This principle is violated in many cases where the semantic contribution Cs of
a syntactic or morphological constituent C includes embedded eventualities, i.e.,
eventualities that do not show up as arguments of Cs. (The tenn 'eventuality'
refers to states of affairs of all kinds; following Davidson (1967), verbs and
their projections have an additional eventuality argument in their semantics.)
Reference to such embedded eventualities by a syntactic or morphological sister
constituent of C should be impossible - but it is not.
Consider e.g. the preferred reading of (1), 'person who usually dances beautifully'. Here the adjective qualifies an eventuality of dancing. This eventuality
is contributed in the semantics of the verb stem, however, it is embedded in the
semantics of dancer (a set of individuals) and thus does not surface as an eventuality argument in the semantic contribution of (1) as a whole (Larson 1998):
(1)
beautiful dancer
(1) has an additional reading 'beautiful person who usually dances'. Here the
adjective pertains semantically to the individual argument of the semantics of
dancer, hence, semantic construction for this reading is trivial.
Another well-known challenge for semantic construction, viz., the so-called
institutive readings of aga/n-sentences can likewise be subsumed under this phenomenon. E.g., the restitutive reading of (2) can be paraphrased as 'Max manipulated the window, and, as a result, the window was open, and it had been open
before':
(2)
Max opened the window again.
Here the idea is that change-of-state verbs like open lexically specify the aftermath or result state of eventualities in their extension. The transitive verb open,
150
Markus Egg
for instance, specifies the result state as a state of the object NP referent (in (2),
the window) being open (Dowty 1979). However, this result state is not an argument of the semantics of the verb as a whole. Again presupposes that a specific
kind of eventuality occurred before. In the restitutive reading, it must be of the
same kind as the result state. I.e., the restitutive reading of (2) follows immediately if we assmne that the modifier again may refer to the embedded result
state of the verb.
In addition, there is another, repetitive reading of (2), in which the adverbial
refers to the mam eventuality argument of the verb. Consequently, the preceding
eventuality which is presupposed in this reading is one of Max manipulating the
window, which results in the window being open (i.e., the whole causation is a
repetition of a previous one).
In analogy to the restitutive reading of (2), (3) can be interpreted as 'Max
manipulated the window and, as a result, it was open for two hours':
(3)
Max opened the window for two hours.
A literal interpretation is barred here due to the aspectual incompatibility of the

durative adverbial for two hours (which selects for unbounded expressions) and
the expression it modifies (roughly, the rest of the sentence, which is bounded). 1
In contrast, the aftermath predicate is unbounded, which allows the application
of the adverbial semantics.
Finally, reference to embedded eventualities may also show up at the morphology-semantics interface, in particular, for the 'bracketing paradox' noted by
Liideling (2001) for nominalisations of German separable prefix verbs like (4)
and analysed by Mller (2003).
(4)
Los- gerenn -e
start iter_nonii rim iter_noni2
'iteration of events of starting to rim'
In the morphology, the order of affixation is circmnfix Ge... e before prefix los-.
In the opposite order, the first part of the circmnfix could not be placed correctly between prefix and stem (Mller 2003). In its semantics, the circmnfix
contributes an iterative operator and maps the semantics of its base onto the
property of being an iteration of a kind of eventuality as determined by the base.
I.e., Gerenne refers to an iteration of running eventualities. If we now assmne
that the prefix contributes an inchoative operator that expresses the beginning
of an eventuality, prefixation of ios- to Gerenne should return an expression
denoting the beginning of an iteration of running eventualities. However, the
1
Properties of eventualities are bounded (formally, in the extension of a property of properties of

eventualities B D ) iff they do not apply to proper parts of elements in their own extension ( Krifka
1992):
(i) VP.BD(P)
Oe^^P(e')
151
prefix refers not to the main eventuality of the iteration, it refers to the embedded
eventualities (that make up the iteration), which returns an expression that refers
to repeated beginnings of running eventualities.
The examples (l)-(4) show that reference to embedded eventualities leads to
ambiguity if it is optional. Thus, (1) and (2) have additional readings in which
the modifier refers to the main eventuality argument of the modified expression.
In contrast, the modifier in (3) and the prefix los- in (4) cannot refer to the main
eventuality argument of its modified expression or its base, respectively, i.e.,
there is no ambiguity.
The structure of the paper is the following. After rephrasing the phenomenon
m a fonnal framework (the -cal cuius) to highlight the common ground between
the presented examples in section 2 , 1 will devote section 3 to a discussion of
previous analyses to the phenomena presented above. Then section 4 provides
a brief introduction into imderspecification formalisms, on which my analysis is
based. The proposed analysis is sketched in section 5, finally, I conclude with a
brief outlook in section 6.
2 Formalisation of the data

In this section, I will offer a fonnal account of the phenomenon. The goal of this
section is also to bring out the common ground between the presented data.
2.1 Agentive nouns
Reconsider the first example (1). Its preferred reading (approximately, 'person
who usually dances beautifully' 2 ) can be derived in two steps. The first step
consists in dividing the semantics of agentive nouns like dancer into the stem
and the affix meaning, where the stem semantics emerges as an argument of the
fimctor which is the semantic contribution of the affix:
(5)
'person who usually'

affix meaning
...
'dance s
stem meaning
The crucial point in this division is that the embedded eventuality argument from
the semantics of dancer is no longer embedded in the stem semantics.
The second step then is to pertain the adjective semantically to the verb stem
only. As a consequence, it can refer to the relevant eventuality argument in
a straightforward fashion, because this argument is open in the verb stem semantics, hence, accessible to a modifier like beautiful. Since the affix meaning
I do not attempt to reconstruct the semantics of these agentive nominis fully, since for the line
of argumentation in the present paper the exact spellout of the affix semantics is not relevant. All
that matters is that it comprises an operator that has the verb stem semantics in its scope.
152
Markus Egg
'person who usually X-es' is applied to the meaning of the stem only after modification by the adjective, the adjective ends up in the scope of the affix -er.
The following fonnal spellout of this basic intuition in the -calculus is based
on the reconstruction of the semantics of the agentive affix -er in (6a) as a function from the stem semantics (a relation between [ G ] individual arguments and one eventuality argument) to a set of individuals. (This is considerably simplified in that issues of argument structure and binding are ignored for
the purposes of this paper.)
Elements of this set of individuals are identical to an individual such that
when participates in an eventuality e (this is expressed by the relation in), then
e is usually a P-eventuality where is the agent. Here ' j ' is shorthand for a
sequence of zero or more individual arguments of the verb.
The definition of the generic quantifier GEN in (6b) adopts one of the versions
of the quantifier that are discussed in Krifka et al. (1995):
(6)
(a)
(b)
XPXz.GEN[e,x](xineAz = x,3_v.P(x,v)(e))
GEN[e,x]((x)(e),C(x)(e) ) iff(x)(e) usually entails C{x){e)
According to (6a), the semantics of dancer is reconstructed in (7a) as the set of

people such that when they are participating in an eventuality, it is usually an
eventuality of them dancing. Here the semantic contribution of the verb stem is
underlined, in this expression, there is a free eventuality argument. 3
If the semantics of the adjective pertains to only this underlined part, the result is the representation (7b) for the preferred reading of ( 1 ). Here the adjective
semantics refers to the embedded eventualities of dancing, thus, the expression
refers to people who are usually dancing beautifully. Its other reading is represented by (7c), which refers to beautiful people who are usually dancing:
(7)
(a)
Xy.GEN[e,x](x in e Ay = x, dance'(x)(e) )
(b)
Xy.GEN[e,x](x in eAy = , dance'()(e) beautiful'(e)
(c)
Xy.GEN[e,x](x in e y = , dance' () {e ) ) beautiful' (y)
2.2 Change-of-state verbs

For change-of-state verbs, I follow the decomposition approach as introduced in
the Generative Semantics tradition and worked out in detail in Dowty (1979).
I.e., the meaning of complex verbs is broken down into simpler meanings that
are linked by suitable operators. Consider e.g. the semantic decomposition for
the transitive verb open:
3
Here and in the following, I will use V\-equality (equivalence of Xx.P(.x) a n d P , if .v does not occur
free in the function P ) to keep semantic representations readable. E.g., the underlined expression
dance'(.Y) in (7a) is equivalent to Xf.dance'(.r)(f ), which highlights the open eventuality argument. In addition, -equality will be used (representations are semantically equivalent if they are
only distinguished by consistent renaming of bound variables ).
(8)
153
?. CAUSE (x, BECOME(Xe'.be-open'(y ) (<-' )))(<?)
In prose, (8) stands for a relation between two individuals and y and an eventuality e iff causes a change of state at e whose result state is y being open. 4
The semantic representation of causative verbs comprises the two operators
CAUSE and BECOME. They are defined in (9), the definition of CAUSE is
simplified in that it does not attempt to pin down the notion of causation. This
would require a much more involved (coimterfactual) definition, see e.g. the
discussion in Dowty (1979).
(9)
(a)
(b)
CAUSE(x,P)(e) iff brings about in e another eventuality e' in the extension

of
BECOME(P)(e) iff e is preceded by an eventuality e' for which - holds
and is succeeded by a P-eventuality e" and there is no smaller eventuality e'"
that also fulfills the first two conditions
The definition of BECOME in (9b) is basically Dowty's; in addition, I follow

Fabricius-Hansen (1975) and assmne that the existence of the preceding <?' is
presupposed.
The (underlined) argument of BECOME is a property of eventualities (called
aftermath predicate) that characterises the result state of the change of state. I.e.,
in this aftermath predicate, the result state occurs as a free eventuality argument.
Thus, by pertaining the modifier again only to this part of the semantics of the
verb, the modifier can refer to the result state, which yields the restiUitive readings of flgfl/77-sentences as well as the interpretation of (3), where the modifier
indicates the length of the result state.
The semantics of the restitutive reading of (2) is then (10a), here tense and
sentence mood are ignored and W is shorthand for the semantics of the window.
In prose: Max brought about a change of state whose result state was the window
being open. Presupposed is a preceding eventuality of the window being open.
In the repretitive reading, the adverbial pertains to the expression it modifies as a
whole, thus, in the semantic representation (10b), a previous eventuality of Max
opening the window is presupposed. The repetitive reading is thus stronger than
the restitutive one because its presupposition entails the presupposition of the
restitutive reading.
(10) (a)
(b )
(c)
CAUSE(max',BECOME(again'(Xe'.be-open'(W)(e') ) ) )
again' ( CAUSE ( max', BECOME {Xe'. be-open' ( W ) (e' ))) )
again'(P)(e) iffP(e), the presupposition is that there is a P-eventuality e' that
precedes e
See Blutner (2000) for an explanation of potential counterarguments to this kind of decomposition
as put forth e.g. in Shibatani ( 1976).
154
Markus Egg
The presupposition is due to again, semantic ally, the adverbial is just the identity
mapping (10c).
2.3 German pre fix-verb nominalisations
The case of German prefix-verb nominalisations is illustrated by formalising
the semantics of Gerenne and Losgerenne in (11a) and (11c), respectively. In
these representations, issues of argument linking and binding are once again
neglected. For Gerenne, the meaning is a property of eventualities e such that
e is an iteration of eventualies where some rims. The semantic contribution
of the stem renn- is underlined in (11a); here the mnning eventuality which the
prefix los- will refer to later is an open eventuality argument.
The semantics of Losgerenne can be obtained from (11a) by pertaining the
prefix meaning ( l i b ) not to the semantics of the base ( l i a ) as a whole, but
only to its underlined part, ( l i b ) maps -ary relations onto the -ary relation which involves the same individual arguments (if any) and the begin of a
P-eventuality (i.e., individual arguments are inherited). The resulting (11c) representation stands for a set of iterations of eventualities where some starts to
nm:
(11) (a)
(b)
(c )
A?.ITER(A?/3;c.run/(;c)(?/))(?)
^.(( )){e)
Xe.ITER( Xe'^x. BECOME ( run' {x)){e')){e)
The operator ITER is defined in ( 12) as a relation between properties of eventualities and eventualities e if e is the convex union (i.e., including anything in
between) of a set of eventualies E, each of whose elements is a P-eventuality. In
addition, e itself may not be a P-eventuality.
(12) WVe.ITER(P)(e) <- 3E.\Je'.e' e E
P(e') \JE = M -nP(e)
In siun, the discussion in this section shows that for each of the presented examples it is possible to divide the semantic contribution of the constituent C\ where
the relevant embedded eventualities occur into two parts. One part is embedded
hito the other one, hi this part, the eventualites embedded hi the semantic contribution as a whole are open arguments. Another constituent C2 can now refer to
the relevant embedded eventuality arguments of C\ by pertaining semantically
only to the embedded part of the semantics of C\.
155
3 Previous approaches
3.1 Agentive nouns
The modification of agentive nouns is the subject of Larson (1998). His analysis
is cast in the framework of Generative Grammar, which distinguishes different
levels of syntactic structure. These levels are related by structural relations like
the notorious 'move a ' . The level on which semantic interpretation is based is
the level of Logical Form and not the directly visible syntactic surface structure.
Larson represents the meaning of his example (13) by (14):
(13) Olga is a beautiful dancer.
(14) Te[Con(oiga', e) dance'(oiga',e)] [beautiful'(e)]
In this representation, T e ' is a generic quantifier for eventualities, 'Con' holds
for an individual a and an eventuality e iff e is contextually relevant and contains
A. In prose, (14) means that usually contextually relevant eventualities where
Olga dances are beautiful.
The underlying syntactic structure on which the semantic construction for
(14) is based is shown in (15):
(15)
NP
DP
I
Olga
\
/
dancer
AP
yx
-f
beautiful
(15) is only a small part of the syntax tree that Larson assmnes for (13), viz.,
the main part of the complement of be. In this subtree, Olga is the subject of
the predicate nommai dancer, therefore it is placed in the specifier position of
the nominal (Chomsky 1995). Larson also claims that the underlying position
of attributive adjectives is after their head nouns. To receive case and to get
agreement with the finite verb and the adjective right, Olga will later move from
SpecN to the specifier position of the AgrP. (Agr is the functional head for
subject-verb agreement. 5 )
The derivation of (14) from (15) is based on Diesing's (1992) Mapping Hypothesis. This hypothesis determines scope and restriction of a strong 6 quantifier
The INFL(ection) node from earlier analyses of verb inflection was split into several heads subsequently, one of these heads is Agr s .
Determiners that introduce strong quantifiers cannot occur in the context of there are..., e.g., all
or most. Accordingly, adverbials like always and mostly introduce strong quantifiers; Fe is strong,
too. Indefinites can introduce strong or weak quantifiers.
156
Markus Egg
ili a sentence (analysed as an Inflection Phrase) by dividing its syntax tree into
two parts. The lower part (the VP) determines the scope of the quantifier, the
upper part, its restriction.
E.g., the meaning of (16) would emerge as (17) if the semantic contribution
of usually is modelled in tenns of the generic operator GEN:
(16) Politicians smile usually.
(17) GEN[x, e] (politician'(x) Ax in e, smile'(x)(e))
In prose: Eventualies in which politicians participate are usually eventualities

in which they smile. Since politicians in (16) is outside the VP, it contributes
the property of being a politician to the restriction of the quantifier, while the
semantic contribution of smile, which is part of the VP, appears in the scope of
the quantifier.
This strategy of semantic construction is now applied to the structure (15).
The generic quantifier (which is contributed by dancer) has its scope determined
by the AP, and its restriction, by the rest of the syntax tree, which yields (14).
I.e., the semantics of dancer comprises both Con(x,e) and (lance'(x,e). (Olga
is an argument of dancer, hence, in the derivation of (14) the meaning of dancer
applies to the meaning of Olga.)
But this raises the question of how Larson would derive the semantic representation (19) for (18) from the syntax tree (20). His interpretation of (18) is
that usually contextually relevant eventualities (where Olga is a participant) are
eventualities where Olga dances:
(18) Olga is a dancer.
(19) re[Con(olga',e)] [dance'(olga',e)]
(20)
N P
/
DP
Olga
I
dancer
It is unclear how to derive (19) from (20) by the Mapping Hypothesis. In particular, it seems difficult to derive the fact that in this example, the semantics of the
nomi must provide both the restriction and the scope for the generic quantifier.
If one adopts the intuitions on the semantics of aga/n-sentences as spelt out in
section 2.2, then the challenge for semantic construction is to pertain a modifier
like again not to its semantics as a whole but only to its aftermath predicate.
157
From the range of previous approaches to the phenomenon I will expound and
discuss meaning postulates and syntactic
decomposition1
3.2.1 Meaning postulates
The use of meaning postulates is advocated by Dowty ( 1979) to derive restitutive
readings of aga/n-sentences. Dowty's postulate claims equivalence between the
repetitive readings of aga/n-sentences and their restitutive readings. The postulate is given in an adapted fonn as (21 ), where the repetitive reading appears on
the left side of the equivalence relation, and the restitutive reading, on the right
side:
(21 ) VxWVe. [again'(CAUSE(x,BECOME(P}))(e)
<- CAUSE (x, BECOME ( again' (P ) ) ) {e )]
This meaning postulate allows one not only to infer the (weaker) restitutive reading from the repetitive reading but also the other way round. But this second,
problematic inference turns out to be eventually responsible for the problems
which Zimmermann (1993) notes for this postulate.
What is more, such a meaning postulate approach is bound to miss the common ground between the cases of reference to embedded eventualities as discussed in this paper for principled reasons: It is only available for expressions
that are ambiguous between an (easily derivable) reading without reference to
embedded eventualities and a (hard to derive) reading where there is such reference. But this strategy breaks down already for modification of change-of-state
verbs by durative adverbials like in (3), it could also not be extended to captine
the case of German prefix-verb nominalisation: For these cases, no meaning
postulate in the spirit of (21 ) could derive a reading with reference to embedded
eventualities from another reading without such a reference, simply because the
latter reading does not exist.
3.2.2 Syntactic decomposition
To preserve the intuitions of Generative Semantics while at the same time avoiding the problems of meaning postulates like (21), von Stechow (1996) assumes
a syntactic decomposition of complex verbs like leave in the framework of Generative Grammar. 8 Consider a more complex example from German, the verb
ffnen 'open', where wieder is the counterpart of again:
Two other approaches to the phenomenon, viz., explication of result states (Dlling 1998) and the
simulation of a scope ambiguity ( Blutner and Jger 2003 ) are discussed in Egg ( 2005 ).
This analysis also aims at explaining the semantic effect of word order variation in wiedersentences. I cannot go into this matter in the present paper, but see Blutner and Jger (2003)
and Egg ( 2005 ) for discussion of this point.
158
Markus Egg
(22) da Max wieder alle Fenster ffnete

that Max again all windows opened
'that Max opened all windows again'
The relevant part of the syntactic structure of (22) (without wieder) is depicted
in (23). In this syntax tree the verb is introduced not in a single leaf node but in
the VoiceP as a whole. The VoiceP is one of the nodes that introduce inflection.
It is the place for the material that determines the voice (see Kratzer 1996). Its
specifier position is the base position of the subject NP (with nominative case).
The adjective offen 'open' in the small clause XP that is the complement in the
VP denotes the property (being open) that holds for the second argument of
the verb in the aftermath of eventualities that can be characterized by ffnen 'to
open'. Its complement is the base position of the object NP (in the accusative
case).
(23)
VoiceP
NPi,
Voice'
Voice
'VP
I
CAUSE
NPL
XP
\
offen
'V
S
BECOME
The cmcial assumption of this analysis is that there are different adjunction positions for an adverbial like wieder in such a syntactic structure, within and without the subtree that corresponds to the complex verb. If the syntactic position
of the adverbial determines its semantic scope, 9 this makes possible a syntactic
reconstruction of the different scope positions of the adverbial: Wieder may be
adjoined to the XP, which gives it narrow scope w.r.t. BECOME. The relevant
fragment of the syntax tree would in that case be (24):
(24)
VP
wieder
XP
\V
XP
BECOME
Alternatively, wieder may be adj oined to VoiceP or a higher phrase, which means
that the whole verb is in its scope.
Formally: A constituent A has scope over a constituent if its node c-commands , the node of
but not vice versa. The relation of c-command is defined in various ways, for ( 23 ) the following
definition suffices: c-commands iff the next branching node above dominates , and does
not dominate or vice versa.
159
This analysis generalizes straightforwardly to the case of aftermath modification by durative adverbials. The next question would be whether it can be
extended to agentive nouns and German prefix-verb nominalisations. For them,
one would have to assume suitable syntactic decompositions; what I have in
mind here is sketched in (25) for beautiful dancer.
(25)
VP
AP
/ \
,
r-r.
beautiful
/
\
I
-er
'
*I
dance
|
Here the idea is that agentive nouns are decomposed in the syntax; the affix -er
is the head of a nominal constituent and subcategorises for a VP. Within this
VP, modification may apply in the usual fashion as adjunction ( simplified here),
which semantically allows reference to the main eventuality argument of the VP,
because this argument is open in the VP semantics. In addition, the affix -er
requires a host, in this case, it forces head-to-head movement of the verbal head,
which is incorporated with the affix.
However, I will not pursue such an approach, because of a problem which
arises already for the syntactic decomposition approach to change-of-state verbs,
e.g., in sentences like (26):
(26) da Max alle Fenster ffnete
that Max all windows opened
'that Max opened all windows'
This sentence is true even in a context where half of the windows were open
before Max started to act, i.e., in a context where he opened only half of the
windows. What is important in this context is that at the end of Max's action, all
windows were open simultaneously. This intuition can be captured by assuming
that (26) has a reading that Max brought it about that all windows were open.
Since the other, dominant interpretation of (26) is that for all windows, Max
brought it about that they were open, the consequence is that the sentence is
ambiguous in that the scope of the universal quantifier and CAUSE/BECOME
is not yet fixed.
But according to standard wisdom of Generative Grammar, subject and object
NPs must move out of their base positions in VoiceP for reasons of case. I.e., in
any case they no longer have a position in the syntax tree that is c-commanded by
anything in the verb. But since syntactic position determines scope in Generative
Grammar, it looks as if there is a major conundrum there if one tries to represent
sentences where an argument NP of a verb has narrow scope w.r.t. operators
within the verb semantics.
160
Markus Egg
In a similar fashion, the problem resurfaces for sentences like (22). As pointed
out by Blutner and Jger (2003 ), this sentence has a restitutive reading that
Max brought it about that all windows were open, a state or affairs that is presupposed to have obtained before. The scope relations are CAUSE/BECOME
< again' < V, where '<" means 'outscopes'. I.e., once again, the universal
quantifier is outscoped by CAUSE and BECOME.
Recently von Stechow ( 2003 ) accepts the restitutive reading for examples that
are like (26 ) except that there is an indefinite object NP:
(27) da Max wieder ein Fenster ffnete
that Max again a window opened
'that Max opened a window again'
He explains this reading by assuming a structual accusative position within the

VoiceP, in the scope of CAUSE and BECOME. This would explain the narrow
scope of the object NP. However, if wide scope of the object NP is explained
in terms of the Mapping Hypothesis, viz., by assuming that NPs with strong
quantifiers must scramble out of the VoiceP, then the new explanation covers
neither the restitutive reading of (22) nor the second reading of (26). In either
case, a strong quantifier is in the scope of a verbal operator.
3.3 German prefix-verb nominalisations
The case of German prefix-verb nominalisations like Losgerenne is analysed in
Mller (2003) in an HPSG framework (Head-driven Phrase Structure Grammar;
Pollard and Sag 1994). This analysis regards separable prefixes, among them
los-, as subcategorised modifiers.
Two features of HPSG are crucial for his analysis: Affixation does not affect
subcategorisation, and a modifier determines the semantics of the constituent
that is built from itself and the expression it modifies ( the so-called head-adjunct
phrase).
Mller assumes that a lexical rule maps an ordinary verb stem like run \ ( suffixes are used for expository reasons) onto a stem run2, which subcategorises for
a separable prefix as a modifier. The prefix semantics becomes the semantics of
the resulting stem run2 It specifies how the semantic contributions of the prefix
and the stem run \ are combined into the semantics of the stem ruri2 Thus, the
semantics of the stem run2 can be paraphrased as 'prefix semantics ( whatever
that may be) applied to the semantics of run\ .
The next step is nominalisation of rim2 by circumfixing Ge... e, which yields
Gerenne. This noun preserves the subcategorisation for the prefix. The semantics of Gerenne can thus be paraphrased as 'ITER applied to run2 semantics'
(i.e., 'ITER applied to the prefix semantics (whatever that may be) applied to
run semantics'). Finally, the prefix semantics is determined once los- fills in
the position of the subcategorised prefix.
161
One might think of generalising this solution to the other phenomena. E.g.,
for beautiful dancer, one would have to assume a lexical rule that maps verb
stemsi to verb stems2 subcategorising for a modifier. The semantics for dance2
would then be something like 'modifier semantics applied to the semantics of
dance\\ Affixation of dance 2 by the agentive -er would still carry this subcategorisation, its semantics would be 'person who usually participates as the agent
in eventualities characterised by modifier semantics applied to the semantics of
dance\ \ which, after saturating the subcategorisation by beautifid, returns the
desired interpretation of the preferred interpretation of beautifid dancer. However, I envisage two drawbacks for this kind of strategy: It would introduce
massive ambiguity into the lexicon and the distinction between subcategorisation and modification would be much less clear-cut as usually assumed.
In sum, the discussion of this section reveals a number of problems, and, what
is more, suggests that there is no straightforward way of generalising one of the
approaches to all of the phenomena that are the topic of this paper. In the next
sections I will show that it is indeed possible to bring out the common ground
between these phenomena in terms of an analysis that regards them as specific
mismatches at the syntax-semantics and the morphology-semantics interface.
4 The formalism
In this section, I introduce the representation formalism which is used for my
own analysis of the presented cases of reference to embedded eventuality arguments. These cases can be described in a suitable underspecification formalism, e.g., Underspecified DRT (UDRT; Reyle 1993), Minimal Recursion Semantics (MRS; Copestake et al. 2005), or Constraint Language for Lambda
Structures (CLLS; Egg et al. 2001). These formalisms allow an adequate representation of structual ambiguity and, what is more, they make possible the
definition of syntax-semantics (and morphology-semantics) interfaces that are
flexible enough to handle reference to embedded eventuality arguments.
Expressions of such a formalism are constraints that describe a set of semantic representations, one for each reading of a structurally ambiguous expression.
In this paper, where I will use CLLS in an abbreviated form, the described semantic representations are -terms. The underspecification of such a constraint
is due to the fact that it deliberately abstracts away from the differences between
the semantic representations that it describes.
The representations that are described by (or compatible with) a constraint
are called its solutions. While solutions in general might comprise much more
material than is explicitly mentioned in a constraint, there is a subset of solutions
that consists only of the material introduced explicitly in the constraint. These
solutions are called constructive', for the purpose of the present paper, only constructive solutions are relevant.
162
Markus Egg
Then constraints can be regarded as a kind of jigsaw puzzle: Parts of a semantic representation are given together with some instructions on how to put them
together. Any possible way of putting them together yields one of the solutions
of the constraint. (More formally, such constraints are a partial order on a set of
fragments of semantic representations.)
The constraint (28) for the semantic representation of (1) will be used in the
remainder of this section as an introduction to the simplified fonn of CLLS employed in this paper. 10
(28)
[[NPS]]: Xy.GEN[e,x] (x i n e Ay = x, ( < ? ) )
il) / beautiful'(Y)
dance'()
(28) comprises the three ingredients out of which the simplified CLLS expressions are constructed, viz., fragments of -tenns, not yet known parts of these
fragments, indicated by 'holes' ( ) , and dominance relations (depicted by dotted lines) that relate fragments to holes. When a fragment is dominated by a
hole it is an (im-)proper part of whatever the hole stands for. Dominance relations model scope. Structures like (28) are called dominance diamonds. (They
are characteristic for quantifier scope ambiguities, too, see section 5 below.)
In (28), there is only a hole on top, which means that we do not know which
-tenn the structure as a whole stands for. But the dominance relations between
this hole and the fragments on the right (the semantic contribution of the modifier) and the left (the meaning of the affix -er) indicate the immediate parts of
this -tenn. Finally, the verb stem semantics shows up in the bottom fragment,
which is dominated by holes in the right and left fragments, i.e., it is in the scope
of both affix and modifier.
Resolving the ambiguity in constraints is modelled as adding information
monotonically, in particular, by strengthening dominance relations between holes
and fragments to identity. For (28), there are two choices: Identifying the affix
fragment with the top hole, the modifier fragment, with the hole in the affix fragment, and the verb stem fragment, with the hole in the modifier fragment yields
(7b). The other choice (starting this procedine with the modifier fragment) retails the -tenn (7c) for the other reading of ( 1 ).
10
In such constraints, ' [[c]] ' indicates the main fragment of a constituent C and ' [[Cs]] the secondary
fragment of C. ' [[c]] :F' expresses that the main fragment of C is defined as fragment F. The notions
of main and secondary fragment are explained at the beginning of section 5.
163
5 The analysis
The proposed analysis puts down reference to embedded eventualities to two
factors: First, a constituent Q , which is combined syntactically or morphologically with a second constituent C2, may have scope only over C'2, a part of Cz,
second, eventualities that were embedded in the semantics of Cz as a whole show
up as open arguments in the semantics of C'2.
E.g., in (1) the adjective may have scope over the verb stem only (it is thus
outscoped by the affix). In the semantics of the verb stem, the eventuality of
dancing is an open argument. The preferred reading of (1) emerges then directly from applying the affix meaning 'person characterised by X ' (where X is
the scope domain of the affix) to the stem meaning (after modification by the
adjective).
The interface derives this kind of semantic representation from surface-oriented syntactic/morphological structures. Semantic reference of one constituent to
only a part of another one is described in tenns ofpotential scope ambiguity. The
semantics of a (lexical or complex) constituent C is divided into a secondary part
(with the embedded eventualities), which must be in the scope of a modifier/affix
of C, and a main part, which need not. The interface rules can handle both kinds
of fragments. Consequently, scope between modifier/affix of C and C's mam
part is deliberately left open, the resulting semantic representations are therefore
ambiguous.
E.g., for dancer, the meaning of the stem is the secondary, and the meaning
of the affix, the main part of its semantic contribution. As shown in the semantic representation of beautiful dancer (28), the scope of affix and adjective is
open, but both outscope the stem. Either of them may have widest scope, which
describes the fact that (1) has the two readings (7b) and (7c).
This strategy returns expressions of the semantic formalism that are already
familiar from imderspecified representations of scope ambiguity: They look just
like the semantic representations of sentences with two scopally ambiguous
quantifying NPs. Here the bottom fragment of the dominance diamond comprises the verb that syntactically subcategorises for the scope-bearing NPs. The
two NPs contribute the two scopally ambiguous fragments of the diamond. See
Egg et al. (2001) or Reyle (1993) for details. 11
5.1 The syntax-semantics interface
The syntactic structure on which the syntax-semantics interface is based is very
surface-oriented. For (1), the underlying syntactic structure is (29):
11
The kind of elaborated syntax-semantics interface that is needed to derive the semantic representations for the phenomena which are analysed in this paper is also required to derive these
representations for quantifier scope ambiguities. I.e., the proposed treatment of these phenomena
does not introduce additional complexity into the syntax-semantics interface. See Egg (2005) for
a detailed exposition of an interface that can handle scope ambiguity.
164
Markus Egg
(29)
NP
AP
dancer
beautiful
It is now the task of the interface to derive the constraint (28) for the semantics
of ( 1 ) from this syntactic structure. The derivation of such constraints uses structured lexical entries like the one for dancer. Here the semantic contribution of
the verb stem is set off in a fragment [[Ns]] of its own. In this fragment, the eventuality argument (for the dancing) that is bound in the semantics as a whole is
free. The affix meaning constitutes the [[N]] fragment: 12
( 3 ) [[N]] : Xy.GEN[<?,A-](A- in eAy = x, (<?))
[[Ns]] : lance'(A)
The general strategy of semantic construction can now be summarised in three
points. First, every constituent inherits the constraints Con and Com of its immediate constituents C\ and Cz- Second, the interface rules specify for each
constituent C how Con and Com are combined into a new constraint Con for
C. Rules may themselves contribute to this Con. Third, the mam and the secondary fragment of a constituent are visible to the process of semantic construction (e.g., as the values of some auxiliary semantic features), they are addressed
in the interface rules, which also determine these two fragments for each new
constituent. 13
Formally, the interface rules specify for a given syntactic construction (on the
left) its semantic impact (on the right), in particular, the main and the secondary
fragment of the emerging constituent. Consider e.g. rule (31), which merely says
that that nonbranching X constituents inherit their fragments from their heads.
Recall that ' [[c]] ' stands for the main and ' [[Cs]] ', for the secondary fragment of a
constituent C; ' [[c]] :F' indicates that the main fragment of C is defined as F:
(SSS)
(3D
12
13
[]
X :X
iXsJJ : HXs.
This twopartite semantic structure can be derived by a rule of the morphology-semantics interface
which combines the stem and the affix semantics. This rule is described as (42) in section 6 below.
This kind of semantic construction is familiar e.g. from MRS Copestake et al. 2005.
165
Next comes the modification interface rule (32), which characterises modification as a process that affects the secondary, but not the main fragment of the
modified expression. The emerging constituent Xi inherits its mam fragment
pi]] from the modified expression. Its secondary fragment pi S ]] is defined as
the modifier fragment [[Mod]] applied to a hole that dominates the secondary fragment p2s]] of the modified expression.
This introduces scope ambiguity between [[Mod]] and pi]] and yields the bottom half of a dominance diamond: Either of these fragments dominates p2s]],
the first due to (32), the second, due to the fact that pi]] (= p 2 ]]) and p 2 S ]]
originate as fragments of the same constituent. Equating the modifier fragments
( [[Mod]] : [[Mods]]) is not necessary, but facilitates reading.
Pis]] : [[Mod]] ( 0 )
(32)
b^ModXa]
(SSS)
:
p2s]]
[[Mod]]: [[Mods]]
P i ] ] : p 2 ]]
The third interface rule correlates the projection of X constituents to XPs on the
syntactic side and the construction of the upper half of the dominance diamond
on the semantic side. The top fragment of XP (also its mam fragment) is a hole
that dominates both fragments of the X constituent:
(33)
[]
(S S)
[[XP]] :
..'
'..
[['XPs]] : [[]]
' S;JJ
I will now show the semantic construction for the three instances of reference to
embedded eventualities that were introduced above.
5.2 Agentive nouns
The derivation of the semantic representation of beautiful dancer on the basis of
(29) makes use of the lexical entries for dancer (30) above and for beautiful (34)
and the rules (31)-(33) to derive the diamond (28).
(34) [[A]], [[AS]]:
.() beautiful' (x)
According to the lexical entry for beautiful, both fragments are identical. 14 Due
to rule (31 ) these fragments are inherited by beautiful as constituent. Then
(33) applies, which returns the following constraint for the AP beautiful:
14
This is simplified in that the modification of modifiers like beautiful is not taken into consideration
here. See Egg (2004) for an account of this phenomenon that is compatible with the analyses
advocated in this paper.
166
(35)
Markus Egg
[[AP]] : 0
[[APS]] : ).() A beautiful'()
The interface rule for modification structures (32) takes (30) and (35) and combines them into the bottom half of a diamond (36) for the meaning of the
constituent beautiful dancer. (36) is then the input to (33), which returns the
complete dominance diamond (28) for beautiful dancer as an NP.
( 3 6 ) p]] : A,y.GEN[e,x] (x in e Ay = x, > ) )
p s ]] : .() beautiful' ()
dance'()
The fact that this resulting dominance diamond has two solutions (viz., (7b) and
(7c)) models the ambiguity of beautiful dancer.
In the same fashion it is possible to construct the semantic representations of
the other examples. The aga/-sentences, for instance, use lexical entries for
change-of-state verbs where the aftermath predicate is set off in a secondary
fragment. Consider e.g. the lexical entry for the transitive open:
( 3 7 ) [[V]] : XyhcXe. CAUSE(x, BECOME [II) (e)
[[VS]]: be-open'(y)
The semantics of again is (38):
(38) [[Adv]] , [[Advs]]:
XPXe.again'(P)(e)
The underlying syntactic structure of (2) is one in which again modifies the rest
of the sentence. If we ignore tense and sentence mood, the resulting semantic
representation of (2) is (39). The semantics of the window is once again abbreviated as the constant W (of type <?):
(39)
[[VP]] . g
[[VPs]]: Xe.CAUSE(max',BECOME( ))(<>)
^.again'('H)(e)
be-open'(W)
167
The semantic representations ( 1 Oa) and ( 1 Ob) for the restitutive and the repetitive
readings of (2) are the two solutions of this dominance diamond.
This immediately begs the question of how to avoid overgeneration for examples where there is no ambiguity. I will illustrate this with the semantic representation of (3): In this representation the semantic contribution of the durative
adverbial is fonnalised as 2hrs', a functor from unbounded properties of eventualities to a bounded property of predicates, roughly, being coextensive to a
P-eventuality and having a duration of two hours.
(40)
[VP]
[[VPS ]]: ^. CAUSE (max', BECOME ( El ) ) ( e )
Xe.2hrs''('S)(e)
be-open'(W)
While this looks at a first glance just like the semantic representation of (2),
which has two readings, the scope ambiguity in (40) is only potential in that
it is immediately resolved by the properties of the involved fragments. The
crucial point here is the aspectual restriction of the adverbial: The hole in the
adverbial fragment may be filled by the unbounded bottom fragment, but not
by the bounded left fragment (all causative verbs are bounded). I.e., the sole
solution for (40) is (41):
(41)
^.CAUSEimax'ÊCOMEi^Mhrs'ibe-open'iWlKe'DKe)
In prose, (41) is the set of eventualities in which Max causes a change of state
whose result state (the window being open) lasts for two hours.
This example shows that the proposed analysis is not in danger of unwanted
overgeneration even though it models reference to embedded eventualities in
tenns of scopai ambiguity: This ambiguity is only potential and can be resolved
directly by the involved fragments if the expression is not ambiguous.
5.4 Gennan prefix-verb nominalisations
In the final part of this section I will show how to extend the proposed analysis from issues at the syntax-semantics interface to a related phenomenon at
the morphology-semantics interface. This extension captures the initial observation that the semantic effect of prefixation in Losgerenne resembles the effect of
modification in examples like (1) and (2). This suggests handling affixation in a
fashion close to the interface rule (32).
The morphology-semantics interface rule for the semantic representation of
affixed nouns is given in (42):
168
(42)
Markus Egg
[ ES ff]
(m
3P h)
[[X]] : [[AfflKXjTO)
;
Ps]]:
M t f )
This rule constracts a structured semantic representation for affixed expressions

in which the main fragment of the affix dominates the secondary fragment of
its base. If the base comprises a main eventuality argument it remains free in
this secondary fragment. Scope between the main fragments of base and affix is
in principle open; it is fixed, though, whenever main and secondary fragments
coincide, which is often the case for roots, as shown e.g., in the lexical entry for
the verbal root renn- 'run':
(43) M ,
[[VS]]:
A-Xe.run'(j)(e)
(42) does not determine the categories of the base ('Bs') and of the resulting
expression ('X'). Furthermore, it is meant as a rule of 'immediate dominance'
and not of 'linear precedence', i.e., it leaves open the ordering of affix and base.
This kind of information must be supplied by the affixes themselves, it is not
part of the interface rule. E.g., Ge... e and -er map verbal bases to nouns, while
los- maps nominal or verbal bases to expressions of the same category.
The semantic representation of dancer (44) [= (30)] is constructed by (42) out
of the affix meaning (6a) and the verb stem meaning, whose main and secondary
fragment consist of dance' (a relation between eventualities and individuals):
( 4 4 ) [KT] : - G E N M ( in e = .. Q(e) )
[[Ns]]: dance'(.)
However, (32) and (42) are not completely parallel: (42) determines the main
fragment of the resulting word as the main fragment of the affix and the resulting
word's secondary fragment as the secondary fragment of the base.
Another difference between (32) and (42) is that the latter anticipates the fact
that affixation may involve argument binding-. Any individual arguments of the
stem are -abstracted in the main fragment, which allows binding by the affix.
The affix itself determines which arguments are bound.
First, -er binds everything but the agentive argument as spelt out in (6a). In
contrast, Ge... e binds every individual argument of its base. Its semantics (45)
maps M-ary relations between an eventuality and 1 individuals to the property of being an iteration of P-eventualities (with possibly different participants):
(45) .('3..()(') )(e)
Finally, semantically transparent prefixes like los- inherit all individual arguments from their bases. The semantics of ios- (46) [= (1 lb)] maps -ary relations
169
onto the n-ary relation which involves the same individual arguments and the
begin of a P-eventuality.
(46) XPXxXe.BECOME(P(x))(e)
Based on rule (42) and the semantic representations for the involved morphemes
semantic construction ior Losgerenne is straightforward. (43) and (45) are combined into the semantic representation (47) for Gerenne:
( 4 7 ) [[N]] : Xe.ITER(Xe'3x. 0 (e'))(e)
[[NS]] :
fe.run'(x)(e)
One more application of rule (42) returns the semantics of Losgerenne, the input
to the rule are this time (46) and (47):
(48)[[N]] ; \((El) (e)
^.ITER(Ie'3x. (?'))(?)
''[[WsJiV.run'ix)^")
In prose, the second affixation augments the constraint with an additional fragment for the prefix semantics ( on the left) that dominates the verb stem semantics
(at the bottom) but not the fragment for the circumfix semantics ( on the right).
I.e., in one of the two solutions of (48), the prefix ends up in the scope of the
circumfix.
This is in fact the only solution of (48), because the other potential solution
( with wide scope of BECOME over ITER) is ruled out due to a selection restriction of los-: My intuition on the semantics of (productive and semantically
transparent) los- is that it requires its argument to refer to an eventuality that
involves a maximal axis in the sense of Lang (1990).
Since movement verbs characteristically refer to such an eventuality, losattaches easily to them, e.g., in loslaufen 'start walking', or losrollen 'start
rolling (intransitive)'. Even weather verbs that involve movement are thus acceptable with los-, e.g., loshageln 'start hailing', losregnen 'start raining'. In
contrast, stative verbs have typically no such maximal axis, hence, derivations
like *losglcinzen 'start gleaming', *losstinken 'start stinking', *loswiinschen 'start
wishing' and so on are ruled out.
Now since an iteration of running eventualities as opposed to these eventualities themselves does not involve such a maximal axis, the sole resolution of
(48) is the one where the right fragment receives widest scope, which yields the
desired semantic representation ( l i e ) for Losgerenne.
170
Markus Egg
This concludes the analyses of the cases of reference to embedded eventualities. What I have tried to show in this section is that all these cases can be
treated in a similar fashion, which brings out the common ground between them
that was expounded in section 2.
6 Conclusion and outlook

In this paper I have described cases where one ( syntactic or morphological) constituent C\ may refer to eventualities embedded in the semantics of another constituent
These cases were analysed as giving rise to (potential) scope ambiguities in their semantics. The semantic contribution of C1 takes scope over
only a part of the semantics of C2, with the rest of C2 's semantics being scopally
ambiguous to the semantics of C\. It is then up to the type-theoretic and/or aspectual properties of the involved semantic representations whether this potential
scope ambiguity actually introduces ambiguity (because either potential scope
ordering is feasible) or whether one of these orderings is immediately ruled out.
The next step for the proposed analysis will be to find further instances of the
phenomenon described in this paper. E.g., the semantic representation (47 ) for
Gerenne is also adequate as the input for the semantic construction of (49 ) by
(32):
(49) schnelles Geremi -e
fast
iter nomi m n iter 1101112
'iteration of fast runnings'
The preferred reading of (49 ) refers to iterations of fast runnings, i.e., the iteration itself need not be fast. This interpretation, where the modifier pertains only
to the stem of its modified expression, can be adequately captured by pertaining the modifier schnell to the embedded fragment of the semantic of Gerenne
( which comprises the stem semantics).
It turns out that the proposed analysis generalises to a number of seemingly
unrelated syntax-semantics puzzles which all can be described in terms of scope
ambiguities. For instance, Egg (2004) extends the analysis to the problem of
modification of modifiers noted by Kasper (to appear ) for expressions like (50):
(50) apparently honest politician
The scope of the modifier of the modifier honest must be prevented f r o m taking
scope over the expression modified by honest: (50) refers to persons whose
honesty but not whose being a politician is doubtful.
I also envisage a wide range of applicability for the analysis in agglutinating
languages like Turkish, where word formation by affixation is extremely productive. E.g., the analysis of (1) generalises directly to the related case of (51):
171
(51) gen at
-li
young horse' furnished.with
'someone with a young horse' or 'a young person with a horse' 15
By representing the meaning of ath as (52)a (the property of having a horse')
one can render the first reading of (51) as (52b):
(52) (a)
(b)
. h orse' (y) with7 (, y )

Xx3y.horse'(y) young' (y) with' (, y)
(52b) can be derived from (52a) by pertaining the modifier gen 'young' only
to the stem at 'horse', whose semantic contribution is underlined in (52a). This
example shows that the proposed analysis promises to be applicable fruitfully to
a wide range of related phenomena. It will be the task of further research to test
and refine the analysis in this way.
References
Blutner, Reinhard (2000): Some aspects of optimality innatural language interpretation.
Journal of Semantics, 17: 189-216.
Blutner, Reinhard and Gerhard Jger (2003): Competition and interpretation: the Geman
adverb wieder ("again"). In Ewald Lang, Claudia Maienbom, and Cathrine FabriciusHansen, (eds.), Modifying adjuncts, pp. 393^116. Mouton de Gruyter, Berlin.
Chomsky, Noam (1995): The minimalist program. MIT Press, Cambridge.
Copestake, Ann, Daniel Flickinger, Carl Pollard, and Ivan Sag (2005): Minimal Recursion Semantics. An introduction. Research on Language and Computation 3, pp.
281-332
Davidson, Donald (1967): The logical forni of action sentences. In Nicholas Reseller,
(ed.), The logic of decision and action, pp. 81-95. Pittsburgh University Press, Pittsburgh.
Diesing, Molly (1992): Indefinites. MIT Press, Cambridge.
Dlling, Johannes (1998): Modifikation von Resultatszustnden mid lexikaliscli-semantischen Reprsentationen. In Petra Ludewig and Bart Geurts, (eds.), Lexikalische
Semantik aus kognitiver Sicht, pp. 173-206. Narr, Tbingen.
Dowty, David (1979): Word meaning and Montague grammar. Reidei, Dordrecht.
Egg, Markus (2004): Mismatches at the syntax-semantics interface. In Stefan Mller,
(ed.), Proceedings of the 11th International Conference on Head-Driven Phrase Structure Grammar, pp. 119-139. CSLI Publications, Stanford.
Egg, Markus (2005): Flexible semantic construction: the case of reinterpretation. CSLI
Publications, Stanford.
Egg, Markus, Alexander Koller, and Joachim Nieliren (2001): The constraint language
15
The first reading is less preferred out of context but its acceptance rises if ( 51 ) is used as a modifier
as in gen ath bir adam 'man with a young horse' (literally, 'young horse-with a man' ).
172
Markus Egg
for lambda-structure s. Journal of Logic, Language, and Information, 10: 457-485.

Fabricius-Hansen, Catherine (1975): Transformative, intransformative und kursive Verben. Niemeyer, Tbingen.
Kasper, Robert (to appear): The semantics of recursive modification. Journal of Linguistics.
Kratzer, Angelika ( 1996): Severing the external argument from its verb. In Johan Rooryck
and Laurie Zaring, (eds.), Phrase structure and the lexicon, pp. 109-137. Kluwer,
Dordrecht.
Krifka, Manfred (1992): Thematic roles as links between nominal reference and temporal
constitution. In Ivan Sag and Anna Sabolcsi, (eds.), Lexical matters, pp. 29-53.
CSLI, Stanford.
Krifka, Manfred, Francis Pelletier, Gergory Carlson, Alice ter Meulen, Gennaro Chierchia, and Godehard Link (1995): Genericity: an introduction. In Gregory Carlson
and Francis Pelletier, (eds.), The generic book, pp. 1-124. University of Chicago
Press, Chicago.
Lang, Ewald (1990): Primary perceptual space and inherent proportion schema. Journal
of Semantics, 7: 121-143.
Larson, Richard (1998): Events and modification in nominis. In Devon Strolovitch and
Aaron Lawson, (eds.), Proceedings from SALT VIII, pp. 145-168. Cornell University, Ithaca.
Ldeling, Anke (2001): On particle verbs and similar constructions in Geman. CSLI
Publications, Stanford.
Mller, Stefan (2003): The morphology of Geman particle verbs: solving the bracketing
paradox. Journal of Linguistics, 39: 275-325.
Pollard, Carl and Ivan Sag (1994): Head-driven Phrase Structure Grammar. CSLI and
University of Chicago Press.
Reyle, Uwe (1993): Dealing with ambiguities by miderspecification: construction, representation, and deduction. Journal of Semantics, 10: 123-179.
Sliibatani, Masayoshi (1976): The grammar of causative constructions: A conspectus.
In Masayoshi Sliibatani, (ed.), The grammar of causative constructions, Syntax and
Semantics 6, pp. 1-42. Academic Press, New York.
von Stechow, Arnim (1996): The different readings of 'Wieder': A structural account.
Journal of Semantics, 13: 87-138.
von Stechow, Arnim (2003): How are results represented and modified? Remarks on
Jger & Blutner's anti-decomposition. In Ewald Lang, Claudia Maienbom, and
Catlirine Fabricius-Hansen, (eds.), Modifying adjuncts, pp. 417-454. Mouton de
Gruyter, Berlin.
Zimmermann, Thomas (1993): Zu Risiken mid Nebenwirkungen von Bedeutungspostulaten. Linguistische Berichte, 146: 263-282.
SECTION III
Event Structure and Situation Aspect
Susan Rothstein (Bar-Ilan University)
Two puzzles for a theory of lexical aspect:

Semelfactives and degree achievements
1 The puzzle about Vendler classes
Vendler (1957/1967), followed by Dowty (1979), distinguished four classes of
verbs, based on empirical evidence such as different patterns of cooccurence
with various kinds of temporal modifiers. The classification is semantic, since
the different verb classes are distinguished by the different properties of the
events in their denotation. The four classes that Vendler distinguishes are
states, such as love, know and believe, activities, such as run, walk and swim\
achievements, such as arrive, die, notice, realise, and reach; and accomplishments, such as read (a book), and build (the house). I discuss the linguistic
properties of these classes in detail in Rothstein (2004); the most important
facts can be summarised as follows. First, states and achievements do not naturally occur in the progressive, while activities and achievements do, as illustrated in (1):
(1)
(a)
#John is believing in the devil.
(b)
#John is noticing the picture.
(c)
Mary is running.
(d)
Mary is building a house.
Second, achievements and accomplishments occur naturally with telic modifiers, while states and activities do not, as illustrated in (2).
(2)
(a)
John believed in the devil #in half an hour/for several years,
(b)
Mary ran #in half an hour/for half an hour.
The first part of this paper is a revision of parts of chapter 8 of Rothstein 2004. The material in
parts 6-8 is entirely new. The material has been presented a number of times: at the Workshop
on Events in Leipzig, at LATL 20, and at colloquia at ZAS, Nanzan University, the University
of Tokyo , SISSA, Tel Aviv University, and the Hebrew University of Jerusalem. I thank audiences at all these events for their comments and discussions. As usual, discussions with Fred
Landman were particularly fruitful. This research was partially supported by the Israel Science
Foundation Grant #951/03.
176
Susan Rothstein
(c)
John noticed the picture in half an hour/#for half an hour.
(d)
Mary built a house in two weeks /#for two weeks.
Third, accomplishments, but not activities, induce the imperfective paradox: in

other words, activities occurring in the progressive consistently allow the entailment pattern illustrated in (3a), while accomplishments don't, as illustrated
in (3b):
(3)
(a)
John was running ENTAILS John ran.
(b)
John was building a house DOES NOT ENTAIL John built a house.
In general, verbs which occur with for a time induce the imperfective paradox,
and those which do not occur with for a time do not. (For a discussion of some
of the apparent counterexamples to all these generalisations see Rothstein
2004.)
The organisation of verb classes which these tests gives us is summed up in
(4), with the features [ occur in the progressive] and [ co-occur with telic
modifiers] allowing us to characterise uniquely each of the four classes.
(4)
[ occur in the progressive]
[ occur with telic modifiers]
States
Activities
Achievements
Accomplishments
However, there are several questions which the above table raises. First, what
about verbs which apparently do not fit into any of these classes? There are
two very obvious classes which do not fit into the classification. The first is the
class of semelfactive verbs, such as kick, knock, jump, skip, and flap (its
wings), which are homonymous with activity verbs but which denote 'single
action events', as when knock denotes a set of events in which one's hand or
an object in one's hand comes into contact with a hard surface only once. As
we will see below, they apparently form a fifth kind of event and thus a fifth
class of verbal predicates, and as such cause a problem for a categorisation
based on two features. The second set of problematic verbs are so called degree achievements, such as cool, warm and redden which appear to fit into too
many classes, since they seem to behave like achievements and like activities,
and sometimes like accomplishments as well, as we will see below.
There is a second question regarding the table in (4): what is the special significance of occurring in the progressive and occurring with telic modifiers
which means that these properties should characterise the verb classes? And a
third issue is, what is the relationship between the table in (4) and Krifka's
Two puzzles for a theory of lexical aspect
177
characterisations of verb meanings in terms of whether or not they are cumulative? In this paper, I shall discuss all of these issues. We will begin with what
is special about the progressive and cooccurence with telic modifiers as tests
for verb classes, and go on to discuss the relation between these tests and
Krifka's tests for cumulativity. We will then, in section 4 and 5, use the answers to these questions to approach the issue of semelfactives and their relation to the other verb classes, and in section 6 we will discuss degree achievements.
2 A theory of lexical aspect (Rothstein 2004)

We start with the question of what the features in (4) mean. These features
reflect fundamental characteristics of events. The property of appearing or not
appearing in the progressive reflects whether a predicate denotes temporally
extended events or (near) instantaneous events, while the property of appearing
with telic modifiers, correlates in (4) with whether or not a predicate denotes
an event of change. We will look at each of these properties in some more
detail.
The property [+ occurring with telic modifiers], which characterises accomplishments and achievements, correlates in our table with the property of denoting an event of change. (Note that this is not the definition of telicity, but a
correlation in the table in (4)). I will not develop a theory of change here, but
will make do with the following working definition:
(5)
"A change from to is an event whose minimal initial part is the last instant i at which
holds and whose minimal final part is the first instant i' at which holds." (Dowty 1979,
Rothstein 2004).
It follows from (5) there are two possible kinds of changes: extended changes
where i and V are not adjacent but separated by an interval, and non-extended
changes where i and i' are adjacent. (An interval is a concave set of instants, or
a set of temporally adjacent instants without gaps.) We characterise the nonextended changes as changes from to .a: on the assumption that either or
id is true of every instant, a change from to .a must be instantaneous. We
characterise extended changes as changes from to where entails -.
Extended changes are accomplishments and non-extended changes are
achievements. The two kinds of non-change events are in the denotations of
state and activity predicates.
The [+occurrence with progressive] characterises activities and accomplishments and distinguishes them from states and achievements. The semantics of the progressive, no matter whose semantic theory one adopts, makes a
sentence such as A is V-ing true if there is a possibly incomplete event in the
denotation of V going on. Landman (1992) argues that the progressive makes
178
Susan Rothstein
reference not just to parts of events, but to stages, where e is a stage of e' if e' is
a development of e. It follows from his theory of stages (developed in Landman 2004) that if e is a stage of e', e and e' must be qualitatively different. We
can use these ideas to explain why the progressive does not occur naturally
with either states or achievements. States are totally homogeneous down to
instants. This means two things. First, the shortest states hold at instants, and
second, each instant or interval at which a state holds is identical in the relevant respects to every other instant or set of instants at which it holds. This
means that states can not have stages, and thus an operation such as the progressive which makes reference to stages cannot apply to states.
Achievements also do not occur with the progressive. Achievements are
events of instantaneous change from to - which consist of two instants, the
last instant i at which holds and the first instant i' at which .a holds, where
crucially there is no instant intervening between i and /' . While these events
are not strictly instantaneous, since they consist of two minimal instants, they
are also not extended since there is no interval intervening between their initial
instant and their final instant. They are thus not naturally divisible into stages,
and not appropriate inputs to the progressive construction. (I show in Rothstein
2004 that where achievements do seem to occur in the progressive, this is
because there has been a shift in the VP meaning from an achievement-type
event to an accomplishment-type event, where the lexical achievement denotes
the event which is the culmination of the accomplishment. For details see
Rothstein 2004 chapter 2.)
Thus, the inability to occur with the progressive reflects a property common
to both state and achievement events, namely that they are not inherently extended in time, and that they are therefore not naturally analysable into stages.
States may hold at instants or a concatenation of instants, while achievements
hold at concatenations of two adjacent instants which cannot be extended. In
contrast, both activities and accomplishment take time. Accomplishments
involve an incremental process (Dowty 1991, Krifka 1992, Rothstein 2004)
and, as Dowty (1979) argued extensively, activity predicates cannot be true of
instants but only of intervals defined by at least two non-adjacent instants.
Thus in order to see whether it is true that John ran, it is necessary to have
evidence as to what occurred at at least two (but possibly more) related but
non-adjacent instants and to compare the state that John was in at each instant,
and then to decide whether the relation between the properties of these two
instants is such that we can conclude that an event in RUN was going on. The
same is true if we evaluate whether Mary built a house is true. The fact that,
for an event e in P, different things are going on at different non-adjacent instants of e is evidence that denotes a set of events that can be analysed as
having stages, and thus the progressive naturally applies.
The table in (6) sums up this discussion:
179
(6)
Minimal events are extended
Event of change
States
Activities
Achievements
Accomplishments
The table expresses that event predicates vary as to whether the (minimal)
events in their denotations are (near-) instantaneous or extended, and as to
whether or not they are events of change. Achievements denote non-extended
events of change; accomplishments denote extended events of change; states
denote non-inherently extended events which do not involve change and
activities denote inherently extended events not involving change. Note that
following Rothstein 1999,1 assume that all verbs have their denotations in the
count domain.
3 Cumulativity
An obvious question to ask is what is the relation between the classification of
verb denotations in (6) and other characterisations of properties of verb meanings. It will be particularly fruitful for our discussion of semelfactives to examine the relation between the table in (6) and Krifka's (1992,1998) classification
of verb meanings in terms of whether they allow cumulative reference. Allowing cumulative reference is a property of states and activities, and therefore
distinguishes between the verb meanings marked [+ telic] and those marked [ telic] in the table in (6).
Cumulativity is a property taken from Link's 1983 discussion of the masscount distinction in the nominal domain. A predicate X is cumulative (allows
cumulative reference) iff:
(7)
Bx5y[X(x) X(y) -, <= y -,y VxVy[X(x) X(y) - X(xuy)]]

"X is cumulative if when two distinct elements and y are in X ,
the sum of and y are also in X."
(7) distinguishes between mass and plural predicates on the one hand and singular predicates on the other: two quantities of water put together gives an
entity in the denotation of water and two pluralities in the denotation of books
summed together give a plural entity in the denotation of books but the sum of
two entities in the denotation of book or dog do not give an entity in the same
set book or dog but in the plural set books and dogs.
Krifka argues that the distinction between cumulative and non-cumulative
predicates is relevant in the verbal domain as well, and that it distinguishes
roughly between predicates which we call atelic and those we call telic. (Krif-
180
Susan Rothstein
ka does not define telicity in terms of cumulativity, but since we are interested
in what cumulativity is, this should not bother us at the moment.) The intuition
is that two separate but temporally adjacent running events can be summed
into an event of running, while two separate, temporally adjacent events of
eating three apples or drinking a glass of milk cannot be summed into an event
in the denotation of eat three apples or drink a glass of milk. This is illustrated
in (8):
(8)
(a)
If John ran from 13.00 to 14.00 and he ran again from 14.00 to 15.00, then he also
ran from 13.00 to 15.00. (= cumulative)
(b)
If John ate (exactly) three apples between 13.00 to 14.00 and then he ate (exactly)
three apples between 14.00 and 15.00, then it is not the case that he ate exactly three
apples between 13.00 and 15.00.
The contrast between (8a) and (8b) is clear; however, a closer look shows that
cumulativity is not the right way to capture it. This is because the definition of
cumulativity makes reference to the effect of operation of summing entities
(singularities and pluralities), and the output of the summing operation is itself
a plurality. In the nominal domain the property of cumulativity distinguishes
between the effects of summing on predicates denoting unmeasured plurals
and mass elements on the one hand, and individual count elements on the other
hand. When the summing operation sums pluralities and non-singular entities,
it gives as output the same kind of plural and non-singular entities that were
the input, but where the summing operation applies to singular entities, it gives
pluralities as output. The distinction shows up clearly because nominis usually encode plurality morphologically, and thus it is clear that a sum of entities
in dog, will result in a plurality which cannot itself be an element of dog, but
only of dogs.
However, the verbal domain does not morphologically encode plurality.
VPs denote sets which include both singular and plural events (see e.g. Landman 2000); so sums of singulars and sums of plurals in the denotation of VP
ought still to be in the denotation of the same VP for both telic and non-telic
VPs. And indeed, they are:
- If John ate three apples and Mary ate three apples, then the plural event of
both of them eating three apples is in the denotation of eat three apples, illustrated in (9a).
- If Mary ate three apples and then ate three apples again, then the sum of those
two events is in the denotation of eat three apples, illustrated in (9b).
(9)
(a)
Who ate three apples? John and Mary ate three apples, (in all six apples were eaten)
(b)
Mary ate three apples twice. (In all six apples were eaten).
181
But this means that so-called telic VPs are cumulative too. If so, then what is
the basis of the distinction between run and eat three apples illustrated in (8a)?
The contrast in (8) expresses the fact that two activity events (or two state
events) can be summed and the result can be treated not just as a sum, or plurality, but as a new singular event, and that this cannot be done with two
events in build a house, eat three apples or arrive. Thus the contrast in (8)
stems from the following fact about language:
- If John ran from 1pm to 2 pm and then from 2pm to 3pm, then there is a
singular event in the denotation of John run which lasted from 1pm to 3pm.
- If John ate exactly three apples between 1pm and 2 pm and then again between 2pm to 3pm, then the sum of these events cannot be formed into, or
conceptualised as, a singular event in the denotation of eat three apples.
The formal property which captures the distinction between run and eat
three apples is what we might call S-cumulativity:
(10)
3e5e'[X(e) X(e') -ne e e' -.e' e e VeVe'[X(e) X(e') R(e,e') -> X ( s (eue'))]]
"A predicate X is S-cumulative if any two distinct instances of X related by the 'R' relation
can be summed, and the sum formed into a singular entity which is itself in the denotation
of X."
We assume that, as far as events are concerned, the R relation is temporal

adjacency, in other words, two events e and e' can be summed and the resulting
sum can be formed into a singular entity if no other event intervenes between
them. (Rothstein 2004, chapter 7, discusses S-cumulativity in the nominal
domain, where the R relation is not temporal adjacency.)
This property of being S-cumulative is what distinguishes states and activities on the one hand from achievements and accomplishments on the other.
Two adjacent events e and e' of loving y (or of running) can be conceptualised as a single event, e" which is itself in the denotation of love (or run). Thus
we can describe the set of activities and states as being closed under the operation of S-cumulativity. Two adjacent events of arriving or of eating an apple
cannot be reanalysed as a single event in the denotation of arrive or eat an
apple and therefore the predicates are not S-cumulative and accomplishments
and achievements denote sets which are not closed under S-cumulativity.
Assuming the correlation between telicity and lack of S-cumulativity is not
coincidental, then we should ask why S-cumulativity does not apply to telic
predicates i.e. predicates of change. Remember that we observed that telic
predicates are events of change. As Kamp (1979a,b) argues, two events in the
denotation of a predicate of change (with the same participants) cannot be
immediately temporally adjacent to each other, since a change from to .a
cannot be immediately followed by another event of change from to -
without an intermediate event of change from back to a. Since Scumulation requires the summed events to be temporally adjacent, it follows
182
Susan Rothstein
that predicates of change (i.e. those marked [+telic] in (6)) cannot be Scumulative.
4 S emelfactives
Semelfactives are verbs such as kick, knock, jump, skip, flap(its wings), wink,
which denote single actions, in the sense that knock(on the door), for example,
may be understood as denoting a single event in which an object is brought in
contact sharply with a door once. These events can be counted: (11a) asserts
that John brought an appropriate object in contact with the door three times
and (1 lb) that he left the ground by jumping three times.
(11)
(a)
(b)
John knocked three times,

John j umped three times.
When they occur in the progressive with a semelfactive reading they induce
the imperfective paradox. Each of the examples in (12) can be used to describe
a situation in which a single knock or a single kick was interrupted:
(12)
(a)
(b)
John was knocking hard when he saw me, so he turned it into a tap instead (and didn't
knock hard).
Bill was kicking him when he saw me, so he stopped midway (and didn't kick him).
The fact that these induce the imperfective paradox indicates that semelfactives are quantized (in the sense of Krifka 1992, 1998). They denote minimal
events such that if e is in the denotation of a semelfactive predicate no part of e
is also in the denotation of that predicate. They also occur with the telic temporal modifiers in a time. In a context of a pole vault or a slow motion film, (13)
is acceptable on the single event reading:
(13)
John jumped in three seconds.
However, all semelfactives are homonymous with activity predicates, and

these activity predicates occur with atelic temporal modifiers, and do not induce the imperfective paradox:
(14)
(a)
(b)
John knocked on the door (repeatedly) for several minutes/half an hour,

John was knocking on the door when I arrived. (So he had knocked.)
Importantly, as we shall see, while all semelfactive predicates have a homonymous activity reading, not all activities have a homonymous semelfactive
reading. Run, swim, and walk have only activity readings.
183
The question is where semelfactives fit into the schema in (6). Smith
(1991), who is probably most responsible for the consensus that semelfactives
constitute a real class of verbs, argues that they are really atelic achievements.
She phrases this by saying that semelfactives are dynamic, atelic, and instantaneous, while achievements are dynamic, telic, and instantaneous. Her intuition
is that semelfactives are, like achievements, single stage events, which, though
they take time/have duration, are conceptualised as instantaneous. They differ
from achievements which are also single stage events, since achievements are
events of change, while semelfactives do not bring about a change. While it
seems clear that Smith is right that achievements and semelfactives differ in
that the former is a predicate of change and the latter not, there are two problems with her account. The first is that it does not explain why semelfactives
occur with in a time (and other telic modifiers). The second is that while
achievements really are non-extended, consisting of two temporally adjacent
instants, semelfactives really take time and are temporally extended. An event
in the denotation of arrive has no internal structure; it consists of the last instant at which is not at location 1 and the first instant at which is at location
1, and there is no gap between these instants. But, semelfactives do have internal structure. Jump, flap a wing, kick and wink all have trajectories, in the
sense that for an event of knocking to occur, several things have to happen at
different non-adjacent instants: an object has to be brought sharply through
space to come in contact with a hard surface. For an event of winking to occur,
an eye has to close and then open again, and so on. If semelfactive events are
defined via trajectories, then they cannot be instantaneous, and, unlike
achievements, they must be [+temporally extended] in the table in (6). So
semelfactives, like activities and accomplishments, hold at minimal intervals
and not at instants. Like accomplishments they are quantized, they induce the
imperfective paradox and they occur with in a time, but unlike accomplishments, they do not denote events of change. They are unlike activities, which
occur with atelic modifiers, and which are not quantized, but they are homonymous with activities. So it seems appropriate to explore the relation between semelfactives and activities further.
As we already pointed out, semelfactives have a related activity reading,
which seems to be an iteration of the single event reading, so that jump can
denote either events of single-occurrences-of-leaving-the-ground or iterations
of these events. However, not all activities are related to semelfactives. The
activity run cannot be used as a semelfactive, and this results in a set of systematic differences between run type predicates and jump type predicates:
i. Counting adverbials can count either the single events or the iterations for
jump type predicates. With run, only extended events can be counted. Compare run with jump :
184
(15)
Susan Rothstein
(a)
Dafna jumped/skipped once/twice,
(b)
Dafna ran once/twice.
ii. When in a time is used as a modifier, it induces the semelfactive reading on

jump type predicates, and measures the time of a single jump. When the same
modifier is used with run, a contextually determined measure for the extended
event is required, and the modifier measures the time of the extended event:
(16)
(a)
Dafna jumped in two minutes,
(b)
Dafna ran in two minutes.
iii. Again and again can modify either the single event or the activity predicate
with jump. The semelfactive reading in (17a) can be paraphrased by (17b).
Since there is no semelfactive reading for run in (17c), again and again can
only modify the extended reading, and thus (17d) is not appropriate as a paraphrase.
(17)
(a)
She jumped again and again
(b)
She jumped for several minutes.
(c)
He ran again and again.
(d)
He ran for several minutes.
iv. Note also that the nominalisations of jump type predicates denote single
events, and occur naturally with the light verb give, while nominalisations of
run denote extended events and occur with have:
(18)
(a)
He gave a jump/a kick/ a wink,
(b)
He had a walk/a run/a swim.
So, activities come in two kinds; those that are related to semelfactives and
those that are not. The kind that are related to semelfactives seem to denote an
iteration or repetition of the single event in the denotation of the semelfactive.
At this point, we go back to Dowty's (1979) discussion of activities. Dowty
(1979), in his discussion of the imperfective paradox, argues that, while John
is running normally entails John ran, it does not have this entailment if the
running event is in its initial stages. He shows that that some minimal interval
must pass at the beginning of an activity event e in before one can say that an
event in has happened, and comes to the conclusion that (i) all activities are
related to a 'minimal' activity event, and (ii) all non-minimal activities can be
seen as concatenations of minimal events. He argues further that it is not normally possible to define the minimal event, but stresses that it holds at an interval and not at an instant. (For more discussion see Dowty 1979 and the
discussion of Dowty in Rothstein 2004.) Dowty does not discuss what opera-
185
tion forms activities from minimal activity events, but we are now in a position
to address this.
An obvious candidate is the concatenation operation. On this account, activities would be concatenations of the single events in the denotations of semelfactives, but this cannot be correct, since concatenation is usually taken to
be additive, which means it applies to non-overlapping entities. The operation
here (as we will see below) applies to overlapping entities. It is a summing
operation: a function from E into E with the standard summing properties,
but it must form a singular entity out of the two singular entities which are
summed. This must be the case since extended events in the denotation of an
activity predicate, which by hypothesis are constructed out of minimal events,
none the less count as single events. Thus two minimal events of walking can
get put together to make a single, non-minimal walking event, and so on. In
general, there will be some condition on which elements can be put together
via S-summing, which we express by saying that elements to be S-summed
must stand in the R-relation, and define the operation S-summing (for singular
summing), as in (19):
(19)
S-sum: Vx,y [S-sum(x,y)
R(x,y) s (xuy)]
The operation defined in (19) is a general operation not restricted to the domain of events. When the entities involved are events, then R(e,e') iff:
(a)
e and e' are temporally overlapping

i.e. x(e) and i(e') overlap, where is a function from events to their running times.
(b)
e and e' have the same participants.
Clearly, S-sum is related to the property of S-cumulativity, which we repeat

here from (10). X is S-cumulative iff:
(10)
3eBe'[X(e) X(e') - , e C e' -,e' C e VeVe'[X(e) X(e') R(e,e')
X ( s (eue'))]]
Clearly, it follows that a predicate is S-cumulative if and only if it is closed

under S-sum. Since activity predicates are S-cumulative, they are closed under
S-sum. It follows that activity predicates denote a set of events which hold at
intervals and which are not conceptualised as changes, and that the set is
closed under S-summing. We assume that all activity predicates, whether or
not they are related to semelfactives, denote sets of minimal events closed
under S-summing in this manner.
We can now explain what the relation between semelfactives and activities
is. Assume that all activity predicates denoting a set A are derived from a set
of basic minimal activity events which we call MinA. In some cases, the
minimal events can be lexically accessed and the predicate is ambiguous be-
186
Susan Rothstein
tween: (a) the semelfactive reading in which it denotes the set of minimal
activity events and (b) the activity reading in which it denotes the set closed
under S-summing. The question is now why some activity predicates are ambiguous in this way and others are not. I suggest that the ambiguous predicates,
those where the minimal events can be lexically accessed, are those where the
minimal events are naturally atomic. A naturally atomic entity is one whose
unit structure is perceptually salient and given by the world. Most objects in
the denotation of non-abstract nominis in the count domain are naturally
atomic in this way: person, cat and cup are all naturally atomic, since in a
situation in which there are a number of humans or cats or cups, what counts
as one of each is in some basic sense given. But even in the domain of concrete
entities, not all count nouns denote sets of naturally atomic entities. Rothstein
(1999, 2004) discusses nouns such as fence, wall, and lawn, which denote nonabstract objects whose unit structure is contextually determined.
A naturally atomic event is one which has a natural beginning and end
point, determined by the trajectory which defines the event. If we look at the
diagrammatic representations of a stretch of jumping and running events, given
below, it is clear that the set of jumping events can naturally be divided into
individual minimal jumping events, with the beginning and the endpoints of
the events indicated by the arrows, representing the points where the jumper
leaves and returns to the ground.
Jump
Run:
In contrast, no such natural intuitive division into atomic minimal events is

possible in the case of run since minimal running events do not have naturally
defined beginnings and endpoints. Instead, any one of a set of overlapping
events could be considered a minimal running event. I suggest that only when
the minimal events in a set of activity events are naturally atomic in this way
are they lexically accessible. When a set of minimal activity events is naturally
atomic, then the predicate is ambiguous between a 'normal' activity reading,
where it denotes the set closed under S-surnming and a semelfactive reading
when it denotes the set of minimal activity events. This is the case with jumptype predicates. When the minimal events are not naturally atomic, then the set
of minimal events is not lexically accessible, and the predicate has only the
reading where it denotes the complete set of activity events.
The analysis that I have just given is an analysis of the semelfactive-activity
relation in English. Nonetheless, it has implications for other languages. The
187
distinction between minimal events and events derived under S-summing

should be a feature of all languages which have activity predicates, and the
implied constraint that minimal events which are not naturally atomic are not
lexically accessible is also a constraint which should not be language specific.
But, there is no reason why other languages should express the contrast between minimal events and extended events in the way English does, namely
via an ambiguous predicate, and we would expect other languages to lexicalise
the distinction differently. Preliminary investigation of some Slavic languages
indicates that the activity/semelfactive distinction is indeed lexicalised differently, with activity verbs having imperfective aspect, and semelfactives having
perfective aspect and being derived from activity predicates via affixation.
5 The S-summing operation

If the S-summing operation is not just a device to relate semelfactives and
activities, then it ought to be a general operation available for deriving lexical
predicates in all lexical classes, and not an operation which is stipulated to
apply to minimal activity events. But if it is a generally available operation,
then we want to explain why its effects are only visible in the relation between
minimal activities and semelfactives. There are two possible explanations.
Either it operates on other lexical classes in such a way that its effects are not
seen, or it is prevented from applying to other classes in some principled way.
A closer look shows that S-summing can be treated as a generally available
operation, and that both of the two possible explanations are relevant. When Ssumming applies to other atelic predicates, its effects are not visible, and it is
prevented from applying to so-called telic predicates in a principled way.
The second non-activity class of atelic predicates is the class of states.
These are cumulative, according to Krifka (1992, 1995) and S-cumulative
according to the definition given in (10) above. Two events of loving y
which are temporally adjacent can be conceptualised as a single event of
loving y, and states are infinitely extendable. However, we do not get the distinction between state predicates denoting the set closed under S-summing and
state predicates denoting sets of minimal state events because minimal state
events are not naturally atomic. They are not naturally atomic for a structural
reason. Minimal states hold at instants, and under the assumption that time is
dense, and that any instant can be split into two instants, there is no set of naturally atomic minimal state events available, and therefore no set of such states
which is lexically accessible.
In contrast, S-summing is prevented from applying to achievements and accomplishments because, as we already noted above, they cannot meet the conditions under which it applies. S-summing applies to two events which are
temporally adjacent. However, the definition of change means that achievements and accomplishments will never be able to meet this condition. Assume,
188
Susan Rothstein
that events have initial and final parts determined by the initial and final intervals of their running times, such that M-IN(e) = the minimal initial interval at
which e holds and M-FIN(e) = the minimal final interval at which e holds. A
change from to is an event whose minimal initial part is the last minimal
interval at which holds and whose final minimal interval is the first minimal
interval at which holds, where entails i (Rothstein 2004). But it follows
from this that an event of change defined in this way will never meet the conditions of S-summing since a change from . to can never be immediately followed by another event of the same type with the same participants
since, as Kamp (1979b) argues, two events of change from i to must be
separated by a change back from to . This corresponds to the basic intuition that, for example, John cannot arrive at the same place twice unless he
leaves after the first event and before the second event. Similarly, the same
house (or puzzle) cannot be constructed twice unless it is taken to pieces after
the first event and before the second event begins.
Kamp (1979a,b) argues that because of the density of time, two apparently
temporally adjacent instants will always be separated by an intervening instant,
and if we accept this, then S-summing will apply only to events which overlap
temporally. Under this assumption, we can state that with activities and states,
for two events e and e' in P, it is possible that the minimal final part of e is the
minimal initial part of e'. With accomplishments and achievements this is not
possible because of the structure of the event of change. This means that Ssumming can not apply to events of change.
We can now draw some conclusions about the table in (6). The features
there give us the possible minimal events. Minimal states are non-extended
and not events of change. Minimal activities are extended verbs of non-change.
Achievements are non-extended verbs of change and accomplishments extended verbs of change. There is a generally available operation of S-summing
which applies freely where conditions of application are met. It cannot apply to
achievements and accomplishments but does apply to states and activities.
Lexical predicates denote sets of minimal events closed under S-summing.
Only naturally atomic minimal events are lexically accessible; thus minimal
states are not lexically accessible, and only some activity predicates have naturally atomic minimal events. When a predicate does denote a set of nonminimal events whose minimal events are naturally atomic, the predicate is
ambiguous between a reading in which it denotes the minimal events - the
semelfactive reading - and the reading in which it denotes the whole set.
Semelfactives then, are minimal atomic events in the denotations of some
activity predicates. An obvious and important puzzle is why, since activity
predicates are not telic, do semelfactives have the properties which we associate with telic predicates? As noted above, they occur naturally with in a time
and other telic modifiers, and they induce the imperfective paradox. In this
they pattern with achievements and accomplishments and not with activities.
189
And yet, according to the analysis developed here, they are firmly located, in
terms of features, with the activity/state group.
The explanation for this apparent paradox is that occurring with telic modifiers is not the defining characteristic of accomplishments and achievements;
rather what characterises them is the property used in the table in (6), namely
that they are predicates denoting events of change. A telic predicate is not one
which denotes an event of change, but one which denotes a set of atomic entities. A predicate is telic if it has as part of its lexical content the information
about what counts as one instance of P, in other words if the individual countable units of are given independent of context, and if we can count, independent of context, how many individuals there are in P. However, while telicity is not synonymous with being a predicate of change, predicates of change
are naturally telic predicates. This is because changes are atomic, and the beginnings and endpoints of events of change are determined by the beginning
and end of the change.
States and activities, we saw, are not telic since they are closed under Ssumming: a state or activity predicate denotes a set containing events which
have as parts events which are also in P, and pairs of events which can be Ssummed into other singular events in P. These are not atomic sets since there is
in principle no context independent way to count the number of events in the
set, since both events and their subparts may, in different contexts, count as
one event.
Semelfactive predicates, however, do denote atomic sets, since they denote
sets of naturally atomic events which, although they may have a very large
cardinality, are countable independent of context. Thus while they are like
achievement and accomplishment predicates in denoting atomic sets, the basis
for the atomicity is different. Nonetheless, all three varieties of predicates are
telic.
Our account has led us not only to an explanation of the relation between
semelfactives and activities, but also to a characterisation of telicity in terms of
atomicity. This means that events of change will be telic, since the characterisation of what a change is itself constitutes criteria for individuating individual
changes (an individual change from to just is the event which begins at the
last moment that holds and stops at the first moment that holds.) However,
this is consistent with their being telic predicates whose atomic units are determined in some different way. Semelfactives denote sets of atoms, but their
atomic elements are determined not by a definition of change but by the natural salience and individuability of the elements which count as atoms. Presumably these follow from properties of the physics of motion and other relevant 'real world' facts.
Semelfactives, then, like accomplishments are minimal atomic events which
hold at intervals. The crucial property distinguishing semelfactives from accomplishments is that, though telic, they are inputs to the operation of S-
190
Susan Rothstein
summing, which means that activities can be formed out of them. It is the
characterisation of accomplishments and achievements as events of change
which means that S-summing cannot apply.
6 Degree achievements
Our analysis makes a prediction: if there are events of change which are not
characterised as changes from to where entails -, and where two events
of change can overlap, then S-summing should be possible, and we should find
predicates which are ambiguous between denoting a set of minimal events and
denoting a set of extended iterated events. I should like to suggest that degree
predicates are examples of exactly this.
Degree achievements are discussed in Dowty (1979), Abusch (1985, 1986)
and more recently in Hay, Kennedy, and Levin (1999) and Kennedy and Levin
(2001). They include verbs such as cool, brighten, redden, widen, and darken
and as was argued by Dowty and Abusch, they appear to belong to more than
one lexical class. To start with, they have the properties of achievements. They
denote instantaneous changes, as in (20), and when modified by telic modifiers
such as in half an hour, the modifier locates the change in time, i.e. at the end
of half an hour, without entailing that the change took place gradually during
half an hour, as in (21):
(20)
(21)
(a)
(b)
(a)
(b)
In an instant, the sky brightened.

When I switched off the light the room darkened.
The sky brightened in half an hour (= at the end of half an hour)
It took half an hour for the sky to brighten.
Furthermore, when used with the futurate progressive, a telic modifier locates
the end point of the event, which Rothstein (2004) argues is a clear test for an
achievement:
(22)
(a)
The room is darkening in a couple of minutes,
(b)
The weather is cooling in three days.
However, in addition to the achievement reading, they can also denote extended events in which they are ambiguous between an activity and accomplishment reading depending on whether the sentence is understood as asserting that the event reached some specified point (Abusch 1985, 1986). Thus
(23 a) can be interpreted as the soup became cooler or the soup became cool
and there is a similar ambiguity in (23b):
(23)
(a)
(b)
The soup cooled.

The room/sky darkened.
191
If we use the imperfective paradox, then we get conflicting entailments, depending on whether the verb is interpreted as reaching a determined culmination or not. If cool is interpreted as become cool, then the soup was cooling
does not entail The soup cooled, since the fact that it was cooling does not
entail that it became cool. If cool is interpreted as become cooler, then the
entailment does go through, since the soup is cooling does entail that it became
cooler. This data supports the clam that cool is ambiguous between an activity
and an accomplishment reading. Further, when cool denotes an extended
event, it can occur with both atelic and telic modifiers. In the first case it
means become cooler and in the second it means become cool.
(24)
(a)
The soup cooled for hours.
(b)
The soup cooled in half an hour.
So an account of the lexical semantics of degree achievements needs to explain

how these achievement, activity and accomplishment readings are all derived.
Degree achievements are almost always derived from adjectives. Adjectives
are naturally analysed as denoting functions from individuals to values on a
specified scale (see Kennedy 2001 and references cited there). The soup is cool
entails that the soup has a value on the temperature scale that is below a certain
value. We assume that adjectives have generally the semantic structure in (25),
denoting functions from individuals to values on a scale, where = represents a
choice between =, < and >, and d is a variable over values on the scale. The
meaning of cool and hot would be represented as in (26), where - C O O L and
s-HOT are the cool-scale and the hot-scale respectively (I assume that scales are
pairs of degrees, or measures, and the domain of the measurement, so that SCOOL will be a pair consisting of cCelsius, Temperaturesx In fact, this
means that S-COOL and S-HOT will be the same scale, but I won't discuss
this any further here.)
S
(25)
(26)
II ADJ II = X x . V s ( x ) = d
(a)
I COOL ||
(b)
HOT||=
= XX.VS.COOL (X)
< d
..(x) > d
The soup is cool will then have the interpretation in (27):

(27)
II The soup is cool ||
= V S . C OOL
(THE
SOUP)
< d
This asserts that the value assigned to the soup on the C O O L scale ( - S E L S I U S ,
is less than a contextually determined standard d.
Adjectival constructions of this kind are inherently comparative, and explicit comparative constructions allow the standard of comparison to be expressed. (28) is an example of this:
C
TEMPERATURE)
192
(28)
Susan Rothstein
(a)
The soup is cooler than the sauce.
(b)
VS.COOL ( T H E SOUP) < VS-COOL (THE S A U C E )
This asserts that the value assigned to the soup on the COOL scale is less than
the value assigned to the sauce on the COOL scale.
Against this, what can we say about the interpretation of the verb cool?
There must be three elements to its meaning. Like the adjective, from which it
is derived, it will relate an object to a value on a scale. It will also involve
some comparison, and, as a verb, it will denote a set of events, presumably
events of change. If (27) compares the temperature of the soup with a standard
value, and (28) compares the temperature of the soup with the value assigned
to the sauce, then the soup is cool compares the temperature of the soup at the
end of the event with the value assigned to it at the beginning of the event. The
verb cool then, denotes the set of events in which an object is assigned a
lower value on the temperature scale at the end of the event than it was assigned at the beginning of the event.
(29)
Il COOLV II = Xe.VS-cooL(x, M-FIN(e)) < VS-COOL(X, M-EN(e))

cool denotes the set of events in which the temperature of at
the minimal final interval of e is lower than the temperature of
at the minimal initial interval of e.
So a cooling event is an event of change from a situation in which is assigned d on the COOL scale, to a situation in which is assigned a value lower
than d on the COOL scale.
How does this allow us to account for the properties of cool, and other degree achievements noted above? Note first that this analysis assigns only one
meaning to cool, that of becoming cooler, since it constrains the value of on
the temperature scale at the end of the event only relative to its value at the
beginning of the event, and not in absolute terms. This means that we are disagreeing with theories such as Abusch (1985, 1986), who argue that these
verbs are ambiguous. The meaning given in (29) specifies the direction of
change of value on the scale, without giving any constraints on or absolute
properties of the final value. Evidence in favour of this is that cool does not
mean the same as become cool since cool dictates the direction of the temperature change, while become cool does not. So while the soup cooled entails that
the temperature of the soup decreased, but does not specify its finale value, the
soup became cool does specify the properties of the final value but does not
constrain the direction of the change. Thus we have the constrast in (30):
(30)
W h e n I took the soup out of the fridge it was so cold that it burned my mouth, but after
some time at room temperature, it had become pleasantly cool/*it had cooled.
193
So cool means 'undergo a decrease in temperature', as in (29) and not 'get a

value in the cool range', while become cool means: 'get to have a temperature
value in the (contextually determined) cool range' without specifying the direction of change.
Now let us look at how the various lexical properties of cool are explained
by this analysis.
(i) Cool is a degree achievement. It denotes a set of instantaneous changes,
where the change is a situation in which is assigned a value d on the cool
scale, to a situation in which is assigned a value lower than d. This change is
not inherently extended. (31) is perfectly coherent:
(31)
When I dropped the ice in, the liquid cooled instantly (although not very much).
This cool denotes a set of minimal (non-extended) changes from to , where

there is no interval intervening between M-IN(e) and M-FIN(e).
We will call the denotation of cool as a degree achievement COOLMIN.
On this reading cool has all the properties of achievements noted above. In
these cases, cooled entails became cooler and not became cool, since (29)
expresses only the relation between the values of at the beginning and end of
the event and not the extent of the change itself.
However, unlike 'normal' achievements, the change events in the denotation of cool are not lexically characterised as a change from to a, but as a
change in values on a scale. Crucially, as well as being the degree of coolness
of at the end of e, d can also be the degree of coolness of at the beginning
of e', where e and e' temporally overlap. Thus for an event e in COOL, V(x, MFIN(e)), the value assigned to on the cool scale can be the starting point for
another event of change e', where e and e' overlap temporarily. In other words,
it is possible that V(x, M-FIN(e)) = V(x, M-IN(e')). This means that the conditions for S-summing are met, and we expect degree achievement predicates
like cool to be closed under S-summing. We expect the set COOLSUM to be
events in the set COOLMIN closed under S-summing, and for the events in
COOLSUM to have the properties of activities.
(ii) Cool is an activity. On this reading it denotes the set COOLSUM, the set
of cooling events closed under S-summing. On this reading, the verb appears
with atelic modifiers and the imperfective paradox will not hold:
(32)
(a)
(b)
The soup was cooling.

The soup cooled for three hours.
(32a) entails that the soup has already cooled (somewhat), and (32b) entails
that the soup cooled for some interval of three hours and for all subintervals of
the three hours. So cool on this reading is an activity, but a special activity
since it is derived from an achievement via the S-summing operation.
194
Susan Rothstein
(iii) Is cool also an accomplishment? Despite the evidence above, we do not

need to argue that cool is also an accomplishment, even when it seems to mean
become cool and to have an inherently determined culmination point. The
grammar already provides us with a way to explain constructions like the soup
cooled in three hours. All activities can be used to head VPs which are telic, in
which case they seem to have the properties of accomplishments, and cool is
just like other activities in this respect. The telicity of activity-headed VPs is
determined either contextually or via extent modification, and the effect is to
make the activity seem like an accomplishment. Cool behaves exactly like run
in this respect. First, if there is a contextually determined extent to the event, it
behaves as if it heads a telic VP:
(33)
(34)
(a)
(b)
(a)
(b)
The soup cooled in half an hour,

John ran in half an hour.
Has the soup cooled yet?
Has John run yet?
These examples imply a contextually determined extent to the cooling and

running events. In such a context, (35) does not imply (36):
(35)
(36)
(a)
(b)
(a)
(b)
The soup is cooling,

John is running.
The soup has cooled,
John has run.
Degree modifiers provide an explicit extent to the activity event, also yielding
a telic VP:
(37)
(38)
(a)
(b)
(a)
(b)
The soup cooled three degrees,

John ran three miles.
The soup cooled 3 degrees in ten minutes,
John ran three miles in half an hour.
When a degree modifier is present, not only is the telic modifier possible, as in
(38), but the atelic modifier is impossible, as (39) shows:
(39)
(a)
(b)
#The soup cooled three degrees for hours,

#John ran three miles for days.
7 Conclusions
Verb classes are defined by two sets of features: whether or not the event in its
denotation is inherently temporally extended, and whether or not it denotes an
event of change. There is an operation of S-summing, which forms singular
195
events in out of sums of temporally adjacent events in P. In the normal case,

S-summing does not apply to verbs of changes, since two events of change
from to where entails - cannot be temporally adjacent. The exception is
with degree predicates, which are characterised as changes in values on a
scale.
A number of other concepts have become important in the discussion. First
is the notion of natural atomicity: some things come in inherently individuable
countable units, and others have their unit structure determined contextually.
Semelfactives denote events that come with natural beginnings and endpoints,
and therefore with a salient unit structure. Minimal events of running and
walking don't have a natural breakdown into minimal parts. Activities, while
they have their denotation in the count domain, do not have a naturally atomic
structure, and what counts as an individuable or countable unit is contextually
determined, or determined via measure phrases as in (35-38) above.
Perhaps the most important idea to come out of this discussion is that telicity crosses what we think of as the natural divisions of the verbs into classes,
since semelfactives, which denote minimal activity events, behave as telic
predicates alongside accomplishments and achievements. The reason for this is
that the telic/atelic distinction really has to do with the distinction between
atomic and non-atomic sets: telic predicates denote sets of entities that are
atomic in the sense that the criteria for what counts as 'one entity' are given,
whereas with atelic predicates the criteria for picking out the units are determined by context, modifiers and so on. Atomic units can be determined in
different ways: accomplishment and achievement predicates denote atomic
events because of the notion of a change is essentially an atomic one (although
in practice things are more complicated: see the discussion of telicity in accomplishments in Rothstein 2004). Semelfactives are atomic because of 'the
way the world is', probably, if one looks at the lexical properties of the class of
semelfactives, because the physics of certain kinds of movements means that
they have natural beginnings and endpoints.
8 Appendix: deriving activities from accomplishments

I argued above that accomplishments cannot be S-summed, and that this is the
crucial difference between semelfactives and accomplishments - both of which
are quantized temporally extended events. This means that activities can be
formed from semelfactives, but not from accomplishments. Apparent counterexamples to the claim that activities are not formed from accomplishments are
sentences like (40a), which are acceptable, contrasting with the unacceptable
(40b):
(40)
(a)
I read Dafna The 101 Dalmatians for an hour before she went to sleep,
(b)
#1 built the house for a week.
196
Susan Rothstein
The reason that this is not a real counterexample is that these activities are not
derived via S-summing. We can see this, because S-summing sums minimal
and bigger than minimal events in P, therefore the output of S-summing is
always bigger than a single event. This we get the entailments in (41):
(41)
(a)
I jumped for an hour ENTAILS I jumped at least once within the hour.
(b)
The bird flapped its wing very slowly for a minute ENTAILS It flapped its wings at
least once within that minute.
However, (40a) does not entail that I read The 101 Dalmatians to Dafna in
under an hour. On the contrary, it asserts that the activity associated with the
accomplishment went on for an hour, but implies that less than one accomplishment event in happened. Thus (42a) is felicitiously assertable, while
(42b) is contradictory:
(42)
(a)
I read The 101 Dalmatians for an hour and stopped in the middle of the fourth
chapter. So we haven't yet read it even once,
(b)
The bird slowly flapped its wings for several minutes, but it was suddenly paralysed
before it could flap its wings even once.
If activity events derived via S-summing must be long enough to include at

least one minimal event, then these activity readings of accomplishments are
derived differently. Since accomplishments are probably best analysed complex events consisting of an activity event and as an incremental process, it
looks as if (40a) should be derived by reducing an accomplishment to a simplex verb by deleting the expression of the incremental process.
References
Abusch, D. (1985): On Verbs and Times. Ph.D. dissertation, University of Massachusetts, Amherst (distributed by GLSA).
Abusch, D. (1986): Verbs of change, causation and rime. CSLI Report #CSLI-86-50.
Dowty, D. (1979): Word Meaning and Montague Grammar. Dordrecht: Ridel Publishing Company.
Dowty, D. (1991): Thematic Proto-Roles and Argument Selection. Language 67: 547619.
Hay, J., C. Kennedy, and B. Levin (1999): Scalar structure underlies telicity in "Degree
Achievements". Proceedings of SALT 9.
Kamp, H. (1979a): Events, instants and temporal reference. In: R. Buerle, U. Egli and
A. von Stechow (eds.) Semantics from Different Points of View. Berlin, Springer.
Kamp, H. (1979b): Some remarks on the logic of change. Part I. In: C. Rohrer (ed.)
Time, Tense and Quantifiers, Niemeyer, Tbingen.
197
Kennedy, C. and B. Levin. (2001): Telicity corresponds to degree of change. LSA talk,
75 th Annual Meeting of the Linguistics Society of America, Washington, DC.
Kennedy, C. (2001): Polar opposition and the ontology of degrees. Linguistics and
Philosophy 24, 33-70.
Krifka, M. (1992): Thematic relations as links between nominal reference and temporal
constitution. In: I. Sag and A. Szabolsci (eds.) Lexical Matters, Stanford: CSLI
Publications.
Krifka, M. (1998): The origins of telicity. In: S. Rothstein (ed.) Events and Grammar,
Dordrecht: Kluwer.
Link, G. (1983): The logical analysis of plural and mass terms: A lattice-theoretic approach. In: R. Buerle et al. (ed.) Meaning, Use and Interpretation of Language.
Berlin: de Gruyter: 302-323.
Landman, F. (1992): The progressive. Natural Language Semantics 1: 1-32.
Landman, F. (2000): Events and Plurality: The Jerusalem Lectures. Dordrecht: Kluwer.
Landman, F. (2004): On the differences between the tense-perspective-aspect system in
English and Dutch, ms Tel Aviv University.
Rothstein, S. (1999): Fine-grained structure in the eventuality domain: the semantics of
predicate Adjective Phrases and Be. Natural Language Semantics 7: 37-420.
Rothstein, S. (2004): Structuring Events: an Essay on the Semantics of Lexical Aspect.
Oxford: Blackwell.
Smith, C. (1991): The Parameter of Aspect. Dordrecht: Kluwer.
Vendler, Z. (1957): Verbs and times. The Philosophical Review LXVI 143-160. Reprinted in a revised version in Vendler (1967).
Vendler, Z. (1967): Linguistics in Philosophy. Ithaca: Cornell.
Eri Tanaka (Osaka)
The notion of 'path' in aspectual composition:

Evidence from Japanese
1
Introduction
The most widely discussed property of events in the literature has been the
notion of telicity, this refers to whether an event in question has a natural
end point or not. It has also often been pointed out that the telicity of an
event may be altered by various linguistic factors. One such factor is the
formation of resultative constructions. These constructions can shift an
atelic event to a telic event by adding a resultative predicate, yielding a
[NP1 V (NP2) XP] construction in English, in which the resultative predicate XP specifies the final state of the event.
As a device for forming a telic event from an atelic event, the resultative
construction provides us with a clue toward elucidating how atelic and telic
events differ and how a telic event is constructed semantically. Two different perspectives have predominated this discussion: one that utilizes the
semantics of the BECOME operator, most notably within a framework that
exploits lexical semantic structure (e.g. Dowty 1979, Levin and Rappaport
Hovav 1995, Kageyama 1996) and another that utilizes a homomorphism
between events and individuals (Krifka 1992; 1998). In this paper, I will
advocate the latter view, and extend the relation to a homomorphism between events and 'paths' (cf. Hay et al. 1999, Wechsler 2001). The crucial
evidence for the homomorphism theory derives from the observation of the
interaction between verbal semantics and the meaning of resultative predicates.
Another related issue presented here is the cross-linguistic variation in the
productivity of the resultative construction. English is considered a language
that licenses the construction relatively freely; Japanese, on the other hand,
has a more restricted distribution than English. I will argue that this typological difference is also captured by the homomorphism theory.
The work presented here is partly supported by the JSPS Research Fellowship for Young
Scientists.
200
Eri Tanaka
This paper is organized as follows: In section 2, I will introduce two typologies of resultative constructions, and will present the quandary that the
two theories create for Japanese resultatives. In section 3, I will propose a
'mixed' theory of BECOME and homomorphism approaches for telic formation, and will show that the quandary is partly solved by it. Section 4
focuses on Japanese adjectival resultatives, which pose a serious problem,
that is in fact solved by the homomorphism theory, incorporating the BECOME theory. Section 5 provides a conclusion to the paper.
2 The quandary
2.1 Two typologies of resultative constructions
The view that resultative constructions do not form a homogeneous class but
consist of several different classes, based mainly on their syntactic properties,
has been advocated by various researchers (e.g. Carrier and Randall 1992,
Levin and Rappaport Hovav 1995). In English, at least two types of resultatives have been identified; CONTROL and ECM resultatives (Carrier and
Randall 1992, Wechsler 2001). The distinction is crucially dependent on
whether the predicational subject of the resultative predicate is subcategorized by the main verb or not. In the control type, the resultative predicate is
predicated of a subcategorized argument of the main verb; while in the ECM
type, the predicational subject is a non-subcategorized argument:
(1)
(2)
(a)
He wiped the table clean. => He wiped the table.
(b)
The water froze solid. => The water froze.
(a)
The dog barked itself hoarse. > *The dog barked itself
(b)
Mary ran the soles off her shoes.
[CONTROL]
[ECM]
*Mary ran the soles.
The fact that the entailments in (2) are not available suggests that the main
verbs are basically (unergative) intransitives.
Recent studies have focused on the semantic differences among the types
of resultative constructions and have shown that these semantic properties
constrain the distribution of the construction (e.g. Rappaport Hovav and
Levin 2001, Wechsler 2001). Most notably, Wechsler (2001) has focused on
the semantic structure of resultative predicates, which has not attracted
much attention in the literature. 1
The resultative constructions consist of two subevents, which are prototypically designated by the main verb and the resultative predicate. Wechsler
(2001) argues that the two types of resultative construction, control and ECM,
contribute to the formation of the event in different ways.
Vanden Wyngaerd (2001) has also presented this argument along similar lines.
The notion of 'path' in aspectual composition
201
Wechsler's (2001) proposal is couched in the 'co-extensiveness' of two

subevents:
(3)
Co-extensiveness condition on control resultatives

1.
The telic event and the path must be (a) homomorphic (parts of the event must
correspond to parts of the path and vice versa) and (b) co-extensive (the event must
begin when the affected theme is at the start of the path and end when the affected
theme reaches the end of the path)
2.
The affected theme must be an argument of the event-denoting predicate.
The 'path' or 'scale' is provided by the resultative predicates; in (la), for

example, the 'path' that the wiping event coincides with is the property scale
of cleanliness. Thus, in (la), as the table-wiping event proceeds, the table gets
cleaner.
Since the co-extensiveness condition in (3) requires that the distance of the
path traversed and the procession of the event coincide, it predicts that the
scalar structure of adjectives and the internal structure of the event must have
congruous properties. Specifically, it predicts the following:
(4)
(a)
Non-gradable adjectives (i.e. adjectives with no scalar structure) and durative
(b)
Non-gradable adjectives can form resultatives with punctual (non-durative) verbs;
(c)
Closed scale adjectives form control resultatives with durative verbs; and
(d)
PPs can form control resultatives with any verbs.
verbs (i.e. verbs without internal structure) cannot form a control resultative;
Modification by very attests to the gradability of adjectives (cf. Hay et al.

1999):
(5)
(a)
Gradable adjectives:
very {long/flat/expensive/straight/full/dull}
(b)
Non-gradable adjectives:
*very {dead/triangular/invited/sold}
Consider the following:

(6)
(a)
*The rabbits had apparently been battered dead.
(b)
The rabbits had apparently been battered to death.
(Wechsler 2001)
In (6a), the adjective dead is a non-gradable adjective, which entails the

non-existence of scalar structure. The co-extensiveness condition explains
the ungrammaticality of (6a) as the incompatibility of non-gradable adjective
dead and durative verb batter, the prediction in (4b) is borne out. In (6b), on
the other hand, the PP specifies the path that continues to the point of death,
which makes the sentence grammatical.
202
Eri Tanaka
Another type of typology of resultative constructions has been presented

by Washio (1997). Washio's classification is based on the notion of 'the
implication of result'; he distinguishes (at least) two types of resultatives,
weak and strong resultatives. 2
(7)
(a)
(b)
WEAK resultatives: the meaning of the main verb contains the result of the action
designated by the verb, and the resultative predicate specifies that result
STRONG resultatives: the meaning of the main verb does not imply the result of the
action, and thus, the meanings of the verb and the resultative predicate are independent of each other
The two types of resultatives are exemplified in the following:

(8)
(9)
(a)
John broke the vase into pieces,
(b)
The water froze solid.
[WEAK]
[WEAK]
(=(lb))
(a)
John hammered the metal flat,
(b)
John beat the man bloody.
[STRONG]
[STRONG]
The breaking event, as a natural course of the process, lexically entails the
broken state of the object. In (8a), therefore, the resultative predicate, into
pieces, designates that resultant state. 3 In (9a), on the other hand, the event of
hammering the metal may or may not result in the metal being flat.
The weak-strong dichotomy is regarded as a criterion for the availability of
the construction varying among languages. English, as shown in (8)-(9),
allows both weak and strong resultatives. Washio (1997), however, observes
that Japanese, unlike English, only permits the weak type:4'5
(10)
(a)
John-ga kabin-o
konagona-ni
J-NOM vase-ACC
pieces-^ broke
kowasita
'John broke the vase into pieces.'

(b)
mizu'umi-ga
kachikachi-ni
kootta
lake-NOM
solidADj-mf
froze
'The lake froze solid.'

(11)
(a)
*John-ga
kinzoku-o
J-NOM
metal-ACC
pechanko-ni
flatADj-,/
tataita
pounded
'John pounded the metal flat.'
Washio (1997) has indeed proposed a third group, spurious resultatives. They are not regarded as true resultatives, as apparent resultative predicates in this group are analyzed as
manner adverbials.
See Torota (1998).
Abbreviations to be used throughout this paper: NOM for nominative, ACC for accusative,
ADJ-m/ for adjectives in infinitive form, LOC for locative, COP for copular.
Cf. Kageyama (1996).
(b)
*John-ga
sono otoko-o
chimamire-ni
nagutta
J-NOM
that man-ACC
bloodyADj-;n/
beat
203
'John beat the man bloody.'
In Japanese adjectival resultative constructions, resultative predicates must

bear the specific morphological marking -ni (or -ku) which is identified as the
infinitive form (called ' r e n ' y o o ' form in traditional Japanese grammar). If
they are not inflected (-da ending, sentence final form) or inflected into another form (-de ending), the sentence is ungrammatical: 6
(12)
(a)
*John-ga
kabin-o
konagona-da/de
kowasita
J-NOM
vase-ACC
pieces
broke
(Intended) 'John broke the vase into pieces.'

(b)
*mizu'umi-ga
kachikachi-da/de
kootta
lake-NOM
solid
froze
(Intended) 'The lake froze solid.'
The examples in (11) are control-strong type resultatives; English allows

ECM-strong resultatives, as shown in (2) above. The Japanese counterpart to
this construction, on the other hand, is ungrammatical: 7
(13)
(a)
*inu-ga
jibun-o
karakara-ni
hoeta
dog-NOM
self-ACC
hoarse^-,,,/
barked
(Intended) 'The dog barked itself hoarse.'

(b)
*hitobito-ga
hodoo-o
pechanko-ni
hasitta
people-NOM
pavement-ACC
flatADj-;/
ran
(Intended) 'People ran the pavement thin.'
The fact that Japanese only permits weak (adjectival) resultatives is referred
to as Washio's generalization throughout this paper.
The question that arises is how the two typologies of the resultative constructions, namely control-ECM dichotomy and weak-strong dichotomy, are
related. The two types of classification are in fact not incompatible with each
other; the relation between the two is summarized in table 1 below. It should
be noted that the category of weak-ECM resultatives is not available, as ECM
resultatives are usually confined to unergative verbs, which are, in turn, semantically not result-implying verbs (i.e. activity verbs in the sense of
Vendler 1957).
6
A d j e c t i v e s with the -de inflection m a y in fact appear in this context, but only with a subject-oriented depictive interpretation:
(i)
John-ga
kabin-o
hadaka-de
kowasita
J-NOM
vase-ACC
nudeADJ
broke
'John broke the vase nude.' [John was nude.]

I will not provide an account of why Japanese lacks the E C M resultatives.
204
Eri Tanaka
Table 1: Control-ECM and Weak-Strong dichotomies

WEAK
STRONG
CONTROL
(8a), (8b)
(9a), (9b)
ECM
None
(2a), (2b)
Both classifications of resultatives make a reference to the aspectual structure

of events: however, the two approaches differ in that the co-extensiveness
condition restricts the way in which the event designated by the main verb is
related to the semantics of the result phrase, while the weak-strong dichotomy
only refers to the event designated by the main verb; the structure of the
resultative predicate is irrelevant.
In the next section, I will demonstrate that there are some cases for which
Washio's generalization does not hold; specifically, when the resultative
predicate is a PP, a strong-type resultative can be observed even in Japanese.
However, it will be shown that in adjectival resultatives, the scalar structure
of resultative predicates is in fact irrelevant, as Washio's generalization indicates.
2.2 Gradability and resultatives in Japanese
A possible resolution to the discrepancy between Washio's generalization
and the co-extensiveness condition is to show that Japanese adjectives, in
contrast to their English counterparts, lack the gradable vs. non-gradable
contrast. This, however, is not true; Japanese adjectives do indeed exhibit this
contrast. Modification by totemo 'very' as in (14) (cf. Uehara 1998,
Takezawa 2001) attests to the gradability in Japanese:
(14)
Gradable adjectives: 8
(a)
totemo
{naga-i/mizika-i/kata-i/taka-i/usu-i/atu-i
very
{long/short/solid/expensive/thin/thick...}
Japanese has two types of adjectives that differ in their morphological markings. One is
usually referred to as '-adjectives, and the other as na-adjectives. The following are the inflectional paradigms of the adjectives:
(i) '-adjectives
[nominal modification]
[infinitive]
naga-i
(b) naga-i
hana
(c) hana-ga
long-I
long-I
nose
[sentence final]
(a) hana-ga
nose-NOM
'The nose is long.'
nose-NOM
'(A) long nose'
nagak-u
nobiru
long-KU
lengthen
'The nose lengthened.'
(ii) na-adjectives
(a) jimen-ga
ground-NOM
'The ground is
taira-da
flat-DA
flat.'
(b) taira-na jimen

flat-NA
ground
'(A) flat ground'
(c) jimen-ga taira-ni naru

ground-NOM flat-NI become
'The ground becomes flat.'
(b)
(15)
totemo
{taira-da/nameraka-da/kirei-da/sinsen-da
very
{flat/smooth/clean/fresh
205
Non-gradable adjectives:
(a)
(b)
totemo
{*sikaku-i/??maru-i/??aka-i/??ao-i....
very
{square/round/red/blue)
totemo
very
{*makka-da/??hetoheto-da/*pechanko-da/*kongona-da....)
{red/tired/flat/fragmented into pieces...)
Among gradable adjectives, closed and non-closed (i.e. open) scale adjectives
are distinguished on the basis of the modification by kanzen-ni 'completely':
(16)
Gradable, closed scale adjectives:

(a)
(b)
(17)
kanzen-ni
{kata-i/usui}
completely
{solid/thin}
kanzen-ni
{taira-da/nameraka-da/kirei-da}
completely
{flat/smooth/clean}
Gradable, non-closed scale adjectives:

(a)
(b)
*kanzen-ni
{naga-i/mizika-i/taka-i/atu-i}
completely
{long/short/expensive/thick}
*kanzen-ni
sinsen-da
completely
fresh
As Washio's generalization indicates, the type of adjectives does not affect

grammaticality, but the type of verbs does:
(18) Gradable adjectives + change of state verbs
(i) closed-scale:
(a) John-ga
J-NOM
kinzoku-o
taira-ni
nobasita
metal-ACC flat-ADW.
flattened
'John flattened the metal.'

(b) aisu criimu-ga
ice cream-NOM
kata-ku kootta
solid- ADJ;,,/
frozen
'The ice cream frozen solid.'

(ii) non-closed scale:
(c) John-ga
J-NOM
pankizi-o
naga-ku
dough-ACC long-ADJI/
nobasita
lengthened
'John lengthened the dough.'

(19)
Non-gradable adjectives + change of state verbs

(a) John-ga
J-NOM
kinzoku-o
sikaku-ku
metal-ACC square-ADJIN/
nobasita
flatten
'John flattened the metal into a square.'

(b) John-ga
J-NOM
kinzoku-o
pechanko-ni
metal-ACC flat -ADhnf
'John flattened the metal.'
nobasita
flatten
206
Eri Tanaka
(20)
Gradable adjectives + activity (=durative) verbs

(i)
closed scale:
(a) *John-ga
kinzoku-o
J-NOM
usu-ku
metal-ACC thin-ADj;/
tataita
pounded
(Intended) 'John pounded the metal thin.'

(b) *John-ga
kinzoku-o
J-NOM
metal
taira-ni
flat-ADj;/
tataita
pounded
'John pounded the metal flat.'

(ii) non-closed scale:
(c) *John-ga
J-NOM
kinzoku-o
mizika-ku
metal-ACC short-ADj;/
tataita
pounded
(Intended) 'John pounded the metal and made it short.'

(21)
Non-gradable adjectives + activity (=durative) verbs

(a) *John-ga
J-NOM
kinzoku-o
pechanko-ni
metal-ACC
flat-ADj;/
tataita
pounded
(Intended) 'John pounded the metal flat.'

(b) *John-ga
J-NOM
hetoheto-ni
hasitta
tired-ADjui/
ran
(Lit.) 'John ran tired.' (Intended) 'John ran himself tired.'
2.3 Control-strong resultatives in Japanese

First observe the following:
(22)
(23)
(a)
John {arrived at/went to) the station,
(b)
John walked/ran to the station.
(a)
John-ga
eki-ni
tuita/itta
J-NOM
station-LOC
arrived/went
'John (arrived at/went to) the station.'

(b)
*John-ga
eki-ni
aruita/hasitta
J-NOM
station-LOC
walked/ran
(Intended) 'John walked/ran to the station.'
The contrast between (22b) and (23b) has been well recognized in the literature (e.g. Y o n e y a m a 1988, Kageyama 1996, Levin and Rappaport Hovav
1995). Since sentence (23b) indicates that the locative phrase headed by -ni
'in/at/to' cannot be associated with manner of motion verbs, which do not
implicate a change of location, its ungrammaticality may be attributed to
W a s h i o ' s generalization. Indeed, these verbs are durative atelic verbs as
shown by the (in)compatibility with time adverbials, while the verb in (23a) is
telic:
(24)
(a)
John-ga
{*5 hunkan/5 hun-de)
tuita
J-NOM
{5 minutes-for/5 minutes-in}
arrived
'John arrived {*for 5 minutes/in 5 minutes).'
207
(b)
John-ga
(5 hunkan/*5 hun-de}
aruita/hasitta/oyoida
J-NOM
{5 minutes-for/5 minutes-in}
walked/ran/swam
'John walked/ran/swam {for 5 minutes/*in 5 minutes}.'
The contrast between (22) and (23) can, thus, be regarded as another instance
of the weak-strong dichotomy. Japanese, however, has another postposition
which can indeed be used with manner of motion verbs to form a telic event
(cf. Tsujimura 1991, Kageyama and Yumoto 1997):
(25)
(a)
(b)
John-ga
eki-made
aruita/hasitta
J-NOM
station-up to
walked/ran
'John walked/ran to the station.'
John-ga
mukoogisi-made
J-NOM
the other side of the river-up to
'John swam to the other side of the river.'
oyoida
swam
The telicity of these sentences is confirmed by the adverbial test as in the

following example:9
(26)
(a)
(b)
aruita/hasitta
John-ga
eki-made
5 hun-de
walked/ran
J-NOM
station-up to
5 minutes-in
'John walked/ran to the station in 5 minutes.'
John-ga
mukoogisi-made
5 hun-de
J-NOM
5 minutes-in
'John swam to the other side of the river in 5 minutes.'
oyoida
swam
This is indeed an ambivalent situation: Washio's generalization holds for

cases in which the resultative predicate is an adjective or -ni locative, while it
does not work for cases in which it is a -made locative.
g
Ueno and Kageyama (2001: 63-64) claim that -made is in fact ambiguous between the telic
and atelic interpretation, because in addition to the frame adverbial 5 hun-de 'in 5 minutes', it
allows the durative adverbial 5 hunkan 'for 5 minutes':
(i) John-ga
eki-made
5 hunkan
aruita
J-NOM station-until
5 minutes-for
walked
'John walked 5 minutes until he reached the station.'
However, the fact that the sentence can license a frame adverb is important, since as Ueno and
Kageyama (2001) note, the -made phrase in (i) may not be a 'locative'. Consider the following example:
(ii) John-ga
Tokyo-made
{3 jikan/*3 jikan-de}
hon-o
yonda
J-NOM Tokyo-until
{3 hours-for/3 hours-in} book-ACC read
'John read the book (for 3 hours/in 3 hours} until he reached Tokyo.'
The -made phrase here is naturally paraphrasable into a sentence with the temporal connective until. In example (ii), a frame adverbial cannot be used. Thus, even if (i) is ambiguous
between telic and atelic interpretations, the present argument is not affected, as the -made in
(i) functions differently (namely as temporal adverbial) from the one in (24).
208
Eri Tanaka
I would like to present a solution to this problem by first proposing that

there are two types of telic formations, both of which obey the
co-extensiveness condition. This proposal untangles the apparent discrepancy
between PPs and adjectives, observed above, and also elucidates the relation
between Washio's generalization and the co-extensiveness condition.
3 Two types of telic formations and path expressions

in Japanese
3.1
Event structures and telicity
The standard theory of telicity of event, especially among the theories that
feature lexical-conceptual structures (LCS), relies on the notion of BECOME
as the origin of telicity. Among Vendler (1957)'s four-way classification of
event schemes, namely State, Activity, Achievement, and Accomplishment,
the latter two categories have a set terminal point (or natural endpoint), while
the former two groups do not. The standard in vs. for adverbial test indicates
this:
(27)
(a)
John loves Mary {for years/*in two years}.
(b)
John walked {for hours/*in an hour).
[State]
[Activity]
(c)
John arrived at the station {*for hours/in an hour).
[Achievement]
(d)
John walked to the station {*for hours/in an hour)
[Accomplishment]
The four categories are represented in the LCS as in (28). The telic eventualities (i.e. Achievements and Accomplishments) include a BECOME operator in their representations: 10
(28)
(a)
[x BE (AT y)]
(b)
[x ACT(-ON y)]
(c)
[x B E C O M E [BE [AT y]]]
(d)
[x ACT(-ON y)] CAUSE [y B E C O M E [BE [AT y]]]
States
Activities
Achievements
Accomplishments
The telicity is thus seen to be due to a BECOME operator, with the semantics
roughly described as a transition from state to state a .
Resultative constructions have been regarded as a linguistic realization of
the scheme in (28d), and the shift in telicity in (27b) to (27d) has been analyzed as pertaining to the introduction of the BECOME operator in the semantic representation. Under this analysis, both at the station in (27c) and to
the station in (27d) specify the final state of the BECOME operator (i.e. [BE
10
I agree with Rothstein(2004: 103-104) that the origin of the telicity in Accomplishment
cannot be due to the CAUSE operator.
209
AT y] component). It is now apparent that the weak-strong distinction corresponds to the distinction between (28c) and (28d). u
One problem for the representations in (28) to be viewed as (the unique)
model for telic formation derives from the following well-known contrast
(Verkuyl 1993, Krifka 1992, 1998, among many others):
(29)
(a)
John ate two apples {*for five minutes/in five minutes}.
(b)
John ate apples {for five minutes/*in five minutes}.
Evidently, the telicity of the examples in (29) is determined by the specificity

of the amount of the object. Example (29a), in which the amount of apples
that John ate has a specific quantity (i.e. two), is interpreted to be telic; while
in (29b), the bare plural does not set a definite boundary to the amount of the
apples, resulting in an atelic event. The coincidence of the telicity of the
described event and the specificity of one of its arguments has been referred
to as an incremental relation (cf. Dowty (1991)): in the present example, the
eating event is incremental with respect to the amount of apples.
The difference in (29a) and (29b) would then be, with the representations
in (28b) and (28d), a result of the BECOME component, and the impact of the
semantics of the object NP could not be defined clearly.
Moreover, the BECOME theory does not explain the contrasts exhibited in
(6a) vs. (6b), and (23a) vs. (23b), as the resultative predicates in these examples uniformly designate the result states of the events (cf. Vanden Wyngaerd 2001).
I now propose that the problem presented here will be solved if we posit
two types of telic formation: one which is the result of the BECOME operator,
and the other which is the result of the end of the path. I take the notion of
'path' to be the entity with which the event in question is related incrementally. Incrementality will be defined below. I will represent both types of telic
events in terms of event semantics, following Parsons (1990) and Landman
(2000): 12
11
Washio (1997) indeed rejects this view, in the light of the fact that huku 'wipe' and niru 'boil'
can form a resultative in Japanese, as in:
(i)
John-ga
teeburu-o
kirei-ni huita
J-NOM table-ACC clean-ADj/. wiped
'John wiped the table clean.'
Huku 'wipe' and niru 'boil', however, may associate either with in or for adverbials, which
suggests that they are indeed ambiguous between a telic and atelic use.
12
In our domain of discourse, I assume in addition to ordinary individuals, events, following

Parsons (1990), Landman (2000). Thus we have in our ontology: (i) individuals, of type d (ii)
events, of type e, and (iii) truth values, of type t.
210
(30)
Eri Tanaka
(a)
(b)
3e [BECOME(P(e)) & Argument(e)=x & 3e' [Cul(e)=e'

& Argument(e' )=x & P(e' )]
Be [P(e) & Path(e)=x]
[BECOME]
[Threshold]
In the proposed semantics, telicity may be due to the semantics of the BECOME operator (30a), or to the boundedness of the path that the event in
question takes (30b).
According to the ambivalent telic model in (30), the telic property of (27c)
is analyzed as a property pertaining to the semantics of BECOME, while the
telicity in (27d) and (29) is dependent on the boundedness of the path. In
(29a), the path (in this case, the object NP) is bounded, thus resulting in a telic
event; in (29b) on the other hand, the path is not bounded, thus leading to an
atelic event. The same relation holds for motion events:
(31)
(a)
(b)
John walked toward the station {for five minutes/*in five minutes}.
John walked along the street {for five minutes/*in five minutes}.
In Krifka (1992, 1998), the incremental relation is couched in the thematic

relation: the relation is understood to be a homomorphism between an event
and one of its arguments. Due to the homomorphism, the structure of the
incremental argument is inherited to the event.
Following Krifka (1992, 1998), I assume part structures with sum operation (described as in (32)) both for the individual and event domain.
Mass/plural-singular distinction in the individual domain corresponds to an
atelic-telic distinction in the event domain. In the individual domain,
mass/plural NPs denote an entity that has a 'part-of relation, singular count
NPs denote an atomic entity which has no internal structure (cf. Link 1983):
(32) Part structure:
abc
ab
=> plural/mass
=> singular (atomic)
Bare plurals and mass NPs feature a cumulative property, which is defined in
(33a). The cumulative property is contrasted with a quantized property, definition of which is given in (33b) (Krifka 1998: 200):
(33)
(a)
(b)
X is CUMULATIVE iff
there are two distinct individuals, x, y such that X(x) and X(y);
Vx, y [X(x) & X(y) -> X(xy)]
X is QUANTIZED iff
Vx, y [X(x) & X(y)
-,y < x]
211
Cumulativity leads to 'unboundedness'; apples in (29b) is cumulative, for

when two distinct entities and y are both apples, then the sum of the two
entities are also apples. On the other hand, two apples in (29c) is not cumulative but quantized, or 'bounded', for the sum of any two objects that can be
called two apples cannot have a proper part, which can be called two apples
again.
In the event domain, the cumulative-quantized contrast leads to an atelic-telic distinction. Atelic events, such as walk, run, and hammer, predicate
over the sum of two events that are also walk, run or hammer, while telic ones,
such as arrive, and hammer the metal flat, cannot. Telic events are regarded
as 'singular-count' events, with no homogeneous internal structure, and form
'atomic' events of their own. It should be noted here that when we talk about
the cumulative-quantized distinction in the event domain, this refers to the
internal structure of 'single' events, not to 'plural' events (cf. S-cumulative in
Rothstein 2004:8-9). When an event is 'pluralized', or an event is interpreted
as occurring several times, even a telic event has a part-whole structure in its
denotation.
Krifka's (1998: 211-213) definition of the incremental relation is as follows:
(34)
(a)
Mapping to Subobjects (MSO)

VxVe, e' [ (x, e) & e' < e > 3y [ (y, e') & y < ]]
(b)
Mapping to Subevents (MSE)

VeVx,y [ (x, e) & y < - > 3e' [ (y, e') & e' < e]]
(35)
Strict Incrementality (SINC)

M S O with Uniqueness: VxVe, e' [ (x, e) & e' < e - 3!y [ (y, e') & y < ]]
M S E with Uniqueness: VeVx,y [ (, e) & y < 3!e' [ (y, e') & e' < e]]
(36)
Cumulativity (CUM)
is cumulative iff
Vx, yVe, e' [ (x, e) & (y, e ' ) -> (xy, ee')
It is now evident that the relation between the eating event and its object
satisfies both SINC and CUM. 13 An event with an incremental relation is
quantized, when its incremental object is quantized; the event is cumulative,
when its incremental object is cumulative (I refer to the details of the proof to
Krifka 1998).
The notion of incrementality defined above builds on an idea very close to
the co-extensiveness condition. However, incrementality based on thematic
relation will not be extended to adjectival resultati ves, since the relation holds
between an event and a property scale, which is not explicitly a thematic
13
(i) For all eat(e), eat(e') such that e' < e and Patient(x, e), there is y such that Patient(y, e') and
y is a part of x.
(ii) For all eat(e), eat(e') and Patient (x, e) and Patient (y, e'), Patient (xSty, ee')
212
Eri Tanaka
argument of the event. Thus, I will modify the notion slightly by implementing the notion of path:
(37)
Strict Incrementality (SINC)

Mapping to Subpath with Uniqueness:
VxVe, e' [Path(e)=x & e' < e
B!y [Path(e')=y & y < ]]
Mapping to Subevent with Uniqueness:

VeVx,y [Path(e)=x & y < -> 3!e' [Path(e')=y & e' < e]]
The thematic relation in (34)-(36) is generalized for the notion of path, which
has to satisfy the following condition:
(38)
An individual is a path of an event iff there is at least one y such that y < x.
3.2 Threshold to path: evidence from PP resultatives in Japanese

Recall that two postpositions, -ni 'in/at/to' and -made 'up to' in Japanese
display a contrast with respect to consistency with the verbs with which they
occur. I will repeat the relevant data below as (39) and (40):
(39)
(a)
John-ga
eki-ni
tuita/itta
J-NOM
station-LOC
arrived/went
'John {arrived at/went to} the station.'

(b)
*John-ga
eki-ni
aruita/hasitta
J-NOM
station-LOC
walked/ran
(Intended) 'John walked/ran to the station.'

(40)
(a)
John-ga
eki-made
aruita/hasitta
J-NOM
station-up to
walked/ran
'John walked/ran to the station.'

(b)
John-ga
mukoogisi-made
oyoida
J-NOM
swam
'John swam to the other side of the river.'
(39) is an instance of BECOME-telicity (=(30a)), while (40) is an instance of

Threshold-telicity (=(30b)). Since the co-extensiveness condition requires a
homomorphism between an event and its path, the ungrammaticality of (39b)
has to be due to the 'non-gradability' of the locative. In this section, I will
show that this is indeed true in Japanese.
The 'non-gradability' or the lack of path in the semantics of the -ni locative
emerges in the following contrasts (cited from Tanaka 2003: 288-289):
(41)
(a)
kodomo-ga genkan-ni
iru
child-NOM entrance hall-LOC
is
'There is/are a child/some children in the entrance hall.'
213
[kodomo 'child': S singular /

(b)
plural]
kodomo-ga genkan-made
iru
child-NOM entrance hall-up to
is
'There are some children who form a line to the entrance hall.'
[kodomo 'child': * singular / S plural]
(42)
(a)
#kono kaidan-wa
2 kai-ni
nandan
ari-masu-ka
this stairs-TOP
second floor-LOC
how many steps
be-POLITE-Q
(Intended) 'How many steps are there to the second floor?'

'How many steps are there on the second floor?'
(b)
kono kaidan-wa 2 kai-made
nandan
nandan
ari-masu-ka
this stairs-TOM second floor-up to
how many steps
be-POLITE-Q
'How many steps are there to the second floor?'
In (41), both -ni and -made are used together with the existential verb iru 'be',
and thus the verbal semantics is neutral in this context. Both postpositions
predicate of the subject NP, kodomo 'child'. In Japanese, which lacks a systematic number marking, a common noun like kodomo 'child' may be either a
singular or a plural, depending on the context. In (41a), where locative -ni is
used, the subject may be interpreted to be either a singular or a plural, while in
(41b), -made requires that the subject NP be a plural. The same contrast also
arises in (42). In (42), the number of the steps to the second floor is asked for:
the fact that -ni cannot be used indicates that it cannot predicate of the path to
a destination.
The data in (41)-(42) suggest that -made takes a path that consists of subparts as its predication subject, while -ni does not, and only designates the
place where the subject exists. The predication subject of -made, unlike -ni,
features an 'ordering' or precedence relation among the subparts. Locative -ni
does not pose such a semantic requirement on its subject. In the light of this
observation, I will propose the following for the semantics of the postpositions:
(43)
(a)
ll-mll = XxXyXe[ni(e)
(b)
\\-made\\ = XxXyXPXe[made(x)(y)
& Theme(e)=y & y is at x]

& P(e), where y satisfies condition (38)]
The semantics given in (43) thus claims that the -ni locative is a predicate
that situates y at x, while -made is combined with a path that has an internal
structure, and gives a bounded path that ends at x.
The lack of path in the semantics of -ni explains the ungrammaticality of
(39b); for (39b) to be a telic event, it must have a bounded path, but -ni cannot
be used as a 'bounder'. In other words, it only designates the final state, and
cannot include the intermediate 'transition'.
On the other hand, -made 'up to' works to bound an intermediate path. In
Japanese, accusative case marker -o may be used to indicate a spatial region
as in (44a,b):
214
Eri Tanaka
(44) (a)
John-ga
Midoosuji-o
aruita/hasitta
J-NOM
Midoosuji Avenue-ACC
walked/ran
'John walked/ran along Midoosuji Avenue.'

(b)
John-ga
kono kawa-o
oyoida
J-NOM
this
swam
river-ACC
'John swam across this river.'
The accusative case marked path cannot appear with verbs like arrive:
(45)
*John-ga Midoosuji-o
tuita
J-NOM
arrived
Midoosuji avenue-ACC
(Intended) 'John arrived through/by way of the Midoosuji avenue.'
The -o marked spatial region can be used with -made phrases, as shown in the
following:
(46)
(a)
John-ga
Midoosuji-o
eki-made
aruita/hasitta
J-NOM
Midoosuji Avenue-ACC
station-up to
walked/ran
'John walked along Midoosuji Avenue to the station.'

(b)
John-ga
kono kawa-o
mukoogisi-made
oyoida
J-NOM
this
swam
river-ACC
'John swam across this river to the other side of it.'
Verbs like iku 'go' and kuru 'come' may be associated with -ni or -made
phrases (47a), and they accept -o marked NPs (47b). Even these verbs,
however, when accompanied by -ni, and not by -made, yield an unacceptable
sentence (47c):
(47)
(a)
John-ga
{eki-ni/eki-made}
itta/kita
J-NOM
{station-LOC/station-up to}
went/came
'John went/came to the station.'

(b)
John-ga
Midoosuji-o
itta/kita
J-NOM
went/came
'John went/came along Midoosuji Avenue.'

(c)
John-ga
Midoosuji-o
(*eki-ni/eki-made)
itta/kita
J-NOM
{station-LOC/station-up to}
went/came
'John went/came to the station along Midoosuji Avenue.'
Another set of data that shows the (non-)existence of an intermediate path in

the semantics of the verbs derives from the partway (partly I halfway) test
given in Tenny (2000:307):
(48)
(a)
Martha partly ate the sandwich.
(b)
Marge ran partway to the drugstore.
(c)
*David put the book partway on the table.
215
(49)
(a)
John-ga
(b)
J-NOM
sandwich-ACC half
ate
'John ate the sandwich halfway.'
John-ga
sandowicchi-o hanbun-made
sandowicchi-o
hanbun tabeta
tabeta
J-NOM
sandwich-ACC half-up to
'John ate half of the sandwich.'
(50)
(a)
(b)
(51)
(a)
(b)
ate
John-ga
eki-made
hanbun aruita/hasitta
J-NOM
station-up to
halfway walked/ran
'John walked/ran halfway to the station.'

John-ga
Midoosuji-o
hanbun
aruita/hasitta
J-NOM
halfway
walked/ran
the Midoosuji Avenue-ACC
'John walked/ran halfway along Midoosuji Avenue.'

John-ga
eki-made
hanbun
itta/kita
J-NOM
station-up to
halfway
'John went/came to the station halfway.'
went/came
*John-ga
itta/kita
eki-ni
hanbun
J-NOM
station-LOC
halfway
went/came
(Intended) 'John went/came to the station halfway.'
(52)
(a)
(b)
*John-ga
eki-ni
hanbun
tuita
J-NOM
station-LOC
halfway
arrived
'John arrived at the station halfway.'

*John-ga
hon-o
hanbun
J-NOM
book-ACC halfway
tukue-ni
desk-LOC
oita 14
put
'John put a book halfway on the desk.'
Table 2 summarizes the observations:

Table 2: Path expressions and verbal semantics
-o marked path
-ni
-made
hanbun 'halfway'
Eat
walk/run/swim
go/come
arrive/put
The correlation observed suggests that 'non-gradable' locatives cannot be

associated with path-taking verbs, and that 'gradable' locatives cannot be
associated with non-path-taking verbs. This is what the co-extensiveness
condition predicts, and the distribution of Japanese PP resultatives is indeed
explained by the theory.
14
Hanbun 'half/halfway' may modify the quantity of books, resulting in the reading 'half of the
books.' This is, of course, not the intended reading here.
216
Eri Tanaka
AP resultatives in Japanese
In the previous section, we saw that PP resultatives in Japanese fulfill the

co-extensiveness condition: Threshold-telicity is provided by a path-taking
verb and a path-bounding PP, while BECOME-telicity is provided by
non-path-taking verb and a non-path-bounding PP. In AP resultatives in
Japanese as observed in section 2, however, the path property seems to be
irrelevant to the (un)grammaticality of the resultative formation; Japanese
only permits weak-control type resultatives, regardless of resultative predicates. In this section, I will provide a solution to the quandary, and argue that
the morphological marking of adjectives is crucial to the interpretation.
4.1
Value and property transitions
An AP resultative construction describes a transition from one state to another. The transition in general can be described as a shift from not A to A.
Vanden Wyngaerd (2001) points out that closed scale and non-closed/open
scale adjectives reveal different types of transition. Consider the following
Dutch examples from Vanden Wyngaerd (2001:72-73):
(53)
(a)
De fies
is
niet
leeg.
the bottle
is
not
empty
'The bottle is not empty'

(b)
Theo
is
niet
blij.
Theo
is
not
happy
'Theo is not happy.'
The adjective leeg 'empty' is a closed scale adjective, which is confirmed by

the compatibility of completely. On the other hand, the adjective blij 'happy'
is an open scale adjective.
(54)
(a)
0
empty
(b)
1/2
a-1
full
--
The negation of empty denotes any points on the empty-full scale, except for
the point of 0 (being the empty point), as depicted in (54a). The negation of
happy, on the other hand, cannot designate an intermediate point; not happy
cannot be located on the happiness scale (see (54b)). Following Vanden
Wyngaerd (2001), I will call these two different transitions value transition
and property transition, respectively.
Vanden Wyngaerd (2001) argues that if all resultatives were analyzed
according to the BECOME-theory, they would permit all types of transitions.
However, as Wechsler (2001) and Vanden Wyngaerd (2001) have observed,
217
open-scale adjectives are not permitted by strong-control resultatives: the

construction requires that the transition included be the one that ranges over
intermediate points (i.e. value transition). This is contrasted with the make
causative in Dutch, an apparent resultative construction, which not only allows open-scale adjectives like sad and happy, but also permits adjectives
modified by very:
(55)
(a)
Die
opmerking
that
remark have
heeft
me erg
verdrietig/blij
gemaakt.
me very
sad/happy
made
'The remark has made me very sad/happy.'

(b)
Dat
medicijn
heeft
me erg
ziek
gemaakt.
that
drug
has
me very
sick
made
'That drug has made me very ill.'
(Vanden Wyngaerd (2001: 74))
Contrast (55) with (56):

(56)
(a)
John hammered the metal {completely/halfway} flat,
(b)
??John hammered the metal very flat.
Vanden Wyngaerd (2001) argues that the make causative construction heads
the BECOME operator in its structure, which will, in principle, license any
kind of transition. Adjectives in resultative constructions, however, are confined to those that feature value transition, and thus, his argument goes, they
do not include a BECOME operator in their semantics and in syntactic
structures.
The same explanation can be given for Japanese AP resultatives. Recall
that Japanese weak adjectival resultatives, as observed in section 2.2, are not
sensitive to the closed-open scale distinction; moreover, as in the case of the
make causative construction in Dutch, adjectives modified by totemo 'very'
are also acceptable as shown in the following:
(57)
(a)
John-ga
kinzoku-o
J-NOM
metal-ACC completely
kanzen-ni
taira-ni
flat-ADjm/
nobasita
flatten
'John flatten the metal completely.'

(b)
John-ga
kinzoku-o
J-NOM
metal-ACC very
totemo
{taira-ni/usuku}
nobasita
{flat-ADJin//thin-ADJm/}
flatten
(Lit.) 'John flatten the metal very flat.'

'John made the metal very flat by pounding it. '
Why then does Japanese lack strong-AP resultatives? I will argue in the next
section that this is due to the morphological marking on the adjectives, following the same explanation we gave for the -ni locative in section 3.
218
Eri Tanaka
4.2
Morphological marking and the type of transition
I have observed in section 3.2 that in combination with the -ni locative, even
path-taking motion verbs such as iku 'go' and kuru 'come' cannot be modified by hanbun 'halfway'. This suggests that the -ni locative excludes path
semantics, because it only denotes a location. A similar observation can be
applied to AP resultatives. Consider first the following:
(58)
(a)
*John-ga
kinzoku-o
J-NOM
metal-ACC halfway
hanbun
taira-ni
flat
"ADjiii/
nobasita
flatten
'John flatten the metal halfway.'

(b)
*mizu'umi-ga
hanbun
kataku
kootta
lake-NOM
halfway
solid
froze
'The lake froze solid halfway.'
The verbs in (58) can, by themselves, be modified by hanbun 'halfway':

(59)
(a)
John-ga
kinzoku-o
J-NOM
metal-ACC halfway
hanbun
nobasita
flatten
'John flattened the metal halfway.'

(b)
mizu'umi-ga
hanbun
kootta
lake-NOM
halfway
froze
'The lake froze halfway.'
Note that the sentences in (59) may be interpreted to mean 'John flattened
half of the metal' or 'The half (region) of the lake has frozen', but this is not
the intended interpretation (see also footnote 14). That which is intended here,
of course, is that the degree of the flattening or freezing is on an intermediate
point.
The crucial factor is the morphological marking on the adjectives, a
property that English lacks. As observed in 2.1, the adjectives in Japanese
resultative constructions must be -ni (or -ku) inflected forms, and forms other
than -ni (i.e. -de or -da) cannot be permitted as a resultative predicate. I
propose now that adjectives in the -ni infinitive form only fit into a type of
property transition:
(60)
taira-ni 'flat'
1
-.flat =>flat
On the flatness scale, not flat ranges over any points except 1, as in English.
However, taira-ni 'flat ADJ ,/ denotes only the final state of the transition, and
cannot include the path to the final state. In other words, taira-ni cannot
219
include the transitional path from not flat to flat, which is revealed in the
hanbun 'halfway' test presented above.
Strong-control resultatives must be formed through the bounding of the
path. To bound a path, as the co-extensiveness condition requires, the
bounding expression itself must tolerate path semantics. The above argument
suggests that Japanese adjectives in infinitive form do not tolerate transitional
paths, as in locative -ni. The lack of path semantics yields the ungrammaticality of strong resultatives in Japanese.
I have noted above that the semantics of BECOME licenses all types of
transitions. However, we have observed in the previous section that motion
verbs such as arrive do not license -made 'up to', as the verb does not include
a path to be bounded by the postposition. This leads to the conclusion that
with BECOME-telicity, depending on the semantics of the verbs, the event
may or may not have a 'duration', and the co-extensiveness condition poses
the restriction on the composed resultative predicate:
(61)
(a)
(b)
event:
========= |
path/scale:
========= |
event:
path/scale:
(61) depicts the possible combinations: I indicates the telic point, and ===
indicates that there is a duration/path, while illustrates the lack of a duration/path.
is a depiction of the resultant state. Thus, with BECOME
semantics, the co-extensiveness condition works as a principle, and the
determination of the grammaticality is reduced to the language specific
features of the lexical items included.
5 Conclusion
I have argued in this paper that the notion of 'path' plays a crucial role in
determining the availability of aspectual composition. I have centered on
resultative constructions in Japanese and English, and have shown the following:
(i)
(ii)
There are two types of telic formation: BECOME and end of path, and
each of them correspond to the weak and strong dichotomies proposed
by Washio (1997).
Japanese PP resultatives exhibit both types of telic formations. The
semantics of the postpositions is crucial in determining the grammaticality of the sentence, in accordance with the co-extensiveness condition
proposed by Wechsler (2001).
220
Eri Tanaka
(iii) T h e c o - e x t e n s i v e n e s s c o n d i t i o n c a n also e x p l a i n J a p a n e s e A P resultatives, which s e e m p r o b l e m a t i c as cases f o r the condition.

T h e p r e s e n t p a p e r h a s , I b e l i e v e , t w o i m p o r t a n t i m p l i c a t i o n s to the t h e o r y of
t h e s e m a n t i c s of e v e n t s . T h e a p p r o a c h t a k e n h e r e f o c u s e s o n t h e i n t e r a c t i o n of
t h e s e m a n t i c s of v e r b s a n d t h e s e m a n t i c s of a d j e c t i v e s and p r e / p o s t p o s i t i o n s .
T h e e v e n t d o m a i n is ' e n r i c h e d ' b y entities of t h e other d o m a i n s , and the w o r k
p r e s e n t e d h e r e h a s a t t e m p t e d to e l u c i d a t e t h e r e l a t i o n s a m o n g t h e m . T h i s
article h a s also d e a l t w i t h t h e s e q u e s t i o n s f r o m a c r o s s - l i n g u i s t i c p e r s p e c t i v e .
It h a s b e e n o b s e r v e d that the d i s t r i b u t i o n of resultatives varies a m o n g lang u a g e s ; b u t at least to m y k n o w l e d g e , t h e r e has n o t b e e n a s t u d y that a t t e m p t s
to e x p l a i n w h y . T h e a n a l y s i s g i v e n h e r e s h o w s that a g e n e r a l p r i n c i p l e such as
t h e c o - e x t e n s i v e n e s s c o n d i t i o n m a y b e r e s p o n s i b l e f o r t h e availability of t h e
c o n s t r u c t i o n , i n t e r a c t i n g with the l a n g u a g e s p e c i f i c s e m a n t i c f e a t u r e s of t h e
lexical items.
References
Carrier, J. and J. H. Randall (1992): 'The Argument Structure and Syntactic Structure of Resultatives.' Linguistic Inquiry 23: 173-234.
Dowty, D. (1979): Word Meaning and Montague Grammar: The Semantics of Verbs
and Times in Generative Semantics and in Motague's PTQ. Dordrecht: Reidel
Publishing Company.
Dowty, D. (1991): 'Thematic Proto-Roles and Argument Selection.' Language 67:
547-619.
Hay, J., C. Kennedy and B. Levin (1999): 'Scalar Structures Underlies Telicity in
'Degree Achievements'.' In T. Matthews and D. Strolovitch (eds.) The Proceedings of Semantics and Linguistic Theory 9. Ithaca: CLC Publications.
Jackendoff, R. (1996): 'The Proper Treatment of Measuring Out, Telicity, and
Perhaps Even Quantification in English.' Natural Language and Linguistic
Theory 14: 305-354.
Kageyama, T. (1996): Doosi Imiron: Gengo to Ninchi no Setten [Verbal Semantics:
The Interface between Language and Cognition], Tokyo: Kurosio.
Kageyama, T. and Y. Yumoto (1997): Gokeisei to Gainenkoozoo [Word formation
and Conceptual Structure], Tokyo: Kenkyusha.
Krifka, M. (1989): 'Nominal Reference, Temporal Constitution and Quantification
in Event Semantics.' In R. Bartsch, J. van Benthem, and P. van Emde Boas
(eds.) Semantics and Contextual Expressions. 75-115. Dordrecht: Foris.
Krifka, M. (1998): 'The Origins of Telicity.' In S. Rothstein (ed.) Events and
Grammar. 197-235. Dordrecht: Kluwer.
221
Krifka, M. (1992): 'Thematic Relations as Links between Nominal Reference and

Temporal Constitution' in: I. Sag and A. Szabolsci (ed.) Lexical Matters. Stanford: CSLI: 29-53.
Landman, F. (2000): Events and Plurality: The Jerusalem Lectures. Dordrecht:
Kluwer.
Levin, B. and M. Rapparpot Hovav (1995): Unaccusativity. Cambridge, MA: MIT
Press.
Link, G. (1983): 'The Logical Analysis of Plurals and Mass Nouns: A Lattice
Theoretical Approach.' R Bauerle, C. Schwarze and A. von Stechow (eds.)
Meaning, Use and the Interpretation of Language. 303-323. Berlin/New York:
Mouton De Gruyter.
Parsons, T. (1990): Events in Semantics of English. Cambridge, MA: MIT Press.
Rappaport Hovav, M. and B. Levin (2001): 'An Event Structure Account of English
Resultatives.' Language 77: 766-797.
Rothstein, S. (2004): Structuring Events: A Study in the Semantics of Lexical
Aspect. Oxford: Blackwell.
Takezawa, K. (1993): 'Secondary Predication and Locative/Goal Phrases.' In N.
Hasegawa (ed.) Japanese Syntax in Comparative Grammar. 45-77. Tokyo:
Kurosio.
Takezawa, K. (2001): 'Nihongo-no Jootaikijutu Nizijutubu to Toogohanchu (Japanese Depictive Secondary Predicates and Syntactic Category).' KLS 22:
257-267. (Proceedings of the 26th Annual Meeting of Kansai Linguistic Society).
Tanaka, E. (2003): 'Event Composition and a Path in Japanese.' In B. Agbyani, P.
Koskinen and V. Smiian (eds.) Proceedings of the 31st Western Conference on
Linguistics. WECOL 14: 282-293.
Tenny, C. (1994): Aspectual Roles and the Syntax-Semantics Interface. Dordrecht:
Kluwer.
Tenny, C. (2000): 'Core Events and Adverbial Modification.' In C. Tenny & J.
Pustejovsky (eds.) Events as Grammatical Objects. 285-334. Stanford: CSLI.
Torota, C. (1998): 'Verbs of Inherently Directed Motion are Compatible with Resultative Phrases.' Linguistic Inquiry 29: 338-345.
Tsujimura, N. (1991): 'On the Semantic Property of Unaccusativity.' Journal of
Japanese Linguistics 13: 91-116.
Uehara, S. (1998): Syntactic Categories in Japanese: A Cognitive and Typological
Introduction. Tokyo: Kurosio.
Ueno, S. and T. Kageyama (2001): 'Ido to Keiro no Hyogen [Expressions of Motion
and Path].' In T. Kageyama (ed.) Doshi-no Imi-to Kobun [Verbal Semantics and
Constructions], 41-68. Tokyo: Taishukan.
Vanden Wygaerd, G. (2001): 'Measuring Events.' Language 77: 61-90.
222
Eri Tanaka
V endler, Z. (1957): 'Verbs and Times.' The Philosophical Review 66: 143-160.
Verkuyl, H. (1993): A Theory of Aspectuality. Cambridge: C a m b r i d g e University
Press.
Washio, R. (1997): 'Resultatives, Compositionality and Language
Variation.'
Journal of East Asian Linguistics 6: 1-49.

Wechsler,
S.
(2001):
'An
Analysis
of
English
Resultatives
Under
the
Event-Argument H o m o m o r p h i s m Model of Telicity.' In Proceedings of the 3rd

Workshop on Text Structure.
Y o n e y a m a , M. (1988): 'Motion Verbs in Conceptual Semantics.' Bulletin of Faculty of
Humanities, Seikei University 22: 1-15.
Eric McCready (Aoyama Gakuin) and Chiyo Nishida (Austin)
Reflexive intransitives in Spanish and

event semantics
1 Introduction
The reflexive clitic se/si in Romance languages assumes a great variety of functions, illustrated in (l)-(4) using Spanish data:
(1)
Reflexive/Reciprocal se:
(a)
(2)
Jos y Maria SE miraron

en el espejo.
Joe and Mary SE looked.at-3pl in the mirror
'Joe and Mary looked at themselves/each other in the mirror.'
(b) Jos y Maria SE mandaron un mensaje.
Joe and Mary SE sent-3pl a message
'Joe and Mary sent themselves/each other a message.'
Middle/Passive se:
(a)
(3)
(4)
Esta camisa SE lava

fcilmente.
This shirt SE washes easily
'This shirt washes easily.'
(b) SE construyeron muchas carreteras el ao pasado.
SE built-3pl
many highways the year past
lot of highways were built last year.'
Anticausative se:
Gracias a Dios, los cristales no SE rompieron.
Thanks to God, the crystals neg SE broke-3pl
'Thank God, the crystals did not break.'
Inherent se:
Mara SE arrepinti
de haber castigado a su hijo.
Mary SE regretted-3sg of having punished to her son
'Mary regretted having punished her son.'
Here all examples have a 3rd person subject, so, the clitic form used is consistently se. The reflexive clitic in fact takes different forms in accordance with the
person/number value of the subject, as shown in (5).
224
(5)
Reflexive Clitic Paradigm in Spanish:
1st
sg
ME
pi
NOS
2nd
TE
OS
3rd
SE
SE
Yo i ME miro.
look at myself'
T i TEi miras.
'You look at yourself'
E/EUcii SEj liira.
'He looks at himself/herself'
Nosotro/asi NOS miramos.

'We look at ourselves'
Vosotro/asi OS miris.
'You look at yourselves'
Elios/EUasi SE miran.
'They looked at themselves'
The Spanish reflexive se is also used to overtly mark telicity in transitive constructions (Nishida 1994, Zagona 1996, De Miguel & Fernndez Lagimilla 2000,
Sanz 2000, i.a. ). Transitive sentences with se (hereafter RTs) must be telic, and
thus require a quantized direct object NP (Krifka 1992), as shown in (6). However, transitive sentences without se show no restrictions on telicity (7):
(6)
(7)
Juan SE fum
{dos piiros/*piiros}
Juan SE smoked-3sg {two cigars/cigars}
'John smoked two cigars/cigars.'
Juan fum
{dos puros/puros}
Juan smoked-3sg {two cigars/cigars}
'John smoked two cigars/cigars.'
The present paper considers a less well-known class of cases which we call
'reflexive intransitives' (Ris, henceforth), where the reflexive clitic se appears
with intransitive verbs, as shown in (8a) and (8b).
(8)
(a)
(b)
Hoy SE march
Juan del
pueblo.
Today SE marched-3sg John from the village
'Today John went away from the village.'
Hoy SE muri
Juan.
Today SE died-3sg John
'Today John died.'
We begin by examining the paradigm of RIs for regularities and show that all
RIs share the property of denoting achievements, which in some cases induces
aspectual coercion. We further show that three issues arise with RIs that are not
present for transitive verbs: the requirement of RIs for quantized subjects, then
introduction of an additional argument that stands in certain relations with the
described event, and then interpretation with respect to transitions of particular
types.
The structure of the paper is as follows. In section 2, we lay out some background on transitive constructions with se (RTs), concentrating on the telicity
requirement for objects. Next, in section 3.1, we provide data on RIs and some
descriptive generalizations about them. Section 3.2 considers some previous
Reflexive intransitives in Spanish and Event Semantics
225
accounts of RIs; the data in section 3.1 shows each of these to be ultimately
inadequate. Our own analysis is presented in section 4. Section 5 concludes.
2 Background: transitive constructions with Se

Nishida (1994) observes that reflexive transitives (RTs) are most commonly distributed among three semantic classes: consumption of a spatial or temporal
object (9), creation of an abstract 'performance object' (10), and acquisition by
an agent of a (possibly abstract) object (11).
comer+se ima manzana
chupar+se un caramelo
fumar+se un puro
pasar+se un da entero en la isla
recitar+se un poema
cantar+se ima cancin
escuchar+se ima simfona
ver+se toda la pelcula
ganar+se ima lotera
robar+se un collar de diamantes
aprender+se toda la leccin
estudiar+se un captulo
(9)
(10)
()
'to eat up an apple'

'to suck up a candy'
'to smoke a cigar'
'to spend an entire day on the island'
'to recite a poem'
'to sing a song'
'to listen to a symphony'
'to see the whole movie'
'to win a lottery'
'to steal a diamond necklace'
'to learn all the lesson'
'to study a chapter'
The majority of the predicates are accomplishments with the direct object constituting an incremental theme (Dowty 1991). However, aspectual se also occurs
with some stative verbs like saber 'to know' and conocer 'to know/to be acquainted with', as shown in (12). Here, the stative is coerced to an inceptive
transition into the corresponding state (cf. de Swart 1998). It has been further
claimed that stative predicates can only appear with aspectual se if the sentence
they appear in can be interpreted as describing the resultative state of an accomplishment. This point will be relevant to the discussion of RIs in the next
sections.
(12) (a)
(b)
Juan SE sabe toda la leccin.

John SE knows all the lesson
'John knows all the lesson.'
Juan SE conoce toda la ciudad.
John SE knows all the city
'John is familiar with the whole city.'
226
3 Intransitive constructions with se: 'reflexive intransitives'

3.1 Data and descriptive generalizations
Reflexive intransitives are comprised of a non-transitive verb (a verb not taking
a direct or indirect object) and se. The base verb for an RI can be purely intransitive (cf. dormir 'to sleep'), take a goal or source argument (cf. ir 'to go'
or salir 'to get out') or take a predicative complement (cf. estar 'to be'). The
data to be presented below show that, like RTs, RIs appear with both dynamic
and stative predicates. Regardless of the aspectual properties of the base verb,
dynamic RIs are achievement predicates denoting transitions, while stative RIs
denote the result states of transitions.
A non-exhaustive list of RIs follows, where dynamic RIs are classified on
the basis of how the transition comes about. 1 For the most part, the essential
meaning of the reflexive intransitive is the same as that of the base verb. If the
se-cliticization yields a meaning different from the base verb, (cf. ir+se and
dormir+se), this meaning is shown in parentheses. 2
Class I: Transition caused by a (possibly abstract) object moving away from/out
of a location:
(13) escapar+se 'escape+se':
El chico SE (les)
escap
de la casa,
the boy SE (CL.dat.3pl) escaped-3sg from the house
'The boy got away from home (on them).'
(14) ir+se 'go+se (go away)':
Jos SE (les)
ie
del pueblo en 1950.
Joe SE (CL.dat.3pl) went-3sg of.the village in 1950
'Joe went away from the village (on them) in 1950.'
(15) marchar+se 'march/go+se (march/go away)':
El chico SE (les)
march
de la casa hoy.
The boy SE (CL.dat.3pl) marched-3sg from the house today
'The boy went away from home (on them) today.'
(16) pasar+se 'pass+se':
SE les
pas
ima gran oportunidad
SE CL.dat.3pl passed-3sg a great opportunity
great opportunity slipped away from them.'
(17) volar+se 'fly+se (fly away)':
SE (les)
vol
el papel.
SE (CL.dat.3pl) flew-3sg the paper
'The paper flew away (from/on them).'
Note that we include only verbs that can appear with se for speakers of both Peninsular and
Latinamerican Spanish.
Note that both animate and inanimate subjects are possible with aspectual se.
(18) caer+se 'fall+se':

El chico SE cay
del
rbol
the boy SE fell-3sg of.the tree
'The boy fell from the tree.'
(19) resbalar+se 'slip+se (slip out)':
SE le
resbal
el jarrn de las manos.
SE CL.dat.3sg slipped-3sg the vase from the hands
'The vase slipped out of his/her hands.'
(20) venir+se 'come+se (come away)':
Jos se vino
de
su patria cuando era
nio
Joe SE came-3sg from Iiis country when was.3sg boy
'Joe came away from Iiis country when he was a boy.'
(21) saltar+se 'jimip+se (pop out)':
SE le
saltaron
los ojos a la mueca.
SE CL.dat.3sg jmnped-3pl the eyes to the doll
'The doll's eyes popped out.'
Class II: Transition caused by an object reaching a location:

(22) venir+se 'come+se':
SE (le)
vino
la noche.
SE (CL.dat.3sg) came-3sg the night
'The night came upon him/her.'
(23) volver+se 'conie.back+se':
SE (le)
volvieron las cartas que haba
enviado.
SE (CL.dat.3sg) retuni-3pl the letter that s/he.had sent
'The letters s/he had sent came back (to/on him/her).'
(24) subir+se 'rise+se':
SE le
subi
el vino a la cabeza
SE CL.dat.3sg rose-3sg the wine to the head
'The wine went up to his/her head.'
(25) trepar+se 'climb+se':
El chico SE trep
hasta lo ms alto del
rbol,
the boy SE clinibed.up-3sg till the very top of.the tree
'The boy climbed all the way up to the very top of the tree.'
Class III: Transition caused by a temporal object reaching an endpoint:

(26) pasar+se 'pass+se (pass by)':
SE les
pas
el tiempo volando.
SE CL.dat.3pl passed-3sg the time
flying
'The time went by flying.'
(27) terminar+se 'end+se':
SE (le)
terminaron las vacaciones muy pronto.
SE (CL.dat.3sg) ended-3pl the vacations very soon
'(His/her)/The vacation came to an end (on him/her) very soon.'
227
228
Class IV: Transition caused by an object coming into being:

(28) aparecer+se 'appear+se (show up)':
SE (le)
apareci
im fantasma por
aqu.
SE (CL.dat.3sg) appeared-3sg a ghost
around here
ghost appeared (on him/her) around here.'
(29) ocurrir+se 'occur+se':
SE le
ocurri
ima idea brillante.
SE CL.dat.3sg occurred-3s an idea brilliant
brilliant idea occurred to him/her.'
Class V: Transition of an object undergoing a change of state'.

(30) dormir+se 'sleep+se (fall asleep)':
El chico SE (le)
durmi inmediatamente,
the boy SE (CL.dat.3sg) slept-3sg immediately
'The boy fell asleep immediately (on/for him/her).'
(31) morir+se 'die+se':
SE (les)
muri
el padre.
SE (CL.dat.3pl) died-3sg the father.
'The/(their) father died (on them).'
(32) quedar+se STATE 'remain+se STATE (become STATE)':
Su marido SE (le)
qued
ciego.
her husband SE (CL.dat.3sg) remained-3sg blind
'Her husband became blind (on her).'
Class VI: Resultative States of Transitions:

(33) e star+se STATE/LOC 'be+se STATE/LOC':
(a)
(b)
El chico
the boy
'The boy
El chico
the boy
'The boy
(34) quedar+se
(a)
(b)
SE (le)
estuvo callado todo el da
SE (CL.dat.3sg) was-3sg silent all the day
was quiet (for her/him) all day long.'
SE estuvo iera mientras hablbamos.
SE was-3sg outside while
we.were.speaking
remained outside while we were talking.'
STATE/LOC 'remain+se STATE/LOC':
Jos SE (le)
qued
dormido por ima hora
Joe SE (CL.dat.3sg) remained-3sg asleep for an hour
'Joe remained asleep for an hour (for him/her).'
La cartera SE qued
en casa todo el da
the wallet SE stayed-3sg at home all the day
'The wallet was left at home all day long.'
Some generalizations can be made about RIs given the above data. First, RIs
cluster around those predicates commonly characterized as imaccusative, which
229
take a theme subject NP. Second, the above predicates are all achievements or
the resulting states of transitions associated with those achievements, cf. the
predicates in Class VI. Finally, unlike RTs, RI sentences do not involve an incremental theme; the homomorphic relation observed in RTs between the object
denoted by the direct object and the event does not exist here. This holds even
for the subject argument, which might be thought to behave like a direct object
given the nature of imaccusative verbs.
However, the requirement for quantization in RTs does appear, although here
it is a requirement on subjects. While RIs are compatible with a quantized subject NP, as in (35a), (35b) shows that nonquantized subjects are dispreferred.
Note that this requirement is limited to RIs; the non-reflexive counterparts take
a quantized or nonquantized subject NP (35c).3
(35) (a)
(b)
(c)
Hoy SE murieron los hombres/dos hombres en la residencia.

today SE died-3pl the men/two men
at the residence
'Today the men/two men died at the residence'
*?Hoy SE murieron hombres en la residencia,
today SE died-3pl men
at the residence
Hoy murieron los hombres/dos hombres/hombres en la residencia
today died-3pl the men/two men/men
at the residence
'Today the men/two men/men died at the residence'
3.2 Previous analyses of RIs

Zagona (1996), Sanz (2000) and De Miguel & Fernndez Lagimilla (2000) attempt to give both RTs and RIs a unified treatment. We consider each of then
attempts in turn.
The focus of Zagona's paper is to establish the link between the semantics
(= eventuality descriptions) and the syntax (= reflexivity) of the constructions at
issue. Zagona claims that "se only occurs with predicates whose object undergoes a change of state that marks the end of an event" (1996:481). According to
Zagona, then, predicates which may appear with se are exclusively drawn from
the class of what Pustejovsky (1995) calls 'transitions', formally represented as
in (36a) by Zagona and as hi (36b) by Pustejovsky himself. Note that Zagona's
eventualities are taken to be specified for spatiotemporal location:
(36) (a)
(b)
3
[ e ! . . . eg ]
[T [ S } } (: Transition; : Process; S: State)
Here another restriction on aspectual se with intransitive Vs should also be noted: it prefers
definite subjects to indefinite ones. Thus in (35), replacing los/dos hombres with unos hombres
'some men' is not so good. The reason, we think, is that the subject must be understood to stand
in some relation with either the utterer of the sentence or some individual already salient in the
discourse; for this relation to be clear, the sentential subject must also be salient, and so introduced
as a definite (cf. Heim 1988). This point is discussed further in a later section. Note also that dos
hombres is most naturally understood partitively here, probably for similar reasons.
230
Zagona explains that for RTs, the direct object argument undergoes the change
of state transition associated with the event, moving in space from e^ to eg,
where the event culminates; as the event culminates, the direct object and the
subject end up in the same temporal location. For RIs, it is the surface subject
that undergoes a change of state (= change of location). Extending Zubizarreta's
analysis of Spanish se (1987), Zagona proposes that se in RTs and RIs is a verbal operator that binds a temporal argument (i.e. eg) instead of a VP internal
argument. Since eg is the final temporal location for the subject in both RTs
and RIs, se as the binder of eg agrees with the subject, giving rise to structural
reflexivity. Zagona's proposal elegantly correlates syntax and semantics, but is
empirically inadequate for RIs. Zagona only considers a small set of motion
verbs: irse 'go away', caerse 'to fall (from somewhere)', subirse 'to go up', and
treparse 'to climb up'. While it is adequate for these verbs, the account does
not generalize in an obvious way to many of the RI examples shown above, in
which the correlation between transitions and movement is less clear.
De Miguel and Fernndez Lagimilla (2000; hereafter DMFL) analyze se in
RTs and RIs as "an aspectual operator that shows that the event culminates in
one point that leads into a change of state". 4 They point out that Zagona's analysis is too coarse-grained, since some predicates denoting transitions can take
se but others cannot. To resolve the problem, they divide Pustejovsky's 'transitions' into five separate event types; for them, only those classes that include
DMFL's redefined 'transition' can take se. Unfortunately, this proposal is not an
improvement on that of Zagona. First, DMFL's criteria for classifying verbs into
categories are unclear, as Martinez (2003) points out. For instance, the similar
predicates llegar 'to arrive' venir (se) 'to come' are placed in different classes,
although both predicates are compatible with both a source and a goal argument.
Thus some distinctions made by DMFL appear unprincipled. Another problem
is common to both Zagona and DMFL. Both assmne that se makes no contribution to the aspectual composition of the RIs. This is true in some cases, but since
V+s<? does sometimes have a different meaning than its clitic-free counterpart,
this prediction is incorrect.
The final analysis we consider is that of Sanz (2000), who proposes a functional projection EventP which sits on top of the TP and whose head is specified
for telicity among other features. She claims that se in both RTs and RIs is a
marker of telicity, more specifically of accomplishments, and is inserted into the
head of the EventP (Evt) to check and erase the feature [+telic]. Although Sanz's
proposal is an interesting one, it is empirically limited. First, she considers an
extremely small set of data, i.e, caerse 'to fall', morirse 'to die', and a few motion verbs. Second, her claim that RI sentences are accomplishments like RTs
cannot hold in view of the data seen above.
"El se es un operador aspectual que seala que el evento culmina en un punto desemboca en un
cambio de estado." (2000: 28).
231
Finally, none of the three analyses have any account for two important properties associated with RIs: first, that the subject NP must be quantized, and that
RIs can co-occur with the dative of interest. The first property can be tied to the
telicity requirement on the sentence introduced by se\ we timi to this issue first
in the analysis presented in the next section.
4 Analysis
Our analysis is meant to account for three basic facts about RIs: the telicity
requirement on the sentence se appears in, which we take to result from a requirement for a quantized subject (similar to the quantized object requirement
on RTs), the possibility of the 'dative of interest' with ie-cliticization, and the
association with a path of the temporal point at which the transition associated
with the verbal achievement occurs. We explore each point in more detail and
provide an analysis in the following sections.
4.1 Telicity and quantized subjects
Why is the reflexive intransitive resistant to a non-quantized subject NP? First
observe the pair of English sentences in (37), where the achivement arrive is the
main predicate.
(37) (a)
(b)
Two travelers from China will arrive at the train station,

Travelers from China will arrive at the train station.
Both sentences are true in situations where the arrivals of the travelers is simultaneous or spread over time. In (37a), the event described by the proposition
under the scope of the modal operator is complete only when the two travelers
have both arrived; thus no subevent of the described event has this property. As
a result, the sentence is interpreted as telic. However, the bare plural subject
in (37b) means that no definite endpoint can be determined for the described
situation, for more travelers may always arrive; thus subevents may also instantiate the sentence, which is therefore atelic. For an intransitive sentence with
an achievement predicate to be interpreted unequivocally as a telic situation, a
quantized subject NP is required, in a way similar to the telicity that results from
a quantized direct object (Krifka 1992).
As shown above, dynamic RIs are achievement predicates that denote a transition (sometimes as a result of aspectual coercion). RIs resist a non-quantized
subject NP like a bare plural because such a subject allows the sentence to be
interpreted as temporally unbounded, i.e. atelic. Ultimately, this means that RIs
must depict telic situations, either an achievement or a series of achievements
that extends over a definite time period.
232
We model this telicity requirement as a presupposition on the events denoted

by sentences. Our specific assumptions are as follows. We follow Miller and
Sag (1997), among others, in analyzing reflexive se as a clitic attached to the
verb: thus e.g. se muri has the structure [v [ci se ] [v murio } }. A piece of
evidence in favor of this move is the fact that se cannot scope over conjoined
VPs:
(38)
*Juan se escap
de la casa y ie
del
pueblo
John SE escaped-3sg from the house and went.3sg from.the village
Intended but unavailable: ' John escaped the house and left the village.'
We further assmne that se is of type ( (ev ,t),(ev,t)),

and so composes with the
denotation of a sentence before existential closure of the event argument.
In order to provide a (weakly) compositional semantics for se, we utilize
the framework of Lexical-Functional Grammar (Bresnan 2001). LFG postulates that syntactic information is represented at multiple levels: among others,
c(onstituent)-structure and f(unctional)-structure. Mappings among the various
levels are accomplished through functional annotations on syntactic nodes. (39)
is an example of how this mapping works, which we simplify by ignoring agreement information. 5 ' 6
(39) (a)
(b)
Juan muri
John died-3sg
'John died.'
S
muno
(c)
'die(subj)'
'JUAN'
PAST
In the sequel we will omit annotations of the form f = r e s t r i c t i n g attention to annotations that
are associated with an f-structural attribute.
Note that morir 'to die' is an unaccusative verb.
233
The 'glue semantics' for LFG (cf. Dalrymple et al. 1997, Dalrymple 2001),
reads semantic information off f-structures like that in (39b) to a third level,
s(emantic)-structure, indicated with a subscript below. We use the version
of glue semantics of Dalrymple (2001); in this theory, semantic objects take
the fonn of -calculus expressions paired with linear logic formulas (Girard
1987). Glue semantics uses the multiplicative fragment of linear logic: the only
connectives are & 'multiplicative conjunction' and 'linear implication.' Note
that LL verifies the following equivalence
A &
C <==>
A(BC)
so implications with conjoined antecedents may be reanalyzed as iterated implications.

Linear logic is resource-sensitive; a premise may be used once and only once
in a derivation. Girard provides the example of buying a pack of cigarettes. If a
pack of cigarettes costs S3 ($3 poc) and one has S3, then one can buy a pack
of cigarettes; doing so, however, eliminates both one's S3 and the possibility of
buying another pack. This derivation is shown in (40).7
(40)
S3, S3
poc \= poc
In glue semantics, semantic composition occurs within the A-expressions in tandem with derivations in the glue logic (using the Curry-Howard isomorphism).
Since LL is resource-sensitive, expressions used in a derivation cannot be used
again, as one would want for deriving the meaning of linguistic expressions.
This also, of course, means that compositionality is upheld.
We now give a derivation for the clitic-free sentence Juan muri 'John died'
shown above in ((41a)). The f-structure provides the following premise set for
the glue logic derivation:
Juan :
SUBJ]ct, Ax.[died(x)] : [ SUBJ]ct of
\= Ax.Ae.[died(e,x)](Juan)
We assmne that (41a) has the structure in (41b), which maps to the f-structure in
(41c):
(41) (a)
Juan se muri
John SE died.3pl
'John died.'
We use the semantic entailment relation |= rather than the proof-theoretic h ; the completeness of
T T m a k e s t h i s an l i n n r o h l e m a t i c m o v e
234
(b)
NP
VP
SUBJ=J.
=]
I
V
CI
Juan
J.GADJ
se
(c)
PRED
'DIE(SUBJ)'
SUBJ
'JUAN'
TENSE
PAST
ADJ
I PRED
'SE'j
Se is analyzed as an adjunct here. As shown below, it semantically applies to

the sentence (= set of events) in its scope. The sentence is defined only if the
sentence denotes a set of telic events. We assume that interpretation functions are
partial, and that presupposition failure causes the sentence to become undefined.
See e.g. Beaver (2002) for discussion of this account of presuppositions.
Given the standard assumption that existential closure operations apply at the
end of the semantic derivation, we need to clarify how se is able to determine
whether a set of events is telic or not. We define two functions to aid us in
making this work out. The first, Einst, is a choice function on sets of events
(where is a set of eventualities, i.e. a function of type (ev, t)):
Einst(Xe.[ip(e)])
= e'
Thus Einst picks a characteristic eventuality from the set. We now need a function that tells us whether this eventuality is telic. We use one standard definition
of felicity: that no event which is a subpart of Einst(Xe[tp(e)]) for a telic predicate is an event of type (Smith 1997). Now we define the predicate telic}
telic{ip) < > Ve'[e' Einst(ip)
> np(e')\
Given these two functions, we can make a first pass at giving a lexical entry
for se. Here | represents the semantic projection of a S-level constituent,
i.e. something of type t. The material within curly braces {} represents the
presupposition of se: that the sentence it applies to is telic.
8
Kjell Siebo notes (p.c.) that this approach to the definition is somewhat inelegant. It suffices for
our purposes here, however.
[se] =
XV{eVt){telic(V)}.[Xe.[V(e)}}
:\
235
Thus, se denotes a function from sets of events that returns the same sets of
events just in case each eventuality in the set is telic.
We now show how this definition applies to examples with and without quantized subjects (ignoring steps of meaning computation irrelevant to se for simplicity). Node annotations are omitted. We also assume an operation of event
existential closure like the assertion operator of Krifka (1992): 9
(42) RI with quantized subject:
(a)
(b)
Dos hombres se murieron

Two men
SE died.3pl
'Two men died.'
S
D
I
Dos
(c)
PRED
I
hombres
Cl
DIE(SUBJ)
SPEC
PRED
TWO
SUBJ
PRED ' M E N '
ADJ
(d)
PRED ' S E '
[XP{ev,t){telic{V)}.[\e.[V{e)}\]
( A e . [ 3 " [ m e n { X ) died(e,X)
=> 3e\X[men(X)
died(e,X)
Card(X)
Card(X)
= 2]])
= 2]]
Since the sentence is telic, it passes the 'filter' imposed by the presupposition.
We now consider an example with a nonquantized subject. For sentences with
bare plurals, SV word order is bad independent of the presence of se\ for this
reason, we use the VS word order, which we assume to be generated with the
subject NP as complement of the (cliticized) verb.
Note that the word order in (42) is used simply for symmetry with other examples. An ordering
in which the subject follows the verbse murieron dos hombresis probably more natural. This
point does not affect the discussion.
236
(43) (a) *Se murieron hombres.

SE died.3pl men
'Men died.'
(b)
S
VP
(c)
NP
CI
se
murieron
hombres
[\V{ev,t){teUc(V)}.[\e.[V{e)\\\
A died(e, X)}})
(Ae.[3A"[men(A")
undefined.
Silice the sentence is not telic, it does not support the presupposition; so the
sentence is undefined.
It would also have been possible to adjoin se to VP and give it a denotation of
a higher type. Our syntactic assumptions did not allow this, but such a derivation
would also have been legitimate.
A nice consequence of modelling se' s telicity requirement as a presupposition
is that it is predicted to still hold when se appears in the scope of semantic
operators, which indeed seems to be the case.1"
(44) (a)
(b)
(c)
* No es el caso que se hayan

muerto hombres
no is the case that SE have.SUBJ.3pl died
men
'It is not the case that men died.'
* Es posible que se hayan
muerto hombres
is possible that SE have.SUBJ.3pl died
men
'It is possible that men died.'
* Juan cree
que se murieron hombres
John believe.3sg.pres that SE died.3pl men
' John believes that men died.'
4.2 Dative of interest

A property possessed by RIs, but not by se-less intransitives, is the possibility of
using the so-called 'dative of interest', which is realized in the fonn of a dative
clitic or a dative elitic/NP pair, as in (45) and many of the sentences used in the
RI classification above.
10
Again, we vary word order here for the sake of naturalness.
(45) (a)
(b)
(c)
237
SE nos
fue
el tiempo volando.
SE CL. lpl went-3sg the time flying
'The time went by on us flying.'
SE le
resbal
el jarrn de las manos.
SE CL.dat.3sg slipped-3sg the vase from the hands
'The vase slipped out of his/her hands.'
El nio SE le
qued
dormido a la mam,
the boy SE CL.dat.3sg remained-3sg asleep to the moni
'The boy fell asleep on the morn.'
The dative of interest in intransitive reflexives represents either an individual

who conceptualizes the event (45a), or one who gets indirectly involved in/affected by the event by holding some relationship of 'interest' with a direct participant in the event: e.g. being the possessor of an involved body part 'the hands'
(45b), or a kinship relation (45c). Interestingly, the addition of the dative argument is only possible with the reflexive version even in cases when there is no
apparent semantic difference between the reflexive and non-reflexive versions,
as illustrated below. These pairs of sentences indicate that the dative of interest in reflexive intransitives is not licensed by telicity itself, because both the
reflexive intransitive and its non-reflexive version depict telic situations.
(46) (a)
Hoy SE le
muri el padre a Jos.
Today SE CL.dat.3sg died-3sg the father to Joe.
'Today Joe's father died (on him)/Joe had Iiis father die on him'
(b) *Hoy le
muri el padre a Jos,
today Cl.dat.3sg died-3sg the father to Joe
(47) (a) Su marido SE le
qued
ciego
en el incendio.
Her husband SE CL.dat.3sg remained-3sg blind-masc.sg in the fire
'Her husband became blind on her in the fire'
(b) *Su marido le
qued
ciego
en el incendio.
Her husband CL.dat.3sg reniained-3sg blind-masc.sg in the fire
Note that this behavior, in which an argument position is added, is the opposite
of that seen in 'argument' clitics, which satinate one argument of their predicates
(see Nishida 1991 on Spanish and Monachesi 1999 on Italian, among many others). We take the above facts to indicate that reflexive se simply adds a possibly
implicit argument position to the predicate it appears with.11 This argument is
related to the event described by the sentence by an underspecified relation R
which receives its value from context. Thus we modify the basic lexical entry
for se provided above, as follows:
11
Dalina Kalluli (p.c. ) points out another possibility: that se introduces an additional event argument
that is associated with an individual then expressed as the clitic. While this approach is interesting,
since the nature of this event is not totally clear to us we choose to analyze le as an additional
argument of the main predicate.
238
(48)
[se] = \x.[\V{ev,t){telic(V)}.[\e.[V(e)
[OBL] c t
- 4
i?(i,e)]]]:
We treat the dative clitic licensed by the se-constructions as an oblique argument subcategorized for by se (a move preserving the intuitions described by
Castao 1999). For cases in which the dative argument is not realized by an
overt argument, we assmne a type-shifting operation Dsat which saturates the
argument place added by se:
(49) Dsat =
ev.t)ev,
[[[0BL] CT ] - 4
- 4
We also assmne an operation of existential closure that applies to the free variable left by this operation in cases where it is not anaphoric to a previous
nommai; the anaphoric case can be left up to standard processes of dynamic
interpretation (Groenendijk and Stokliof 1991, Kamp and Reyle 1993, i.a. ).
These assumptions give the following representation for a simplified variant
of (46a):
(50) (a)
(b)
(c)
Juan se le
muri
John SE Cl-dat.3sg died.3sg
'John died on him/her.'
S
NP
VP
1'SUBJ=J.
le
[ S e]([Ze])([mri 0 ](L/an]))
= 3e[died(e,john) 3x[(x, e)]] since died(e,john)
is telic
The case where the dative argument is unexpressed will be similar, except that
Dsat will apply to se before composition with the sentence denotation.
It is also possible to 'double' the dative argument, though only if the dative
clitic le is present (Gutierrez-Rexach 1999). We do not consider the doubling
facts in detail in the present paper due to space constraints, noting only that the
phenomenon can be related to the necessity of doubling certain clitics when they
are intepreted as event participants (Bleam 2000).
239
(51) (a)
Juan se le
nuiri a Mara
Juan SE Cl.dat.3sg died.3sg to Maria
'Juan died on Maria.'
(b) *Juan se muri a Mara
John SE died.3sg to Maria
4.3 Spatiotemporal paths

If Ris must denote a transition, what function does se assume in the aspectual
composition of RIs? The majority of verbs that combine with seall in Class
I and IIare predicates denoting spatial movement with an inherent direction.
RI sentences with movement verbs denote either the onset (Class I) or the final
endpoint of the spatial path (Class II). Thus, we can say that se takes a process
consisting of a directed path as its argument and focuses on one location, either
on the onset or the final endpoint of that movement. Which endpoint of the path
is focused on seems to depend on whether the movement is inherently sourceoriented or goal-oriented. 12 Note further that verbs without se, as they combine
with a source or a goal argument, can denote a transition, the onset or the final
endpoint of the process.
We model the observations above using meaning postulates on verb classes
(conditions on admissible models for sentences containing se). In the following
and d are variables over paths and locations respectively. Paths are totally
ordered sets of spatial points. We assmne the availability of two operators and
E on paths:
B(p)
= Ld[d G A W'[d'
e d + d ' d ' >
d]]
E(p)
= id[d
e A d d!
d}}.
GpA
W'[d'
d'<
thus picks out the initial point of a path, and E the final endpoint. We also
need an operator L, which is a function of type (ev, d) that maps events to the
locations at which they occura spatial analogue of the temporal trace function
r . Using these three operators, we can now define meaning postulates on the
verb classes that describe their behavior with respect to their associated path
arguments.
(52) (a)
(b)
MP for Class I verbs V: Ve[se(V(z, e))

MP for Class II verbs V: Ve[se(V(z, e))
3p[L(e) = B{p)\\
3 p [ L ( e ) = E(p)\]
Given these postulates, any model verifying a sentence containing a Class I

verb+se will contain a spatial path whose beginning is cospatial with the spa12
While this statement holds in general, there are some exceptions, such as caer 'to fall', which
is not clearly source- or goal-oriented in isolation, but with se is focused on the path onset. We
assume cases of this sort to be due to lexical idiosyncracies, and will not consider them in detail
here.
240
tial location of the event described by the sentence; models verifying sentences
with Class II verbs+se also contain a path, but here the location of the event is
cospatial with the path's final endpoint.
Se can also focus on a location of a process not involving spatial movement,
e.g. on the final point in a temporal path, as in some class IV examples like that
in (53) below.
(53) SE (le)
terminaron las vacaciones muy pronto
SE (CL.dat.3sg) ended-3pl the vacations very soon
'(His/her)/The vacation came to an end (on him/her) very soon'
The path-based semantics above can be extended to account for the way morir+se
'die+se' differs from morir 'to die'. While nonreflexive morir can refer to
any kind of death, morir+se describes one that is associated with a preparatory phase, as after an illness (cf. Sanz 2000). Thus it is incompatible with
expressions implying a sudden accidental death, as shown in ( 54 ).
(54) (a)
El colonel se muri despus de aos de sufrimiento.

the colonel SE died-3sg after
of years of suffering
'The colonel died after years of suffering.'
(b) *E1 colonel se muri en un accidente de coche
the colonel SE died.3sg in a accident of car
'The colonel died in a car accident.'
These facts indicate that se is interpreted as introducing a path argument in the

case of morir as well as that of some verbs in class IV such as terminar, here,
however, it is a temporal path. This idea can be modelled in a manner analoguous to what was done for Hie spatial paths above, simply by allowing d
to range over paths that have both spatial and temporal coordinates. 13 Now
maps paths to their spatiotemporal initial points, and E to their spatiotemporal
final points. We can make use of to define a meaning postulate for verbs of
this type. The following MP requires that all models verifying sentences with
se contain a spatiotemporal path whose endpoint is temporally coextensive with
the event's runtime.
(55) MP for morir-type verbs V:
Ve[se(F(.T,e))
3p[r(e) = E(p)]]
In other cases, the verb combined with se focuses the beginning of a nondirected
process or on the inchoative onset of a state. It may be that these cases can
be analyzed as involving a 'minimal path' (Denis et al. 2003), which occupies
only a single spatiotemporal instant and is thus equivalent to an achievement
13
We unify the two into spatiotemporal points, though it would also be possible to separate them
into an ordered pair, as in Faller (2003).
241
(= instantaneous transition). It is not clear to us at present whether an approach

based on meaning postulates can capture the facts in a satisfying way.
5 Conclusion
To summarize, RI s have three distinctive properties. They require a quantized
subject NP, they license a dative argument which stands in some relation with
the event denoted by the sentence, and they coerce the verbs they appear with
to achievements denoting a transition, the onset or the final end of a process
or a state. We modelled these facts by introducing a presupposition of telicity
on the VP associated with reflexive se, allowing RIs to be associated with a
(possibly implicit) additional argument, and placing meaning postulates on verb
types associating the temporal interval of the event with the initial or final point
of a path.
We close with a final point. While we have shown that, in general, se requires
quantized subjects, there are cases of habitual sentences like those in (56) in
which nonquantized subjects are possible: 14
(56) (a)
(b)
Se le
caen cosas
SE CL.dat.3sg fall.3pl tilings
'Tilings fall on/from him/her.'
Si se caen gotas de vino, se manchar
el sof.
If SE fall.3pl drops of wine, SE will.stain.3pl the sofa
'If drops of wine fall down, the sofa will get stained.'
We believe that these cases are special in that, as habituais, they involve an
implicit quantification over events ( cf. Smith 1997 ). Each of these subevents
can be analyzed as involving a discrete amount of things or wine; each subevent,
then, will be telic. We leave a detailed explication of this idea to future work.
6 Acknowledgements
Thanks to Rajesh Bhatt, Pascal Denis, Fred Hoyt and Brian Reese and audiences at LSRL 34 and the Event Structures in Linguistic Form and Interpretation
Workshop (especially Dalina Kalluli and Kjell Sasbo) for discussion.
References
Beaver, D. (2002): Presupposition and Assertion in Dynamic Semantics. No. 16 in
Studies in Logic, Language and Information. Stanford, CA: CSLI/FoLLI.
Bleam, T. (2000): Leista Spanish and the Syntax of Clitic Doubling. Ph.D. dissertation,
University of Delaware.
14
This fact was pointed out to us by Cristiana Martnez-Benito (p.c.).
242
Bresnan, J. (2001): Lexical-Fimctional Syntax. Mass.: Blackwell.

Castao, J. (1999): Spanish clitics and event structure. In V. Kordoni, ed.. Lexical
Semantics and Linking in Constraint-Based Theories, pages 126-140.
Dalrymple, M. (2001): Lexical-Functional Grammar. New York: Academic Press.
Dalrymple, M., J. Lamping, F. Pereira, and V. Saraswat (1997): Quantifiers, anaphora
and intensionality. Journal of Logic, Language and Information 6:219-273.
De Miguel, E. and M. Fernndez Lagimilla (2000): El operador aspectual se. Revista
Espaola de Linguistica 30:1343.
de Swart, H. (1998): Aspect shift and coercion. Natural Language and Linguistic Theory
16:347-85.
Denis, P., J. Kuhn, and S. Wechsler (2003): V-PP goal motion complexes in english: an
HPSG account. In The Linguistic Dimensions of Prepositions, pages 121-132.
Dowty, D. (1991): Thematic proto-roles, argument selection, and the lexicon. Language
67(3):547-619.
Faller, M. (2003): Propositional- and illocutionary-level evidentiality in Cuzco Quechua.
In J. Anderssen, P. Menendez-Benito, and A. Werle, eds., Proceedings of SULA 2.
Amherst: GLSA.
Girard, J.-Y. (1987): Linear logic. Theoretical Computer Science 50:1-102.
Groenendijk, J. and M. Stokhof (1991): Dynamic predicate logic. Linguistics and Philosophy 14:39-100.
Gutierrez-Rexach, J. (1999): The formal semantics of clitic doubling. Journal of Semantics 16:315-380.
Heim, I. (1988): The Semantics of Definite and Indefinite Nomi Pirrases. Outstanding
dissertations in linguistics. New York: Garland. 1982 doctoral dissertation.
Kamp, Hans and Uwe Reyle (1993): From Discourse to Logic. Dordrecht, Reidel:
Kluwer.
Krifka, M. (1992): Thematic relations as links between nominal reference and event
domains. In I. Sag and A. Szabolsci, eds., Lexical Matters. Stanford, CA: CSLI
Publications:29-53
Martnez-Benito, C. (2003): Aspectual se: an interface of syntax and semantics. M A
Thesis, University of Texas at Austin.
Miller, P. and I. Sag (1997): French clitic movement without clitics or movement. Natural
Language and Linguistic Theory 15(3):573-639.
Monachesi, P. (1999): A Lexical Approach to Italian Cliticization. Stanford: CSLI
Publications.
Nishida, C. (1991): A non-transformational analysis of clitic climbing in Spanish. In
A. Halpern, ed., Proceedings of the Ninth West Coast Conference on Formal Linguistics. CSLI Publications: 395-409
Nishida, C. (1994): The Spanish reflexive clitic se as an aspectual class marker. Linguistics 32:425-58.
Pustejovsky, J. (1995): The Generative Lexicon. Cambridge, Mass.: MIT Press.
Sanz, M. (2000): Events and Predication. Amsterdam/Philadelphia: John Benjamins.
Smith, C. S. (1997): The Parameter of Aspect. No. 43 in Studies in Linguistics and
Philosophy. Dordrecht: Kluwer. Second Edition; First Edition 1991.
Kluwer.
Zagona, K. (1996): Compositionality of aspect: Evidence from spanish aspectual se.
243
In E. Parodi et al. (eds.) Aspects of Romance Linguistics: Selected Papers from the
Linguistic Symposium on Romance Languages XXIV, pages 475^188. Washington
D.C.: Georgetown University Press.
Zubizarreta, Maria Luisa (1987): Levels of Representation in the Lexicon and in Syntax.
Dordrecht: Foris.
John Beavers (Washington)
Scalar complexity and the structure of events *

1 Introduction
this paper I examine the aspectual behavior of dynamic predicates, i.e. predicates that involve some "change" or potential change in one participant, including change-of-state, motion, and consumption/destruction predicates. I focus on
the factors governing durativity in dynamic predicates, which I integrate into
previous work on telicity to build a broader picture of their aspectual behavior.
A considerable amount of recent work has argued that the telicity of dynamic
predicates arises from a homomorphism between the event and some participant
in the event for which the predicate supplies an explicit bound. For example,
Tenny (1987, 1992, 1994) argues that events are "measured out" via a mapping
between the event and one of three entities: incremental themes of what she
calls incremental theme verbs as hi (la) (e.g. verbs of consumption), a property
of an affected participant in change-of-state predicates as hi (lb), or the path of
motion verbs as in (lc).
(1)
(a)
J o h n ate t h e a p p l e .
(b)
J o h n w i p e d t h e t a b l e clean.
(c)
J o h n h i k e d t h e B a r t o n S p r i n g s trail.
( P r o g r e s s o f event -- a p p l e )
( P r o g r e s s o f event -- c l e a n l i n e s s of t h e t a b l e )
( P r o g r e s s o f event -- B a r t o n S p r i n g s trail)
Telicity is derived from the "measuring out constraint" which ties the bound
imposed on the event to the bound imposed on the incremental theme, property, or path (see also Krifka 1989, 1992, 1998, Dowty 1991, Jackendoff 1996,
I would like to thank Beth Levin both for her extensive discussion and insights on this topic and
also for supplying me with her hand collected database of resultatives from which my initial investigations were drawn. I would also like to thank Hana Filip, Chris Kennedy, Paul Kiparsky,
Manfred Krifka, Andrew Koontz-Garboden, Chris Pin, Ivan Sag, Judith Tonhauser, and Steve
Wechsler as well as audiences at the 76th Annual LSA Meeting, the Stanford Semantics Workshop, and the Workshop on Event Structures at the University of Leipzig. This is based on my
previous and currently unpublished work on prepositional resultatives (Beavers 2002). Steve
Wechsler (2001, 2005 ) arrived at similar conclusions working primarily on adjectival resultatives,
and I draw heavily on his insights even though our models differ in certain respects. However,
any mistakes or deficiencies are purely my own.
246
John Beavers
Kratzer 2004 for various similar approaches). More recently, Hay et al. (1999)
and Kennedy and Levin (2001) collapse these three types of felicity under one
rubric, arguing that in each case felicity derives from constraints imposed on
a scale that measures the change undergone by the incremental theme, patient,
or figure participant. For example, in (la) the progress of the event is correlated with the volume of the apple, not the apple itself. The predicate inherently
bounds this scale by the value zero, thus the event ends when the apple's volume
reaches that point. Likewise, (lb,c) correlate the progress of the event with the
cleanliness of the table and the position of the figure on the path respectively, for
which each predicate supplies an appropriate bound. Since the homomorphism
is always between the event and a scale, this approach collapses the various previously heterogeneous mappings in (1) into a single mapping. 1 For the remainder of the paper I assmne this version of the homomorphic approach, though I
recast it in different tenns.
One prediction of this approach concerns which result pirrases are acceptable
m resultatives. Wechsler (2001, 2005), Wyngaerd (2001), and Beavers (2002)
suggests that only certain kinds of scalar XPs provide appropriate scalar bounds.
For example, Wechsler argues that only some adjectives ("non-gradable" and
"maximal endpoint closed scale gradable" adjectives) may serve as result XPs
since they describe appropriate culmination points for the relevant scales. The
adjective flat provides an appropriate bound for the scale it describes while long
does not (cf. the completely test, completely flat/*long\ Kennedy and McNally
1999, 2005). Thus long may not serve as a result XP:
(2)
John pounded the metal flat/*long.
Intuitively, this is because resultatives are telic and thus acceptable result XPs
must provide specific enough bounds from which telicity may be determined. 2
However, this does not determine all of the aspectual constraints on resultatives,
as shown in (3) (cf. Wechsler 2001, Beavers 2002). 3
1
Technically speaking, Kennedy and Levin (2001 ) do not assume a homomorphism between the
scale and the event, just that boundedness ( "quantization" ) imposed on the scale determines telicity (Chris Kennedy, p.c.). For determining telicity this may well be the case, although nothing
in their approach is incompatible with a homomorphism between the event and scale at all. For
the aspectual properties I discuss here this homomorphism is in fact crucial, and I will assume a
version of their theory involving one.
The claim that resultatives must be telic is modulo the effects of mass or bare plural arguments of
the verb, which induce atelicity in otherwise telic predicates, an issue not particular to resultatives
(Garey 1957, Verkuyl 1972, 1993, Dowty 1979). Goldberg and Jackendoff (2004) dispute the
claim that resultatives must be telic on the basis of result phrases of the form X-er and X-er or
ever X-er, as in For years, Penelope wove the shawl longer and longer ( Goldberg and Jackendoff,
(23c), p. 543). I ignore such examples since they necessarily require "resultatives" of a highly
particular morphology and thus are not indicative of the larger tendency of resultatives towards
telicity.
Most of the data from Beavers (2002) are from a corpus of resultatives including about 1,700 with
prepositional result phrases collected from various print media by Beth Levin, although some
(3)
(a)
(b)
(c)
(d)
247
The outlaw knocked/beat the sheriff senseless.

The outlaw knocked/*beat the sheriff dead.
The flames will lick the room to/into a semblance of comfort.
Georgiana was surprised into/*to cordiality.
All of the result XPs in (3) are bounded adjectives or goal-marking prepositions.
Yet senseless and into-PPs occur with verbs that dead and to-PPs are unacceptable with. Adopting Krifka's (1989,1992,1998) mereological model of telicity,
Wechsler (2001,2005) and Beavers (2002) argue that this follows from a correlation between the durativi ty of the event and the gradability of the scale: durative
events require gradable scales, punctual events require non-gradable scales. Expressions denoting events and scales must have compatible durativity/gradability
properties. In this paper I refine and expand this analysis, making the following
claims:
Claim #1: Durativity and gradability reflect two mereological complexity
types: bipartite structures and greater than bipartite structures.
Claim #2: All dynamic predicates correlate durativity with gradability.
Claim #3: The appropriate homomorphism to explain both the scalar boundedness/telicity correlation and the gradability/durativity correlation
is an abstract movement relation between the event and scale of
change, which preserves the relevant mereological properties of
each.
Claim #4: Movement relations are the core property of dynamic predicates.
In 2-3 I review the findings of Wechsler (2001, 2005) and Beavers (2002) on
the durativity/gradability correlation in resultatives, expanding and generalizing
their results. In 4 I generalize this further, showing that this correlation is a
general property of all dynamic predicates even without result phrases. In 5 I
argue that this correlation is best understood as a fonn of abstract "motion". I
outline the relevant lexical, pragmatic, and contextual constraints on durativity
and gradability and show how a generalized movement relation between events
and scales explains the gradability/durativity correlation. In 6 I discuss the
origins of gradability, durativity, and the correlation of the two, arguing that it is
the defining feature of dynamic predicates.
2 Beyond telicity - durativity and resultatives

I define durativity informally in tenns of the "subdividability" of an event (Engelberg 1999, 2000, Beavers 2002). 4 Durative events have multiple discernible
4
examples have been modified here for brevity or clarity.

Unlike telicity, which is a property of predicates, durativity is a property of events, although
predicates may encode sortal constraints on durativity. Thus I refer to predicates that select for
durative or punctual events as "durative predicates" and "punctual predicates" respectively.
248
John Beavers
subparts. For instance, build describes an event that has a beginning point at
which nothing has been built, a final point at which a full entity has been built,
and a series of intermediate subevents corresponding to different degrees of
being built. Thus durative predicates involve three subevents: a beginning, a
middle, and an end. Punctual events have only two of these. Verbs like notice just describe transition events, from not noticed to noticed, thus requiring
only two subevents. This definition may seem counterintuitive at first, since one
might expect to define durative events as those that are subdividable (having any
subevents at all) and punctual events as those that are not. But as discussed
in Dowty (1979, pp. 168-173) (following Taylor 1977) punctual events require
multiple observation points to discern. For instance, an event of stepping once
viewed at an instant might simply be an event of standing. It requires at least two
snapshots, so to speak, to understand it as a stepping event, and more complicated events will require more snapshots. Perhaps the only situations that may
be evaluated at an instant are what Dowty calls "momentary" statives, such as
the ball is on the table, the truth of which one can evaluate on the basis of a
single snapshot. In smn, punctual events are composed of two subevents, a beginning and an end, and durative events additionally have medial subevents (see
Dowty 1979, p. 181 for a similar distinction).
Tests for durativity are less common in the aspectual literatme than tests for
felicity. Following Kearns (2000, p.206), I test for durativity using for/in temporal adverbials in combination with the future tense. 5 Atelic predicates acceptable with for adverbials as in (4a) are necessarily durative since for inherently
imposes a duration on the event. Telic predicates occurring with in either have
only an after reading (that the event occurred after Xtime) or else are ambiguous
between an after reading and a durational reading (similar to for with atelic predicates). Ambiguous predicates are durational while predicates admitting only the
after reading are punctual as in (4c,d) respectively.
(4)
(a)
(b)
(c)
(d)
John will run for five minutes.

#John will blink once for five minutes.
John will build the house in two years.
John will notice the painting in five minutes.
(duration/a/te/)
{after)
Some predicates are imderspecified for durativity. For example, semelfactives

such as blink, tap, and stab (Comrie 1976, Smith 1991, Beavers 2002, Rothstein
2004) can be either punctual or iterative. Likewise, some non-semelfactives
allow both durative and punctual readings. The verb cross has a punctual or
durative reading depending on the path denoted by its object:
(5)
(a)
(b)
The settler will cross the desert in two hours.

The settler will cross the border in two hours.
(duration//?')
{after)
This test works in the past tense as well but the future tense seems to draw out the distinction
more clearly for mysterious reasons.
249
With this working diagnostic for durativity, we can re-examine the resultative
data in (3). I look first at the results of Wechsler (2001, 2005) on adjectival resultatives and then timi to Beavers (2002) on prepositional resultatives, in both
cases summarizing and expanding the earlier results. Wechsler observed distinctions similar to the following among adjectival resultatives. 6
(6)
(a)
(b)
The sheriff beat/battered the outlaw senseless/black and blue,

The sheriff shot/??battered the outlaw dead.
What appears to govern the imacceptability of dead hi (6b) is durativity. Specifically, dead may only occur in punctual event descriptions while the other adjectives may occur hi durative event descriptions, as (7) confirm:
(7)
(a)
(b)
The sheriff will beat the outlaw senseless/black and blue in five minutes.
(duration//?')
The sheriff will shoot the outlaw dead in five minutes.
{after)
By the in adverbial test, dead occurs in descriptions allowing just the after reading, whereas senseless and black and blue also allow durational readings. However, something not discussed by Wechsler is that adjectives occurring hi durative contexts may also occur hi punctual contexts:
(8)
With one solid punch, the sheriff will knock the outlaw senseless/black and blue in
five minutes.
{after)
Here senseless and black and blue are both compatible with the semelfactive
btock on a punctual reading. So it seems that certain adjectives are sensitive to
durativity while others are not. Turning to prepositional resultatives, Beavers
(2002) notes similar distinctions between to and into result phrases:
(9)
(a)
(b)
(c)
(d)
Kim polished the shoes into/to a sombre, imscuffed shine.

The gray sky dimmed into/to dusk and the snow started up again.
I was startled into/??to indiscretion.
I ducked into/*to the cave.
Once again durativity correlates with these distributional differences:

(10) (a)
(b)
The gray sky will dim into/to dusk in ten minutes.

I will duck into the cave in two minutes.
(duration//?')
{after)
Thus both to and into are acceptable with durative event descriptions, whereas
6
I focus here on what Wechsler ( 1997) calls "control" resultatives, where the verb and the result
phrase share an argument, as opposed to " E C M " resultatives (involving fake objects/reflexives)
where the verb is not subcategorized for the subject of the result phrase.
250
John Beavers
only into is also compatible with punctual event descriptions, shown in (10b).7
Furthermore, something not discussed in my previous work is that some prepositions seem to require purely punctual interpretations:
(11) (a)
(b)
(c)
John will slap the poster on the wall in five minutes.

John will roll the poster onto/??on the wall in ten seconds,
John will push the cart at the wall in five minutes.
{after)
(duration//?')
{after)
On occurs with the semelfactive slap only on a punctual reading and is incompatible with durative roll, although onto is acceptable here. At is compatible
with push but only on a punctual reading, where John gave the cart a quick
shove. Thus some prepositions behave like dead in requiring punctual events,
while others behave like senseless in appearing in either durative or punctual
descriptions. The curious exception is to, which only occurs in durative event
descriptions. In fact, we see a striking contrast between dead and to death in a
context with a preceding non-resultative atelic expression:
(12) (a)
(b)
After firing several shots, the sheriff finally shot the outlaw dead.
#After firing several shots, the sheriff finally shot the outlaw to death.
The sentence in (12a) quite felicitously means that the sheriff shot at the outlaw
for a while until finally hitting and killing him, while (12b) has only the pragmatically bizarre reading that the sheriff fired several shots at the outlaw before
finally firing several more shots in a genuine effort to kill him. In sum, we see
the correlations in (13) between result phrases and durativity.
(13) (a)
(b)
(c)
dead, on, at
to
All others
s
s
s
punctual event
durative event
durative or punctual event
In the next section I examine these scalar expressions more closely and show
that these constraints are predictable from the gradability of the scale.
3 The durativity/gradability correlation

Gradability is the subdividability of a scale, i.e. whether it is binary or multivalued (see for example Kennedy 2001; see Sapir 1944 for an early discussion
of gradability). Intuitively, dead describes a point on a binary scale (dead vs.
not dead). Flat on the other hand describes a point in a multi-valued space of
flatness: there are many degrees offlatness other than just flat. An operational
7
Denis et al. ( 2003 ) and Wechsler ( 2003 ) argue that into is only compatible with punctual contexts.
However, examples like ( 10) as well as motion events that favor durative contexts such as After
hitting the top of the roof the branch rolled gently into the gutter suggest that into does allow
durative readings, though perhaps it favors punctual readings.
251
test for this is the acceptability of an adjective with comparative morphology

(Kennedy and McNally 2005). Gradable adjectives are acceptable with comparative morphology while non-gradable adjectives generally are not:
(14) dirtier, wetter, straighter, more bent, more senseless, more black and blue, #deader,
#more dead
Although it has been less noted in the literatme, it does appear that different
classes of prepositions are also amenable to this distinction:
(15) This road cuts more into the woods than the highway, John walked more to/toward/
across the valley than Bill, #Bill is more at the store than John, #This fly is more on
the wall than that one.
The prepositions in ( 15) that are unacceptable with comparative morphology are
those that are largely locative and point-based in natine. The prepositions that
are acceptable are those that are directional or else locative but involve comparisons of extended spatial regions. On the basis of (15) I tentatively assiune that
gradability is a general property of scalar expressions exhibited by both adjectives and prepositions. With this operational definition, we can re-examine the
correlations in (13). Recall that dead is unusual among the adjectives discussed
above in that it occurs only in descriptions of punctual events. Note, however,
that in (14) dead is the only one of those that is non-gradable. On the basis
of evidence tike this, Wechsler proposes that non-gradable adjectives tike dead
can only occur with verbs describing punctual events. Gradable adjectives on
the other hand may cooccur with either durative or punctual event descriptions
as seen above. Among prepositions, into/onto pass the test for gradability and
likewise cooccur with durative and punctual predicates. On/at do not pass the
gradability test and only occur in punctual event descriptions. To is unusual
m that it is gradable but strictly requires durative readings. Thus we have the
following attested patterns:
(16) (a)
(b)
(c)
Durative verb + gradable scalar

Punctual verb + non-gradable scalar
Punctual verb + gradable scalar except to
From just (16a,b) one could suppose that durativity and gradability are somehow correlated with one another as are punctuality and non-gradability. But the
peculiar case in (16c) poses problems for this clean story, where punctual verbs
may also occur with gradable scalars (with the exception of to, another puzzle).
However, (16c) may not pose much of a problem at all once we look at how
gradable scalars are interpreted with durative and punctual verbs. Consider the
use of the gradable adjective flat with stamp.
252
John Beavers
(17) (a)
(b)
With one quick motion, John will stamp the tulips flat in two minutes, {after)
John will stamp the tulips flat in two minutes.
(duration//?')
Stamp is semelfactive in (17a), where the context favors a punctual interpretation, and iterative in the durational context of (17b). In both cases flat is acceptable. However, commensurate with the punctuality in (17a), there is a nongradable interpretation of the scale. In this context the only relevant distinction
is between flat and not flat, not admitting intermediate degrees of flatness. In
(17b), however, it is possible to isolate intermediate timeslices of the event at
which we see intermediate degrees of flatness. We likewise see a similar correlation with into result phrases:
(18) (a)
(b)
With one sputter, the generator will cough into life in two minutes.
{after)
The generator will cougli into life in thirty seconds.
( durati on/(<7/fe7')
In ( 18a) the generator goes from "dead" to "alive" without intermediate steps,
whereas in (18b) it presmnably hovers between death and life before purring
completely into life. Therefore it appears that most of the purportedly gradable scalars described here are in fact underspecified for gradability: they describe either gradable or non-gradable scales, depending on context. These are
distinguished from purely non-gradable scalars like dead, pregnant, and unanimous, which never admit a gradable reading. Likewise, we could further suppose
that to represents a class of strictly gradable scalars which never admit a nongradable reading. I summarize all of the possible combinations in the following table (where each cell represents the durativity/gradability of the verb/scalar
combination and indicates an ungrammatical combination):
Verb type
Durative
Underspecified
Punctual
Non-gradable
X
pimct/non-grad
pimct/non-grad
Scalar gradability type

Underspecified
dur/grad
dur/grad or pimct/non-grad
pimct/non-grad
Gradable
dur/grad
dur/grad
X
Ahnost every combination is allowed, except strictly non-gradable scalars with

durative verbs and strictly gradable scalars with punctual verbs. Likewise, verbs
underspecified for durativity take on durative readings with gradable scalars and
punctual readings with non-gradable scalars and vice versa. Counting occurrences of different result pirrases in the resultative data from the BNC collected
by Boas (2003, Appendix A) further supports these conclusions. Dead has 429
occurrences as a result pirrase, all with verbs that may describe punctual events.
Likewise to death occurs 498 tunes and to sleep 45 times, all with verbs that
allow durative readings. 8 Other result pirrases have mixed possibilities. For ex8
Steve Wechsler (p.c.) points out that some examples in the BNC of shoot to death may allow
punctual interpretations as well as durative ones. Likewise, in Boas's corpus to pieces occurs more
253
ample, shut occurred 202 times, 97 with necessarily punctual verbs, 91 with
necessarily durative verbs. From this summary we can conclude that a durative interpretation of an event described by a resultative is correlated with a
gradable interpretation of the result pirrase and punctuality is correlated to nongradability. In the next section I demonstrate that this generalization has consequences beyond the class of resultatives.
4 Durativity and inherent scales of change

Even without result pirrases, dynamic predicates show durativity/gradability correlations. This can be seen by examining two kinds of data. First are sentences
without result phrases but with overt scale denoting expressions (e.g. traversal
objects of motion verbs), such as (5), repeated here:
(20) (a)
(b)
The settler will cross the desert in 2 hours.

The settler will cross the state line in 2 hours.
(duration//?')
{after)
When the path is non-extended the event is understood as punctual, and when
the path is extended the event is durative. Second, even when the path is not
overtly expressed, for instance in a motion event with a goal-denoting object
rather than a traversal object, changing the context so that the implied path must
be construed as either extended or non-extended has an effect on the durative or
punctual interpretation of the whole sentence:
(21) (a)
(b)
[John is standing just outside the entrance of a cave]

John will enter the cave in 30 seconds.
[John is standing outside a tunnel leading into a cave]
John will enter the cave in 30 seconds.
{after)
(duration//?')
Likewise, even when the scale is not a path, the same interpretive correlation
holds. Consider (22), where the same sentence is durative in one context but
punctual in another depending on how the covert scale is interpreted.
(22) (a)
(b)
[In a context of turning a knob that dims the lights]

The stagehand will lower the house lights by 3/4 in five minutes.
(duration//?')
[In a context of flicking a switch that cuts the lights by 3/4]
The stagehand will lower the house lights by 3/4 in five minutes.
{after)
readily with strictly punctual verbs such as break. However, in both cases I think there is a degree
of lexical idiosyncrasy with to death and to pieces due to their semi-conventionalization (e.g. both
have non-literal uses as in Hove you to death and The review tore his symphony to pieces). The
general trend in both the BNC and of my own proprietary corpus searches have confirmed that
/o-PPs are overwhelmingly durative in nature, setting these two potential exceptions aside.
254
John Beavers
Thus the general conclusion for all dynamic predicates, either with overt scales
of change as in resultative expressions or else covert scales of change as in inherent change-of-state, is given in (23).
(23)
Interpretation of event
Interpretation of scale
Durative
Gradable
Punctual
Non-gradable
Intuitively this is what one would expect on a truly homomorphic model of

events to scales: progress of the event is determined by progress along the scale,
so the complexity of one should somehow mirror the complexity of the other,
at least at some level of granularity.9 In the next section I integrate this into a
specific model of the event-scale homomorphism.
5 A complete event homomorphism model

Although durativity and telicity are separate aspectual features, we can nonetheless understand the durativity/gradability and telicity/boimdedness correlations m tenns of the same homomorphism. In this section I argue for a particular
model of the event-to-scale homomorphism based on the mereological event-topath homomorphism defined by Krifka (1998) for motion and change-of-state
descriptions. Before discussing the properties of this homomorphism, I first define durativity and gradability mereologically. I assumed above that punctual
events and non-gradable scales involve just two pieces (a beginning and an end)
while durative events and gradable scales involve three or more pieces (a beginning, an end, and some middle portion(s)). Following Beavers (2002), we can
generalize over these two types by referring simply to what I call "minimally
complex" and "complex" objects as in (24), producing the two relevant binary
distinctions for events and scales in (25).111
(24)
(25)
10
(a)
M i n i m a l l y C o m p l e x O b j e c t s (MCO)
(b)
C o m p l e x O b j e c t s (CO)
h a v e exactly t w o subparts,
have at least three subparts.
F o r event e a n d scale s:
MCO(e):
Punctual event
CO(e):
Durative event
MCO(s):
CO(s):
N o n - g r a d a b l e scale
G r a d a b l e scale
Evidence from English deverbal adjectives supports this general correlation (Kennedy and McNally 2005). Punctual verbs like stun tend to yield non-gradable deverbal adjectives while durative verbs like build tend to yield gradable deverbal adjectives.
Most of the terms and formal notions I introduce here are given more explicit model-theoretic
definitions in Beavers ( 2002 ) based quite heavily on the mereological framework of Krifka ( 1998 ).
For purposes of clarity and ease of presentation I keep the discussion here considerably more
informal, except where needed. Although I rely here on purely mereological properties of objects,
see Pin ( 1997) for a formalization of punctuality (of achievement events) based on a notion of
a "boundary" as a separate mereological entity. While ontologically distinct (making different
predictions about sorts of punctual events), this discussion could be recast in those terms.
255
Complex objects do not necessarily have just three subparts. They only have a
minimum of three subparts. Why are higher complexity types irrelevant? We can
view this in tenns of "granularity". For any mereological entity, different subparts may be important for different contexts. One could conceive of an event of
eating a sandwich as consisting of a series of bites or a series of singular chewing events (movements of the jaw). A path from San Francisco to New York
could be viewed in tenns of each individual mile or else hi tenns of just city-tocity segments, for instance the various ways different airlines calculate frequent
flyer miles. Each represents a different "take" on a particular entity, at different
levels of granularity. Likewise, only certain subparts or granularities may be
relevant for different grammatical phenomena. Recall that on nearly any komomorpliic model of telicity the endpoints of the event/scale are the most crucial
for ascribing bounds; the rest of the event/scale is irrelevant. Wiat I maintain
here is that for the grammatical distinction between durative/punctual events and
gradable/non-gradable scales the relevant granularity is one that separates beginnings and endings and lumps any medial subobjects together.
Turning now to the correlations between events and scales, consider first descriptions of motion. In an event described by John walked to the store, there
is a homomorphic relationship between the event and the position of the figure
on the path that can be verified visually. John's progress is measured first by Iiis
initial departure from his starting location. He then progresses along the path,
which he traverses in a spatially adjacent fashion though he may go hi any direction and even backtrack. Finally he anives at the store, designating the end
of the event. Krifka (1998, (71), p.225) describes this in tenns of a Movement
Relation (MR) entailed by a motion verb between its event argument e and a
path argument for a figure x. Informally speaking, MRs are functions from e
to with the following properties:11
(26)
(a)
Coextensiveness:
T h e initial a n d final s u b e v e n t s of e are m a p p e d to t h e initial
a n d final s u b p a t h s o f respectively.
(b)
Adjacency.
T e m p o r a l l y a d j a c e n t s u b e v e n t s o f e m a p t o spatially a d j a c e n t s u b -
p a t h s o f p.
(c)
Surjective
Functionhood:
A l l s u b e v e n t s o f e are m a p p e d t o a single s u b p a t h
o f a n d all s u b p a t h s o f c o r r e s p o n d t o at least o n e s u b e v e n t of e.
(d)
Minimality.
The event begins w h e n leaves the source and ends w h e n
arrives at t h e goal.
11
MRs are a generalization of what Krifka ( 1998, (69), p.224) refers to as Strict Movement Relations (SMRs). The difference is that SMRs encode motion involving constant progress with no
backtracking, circles, or stopping. MRs embed SMRs but allow these extra types of motion. Note
that there is a small technical difference between how I discuss paths and how Krifka defines
them. For Krifka, sources and goals are limits on a path, defined as the boundaries that the initial
and final subpaths of a path are adjacent to. But they are not part of the path per se. On my
approach, they are the minimal (atomic/non-divisible) endpoints on a path, i.e. they are part of
the path. I take this approach largely for expository purposes. It is not obvious to me that this
difference is crucial or that Krifka's formal definitions would need to change in any non-trivial
way.
256
John Beavers
Intuitively these properties capture the relationship of motion events to paths

as described above. The last property, minimality, deserves some comment.
Consider a context in which John declares Ms intention to walk to the store,
but does not actually budge. Is the sentence John is walking to the store true?
Clearly not. Although he may retimi to the house during the course of the event,
the event does not begin until he leaves for the first time. Likewise, once John
reaches the store, the event ends. If he reaches the store and then leaves again,
one would no longer say John is walking to the store, at least not to describe
the same event. Minimality is the property that ensures events begin and end
promptly upon departure and arrival.
Coextensiveness determines telicity of motion predicates when a source and
goal are specified (see Krifka 1998, p.228). But what is of interest here is what
MRs say about the internal complexity of e and p. If each has just two parts then
the MR maps them to one another in the expected fashion. But what if either
is more complex? I examine this on a case by case basis. If e has a non-initial,
non-final subevent e' then there must exist a non-initial, non-final subpath p' of
to which it maps. This is because MRs are functions, and all subevents must
map to some subpath. This subpath cannot simply be either the initial or final
subpath of since e' would then map to the same subpath as either the initial
or final subevent of e, violating the minimality of MRs. So p' must be distinct
from the initial and final subpaths of p. Therefore is also complex. Likewise,
if there exists some non-initial, non-filial subpath p' of there must exist a noninitial, non-final subevent e' of e. This is due to the surjectivitiy of MRs, which
ensures that each subpath has a corresponding subevent. This cannot be either
the initial or final subevent of e since these are already mapped to the initial and
final subpaths of and each subevent may only be mapped to one subpath by
the fimctionliood of MRs. Thus e must be complex as well. To recast this more
informally, if you spend time moving between a source and a goal there must be
some place you spent that time moving, and likewise if you hit any part of a path
that is not the source or goal there must be some portion of the event you spent
traversing it.
Thus MRs are isomorphic up to tripartite structures with respect to mereolog-
ical complexity. This means that e has two subparts iff has two subparts, and e
has three or more subparts iff has three or more subparts. Note that this does
not mean that MRs are isomorphic with respect to temporal/spatial precedence,
i.e. temporal precedence does not necessarily reflect spatial precedence. This allows for the fact that once a figure leaves the source point on a path, it may return
to it multiple tunes dining the course of the event before reaching the goal, in
which case the figure lias left the medial subpath during a medial subevent to return to the initial subpath. MRs are also not isomorphic for structures of greater
than three subparts. A figure could spend a lot of time (a multi-part e) walking
back and forth along a very small before reaching its destination. In this case
257
there are arguably considerably more subevents than subpaths. 12 All MRs do
is preserve isomorphy with respect to mereological complexity for bipartite and
greater than bipartite structures.
But of course, this isomorphy is exactly what underlies the correlation of
durativity/gradability, which therefore falls directly out of the MR:
(27) MRs are isomorphic up to tripartite mereological complexity, i.e.:
(a)
(b)
MC'O(e) MCO(p)
CO(e) ~ C'0{p)
Thus MRs are responsible for the durativity/gradability correlation for motion
descriptions. Recasting this in scalar terms, the MR could be viewed instead as
a homomorphism between the event and a scale of position along the path, from
which the same correlations follow for the same reasons. 13 Of course, there is
no a priori reason why the durativity/gradability correlations of other dynamic
predicates amenable to a scalar analysis should not follow from the same kind of
relation. For an expression like John built a model airplane, as John builds the
(initially non-existent) airplane it progresses adjacently through different points
on a scale of builtness until finally reaching completely built. Again, beyond
tripartite complexity there is no one-to-one correlation between the event and
scale. If the model came in three pieces, John could conceivably put two pieces
together and take them apart again ad infinitum, but the model will not be built
until he snaps on the final piece. Therefore I propose that all dynamic predicates
have the same general form:
(28) is a dynamic predicate iff predicates over an event e, a force-recipient , a
scale of change s, and possibly other entities.14
Dynamic predicates are descriptions of an event e that relate e to a theme
which is (potentially) changed and a scale s that describes the change. The relationships of , s, and other participants to e are determined by what Krifka
(1998) refers to as (^-relations, relationships implied by that relate these entities back to e (similar to Parson's 1990 thematic role relations). Following
Beavers (2002), I propose that the 0-relation relating s to e is a type of MR that
relates events to scales rather than just paths, which I refer to here as a Generalized Movement Relation ( GMR). What kind of dynamic predicate is depends
on the nature of the scale as discussed in 1 : scales corresponding to positions
along paths are appropriate for motion descriptions, scales corresponding to the
12
13
14
The reverse is not true. If is multipart it requires a separate subevent for each subpath.
Alternatively (or perhaps preferably) the path could be viewed as a type of scale itself.
"Force recipient" is the term Rappaport Hovav and Levin (2001) apply to participants in events
that have potential for change.
258
John Beavers
extent of entities are appropriate for creation/destruction predicates, scales corresponding to non-spatial properties are appropriate for change-of-state predicates,
etc. 15 The scale appropriate for a given expression is determined by a combination of lexical, contextual, and pragmatic factors.
Before examining how GMRs explain the data discussed in 3-4, I briefly
review where CO and MCO constraints come from. As noted above, lexemes
impose complexity constraints as shown in (29) and (30).
Verb
Strictly durative
Underspecified
Strictly punctual
Example
run, walk, drive
tap, cross, enter
stun, shock, die
Constraints Imposed
VO{e)
Scalar
Strictly gradable
Underspecified
Strictly non-gradable
Example
to
dry, clean, onto
dead, stunned, at
Constraints Imposed
C'O(s)
MCO (e)
MC'O(s)
Presumably, these constraints are not basic but are derived from more basic properties of different classes of lexemes, something I return to in 6. Furthermore,
context may determine complexity constraints:
(31 ) [John is standing just outside the office]
#John walked to the office.
The sentence in (31) is of course grammatical, where walk. imposes CO(e) and
to imposes CO(s).
But the context inherently involves a simplex path, thus
contributing an MCO(s) constraint, contradicting the constraint imposed by to.
Finally, pragmatics can influence the mereological interpretation of an event or
scale, as in the following examples (inspired by Verkuyl 1993):
(32) (a)
(b)
John drew a circle (#instantly).

The (special new) printer drew a circle (instantly).
Pragmatically, humans draw in a sequenced manner and thus such an event must
be complex. But if the agent is a printer with a circular jet that prints all points
of the circle at once, then this constraint vanishes. This suggests that draw does
15
The G M R t h u s expands Krifka's MR analysis, which he only applies to motion expressions and
change-of-state predicates, to incremental theme verbs, thus implementing the scalar approach
of Kennedy and Levin (2001) in Krifka's mereological framework. Wechsler (2005) argues for
a similar generalization of a movement-based homomorphism, but does not discuss the nature
of the homomorphism in terms of mereological complexity, which I argue must essentially and
specifically be a kind of MR.
259
not inherently describe events of a certain complexity type, although in practice

drawing events are durative. Putting everything together, dynamic predicates
consist of a verb V predicating over an event e, theme x, and scale s and zero or
more scalar XPs predicating over s. Any of these elements, as well as the context
and pragmatics, determine CO and MCO constraints on e and s. The G M R
implied by the predicate ensures compatibility of these constraints. From this we
can derive the interpretative and grammatical patterns we saw above for different
types of predicates. Durative verbs are compatible with gradable scales and
punctual verbs with non-gradable scales:
(33) (a)
(b)
Jolm walked to the store.

Jolm shot the sheriff dead.
(CO{e) C'O(s))
(MC'O(e) MC'O(s) )
In (33a) walk imposes CO{e), compatible with the CO{s) constraint of to. A
different situation occurs in (33b), where dead imposes MCO(s).
The semelfactive shot is underspecified for complexity constraints, but the G M R ensures
MCO(e) to satisfy the preservation of complexity. Similarly, va. John stunned
Mary into silence, the achievement stun imposes MCO(e) but into is underspecified for complexity, which the G M R resolves to MCO(s).
Even if no lexeme
imposes constraints, context alone can, as in (21), repeated here:
(34) (a)
(b)
[Mm is standing just outside the entrance of a cave]

M m will enter the cave in 30 seconds.
[Min is standing outside a tunnel leading into a cave]
M m will enter the cave in 30 seconds.
(after)
(duration//??;-)
Neither the verb nor the goal object in (34) imposes any complexity constraints
on e or s. But different contexts favor different complexity constraints on s
which likewise favor certain constraints on e, explaining the judgments on the
in test. Thus the homomorphic approach explains a variety of interpretive facts
about dynamic predicates. This approach also explains the grammaticality facts
discussed above ( the unacceptability of some verbs with some scalare) in terms
of conflicting constraints. This situation is illustrated in (35 ).
(35) (a)
(b)
* Mill stunned Mary to silence.

*The sheriff battered the outlaw dead.
( M CO (e) CO (s))
(CO(e) MCo\s))
Stun requires a bipartite event, but to requires a tripartite scale, thus yielding
a contradiction by (27). Likewise, batter requires a tripartite event but dead
requires a bipartite scale, leading to another contradiction. To see exactly where
the failure is, it is helpful to view this visually:
260
John Beavers
(36) Ungrammatical diirativity/gradability correlations:

(a)
e'
//
e'"
^
CVpunct + gradable scale)
///
(b)
e'
e"
e'"
///
(*Vdur + non-gradable scale)
In both cases, the condition of up to tripartite isomorpky is violated: there is either not enough scale for the event or not enough event for the scale. In (36a) it
is not possible for either e' or e'" to map to s", since this violates the condition
that the GMR be a function. Likewise, in (36b), e" may not map to either s'
or s'". While m general a subscale can correspond to multiple subevents, the
minimality of MRs ensures that nothing adjacent to e' or e'" can share a subscale with it. Thus the imgrammaticality of certam result pirrases with certain
predicates follows from violations of the relationship between the scale and the
event. I summarize all of these conclusions in tlie next section and discuss tlie
origins of tlie various lexical constraints assumed liere.
6 Conclusion - the underlying nature of complexity constraints

I argued above tliat a liomomorpliism model designed to explain tlie nature of
felicity of motion predicates (a) also explains tlie diirativity/gradability correlation argued for by Wechsler (2001, 2005) and Beavers (2002) and expanded on
above and (b) can be generalized to cover a range of dynamic predicates following the scalar approach to felicity hi Hay et al. (1999) and Kennedy and Levin
(2001). Crucially, the event-to-scale liomomorpliism preserves complexity at
exactly the level of granularity relevant for this correlation. Before concluding,
I examine more closely the lexical complexity constraints discussed above and
offer some tentative suggestions as to their origin, since it seems unlikely that
these constraints should be basic. First is the distinction between durative and
punctual predicates. Olsen (1994) proposes that durativity is a privative feature
of verbs, i.e. some verbs lexically select for durative events whereas all others are
imderspecified. If this were true then we would have only to explain why some
verbs are durative. But prima facie, this claim is not correct, since achievements
only admit punctual readings.
(37) (a)
(b)
(c)
The art critic will notice the painting in five seconds.

John will die in five seconds.
John will stun Mary in five seconds.
{after)
{after)
{after)
While most punctual verbs may also be durative (such as semelfactives or pathof-motion verbs like cross), achievements steadfastly reluse to be durative. But
notice that achievements are associated with scales that are (a) inherently non-
261
gradable (cf. *more noticed, *more dead, llmore stunned', see Kennedy and
McNally 2005) and (b) non-iterable, since they entail results. This is unlike
path-of-motion verbs which may have gradable scales and semelfactives which
may iterate because they do not entail results. In principle, if a context were
devised favoring a gradable reading of the inherent scale or iterability of the
result state we might expect a durative reading of some sort. Consider a context
where we might get iterability:
(38) [In a context of Nancy suffering from acute amnesia]
?Sid's Mohawk will stun Nancy over and over again for five minutes.
This (admittedly forced) context seems to allow a durative reading of stun. Arguably then punctuality can be derived from properties of the underlying scale.
So is there no genuine punctuality? Hana Filip (p.c.) notes the unique behavior
of semelfactives in the progressive:
(39) (a)
(b)
The bunny is sleeping.

The bunny is hopping.
(durative, non-iterative)
(necessarily iterative)
In (39a) we get a regular durative reading, i.e. that the bunny is in the middle of a
sleeping event. But in (39b) we necessarily have an iterative reading. This would
imply that there is still something "punctual" about semelfactives such that we
can only construe them durationally by stringing together lots of punctual events.
But note of course that pragmatics and context play a role, since The horse is
hopping (over the fence) may be non-iterative when the path of the hopping
is long enough. This suggests that what is at play here is not a grammatical
constraint of punctuality so much as a tendency to apply to shorter events. If
this is the case, it might be that Olsen (1994) is indeed right and punctuality is
not a lexical constraint, leaving only durativity.16 But is there something we can
reduce this to? Consider the data in (40).
(40) (a)
(b)
(c)
(d)
The wizard rolled down the hill (#in an instant).

The wizard moved down the hill (?in a flash of light).
The alien slurped the bowl of soup down (#in an instant).
The alien consumed the bowl of soup (in an instant).
In (40a,b) there is a contrast between two verbs describing motion along the
same path, where one is necessarily durative but the other allows the possibility
of "instantaneous" motion given a situation involving a wizard who can move
magically. Likewise in (40c,d) slurping down vs. consuming a bowl of soup in
an instant (through some alien power or technology) have different durativity
16
The one counterexample to this I have found is step, which steadfastly resists iteration despite
not entailing a result state (cf. *He stepped to the office, Beavers 2002). This may be due to a
blocking effect of sorts with walk, which is a lexicalized iterative stepping.
262
John Beavers
constraints. Although these verbs can describe the same situations, only roll and
slurp down must be durative. Why is this the case? It seems that necessarily durative verbs are those with some associated maimer that precludes instantaneous
readings. Dowty (1979) proposes that some verbs require multiple "snapshots"
to evaluate (e.g. one snapshot of walking might just be standing, two might be
stepping, but three would be walking). Roll and slurp down involve maimers
that are sequenced in such a way that multiple "snapshots" are required to discern them as a pattern, whereas move and consume do not have maimers. Thus
durativity constraints may be derivative of maimer, i.e. [+mamier] > [+durative] in Olsen's featural tenns, suggesting that neither punctuality nor durativity
is really basic to event descriptions.
Turning to scalars, the difference between non-gradable and imderspecified
scales is perhaps conceptually based. Certain scalars (like dead or pregnant) are
lexicalized as non-gradable because they correspond to real-world binary contrasts. Other scalars however correspond to contrasts that could be binary or
have various shades. Thus gradability types are perhaps basic in the sense that
they correspond to specific real world contrasts. But this leaves to unexplained,
which is unique in being the only scalar to impose a durativity constraint. W i y
should this be the case? The short answer is that I have no idea. The long answer
is to tentatively suggest that this is somehow correlated to the motion typology of
Talmy (1975,1985,2000). The core of the typology is what Talmy (2000) refers
to as a "framing event", the semantic structure that defines, among other things,
the aspectual and scalar structure of the event. Tahny distinguishes between
verb-framed languages in which verbs define the framing event and satelliteframed languages, where satellites to the verb (e.g. particles and presmnably
prepositions) define the framing event. This pans out most conspicuously in expressions of motion in different languages, where manner-of-motion verbs like
saunter and dance may cooccur with goal-marking satellites in satellite-framed
languages but not in verb-framed languages. Wiat is unique about to in English,
a satellite-framed language, is that it defines framing events. Verb-framed languages (like Japanese and French) tend to lack elements with the distribution and
semantics of to. Perhaps imposing strict gradability and boimdedness, thereby
providing a frame for dynamic predicates, is part and parcel of to's ability to
define framing events. This is not an explanation of to's behavior, but merely a
correlation that might offer clues into why to behaves as it does. 17
Finally, there is the question of why dynamic predicates impose a komomorpliism at all. Again I have no definitive answer, except to suggest that rather
than thinking of the homomorphism as something "extra" needed to explain the
durativity/gradability and telicity/boimdedness correlations, we instead think of
17
Note that some verb-framed languages (e.g. Japanese, Korean, French, Spanish, and Turkish)
permit ///-markers to mark goals with manner-of-motion verbs. These are not goal-markers per
se, having a more general semantics (occurring also as temporal, spatial, and numerical boundary
markers). Interestingly, they exhibit the same complexity constraints as to, perhaps for similar
reasons. See Beavers ( 2004 ) for further discussion.
263
it as the defining property of dynamic verbs. That is, a komomorphism is what

all dynamic predicates are about: force-dynamic relationships involving measurable change. Different event types are assigned to this category w h e n they
involve measurable changes, and the allegedly basic telicity/boimdedness and
durativity/gradability correlations follow, all under one rubric. We would of
course expect further properties of dynamic predicates to fall out of the same
komomorphism, a matter I leave to future work.
References
Beavers, John (2002): Aspect and the distribution of prepositional resultative pirrases in
English. LinGO Working Paper #2002-7, CSLI, Stanford University, Stanford, CA.
Beavers, John (2004): On the nature of goal markers and event delimiters: Evidence from
Japanese. Unpublished Ms., Stanford University.
Boas, Hans C. (2003): A Constructional Approach to Resultatives. CSLI Publications,
Stanford, CA.
Comrie, Bernard (1976): Aspect. Cambridge Univeristy Press, Cambridge.
Denis, Pascal, Jonas Kuhn, and Stephen Wechsler (2003): V-PP goal motion complexes
in English: An HPSG account. In Patrick St. Dizier, (ed.), Proceedings of the ACLSIGSEM Workshop on the Linguistic Dimensions of Prepositions and their Use in
Computational Linguistics Formalisms and Applications. IRIT, Toulouse, France.
Dowty, David (1979): Word Meaning and Montague Grammar. Reidel, Dordrecht.
Dowty, David (1991): Thematic proto-roles and argument selection. Language, 67(3):
547-619.
Engelberg, Stefan (1999): 'Punctuality' and verb semantics. Proceedings of the 23rd
Annual Penn Linguistics Colloquium, Univeristy of Pennsylvania Working Papers in
Linguistics.
Engelberg, Stefan (2000): The magic of the moment: What it means to be a punctual
predicate. In Proceedings of Berkeley Linguistics Society 25, pp. 109-121.
Garey, Howard B. (1957): Verbal aspects in French. Language, 33(2): 91-110.
Goldberg, Adele E. and Ray Jackendoff (2004): The English resultative as a family of
constructions. Language, 80(3): 532-569.
Hay, Jennifer, Christopher Kennedy, and Beth Levin (1999): Scalar structure underlies
telicity in degree achievements. In The Proceedings of SALT IX, pp. 127-144.
Jackendoff, Ray (1996): The proper treatment of measuring out, telicity, and perhaps
event quantification in English. Natural Language and Linguistic Theory, 14: 305354.
Kearns, Kate (2000): Semantics. St. Martin's Press.
Kennedy, Christopher (2001): Polar opposition and the ontology of'degrees'. Linguistics
and Philosophy, 24: 33-70.
Kennedy, Christopher and Beth Levin (2001): Telicity corresponds to degree of change.
Seminar presentation, Stanford University.
264
John Beavers
Kennedy, Christopher and Louise McNally (1999): From event structure to scale structure: Degree modification in deverbal adjectives. In The Proceedings of SALT IX,
pp. 127-144.
Kennedy, Christopher and Louise McNally (2005): Scale structure, degree modification,
and the semantics of gradable predicates. Language, 81(2), pp. 345-381
Kratzer, Angelika (2004): Telicity and the meaning of objective case. In J. Guron and
J. Lecarme, (eds.), The Syntax of Tense, pp. 389-423. MIT Press.
Krifka, Manfred (1989): Nominal reference, temporal constitution and quantification in
event semantics. In Renate Bartsch, Johan van Benthem, and Peter van Emde Boas,
(eds.), Semantics and Contextual Expressions, pp. 75-115. Foris Publications, Dordrecht.
Krifka, Manfred (1992): Thematic relations as links between nominal reference and temporal constitution. In Ivan Sag and Anna Szabolcsi, (eds.), Lexical Matters. CSLI
Publications, Stanford, CA. pp. 29-53
Krifka, Manfred (1998): The origins of telicity. In Susan Rothstein, (ed.), Events and
Grammar, pp. 197-235. Kluwer Academic Publishers, Dordrecht.
Olsen, Mari Broman (1994): The semantics and pragmatics of lexical aspect features.
Studies in the Linguistic Sciences, 24(2), pp. 361-375
Parsons, Terence (1990): Events in the Semantics of English. Massachusetts Institute of
Technology Press, Cambridge, MA.
Pin, Christopher (1997): Achievements in an event semantics. In Proceedings of SALT
VII, pp. 276-293.
Rappaport Hovav, Malka and Beth Levin (2001): An event structure account of English
resultatives. Language, 77(4): 766-797.
Rothstein, Susan (2004): Structuring Events. Blackwell, Oxford.
Sapir, Edward (1944): Grading: A study in semantics.
93-116.
Philosophy of Science, (11):
Smith, Carlotta (1991): The Parameter of Aspect. Kluwer Academic Publishers, Dordrecht.
Talmy, Leonard (1975): Semantics and syntax of motion. In John P. Kimball, (ed.),
Syntax and Semantics, volume 4, pp. 181-238. Academic Press, New York.
Talmy, Leonard (1985): Lexicalization patterns: Semantic structure in lexical forms. In
T. Shopen, (ed.), Language Typology and Syntactic Description Vol. 3: Grammatical
Categories and the Lexicon, pp. 57-149. Cambridge University Press, New York.
Talmy, Leonard (2000): Toward a Cognitive Semantics: Typology and Process in Concept
Structuring, volume 2. MIT Press, Cambridge, MA.
Taylor, Barry (1977): Tense and continuity. Linguistics and Philosophy, 1: 199-220.
Tenny, Carol (1987): Grammaticalizing Aspect and Affectedness. Ph.D. thesis, MIT,
Cambridge, MA.
Tenny, Carol L. (1992): The aspectual interface hypothesis. In Ivan A. Sag and Anna
Szabolcsi, (eds.), Lexical Matters. CSLI Publications, Stanford, pp. 490-508
Tenny, Carol L. (1994): Aspectual Roles and the Syntax-Semantic Interface.
Academic Publishers, Dordrecht.
Kluwer
Verkuyl, Henk J. (1972): On the Compositional Nature of the Aspects. Reidel, Dordrecht.
265
Verkuyl, Henk J. (1993): A Theory of Aspectiiality: The Interaction between Temporal

and Atemporal Structure. Cambridge Unversity Press, Cambridge, UK.
Wechsler, Stephen (1997): Resultative predicates and control. In The Proceedings of the
1997 Texas Linguistics Society Conference, pp. 307-321. Texas Linguistic Forimi.
Wechsler, Stephen (2001): An analysis of English resultatives under the event-argument
homomorphism model of telicity. In Proceedings of the 3rd Workshop on Text Structure. University of Texas, Austin.
Wechsler, Stephen (2003): Serial verbs and serial motion. In Dorothee Beermann and
Lars Hellan, (eds.), Proceedings of the Workshop on Multi-Verb Constructions. Trondheim Summer School, Norwegian University of Science and Technology, Trondheim.
Wechsler, Stephen (2005): Resultatives under the 'event-argument homomorphism' model
of telicity. In Nomi Erteschik-Shir and Tova Rapaport, (eds.), The Syntax of Aspect.
Oxford University Press, Oxford, pp. 255-273
Wyngaerd, Guido Vanden (2001): Measuring events. Language, 77(1): 61-90.
SECTION IV
Event Structure and Plurality
Angelika Kratzer (Amherst)

1 Introduction
This paper pursues some of the consequences of the idea that there are (at
least) two sources for distributive/cumulative interpretations in English. One
source is lexical pluralization: All predicative stems are born as plurals, as
Manfred Krifka and Fred Landman have argued. Lexical pluralization should
be available in any language and should not depend on the particular make-up
of its DPs. I suggest that the other source of cumulative/distributive interpretations in English is directly provided by plural DPs. DPs with plural agreement
features can 'release' those features to pluralize adjacent verbal projections. If
there is a lexical source for distributive/cumulative interpretations, there
should be instances of such interpretations with singular DPs. But there should
also be cases of distributive/cumulative interpretations that require the presence of DPs with plural agreement morphology.
What is the role of events in all of this? Events have played a major role in
the semantics of plurality since the pioneering work of Barry Schein and Peter
Lasersohn. Yet to the present day, there is no consensus about the need of
event-based accounts of plurality. Non-event-based analyses of plural phenomena continue to be proposed. The phenomena discussed in this paper all
present small or not so small conceptual problems for event-less analyses, but
can be given elegant accounts within frameworks that incorporate some version of a Davidsonian event semantics. The hope is, then, that an event semantics for plurals might at least be a good bet about reality.
2 Pluralization
Are there plural verbs? And if there are, how did they become that way? Take
the verb fall. Fall denotes a relation between individuals and events: individuals who fall are being related to their falls. I don't know why, but most of us
grew up believing that verb meanings start out 'singular': in the case of fall,
singular individuals are being linked to singular events. My views changed
when Manfred Krifka (1992) and Fred Landman (1996) suggested that verbs
are born as plurals. Fall could then also link plural individuals to plural events
270
Angelika Kratzer
from the very start1. This cannot be the end of the story, however. VPs and
bigger verbal projections can be plural, too, and their plurality cannot always
be inherited from the plurality of their verbs. There must be another source of
pluralization, then. Sternefeld (1998), Sauerland (1998), Beck (2000), and
Beck and Sauerland (2000) have proposed that there is an optional and freely
available operator in the syntax that pluralizes predicates, both those that are
basic and those that are syntactically derived. If they are right, the plurality of
verbs is just a special case of a much more general phenomenon. In this paper,
I will argue that there is a distinctive theoretical place for lexical pluralization,
and that pluralization of phrasal verbal projections is not at all unconstrained.
It can only occur in the immediate neighborhood of a DP with plural agreement morphology.
Before we begin, let us get the technicalities out of the way. How do you
pluralize a predicate? Here is a recipe. First, our basic domains have to be
right. The domain of entities D e should contain both singular and plural individuals. Following Link (1983), we construe plural individuals as sums and
assume that D e is cumulative, that is, closed under sum formation: whenever
and y are in D e , so is x+y, the sum of and y. In addition to D e , we need a
domain of events D s . The sum operation is also defined for events, and, consequently, D s can be assumed to be cumulative, too. Following Krifka (1989), we
extend the sum operation to ordered pairs and other tuples built from members
of D e and D s . The sum of the pairs <Mary, fall,> and <John, fall 2 >, for example, would be <John+Mary, falli+fall 2 >. Pluralization can now be defined as
an operation * that maps sets that come with a sum operation to their smallest
cumulative superset.2 Here is an illustration of what * might do to the extension of fall. Suppose there are just two falls, one by Mary, and one by John.
We have then:
(1)
(a)
[[fall]] = {<John, falli>, <Mary, fall 2 >)
(b)
[[*///]] = {<John, falli>, <Mary, fall 2 >, <John+Mary, falli+fall 2 > }
Having learned how to pluralize, we can begin to think about the hard questions: Where do pluralization operators show up? Why do they show up where
I am assuming a weak notion of plural, where singularities are special cases of pluralities (Link
1983). See Sauerland, Anderssen, and Yatsushiro ( 2 0 0 4 ) for recent support (including processing and acquisition data) of this assumption.
See Krifka ( 1 9 8 9 ) and Landman (1996). In my illustration, I am not working with Schnfinkeled verb denotations. I am using set talk, which allows a simple definition of the pluralization operation for predicates whose denotations let us define a plausible sum operation. The
transition to Schnfinkeled denotations is straightforward. The operation that pluralizes functions of type D < e < ! , , for example, can be defined as AR-IJ-'IU;,.,,, [ * x , o
{ *cx,ci-
R(x)(e) } ]. I will mostly use Schnfinkeled denotations, and 1 will then use " * " also as the
symbol for the corresponding cross-categorial pluralization operation. Basic types used in this
paper: e (individuals), t (truth-values), s (events).
271
they do? How are they related to plural morphology on nouns and verbs? What
is their semantic effect? And finally, if there is pluralization, shouldn't there be
singularization, too? I will start with the last question, and suggest that there is
no such thing as singular number. If the suggestion is correct, we do not expect
operators that 'singularize', and we are entitled to focus our attention on pluralization alone. To make my point, I have to briefly discuss the interpretation
of number marking with nouns.
3 Nominal number. Eliminating [singular] as a number feature

Over the last 15 years or so,3 Manfred Krifka has explored cumulativity as an
important property of nominal and verbal predicates, and in the course of this
work, the possibility emerged that cumulativity might correspond to a significant semantic universal: "simple predicates in natural language typically are
cumulative". 4 Blatant counterexamples to Krifka's universal seem to be singular count nouns like child, chair, or chin. Following Link, the extensions of
singular count nouns are usually taken to be sets of singularities, hence could
not be cumulative. If Josephine is a child, and Beatrice is too, the sum of Beatrice and Josephine is not a child. Those two girls are children. However, child,
chair, or chin are not necessarily simple predicates. They may already be complex by the time we get to see or hear them. They each might consist of a root
(Schild, Vchair, Schiri) and a piece of nominal inflection. Some have argued
that common noun roots have predicative, number-neutral ('transnumeral')
denotations (Mller 2000, Rullmann and You 2003). On that proposal, the root
Schild, for example, would denote the set consisting of all singular children
and their sums. The denotations of common noun roots would be cumulative,
then, and would thus satisfy Krifka's universal. Alternatively, we might consider the possibility that noun roots are referential and refer to kinds, following
Krifka (1995) and Yang (2001), who both build on Carlson (1977). Not being
predicative at all, common noun roots would satisfy Krifka's universal trivially. It would now be part of the job of nominal inflection to turn those roots
into predicates. This last proposal is very much in the spirit of Borer (2005),
and is especially attractive since it does not assume any particular mode of
individuation or portioning for the denotations of noun roots. Crosslinguistically, the function of individuating and portioning is often carried by
classifiers. On an analysis where English common noun roots do not yet denote properties (that is, sets containing portions or individuals), there would be
a stage in the derivation of English nouns where they look exactly like their
Chinese or Japanese cousins. We would now have to look for classifiers in
3
4
Krifka (1987, 1989, 1992, 1998).

Krifka (1998), 200. See also Landman (1996, 2000).
272
Angelika Kratzer
English, that is, pieces of inflection that could be held responsible for mapping
kinds into sets of individuals or portions.
Following up on Krifka (1995), I want to suggest that English has a multiply ambiguous non-overt classifier, and that the noun forms that are usually
categorized as 'singular' are in reality roots with an incorporated classifier.
Here is an illustration of the proposal for count nouns.
(2)
(a)
(b)
[[Vzebra]] = 'zebra'.
[[CLinJ] = [kind(x) & individual(y) & y < x]
[[CLkmd]] = XxXy [kind(x) & kind(y) & y < x]
According to the proposal sketched in (2), the word zebra is a 'singular' predicate by the time we see or hear it. It was turned into a predicate by an incorporated ambiguous classifier, and is therefore ambiguous, too. It can denote a set
of individual zebras (with CL ind ). or a set of subspecies of the species 'zebra'
(with CLkind). Zebra is ambiguous in this way, as shown in (3).
(3)
(a)
(b)
This zebra has not been fed.

This zebra is almost extinct.
On one interpretation, the incorporated classifier maps a kind to the set of its
individual realizations. On the other interpretation, a kind is mapped to the set
of its subkinds. You might call the result a 'singular' predicate, but we have to
be careful if we talk that way. It is precisely those 'singular' predicates that are
the input for pluralization. As shown in (4a) and (b), the plural noun zebras is
ambiguous in the very same way as the 'singular' noun zebra is.
(4)
(a)
(b)
Those two zebras have not been fed.

Those two zebras are almost extinct.
(4a) talks about two individual zebras that have not been fed, and (4b) about
two subspecies of the species zebra that are almost extinct. Interestingly, the
same ambiguous classifier can sometimes attach to certain mass nouns. If it
does, the resulting predicates show the expected ambiguity, and so do the corresponding plural forms:
(5)
(6)
(a)
(b)
(a)
(b)
This wine is for table 8.

You dropped two red wines.
This Pinot Noir is rare.
We tasted five different Pinot Noirs.
(5a) may be used to inform a waiter that a particular glass or bottle of red wine
has to go to table 8. (5b) could be a complaint about you having dropped two
particular glasses or bottles of red wine. (6a) and (b) illustrate the subkind
273
reading of Pinot Noir and its plural Pinot Noirs. There is some indication,
then, that English has an unpronounced classifier that builds predicates from
names for kinds, and that the resulting predicates can then be submitted to
pluralization.
So far, we haven't seen any need for a nominal number feature [singular]. If
there was such a feature in English, there would be no semantic job for it to do.
Since there is also no overt morpheme marking nominal [singular] in English,
we might suspect that there is no such feature to begin with. Alternatively, we
might say that English has an ambiguous classifier [singular]. We can save
[singular] in this way, but we are removing it from the list of number features.
Be this as it may, here is what the immediate projection of an English count
noun root might look like:
([pit-
Figure 1
Mass nouns in English are also predicative by the time we see them, hence
should come with an obligatory classifier, too. Following Chierchia (1998),
that classifier should map a kind into the set of all of its singular or plural realizations. Predicative mass nouns are already pluralized by their classifier,
then, and are therefore not submitted to further pluralization. Unless they combine with an individual or a kind classifier, mass nouns cannot project [plural].
Both mass nouns and non-plural count nouns trigger 'singular' agreement in
English. We now understand why. Both types of nouns have only projected a
classifier and therefore lack [plural].
That plural marking and agreement can co-occur with a classifier in a language might seem a typological anomaly. Greenberg (1972) and Sanches and
Slobin (1973) have explicitly associated the existence of numeral classifiers
with the absence of obligatory nominal plural marking and agreement. Borer
(2005) takes that typological connection to be a major argument in favor of
analyzing the [plural] feature itself as a classifier. However, Aikhenwald's
typological study of classifiers (Aikhenwald 2003) does not support a necessary connection between the presence of numeral classifiers and the absence of
plural marking and agreement. She mentions a respectable number of exceptions, including (among others) Yuki, Nootka, Tlingit, Tucano, North Arawak,
and South Dravidian languages. 5 If the [plural] feature was responsible for
Aikhenwald (2003), 100, 249.
274
Angelika Kratzer
both individuation and pluralization, as Borer (2005) proposed, it would be a

mere accident that in English, pluralized predicates rely on the very same kinds
of individuation as their 'singular' counterparts. On the other hand, if pluralization operates over the alleged 'singular' predicates, we understand why 'singular' and plural nouns in English use the same modes of individuation. To
make this explanation palatable, I suggested that there is no feature [singular]
in English. What we have been calling 'singular' might literally be the absence
of plural - not only in the morphology, but in the semantics as well. Non-overt
classifiers are responsible for giving us 'singular' predicates.
Nominal plural is not the main topic of this paper, so I will not be able to
pursue the full range of relevant consequences of the suggestion that English
has silent classifiers that create predicates for the nominal [plural] feature to
operate on. The proposal is compatible with Chierchia (1998) in that it also
assumes that English count nouns are predicative by the time we see them. It is
that property that prevents them from occupying argument positions. However,
in the spirit of Krifka (1995) and Borer (2005), I have suggested that predicative count nouns are syntactically constructed from names for kinds with the
help of classifiers. What distinguishes English from Chinese or Japanese, then,
is that it has obligatory incorporated classifiers for its nouns. The empirical
gain of a constructional approach to the mass/count distinction is that, in contrast to global parameters of the kind Chierchia proposed, it allows for microvariation and variability within a single language.
Returning to our topic, the main goal in this section was to give at least
some support to the idea that there might not be a number feature [singular] in
languages like English. That point is important for my overall argumentation,
and this is why I spent some time on motivating it. Agreement phenomena
show that there is a tight connection between nominal and verbal number, and
if there is no nominal [singular], it seems safe to conclude that there is no verbal [singular] either. We expect the connection between nominal and verbal
number to be established via [plural] alone, then. When we see 'singular'
agreement, what we see is inflection that is there because of the absence of
[plural]. I will have to be serious about pluralization, then, but no apologies are
needed for my complete neglect of 'singularization'.
4 On the plurality of verbs. Lexical cumulativity

In the nominal domain, we were able to maintain Krifka's cumulativity universal by positing referential denotations for noun roots. Noun roots are never
predicative, then, hence satisfy Krifka's universal trivially. According to the
story developed in the last section, English nouns may become predicative at
some stage in the course of a syntactic derivation, but when they do, they are
no longer simple. What about verb roots and verb stems? As a class, verbs
have the characteristic property of taking arguments. Some of those arguments
275
seem to be syntactically acquired in the course of a derivation. Marantz (1984)

and Kratzer (1996) have argued that external arguments are always added in
the syntax. Pylkknen (2001, 2002) makes the same point for applicative arguments. Some direct internal arguments seem to be introduced syntactically,
too, via secondary predicates or serialization, for example. 6 But there are also
transitive and unaccusative verbs with inherently relational meanings: relate,
connect, resemble, surpass, outdo, depend, hinder, cause are in this group. It is
hard to imagine that those verbs could merely characterize kinds of states or
events without relating them to at least one of their participants. With many
transitive and unaccusative verbs, the kind of event described varies with the
kind of direct internal argument in sometimes erratic ways. Take pick.7 Picking
a pumpkin, picking cat hair off your pants, picking a lock, or picking someone's pocket are quite different kinds of activities. And so are activities like
plucking apples from a tree, plucking a goose, plucking guitar strings, or
plucking a hair from your soup. What popping events are depends on what it is
that pops: corks, balloons, eyes, or neighbors who just pop in. The verb rise
hardly picks out a natural class of events either: you rise when you stand up or
get up from bed, smoke may rise above the trees, rebels may rise against a
cruel tyrant, lakes may rise, and prices and bread dough, too. If the meanings
of those verbs are inherently relational and relate individuals and events, we
understand why the type of the event described can depend on the type of
direct internal argument in not entirely predictable ways.8 It would be hard to
account for this dependency under the assumption that a verb's argument
structure is always syntactically constructed. 9 There seems to be a large group
of inherently relational verb roots, then, and this suggests that as a class, verb
roots might be predicative from the start. If they are, we would expect them to
fall under Krifka's generalization, and they should have cumulative denotations. And if external and applicative arguments are not true arguments of their
verbs, but are added in the syntax, we need thematic role predicates like agent
or goal to introduce them. Those predicates should fall under Krifka's universal as well, and their denotations should be cumulative, too.
There is a major prediction of the 'cumulativity from the start' hypothesis
for verbs and thematic role predicates, which has made it attractive for many
researchers since it was first proposed in Krifka (1992). 10 If the denotations of
verbs and thematic role predicates are cumulative from the start, the effortless
10
See Kratzer (2005) for a proposal about resultatives.

The following examples are all taken from the McGraw-Hill Children's Dictionary.
Dependencies of this kind do not seem to exist between external arguments and the events
described by their verbs. This was the main motivation for Marantz (1984) to suggest that external arguments are not true arguments of their verbs.
See Kratzer (forthcoming) for more arguments supporting the conclusion that not all of a
verb's arguments are introduced in the syntax. Borer (2005) presents an opposing view.
See in particular Landman (1996, 2000).
276
Angelika Kratzer
availability of a cumulative 11 interpretation for sentences like 7(a) below is

expected. We can represent that interpretation as in 7(b):
(7)
(a)
Twenty children ate ten pizzas.
(b)
Be3x3y [*child(x) & /x/ = 20 & *agent(x)(e) & *pizza(y) & /y/ = 10 &
*eat(y)(e)]
On its cumulative interpretation, 7(a) can be true in a wide range of situations,

as long as 10 pizzas were eaten in all, and 20 children did the eating. It does
not matter how the 10 pizzas were shared among the children. For the 20 children to count as the plural agents of the pizza-eating event, they each have to
have eaten at least part of one of the pizzas, and together, they must have consumed all 10 of them. 7(b) captures such scenarios. All predicates in 7(b) have
cumulative denotations. As in Landman (1996, 2000), the basic predicates of
our metalanguage are singular predicates that are pluralized with the *operator, which maps sets that come with a sum operation to their smallest
cumulative superset. If every basic verb and thematic role predicate has a cumulative denotation from the start, there is no need to repeat that information
for every lexical item, of course. However, using the *-operator even for those
predictable cases is still pedagogically useful as a reminder that we are dealing
with cumulative denotations. I will follow this practice for clarity.
To see the impact of verb cumulativity in an event semantics in more technical detail, I will work through a very simple illustration, sentence (8):
(8)
Two children lifted two boxes.
On the intended cumulative interpretation, (8) is again compatible with a wide

range of situations, as long as two children did the lifting and two boxes were
lifted in all. The children might have acted individually or jointly. The boxes
might have been lifted one at a time or both together. And either box or the
two boxes together might have been lifted once or several times. How does (8)
manage to cover so many different kinds of situations? Suppose the two children are Casey and Stacey, and the two boxes are Red and Green. Casey lifted
Red on her own once, and Stacey did so twice. In addition, Casey and Stacey
jointly lifted Green. We have four events, e l5 e 2 , e 3 , and e 4 , then, which can be
characterized as in Table 1 :
11
The term is due to Scha (1981,1984). Kroch (1974) coined the name 'serially distributive
reading' for the salient interpretation of The men in the room are married to the girls across
the hall (p. 204 f.). Sauerland (1998) uses the term 'co-distributive reading'.
For any individual x, /x/ is only defined if there is a set of atomic individuals that is the sum
of. If defined, /x/ is the number of atomic individuals that is the sum of, i.e., /x/ = /{y: y <
& atom(y)}/.
277
Table 1
Box lifted
Box lifter
ei
Red
Casey
e2
Red
Stacey
e3
Red
Stacey
e4
Green
Casey+Stacey
In truth-conditional semantics, the extensions of predicates depend on relations

that hold in the actual world. A customary, non-cumulative, extension for lift
would pair actual lifting events with the objects actually lifted, for example.
Disregarding Schnfinkelization, the relation would include the pairs listed in
(9) (assuming our scenario is true):
(9)
Extension of lift
{<ei, Red>, <e2, Red>, <e3, Red>, <e4, Green>,
The customary extension of 'agent' would include the pairs in (10):

(10)
Extension of agent
{<ei, Casey>, <e2, Stacey>, <e3, Stacey>, <e4, Casey +Stacey>,
So far, we have a close match between what might be 'basic' relations in the
actual world and the relations in the extensions of the predicates. Intuitively,
there are four box lifting events and three different agents. One of the agents is
a plural individual, and that means that there is collective action. These kinds
of denotations reflect nicely what is going on in the world as we see it. At this
stage, extensions satisfy what Fred Landman has called the 'Collectivity Criterion'. 13 All plural individuals paired with an event are collectively involved in
that event. All plural agents are collective agents, then. Enters Cumulativity,
and our extensions seem to turn to mush:
(11)
(a)
(b)
13
Extension of *lift
{<ei, Red>, <e2, Red>, <e3, Red>, <e4, Green>, <ei+e2, Red>, <ei+e3, Red>, <ei+e4,
Red+Green>, <e2+e3, Red>, <e2+e4, Red+Green>, <e3+e4, Red+Green>,
<ei+e2+e3, Red >, <ei+e2+e4, Red+Green>, <ei+e3+e4, Red+Green>,
<e2+e3+e4, Red+Green>, <ei+e2+e3+e4, Red+Green>,
}
Extension of *agent
{<ei, Casey>, <e2, Stacey>, <e3, Stacey>, <e4, Casey+Stacey>, <ei+e2, Casey+Stacey>, <ei+e3, Casey+Stacey>, <ei+e4, Casey+Stacey>, <e2+e3, Stacey>,
Landman (1996, 2000).
278
Angelika Kratzer
<e2+e4, Casey+Stacey >, <e3+e4, Casey+Stacey>, <ei+e2+e3, Casey+Stacey>,
<ei+e2+e4, Casey+Stacey>, <ei+e3+e4, Casey+Stacey>, <e2+e3+e4, Casey+Stacey >,
<ei+e2+e3+e4, Casey+Stacey >,
}
The cumulative extensions in ( I l a ) and (b) include more than just the basic
relations between individuals and events we might be prepared to recognize.
There are more lifting events than we ever dreamed of, and, strangely, the sum
of Casey and Stacey is the plural agent of most of them. There is nothing intrinsically bad about this state of affairs, however, as long as the truth conditions we predict are right. Are they?
Assuming the cumulative denotations partially listed in (11), the logical representation 12(a) correctly comes out true. The open sentence (12b) is satisfied
by several variable assignments, including the one in (12c):
(12)
(a)
(b)
(c)
Be3x3y [*child(x)
[*child(x) & /x/ =
'e'
'x'
>
'y'
>
&/x/ = 2 & *agent(x)(e) & *box(y) & /y/= 2 & *lift(y)(e)]

2 & *agent(x)(e) & *box(y) & /y/= 2 & *lift(y)(e)]
ei+ e2+ e3+ e4
Casey+Stacey
Red + Green
Having cumulative denotations yields correct results not only for plural VPs,
as in (13a) and (b) below, but also for singular VPs, as in (13c) and (d). Assume the same scenario as before and look at the following sentences:
(13)
(a)
(b)
(c)
(d)
Casey and Stacey lifted Red.

Casey and Stacey lifted Green.
Casey lifted Red (at least) once.
Stacey lifted Red (at least) twice.
(13a) to (d) should all come out true, and they do. (13a) is verified by ei+e 2 ,
e!+e 3 , and ei+e 2 +e 3 . (13b) is verified by e 4 . The fact that Stacey, but not Casey,
lifted Red twice is in principle retrievable from (13a) and (b) as well. There is
an event (namely e 2 +e 3 ) that has Stacey as its agent, and also has two proper
subevents, each of which is a lifting of Red by Stacey. As for Casey's liftings
of Red, there is only one such event, ei. We can also retrieve the information
that Casey and Stacey lifted Green collectively. 14 There is one event in which
Green alone was lifted, e 4 , and that event has a plural agent, Casey and Stacey,
but no subevent in which Green was lifted by Casey or Stacey alone.
Some breaks have to be built in to prevent the sentences below from winding up true on our scenario:
14
Lasersohn (1988, 1990, 1995).
(14)
(a)
(b)
279
Red was lifted fourteen times.

Casey and Stacey together did eleven liftings.
True, there were fourteen events in which Red was lifted, and there were
eleven liftings whose agents were Casey and Stacey. A basic principle of
counting says that if I count you as an entity, I can't count your head separately. The same principle applies to counting events. If I count e b e 2 , and e 3 as
events of lifting Red, none of the other events in which Red was lifted can be
counted, since they all contain at least one of those three events as a part. If
that was the whole story, though, what would prevent me from claiming that
Red was lifted exactly once, pointing to ei+e 2 +e 3 as my verifying event? Or
exactly twice, with ei+e 2 and e 3 as the relevant verifying instances? What really
seems to count in counting is atomicity. The extension of *lifi contains exactly
three atomic pairs that connect Red to a lifting event. Red was lifted exactly
three times, then. And the extension of *agent contains exactly one atomic pair
that connects Casey and Stacey to a lifting event. They did exactly one lifting
together, then. All in all, it looks like cumulation preserves the information we
want to extract from a verb's extension. Within an event semantics, cumulating
predicate extensions does not lead to a loss of information we might need to
get the semantics of adverbs like twice or three times, or individually or together right, for example. 15
5 How many readings? Phrasal cumulativity

There is an ongoing debate about the number of readings for sentences like (8),
repeated below as (15):
(15)
Two children lifted two boxes.
Our analysis so far says that there is a reading of (15) that lumps together what
are traditionally called 'collective' and 'cumulative' interpretations, and
doesn't distinguish between one-time and repetitive liftings. (15) can truthfully
describe any singular or plural event of lifting two boxes, as long as two children did the lifting. It doesn't matter how they did it. Is it right to lump together all those interpretations that others have taken pains to distinguish? 16
15
A more complete test run is needed, though, see Kratzer (forthcoming), chapter 4. The pioneering work in this area was done by Peter Lasersohn (Lasersohn 1988, 1990, 1995). The toughest
examples were constructed by Roger Schwarzschild (1991, 1992, 1993-94, 1996). And some
of the crucial insights come from Barry Schein's work (1986, 1993).
See e.g. Landman (2000). Schwarzschild has consistently argued against multiplying readings
in his works on plurality (Schwarzschild 1990, 1991, 1992, 1993-94, 1996). I am aspiring to
Schwarzschild's superego, but am exploiting the resources of an event semantics along the
lines of Schein (1986, 1993).
280
Angelika Kratzer
VP-ellipsis constructions provide the best test I know of for individuating

readings. 17 In the interpretation of VP-ellipsis constructions with ambiguous
VPs, the ambiguity in the overt and in the silent VP must be resolved in the
same way. This is illustrated in (16):
(16)
I went to the bank, and you did, too.
Since bank is ambiguous the overt VP in (16) is ambiguous. I might have gone
to a bank to deposit a check, for example, or to the bank of the Connecticut
River to relax. Whatever interpretation you pick for the first VP, you have to
pick the same interpretation for the second VP. Now consider (17):
(17)
The two boys lifted the two boxes, and the two girls did, too.
Is (17) true in a situation in which the two boys jointly lifted each of the two
boxes, but the two girls each lifted a different one of the two boxes on her
own? I think the answer is 'yes', which shows that we are right in lumping
together collective and cumulative interpretations into a single reading. Other
combinations yield similar results. (17) would also be true, for example, if the
boys lifted the two boxes just once, but the girls lifted them multiple times, and
so on.
In addition to the cumulative interpretation we have been investigating, (15)
has two distributive interpretations. (15) can also describe events where each
one of two boys lifted two boxes. Such events might involve up to four boxes.
And for some people, (15) might also be used to describe events where two
boxes were each lifted by two boys. This time round the events might include
up to four boys. Landman (1989) argued that when a plural DP produces distributive interpretations of this kind they should be derived by pluralizing its
sister predicate. 18 The two distributive interpretations of (15), for example, can
be produced as shown in (18) and (19):
(18)
17
(a)
(2 children) * [lifted 2 boxes]
(b)
(2 boxes) * ! [2 children lifted tj]
Zwicky and Sadock (1975), Cruse (1986). Schwarzschild (1996) uses VP-ellipsis cases to
argue that the distributive-collective 'ambiguity' is essentially a pragmatic one (chapter (5)).
Landman draws a distinction between sums and groups in all of his works on plurality, and the
analysis of Landman (1996, 2000) actually produces 8 readings for sentences like (15).
Schwarzschild (1993-94), who is not working within an event semantics, argues explicitly for
lumping together collective, cumulative, and distributive interpretations by staning all plural
VPs, but builds context dependence into the definition of cumulativity. I am following the
spirit of Schwarzschild throughout this paper, but in line with Schein (1986, 1993), Moltmann
(1996), and Beck (2002), I expect the necessary contextual distinctions to come in via contextually restricted quantification over events.
O n the plurality of verbs
(19)
(a)
(b)
* Xxke3y
281
[*box(y) & /y/ = 2 & *lift(y)(e) & *agent(x)(e)]
* x e y [*child(y) & /y/ = 2 & *lift(x)(e) & *agent(y)(e)]
19(a) is the denotation of the predicate that can be obtained by starring the
subjects sister constituent, as in 18(a). 19(b) is the denotation of the predicate
that results from movement of the direct object over the subject and starring
the object's sister constituent, as shown in 19(b). Since starring a predicate
always extends the original extension, both 18(a) and (b) still cover all the
scenarios we discussed before. That is, 18(a) and (b) lump together the collective, cumulative, and repetitive interpretations of (15) with one of its two distributive interpretations. 18(a) can now also describe events in which up to two
boxes were lifted. And the events picked out by 18(b) might include some
where up to four boys did the lifting. If pluralization of verbal predicates is the
correct way of accounting for distributive interpretations, we are committed to
lumping together interpretations in a particular way. Here is the clustering of
interpretations we predict for (15) so far:
The cumulative/collective/repetitive interpretation for (15) can be derived with

lexical cumulativity alone. The subject distributive interpretation requires
starring of the subject's sister predicate, and the object distributive interpretation requires movement of the object over the subject and starring of the resulting sister predicate. If starring of a plural DPs sister node is obligatory, as I
will suggest below, we only expect two truly distinct readings for sentence
(15), one of which is highly dispreferred. The difference boils down to whether
or not we move the object over the subject. That kind of movement is bound to
be costly, hence object distributive interpretations are expected to be dispreferred, if they are available at all. The LF 18(a) should then represent the
default interpretation of (15). As far as grammar goes, no distinction is made
between subject distributive, cumulative, collective, and iterative interpretations. All those different ways of understanding (15) correspond to a single
282
Angelika Kratzer
reading that can be computed in a straightforward way from a single syntactic

representation. We avoid explosion of computational complexity in this way,
not only in the syntax, but also in the semantics. If distinctions between distributive, cumulative, or collective action are needed for the semantics of adverbs like together or individually, for example, they can be easily retrieved in
an event semantics, as Peter Lasersohn has shown. 19
The usual ambiguity tests should tell us whether we have individuated our
readings correctly. In ellipsis constructions, for example, it should be possible
to mix distributive and cumulative or collective interpretations. 20 Look at the
following example:
(20)
The two chefs cooked a stew, and the two students did, too. The chefs were very experienced, so they each prepared a Moroccan tagine. The two students worked together on a
Boeuf Bourguignon.
The text in (20) does not feel inconsistent. It can only be perceived as consistent, though, if we are allowed to mix distributive and collective interpretations
for the two conjuncts in the VP-ellipsis construction.
Roger Schwarzschild has observed that separating distributive and collective/cumulative interpretations can have undesirable consequences in the scope
of negation. Here is a variation of one of his examples. 21
(21)
Beasly, better make sure those guys don't win a car this week!
Like Schwarzschild's original example, (21) is to be understood as a demand

made by a head mobster on one of his "flunkies". The relevant observation is
that "Beasly's goose is cooked" if those guys win a car, whether as a group or
individually. Beasly cannot save himself by arguing that he understood the
demand only distributively or only collectively.
In this section, we have seen what looked like initial support for lexical cumulativity, but we have also seen that lexical cumulativity alone is not enough.
Phrasal cumulativity is needed to account for certain cases of distributive interpretations. We need *-operators that can pluralize phrases, then. Once we
have those *-operators, do we still need Lexical Cumulativity? Can't those *operators alone do the jobs we thought Lexical Cumulativity was responsible
for?
19
20
Without events, retrieving those distinctions is a bit of a trial, but Schwarzschild has done it.
Schwarzschild (1993-94).
Schwarzschild (1996), chapter 5.
Schwarzschild (1993-94), 232, example (72). I changed the example to one that cannot be
reduced to lexical cumulativity.
283
6 Evidence for lexical cumulativity

The assumption that verbs start out with cumulative denotations has come
under attack. Not because it was shown to be wrong, but because the generalizations it expresses looks like a special case of a much more general phenomenon. Several authors, most prominently Wolfgang Sternefeld, Uli Sauerland,
and Sigrid Beck, 22 have argued that the denotations of verbs can be rendered
cumulative through the freely available optional presence of syntactically represented *-operators that can pluralize any kind of verbal predicate, whether it
is lexical or phrasal, basic or syntactically derived.
My immediate goal in this section is to show that the proposal of Sternefeld,
Sauerland, and Beck overgenerates, and that there is still a theoretically distinguished place for Lexical Cumulativity and Krifka's Cumulativity Universal.
Look at the following examples, which all have singular indefinite objects
and describe iterated events:
(22)
(23)
What does this intern do?

(a)
(b)
(c)
(d)
(e)
(a)
(b)
She guards a parking lot.

He cooks for an elderly lady.
She waters a garden.
He watches a baby.
She cleans an office building.
I dialed a wrong phone number for 5 minutes. 23
She bounced a ball for 20 minutes.
(c)
(d)
He kicked a wall for a couple of hours.

She opened and closed a drawer for half an hour.
(e)
I petted a rabbit for two hours.
What is remarkable about those sentences is that the singular indefinite objects
invariably fail to distribute. They look as if they were taking wide scope over
an operator that pluralizes events (Zucchi and White 2001, van Geenhoven
2004): A single parking lot is guarded habitually, a single elderly lady is
cooked for repeatedly, a single ball is bounced again and again, and so on. This
phenomenon shouldn't exist if we allowed free optional insertion of unpronounced star operators. If *-operators could be inserted freely, they could immediately produce (24b) from (24a), for example, hence derive unattested
interpretations for the sentences in (22) and (23):
(24)
22
23
(a)
(b)
[ball(x) & *bounce(x)(e)]

* [ball(x) & *bounce(x)(e)]
Stemefeld (1998), Sauerland (1998), Beck (2000, 2001), Beck and Sauerland (2000).
Zucchi and White (2001), van Geenhoven (2004).
284
Angelika Kratzer
(24b) describes possibly repeated events in which more than a single ball
might be bounced. In contrast, since ball is 'singular' (hence only describes
singular balls), each event described by (24a) can only have a single ball in it.
In an event semantics, the facts in (22) and (23), fall out if verbs are born as
plurals. No obligatory scoping or a narrow scope "frequentative aspect" operator (van Geenhoven 2004) has to be stipulated. To see this more clearly look at
the computation of the denotation of the VP in (25):
(25)
[ bounce a ball]vp
(a)
*bounce(x)(e)
(b)
^ , 3 [ball(x) & R(x)(e)]
(c)
[ball(x) & bounce (x)(e)]
'being a possibly plural event e such that there is a ball and e is an event of
bouncing x'
In a Davidsonian event semantics, the events described are always 'minimal'

in the sense that an event of bouncing this ball, for example, is an event in
which this ball is being bounced and which contains nothing above and beyond
that ball and whatever it takes for it to be bounced. Crucially, it can't have a
second ball in it. Since the relation described in (25a) is cumulative, the property (25c) can describe an iterated event made up of events which themselves
have the property (25c). However, whenever (25c) is true of an event e and a
subevent e' of e, the ball in e' is bound to be the same as the ball in e. Otherwise, e would have two balls, rather than one. Each event in the iteration, then,
has the same ball in it. This is a direct consequence of standard Davidsonian
event predication. This is how event predication is understood.
I haven't quite explained the facts in (22) and (23) yet, however. The assumption that verbs are born with cumulative denotations doesn't mean that
verbs have to describe plural events, of course. The way we defined the *operator corresponds to a weak notion of plurality, where pluralities always
have singularities as special cases. That the sentences in (22) and (23) necessarily describe iterated events does therefore not follow from lexical cumulativity alone. In (22), habitual aspect seems to be responsible for the necessarily
iterative interpretation, in (23), durativity plays a similar role. What is interesting for our story is that given lexical cumulativity, we still predict the facts in
(22) and (23), even if the habitual operator and durational adverbs take scope
over the indefinite direct object. In German, for example, both (26a) and (26b)
imply that a single phone number was dialed for five minutes. That even (26a)
should have that interpretation is surprising in a language that otherwise marks
scope relationships between adverbs and DPs overtly.
(26)
(a)
Ich hab' fnf Minuten lang eine falsche Telefonnummer gewhlt.

I have five minutes long a wrong telephone number dialed.
I dialed a wrong phone number for five minutes.
(b)
Ich hab' eine falsche Telefonnummer

I
have a
285
fnf Minuten lang gewhlt.
wrong telephone number five minutes long dialed.
I dialed some wrong phone number for five minutes.
Lexical cumulativity allows us to explain apparent 'scope' puzzles of the kind

seen in (22), (23), or (26). I will spell this out for durative adverbials. The
habitual cases should be amenable to a very similar analysis. Suppose the denotation of durational adverbials like for 5 minutes is as in (27):
(27)
(a)
<> [P(e) & e = ' [P(e' ) & e' < e] & f nlnulc (e) = 5] 2 4
The definitions in (27) use Link's -operator. In our case, the operator maps
the events in the set { e ' : e' < e & P ( e ' ) } to their supremum - if it exists. The
operation is undefined otherwise. W e are talking about the sum of all events e'
that are proper parts of e and have the property P. The requirement is that that
sum be identical to e. Following Morzycki's Program of Modified Modification (Morzycki 2 0 0 4 ) and the independently developed analysis of durational
adverbs in van Geenhoven (2004), 2 5 we would eventually want to split up the
denotation of durational adverbials like for five minutes into at least two parts:
(28)
(a)
^ [P(e) & = ' [P(e' ) & e' < e]]
(b)
fmmucc(e) = 5
(28a) contributes iterativity/continuity and could be realized by a non-overt

inflectional head in English. (28b) is the contribution of the adverbial proper,
which merely tells us how long the event lasted. The denotation of dial a num24
The function fmimue in (27) is a measure function that measures the time of an event in minutes.
If durational adverbs were main concern of this paper, more would have to be said about
the properties of such measure functions. When I say that I slept for two hours today, for example, we usually understand this as saying that was a 2-hour interval during which I
slept. The time of the event described is an interval, then. This dosn't have to be so, however.
If I am paid by the hour, I may ask for my pay by informing you that I worked on your gutters
for 2 0 hours. In that case, there is no implication that the time of the sum of all events where I
work8d on your gutters is an n e r v a l . Gaps are allowed. I might have worked on your gutters
on several distinct occasions. Sometimes, what treasure function has to measure is the
minimal interval that includes the times of all the subevents of the event whose time is b8ing
measured. When I report that I saw Dr. Spck for 5 years, for , what I seem to be saying is that the minimal interval that includes times of all of my visits to Dr. Spck' s office
is 5 yars.
Van Geenhov8n's pap8r includ8s detailed discussion of West 0 8 ^ : , where some of
relevant operators are overt. It seems 3 the overt West Gre8nlandic operators discussed by
van Geenhoven are related to iterativity/continuity operators ("frequentative aspect" in van
Geenhoven's terminology), and are thus not counterparts of our lcxical ^-operator, which
merely indicates lexical cumulativity, a property that should be a universal prop8rty of verb
stems if Krifka is right. W e should not automatically a direct connection between
lexical ^-operator and overt pluractional operators, .
286
Angelika Kratzer
ber for 5 minutes, for example, can now be computed by applying the denotation of for 5 minutes to the denotation of dial a number. The VP dial a number
is thus clearly in the scope of for 5 minutes. The result is the denotation in
(29):
(29)
Xes3x [number(x) & *dial(x)(e) & e = ' [mimber(x) & *dial(x)(e' ) & e' < e]
& fminule(e) = 5 ]
(29) says that there was an event of dialing some phone number that was composed of proper subevents of dialing a phone number and lasted for five minutes. Given Davidsonian event predication, this implies that the same phone
number was dialed throughout the event. If a sum of events involves just one
phone number, none of its subevents can involve more than one phone number.
We have found a non-trivial consequence of the Lexical Cumulativity hypothesis, then. Assuming Lexical Cumulativity, iterative interpretations for
verbs are possible from the very start, and iterativity without concurrent 'object distributivity' is the automatic result of introducing an ordinary singular
indefinite in the early stages of a syntactic derivation. Given Lexical Cumulativity, habitual operators and durational adverbs do no longer have to pluralize
the predicates they operate over, or introduce quantification over subevents.
They merely have to make sure that those predicates do not describe any
singular events, but are properly plural in a lexically defined sense. This means
that we do no longer have to stipulate obligatory narrow scope for such operators. The desired interpretations can be derived, even if the relevant aspectual
operators are sitting above direct objects.
The same data that provided evidence for Lexical Cumulativity also showed
that *-operators cannot be inserted freely. If they could, we wouldn't expect
the 'failure of distribution' effect illustrated in (23) and (24). But if ôperators
cannot be inserted freely, we are left wondering where phrasal ^-operators
might come from. What is the force that produces phrasal cumulativity, hence
many cases of distributivity?
7 The source of phrasal cumulativity

What makes phrasal cumulativity possible? Here is what looks like an obvious
answer. It cannot be an accident that none of the sentences in (23) or (24) contained any plural DPs. Maybe phrasal *-operators are necessarily tied to the
presence of plural DPs in some way or other. Schwarzschild (1993-94), for
example, proposed that all plural VPs are obligatorily translated with the *operator, hence always have cumulative denotations.
287
My star, ..., appears on the translations of all plural verb phrases. I

also differ here from Landman (1989), who optionally translates plural verb phrases with a star. My star is obligatory: it is there whenever the verb is plural.
Schwarzschild 1993-94, p. 206
Within the current framework of assumptions, Schwarzschild's proposal has to
be rephrased so as to allow for a more articulated sentence structure where VPs
project a hierarchy of functional heads, including habitual operators and heads
that introduce external or applicative arguments. The relevant condition could
now be stated as requiring that at the level where semantic interpretation takes
place, sister constituents of plural DPs are pluralized, regardless of whether
they are still in their base position or have moved away. Here is one particularly attractive way of executing this proposal. So far, I have only talked about
the syntactic structure of plural NPs. Plural DPs are projected from determiners, and following Sauerland (2005), let us assume that when a plural DP is
built from a determiner and a plural noun, for example, both the noun and the
determiner come with their own number projection. 26 We have structures of
the following kind, then:
Figure 3
In Figure 3, there are two [plural] projections. The lower [plural] feature is
responsible for the pluralization of the noun. According to Sauerland, the
higher occurrence of [plural] could be there even if the plural DP doesn't contain any plural NP at all. DPs like Spencer and Webster for example, show all
the behavior of plural DPs. They can trigger verbal agreement and can have
distributive interpretations like other plural DPs. On Sauerland's proposal, the
higher [plural] projection can be held responsible for those properties. How
26
Sauerland assumes the existence of a number feature [singular], though, and takes [plural] to be
the unmarked case. The idea I am adopting from Sauerland is that DPs may have two projections of a number feature. For us, that number feature could only be [plural]. If there is a feature [singular] at all, it has to be a classifier, and not a number feature.
288
Angelika Kratzer
exactly? Here is a possible story. While the lower occurrence of [plural] in

Figure 3 pluralizes the noun, the higher occurrence of the feature can't seem to
be interpretable within DP. Suppose that nominal [plural] is always interpretable, and it always carries the cross-categorial plural operator. Since a higher
[plural] feature can't be interpreted within its DP, it is forced to move out before semantic interpretation takes place. Moving as little as possible, it could
become a verbal inflectional head right below its DP. In this way, a DP could
literally create its own agreement projection, possibly on top of another verbal
projection like voice, aspect, or tense. Heim and Kratzer (1998) invoke the
very same syntactic mechanism for the index of a moved DP that can be
reparsed as a binder index for the DP's trace. If indices are agreement features,
as proposed in Kratzer (2004), Heim and Kratzer's reparsing of the DP index
means that moved DPs create their own verbal agreement projections by 'releasing' one of their features that could not be interpreted otherwise. This allows us to hold on to the generalization that indices are always interpretable.
Likewise, allowing plural DPs to create their own verbal [plural] projection
makes it possible for us to maintain the generalization that nominal [plural] is
always interpretable. We can cut down on uninterpretable features that way.
When [plural] migrates out of its DP, we get a ^-operator that pluralizes the
DP's sister node, possibly showing up as overt verbal agreement. We thus
have the following configuration:
[plural]
=
Figure 4
An immediate prediction of this proposal is that pluralization of phrasal verbal
projections should require the presence of DPs with [plural] agreement features in English. But distributive/cumulative interpretations that can be produced by Lexical Cumulativity alone, should also be available for singular
DPs. Here are some examples that seem to have the right properties to test our
prediction.
(30)
(31)
(a)
(b)
(a)
(b)
She
She
She
She
sent her
sent her
sent her
sent her
offspring to 5 different boarding schools,

offspring to a boarding school.
children to 5 different boarding schools,
children to a boarding school.
289
(30a) looks like a sentence that has a run-of-the-mill cumulative interpretation.

If she had 5 children, the sentence allows for all of the usual cumulative scenarios. Each child might have gone to a different boarding school. All five of
them might have gone to all five of the boarding schools. Two of the children
might have each gone to three of the boarding schools, and the remaining three
children might have gone to the remaining two boarding schools ... and so on.
This results follows directly from Lexical Cumulativity and the assumption
that her offspring is the sum of her five children, as it should be. But why is
(32) bad, then?
(32)
*Her offspring each went to a boarding school.
One possibility is that floated each needs to agree with [plural]. Interestingly,
Oh (2001) has argued that the apparent Korean distributivity marker ssik is not
itself a distributivity operator, but is a particle that must be in the scope of a
distributivity operator. If each is submitted to such a condition as well, we rule
out (32) because of the lack of [plural] in offspring. We might be able to maintain, then, that her offspring denotes a semantic plurality, namely the sum of
her descendants.
Since offspring is singular, we do not expect it to produce phrasal cumulativity. Not surprisingly, then, (30b) implies that all of her children went to the
same boarding school. In contrast, the sentences in (31) have a plural direct
object, and we immediately see phrasal cumulativity pop up. In addition to a
cumulative reading, (31a) also has a reading where each of her children was
sent to a different boarding school. And (31b) is compatible with a scenario
where each of her children went to a different boarding school. This interpretation, too, is the effect of phrasal cumulativity.
The examples in (30) are not isolated cases. Cumulative interpretations are
generally available for mass nouns. But if they are, those interpretations were
produced by Lexical Cumulativity.
(33)
(34)
(a)
(b)
(a)
(b)
All that furniture was loaded onto five tracks,

All that furniture was loaded onto a truck.
Her offspring inherited all her jewelry,
Her offspring inherited a villa in Tuscany.
Conjoined mass nouns sometimes allow singular agreement, 27 and can then
produce distributive/cumulative interpretations. Here are some cases that
sound acceptable to the ears of the native speaker I consulted:
27
My dialect of German allows those kinds of singulars quite freely when two NPs with identical
mass nouns are conjoined. The judgments of the one other native speaker of German I consulted (a North German speaker) went more in the direction of my Standard American English
consultant, whose use of those singulars is rather restricted. Agreement facts are, of course, a
290
(35)
Angelika Kratzer
(a)
(b)
The moss on the rocks and the moss on the trees is blighted,
Jane's china and Alice's china was stored in separate closets.
(35a) and (b) have distributive/cumulative interpretations. Since the DPs are
singular, Lexical Cumulativity must be responsible. Good cases of essentially
phrasal cumulativity can't seem to be produced in the absence of plural DPs.
Sentences (36a) to (c) lack distributive interpretations. (36a) says that the two
kinds of sugar were stored in the same jar, and (36b) implies that the silverware went to the same cousin.
(36)
(a)
(b)
The sugar for the coffee and the sugar for the cake was stored in a plastic jar.
Jane's silverware and Patsy's silverware was sent to a cousin.
Another consequence of the proposed account of phrasal verbal cumulativity is

that it should not be possible to simultaneously cumulate two non-event arguments. Pluralizing predicates with more than one non-event argument is not an
option within the theoretical framework assumed here. In a theory where external and applicative arguments are introduced by independent heads, we do
not have lexical predicates with more than two non-event arguments to begin
with, and plausible assumptions about movement do not seem to allow us to
derive any such predicates in the syntax. 28 A DP's sister constituent can only
be of type <et> (properties of individuals) or <e<st>> (relations between individuals and events), then, hence the pluralization operation could not affect
any other non-event argument position apart from the one that is about to be
saturated by the DP triggering the pluralization. Consider now the following
example from Beck and Sauerland (2000): 29
(37)
These 5 teachers gave a bad mark to those 20 protesting students.
(37) has a cumulative interpretation. It could be true in a situation, for example, where each of the students got a bad mark from only one of the teachers.
Not assuming an event semantics, Beck and Sauerland argue that the intended
28
29
prime target for normative grammars, which might explain the more liberal judgments for a
speaker from South Germany, who grew up in an area where various registers of regional dialects were spoken. Interestingly, both of my consultants reported that in many of those cases,
they wouldn't want to use the plural either, and would therefore try to avoid those constructions altogether.
Sauerland (1998) and Beck and Sauerland (2000) invoke a special mechanism for the creation
of 2-place predicates in the syntax. Seen from a perspective where moved DPs can create their
own agreement projections by 'releasing' their index to become a binder index, that mechanism would allow a DP to move into the agreement projection created by another DP, and set
up its own agreement projection within it. Nested agreement projections of this kind are interpretable, but are likely to be ruled out by general constraints on movement.
Beck and Sauerland (2000), p. 356.
291
interpretation of (37) can only be derived by pluralizing the 2-place relation

xy3z [bad-mark(z) & gave-to(y)(z)(x)].
In an event-based semantics, the intended interpretation of (37) can be derived without pluralizing more than one non-event argument at a time by
adopting an analysis along the lines of Schein (1986, 1993), who extensively
discussed cumulative interpretations of this kind. The key for (37) is neoDavidsonian association of the agent argument, coupled with movement of the
indirect object to a position right above the direct object. The moved DP's
sister predicate would now be pluralized and would wind up with the denotation in (38):
(38)
*XyXe 3z [bad-mark(z) & *gave(z)(e) & *goal(y)(e)]
The pluralized predicate in (38) is of type <e<st>>, hence only has one nonevent argument. Applied to a plurality like those 20 protesting students, the
predicate in (38) yields a property that can be true of events in which each of
those 20 students received a bad mark, for example: 30
(39)
(* 5z [bad-mark(z) & *gave(z)(e) & *goal(y)(e)]) (those 20 protesting students)
Next, we add the agent argument and saturate it:

(40)
[*agent(those five teachers)(e) & (* y e [bad-mark(z) & *gave-to(y)(z)(e)])

(those 20 protesting students)(e)]
The interpretation captured in (40) says that those five teachers were the agents
of an event in which those 20 protesting students received one or more bad
marks. This is the desired result.
A surprising consequence of the current analysis of phrasal verbal cumulativity relates to an example presented in Winter (2000). Winter's example is
meant to show that theories that account for distributive interpretations of
plural DPs by pluralizing their sister predicates overgenerate. They seem to
predict interpretations that do not in fact exist. Interestingly, Winter's objection applies to event-less versions of the 'nominal distributivity via verbal
cumulation' idea, but not to the event-based account proposed here. Let us see
why.
Winter asks us to judge the truth of sentence (41) in the scenario depicted in
figure 5:
30
Note that the ^-operator blocks -conversion. Like the corresponding unstarred predicate, the
starred predicate in (39) is of type <e<st>>, hence yields a predicate of type <st> after being
applied to those 20 protesting students.
292
(41 )
Angelika Kratzer
The children are holding a wheel.
Winter (2000), 63
Figure 5
Winter observes correctly that in situations of this kind, (41) is false or at least
"highly strange". On Winter's own account, all non-lexical cases of
distributivity are due to covert monadic distributivity operators that enforce
atomic distribution. (41) would then be true just in case each child is holding a
wheel. This condition is not satisfied in Winter's scenario. Winter predicts (41)
to be false in his scenario, then. Winter's point is that analyses where plural
DPs trigger cumulation of their sister make false predictions in this case. Let's
forget about events for a moment, and reconstruct Winter's argument. Boy!
and Boy 2 are holding a wheel, and so do Boy 2 and Boy 3 . The denotation of the
unstarred VP in (41) is therefore true of the two pluralities Boyi+Boy 2 and
Boy 2 +Boy 3 . If the plural subject the children induces starring of the VP, the
denotation of that VP is true of Boyi+Boy 2 +Boy 3 , and hence of the children.
(41) is thus predicted to be true on Winter's scenario.
Does our event-based scenario fare any better here? On our account, the sister constituent of the plural subject in (41) expresses a relation between individuals and events, and it is that relation that is cumulated.
(42)
(a)
[*agent(x)(e) & 3y [wheel(y) & *hold(y)(e)]]
(b)
* [*agent(x)(e) & 3y [wheel(y) & *hold(y)(e)]]
Does the pair consisting of the three boys and the event e represented in Winter's scenario satisfy the starred relation in (42b)? It could only do so if there
are pairs <X], ei> and <x2, e 2 > that satisfy the relation in (42a), where xi+x 2 =
the children and ei+e 2 = e. However, the event represented in figure 5 is most
naturally conceptualized as a single event. There are no natural, but only
'strange' or artificial ways of conceptualizing it as the sum of two subevents.
The subevents singled out in Figure 6, for example, do not seem to be among
the atoms in our domain of events:
293
Figure 6
Any serious semantics relies on domains for the basic entities that provide the
building blocks for the whole repertoire of denotations. In the extensional
semantics assumed here, we have three basic types, for example: e for individuals, s for events, and t for truth-values. The truth-values are just True and
False. For the domains of individuals and events, the subdomains containing
the atoms play a special role. They contain the individuals and events that
singular quantifiers quantify over. Among the atoms in the domain of individuals are the cups in my cupboard, for example. Those cups have parts, of
course, and sometimes, we want to quantify over those parts, too. Sometimes.
Usually, we do not recognize the parts of a cup as separate individuals - at
least not the weird parts. The handles might be the topic of a discussion in
some contexts. But there are many oddly individuated parts of those cups that
hardly ever qualify. Take the parts of that cup over there that could instantly
acquire a more respectable status if the cup was dropped and broke. The cup
has all of those parts already, but they don't usually make it into our domain of
atomic individuals. The parts of events behave no differently. True, the individuation conditions for events are a bit looser than those of most individuals,
but that doesn't mean that anything goes. We can't assume that weirdness of
parts should not play a role for events at all. The parts highlighted in Figure 6
are weird in most contexts. The prediction of our event-based account is then
that (41) should be judged false or highly strange. There is a real question
whether the two required subevents ei and e 2 exist in our domain of atomic
events.
If the problem with Winter's example is related to the lack of properly individuated subevents in the original scenario, we expect (41) to be judged true in
294
Angelika Kratzer
scenarios where the relevant subsituations are individuated more clearly.31

This is so:
S
Mt
>)
11
0!
'J
I I
iL
%
<
V* 1
'
Ui
/kt
Figure 7
Sentence (41) is clearly true in the scenario of Figure (7). Rather than presenting a challenge to our account, Winter's example provides a surprising piece of
support.
I conclude, then, that plural DPs are themselves sources of phrasal cumulativity - or more concretely, their higher [plural] features are (in the sense of
Sauerland (2005)). Pluralizing their DP's sister node seems to be the only way
for those features to be put to semantic use. Within an event semantics, a DP's
sister node often denotes a relation between individuals and events. Consequently, judgments about the truth of sentences like (15) are bound to be sensitive to the individuation of events.
8 Conclusion and questions for further research

It seems, then, that there are indeed at least two pluralization mechanisms at
work in languages like English. One is Lexical Cumulativity, which seems to
be universal. The other one is carried by the inflectional feature [plural]. I have
argued that [plural] is always interpretable, and always denotes the crosscategorial ^-operator. Moreover, I have suggested that [plural] always origi-
31
See Schein (1986, 1993), Schwarzschild (1991, 1996), and Beck (2002) for discussion of
related cases. On Schwarzschild's account, matters of event individuation can indirectly influence the subpluralities we consider via contextually provided covers. His theory then predicts
the same judgment for Winter's example as the event-based one. For Schwarzschild's proposal
to work, cumulation has to be constrained so as to mirror event individuation. See
Schwarzschild (1996), p. 96, footnote 27.
295
nates within a DP and pluralizes nominal or verbal projections, depending on

whether it occupies a high or a low position within its DP. The low position
provides access to a noun, the high position provides access to a verbal projection. A single inflectional feature can thus create subtle variations in the availability of distributive/cumulative interpretations, even within a single language. In the long run, the behavior of [plural] in English might provide us
with a general model of how essentially nominal features can provide operators for verbal projections.
While this study has focused on English, the results obtained generate expectations for other languages. Take the Chinese sentence in (43).
(43)
Tamen mai-le yi-bu chezi.

They buy-Asp one-CL car
'They bought a car.'
Lin (1998), 201.
(43) only has a collective interpretation. ".. .the entire group of people denoted
by tamen 'they' collectively bought a car." 32 For (43) to receive a distributive
interpretation on the current account, its VP would have to be pluralized. This
is expected to be impossible if DPs like tamen do not have a higher [plural]
projection. If that projection is linked to agreement morphology, it is not necessarily available for all DPs that are semantically plural. To get a distributive
interpretation for Chinese sentences like (43), the overt distributivity operator
dou has to be used. In Chinese, then, dou is a carrier of the *-operator (Lin
1998, Yang 2001). If Chinese quantifiers generally quantify over pluralities, as
Lin and Yang have argued, we expect dou to co-occur with quantifiers and
create the distributivity effects that come with them. Those distributivity effects are not properties of the Chinese quantifiers themselves.
Natural languages are also known to have operators that exclusively pluralize properties of events. German jeweils is an example. 33 With the use of
jeweils, we can again produce cumulative/distributive interpretations for sentences that do not have any plural DPs at all. Here is an example:
(44)
Eine Kanne Milch hat jeweils ein Pfund Kse produziert.

One can
milk has jeweils one pound cheese produced.
O n each occasion, one can of milk produced one pound of cheese.'
(44) is interpreted as talking about a situation that is the sum of multiple events
where one can of milk produced one pound of cheese. Jeweils may thus be
32
33
Lin (1998), 201.1 am indebted to Jo-Wang Lin for reminding me of this fact.
Link (1998), chapter 5, Zimmermann (2002). See also the discussion of Chinese event classifiers in Yang (2001), and Matthewson (2000) on Salish.
296
Angelika Kratzer
given the following interpretation, which is almost, but not quite, the denotation we posited for the iterativity/continuity component of durative adverbials.
(45)
<51> [e = ' [P(e') & e' < e] ]
The lesson from (43) and (44) is that crosslinguistically, phrasal plurality is not
always linked to nominal [plural]. The feature [plural] does not have to be the
one and only possible source of phrasal plurality, even in a language that also
has [plural]. The exact source of phrasal plurality may thus not always be easy
to determine for a given language. We may have to pay close attention to subtle differences between possible pluralization operations.
There is some indication that subject-distributivity is hard to get when the
subject is left in a low position, as in the German sentence (46).
(46)
Am
Nebentisch rauchten vier Mnner eine Zigarre.
At+the next table smoked four men a
cigar
'At the next table four men were smoking a cigar'
(46) strongly suggests that the four men were sharing a single cigar. Subjects
sitting in low positions are also known to have different agreement properties
in some languages, e.g. French II est arriv des enfants ('there is arrived children'). Maybe those low plural subjects also lack the higher [plural] projection,
in which case they wouldn't be able to pluralize their sister constituent on the
current account. The question is then why those sentences still show overt
plural agreement in German. What exactly is the relation between [plural] and
verbal agreement? I have to leave serious investigation of this issue to another
occasion.
References
Aikhenvald, A. Y. (2003): Classifiers. A Typology of Noun Categorization Devices.
Oxford, Oxford University Press.
Beck, S. (1999): Plural Predication and Partitional Discourse. 12th Amsterdam Colloquium, University of Amsterdam.
Beck, S. (2000): Star Operators. Episode 1: Defense of the Double Star. U M O P 23:
Issues in Semantics. K. Kusumoto and E. Villaita. Amherst/Mass., GLSA, UMass
Amherst.
Beck, S. (2001): Reciprocals are Definites. Natural Language Semantics 9(1): 1-69.
Beck, S. (2002): Event-Based Cumulation. Manuscript, University of Connecticut.
Beck, S. and U. Sauerland (2000): Cumulation is needed: A Reply to Winter 2000.
Natural Language Semantics 8(4): 349-371.
297
Borer, H. (2005): In Name Only. Structuring Sense, Volume 1. Oxford, Oxford University Press.
Carlson, G. N. (1977): Reference to Kinds in English, University of Massachusetts at
Amherst Doctoral Dissertation.
Chierchia, G. (1998): Reference to kinds across languages. Natural Language Semantics 6(4): 339-405.
Cruse, D. A. (1986): Lexical Semantics. Cambridge, Cambridge University Press.
Davidson, D. (1967): The Logical Form of Action Sentences. The Logic of Decision
and Action. N. Rescher. Pittsburgh, University of Pittsburgh Press: 81-95.
Geenhoven, V. v. (2004): For-Adverbials, Frequentative Aspect, and Pluractionality.
Natural Language Semantics 12(2): 135-190.
Greenberg, J. H. (1972): Numeral Classifiers and Substantival Number: Problems in the
Genesis Type. Working Papers in Language Universals.
Heim, I. and A. Kratzer (1998): Semantics in Generative Grammar. Oxford, Basil
Blackwell.
Kratzer, . (1996): Severing the External Argument from its Verb. Phrase Structure
and the Lexicon. J. Rooryck and L. Zaring. Dordrecht, Kluwer: 109-137.
Kratzer, . (2004): Telicity and the Meaning of Objective Case. The Syntax of Tense.
J. Guron and J. Lecarme. Cambridge/Mass., The MIT Press: 389-423.
Kratzer, A. (2005): Building Resultatives. Events in Syntax, Semantics, and Discourse.
C. Maienborn and A. Wllstein-Leisten. Tbingen, Niemeyer.
Kratzer, A. (forthcoming): The Event Argument and the Semantics of Verbs, ms. University of Amherst.
Krifka, M. (1987): Nominal Reference and Temporal Constitution. Proceedings of the
Sixth Amsterdam Colloquium, Amsterdam, ITLI, University of Amsterdam.
Krifka, M. (1989): Nominalreferenz und Zeitkonstitution. Zur Semantik von Massentermen, Pluraltermen und Aspektklassen. Mnchen: Fink.
Krifka, M. (1992): Thematic Relations as Links Between Nominal Reference and Temporal Constitution. Lexical Matter. I. Sag and A. Szabolsci. Stanford, CSLI: 29-53.
Krifka, M. (1995): Common Nouns: A Contrastive Analysis of Chinese and English.
The Generic Book. G. N. Carlson and F. J. Pelletier. Chicago, The University of
Chicago Press: 398-411.
Krifka, M. (1998): The Origins of Telicity. Events in Grammar. S. Rothstein.
Dordrecht, Kluwer Academic Publishers: 197-235.
Kroch, A. S. (1974): The Semantics of Scope of English. MIT Doctoral Dissertation.
Landman, F. (1989): Groups, Part 1 and Part 2. Linguistics and Philosophy 12: 559605, 723-744.
Landman, F. (1996): Plurality. The Handbook of Contemporary Semantic Theory. S.
Lappin. Oxford, Blackwell: 425-457.
298
Angelika Kratzer
Landman, F. (2000): Events and Plurality. The Jerusalem Lectures. Dordrecht, Kluwer
Academic Publishers.
Lasersohn, P. (1988): A Semantics for Groups and Events. Ohio State University Doctoral Dissertation.
Lasersohn, P. (1990): Group Action and Spatio-Temporal Proximity. Linguistics and
Philosophy 13: 179-206.
Lasersohn, P. (1995): Plurality, Conjunction and Events. Dordrecht, Kluwer Academic
Publishers.
Lin, J.-W. (1998): Distributivity in Chinese and its Implications. Natural Language
Semantics 6(2): 201-243.
Link, G. (1983): The logical analysis of plurals and mass terms: A lattice-theoretic
approach. Meaning, Use, and Interpretation of Language. R. B. e. al. Berlin, De
Gruyter: 302-323.
Link, G. (1998): Algebraic Semantics in Language and Philosophy. Stanford: CSLI
Publications.
Marantz, A. (1984): On the Nature of Grammatical Relations. Cambridge/Mass., MIT
Press.
Matthewson, L. (2000): On Distributivity and Pluractionality. Proceedings of SALT X.
B. Jackson and T. Matthews. Ithaca/N.Y., CLC Publications.
Moltmann, F. (1996): Parts and Wholes in Semantics. Oxford, Oxford University Press.
Morzycki, M. (2004): Modifier Interpretation and Functional Structure. University of
Massachusetts Doctoral Dissertation.
Mller, A. (2000): The Expression of Genericity in Brazilian Portuguese. U M O P 23:
Issues in Semantics. K. Kusumoto and E. Villalta. Amherst/Mass., GLSA: 137-154.
Oh, Se-Rang (2001): Distributivity in an Event Semantics. Proceedings of SALT XI. R.
Hastings, B. Jackson and Z. Zvolenszky. Ithaca/N.Y., CLC Publications.
Pylkknen, L. (2001): What applicative heads apply to. Proceedings of the 24th Annual
Penn Linguistics Colloquium. M. Fox, A. Williams and E. Kaiser. Philadelphia,
Department of Linguistics, University of Pennsylvania.
Pylkknen, L. (2002): Introducing Arguments. M I T Doctoral Dissertation.
Rullmann, H. and A. Yu (2003): General Number and the Semantics and Pragmatics of
Indefinite Bare Nouns in Mandarin Chinese. University of Calgary.
Sanches, M. and L. Slobin (1973): Numeral Classifiers and Plural Marking: an Implicational Universal. Working Papers in Language Universale 11: 1-22.
Sauerland, U. (1998): Plurals, Derived Predicates and Reciprocals. The Interpretative
Tract. U. Sauerland and O. Perus. Cambridge/Mass., M I T Working Papers in Semantics. 25: 177-204.
Sauerland, U., J. Anderssen, and K. Yatsushiro (2004): The Plural Involves Comparison. Tbingen Conference on Linguistic Evidence.
299
Sauerland, Uli (2005): On the Semantic Markedness of -Features. Manuscript, ZAS

Berlin.
Scha, R. (1981): Distributive, Collective and Cumulative Quantification. Formal Methods in the Study of Language, Part 2, Mathematical Center Tracts 136. J. Groenendijk, T. Janssen and M. Stokhoff. Amsterdam, University of Amsterdam: 483512.
Scha, R. J. H. (1984): Distributive, Collective and Cumulative Quantification. Truth,
Interpretation, and Information. J. Groenendijk, T. Janssen and M. Stokhof.
Dordrecht, Foris: 131-158.
Schein, . (1986): Event Logic and the Interpretation of Plurals. MIT Doctoral Dissertation.
Schein, . (1993): Plurals and Events. Cambridge/Mass., MIT Press.
Schwarzschild, R. (1990): Against Groups. Proceedings of the Seventh Amsterdam
Colloquium. M. Stokhof and L. Torenvliet. Amsterdam, Institute for Logic, Language and Information, Universiteit van Amsterdam: 475-493.
Schwarzschild, R. (1991): On the Meaning of Definite Plural Noun Phrases. University
of Massachusetts Doctoral Dissertation.
Schwarzschild, R. (1992): Types of Plural Individuals. Linguistic and Philosophy 15:
641-676.
Schwarzschild, R. (1993-94): Plurals, Presuppositions, and Sources of Distributivity.
Natural Language Semantics 2: 201-248.
Schwarzschild, R. (1996): Pluralities. Dordrecht, Kluwer Acadmic Publishers.
Sternefeld, W. (1998): Reciprocity and Cumulative Interpretation. Natural Language
Semantics 6: 303-337.
Winter, Y. (2000): "Distributivity and Dependency." Natural Language Semantics 8(1):
27-69.
Yang, Rong (2001): Common Nouns, Classifiers, and Quantification in Chinese. Rutgers University Doctoral Dissertation.
Zimmermann, M. (2002): Boys Buying Two Sausages Each. On the Syntax and Semantics of Distance Distributivity. University of Amsterdam Doctoral Dissertation.
Zucchi, S. and M. White (2001): Twigs, Sequences and the Temporal Constitution of
Predicates. Linguistics and Philosophy 24(2): 223-270.
Zwicky, A. M. and J. M. Sadock (1975): Ambiguity tests and how to fail them. Syntax
and Semantics 4. J. Kimball. New York, Academic Press: 1-36.
Kimiko Nakanishi (Calgary)
Event quantifcation and distributivity*

1 Introduction
It is well known that sentences with plural subjects involve a semantic ambiguity that is not available with singular subjects. For example, while John made a
chair simply states that John made a chair, John and Bill made a chair could
mean that John and Bill together made a chair (collective reading) or that each
of them made a chair (distributive reading). Similarly, the Japanese and German sentences in (la) and (lb) have the collective reading that three boys
together made one chair and the distributive reading that each of the three boys
made a chair. 1 The empirical data examined here come from what I call split
quantifier constructions in (2), where a quantificational expression appears
away from its host noun. It has been observed in the previous literature that the
Japanese split quantifier construction (i.e. so-called floating quantifier construction) allows for a distributive reading, but not for a collective reading
(Terada 1990, Kitagawa and Kuroda 1992, Ishii 1999, Kobuchi-Philip 2003,
among others): unlike the non-split quantifier construction in (la), the split
quantifier construction in (2a) has the distributive reading only. 2 The same
contrast can be observed in German, as in (lb) and (2b).
(1)
(a)
[Otokonoko
san-nin]-ga
kinoo
isu-o
tukut-ta.
[boy
three-CL]-NOM
yesterday
chair-ACC
make-PAST
'Three boys m a d e a chair yesterday.'

(b)
[Drei
Jungen]
haben
einen
Stuhl
gebaut.
[three
boys]
have
chair
built
^/collective, ^ d i s t r i b u t i v e
I w o u l d l i k e to t h a n k M a r i b e l R o m e r o , K l a u s Abels, A n g e l i k a Kratzer, a n d B e a t r i c e Santorini

f o r v a l u a b l e c o m m e n t s a n d d i s c u s s i o n s . T h a n k s are also d u e to the a u d i e n c e at W o r k s h o p o n
E v e n t S t r u c t u r e s in L i n g u i s t i c F o r m a n d Interpretation.
A b a r e n o u n in J a p a n e s e c a n b e interpreted as singular or plural. T h u s , the b a r e n o u n isti in (1)
is a m b i g u o u s b e t w e e n ' a c h a i r ' a n d ' c h a i r s ' . S e e section 3.2 b e l o w f o r f u r t h e r d i s c u s s i o n .
In s o m e p r e v i o u s studies, a distinction b e t w e e n distributive a n d n o n - d i s t r i b u t i v e r e a d i n g s is
m a d e d e p e n d i n g on w h e t h e r or n o t m u l t i p l e e v e n t s o c c u r s i m u l t a n e o u s l y ( K i t a g a w a a n d K u r o d a 1992, Ishii 1999, f o r e x a m p l e ) . T h e distributive-collective d i s t i n c t i o n h e r e has n o t h i n g to
d o w i t h a t e m p o r a l relationship, b u t it is d u e to a g e n t h o o d : a distributive r e a d i n g obtains w h e n
e a c h i n d i v i d u a l in the e x t e n s i o n of the p l u r a l s u b j e c t is t h e a g e n t of the r e l e v a n t event, a n d a
collective r e a d i n g o b t a i n s w h e n a g r o u p r e p r e s e n t e d b y t h e p l u r a l s u b j e c t is t h e agent.
302
(2)
Kimiko Nakanishi
(a)
(b)
Otokonoko-ga
kinoo
san-nin
isu-o
tukut-ta.
boy-NOM
yesterday
three-CL
chair-ACC
make-PAST
Jungen
haben
drei
einen
Stuhl
gebaut.
boys
have
three
chair
built
??collective, ^/distributive
In this paper, I argue that split quantifier constructions involve the measurement of events. The mechanism of event measurement requires a homomorphism h (a structure-preserving function) from events to individuals and that a
measure function applies to the range of h, i.e to individuals mapped from
events. In this way, the measure function in split quantifier constructions indirectly measures events by measuring individuals. I show that this mechanism
satisfactorily accounts for why split quantifier constructions lack collective
readings in (2). Furthermore, I present some examples where a collective reading is available in split quantifier constructions and show that the proposed
analysis is able to account for these cases.
The organization of the paper is as follows. In section 2, I briefly summarize previous studies on distributive and collective readings. In section 3, I
examine various empirical data on collective and distributive readings of split
quantifier constructions. I argue that split quantifier constructions involve
measurement in the verbal domain. Section 4 shows that the proposed analysis
is capable of handling the whole range of data. Section 5 concludes the paper.
2 Previous studies on distributivity

Before examining the data on split quantifier constructions, I briefly summarize the previous literature on distributivity. First, I introduce the lexical distinction between distributive and collective predicates, and then I move onto
Landman's theory of distributivity as semantic plurality (Landman 1989a,
1989b, 1996, 2000). 3
2.1 Distributive and collective predicates
As mentioned in section 1, sentences with plural subjects sometimes evoke
semantic ambiguity between distributive and collective readings (see (1)).
Sentences with plural subjects may also be unambiguous, allowing either only
a collective reading, as in (3a), or only a distributive reading, as in (3b).
(3)
(a)
The students are numerous. / John and Mary are a couple,
(b)
The babies are asleep. / Ann and Beth are pregnant.
See Winter (2001) for a possibly relevant distinction between atom and set predicates, which is
proposed as an alternative for the distinction between distributive and collective predicates.
303
These observations lead some researchers to the assumption that there are
inherently collective, inherently distributive, and ambiguous predicates (Link
1984, Dowty 1987, Roberts 1990, Lasersohn 1995, in particular). Under this
view, a distributive reading arises when a distributive operator is present on the
VP, while a collective reading arises when there is no such operator. The distributive operator D is defined in (4), where < is a part-of relation and ATOM
stands for the property of being an atomic element la Link (1983). 4 This
operator makes the relevant verbal predicate apply to all atomic members of
the plural subject x. Following Link, I assume that the extension of plural
subjects is the sum of individuals (Ui). For instance, the extension of John and
Bill is jUib, and the extension of the students would be a u j b u j c when there are
three students a, b, and c. With the D-operator, we obtain (5) as the truth
conditions of {John and Bill / the students} lifted the piano.(5a) means that all
atomic members of j u i b , namely, John and Bill, lifted the piano, yielding the
distributive reading 'John lifted the piano and Bill lifted the piano'. A collective reading obtains when there is no distributive operator and lift the piano
simply applies to the plural subject j u ^ , meaning John and Bill together lifted
the piano, i.e. ftJohn and Bill lifted the piano]] = lift.the.piano(jUib).
(4)
(5)
(a)
lift.the.piano(juib) = Vy [ y < J U I B
(b)
lift.the.piano(aU!bUic) = Vy [ y < aUibUiC ATOM(y) -> lift.the.piano(y) ]
P(x) = Vy [ y < ATOM(y) -> P(y) ]
ATOM(y)
lift.the.piano(y) ]
2.2 Plurality of events and Landman's distributivity

Landman (1989a, 1989b, 1996, 2000) proposes a novel analysis that distributivity is reduced to a semantic pluralization of verbal predicates. First, Link's
(1983) operation of semantic pluralization * needs to be introduced. The *operator (the 'star'-operator) applies to a one-place predicate and generates
all the individual sums of members of the extensions of P. For instance, if the
denotation of dog is {x, y, z}, the denotation of *dog or dogs would be {x, y, z,
xuiy, xUiZ, yuiz, xuiyuiz}. *P is closed under sum formation: any sum of
parts that are *P is also *P. Link treats nominal predicates such as are boys as
inherently distributive, based on the assumption that their interpretation is the
same as the plural NP that they are based on, i.e. the denotation of are boys is
*boy. Then the denotation of John and Bill are boys is *boy(ju!b). Landman
shows that (6) is guaranteed by the definition of * and by the fact that the domain D is an atomic part-of structure. Moreover, (6) in turn guarantees that
John and Bill are boys is distributive, i.e. John is a boy and Bill is a boy. The
same paraphrase obtains for distributive readings in general; the denotation of
Link's original definition is given in (i).

(i) DistriP) <-> Vx ( P(x) -> ATOM(x) )
(Link 1983:309)
304
Kimiko Nakanishi
John and Bill made a chair is *make.a.chair(juib), yielding a distributive

reading that John made a chair and Bill made a chair.
(6)
FACT:
if is a set of atoms, thenct s *P iff Va e (): a G
(Landman 2000:148)
Landman's analysis is further supported by the fact that so-called collective

predicates such as meet can have distributive readings; the boys and the girls
met has a distributive reading where the boys met and the girls met (in a different room). Landman proposes the group-forming operation that maps a sum
of individuals (e.g. the sum of the boys; xL^y) to an atomic group individual
(e.g. the boys as a group; T(xUiy)). Under a distributive reading, a pluralized
predicate *meet takes the sum of the two groups in its extension, as in
*meet((xUiy)UiT(aUibU!c)), where T ( x u i y ) is the boys as a group and
T(auibuic) is the girls as a group. Since groups are atomic elements, the
predicates distribute to the two groups, and not all the way down to the individuals in the groups. In this analysis, a collective reading obtains when a
group applies to a singular predicate, as in make.a.chair((juib)) for John and
Bill made a chair. Crucially, Landman claims that "basic predicates never take
sums in their extension" (Landman 1989a:593), hence make.a.chair(juib) is
not a legitimate denotation. With this claim, Landman reaches to a simple
generalization: when a verbal predicate is singular, it applies to an atomic
(singular or group) individual. When a verbal predicate is pluralized, it applies
distributively to a plural sum of atomic individuals.
3 Split quantifier constructions and distributivity

Having seen some previous studies on distributivity, I now turn to various data
on collective and distributive interpretations of split quantifier constructions.
3.1 The data
In the introduction, we have seen that split quantifier constructions allow only
for distributive readings, while their non-split counterparts allow for both distributive and collective readings (see (1) and (2)). I provide here further examples to illustrate this point. (7b) and (8b) only mean that each student found a
ten-dollar bill, although (7a) and (8a) have the reading where three students
together found one bill. Similarly, (9a) and (10a) can mean that each of the two
friends got married to someone (there were two couples) or that two friends
married each other (there was one couple), while (9b) and (10b) have the former distributive reading only.
(7)
(a)
[Gakusei
san-nin]-ga
mitibata-de juu-doru-satu-o
hirot-ta.
[student
three-CL]-NOM
roadside-by ten-dollar-bill-ACC
find-PAST
305
(b)
(8)
(a)
'Three students found a ten-dollar bill by the roadside.'

Vdistributive, Vcollective
Gakusei-ga
mitibata-de san-nin
juu-doru-satu-o
student-NOM
roadside-by three-CL
ten-dollar-bill-ACC
Vdistributive, ??collective
[Drei Kinder]
haben
einen Zehn-Euro-Schein im Straengraben
[three children] have

a
ten-Euro-bill
'Three children found a ten-Euro bill in a gutter.'
(b)
hirot-ta.
find-PAST
in street-ditch
Kinder haben
drei
einen Zehn-Euro-Schein im Straengraben
children have
three a
ten-Euro-bill
in street-ditch
gefunden,
found
gefunden,
found
Vdistributive, Collective
(9)
(a)
[Tomodati
huta-ri]-ga
kyonen
kekkonsi-ta.
[friend
two-CL]-NOM
last year marry-PAST
'Two friends got married last year.'
(b)
Tomodati-ga
kyonen
friend-NOM
last year
huta-ri
two-CL
kekkonsi-ta.
marry-PAST
Vdistributive, ""collective
(10)
(a)
(b)
[Zwei Bekannte]
haben
gestern
geheiratet,
[two acquaintances] have

yesterday
'Two acquaintances manied yesterday.'
married
Bekannte
haben gestern
acquaintances have yesterday
geheiratet,
zwei
two
married
Vdistributive, Collective
3.2 The analysis

The data in the previous section revealed that split quantifier constructions are
semantically more restricted than their non-split counterparts in terms of collective readings. In Nakanishi (2003, 2004, 2005), I presented further data
showing that semantic restrictions of split quantifier constructions are manifested in terms of different semantic properties. First, unlike non-split quantifier constructions, split quantifier constructions are incompatible with singleoccurrence events such as kill Peter, as in (11) and (12), and with individuallevel predicates, as in (13) and (14).
(11)
(a)
(b)
[Gakusei
san-nin]-ga
kinoo
Peter-o
korosi-ta.
[student
three-CL]-NOM
yesterday
Peter-ACC
kill-PAST
'Three students killed Peter yesterday.'

?? Gakusei-ga
kinoo
san-nin
student-NOM yesterday
three-CL
Peter-o
korosi-ta.
Peter-ACC
kill-PAST
306
(12)
Kimiko Nakanishi
(a)
[Drei
Studenten]
haben
Peter
umgebracht,
[three
students]
have
Peter
killed
'Three students killed Peter.'

(b)
?? Studenten haben
students have
(13)
(a)
Peter
drei
umgebracht.
Peter
three
killed
Kono
kurasu-de
[gakusei
san-nin]-ga
kasikoi.
this
class-in
[student
three-CL]-NOM
smart
'Three students are smart in this class.'
(b)
(14)
(a)
?? Gakusei-ga
kono
student-NOM this
kurasu-de
class-in
[Drei
[three
sind
are
Feuerwehrleute]
firemen]
san-nin
three-CL
kasikoi.
smart
intelligent,
intelligent
'Three firemen are intelligent.'

*Feuerwehrleute
firemen
sind
are
drei
three
intelligent,
intelligent
(b)
I argued that this is because the measure function in split quantifier constructions measures events, while the measure function in non-split quantifier constructions measures individuals (see Nakanishi 2003, 2004, 2005 for details).
The extension of single-occurrence events is a singleton, hence it cannot be
measured by the measure function associated with a split quantifier. This
analysis can further account for the fact that split quantifier constructions are
incompatible with I-level predicates. Kratzer (1995) argues that, unlike S-level
predicates, I-level predicates lack event arguments in their denotation. It follows that a split quantifier is not compatible with I-level predicates that lack
event arguments.
However, the situation is not so simple. On the one hand, a split quantifier
contains a classifier or a measure word that correlates with the host NP. For
instance, in (7b) above, the split quantifier contains a classifier -nin, which
semantically agrees with the host NP gakusei 'student', indicating that san-nin
'three-classifier' must express the cardinality of the students. On the other
hand, the incompatibility with single-occurrence events and with I-level predicates clearly indicates some restriction in the verbal domain, which can be
explained straightforwardly if we assume that the measure function in split
quantifier constructions applies to events. To solve this dilemma, I propose a
mechanism that maps events to individuals and, with the help of this mapping,
the measure function in split quantifier constructions applies to individuals
mapped from events. In this way, split quantifier constructions indirectly
measure events by measuring individuals. This mechanism is motivated by
Krifka's (1989) analysis of temporal adverbials like for two hours in John slept
for two hours (see also Lasersohn 1995). Krifka claims that temporal adverbials cannot apply to events directly, but they can apply to entities which bear
a relation to events, most notably times. That is, for two hours indirectly measures the sleeping event by measuring the run time of the event. Formally, he
307
assumes that there is a homomorphism from events E to event run times T. A

homomorphism A is a function that preserves some structural relation defined
on its domain in a similar relation defined on the range, as in A(eiUEe2) =
/(ei)uT/(e2), where u E and u T are sum operators for events and times, respectively.5 Krifka claims that, given a measure function for times and h from E
to T, we can construe a derived measure function ' for events, as in (15). A
derived measure function describes the transfer of a measure function from one
domain to another. In (15), ' is defined by and h\ for all events, the amount
of the event e measured by ' in E is equal to the amount of /(e) measured by
in T.
(15)
Ve [ '() = (()) ]
(Krifka 1989:97)
Extending Krifka's analysis to the Japanese data, I argue that there is a homomorphism h from events in E denoted by the VP to individuals in I denoted by
the host NP, satisfying /i(e,U[:e2) = /i(e l )u I /i(e 2 ). From the data on non-split
quantifiers, it is clear that measure functions can apply to individuals (e.g. in
three liters of water, the measure function applies to water). Following Krifka,
given a measure function for individuals and h from E to I, we can derive a
measure function ' for events. In (16), a measure function associated with a
non-split quantifier directly applies to a set of individuals (the grey-shaded
area in (16)) and returns measured amounts. In contrast, the measure function
' associated with a split quantifier in (17) applies to a set of events (the greyshaded area in (17)) and returns measured amounts. As in (15), since ' for
events in (17) amounts to (()), the same measurement as (17) can be represented as in (18); '() in (17) (the measured amount obtained by ' applying
to events) is equal to (()) in (18) (the measured amount obtained by applying to individuals mapped from events), that is, ' for events is a combination of h and for individuals. The measure function in (18) associated with
a split quantifier applies to individuals mapped from events by h, i.e. the range
of h (the grey-shaded area in (18)), indicating that indirectly measures events
by measuring individuals mapped from events by h.
(16)
A measure function associated with a non-split quantifier

\
A homomorphism of the semilattice S, = <S, "> into the semilattice S 2 = <S2, > is a mapping F: S,
S 2 such that F(ab) = F(a)F(b), where
tions (Partee, ter Meulen and Wall 1990:286).
denotes a composition of two func-
308
(17)
(18)
Kimiko Nakanishi
A measure function associated with a split quantifier
cz>
'
measured amount
A measure function associated with a split quantifier
C3
E
measured amount
An issue arises as to what kind of function can serve as a homomorphism h

from events to individuals. Suppose there are two elements of the sort S. Then
h from S to S ' maps the sum of and y in S, i.e. x u s y , to the sum of h(x) and
h(y) in S ' , i.e. ft(x)uS'/i(y), as defined in (19). That is, h must be a function and
it must be structure preserving. The structure preserving nature of h is reminiscent of the property of cumulativity. Cumulativity as a property of individuals
or events is defined in (20a): if we add two elements in the extension of some
predicate, the sum of the two is also in the extension of the same predicate (cf.
(22) below). Cumulativity as a property of relations between two sorts is defined in (20b). We can further define cumulativity as a property of functions
from elements in the sort S to elements in the sort S ' , as in (20c). Functional
cumulativity in (20c) essentially expresses the defining property for h in (19):
F(aUsb)=F(a)Us'F(b). Thus, any relation that is functional and cumulative can
serve as h. The agent function is functional and cumulative, as in (21a), which
is equivalent to (21b) (see Kratzer to appear). It follows that, when the host NP
of a split quantifier is an external argument, the agent function serves as h from
events to individuals.
(19)
Vft Vx,ye Ds [ A(xu s y) = A(x)u s -/i(y) ]
(20)
(a)
V P V x , y e D s [ [P(x) P(y)]
(b)
V R Va,bs D s Vx,ye Ds [ [R(a)(x) R(b)(y)] - > R ( a u s b ) ( x u s y) ]
(c)
V R Va,be Ds V x , y s Ds- [ [F(a) = F(b) = y] - > F ( a u s b ) = x u s y ]
(21)
P(xusy) ]
(a)
Ve.e'EE Vx,ye A [[Agent(e) = Agent(e') = y] > Agent(eu E e') = xuiy]
(b)
Agent(eiUEe 2 ) = Agent(ei)UiAgent(e 2 )
Before applying the current analysis to split quantifier constructions, I introduce a model-theoretic method of representing extensions of NPs and of VPs.
Link (1983) claims that NPs can be divided into two classes, mass and plural
count NPs on the one hand and singular count NPs on the other. The NPs in
the first class, but not the ones in the second, have cumulative reference (Quine
1960), as in (22) (< is a part-of relation). Link proposes to capture this fact
model-theoretically using a lattice, which is a partially ordered set ordered by a
reflexive, anti-symmetric, and transitive relation. Assuming that the denotation
of NPs is a set of individuals, the cumulative reference of mass and plural
309
count NPs can be expressed by ordering the individuals in the extension. For
instance, consider (23a), where x, y, and are singular individuals, Ui is an
individual sum operator, and the lines indicate the part-of relation < 6 If the
extension of a given NP is a lattice, some members of the denotation of an NP
are a subpart of some other members. For example, suppose that x, y, and are
water, then their sums ( x u ^ , xUjz, yuiz, xUiyUiz) are also water due to the
cumulative reference property, i.e. Jvvaterj is {x, y, z, xUiy, xuiz, yuiz,
XUiyUiz}. Thus, the extension of a mass NP can be modeled as a lattice, as in
(23a). The same argument holds for plural count NPs. In contrast, the denotation of a singular count NP is a set of singular individuals, hence no member is
a subpart of others. That is, unlike the extensions of plural count and mass NPs,
the extension of singular count NPs is not a lattice.
(22)
has cumulative reference iff:
(23)
(a)
VxVy[ P(x) P(y) -. < y
XUiyUiz
XL
(b)
J\Z
P(xusy) ]
eiUEe2UEe3
eiU
JEe3
The denotation of VPs can be expressed by using a lattice if we introduce

event arguments and assume that events can form a lattice, as in (23b), where
e b e 2 , and e 3 are singular events, u E is an event sum operator, and the lines
indicate the ordering part-of relation < (Rrifka 1989, 1992, 1998, Landman
1996, 2000, in particular). Different theories have been proposed depending on
how event arguments are associated with verbs, as in (24) (v for the type of
events). Among these, I adopt Kratzer's (1996, to appear) claim in (24c) that
external arguments are introduced by a neo-Davidsonian method both in the
syntax and at conceptual structure.
(24)
(a)
Davidsonian:
Xx e Ay e Aev. see(x,y,e)
(b)
Neo-Davidsonian:
xe.ye.ev. see(e) Theme(x,e) Agent(y,e)
(c)
Kratzerian:
.. see(x,e)
Under Kratzer's analysis, at the level of the VP, all the internal arguments are
saturated and the VP denotes a set of events of type <v,t> (see (24c)). Take the
atelic VP drive a car, for instance (John drove a car {for/*in} one hour),
where ([drive a carI is Xev.drive(a.car,e). If the members in this set stand in the
part-of relation, the extension of drive a car is a lattice of events, as in (23b).
In Link (1983), besides singular individuals like John, there are plural individuals or individual
sums of type e like JohnUiBill that are different from sets like {John, Bill) (see Schwarzschild
1996 for alternative approaches).
310
Kimiko Nakanishi
Suppose that ei ,e2, e 3 are drive-a-car events. 7 Atelic VPs are like mass NPs in
terms of cumulative reference; if we have two driving-a-car events, the sum of
the two is also a driving-a-car event. Thus, the sums of ei ,e2, e 3 , i.e. e!U E e 2 ,
e]U E e 3 , e 2 u E e 3 , eiU E e 2 u E e 3 , are also drive-a-car events. That is, [[drive a carj
is {ei, e 2 , e 3 , eiU E e 2 , eiU E e 3 , e 2 u E e 3 , eiU E e 2 u E e 3 }. Since the members can be
ordered by the part-of relation, the extension of this atelic VP is a lattice of
events, as in (23b). Telic VPs are analogous to count NPs; in the same way as
a singular count NP like dog denotes a set of atomic individuals, a singular
telic VP like break a car denotes a set of atomic events. Telic VPs can be
pluralized by applying the semantic pluralization operation * used for pluralization in the nominal domain (see section 2.1, see also Landman 1989a,
1989b, 2000). With the help of the *-operator, telic VPs can be semantically
pluralized. A plural telic VP denotes a set containing atomic events and their
sums, just like a plural count NP denotes a set containing atomic individuals
and their sums. For instance, when $?reak a carj is {ei, e 2 , e 3 }, fbreak a car]\
is {ej, e 2 , e 3 , ejU E e 2 , e!U E e 3 , e 2 u E e 3 , eiU E e 2 u E e 3 }, which can be modeled as a
lattice of events, as in (23b). Note that a telic VP that denotes a singleoccurrence event like kill Peter can never denote a lattice: assuming that Peter
dies only once, the killing-Peter event can occur only once. That is, even if we
pluralize it, the extension of kill Peter is always a singleton. In sum, the extensions of a plural telic VP and of an atelic VP are a lattice of events, while the
extension of a singular telic VP is not.
Let us now apply the proposed analysis to some empirical data. In (25), for
example, a homomorphism h maps coughing events to their agents. The measure function then applies to the range of h and picks out a sum of students
whose cardinality is two. (26) and (27) illustrate legitimate h from a lattice of
coughing events to a lattice of students. 8 Importantly, the atomic coughing
events e, e 2 , and e 3 are never be mapped to the sums of individuals xuiy, xuiz,
yuiz, and xuyuiz. This is due to Landman's (1989a) claim that basic predicates never take sums in their extension. Hence, h must map atomic events to
atomic (singular or group) individuals. The measure function applies to a set of
individuals mapped from events, that is, {x, y, z, xuiy, xuiz, yuiz, xuiyuiz}
in (26) and {x, y, x u j y } in (27). Among these sets, the split quantifier picks up
a sum whose cardinality is two, i.e. x u ^ , xuiz, or y u ^ in (26) and xUjy in
(27).
Just like mass NPs, minimal parts of atelic VPs are somewhat vague (see Rothstein 2004, for
instance). In this paper, I simply assume that bottom elements in a lattice of an atelic VP are
events which have the same property as the relevant atelic VP.
A lattice of coughing events can be much larger than the ones in (26) and (27) in that these
events can take individuals who are not in the denotation of *student as an agent. Since what is
relevant for the denotation of (25) is individuals who are students and coughed, I only consider
the relevant portion of the lattice of coughing events.
311
(25)
Gakusei-ga
kono
jup-pun-de
huta-ri
seki-o
si-ta.
student-NOM
this
ten-minute-in
two-CL
cough-ACC
do-PAST
'Two students coughed within the last ten minutes.'
(26)
e 1 u E e 2 u E e3
eiU E 2
ei
eiU E 3
e2uEe'3
e2
e3
xujyuiz
..S. -Ul.!.'.-.
\[*cough]\= {ei, e2, e3, e]U E e 2 , eiU E e 3 , e 2 u E e 3 , e i U ^ u ^ }

* student^ = {x, y, z, xuiy, xuiz, yUiz, xUiyUiz)
(27)
Wcoughl = {ei, e2, e i U ^ }

[[*sludent]] = {x, y, z, xUiy, xUiZ, yUiz, xUiyUiz}
The definition of h permits a one-to-many mapping; as in (28), multiple atomic

events can be mapped to one atomic individual, where y corresponds to two
events. Indeed, (25) is compatible with a scenario where, among two students
who coughed within the last ten minutes, one of them coughed twice. In this
scenario, the number of students who coughed is two, but the number of
atomic coughing events is three. The situation in (28) is compatible with (25),
just like (27) is: the measure function applies to {x, y, xuy}, and it picks up a
sum whose cardinality is two, that is, xUiy.
Note that a one-to-many mapping is not legitimate when the relevant VP has a
singular count NP as an internal argument. Consider first the German example
in (29b), where the extension of ate a cake forms a lattice of events. This example is not compatible with (28), since, in (28), one of the two boys, i.e. y, is
an agent of two eating-a-cake events, that is, y ate two cakes. Unlike (29b),
(29a) in Japanese is compatible with (28). This is because bare nouns in Japa-
312
Kimiko Nakanishi
nese can be interpreted as singular or plural (see footnote 1); (29a) doesn't say
anything about how many cakes the two boys ate. Although the plural interpretation of Japanese bare nouns hasn't been discussed so far, it is certainly available in any example with a bare internal argument. For instance, in (2a), isu
'chair' can be interpreted as singular or plural, hence (2a) means that each boy
made one chair or chairs, yielding three or more chairs as a result. Crucially,
(2a) still lacks the collective reading that three boys together made one chair or
chairs.
(29)
(a)
Otokonoko-ga
sono kafe-de
huta-ri
keeki-o
tabe-ta.
boy-NOM
that cafe-at
two-CL
cake-ACC
eat-PAST
'Two boys ate a cake / cakes at that cafe.'

(b)
Jungen
haben
zwei
einen Kuchen
gegessen,
boys
have
two
eaten
cake
The proposed analysis captures the semantic differences between (25) and (30),
where ni-kai 'twice' is simply counting the number of events without being
associated with the number of students (cf. Doetjes 1997). Crucially, while
(25), with the split two-CL, must involve two students, (30) does not have to:
(30) means that a student or students whose cardinality is unspecified coughed
twice. This semantic difference indicates that the measure function in (25) is
not applying to events directly.
(30)
Gakusei-ga
kono
jup-pun-de
ni-kai
seki-o
si-ta.
student-NOM
this
ten-minute-in
two-time
cough-ACC
do-PAST
student/students coughed twice within the last ten minutes.'
Let us now consider the observation that split quantifier constructions allow
only distributive readings (see section 3.1). Suppose that, in (2) above, a lattice
of making-a-chair events is mapped to a lattice of boys by a one-to-one mapping h, as in (31). The measure function applies to the range of h, i.e. a set of
agents {x, y, z, xUiy, xuz, yuiz, xujyuiz}. The split quantifier picks out the
member whose cardinality is three, i.e. xuiyuiz. xUiyUjZ consists of x, y, z,
each of whom is an agent of an atomic making-a-chair event ei, e 2 , e 3 , which
yields a distributive reading. As discussed above, Japanese further allows a
one-to-many mapping like the one in (28). Suppose that there are four makinga-chair events ei, e 2 , e 3 , e 4 , and C| is mapped to x, e 2 is to y, and e 3 and e 4 are to
z. The proposed mechanism yields the reading that three boys x u i y u i z are
agents of four building-a-chair events, where and y built one chair each and
built two chairs. Indeed, (2a) is compatible with such a situation.
313
(31)
e1uEe2uEe3
ei
e2
e3
xuiyuiz
l*make a chairj = {eh e2, e3, e i U ^ , eiUEe3, e2UEe3, e i U ^ u ^ )

l*boyJ = { , y, z, xUiy, xu, yuiz, xUiyUiZ )
Under Landman's analysis discussed in the previous section, a collective reading obtains when a predicate is not pluralized and it takes a group individual as
an agent. In my analysis, this would be the case with h from a singleton containing an atomic making -a-chair event e to the group of three students
T(xU[yUiZ). In this case, although the split quantifier needs to pick out the
member whose cardinality is three, there is no such element in the range of h;
the range of h only has ( ^ ( ) whose cardinality is one. In this way, the
proposed analysis correctly rules out the collective reading.
4 Split quantifier constructions and collectivity

In the previous section, I proposed a mechanism where, with the help of a
homomorphism h from events to individuals, the measure function associated
with a split quantifier applies to individuals mapped from events, i.e. the range
of h. This mechanism accounts for why split quantifier constructions lack
collective interpretations. In this section, I show that the proposed analysis
extends to a wider range of data on distributive and collective readings. In
section 4.1,1 present novel data indicating that, depending on the aspectuality
of the VPs, collective readings are available with split quantifier constructions.
Section 4.2 deals with another instance of collective readings, i.e. collective
readings obtained with a collectivizing adverb such as together. In section 4.3,
I introduce another type of reading, namely, the so-called cover reading, which
can also be explained by the current analysis. Section 4.4 discusses collective
interpretations allowed by split quantifier constructions where the host NP is
an internal argument. I propose that these collective readings are not genuinely
collective, hence they are not problematic to the present analysis.
4.1 Collectivity with progressives
In section 3.1,1 showed that, while non-split quantifier constructions have both
collective and distributive readings, split quantifier constructions have only
distributive readings. However, it is not the case that collective readings are
always unavailable in split quantifier constructions: when a VP is in a progres-
314
Kimiko Nakanishi
sive form, a collective reading obtains. Japanese has a morpheme -teiru, which
attaches to a verb and expresses that the relevant event is progressing, like -ing
in English (see Ogihara 1998). The example in (32a) illustrates that, with -teiru,
a telic VP such as make a chair permits a collective reading even in the split
quantifier construction (cf. (2a)). In German, progressive aspect can be expressed by a construction with the preposition an.9 For example, in (32b), the
an construction roughly translates as the -ing progressive in English. In this
progressive context, the split quantifier construction allows ambiguity between
a distributive and a collective reading (cf. (2b)).
(32)
(a)
Otokonoko-ga
kinoo
san-nin
isu-o
tukut-tei-ta.
boy-NOM
yesterday
three-CL
chair-ACC
make-PROG-PAST
'Three boys were making a chair yesterday.'

(b)
Jungen
boys
haben
have
Vcollective, ^/distributive
drei
an
einem
Stuhl
gebaut.
three
PREP
chair
built
Vcollective, ^distributive
In the same vein, although the split quantifier constructions in (33b) and (34b)
with the collective VP build Tokyo Tower are unacceptable, their progressive
counterparts in (33c) and (34c) are acceptable.
(33)
(a)
(b)
(c)
[Sagyooin
hyaku-nin]-ga
Tokyo-de
Tokyo Tower-o
tate-ta.
worker
100-CL]-NOM
Tokyo-de
Tokyo Tower-ACC
0 0 workers built Tokyo Tower in Tokyo.'
build-PAST
*Sagyooin-ga
worker-NOM
Sagyooin-ga
tate-ta.
Tokyo-de
Tokyo-in
Tokyo-ni
hyaku-nin
100-CL
hyaku-nin
Tokyo Tower-o
Tokyo Tower-ACC
Tokyo Tower-o
worker-NOM
Tokyo-in
100-CL
Tokyo Tower-ACC
'100 workers were building Tokyo Tower in Tokyo.'
(34)
(a)
build-PAST
tate-tei-ta.
build-PROG-PAST
[100
Arbeiter]
haben
den Tokio-Tower
ir
Tokio
gebaut,
[100
workers]
have
the Tokyo-Tower
ir
Tokyo
built
ir
ir
Tokio
gebaut,
Tokyo
built
0 0 workers built Tokyo Tower in Tokyo.'

(b)
(c)
*? Arbeiter haben
workers have
100
100
den Tokio-Tower
the Tokyo-Tower
Arbeiter haben
100
an
dem Tokio-Tower
in
Tokio
gebaut,
workers have
100
PREP
the
in
Tokyo
built
Tokyo-Tower
' 100 workers were building Tokyo Tower in Tokyo.'
The an construction is rather restricted in its distribution. For instance, predicates such as read
a chapter, eat a cake, drink a glass of wine are not permitted in the an construction, as in (i).
(i) (a) *Die
Studenten
haben an
einem Kapitel gelesen,
the
students
have
PREP
a
chapter read
(b) *Die
Studenten
haben an
einem Kuchen gegessen,
the
students
have
PREP
a
cake
eaten
(c) ??Die Studenten haben an
einer
Flasche Wein
getrunken,
the
students
have
PREP
a
bottle wine
drunk
315
The question to be addressed is then why split quantifier constructions permit

collective readings with progressive VPs. It has been noted that verbal predicates in progressives are tied to a notion of partiality, as informally defined in
(35) (Bennett and Partee 1972, Krifka 1992). With this definition, the extension of progressive VPs is considered to be a lattice of subevents. For instance,
the VP make a chair in a progressive form may have subparts e', e", e " \ Crucially, e', e", e'" are subevents of a singular making-a-chair event e. These
subevents and their sums form a lattice of subevents, where e ' u E e " u E e " ' corresponds to e. Then the extension of the making-a-chair event is a lattice, as in
(36).
(35) PROG = <,(>'. [ P(e) e'< e e' is not the final subevent of e]
(36)
e uEe uEe '

e'uEe"
e'uEe"'
e"uEe"
[ be making a chairj =
{ e',e",e"',e'Une",e'UEe'" ,e"UEe"',e'uEe"UEe"
In section 3.2 above, I discussed why split quantifier constructions lack collective interpretations. The measurement of events is done by measuring individuals through events with the help of a homomorphism from events to individuals. One possible case is to make use of a one-to-one homomorphism, as
illustrated in (31) above, which necessarily yields a distributive reading: each
making-a-chair event e b e 2 , e 3 is mapped to its agent x, y, z, respectively. Extending this approach to split quantifier constructions with progressive VPs,
we could postulate h in (37), where each subpart e', e", e"'of a making-a-chair
event is mapped to its agent x, y, z, respectively. Crucially, a singular makinga-chair event e, that is, e ' u E e " u E e " ' is mapped to xUjyUiZ, which yields a
collective reading, although (37) is still distributive in that each individual is
an agent of a different subevent.
(37)
e uEe uEe
e ' u E e ; ; " e'u E e'""" e " u E v "
xuiyuiz
xuiy " xUiz
yuiz
Ibe making a chairj = {e\ e", e"\ e'UEe", e'UEe'", e"UEe"', e'UEe"UEe"'}
fboyl = { x, y, z, xuiy, x<uz, yuiz, xUiyU[Z }
The analysis that progressives create subevents receives supporting evidence
from the following Japanese data. Recall the claim in section 3.2 that split
316
Kimiko Nakanishi
quantifier constructions are incompatible with VPs denoting a singleoccurrence event. These examples become acceptable when VPs are in a progressive form, as in (38) and (39). 10 This pattern follows naturally from the
current analysis: in (38a) and (39a), the extension of the VPs is a singleton,
while the extension of the progressive VPs in (38b) and (39b) is a lattice of
subevents. W e further predict that these sentences allow a collective reading.
In fact, since the events of breaking-that-table and killing-Mary can occur only
once, the collective reading is the only reasonable interpretation.
(38)
(a)
llGakusei-ga
kinoo
student-NOM yesterday
(b)
san-nin
sono isu-o
three-CL
that chair-ACC break-PAST
'Three students were breaking that chair yesterday.'

Gakusei-ga
kinoo
san-nin
sono isu-o
kowasi-ta.
kowasi-tei-ta.
student-NOM
yesterday
three-CL
that chair-ACC break-PROG-PAST
'Three students were breaking that chair yesterday.'
(39)
(a)
??John-wa
[gootoo-ga
sokode
san-nin Mary-o
korosi-ta]-to
itta.
John-TOP [robber-NOM there

three-CL Mary-ACC kill-PAST]-C0MP said
'John said that three robbers were killing Mary over there.'
(b)
John-wa
[gootoo-ga
sokode
san-nin Mary-o
korosi-tei-ta]-to
itta.
John-TOP [robber-NOM there

three-CL Mary-ACC kill-PROG-PAST]-COMP said
'John said that three robbers were killing Mary over there.'
In this way, the mechanism of a homomorphism proposed in section 3 can

account for the fact that split quantifier constructions permit collective readings in progressive forms.
4.2 Collectivity with together

Directly relevant to the data presented in this section are so-called "collectivizing" adverbials in English such as together (also as a group,
collectively,
jointly, etc.) (Lasersohn 1995, in particular). When these adverbs appear in
sentences which are otherwise ambiguous between distributive and collective
readings, only a collective reading is available, as in (40). Likewise, in Japanese and German, when collectivizing adverbs co-occur with split quantifiers,
as in (41), only collective readings are available.
10
The German equivalent of these examples is unacceptable, as in (i). However, this may be
because the distribution of the an construction is much more restricted in the first place, as
mentioned in footnote 9.
(i) (a) *Die
Studenten
haben an
Peter
umgebracht,
the
students
have
PREP
Peter
killed
(b) *Die
Studenten
haben an
einem Haus
vernichtet,
the
students
have
PREP
a
house
destroyed
317
(40)
(a)
(b)
(41)
(a)
John and Mary built a table.

John and Mary built a table together.
Otokonoko-ga
kinoo
san-nin
issyoni
isu-o
tukut-ta.
boy-NOM
yesterday
three-CL
together
chair-ACC
make-PAST
'Three boys made a chair together yesterday.'

(b)
Otokonoko-ga
san-nin(-hito-kumi)-de isu-o
boy-NOM
three-CL(-0ne-CL)-C0P
chair-ACC
'Boys made a model boat by three (as a group).'
(c)
Jungen
boys
haben
have
drei
three
zusammen
together
einen
a
tukut-ta.
make-PAST
Stuhl
chair
gebaut,
built
I propose that these collectivizing adverbs rely on the group formation operator
proposed by Landman (1989a, 1989b, 2000), which was introduced in section 2.2. is a type-shifting operator that maps a sum of individuals (e.g.
xuiyuiz) to an atomic group individual (e.g. T(xUiyUiz)). For instance, in (41),
t forms a group of three boys. That is, with t , the split quantifier indicates the
cardinality of individuals in the group. Then (41) means that one group consisting of three boys made a chair, where there was only one agent, namely, a
group of three boys. It follows that the sentences with these collectivizers have
the same status as sentences with the split one, as in (42). 11 1 propose that the
extension of the VP is a lattice of events even with the split one. For instance,
in (42), a lattice of hitting-Peter events is mapped to a lattice of students, and
the split quantifier picks out a member whose cardinality is one. The role of
split one is simply to say that, out of a lattice of indivituals mapped from
events, there is a relevant individual /(e) whose cardinality is one.
(42)
Gakusei-ga
kinoo
hito-ri
Peter-o
tatai-ta.
student-NOM
yesterday
one-CL
Peter-ACC
hit-PAST
O n e student hit Peter yesterday.'
Similarly, I assume that the extension of the VP must be a lattice of events

even with a collectivizing adverb. In (41), each atomic event is mapped to an
atomic group consisting of three boys, as illustrated in (43), which illustrates a
legitimate homomorphism from events to groups.
11
The German ST with one is generally ungrammatical, as

special contexts (e.g. es gibt construction in (ii), etc.).
(i) *Student / Studenten hat
Peter
einen
student / students
has
Peter
one
hit
(ii) Kopiergerte
gibt
es hier
nur
photocopiers.ACC
there
is here
only
"There is only one photocopier here.'
in (i), except for the examples in some

geschlagen,
eines.
one.ACC
(Nolda 2000:3)
318
(43)
Kimiko Nakanishi
eiU E e 2 u E e3
T(aUibu ! c)UiT(dUieUif)u I T(gUihUii)
eiU E e 2 eiU E e 3 e 2 u E e 3
t(auibu,c)u]
t(auibuic)
T(du!euif)
T(gu t huii)
Note that the analysis proposed here is analogous to Landman's analysis of

collective predicates like meet. He points out that these predicates allow distributive readings: the boys and girls met has a reading where a group of boys
met and a group of girls met independently (see section 2.2). This is because
the extension of the boys and girls can be a lattice consisting of two group
atoms, i.e. a group of boys and a group of girls. In the same vein, in split quantifier constructions with collectivizers, the extension of agents can be a lattice
consisting of group atoms, as illustrated in (43).
In section 3.1, I showed that split quantifier constructions lack collective
readings. What is relevant here is that the degree of unacceptability seems to
vary among the informants. In fact, some informants said that collective readings do not seem to be completely unacceptable. I suggest that this is because
the informants manipulate the existence of collectivizing adverbs. This is especially plausible for Japanese. As in (41b), by simply having -de ' - c o p u l a '
following the split quantifier, collective interpretations obtain. Hence, even
when the split quantifier is not followed by -de, speakers may obtain 'illusive'
collective interpretations by positing a covert -de.
4.3 Cover readings

There is another reading relevant to the discussions here, namely, cover readings (Gillon 1987, Schwarzschild 1991, 1996, Verkuyl and van der Does 1991,
van der Does 1992, Brisson 1998, 2003). For instance, (44a) is true when,
among three composers Ann, Beth, and Colin, Ann wrote musicals, and Beth
and Colin together wrote musicals. Another example is given in (44b), where
Landman (2000) claims that the most plausible reading is a cover reading:
different groups of firefighters (maybe overlapping) put out fires, and those
groups together make up 400 firefighters. The distributive and collective readings are less plausible: each of 400 firefighters put out the fires or a single
group of 400 firefighters put out the fires. Indeed, the Japanese non-split quantifier construction in (45a) shows the same preference. However, the corresponding split quantifier construction in (45b) does not allow the cover reading
nor the collective reading. Rather, it only allows the distributive reading where
each of 400 firefighters put out fires. Thus, (45b) is most naturally used in a
319
scenario such as follows: there was a contest participated in by 400 firefighters

where each of them put out fires.
(44)
(a)
(45)
(b)
(a)
(b)
(Gillon 1987)
Three composers wrote musicals.

400 firefighters put out the fires in Colorado.
[Syooboosi yonhyaku-nin]-ga
kororado-de
kaji-o
kesi-ta.
[firefighter 400-CL]-NOM
fire-ACC
put OUt-PAST
Colorado-in
(Landman 2000:125)
'400 firefighters put out the fires in Colorado.'

Syooboosi-ga
kororado-de
yonhyaku-nin
kaji-o
kesi-ta.
firefighter-NOM Colorado-in
fire-ACC
put out-PAST
400-CL
In Japanese, the classifier -nin is used for individual persons; in (45b), -nin
must count individual firefighters, that is, -nin must be associated with individual atoms. Thus, in the extension of the host NP, an atomic element in a
lattice must be an individual firefighter. The collective reading would obtain
when the extension of the NP contains a group consisting of 400 firefighters.
Although the split quanitfier picks out an element whose cardinality is 400,
cardinality of the group is one. Hence, the collective reading is unavailable
(see section 3.2 for details). The cover reading obtains when each atomic element is a group of unknown number of firefighters and '400-CL' indicates the
total number of firefighters. For example, take the situation illustrated in (46),
which is analogous to the one proposed in (43) for together. To achieve the
cover reading, we need to ensure that the total number of individual firefighters is 400. However, in (46), there are no individual atoms that can be associated with -nin, hence the cover reading is unavailable. The closest reading
available would be the one presented in (47). Besides a classifier for individual
atoms, Japanese has the classifier -kumi that is used for group atoms. For example, in (47), -kumi must count groups of firefighters, that is, -kumi must be
associated with group atoms. In this way, with -kumi, it is possible to count the
number of group atoms in (46).
(46)
e1uEe2uEe3
eiUE2
ei
(47)
eiU E e 3
e2
(au1bu1c)uit(duie)u1(fu]g)
e 2 u E e 3 T(auibuic)uit(du!e)
e3
t(auibUic)
...
t(duie)
(fUig)
Syooboosi-ga
kororado-de
yonhyaku-kumi
kaji-o
kesi-ta.
firefighter-NOM
Colorado-in
400-CL
fire-ACC
put out-PAST
'400 groups of firefighters put out the fires in Colorado.
320
Kimiko Nakanishi
4.4 Collectivity with internal arguments

So far, I have examined split quantifier constructions where the host NP is a
subject, or more specifically, an external argument. As shown in (48), the host
NP in these constructions can be an internal argument.
(48)
(a)
(b)
san-satu
yon-da.
John-NOM book-ACC yesterday

'John read three books yesterday.'
John-ga
hon-o
kinoo
three-CL
read-PAST
Bcher
hat
Hans
gestern
drei
gelesen,
books
has
Hans
yesterday
three
read
In the following, depending on whether a split quantifier is associated with an

external argument or an internal argument, split quantifier constructions are
referred to as external split quantifier constructions or internal split quantifier
constructions. In this section, I first show that internal split quantifier constructions seem to allow a collective reading. However, this reading is different in
nature from the collective reading prohibited in external split quantifier constructions. I propose that the asymmetry between external and internal arguments discussed in section 4.3 is responsible for the difference in collective
readings.
The categorization of predicates based on distributivity is generally defined with respect to plural subjects, or more precisely, external arguments.
Little has been said on distributivity with respect to plural objects (or internal
arguments). The only previous references I have found are Dowty (1987),
where enumerate and count are categorized as collective predicates and summarize as an ambiguous predicate, and Landman (2000), where combine is
argued to be collective in its object position. As an ambiguous predicate, I use
mix, as in (49). The distributive reading is that John made three different cocktails separately, say, martini, margarita, and pia colada. The collective reading
is that John put together three cocktails and made a mysterious drink. The split
internal quantifier constructions in (49) have these two readings. The examples
in (50) show that they are compatible with a collective predicate such as pile
(49)
(a)
John-ga
kakuteru-o
syeikaa-de
san-bai
maze-ta.
John-NOM
cocktail-ACC
shaker-by
three-CL
mix-PAST
'John mixed three cocktails by shaker.'

(b)
Getrnke
drinks
hat
has
Hans
Hans
drei
three
gemixt,
mixed
'Hans mixed three drinks.'
12
Other collective predicates with respect to a plural internal argument I came up with are collect,
assemble, gather, sum, accumulate, separate, and divide.
(50)
(a)
John-ga
hako-o
heya-ni
juk-ko
tumikasane-ta.
John-NOM box-ACC
room-in
ten-CL
'John piled up ten boxes in the room.'
(b)
Ksten
hufte
Hans
gestern
boxes heaped Hans

yesterday
'Hans piled up ten boxes yesterday.'
321
pile up-PAST
zehn
an.
ten
on
It seems then that internal split quantifier constructions allow collective readings, unlike external split quantifier constructions. I argue that the collective
readings with respect to a plural internal argument are different in nature from
collective readings with respect to a plural external argument. For instance, in
(50), if John piled up ten boxes, then he must have piled up nine boxes, eight
boxes, and so on. Compare this with ten boys built the statue, where there is no
entailment that the smaller number of boys could have built the statue. In this
sense, although pile up is 'collective' in that it is incompatible with one (*John
piled up one box), it is different from genuinely collective predicates with
respect to an external argument where there is no entailment. Consider further
the distributive predicate sleep. If ten boys slept, then nine boys must have
slept, eight boys must have slept, and so on. This suggests that a collective
reading in internal split quantifier constructions is not really 'collective', but
rather distributive. There seem to be no genuine collective readings that do not
have this entailment with respect to a plural internal argument. A piece of
supporting evidence for this generalization comes from the observation that the
distributive quantifier every is inappropriate in subject position of collective
predicates (Roberts 1990), but not in object position (Landman 2000), as in
(51).
(51)
(a)
?Every boy meets.
(b)
In this class I will try to combine every semantic theory that has been proposed in the
literature.
(Landman 2000:83)
Assuming that the generalization that internal split quantifier constructions

lack genuine collective readings is on the right track, the question is why. I
propose that this is because there is an inherent incremental relationship between an event and its internal argument, but not between an event and its
external argument. 13 For instance, in eat an apple, there is an incremental
relationship between the eating event and its internal argument an apple in that,
as the eating event proceeds, the amount of apple consumed increases. In contrast, it has been claimed that there is no such incremental relationship found
13
Previous studies have argued for various types of incremental relationships: "ADD-TO" property (Verkuyl 1972, 1993), "measuring out" (Tenny 1987, 1994), "graduality" (Krifka 1989,
1992), "incremental theme" (Dowty 1991), and "structure-preserving binding relations"
(Jackendoff 1996) (see Krifka 1998:198-199 for summary).
322
Kimiko Nakanishi
between an event and its external argument (Tenny 1987, 1994, in particular). 14 Let us now go back to the data on distributivity. In John piled up ten
boxes, there is an incremental relation between the piling up event and boxes,
that is, as the piling up event proceeds, the number of boxes increases. This
amounts to the entailment described above. In contrast, in ten boys built the
statue, there is no incremental relationship between ten boys and the buildingthe-statue event, hence there is no entailment of the kind found with pile up ten
boxes. Summing up, since an internal argument always has an incremental
relationship with an event, it never allows a genuine collective reading without
entailment. I take this to mean that the seemingly collective readings in (49)
and (50) are not genuinely collective, hence they are different in nature from
genuine collective readings with external arguments.
5 Conclusion
In this paper, I presented various empirical data on collective and distributive
interpretations of split quantifier constructions, which is summarized in (52). I
proposed a mechanism of event measurement that makes use of a homomorphism from events to individuals. This mechanism is capable of handling the
whole range of data.
(52)
(a)
Split quantifier constructions generally disallow collective readings.
(b)
(c)
(d)
(e)
Split quantifier constructions allow collective readings with a progressive VP.

Split quantifier constructions allow collective readings with a collectivizer.
Split quantifier constructions disallow cover readings.
Internal split quantifier constructions lack genuine collective reading.
14
Dowty (1991) and Krifka (2001) argue that there can be an incremental relationship between
an event and its external argument, as in (i) and (ii). However, these examples are not strong
counter-examples to the generalization that the incremental relationship holds only between an
event and its internal argument. For (i), we could say that the external argument of a movement
verb is special in that it has a double role of being an agent and a moved object, as claimed in
Krifka (2001:7). As for (ii), the incrementality found with quantified NPs such as, fifty customers is different in nature from the incrementality with singular NPs, in that, in the former case,
the incrementality can be forced by pluralizing both individuals and events, yielding a homomorphic relation between the two (cf. Dowty 1991:570). Thus, a singular external argument is
different from a singular internal argument in that it does not have an incremental relationship
with an event.
(i) (a) John entered the icy water (very slowly).
(b) John crossed the desert in a week.
(Dowty 1991:570-571)
(ii) Fifty customers complained about the product in two days.
(Krifka 2001:7)
323
References
Bennett, M . and B. Partee (1972): T o w a r d the Logic of T e n s e and A s p e c t in English.
Santa Monica: System Development Corporation.
Brisson, C. (1998): Distributivity, M a x i m a l i t y , and Floating Quantifiers. P h . D . dissertation, R u t g e r s University.
Brisson, C. (2003): Plurals, all, and the n o n u n i f o r m i t y of collective predication. Linguistics and P h i l o s o p h y 26, 129-184.
van der D o e s , J. (1992): Applied Quantifier Logics. P h . D . dissertation, University of
Amsterdam.
D o e t j e s , J. (1997): Quantifiers and Selection: O n the Distribution of Q u a n t i f y i n g Exp r e s s i o n s in F r e n c h , D u t c h and English. H a g u e : H o l l a n d Institute of Generative
Linguistics.
D o w t y , D. (1987): Collective predicates, distributive predicates, and all. T h e Proceedings of E a s t e r n States C o n f e r e n c e o n Linguistics ( E S C O L ) '86, 9 7 - 1 1 5 .
D o w t y , D. (1991): T h e m a t i c proto-roles and a r g u m e n t selection. L a n g u a g e 67, 547619.
Gillon, B. (1987): T h e r e a d i n g s of plural n o u n phrases in English. Linguistics and
P h i l o s o p h y 10, 199-220.
Ishii, Y. (1999): A n o t e o n floating quantifiers in Japanese. In: M . M u r a k i and E. Iwam o t o : Linguistics: In Search of the H u m a n M i n d , A Festschrift for K a z u k o Inoue.
T o k y o : Kaitakusha, 2 3 6 - 2 6 7 .
J a c k e n d o f f , R. (1996): T h e proper treatment of m e a s u r i n g out, telicity and p e r h a p s even
quantification in English. Natural L a n g u a g e and Linguistics T h e o r y 14, 3 0 5 - 3 5 4 .
K i t a g a w a , Y. and S.-Y. K u r o d a (1992): Passive in Japanese. M s . University of R o c h e s ter and University of California, S a n Diego.
K o b u c h i - P h i l i p , M . (2003): Distributivity and the J a p a n e s e Floating N u m e r a l Quantifier. P h . D . dissertation, T h e City University of N e w York.
Kratzer, A. (1995): Stage-level predicates and individual-level predicates. In: G. Carlson and F. J. Pelletier: T h e G e n e r i c B o o k . Chicago: T h e University of C h i c a g o
Press, 125-175.
Kratzer, A. (1996): Severing the external a r g u m e n t f r o m its verb. In: J. R o o r y c k and L.
Zaring: P h r a s e Structure and the L e x i c o n . D o r d r e c h t : Kluwer, 109-137.
Kratzer, A. (to appear): T h e E v e n t A r g u m e n t and the S e m a n t i c s of V e r b s .
K r i f k a , M . (1989): N o m i n a l reference, temporal constitution and quantification in event
semantics. In: R. Bartsch, J. van B e n t h e m and P. van E m d e Boas: S e m a n t i c s and
C o n t e x t u a l E x p r e s s i o n . D o r d r e c h t : Foris, 7 5 - 1 1 5 .
K r i f k a , M . (1992): T h e m a t i c relations as links b e t w e e n n o m i n a l r e f e r e n c e and temporal
constitution. In: I. A. Sag and A. Szabolcsi: Lexical Matters. Stanford, C A : CSLI,
29-53.
324
Kimiko Nakanishi
Krifka, M. (1998): The origin of telicity. In: S. Rothstein: Events and Grammar.
Dordrecht: Kluwer, 197-235.
Krifka, M. (2001): The mereological approach to aspectual composition. Handout for
Perspectives
on Aspect. University of Utrecht.
Landman, F. (1989a): Groups I. Linguistics and Philosophy 12, 559-605.

Landman, F. (1989b): Groups II. Linguistics and Philosophy 12, 723-744.
Landman, F. (1996): Plurality. In: S. Lappin: Handbook of Contemporary Semantics.
Oxford: Blackwell, 425-457.
Landman, F. (2000): Events and Plurality. Dordrecht: Kluwer.
Lasersohn, P. (1995): Plurality, Conjunction and Events. Dordrecht: Kluwer.
Link, G. (1983): The logical analysis of plural and mass terms: A lattice theoretic approach. In: R. Bauerle, C. Schwarze and A. von Stechow: Meaning, Use and Interpretation of Language. Berlin: de Gruyer, 302-323.
Link, G. (1984): Hydras: On the logic of relative clause constructions with multiple
heads. In: F. Landman and F. Veltman: Varieties of Formal Semantics (GRASS
vol. 3). Dordrecht: Foris, 245-257.
Nakanishi, K. (2003): The semantics of measure phrases. The Proceedings of the 33rd
Conference of the North East Linguistic Society (NELS 33), 225-244.
Nakanishi, K. (2004): Domains of Measurement: Formal Properties of Non-Split/Split
Quantifier Constructions. Ph.D. dissertation, University of Pennsylvania.
Nakanishi, K. (2005): Semantic properties of Split Topicalization in German. In: C.
Maienborn and A. Wollstein: Event Arguments:Foundations and Applications.
Tubingen: Niemeyer, 331-356.
Ogihara, T. (1998): The ambiguity of the -te iru form in Japanese. Journal of East Asian
Linguistics 7, 87-120.
Partee, ., A. ter Meulen and R. Wall (1990): Mathematical Methods in Linguistics.
Dordrecht: Kluwer.
Quine, W. V. O. (1960): Word and Object. Cambridge, MA: M I T Press.
Roberts, C. (1990): Modal Subordination, Anaphora, and Distributivity. New York:
Garland.
Rothstein, S. (2004): Structuring Events: A Study in the Semantics of Lexical Aspect.
Oxford: Blackwell.
Schwarzschild, R. (1991): On the Meaning of Definite Plural Noun Phrases. Ph.D.
dissertation, University of Massachusetts, Amherst.
Schwarzschild, R. (1996): Pluralities. Dordrecht: Kluwer.
Tenny, C. (1987): Grammaticalizing Aspect and Affectedness. Ph.D. dissertation, MIT.
Kluwer.
Terada, M. (1990): Incorporation and Argument Structure in Japanese. Ph.D. dissertation, University of Massachusetts, Amherst.
325
Verkuyl, H. (1972): On the Compositional Nature of the Aspects. Dordrecht: Reidel.

Verkuyl, H. (1993): A Theory of Aspectuality: The Interaction between Temporal and
Atemporal Structure. Cambridge: Cambridge University Press.
Verkuyl, H. and J. van der Does (1991): The semantics of plural noun phrases. In. ITLI,
University of Amsterdam.
Winter, Y. (2001): Flexibility Principles in Boolean Semantics: The Interpretation of
Coordination, Plurality, and Scope in Natural Language. Cambridge, MA: MIT
Press.
Alexis Dimitriadis (Utrecht)
The event structure of irreducibly

symmetric reciprocals *
1 Introduction
It is often said that reciprocals express a "symmetric" relationship between participants. This is true in the sense that if a reciprocal sentence involves just two
participants, it will (in the usual case) express a symmetric relationship between
them: each stands as both originator and receiver of some event of the type described. But if we focus on the individual events comprising a reciprocal situation, we find that there is a distinction between the following reciprocal sentences:
(1)
(a)
(b)
The boys saw each other,

The boys met (each other).
A sentence like (la) describes a plurality of events, each of which might be an

event of asymmetric seeing: the reciprocal predicate is true just if for each boy
there is some event of seeing and some event of being seen. These events may
well involve different co-participants. 1 Such a state of affairs is not possible with
events of meeting: There can be no event of John meeting Mary without Mary
meeting John at the same time. I will argue that we can in fact make a stronger
claim: That we can say, in a linguistically meaningful sense, that there can be
no event of John meeting Mary without that same event also being an event of
Mary meeting John. I will refer to events of this type as symmetric events, and to
predicates that can only be symmetrically true of their participants as irreducibly
symmetric predicates.
It is well-known that so-called "covert reciprocals" in English are irreducibly
symmetric in this sense (Gleitman et al. 1996). While example (2a) could refer
I am grateful to Tanya Reinhart, Martin Everaert, Tal Siloni, Marijana Marelj and Gyrgy Rkosi
for their comments and suggestions, and to Patrick Brandt, Kristina Riedel, Maria Pinango,
Marika Lekakou, and Tanja Milicev for language judgements. Earlier stages of this work were
presented at the Workshop on Argument Structure and Reflexivization (Utrecht, September 2002),
at the Conference on Cross-linguistic Data and Theories of Meaning (Nijmegen, May 2003 ), and
at the 6th International Conference on Greek Linguistics ( Rethymno, September 2003 ). The Bantu
data derive from joint work with Amanda Seidl ( Seidl and Dimitriadis 2003 ).
We assume the semantics of weak reciprocity for the time being.
328
Alexis Dimitriadis
to a sequence of asymmetric kisses, the covert reciprocal (2b) can only refer to
a symmetric kiss, i.e., on the lips.
(2)
(a)
(b)
John and Mary kissed each other.

John and Mary kissed.
Example (2b) describes a single event of symmetric kissing, in which John and
Mary have identical participation: each of them is both kisser and kissed, i.e.,
each is both Agent and Patient in the event.
A variety of linguistic phenomena are sensitive to irreducible symmetry. One
of them is the discontinuous reciprocai construction, found in numerous languages around the world, in which the logical subject of a reciprocal verb appears to be split between the syntactic subject and a with- phrase, henceforth
comitative argument. (Dimitriadis 2002, to appear)
(3)
O Giannis
filithike
me ti Maria
the John
kissed-Recip.Sg with the Maria
'John and Maria kissed each other'
(Greek)
In a wide range of languages, discontinuous reciprocals can be fonned only

from reciprocal verbs that are irreducibly symmetric in meaning. In addition
to providing evidence for the relevance of the notion, this construction provides
us with a glimpse into the argument structure of symmetric events. It is shown in
Dimitriadis (to appear) that discontinuous reciprocals cannot be treated as oneplace predicates: their two arguments, subject and comitative oblique, must be
treated as distinct arguments of the verb at all stages of the derivation. Evidence
from such constructions shows that irreducibly symmetric events also involve
two separate participants, and therefore cannot be treated as simple collectives
(as proposed, for example, by Carlson 1998).
However, there are well-known technical problems with assigning the same
thematic role to multiple participants in a single event, as suggested by the claim
that John and Mary are separately Agents of the event described by example
(2b). The conflicting requirements of keeping the subject and comitative participants distinct while preserving uniqueness of participants do not admit a straightforward solution. I propose to treat symmetric events as complex events that are
"specified", in the sense of Link (1998), by sub-minimal events expressing the
distinct relations of each participant to the symmetric event.
2 Symmetric predicates and symmetric events

By definition, a two-place predicate is symmetric if exchanging its two arguments always preserves truth values; so X met Y is symmetric, but X saw Y is
329
not (since X might see Y without Y seeing X). 2 Reciprocals can in general be
fonned from either type of predicate:
(4)
(a)
(b)
The boys met each other,

The boys saw each other.
If a reciprocal sentence involves just two participants, it will (in the usual case)3
express a symmetric relationship between them: each stands as both originator
and receiver of the activity described. But at the level of the individual events
comprising a reciprocal situation, there is still a distinction between the two
reciprocal sentences above. Sentence (4b) describes a plurality of events, each
of which might be an event of asymmetric seeing; the reciprocal predicate is true
just if for each participant there is some event of seeing and some event of being
seen.4 Such a state of affairs is not possible with events of meeting: There can be
no event of John meeting Mary without that same event also being an event of
Mary meeting John. I will refer to events that have this property as (irreducibly)
symmetric events, and to predicates that are only true of symmetric events as
irreducibly symmetric predicates.5 We summarize the definition as follows:
(5)
Definition. A predicate is irreducibly symmetric if (a) it expresses a binary relationship, but (b) its two arguments have necessarily identical participation in any
event described by the predicate.
At this point I want to remain vague about the notion of event alluded to above;
certain formalizations of events, the "eventualities" of Parsons (1990) among
them, do not allow the same thematic role to be assigned to two distinct participants. For the time being our concern will be with showing that irreducibly
symmetric events, as defined above, are treated as real entities by a number of
linguistic constructions. How they might be fonnalized in the context of a Parsonean theory of events will be discussed in section 7.
A predicate that is not symmetric will be called non-symmetiic. Such predicates are neutral with
respect to symmetry: some symmetric pairs may or may not exist in their extension. Lack of
symmetry must be distinguished from the property of being asymmetric, which holds for a relation
if xRy <yRx. For example, see is non-symmetric but precede is asymmetric.
The exceptions involve so-called "chained" or "asymmetric" reciprocals such as The children
followed each other into the room.
For ease of exposition, we gloss over the variety of possible reciprocal situations identified by
Langendoen ( 1978), Dalrymple et al. ( 1998), and others. We assume the semantics of weak reciprocity.
The "symmetry" of reciprocal predicates, therefore, should not be confused with the property
of irreducible symmetry. The reciprocal "X and Y saw each other" is symmetric on the X and
Y positions, since these can be exchanged without loss of truth (as a matter of fact, this is true
of almost any predicate with a conjoined subject). Nevertheless this predicate does not involve
symmetric events. To avoid confusion I will not refer to reciprocal predicates as "symmetric"
unless the underlying events are irreducibly symmetric.
330
Alexis Dimitriadis
While meet is irreducibly symmetric even when used transitively, it has long
been known that other English verbs acquire irreducibly symmetric meaning,
with a greater or lesser meaning shift, when used in a covert reciprocal (cf.
Gleitman et al. 1996, Schwarzschild 1996 for discussion). For example, talk
is not irreducibly symmetric when used transitively, as in (6a): The students are
not talking to the teacher while she's talking to them. But the covert reciprocal
(b) can only be understood symmetrically: It says only that John and Mary are
engaged in conversation (not, for example, that they are addressing each other
but not in the context of a conversation). 6
(6)
(a)
(b)
The teacher is talking to the students,

John and Mary are talking.
Some verbs can refer to either symmetric or non-symmetric events. An example,

discussed by Gleitman et al., is the transitive verb to kiss. As they put it:
(7)
"Not all kissing is reciprocal (the flag never kisses one back), and reciprocal kissing
is not always symmetrical kissing." (Gleitman et al. 1996).
In other words, the denotation of (transitive) kiss includes both symmetric and
non-symmetric kisses. This is also true of reciprocals fonned with each other,
which do not appear to change the event type under consideration. Example (8a)
is as vague as the transitive verb kiss. It might refer to one or more symmetric
kisses, or to a series of asymmetric kisses: on the hand, cheek, or top of the head.
But when used as a covert reciprocal, kiss becomes irreducibly symmetric; so
example (b) can only refer to one or more kisses with symmetric participation,
i.e., on the lips.7
(8)
(a)
(b)
John and Mary kissed each other,

John and Mary kissed.
We find the same behaviour in other languages. Many have reciprocal strategies
that can create irreducibly symmetric predicates out of (possibly) non-symmetric
base verbs, either obligatorily or optionally. Such strategies always appear to
involve a verbal affix or clitic; I am aware of no argument reciprocals that change
the event type of the verb they modify.
For verbal reciprocal strategies, there are several possibilities: First, a reciprocal strategy might always impose irreducibly symmetric semantics on its output (even if the base verb was non-symmetric). Such strategies are typically
restricted to a subset of all verbs in the language (they are "middle strategies",
6
We ignore the irrelevant, non-reciprocal reading of sentence (6b), paraphrasable as "John is talking and Mary is talking."
Example (b) could also refer to a sequence of kisses exchanged in greeting; in that case the
"kissing" refers to the entire greeting ritual, which is itself symmetric when taken as a whole.
331
ill the terminology of Faltz 1977).8 Greek, Hebrew, Hungarian and English are
in this category.
A second category includes strategies that introduce irreducibly symmetric
semantics for some, but not all of the verbs they apply to. Such a strategy may
apply to all, or almost all transitive verbs in its language, but it only imposes
irreducibly symmetric semantics to some of them. German, French, Serbian,
Lao and Swahili have reciprocals of this type.
The third possibility is that a reciprocal strategy may not be compatible with
irreducible symmetry at all; such strategies always co-exist with another strategy
that must be used with irreducibly symmetric verbs. (I am aware of no language
that only has asymmetric reciprocals, and it would be surprising if one exists). 9
Some argument reciprocals are also incompatible with irreducible symmetry;
the Serbian reciprocal jedan drugog 'each other' is one such case. For other
argument reciprocals, irreducibly symmetry is simply irrelevant. This is the case
with each other in English, which applies identically to symmetric and nonsymmetric transitive verbs. We now consider each possibility in tan.
3 Reciprocals and symmetiy

3.1 The obligatorily symmetric reciprocals
Greek, Hebrew and Hungarian have verbal reciprocals that obligatorily refer to
symmetric events; I will refer to them as obligatorily symmetric reciprocals for
short. In each case, the reciprocal fonn of the verb kiss can only refer to symmetric kisses. As already mentioned, on the other hand, argument reciprocals do
not change the event type of the verb.
(9)
(a)
(b)
O Yanis kje i Maria filithikan.

the John and the Maria kissed-Rcp
' John and Maria kissed.' (Symmetric only)
(Greek)
O Yanis kje i Maria filisan o enas ton alo.
the John and the Maria kissed the one the other
'John and Maria kissed each other.' (Symmetric or non-symmetric)
In Hungarian, the reciprocal fonn of kiss can only denote "the sexual type of
kissing where the two tongues are involved", as Rkosi (2003) puts it, while the
transitive verb can denote any kind of "intensive" kissing activity.
Faltz's classification was intended for reflexive constructions, but can be naturally extended to
reciprocals; see Dimitriadis and Everaert 2004.
The first and second category correspond to the languages identified by Reinhart and Siloni ( 2003 )
as having reflexivization and reciprocalization operations that apply in the lexicon and in the
syntax, respectively.
332
Alexis Dimitriadis
(10) (a)
(b)
ns a bty-m
meg-cskol-t-iik egyms-t.
I and the brother-lsg Prt-kiss-Past-lpl each.other-Acc
and my brother kissed each other.'
Jnos s Kati cskol-z-t-ak.
John and Kate kiss-Rcp-Past-3pl
'John and Kate were involved in a mutual sexual type of kissing.'
These reciprocalization strategies can only be applied to particular verbs (they

are "middle strategies" in the sense of Faltz (1977)); the resulting reciprocals
usually describe social interactions and other "naturally reciprocal" relationships.
It is common for some reciprocal verbs to take on idiomatic, non-compositional meanings, typically related to social interactions; these, too, are irreducibly symmetric. In such cases the base verb might not even describe a "naturally reciprocal" activity, but the reciprocal fonn will have all the typical properties of irreducibly symmetric reciprocals. The argument reciprocal in example
(11a) can describe a series of blows, simultaneous or at different times, while
sentence (b) can only describe a physical fight. Example (12b) involves a more
extreme case of non-compositionality: The verb tsakono 'to catch' in its transitive fonn is used to mean 'to catch someone in the act', but its reciprocal
fonn means 'to argue, to have a falling-out'. Similarly the verb diastavrono 'to
cross (combine, interbreed two things)' has the reciprocal fonn diastavronome
'to cross paths'. Such behavior is common cross-linguistically.
(11) (a)
(b)
(12) (a)
(b)
O Yorgos kje i Maria xtipisan o enastonalo.

the Yorgos and the Maria hit
the one the other
'Yorgos and Maria hit each other.'
(Greek)
O Yorgos kje i Maria xtipithikan.
the Yorgos and the Maria hit.Rcp
'Yorgos and Maria came to blows (with each other).'
O Nikos kje o Andonis tsakosan o enas ton allo (nakimate).
the Nick and the Anthony caught the one the other (to sleep)
'Nick and Anthony caught each other sleeping.'
O Nikos kje o Andonis tsakothikan.
the Nick and the Anthony caught.Rcp
'Nick and Anthony argued.'
We find the same meaning shift in Hungarian. Example (13a) might be true if
John and Peter were taking turns delivering blows at each other, but example (b)
denotes an activity in which "the hits cannot be seriated or even individuated in
any meaningful way" (Rkosi 2003).
(13) (a)
(b)
Jiios
John
' John
Jnos
John
' John
333
s Pter ver-t-k
egyms-t.
and Peter beat-Pst-3pl each.other-Acc
and Peter were beating each other'
s Pter ver-eked-t-ek
and Peter beat-Rcp-Pst-3pl
and Peter were fighting/wrestling'
3.2 The optionally symmetric reciprocals

The second type of reciprocals are those in which we find the irreducibly symmetric meaning with some, but not all verbs. I will refer to such strategies as optionally symmetric. They are found in German, French, Spanish, Serbian, Lao,
Swahili, Chichewa, and elsewhere. The (b) examples below either require or
strongly favor symmetric kisses, while the (a) examples, which involve argument reciprocals, do not impose a requirement for irreducibly symmetric events.
(14) (a)
(b)
(15) (a)
(b)
Jean et Marie se sont embrasss l'un l'autre.

John and Mary Rep were kissed
each other
'John and Mary kissed each other.'
(French)
Jean et Marie se sont embrasss.
John and Mary Rep were kissed
' John and Mary kissed.'
Hans und Maria haben einander gekt.
Hans und Maria haben sich gekt.
(Geman; Kemmer 1993, 112)
In other cases, the resulting reciprocal does not have an irreducibly symmetric
interpretation. In German, for example, the verbal reciprocal sich can be used
with the verb vergttern 'to idolize'. Idolizing is evidently not a naturally reciprocal activity, at least as far as Geman is concerned, and example 16 does not
have irreducibly symmetric meaning.
(16) Johann und Maria vergttern sich.
Johann and Maria idolize
Refl/Rcp
'Johann and Maria idolize each other (or: themselves).'
(Geman)
That vergttern is not irreducibly symmetric can be demonstrated by the fact

that it is incompatible with the discontinuous reciprocal construction; this is discussed in section 4.1.
It can be seen that German sich, French se, and analogous optionally symmetric strategies in other languages can function in two ways: they can behave
like the symmetricizmg reciprocals in Greek or Hebrew, or they can generate
non-symmetric reciprocals more akin to each other in English.
334
Alexis Dimitriadis
3.3 Other strategies

Besides the obligatorily and optionally symmetric strategies, there are reciprocal types that do not introduce irreducibly symmetric semantics when they apply.
Even some of these show a sensitivity to the factor of irreducible symmetry, usually by being incompatible with it. For example, the Serbian argument reciprocal
jedan drugog 'each other' cannot be applied to verbs with irreducibly symmetric
meaning; the verbal reciprocal se must be used instead.
(17) (a)
(b)
*Petari
Marko su sreli jedan drugog.
Peter and Marko Aux met each other
'Peter and Marko met each other.'
Petar i Marko su se sreli.
Peter and Marko Aux Rcp met
'Peter and Marko met.'
Similarly, Rothmayr (2004) reports that the reciprocal sich gegenseitig is (at least
m some dialects of German) incompatible with irreducibly symmetric verbs:
(18) (a)
(b)
weil die Toni und die Inni einander treffen/umarmen.

'because Tony and Inni meet/embrace each other.'
?weil die Toni und die Inni sich gegenseitig treffen/umarmen,
'because Toni and Inni meet/embrace each other.'
Conversely, sich cannot be used with verbs whose meaning excludes symmetric
situations:
(19) Die Kinder folgten einander/*sich ins Zimmer.
'The children followed each other into the room.'
German thus appears to exclusively assign the two ends of the symmetry spectrum, irreducibly symmetric and asymmetric verbs, to distinct verbal reciprocal
strategies. The middle ground, those verbs that may or may not be symmetrically true in a situation, are compatible with either fonn; and the entire range is
compatible with the argument reciprocal einander.
These effects appear to be idiosyncracies of the various strategies, since they
are language-particular; for example, einander and each other can be used with
irreducibly symmetric verbs like meet, unlike then Serbian counterpart; and in
contrast to sich, the French verbal reciprocal se can be used with asymmetric
predicates:
(20) Les enfants se sont suivi,
the children Rcp are followed
'The children followed each other.'
335
It can be seen that many reciprocal strategies are sensitive, in diverse ways, to the
parameter of irreducible symmetry or to symmetry in general. To others, such as
each other in English, it seems simply irrelevant.
4 Discontinuous reciprocals
Alongside ordinary reciprocals, many languages allow the discontinuous reciprocal construction, in which the logical subject of a reciprocal verb appears to
be split between the syntactic subject and a comitative argument.
(21) (a)
(b)
O Gianni s kj e i Maria filithikan

the John and the Maria kissed-Recip.Pl
'John and Maria kissed each other.'
O Giannis
filithike
me ti Maria
the John kissed-Recip.Sg with the Maria
'John and Maria kissed each other.'
(Greek)
In this section we summarize the analysis of discontinuous reciprocals presented

m Dimitriadis (2002, to appear), and apply it to the investigation of irreducible
symmetry.
The discontinuous reciprocal is a construction specific to certain reciprocalforming strategies; it is possible with sich in German, with se in Serbian, and
with the Greek verbal reciprocal shown in 21, but not with the "argument" reciprocals of the same languages, respectively einander,jedan drugog, and o enas
ton alo. In fact, it seems to be restricted to verbal reciprocals; of the many languages discussed in Dimitriadis (to appear) that have the discontinuous construction, none allow it with argument reciprocals. 1 "
We can add to our list of discontinuous reciprocals the covert reciprocals of
English, many of which can be used discontinuously. Once again, the argument
reciprocal each other cannot be used discontinuously.
(22) (a) John met/argued/talked/collided with Mary,
(b) *John met each other with Mary.
Following Reinhart and Siloni (2003), I consider covert reciprocals to be derived from transitive verbs through a morphologically null argument structure
operation. 11 English should be grouped with the "obligatorily symmetric" lan10
11
For evidence that se and sich are verbal reciprocals, see Zee ( 1985), Reinhart and Siloni (forthcoming), and the discussion in Dimitriadis (to appear).
English covert reciprocals have been recognized as reciprocals since the early days of the generative literature, when the question of whether they can be transformationally related to each-other
reciprocals was debated at some length. (Gleitman 1965, Fiengo and Lasnik 1973, Dougherty
1974, Langendoen 1978).
336
Alexis Dimitriadis
guages like Greek and Hungarian, since covert reciprocals must be irreducibly
symmetric. But because covert reciprocals are not morphologically marked, it
is impossible to know when reciprocalization has applied and when we have an
imderived verb with sufficiently similar semantics. For this reason the English
facts must be approached with caution, and are not used as grounds for any conclusions in this work.
It is common to analyze discontinuous reciprocals by reducing them to the
corresponding "simple reciprocal" sentences, either by deriving the former from
the latter via syntactic movement or at the level of interpretation (Vitale 1981,
Mchombo and Ngimga 1994, Siloni 2001). However, it can be shown that the
semantics of discontinuous reciprocals is more specific, that is, more expressive,
than the semantics of the corresponding simple reciprocals (Dimitriadis to appear). To see this, we must consider discontinuous examples in which either the
syntactic subject or the comitative argument is plural.
(23) (a)
(b)
O Yanis, o Nikos kje i Maria tsakothikan

the John the Nick and the Maria argued.Rcp
'John, Nick and Maria argued.'
O Yanis kje o Nikos tsakothikan me ti Maria
the John and the Nick argued.Rcp with the Maria
'John and Nick argued with Maria.'
(Greek)
Example (23 a) describes a situation of conflict between the three members of

the subject, with no specification of which party or parties were in conflict with
whom. But (23b) is either about an argument between John and Nick on the
one part and Maria on the other, or possibly about two different arguments between Maria and each of the two men. In each case, the reciprocal relation must
involve pahs consisting of one participant (possibly plural) from the syntactic
subject, and one participant from the comitative argument. The simple reciprocal sentence (a) could also have been used to describe this situation, but it could
not refer only to these possibilities; the meaning of (b) is therefore more specific than that of (a), and is not semantically reducible to it. More generally: The
meaning of the discontinuous reciprocal is not reducible to the meaning of the
corresponding simple reciprocal. To express the meaning of (b) it is necessary to
treat the two positions, subject and comitative, as distinct arguments of the verb
at all stages of the derivation. In other words, discontinuous reciprocals must be
analyzed as two-place predicates.
We have demonstrated that subject and comitative must be distinguishable in
the semantics. It remains to rule out the possibility that such constructions involve (at some appropriate syntactic or semantic levels) a structured entity that
is subdivisible hito the appropriate subparts: the subject of (23b) might be the
"group" Y a m s N i k o s > M a r i a > , which can be subdivided into the appropriate top-level subgroups, <YanisffiNikos> and <Maria>. However, the
work of Schwarzschild (1996) shows that such an analysis is not sustainable.
337
The splitting of conjoined NPs into the parts of the conjunction is a discourse
effect that can be overridden by providing an explicit criterion for grouping. For
example, the most obvious interpretation of example (24a) is that the animals
were separated into two groups, one consisting of the cows and the other consisting of the pigs; but if we add the pirrase "according to color", example (24b)
states that the animals were separated by color, regardless of species.
(24) (a)
(b)
(c)
The cows and the pigs were separated (from each other).
The cows and the pigs were separated according to color.
The animals were separated according to color.
Therefore, Schwarzschild argues, if the cows and the pigs are the only animals
then the subject of (24a) should be analyzed identically to that of (24c) both
must be treated as plural individuals consisting of a number of atoms, with no
intermediate structure.
On the other hand, the division of the discontinuous (25a) into subject and
comitative cannot be overridden in this way, as the ungrammaticality of (25b)
shows.
(25) (a)
(b)
The cows were separated from the pigs.

# The cows were separated from the pigs according to color.12
The pairing structure we find in discontinuous covert reciprocals, in other words,

is imposed by syntactic structure rather than by discourse effects, and cannot
be overridden by manipulating the context. This means that we cannot extend
Schwarzschild's analysis to discontinuous reciprocals: the tests on which his
argumentation rests will fail for discontinuous reciprocals.
The same applies to languages with overt discontinuous reciprocals: The relation being described must hold between parts of the syntactic subject and parts
of the comitative oblique. Let us consider the relevance of the covers analysis
with a new example: Greek sentence 26 can only describe hugs between a boy
and a girl, not hugs between boys or between girls.
(26) Ta agorja angaljastikan me ta koritsja.
the boys hugged-Rep with the girls
= Each boy shared hugs with some (all?) girls.
To see that manipulation of the context cannot override this reading, consider a
scenario in which a group of students has gone to a competition where participants compete in teams of two. Assmne for now that some teams consist of a boy
12
Many speakers find this sentence acceptable if it so happens that all cows were one color and all
pigs were another, so that the two species were separated from each other as a result of separating
by color. In this case the division required by the syntactic structure (separation according to
species) is respected by the explicitly stated criterion, color.
338
Alexis Dimitriadis
and a girl, while others consist of two boys or two girls. At the end of the competition, the entire group is praised for having done well, and each contestant
hugs his or her teammate. We could then say (27a), but not (27b).
(27) (a)
(b)
Ta agoija kje ta koritsja angaljastikan (o kathenas me to teri

tu).
the boys and the girls
hugged-Rep the each
with the partner his
'The boys and the girls hugged (each with their partner).'
Ta agoija angaljastikan me ta koritsja (o kathenas me to teri
tu),
the boys hugged-Rcp with the girls
the each
with the partner Iiis
'The boys shared hugs with the girls (each with their partner).'
Sentence (27a) says simply that each boy or girl hugged his or her teammate; instead of an unstructured assortment of hugs, the context tells us that each person
hugged just one other, appropriate person. But sentence (27b) cannot be used
felicitously. In this context it is only acceptable if, contrary to our earlier assumption, each team consisted of one boy and one girl: then it would be possible
to simultaneously respect syntactic structure and the requirements of the context, and the sentence would be acceptable. (Compare example (25b) above).
Thus the division into subject and comitative oblique cannot be overridden by
the context.
Our example shows that manipulation of the context can affect the interpretation of our sentence, but only if it respects the distinctness of the two reciprocal
positions. This is exactly what we expect if we adopt Schwarzschild's system but
consider the subject and the comitative oblique to be two separate arguments. 13
4.1 The role of symmetry
In a great number of languages, irreducible symmetry plays a prominent role in
the distribution of discontinuous reciprocals. In particular, it is shown in Dimitriadis (to appear) that the discontinuous construction can only be used with
reciprocal verbs that are irreducibly symmetric in meaning. In Serbian, for example, the reciprocal fonn of kiss can be used discontinuously, with irreducibly
symmetric semantics, while the reciprocal of hear cannot; but the latter verb can
be used discontinuously with the symmetric, lexicalized meaning to talk to each
other. Other verbs that allow the reciprocal se but cannot be used discontinuously
are help, praise, etc. 14
(28) (a)
13
14
Jovan i Marija
se ljube.
John and Mary.Nom Rep kiss
'John and Mary kissed.'
A Schwarzschild-style analysis of sentence (27b) will involve a paired cover, which Schwarzschild defines precisely to account for dependencies between the arguments of two-place predicates. See Dimitriadis (to appear) for more details.
Note that it is the symmetry of the derived (reciprocal ) form that matters, not of the basic transitive
verb. Neither kiss nor hear are symmetric in their transitive form.
(b)
(29) (a)
(b)
339
Jovan
se ljubi sa Marijom.
Jovan.Nom Rep kisses with Marija.Inst
'John and Mary kiss.'
Jovan i Marija
se cuju.
Jovan and Marija.Nom Rep hear.3Pl
'John and Mary hear each other.'
* Jovan se cuje sa Marijom
Jovan Rep hears with Marija.Inst
(Ok with secondary meaning: 'John and Maria talk (to each other).')
Similarly, most verbs in German can form a sich reciprocal; but while sich
schlagen 'to fight' and sich Missen 'to kiss' can be used discontinuously, sich
vergttern 'to idolize each other' cannot.
(30) (a)
(b)
(31) (a)
(b)
(32) (a)
(b)
(33) (a)
(b)
Johann und Maria schlugen sich.

Johann and Maria hit
Rcp/Refl
'Johann and Maria hit each other/themselves.'
Johann schlug sich
mit Maria
Johann hit
Rcp/*Refl with Maria
'Johann and Maria hit each other/*themselves.'15
Hans versteht sich mit Maria.
'Hans and Maria understand each other'
Hans vertrgt sich mit Maria.
'Hans and Maria get along'
Johann und Maria vergttern sich.
Johann and Maria idolize
Refl/Rcp
'Johann and Maria idolize themselves/each other.'
* Johann vergttert sich mit Maria.
*Hans mag sich mit Maria.
'Hans and Maria like each other'
*Hans hat sich mit Maria.
'Hans and Maria hate each other'
In the "obligatorily symmetric" languages, the required reciprocal construction

can itself only be used if the result is irreducibly symmetric. Greek and Hungarian are in this category. In such languages the generalization is that if a verb can
be reciprocalized, it can also be used discontinuously. For example, the Greek
verbs eklego 'elect', proslavmano 'hire', and didasko 'teach' cannot f o n n this
type of verbal reciprocal at all; but sinando 'meet', sproxno 'push' and tilefonao
'telephone ' all have irreducibly symmetric verbal reciprocals, and all can be used
discontinuously.
15
This sentence also has an irrelevant instrumental reading, which says that Johann used Maria as a
club to hit himself.
340
Alexis Dimitriadis
(34) (a)
(b)
(35) (a)
(b)
O Nikos kje o Andonis tsakothikan.

the Nick and the Anthony caught.Rcp
'Nick and Anthony argued.'
O Nikos tsakothike me tonAndoni.
the Nick caught.Rcp with the Anthony
'Nick got in an argument with Anthony.'
Jnos s Kati cskol-z-t-ak.
John and Kate kiss-Rcp-Past-3pl
' John and Kate were kissing.'
Jnos cskol-z-ott Kati-val.
John kiss-Rcp-Past Kate-with
' John and Kate were kissing.'
(Greek)
(Hungarian)
English, which is in principle in this category, presents a problem: Some covert

reciprocals do not allow the discontinuous construction as expected. For example, John kissed/married with is not very good. But since there is no visible
exponent of a reciprocalization operation, it is not clear what we should make of
this observation.
In both types of languages considered here, the discontinuous construction
is restricted to predicates that are irreducibly symmetric. But it should be mentioned here that this correlation does not hold universally. The Bantu languages
Swahili, Chichewa and Ciyao allow the discontinuous reciprocal construction,
but irreducible symmetry is not required. The following example is a classic
example of a "chained reciprocal", in which the relationship holding between
participants is asymmetric.
(36) Ugonjwa hu-fuat-ana
na upotevu wa maisha.
sickness SM-follow-Rcp with waste of life
'Sickness follows from a life of profligacy.'
(Swahili; Johnson et al. 1939)
5 Counting symmetric events

We have seen that symmetric events are responsible for licensing the discontinuous construction, and that in many languages the result of a reciprocalization operation must be irreducibly symmetric. But it is reasonable to wonder if
irreducibly symmetric predicates might not simply describe pairs of ordinary,
"asymmetric" events. This would simplify the task of analyzing such predicates,
and in fact seems necessary at some level of formalization, but it does not match
the way we talk about events of this sort. As Siloni (2002) points out, symmetric
verbal reciprocals do not show the counting ambiguities that characterize their
argument reciprocal counterparts. In sentence (37a), the count "five times" can
be understood as counting either the total number of kicks or the kicks delivered
341
by each of John and Mary. But sentence (37b) can only be about five kicking
occasions (each involving an indeterminate, and irrelevant, number of kicks).
(37) (a)
(b)
(38) (a)
(b)
(39) (a)
(b)
O Yanis kje i Maria klotsisan o enas ton alo pende fores.

the John and the Mary kicked the one the other five times
(i)
John and Mary kicked each other; there were a total of five kicks, all
together.
(ii)
John kicked Mary five times; Mary kicked John five times. There were
a total of ten kicks.
O Yanis kje i Maria klotsithikan pende fores,
the John and the Mary kicked.Rcp five times
(i)
John and Mary kicked each other. There were a total of five kicks, or
five kicking matches, all together.
Dan ve-Ron nisku exad et ha-seni xames pe'amim.
Dan and-Ron kissed each Acc the-other five times
(i)
There were five mutual kissing events.
(ii)
There were ten kissing events: five by Dan and five by Ron.
Dan ve-Ron hitnasku xames pe'amim.
Dan and-Ron kissed five times
(i)
There were five mutual kissing events. (Symmetric only)
(Hebrew; Siloni 2002)
John and Mary kissed each other five times.
(i)
There were five kissing events.
(ii)
There were ten kissing events: five by John and five by Mary,
John and Mary kissed five times.
(i)
There were five mutual kissing events. (Symmetric only)
The source of this contrast is not the difference between verbal and argument
reciprocals per se, but the difference between irreducibly symmetric and nonsymmetric predicates: When we count asymmetric events, we can choose between counting the total number of events or counting the number of events
attributable to each participant; but when we count symmetric kisses (or symmetric altercations involving kicking), we can count them only once: the symmetric kiss given by Dan to Ron cannot be counted as distinct from a symmetric
kiss given at the same moment by Ron to Dan. In other words, symmetric events
are atomic as far as this test is concerned.
To see that argument reciprocals are not in themselves the reason for the ambiguous readings, it is enough to consider examples with an irreducibly symmetric base verb:
(40) (a)
(b)
John
(i)
(ii)
John
(i)
and Mary met each other five times.

There was a total of five meetings.
*There was a total of ten meetings,
and Mary met five times.
There was a total of five meetings.
342
Alexis Dimitriadis
The contrast we found in example 39 has disappeared. Sentence (40a) lacks the
ambiguity, even though it uses the reciprocal each other, which readily gives rise
to scope-like ambiguities elsewhere.
In languages whose verbal reciprocals are not obligatorily symmetric, we predict that non-symmetric verbal reciprocals will be ambiguous, like argument
reciprocals. This is indeed the case in Geman and Serbian, as the following examples show. The non-symmetric verbal reciprocals in the (b) sentences pattern
just like the non-symmetric argument reciprocals in the (a) sentences.
(41) (a)
(b)
(42) (a)
(b)
Johann und Maria traten einander fnfmal vors

Schienbein
Johann and Maria kicked each.other five times against.the shinbone
(i)
John and Mary kicked each other. There were a total of five kicks.
(ii)
(Geman)
Johann und Maria traten sich
fnfmal vors
Schienbein
Johann and Maria kicked each.other five times against.the shinbone
(i)
John and Mary kicked each other. There were a total of five kicks.
(ii)
Petar i Marko su se udarili pet puta.
Peter and Marko Aux Rep kick five times
'Peter and Marko kicked each other five times.'
(Serbian)
(i)
?Peter and Marko kicked each other. There were a total of five kicks.
(ii)
Peter kicked Marko five times; Marko kicked Peter five times. There
were a total of ten kicks.
Petar i Marko su udarili jedan drugog pet puta.
Peter and Marko Aux kick each other five times
(i)
(ii)
*Peter and Marko kicked each other. There were a total of five kicks.16
Peter kicked Marko five times; Marko kicked Peter five times. There
were a total of ten kicks.
Verbs like meet, which are irreducibly symmetric regardless of the reciprocal's
semantic contribution, behave just like in the obligatorily symmetric languages:
the ambiguity disappears. In the following examples, the ten-event reading is
ruled out for the argument reciprocal and the verbal reciprocal alike; the presence
of irreducible symmetry blocks it, regardless of the fonn of the reciprocal. 17
16
17
While there is some variation and noise in the judgements, the status of the crucial ten-event
readings was clear: My Serbian consultant found that ten kicks were perfectly acceptable with
either reciprocal, and ten meetings were clearly impossible.
The argument reciprocal is incompatible with irreducibly symmetric verbs in Serbian, hence example (44a) is ungrammatical.
343
(43) Johann und Maria trafen einander/sich fnf mal.

Johann and Maria met each.other five times
(a)
(b)
(44) (a)
(b)
There were a total of five meetings.

*There were a total of ten meetings.
*Petari
Marko su sreli jedan drugog pet puta.
Peter and Marko Aux met each other five times
Petar i Marko su se sreli pet puta.
Peter and Marko Aux Rep met five times
(i)
There were a total of five meetings.
(ii)
*There were a total of ten meetings.
Siloni (2002) gives a scopai account of the two readings of (38a), following
Heim et al.'s (1991) analysis of sentences like John and Maty won $100. Siloni
argues that the reciprocals we have identified as irreducibly symmetric are fonned
in the (computational) lexicon; syntactic reciprocals can undergo QR and give
rise to ambiguities of this sort, but lexicon reciprocals cannot. Siloni's analysis
makes substantially the same predictions as the account presented above where
verbal reciprocals are concerned, but the two accounts diverge when we consider argument reciprocals: Only a symmetry-based analysis can explain why
irreducibly symmetric base verbs like meet never give rise to ambiguous counts,
even with argument reciprocals (which are necessarily fonned in the syntax). A
scopai account would predict that argument reciprocals should always give rise
to the ambiguity.
The crucial factor, then, is not the type of reciprocal but whether the events
described are symmetric. A sentence about non-symmetric events is ambiguous
because it can be taken to count the actions of each participant or the total number of actions; but symmetric events cannot be counted twice (once for each
participant), and so the ten-event reading is not possible. No such effect would
be expected if an event of meeting, or a symmetric kiss, in fact consisted of two
asymmetric events. This proves what we set out to show in this section: that
"symmetric events" truly behave as a single, symmetric event, rather than as a
pair of simultaneous events that entail each other.
5.1 Scope-like ambiguities
The issue of individuating symmetric events has also been addressed, with generally similar results, by Carlson (1998). In this section I summarize some of his
findings before reconsidering some of them in light of additional evidence. Carlson concludes, as we have done, that a symmetric covert reciprocal describes
only a single event. For example, sentence (45) describes only a single event of
meeting.
(45) John and Bill met in Cleveland.
344
Alexis Dimitriadis
Carlson's argumentation is based on an asymmetry between sentences of the

following sort:
(46) (a)
(b)
Bill and Mary (each) thought that they had kissed each other.
Bill and Mary (each) thought that they had kissed.
Such constructions, in which the subject of the reciprocal verb is a pronomi

bound by a higher antecedent, were examined in the well-known study of reciprocals by Heim, Lasnik, and May (1991). Sentence (46a) can have two readings,
known as the "we" and the "I" readings. According to the "we" reading, Bill
thought "we have kissed each other", and Mary thought the same thing. In the
"I" reading, Bill thought "I have kissed Mary", and Mary thought "I have kissed
Bill". 18 In the "we" reading, the pronomi serving as subject of the embedded
clause is coreferent with the plural individual John and Maty. In the "I" reading,
the pronomi is bound by a distributive operator ranging over the members of the
matrix subject. 19
The ambiguity found with (46a) disappears when we substitute a covert reciprocal. Sentence (46b) only has the "we" reading. Carlson argues that such
symmetric verbs do not distribute over their subject but have only a group reading, like collective predicates such as gather. Consequently they describe only a
single event. Example 45, for example, describes a single event of meeting.
Carlson's goal was to show that verbs, by themselves, denote only singular
events. Reference to multiple events can be introduced by an external quantifier,
such as each other. This is clearly the case in examples like (47a), which involves
a multitude of kicking events; but although Carlson assumed that multiple events
are also involved when an irreducibly symmetric verb is involved, as in (47b),
there is some evidence to the contrary.
(47) (a)
(b)
John and Bill kicked each other.

John and Bill met each other in Cleveland.
The event-counting test from the previous section does not distinguish between
meet and meet each other, neither of the following sentences is compatible with
a situation in which there were two meetings.
18
Carlson also mentions a so-called "you" reading for sentence (46a), but this may be inaccurate;
Heim et al. report the "you" reading only for sentences like ( i ), in which the first reciprocal serves
as the antecedent of the embedded pronoun. According to this reading John told Mary that she
should leave, and vice versa. The "you" reading does not appear to be available for example (46a).
19
It is immaterial whether the ambiguity is viewed as due to different scope possibilities for the
reciprocal, as Heim et al. propose, or as the result of different possible antecedents for the dependent pronoun, as Williams ( 1991 ) argues. (The latter option is also defended in Dimitriadis ( 1999,
(i) John and Mary told each other that they should leave.
2000)).
(48) (a)
(b)
345
John and Bill met once.

John and Bill met each other once.
Each of the above sentences only allows one meeting, while a non-symmetric
verb in the same construction is ambiguous. Example 49 most likely means that
there were two visits, one by each participant.
(49) John and Bill visited each other once.
This is not to say that the two sentences in 48 necessarily have the same logical
fonn; it is quite plausible that each other introduces the possibility of reference
to multiple events, as Carlson assmnes; but the symmetric semantics of meet
force the identity of the referred-to events, so that only a single event is involved
after all. The result is that the different potential readings of (48b), if they may
be called that, are indistinguishable.
We find additional support for this conclusion if we substitute a symmetric
verb in place of kiss in example 46. Recall that kiss is not symmetric as a transitive, but becomes so when used as a covert reciprocal. This is why sentence
(46a) is ambiguous but sentence (b) is not. With meet, no ambiguity is possible
with either sentence: If John believes that he met Bill, he must believe that he
and John met. 2 "
(50) (a)
(b)
John and Bill believed that they had met each other,
John and Bill believed that they had met.
We conclude that while covert reciprocals only refer to a single event, as Carlson
argues, argument reciprocals need not always refer to multiple events. This does
not affect any other aspects of Carlson's analysis; in particular, it is consistent
with his position that every verb must introduce reference to just one event. 21
20
Carlson states that the symmetric collide, many and exchange glances give rise to ambiguities
like kiss: but the ambiguity only seems to arise if we construe these verbs as non-symmetric (e.g.,
if we take collide to mean "crash into" ). Otherwise there is no truth-value difference between the
"I" and "we" construals. For example, the "I" reading of (i), Carlson's (25a), is given in (i.a); it
is equivalent to the "we" reading in (i.b), unless furtively in (i.a) can be taken to describe Beth's
manner only, not Sue's. But it is not at all clear that this is the case.
(i) Beth and Sue believed that they had exchanged glances with each other furtively.
(a)
(b)
21
Beth believed that she had exchanged glances with Sue furtively.
Beth believed that she and Sue had exchanged glances furtively.
The one-event condition might have to be relaxed to mean "one event for any combination of
participants." Sentences ( ii ) and ( iii ) are potential counterexamples to the stronger claim:
(i) The committee members hugged.
( ii ) The committee members kissed.
(iii)Beth, Sue and Jake exchanged glances.
While sentence (i) does suggest a group hug, as Carlson predicts, the other two seem to involve
multiple ( symmetric ) events in their interpretation. This seems to be a type of accommodation
346
Alexis Dimitriadis
6 The thematic roles of symmetric participants

Each participant to an ordinary event fulfils a different role: a kiss involves the
kisser or Agent and the kissed or Patient. But as we have seen, a symmetric
kiss must be described as a single event, in which the participants are identically
involved. Example (51) refers to a single event of kissing, each of whose participants was both kissing and being kissed. The direct object of kiss can easily be
any sort of inanimate object, but the comitative used here must be animate.
(51) O Yanis filithike me ti Maria,
the John kissed.Rcp with the Maria
'John kissed with Maria.'
While we have defined irreducibly symmetric predicates as those whose two

arguments must have necessarily identical participation, the two arguments of
discontinuous reciprocals are not identical in all respects. When there is considerable difference in the stafiis of the participants, for example, it is often possible
to use a symmetric predicate discontinuously where its simple fonn would be
odd.
(52) (a)
(b)
(53) (a)
(b)
The car collided with the tree,

# The car and the tree collided.
The bicycle is near the garage.
# The bicycle and the garage are near each other.
But this does not mean that the two arguments are thematically different. As
Gleitman et al. (1996) show, there are measurable differences between the two
arguments of even logically symmetric predicates like be equal to, due to the
different syntactic prominence of the arguments and to discourse structure effects. (cf. also Dowty 1991, Carlson 1998). Gleitman et al. suggest that symmetrical comparisons, like ordinary predicates, have a Figure-Ground structure;
whichever participant appears on nonsubject position becomes the Ground. In
similarity comparisons, the subject is understood to have some property that is
characteristic of the Ground; therefore example (54 a) might be understood to
say that Clima is isolationist like North Korea, while example (b) might be saying that North Korea shares some salient property of Clima. Gleitman et al. show
that if we explicitly include the standard of comparison, as hi (55), the difference
between the two versions disappears.
triggered to rescue the sentences, which would otherwise be weird: only two people can participate in a single exchange of glances, and it is difficult to conceive of (for example) a five-way
(54) (a)
(b)
(c)
(55) (a)
(b)
347
China is similar to North Korea.

North Korea is similar to China.
North Korea and China are similar.
North Korea is similar to China in size,
North Korea and China are similar in size.
Such contrasts are clearly non-thematic, and we can safely attribute them to
structural differences between the two argument positions.
The discontinuous construction is doubtless useful as a way to assign unequal
discourse status to the participants in a single symmetric event. The construction
also provides the opportunity to use modifiers that target the subject only (such
phenomena provide additional evidence that the two positions are distinct arguments; cf. Dimitriadis to appear).
(56) Peter kte
sich geme mit Maria.
Peter kissed.Sg Rep gladly with Maria
'Peter liked to get kissing with Maria.'
(Geman; Behrens et al. 2003)
There is also some evidence that the two positions, subject and comitative oblique, differ subtly in the degree of agency they require. Note that it is odd to say
(57a) if John forced the kiss on Mary. It is also odd to say (57b) in a situation
where John walks up to a statue, embraces it, and plants a kiss on its lips: it
seems that the subject position requires intentional participation in the act being
described.
(57) (a)
(b)
#John and Mary kissed (although Mary resisted),

#John and the statue kissed.
While the English verb kiss cannot be used discontinuously, its Greek equivalent
can. The non-discontinuous (58a) is odd, just like its English counterpart, but
many Greek speakers find the discontinuous (58b) to be acceptable.
(58) (a)
(b)
# 0 Nikos kje to aghalma filithikan.

the Nick and the staUie kissed.Rcp
'Nick and the staUie kissed.'
O Nikos
filithike
me to aghalma.
The Nick kissed.Rcp.Sg with the staUie
'Nick engaged in a mutual kiss with the statue.'
It seems that Nick should be acting as if the statue is also participating in the kiss.
This is a subtle effect that does not seem to hold universally in other languages.
My consultants reported the Hebrew and Serbian equivalents of (58b) to be illformed; Rkosi (2004) reports that while he initially disliked the same example
in Hungarian, he later came to consider (59b) well-formed.
348
Alexis Dimitriadis
(59) (a)
(b)
#Jnos
s a szobor
cskol-z-t-ak.
John.Nom and the statue.Nom kiss-Rcp-Pst-3pl
' John and the statue kissed.'
Jnos
rszegen cskol-zo-tt a szobor-ral.
Jolin.Nom drunk
kiss-Rcp-Pst the statue-with
'John kissed with the statue while drunk.'
(Hungarian)
There may also be clearer cases. Behrens et al. (2003) report that in Tetim Dili
(East Timor), "in cases where one of the participants is presented as the instigator, the subject refers to the instigator [...] and the secondary participants are
introduced by ho 'with'." (Cited from Williams-van Klinken et al. 2002,60-61 ).
(60) (a)
(b)
Joo ho
Maria istori malli.
John and/with Maria quarrel Rcp
' John and Maria quarreled (no indication as to who started it).'
Joo istori malli ho
Maria.
John quarrel Rcp and/with Maria
' John quarreled with Maria (he started it).'
In each case, it seems that intention or "instigation" is distinguished from participation in the act itself; the subject position attributes both instigation and
participation to the subject, while the comitative position only attributes participation.
While the topic clearly merits further investigation, I will assume here that the
two positions are thematically identical, in the sense of having the same thematic
relationship with the lexical verb; I will assume that additional requirements on
the subject, such as differences in instigation or degree of participation required,
are associated with its syntactic position (for example, we might treat them as
contributed by some functional head rather than by the verb root).
7 Formalizing symmetric events

We now come to the question of how to fonnalize our notion of symmetric events
in the context of a general event semantics. In particular, I assume the common
"neo-Davidsonean" system of the type proposed by Parsons ( 1990), which relies
crucially on the notion of thematic roles in addition to the notion of an event
variable. For the purposes of this discussion I will take it for granted that the
participants in a symmetric event have thematically identical relations to the
event, and that any differences are non-thematic in natine as already discussed.
In pursuing an analysis we can identify two distinct but interrelated questions:
First, what is the thematic role assigned to the subject of a symmetric reciprocal;
and second, what is the relationship between the role assigned to the subject and
the role assigned to the comitative of a discontinuous reciprocal.
349
The challenge is how to formalize the idea that there are two participants
with thematically identical participation, without running afoul of the problems
inherent in assigning the same thematic role to multiple participants in a single
event. We take as our starting point the analysis of discontinuous reciprocals articulated by Siloni (2 0 02). 22 Following an analogous analysis of reflexive verbs
(Reinhart and Siloni 2003), Siloni argues that lexical reciprocalization works by
bundling the two theta roles of the underlying transitive predicate into a single
complex theta role, e.g., [Agent-Theme].
(61) Bundling of roles by 0-unification: V[Agent],[Theme] * V[Agent-Theme].
In the case of reflexives, the application of this is straightforward; Reinhart and

Siloni interpret a bundled role as the "distributive conjunction" of the two roles,
i.e., assigning a bundled role is just like assigning the two component roles to
the same participant. So if John is an Agent-Theme of e, then John is an Agent
and a Theme of e. But this is not appropriate for reciprocals. For example, assigning the bundled role to the participants of kiss would entail that John kissed
himself. Siloni proposes that bundled reciprocal roles are assigned to pairs of
participants, who can then be assigned to the component roles in either order
(reflecting the symmetry of the relationship).
This allows Siloni to formulate the correct semantics for discontinuous reciprocals, by requiring each pair to contain an element from the subject and an
element from the comitative:
(62) V i SVy
(3e (kiss(e) & [Agent-Theme](e, { i , y})))
However, this approach does not escape the problem of imdesired entailments.
Suppose that the bundled role is interpreted distributively, with each component
assigned arbitrarily to one of the pair's elements. In that case, we can expand an
Agent-Theme relation as in (a) and eliminate some conjuncts, getting (b). This
gives us the imdesired entailment that, for example, John kissed himself.
22
An alternative approach, adopted by Rkosi ( 2003 ), is to give the second argument of the discontinuous reciprocal the special role Partner. The asymmetries in initiative and participation between
the two arguments lead Rkosi to reject the proposal that the two positions are thematically identical; he treats the subject as a simplex Agent, not a combined Agent-Theme. The Partner is a
Theme-like argument that is intended not as a semantic role relation, but as a syntactic label for
an underspecified patient-like role, which Rkosi describes as somewhat similar to Experiencer.
While this solution is consistent with Rkosi's assumptions, it does not account for the symmetric entailments that do arise in such constructions; thus it does not help us with our goal of
formalizing the notion of participants with thematically identical involvement.
350
Alexis Dimitriadis
(63) (a)
(b)
[Agent-Theme] (e, {J, Ai}) s

(Agent (e, J ) & Theme (e, M)) & (Agent (e, M) & Theme (e, J)) -*
( Agent (e, J ) & Theme (e, J))
It would not help to simply add a non-identity condition (to the effect that the
Agent cannot be equal to the Theme) to the translation of the reciprocal; this
would not block the derivation of (63b), but would lead to a logical contradiction
in combination with it. (This means that our reciprocal formula would also be
logically inconsistent, since it entails a contradiction).
I can only see two ways to rescue the analysis: either the Agent-Theme role
must remain unanalyzed, or the event variable must be decomposed into smaller
events. We pursue the second approach here.
We have seen that there are, in some linguistically real sense, symmetric
"events" of kissing, colliding, arguing, etc. Such events are treated as atomic
by tests such as counting and scope-like ambiguities, and can be shown to have
two distinguishable event positions. But our toolbox of neo-Davidsonean events
cannot express multiple assignments of the same thematic relation. To do so, we
must resort to a level of representation below the level to which our linguistic
tests have access. We do so by adapting the relation of specification defined by
Link (1998).
(64) Event specification (Link 1998, 251-261).
(a)
(b)
The set of eventuality variables E forms a complete atomic lattice, intrinsically ordered by <.
There is a 2-place relation S on the set E of atomic events.
e Se' means "e specifies e .
Specification is meant to express relationships between eventuality variables representing the same real-world event. We use it to model a symmetric event as an
eventuality that is specified by two eventualities of the same type, with permuted
roles for its participants. We stipulate that events specifying some superior eventuality are obscured by it; when we count events, we only count eventualities that
are (possibly) specified but do not themselves specify some "larger" eventuality.
We can then define the expansion of Siloni's [Agent-Theme] bundle as follows:
(65) [Agent-Theme] (e, A) =
Vi A 3y ({, y} = A h 3e' {e'Se & Agent(e', ) & Patient(e', y)))
If A is a two-element set as in 62, this formula introduces a separate eventuality

variable for each element, with both eventualities specifying e. The two eventualities represent, for example, John's kissing Mary and Mary's kissing John.
They are components of a composite, symmetric event which they both specify,
but the simplex roles are only defined at the lower level. Since the two Agent
351
roles modify different variables, the problem of incorrect entailments does not
arise.
This brief sketch has not addressed the question of which sentence modifiers
can make reference to which eventuality variables. Various agent-oriented modifiers target only the syntactic subject when applied to a discontinuous reciprocal
(cf. example 56), and in such cases the association should be preserved in our
semantics.
8 Conclusions
All reciprocals describe situations that are in some maimer reciprocated between
participants; but irreducibly symmetric predicates fonn a distinct class within
them, and a variety of syntactic constructions are sensitive to the distinction.
But formalizing the notion poses challenges, especially since the properties of
discontinuous reciprocals require us to treat symmetric reciprocals as two-place
predicates.
The analysis outlined in the previous section imposes a certain burden of
complexity; it requires a layered event structure of sub-events specifying our
symmetric event, and an elaboration of argument-passing mechanisms to allow
a two-element set to be assigned to the bundled role. But it can be seen that
the conflicting requirements of uniqueness of thematic roles and symmetry of
the predicate require something along these lines, if the symmetry of the event
participants is to be expressed. However, Carlson (1998) argues that differences
between the argument and comitative positions, even if they are not per se thematic, are sufficient to differentiate the two roles; and therefore that in all cases
uniqueness of roles is preserved.
For truly symmetric predicates, Carlson provides an ingenious analysis: they
are one-argument predicates that are interpreted collectively; so John and Mai-y
met is true of the group John and . For all its merits, Carlson's analysis
unfortunately cannot account for the interpretation of certain discontinuous constructions. Consider again example (23b), repeated below. This could describe
either two separate conflicts between Maria and one of the men, or a single
conflict in which John and Nick, together, are in conflict with Maria. This last
reading cannot be expressed if we treat argue as a one-place collective predicate,
since no subgroups could be retained among the parts of its argument.
(23) (b)
23
O Yanis kje o Nikos tsakothikan me ti Maria

the John and the Nick argued.Rcp with the Maria
' John and Nick argued with Maria.'
A benefit of the formulation in 65 is that it can be used for reflexives as well: if A only has a single
element, the formula is satisfied if we choose = y, and the result expresses the appropriate
reflexive meaning (the introduction of a spurious e ' is harmless).
352
Alexis Dimitriadis
It does not appear that irreducibly symmetric reciprocals are simply collectives.
Our conclusion is supported by Hackl (2002), who shows that "essentially plural" relational nouns are semantically and syntactically distinct from genuine
collectives. As he points out, these are in fact symmetric nominal reciprocals,
and they share many properties with the symmetric reciprocals discussed here.
There is a final alternative that one could pursue, and this is that thematic
roles of this sort are simply the wrong way to look at this phenomenon. Dowty
(1991) argues that thematic roles are not necessary in semantics, and that syntax
only needs them as a means of indexing semantic arguments against syntactic
projections. He shows that the latter can be accomplished by positing just two
"proto-roles", proto-Agent and proto-Patient, that are associated with the subject
and direct object positions respectively. The argument with the most Agent-like
characteristics (Dowty provides a list) is identified as the proto-Agent and appears as the subject, and the one that is most Patient-like becomes the subject.
The existence of only two role types means that uniqueness of roles is not required; in our case, this would mean that a symmetric event is, simply, a twoplace predicate that makes (near-)identical entailments of its two participants.
If the two positions are truly identical, initiative-related entailments for the subject are due to its syntactic position (Dowty recognizes this factor, and expresses
uncertainty about just which properties are thematic and which are due to syntactic position); if the factor of initiative is thematic, it is enough to allow that
participant to claim the proto-Agent role.
If this is the case, a symmetric predicate is simply a symmetric predicate and
there is not much more to say. The problem, from our perspective, is that the absence of unique roles prevents the adoption of a simple neo-Davidsonean event
semantics: As Parsons made clear, Ms system relies on the existence of a unique
thematic relation for each event participant. If we want an explicit system of
eventuality variables and binary role relations, we will still need the kind of system developed for the bundled role account. This is simply a consequence of the
neo-Davidsonean system's need to individuate event participants by associating
them with distinct theta roles.
References
Behrens, L., N. Evans, and R. Nordlinger (2003): Single reciprocals. Paper presented at
the Conference of the Australian Linguistic Society, Newcastle.
Carlson, G. (1998): Thematic roles and the individuation of events. In: S. Rothstein (ed.)
Events and Grammar, 35-51. Dordrecht: Kluwer.
Dalrymple, M., M. Kanazawa, Y. Kim, S. Mchombo, and S. Peters (1998): Reciprocal
expressions and the concept of reciprocity. Linguistics and Philosophy 21, 159-210.
Dimitriadis, A. (1999): Reconciling dependent plurals with each other. In: T. Matthews
and D. Strolovitch (eds.) Proceedings of SALT 9. Ithaca, NY: CLC Publications.
353
Dimitriadis, A. (2000): Beyond Identity: Problems in Pronominal and Reciprocal Anaphora. Ph.D. thesis. University of Pennsylvania.
Dimitriadis, A. (2002): Discontinuous reciprocals and symmetric events. Paper presented
at the Anaphora Typology Workshop on Reciprocals, Utrecht University.
Dimitriadis, A. (to appear): Discontinuous reciprocals. Ms., Utrecht institute of Linguistics OTS. Submitted for publication.
Dimitriadis, A. and M. Everaert (2004): Typological perspectives on anaphora. Proceedings of the International Symposium on Deictic Systems and Quantification in
Languages Spoken in Europe and North and Central Asia. (Izhevsk, Russia, 2001).
Dougherty, R. C. (1974): The syntax and semantics of each other constructions. Foundations of Language 12, 1-47.
Dowty, D. R. (1991): Thematic proto-roles and argument selection. Language 67(3),
547-619.
Faltz, L. M. (1977): Reflexivization: A Study in Universal Syntax. Ph.D. thesis, University of California at Berkeley.
Fiengo, R. W. and H. Lasnik (1973): The logical structure of reciprocal sentences. Foundations of Language 9, 447^168.
Gleitman, L. H. (1965): Coordinating conjunction in English. Language 41, 260-293.
Gleitman, L. R., H. Gleitman, C. Miller, and R. Ostrin (1996): Similar, and similar concepts. Cognition 58, 321-376.
Hackl, M. (2002): The ingredients of essentially plural predicates. In: M. Hirotani (ed.)
Proceedings of N E L S 32, 171-182. Amherst, MA: GLSA.
Heim, I., H. Lasnik, and R. May (1991): Reciprocity and plurality. Linguistic Inquiry
22(1), 63-101.
Johnson, F. et al. (eds.) (1939): A Standard Swahili-English Dictionary. Nairobi: Oxford
University Press, first ed.
Kemmer, S. (1993): The Middle Voice. No. 23 in Typological Studies in Language.
Amsterdam: John Benjamins.
Langendoen, D. T. (1978): The logic of reciprocity. Linguistic Inquiry 9(2), 177-197.
Link, G. (1998): Algebraic Semantics in Language and Philosophy. Stanford: CSLI Publications.
Mchombo, S. A. and A. S. A. Ngimga (1994): The syntax and semantics of the reciprocal
construction in Ciyao. Linguistic Analysis 24, 3-31.
Parsons, T. (1990): Events in the Semantics of English: A Study in Subatomic Semantics.
Cambridge, MA: MIT Press.
Rkosi, G. (2003): Comitative arguments in Hungarian. In: UiL-OTS Yearbook 2003,
47-57. Utrecht: Utrecht institute of Linguistics OTS.
Rkosi, G. (2004): The argument structure of inherent reflexive and reciprocal predicates
in Hungarian. Paper presented at the Workshop on Reciprocity and Reflexivity, Freie
Universitt Berlin, October 2004.
Reinhart, T. and T. Siloni (2003): Thematic arity operations and parametric variations.
Ms..
Reinhart, T. and T. Siloni (forthcoming): Thematic arity operations and parametric variations. To appear in Linguistic Inquiry.
Rothmayr, A. (2004): Zwei Reziprokausdrcke im Deutschen. Ms..
Schwarzschild, R. (1996): Pluralities. No. 61 in Studies in Linguistics and Philosophy.
Dordrecht: Kluwer.
354
Alexis Dimitriadis
Seidl, A. and A. Dimitriadis (2003): Statives and reciprocal morphology in Swahili. In:
P. Sauzet and A. Zribi-Hertz (eds.) Typologie des langues d'Afrique et imiversaux de
la grammaire, vol. 1. Paris: L'Harmattan.
Siloni, T. (2001): Reciprocal verbs.
In: Y. N. Falk (ed.) Proceedings of the Israel Association of Theoretical Linguistics.
Online publication, available at
http://atar.mscc.huji.ac.il/ênglish/IATL/17/.
Siloni, T. (2002): Active lexicon. Theoretical Linguistics 28(3), 383^100.
Vitale, A. J. (1981): Swahili Syntax. Dordrect: Foris.
Williams, E. (1991): Reciprocal scope. Linguistic Inquiry 22(1), 159-173.
Williams-van Klinken, ., J. Hajek, and R. Nordlinger (2002): Tetim Dili: A Grammar
of an East Timorese Language. Canberra: Pacific Linguistics.
Zee, D. (1985): Objects in Serbo-Croatian. In: M. Niepokuj et al. (eds.) Proceedings of
the 11th Annual Meeting of the Berkeley Linguistics Society, 358-371.
Sheila Glasbey (Birmingham)
Existential readings for bare plurals

in object position
1 Introduction
In this paper we take a close look at exactly which predicates (both verbal and
adjectival) fail to allow existential readings for their bare plural objects. We
show that, among the verbs, rather than it being the individual level predicates
(as characterised by previous accounts going back to Carlson 1977) that fail to
allow existential readings, it is a narrower set of verbs which we identify as
corresponding to psychological verbs with experiencer subjects (the psych-ES
verbs). On the other hand, we show that the majority of adjectival predicates
disallow existential readings for their bare plural subjects, except in cases
where a particular kind of contextual support is present. We present an explanation of these observations which relies on the distinction between verbal
predicates, which generally have an event argument, and adjectival predicates,
which do not. We propose, roughly speaking, that an existential reading is
made possible by the presence of a "localising situation", which may be provided either by the event argument of the verb or by an appropriate context.
We propose that the psych-ES verbs are distinctive among verbs in not possessing an event argument, which accounts for their lack of existential readings
for bare plural objects, and we explain why it is those psych-verbs which lack
event arguments tend to have experiencer subjects. Concerning adjectival
predicates, we show how our analysis explains the context-dependency of the
availability of existential readings for their bare plural subjects. This allows us,
too, to explain why 'John hates lawyers' (as discussed in Cohen & ErteschikShir 2002), while not normally allowing an existential interpretation of 'lawyers', is able to do so in certain very restricted contexts. Finally we offer a
brief and very tentative proposal concerning the nature of psych-ES verbs
which may explain their lack of event arguments.
356
Sheila Glasbey
2 'Hate' and ' K n o w '

We begin with a previously noted problem (see, for example, Cohen &
Erteschik-Shir 2002):
(1)
(2)
John knows lawyers.

John hates lawyers.
Why is an existential reading of the bare plural object 'lawyers' available in

(1) but very difficult, if not impossible, in (2)?
Most analyses of the English bare plural fail to account for this observation.
In Carlson (1977), 'know' and 'hate' are both individual level (i-level) predicates, and for this reason the bare plural must be interpreted as a "kind", giving
rise to a generic reading for 'lawyers' in both (1) and (2). In Kratzer (1995)
'know' and 'hate' are, once again, classified as i-level. In Kratzer's analysis,
based on a syntactically induced restrictor-scope partition, direct objects are
mapped into the scope and are thus predicted to receive existential readings.
Hence the observed reading is predicted for (1) but not for (2). Kratzer uses a
mechanism known as 'scrambling' which allows objects to move into the restrictor. This would allow 'lawyers' in (2) to receive a generic reading. However, since scrambling is an optional process, we have no explanation of why
'lawyers' in (2) apparently must scramble and thus can only be interpreted as
generic.
A number of recent analyses propose that the availability of existential readings for bare plurals require the predicate in question to be spatially located (in
a sense that varies among different accounts). (See, for example McNally
1998, Dobrovie-Sorin 1998.) Although we believe that there are important
intuitions here, spatial locatedness does not appear to work in (1) and (2)
above. We would need some reason to say that "know" eventualities 1 are spatially located while "hate" eventualities are not, and we can think of no convincing way to justify this. Examples like (3) are sometimes cited to try to
show that "know" eventualities are spatially located:
(3)
I know lawyers in Durham.
However, it is not clear that the possibility of using the spatial modifier 'in
Durham' here provides any evidence of spatial-locatedness. One very natural
paraphrase of (3) is know lawyers who live/work in Durham', where, of
course, it is the living (or working) that is spatially modified, rather than the
We use the term 'eventuality' to refer to events and states, as in Bach (1986).
357
knowing. Another paraphrase is apparent if we change the example to past

tense, as in (4):
(4)
I knew lawyers in Durham.
(4) can readily be paraphrased as knew lawyers when I lived/worked in Durham'. Here, the spatial modification appears to be temporal rather than spatial.
For further discussion of this issue, see Cooper (1985, 1986), Glasbey (1994).
Our point here is simply that it is very hard to argue that examples like (3)
show "know" eventualities to be spatially located. And, furthermore, even if
we were to convince ourselves that they were, we would also need to show,
somehow, that the "hate" eventuality described by (5) is not spatially located:
(5)
I hate lawyers in Durham.
Since there appear to be no sound reasons for classifying "know" but not
"hate" eventualities as spatially located, we will abandon this line of explanation in its present form - although there may well be connections between it
and the account we will shortly propose.
3 Cohen and Erteschik-Shir's account

Recently, Cohen and Erteschik-Shir (2002) (henceforth C & E-S) presented an
analysis of bare plurals that relies on the topic-focus distinction, as characterised, for example, in Erteschik-Shir (1997). C & E-S claim that bare plural
objects that are m focus position have an existential reading. Thus in (6), where
'John' is the topic, 'lawyers' has an existential reading by virtue of being in the
focus:
(6)
[John],opc knows lawyers.
But of course, 'John' can similarly be in topic position within a similar sentence with 'hate':
(7)
[John]loplc hates lawyers.
So why does 'lawyers' lack an existential reading in (7)? C & E-S explain this
using the phenomenon of "incorporation" from van Geenhoven (1996). According to van Geenhoven and others, bare plurals on the non-generic interpretation denote properties. C & E-S (p. 151) explain how, according to van
Geenhoven, a verb such as 'see' has a non-incorporating version, (a), whose
arguments are individuals, and an incorporating version, (b), whose argu-
358
Sheila Glasbey
ments are an individual and a property respectively. The two versions can be
represented as follows:
(a)
(b)
y. . see(x,y)
. . 3y: P(y) & see(x,y)
non-incorporating version
incorporating version
C & E-S (p. 152) propose that (b) is related to (a) by type-shifting. Bare plural
objects on a non-generic reading are interpreted according to (b). Consider (8):
(8)
John saw spots.
This is represented as:

3y: spot(y) & see(j, y)
Let us see how this works for (1). Using the incorporating version of 'know',
C & E-S give the following Discourse Representation Theory (DRT: Kamp &
Reyle 1993) representation for (1):
John(x)
know-lawyers(x)
The bare plural 'lawyers', according to C & E-S, does not introduce a discourse referent. Type-shifting then takes place, to give the discourse representation structure (DRS):
John(x)
BY(lawyers(Y) & know(x, Y))
Note that the existential quantifier here does not correspond to a discourse referent.
C & E-S claim that it is impossible to construct a similar DRS for (2). The
reason they give is that 'hate' introduces the presupposition that hate(x, y) presupposes know(x, y).2 They represent this using van der Sandt's (1992) treat-
It seems that this presupposition is at least open to question. We would contest the statement
that 'x hates y' presupposes 'x knows y'. However, we do not discuss the issue here, since we
give below what we believe are clearer reasons for rejecting C & E-S's analysis.
359
ment of presupposition in DRT, as below, where the presupposed material is

enclosed in the dotted box:
John(x)
hate-lawyers(x)
1 3 Y(lawyers ( Y) _& ko w(x, _Y))
After type-shifting, this becomes:
John(x)
3Z(lawyers(Z) & hate(x, Z))
[ H M a ^ r s ( Y ) & knowx". Y))
C & E-S point out that the problem here is that and Y are distinct variables
and there is thus no way to ensure that the lawyers John hates are the same
lawyers that he knows. This, they claim, blocks the existential reading for
'lawyers' in (2).
That is, according C & E-S the presupposition 'John knows lawyers' blocks
the type-shifting responsible for the existential interpretation of 'lawyers'. Or,
perhaps, it does not necessarily block the type-shifting as such, but it prevents
it from happening in the way that would give the required interpretation.
Apart from our doubts about the existence of the presupposition in question,
we note that there may be a way round this problem. Why not assume that the
presupposition is initially expressed in "pre-type-shifted" form, giving:
John(x)
hate-lawyers(x)
; know-lawyers(x)
And now, why not assume that the type-shifting occurs simultaneously in both
cases, giving rise to:
360
Sheila Glasbey
X
John(x)
3Y(lawyers(Y) & hate(x, Y
13yYlawye"r7(Yy& know(x~. Y))'
We are not aware of anything that might prevent such simultaneous typeshifting, which would give rise to a single variable Y, as shown, and hence
provide the desired reading.
Even if simultaneous type-shifting is not permitted for some reason, C & ES's account has a further problem. As they point out, in certain contexts (1)
does have a reading where 'lawyers' is existential. Imagine a scenario where
John, for some reason, makes a list of the people he hates. Mary, on reading
his list, remarks to someone else:
(9)
Ah - 1 see John hates lawyers...
In this context, 'lawyers' in (9) can clearly have an existential interpretation.

Let us assume that this change in interpretation is a purely contextual effect,
rather than being in any way due to the additional material added in (9). This
seems reasonable, given that Mary could just have (say) nodded and uttered
'Ah - John hates lawyers....'.
C & E-S try to explain this reading (which we will call the 'hate-list' reading) by claiming that, in this context, there is no presupposition that John
knows the lawyers on his list. But this seems wrong. We would argue that exactly the opposite is true. In order to make a list of names of people he hates,
there is a strong presupposition that John knows these people - certainly
stronger than in the case of an utterance of (1) in a null context. In the null
context, John may simply hate lawyers en masse, on the basis of his received
impression of the legal profession, without needing to know a single individual
lawyer. In the hate-list context, however, he would presumably need to know
those lawyers whose names he writes down. Of course there is a degree of
vagueness in the meaning of 'know' - for example, over the question of
whether one needs to have met someone in person in order to know them, or
whether it is enough to have had, say, an extensive correspondence with them.
But this does not affect our point here.
361
4 Some further data: apples and actresses

Consider the following examples:
(10)
(11)
My uncle loved apples.

My uncle loved actresses.
Both contain the verb 'love'. Yet (10) has only a generic reading of 'apples',
while (11) allows us to interpret 'actresses' either as generic or, in certain contexts, as existential. To appreciate the latter, imagine that my uncle has been
dead for a number of years and I am discussing his life history and, perhaps,
his reputation as a womaniser. I may then use (11) to convey that during his
lifetime, my uncle underwent periods of loving, or "being in love with" 3 a
number of actresses. For example, he may have loved Dame Margaret Rutherford for a few years, and Dame Peggy Ashcroft for a further time (leaving
open the possibility of these time periods overlapping).
What explanation could C & E-S give for this? In order to exclude the existential reading of 'apples' in (10), they would presumably have to introduce
some kind of presupposition - e.g., to love an apple, one must have tasted it
(or, perhaps, seen it). This seems very odd. Speaking of loving a particular
apple strikes one as rather unnatural, anyway. We are much more likely to
speak of loving apples as a type of fruit or foodstuff. Furthermore, in order to
explain the existential reading for 'actresses' in (11), we would then need to be
able to show that in this particular context, there is no relevant presupposition
of any kind. This would be very difficult - for example, we might well want to
say that in order to love an actress in the sense intended, one would need to
know that actress, in some relevant sense of 'know'. Unless we have an independent way to rule out any such presupposition, the explanation becomes
circular.
We conclude, therefore, that C & E-S's explanation of the different interpretations for the bare plurals in (1) and (2) is unsatisfactory. In the next part
of the paper we will take a closer look at the data, in order to discover exactly
which verbs do and do not give existential readings for their bare plural objects. On the basis of our findings, we will offer an alternative analysis.
5 Towards an improved analysis - some pointers

Here are some points to be considered as we seek a better analysis:
We have seen that it is not just the verb that counts. The same verb,
e.g. 'hate' or 'love', may or may not have an existential bare plural
Not necessarily in the sense of being loved in return.
362
Sheila Glasbey
object, depending on contextual factors and the nature of the bare plural object.
We need to consider more carefully what it is about the 'hate-list' and
the 'uncle's-love-life' contexts described above that allow them to
give an existential reading to bare plural objects.
We need to look more closely at:

(i) exactly which verbs allow existential readings for their bare plural
objects with no need of contextual support;
(ii) which verbs allow such readings only with contextual support;
(iii) which verbs, if any, never allow existential readings for their bare
plural objects, no matter what the context.
6 A closer examination of the data

What exactly is the data to be explained? It is often taken as read that those
verbs classified by Carlson (1977) as i-level give only generic interpretations
for their bare plural objects. Sometimes a few exceptions - e.g. 'know', as discussed above, are admitted. Yet a closer look shows this accepted picture to be
false. Notice, for example, that many verbs that might well be classified as ilevel (in that they are lexically stative, tendentially stable predicates 4 ) allow
existential interpretations. Examples include 'know', 'own', 'include', 'give'
'support' and 'provide', as shown in the examples below, all of which allow
existential interpretations of their bare plural objects:
(12)
(13)
(14)
(15)
(16)
John owns racehorses.

Mary's lecture notes include examples.
This article gives counter-examples.
The party manifesto supports controversial policies.
This letter provides counter-arguments.
All these verbs are lexical statives (shown by the fact that they do not combine
readily with the progressive, and they can be used in the simple present without receiving a habitual interpretation. See Smith 1991, for example). All may
be regarded as tendentially stable. Thus they are good candidates for i-level
predicates.
Of course, we could simply decide that, on the basis of the existential bare
plural readings, these verbs are by definition not i-level. But if we cannot provide independent criteria for the classification, the i- and s-level distinction
loses its explanatory value and the explanation is circular. So we will avoid
taking this step (and we will, eventually, discard the i/s distinction anyway).
4
See (Chierchia 1995), for example.
363
So, can we identify and offer independent criteria to classify those verbs
which do not give existential readings for bare plural objects? Do these verbs
form a coherent group? We have seen that 'hate' and 'love' do not allow existential bare plural objects (if we leave aside for the moment the exceptional
contexts that allow such readings). 5
From a preliminary study of verbs as classified in Levin (1993), we have
identified the following 6 as verbs which do not (excluding exceptional contexts
of the kind discussed above) allow existential readings for their bare plural
objects: hate, love, like, adore, respect, despise, deplore, envy, ... These verbs
all fall into Levin's (1993) class of "verbs of psychological attitude with experiencer subjects" (henceforth 'psych-ES verbs'). As far as we can tell from
our preliminary investigation, almost none of the psych-ES verbs allow existential readings for their bare plurals objects, except in the exceptional ("hatelist") contexts. 7 Moreover, we have not, so far in our preliminary survey, found
any verbs other than the psych-ES verbs that do not allow existential readings
for their bare plural objects (discounting, once again, the exceptional contexts).
We would, however, recommend a detailed corpus study in order to investigate
the correctness of this generalisation.
Let us proceed on the basis of what is, at least, suggestive evidence. We
may now ask why it should be that the psych-ES verbs fail to give existential
readings for their bare plural objects? We will present a detailed explanation
below. In brief, our explanation will rest on the fact that such verbs, unlike all
other verbs, do not possess an eventuality argument. Of course, this is reminiscent of earlier accounts, such as Kratzer (1995). However, we will not give a
syntax-based explanation like Kratzer's ; our account will be primarily a semantic one. We will also offer some suggestions as to why it should be the
psych-ES verbs, in particular, that show this property. Our analysis, as will
become clear, is a further development of the account of bare plural subjects
given in Glasbey (1998) and Glasbey (1999). We also borrow insights from a
number of previous works including Kratzer (1995), Chierchia (1995) and
McNally (1998). We will offer an explanation not only for the bare plural object data considered so far, but will give a brief treatment of bare plural subjects. In particular, we will show how we can explain the context-dependency
of readings of bare plural subjects of adjectival predicates (this will expand
upon an earlier treatment in Glasbey 1999).
Are we justified in regarding such contexts as exceptional? We believe so, and ask the reader
to bear with us at this point. Our analysis will show that this was justified. If we do not take
this step, then we may well find that for any verb we can always find some context that would
allow an existential bare plural object, and we would forgo any possibility of distinguishing
those verbs that need supportive contexts from those that don't, and may thereby miss important generalisations.
Note that this is not the full set, but a subset given here for illustrative purposes.
'Admire' is the only clear exception we have identified. See the discussion of 'admire' below.
364
Sheila Glasbey
7 Classification of predicates
We begin our analysis by dividing predicates into two groups - verbal predicates such as 'love', 'admire' and 'run' and adjectival predicates such as
'happy', 'hungry' and 'intelligent'. We propose that verbal predicates, in general, have what we will call an 'eventuality argument' (we will flesh this out in
more detail below). Adjectival predicates have, in general, no such eventuality
argument.
Thus we have:
e.g. 'love',
'eat', 'run',
'admire'
e.g. 'happy', 'intelligent',

'hungry'
In the above diagram, +e denotes the possession of an eventuality argument in

the verb's argument structure; -e denotes the lack of one.
We may choose to represent the eventuality as a verbal argument using a
notation based on (Davidson 1967):
e a t (e, , y)
Thus (17):
(17)
John ate cake.
would be represented as e a t ( e , j , c ) , where e is the eventuality, j is John

and c is cake. Note that we do not attempt to represent tense, proper names or
mass nouns here. Similarly (18):
(18)
John knows Mary.
would be represented as know ( e , j , m).

By making our primary classification of predicates into verbal and adjectival, we are cutting across the classification proposed by Carlson (1977) where
365
the primary division is between i-level and s-level predicates. We believe that
this revision is justified by the analysis that it allows - see below.
We will adopt a form of representation based on situation theory (Barwise
& Perry 1983) and its applications to natural language semantics, often known
as situation semantics (see, for example, Cooper 1985, Cooper 1986, Barwise
& Cooper 1991, Glasbey 1994). Our reasons for choosing this framework will
become clear below.
8 Situation semantic analysis

We can represent the meaning of (17) in a situation semantic framework as:
e
eat(j, c)
This is the proposition that the situation 'e' supports the infon e a t ( j , c ) .
We use the Extended Kamp Notation (EKN) of Barwise & Cooper (1993). For
a detailed explanation of EKN and of situation semantics more generally, see
Cooper (1992). For examples of further use of this notation, see Glasbey
(1994) and Glasbey (1998). In situation theory, situations are parts of the
world, which may support (or "make true") units of information, known as
'possible facts' or 'infons'. An infon may be thought of as roughly equivalent
to a condition in DRT - except, of course, that standard DRT does not employ
situations.
In the above, 'j' represents the individual 'John', and c 'cake'. Once again,
we have not attempted to represent the tense of the sentence nor dealt with the
NPs in any principled way. See Glasbey (1994) for a treatment of tense and
aspect in situation theory, and Cooper (1993) for a treatment of NPs, proper
nouns, etc.
We regard an eventuality as a particular type of situation. We assume here
that e is the minimal or "smallest" situation supporting the infon e a t ( j , c ).
This may be thought of, roughly, as the situation that supports this particular
infon and no other. The notion of minimal support is actually a little more
complicated than this (see Glasbey 1994 for discussion) but the simplified version suffices here.
Let us now move onto the situation semantic representations of sentences
containing bare plural objects. We begin with (1), repeated here as (19):
(19)
John knows lawyers.
366
Sheila Glasbey
We follow van Geenhoven 1995 in treating bare plurals as denoting properties.

The situation semantic representation is:
e
know(j, lawyer)
The supporting situation (eventuality) e is in this case a state, since the verb
'know' is (lexically) stative, 'j' is the individual named 'John', and 'lawyer' is
the property of being a lawyer.
Where does the existential reading for 'lawyers' come from? 8 It comes, we
propose, from a process known as the 'existential inference'. The existential
inference is licensed by a situation - in this case by the eventuality e. The existential inference yields individual situations - one for each "knowing" of an
individual lawyer by John. It may thus be thought of as a kind of type-shifting
process - but one that can take place only under specially-licensed circumstances - the presence of an appropriate situation. We can represent the existential inference as follows:
e
know(j, lawyer)
Existential inference
Be', 1 such that:

e
_ J
know(j, 1)
In the above diagram, e' is an individual situation supporting the infon

know ( j , 1 ), where 1 is an individual lawyer.
g
We may also ask where the generic reading, to the extent that there is one available for this
example, comes from. This chapter does not attempt to deal with generic readings. See Glasbey
(1998) for a possible treatment of generic readings within a channel theoretic related framework. Note, however, that alternative treatments of generics may well be compatible with the
account of existential readings given in the current paper.
367
The existential inference is, as stated above, licensed by e - the "overall

situation". Thus, it is the fact that 'know' has an eventuality argument that
makes the existential inference possible.
Notice that the primary outcome of the existential inference is the yielding
of individual situations e'. Individual lawyers, 1, are derived, one per situation,
as a secondary effect.
We will see later that the existential inference is also responsible for existential readings for bare plural subjects.
Since we claim that verbs in general have eventuality arguments, this predicts that all verbs will allow existential readings for their bare plural objects
(and subjects). Clearly, this is incorrect in the case of the psych-ES verbs. At
this stage, we will simply propose that psych-ES verbs do not have eventuality
arguments. Thus there is no situation provided by the verb to license the existential inference, and therefore no existential readings for bare plural objects.
However, we still need to explain, of course, why these particular verbs lack
eventuality arguments. We will return to this towards the end of the paper.
Notice that our account, as it stands, also predicts that since adjectival
predicates do not have eventuality arguments, none of these will give existential readings. This is clearly wrong, and we will address it in detail below.
For the time being, however, our classification of predicates looks as follows:
Predicates
Adjectival
(-e)
Psych-ES verbs
(-e)
No existential bare
plural objects
"the rest"
(+e)
Existential bare
plural objects
Let us look at the psych-ES cases, where no existential inference is available,

in a little more detail. We will represent (2), repeated here as (20), as shown
below:
(20)
John hates lawyers.
368
Sheila Glasbey
hate(j, lawyer)
Notice that no situation supports the infon here. Presumably, in that case, we
might consider the infon to be supported by the maximal or "world" situation
that we will call 'w':
w
hate(j, lawyer)
We may ask why it is that ' w \ the "world situation" cannot support the existential inference. We offer two possible answers here. The first is that
h a t e ( j , l a w y e r ) is not an infon at all, and is therefore not supported by
any situation. In situation theory, there are two types of predicates, "relations"
and "types". Relations are the ones we have used so far in this discussion; we
have thought of them roughly as being denoted by verbs. Relations combine
with arguments to produce infons, which are then supported by situations, as
we have seen. The second kind of situation theoretic predicate, types, do not
form infons. A type has one or more arguments, once again corresponding to
the arguments of the verb. If 'hate' is a type, then we represent the meaning of
'John hates Mary' as:
j,m
hate
This is the proposition that the individuals j ('John') and m ('Mary') are classified by the binary type 'hate'. The information encoded here can be thought of
in some sense as 'non-local'. The fact that John hates Mary is not supported by
any particular situation (or part of the world) - it may be roughly described as
general, non-localised information.
This allows us to say that, since there is no situation here, the existential inference cannot be licensed, and hence there is no existential reading for the
bare plural. Spelling this out in detail, for (20), we now have:
369
j, lawyer
hate
Here, we are now considering the binary type 'hate' to hold of the individual j
('John') and the property 'lawyer'. Now we can see that there is no supporting
situation, and thus no existential inference and no existential reading for 'lawyers'.
Alternatively, we may continue to regard 'hate' as a relation, i.e., as forming infons which are supported by situations. We then go back to saying that
the world situation, w, supports the infon, i.e.:
w
hate(j, lawyer)
But if we do this, we must explain why w does not permit the existential inference. We might perhaps say that the existential inference relies on some notion
of localisation. In other words, it is the localised nature of the situation that
somehow results in the inference to particular situations. But we realise that
this is an aspect of our analysis that is incomplete, whichever route we take.
Further work is needed to clarify the nature of the existential inference and its
connection with localisation. (Indeed, what exactly is localisation? Is it spatial,
or spatiotemporal, for example? We suspect but cannot currently prove the
latter.) However, in spite of being unable to say at this point exactly what the
existential inference is, we believe it is a useful starting point for further investigation and would welcome further research on the subject. We will return to
the relation/type distinction when we consider adjectival predicates below.
Before moving on, we need to refine our classification of predicates
slightly, in order to incorporate the event/state distinction. It is not essential to
our analysis that we make this distinction, since we are saying that both (lexical) event and (lexical) state verbs can have eventuality arguments - but the
refinement will make things clearer and easier to compare with other accounts.
Our revised classification is thus:
370
Sheila Glasbey
Predicates
Verbal
'
Events
(+e)
All have
existential
bare plural
objects
Adjectival
(-e)
States
psych-ES verbs
(-e)
No existential bare
plural objects
"the rest"
(+e)
Existential bare
plural objects
Here, we simply divide verbal predicates into (lexical) events and (lexical)
states. Event verbs all have an eventuality argument, and hence these will always give existential readings for bare plural objects (and subjects) - as indeed
has been shown to be the case in much previous work. State verbs divide into
two categories - the psych-ES verbs, which do not have an eventuality argument, and the rest of the stative verbs, which do.
In order to develop our analysis further, we now need to consider (briefly)
the interpretation of bare plural subjects. This has been alluded to a number of
times so far in our discussion: it is now time to make things clear. What follows in the remainder of this section is a slightly revised account of part of
Glasbey (1998).
Consider (21), which has both a bare plural subject and bare plural object:
(21)
Dogs chased cats in my garden last night.
Note that the temporal and locational adverbials are included solely to ensure
that the desired interpretation is selected - one where there was a particular
situation (event) consisting of chasings of cats by dogs at a particular spatiotemporal location.
We can represent the meaning of this as follows, ignoring the past tense and
the modifiers in order to keep things simple:
371
3c', d, c such that
e'
chase(dog, cat)
chase(d, c)
In the above, d is an individual dog, c an individual cat and e' is an individual

situation (event) of a dog chasing a cat. 9 The existential inference is licensed
here by virtue of 'chase', an eventive verb, having an event argument, and
therefore providing a situation to license the existential inference.
Our treatment of bare plural subjects predicts that existential readings will
be available for bare plural subjects in exactly the same linguistic environments as those that allow existential readings for bare plural objects. Any observed difference in available readings will have to be accounted for via the
contextual effects to be discussed further below.
9 Adjectival predicates
We turn now to adjectival predicates. We stated above that we take these to
lack eventuality arguments. The prediction is, then, that none of them will give
existential readings. But this is clearly false - as shown, for example by one of
the examples from Carlson (1977) which helped to motivate his i/s-level distinction.
(22)
Firemen are available.
Clearly 'firemen' may be interpreted as existential, and indeed this is the more
salient reading. How can we explain this?
We should note before proceeding that many adjectival predicates - including many of a temporary nature that we might otherwise want to classify as slevel - do not at all easily give existential readings for bare plurals. This has
been noted before - see Greenberg (1994) and Glasbey (1999) for further examples and discussion.
Consider the following (both from Glasbey 1999):
We have glossed over an important issue here regarding collective and distributive readings.
Presumably, scenarios are possible, where, for example, two dogs together chase one cat. Thus,
not all the individual situations will necessary be one dog: one cat chasing events. Indeed, in
the extreme, the whole bunch of dogs may collectively chase the whole bunch of cats. We do
not believe that this causes any problems for the notion of existential inference, since there appears to be no problem with the existential inference yielding just one individual situation.
However, the details would need to be worked out.
372
(23)
(24)
Sheila Glasbey
Children are hungry.

Plates are dirty.
In a neutral context, it is hard to get any reading (whether existential or generic) for these sentences. A generic reading is available for (23) if one takes it
to mean that children have a tendency to be (continually) hungry. Such a reading is difficult for (24) or is perhaps simply untrue, since plates do not have an
inherent tendency to be dirty. We will not consider generic readings further
here. Of more concern to us is the question of why the existential readings are
difficult or impossible. 'Hungry' and 'dirty' are both most naturally interpreted
as temporary, non-tendentially stable predicates. They are thus prime candidates for s-level, and presumably would be classified by Carlson (1977) as
such, and therefore predicted to give existential readings.
Now consider the effect of context. Imagine someone describing her recent
visit to a rather badly run children's camp. She might say the following:
(25)
The place was in an appalling state. Children were hungry and raiding the kitchens for food.
Plates were dirty, knives and forks were all over the floor, carpets were filthy - and the
adults were all sitting in the next room watching TV.
See also the examples in Glasbey (1999 p.86 (3) and (4) and C & E-S (2002
p. 142 (51a-g).
In this example it is not difficult to interpret 'plates' and 'children' as existential. (25) could be true if only some of the children were hungry, and/or
some of the plates were dirty, and/or some of the knives and forks were on the
floor, and/or some of the carpets were filthy. Why, we may ask, has the existential reading suddenly emerged here?
Intuitively speaking, the effect appears to have something to do with the detailed descriptions of the surrounding context. The fact that we have switched
to past tense also seems to help, although this does not appear to be enough on
its own. We are being led to consider, in other words, a "local", localised situation - that of the children's camp. But why should this make the existential
reading possible? 10
Interestingly, there are other adjectival predicates where no amount of contextual manoeuvring seems to make an existential reading available. Consider,
for example:
(26)
I really liked the school where I did my teaching practice. Parent helpers were available.
Children were intelligent and hardworking.
10
The reader may be reminded of the 'hate-list' examples considered earlier. Indeed, our final
explanation will cover both cases.
373
Notice that, while 'parent helpers' can be interpreted as existential, it is difficult, if not impossible, to give 'children' an existential interpretation - even
though we have provided the kind of localised context that we did in (25). It
looks as though some adjectival predicates will provide existential readings
irrespective of context, as in (22), while others require contextual support to
give such readings, as in (25), and others will never give existential readings,
even in what is presumably a supportive context as in (26). How can we explain this?
We begin by adopting the situation semantic proposal discussed earlier that
predicates come in two kinds - relations and types. Relations, the reader will
remember, combine with arguments to form infons, which may be supported
by situations. The proposition that a given situation supports an infon, e.g.
s
hungry (j)
is known as an Austinian proposition, after the philosopher J. Austin. The

other kind of predicate, types, form propositions which simply classify one or
more individuals (or properties) as being of that type. Such a proposition is
known as a Russellian proposition, after Bertrand Russell. An example is:
j
intelligent
Here, we have got ahead of ourselves slightly by proposing that "intelligent"

should be a type. But this, of course, is what we require in order to explain
why 'children' in (27) does not have an existential reading, no matter how
much, it would appear, we manipulate the context.
(27)
Children are intelligent.
By classifying "intelligent" as a type, we are blocking the existential inference

as discussed above. We propose that "dangerous", "tall", etc., be classified as
types. This category therefore corresponds to what Carlson (1977) classified as
adjectival i-level predicates.
Now let us turn to the adjectival predicates like "hungry" and "dirty" which
seem to need contextual support in order to give existential readings. Why
should this be so?
374
Sheila Glasbey
Remember that we proposed that adjectival predicates in general do not

have eventuality arguments. Let us suppose that such predicates are situationtheoretic relations (while at the same time not having eventuality arguments).
This means that they form infons, which may be supported by situations. But
where are these situations to come from? If the predicate has no eventuality
argument, then the situation clearly cannot come from here. Now, suppose
that, in some case, an appropriate context were to supply the situation. In (25)
above, the context, as we already noted, could be characterised as something
like "the state of the children's camp". If we consider this context to provide a
situation (it is, after all, part of the world, as identified by the description), then
we have a situation, provided by context, that can support the infon and
thereby license the existential inference. Thus for (28):
(28)
Children were hungry,
we have:
3s', c such that:

s'
s
hungry(child)
hungry(c)
where s' is an individual situation and c is an individual child.

Now, of course, this is only possible because we made "hungry" a relation.
We therefore have an explanation for the context dependency of existential
readings in a certain class of adjectival predicates. We have also dealt with
those adjectival predicates that never give existential readings, by classifying
them as types. Finally, we need to deal with the third class of adjectival predicate - those like "available" that appear to give existential readings very easily,
without the need for any supportive context.
We propose that predicates like "available" are exceptional among adjectival predicates in that they have something akin to an eventuality argument,
which therefore supplies a situation and licenses the existential inference.
Other predicates in this class are "present" and "visible". However, we are not
entirely clear whether this should be considered to be an eventuality argument
or whether it might make more sense to consider it as a situation argument of
some other kind - perhaps as a spatiotemporal location. We will not spend any
more time on the matter here. There do not appear to be many of these predicates, and it seems clear that they should be treated as exceptional rather than
mainstream within the class of adjectival predicates. See (Glasbey 1999 p.101)
for further discussion.
We can now add our treatment of adjectival predicates to our classification
of predicates.
375
10 Final classification of predicates

Our final classification of predicates is given below:
Predicates
Verbal
(mainly +e)
Relations
Events
Adjectival
(mainly -e)
States
Relations
Types
(+e)
Psych-ES verbs 'other states"
(+e)
(+e/s)
(-e)
(-e)
Existential
No existential
Existential
Existential
Existential
No
bare plurals
bare plurals
bare plurals
bare plurals
bare plurals
walk, eat,
love, hate,
own, have,
available,
if context
like, despise
know
present
provides a
intelligent,
situation
dangerous,
(-e)
hungry,
existential
bare plurals
dirty
tall
But we have not yet finished our account. Note that we still have to do the following:
(i)
Explain why 'hate' gives an existential bare plural object reading in

the 'hate-list' context;
(ii) Explain why 'My uncle loves apples' does not give an existential
reading while 'My uncle loved actresses', in certain contexts, does.
(iii) Explain why the psych-ES verbs lack eventuality arguments.
We will now give an account of (i) and (ii) and make some preliminary and
tentative suggestions about (iii).
Firstly, then, let us return to the hate-list examples. To recap, we saw earlier
that although (2) (repeated here as (29)):
(29)
John hates lawyers.
376
Sheila Glasbey
does not normally give an existential reading for 'lawyers', in certain contexts
such as John's writing a list of people he hates, the existential reading does
become available. We can give an explanation similar to the one we gave for
the adjectival predicate examples like (23), repeated here as (30):
(30)
Children are hungry.
We classify "hate" as a relation, since we take all verbs to express relations.

Because it is a relation, if we can find a context that will provide a situation to
support the infon, then the existential inference will be possible and we will
have an existential reading. Now it seems very plausible that the hate-list context can provide exactly the situation we need. We could think of the situation
as corresponding to the list itself, or to John's action in writing the list. Then
the existential inference can take place as follows:
3s', 1 such that:

7
hate(j, lawyer)
hate(j, 1)
That is, our account predicts that whenever context provides a suitable "localising" situation, for an infon from a verbal predicate, the existential reading of
the bare plural will become available.
Now let us turn to (10) and (11), repeated here as (31) and (32):
(31)
(32)
My uncle loved apples.

My uncle loved actresses.
Our task is to explain why 'apples' lacks an existential reading, while 'actresses', in certain contexts, has one.
Earlier, in order to point the reader towards an existential reading in (11/32),
we introduced the scenario where my uncle's life (and possibly his womanising) was under discussion. Thus, once again, it looks as though we can see this
scenario ("my-uncle's-love-life") as a "localising" situation which supports the
relevant infon and makes possible the existential inference.
To summarise briefly at this point: we have introduced a two-way distinction
(i) between types and relations;
(ii) between predicates with eventuality arguments and those without.
This allows us to explain the previously problematic context dependency of
existential readings for the bare plurals of both verbal and adjectival predicates. In particular, it allows us to give the same explanation for both.
377
Now for our final challenge. Can we explain why it is that the psych-ES
verbs should lack eventuality arguments?
11 Why do the psych-ES verbs lack eventuality arguments?

The list of psych-ES verbs given in Levin (1993, p.191) is shown below. We
present the list as given in Levin (1993), divided into two sub-classes, "positive verbs" expressing a positive attitude on the part of the experiencer, and
"negative verbs" expressing a negative attitude on the part of the experiencer.
Positive verbs:
Admire, adore, appreciate, cherish, enjoy, esteem, exalt, fancy, favour, idolise,
like, love, miss, prize, respect, relish, revere, savour, stand, support, tolerate,
treasure, trust, value, venerate, worship.
Negative verbs:11
Abhor, deplore, despise, detest, disdain, dislike, distrust, dread, envy, execrate,
fear, hate, lament, loathe, mourn, pity, regret, resent.
Note, firstly, that none of these verbs appear, informally speaking, particularly "event-like". We would not naturally speak of an event of Mary respecting Jane or of Fred resenting Bill, for example. (An interesting exception may
be 'enjoy', where it does perhaps appear quite natural to speak of an event of,
say, Mary enjoying an ice-cream. We suspect that the event-like quality may
come from the contribution of the eventive verb eat' here, given that 'Mary
enjoyed the ice-cream' means something like 'Mary enjoyed eating the icecream'. This should be investigated further, but we will not do so here.)
How can we be more precise about what we mean by 'event-like'? One approach is to take the fairly standard view of events as involving a change of
state, sometimes in one or more of the participants. One stereotypical type of
event involves one participant (often called the 'agent') taking part in some
action which results in a change in the other participant (the 'patient'). Dowty
(1991) identifies sets of properties typically associated with agents and patients, and from this develops a characterisation of thematic proto-roles (shown
below). In particular, note that a change of state in a participant is a contributing factor for that participant to be a proto-Patient.
In Dowty's framework, the linking between syntactic and semantic representations is determined by clusters of verbal entailments, known as protoAgent and proto-Patient properties. According to Dowty (1991 p. 572), contributing factors for the Agent proto-role are as follows:
(a) Volitional involvement in the event or state;
(b) Sentience (and/or perception);
(c) Causing an event or change of state in another participant;
11
One verb 'rue' from this class was prefaced by a question mark in (Levin 1993) and we have
omitted it.
378
Sheila Glasbey
(d) Movement (relative to the position of another participant;

(e) (Exists independent of the event named by the verb.)
Contributing factors for the Patient proto-role are:
(a) Undergoes change of state;
(b) Incremental theme;
(c) Causally affected by another participant;
(d) Stationary relative to movement of another participant;
(e) (Does not exist independently of the event of the event, or not at all.)
Now Dowty's argument selection principle (Dowty 1991 p. 576) is as follows:
"ARGUMENT SELECTION PRINCIPLE: In predicates with
grammatical subject and object, the argument for which the predicate entails the greatest number of proto-Agent properties will be
lexicalised as the subject of the predicate; the argument having the
greatest number of proto-Patient properties will be lexicalised as the
direct object."
Filip (2000) explains how this works for the related pair of psychological
predicates, 'frighten' and 'fear'. In her words:
"Although the Experiencer argument of the fear and frighten classes
are equal in Agent properties, they are unequal in that the Experiencer of the frighten class denotes an entity that undergoes a change
in the denoted event, and hence it is a 'better' patient. Therefore, it
must be the direct object (cf. Dowty 1991: 580)." (Filip 2000 p. 9.)
Notice, therefore, that in general those verbs which refer to events will be ones
where one participant (which we will call A) causes a change of state in the
other participant, B. The A participant will therefore tend to be a proto-Agent,
and will be realised as the subject, and the participant will tend to be a pro toPatient and be realised as the direct object. This explains, for example, why in
the case of the eventive verbs 'chase' and 'eat', the chaser and the eater are
realised as subjects, while the chased and the eaten are direct objects.
Now consider the psychological verbs (psych-verbs). Let us begin by assuming that some of these "describe events" and some do not. To be explicit,
what I mean by saying that a verb "describes an event" is that it has an event
argument, and that it therefore allows existential readings for subject and object bare plurals (the latter having been the focus of our discussion in this paper). According to our analysis given above, it is the presence of the event argument that licenses the existential reading for bare plurals. Let us also sup-
379
pose that there are some psychological verbs that do not, in the sense above,
describe events - that is, they will not allow existential readings for bare plurals. We will refer to these two classes of psych-verbs as 'eventive' and 'noneventive' respectively.
Let us turn first to the eventive psych-verbs. We might consider a possible
"standard" scenario here, in which one participant (we will assume transitive
verbs here) does something as a result of which the other participant undergoes
a change in psychological state - e.g. the scenario described by 'Fred amused
Mary', on a single-event reading where Fred carried out some action which
resulted in Mary becoming amused. Now, because the experiencer (Mary) undergoes a change of state and is causally affected by the other participant (both
of which are listed by Dowty's as proto-Patient properties), and because the
"amuser" (Fred) is volitionally involved, sentient, and causes a change of state
in the other participant (all of which are included in Dowty's pro to-Agent
properties), there is presumably a strong tendency for the experiencer to be
realised as the direct object, and the "amuser" to be the subject (according to
Dowty's ASP). And indeed, of course, this is exactly what we observe in the
case of 'amuse', which therefore falls into the class of psych-verbs with experiencer objects, or psych-EO verbs. There are many others in this class, and
if we examine the list of psych-EO verbs in Levin (1993) we find that without
exception these all fit the scenario described above - they describe events
where one participant does something that results in change of state in the
other, the experiencer participant. To take a couple more examples, 'frighten'
and 'entertain' express scenarios which are naturally construed as events. For
example, 'John frightened Mary' can be readily understood to mean that John
did something specific which caused Mary to feel fear (although there are
other readings - for example the one where John had a disposition to cause
fear in Mary).
Now let us turn to the class of non-eventive psych-verbs. We assume that
these do not follow the stereotypical agent/patient scenario described above.
Consider the verb 'love'. If John loved Mary, for example, it does not seem
possible to identify any "events" here. The crucial thing with these noneventive verbs is that the experiencer is not construed as undergoing a change
of state. Nor does the other participant, necessarily carry out any action which
causes a change of state in the experiencer (Mary does not necessarily "do anything" which results in John loving her). The experiencer, here, by virtue of
sentience, if not quite of volitional involvement, has a number of proto-Agent
properties and no proto-Patient properties. Dowty's ASP would therefore predict that the experiencer will be realised as subject, which is indeed the case
with 'love'.
So we see a pattern - non-eventive psych-verbs tend to have experiencer
participants which do not undergo a change of state - and hence tend to be low
in pro to-Patient properties and therefore more likely to be realised as subjects.
380
Sheila Glasbey
But, of course, non-eventive verbs (or, more strictly, those with no eventuality
argument) are precisely those which lack existential readings for bare plural
objects.
Eventive psych-verbs, on the other hand, have experiencer participants
which are relatively high in proto-Patient properties and therefore tend to be
realised as objects. So we have a putative explanation of our observation that
the psych-ES verbs lack existential bare plural object readings and the psychEO verbs have such readings. It is the presence (or otherwise) of an event argument that both determines the possibility of existential bare plurals and exerts a strong influence over whether the experiencer is realised as subject or
object.
Note that we say 'exerts a strong influence' here. Dowty's ASP works on
the basis of prototypical properties, and relies on a sense of "other things being
equal". It thus predicts tendencies rather than absolutes. Thus we perhaps
should not be surprised if we encounter a few exceptions to the generalisation
above. It should be said, however, that in our preliminary survey we have
come across very few such exceptions - that is, of either psych-ES verbs that
do allow existential bare plural objects, or psych-EO verbs that do not. As
mentioned above, however, we recommend a detailed and intensive survey of
these verbs.
One clear exception needs to be discussed - the case of 'admire' mentioned
earlier. Consider (33):
(33)
John admired Mary's paintings.
There are at least three readings here. The first is a single event reading, where
perhaps John went to an exhibition of Mary's paintings and stood gazing at
some of them in admiration for a while. Here, we clearly interpret the bare
plural as existential. There is a second existential reading, where John made a
habit, perhaps, last summer, of visiting galleries displaying Mary's work, and
each time gazed at her paintings. Notice that in this case he does not need to
have admired all or most of Mary's paintings, even among those he saw, so
once again we have an existential reading, although of a different kind. Thirdly
there is a clearly generic reading, where John has a dispositional attitude to
admire Mary's work.
Our point here is simply to note that 'admire' does allow readings where the
bare plural is interpreted as existential. Yet, notice that 'admire' is a psych-ES
verb - the experiencer, or "admirer" is realised as the subject. This potentially
upsets our account, which predicted that if a verb allows existential bare plurals then it must be an eventive verb, and if it is an eventive verb then it will
tend to have an experiencer object. Notice, too, that the existence of the three
readings described above, one of which is clearly identified as single-event
381
reading, offers strong confirmation that 'admire' is an eventive verb. Why,

then, should it have an experiencer subject?
We observed above that Dowty's ASP describes a tendency rather than an
absolute. Let us look more closely at the event of John admiring one or more
of Mary's paintings. Notice, first of all, that John is not easily seen as a causally affected participant. It is not that a painting "does something" that makes
John admire it. And indeed it may be a moot point as to whether we want to
think of John as undergoing a change of state by virtue of his admiring a painting (of course there are no strict real-world criteria that we can appeal to here such judgments reflect the way that language construes the world, not objective facts about the world). The point is that John, as an admirer, is not highlyendowed with proto-Patient properties. Nor is the painting, as the "admired"
participant, well-endowed with proto-Agent properties. In fact, given that John
is sentient and taking part in a volitional act, John is better-endowed with
proto-Agent properties and it thus seems entirely reasonable that he should be
the subject.
There is, of course, a verb 'impress' which has a roughly similar meaning to
'admire' except that the two participants are reversed:
(34)
Mary's paintings impressed John.
Here the sentient participant is the object, showing that in this case things can
go either way. Note, however, that the psychological meaning of 'impress'
may be seen as a metaphorical version of the literal 'impress' whereby a
physical mark is made on the participant. Assuming that the metaphorical version would maintain the same lexical form as the original literal version
(which seems a reasonable assumption to make), then we may have an explanation here of why 'impress' is a psych-EO verb. And indeed, a number of
psych-EO verbs may be regarded as metaphorically derived from literal verbs
where the event is much more stereotypical in terms of agent patient structure.
Further work is needed here.
Apart from 'admire', and possibly 'enjoy', 12 we have not found any verbs in
the psych-ES group which allow existential bare plural readings.
We believe, however, that a detailed study of the psych-verbs is needed to
test our analysis, and indeed plan to carry out such a study involving corpus
investigations.
A question that remains open is why this class of non-eventive psych verbs
should exist at all. According to our account, all other verbs have eventuality
arguments (though some of these are states, rather than events, as we saw ear-
12
We mentioned earlier that a sentence like 'Mary enjoyed the ice-cream' might well acquire its
eventive status from the implied underlying event of "eating", and recommend further investigation.
382
Sheila Glasbey
lier). Why should the psych-ES verbs, as opposed to the rest of the verbs, not
be associated with events?
We believe that a fruitful approach may be to consider the psych-ES verbs
as "inherent generics" - an idea that goes back to Chierchia (1995). We would
wish to modify Chierchia's proposal, however, since he suggested that i-level
verbs in general were inherent generics, and we wish to restrict this to the
psych-ES verbs. We do not have time to investigate this further here, and will
reserve a detailed account for another occasion.
12 Conclusion
We have presented an analysis of the English bare plural which explains,
among other things, why 'lawyers' in (1) has an existential reading, while
'lawyers' in (2) does not, except in special contexts. We explained why a context such as the 'hate-list' context for (2) allows an existential reading for
'lawyers'. Our account of bare plural objects is part of a broader analysis,
which relies, firstly, on a distinction between predicates which have eventuality arguments and those that do no not, and, secondly, on a distinction between
predicates which are relations (in the situation-theoretic sense) and those
which are types. We have shown how, by making both these distinctions, we
can explain the effects of context on the availability of existential readings of
bare plurals for both adjectival and verbal predicates. We consider it a strength
of our account that we can provide effectively the same explanation for both
types of predicate.
Having taken a closer look at the data on the availability of existential readings for bare plural objects, we identified those verbs which do not give existential readings (except in special contexts) as being the psych-ES verbs. We
offered a brief explanation of why things should pattern in this way, using
Dowty's argument selection principle (Dowty 1991) to explain why this class
of verbs, which lack event arguments, tend to have experiencer subjects.
In summary, our classification of predicates depends on a fundamental dichotomy between verbal predicates (which have eventuality arguments) and
adjectival predicates, which, in general, do not. We have effectively abandoned the traditional i-level/s-level distinction, although our account owes
many insights to previous analyses that used this distinction. We consider the
verbal/adjectival distinction to be the primary distinction among predicates,
while showing that a distinction between relations and types is also needed.
The distinction between eventualities (which may be arguments of verbs
and include both events and states) and the more general notion of situation
(which includes eventualities but is a broader notion, and may be made available by suitable contexts) is a further essential feature of our account. In analyses that make use of situations, eventualities and situations are often seen as
383
the same thing. We have shown that it is important to distinguish between

them.
Further work is needed to check that the generalisation identified (that it is
the psych-ES verbs which lack existential readings for bare plural objects) is
indeed correct. We recommend a detailed corpus study involving the psych-ES
verbs, the psych-EO verbs and indeed as many verbs as possible from other
classes. Clearly this is a substantial endeavour but one which we believe will
repay the effort, especially as it appears that a number of previous accounts of
bare plurals have gone astray by failing to pay close attention to exactly which
predicates do and do not allow existential bare plurals.
We merely hinted in the final section at a possible answer to the question of
why it should be the psych-ES class of verbs that lack event arguments. A detailed account along the lines suggested (or indeed some other lines) remains
to be worked out and tested.
Finally, while we believe the notion of "existential inference" to be a powerful and useful one, inspired by a number of earlier accounts of bare plurals
(such as those that rely on a notion of located vs. unlocated predicates, and
indeed on Carlson's original notion of "stage"), we acknowledge that it remains to be fully worked out and understood. Further investigation is clearly
needed here.
References
Bach, E. (1986): The algebra of events. Linguistics and Philosophy 9(1), pp 5-16.
Barwise, J. & J. Perry (1983): Situations and Attitudes. Bradford Books.
Barwise, J. & R. Cooper (1991): Simple Situation Theory and its Graphical Representation. Indiana University Logic Group Reprint No. IULG 91-8.
Barwise, J. & R. Cooper (1993): Extended Kamp Notation: a graphical notation for
situation theory. In P. Aczel, D. Israel, Y. Katagiri and S. Peters, eds., Situation
Theory and its Applications, Vol. 3, pp 29-53. Stanford, Ca.: CSLI.
Carlson, G. (1977): Reference to Kinds in English. PhD dissertation, University of
Massachusetts, Amherst.
Chierchia, G. (1995): Dynamics of Meaning. Chicago: The University Press of Chicago
Cohen, A. and N. Erteschik-Shir (2002): Topic, focus, and the interpretation of bare
plurals. Natural Language Semantics, 10:2, pp 125-126. Kluwer Academic Publishers.
Cooper, R. (1985): Aspectual classes in situation semantics. Report CSLI-84-14C,
CSLI, Stanford, Ca.
Cooper, R. (1986): Tense and discourse location in situation semantics. Linguistics and
Philosophy, 9(1), pp 17-36.
384
Sheila Glasbey
Cooper, R. (1992): A working person's guide to situation theory. In Topics in Semantic

Interpretation, eds. Steffen Leo Hansen & Finn Soerensen, Samfundslitteratur, Frederiksberg, Denmark.
Cooper, R. (1993): Generalized quantifiers and resource situations. In P. Aczel, D. Israel, Y. Katagiri and S. Peters, eds., Situation Theory and its Applications, Vol. 3.
Stanford, Ca.: CSLI.
Davidson, D. (1967): The logical form of action sentences. In N. Rescher, ed., The
Logic of Decision and Action. Pittsburgh: University of Pittsburgh Press.
Dobrovie-Sorin, C. (1998): Types of predicate and the representation of existential
readings. In: A. Lawson (ed.), Proceedings of SALT 7. Cornell University Press,
Ithaca, N.Y.
Dowty, D. (1991): Thematic proto-roles and argument selection. Language 67(3), pp
547-619.
Erteschik-Shir, N. (1997): The Dynamics of Focus Structure. Cambridge: Cambridge
University Press.
Filip, H. (2000): Psychological predicates and the syntax-semantics interface. In: Goldberg, A.E. (ed.), Conceptual Structure, Discourse and Language. Stanford: CSLI.
Available at: http://semanticsarchive.net/Archive/TI2YmU30/filip.psychverbs.pdf
(consulted 27 August 2004).
Glasbey, S. (1994): Event Structure in Natural Language Discourse. Unpublished doctoral dissertation, Centre for Cognitive Science, University of Edinburgh.
Glasbey, S. (1997): I-level predicates that allow existential readings for bare plurals. In:
Aaron Lawson, ed., Proceedings of SALT VII : Semantics and Linguistic Theory
pp 169-179, CLC Publications, Cornell University.
Glasbey, S. (1998): A situation-theoretic interpretation of bare plurals. In: J. Ginzburg,
Z. Khasidashvili, C. Vogel, J-J. Levy, and E. Vallduvi, eds, The Tbilisi Symposium
on Logic, Language and Computation: Selected Papers, pp. 35-54. CSLI Publications, Stanford, Ca.
Glasbey, S. (1999): Bare plurals, situations and discourse context. In: L. Moss, J.
Ginzburg and M. de Rijke, Logic, Language and Computation, Volume 2, pp. 85105. CSLI Publications, Stanford, Ca.
Greenberg. Y. (1994): Hebrew nominal sentences and the stage/individual level distinction. Master's thesis, Bar Ilan University, Israel.
Kamp, H. & U. Reyle (1993): From Discourse to Logic: Introduction to Modeltheoretic
Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Dordrecht: Kluwer.
Kratzer, . (1995): Stage-level and individual-level predicates. In G. Carlson & F.
Pelletier (eds.), The Generic Book. Chicago: University of Chicago Press.
Levin, B. (1993): English Verb Classes and Alternations. University of Chicago Press,
Chicago.
385
McNally, L. (1998): Stativity and theticity. In S. Rothstein (ed.), Events and Grammar.
pp 293-307. Kluwer Academic Publishers, Dordrecht.
Smith, C. (1991): The Parameter of Aspect. Dordrecht: Kluwer.
Van der Sandt, R. (1992): Presupposition projection as anaphora resolution. Journal of
Semantics 9, pp 333-377.
Van Geenhoven, V. (1996): Semantic Incorporation and Indefinite Descriptions, PhD
dissertation, University of Tuebingen. Published 1998 by CSLI, Stanford.
SECTION V
Event Structure and Temporal Location

(Potsdam and Berlin)
Tense and adverbial quantification*

1 Novel data
Consider (la) and (lb) below: In (la), a Q(uantificational)-adverb is combined
with an indefinite DP, while (lb) contains a quantificational determiner, the
quantificational force of which corresponds to that of the Q-adverb in (la). It is
often assumed (cf. Lewis 1975, Kamp 1981, Heim 1982, Diesing 1990, von
Fintel 1994, Chierchia 1995a, Kratzer 1995, Herburger 2000 and many others)
that such sentences essentially receive the same interpretation. This is generally referred to as the quantificational variability effect (QVE) (Berman 1991).
(1)
(a)
(b)
A police car is usually blue,

Most police cars are blue.
But whereas (la) indeed seems to have the same meaning as (lb) at an intuitive level, there are cases where the correspondence exemplified by this pair of
sentences breaks down. Consider the contrast between (2a) and (2b) below:
(2)
(a)
(b)
11
A car that was bought in the eighties is usually blue,

Most cars that were bought in the eighties are blue.
While (2b) is perfectly acceptable, most speakers perceive (2a) to be very

strange. It seems that the indefinite DP can only be understood as having scope
This paper was presented at the workshop on event structures in Leipzig (March 2004) and at
the Milan meeting on covert variables at LF (June 2004). We would like to thank the participants of both events for comments and suggestions. Furthermore, we would like to thank Ariel
Cohen, Christian Ebert, Alex Grosu, Andreas Haida, Shinichiro Ishihara, Manfred Krifka,
Marko Malink, Philippe Schlenker, Roger Schwarzschild, and Malte Zimmermann for discussions and valuable comments. In particular, we thank Philippe Schlenker for pointing out the
empirical facts discussed in section 4.3, Roger Schwarzschild for drawing our attention to the
work of Renate Musan, and Christian Ebert for helping us to spell out the formal details of our
approach. Furthermore, we also thank the anonymous reviewers for helpful comments on an
earlier draft of this paper.
This work was funded by the Deutsche Forschungsgemeinschaft as part of the Sonderforschungsbereich 632 (Information Structure).
390
over the Q-adverb. But such an interpretation is very strange, as the property of
being of a particular color is stable for a given car under normal circumstances,
i.e. the predicate to be blue is usually interpreted as an individual level predicate with respect to cars.1 This raises the question as to why a reading, in
which the Q-adverb has scope over the indefinite DP, is blocked in the case of
(2a), while it is easily available in the case of (la).
Interestingly, (3a) is perceived as much better than (2a), while (3b) is
deemed to be just as good as (2b), though different in interpretation.2
(3)
(a)
A car that was bought in the eighties was usually blue,
(b)
Most cars that were bought in the eighties were blue.
The improved status of (3a) seems to be due to the fact that in contrast to (2a),
the indefinite DP in (3a) can be interpreted as falling within the scope of the Qadverb. The same holds for (4), where both the relative clause verb and the
matrix verb are marked for present tense.
(4)
A car that pleases Peter is usually blue.
Our data raise the question of why adverbially quantified sentences that contain indefinite DPs modified by relative clauses, in the absence of intervening
factors (cf. sections 4.3, 4.4 and section 5), only show QVEs if the tense of the
relative clause verb agrees with the tense of the matrix verb?
2 Existing analyses
In this section we will discuss three different accounts of QVEs and show that
none of them is able to account for the contrast in acceptability between (2a)
and (2b). Due to limitations of space, we will have to gloss over many details.
2.1 Q-adverbs as unselective binders
The theories of Heim (1982) (based on Lewis 1975), Diesing (1990), and
Kratzer (1995) share the following assumptions:
Indefinites provide a restricted variable that must be bound by an adverbial quantifier.
Of course, cars can change their color when they are repainted, which means that strictly
speaking, blue is not a real individual level predicate in this context. Yet, we will ignore this
complication throughout this paper.
We will discuss the interpretative difference in section 4.2.
391
Adverbial quantifiers are unselective binders that bind every free variable in their scope, i.e. individual as well as situation/event variables.
If a sentence does not contain an overt Q-adverb, the restricted variable
introduced by an indefinite is bound by a covertly inserted quantifier
with either existential or generic force.
Furthermore, according to Kratzer (1995), stage level predicates (which ascribe transitory properties to their individual arguments) come with a spatiotemporal argument, whereas individual level predicates (which ascribe stable
properties that typically last a whole lifetime to their individual arguments) do
not.
Despite its strangeness, (2a) (repeated below as (5a)) acquires a perfectly
coherent interpretation according to these approaches, as there is a free variable (provided by a car) which could be bound by the adverbial quantifier.
(5)
(a)
" A car that was bought in the eighties is usually blue,
(b)
MOST, [car(x) bought in the 80s (x)] [blue(x)].
This is exactly the same interpretation that is assigned to (2b) (repeated below
as (6a)):
(6)
(a)
Most cars that were bought in the eighties are blue,
(b)
MOST, [car(x) bought in the 80s (x)] [blue(x)].
This means that the clear contrast in acceptability between the two sentences
cannot adequately be accounted for by these theories.
2.2 Q-adverbs as topic-sensitive binders
Chierchia (1995a) differs from the above view in two respects: Firstly, indefinites are interpreted as regular existentially quantified DPs. When they are
topical (which is signalled by de-accentuation), they are turned into predicative
expressions via an operation called existential disclosure (Dekker 1993) and
can later be bound by a c-commanding adverbial quantifier. And secondly,
individual level predicates also come with a spatio-temporal argument, but in
contrast to the argument introduced by stage level predicates, this needs to be
bound by the generic quantifier. So, if the indefinite is de-accented, (2a) (repeated below as (7a)) is interpreted as in (7b) below.
(7)
(a)
"A car that was bought in the eighties is usually blue.
(b)
MOST, [car(x) bought in the 80s(x)] [GEN, [in (x, s)] [blue(x, s)]]
392
But (7b) is of course equivalent to (5b). So again, no reason is offered why

(2a) should be unacceptable.
2.3 Situation and event semantic approaches
The theories of Berman (1987), de Swart (1993), von Fintel (1994) and Herburger (2000) share the following assumptions:
Q-adverbs only quantify over (sets of) situations/eventualities. 3

Restrictor and nucleus are determined on the basis of information structure: The denotation of the whole clause minus the Q-adverb is mapped
onto the nuclear scope, while the denotation of the non-focal/topical
constituents is mapped onto the restriction.
Indefinites are interpreted as existentially quantified DPs.
There is no difference between stage level and individual level predicates with respect to the introduction of situation/eventuality variables.
QVEs then come about in the following way: If an indefinite is de-accented, its
denotation is mapped onto the restriction of a Q-adverb. Furthermore, the
value assigned to the individual variable bound by the existential quantifier
may vary with the value assigned to the situation/event variable bound by the
Q-adverb. The combination of these factors enables sentences like (la) (repeated below as (8a)) to be interpreted as in (8b) or - equivalently - (8c). 4 5
(8)
(a)
A police car is usually blue.
(b)
(c)
MOSTe [3x. police car () Arg (e, )] [3. police car () Arg (e, ) blue(e)]
MOST [3x. police car () Arg (e, )] [blue(e)]
The problem with these theories, however, is that they also predict sentences
like (2a) (repeated below as (9a)) to have well formed semantic representations
like the one given in (9b) below:
The term eventuality is meant to encompass dynamic as well as static "events".

In the following, we will only give simplified representations like (8c), where material that
already occurs in the restrictor is not repeated in the nucleus. This is justified on the assumption that theta-roles are assigned exhaustively (cf. Herburger 2000), i. e. that each eventuality
contains only one agent, theme, etc. This last assumption furthermore avoids the requantification problem (von Fintel 1994, Rooth 1995), which arises in a situation/event semantics approach as a result of the fact that topical indefinites are interpreted twice: Once in the restrictor
and once in the nucleus (cf. Krifka 2001, who offers a different solution).
In the following semantic representations, we will use the term Arg(ument) for the external
arguments of stative predicates such as blue, intelligent etc., because neither of the existing
theta roles seems to be applicable in these cases. Nevertheless, we assume that these roles are
also assigned exhaustively (s. footnote 4).
(9)
(a)
" A car that was bought in the 80s is usually blue.
(b)
MOSTe [Bx. Arg(e, ) car(x) [Be', buy (e') theme(e', ) in 80s(e')]
393
[blue(e)]
This means that without further assumptions, the existing situation/event semantics accounts of QVEs cannot explain the observed contrasts either.
3 Conceivable solution strategies

As has been shown in the preceding chapter, none of the existing theories can
explain the difference between (2a) and (2b). Before we will present our own
account, we wish to mention briefly some conceivable solution strategies that
could come to mind, and argue why they cannot be maintained.
3.1 Natural classes?
One could speculate that, for some reason, QVEs only arise with indefinites
that pick out individuals from a well defined class (cf. the work on generics
and natural classes in Krifka et al. 1995 and the work of Cohen 2001, Greenberg 2002 and Greenberg 2003 6 on the different behaviour of singular indefinites and bare plurals in generic sentences, i.e. sentences that do not contain an
overt Q-adverb). But the fact that sentence (10) is perfectly acceptable shows
that this cannot be the correct generalization for the cases discussed here.
(10)
A French linguist with green hair and six toes is usually intelligent.
It will be hard to argue that the class of French linguists with green hair and
six toes is a natural one or even that this should be a more natural class than the
one of cars that were bought in the eighties.
3.2 Specificity?
Alternatively, it could be argued that for some unknown, yet compelling reason, temporally fixed indefinites have to be interpreted specifically. But this
assumption is not borne out either, as the generalization does not hold for nonQV environments:
(11)
It is possible that a car that was bought in the eighties had an accident today.
(12)
Every customer recognized a car that was on exhibition in this shop window yesterday.
W e would like to thank Angelika Kratzer for drawing our attention to the work of Yael Greenberg.
394
In (11), the speaker does not need to have a particular car in mind, and in (12)
the cars may vary with the customers.
4 A pragmatic account
We follow von Fintel (1994) and Herburger (2000) in the assumption that
D(eterminer)-quantifiers take sets of individuals as arguments, while
A(dverbial)-quantifiers take sets of eventualities. The arguments of Dquantifiers are determined grammatically, while the restriction of A-quantifiers
must be determined solely on the basis of information structure (or contextual
information).
We also assume that every quantification entails covert domain restriction
(cf. von Fintel 1994, Stanley 2000 and Marti 2003). For D-quantifiers this
means that the restrictor set has to be intersected with the set characterized by
a covert predicate that is determined by the context. In a context like the one
given in (13a), a sentence such as (13b) would not be about all the apples in
the world, but about all the apples that have been introduced in the previous
sentence, i.e. all the apples that Peter bought the day before:
(13)
(a)
(b)
Yesterday, Peter bought apples,

Every apple tasted awful.
Analogously, domain restriction for events entails, among other things, locating the respective events in time (cf. Partee 1973, Lenci and Bertinetto 1999).
In a context such as (14a), the event of drinking beer in (14b) is automatically
interpreted as having taken place at some interval that lies within the running
time of the eventuality referred to in (14a), i.e. the beer drinking is understood
to have occurred at Mary's party (cf. Partee 1973):
(14)
(a)
Yesterday, Peter had a good time at Mary's party,
(b)
He drank a lot of beer.
We thus claim that the unacceptability of (2a) can be explained by the fact that
there is a conflict between the tense information given by the relative clause
verb and the tense information given by the matrix verb.
4.1 Technical preliminaries
We will first explain our technical apparatus in a discussion of example (2b)
(repeated below as (15a)). We will show that our approach actually predicts
that this is a felicitous sentence for which there exists a sensible interpretation.
Due to the presence of the D-quantifier most, the sentence is interpreted as
quantifying over individuals x. Every quantifier is connected to a domain re-
395
striction, including the quantifier most in our example, which introduces the
conjunct C{x)J Note that every verbal predicate introduces an eventuality
variable which in the absence of an overt Q-adverb is bound by a covert existential quantifier (or by a covert generic quantifier if the respective sentence
requires a generic interpretation). Of course, each covert quantifier is connected to a domain restriction.
(15)
(a)
(b)
Most cars that were bought in the 80s are blue.

MOST, [ car(x) [Be', buy (e') Theme(e\ ) past (e')
in 80s(e') C'(e')] C " ( X ) ]
[3e. Arg(e, ) pres(e) blue(e) C(e)]
In case of quantification over individuals, the restriction, among other things,

serves to locate an eventuality e within an interval ie. This means that C(e)
takes on the form e @ ie.
(16)
(a)
(b)
MOST, [ car(x) [3e\ buy (e') Theme(e', ) past (e')
in 80s(e')
e' @ ie]
C"(X)]
[. Arg(e, ) pres(e) blue(e) e @ ie]
Temporal location of an event within an interval is defined as follows:

(17)
e @ ie := "C(e) i e ,
where t(e) denotes the running time of e.
In words, e @ means that e, in the case of verbs denoting dynamic eventualities (i. e. achievements, accomplishments and activities, cf. Vendler (1957)),
takes place at some time during the interval ie or, in the case of a stative verb/a
property, exhausts ie.8
We assume the following (simplified) semantics for tense information relative to the speech time t0\
C is to be understood as a variable ranging over contextually inferred predicates. Note that in
contrast to von Fintel (1994) and Marti (2003), we assume that this domain restriction is added
at the latest possible position, because it is determined by overt information that has been introduced previously.
Following Bach (1986) (among many others, s. Rothstein 2003 and references therein for
recent discussion), we assume that statives (as well as activities) are homogenous with respect
to their internal structure. In the case of stative verbs such as to be French, the state of being
French for a given individual denotes an infinite set of being French eventualities, the largest
of which is the maximal eventuality in which the property of being French holds for the individual under consideration. Under this view, it follows directly that e @ ie picks out only those
subeventualities of the state under discussion that lie in the interval ie. Analogous to activities,
only the maximal eventuality (i.e. the one exhausting the whole interval) is taken into account
when computing the truth conditions of the sentence.
396
(18)
(a)
pres(e) := t o e ()
(b)
past(e) := x(e) < to
4.2 The interval resolution strategy

The free interval variables i in (16b) need to be resolved, requiring the use of
both overtly given and contextually inferable information.
More specifically, we assume that there is a pragmatic strategy that determines how available information is used in order to locate eventualities temporally. W e call this the interval resolution strategy. It works according to the
following principles:
(19)
1.
2.
Take overt information.

If not available: Take contextual information from the same domain (restrictor vs.
nucleus), i. e. the running time of another salient eventuality.
3.
If not available: Take contextual information from the other domain, or take the
default time interval i,wrid, which denotes the whole time axis.
The principle behind this strategy is as follows: If there is overt information

concerning the time when an event e takes place, this information must be used
in order to instantiate the interval ie. This is the case in example (15a), where
the event of Peter's having a good time at Mary's party must be located during
the interval denoted by yesterday. In (14b) on the other hand, there is no overt
material that denotes an interval in which the beer drinking event has to be
located. In this example, contextual information must be taken into account,
corresponding to point (2.) of the interval resolution strategy. According to the
strategy, the event of Peter's drinking beer has to be located at some contextually given time interval. In the example, this is the running time of another
contextually given salient event, i.e. the time during which Peter was at Mary's
party. Local proximity is an important concept in this analysis; contextual
information mentioned immediately before the event to be located is more
appropriate as a restriction to the respective event than material that has been
presented much earlier. This is reflected in the interval resolution strategy,
according to which local information (point 2.) is required to be preferred over
non-local information (point 3.).
4.2.1 Quantification over individuals

In the case of (16), repeated here as (20), there are two intervals which are to
be resolved: ie and ie.
(20)
(a)
(b)
397

MOSTx [ car(x) [Be'. buy (e' ) Theme(e', ) past (e' )
in 80s(e')
e' @ i]
C"(X)]
[3e. Arg(e, ) pres(e) blue(e) e @ ie]
Overt information is provided in connection with the relative clause events e',
which must be located in the interval ie: the interval denoted by the PP the
eighties. Therefore, ie needs to be instantiated with this interval. For /,,, on the
other hand, there is neither a constituent that denotes an interval nor any other
indirect interval information given within the same domain (which is the nucleus). Point (3.) of the interval resolution strategy given in (19) therefore
gains relevance. According to this principle, the first option to resolve ie would
be to instantiate it with the running time of the relative clause events (this
counts as information from the other domain, i. e. from the restrictor). 9 This
would result in the following representation:
(21)
(a)
(b)
MOSTx [ car(x) [Be'. buy (e') Theme(e', ) past (e')
in BOs(e')
e' @ 80s!
C"(X)]
[3e. Arg(e, ) pres(e) blue(e) e @ T(e')l
The events e would then be interpreted as being located within the same interval as the events e' - i.e. the eighties. But this would directly clash with the
semantics of present tense:
(22)
(a)
(b)
#MOSTx [ car(x) [Be', buy (e') Theme(e\ ) (e') < to

in 80s(e') e' @ 80s! C " ( X ) ]
[Be. Arg(e, ) FOET(e) blue(e) I(E) er T(e')1
As the speech time t0 is not contained within the eighties, the tense specification within the nucleus is contradictory:
to c x(e) c x(e') : 80s, contradicting 80s < to.
The other option specified in point (3.) of the interval resolution strategy must
therefore be taken: ie has to be instantiated with the whole time axis iworu The
resulting representation is given in (23):
We assume here and in all the formulas to follow that the variable e ' mentioned in the tense
specification e @ t(e') is dynamically bound by the existential quantifier that binds the variable introduced by the relative clause verb (cf. Staudacher 1987, Groenedijk and Stokhof 1991
and Chierchia 1995b for details with respect to the principles of dynamic binding). Note that
this causes the running times of the matrix eventualities to vary along with the running times of
the relative clause eventualities.
398
(23)
(a)
(b)

MOST* [ car(x) [3e\ buy (e') Theme(e', ) (e') < to
in 80s(e') e' @ 80s] C " ( X ) ]
[Be. Arg(e, ) TO e () blue(e) e @ i w o r i d ]
Let us now consider (3b) (repeated below as (24a)), the variant of (2b) in
which the matrix predicate is in the past tense. In this case, there is no difficulty in taking the first option specified in point (3.) of the interval resolution
strategy. The running times of the matrix eventualities e can be equated with
the running times of the relative clause eventualities e', as there is no tense
clash as a result of the past tense marking of the matrix verb:
(24)
(a)
(b)

MOST, [ car(x) [Be'. buy (e' ) Theme(e', ) (e' ) < TO
in 80s(e') e' @ 80sl C " ( X ) ]
[Be. Arg(e, ) t(e) < tn blue(E) e @ t(e')l
The meaning is thus: Most cars bought in the eighties were blue when they
were bought. Note that we do not get to know whether the respective cars are
still blue today. This is simply left open.
It is also possible to take the second option specified in point (3.) above, and
instantiate the matrix interval with the whole time axis. This leads to a different reading of the sentence, which indeed seems to be available:
(25)
(a)
(b)

MOST* [ car(x) [Be'. buy (e' ) Theme(e', ) (e' ) < to
in 80s(e') e'@ 80s] C " ( X ) ]
[Be. Arg(e, ) past(e) blue(e) e @ i w o r i d l
The past tense demands zfej, i. e. the time of being blue, to end before the
speech time t0\
(26)
(a)
(b)

MOSTx [ car(x) [Be'. buy (e' ) Theme(e', ) (e' ) < to
in 80s(e') e'@ 80s] C " ( X ) ]
[Be. Arg(e, ) t(e) < to blue(e) e @ j .worldJ.
This means that the eventuality of being blue has to have ended before the
speech time. Under the assumption that blue is regarded as an individual level
predicate with respect to cars, this triggers the hearer's expectation that the
respective cars no longer exist.
We take this to be a consequence of our analysis of individual level predicates. On the one hand, only the maximal eventualities of cars being blue that
lie within the respective interval (which in this case is iwori) may be picked
399
out. On the other hand, the past tense marking of the matrix verb requires those
eventualities to end before the speech time. Both requirements are only met if
the cars quantified over no longer exist. There would otherwise be a greater
eventuality of those cars being blue that lies within the interval iworU, i.e. one
comprising the whole time of the cars' existence, which would then extend
beyond the speech time.
This means that using the past tense, one does not provide as much information as possible with respect to the chosen interval (which is iworid) if the cars
quantified over still exist. If, on the other hand, those cars no longer exist, a
past tense marking provides the greatest eventuality of the respective cars
being blue that lie within this interval. The hearer therefore automatically assumes that the cars quantified over no longer exist. 10
This effect is reminiscent of the facts discussed by Kratzer (1995) and
Musan (1997) as life time effects. Consider the sentence below:
(27)
Gregory was from America.
If (27) is uttered out of the blue, it implicates that Gregory is dead at the
speech time. If, on the other hand, the sentence is embedded in a context like
the one given in (28a), no such implication arises. Phrased in our terms, this
difference could be explained as follows (cf. Musan 1997 for a very similar
solution): In (27), the eventuality of being from America is located within ,vorU.
In (28b), on the other hand, it is most likely understood as coinciding with the
running time of the event in (28a).
(28)
(a)
Yesterday, I met Gregory and Paul.
(b)
Gregory was from America, (while Paul was from Australia).
To summarize the results of this section, we claim that (2b) is acceptable for
the following reasons:
10
D-quantification does not bind eventualities.

The predicate to be blue in the nuclear scope introduces an existentially
bound eventuality variable e.
This eventuality can be located in an interval that is independent of the
one given in the relative clause.
As has been pointed out to us by Manfred Krifka, there is another way to resolve ie in the case
under discussion. If the sentence is embedded in a certain context such as the one given in (a)
below, ie could also be resolved to the time specified by this context,
(i) (a) There was a second hand car market in this town in 1995.
(b) Most cars that were bought in the eighties were blue.
In this case, ie can be set to the time when the second hand car market mentioned took place.
This is predicted by our approach because according to point (3.) of the interval resolution
strategy, non-local contextual information can be taken into account.
400
There is no interval information in the nuclear scope.

The interval ie can be set to the default interval iworid4.2.2 Quantification over eventualities
In the case of (2a) (repeated below as (29)), matters are different:

(29)
" A car that was bought in the eighties is usually blue.
In adverbial quantification, it is not the syntax that determines restrictor and

nucleus, but the information structure (or contextual information). Nonfocal/topical material is mapped onto the restrictor, while the focal material is
mapped onto the nuclear scope (cf. among others Chierchia 1995a, Krifka
1995, Partee 1995, Rooth 1995, Herburger 2000). More specifically, we assume the mapping algorithm of Herburger (2000), already discussed in section
2.3.
In the case of (2a), at least if it is uttered out of context, the most natural assumption is that the matrix predicate blue is focused. It is therefore mapped
onto the nuclear scope. Furthermore - and this is crucial for our account - the
eventuality variable introduced by blue is bound by the adverbial quantifier
usually in the restrictor as well as in the nuclear scope. This has the consequence that in (2a) the eventuality variable introduced by the matrix verb ends
up in the same domain as the eventuality variable introduced by the relative
clause verb - i.e. in the restrictor of the adverbial quantifier usually.
This contrasts with the situation in (2b), in which the two variables are interpreted in different domains. The variable introduced by the verb of the relative clause is interpreted in the restrictor of the determiner quantifier most,
while the variable introduced by the matrix verb ends up in the nuclear scope
of the quantifier. This, together with the fact that the matrix eventuality variable also needs to be restricted by a time interval, leads to the interval resolution strategy working differently in the two cases. Now consider the semantic
representation of (2a) (repeated below as (30)) in detail:
(30)
(a)
,?
(b)
MOST e [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', x)
A car that was bought in the eighties is usually blue.
past(e') in 80s(e') C'(e')] C ' ( x ) C(e)] [pres(e) blue(e)]
As mentioned above, the domain restriction C(e) for the adverbial quantifier
usually must include the constraint e @ ie, where ie is to be resolved. As there
is no overt information with respect to ie in the matrix clause, the only available interval information originates from the information concerning the events
e' in the relative clause. This is information originating from the same domain,
401
i.e. from the restrictor, and according to the interval resolution strategy, ie must
be equated to the interval denoted by the running time of the events e': 11
(31)
(a)
?? A
(b)
MOSTe [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', )
car that was bought in the eighties is usually blue.
past(e') in 80s(e') e' @ 80s] C " ( X ) e @ (')] [pres(e) blue(e)]
As the events e take place in the eighties and the events e are located during
the running times of the events e', only events located in the eighties, i.e. before the speech time to, will be considered in the restrictor, whereas the nucleus
requires e to include the speech time:
(32)
(a)
(b)
" A car that was bought in the eighties is usually blue.

MOSTe [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', )
X(e') < to in 80s(e') e' @ 80sl C " ( X ) e @ t f e ' l ito cz T(e) blue(e)]
This by necessity yields an empty intersection of restrictor and nucleus and

thus accounts for the strangeness of (2a).
As this strangeness is not due to a grammatical, but rather a pragmatic principle, it is to be expected that the unacceptability is not absolute. For some
speakers it might be possible to construct contexts in which the sentence is
fine. Still, (2a) is much less natural than (2b), where it is not necessary for the
hearer to construct a matching context in order to be able to interpret the sentence adequately.
Obviously, if the information in the matrix clause is non-contradictory in
this respect, one expects the utterance to be felicitous, which is in fact borne
out. This can be seen in (3a), repeated below as (33a):
(33)
(a)
A car that was bought in the eighties was usually blue.
(b)
MOSTu [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', )

-c(e') < TO in 80s(e') e' @ 80s! E ' '() e @ t ( e ' l t(e) < TO blue(e)]
In this case, instantiating ie with the running time of the respective eventuality
e' (which must be located in the eighties) does not lead to a contradiction, as
the past tense information in the nucleus requires the events quantified over to
be located at an interval that is prior to the speech time.
To summarize the results of this section, our approach predicts (2a) to be
unacceptable for the following reasons:
11
Compare this to our example (14b), in which, in the context given, the event of Peter's beer
drinking must be interpreted as being located in the interval denoted by the running time of the
immediately preceding sentence - due to the local proximity of the two sentences. Obviously,
local proximity also plays a role in the example under discussion, as the running time of the
relative clause is salient local information.
402
A-quantification binds the eventuality variable e in the restrictor and in

the nuclear scope.
Domain restriction forces e to be located in an interval ie.
Due to contextual information in the restrictor, ie has to be resolved to
the running time of e '.
This clashes with the present tense information in the nuclear scope.
The intersection of restrictor and nucleus is inevitably empty.
4.3 Explicit interval setting
Interestingly, (34a) is acceptable in spite of the fact that it is structurally almost

identical to (2a). The matrix verb is marked for present tense, while the relative
clause verb is marked for past tense. As can be seen by comparing (34a) to the
minimally distinct example (34b), what makes the difference is the presence of
the adverb nowadays in the matrix clause, not the nature of the matrix verb
itself.
(34)
(a)
A car that was bought in the eighties is usually rusty nowadays,
(b)
11
A car that was bought in the eighties is usually rusty.
The initial representation of (34a) is the following:

(35)
(a)
A car that was bought in the eighties is usually rusty nowadays.
(b)
MOSTe [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', ) past (e')
in 80s(e') e' @ 80s] C"(X) nowadays (e) e @ i j [pres(e) rusty(e)]
Let us assume for the sake of concreteness that nowadays introduces an interval of contextually specified size that is constrained to include the speech time,
and locates the eventuality introduced by the verb it modifies within this interval. 12 Furthermore, it is intuitively clear that this interval does not extend far
enough into the past to include the interval introduced by the internal adverb of
the relative clause the eighties, i.e. the local context seems to influence the
choice of the interval denoted by nowadays.
As the adverb nowadays counts as overt information, (34a) is predicted to
be acceptable in accordance with the interval resolution strategy. The interval
ie must not be set to the duration of the respective eventuality denoted by the
12
As has been pointed out to us by M a n f r e d Krifka and Alex Grosu (p. c.), it is not clear why
nowadays introduces such an interval whereas the present tense marking of the matrix verb
does not, and therefore does not lead to an interval resetting. O n e obvious solution would be to
assume that this is d u e to the fact that the denotation of nowadays is (most plausibly) mapped
onto the restrictor, while the denotation of the matrix verb (including the tense specification) is
mapped onto the nucleus. Point (1.) of the interval resolution strategy would then have to be
changed accordingly: T a k e overt information f r o m the same domain.
403
relative clause verb, but - according to point (1.) of the interval resolution
strategy - needs to be set to the interval denoted by nowadays. In this case,
there is no clash between the temporal information in the restrictor, and the
temporal information that the present tense marking of the matrix verb contributes to the nuclear scope. The sentence is therefore felicitous:
(36)
(a)
A car that was bought in the eighties is usually rusty nowadays.
(b)
in 80s(e') e' @ 80s] C"(X) nowadays (e) e @ nowadays 1
fpres(e) rusty(e)]
An obvious question is whether this also works with our initial example, i. e.
whether the addition of the adverb nowadays also improves the status of (2a)
(repeated below as (37a)). This seems to be the case, as is evidenced by the
fact that (37b) is at least more acceptable than (37a):
(37)
(a)
(b)
A car that was bought in the eighties is usually blue.

A car that was bought in the eighties is usually blue nowadays.
Nevertheless, in spite of its improved status (in comparison to (37a)), (37b) is

still strange. This seems to be due to the fact that (37b) strongly implicates that
the respective cars were not already blue at the time when they were bought, i.
e. blue can no longer be interpreted as a true individual level predicate with
respect to cars.
Note that the same implicature is triggered in the case of (36a) - the only
difference being that it is quite natural to assume that cars have not already
been rusty at the time when they were bought. One possible explanation for
this implicature is that adding the adverb nowadays causes interval resetting in
the restrictor (otherwise ie would simply be instantiated with zfe'|), with the
hearer assuming that there is a reason why it has been added in the first place.
4.4 Interval resetting induced by presuppositions
Consider (38) below, which is just as acceptable as (34a) - in spite of differing
tenses in matrix and relative clause. In this case, the presence of the adverbial
still in the matrix clause seems to be the relevant factor.
(38)
A car that was bought in the 80s is usually still roadworthy.
We assume that still is similar to nowadays in that it introduces an interval t in

which the matrix eventuality e has to be located. Besides that, it does not add
much to the semantic content:
404
(39)
still (P, e) = P(e) e @ t,

where e is the eventuality variable introduced by the respective verb (here, be
roadworthy),
while is the (denotation of the) intermediate projection of this verb (see below).
As shown in (39), we assume that still takes two arguments: First, it takes an
eventuality predicate P, where is the (denotation of the) intermediate projection of the verb that results from applying the denotation of this verb to its
individual argument(s). 13 Therefore, denotes a function from eventualities to
truth values.
The second argument is the eventuality variable introduced by the respective verb. In line with Kratzer (1995), we assume that the eventuality arguments of verbs are represented directly in the syntax: They are generated in the
outermost specifier position of the verbal projection. Under the assumption
that still is adjoined directly below the eventuality argument, it first combines
with the denotation of the intermediate verbal projection below it (i. e. P), and
in the next step combines with the respective eventuality variable.
It is crucial for our purposes that apart from its rather trivial assertive content, still also triggers a presupposition (cf. Lbner 1999, Smessaert and ter
Meulen 2004, among others; see also Zybatow and Malink 2003), which is
given below:
(40)
3t'. salient(t') a t'< t Vt".[t' < t " < t - 3 e ' . e'@ t " AP(e')],
where t is the time interval introduced by the lexical content of still (cf. (39)).
For this presupposition to be satisfied in the case of (38), there has to be a

salient time interval t' which is located before t, t being the interval for which
the be roadworthy eventuality e holds with respect to the cars introduced by
the indefinite. Furthermore, this property must persist during the period until t
starts. In this example, the explicitly mentioned interval denoted by the adverb
the eighties can serve to satisfy the presupposition locally. It is plausible to
assume that the respective cars already had the property of being roadworthy at
the time when they were bought.
As discussed above, the overtly introduced interval t (originating from the
semantic content of still) serves to determine the interval ie. As t follows t',
which is set to the eighties due to presupposition binding (cf. van der Sandt
(1992)), t is an interval following the eighties and can thus include the speech
time. 14
13
We assume that these arguments are base generated inside the verbal projection (cf. Koopman
andSportiche 1991).
Sentence (41) is phrased as similarly to our initial example (2a) as possible. But since the
sentence cannot reasonably be uttered with a true individual level predicate (which blue is assumed to be with respect to cars; cf. the discussion of (37b) above), the matrix predicate had to
(41)
(a)
A car that was bought in the eighties is usually still roadworthy.
(b)
405
in 80s(e') e' @ 80s] C"(X) e @ ti Ipres(e) roadworthy(e)],

where t follows the eighties d u e to presupposition binding.
5 Causally related eventualities

The following examples are all felicitous, in spite of the fact that each of them
exemplifies the constellation that led to pragmatic deviance in our initial set of
examples, i.e. the relative clause verbs are marked for past tense, while the
matrix verbs are marked for present tense, and there is no overt interval setting:
(42)
A car that was m a d e in the eighties is usually blue.
(43)
A house that was built in the 19 lh century usually has a gabled roof.
(44)
A lawyer w h o was educated in Berlin is usually competent.
(45)
A m a n w h o was in jail during the eighties usually has a Bruce Lee tattoo.
What all the sentences have in common is that the states denoted by the matrix
verbs can plausibly be interpreted as being (at least indirectly) caused by the
relative clause eventualities.
In examples (42) - (45), the relative clause internal predicate denotes a set
of telic events. The sentences all require an interpretation for which the culmination point of the respective telic event coincides with the respective matrix
state. With verbs of creation such as the ones given in (42) and (43), this is
trivially true, because properties are usually only ascribed to existing entities.
In (44), this is due to the specific relation between the relative clause event and
the matrix state.
In (45), where the internal predicate of the relative clause denotes a state
without a culmination point, the sentence is still interpreted to mean that, with
respect to each man, the matrix state is the result of an (unspecified) event that
happened at some point during the running time of the relative clause state.
This means that the respective individual cannot have been in the state denoted
by the matrix verb at a time before the eventuality of the relative clause began,
and furthermore that there is some - however indirect - connection between
the two eventualities.
Interestingly, sentences become strange if a predicate is chosen in the matrix clause that cannot be interpreted as denoting a state indirectly caused by
be substituted. As can be seen below, the sentence is unacceptable with a true individual level
predicate:
(i) ?? A car that was bought in the 80s is usually still a B M W .
W e assume that this is due to the fact that still is superfluous as it only adds a presupposition
which is already part of the meaning of the individual level predicate be a BMW.
406
the respective relative clause eventuality. Compare (44) to (46) and (45) to
(47):
(46)
77
A lawyer who was educated in Berlin is usually blond.
(47)
77
A man who was in jail during the eighties usually has blue eyes.
We assume that (42) - (45) are felicitous for the following reason: It is impossible to convey the intended meanings of the sentences by using past tense in
both relative and matrix clause, due to the interval resolution strategy. On the
other hand, overt interval resetting by using the adverb nowadays (as in (34a))
is not an option either, because of the temporal proximity of the relative clause
eventualities and matrix eventualities.15
Consider in detail what happens if the matrix verb in example (43) is set to
past tense:
(48)
A house that was built in the 19'h century usually had a gabled roof.
This sentence gets the following reading: Most houses in the 19th century were
built with a gabled roof, and it implies that at least some of those houses do not
exist any more at the speech time (this is due to a lifetime effect, as described
for (27)). Note, however, that the interval resolution strategy predicts a different reading which is virtually impossible to get: According to this reading,
most houses that were built in the 19th century had a gabled roof before they
were built. Furthermore, nothing is implied about the existence of those houses
at the speech time.
This second reading is predicted by the interval resolution strategy for the
following reason: If e (where e is the eventuality of having a gabled roof) is
interpreted as holding at the same time as u(e') (where n(e') denotes the running time of the relative clause event), the corresponding representation for
(48) is as follows:
(49)
(a)
A house that was built in the 19lh century usually had a gabled roof,
(b)
MOST, [3x. Arg(e, ) house(x) [Be', build(e') Theme(e\ )

past (e') 19c(e') e' @ 19c] C"(X) e @ (')] [past(e) gabled_roof(e)]
This would imply that the gabled roof was already a property of the respective
houses before the process of building them was finished. But this is highly
implausible, and certainly not what sentence (48) is supposed to express.
If, on the other hand, the third step of the interval resolution strategy is
taken, and the matrix interval is set to the whole time axis, the sentence comes
15
Recall from the discussion of (34a) that the starting point of the interval denoted by nowadays
is automatically interpreted as lying at a certain distance from the end point of the relative
clause eventuality.
407
to mean that most (maximal) eventualities thematically related to houses built

in the I9'h century are eventualities of having a gabled roof that end before the
speech time. 16 This however implies that the respective houses no longer exist,
and a lifetime effect obtains.
Let us now suppose a speaker neither wants to express the implausible reading given in (49), nor imply that most houses with gabled roofs that were built
in the 19th century do not exist anymore at the speech time. Instead, she wants
to make a generalization about (probably still existing) houses that were built
in the 19th century. In this case, the only natural strategy seems to set the matrix verb to present tense, and directly take the last step of the interval resolution strategy given in (19), i. e. to instantiate the interval ie with the whole time
axis.17
Thus, sentence (43) (repeated below as (50a)) is interpreted as shown in
(50b):
(50)
(a)
A house that was built in the 19 lh century usually has a gabled roof,
(b)
MOSTe [Bx. Arg(e, ) house(x) [Be'. build(e') Theme(e', )

past (e') 19c(e') e' @ 19c] C"(X) e @ iWORID] [pres(e) gabledroof(e)]
According to (50b), the sentence states that most (maximal) eventualities that
are related thematically to a house that was built in the 19th century (in a specific manner) are eventualities of having a gabled roof that include the speech
time. This seems to be the correct meaning.
The same logic applies to the other examples in (42) - (45): In each case, the
strategy that was helpful in the case of (3a) - i.e. to set the matrix verb to past
tense - is not possible if one wishes to express an (at least indirect) causal
relation between the relative clause eventuality and the matrix eventuality,
because
(a) following the interval resolution strategy would force a reading according
to which the running time of the matrix eventualities is included in the running
time of the relative clause eventualities, and
(b) violating the interval resolution strategy and instantiating ie with the whole
time axis would result in an unintended lifetime effect.
Therefore, the only available option is to set the respective matrix verbs to
present tense, and instantiate ie with iworl - in violation of the interval resolution strategy.
The proposed mechanism seems to be confirmed by the following facts:
16
The fact that (48) requires such a reading shows that the interval resolution strategy may be
violated if there is an obvious reason for violating it: Obeying it would result in a reading that
is obviously not the intended one.
Point (1.) of the interval resolution strategy cannot be applied, because there is no overt information. Point (2.) is not an option either, as this would lead to the same contradiction as shown
for example (2a).
408
(51)
A lawyer who was educated in Berlin was usually competent.
(52)
A man who was in jail during the 80s usually had a Bruce-Lee tattoo.
In both (51) and (52), either a lifetime effect is triggered or the sentences get a
reading according to which the matrix states are already true of the respective
individuals at the time when the relative clause eventualities start. Thus, they
are no legitimate alternatives to (44) and (45) respectively if, on the one hand,
a causal relation between the two eventualities is to be expressed, and if, on the
other hand, the speaker does not seek to trigger a lifetime effect.
In this section we have shown that the interval resolution strategy may be
18
violated if this is the only way to express a certain meaning.
6 Summary
Based on a set of new observations, we have argued for an analysis of Qadverbs as exclusive binders of eventuality variables. We have shown that the
availability of QV-readings in sentences with indefinite DPs containing a relative clause is sensitive to the interaction of the tense markings of the respective
clauses (matrix clause vs. relative clause). QV is generally only possible if the
tenses agree.
We have argued for the existence of a pragmatic strategy that temporally locates the eventualities bound by the Q-adverb in an interval determined on the
basis of available information. This pragmatic mechanism is sensitive to locality considerations. In the absence of overt information, the eventualities that
are quantified over are located within the same interval as the running times of
the respective relative clause eventualities, since these count as interval information originating from the same domain (i.e. the restrictor). If this information concerning the temporal location of the respective eventualities contradicts the information constituted by the tense marking of the respective matrix
verbs (which are interpreted in the nuclear scope), the resulting structures are
semantically vacuous.
18
As Graham Katz (p.c.) has pointed out to us, there are related data that are problematic for our
account:
(i) (>>A song that was popular in the eighties usually has electronic beats in it.
Though it is not only possible, but necessary that the respective songs already had electronic
beats in them when they were popular, the sentence is still quite acceptable. We can only
speculate why this should be so. Perhaps the intended meaning cannot adequately be expressed
by using the past tense variant in this case either, as it is relevant to state explicitly that the respective songs still exist at the speech time. (If the past tense were to be used, and the running
times of the eventualities quantified over were set to the running times of the relative clause
eventualities, it would simply remain unclear whether the respective songs still exist at the
speech time).
409
F u r t h e r m o r e , w e h a v e explained why in certain well d e f i n e d cases the interval resolution strategy does not rule out the otherwise infelicitous structures
m e n t i o n e d above. This was either d u e to the p r e s e n c e of adverbs that overtly
introduce an interval in w h i c h the eventualities quantified over could be located, or to a specific relation holding b e t w e e n the relative clause and the m a trix eventualities: If matrix eventualities can naturally b e interpreted as having
b e e n (at least indirectly) caused by the relative clause eventualities, the respective sentences are felicitous. W e h a v e accounted f o r this e f f e c t by s h o w i n g that
skipping an o t h e r w i s e obligatory step of the interval resolution strategy and
resolving the contextual variable r e s p o n s i b l e f o r the t e m p o r a l location of eventualities to the w h o l e time axis is the only way to express the intended m e a n ings of the respective clauses, i.e. to express the causal relations b e t w e e n the
r e s p e c t i v e relative clause and matrix eventualities.
References
Bach, E. (1986): The Algebra of Events. Linguistics and Philosophy 9, 5-16.
Berman, S. (1987): Situation-Based Semantics for Adverbs of Quantification. In: J.
Blevins and Anne Vainikka (eds): University of Massachusetts Occasional Papers
12. GLSA, University of Massachusetts, Amherst.
Berman, S. (1991): The semantics of Open Sentences. Ph. D. dissertation, University of
Chierchia, G. (1995a): Individual Level Predicates as inherent Generics. In: G. Carlson
and F. J. Pelletier (eds): The Generic Book. Chicago: University of Chicago Press.
Chierchia, G. (1995b): Dynamics of Meaning. Chicago: University of Chicago Press.
Cohen, A. (2001): On the Generic Use of Indefinite Singulars. Journal of Semantics 18,
183-209.
de Swart, H. (1993): Adverbs of Quantification: A Generalized Quantifier Approach.
New York: Garland.
Dekker, P. (1993): Existential Disclosure. Linguistics and Philosophy 16, 561-587.
Diesing, M. (1990): Indefinites. Linguistic Inquiry Monographs. Cambridge (Mass.):
MIT Press.
Greenberg, Y. (2002): Two Kinds of Quantificational Modalized Genericity, and the
Interpretation of Bare Plural and Indefinite Singular NPs. In: B. Jackson (editor):
Proceedings of SALT XII. Cornell University: Cornell Linguistics Circle.
Greenberg, Y. (2003): Manifestations of Genericity. Outstanding Dissertations in Linguistics. New York and London: Routledge.
Groenendijk, J. and M. Stokhof (1991): Dynamic Predicate Logic. Linguistics and
Heim, I (1982): The Semantics of Definite and Indefinite Noun Phrases. PhD thesis.
University of Massachusetts, Amherst.
410
Herburger, E. (2000): What counts. Focus and Quantification. Cambridge (Mass.): MIT
Press.
Kamp, H. (1981): A Theory of Truth and Semantic Representation. In: J. A. G. Groenendijk, T. M. V. Janssen and M. B. J. Stokhof (eds): Formal Methods in the
Study of Language, 277-322. Amsterdam: Mathematisch Centrum, University of
Amsterdam.
Koopman, H. and D. Sportiche (1991): The Position of Subjects. Lingua 85, 211-258.
Kratzer, A. (1995): Stage-Level Predicates and Individual-Level Predicates. In: G.
Carlson and F. J. Pelletier (eds): The Generic Book, 125-175. Chicago: University
of Chicago Press.
Krifka, M. (1995): Focus and the Interpretation of Generic Sentences. In: G. Carlson
and F.J. Pelletier (eds): The Generic Book, 238-264. Chicago: University of Chicago Press.
Krifka, M., F.J. Pelletier, G.N. Carlson, A. ter Meulen, G. Chierchia, and G. Link
(1995): Genericity: An Introduction. In: G. Carlson and F. J. Pelletier, (eds): The
Generic Book, 1-124. Chicago: University of Chicago Press.
Krifka, M. (2001): Non-Novel Indefinites in Adverbial Quantification, In: c. condoravdi and G. Renardel de Lavelette (eds): Logical Perspectives on Language and
Information. Stanford: CSLI Publications.
Lenci, A. and P. M. Bertinetto (1999): Aspect, Adverbs, and Events: Habituality vs.
Perfectivity. In: F. Pianesi, J. Higginbotham and A. C. Varzi (eds): Speaking of
Events, 245-287. Oxford: Oxford University Press.
Lewis, D. (1975): Adverbs of Quantification. In: E. L. Keenan (editor): Formal Semantics of Natural Language, 3-15. Cambridge: Cambridge University Press.
Lbner, S. (1999): Why German schon and noch are still duals: A reply to van der
Auwera. Linguistics and Philosophy 22, 45-107.
Marti, L. (2003): Contextual Variables. PhD thesis. University of Connecticut, Connecticut.
Musan, R. (1997): Tense, Predicates and Life-Time Effects. Natural Language Semantics 5, 271-301.
Partee, . (1973): Some Structural Analogies Between Tenses and Pronouns in English.
The Journal of Philosophy, 70, no. 18, 601-609.
Partee, . (1995): Quantificational structures and Compositionality. In: A. Kratzer, E.
Bach, E. Jelinek and . Partee (eds): Quantification in Natural Languages, 541-602.
Dordrecht: Kluwer.
Rooth, M. (1995): Indefinites, Adverbs of Quantification and Focus Semantics. In: G.
Carlson and F. J. Pelletier (eds): The Generic Book, 265-299. Chicago: University
of Chicago Press.
Rothstein, S. (2003): Structuring Events. A Study in the Semantics of Aspect. Maiden:
Blackwell.
411
Sandt, R. van der (1992): Presupposition Projection as Anaphora Resolution. Journal of

Semantics 9, 333-377.
Smessaert, H. and A. ter Meulen (2004): Temporal Reasoning with Aspectual Adverbs.
Stanley, J. (2000): Context and Logical Form. Linguistics and Philosophy 23, 391-434.
Staudacher, P. (1987): Zur Semantik Indefiniter Nominalphrasen. In: . AsbachSchnittker and J. Roggenhofer (eds): Neuere Forschungen zur Wortbildung und
Historeographie der Sprache. Festgabe fr Herbert Brekle, 239-258. Tbingen: Narr
Verlag.
Vendler, Z. (1957): Verbs and Times. The Philosophical Review LXVI, 143-160.
von Fintel, K. (1994): Restrictions on Quantifier Domains. PhD thesis. University of
Zybatow, T. and M. Malink (2003): Verbklassen und Phasenpartikeln. In: M. Weisgerber (editor): Proceedings of the Conference SuB 7 - Sinn und Bedeutung", 327351. Arbeitspapier 114, Fachbereich Sprachwissenschaft, Universitt Konstanz.
Marko Malink (Berlin)
Phase structures and quantification*

1 Introduction
This paper deals with the interplay between phase particles and quantification
in Geman. We shall give a semantic analysis of phase structures denoted by
sentences such as (1) in which one of the phase particles in (2a) is applied to a
proposition whose subject is one of the quantifiers in (2b):
(1)
(a)
E i n i g e s i n d n i c h t m e h r da.
(b)
A l l e s i n d s c h o n da.
S o m e p e o p l e are n o t t h e r e a n y m o r e ,
E v e r y o n e is a l r e a d y there.
(2)
(a)
schon
(already)
(b)
alle (all)
noch (still)
einige
noch nicht (not y e t )
nicht alle (not all)
(some)
nicht mehr (no l o n g e r )
niemand
(nobody)
The plan of the paper is as follows. In the second section we will set up a suitable fonnal framework by formalizing the basic meaning of phase particles as
operating on predicates of times. In the third section we will give a detailed analysis of the scope relations between phase particles and quantifiers. This analysis
turns out to require a somewhat intricate procedine of extracting negations from
phase particles and quantifiers. The scope analysis will allow us to divide the
32 combinations consisting of one phase particle and one quantifier into eight
equivalence classes each of which contains four combinations.
In the fourth section the formalization developed so far is shown to fail to
account for an important feature of the intuitive meaning of the phase structures
under consideration. In the fifth section I propose a solution to this problem
within a generalized quantifier framework, giving up the assumption that phase
particles operate on predicates of times.
I am indebted to Regine Eckardt, Markus Egg, Cornelia Endriss, Tatjana Heyde-Zybatow and Ingolf Max for valuable remarks on this paper. Especially I would like to thank Christopher Pin
whose thorough comments on an earlier draft prevented me from saying a lot of unconvincing
things, and who suggested to me to use generalized quantifiers in this paper. All errors and obscurities are of course only the author's responsibility.
414
Marko Malink
2 Formalizing phase particles

The basic meaning of phase particles is a temporal one. It is confined to sentences in which phase particles take wide scope over an 'imperfective' (atelic)
time-relative proposition such as PETER-SLEEP(). In the basic meaning, that
is, phase particles are applied to predicates of times without focussing on the
constituents of these predicates (Lbner 1989, 1990).
(3)
Peter is already asleep.
We will assmne that the basic meaning of phase particles consists of an assertion
and a presupposition. A sentence such as (3) asserts that Peter is asleep at the
speech time, and presupposes that he was not asleep at some time in the recent
past. More generally, the assertion of already states that the argument proposition P(t) holds at the time of assertion, that is, at the reference time tr (which is
identical to the speech time in the present tense). The presupposition of already states that P(t) did not hold at some time before tr, giving rise to a phase
structure consisting of a negative and an ensuing positive phase of P ( t ) . During
the negative phase, P(t) is false, and during the positive one, P(t) is true, the
reference time tr being part of the positive phase. Not yet, on the other hand,
presupposes (the expectation of 1 ) a change from a negative phase to a positive
one as well, but asserts that the reference time tr is part of the negative phase.
By analogy, still asserts that the argument proposition is true at the reference
time, but presupposes (the expectation of) a change from positive to negative
polarity. That is, still presupposes (the expectation) that a negative phase of P ( t )
will follow the positive one. Again, no longer shares this presupposition, but
asserts that the reference time belongs to the negative phase. Thus the phase
structures triggered by the four phase particles can be illustrated by the following
diagram:
It is a moot point whether the phase structure after the reference time tr is to be regarded as truthconditionally presupposed or rather as expected or implicated (see Doherty 1973, p. 155; Knig
1977, pp. 192f; Lbner 1989, p. 176; 1990, p. 118; 1999, p. 60; van der Auwera 1998, pp. 39f;
Smessaert and ter Meulen 2004, p. 237 ). I shall not discuss this issue in this paper. For the sake of
simplicity, I shall neglect epistemic and pragmatic differences between the phase structure before
and after the reference time, giving a logical account of the symmetric full-blown phase structures
displayed in the diagram in (4 ) below.
415
Phase structures and quantification
(4)
already
stili
not yet
no longer
reference
time
nP(i)
Pit)
In formalizing the basic meaning of the phase particles we will use a two-sorted
first-order language containing variables t, t1, i * . . . of type (i) standing for time
points, and variables z,j, m... of type (e) standing for individuals such as John
and Mary. Moreover, we will need a binary relation of temporal precedence
t <ti (t is before ti ) obtaining between time points.
Given this relation, the change from negative to positive polarity presupposed
by already can be described by stating the existence of a final time point i* of
the negative phase, such that for all times t, the argument proposition P(t) holds
if and only if t is alter i* :2
3i*Vf(f* <t
(5)
P(t))
Not yet is similar to already in that it presupposes a change from negative to

positive polarity, but differs from it in that the reference time tr is part of the
negative phase. The two phase particles share the same presupposition (5), but
differ from each other in asserting ->P(tr) or P{tr) respectively. Still and no
longer, on the other hand, presuppose a change from positive to negative polarity. Such phase structures can be expressed by stating that there is a final time
point i* of the positive phase such that P(t) is false if and only if t is alter i* :
3 i*Vf(f*<f
(6)
Of course, the phase structure triggered by the phase particles holds only for a
contextually relevant time interval. This interval may cover a number of years
like in (7a), or only a few seconds like in (7b).
(7)
(a)
(b)
Peter is still a child.

The traffic light is already green.
Formally, the fact that phase particles are related to a contextually relevant time
interval could be captured by restricting both quantifiers in (5) and (6) to a certain
time interval. In this case, we would have to add the requirement that the final
2
For a similar analysis of the presupposition of already, see Krifka ( 1995, p. 242 ).
416
M a r k o Malink
time * of the negative or positive phase be neither an initial nor a final point
of that interval, as otherwise we would obtain trivial phase structures lacking
any change of polarity. For the sake of simplicity, however, we will not explicitly represent the contextually relevant interval in our formalization of the phase
particles. Instead, we will tacitly assmne that the domain (i) of time points is already restricted to a contextually relevant time interval, and that there are always
time points before and after i*.
In order to describe the presuppositional structure of the phase particles, we
will use a two-dimensional framework.3 That is, the phase particles will be
represented as ordered pairs of classical first-order expressions:4
,Q\
^
a s s e r t i o n
'
p r e s u p p o s i t i o n
The relation between assertion and presupposition in such two-dimensional formulae may be thought of as the relation of standard classical conjunction, displaying non-standard behavior only under negation. Thus, -prefixes are allowed to be attached to two-dimensional formulae, the rules governing these
prefixes being the same as those governing prefixes of classical conjunctions.
Given this framework, the basic meaning of the phase particles can be described
as follows:
(9)
Let tr be a variable of type (i) and a predicate of type (?', t):

ALREADY
d f
NOT-YET
X P X t
P ( t
)
J ; L p
-*P(tr)
f
X P X t r -
3i*Vf(f*<f
P { t r )
STILL
-df
NO-LONGER
=df
X P X t r .
X P X t
3 i * V f ( f * <t
-iP(i))
-.P(ir)
3 t M { t * < t
P{t))
-iP(i))
Suggestions of such two-dimensional frameworks include Karttunen and Peters ( 1979 ), Bergmann
(1981).
Semantically, such two-dimensional formulae are evaluated relative to standard classical models:
the formula
is true in a classical model if both A and are true in it; it is false if A is
false and is true; otherwise it is incorrect.
417
We want the default negation of two-dimensional formulae to negate only the

assertion and to preserve the presupposition. Moreover, we want this negation
to apply also to -prefixed formulae. This leads to the following definition of a
default negation ->a, with XX being a possibly empty string of A-abstractions:
(10)
XX.
"
"
,
= XX.
-A '
Finally, an equivalence relation ^ obtaining between (possibly -prefixed) twodimensional formulae can be defined as follows:
(11) Let " be a possibly empty string of -abstractions. Then the equivalence relation
-4i
Aa
holds true if and only if A i is classically equivalent
XX
XX
Si
B2
to A2, and B i is classically equivalent to B2.
This equivalence relation allows us to observe that the negation operator -

neatly fits the natural language default negation of phase particles illustrated in
(12):
(12) (a)
(b)
(13) (a)
(b)
Peter is already asleep. - No, he is not asleep yet.

Peter is not asleep anymore. - No, he is still asleep.
-i a ALREADY NOT-YET
-i a STILL NO-LONGER
We may also observe a kind of duality obtaining between ALREADY and STILL:
(14) (a)
(b)
-. A ALREADY(A,-.P()) STILL(AF.P(F))
-i A STILL(Af.-iP(F)) - ALREADY(A.P())
Finally, the equivalences in (13) and (14) imply the following equivalences:
(15) (a)
(b)
ALREADY(A.-IP()) - NO-LONGER(A.P())
STILL(A.-IP()) -(.())
3 Scoping quantifiers and phase particles

There are 32 combinations consisting of one of the four Geman phase particles
{schon, noch, noch nicht, nicht mehr} and one of the four Geman quantifiers Q G {alle,
einige, nicht alle, niemand}. In sixteen of them, the phase particle precedes the quantifier,
e.g., schon einige, noch einige. These combinations often contain the expletive subject es
(it):
(16) Es sind schon einige da.
There are already some people there.
418
Marko Malink
(17) Es sind noch einige da.

There are still some people there.
In the remaining sixteen combinations, the quantifier precedes the phase particle,
e.g., einige schon, einige nocir.
(18) Einige sind schon da.
Some people are already there.
(19) Einige sind noch da.
Some people are still there.
Throughout this paper, I confine myself to considering the default word order in
neutral sentences, neglecting leftward movements caused by information structure such as in the following example:
(20) ALLe sind noch Nicht da, aber einige.
Not everyone is there yet, but some of them.
If we assmne that there is a close connection between default word order and
semantic scope, the sixteen PQ combinations in which the phase particle precedes the quantifier are - at first glance - improblematic. It does not seem to be
relevant that the subject of the proposition to which the phase particle is applied
is a quantifier. The phase particle is applied to such a proposition in the very
same way as to any other proposition. Applying ALREADY to Einige sind da,
that is, to t . 3 z R ( z , t), yields the following formula:
(21)
A L R E A D Y ( . 3 S ( S , ) ) = tr
zR(z,tr)
3 f V f ( f , <t
3zR(z,t))
This seems to be a correct paraphrase of the meaning of (16). (21) predicts

that dming the presupposed negative phase, t . 3 z R ( z , t) is false, i.e., no people
are there, and that during the positive phase after , t . 3 z R ( z , t) is true, i.e.,
some people are there. The reference time tr is part of the positive phase, as
3zR(z, tr) is true. By analogy, (22) yields a correct paraphrase of (17), predicting that during the positive phase, which is located before i* and which contains
the reference time, t . 3 z R ( z , t) is true, i.e., some people are there, and that no
people will be there during the ensuing negative phase:
(22)
STILL(.3S(S,)) =
Xtr.
3zR(z, tr)
3 t M ( t *
<
-.3 z R ( z , t))
Thus, the two P Q combinations in (16) and (17) can be correctly analyzed by
assuming that the phase particle takes wide scope over the quantifier, that is, that
the default word order corresponds to semantic scope. However, the correspon-
419
dence between word order and semantic scope breaks down when the quantifier
precedes the phase particle. In ( 18) and (19) the quantifier einige obviously does
not take wide scope over the phase particle, even though preceding it. Otherwise
(18) would mean that there is at least one person, say John (j), who meets the
phase structure of schon. That means that only John must be absent during the
preceding negative phase presupposed by schon while all other people may be
present:
(23)
ALREADY(A.I?(J,F))
=Xtr
R(j,tr)
3i*Vf(f, <t
R(j,t))
Clearly, this is not what we mean by (18) (Einige sind schon da). If John is
the only one still missing at a party, it would be rather odd to utter (18) after
John has arrived, even though there is at least one person, namely John, who
meets the phase structure of schon. Instead, (18) presupposes that during the
preceding negative phase no people are present, just like the sentence (16) using
the same phase particle and the same quantifier in the reversed P Q word order.
Both sentences are equivalent to the wide-scope construction (24), and can, for
our purposes, be taken to have exactly the same meaning.
(24) Es ist schon der Fall, dass einige da sind.
It is already the case that some people are there.
In the same way, (19) (Einige sind noch da) does not mean that only some people
will not be present during the ensuing negative phase, but that no people will be
present. Therefore, (19) has the same meaning as (17), and both sentences are
correctly rendered by the wide-scope paraphrase (25):
(25) Es ist noch der Fall, dass einige da sind.
It is still the case that some people are there.
We can conclude that the word order of the phase particle and the quantifier does
not matter in (16)-(19). In all of these examples, the phase particle takes scope
over the quantifier, causing a mismatch between word order and scope in those
cases when the quantifier precedes the phase particle ((18) and (19)).
However, the order of the phase particle and the quantifier does make a difference when the phase particle is non-factive, that is, when it contains a negation
{noch nicht, nicht mehr). (26) is not equivalent to (27):
(26) Es sind noch nicht (einmal) einige da.
There aren't any people there yet (at all).
(27) Einige sind noch nicht da.
Some people are not there yet.
420
Marko Malink
The first sentence states that no people are there at the reference time (=speech
time), while the latter only states that some people are not there. Since indefinites
such as einige usually cannot stand immediately after a negation, (26) sounds
somewhat odd in the absence of einmal, but nevertheless the difference between
(26) and (27) is perfectly understandable. The non-factive phase particle does
not take scope over the quantifier in the Q P sentence (27). The same is true for
the following Q P sentence containing a non-factive phase particle:
(28) Einige sind nicht mehr da.
Some people are not there anymore.
Neither (27) nor (28) can be correctly rendered by the wide-scope paraphrases
(29) and (30), respectively. These paraphrases state that no people are present at
the reference time (=speech time), whereas (27) and (28) only require that some
people be absent at the reference time.
(29) Es ist noch nicht der Fall, dass einige da sind.
It is not yet the case that some people are there.
(30) Es ist nicht mehr der Fall, dass einige da sind.
It is not the case anymore that some people are there.
In view of the fact that the non-factive phase particle does not take wide scope
over the quantifier in (27) and (28) one might conclude that the quantifier takes
scope over the phase particle. In this case, however, both sentences would be
true as soon as some people meet the phase structure of noch nicht or nicht mehr
respectively, that is, if only some people are present during the presupposed
positive phase. But clearly, (27) and (28) presuppose that all people are present
during the (preceding or ensuing) positive phase. Hence the quantifier cannot
take wide scope over the non-factive phase particle in these sentences.
In order to obtain a correct paraplnase of (27), we need to split up the phase
particle noch nicht hito a factive, purely positive part noch and an 'internal'
negation nicht. The internal negation remains within the scope of the quantifier,
whereas the factive part of the phase particle is given wide scope over the quantifier like in (18) and (19). This yields (31), which is a correct paraplnase of
(27).
(31) Es ist noch der Fall, dass einige nicht da sind.
It is still the case that some people are not there.
In the case of (28) it may not be obvious how to split up nicht mehr (no longer)
hito a factive phase particle and an internal negation: mehr is an NPI substitute
for noch5 while nicht is an external presupposition preserving negation of noch
5
In certain contexts mehr and noch are interchangeable, for instance, after kaum (hardly). Both
kaum mehr and kaum noch are acceptable. Compare also the interchangeability of Dutch niets
421
(see (13b)). From a logical point of view, however, nicht mehr is equivalent to
schon nicht (already not) (see (15a)). This is confirmed by the fact that in many
Slavic languages, the phase structure of no longer is expressed by already not
(for instance, Czech uz ne). Thus nicht mehr can be split up into the factive
phase particle schon and an internal negation. As in the case of (27), the correct
paraphrase of (28) is obtained by giving wide scope to the factive phase particle
and leaving the internal negation within the scope of the quantifier:
(32) Es ist schon der Fall, dass einige nicht da sind.
It is already the case that some people are not there.
The above way of splitting up phase particles can be symbolized by two functions and which yield the factive part of the phase particle P and, if it
exists, the internal negation P :
p ^
(33)
schon
noch
noch nicht
nicht mehr
ALREADY
STILL
STILL
ALREADY
Finally, the situation becomes even more complex when we consider the monotone decreasing quantifiers niemand and nicht alle in QP-sentences. In this case,
the factive (part of the) phase particle does not take scope over the whole quantifier: (35) is not a correct paraphrase of (34) nor is (37) a correct paraphrase of
(36).
(34) Niemand ist noch da. 6
Nobody is still there.
(35) Es ist noch der Fall, dass niemand da ist.
It is still the case that nobody is there.
(36) Nicht alle sind nicht mehr da.
Not everyone is no longer there.
(37) Es ist schon der Fall, dass nicht alle nicht da sind.
It is already the case that not everyone is not there.
It is helpful here to remind ourselves that every monotone decreasing quantifier

can be thought of as the negation of a monotone increasing quantifier (Barwise
and Cooper 1981, p. 186). For instance, niemand (nobody ) is nicht einige (not
some). The correct paraphrase of (34) and (36) is obtained by giving to the
factive part of the phase particle scope over the monotone increasing part of the
nog(nothing still) and niets meer (nothing anymore) described by van der Auwera (1998, p. lOlf).
Noch is a positive polarity item (see p. 423 below). Therefore Niemand ist noch da sounds somewhat odd. One would prefer Niemand ist mehr da instead. For present purposes, we may neglect
this difficulty as the meaning of Niemand ist noch da is none the less perfectly understandable.
422
Marko Malink
quantifier (alle and einige) while the negation of the quantifier is given wide
scope over the whole sentence including the factive phase particle: 7
(38) Es ist nicht der Fall, dass noch einige da sind.
It is not the case that still some people are there.
(39) Es ist nicht der Fall, dass schon alle nicht da sind.
It is not the case that already all people are not there.
In Q sentences, the external negation Q of the quantifier takes scope over

sentences that contain a presupposition triggered by a phase particle. That is,
within the two-dimensional framework introduced above, Q is applied to a
two-dimensional formula. In this case, Q can be represented by the presupposition preserving default negation -i a defined in (10). In P Q sentences, on the
other hand, the negation Q does not take scope over the phase particle, but is
applied to a one-dimensional formula. In this case, -<a will be understood to be
the standard classical negation -i. 8
Q
Q Q
(40)
alie
einige
nicht alie
Mz
3z
'
V
niemand
3z
Given the procedure of splitting up phase particles and quantifiers specified in

(33) and (40), the scope behavior of Q sentences can be described by the following diagram:
/
(41) Q P
Q P s Q P
The mismatch between word order and scope is confined to the 'positive' parts
P and Q of the phase particle and the quantifier. These have to be interchanged in order to get the correct scope relations. The internal negation of the phase
particle and the external negation of the quantifier Q are not affected by
this inversion. In P Q sentences, on the other hand, the order of the 'positive'
and 'negative' parts of the phase particles and quantifiers remains unchanged
and corresponds to the word order:
(42) P Q
P Q Q
As an example, the scope analysis specified in (41) and (42) will be carried out
for the combinations niemand noch nicht and schon alle, which happen to yield
equivalent phase structures:
7
8
For a similar phenomenon, see the 'split readings' of quantifiers in de Swart (2000).
That is, - is understood to be applicable to both one- and two-dimensional formulae. When
being applied to one-dimensional formulae, <a reduces to classical negation.
423
(43) niemand noch nicht

-a STILL
-1
niemand" noch nicht niemand noch nicht"
This m e a n s :
niemand noch nicht

Xtr.
\t
jSTILL (\t.3z^R(z,
t))
i3ziR(z,tr)
3V( < t
- . 3 z ^ R ( z , t))
\/zR(z,
3tM(t* <t
tr)
VzR(z,t))
a l r e a d y ( A t . V z R ( z , t))
/fe0 schon //e schon"
0 ALREADY V; 0
alle schon
schoif schon"alie" alie
ALREADY 0 0 \/
schon alie
Wlien applying this scope analysis to all the 32 combinations consisting of a phase particle and a quantifier, it turns out that there are eight equivalence classes.9
This means that German is able to express only eight different phase structures by means of the four quantifiers and the four phase particles considered in
this paper. Each of the equivalence classes contains four combinations. These
combinations are listed in the first row of the table on p. 424f. The second row
gives the fonnal representation obtained by the scope analysis specified in (41)
and (42). The third row illustrates the intuitive meaning of the given equivalence
class by a diagram.
Those combinations which cannot be uttered felicitously in neutral contexts
are marked by ? ? . Within the equivalence classes 1-4, the combinations marked by ? ? are inappropriate because they contain two negations while there are
two equivalent combinations which do not contain any negation. Thus, it would
violate pragmatic principles to use the doubly negated combinations in neutral
contexts. Within the equivalence classes 5-8, however, both correct and incorrect
combinations contain only one negation. Hence there must be other reasons for
the incorrectness in these equivalence classes than the pragmatic constraints explaining the incorrectness in the equivalence classes 1-4. First, schon and noch
are positive polarity items 1 " (PPI) which usually cannot stand within the scope of
monotone decreasing quantifiers ( niemand noch,''nicht alle noch,''niemand
schon,??nicht alle schon). Second, einige shows PPI properties as well 11 (""nicht
mehr einige, ""{noch nicht} einige). Moreover, alle normally does not allow for
negations within its scope 12 {""alle nicht mehr, ""alle noch nicht). The same
is true for schon, which does not allow for negations within its scope either
{""schon niemand, ""schon nicht alle). Finally, there is an interesting contrast
between {noch nicht} alle and ""{noch} nicht alle, but it would need to digress
too much to have a closer look at it.
9
10
11
12
That is, equivalence classes in the sense of the equivalence relation

defined in ( 11 ).
See e.g. Krifka (1995, p. 242), van der Wouden ( 1997, p. 118 and p. 131).
See Jacobs ( 1982, p. 149).
See Jacobs ( 1982, p. 193).
424
Marko Malink
Ti
ce
bO
>
M
p
<1)
-
'S
fi
o nfi
o
fi
C-
ho
te
te;

'3
>
te1
t
<
O
>
>
i
\
>
m
<
>
<
'
0)

fi
-
-
'fi
fi
-
0) Td

0)
'fi
'fi
te
>
te
'
>
I
^
V*
>
m
>
<
>
425
-it;
-
- I
>r
-n
41
ho
te
-n
^ >r
i
v.
ho
'3
) -
m
te
r
426
Marko Malink
4 The problem
There is an important feature of the intuitive meaning of the phase structures
illustrated in the third row of the table above which is not captured by the formalizations m the second row. Take, for instance, einige schon (some already) in
equivalence class 3. The presupposition (44) of the formalization predicts correctly that during the presupposed negative phase (that means at all time points t
which are not after i), we have ->3zR(z, t.). In tenns of our example, no people
are present at t for all t < i* :
(44) 3Wit{U < t 3 z R { z , t ) )
As soon as we enter the positive phase and t is after
the existentially quantified formula 3zR(z, t) becomes true. The problem is that the existential quantification can be made true by entirely different persons during the positive phase.
Thus, the formalization allows for phase structures like (45) or (46). In both of
them, the existential quantification is true during the positive phase after t*, even
though people are arriving and leaving again. The number of people present may
even decrease, as shown in (46).
(45)
BEG
END
(46)
BEG
**
tr
END
BEG
**
tr
END
(47)
But clearly, what we intuitively mean by einige schon in sentences like (18) is
the phase structure (47). In this phase structure nobody leaves once they arrive,
427
and thus the number of persons present increases constantly with every person
who arrives. Granted that a sentence such as (18) is typically not intended to
explicitly exclude the possibility of someone leaving a (possibly big) party. But
nevertheless there is a strong intuition that the expected continuous increase is
due to the presupposition that nobody will leave (during a contextually relevant
interval indicated by BEG and END in the above diagrams 13 ). We want the formal representation of (18) to account for this intuition and to rule out unwanted
models such as (45) or (46).
Similar problems arise for each of the eight equivalence classes listed on p.
424f because in each of them, an existential proposition 3zR(z, t) or -i \/zR(z, t)
is required to be true either during the positive or during the negative phase. For
instance, the fonnal representation of nicht mehr alle (no longer all) in equivalence class 6 allows for unwanted phase structures such as (48) as well as for
correct phase structures such as (49).
V;
Hz
Nz
(48)
BEG
**
V;
tr
-iVs
END
<\/z
(49)
BEG
**
tr
END
5 A solution
In order to rule out unwanted phase structures such as illustrated in the diagrams
(45), (46) and (48), we have to take into account every horizontal Ime in these
diagrams separately. That means, we have to take into account not only quantified formulae such as VzR(z, t) or 3zR(z, t), but also non-quantified formulae
such as R(J, t),R(m, t),... stating that John (j) is there, Mary (in) is there and
so on.
There are several ways to describe and rule out the unwelcome features of
the phase structures (45), (46) and (48). One way would be to require that for
every person z, it must not be the case that the (contextually relevant) domain
13
The contextually relevant interval indicated by BEG and END is not made explicit in the formalizations in the second row of the table on p. 424f (see p. 415 ).
428
Marko Malink
of times contains both the beginning of a maximal positive phase of R(z, t) and
the end of a maximal positive phase of R(z, t). This condition precludes phase
structures such as (45) and (46) in which for some there is a maximal positive
phase of R(z, t) such that both the beginning and the end of this maximal phase
belong to the relevant domain of times. Moreover, this condition precludes phase structures such as (48) in which for some there are two distinct maximal
positive phases of R(z, t) such that the end of the first phase and the beginning
of the second one belong to the relevant domain of times.
The unwanted phase structures could be precluded by adding the above condition as a further presupposition to the definition of the phase particles. However,
the introduction of such an additional presupposition may appear to be an ad hoc
solution which does not get to the heart of the problem. For instance, we would
need to assmne that the additional presupposition is also present when the phase
particle is applied to sentences which do not contain any quantifiers. However, it
seems to me that our problem is closely related to the fact that the phase particle
is applied to a sentence whose subject is a quantifier. In what follows, I wish
to propose one way of solving our problem without adding a new presupposition; instead, we will modify the given presupposition such that the solution of
the problem is related to the fact that the phase particle is applied to a sentence
whose subject is a quantifier.
Consider again the phase structure (47) denoted by einige schon (some already) in equivalence class 3. The fonnal representation (44) of this phase structure
requires that the quantified formula 3 z R ( z , t ) meet the presupposition of ALREADY, that is, that there be exactly one preceding negative phase during which
3zR(z, t.) is false and exactly one ensuing positive phase during which this formula is true. Now, the unwanted phase structures (45) and (46) can be ruled out
by requiring that not only 3zR(z, t) meet the presupposition of ALREADY but
also the non-quantified formula i?(z,i) for every z. That is, for every person there must be exactly one negative phase during which she is absent and exactly one
ensuing positive phase during which she is present. The duration of the negative
phase may differ from person to person; for every there may be a different final
point i* of the negative phase. Formally, this amounts to:
(50) Vz3UVt(U < t
R(z,t))
This condition precludes the unwanted structures (45) and (46) in which there
are for some persons two negative phases interrupted by a positive one. However, the condition in (50) is too strong, as the well-behaved phase structure (51)
(=(47)) does not meet it either. The reason is that in (51 ) not all individuals meet
the presupposition of ALREADY, but only the four middle individuals b, c, d and
e. The two outer individuals a and / fail to meet the presupposition of ALREADY
because they do not possess a positive phase.
3z
429
3s
(51)
BEG
END
Thus, we have to exclude the two irrelevant outer individuals, only requiring
that the remaining relevant individuals meet the presupposition of A L R E A D Y . In
(51), the irrelevant individuals are those which are in a negative phase during all
the time. Hence the remaining relevant individuals b, c, d and e can be picked
out by the following formula:
(52) 3 t R ( z , t )
The same strategy works for the equivalence classes 4, 5, and 7. In all these
equivalence classes, the irrelevant individuals do not possess a positive phase so
that the relevant ones can be picked out by (52). Given this description of relevant
individuals, we may require that not only the quantified formula 3zR(z,t)
meet
the presupposition of the phase particle, but also that every relevant individual
meet this presupposition.
In the remaining equivalence classes 1, 2, 6 and 8, the irrelevant individuals
are those which do not possess a negative phase, so that the relevant individuals
can be be picked out by the following formula:
(53) 3 t ^ R ( z , t )
Apart from this difference, the strategy introduced for the equivalence classes 3,
4, 5 and 7 can also be applied to the equivalence classes 1, 2, 6 and 8. In the case
of nicht mehr alle in equivalence class 6, for instance, we have to require that
not only the quantified formula \/zR{z,t)
meet the presupposition of S T I L L , but
also the relevant individuals picked out by (53).
For all eight equivalence classes, the irrelevant individuals can be characterized as those which do not undergo any change of polarity. Hence, the relevant
individuals are those which possess a positive phase and a negative phase. In all
eight equivalence classes, the individuals , relevant with respect to the relation
R(z,t) can be picked out by the following formula:
(54)
REL(s,
R)
=df
3tR(z, t)
3t^R(z,
t)
As an aside, we might note that our formalization of the eight equivalence classes
(the second row of the table on p. 424f) does not require that there exist any
430
Marko Malink
irrelevant individuals, though the diagrams in the third row of the table on p.
424f always contain irrelevant individuals (the two outer ones).
After having characterized the relevant individuals in all equivalence classes
by (54), we now proceed to extend the definition of the phase particles in such
a way that the presupposed phase structure holds not only for quantified formulae such as VzR(z, t) or 3zR(z, t), but also for non-quantified formulae such
as R(z,t) for all relevant individuals z. According to definition (9), the phase
particles are applied to a one-place predicate Xt,.P(t) of times. This reflects the
idea that in then basic meaning, phase particles take wide scope over a whole
time-relative proposition (that is, a predicate of times) without focussing on the
constituents of the proposition. This idea, I submit, has to be given up in view of
the problems described above. In order for these problems to be solved, we have
to take into account not only quantified predicates of times such as Xt.VzR(z, t)
or t . 3 z R ( z , t), in which the subject position is bound by a quantifier, but
also several non-quantified predicates of times t . R ( z , t), in which the subject
position is occupied by free variables. This is impossible as long as the phase
particles are applied to quantified predicates of times such as Xt.VzR(z, t), as
there is no way of removing the quantifier so as to obtain non-quantified predicates.
Instead, the phase particles should be applied separately to a two-place relation XzXt,.R(z, t) obtaining between individuals and times and to a subject (external argument) occupying the left -argument of this relation. This subject can be
specified either as one of the quantifiers Vx, 3x or as a relevant individual. That
is, the one-place time-relative proposition Xt.P(t) is split up into a subject and
a two-place relation XzXt.R(z, i). The phase particle may be seen as providing
the connecting link between the two parts of the time-relative proposition.
According to this view, phase particles do not take wide scope over a whole time-relative proposition in the natural language sentences considered in this
paper. Rather, they resemble the behavior of a focus particle in that there is an
interaction with a certain constituent of a proposition, that is, with the subject
constituent realized by quantifiers such as alle and einige. Hence we shall assume that phase particles are applied separately to a subject and to a two-place
relation of type (e, {i, t}} obtaining between individuals and times.
In order to make this strategy formally work, the quantifiers Vx and 3x should
be treated as entities of the same logical type as individuals, since both individuals and quantifiers should occupy the subject position of XzXt.R(z, t) in the
same way. This can be done by treating quantifiers and individuals as generalized quantifiers of type ((e, {i, t)), {i, t)). Such generalized quantifiers are applied to a two-place relation XzXt.R(z, t) of type (e, {i, t)), and yield a one-place
predicate of times by binding the -argument of type (e) in this relation. The definitions of the generalized quantifiers t V , 1 3 1 and 2 are straightforward:
(55) (a)
(b)
(c)
V =df
3 =df
s =df
431
XRXt.VzR(zj)
\RM3zR(z,t)
XRXt.R(z,t)
For example, the assertion VzR(z, tr) of alle schon ili the first equivalence class
reads (t V | (R))(t r ) in tenns of generalized quantifiers. Taking and as
variables for generalized quantifiers of the type defined in (55), we can introduce
a relation ^ R Q between generalized quantifiers which states that is either
identical 14 to or that is (identical to a generalized quantifier corresponding
to) an individual relevant with respect to the relation R in the sense of (54):
(56)
=df
= 3^(,.)=>)
Now, the presupposition of ALREADY and NOT-YET, formulated in (5), can be
redefined in the following way:

(57) [
Q D 3tM{t*<t
(())())]
or shortly:
(58) ~ 3fVf(f<f (())())
This formula states that for all generalized quantifiers which are either identical to or which are an individual relevant with respect to R, the formula
((_))() meets the presupposition of ALREADY. In the case of alle schon in
the first equivalence class, for instance, is specified as V | and (58) implies,
first, that the formula \/ \{R)){t) (i.e. VzR(z, t)) meets the presupposition of
ALREADY, and, second, that the formula (T Z \(R))(t) (i.e. R(z, t)) meets this
presupposition for every individual which is relevant with respect to R in the
sense of (54).
By analogy, the presupposition of STILL and NO-LONGER is obtained f r o m
(58) by inserting a negation immediately before the formula ((_))(). In the

case of einige noch in the fourth equivalence class, this presupposition states,
first, that the formula
3 \{R)){t) (i.e. 3 z R ( z , t ) ) meets the presupposition
of STILL and, second, that the formula (T z \(R))(t) (i.e. R(z,t)) meets this
presupposition for every relevant individual z.
Thus, the presupposition in (58) and its counterpart for STILL do what we
want them to do, and the four phase particles can be redefined in tenns of generalized quantifiers as follows:
14
Identity is to be taken here as the extensional identity of set theory.
432
Marko Malink
(59) Let tr be of type (i), Q of type ((e, (i, t)), (i, t)) and R of type (e, (?', t)):
ALREADY =df
XRXQXtr.
(Q())(ir)
~ * 3fVf(f<f ( (#))(*))
NOT-YET =df
XRXQXtr.
-(Q())(ir)
~ 3*V(* <t ^ ((i?.))M)
STILL = df
XRXQXtr.
(Q())(ir)
~ 3*V(*< -.((?))())
NO-LONGER = d /
XRXQXt,.
n(Q())(ir
3*v(* < -.((?))())
I wish to conclude this paper by showing that the revised definition (59) does not
only imply the standard wide scope definition of the phase particles in (9), but
that it is even equivalent to this simpler definition under certain circumstances.
More precisely, if the external argument Q is not a quantifier such as alle or
einige ( V | or | 3 ) but a single individual such as John (| j ), the revised
definition of the phase particles can be slightly modified such that it is equivalent
to the standard wide scope definition. To this end, we have to take into account
the domain of the generalized quantifier Q, and to ensure that the relation ^ R
O. introduced in (56) holds only if is or if is a relevant individual which
belongs to the domain of Q.
Intuitively, the domain of V | and 3 is the whole domain of individuals of
type (e), the domain of | a l l men f and | some men f is the subset of men in
the domain of type (e), and, crucially, the domain of j is the singleton {j}. 15
Formally, the domain of a generalized quantifier 0. is the smallest set 0. lives on
(for this notion see Barwise and Cooper 1981, p. 1781). A generalized quantifier
XRXt.Q
of type ((e, { i , t)), { i , t}} lives on a predicate .E of type (e, t) iff for
all times t and for all XzXt.R, it makes no difference whether Q is applied to
the relation XzXt.R or to the relation obtained from XzXt.R by restricting the
-argument to the predicate E:
15
Generalized quantifiers can be seen as NP-denotations resulting from the application of a determiner to an N-denotation, the N-denotation being a set of individuals (Barwise and Cooper
1981). For example, the determiners a l l and s o m e are \E\R\t.*iz(E(z)
D R(z,t))
and
XE\RXt3z{E{z)
AR(z, t) ) respectively. Then \ V \ is a l l ( . = ), \ 3 \ is s o m e ( z . j =
), \ s o m e men \ is s o m e f A i . m a n ^ ) ) , and \ j \ is a l l ( A ^ . j = z). Given this analysis of generalized quantifiers, the domain of \ V f, \ 3 \ and \ j \ is exactly the N-denotation . =
or \z.j = to which the determiner is applied in order to obtain the generalized quantifier in
question.
433
(60)
Q LIVE-ON E
=df
VtVR{[Q(\z\t.R(z,t))](t)
<-> [Q(XzXt.R(z,t)
E(z))](t)}
The domain ["}] o f a generalized quantifier 3 is the smallest set Q lives on.
That is, [ 0 ] is an improper subset o f all predicates on which 3 lives. Or else,
[ ] (z) is true iff belongs to all sets on which Q lives:
(61)
\Q](z)
WE{{Q
d f
L I V E - O N E)
E(z))
It is not hard to verify that the following holds: 1 6

(62)
(a)
rmi(z)
3)
(b)
[Tj]()
=3
Finally, the relation

Q is redefined by adding the condition [}] (z) such
that individuals which do not belong to the domain o f Q are disregarded:
(63)
* P
=df
Q = p V
3z(rel(z,R)A\Q](z)A^=U\)
I f Q is I V I or I 3 f , the addition o f the condition [ 0 ] ( s ) does not have any

consequences, as pQ] (z) is equivalent to the trivial condition = z. However,
i f Q is the generalized quantifier } j | corresponding to a single individual j,
\] () is tantamount to = j . Consequently, the only quantifier such that
is j itself: by (63),
j means that is either identical to
'j j j or a relevant individual belonging to the domain o f ] j 1 ; but the domain
o f j j j is the singleton {j}. Hence, the presupposition (64a) can equivalently
be reduced to the standard presuppositions (64b) and (64c) in which the phase
particle can be taken to apply to a one-place predicate Xt.R(j, t) o f times:
(64)
(a)
(b)
(c)
16
Tj T
3uvt(u<t -
({R))(t))
at.vt(t.<i~(m(fl))(t))
3 t M { t , < t ^
R(j,t))
To verify (62a), we have to show that neither 1 V f nor f 3 f lives on any \z.E(z)
such that
HzE{z). To this end, assume that there is an such that -<E(x). Take XzXt.R to be the relation
XzXt.z = t = t. In this case, we have for every time f: | V f ( A z X t . R ( z , t))(t), whereas
fV t ( X z \ t . R ( z , t) E(z))(t)
fails to hold. In view of (60), this means that | V | does not live
on Xz.E(z).
Moreover, take XzXt.R to be XzXt.z = = f. In this case, we have for every
time i: t 3 ( X z X t . R ( z , t ) ) ( i ) , whereas 3 f ( X z X t . R ( z , t) E{z)){t)
fails to hold. This
means that f 3 f does not live on . ( ) .
To verify (62b), we have to show, first, that f j f lives on Xz.z = j, and, second, that | j |
lives on no Xz.E(z) such that <E(j). The first claim is obvious. To show the second claim, we
assume iE(j) and take XzXt.R to be the relation XzXt.z = j = t. In this case, we have
for every time t: j (XzXt.R(z,
t))(t), whereas f j | (XzXt.R(z.
t) E(z))(t)
fails to hold.
This means that f j f does not live on
Xz.E(z).
434
Marko Malink
T h u s , t h e s t a n d a r d w i d e s c o p e p h a s e p a r t i c l e s in (9) t u r n o u t to b e a s p e c i a l ins t a n c e o f t h e r e v i s e d p h a s e p a r t i c l e s in (59). If t h e s u b j e c t o f t h e n a t u r a l l a n g u a g e

s e n t e n c e is a single i n d i v i d u a l s u c h as J o h n , t h e p h a s e p a r t i c l e s in ( 5 9 ) r e d u c e
to t h e w i d e s c o p e o p e r a t o r s d e f i n e d in (9). O n l y w h e n t h e s u b j e c t is a q u a n t i f i e r
s u c h as alle o r einige,
w e n e e d to d i s t i n g u i s h t h e s u b j e c t f r o m t h e s u b j e c t l e s s
time-relative proposition, and to take into account the interaction b e t w e e n the

p h a s e p a r t i c l e a n d t h e s u b j e c t quantifier. If t h e s u b j e c t is n o t a quantifier, w e c a n
s a f e l y c o n t i n u e a s s u m i n g that, in t h e i r b a s i c m e a n i n g , p h a s e p a r t i c l e s t a k e w i d e
s c o p e o v e r a w h o l e t i m e - r e l a t i v e p r o p o s i t i o n , t h a t is, o v e r a p r e d i c a t e o f t i m e s .
References
Barwise, John and Cooper, Robin (1981 ): Generalized Quantifiers and Natural Language.
Linguistics and Philosophy 4: 159-219.
Bergmann, Merrie (1981): Presupposition and Two-Dimensional Logic. Journal of Philosophical Logic 10: 27-53.
de Swart, Henriette (2000): Scope Ambiguities with Negative Quantifiers. Pp. 109-132
in Reference and Anaphoric Relations, edited by Klaus von Heusinger and Urs Egli.
Dordrecht: Kluwer.
Doherty, Monika (1973): Noch and schon and their Presuppositions. Pp. 154-177 in Generative Grammar in Europe, edited by F. Kiefer and N. Ruwet. Dordrecht Reidel.
Jacobs, Joachim (1982): Syntax und Semantik der Negation im Deutschen. Mnchen:
Wilhelm Fink Verlag.
Karttunen, L. and Peters, S. (1979): Conventional Implicature. Pp. 1-56 in Syntax and
Semantics 11: Presupposition, edited by Choon-Kyu Oh and D. Dineen. New York:
Academic Press.
Knig, E. (1977): Temporal and Non-Temporal Uses of schon and noch in German. Linguistics and Philosophy 1: 173-198.
Krifka, Manfred (1995): The Semantics and Pragmatics of Polarity Items. Linguistic
Analysis 25: 209-257.
Lbner, Sebastian (1989): German schon - erst - noch: An Integrated Analysis. Linguistics and Philosophy 12: 167-212.
Lbner, Sebastian (1990): Wahr neben Falsch. Duale Operatoren als die Quantoren natrlicher Sprache. Tbingen: Max Niemeyer Verlag.
Lbner, Sebastian (1999): Why German schon and noch still are Duals: A Reply to Van
der Auwera. Linguistics and Philosophy 22: 45-106.
Smessaert, Hans and ter Meulen, Alice G. B. (2004): Temporal Reasoning with Aspectual
Adverbs. Linguistics and Philosophy 27: 209-261.
van der Auwera, Johan (1998): Phasal Adverbials in the languages of Europe. Pp. 25-145
in Adverbial Constructions in the Languages of Europe, edited by J. van der Auwera
and D. P. O Baoill. Berlin, New York: Mouton De Gruyter,
van der Wouden, Ton (1997): Negative Contexts. London and New York: Routledge.
Alice G. . ter Meulen (Groningen)
Cohesion in temporal context:

The role of aspectual adverbs
1 Introduction
Communicating information between agents, natural or artificial, may be formally modeled as non-deterministic, internally structured update procedures,
adding new information issued by the sender or speaker to the given information state of its recipient, barring logical inconsistency or cognitive incomprehensibility. The widened perspectives of this research program of dynamic
natural language semantics integrate issues of information sharing and belief
revision, commonly considered epistemic in nature, with its more traditional
logical concern about the narrow truth-functional aspects of meaning and interpretation. Mere consistency of an information state (= truth of all sentences
(at an index) in a model) does not suffice to properly characterize the dynamics
of human communication. Satisfaction of (open) propositions by the given,
supposedly current, variable assignment function, extensions thereof or alternative assignments to it, in the intended model (at an index) will not do either,
although partial variable assignments have a central role to play in modeling
how information is accumulated in the common ground, holding information
the sender and recipient jointly assume to be true.
Coherence is the more interesting structural semantic notion, understood as
a property of dynamic information states and their update relations. In order to
determine what makes information coherent or an exchange of information a
coherent process, it shall be necessary to study in more linguistic detail how
factual information may be expressed in natural language together with subjective or personal information, i.e. how information about a situation is mixed
with information about epistemic states regarding the situation in one and the
same clause. Towards this greater goal of understanding coherence of content
the present paper presents an account of English aspectual adverbs and their
usage indicating the user's subjective attitudes towards the course of events
described. Their supposedly objective temporal descriptive content interacts
with the subjective information about the user's attitudes towards the described
change, creating informational dependencies between facts and beliefs, which
contextually constrain inferences, presuppositions and entailments. Studying
436
Alice G. . ter Meulen
how asserting new information differs in its dynamic force from presupposing,
entailing or otherwise inferring information will help us understand the way
communication situates its users in time. This may get us one step closer towards a generally desired dynamic logic of indexicals and demonstrative reference.
2 The semantics of aspectual adverbs

The semantics of aspectual adverbs is captured by the indexical interaction
between reference times, polarity transitions START and END, and their temporal anaphoric dependencies created by the adverbs SINCE and UNTIL,
constituting a coherent context (cf. Smessaert & ter Meulen 2004). For instance, if John is not yet asleep, John is obviously now not asleep, but he also
must be falling asleep, resulting in his being asleep later (cf. la). In addition to
the information expressed by the plain atemporal negation John is not asleep,
the aspectual adverb not yet adds the information that, although the negative
sleeping state currently holds of John, its polarity reversal is imminent. The
primary use of aspectual adverbs is hence to modify a description of a state,
indicating it is about to change or has just changed, without actually referring
to the event causing this change. Obviously, the corresponding event, causing
the transition of the negative polarity of John's not being asleep to a positive
one, must be currently going on, which we would express in English as falling
asleep. If r0 represents the current reference time, it must be temporally
included in the event of John falling asleep fall asleep'(e, j, +), which
presupposes that he is not asleep, while a future reference time r temporally
locates the onset of John's sleeping, i.e. the time John has fallen asleep,
asleep'(e, j, +). The indexical, context dependent temporal adverb until relates
these two reference times, binding in effect r to be the first moment at which
John is asleep after the current time r 0 , disregarding any later times at which he
may also be asleep as irrelevant or at least excluded from the current context.
The semantic rule for of the aspectual adverb not yet could accordingly be
specified as follows:
[INFL not yet [ w () ]] => , , , Xs [P'(s, , -) A r0 &
PERF (CAUSE P'(e, x, +)) ! & r0< r, & UNTILA, (CAUSE P'(e, x, +))) a r0]
Since polarity reversal on the state is all that affects the temporal reasoning
in these contexts this semantic rule for the basic aspectual adverbs can be simplified, disregarding the causative event reversing the polarity of the corresponding static property. Correspondingly, in (1) below the semantics of the
basic aspectual adverbs is presented only in terms of polarity transitions
START () and END () of static properties and the indexical binding adverbials SINCE and UNTIL, well known in most systems of temporal logic. This
C o h e s i o n in t e m p o r a l
437
context
clarifies the logical relationships between the four basic aspectual adverbs best,
showing the compositional interaction between positive and negative polarity,
their transitions and these two indexical adverbs. If desired, the semantics
could be defined in equivalent terms of causative events and the perfect operator, indicating the resulting state.
Analogously, if John is already asleep, he must have fallen asleep in the
past, hence the start of the positive phase must be past, and he must have been
asleep since (cf. lb). The aspectual adverbs still and not anymore are now seen
to constitute the obvious logical counterparts to not yet and already respectively (cf. le and Id). The parameters listed before the bar, I , are reference
markers, supposedly existentially quantified as in DRS representations, and the
conditions listed after I present the truth functional content.
(1)
Basic aspectual
(a)
[ip x
adverbs
[ i n f l not
yet
UNTELO,, (P'(s, ,
(b)
[>
[infl
[vp ] ] ]
[ r 0 , , s , I ' ( s , , - ) & s a
already
-))) a
[vp
(c)
[ip x
[infl
[infl
r0 & r a
S T A R T ( P ' ( s , x, +)) & n < r0
&
r0]
still [ ]]] =>
U N T I L ( r i , ( P ' ( s, x,
[ x
&
/>]]] =>
[r0, r , , s, I P ' ( s , x, + ) & s a r
(d)
E N D ( ' ( s , , -)) & r 0 < r,
r]
[r0, , s, I P ' ( s , , + ) & s 2

S I N C E ( r , , (P'(s, x, +))) a
>
r 0 & [ 2
& r i a E N D ( P ' ( s , , + ) ) & r 0 < r,
&
+ ) ) ) a r]
not [ v p x P] anymore]] =>
[ r o , , s , I P ' ( s , ,
-) & s a
S I N C E ( r i , (P'(s, x,
-)))ar0]
r0 & r , a S T A R T ( P ' ( s , x, - ) ) & [< r 0
&
These four basic aspectual adverbs constitute a logical polarity square in the
temporal domain of events, showing the basic logical interaction between the
current, past or future reference times, related by since and until. Limitations
of space prevent me from discussing other accounts of the semantics of aspectual adverbs in the literature in any detail here. The interested reader is referred
to Smessaert & ter Meulen (2004) for a similar account with some minor differences, and our rebuttal of different semantic theories of aspectual adverbs.
3 Prosodie meanings of aspectual adverbs

In English aspectual adverbs may be effectively used to express not only factual information about the onset of an event or its termination, but, if uttered
with marked high pitch prosody, they also convey the speaker's attitude regarding the flow of events or its perceived speed. Other languages may express
this mix of factual and subjective information differently with aspectual verbs,
using, for instance, lexical composition (Dutch) or word order variation (German). Interesting issues of linguistic variability may arise in studying the ex-
438
pressive range of such mixed temporal information, but this paper is limited to
very simple cases of English aspectual adverbs. If a speaker feels annoyed or
surprised that something is not yet the case, he may of course describe his
attitude explicitly stating in a full clause that he is annoyed, surprised or whatever at its not yet being the case. But in English such attitudes may be very
effectively indicated with high pitch prosody on aspectual adverbs. Though the
basic externally negated aspectual adverb not yet may accept certain marked
prosody other than high pitch, its logically equivalent counterpart with internal
negation still not more readily accepts marked high pitch, here simply indicated with capital letters as STILL not. Pitch marking of expressions is well
known from studies on focus and information structure, where high pitch
serves to demarcate new information from what is already assumed, given or
otherwise included in the common ground. Along similar lines, the informational purpose of pitch marking aspectual adverbs is to present the subjective
content as new, relegating all other supposedly factual information to the
background, as if it were already incorporated into the common ground and
familiar, hence not at issue in the communication. The speaker uses pitch
marked STILL not when he had expected for one reason or another the described, topical state to have started earlier and wishes to express his dismay or
surprise at it not yet being the case. For instance, if the speaker says that John
is STILL not asleep, he must counterfactually have expected John to be asleep
by now, hence to have started sleeping or to have fallen asleep in the past. To
capture this counterfactual expectation of the speaker in terms of a truth functional operator, a modal operator ALT taking as arguments the speaker (sp),
the current reference time and a set of conditions, is interpreted as quantifying
over ALTernatives to the current course of events, subjectively dependent
upon the speaker's epistemic state.
In (2a) the future (r0< rO endpoint r of the continuing current (s r0) negative phase of the P-state (P'(e, x, - )) should be past (ri < r 0 ) according to the
speaker's alternative course of current events. Similarly, if a speaker pitchmarks alREADY, he indicates that the actual onset of the current positive phase
of took place earlier than he had expected. Again, the pitch marked version
STILL is a polarity counterpart of STILL not, both forward looking towards a
later alternative polarity transition, and no LONGER lexicalizes the pitch
marked version of not anymore in (2d), looking back to the past transition,
considered early. The analysis is more fully explained in ter Meulen (2000),
here simplified considerably. Bold face indicates the primary focus information, even in contexts where the remainder of the content is new to the recipient and hence may be considered secondary focus.
(2)
The semantics of pitch marked aspectual adverbs,

(a)
[INFL STILL not [w x />]]] =>

[r0, ri, s, I '(s, , -) & s 3 r0 & r, 3 END (F(s, x, -)) & r(,< r, &
[IP
Cohesion in temporal context
439
UNTIUr,, P'(s, X, -)) 2 r0 & ALT(sp, r0, [ r, < r0 & SINCE(ri, P'(s, x, +)) 3 r,,])]
(b)
(c)
(d)
[ip [infl alREADY [vp ]]] =>

[r0, n, s, I P'(s, x, +) & s 3 r0 & ri 2 START (P'(s, x, +)) & r ! < r0 &
SINCE(ri, P'(s, x, + ) ) D t , 4 ALT(sp, r0, [ r0< & UNTIL(ri, P'(e, x, -)) 2 r0])]
[ipX [infl STILL [w / ] ] = >
[r0> r,, s, I P'(s, , +) & s 2 r0 & ri 2 END ('(s, , +)) & r,,< r, &
UNTILCn, P'(s, x, +)) 2 ro & ALT(sp, r, [ r, < r0 & SINCEO, P'(s, x, +)) 2 r0])]
[ip x [infl no LONGER [vp x />]]] =>
[r0, , s, I P'(s, x, -) & s 2 Io & ri 2 START ('(s, x, -)) & r , < r0 &
SINCE(ri, P'(s, x, -)) 3. io & ALT(sp, r0, [ r0< ri & UNITLCn, P'(e, x, -)) 3 r0])]
Assuming the semantics of aspectual adverbs in (1) and the additional information their pitch marked variants express in (2), the remainder of this paper is
concerned with applications of their semantic properties in question-answer
dialogue and the constraints they induce on the accommodation of presupposed information in a given context. The paper concludes with a discussion of
the notion of logical consequence in a dynamic semantics with structured information states, arguing that only states described by perfect tense clauses
with locally consistent presuppositions may always be coherently asserted as
new information or primary focus, since they lack the dynamic power to affect
context-change. Different contexts impose assertability constraints associated
with simple past tense clauses referring to events. Asserting information as
primary focus that is already presupposed, entailed or otherwise part of the
common ground is much more constrained, if preservation of coherence is
required. Sharing information is best represented by an onion-layered common
ground with constraints, relating to cognitive effort or complexity, on how
deep down into the different layers speakers may have access to revise their
information.
4 Aspectual adverbs in dialogue

Proper or felicitous answers to polarity questions with aspectual adverbs must,
as usual, share the presuppositions of the question, as in (3a, b). If the presuppositions of the questioner A are not acceptable to the answerer , must use
another, stronger form of negation, i.e. denial as in (3d), to cancel A's presuppositions. In (3a, b) still and no longer share the presupposed information that
John was asleep before Mary arrived, not shared by already in (3c), and cancelled by not yet in (3d). The questioner A requests an update from about the
temporal relation between the endpoint of John's sleeping and Mary's arrival,
positively suggesting John woke up after Mary's arrival.
(3)
(a)
A: Was John still asleep, when Mary arrived?
(b)
B: No, he was not asleep anymore.
440
(c)
: * No, he was already asleep.
(d)
: No, he had not (even) fallen asleep yet.
It should be evident from the relative incoherence of (3a+c) that temporal

presuppositions of polarity questions cannot simply be cancelled by asserting
an answer with partially conflicting presuppositions. A's question (3a) asks
to supply only information about the temporal relation between the two reference times it introduces, i.e. (ri< r 2 o r r 2 < r 0 , as indicated in bold face below
in (4a). The remaining content of (3a) must be considered common ground, i.e.
information A assumes is already shared by A and B. The aspectual adverb
still in (3a) partitions the content of A's question into the narrow interrogative
focus information ?([< r 2 o r r 2 < r 0 , relegating all remaining content to background information, with which A supposes to agree already. The questioned focus information is indicated in (4) with ?(ri< r 2 o r r 2 < r,), informally
meaning that the truth-value of ( r ^ r 2 ) or (r 2 < r ^ , is requested in all situations
which support the information in the common ground, constituting the antecedent. The formal semantics of conditionals with background information as
antecedent and questioned content as consequent needs to be made explicit,
but this would lead us much beyond the intended scope of the present paper. 1
In Situation Semantics questions could simply be implemented with a parameter for the polarity in the event-type to be resolved into true or false by an
update with the answer. Whatever precise form the dynamic semantics of interrogatives may take, the insight should be implemented that presuppositions
and background information constitute constraints on the situations or worlds
that must be satisfied before the interrogative focus information is evaluated.
Clearly, updating a context with the interrogative focus requires a more constrained accommodation procedure of its presuppositions than has customarily
been adopted.
(4)
(a)
A: [I P John [INFL still [ w sleep (j)] [ADVP when M. arrived]]] =>

[ro, , 2, s, j, m I sleep'(s, j, +) & s 3r 0 &
r, g END (sleep'(s, x, +)) & r0< r, & UNTILCr,, (sleep'( s, x, +))) 2 r0 &
r2 2 arrive'(m) => ?(ri< r 2 or r2< ri)]
In negatively answering (3a) with (3b) updates A's information state only
by asserting as new information that r t < r 2 , i.e. that Mary arrived after John
had woken up. Subsequently A revises his information by eliminating the other
option he held to be possibly true i.e. r 2 < r r
To cancel A's presuppositions by answering with (3d/5d), systematically
resets the positive polarity parameter in the presupposed information of
See Asher & Lascarides (1998) for an excellent exposition of various dynamic semantics of
questions in dialogue. Their SDRT account would square well with my analysis.
441
(3a/4a), restructuring its content as if its presuppositions were asserted focus

information and relegating the information that Mary arrived to shared information in the common ground, thereby identifying its reference time r 2 with
the given reference time r 0 . Accordingly, B ' s answering with (3d/5d) constitutes a refusal to make the temporal relation between r and r 2 the issue under
consideration.
(5)
(d)
B: [n> john [INFL not yet [ w sleep'(j)]]] =>

[r0, , 2, s, j, I sleep'(s, j, -) & s 3 ro & r i 2 END (sleep'(s, j, -)) & ro< ri &
UNTEL(ri, (sleep'( s, j, -))) 3 r0 & r 2 3 arrive'(m) & r(> = r2 ]
The incohesive answer in (3c/6c) creates havoc with

ground/interrogative focus partitioning induced by A's use of still.
(6)
(c)
the
back-
[IP john [itm. already [wf sleep (j)]]] =>

[ro, ri, s, j I sleep'(s, x, +) & s 20 &
r a START(sleep'(s, x, +)) & r,< r0 & SINCE(ri, (sleep'(s, x, +))) 3 r0]
In (3c/6c) informs A that A is right on only some presupposed information,

i.e. that John was asleep at the given reference time r 0 , i.e. sleep'(s, x, +) &
s 2 r 0 . But also indicates that A was wrong in questioning the end point of
the state s. corrects A by adding information about the past starting point of
the state e, i.e. John's falling asleep. Clearly, A had not intended to make an
issue of the temporal relations of the starting point, although he obviously also
secondarily presupposed it. Such secondary presuppositions of presuppositions
cannot be cancelled by asserting (3c/6c), since B's correction of A ' s presuppositions creates incoherence in the context. If secondary presuppositions must
be addressed, they must first be brought into the dialogue by restructuring the
relevant information in the common ground using other, more direct communicative means. Secondary presuppositions are otherwise inaccessible for revision, constituting a layer within the common ground that is considered a closed
issue.
Characteristically, English aspectual adverbs may also be used interrogatively in dialogue with marked prosody, in B ' s reaction to A's assertion describing a current factual state. In (7b) the questioner seeks A's agreement
with B ' s subjective assessment of the timing of what A described as the factual
course of events, i.e. with the focus information expressed with the high pitch
STILL (cf. 2c).
(7)
(a)
(b)
(c)
A: John is sleeping,
B; A1READY?
B: STILL?
442
In (7a, b) conditionally accepts A's assertion into the common ground and
solicits A's agreement with B's subjective perception that John was early to
fall asleep. In uttering (7b) wants A to agree with B's first envisaged, but
now counterfactual view that John should have been awake by now. In responding with the question (7c) wants A to share B's judgement that John is
late to wake up. To develop the dynamic semantics of this interrogative usage
of aspectual adverbs in discourse would take us much beyond the present paper, but in outline it should be quite clear already that focus information may
be interrogatively used. Pitch-marked aspectual adverbs apparently manage to
do a lot of dynamic work in (re)structuring factual and subjective information
states, partitioning new information into an update of the common ground and
primary focus.
5 Coherence and constraints on accommodation

If (8a) is assumed to constitute a coherent continuous monologue, the presupposition ( John fell asleep) of the first clause (John was already asleep) cannot
apparently be felicitously asserted again in a subsequent simple past tense
sentence, even though the information it contains must either already be part of
the common ground or properly introduce a new, later reference time at which
John fell asleep for a second time. Analogously, in (8b) asserting first that John
was no longer asleep prevents the speaker to coherently continue by adding
already presupposed information. It is not straightforwardly possible to interpret (8b) as referring to two distinct episodes of John's sleeping, separated by a
period of his being awake. To do so, either the presuppositional adverb again
must be used to make the disjoint reference to two events explicit or the monologue itself must be temporally discontinuous.
(8)
(a)
?* John was already asleep. John fell asleep,
(b)
?* John was no longer asleep. John was asleep.
What may be accommodated into the common ground or asserted as new,

primary focus is apparently effectively constrained by aspectual adverbs. For
instance, (8a) cannot be interpreted to mean that John was asleep twice, coercing accommodation of the presupposition of the second clause that he was not
asleep for some preceding period. Cohesion of the discourse in (8 a, b) may of
course be easily restored by explicitly asserting a polarity reversal (he woke
up/he fell asleep) to shift to a new reference time in (9 a, b). Only such asserted information can coerce disjoint temporal reference of the clauses to
ensure that the second clause no longer constitutes the presupposition of the
first clause, but describes a new, later event of the same event type.
(9)
(a)
John was already asleep. He woke up and fell asleep (again).
(b)
443
John was no longer asleep. He fell asleep (again). He was asleep.
The presuppositions of adverbs such as again serve to coerce such temporal

disjoint reference, but their dynamic force is by itself not sufficiently strong to
restore cohesion in (8a) or (8b). In the perfect clause (10), that describes the
state resulting from the corresponding event, again takes widest scope to introduce two events of the same type, i.e. John falling asleep, where the current
reference time is included in the state resulting from the later event. The aspectual adverb already in (10) indicates that this second event occurred earlier
than expected. In English aspectual adverbs are odd, if not downright unacceptable, with again in the lexically stative clauses (11) and (12). In the richer
adverbial structure of Dutch the morphologically complex alweer/alreadyagain is perfectly acceptable in either the aspectually stative perfect clause
(10) or the lexically stative (11), and even in the composition of aspectual
adverbs with again in (12).
(10)
E:
John has already fallen asleep again.
D:
Jan is alweer in slaap gevallen

J. is already-again in sleep fallen-perf part
??John is already asleep again.
Jan slaapt alweer.
J. sleeps already-again
*?John is {again no longer asleep/ no longer asleep again}
Jan slaapt (al)weer niet meer.
J. sleeps (already) again no more.
(11)
E:
D:
(12)
E:
D:
Assuming an overall constraint of local coherence of contexts, updating the

common ground by presupposition accommodation must hence be sharply
distinguished from updating it by asserting new, primary focus information.
Asserting information as if it were new, even though it is already entailed or
presupposed by the immediately preceding clause, creates an incoherent context. In other words, updating with asserted information is only possible when
the information to be added is genuinely new in the current context. Accordingly, coercion by context shift is not a realistic strategy in natural language
semantics to ensure that updating with asserted information remains always
possible, as preservation of coherence prevails.
6 Logical consequence in dynamic semantics

The classical notion of logical consequence in DRT in (13) explains why dynamic temporal presuppositions cannot be asserted at the reference time of the
subsequent state, even though they are entailed, since logical entailments may
never introduce new reference times.
444
(13)
DRT Definition of logical consequence

Let , K' be pure (...) DRSs. K' is a logical consequence of ( 1= ') iff. the following
condition holds: Suppose M is a model and f is a function from Uj u Fr(K) Fr(K') into
U[y[. s.t. M l=f K, then there is a function g 2 U K ' f s u c h ' h 3 ' M l=g K'.
(from Kamp 1993: 305.)
When an event must be referred to again in a later context, one may always do
so using the static perfect tense in a continuous, coherent monologue, since
perfect tense clauses are immune to shifting the reference time, as in (14 a, b).
The perfect state that results once an event terminates, endures forever after
and referring to it will never affect the reference time. For instance, that John
had fallen asleep remains true, not only during John's sleep, but also at any
arbitrary later moment after he woke up. The semantics of the past perfect
requires merely that the event causing the perfect state must precede the
speech time as well as the contextually determined reference time. Hence in
(14a) the second past perfect clause he had fallen asleep entails the existence
of an arbitrary past event of John's falling asleep. To understand (14a) as cohesive discourse, that arbitrary event is identified as onset of the state described by the first simple past tense clause John was already asleep, and it is
claimed by the speaker to have occurred early. Logically, the first clause entails the second one, but the second only entails the existence of some prior
event of falling asleep.
(14)
(a)
John was already asleep. He had fallen asleep,
(b)
John was no longer asleep. He had been asleep.
Preserving coherence, the assertability constraints associated with states described by perfect tense clauses must be distinguished from those associated
with states described by simple past tense clauses. Only the perfect tense
clauses may always coherently be asserted as new information, regardless of
the common ground, even when their content is already logically entailed by
immediately preceding clauses in the discourse.
This outline of the dynamic semantics of tense, aspect and aspectual adverbs and its associated, classical notion of logical consequence still needs to
be supplemented with a module containing natural deduction style inference
rules for temporal reasoning. An inference rule 'PERF introduction' should
characterize which transformations of contexts updated by past tense clauses
are required to report their content using past perfect tense (cf. ter Meulen
1995, ter Meulen 2000). In such a natural deduction system validity of temporal reasoning may be characterized without having to appeal to slippery notions such as a 'normal' course of events, to rhetoric relations or to common
sense about what the world is or should be like or any understanding of the
psychologically slippery notion of causality.
445
References
Beaver, D. (1997): Presupposition. In: J. van Benthem, and A. ter Meulen (eds) (1997).
Handbook of Logic and Language. Elsevier Science, Amsterdam, & MIT Press,
Cambridge, 939-1008.
Kamp, H. and U. Reyle (1993): From discourse to logic. Kluwer, Dordrecht.
Lascarides, A. and N. Asher (1993): Temporal Interpretation, Discourse Relations and
Commonsense Entailment, Linguistics and Philosophy 16.5, 437-493.
Lascarides, A. and N. Asher (1998): Questions in Dialogue. Linguistics and Philosophy, 21, 237-309.
ter Meulen, A. (1995): Representing Time in Natural Language. The dynamic interpretation of tense and aspect. MIT Press, Cambridge,
ter Meulen, A. (2000): Chronoscopes: the dynamic representation of facts and events.
In: J. Higginbotham et al. (eds). Speaking about events. Oxford U.P., 151-168.
Smessaert, H. and A. ter Meulen (2004): Dynamic reasoning with aspectual adverbs.
Linguistics and Philosophy 27.2, 209-261.
van Eijck, J. and H. Kamp (1997): Representing discourse in context. In: J. van Benthem, and A. ter Meulen (eds). Handbook of Logic and Language. Elsevier Science,
Amsterdam, & MIT Press, Cambridge, 179-237.
Hooi Ling S oh and Meijia Gao

(Minneapolis)
Mandarin sentential -le, perfect

and English already
1 Introduction
In previous work (Soh and Gao 2006), we have argued that Mandarin sentential -le is a transition marker, and that it triggers a presupposition about an
immediate past situation in opposition to the one described by the sentence.
The present paper further explores and clarifies the meaning of sentential -le
with specific focus on its relation to perfect and English already. We propose
that sentential -le encodes two types of information: (i) an assertive meaning
that the situation described is realized prior to a reference time; (ii) a presupposition that a situation opposite to the one described by the sentence exists immediately before the point of realization (compare J.W. Lin 2003, Lai 1999).
We claim that the reference time is the speech time, unless the particle jiu is
used, in which case, it may be a specified time in the past or future. Evidence
for our proposal comes from the readings associated with sentential -le in
different situation types, and a restriction in the occurrence of sentential -le
with zhi 'only' and budao 'less than'. Previous authors have proposed a connection between sentential -le and perfect, noting similarities between sentential -le and the English perfect in terms of the presence of a result state and a
continuative reading (e.g. Li, Thompson and Thompson 1982, Q. Zhang 1997,
J.W. Lin 2003). We clarify the similarities that have been noted and show that
neither the result state nor the continuative reading is entailed by sentential -le.
We would like to thank David Beaver, Bridget Copley, Kai von Fintel, Jeanette Gundel, Nancy
Hedberg, Irene Heim, Michael Kac, Ding-cheng Li, Jimmy Lin, Jo-wang Lin, Ted Pedersen,
Norvin Richards, Susan Rothstein, Christina Schmitt, Carlota Smith, Shiao-Wei Tham and Ron
Zacharski for discussions at various stages of this paper. We also thank the participants at the
Workshop on Event Structures in Linguistic Form and Interpretation at Leipzig (March 17-19,
2004) for questions and comments. We are grateful to Jeanette Gundel for reading and commenting on our paper, and David P. Slovut for proof-reading our paper. All errors are ours.
This research was funded by a grant from the Chiang Ching-Kuo Foundation for International
Scholarly Exchange and a University of Minnesota Single Semester Leave, both awarded to
Hooi Ling Soh. Their support is gratefully acknowledged.
448
We suggest that the result state reading is a relevance-based implicature, while

the continuative reading derives from the way the interval expressed by a duration phrase is temporally located (following Portner 2003). We suggest that
sentential -le shares its assertive meaning with perfect, and its presupposition
with English already. The established connection between sentential -le and
perfect enables us to provide further evidence for the claim that completion is
not part of the meaning of perfect (Iatridou et al. 2001, Portner 2003). Our
analysis also supports the claim that already presupposes the existence of a
prior situation that is in opposition to the one described by the sentence (Lbner 1989, 1999, van der Auwera 1993, but see Mittwoch 1993).
Before we proceed, it is important to distinguish -le when it appears at the
end of the sentence from when it appears immediately after the verb. 1 We refer
to the former as sentential -le and the latter as verbal -le.
(1)
(a)
(b)
Women daoda shan-ding

le.2
we
reach mountain-top LE
'We reached the top of the mountain.'
Women daoda le shan-ding.
we
reach LE mountain-top
Although it remains controversial whether verbal -le and sentential -le are
instances of the same morpheme or distinct morphemes, we assume the result
of our work in Soh and Gao (2006), where we argue that they are distinct morphemes, following Li and Thompson (1981), Ross (1995), Smith (1997), Sybesma (1999), Zhang (1997) (see Rohsenow 1978, Shi 1990, Huang and Davis
1989, Kang 1999, Lin 2003 for an alternative view). In particular, we assume
that verbal -le is a perfective aspect marker (following Li and Thompson 1981,
Ross 1995, Smith 1997, Zhang 1997). Our focus in this paper is on the status
of sentential -le.
The paper is organized as follows: In section 2, we present the readings associated with sentential -le in different situation types, and a restriction found
between sentential -le and zhi 'only' and budao 'less than'. Our analysis is
presented in section 3. In section 4, we clarify the similarities that have been
noted between sentential -le and perfect, relating to result state and continuative reading, and show that the relevant readings are not entailed by sentential
Verbal -le and sentential -le also may appear simultaneously. See Soh and Gao (2006) for an
analysis.
(i) Women daoda le shan-ding
le.
we
reach LE mountain-top LE
The following abbreviations are used in glossing examples: CL
classifier; POSS possessive marker; PROG progressive aspect; DOU plural
marker; Q question particle; BA extraposition marker; ASP aspect marker.
449
-le. In section 5, we discuss the relation between sentential -le, perfect and
already. The implications and conclusions are presented in section 6.
2 Data
2.1 Readings associated with sentential -le
In the following subsections, we demonstrate that sentential -le gives rise to a
completive reading when the sentence describes a telic situation, and an inchoative reading when the situation is atelic. The completion or the beginning
of the event occurs before the speech time. It may occur prior to a specified
past time or future time when the sentence contains the particle jiu. We show
that in the absence of the particle jiu, sentential -le may appear with a time
adverbial indicating a past time (except when the sentence describes a state, for
reasons unknown to us), but not with a time adverbial expressing a future time.
2.1.1 States
Sentential -le appears freely in stative sentences and provides a change of
state/inchoative reading to the sentence as shown in (2) and (3).
(2)
(a)
(b)
Ta xiang
baba.
he resemble dad
'He resembles dad.'
Ta xiang
baba le.
he resemble dad LE
'He resembles dad now, (which he did not before).'
(3)
(a)
(b)
Ta danxin ta de anquan.
he worry he Poss safety
'He worries about his safety.'
Ta danxin ta de anquan le.
he worry he Poss safety LE
'He worries about his safety, (which he did not before).'
The change of state is interpreted as having occurred before the speech time.
The change of state may hold before a future time or a past time. In such cases,
the particle jiu is required (cf. Li and Thompson 1981:242-256, Lai 1999).
Note that with states, sentential -le may not appear with a time adverbial without jiu.
(4)
(a)
Ta zai
guo liang nian jiu xiang
baba le.
he further pass two year JIU resemble dad LE
'He will resemble dad in two years.'
450
(5)
(b)
*Ta zai
guo liang nian xiang
baba le.
he further pass two year resemble dad LE
(a)
Wo qu-nian jiu danxin ta de anquan le.

I last-year JIU worry he Poss safety LE
had begun worrying about his safety last year.'
*Wo qu-nian danxin ta de anquan le.
I last-year worry he Poss safety LE
(b)
Evidence that the change of state occurs before (as opposed to at) the speech
time or a specified time in the future or past is provided by the acceptability of
the following sentences (expanding on Lai 1999: 634-635):
(6)
(a)
(b)
(c)
Ta zai bangongshi le. Shishishang, ta yijing lai-le

wu fengzong le.
he at office
LE in fact
he already come-Perf five minute LE
'He is in his office now. In fact, he arrived five minutes ago.'
Ta ba dian
jiu zai bangongshi le. Shishishang ta shi qi
dian
lai
de.
he eight o'clock JIU at office
LE in fact
h e b e seven o'clock come DE
'He had already been in his office at eight o'clock. In fact he came at seven.'
(Lai 1999: 634-635, modified version of her example (13))
Ta mingtian ba dian
jiu zai bangongshi le.
he tomorrow eight o'clock JIU at office
LE
Shishishang ta dasuan qi
dian
dao de.
in fact
he plan seven o'clock arrive DE
'He will have already been in his office at eight o'clock. In fact, he plans to arrive at
seven.'
2.1.2 Achievements
When sentential - l e occurs with an achievement, it contributes a completive
reading to the sentence. This is, however, difficult to demonstrate as the sentences with and without sentential -le seem to have the same interpretations as
shown in (7) and (8).
(7)
(a)
(b)
Tarnen ganggang daoda shan-ding.3

they just
reach mountain-top
'They just reached the top of the mountain.'
Tamen ganggang daoda shan-ding
le.
they just
reach mountain-top LE
'They have just reached the top of the mountain.'
A better minimal pair would be one without ganggang 'just' to ensure that the completive
reading does not come from the adverb. However, the sentence tamen daoda shan-ding 'they
reached the top of the mountain' does not sound natural.
(8)
(a)
(b)
451
Ta zhong tou jiang. (Wo de anwei

jiang)
he hit
first prize I get consolation prize
'He hit the jackpot. (I got the consolation prize.)'
Ta zhong tou jiang le.
he hit
first prize LE
'He has hit the jackpot.'
The events described in (7) and (8) are interpreted as having been completed
before speech time. The event may be completed before a time distinct from
the speech time, either in the past or in the future. In these cases, jiu is required. Although (9b) is acceptable without jiu, it elicits a reading in which the
event is completed before speech time, and the time adverbial locates the event
time. In order to obtain an interpretation where the event is completed before a
past reference time indicated by the time adverbial, // is necessary as in (9a). 4
(9)
(a)
(b)
Tamen zuotian jiu daoda shan-ding

le.
they yesterday JIU reach mountain-top LE
'They had reached the top of the mountain yesterday.'
Tamen zuotian daoda shan-ding
le.
they yesterday reach mountain-top LE
'They have reached the top of the mountain yesterday.'
Similarly, in (10a), the event can be interpreted, with jiu, as having been completed before a future reference time provided by the time adverbial. Without
jiu, the sentence is unacceptable as shown in (10b).
(10)
(a)
(b)
Tamen mingtian jiu hui daoda shan-ding
le.
they tomoiTow JIU will reach mountain-top LE

'They will have reached the top of the mountain tomorrow.'
T a m e n mingtian hui daoda shan-ding
le.
they tomorrow will reach mountain-top LE
2.1.3 Accomplishments
The situation is more complicated for accomplishment events. That which is
generally regarded as accomplishment sentences fall into two types in Mandarin, which differ as to how they interact with the perfective aspect marker
4
That the time adverbial can locate a past reference time is evidenced by the acceptability of (),
(i) Wo kending tamen liang-dian jiu daoda shan-ding
le.
I certain they two o'clock JIU reach mountain-top LE
Tamen daodi zhenzheng shi shenme shihou daoda de, wo ke bu-shi hen qingchu.
they actual exactly
be what time reach DE I but not-be very clear
am certain that they had reached the top of the mountain at two.
Exactly what time they actually reached the top of the mountain, I am not sure.'
452
(verbal -le) (Chu 1976, Tai 1984, Smith 1994, 1997, Sybesma 1997, 1999,
Klein et. al 2000, Soh and Kuo 2005). In one type, the presence of perfective le does not necessarily indicate completion. The event can be terminated without reaching the inherent end point. This is shown by the fact that it is not
contradictory to conjoin the first sentence in (11a) and ( l i b ) with an assertion
that the event is not complete (Tai 1984).5
(11)
(a)
Wo zuotian
(b)
I yesterday write LE one-CL letter but not write-finish

started writing a letter yesterday, but I didn't finish writing it.'
Ta hua-le yi-fu hua, keshi mei hua-wan.
he draw-LE one-CL picture but not draw-finish
'He started drawing a picture, but he didn't finish drawing it.'
xie
le yi-feng xin, keshi mei xie-wan.
In another type, with the presence of a completive marker such as wan 'finish'
after the verb, verbal -le must indicate that the event has been completed, and
not merely terminated (Tai 1984, Smith 1994, 1997).6 This is shown by the
fact that an accomplishment sentence with a completive marker and verbal -le
cannot be followed by an assertion that the event is not complete.
(12)
(a)
#Wo zuotian xie-wan

le yi-feng xin, keshi mei xie-wan.7
I yesterday write-finish LE one-CL letter but not write-finish
wrote a letter yesterday, but I didn't finish writing it.'
Given that completion is not necessary, one may wonder if xie 'write' and other verbs of this
group should be considered to be activity predicates (cf. Tai 1984) rather than accomplishment
predicates. There is reason to maintain that xie 'write' and other verbs of this group are accomplishment predicates. Soh and Kuo (2005) show that completion is required with some created
objects, namely those that cannot be considered an instance of the object until the creation
process has reached its inherent end point. There is a contrast between the created object yifeng xin 'a letter' andyi-ge zi 'a character'.
(i)
Ta xie-le
yi-feng xin/ #yi-ge zi,
keshi mei xie-wan.
he write-LE one-CL letter/one-CL character but not write-finish
'He wrote a letter/a character, but he didn't finish writing it.'
While a partially written letter can be considered an instance of a letter, a partially written
character cannot be considered an instance of the relevant character. Soh and Kuo (2005) propose that in creation events, -le indicates the completion of the event leading to the creation of
an object that qualifies as the relevant object.
It is possible that sentences with the completive marker wan 'finish' denote achievements
rather than accomplishments. In fact, there is reason to believe that they may denote achievements given that they may not appear with the progressive.
(i) *Ta zai xie-wan
yi-feng xin.
(ii) *Ta zai hua-wan yi-fu
hua.
he Prog write-finish one-CL letter
he Prog draw-finish one-CL picture
'He is finishing a letter.'
'He is finishing a picture.'
However, we include them under accomplishments following previous work.
# here indicates that the conjunction is unacceptable, though each conjunct is acceptable on its
own.
(b)
453
#Ta zuotian hua-wan

le yi-fu
hua, keshi mei hua-wan. 8
he yesterday draw-finish LE one-CL picture but not draw-finish
'He drew a picture yesterday, but he didn't finish drawing it.'
Like accomplishment sentences with a completive marker, accomplishment

sentences with a numeral object must also be completed when verbal -le is
present (Soh and Kuo 2005, Liu 2003). Soh and Kuo (2005) suggest, following Li and Thompson (1981: 132) and Gundel et. al (1993: 290-291) that yi
'a/one' is ambiguous between a numeral and an indefinite determiner in Mandarin. The effect of numeral objects is therefore not easily detected with the
numeral yi 'one'. The numeral object liang-ge dangao 'two cakes' and liang-fu
hua 'two pictures' are contrasted with definite noun phrase objects in (13) and
(14). Verbal -le adds a completive reading to the sentence that contains a numeral object, but only a terminative reading in the sentence without a numeral
object.
(13)
(a)
(b)
(14)
(a)
(b)
Ta chi le na-ge dangao, keshi mei chi-wan.

he eat LE that-CL cake
but not eat-finish
'He started eating that cake, but he did not finish eating it.'
#Ta chi le liang-ge dangao, keshi mei chi-wan.
he eat LE two-CL cake
but not eat-finish
'He ate two cakes, but he did not finish eating them.'
Tahua le na-fu hua, keshi mei hua-wan.
he draw LE that-CL picture but not draw-finish
'He started drawing that picture, but he didn't finish drawing it.'
#Ta hua le liang-fu hua, keshi mei hua-wan.
he draw LE two-CL picture but not draw-finish
'He drew two pictures, but he didn't finish drawing them.'
Sentential -le gives rise to different readings in these two types of accomplishment sentences. In accomplishment sentences that have an explicit completive marker and the ones that contain a numeral object, sentential -le provides a completive reading, patterning like verbal -le.
(15)
(a)
Ni he wan tang.
you drink finish soup
'You finish drinking the soup.'
The presence of wan 'finish' after the verb does not always indicate that the event is completed. This is shown in (i).
(i) Ta yao wo xie-wan
zhe-feng xin.
he want I write-fmish this-CL letter
'He wants me to finish writing this letter.'
454
(b)
(c)
(16)
(a)
Ni he wan le tang,
you drink finish LE soup
'You finished the soup.'
Ni he
wan tang le.
you drink finish soup LE
'You have finished drinking the soup.'
Tahua le san-ge quanquan. (Wohua liang-ge.)
he draw LE three-CL circle
I draw two-CL
'He draws three circles. (I draw two.)'
(b)
Ta hua le san-ge
quanquan.
he draw LE three-CL circle
'He drew three circles.'
(c)
Ta hua san-ge quanquan le.

he draw three CL circle
LE
'He has drawn three circles.'
In other accomplishment sentences, sentential -le indicates that the event has
begun. Whether the event has terminated or not is left open. This contrasts
with sentences with the perfective -le, which has a terminative reading.
(17)
(a)
(b)
(c)
(18)
(a)
(b)
(c)
Taxie na-feng xin.

he write that-CL letter
'He writes that letter.'
Ta xie le na-feng xin.
he write LE that-CL letter
'He wrote that letter.
Ta xie na-feng xin le.
he write that-CL letter LE
'He has started writing that letter.'
Tahua na-fu
hua.
he paint that-CL picture
'He paints that picture.'
Ta hua le na-fu hua.
he paint LE that-CL picture
'He painted that picture.'
Ta hua na-fu hua
le.
he paint that-CL picture LE
'He has started painting that picture.'
The difference between verbal -le and sentential -le can be brought out by a
sentence that explicitly indicates the non-termination of the event. In (19a), for
example, the sentence with verbal -le cannot be followed by an assertion that
the event of letter writing is not terminated. This contrasts with (19b) with
sentential -le, which can be followed by an assertion that the event is not terminated.
(19)
455
(a)
#Ta jintian zaoshang xie le na-feng xin. Xie dao xianzai hai bu ken
ting.
he today morning write LE that-CL letter write till now still not willing stop
'He wrote the letter this morning. He has been writing till now and is still unwilling
to stop.'
(b)
Ta jintian zaoshang xie na-feng xin le. Xie dao xianzai hai buken
ting,
he today morning write that-CL letter LE write till now
still not willing stop
'He started writing the letter this morning. He has been writing till now and is still
unwilling to stop.'
Similar to the previous situation types, (15c) and (17c) respectively describe
that the event is completed or has started before speech time. In order for the
sentence to describe a situation in which the completion or the beginning of the
event holds prior to a time distinct from the speech time, jiu is necessary.
(20)
(21)
(a)
Ni zuotian
(b)
you yesterday JIU drink finish soup LE

'You had finished drinking the soup yesterday.'
Ni zuotian
he
wan tang le.
you yesterday drink finish soup LE
'You finished drinking the soup yesterday.'
(a)
(b)
jiu he
wan tang le.
Ta zuotian jiu xie na-feng xin le.

he yesterday JIU write that-CL letter LE
'He had started writing that letter yesterday.'
Ta zuotian xie na-feng xin le.
he yesterday write that-CL letter LE
'He started writing that letter yesterday.'
Although (20b) and (21b) are acceptable without jiu, they only have the interpretation that the completion or the beginning of the event is prior to the
speech time, rather than the past time specified by the time adverbial. In (22a)
and (23a), with jiu, the event can be interpreted as having been completed or
begun before a future reference time provided by the time adverbial. Without
jiu, the sentence is unacceptable as shown in (22b) and (23b).
(22)
(a)
(b)
(23) (a)
(b)
Ni mingtian jiu huihe

wan tang le.
you tomorrow JIU will drink finish soup LE
'You will have finished drinking the soup tomorrow.'
*Ni mingtian hui he
wan tang le.
you tomorrow will drink finish soup LE
Ta mingtian jiu hui xie na-feng xin le.
he tomorrow JIU will write that-CL letter LE
'He will have started writing that letter tomoiTow.'
*Ta mingtian hui xie na-feng xin le.
he tomorrow will write that-CL letter LE
456
2.1.4 Activities
In combination with activities, sentential -le expresses the idea that the event
has begun and may or may not have ended.
(24)
(a)
(b)
(25)
(a)
(b)
Ta you yong. (Wo pao bu.)

he swim swim I ran step
'He swims. (I ran.)'
Ta you yong le.
he swim swim LE
'He started swimming.'
Ta da ta de meimei.
he hit he Poss sister (younger)
'He hits his younger sister.'
Ta da ta de meimei le.
he hit he Poss sister LE
'He started hitting his sister.'
As shown below, sentential -le does not entail termination, in contrast to the
perfective -le.
(26)
(a)
(b)
(27)
(a)
(b)
#Ta you le yong. Cong zaoshang you dao xianzai hai zai you.
he swim LE swim from morning swim till now
still Prog swim
'He swam. Starting from this morning till now, he is still swimming.'
Ta you yong le. Cong zaoshang you dao xianzai hai zai you.
he swim swim LE from morning swim till now still Prog swim
'He started swimming. Starting from this morning till now, he is still swimming.'
#Ta ma le ta de haizi. Cong zaoshang ma dao xianzai hai zai ma.
he scold LE he Poss child from morning scold till now
still Prog scold
'He scolded his child. Starting from this morning till now, he is still scolding.'
Ta ma ta de haizi le. Cong zaoshang ma dao xianzai hai zai ma.
he scold he Poss child LE from morning scold till now
still Prog scold
'He started scolding his child. Starting from this morning till now, he is still
scolding.'
As in other situation types, (24b) and (25b) describe situations that begin prior
to the speech time. In order for the sentence to describe a situation that begins
at a reference point before speech time or after speech time, jiu is necessary.
Again, note the difference in interpretation between (28a) and (28b), and the
contrast in acceptability between (29a) and (29b).
(28)
(a)
Ta zuotian jiu ma ta de
haizi le.
he yesterday JIU scold he Poss child LE
'He had started scolding his child yesterday.'
(b)
(29)
(a)
(b)
457
Ta zuotian ma ta de haizi le.

he yesterday scold he Poss child LE
'He started scolding his child yesterday.'
Ta mingtian jiu hui ma ta de haizi le.
he tomorrow JIU will scold he Poss child LE
'He will have started scolding his child tomorrow.'
*Ta mingtian hui ma ta de haizi le.
he tomorrow will scold he Poss child LE
2.1.5 Summary
Table 1 summarizes the readings of sentential -le.
Situation Type
Contribution of Sentential -le
States
Inchoative
Activities
Inchoative
Completion
Achievements
Completion
Accomplishments (with completive marker
or numeral object)
Accomplishments (without completive
Inchoative
marker or numeral object)
Table 1 : The semantic contributions of sentential -le
2.2 A restriction with zhi 'only' and budao 'less than'

There is an interesting restriction in the occurrence of sentential -le in sentences containing zhi 'only' and budao 'less than', which to our knowledge has
not been noted previously. Sentential -le can co-occur with zhi 'only' and
budao 'less than' with some numeral phrases, but not others. 9 In particular, it
may occur with zhi 'only' and budao 'less than', when the numeral phrase
expresses a quantity that may increase or decrease over time, but not when the
numeral phrase expresses a quantity that may only increase over time. For
example, when the numeral phrase expresses an amount of money that is in the
possession of the speaker (which may increase or decrease over time) as in
(30) and (31), sentential -le may occur with either zhi 'only' or budao 'less
than'.
g
Zhi 'only' can also appear in sentences without a numeral phrase, and be followed by sentential
-le.
(i) Ta zhi ting ta de hua.
(ii) Ta zhi ting ta de hua le.
he only listen he Poss word
he only listen he Poss word LE
'He only listens to him.'
'He only listens to him now, which he did not do
before.'
Budao 'less than' needs to be followed by a numeral phrase and cannot appear in sentences
without one.
458
(30)
(a)
(b)
(31)
(a)
(b)
Wo zhi you wu kuai qian.

I only have five dollar money
have only five dollars.'
Wo zhi you wu kuai qian
le.
I only have five dollar money LE
have only five dollars now, which was not the case before.'
Wo hai sheng bu dao wu kuai qian.
I still left not until five dollar money
have less than five dollars left.'
Wo hai sheng bu dao wu kuai qian
le.
I still left not until five dollar money LE
have less than five dollars left, which was not the case before.'
On the other hand, when the numeral phrase expresses the duration of the
speaker living in a certain place (which may only increase over time) as in (32)
and (33), sentential -le may not occur with either zhi 'only' or budao 'less
than'.
(32)
(a)
(b)
(33)
(a)
(b)
Wo zai zher zhi zhu le wu nian.

I at here only live LE five year
only lived here for five years.'
*Wo zai zher zhi zhu (le) wu nian le.
I at here only live LE five year LE
Wo zai zher zhu le bu dao wu nian.
I
at here live LE not until five year
lived here for less than five years.'
*Wo zai zher zhu (le) bu dao wu nian le.
I at here live LE not until five year LE
Numeral phrases indicating the number of times an event has occurred or the
distance that has been traveled behave like a duration phrase in that the quantity may increase but not decrease over time.
(34)
(a)
(b)
(35)
(a)
(b)
(36)
(a)
Wo zhi qu le liang ci.

I only go LE two time
only went there twice.'
*Wo zhi qu (le) liang ci le.
I only go LE two time LE
Wo qu le budao liang ci.
I go LE less than two time
went there less than two times.'
*Wo qu (le) budao liang ci le.
I go LE less than two time LE
Wo zhi pao le yi yingli.
I only run LE one mile
only ran one mile.'
(b)
(37)
(a)
(b)
459
*Wo zhi pao (le) yi yingli le.

I only run LE one mile LE
Wo pao le budao
yi yingli.
I run LE less than one mile
ran less than one mile.'
*Wo pao (le) budao yi yingli le.
I run LE less than one mile LE
3 Analysis
In this section, we propose that sentential -le has the assertive meaning that the
situation described is realized prior to a reference time (adopting a part of J.W.
Lin's (2003) analysis). 10 We claim that the reference time is the speech time,
unless the particle jiu is used, in which case, it may be a specified time in the
past or future. In addition, we claim that sentential -le presupposes the existence of a situation opposite to the one described by the sentence immediately
before the point of realization (compare Lai 1999). 11 We show how our proposal accounts for the readings associated with sentential -le and its restrictions with zhi 'only' and budao 'less than'. 1 2
10
J.W.Lin (2003) argues that sentential -le indicates event realization plus result state, while
verbal -le expresses event realization alone. Unlike J.W. Lin (2003), we do not argue that verbal -le indicates event realization (see Soh and Gao, to appear). While we agree that sentential
-le marks event realization, we do not assume that the result state is part of the conventional
meaning of sentential -le. See section 4 for discussion.
It is beyond the scope of this paper to determine the source of the presupposition. We assume
that the presupposition is part of the lexical meaning of sentential le. We are aware that this
assumption raises the question of how presupposition is distinguished from assertion in the
lexical entry (Levinson 1983).
In this paper, we focus on the temporal use of sentential -le. Sentential -le also has nontemporal uses indicating non-temporal scales. An example from Li and Thompson (1981: 242)
is given below.
(i) Zhei ge gua
hen tian le.
this CL melon very sweet LE
'This melon is very sweet.'
While the sentence can be interpreted as expressing a transition from a previous state where the
melon was not very sweet to the current state where it is, this is not the only possible
interpretation. According to Li and Thompson (1981: 243), sentential -le in (i) expresses the
meaning that the sweetness of the melon is relevant to the current situation. The sentence can
be used in a situation in which "one had guessed that the melon would be sweet and found
upon eating it that it was indeed sweet, or if, conversely, one had guessed that it wouldn't be
sweet, but discovered while tasting it that the guess was wrong" or in which "one wanted to
announce a new 'discovery' of the melon's sweetness or if one wanted the hearer to discover
its sweetness".
460
3.1 The readings associated with sentential - l e

As shown in section 2.1, sentential -le gives rise to an inchoative reading when
the situation is atelic and a completive reading when the situation is telic. We
suggest the readings are due to the fact that sentential -le encodes the information that the situation described is realized prior to a reference time (cf. Lai
1999, J.W. Lin 2003). The concept of event realization as in (38) is defined in
Bohnemeyer and Swift (2001).
(38)
, e, t c E [REALE (P, e, t) o P(e) Be' [P(e') e' < E e tSiT(e') t]]
(38) states that "for an event e denoted by to be realized at a (topic) time

interval t, t must contain the run time tSIT of a part e' of e such that e' is also a
P" (Lin 2003: 270). As Lin (2003: 270) notes, a consequence of the definition
is that a telic event is realized only when the inherent end point of the event is
reached, but an atelic event can be realized as long as a subpart of it holds. As
an event realization marker, sentential - l e gives rise to an inchoative reading
with atelic situations because, for an atelic situation to be realized, the situation
only needs to begin. Sentential -le gives rise to a completive reading with telic
situations because, for a telic situation to be realized, the inherent end point of
the situation must be reached.
We claim that the reference time is the speech time, unless the particle jiu
is used, in which case, it may be a specified time in the past or future. The
completion or the beginning of the event occurs prior to the speech time,
unless the sentence contains the particle jiu, in which case it may occur prior to
a specified past time or a specified future time. As shown above, while sentential -le may appear with a time adverbial indicating a past time when the sentence does not also contain jiu (except, for reasons unknown to us, when the
sentence describes a state), it may not appear with a time adverbial indicating a
future time when the sentence does not also contain jiu. We claim that the
restriction holds because without jiu, the time adverbial can only locate the
situation time. In sentences with a time adverbial expressing a future time, the
situation is interpreted as occurring in the future. This future situation time is
incompatible with sentential -le, which entails that the situation is realized
before speech time, in the absence of jiu.
There is a sense that the situation is realized in the recent past in relation to
the reference time. We believe that the recent realization of the situation is not
encoded as part of the conventional meaning of sentential -le, but rather is a
relevance-based implicature (see Lbner 1999). This is because, when connected to an appropriate context, the realization of the situation does not need
to be recent. The sentence in (39), for example, can be used to describe a situation in which a person has returned recently, say from another country, or has
returned for a long time.
(39)
(a)
(b)
461
Ta hui lai
le. Women lai
gei ta xichen.
he return come LE we
come give him welcome dinner
'He has returned. Let's give him a welcome dinner.'
Ta hui lai
le. Keshi hui
lai zhemeduo nian le
he return come LE but
return come this many year LE
dou mei lai zhao wo.
DOU not come seek me
'He has returned. But after so many years, he still has not come to visit me.'
3.2 A restriction with zhi 'only' and budao 'less than'

As shown in section 2.2, sentential -le can co-occur with zhi 'only' and budao
'less than' when the numeral phrase expresses a quantity that may increase or
decrease over time, but not when the numeral phrase expresses a quantity that
may only increase over time. We propose that this restriction holds because
sentential -le presupposes the existence of a situation opposite to the one described by the sentence immediately before the realization of the event described by the sentence (compare Lai 1999).13 We note the transition as
-, E
E (Dowty 1979, citing von Wright 1963, 1968, Pustejovsky 1991, cf.
Huang 1987, Huang and Davis 1989, Sybesma 1999).
To illustrate how our analysis works, consider example (32b) (repeated below as (40a)) involving zhi 'only', and a duration phrase in a situation in which
the duration can only increase over time. Sentential -le presupposes the existence of a situation immediately before the point of realization that is in opposition to [the speaker has only lived here for five years]. As shown in (40b), the
contrary of [the speaker has only lived here for five years] is equivalent to [the
speaker has lived here for more than five years],
(40)
(a)
*Wo zai zher zhi zhu (le) wu nian le.

I at here only live LE five year LE
13
Lai (1999) proposes that jiu presupposes the change of state of a proposition and that the
change happens earlier than expected. She notes that jiu is like German schon and English already. It is also noted that jiu and sentential -le can co-occur (and in fact must co-occur when
the sentence denotes a past event) (Lai 1999: 635-636), while a different particle cai, which is
claimed to be like German erst and English only cannot co-occur with sentential -le. To account for why jiu may co-occur with sentential -le, but cai may not, Lai ( 1999) proposes that
sentential -le asserts that a proposition becomes true before a reference time, while jiu asserts
that a proposition becomes true either before the reference time or at the reference time. Because the meanings of sentential -le and jiu are compatible with each other, they may co-occur.
While our proposal is very close to Lai (1999), we differ in that we do not think that both jiu
and sentential -le have such similar meanings. We attempt to clarify how each of these elements contributes to the overall meaning of the sentence.
462
(b)
Past
Present (speech time)
>[I only live here for five years]
[I only live here for five years]
=[I have lived here for more than five years]
(40a) is unacceptable because the assertion and the presupposition cannot both
be true. Sentential -le encodes the assertion that the situation [the speaker has
only lived here for five years] is realized at a point prior to the speech time,
and presupposes that the situation [the speaker has lived here for more than
five years] immediately precedes the point of realization. Given that the duration associated with living at a certain place can only increase and not decrease
over time, (40a) is ruled out.
The same holds for (33b) (repeated below as (41a)) involving budao 'less
than' and a duration phrase in a situation in which the duration can only increase over time.
(41)
(a)
(b)
*Wo zai zher zhu (le) bu dao wu nian le.

I at here live LE not until five year LE
Past
Present (speech time)
i[I live here for less than five years]
[I live here for less than five years]
=[I have lived here for five years or more]
Sentential -le presupposes the existence of a situation immediately before the

point of realization that is contrary to [the speaker has lived here for less than
five years]. As shown in (41b), the opposite of [the speaker has lived here for
less than five years] is equivalent to [the speaker has lived here for five years
or more]. As in the previous example, (41a) is unacceptable because the assertion and the presupposition cannot both be true. Sentential -le asserts that the
situation [the speaker has lived here for less than five years] is realized at a
point prior to the speech time, and presupposes that the situation [the speaker
has lived here for five years or more] immediately precedes the point of realization. Given that the duration associated with living at a certain place can
only increase and not decrease over time, (41a) is ruled out. The same explanation can be given for cases involving frequency phrases and phrases expressing
distance traveled.
A quantity numeral that may increase or decrease with time can easily appear in a sentential -le sentence containing zhi 'only' or budao 'less than'.
Consider (30b) repeated below as (42a).
(42)
(a)
Wo zhi you wu kuai qian
(b)
I only have five buck money LE

have only five bucks now, which was not the case before.'
Past
Present (Speech time)
i[I only have five bucks] -> [I only have five bucks]
=[I have more than five bucks]
le.
463
Sentential -le presupposes the existence of a situation immediately before the

point of realization that is contrary to [the speaker has only five bucks]. As
shown in (42b), the opposite of [the speaker has only five bucks] is equivalent
to [the speaker has more than five bucks]. The sentence is acceptable since
both the assertion and the presupposition can be true. Sentential -le asserts that
the situation [the speaker has only five bucks] is realized prior to the speech
time, and presupposes that the situation [the speaker has more than five bucks]
immediately precedes the point of realization. Given that the amount of money
one owns may increase or decrease over time, the assertion does not contradict
the presupposition. The same holds for (31b), repeated below as (43a), involving budao 'less than'.
(43)
(a)
Wo hai sheng bu dao wu kuai qian
(b)
I still left not until five buck money LE

have less than five bucks left, which was not the case before.'
Past
i[I have less than five bucks]
[I have less than five bucks]
=[I have five bucks or more]
le.
Our claim that the existence of a past opposite situation is presupposed, rather
than asserted or implied, is evidenced by the fact that the presupposition continues to hold when the sentence is questioned as in (44b), or when it appears
within a conditional as in (44c). We do not use negative sentences as a test
because sentential -le does not occur within the scope of negation.
(44)
(a)
Ta de shenti hao le.

he Poss body good LE
'He is healthy now.'
Presupposition: He was not healthy before.
(b)
Ta de shenti hao le ma?

he Poss body good LE Q
'Is he healthy now?'
(c)
Yaoshi ta de shenti hao le, ta hui hui

Beijing yi tang,
if
he Poss body good LE he will return Beijing one time
'If he were healthy now, he would return to Beijing for a visit.'
Additional support for our analysis comes from the interaction between sentential -le and hai 'still', which has its own presupposition. 14 As (45b) shows,
sentential -le may not occur with hai 'still'.
14
We thank David Beaver for pointing out the connection with hai 'still', and Shiao-wei Tham
for discussion.
464
(45)
(a)
(b)
Ta hai chou yan.

he still smoke cigarette
'He still smokes.'
*Tahai chou yan
le.
he still smoke cigarette LE
The sentence is ruled out because of the incompatibility between the presupposition of hai 'still' and the presupposition of sentential -le. As shown in (46),
sentential -le presupposes that the situation [he does not smoke] immediately
precedes the realization of the situation described, while hai 'still' presupposes
that the situation [he smokes] immediately precedes the situation described.
The sentence is ruled out because the presuppositions cannot both be true.
(46)
Past
Sentential -le: -i[He still smokes]

[He still smokes]
=[He does not smoke]
hai 'still' :
[He smokes]
Furthermore, as expected, sentential -le cannot occur in sentences that describe mathematical truths as shown in (47), given that its presupposition cannot be satisfied. 15
(47)
*Yi jia yi dengyu er le.

one plus one equals two LE
'One plus one equals two.'
4 Similarities with the English perfect

Previous authors (e.g., Li, Thompson and Thompson 1982, Zhang 1997, Lin
2003) have noted similarities between sentential -le and the English perfect.
Among the main similarities are properties involving the existence of a result
state at the speech time or a reference time, and the presence of a continuative
reading in some sentences with a duration phrase. In this section, we clarify the
facts relating to the result state and the continuative reading, and show that the
relevant readings are not entailed by sentential -le. We suggest that the result
state reading is a relevance-based implicature, while the continuative reading
has its source from the way the interval expressed by a duration phrase is temporally located (following Portner 2003).
15
We thank Manfred Krifka (personal communication) for pointing out this prediction of our
analysis. See also Lbner (1989).
465
4.1 Result state

Lin (2003) claims that sentential -le entails that the result state of the eventuality described overlaps with the speech time, while Zhang (1997) allows the
result state to hold at the speech time or the reference time when another situation is explicitly mentioned. Although neither Lin (2003) nor Zhang (1997) is
explicit about the notion of result state, the examples used seem consistent
with the idea that the subject acquires a special property as a result of its participation in the situation, a meaning attributed to perfect constructions in
Smith (1997: 106). The connection between sentential -le and the existence of
a result state has led Zhang (1997) to claim that sentential -le is a perfect
marker in Mandarin Chinese (see also Li, Thompson and Thompson 1982),
and has led Lin (2003: 281) to suggest that sentential -le entails the realization
of an event that brings about a result state, and that the result state overlaps
with the speech time.
As evidence for the claim that sentential -le entails a result state that holds
at the speech time, Lin (2003: 279) points to the contrast between (48a) and
(48b) below, and notes that (48b) implies that at speech time, Zhangsan still
has the car that he bought, and as a result, the discourse in (48b) is incoherent,
in contrast to (48a) involving verbal -le.
(48)
(a)
(b)
Zhangsan zuotian mai-le yi-liang xin che,

Zhangsan yesterday buy-Asp one-CL new car
keshi jintian jiu ba chezi mai-gei-le bieren.
but today then BA car sell-to-Asp other-people
'Zhangsan bought a new car yesterday, but he sold it to some other person today.'
?? Zhangsan zuotian mai yi-liang xin che le,
Zhangsan yesterday buy one-CL new car LE
keshi jintian jiu ba chezi mai-gei-le bieren.
but today then BA car sell-to-Asp other-people
'Zhangsan bought a new car yesterday, but he sold it to some other person today.'
However, as Zhang (1997) points out, the result state does not need to hold at
the speech time. This is demonstrated in the contexts in (49) and (50). Within
the context in (49), the result of speaker giving up smoking holds at the
speech time. However, within the context in (50), the result state of speaker
B's going out to buy things (i.e., not being at home), does not hold during the
time of speech, as speaker is found sitting on the sofa at home during speech
time.
(49)
Context: A is offering a cigarette.

A:
Lao Zhang,
chou-gen yan.
Old Zhang (address term), smoke - C L cigarette
O l d Zhang, have a smoke.'
466
(50)
Wojie
yan
le.
I give up cigarette LE
have given up smoking.'
Context: A, B's mother came back home and saw sitting on the sofa.
A:
Wo gang-cai wangjia
li da dian-hua zenme mei ren
jie?
I just-now to
home in call telephone why no person answer
called home just now, why did nobody answer?'
B.
Wo chu-qu mai dongxi le.
I out-go buy thing LE
went out to buy things.'
While we agree with Zhang (1997) that the result state does not need to hold at
the speech time, unlike Zhang (1997), we interpret (50) as indicating that the
result state reading is not entailed by sentential -le, as opposed to indicating
that the result state holds at a reference time distinct from the speech time. Our
reasoning is that a reference time distinct from the speech time (in which the
situation described is realized prior to it) cannot be introduced in sentential -le
sentences without the particle jiu. Instead, we suggest that the result state reading is a relevance-based implicature arising from the use of sentential -le in
achievement and accomplishment situations. In sentences that describe activities and states, it is difficult to determine if such an implicature is available
because it is not clear what the result state of an on-going situation might be. 16
4.2 Continuative reading

For sentences with a duration phrase, J.W. Lin (2003) suggests that the eventuality described must overlap with the speech time, while Zhang (1997) suggests that it does not. J.W. Lin (2003: 279), for example, notes that (51b) is
only compatible with a situation in which the speaker still lives in the United
States at speech time, but not one in which the speaker no longer lives in the
United States. This is in contrast to (51a) with verbal -le, which is compatible
with either situation.
(51)
(a)
(b)
16
Wozaimeiguo zhu-le ershi nian, cong mei tingshuo-guo zhe-zhong shi.

I in America live-Asp twenty year ever not hear-Asp
this-kind thing
(have) lived in America for 20 years and (have) never heard this kind of thing,
Wo zai meiguo zhu
ershi nian le, cong mei tingshuo-guo zhe-zhong shi.
I in America live-Asp twenty year LE ever not hear-Asp
this-kind thing
have lived in America for 20 years and have never heard this kind of thing.'
Iatridou et al. (2001: 192) note that perfect of result is considered possible only in connection
with telic predicates. On the other hand, J.W. Lin (2003: 281) suggests that the result state of a
state is the state itself. Lin's analysis allows eventualities that began in the past, but which are
still going on, to have current results. See also Musan (2001).
467
For Zhang (1997), the first sentence in (51b) has two possible interpretations.
One is that the speaker is now in the state resulting from his/her previous living in the United States for twenty years. Another is that the speaker is now in
the state of living in the United States and this state has been going on for
twenty years at the time of the utterance. We agree with Zhang that the eventuality described in (51b) does not need to overlap with the speech time. The
exchange in (52) supports our intuition that the continuative reading is not
entailed by sentential -le.
(52)
A:
Ni weishenme bu zai na
B:
you why
not at there live LE
'How come you don't live there anymore?'
Wo zaina zhu ershi nian le. Gou
jiu de le.
I at there live twenty year LE enough long Prt. LE
lived there for twenty years. (I (have) lived there) long enough already.'
zhu le?
Further examples to illustrate this are given below. (53a), for example, can be
uttered in a situation in which the waiting continues as in (53b) or one in which
the waiting no longer continues as in (53c).
(53)
(a)
Wo deng ni liang xiaoshi le.

I wait you two hour LE
have waited for you for two hours.'
(b)
Ni zenme
xianzai cai dao? Wo deng ni liang xiaoshi le.
you how come now only come I wait you two hour LE
'How come you are here only now? I have waited for you for two hours.'
Ni zenme
xianzai hai mei dao? Wo deng ni liang xiaoshi le.
you how come now still not come I wait you two hour LE
'How come you are still not here? I have been waiting for you for two hours.'
(c)
As with sentential -le sentences, the English perfect, when combined with a
duration phrase as in (54), can also have a continuative or a non-continuative
reading (Portner 2003, Hitzeman 1997).
(54)
Mary has lived in London for five years.
In our analysis, sentential -le leaves the continuation of a situation open and is
hence consistent with both continuative and non-continuative interpretations.
In (51b), for example, sentential -le expresses that the situation involving the
speaker living in the United States for 20 years was realized prior to the speech
time. It presupposes that the situation described by the sentence is preceded by
a situation in which the speaker has lived in the United States for less than 20
years. Nothing is said about whether the situation continues or not.
468
While the continuative reading is not entailed by sentential -le per se, there
is a strong sense that the event described continues at speech time in sentences
with sentential -le, in contrast to sentences with verbal -le. What is the source
of the continuative reading? Following Portner (2003), we suggest that the
source of this reading is the way in which the interval expressed by the duration phrase is temporally located (see also Iatridou et. al 2001). Portner (2003),
elaborating on Hitzeman (1997), shows that a sentence like (54) allows for an
overt specification of the temporal location of the interval expressed by the
duration phrase.
(55)
(a)
(b)
Mary has lived in London for five years as of now.

Mary has lived in London for five years as of January 1985.
Portner (2003) suggests that when the temporal location is not expressed, there
are two ways to interpret the temporal location. One is to situate the temporal
location with a value from the context, and in the case of the present perfect,
this value is the speech time. The other option is that it is existentially quantified.
(56)
(a)
Mary has lived in London for five years as of now.
(b)
Mary has lived in London for five years as of some time.
The "now" reading of the adverbial provides a continuative reading of the

perfect and the existential reading provides a non-continuative reading. We
suggest that the same account can be extended to the continuative/noncontinuative reading in sentential -le sentences with a duration phrase. The
continuative reading arises when the temporal location is situated at the speech
time, and the non-continuative reading arises when it is existentially quantified. In other words, the continuative/non-continuative distinction derives from
the way in which the temporal location is interpreted (Portner 2003). Sentences
with sentential -le allow the interval expressed by the duration phrase to be
located at speech time, in contrast to sentences with verbal -le.
5 Sentential -le, perfect and already

In this section, we discuss the relation between sentential -le, perfect and already. We show that the assertion of sentential -le resembles perfect, and the
presupposition of sentential -le resembles English already.
5.1 Sentential -le and perfect
While there are various approaches to the semantics of perfect (see Portner
2003, Alexiadou et. al 2003), it seems to be generally accepted that the perfect
469
temporally locates a situation relative to a reference time. Comrie (1985: 6482), for example, proposes that the meaning of pluperfect (past perfect) and
future perfect is that there is a reference point in the past or future respectively,
and that the situation is located prior to that reference point. 17 Previous authors have noted that perfect interacts differently with different situation types
(Iatridou et al. 2001, Musan 2001, Portner 2003). For example, German present perfect expresses inchoativity with atelic situations and completion with
telic situations (see Musan 2001:362-364). The same seems to be true for
Greek perfect as discussed in Iatridou et al. (2001: 207-208, 213). Similarly, an
English perfect sentence that describes a telic situation is interpreted as completed (past with respect to the reference time or speech time), while a sentence that describes a state (atelic) is interpreted as beginning (either past with
respect to the reference time/speech time or overlapping with the reference
time/speech time) (Portner 2003).
(57)
(a)
Mary has reached the finish line.
(b)
(c)
Mary has understood the issue.

Mary has been angry all day.
However, there is a question about whether activities, which are atelic, pattern
like states or not (see Portner 2003: 462-463, Iatridou et. al 2001:210-211). An
activity in the perfect is usually interpreted as terminated or past with respect
to the reference time or speech time. Upon hearing (58a) and (58b), for example, someone would normally infer that the activities described have terminated.
(58)
(a)
(b)
She has searched for the ring,

He has exercised this morning.
However, the termination of the activities can be cancelled as shown in the

exchanges in (59), indicating that the relevant reading is an implicature rather
than an assertion of the sentence. 18
(59)
(a)
A:
B:
(b)
17
18
I don't think she cares about you. She doesn't even care that the ring you gave
her was missing.
Yes, she does. She has searched for the ring. In fact, she may still be searching
for it now.
Background: Mr. Anderson is a difficult patient. He is supposed to exercise every

morning, but he often skips his exercise.
Nurse A:
I bet you Mr. Anderson hasn't exercised this morning.
Comrie (1985:78) treats present perfect differently from past and future perfect.
We thank Jeanette Gundel for discussions.
470
Nurse :
that now.
Actually, he has exercised this morning. In fact, he may still be doing
We think that the English perfect behaves like the German and Greek perfect,
and it can be said that perfect in these languages expresses the realization of a
situation prior to a reference time. Given this, the assertive meaning of sentential -le is the same as perfect in these languages. By specifying that sentential
-le carries the same assertive meaning as perfect, and clarifying that the English perfect does not entail the termination of an activity, we provide further
evidence for the claim that the notion of completion that is often associated
with perfect is not part of the meaning of the perfect (Iatridou et al. 2001,
Portner 2003).
5.2 Sentential -le and already
While sentential -le has the same assertive meaning as perfect, its presupposition is not shared by perfect. The English perfect does not exhibit the same
restrictions as sentential -le with only and less than. There is no contrast between (60) and (61) and (62).
(60)
(61)
(62)
(a)
I have only five dollars.
(b)
(a)
(b)
(a)
(b)
I have less than five dollars.

He has lived here for only five years.
He has lived here for less than five years.
I have visited there only twice.
I have visited there less than two times.
On the other hand, the adverb already does. Already may occur with only and
less than when the numeral phrase expresses a quantity that may increase or
decrease over time, but not when the numeral phrase expresses a quantity that
may only increase over time. This is shown in the contrast between (63) and
(64) and (65).
(63)
(64)
(65)
(a)
(b)
(a)
(b)
1 already have only five dollars.

I already have less than five dollars.
*I (have) already lived there for only three years.
*I (have) already lived there for less than three years.
(a)
(b)
*I (have) already visited there only twice.

*I (have) already visited there less than two times.
That English already patterns like sentential -le suggests that already shares
sentential -le' s presupposition, and provides support for the claim that already
presupposes the existence of a prior situation that is in opposition to the one
471
described by the sentence (Lbner 1989, 1999, van der Auwera 1993, but see
Mittwoch 1993).19
6 Conclusions
In this paper, we have shown that sentential -le encodes the assertive meaning
that the situation in question is realized prior to a reference time, and the presupposition that a situation opposite to the one described by the sentence exists
immediately before the point of realization (compare J.W. Lin 2003, Lai
1999). We claim that the reference time is the speech time, and that it may be a
specified time in the past or future when the particle jiu is used. We have
shown how our proposal accounts for the readings associated with sentential le, and its restriction with zhi 'only' and budao 'less than'. We have also attempted to clarify the similarities that have been noted between sentential -le
and perfect, and have argued that neither the result state nor the continuative
reading is entailed by sentential -le. Instead, we suggest that the result state
reading is a relevance-based implicature, while the continuative reading has its
source from the way the interval expressed by a duration phrase is temporally
located (following Portner 2003). We show that sentential -le shares its assertive meaning with perfect, and its presupposition with English already. Our
analysis provides further evidence for the claim that completion is not part of
the meaning of perfect (Iatridou et al. 2001, Portner 2003), and supports the
claim that already encodes a presupposition about the existence of a prior
situation in opposition to the one described by the sentence (Lbner 1989,
1999, van der Auwera 1993, but see Mittwoch 1993). By clarifying the source
of the meanings of sentential -le, and the way in which sentential -le is related
to perfect and already, we hope to have enriched the empirical base necessary
for the construction of a universal theory of perfect.
References
Alexiadou, ., M. Rathert and A. von Stechow (2003): Introduction: the Modules of
Perfect Constructions, June 2003,
http://vivaldi.sfs.nphil.unituebingen.de/~arnimlO/Aufsaetze/vorwortperfect.pdf.
Bohnemeyer, J. and M. Swift (2001): Default Aspect: The Semantic Interaction of
Aspectual Viewpoint and Telicity. Proceedings of Perspectives on Aspect, Utrecht
Institute of Linguistics.
Comrie, B. (1985): Tense. Cambridge: Cambridge University Press.
Chu, C. (1976): Some Semantic Aspects of Action Verbs. Lingua 40, 43-54.
19
Mandarin Chinese also has an adverb yijing 'already'. We leave the examination of the relation
between yijing 'already' and sentential -le for future work.
472
Dowty, D. (1979): Word Meaning and Montague Grammar: The Semantics of Verbs
and Times in Generative Semantics and in Montague's PTQ. Dordrecht: Reidel.
Gundel, J., N. Hedberg and R. Zacharski (1993): Cognitive Status and the Form of
Referring Expressions in Discourse. Language 69, 274-307.
Hitzeman, J. (1997): Semantic Partition and the Ambiguity of Sentences Containing
Temporal Adverbials. Natural Language Semantics 5, 87-100.
Huang, L. (1987): Aspect: A General System and Its Manifestation in Mandarin Chinese. Doctoral Dissertation, Rice University.
Huang, L. and P. Davis. (1989): An Aspectual System in Mandarin Chinese. Journal of
Chinese Linguistics 17, 128-166.
Iatridou, S., E. Anagnostopoulou, and R. Izvorski (2001): Observations about the Form
and Meaning of the Perfect. In M. Kenstowicz (ed.): Ken Hale: A Life in Language.
Cambridge, MA.: M I T Press.
Kang, J. (1999): The Composition of the Perfective Aspect in Mandarin Chinese. Doctoral Dissertation, Boston University.
Klein, W., P. Li, and H. Hendriks (2000): Aspect and Assertion in Mandarin Chinese.
Natural Language and Linguistic Theory 18, 723-770.
Lai, H. (1999): Rejected Expectations: The Scalar Particles cai and jiu in Mandarin
Chinese. Linguistics 37, 625-661.
Levinson, S. (1983): Pragmatics. New York: Cambridge University Press.
Li, C. and S. Thompson (1981): Mandarin Chinese: A Functional Reference Grammar.
Berkeley: University of California Press.
Li, C., S. Thompson and R. Thompson (1982): The Discourse Motivation for the Perfect Aspect: The Mandarin Particle LE. In: P. Hopper (ed): Tense-Aspect: Between
Semantics and Pragmatics. Amsterdam: John Benjamins Publishing Company.
Lin, J. (2004): The Mandarin Verbal System. Manuscript, MIT.
Lin, J.W. (2003): Temporal Reference in Mandarin Chinese. Journal of East Asian
Linguistics 12, 259-311.
Liu, F. (2003): Definite NPs and telicity in Chinese. Snippets 7, July 2003.
http://www.ledonline/snippets.
Lbner, S. (1989): German schon-erst-noch: An Integrated Analysis. Linguistics and
Lbner, S. (1999): Why German schon and noch Are Still Duals: A Reply to Van der
Auwera. Linguistics and Philosophy 22, 45-107.
Mittwoch, A. (1993): The Relationship Between schon/already and noch/still: A Reply
to Lobner. Natural Language Semantics 2, 71-82.
Musan, R. (2001): The Present Perfect in German: Outline of Its Semantic Composition. Natural Language & Linguistic Theory 19, 355-401.
Portner, P. (2003): The (Temporal) Semantics and (Modal) Pragmatics of the Perfect.
473
Pustejovsky, J. (1991): The Syntax of Event Structure. Cognition 41, 47-81.

Rohsenow, J. (1978): Perfective le: Aspect and Relative Tense in Mandarin Chinese. In
R. Cheng, Y. Li and T. Tang (eds): Proceedings of Symposium on Chinese Linguistics. 269-291. Taipei: Student Book Co.
Ross, C. (1995): Temporal and Aspectual Reference in Mandarin Chinese. Journal of
Chinese Linguistics 23, 87-135.
Shi, Z. (1990): Decomposition of Perfectivity and Inchoativity and the Meaning of the
Particle LE in Mandarin Chinese. Journal of Chinese Linguistics 18, 95-123.
Smith, C. (1994): Aspectual Viewpoint and Situation Type in Mandarin Chinese. Journal of East Asian Linguistics 3, 107-146.
Smith, C. (1997): The Parameter of Aspect (2nd Ed). Dordrecht: Kluwer Academic
Publishers.
Soh, H. and M. Gao (2006): Perfective Aspect and Transition in Mandarin Chinese: An
Analysis of Double -le Sentences. In Pascal Denis et al. (eds), Proceedings of the
2004 Texas Linguistics Society Conference, Somerville, MA: Cascadilla Proceedings Project. 107-122. www.lingref.com, document #1510.
Soh, H. and M. Gao (to appear): It's Over: Verbal -le in Mandarin Chinese. In N. Hedberg and R. Zacharski (eds): Topics on the Grammar-Pragmatics Interface: Essays
in honor of Jeanette . Gundel. John Benjamins.
Soh, H. and J. Kuo (2005): Perfective Aspect and Accomplishment Situations in Mandarin Chinese. In A. van Hout, H. de Swart and H. Verkuyl (eds): Perspectives on
Aspect, Dordrecht: Springer. 199-216.
Sybesma, R. (1997): Why Chinese Verb-LE Is a Resultative Predicate. Journal of East
Asian Linguistics 6, 215-261.
Sybesma, R. (1999): The Mandarin VP. Dordrecht: Kluwer Academic Publishers.
Tai, J. (1984): Verbs and Times in Chinese: Vendler's Four Categories. In D. Testen,
V. Mishra and J. Drogo (eds): Papers from the Parasession on Lexical Semantics,
Chicago Linguistic Society 20. 289-296.
van der Auwera, J. (1993): 'Already' and 'Still': Beyond Duality. Linguistics and Philosophy 16, 613-653.
von Wright, G. (1963): Norm and Action. New York: Humanities Press.
von Wright, G. (1968). An Essay in Deontic Logic and the General Theory of Action.
Acta Philosophica Fennica 11.
Zhang, Q. (1997): The Semantics of the Verbal Suffix -le and Sentence Final -le.
Manuscript, Stanford University.
Zhu, D. (1982): Yu Fa Jiang Yi. Beijing: Commercial Press.
SECTION VI
Event Structure and Natural Language Ontology
Regine Eckardt (Gttingen)
The lower part of event ontology*

1 Conflicting applications of event ontology
In this paper, I will address an apparent conflict between two applications of
event ontology for natural language semantics. The conflict arises within the
lower part of event ontology, and consists, briefly, in the following: Scholars
who aim at modelling tense and aspect, and in particular the distinction between telic and atelic predicates, commonly assume that certain properties of
events are inherited by all their parts. These are called homogeneous properties. On the other hand, recent proposals to model negative polarity items have
to assume that there is a level where the parts of events are so small that they
can no longer reasonably inherit any property that can be denoted by a natural
language predicate. In the following two sections, I shall recapitulate the respective positions and list some armchair assumptions about events that come
along with either one. In section 1.3. I will name two possible ways out of the
dilemma, which will be elaborated in this paper.
1.1 Aspect
Krifka (1989), following earlier work by Link (1983), can be seen as the
groundbreaking elementary proposal to model the distinction between telic and
atelic predicates on the basis of events. It is assumed that sentence radicals
denote properties of events P. If the property is quantized, the sentence
makes a telic statement. If the property is homogeneous, then the sentence
makes an atelic statement. The simplest linguistic correlate to this distinction is
the test of whether the duration of the eventuality described will be specified
with an in-PP (telic) or a for-PP (atelic sentence). The simplest definitions of
"being quantized" and "being homogeneous" are given in (1) and (2). Further
refinements were discussed in subsequent literature but are of no immediate
concern here. I use cr for the part-of relation in the domain of events.
I would like to thank Manfred Krifka and, in my former life, Ulrich Felgner and Frieder Haug
for teaching me model theory. They are not to blame if they did not succeed.
478
Regine Eckardt
(1)
QUANT(P) <=> 3e( P(e) & Ve '( e ' c= e -> -<P(e ') ) )
(2)
HOM (P) <=> Ve Ve' ( P(e) & f ' c e - > P(e ') )
It is easy to see that these definitions rest crucially on further assumptions

about event ontology. In particular, for QUANT to be meaningful we need to
ensure that the event ontology as such does not have an atomic level. Events
are conceived of much like the set of time intervals on the rationals, which
does not have an atomic level of smallest intervals. Indeed, at least nonstative
events e have a running time z(e) which is an interval on the time line. The
following assumptions, hence, seem to be uncontroversial for this kind of theory, even if single authors do not care to list them all explicitly. Note that the
symbol < stands for temporal precedence on the time line, as well as temporal
precedence between two (temporally located) events. The relation c stands for
the part-of relation between two events as well as the subset relation between
time intervals. Clearly, if e cz e' then zie) c z(e').
(3)
There is no lower boundary to events:

Ve3e'( e' e e )
(4)
Boolean Structure: There is a summation operation defined on events that adds up adjacent events (incl. overlapping events) to larger events:
Ve Ve' ( - , 3e*( <e) < <e*) < He")) - /( e@e'=f) )
(< on time intervals is the partial ordering defined as / < J
iff Vi V/(e / & je J -* i<j) )
(5)
Betweenness: Between any two events, there is another one.

VeVe'( e ' c e - > 3 e * ( e ' c c * c e ) )
(6)
Differences: If e' is part of e, then there are non-overlapping e", e'"

that add up e ' to e :
VeVe' ( t ' c e - )
[ 3e"( e' e"=e 3e*(e*ae' e*<=e" ) )
3e"e"'( e' e" e"'=e -de*(e*<z e' & e*a e" ) -i3e*(e*C e' & e*cz e'" )]
Atelic predicates are modelled as homogeneous predicates (Link 1983, Krifka

1989 and subsequent, Pin 2000 and others). Link (1983) pointed out that, at
least in the domain of real matter and things, these assumptions are in fact
wrong, physically speaking. He notes that the matter gold, for instance, has an
atomic level (namely, the level of single gold atoms) even though the natural
language term 'gold' behaves as if it denoted a homogeneous property. Link,
as well as later authors, assumes that natural language ontology need not be
isomorphic to a physically tenable model of the world, because the facts and
phenomena in natural language have not been shaped by modern quantum
physics but by folk views about nature and matter. Krifka (1989) makes similar remarks with respect to the lower end of event ontology. The general strategy hence seems to be, to use folk models of the world and time because we
479
want to model the grammatical effects of folk notions about the world and
time.
2.1 Negative Polarity Items

In a series of papers that go back to Fauconnier (1975), the behavior of negative polarity items is modeled based on the fact that they describe the weakest
possible case of a set of salient alternatives (Krifka 1995, Lahiri 1998, Eckardt
2003). A sentence like (7) is predicted to be well formed because lift a finger
denotes the smallest possible way of lending help.
(7)
Tom did not (even) lift a finger in order to help me.
Interestingly, many languages distinguish between so-called weak and strong

negative polarity items where strong NPIs are virtually restricted to negated
contexts and rhetorical questions. The NPI in (7) is commonly assumed to be
of the strong type. Informants agree that downward entailing contexts like/ew,
rarely, etc. do not license it (see (8)) and that a question like (9) can only be
used as a rhetorical question.
(8)
*Few people (even) lifted a finger in order to help me.
(9)
Did Tom even lift a finger in order to help you?
There is a general consensus on how the data in (7)-(9) should be derived from
the lexical meaning of the NPI, which is that expressions like lift a finger, bat
an eyelash, drink a drop etc. denote irrelevantly small events or objects. This
general idea has received different analyses by various authors. Krifka (1995)
models it in probabilistic terms and stipulates the following strict inequality of
probabilities:
(10) p( A Tom did not do a in order to help me) > p(Tom did not lift a finger)
where the conjunction A ranges over all possible alternative ways in which Tom
could have helped me.
The distinct distribution of strong NPIs is derived from this inequality. The
details of the theory will not be discussed further in this paper. While this
probabilistic account for strong polarity sensitivity is logically consistent with
other assumptions about event ontology, inequalities such as the one in (10)
are hard to relate with intuitions about events and their sub-events. One may
suspect that a full model-theoretic account for (10) needs to address similar
issues like the ones treated in this paper.
In a related but different vein, van Rooy (2003) explains the rhetorical quality of the question in (9) essentially by remarking that a positive answer to this
480
Regine Eckardt
question would be downright absurd, given that the question, according to his
background theory, is such that a positive answer would pragmatically implicate that "lifting a finger" is also the maximal and hence the only thing that
Tom undertook in order to help me. He states that this would not be a relevant
act of helping, without further discussion. Eckardt (2005: chap. 5, 2006) takes
up this position and elaborates the distinction between weak and strong negative polarity items on the basic assumption that weak NPIs denote small objects and events but ones that are still reasonable things to do. Strong NPIs, in
contrast, denote eventualities that are so small that they no longer fall in the
right kind of category. For example, lifting one's finger may be a subevent of
events of helping, but it is not an event of helping itself, and it can not occur in
isolation (i.e. without an appropriate superevent of reasonable size). This relates to the observation that natural language terms like Dutch ook maar or
German auch nur, if used in questions, lead to rhetorical questions that cannot
possibly receive a positive answer.
The ontological implementation of the strong-weak distinction suggests the
plausibility of axioms like the ones in (11) and (12). These hold in particular
also for properties of events that would be regarded as homogeneous in an
aspectual theory.
(11) If you really look down into the lower end of ontology,
some events are just too small to count:
Ve( P(e) - 3 ( -,() ) )
(12) Canwe tell where?
(a) 3e(/ , (e)AVe'(e'ce^-,P(e')))
('yes')
(b) Ve(P(e)->3e'(e'ce A P(e)))
('no')
It is evident that assumptions like (11) are fatal for a predicate that an account
of aspect would predict to be homogeneous. On the other hand, we might
claim that negative polarity items which denote "minimal objects or events" do
exactly what one should not do according to the general guidelines of aspect
theory, which is to zoom into the lower part of event ontology which is simplified and idealized in this kind of modelling.
Note that the present conflict is not an easy one between folk theory and
physical theories about the world. Both ways to view the lower end of event
ontology are supported by linguistic facts. Hence, there seem to be two different folk theories about very small events. What kind of viewpoint shift is occuring here?
As the summary above already suggests, several analyses of "P-events too
poor to mention" can be imagined. For present purposes, I will hypothetically
adopt the
481
Strong position: There are events e below P-events that are not themselves in - even if is intuitively a homogeneous predicate (for instance 'walk a single step ' is not something that is an event of walking.)
I will not defend the strong position as the best, or only possible one. The aim
of this paper is to demonstrate how this position can be carried out. Before we
turn to the details, I will lay out the roadmap of the paper in the following
section.
1.3 Possible solutions
What kind of "blindness" makes speakers prefer one kind of expectations on
one occasion and another on another occasion? What kind of change in our
world view takes place once we zoom in the lower end of ontology? Somewhat
surprisingly, there are even two consistent answers to this question. The first
one elaborates the idea that we make bold universal statements about events
(like HOM) because we ignore some events. If we really take all events into
account, we are forced to retract these strong universal statements. The step
between view one and view two hence consists in increasing or reducing the
underlying domain of events. From a superficial view, so to speak, we can not
see all events and hence feel inclined to universal statements like HOM(P).
The surprising part of this idea is that we seem to see a great many small
events even before we took that closer look. How could we have overlooked so
many of them? In section 2, I offer an application of a model theoretic construction to the domain of events which shows that this is logically possible. In
section 3, I compute an actual example that might be useful as an illustration,
or for concrete applications.
In section four, I turn to a second kind of explanation which rests on the assumption that we face an instance of the Sorites paradox. This view comes
down to the claim that we make bold universal statemements because we idealizingly assume wrong properties for some minor events. I will discuss one
spellout of this view and turn to a final comparison in the last section.
2 Infinitesimal events
Let L be a first order language that contains relations and functions appropriate
to event ontology. Specifically, I will use a sortal distinction between events
and time intervals (along with the classical sorts for individuals; I will ignore
extensions to higher order logic in the following). The unary function symbol
will be interpreted as the function that maps each event onto its running time.
The binary relation c will be defined both on the set of events as well as the
482
Regine Eckardt
set of time intervals. The binary relation < is defined primarily as the earlierthan relation on the set of time intervals. It can be shifted to the domain of
events by assuming that e<e2 iff z(e;)<i(e->). Finally, the binary function is
interpreted as event summation. For the present purposes I will assume that
summation is restricted to temporally adjacent events. Nothing depends crucially on this assumption, but it is in the spirit of the general enterprise to see
how two perfectly natural but contradictory views of event ontology relate to
each other.
Let E = (, , , <, ci) be an event structure for such a first order language,
and one that specifically verifies the L-axioms (3) to (6) above. We assume
moreover that there is at least one homogeneous predicate that lives on E.
The model theoretic construction will be spelled out with reference to . It can
easily be modified so as to extend to further homogeneous predicates.
Definition: Let (e,)ISEN be a sequence of events in E. We call (e)1= N zeroconvergent iff
ViV/( i<j -êjcz e )
3e Vi( e c e ( )
Let () be the set of all zero-convergent sequences in E.

Definition: Let = be a relation on () that is defined as follows:
(e)eN ~ (/)eN iff
Ve, \fj(fjCied
V/t 3er ( er c fk)
As a consequence, = is an equivalence relation on ().
Next, we will augment L to a richer language L(E) by adding constant
names for all events in E. Formally, we could do so by taking the respective
domain of E, indexing all its elements e, for example, as e, in order to avoid
confusion between objects and language, and add these indexed elements as
new constant symbols to L.
Now, consider the following sets of sentences in L(E):
(13) [(,)] := { * <= e, I ie } {-> \fe(x c e) }
An object that would make all statements in [(e,), eN ] true would be part of all
events in (e,) ieN without being the zero event. The next goal we need to
achieve is
1) to construct an event structure
2) that extends the original event structure E and
483
3) contains new elements such that

4) for each one of the sets of sentences as in (13), there is some for
which all the formulae in hold true at once.
The construction we are aiming for should add such very small elements for all
zero-convergent sequences in E. However, we also must account for cases in
which two such sequences converge to "the same point". In order to avoid
contradictions, we need to ensure that only one infinitesimal element will be
added in such cases. Therefore, we will first identify all co-convergent sequences.
First, we will choose some representative sequence (e,), eN for each of the
equivalence classes modulo ~ in (). Remember that this sequence now
stands as the representative for all further sequences that consist of different
events but eventually dove-tail with this representative in such a way as to
converge to the same (so far: abstract) mini-event. For each one of these representatives, we now take the respective set of formulae 0[(e,j, e N ] as in (13) and
keep it in stock. We need to keep the free variables in each of the 0[(e,), eN ]
distinct. For the purpose of exposition here, I will use different letters x, y,
for different sets of formulae 0[(e,), cN ], 0[(/j) i e N ] Generally, we must use
variables in the definition of 0[(e,), e N ] that are indexed with the respective
sequence. I will not carry this out for obvious typographical reasons. Formally:
Definition: Choose a fixed set of representatives for the equivalence classes in
()/~
Let
:= u{<B[(e);eN] I (e) the representative
of some equivalence
class in
()/~}
We now need to conjoin these formulae with the elementary theory of E: Let
therefore
Th(E) := { \ âtomic sentence in L(E) and E 1= )
Hence, Th(E) offers a full description of all elements in E. Any model for
Th(E) will therefore contain a substructure that is isomorphic to the original
structure E.
Finally, let us add the requirement that events between two P-events are
again P-events:
VeVe'Ve*( e' c e* c e & P(e) & P(e')
P(e*) )
We can now turn to the construction of an event structure which extends E in

a conservative manner, which contains infinitesimal events, and where these
infinitesimal events are not in the extension of even though they might be
parts of larger events that are in the extension of P.
484
Regine Eckardt
First observe that the following set of formulae is finitely consistent:

U Th(E) U { V e V e V e * ( e' c e* c e & P(e) & Pie") -> P{e*) )}
If we take any finite subset of this set of formulae, we can prove its consistency by interpreting it in the old event structure E f r o m which we started.
M o r e specifically, there is an interpretation I of the constant symbols and a
variable assignment g for the free variables in such that E \=' s for all formulae in . W e can simply interpret all constant names in by the respective
element e in E and interpret the variables y, , as events that are part of all
those larger events that are mentioned in , among the formulae collected in
. Because there are only a finite number of such statements in and the decreasing sequences (e,), EN were assumed to be infinite, we can always find
events that are smaller than a finite part of the infinitely decreasing sequence.
As all finitely consistent sets of formulae are also consistent, there exists a
model of Th(E). I will use , ' etc. as meta-variables for elements that
realize one of the types of infinitesimal objects, 0 [ ( e , ) i e N ] .
As Th(E) contains only atomic sentences in L(E) (i.e. importantly, not the
clause about the homogeneity of P ) we can moreover consistently assume that
() for all infinitesimal objects. (This step of the construction will be formally legitimised below by a model construction that proves its consistency.)
W e can n o w f o r m the set of all infinitesimal small objects below that are
too small to b e themselves: ( 3e(P(e) & ) & {) ). For convenient
reference, let us call this area the infinitesimal part below P .
INF(P)W<H>3e(P(e)
& c e ) & -J>(e) )
T h e model hence comes up to our expectations about "zooming into" the

lower end of event ontology in the following way: W e maintain everything
that we believed about previously recognized events (E is a substructure of ).
All previous P-events as well as those that are between earlier P-evcnts remain
P-events. However, there is a lower level of previously unrecognized events
that are not P .
3 An Example
In order to exemplify the above construction, I will repeat it on the basis of
c o m m o n mathematical structures. W e will start with the real numbers R and
the set of all open intervals I over R . Let us call this set E, in order to stress
that we are not supposed to cnsider the internal structure of the objects in question f r o m n o w on.
W e can n o w take E to be the domain of events of our event structure and
extend this set to a full event structure; specifically by adding the linearly or-
485
dered real numbers as our domain of time points. Let me define the basic relations and functions on events in E as follows:
The timeline will consist of the real numbers (R, <).

For all e in E, i(e) :-1 iff e=/ (remembering e's internal structure for
a moment)
For all e, e' in E\ ee' is defined iff e- ]x;y[ and e'=]w;z[ and
either ~\x;y[ and ]w;z[ have nonempty intersection
or y=w.
In case (i), ee'
the event represented by ]x;y[ ]w;z[
In case (ii), ee' :- ]x;z[
(If desired, the operation can be made commutative)
An event e is a mereologialc part of another event e', e c e' iff, seen
as intervals in R, e <z e'.
W e can now choose the extension of a homogeneous predicate in E , starting

from some maximal P-event e.
P(e*) iff e * c e
Let us check that the structure E conforms to axioms (2) to (6) above:
(3)
There is no lower boundary to events:
\/e3e'( e' ce)

This holds true, because for each interval / in R there are more open intervals
that are true parts of I.
(2)
Homogeneous predicates apply to events that consist of P-parts all the way down:
HOM(P)
VeVe'( P(e) A t ' c M f f O )
This holds true due to definition of the extension of P.

(4)
Boolean Structure: There is a summation operation defined on events that adds

temporally adjacent events (incl. overlapping events) to larger events:
Ve Ve ' ( -i 3e*(
< H>*) <
-> 3/( ee'=f) )
This holds true due to the definition of . Of interest to us are are events with
a non-intersecting temporal extension {]x;y[ and ]y;z[ with y in the temporal
extension of neither). Here, the addition of events diverges from simple set
union in R.
486
Regine Eckardt
(5)
Betweenness: Between any two events, there is another one.
(6)
Differences: If e' is part of e, then there are non-overlapping e", e'" that add up e' to e:
VeVe'( e' c. e 3e*( e' cz e* a e) )

V e V e ' ( e' e e >
[ Se"{ e' e"=e Be*(e*cie' e*ae" ) )

3e"e"'( e' e"e"'=e 3e*{e*c:e' & e*ae" ) 3e*(e*cze' & e*CLe"' )]
Both (5) and (6) hold true due to construction.
The construction of infinitesimal elements over this initial event structure E
will result in the introduction of events that would correspond to single points
in R. Each zero-convergent sequence (e)eN in E corresponds to a convergent
sequence of intervals in R:
( ]*,; y I ) , convergent series of intervals.
It is a theorem in R that the limes element of such sequences exists.
arises from E by adding (closed) intervals that consist of one point only. The
closure over these will yield - {all open and closed intervals over R}. We
can consistently assume that all events that correspond to single points in R
(\x;x\) are not in the extension of P.
The property is therefore not homogeneous in the strong sense in that
each and any part of a P-event is again a P-event. However, homogeneity can
be stated in the following weaker form:
HOM(P)
Ve(P(e)
[ VeVe'i P(e) & e t e

3e'( e t e & P(er)) ]
(e") INF(P)(er) )
Note that the structure as it is defined so far does not support axiom (3). Even
though there is no lower limit to P-events, there are smallest events that have
no proper parts, namely the events that correspond to single points.
In order to obtain a structure that supports (3), the construction would need
to adopt the assumption that single points in fact hide another infinity of
events. A concrete structure that illustrates this step can be built on the basis of
tuples of real numbers. If we call E 0 the part in below some infinitesimal
event e, we can set:
E 0 := { ](x,a); (x,b)[ I x, a, b e R and a<b }
For all events e, e' in E 0 :
e'<e :<=> e=](jt,a); (x,b)[ and e' = ](x,ay, (x,0[ and ]a';b'[ ]a;b[.
I will not further explore whether we can faithfully assume that the temporal
extension of all these events comes down to the same point in time (plausibly).
487
If we decide that we cannot, if we, in other words, maintain that events have
unique temporal extension, then we are forced to add infinitesimal elements to
the time line as well (see Robinson 1974: 244).
As an aside, I would like to mention that the initial event structure in this
example appears to shed light on a paradox about time that was posed by
Sebastian Lbner (p.c. in 1997). He pointed out that we have conflicting intuitions about time. On the one hand, we have a notion that there can be two
immediately adjacent but nonintersecting time phases. On the other hand, we
usually assume that between any two distinct time points there must be a third
one, distinct from both ( i.e. density). These intuitions are in fact not both supported in the same model, in the present construction. However, this model
construction can explain how we shift between two possible conceptualizations
of eventualities where one view supports assumption (i) and the other supports
assumption (ii).
4 Solution two: Sorites

The introduction of infinitesimally small events has turned out to be a consistent way to explain the conflicting intuitions listed in section one. However,
you might object that the solution locates the "hazy phase" at the wrong point.
You might maintain that the intuition that "all walkings consist of smaller
walkings" does not come about as our failure to see small events. Let us take a
closer look into this example. In fact, events for which we would cease to think
about a walking are still quite macroscopic. Certainly, one step is not a walking. Certainly, two steps are not sufficient for a healthy walking either. Certainly, there seems to be some boundary somewhere between three and 100
steps (very losely speaking) where the single steps end, and the real walking
starts?
Looking at it from the upper end, we might likewise propose that the intuition that "all walkings consist of smaller walkings" comes about differently.
Perhaps it means something like "if e is a walking, and if I take away one step
of e to get to e\ then e' will be a walking as well". We are not able to imagine
the case where a walk e minus one step e' results in something too small for a
walking. Cases like these have been discussed as the heap paradox in logic
and philosophy. We can recast it as weak homogeneity. Consider the following
condition.
(WH)
VeVe V e "( P(e) & e=e 'e "
(P(e ") P(e "))
Condition (WH) is more cautious than full homogeneity. It reflects an intuition

something like "if a P-event can be subdivided into two parts, then at least one
(the larger one?) is again P " . Let us assume that for each P, there is a uniform
measure which distinguishes those parts of P-events that are too small to be P,
488
Regine Eckardt
e.g. 'step' for walking, 'make a sound' for 'say something' etc. Assume moreover that all ordinary f-events in any ontology consist of a finite sequence of
such STEP(P)-events. Then, by induction, we will obtain instances of sorites
sequences (see Graff, 2000):
(14)
P(e)
(15)
If P(e) and e=e,e2 and STEP(P)(e2)
(16)
Any e in P{e) is linked to some event such that there is a finite sequence
e=e, e2, e3, ..., eê
and such that
then P(e,) still.
where all e e,+1 are linked by the sorites relation in (ii),
STEP(P)(e).
This shows that (WH), even though it was a careful assumption about homogeneity, can not be maintained once we spell out all assumptions that are characteristic for a heap paradox case. Cases like these have received extensive
discussion in the literature, and I refrain from recapitulating all the solutions
that were proposed. Instead, I will base my discussion on work by Graff, specifically Graff (2000). She offers a solution to the heap paradox that rests on
classical two-valued logic. This is advantageous for semantic modelling, because we need not burden semantic theory with controversial many-valued
logics. More importantly, however, G r a f f s solution is particularly relevant to
the present case insofar as it makes essential use of blind spots of the categorizing individual. Let me briefly outline her proposal.
Graff claims that, for any pair of objects (in our case: events e, e2) that are
immediately linked by the sorites relation in question, the following cognitive
effect occurs: Once we focus our attention on these two objects, their similarity
is so salient that we cannot, subjectively, judge one to have property but not
the other. This is a subjective and essentially context-driven judgement, as
Graff argues. If we decide for two events e and e2 where e is a walking and e2
is just one step shorter than e that e2 is likewise a walking, we tacitly expect
that the two events e' and e" which are a walking, a non-walking and separated
by just one step are just somewhere lower on the scale of ever smaller events.
This holds similarly for the dual case of two non-walkings. Globally speaking,
therefore, there exists a borderline, i.e. two events e, e2 such that
-,?(/) ->P(2) P{iS2)
Looking at things locally, however - and this seems to be the kind of perspective that feeds our armchair intuitions about event ontology - we maintain
principle (WH). Like for the previous solution, the condition on homogeneous
predicates needs to be adapted:
HOM(P) ^
[ VeVe'( P(e) & e'ee
Pie') INF(P)(er) )]
489
Note that in this case, we can not safely assume that all P-events have at least
some parts that are again P. There is a strict boundary somewhere that separates from INF(P). We are just unable to locate it precisely:
3e( P(e) Ve'( e t e
-,P(e) )
Hence, the sorites solution and the infinitesimal construction, even though both
capture our armchair intuitions about events, can be clearly distinguished by
the logical truths that are supported by either kind of model.
5 Outlook and summary

Both the construction of infinitesimal events and the sorites explanation appear
to capture some of the essence of how we think about very small events. At
present, I have no conclusive argument to favour one or the other treatment.
However, the existence of two logically distinct ways to fine-tune the notion
of homogeneity could be put to work to distinguish cases that could not be
differentiated by earlier theories. This is particularly interesting for cases
where predicates appear to be homogenous, but are not so perceived by speakers.
For example, Zucchi and White (Zucchi 2001) investigate the so-called
twigs and sequences puzzle. It has been observed that a sentence like (17) is
ill-formed, although geometry tells us that the initial segment of a line is again
a line and hence, each line consists of an infinity of shorter lines.
(17)
* John drew a line for 2 minutes.
The difference between a case like (17) and a (well-formed) sentence like
'John took a nap for 3 minutes' could be located in the different ways in which
we think about smaller parts of a nap, and smaller parts of lines. For example,
we could assume that objects like lines, sequences etc. are viewed as soriteshomogeneous but not infinitesimally homogeneous.
HOMAP) <-4 [ VeVe'( P(e) & e t e
3e( P(e) A Ve'( e t e - 4 -,P(e) )
P(e') INF{P){er) )]
Atelic predicates in the sense of aspect semantics, by contrast, could be required to be homogeneous in the strict sense.
HOM2(P) ^ [ VeVe'( P(e) & e t e - 4 P(e') INF(P){er) )
Ve(P(e) ^ 3e'( e t e & P{e')) ]
490
Regine Eckardt
This opens up a new possible way to distinguish between *John drew a line
for 2 minutes and John ate beans for 10 minutes, and hence could explain their
different behaviour.
To summarize, in this paper I drew attention to conflicting assumptions
about the lower end of event ontology that are suggested by different linguistic
phenomena. Homogeneity (as required in the modelling of aspect) suggests
that some properties apply to large events and all their smaller parts, no
matter how far down we look. Minimal-event-NPIs on the other hand suggest
that events can indeed be too small to count as an element in the extension of
(for the same, or similar, properties ). I suggested that the dilemma can be
resolved in two different ways.
The Infinitesimal Event construction rests on the assumption that there is a
conceptual 'blind spot' of speakers that drives them to make inconsistent assumptions about event ontology on different occasions. It essentially consists
in ignoring irrelevant material. As soon as we are forced to acknowledge the
existence of extremely small events, we enrich our ontology, and readjust
notions like HOM accordingly.
The starting point of the sorites solution is the hypothesis that we make idealised assumptions about the properties of very small events in everyday reasoning, just in order to keep matters simple. As soon as we are forced to think
seriously about these minute eventualities, we acknowledge our idealisation as
false, and readjust notions like HOM accordingly.
It appears very difficult to devise definite arguments in favour of one or the
other of these two options. However, their joint existence opens up new perspectives in the investigation of aspect and related issues.
References
Eckardt, R. (2003): Eine Runde im Jespersen-Zyklus: Negation, emphatische Negation
und
negative
polare
Elemente
im
Altfranzsischen.
Internetpublikation:
http://www.ub.uni-konstanz.de/kops/volltexte/2003/991.
Eckardt, R. (2005): Too poor to mention. Subminimal Eventualities and Negative Polarity Items. In Maienborn, C., A. Wllstein-Leisten (eds.): Events in Syntax, Semantics and Discourse. Tbingen: Niemeyer Verlag: 301-330.
Eckardt, R. (2006): Meaning Change in Grammaticalization. An Inquiry into Semantic
Reanalysis. Oxford University Press.
Fauconnier, G. (1975): Pragmatic Scales and Logical Structures. Linguistic Inquiry 6:
353-375.
Graff, D. (2000): Shifting Sands. An Interest-Relative Theory of Vagueness. Philosophical Topics 28.1:45-81.
491
Graff, D. (2003): Gap Principles, Penumbral Consequence, and Infinitely Higher-Order

Vagueness. In J.C. Beali (ed.): Liars and Heaps: New Essays on Paradox. Oxford:
Oxford University Press.
Krifka, M. (1989): Nominalreferenz und Zeitkonstitution: zur Semantik von Massentermen, Pluraltermen und Aspektklassen. Mnchen: Fink.
Krifka, M. (1995): The semantics and pragmatics of polarity items. Linguistic Analysis
25: 209-257.
Link, G. (1983): The logical analysis of plurals and mass terms: A lattice theoretic
approach. In: Buerle, R., U. Egli, Ch. Schwarze and . von Stechow (eds.): Meaning, Use and Interpretation of Language. Berlin: de Gruyter.
Lahiri, U. (1998): Focus and Negative Polarity in Hindi. Natural Language Semantics
6: 57-123.
Pin, Ch. (2000): The syntax and semantics of vgig. In G. Alberti and I. Kenesi
(eds.): Papers from the Pcs conference, Approaches to Hungarian. Szeged, Hungary: 201-236.
Robinson, A. (1974): Introduction to Model Theory and to the Metamathematics of
Algebra. Amsterdam: North Holland.
Van Rooy, R. (2003): Negative Polarity Items in Questions: Strength as Relevance.
Jounal of Semantics: 239-274.
Zucchi, S. (2001): Twigs, Sequences and the Temporal Constitution of Predicates.
Linguistics and Philosophy 24.2, 223-270.
Christopher Pin (Budapest)
Verbs of creation *
1 Introduction
Broadly speaking, verbs of creation denote the coming into being of the referent
of their direct internal argument as a result of the event named by them. Such
verbs are therefore often said to take 'effected objects'. Examples in which the
entity created is a physical object are shown in (1).
(1 )
(a)
R e b e c c a built a V i c t o r i a n style h o u s e .
(b)
Sarah compiled a program written in Scheme.
(c)
D a n i e l m a d e a C a e s a r s a l a d f o r dinner.
(d)
R e b e c c a d r e w a right t r i a n g l e .
(e)
Sarah painted a picture of the Hungarian parliament building.
(f)
Daniel wrote a paper o n verbs of creation.
The physical medium may sometimes take the fonn of a file saved on a computer
(e.g., in (lb), and possibly in (Id) and (11) as well), and yet computer files also
count as physical objects for present purposes, for they can be modified, copied,
deleted, misplaced, etc.
The deliberately broad characterization given above is intended to cover socalled performance verbs as well, though in this case the entities created are
events (namely, performances) and not physical objects:
(2)
(a)
R e b e c c a s a i d a p r a y e r f o r dinner.
(b)
S a r a h s a n g a s a d song.
(c)
Daniel recited a p o e m by E. E. C u m m i n g s .
(d)
Rebecca read
Fatelessness.
In (2), the events named by the verbs are themselves the performances created,
but the performances count as instances of the entities described by the object
I presented an earlier version of this paper at Workshop on event structure in linguistic form and
interpretation at Universitt Leipzig on 17 March 2004 and appreciate the questions raised and
comments made by that audience. I am grateful to Fabienne Martin for valuable discussions of
verbs of creation and for her insightful comments on a prefinal draft. I also thank Eric McCready
for useful feedback. Finally, I am indebted to the editors, Hannes Dlling, Tanja Heyde-Zybatow,
and Martin Schfer, for a wild patience. This work was supported by the Hungarian Scientific
Research Fund (OTKA TS 049873). The author's web address is (http://pinon.sdf-eu.org).
494
Christopher Pin
nomi phrases. In (2a), for example, Rebecca creates a prayer performance of a

prayer for dinner in saying such a prayer. Performance verbs may not be canonical verbs of creation, but insofar as they describe the creation of a performance
it is sensible to view them as a species of verbs of creation.
Finally, the characterization above is also meant to accommodate verbs whose
internal argument denotes an abstract entity that intuitively comes into being as
a result of the event in question:
(3)
(a)
(b)
(c)
(d)
Rebecca composed a symphony.

Sarah designed a Victorian style house.
Daniel invented a new salad.
Rebecca fabricated a story.
Normally, the composer of a symphony writes it down, but this is a practical

necessity given the usual complexity of symphonies and is not strictly necessary
for the truth of a sentence such as (3a). Similarly, an architect who designs a
house generally produces a blueprint of it, but again this is not necessary for the
truth of a sentence like (3b). Given the relative simplicity of salad recipes, it is
perhaps clearest in the case of (3c) that the invented salad need not be written
down (consider also (3d) in this respect). Even so, this is not to say that the
symphony, the house, the salad, and the story qua abstract entities need not be
physically 'anchored' or represented in some way. On the contrary, such entities are minimally physically represented in the brains of their respective creators immediately following the corresponding creation events,1 independently
of whether or not they acquire written representations as well. However, since
people tend to easily forget things, including things that they themselves have
created, the main condition for abstract entities which are created is that they be
represented in some physical medium, for otherwise it would be unclear what
their 'coming into being' amounts to.2
In sum, verbs of creation fall into three subclasses, depending on the semantic character of their direct internal argument: those denoting the creation of
a physical object, those denoting the creation of an event (henceforth, 'performance verbs of creation'), and those denoting the creation of an abstract entity.3
1
While it is an empirical question how ( e.g. ) a salad recipe is physically stored or neurally encoded
in a person's brain, I take such a neural configuration to be a physical object, much on a par with
a representation in terms of collections of bytes in computer memory or on a disk.
I thereby reject a Platonist view according to which preexisting abstract symphonies, houses, and
salad recipes are merely 'discovered' and not veritably created. On such a view, the verbs in (3 )
would not be verbs of creation.
According to Levin ( 1993, sect. 26), the verbs in ( la )( lc) are 'build verbs', those in ( Id H If),
( 2 b H 2 c ) , and (3a) are 'performance verbs' (she does not mention the use of say in (2a), nor
does she take read to be a verb of creation), and those in (3b)-(3d) are 'create verbs' (compose
also falls into this category but not as used in (3a)). Levin uses a combination of semantic and
morphosyntactic criteria for her classes, but the morphosyntactic criteria do not always obviously
yield semantically coherent classes. For example, it is odd to take write to be a 'performance verb'
in the same sense that sing is, and by Levin's morphosyntactic criteria alone eat would also count
Verbs of creation
495
fact, since physical objects and events are both concrete entities, the first two
subclasses fonn a natural subclass against the subclass of verbs denoting the
creation of an abstract entity. Nothing prevents a verb from belonging to more
than one subclass, and the following two sets of examples suggest that build and
make have a reading on which their internal argument refers to an abstract entity:
(4)
(5)
(a)
R e b e c c a built a V i c t o r i a n style h o u s e that S a r a h d e s i g n e d ,
(b)
S a r a h d e s i g n e d a V i c t o r i a n style h o u s e . R e b e c c a built it.
(a)
S a r a h m a d e a n e w s a l a d that D a n i e l i n v e n t e d ,
(b)
D a n i e l i n v e n t e d a n e w salad. S a r a h m a d e it.
In (4a), the object nomi pirrase of build appears to designate an abstract house
(namely, a house design) due to the relative clause with design (recall (3b)). A
syntactic variant on this is given hi (4b), where it is anaphorically dependent
on the object nomi pirrase of design. Either way, it is difficult to escape the
conclusion that the internal argument of build can sometimes refer to an abstract
entity. The pahs of sentences hi (5) point to the same conclusion for make.4 But
even granting that build and make are ambiguous with respect to the character of
their internal argument (physical object vs. abstract entity), the two meanings in
question are nevertheless intimately related, and any analysis should make this
explicit, especially because on both readhigs a physical object is created.
The idea that performance verbs of creation may take an abstract entity as
their internal argument is perhaps more evident. Indeed, this is the only way of
construing the sentence hi (2c), but it is the natural way of understanding those hi
(2a) and (2b) as well. In (2a), for example, Rebecca probably had either a fully
specified or at least a partially specified prayer hi mind to say, and the same pohit
applies to Sarah and the song in (2b). Pahs analogous to those in (4) and (5) can
also be provided for performance verbs of creation: 5
(6)
(7)
(a)
R e b e c c a said a p r a y e r f o r d i n n e r that S a r a h w r o t e ,
(b)
S a r a h w r o t e a p r a y e r f o r dinner. R e b e c c a said it.
(a)
D a n i e l p l a y e d a p i e c e f o r t h e p i a n o that R e b e c c a c o m p o s e d ,
(b)
R e b e c c a c o m p o s e d a p i e c e f o r t h e p i a n o . D a n i e l p l a y e d it.
as a 'performance verb'it is ruled out because its object is not 'effected', a semantic criterion.
However, such discrepancies need be examined more closely before reliable conclusions about
the semantic coherence of morphosyntactic criteria can be drawn.
If verbs denoting the creation of a physical object and performance verbs of creation form a
natural subclass (since they both denote the creation of concrete entities, as suggested in the text),
then we might expect any ambiguities to be between this natural subclass and the subclass of
verbs denoting the creation of an abstract entity. This expectation seems to be borne outat
least I could not find a verb of creation taking either a physical object or an event as its internal
argument.
Note that write in (6) is used in the sense of 'to author', and in this sense it is a verb denoting the
creation of an abstract entity, like those in (3).
496
Christopher Pin
A reasonable conjecture is that if a performance verb of creation heads a clause

that is aspectually an accomplishment (which is the intended reading of the sentences in (2) and (6)-(7)), then its internal argument denotes an abstract entity.
In this paper, I present a new approach to verbs of creation that pays equal
attention to each of the aforementioned subclasses. The analysis is cast in an
event semantic framework that is extended with an existence predicate and three
sorted domains of templates. In sections 2 and 3,1 develop the details of the new
account of verbs of creation, and in section 4 I briefly compare it to previous
analyses due to Dowty (1979), Krifka (1989, 1992), von Stechow (2001), and
McCready (2003a,b), arguing that the new account is more satisfactory than the
previous ones.
2 Creating physical objects

In this section, I propose a treatment of verbs denoting the creation of a physical
object (see (1)), which constitute the 'classical' case of verbs of creation. The
treatment proceeds in two steps: in section 2.1, I introduce a formal language
Lc for the analysis of verbs denoting the creation of a physical object, and in
section 2.2 I show how such verbs are analyzed with the help of Lc.
2.1 The model structure
The semantic analysis in section 2.2 will be formulated with the help of a standard higher order extensional type theoretical language Lc with lambda abstraction, identity, and the iota operator. A model for Lc is a pair Ji =
where
.y is a model structure and / is an interpretation function, 5? is in turn a tuple
(D, ,, T, C, -<, trace, exist, d0)
where D, , E, and are nonempty sets of individuals,6 c , -<, trace, and exist
are distinguished relations on one or more of these sets, and d0 is a special nil
individual in D. In this section, I elaborate on these ingredients of the model
structure.
The sets , E, and are pairwise disjoint and each fonns a subset of D.
Intuitively, O is a set of physical objects, E is a set of events, and is a set of
times. It is useful to introduce sorted variables for the elements of each of these
three domains:
(8)
: , y, , . . . (physical objects)
E: e, e', e", . . .
(events)
: t, t', t", ...
(times)
Henceforth, I employ the term
individual in the
sense of 'entity'
497
Verbs of creation
E includes states as well as events proper, and contains both instants and intervals. The unsorted variables a,b,c,...
range over the elements of D, which also
contains 'mixed' individuals that are composed of different sorts of individuals,
as we will see below.
The relation c on D D is a mereological relation ofproper part {a c b 'a is
a proper part of b'). It is a strict partial order (i.e., irreflexive, asymmetric, and
transitive), and the following notions are based on it and identity (in (9c), is a
one-place predicate with an extension in D) :
(9)
(a)
(b)
(c)
def
a C b = a C b V = b (a is a part of b)
def
e o i = 3 c [ c ( i A c b] (a and b overlap)
def
sum(7,P) = Vb[boa <-> 3c[P(c)Aco6]] (a is a sum o f P )
The overlap relation in (9b) allows the following witness principle for proper
part to be stated more compactly:
(10) Axiom. VaV6[a C b > 3c[c C bA -i(co a)]
(one proper part implies another)
This axiom excludes the possibility that an individual has a single proper part.
The smn relation in (9c) is demonstrably functional with respect to its individual argument: 7
(11) Fact. VaV6VP[sum(a,P) sum (,P)) > a = b] (uniqueness of sums)
This fact allows us to introduce iota tenns for siuns in case they exist:
Hpf
(12) () = ia[sum(a,P)]
(the smn of P)
A special case of smn is when two individuals are summed:

def
(13) a()b = ct(Ac[c C a Ve C b] ) (the smn of a and b)
The final mereological principle guarantees the existence of siuns whenever the
extension of is nonempty:
(14) Axiom. 3a[P(a)] > 3<7[sum(<7,P)]
(existence of sums)
This axiom has the consequence that D also includes 'mixed' individuals such
as siuns of physical objects and events and siuns of events and times. Although
such siuns do no harm, they do not belong to O, E, or P, in contrast to the 'pure'
individuals and then siuns from these subdomains. Letting \] designate the
7
This ultimately follows from the definition of sum and the antisymmetry of the part relation in
(9a).
498
Christopher Pin
closure of X under the sum operation, for a given set X, we may now define D
to be the closure of the union of O, E, T, and {i/o} :
Hef
(15) D = [OUEUTU{d0}]<j
(D is the closure of the union of O, E, T, and {ifo} under sums)
The relation -< is temporal precedence, which is a strict partial order on
{}
[\
i.e., a two-place relation on the closure of E U under the smn operation. At this
point, it is expedient to introduce sorted variables for the elements of [E U ],
all of which are temporal individuals (namely, events or times or any of their
sums):
(16) [E Li ]: s, s', s", ...
(temporal individuals)
The following principle states that temporal precedence is incompatible with

overlap (s -< s' 's temporally precedes s") :
(17) Axiom. VjVj'[j -< s'
> n ( s o j ' | ]
(temporal precedence excludes overlap)
In contrast to O and , has a linear structure, which means that any two times
either stand in the precedence relation or overlap: 8 :
(18) Axiom. \/t\/t'[t <t'\lt' -<tVtot' \j3t\3t2[t\ E t At2 E t'[h ~< 2 V2 -! il]]]
(linearity of times)
Finally, instants are times without proper parts:
(19) instant)/) =' St'[t' C t] (instant)
The relation trace on [E U ] is postulated to be functional with respect to
its time argument, as stated in (20a), and supplies the time (or temporal trace)
of a temporal individual ( t r a c e r s ) is the temporal trace of s'). Furthermore,
and unsurprisingly, the time of a time is simply that time, as poshilated in (20b).
(20) (a)
(b)
Axiom. VVVy[(trace(,.y)Atrace(/,.y)) -> t = t'\

(uniqueness of temporal traces)
Axiom. Vi [trace(t,t )] (the temporal trace of a time is that time)
Given the functionality of trace with respect to its time argument, we may speak
of the temporal trace of a temporal individual:
8
The final clause is needed because also contains sums of disconnected times (i.e., times that are
neither instants nor intervals ).
499
Verbs of creation
(21) ) = i[trace(,j)]
(the time of an event or a time)
The availability of enables the following two axioms to be stated more succinctly:
(22) (a)
(b)
Axiom. V j V j ' [ t ( j j') = ) ' ) ]

(the time of a sum of temporal individuals is the sum of their times)
Axiom. \/s\/s'[s -! s' > ) -< ' ) ]
(precedence of temporal individuals implies precedence of their times)
Temporal individuals that do not stretch infinitely into the future have an end.
The following definition determines what an end of a temporal individual is:
Hpf
(23) end(M) = instant)/) At C ) AV'[(' C (s) (' o ) ) ' -< t\

(t is an end of s)
If a temporal individual has an end, then it demonstrably has a unique end, but
since there may be temporal individuals that stretch infinitely into the future,
end is not a total function:
(24) Fact. VtVt'Vs[(end(t^)Aend(t',s))
- > =t'\
(uniqueness of ends of temporal individuals)
This fact allows us to speak of the end of a temporal individual, provided that it
has an end:
Hpf
(25) e{s) = l [end(, J)]
(the end of s)
The iota operator plays a role in definitions of stun (; (12)), temporal trace
(; (21)), and end (; (25)) above, and the question arises about what happens
when descriptions fonned with the help of the iota operator fail to be defined. I
adopt a Fregean strategy to this question and posfitlate a nil individual d0 as the
denotation of such undefined descriptions (see also Gamut 1991, chap. 5.2). This
appeal to do is simply a technical convenience (or hack) that enables Lc to remain
bivalent. As long as do is excluded from the denotation of most predicates that
we are interested in, then most claims about do will be false. For example,
given a one-place predicate house which denotes the set of houses, the statement
house(i/0) is false, because d0 is not a house. For the sake of completeness, the
semantics for the iota operator is given as follows (with respect to a model J
and an assignment function g) :
(26)
is the unique individual b of D such that

such an individual; otherwise
= do-
g\a^b] = ^ if there is
500
Christopher Pin
A temporal individual exists at a time just in case the value of applied to it is

precisely that timein this sense, a temporal individual has its time of existence
'built into it'. But so far there is no way of expressing the idea that a physical
object exists at a certain time but not at another. Naturally, given a canonical
physical object a, it is certainly the case that a exists at some time or other, but
this is weaker than saying that a exists at some specific time t. The relation exist
on (exist(A, t) 'a exists at t ') fulfills the need to talk about physical objects
existing at certain times. This relation is required to be divisive with respect to
its time argument:
(27) Axiom. VxVi[exist(x,t) > W[t' t > exist(x,i')]]
(existence of a physical object at a time implies its existence at all subtimes)
According to this axiom, if a exists at t, then a exists at any part of t. The second
principle states that exist is divisive with respect to its physical object argument
as well:
(28) Axiom. VxVi[exist(x,t) > Vy[y i - t existai)]]
(existence of a physical object at a time implies existence of its parts then)
This axiom asserts that if a exists at t, then every part of a exists at t, which is a
way of saying that physical objects lack temporal parts.
With the help of exist, we can define tensed predicates. For example, a tensed
version of c is defined as follows:
def
(29) \Zt y = C y A exist(x, t ) exist(y, t ) (x is a proper part of y at t)
In contrast to C, c f requires both of the physical objects to exist at t.
A simple example helps to illustrate the role of exist in this model structure.
Consider the partial model described in (30), where O contains seven individuals
and T, three. Note that the smn individuals in each set are guaranteed to exist
by the smn principle in (14). Moreover, is constrained so that t precedes t'.
Finally, the extension of exist at each of the times in is as specified.
(30) 0 = { , , , , , , }
=
{t,t',t()t'}
t -< t'
[ A x [ e x i s t ( x , g = {,,}
[Ax[exist(x,?')]]. #ig = {,,}
[Ax[exist(x,??')]J.#,g = M
With respect to this model, the statements y ( ) and y\Zt> {y ) are true,
but j e , (y ) is false because y does not exist at If y existed at t, then
by the principle in (28) both y and would exist at t, but this would contradict
501
Verbs of creation
the assumption that does not exist at t. The formula e x i s t ( x , t t') is also false:
if it were true, then by the axiom in (27) a would exist at t', and yet this would
contradict the premise that a does not exist at t'. More strikingly, neither
nor j exists at any time. Although both and y exist in the
sense of being elements of O (and hence fall within the range of the existential
quantifier 3), they do not exist at any time despite the fact that each of their parts
exists at some time. Consequently, statements such as a- c f are false, for any
value of t. This differentiation in tenns of existence among the siuns of O nicely
accounts for the intuition that - y and y are 'more natural' siuns than
or ^ : they are 'more natural' precisely because they exist at some time,
whereas the latter two do not. For example, if the elements of O were houses
and siuns of houses, then y and y would be siuns of coexistent houses,
whereas and y would be siuns of temporally disj oint houses.
2.2 The semantic analysis, I
With Lc at our disposal, we can turn to the analysis of verbs denoting the creation
of a physical object. The idea is that all of these verbs share a thematic relation
o n O as their common core of meaning. I begin by defining four properties
that a thematic relation may have and will then discuss the particular thematic
relation in question for verbs of creation. 9
A thematic relation R satisfies the property uniqueness of physical objects just
in case it is functional with respect to its physical object argument, as defined
hi (31a). This is an expression of thematic uniqueness, familiar from syntactic
theories: the thematic role hi question may be assigned to at most one argument.
The relation R satisfies the property uniqueness of evento just hi case it is functional with respect to its event argument, as stated in (3 lb). This in turn encodes
a prohibition against iterativity: the physical object may stand in this relation at
most once to a event.
(31)
(a)
UNI-O() = f V e V x V v [ ( ( e , x ) A ( e , . v ) )
= v]
(R satisfies u n i q u e n e s s of physical objects)

(b)
UNI-E() = f V e V e ' V x [ ( ( e , x ) R ( e ' , x ) )
e = e']
(R satisfies u n i q u e n e s s of events)
The relation R satisfies the property weak mapping to physical objects just hi
case any subevent of its event argument e is a part of a subevent of e that stands
hi the relation R to a part of the physical object argument of R, as shown hi (32a).
Notice that this property does not require every subevent of e to be mapped to a
part of a, but only that every subevent of e be covered by such a mapping. The
converse of this property is weak mapping to events, which is fulfilled by R only
9
Anyone familiar with Krifka's (1989, 1992) approach will notice that my analysis is similar in
spirit to his. Even so, there are differences as far as the treatment of verbs of creation is concerned,
as I will point out in section 4.2.
502
Christopher Pin
if any part of its physical object argument A- is a part of a part of A that stands in
the relation R to a subevent of the event argument of R, as formulated in (32b).
As before, this does not require every part of A to be mapped to a part of e, but
only that every part of A be included in such a mapping.
(32) (a)
(b)
WMAP-0()= f
VeVe'Vx[((e,x) Ae' C e H 3e"3y[e' e" Ae" C e Ay C x A ( e " , v ) ] ]
(R satisfies weak mapping to physical objects)
WMAP-E() *==
VeVxVy[((e,x) _v C )
3z3e'[v zAz
(R satisfies weak mapping to events)
C xAe'
C e A R(e' ,z)]]
The four properties in (31) and (32) captine a sense in which a physical object
may participate incrementally in an event. The next step is to introduce a particular thematic relation incremental that is postulated to have these properties:
(33) Axiom. UNi-o(incremental ) UNI-E(incremental) WMAP-o(incremental )
WMAP-E(incremental )
(incremental satisfies the four properties in (31) and (32))
Observe that the relation incremental is not tensed; it says nothing about whether
or not its physical object exists at a given time, hence it is neutral with respect
to whether or not its event argument designates a creation event. The thematic
relation created is a tensed version of incremental that requires the physical
object to exist at the end of the event in question and at no time during the event
before its end:
(34) created(e,x) = '
incremental(e,x) Aexist((e),x) \[( C x(e) At -< e(e) ) > êxist(i,x)]
(x is created in e)
Since created is partly defined in tenns of incremental, it clearly inherits the

four thematic properties that the latter has:
(35)
- o ( created ) - E ( created ) A w - o ( created ) A w A - E ( created )

(created satisfies the four properties in (31) and (32))
The common meaning component of verbs denoting the creation of a physical

object is precisely the relation created. For an illustration of created in action,
consider the proposed analysis of the sentence in (la), which is headed by build.
Although the details of the semantic composition may be worked out in various
ways, for present purposes I adopt Kratzer's (1996) proposal that the external
argument of a verb is not included in its semantic representation but rather enters
503
Verbs of creation
the semantic composition at a higher syntactic level. 10 This means that transitive
verbs (e.g., build) are treated as two-place relations between events and physical
objects and not as three-place relations that include an agent argument as well.
With this background, the verb build, the agentive element, and the two noun
pirrases of (la) are analyzed as follows (ignoring tense):
(36)
AyA<?[build (e ) created (<?, y )]
(a)
build
(b)
AGENT ' XPXxXe[P(e)
(c)
a Victorian style h o u s e
(d)
Rebecca
Aagent(e,x)]
XRXe\y[R{e,y)
victorian-style-house(_y)]]
rebecca
Assuming that the sentence in (la) has the schematic syntactic structure indicated in (37a), its corresponding event predicate is shown in (37b), which is the
straightforward result of type-driven functional application:
(37)
(a)
(b)
[(Rebecca) [(AGENT) [(build) (a Victorian style house)]]]

Ae[3y[build(e) A created (e,i) victorian-style-house(y)] A
agent(e, rebecca)]
The event predicate in (37b) denotes the set of events in which Rebecca builds a
Victorian style house. Suppose now that one of these events is <?" and the house
that she builds in e" is z, as depicted in Figure 1. Due to the role of created,
exists at the end of e" (= (<?")) but not at any time earlier in e". However, this
still allows for various proper parts of to exist earlier. As shown, is created
in e and begins to exist at the end of e, and y is likewise created in e' and begins
to exist at the end of <?', where and e are proper parts of y and <?', respectively.
Note that weak mapping to physical objects (see (32a)) does not require every
subevent of e" to be a creation event. The mixing of cement, the sawing of
wood, and the plastering of walls are all subevents of e", yet none of these are
creation events per se. What weak mapping to physical objects requires is that
each of these events be a part of a creation subevent of <?", which is plausible.
For example, the building of the foundation of is a creation subevent of e" that
has the mixing of cement as a subevent even though the latter is not a creation
subevent. Conversely, weak mapping to events (see (32b)) does not demand that
every part of be created in a subevent of e". The door and windows of were
not created in e", because Rebecca bought them prebuilt, ready to be installed.
What weak mapping to events demands is that they each be a part of a part of
that is created, which seems correct. For instance, the three windows are parts
of the facade of z, which is created in a subevent of e".
10
For Kratzer, the higher level is a so-called Voice Phrase, but the exact label of this syntactic
projection is not crucially relevant here.
504
Christopher Pin
e{)
e(e>)
I
created (e,*)
exist(e(e),x)
(")
created (<?',y)
exist^fV),;;)
.Mol
created (e",)
exist(e(e"),z)
Figure 1: Building a house
Although aspectual issues are not the mam focus here, I point out that the
event predicate in (37b) is demonstrably quantized (which is characteristic of
accomplishments). This is a consequence of the fact that the nommai predicate
victorian-style-house is quantized and of the properties of the thematic relation
created. Quantized reference for one-place predicates of individuals is defined
in (38a), and the corresponding result for the event predicate in (37b) is given in
(38b). 11
Hpf
(38) (a)
(b)
QUA(P) = \ / a \ / b [ { P ( a ) A b \ Z a ) ^ ^ P ( b ) ] (P is quantized)
Fact. QUA(Ae[3y[build(e) created {e,y) victorian-style-house (y)] A
agent(e, rebecca)])
(the event predicate in (37b) is quantized)
A remark for those familiar with Krifka's theory is that the proof of the quantization of an event predicate based on a verb denoting the creation of a physical
object together with a quantized nommai predicate restricting its internal argument does not depend on Krifka's stronger property of mapping to objectsthe
weaker one in (32a) is sufficient.
The analysis of the other sentences in (1) are analogous to that of (la). In
each case, the thematic relation created is employed to connect the physical
object created to the event in question. As mentioned in section 1, the physical
object created may be a collection of bytes, e.g., a binary file saved on a disk,
like the program that Sarah compiles in (lb), but it counts as a physical object
nevertheless.
11
The proof makes use of uniqueness of objects, uniqueness of events, weak mapping to objects,
and of course the fact that victorian-style-house is quantized (compare Krifka 1992, TI 1, p. 41 ).
Verbs of creation
505
3 Creating events and templates

In this section, I propose a treatment of performance verbs of creation (see (2))
and those denoting the creation of an abstract individual (see (3)). As before, the
treatment proceeds in two steps: in section 3.1,1 extend the model structure for
Lc with three domains of templates and two new relations, naming the extended
language L+, and in section 3.2 I show how verbs of creation belonging to these
two subclasses are analyzed with the help of L+.
3.1 Extending Lc
To help fix intuitions, consider the three individuals depicted in Figure 2. At the
left, denoted by A-, is the Victorian style house that Rebecca built and now lives
in. At the right, designated by y, is the architectural plan of the abstract house
design realized on a sheet of paper (imagine a blueprint). Finally, at the bottom,
named by x, is the abstract house design (or a house template, see below). 12
Clearly, these are three different things, for both the physical house and the
architectural house plan could get destroyed in a fire, but no fire could touch the
abstract house design per se. 13 Furthermore, Rebecca lives in the physical house
but she could not live in the architectural house plan or in the abstract house
design. Yet all three have in common that they may be created.
The relation between the physical house and the abstract house design in Figure 2 is one of instantiation, symbolized as > : A>X 'A instantiates x'. The
relation between the architectural house plan and the abstract house design is
one of representation, symbolized as =x y
x 'y represents '. Both of these
relations are irreflexive, asymmetric, and intransitive, thus the abstract house
design neither instantiates the physical house nor represents the architectural
house plan. With the help of these two relations it is possible to define a notion
of derivative instantiation, designated by > ' , which relates the physical house to
the architectural house plan: \>'y 'A derivatively instantiates y\ The physical
house derivatively instantiates the architectural house plan because there is an
abstract house design that the fonner instantiates and the latter represents.
A conspicuous difference between the physical house and both the abstract
house design and the house plan in Figure 2 is that the fonner has, but the latter
two lack, a window from the second story. The idea is that a concrete individ12
For convenience, the abstract house design in Figure 2 is displayed in the form of an image of
a house. But this could be misleading, because the abstract house design is not an image, and
it would be more apt to think of it as a set of propositional functions describing the design in
question, e.g., {the facade of has at least two windows, has a slanted roof, . . . } . Naturally, the
design may be more or less specified, and less specified designs would in this way be treated as
subsets of more specified designs.
13
As we will see below, especially in connection with (41 ) and (42), an abstract house design ceases
to exist at a time in a certain sense if it is not represented by a physical object existing at that time.
Thus, a fire may indirectly affect the existence of an abstract house design by destroying all of its
representations, but the point remains that a fire cannot touch an abstract house design directly.
506
Christopher Pin
".
y-
(>0
>
Figure 2: A house (), a house template (), and a house plan (y)
ual may be (in fact, usually is) more detailed than an abstract individual that it
instantiates, provided that its extra detail does not conflict with the information
that the abstract individual specifies. Thus, although the abstract house design
does not specify a window from the second story, it also does not specify that
there is no such window, hence the physical house may have such a window
and not conflict with the abstract house design. This pennits a concrete individual to instantiate many different abstract individuals, where the latter differ
from each other according to the information (greater or less detail) that they
specify. In contrast, the relation of representation as construed here is much less
liberal and requires a tight fit between the representing individual and the represented individual. This means that if the house plan y had a window from the
second story, then it would not represent this particular abstract house design x.
A way to capfiue this is to say that any concrete individual represents at most
one abstract individual (see (43)).
Nafiually, and as just suggested, there may be many physical houses that instantiate the abstract house design in Figure 2, just as there may be many representations of it. Nor does every representation of an abstract house design have
to be realized on paper, though this is probably the standard way of representing
house designs. If Sarah, who is an architect, creates an abstract house design,
she may initially only have it 'in her head', so to speak, before she gets a chance
to make a blueprint. But however exactly this abstract house design is neurally
encoded in her brain, the particular neural configuration also counts as a physical
representation of the abstract house design that she created, though obviously it
is one that only she has access to. Her abstract house design can also be represented by a computer file that is created with the help of a draw program. There
may also be many abstract house designs, which are individuated in tenns of
the information they specify. An abstract house design that specified one tall
Verbs of creation
507
window on the facade instead of two small windows would be distinct from the
abstract house design x, despite the fact that they would have everything else in
common.
On the present conception, the abstract house design in Figure 2 is an abstract
individual and not a (first order) property or a kind. This is a somewhat delicate
distinction, because properties and kinds may be treated as individuals, and yet
such a possible treatment should not affect the distinction in question. For example, Dlling (2001) analyzes (first order) properties as (first order) individuals,
calling them 'kinds'. He relates ordinary individuals to kinds with the help of a
relation INST 'instance of'. For instance, he would fonnalize the statement that
is a house as 'A INST house', which is paraphrasable as 'A is an instance of the
kind house'. More generally, his kinds play the same role that (first order) predicates play in L+ (and Lc). However, Dlling's strategy of treating properties as
kinds qua individuals is orthogonal to (and hence compatible with) the present
point that the abstract house design is an abstract individual but not a kind qua
individual. In L+, the formalization of the statement that is a house (namely,
an abstract house design) would be ' H O U S E ( X ) ' , where H O U S E is a (first order)
predicate of abstract house designs. Observe that if we adopted Dlling's approach here and treated H O U S E as a kind qua individual, the formalization of the
previous statement would be 'x INST H O U S E ' , which would also bring home the
point that is being treated as a particular individual (albeit abstract) and not as
a kind qua individual.
The three-way distinction drawn for houses in Figure 2 is more generally
applicable. Take salads: a physical salad is something that can be eaten, the
salad recipe is something that it instantiates, and the salad recipe in a recipe
book is a representation of the recipe. More subtle are computer programs: a
physical program is a binary file that can be executed, it instantiates an abstract
program, and the source code saved in a file represents the abstract program.
Or consider prayers: an event in which a prayer is said instantiates the abstract
prayer, which is in fiun represented by the prayer in a prayer book. Clearly,
songs, poems, and symphonies are analogous to prayers, differences in structure
aside.
In order to be able to talk about abstract individuals like the abstract house
design in Figure 2, I extend the model structure for Lc with three pairwise
disjoint nonempty sets of templates:14
14
The term 'template' may not be ideal, but I prefer it to 'type', which would have unintended
connotations in the present context. With other applications in mind, Levy and Olson (1992)
construe templates as physical objects that determine artifacts of a certain type. For example, a
cookie cutter is a template for them, because under the right conditions it determines cookies of
the same size and shape. The templates that I appeal to, although abstract individuals, are much
more akin to cookie cutters than to properties or universale.
508
Christopher Pin
(39) O,,, : , y, , . . . (templates for physical objects)

E,: e, e', e", . . . (templates for events)
T, : t, t', t", . . . (templates for times) 15
Defining the set M to be the union of these three sets of templates, we then
introduce variables for templates of any sort:
def
(40) (a)
(b)
M = OmyjEmyjTm
M: m, m', m", . . .
(M is the union of Om, Em, and Tm)

(templates)
A notion of derivative existence at a tune for templates can be defined in tenns

of the existence at a time of physical objects that represent them, as in (41).
The notion of representation (=>), a relation on O x M , was introduced above to
relate the house plan to the house design in Figure 2.
(41)
def
E X I S T ( m , i ) = 3 x [ e x i s t ( x , i ) A x => m ]
( m d e r i v a t i v e l y exists at t)
In Figure 2, the house design derivatively exists at a tune t if the house plan
exists at t. The following principle requires every template to derivatively exist
at a tune:
(42) Axiom. Vm[3[EXIST(m,)]]
(templates derivatively exist at a time)
This axiom requires every template to be existentially anchored to a physical

object that represents it. Without this axiom, templates could just as well be
Platonic objects with no necessary existential tie to physical objects. Note also
that a template ceases to exist at a tune (in the sense of EXIST) once there is no
longer any physical object representing it that exists at that time.
The tight fit between a physical object and a template that it represents (recall
the discussion above) is captured by the following principle, which states that
is functional with respect to its template argument:
(43) Axiom. VxVmVm'[(x = > m A i = > m ' ) - n n = m']
(uniqueness of templates in representation)
However, the converse should not hold, because a given template may be represented by more than one physical object (e.g., imagine several blueprints of a
house design).
Although every template is represented by a physical object, it need not be
instantiated by any individual. 16 The notion of instantiation (> ), which is a rela15
16
The templates for times are included for the sake of completeness, though it is admittedly not so
clear whether they are really needed. Perhaps theories of time (e.g., a theory of linear time vs. a
theory of branching time ) are examples of templates for times.
In terms of houses, this means that there could be an abstract house design and a house plan that
509
Verbs of creation
tion on (O U E U ) M, was introduced above to connect the physical house to

the house design in Figure 2. If a template is instantiated by a concrete individual, then the concrete individual has to be of the appropriate sort. Specifically,
if a template for physical objects is instantiated, it is instantiated by a physical
object, if a template for events is instantiated, it is instantiated by an event, and
if a template for times is instantiated, it is instantiated by a time:
(44) Axiom. ViVm[i7 > m > (3x[m = ] <-> 3x[a = ] ) (3e[m = e] <-> 3e[a = e] ) A
(3t[m = t] <->3[ = ])]
(sortal correspondence for the instantiation of templates)
A notion of derivative instantiation, a relation o n ( 0 u u r ) x 0 , can be defined
in terms of instantiation and representation, as in (45). This notion was appealed
to above to relate the physical house to the house plan in Figure 2.
(45) a l>'x =' 3m[(7 > mAx => m]
(a derivatively instantiates x)
Templates may have subtemplates. For instance, the house template in Figure 2
has a subtemplate that leaves out the information about the door and the windows. In Ime with the present strategy of tying templates as tightly as possible
to their physical representations, we define a notion of proper subtemplate (')
in tenns of representation and proper part:
def
(46) m c ' m' = 3x3_y[x = m Ay = m' Ax C ]
(m is a proper subtemplate of m')
With respect to Figure 2, this definition states that any template is a proper subtemplate of the house template just in case it is represented by a proper part of
the house plan.
With the notion of proper subtemplate in hand, it is straightforward to define
template analogues of the mereological relations in (9), (12), and (13) (where Q
in (47c) is a one-place predicate of templates):
(b)
def
m C' m' = m c ' m' V m = m' (misa subtemplate of m')
def
m o' m' = 3m" [m" C' m m" C' m'] (m and m' overlap)
(c)
SUM(m,>) = Vm'[m'o'm<-> 3m"[>(m") Am"o'm / ]]
(d)
(e)
a'(Q) ='Hpf
im[sUM(m,)] (the sum of )
m m' = a ( l m " [m" C' m V m" C' m'] ) (the sum of m and m')
(47) (a)
(m is a sum of Q)
While the proper subtemplate relation inherits the properties of the proper part
relation (hence it provably is a strict partial order and satisfies the template analogue of the witness principle in (10)), the template analogue of the smn princirepresents it without there also being a physical house that instantiates the abstract house design.
510
Christopher Pin
pie in (14) does not automatically follow. 17 Consequently, the existence of smns
of templates has to be ensured separately: 18
(48) Axiom.
(3m[(m)] ( V m [ ( m )
3 x [ m = x]] V V m [ ( m )
3 e [ m = e]] V
Vm[>(m) > 3t[m = t]] ) ) > 3 m [ s U M ( m , ) ]

(existence of sums for templates)
We also have to allow for the possibility that descriptions of smns of templates
fonned with the help of the iota operator in (47d) are not defined, which is the
case if the denotation of Q is empty. Parallel to the semantic clause in (26), I
assmne that such descriptions denote the nil individual d().
Three mapping principles regulate the relations of instantiation and representation. The first, mapping from templates to instantiations, states that if an
individual a instantiates a template m and m' is a subtemplate of m, then there
is a part of a that instantiates m', as in (49a). The second principle, mapping
from templates to representations, asserts that if a physical object represents a
template m and m' is a subtemplate of m, then there is a part of that represents
m', as in (49b). Finally, the third principle is the converse of the previous one
and states that if a physical object represents a template m and y is a part of x,
then there is a subtemplate of m that y represents, as in (49c).
(49) (a)
Axiom.
V a V m V m ' [ ( a > m A m ' ' m ) - > 3b[b C a A f c > m ' ] ]
(mapping f r o m templates to instantiations)

(b)
Axiom.
V x V m V m ' [ ( x => m > m ' C ' m ) > 3_y[y = m']]
( m a p p i n g f r o m t e m p l a t e s to r e p r e s e n t a t i o n s )
(c)
Axiom.
V x V y V m [ ( x = > m > y C x ) > 3 m ' [ m ' C ' m A y => m']]
( m a p p i n g f r o m r e p r e s e n t a t i o n s to t e m p l a t e s )
The converse of the principle in (49a) would not be desirable, because instantiations may be more detailed than the templates that they instantiate, as discussed
in connection with the window from the second story in Figure 2, and the converse would require every part of an individual to correspond to a subtemplate of
a template that it instantiates, which would be too strong. However, hi the case
of representations, both the principle in (49b) and its converse hi (49c) are desirable, precisely because of the tight fit between physical objects and the templates
that they represent.
17
18
It would follow if there were a principle guaranteeing that for every physical object there is a template that it represents. However, such a principle would make the connection between physical
objects and templates even tighter than envisioned here. In particular, the present approach allows
for there to be more physical objects than templates, because there may be physical objects that
do not represent templates but by (42) every template is represented by a physical object (that by
(43) represents only it).
For a more general formulation, we would have to allow for 'mixed' templates consisting of
templates of different sorts, not dealt with here. The principle in (48) only guarantees the existence
of sums of templates of the same sort.
511
Verbs of creation
Taking stock, the model structure forL c has been extended with three domains
of templates and two new relations connecting templates to concrete individuals.
The extended language is L+, and a model forL+ is a pair J = {-y J ) , where
S^ is a model structure and / is an interpretation function. S? is now a tuple
{D,0,E,T,0m,Em,Tm,n,~<,
trace, exist, >,
, d0)
where Om, E Tm are nonempty sets of templates for physical objects, templates
for events, and templates for times, respectively, > is a relation of instantiation
(between concrete individuals and templates),
is a relation of realization (between physical objects and templates), and the other components of S? are as
they are in the model structure for Lc. Clearly, L+ is more expressive than Lc,
though not in a logical sense but rather in the sense that L+ can express things
about sorts of individuals (namely, the three sorts of templates) that Lc cannot
say things about. I will make use of this greater expressibility in the next section.
3.2 The semantic analysis, II
The idea about performance verbs of creation and verbs denoting the creation
of abstract individuals is that they both take an internal argument designating
a template. To implement this idea, we need to introduce a thematic relation
INCREMENTAL between events and templates that is the analogue of the relation
incremental between events and physical objects. Thematic properties corresponding to those in (31a) and (32a) are also called for, which are defined as
follows (where S is a two-place relation between events and templates):
(50)
(a)
UNI-O(S) = f V e V m V m ' [ ( S ( e , m ) A S ( e , m ' ) ) ^ m = m']

(,S satisfies uniqueness of templates)
(b)
WMAP-O(S') ^ ' V e V e ' V m ^ ^ m l A e '
C c h
3e"3m' [e' C e" Ae" C e Am' C' mAS(e",m')]]

(,S satisfies w e a k mapping to templates)
(c)
WMAP-E(S') = f V e V m V m ' [ ( S ( e , m ) A m '
m)
3m"3e' m' C' m" A m" C' m A e' C e A S(e', m" )]]
(,S satisfies w e a k mapping to events)
The relation INCREMENTAL is poshilated to have these three thematic properties:

(51) Axiom. UNI-0(|NCREMENTAL)AWMAP-0(|NCREMENTAL)A
WMAP-E( INCREMENTAL)
(INCREMENTAL satisfies the three properties in (50))
Although we could define a template analogue of uniqueness of events (see

(31b)), the relation INCREMENTAL should not have this property, because it is
possible to create a given template more than once. For example, if Sarah designs a Victorian style house (say the house template in Figure 2), it is certainly
512
Christopher Pin
possible for someone else to independently design exactly the same house on
another occasion. Indeed, even Sarah herself may design the same house twice,
especially if she forgets and loses all record of her first design.
The next step is to define a template variant of the thematic relation created
introduced in (34). The new relation is CREATED, defined with the help of
INCREMENTAL and created as follows:
(52)
def
CREATED(e,m) =
I N C R E M E N T A L ^ , m ) 3 x [ c r e a t e d ( e , x ) A x = m ]
(m is created in e)
This says that a template m is created in an event e just in case m is incremental

in e and there is a physical object a created ili e which represents m.
For an application of the relation CREATED, consider the analysis of the sentence in (3b), headed by design (compare (36)). The verb design denotes a twoplace relation between events and templates for physical objects, as in (53a). The
agentive element is given in (53b) (repeated from (36b)), and the nomi phrase
a Victorian style house, is analyzed as an existential quantifier over house templates, as in (53c). This nomi phrase (among many others) is thus treated as
systematically ambiguous between a existential quantifier over physical houses
(see (36c)) and one over house templates, where the index 'zw' marks the latter.
Last but not least, Sarah is treated as a constant, as in (53d).
(53)
(a)
d e s i g n s A x A e [ d e s i g n ( e ) CREATED(e,x)]
(b)
(c)
AGENT ' XPXxXe[P(e)

Aagent(e,x)]
a Victorian style house,
ASAe[3x[S(e,x) AVICTORIAN-STYLE-HOUSE(x)]]
Sarah
sarah
(d)
Given the schematic syntactic structure displayed in (54a), the corresponding

event predicate is shown hi (54b) (compare (37)).
(54) (a)
(b)
[(Sarah) [(AGENT) [(design) (a Victorian style house, )]]]

Ae[3x[design(e) ACREATED(e,x) AVICTORIAN-STYLE-HOUSE(x)] A
agent(e, sarah)]
This event predicate denotes the set of events in which Sarah designs a Victorian
style house template. Due to the definition of CREATED, a physical object representing the Victorian style house template is created as a result of such an event,
but note that design crucially denotes a relation between designing events and
templates, and not between designing events and representations of templates.
Moreover, there is no entailment that a physical house instantiating the house
template is created.
The other sentences in (3) with compose and invent receive the same kind of
analysis. However, one difference is that compose takes a template for events as
513
V e r b s of c r e a t i o n
its internal argument. Consider an analysis of the verb phrase of the sentence in
(3a):
(55)
(a)
c o m p o s e s AeAe[compose(e)
(b)
a symphony,,,
(c)
compose a symphony,,,
CREATED(e,e)]
ASAe[3e[S(e,e)
SYMPHONY(e)]]
Ae[3e[compose(e,e) CREATED(e,e) SYMPHONY(e)]]
The event predicate in (55c) denotes the set of events in which a symphony template is created. By the semantics of C R E A T E D , a representation of the symphony
is also created, but of course no instantiation of the symphony (i.e., no performance) is entailed.
Performance verbs of creation such as recite (see (2)) differ from those denoting the creation of a template in that they entail an instantiation of the template
m question. More precisely, such verbs take an internal argument denoting a
template for events and they assert an instantiation of this template. However,
the relation C R E A T E D cannot be used to captine this, precisely because no template is createdthe individual created is the event (i.e., the performance) itself.
To till the gap, a new thematic relation P E R F O R M A N C E may be defined in tenns
of I N C R E M E N T A L and instantiation:
(56)
PERFORMANCE^,e)
def
=
I N C R E M E N T A L ^ , e) e O e
(e is a p e r f o r m a n c e o f e)
An event e is a performance of a template e just ili case e is incremental ili e and

e instantiates e. In the case of performances, no physical object is created, hence
there is no need to appeal to the relation exist.
With the relation P E R F O R M A N C E available, consider the analysis of the sentence m (2c), headed by recite.
(57)
(a)
recite
(b)
a poem,,, b y E. E. C u m m i n g s
A e A e [ r e c i t e ( e ) P E R F O R M A N C E ^ , e)]
(c)
Daniel
ASAe[3e[S(e,e)
POEM-BY-E.E.-CUMMINGS(e)]]
daniel
Note that the noun phrase a poemm byE. E. Cummings is an existential quantifier
over templates of poems by E. E. Cummings (and poem templates are templates
for events). Given the syntactic structure sketched hi (58a), the resulting event
predicate for the sentence is shown in (58b).
(58)
(a)
[ ( D a n i e l ) [(AGENT) [(recite) (a poem,,, b y E . E .
(b)
Ae[3e[recite(e) PERFORMANCE^,e) POEM-BY-E.E.-CUMMINGS(e)]
Cummings)]]]
a g e n t ( e , d a n i e l )]
The other sentences hi (2) are treated in a similar fashion.
514
Christopher Pin
I began in section 1 with three subclasses of verbs of creation and have shown
how the verbs of each subclass are handled in the present approach. Verbs denoting the creation of a physical object (see (1)) are analyzed as relations between
events and physical objects with the help of the thematic relation created (e.g.,
(36)). Performance verbs of creation (see (2)) are treated as relations between
events and templates for events (e.g., (57)) with the aid of the thematic relation
PERFORMANCE. Finally, verbs denoting the creation of a template (see (3)) are
analyzed as relations between events and templates (e.g., (53) and (55)) with the
assistance of the thematic relation CREATED. While these are indeed the primary
analyses, the data indicate the need for sort shifters that are able to shift the
internal argument of a verb from one sort to another.
Recall that the pairs of sentences in (4) and (5) suggest that verbs denoting
the creation of a physical object sometimes appear to be able to take templates as
their internal arguments. In (4a), if the noun plnase a Victorian style house that
Sarah designed is treated as an existential quantifier over house templates, which
is reasonable in the light of design (see (53a)), then it will not be able to combine
with build as analyzed in (36a) due to a sortal conflict. A solution is to postulate
a particular sort shifter (SSH- 1 ) that applies to a verb denoting a relation between
events and physical objects and yields a verb denoting a relation between events
and templates such that the templates are instantiated by a physical object:
(59)
S S H - 1 -VA
kRkxXe[3y[R{e,y) Ay
> x]]
( s o r t s h i f t e r 1)
Applying the shifter SSH- 1 to build, we get:

(60)
S S H - 1 ( b u i l d ) ' V A A x A e [ 3 _ y [ b u i l d ( e ) c r e a t e d (e,_y) A y O x ] ]
Assuming that a Victorian style house that Sarah designed is analyzed as the existential quantifier over house templates in (61 a) and that the sentence in (4a) has
the schematic syntactic structure in (61b), then the corresponding event predicate is displayed in (61c).
(61)
(a)
a Victorian style house, that S a r a h
designed
A S A e [ 3 x [ S ( e , x ) VICTORIAN-STYLE-HOUSE(x) A 3e'[design(e')
CREATED^',)
(b)
[(Rebecca)
Aagent(e',sarah)]]]
[(AGENT) [(SSH-1 (build)) (a V i c t o r i a n style house,,,
that
Sarah
designed)]]]
(c)
A e [ 3 x [ 3 _ y [ b u i l d ( e ) A c r e a t e d (e,_y) _ > ] V I C T O R I A N - S T Y L E - H O U S E ( x )
3 e ' [ d e s i g n ( e ' ) C R E A T E D ^ ' , ) A a g e n t ( e ' , s a r a h )]] a g e n t ( e , r e b e c c a )]
This event predicate denotes the set of events in which Rebecca builds a physical
object that instantiates a Victorian style house that Sarah designed. The analysis
of the second sentence in (4b) would also make use of S S H - 1 ( Z W / W ) , and the
pronomi it would refer to the Victorian style house template that Sarah designed
Verbs of creation
515
that is introduced in the first sentence.

But SSH-1 is not the only sort shifter needed. Imagine the following sentence
m the context of an architect's office, where the house on the M'ali refers to the
blueprint of a house hanging on the wall:
(62) Rebecca built the house on the wall.
To handle this case, we need a version of build that takes a physical object representing a template as its internal argument and asserts that this representation
is derivatively instantiated (see (45)). Such a version is derived with the aid of
the sort shifter in (63a) (SSH-2), which is applied to build in (63b).
(63) (a)
(b)
SSH-2
XRXzXe\3y[R{e,y) Ay\>'z]]
(sort shifter 2)
SSH-2(build) 'v AzAe[3y[build(e) created (e,y) Ay >'z]]
Applied to a physical object z, the relation in (63b) denotes the set of events in
which a physical object y is built that derivatively instantiates z.
A sort shifter is not required for the analysis of the pairs of sentences in (6)
and (7), because performance verbs of creation (e.g., say, play) already receive
a primary treatment in which they assert that a template for events is instantiated. However, the following example suggests that such verbs sometimes take
a physical object representing a template for eventswhich, by the axiom in
(43), is uniqueas then internal argument:
(64) Rebecca said the prayer on page 25 of the prayer book.
To handle this use of say, a sort shifter is needed (SSH-3) that applies to aperformance verb of creation and yields a verb taking a physical object as its internal
argument that represents a template for events which the events denoted are performances of. The shifter SSH-3 is defined in (65a), the primary analysis of say
as a performance verb of creation is given in (65b), and the result of applying
SSH-3 to say is displayed in (65c).
(65) (a)
(b)
(c)
SSH-3 -VA ASAj'Ae[3e[S(e,e)Aj'=>e]] (sort shifter 3)

say ' AeAe[say(e) A PERFORMANCE(e,e)]
SSH-3(say) ^ AyA<?[3e[say(<?) A PERFORMANCE^, e ) A_y = e]]
Applied to a physical object y, the relation in (65c) denotes the set of events
which are saying performances of a template for events e that y represents.
I conclude with a brief mention of yet another sort shifter (SSH-4) that takes
a verb denoting the creation of an abstract individual and yields a verb taking a
physical object as its internal argument which instantiates a template for physical
objects. This sort shifter is needed for examples such as the following:
516
Christopher Pin
(66) Sarali designed the house on the comer.
The definition of SSH-4 is given in (67a) and its application to design (see (53a))
is shown in (67b).
(67) (a)
(b)
SSH-4 ~i-ASAj'Ae[3X[S(E,x)AJ'>x]] (sort shifter 4)

SSH-4(design) ^ AyA<?[3x[design(<?) A CREATED(e,x) Ay > x]]
Applied to a physical object y, the relation in (67b) denotes the set of events in
which a template for physical objects is designed such that y instantiates x.
4 Comparisons
In this section, I briefly contrast my proposal for verbs of creation with four
previous ones due to Dowty (1979), Krifka (1989, 1992), von Stechow (2001),
and McCready (2003a,b), respectively. My aim is not to provide an extended
commentary on any of these approaches (which would take me far afield) but
rather to highlight the ways in which theirs differ from mine and are arguably
less satisfactory as accounts of verbs of creation.
4.1 Dowty (1979)
Dowty suggests in passing that verbs of creation are semantically decomposed
with the help of the predicates CAUSE and BECOME, which are used for the
analysis of accomplishments in his framework:
(68) John painted a picture.
[[John paints] CAUSE [BECOME [ picture existe]]]
(Dowty 1979, (98), p. 91)
Without going mto the technical question of how CAUSE and BECOME are interpreted, the intuitive meaning assigned to this representation is that John's painting activity causes a picture to come into existence.
As von Stechow (2001, sect. 4) points out in detail, the major flaw in Dowty's
analysis in (68) is that the corresponding truth conditions prohibit any picture
at all from existing at the beginning of the interval of painting, and yet this is
clearly too strong, because there may well be (other) pictures that exist in the
world at the beginning of this interval. Von Stechow also argues that this flaw is
not so easy to fix in Dowty's framework.
But putting this difficulty aside, I point out that Dowty's treatment does not
handle performance verbs of creation (see (2)) or those denoting the creation of
an abstract individual (see (3))at best it serves for verbs denoting the creation
of a physical object. Interestingly, Dowty is aware of this shortcoming. For
example, he is concerned (pp. 186-187) that perform a sonata cannot plausibly
be analyzed as [CAUSE [BECOME [a sonata eris]]]. He then suggests that
Verbs of creation
517
John performs a sonata might be treated as having the fonn [[John a c t e ] CAUSE
[ T R A N S P i R E ( a sonata)]] but leaves this as 'a mere speculation'.
Fortunately, the verb pirrase perform a sonata does not pose any special difficulty for the present approach. The verb perform is analyzed using the relation
19
PERFORMANCE from (56), as in (69a), the noun pirrase a sonatam as a quantifier over sonata templates (which are templates for events), as in (69b), and the
resulting event predicate f o r p e r f o j m a sonatam is shown in (69c).
(69)
(a)
p e r f o r a i ^ AeAe[PERFORMANCE(e,e)]
(b)
(c)
a sonata,,,
ASAe[3e[S(e,e) ASONATA(e)]]
perforai a sonata,,, ^ Ae[3e[PERFORMANCE(e,e) ASONATA(e)]]
4.2 Krifka (1989,1992)

Strictly speaking, Krifka does not offer a treatenent of verbs of creation. His
notion of graduality (Krifka 1992, p. 42), which characterizes thematic relations
that satisfy the properties uniqueness of objects, mapping to objects, and mapping to events in his framework, does not distinguish between 'effected patients'
and 'consumed patients'. However, even disregarding this, there is a significant
difference between his approach and mine in the strength of the mapping properties appealed to, as I hinted at in section 2.2. In particular, his properties of
mapping to objects and mapping to events are stronger than the properties of
weak mapping to physical objects and weak mapping to events that I define in
(32). For example, in the case of build a house, his mapping to events would
require every part of the house to be built in the building event, yet this is implausible as a general requirement, because many parts of the house (e.g., the
doors and windows) may be prebuilt and therefore simply installed in the course
of the building event. The property of weak mapping to events in (32b) allows
for this, as I pointed out in section 2.2. His mapping to objects is also too strong,
because it would require every subevent of the building event to be an event in
which a part of the house is built. 20
Krifka (1992, p. 46) notes in passing a possible extension of his approach
for performance verbs of creation such as play in play a sonata (see also Krifka
1989, pp. 198-199). He suggests introducing a domain of types and a relation of
realization between tokens and types so that play could describe the realization
of a type. Since he does not spell out the details, it is hard to make specific
comparisons, but the role of instantiation in my analysis of performance verbs
of creation is clearly very much in this spirit. Nevertheless, he would still lack
an analogue to the relation of representation that I use for the analysis of verbs
denoting the creation of an abstract individual.
19
20
Arguably, PERFORMANCE constitutes the sole descriptive content of perform, but an additional
restriction on the events in its extension could be specified if necessary.
Krifka ( 1992, pp. 4546 ) is aware of this problem and suggests a somewhat intricate solution for
build, but I think that his mapping properties are unrealistically strong to begin with.
518
Christopher Pin
4.3 Von Stechow (2001)

Von Stechow begins with a critique of Dowty's account of verbs of creation and
proceeds by proposing three possible theories to replace it with. Unfortunately,
at least on my reading, he does not unambiguously reveal which of the three
theories he is most committed to, but since the third theory is the most similar to
a Krifka style approach, I will mention it. The idea is that the analysis of verbs
of creation makes use of a thematic relation I-Them e for the internal argument,
defined as follows (p. 310):
(70) I-Themed=
XwXeXx[BECOMma^ve){Xw'Xt[exist(wt^(x)]]
( von Stechow 2001, (99), p. 310)
In light of von Stechow's definition of B E C O M I N G (p. 290), this says that is an
I-Theme in w of e just in case does not exist in w at the beginning of e, exists
in w at the end of e, and is undefined for existence in w at any time properly
between the beginning and end of e.21
As far as von Stechow's I-theme is concerned, I do not understand the motivation for saying that is undefined for existence between the beginning and
the end of e. Since undefinedness is both a conceptual and technical hassle, it
should be strongly motivated. The relation created (see (34)) that I use does not
appeal to undefinedness, and as I argued, the parts of the object created come
into existence piecemeal in the course of the creation event but before coming
into existence they do not exist.
A final point is that von Stechow's treatment covers only verbs denoting the
creation of a physical objecthe does not mention performance verbs of creation or verbs denoting the creation of an abstract individual, hence it is at best
rather incomplete as an account of verbs of creation. 22
21
Von Stechow adds (fn. 17, p. 310) that \ is an I-Theme of e iff there is a bijection /', such that
for any part e' of e: f(e') is a part of .r & f(e') does exist [sic] at BEG(f(e')), but f{e') exists
at END(f(e)) [sic].' (There are two unfortunate typos here: the first should be corrected as 'does
not exist', and the second, as 'END(f(e'))'.) The condition on existence aside, this amounts to
Krifka's thematic properties of uniqueness of objects, uniqueness of events, mapping to objects,
and mapping to events, though von Stechow does not explicitly make this connection. However,
it is also unclear how these added requirements are related to the putative definition of I-Theme in
(70), which does not mention any such function / . If such an / is needed to characterize I-Theme
(and something like it is needed, though 1 would advocate weaker mapping properties), then it
should properly appear in the definition of I-Theme.
22
Von Stechow's paper also touches upon many other topics loosely related to verbs of creation,
something that I have not conveyed here. I have focused on the substance of the third theory that
he presents.
Verbs of creation
519
4.4 McCready (2003a,b)

McCready focuses on the interaction between progressivized verbs of creation
and anaphoric reference to partially created objects. He aims to account for
contrasts such as the following:
(71) (a)
(b)
John was painting a picture. #It was a masterpiece.

(McCready 2003b, (2a), p. 328)
John was building a house. His brother designed it.
(McCready 2003b, (4a), p. 328)
McCready's idea is that it in (71a) cannot refer back to the partially completed
picture that John was painting because the noun masterpiece may only apply to
completed objects. In contrast, it in (71b) refers to an abstract object (namely, a
house design) and not to the partially completed house that John was building.
For the analysis of verbs denoting the creation of a physical object (e.g., paint
in (71a)), McCready basically employs von Stechow's I-Theme, but he takes
its definition to be what is in von Stechow's fn. 17 (see my fn. 21), silently
correcting the typos and discreetly discarding what appears in (70): 23
(72) I-Theme(e,x) =f
3 / V e Y C t - ( f ( e ' ) -i e r i s ( E G ( e' ), / ( e' ) ) e i s ( E N D ( e ' ) , / ( e ' ) ) ) ]
(McCready 2003b, (9), p. 330)
In order to treat build in (71b), McCready (2003a, fn. 25, p. 37) takes verbs of
creation "to be ambiguous between a reading in which the verb acts as a 'verb
of realization,' which selects for a property complement, and a reading taking an
actual object, which describes an actual creation event." The reading of build in
(71b) is the one on which it takes a property complement. McCready represents
the nonprogressivized version of the first sentence in (71b) as the following event
predicate, which serves as the input to the progressive operator: 24
(73) John b u i l d - a house:
Xe[build(jolm,Xx{house(x)])]
McCready's approach is congenial to mine in its attempt to cover a wider range

of verbs of creation than the other approaches discussed above (though it does
not address performance verbs of creation). However, his conception also differs from mine in that he takes the abstract objects of verbs of creation to be
properties (extensionally, sets) and not bona fide first order individuals, albeit
23
24
McCready remarks that'[t]his definition states that is an I-Theme of e iff there is a bijection that
maps every subevent of the creation to a subpart of its object, and, until the completion of each
subevent, its corresponding object subpart does not exist.' I simply note that, strictly speaking,
the definition in (72) says nothing about whether or not .v exists between the beginning and the
end of e.
Unfortunately, since McCready does not present detailed derivations, the fonnula in (73) is the
result of a bit of extrapolation, based on what he does present.
520
Christopher Pin
abstract. 25 Yet consider (e.g.) Sarah designed a house: it would be incorrect

to say that Sarah designed the property of being a housecertainly she did not
manage to do that. To get around this, McCready could say that she designed a
subproperty of the property of being a house, which would extensionally amount
to a subset of the set of houses. But if no one ever built the house that she designed, then she would have effectively designed the empty set, which is a very
counterintuitive result. He could then try to get around this by intensionalizing
the property complement that design takes (e.g., by construing it as a function
from possible worlds to sets of individuals), but this move would lack independent support in that design otherwise behaves like an extensional verb (unlike
seek, for example). But even putting this problem aside, McCready still has to
clarify more explicitly the connection between (e.g.) build as a verb that takes
a property complement (see (73)) and build as a verb that takes an individual
argument, because from the formula in (73) it does not follow that a physical
house is built (and the addition of a simple tense operator will not guarantee this
either).
The present approach does not face these difficulties. The internal argument
of design denotes a template for physical objects that is created (in the sense of
CREATED; see (53b)). If such a template is created, it is represented, but it may
or may not be instantiated. Templates are first order individuals, albeit abstract,
and behave as individuals for the purposes of quantification, anaphoric reference,
and the like. Finally, the two senses of build that McCready is concerned with
are analyzed in (36a) and (60), being explicitly related with the help of the sort
shifter SSH-1 in (59).
4.5 Conclusion
Verbs of creation come in three sorts: verbs that denote the creation of a physical object (e.g., build), performance verbs of creation (e.g., sing), and those that
denote the creation of an abstract individual (e.g., design). I have presented a
new analysis of verbs of creation that does equal justice to each of these sorts.
The new analysis presupposes a model structure that has an existence predicate
and distinguishes between physical objects, events, times, and three kinds of
templates (templates for physical objects, templates for events, and templates
for times). Templates are abstract (first order) individuals that are existentially
anchored to physical objects with the help of a relation of representation and
which may also be connected to concrete individuals by means of a relation of
instantiation. For example, a house template (a house design) may be represented by a blueprint, whereas it may be instantiated by a physical house. This
framework offers straightforward treatments of the three sorts of verbs of creation. Furthermore, the new analysis provides a set of sort shifters that serve
to capture systematic ambiguity among verbs of creation (e.g., the distinction
25
Recall my remarks in connection with Dolling (2001) in section 3.1.
Verbs of creation
521
between build as a verb denoting the creation of a physical object and build as
a verb denoting the instantiation of a template). Finally, I have argued that the
new approach fares better than the competition as a general account of verbs of
creation.
References
Dlling, J. (2001): Ontological domains, semantic sorts, and systematic ambiguity. In:
Systematische Bedeutungsvariationen: Semantische Form und kontextuelle Interpretation (= Linguistische Arbeitsberichte 78). 71-92. Institut fr Linguistik, Universitt
Leipzig.
Dowty, D. (1979): Word meaning and Montague grammar. Dordrecht: D. Reidel.
Gamut, L. T. F. ( 1991): Logic, language, and meaning. Vol. 1: Introduction to logic. The
Kratzer, A. (1996): Severing the external argument from its verb. In: J. Rooryck and L.
Zaring (eds): Phrase structure and the lexicon. 109-137. Dordrecht: Kluwer.
Krifka, M. (1989): Nominalreferenz und Zeitkonstitution: Zur Semantik von Massentermen, Pluraltermen und Aspektklassen. Munich: Wilhelm Fink Verlag.
Krifka, M. (1992): Thematic relations as links between nominal reference and temporal
constitution. In: I. Sag and A. Szabolcsi (eds): Lexical matters. 29-53. Stanford: CSLI
Publications.
Levin, B. (1993): English verb classes and alternations: A preliminary investigation.
Levy, D. M. and K. R. Olson (1992): Types, tokens and templates. Report No. CSLI-92168. Stanford: CSLI Publications.
McCready, E. (2003a): Discourse anaphora, verbs of creation, and the progressive. Qualifying paper, Department of Linguistics. University of Texas at Austin.
McCready, E. (2003b): Anaphora and (un)finished objects. In: G, Garding and M. Tsujimura (eds): WCCFL 22 Proceedings. 328-341. Somerville, MA: Cascadilla Press,
von Stechow, A. (2001): Temporally opaque arguments in verbs of creation. In: C. Cecchetto, G. Chierchia, and M. Guasti (eds.): Semantic interfaces: Studies offered to
Andrea Bonomi on the occasion of his sixtieth birthday. 278-319. Stanford: CSLI
Publications.
Short Portraits of the Authors

Beavers, John
PhD in 2006 from Stanford University. Visiting Assistant Professor at Georgetown University. Research interests: lexical semantics, syntax, and the syntax/semantics interface; computational linguistics.
Cole, Peter
Professor, Department of Linguistics at the University of Delaware. Research
interests: syntax and semantics, typology and universals, structure of the Austronesian languages of Indonesia, Chinese, Quechua, Hebrew.
Dimitriadis, Alexis
PhD in 2000 from the University of Pennsylvania. Lecturer at the Utrecht
institute of Linguistics OTS. Research interests: Semantics and typology of
reciprocals; argument structure; anaphora; semantics of plurals; event semantics; referentiality and philosophy of reference; information technology for
linguistics; Greek linguistics; Bantu linguistics.
Eckardt, Regine
Full professor at the Department of English/Linguistics at the University of
Goettingen. Special fields: semantics/pragmatics, language change and variation. Research interests: NPI licensing, questions, discourse markers, and semantic reanalysis in language history.
Egg, Markus
Associate Professor, Center for Language and Cognition, University of Groningen. Research interests: Theoretical and computational semantics, pragmatics and discourse, syntax-semantics and semantics-pragmatics interface.
Endriss, Cornelia
Research Assistant in project A2 on "Quantification and Information Structure" of the SFB 632, Linguistics Department at the University of Potsdam.
Research Interests: formal semantics, information structure, esp. quantifier
semantics.
Gao, Meijia
MA (Linguistics) in 2003 from University of Minnesota. Research interests:
syntax, semantics.
524
Sheila Glasbey
PhD in 1994 from the University of Edinburgh. Lecturer in the School of
Computer Science at the University of Birmingham, U.K. Research interests:
Natural language semantics of tense, aspect, indefinites and generics; augmentative and alternative communication; e-drama; computational semantics and
pragmatics of metaphor interpretation and reasoning; temporal metaphor and
its contribution to the temporal structure of discourse.
Hinterwimmer, Stefan
Research Assistant in project A2 on "Quantification and Information Structure" of the SFB 632, Linguistics Department at the Humboldt University of
Berlin. Research Interests: formal semantics, syntax-semantics-interface, information structure, esp. adverbial quantification.
Kratzer, Angelika
Professor of Linguistics, University of Massachusetts at Amherst. PhD in 1979
from the University of Konstanz. Research interests: formal semantics of natural languages and the syntax-semantics interface.
Lin, Jo-wang
Professor of Linguistics, Department of Foreign Languages and Literatures at
National Chiao Tung Universtiy. Research interests: semantics, syntax and
syntax-semantics interface.
Malink, Marko
M.A. in 2004 in Leipzig. Research interests: formal aspects of natural language semantics, Ancient logic, Aristotle
McCready, Eric
Instructor in the Department of English at Aoyama Gakuin University, Tokyo,
Japan. Research interests: semantics of modals, particles, and evidentials.
Morzycki, Marcin
Assistant professor of linguistics, Michigan State University. Research interests: semantics, syntax, and their interface, especially the grammar of modification.
Nakanishi, Kimiko
PhD in 2004 from the University of Pennsylvania. Assistant professor at the
Department of Linguistics at the University of Calgary. Research interests:
formal semantics, syntax-semantics interface, prosody.
525
Nishida, Chiyo:
Associate Professor of Spanish, Department of Spanish and Portuguese at the
University of Texas at Austin. Research interests: Spanish syntax and second
language acquisition of Spanish syntax.
Pin, Christopher
Researcher, Research Institute for Linguistics, Hungarian Academy of Sciences. Research interests: semantics; more specifically, agentivity, aspect,
lexical semantics, modality, tense, and the semantics - pragmatics interface.
Rawlins, Kyle
Doctoral Candidate in the Department of Linguistics at University of California, Santa Cruz. Research interests: Formal semantics, the syntax/semantics
interface, the semantics of adverbials, concessives and conditionals, definiteness.
Rothstein, Susan
Professor of Semantics, Gonda Brain Research Center and English Department, Bar-Ilan University, Israel. Research interests: formal semantics, semantics-syntax interface, language and cognition.
Saeb0, Kjell Johan

Professor, Department of European Languages at the University of Oslo. Research interests: semantics; modality and temporality, context theory; pragmatics.
Soh, Hooi Ling

PhD (Linguistics) in 1998 from Massachusetts Institute of Technology. Associate Professor, Linguistics, University of Minnesota. Research interests: syntax, semantics.
Son, Minjeong
Dr. Ling.; Department of Linguistics at the University of Delaware. Research
interests: semantics-syntax interface, aspect, complex predicates (causatives,
resultatives, etc.), argument structure, Austronesian languages of Indonesia,
Korean.
Tanaka, Eri
Research Fellow of Japanese Society for the Promotion of Science, Osaka
University. Research interests: formal semantics, lexical semantics, Japanese
and English linguistics.
526
ter Meulen, Alice G.B.

PhD in 1980 at Stanford University. Formerly University of Washington, Seattle and Indiana University, Bloomington in the US. Now professor of Language and Cognition at the University of Groningen. Member of the Royal
Academy of Arts and Sciences (KNAW) and Governing Board of the Dutch
National Science Foundation (NWO). Research interests: natural language
semantics, computational linguistics, cognitive science and more specifically
temporal reasoning.
Williams, Alexander
Postdoctoral fellow, University of Maryland. Research interests: verb semantics, Sino-Tibetan languages.
Index
-le 3f, 6, 10-13, 15-17, 21f, 24-26, 29,

447-474
-made 206, 208, 213-216, 218, 220
-m 207-216, 216-220
accomplishment 37, 46, 52, 54, 144f,
175-179, 181, 183, 187-191, 194196, 209, 225, 230, 395, 452-454,
457,467,504,516
achievement 35, 37, 40, 46, 48, 127,
131, 144f, 147, 175-179, 181, 183,
187-193, 195f, 209, 220f, 224,
226, 229, 231, 240f, 254, 259f,
263f, 395, 450, 452, 457, 467
activity 35, 38, 144, 176-178, 181191, 193-196, 204, 206, 209, 329,
331-333,452, 470,516
adjective
- adjective 74, 90-95, 103-109, 112,
115, 117-125, 149, 151f, 155, 158,
163, 191f, 197, 200-206, 208, 215219, 243f, 249, 251,254, 264,
- extended adjectival projection 104
- gradable 112, 201, 205f, 246, 251
- non-gradable 201, 205f, 251
adverb
- adverb 12, 34, 44, 47, 52f, 57f, 76,
81-87, 90-98, 100-126, 150, 153155, 157-159, 167, 171, 183, 202,
207-209, 220, 248f, 279, 285f,
296, 307, 313, 317f, 370, 389-393,
399-407, 409-411, 434-445, 449451, 455f, 461, 479, 471f, 523f
- aspectual 411, 434, 445
- durational 52, 285f
-manner 82, 87, 102, 121,202
- subject-oriented 120-124
again 448, 452f, 457, 459f
argument
- event(uality) 40f, 57, 60, 63, 73,
87-89, 98, 149-154, 159, 161, 164,
168, 232, 237, 255, 290f, 297, 306,

309, 324, 355, 363f, 367, 371, 374377, 379f, 382, 404, 501f
- external 36f, 40, 53, 57-62, 64f,
77f, 87, 89, 101, 139, 147, 172,
275, 297, 308f, 320-322, 324, 430,
432,502,521
- internal 19, 60, 72, 230, 275, 310,
312, 314, 320-322, 493-496, 504,
511,513-515,518,520
argument selection principle 378, 382
aspect
- lexical 46f, 175, 177
aspectual composition 44, 46-48, 54,
199,218,230, 239
assertion 408, 414, 416f, 431, 441f,
452, 455, 460, 462f, 469f
asymmetric 327-331, 334, 340f, 343,
497, 505
atelic 35, 42f, 5 If, 144, 179, 182f,
187, 191, 193-195, 199, 207-211,
231, 246, 248, 250, 310, 414, 449,
460, 469f, 477f, 479
atom 55, 58, 186-189, 195, 211, 255,
276, 279, 292-294, 302, 310f, 313,
317-319, 337, 341, 350, 353, 478,
483f
Austinian proposition 373
BECOME 20f, 36, 55, 62, 90, 99,
122f, 132f, 134f, 141, 144f, 153f,
157-160, 166f, 169, 191-195, 199f,
205, 208-210, 212, 214, 216, 218,
228,330,516
benefactive 65-76
by-phrase 128, 131-139, 141-146
c-variables 409, 411
causative
- lexical 64-65
- morphological 55, 57f, 62, 65f, 76,
78
528
Index
CAUSE 11, 19-23, 36, 55-57, 59, 6165, 73f, 76f, 132, 153, 158, 160,
209f, 436, 516f
change 23, 37, 177-179, 181-183,
187-190, 192f, 195, 245f, 253-255,
257,414-416, 429, 481
change of state 3If, 35, 40f, 49, 55,
74, 127, 144, 153, 167, 206, 228231, 377-381, 449f, 461
classifier 33, 271-274, 287f, 296, 306,
319, 448
clitic 24, 223f, 226, 230f, 233, 235238, 330
coherence 435, 439-444, 495
collectivity 277, 313f, 317f, 320, 323
comitative 328, 335-338, 346f, 351
completive 449f, 452f, 457, 460
contextual
- contextually-salient 114
- domain restrictions 110, 113
- information 397, 399f, 402
conversational background 85f, 88f,
91-93, 96-98
count noun 186, 271-274
cover reading 313, 318f, 323
covert reciprocal 327f, 330, 335-337,
340, 343-345
creation 225, 257, 291, 406, 452, 493497, 499, 501-505, 507, 511, 513521
cumulati vity
- cumulativity 47, 177, 179-181, 185,
21 If, 271, 274, 276f, 279, 281283, 285-287, 289f, 295, 308
- lexical 274, 282f, 285-287, 289f,
295
- phrasal 279, 283, 287, 290
dative of interest 236f
decomposition 37, 51, 55, 57, 59, 65,
72, 76, 104, 120, 133f, 141, 145,
152f, 157, 159
degree achievement 175, 177, 190f,
193
degree word 103, 105f, 117, 120, 122
degrees 104, 112-116, 120, 124, 191,

194, 197, 248, 250, 252, 263
discontinuous reciprocal 328, 333,
335-338, 340, 346, 348f, 351, 353
discourse relation 143, 146, 445
discourse representation theory 133,
358
distributivity 286f, 289, 292, 295f,
298-304, 320, 322
domain widening 104, 112, 114, 119
double object construction 68f, 72
durative phrase 31-35, 37-46, 48f, 5If
durativity 245, 247-254, 257, 260263, 285
elaboration 143, 146, 351
entailment 86f, 90, 97, 99, 114, 184,
191, 196f, 200, 232, 32If, 349,
35 If, 377, 435,444f, 512
evaluative 103f
event
- causally related 405
- event-to-path homomorphism 221,
254, 260, 265
- event-to-scale 254, 260
- punctual 247-252, 254f, 261
- specification 350
- stage 183
- structure 37, 49, 55, 65, 72f, 76,
208, 327, 351, 482-484, 486f
event semantics 3, 61, 86, 89, 210,
269, 276, 279-282, 284, 291, 295,
348, 352, 392f
exclamative 104, 108-114
existential inference 366-369, 371,
373f, 376, 383
experiencer 349, 355, 363, 377-382,
391
factual information 435, 438
functional head 33, 36, 60, 71, 94f,
101, 123f, 133, 155,287, 348
generalized movement relation 247,
257
German 35, 37, 57, 150, 154, 157,
159f, 167, 285, 290, 296-299, 301,
Index
312, 314, 316f, 331, 333-335, 339,

342, 347, 413, 417, 423, 462, 469f,
480
gradability 82f, 99f, 201, 204f, 247f,
250-254, 257, 260, 262f
Greek 328, 331-333, 335-337, 339f,
347,469f
heap paradox 487f
homogeneity 37-40, 51, 484, 486f,
489f
homomorphism 199f, 211, 213, 245247, 254, 257f, 260, 262f, 302,
307f, 310, 313,315, 317f
Hungarian 33 If, 336, 339f, 347f
Igbo 3-11, 13-30
imperfective paradox 176, 182-184,
188, 191, 193
inchoative 62, 150, 240, 449, 457,
460
incoherence 440f
incorporation 136, 357
incremental theme 32f, 41, 44-52,
225, 229, 245f, 258, 322, 378
incrementality 210, 212, 322,
indexical 436f
individual
- abstract 505-507, 511, 515-518,
520
- temporal 598-500
individual level (i-level) 91, 355f,
384, 390-392, 399, 403, 405
infinitesimal 481, 483f, 486f, 489f,
information structure 392, 394, 400,
418,438
instantiation 57, 65, 76, 91, 184, 186190, 193, 197, 505, 508-511, 513,
517,520
instrumental adjuncts 128, 146
interpretation
- collective 314
- cumulative 269, 276, 280f, 282,
289-291, 295
- distributive 280-282, 288, 290, 292,
295f, 304
- habitual 362
529
- intersective 115, 119, 121

- iterative 31, 282, 285f
- repetitive 28 If
interval resolution strategy 396-398,
40If, 403, 406-409
Japanese 78, 199f, 202-205, 207-209,
213f, 214-218, 220, 262f, 272,
274, 301, 307, 312-314, 316, 318f,
323-325
lattice 307, 309f, 315-319
linear logic 232f
localising situation 355
Mandarin 3-11, 13-30, 447, 452f,
465,471
Mandarin v-de construction 24f
manner 50, 81-83, 87, 91f, 94f, 103,
121, 127f 131, 138, 141-145, 202,
207
mass noun 220, 272f, 290, 364
measure function 285, 302, 306-308,
310-313
measuring out 245, 322
modification
- adverbial 57, 87, 90, 103-105, 120,
124
movement relation 247, 255, 257
negation 34, 96, 109, 138, 215, 282,
413, 416f, 419-423, 431, 436,
438f, 464
negative polarity item 420, 477, 479f,
490, 522
neo-davidsonian 57, 60, 87, 291,
309f, 348, 350, 352
nonprojectionist 4f
object
- bare plural object 355f, 361-363,
365, 367, 37Of, 375, 380, 382f
ontology 115, 210, 477-482, 484,
488, 490
operator
- deontic modal 85
- distributive 303, 344
-modal 85, 100, 231,438
- partitive 48
paired cover 338
530
Index
path
- path 199, 20If, 208, 210, 210-218,
231, 239-241, 245f, 248, 253f,
258, 260f
- path of motion 245, 260
- spatiotemporal 239-240
patient 3-23, 25-28, 212, 246, 328,
346, 349f, 352, 377-381, 470, 517
phase particle 413-423, 429-434
physical object 493-497, 500-505,
507-520
plural 179f, 197, 210f, 231, 246, 269274, 276-282, 284-290, 292f, 294,
296, 301-304, 309-312, 320f, 323325, 337, 344, 355-358, 361-363,
365, 367, 370f, 380-383
plurality 180, 269f, 274, 280f, 285,
289, 291, 296, 302f, 327, 329
pluralization
269-274, 281, 288f,
291,295f, 303, 310
positive polarity item 421, 423
predicate
- abstract 128, 132- 137, 143, 145f
- adjectival 91, 355, 363f, 369, 371374, 376f, 382
- causative 58, 62, 65, 127-129, 131,
133f, 138f, 141, 143
- creation/destruction 257
- criterion 127-129, 131-134, 139,
141-143, 145
- dynamic 245, 247, 253f, 257-260,
262f
-homogeneous 478, 48 If, 485, 488
- irreducibly symmetric 327, 329f,
340, 346, 351
-motion 256,260
- quantized 47f
- resultative 22, 199-204, 208, 210,
215, 217f
- verbal 46, 48, 54, 88, 91, 127, 176,
271, 281, 283, 303f, 315, 364, 370,
376, 382, 395
presupposition 86, 142, 146, 153f,
231, 234-236, 241, 358-361, 404f,
414-417, 420, 422, 426-429, 431,
433, 435, 439-445, 447f, 460, 462465, 469-472

process-related 32
progressive 3, 144f, 175-178, 182,
190, 261, 314-317, 323, 362, 448,
452,519
projectionist 4f, 9
proto-agent 352, 378f, 381
proto-patient 352, 377-381
psychological verb (psych-verb) 355,
378-381
punctuality 251-254, 261f
quantification
-adverbial 113,400
- quantificational determiner 389
- quantificational variability effect
389
quantifier 57, 87, 90, 97, 100, 113,
152, 155f, 159f, 162f, 293, 295,
30If, 304-309, 311-323, 344, 358,
390-392, 394f, 400f, 413, 415,
417-423, 428, 430-434, 501, 512514,517
quantization 46f, 229, 246, 504
quantized reference 504
reading
- collective 302, 305, 313, 317f, 320f
- existential 355-357, 359, 361-363,
365-380, 382-384, 469
- generic 356f, 361, 366, 372, 380
- repetitive 36f, 59, 63-65, 150, 153,
157, 167
- restitutive 36, 59, 63f, 149f, 153,
157, 160
reciprocal 223, 327-336, 338-340,
342-344, 349-351
reference time 414f, 418, 420, 436,
438, 441-444, 447, 451, 456, 459462, 465f, 469-471
reflexive 223-226, 229, 231, 237,
240f, 249, 309, 331, 349, 351, 497
relational noun 352
result
- phrase 73, 204, 246f, 249f, 252f,
260
Index
- result-related 32
- result-state-denoting 56f, 76
- state 17, 20, 33, 35-37, 39-43, 48f,
5 If, 57-59, 64, 73-75, 149f, 153,
157, 167, 210, 226, 261, 447f, 459,
465-467, 472
resultative
- construction 3-5, 7f, 16-23, 199f,
202f, 209, 215-218
- weak and strong 202, 218
Russellian proposition 373
satellite-framed language 262
scalar 97, 99, 201, 204, 245-247, 250252, 257, 260, 262
scale of change 247, 257
scope
- scope 34, 36, 47, 59, 76, 81f, 90,
93f, 97f, 100, 132, 143, 15 If, 155160, 162f, 167-170, 23If, 234,
236, 282, 284-286, 289, 342-344,
350, 356, 389-392, 400, 402f, 409,
413f, 418-423, 430, 432-434, 440,
443,464
cope ambiguity 55, 59, 63-65, 157,
163, 165, 167, 170
semelfactive 175-177, 179, 182-189,
195, 248-250, 252, 259-261
Serbian 331, 333-335, 338, 342, 347
sorites 481,487-490
stage level (s-level) 306, 362, 365,
371f, 382, 391f
star operator 284
subject
- bare plural subjects 355, 363, 367,
370f
531
subjective information 435, 437, 442

sum
-s-summing 185-190, 193, 195f
- sum operation 209, 270, 276, 498
symmetric relationship 327, 329
symmetry 130, 133f, 145f, 237, 328f,
331, 334f, 338, 340, 342f, 349, 351
telic
- modifier 175-177, 183, 188-191,
193f
telicity 44, 46, 51, 129, 177, 180f,
189, 194f, 199, 208-210, 213f,
218f, 224, 230f, 234, 236f, 241,
245-248, 254-256, 260, 262f
template 496, 505-517, 520f
temporal reasoning 436, 444
tense
- marking 398f, 403, 409
- past 98, 100, 248, 370, 372, 398f,
402, 405f, 408, 439, 442, 444f
- perfect 439, 444f
thematic role 24, 57, 257, 275f, 328f,
349, 348f, 35 If, 501
transition 49, 209, 214, 215-218, 224231, 239, 241, 248, 270, 436-438,
447, 460f
twigs puzzle 489
typeshift 47, 94f, 97, 100, 116, 360
unification 92, 128, 133, 136-138,
140-143, 145f, 349
uniform projection 9, 21, 24, 26
valence 4, 8 , 2 1 , 2 3 , 30,71
Vendler classes 175
verb-framed language 262
zero-convergent sequence 482f, 486

Dölling Etc., Language, Context, and Cognition

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dölling Etc., Language, Context, and Cognition

Uploaded by

Copyright:

Available Formats

Event Structures in Linguistic Form and Interpretation

Language, Context, and Cognition

Printed on acid-free paper which falls within the guidelines

Library of Congress Cataloging-in-Publication

Event structures in linguistic form and interpretation / edited by

The Deutsche Nationalbibliothek lists this publication in the Deutsche

Copyright 2008 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin

Section I: Event Structure and Syntactic Construction

Event decomposition and the syntax and semantics

Syntactic decomposition of events in Korean and Standard Indonesian

Minjeong Son and Peter Cole

Section II: Event Structure and Modification

Adverbial modification of adjectives: Evaluatives and a little beyond

The structure of criterion predicates

Reference to embedded eventualities

Section III: Event Structure and Situation Aspect

Reflexive intransitives in Spanish and event semantics

Scalar complexity and the structure of events

Section IV: Event Structure and Plurality

Event quantification and distributivity

The event structure of irreducibly symmetric reciprocals

Existential readings for bare plurals in object position

Section V: Event Structure and Temporal Location

Phase structure and quantification

Cohesion in temporal context: The role of aspectual adverbs

Mandarin sentential -le, perfect and English already

Hooi Ling Soh and Meijia Gao

Section VI: Event Structure and Natural Language Ontology

Portraits of the Authors

Section I: Event Structure and Syntactic Construction

syntax. According to them, such an approach provides satisfactory explanation

Section II: Event Structure and Modification

Section III: Event Structure and Situation Aspect

Section IV: Event Structure and Plurality

structure must play a major role in accounts of plurality, including distributive,

predicate like meet that it expresses a relationship between two arguments

Section V: Event Structure and Temporal Location

Section VI: Event Structure and Natural Language Ontology

Regine Eckardt addresses an apparent conflict between two applications of

Alexander Williams (Maryland)

Patients in Igbo and Mandarin

The grammar of these constructions demonstrates that, characteristically in Igbo

2 Projectionist and nonprojectionist models

\A1 pounded the cutlet\ = 3e.pound{e) PAT(e) = c AG(e) = a

Patients in Igbo and Mandarin

\pound\ = \e.pound {e)

[ [VP V D P ] ] = Ae. [ V ] (e) P A T ( e ) =

Ve [pound (e) s 3j/.PAT(e) = y]

Thus the choice between projectionist and nonprojectionist models is primarily

Al pounded the cutlet flat.

Al pounded the cutlet flat,

Patients in Igbo and Mandarin

to emerge which are otherwise obscured by independent differences in syntax

4 The relevance of resultative constructions

(linear order irrelevant)

5 English resultative constructions and verbal valence

7Al hammered his wrist sore.

Al cut *(the frozen meat).

*Al cut the knife dull.

*The bears frightened the campground

Patients in Igbo and Mandarin

Uniform projection is explained if argument requirements are stated as lexical

of argument relations, most emphatically in Levin and Rappaport Hovav 1995