Professional Documents
Culture Documents
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
Data
ISBN 978-3-11-019066-3
Bibliographic information published by the Deutsche
Nationalbibliothek
Contents
Introduction
IX
31
55
81
103
127
149
Markus Egg
175
199
VI
Contents
223
245
John Beavers
269
301
327
355
Sheila Glasbey
389
413
435
447
477
Verbs of creation
Christopher Pin
493
Contents
VII
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.
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
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
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.
Introduction
XV
XVI
Introduction
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.
Introduction
XIX
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
(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)
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.
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)
(b)
[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)
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)
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)
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.
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.
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)
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)
(19) (a)
(b)
(20) (a)
(b)
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.
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
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
*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)
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
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)
(b)
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
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)
(48) (a)
(b)
-FACT ( B V C )
(BVC)
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.'
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
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)
ya.
y a m t h a t d i g . o u t -FACT 3s
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
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 )
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.
Mandarin 'cut':
Igbo 'cut':
\qie\
[6]
=
=
Xe.cut(e)
e. cut. (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) =
\x\e.cut{e)
D P qi]]
[DP]
] = [ [, I'AG [ V ' D P bi ]] ] =
=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)
(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)
15
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)
21
(b)
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).
(b)
(c)
Mandarin'cut': \_qie\ =
Xe.cut.(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)
(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 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) =
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
[MR] =
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
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
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
*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
le.
PRT
'Flattering Lao Wei, I praised [him] such that even his wife got embarrassed.'
*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
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)
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.
27
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
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
Ta yijing jiehun
san
nian le
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
Wo
yijing
kai
ershi
already
drive
twenty year
nian jichengche
taxi
le
PAR
(a)
Women yijing
dida
zhongdian
shi
fengzhong
le
we
reach
destination
ten
minute
PAR
already
*Women
we
yijing
dida
shi
fengzhong
zhongdian
le
already
reach
ten
minute
destination
PAR
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)
(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.
[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
36
Jo-wang Lin
(b)
AH Baba
Subj
(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
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
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
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
[AgrsP
women k
[Volccp
tk
[voice 1
ten minute
destination
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.
'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)
[[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', , )]
With the above definitions in mind, I propose that a resultative verb such as dida
'reach' in (4a) denotes something like (16):
(16)
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.
43
Event decomposition
Yuehan
ba
men
guan-le
John
BA
door
close-Asp three-CL
san-ge
xiaoshi
le
hour
PAR
(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
10
*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.
(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
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
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:
(Fililp 2001,474)
(Fililp 2001, 475)
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)
(b)
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.
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
(a)
[ x CAUSE
(b)
[ y BE BROKE ] CHANGE]]
(c)
Carter (1976)
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)
Causing Event
NP
V'
John
Caused Event(uality)
V'
cup
AP
break
Unaccusative
Causatives
(a)
kwis
'to fall'
Vkwis-ts
(b)
t'iq
Vt'iq-s
(c)
us
Vus-W
Unaccusative
Causatives
(a)
t-um-umba
m-pag-tumba
(b)
s-um-abog
'x explode'
m-/>ag-sabog
(c)
1-um-uwas
m-/>ag-luwas
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-
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.
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
Chelswu-ka
sakwacup-ul
nok-i-ess-ta.
Chelswu-NOM
apple juice-ACC
melt-CAUSE-PST-DEC
(a)
aiskrim-i
el-ess-ta.
ice cream-NOM
freeze-CAUSE-PST-DEC
Yenghi-ka
aiskrim-lul
el-//-ess-ta.
Yeonghi-NOM
ice cream-ACC
freeze-CAUSE-PST-DEC
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)
(b)
Yenghi-ka
aiskrim-lul
tasi
el-//-ess-ta.
Yenghi-NOM
ice cream-ACC
again
freeze-CAUSE-PST-DEC
(i)
59
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)
Restitutive reading
(b)
Repetitive reading
60
<e <s,t>
<s,t>
J
Ag
h
<e, <s,t>>
^ by Event
/ <e<s,t
XxXc.Ag(e.x) wash
clothes
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)
: . (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
VP (melticc)
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
apple juice
CAUSE-Voice
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
repetative^^
Chelswu
'
(b)
VP
tasi
restitutive^-^'
apple juice
_
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)
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
(20)
Chelswu-ka
mwun-ul
tasi
yel-ess-ta.
Chelswu-NOM
door-ACC
again
open-PST-DEC
(a)
(b)
[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)
[Restitutive]
(b)
[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.
(a)
Cangkir-nya
cup-3
pecah.
break
cangkir-nya.
cup-3
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.
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 di- in 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
roti
itu.
Tika
bread
the
meN-bake-KAN
Eric
(a)
Eric
membuat
rumah-rumahan
itu (untuk
anak-nya).
Eric
meN-make
RED-house-AN
the
child-3
for
Eric
membuat-*(kan)
anak-nya
rumah-rumahan
itu.
Eric
meN-make-KAN
child-3
RED-house-AN
the
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)
*Roti
itu
bread
the DI-bake-KAN
dipanggang-kan
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
Eric membuat-kan
rumah-rumahan
itu *(untuk
Eric meN-make-KAN
RED-house-AN
the
for
anak-nya).
child-3
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
untuk
Eric.
for
Eric
Rumah-rumahan
itu
dibuat-kan
RED-house-an
the
DI-make-KAN
untuk
for
anak-nya.
child-3
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
Rumah-rumahan itu
RED-house-an
the
dibuat-kan.
DI-make-KAN
someone.'
68
(a)
Roti
itu
bread the
dipanggang.
DI-bake
Rumah-rumahan
itu
dibuat.
RED-house-an
the
DI-make
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
memanggang-kan
roti
itu
untuk Eric.
Tika
meN-bake-KAN
bread
the
for
Eric
Tika
memanggang
roti
itu
untuk
Eric.
Tika
meN-bake
bread
the
for
Eric
(a)
(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)
69
(b)
(Ambiguous)
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)
(a)
*Saya
men-(p)anggang-Ara/i
perayaan
1SG
meN-bake-KAN
celebration birthday
ulangtahun
Eric
biskuit
itu
Eric
biscuit
the
*Dia
mem-buat-o
perayaan
Halloween
rumah-rumahan
itu
3SG
meN-build-KAN
celebration
Halloween
RED-house-AN
the
(a)
meN-bake-KAN
ulangtahun Eric
celebration birthday
Eric
Dia
mem-buat-
3SG
meN-build-KAN
rumah-rumahan
itu
RED-house-AN the
untuk perayaan
Halloween
for
Halloween
celebration
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
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
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
72
(40)
Benefactive Constructions
(a) NP+NP structure
TrP
/ \
Spec
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
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)]
(a)
Cangkirnya
pecah
cup-3
break
(b)
Janet memecah-kan
cangkirnya
(a)
(b)
Paraphrase: the event such that Janet is the agent of doing something brings about a
VOICEP
[AP
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
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
VoiceP
. 3e'[Result(e', e) & in possession (e1, toy house, his child) & making(e) &
Ext
Eric
'
-Voice
r-
by Event
Identification
. Ext (e,x)
.
ru
rumah-rumahan
'the toy h o u s e '
'the
-kan
itu
Th(e,y)
DP
anak-nya 'his child'
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
References
Arka, I.W. (1992): The -kan
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
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)
(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
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.
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.
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
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
[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
\
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)
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
[ 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)
(
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
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
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)
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
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
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
/
Xx De
is red in w
is a house 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)
. 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)
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
XR
. \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.
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
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
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
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.
(a)
(b)
Similarly, evaluative readings are not possible to the right of an adjective, though
domain-adverb readings1 are:
(2)
(a)
(b)
(c)
(d)
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)
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.
105
(a)
(b)
(c)
(d)
(a)
(b)
(a)
(b)
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)
(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)
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)
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
109
This contrast holds under embedding as well. Embedded clauses that are relatively clearly indirect questions do not admit vety:
(13) (a)
(b)
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)
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)
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)
More generally, then, (19a) will denote the set of true propositions of the fonn
'he eats x' for some (surprising) value of x:
(20)
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)
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)
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
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)
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)
(not final)
114
Marcin Morzycki
[ Clyde is remarkably
talle ]
(not final)
[ Clyde is remarkably
talle ]
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.
This presupposes that the standard will always be in the domain of quantification - an assumption
natural at least, and perhaps unavoidable.
116
Marcin Morzycki
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)
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.
])
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 =
] )( [ 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))
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.
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.
'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
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)
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)
122
Marcin Morzycki
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
(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)
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)
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
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
126
Marcin Morzycki
(a)
(b)
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)
128
(4)
(5)
(c)
129
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)
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
(15) (a)
(b)
(c)
(b)
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)
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)
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
> 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)
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)
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)
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)
(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) )
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
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
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
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.
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
(27) (a)
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.
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
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
That is, there is no theoretical problem; in practice, however, such a representation will easily
become very complex.
142
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 ^
} 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)
(e, ) I ,
spew...
(e)
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
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)
144
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)
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
( , indefinite } J
P(e)
ay7/?e(Agent(e))(ei)
C ause (escape ( Agent ( e ))( e ))( ( e ) )
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
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.
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)
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)
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
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.
...
'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)
(a)
Xy.GEN[e,x](x in e Ay = x, dance'(x)(e) )
(b)
(c)
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
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(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
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))
/
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.
3.2 Change-of-state verbs
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
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'
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)
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]]:
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.
5.3 Change-of-state verbs
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
^.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]
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'^ECOMEi^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 )
[[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.
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)
(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
SECTION III
Event Structure and Situation Aspect
(a)
(b)
(c)
Mary is running.
(d)
Second, achievements and accomplishments occur naturally with telic modifiers, while states and activities do not, as illustrated in (2).
(2)
(a)
(b)
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)
(d)
(a)
(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]
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
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.
"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)
(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."
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)
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)
(a)
(b)
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)
(b)
(a)
(b)
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)
(b)
(c)
(d)
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)
(b)
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)
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)
(b)
X ( s (eue'))]]
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:
187
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)
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)
(b)
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)
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)
(b)
(25)
(26)
II ADJ II = X x . V s ( x ) = d
(a)
I COOL ||
(b)
HOT||=
= XX.VS.COOL (X)
< d
..(x) > d
= 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)
(b)
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)
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
When I dropped the ice in, the liquid cooled instantly (although not very much).
(a)
(b)
(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
(a)
(b)
(a)
(b)
(a)
(b)
(a)
(b)
Degree modifiers provide an explicit extent to the activity event, also yielding
a telic VP:
(37)
(38)
(a)
(b)
(a)
(b)
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)
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
(a)
I read Dafna The 101 Dalmatians for an hour before she went to sleep,
(b)
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.
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.
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)
(b)
(a)
The dog barked itself hoarse. > *The dog barked itself
(b)
[CONTROL]
[ECM]
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.
201
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.
(a)
(b)
(c)
Closed scale adjectives form control resultatives with durative verbs; and
(d)
verbs (i.e. verbs without internal structure) cannot form a control resultative;
(a)
Gradable adjectives:
very {long/flat/expensive/straight/full/dull}
(b)
Non-gradable adjectives:
*very {dead/triangular/invited/sold}
(a)
(b)
(Wechsler 2001)
202
Eri Tanaka
(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
(a)
(b)
[WEAK]
[WEAK]
(=(lb))
(a)
(b)
[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
mizu'umi-ga
kachikachi-ni
kootta
lake-NOM
solidADj-mf
froze
(a)
*John-ga
kinzoku-o
J-NOM
metal-ACC
pechanko-ni
flatADj-,/
tataita
pounded
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
(a)
*John-ga
kabin-o
konagona-da/de
kowasita
J-NOM
vase-ACC
pieces
broke
*mizu'umi-ga
kachikachi-da/de
kootta
lake-NOM
solid
froze
(a)
*inu-ga
jibun-o
karakara-ni
hoeta
dog-NOM
self-ACC
hoarse^-,,,/
barked
*hitobito-ga
hodoo-o
pechanko-ni
hasitta
people-NOM
pavement-ACC
flatADj-;/
ran
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
204
Eri Tanaka
STRONG
CONTROL
(8a), (8b)
(9a), (9b)
ECM
None
(2a), (2b)
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
nose-NOM
nagak-u
nobiru
long-KU
lengthen
(ii) na-adjectives
(a) jimen-ga
ground-NOM
'The ground is
taira-da
flat-DA
flat.'
ground
(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)
(17)
kanzen-ni
{kata-i/usui}
completely
{solid/thin}
kanzen-ni
{taira-da/nameraka-da/kirei-da}
completely
{flat/smooth/clean}
*kanzen-ni
{naga-i/mizika-i/taka-i/atu-i}
completely
{long/short/expensive/thick}
*kanzen-ni
sinsen-da
completely
fresh
kinzoku-o
taira-ni
nobasita
metal-ACC flat-ADW.
flattened
kata-ku kootta
solid- ADJ;,,/
frozen
pankizi-o
naga-ku
dough-ACC long-ADJI/
nobasita
lengthened
kinzoku-o
sikaku-ku
metal-ACC square-ADJIN/
nobasita
flatten
kinzoku-o
pechanko-ni
nobasita
flatten
206
Eri Tanaka
(20)
closed scale:
(a) *John-ga
kinzoku-o
J-NOM
usu-ku
metal-ACC thin-ADj;/
tataita
pounded
kinzoku-o
J-NOM
metal
taira-ni
flat-ADj;/
tataita
pounded
kinzoku-o
mizika-ku
metal-ACC short-ADj;/
tataita
pounded
kinzoku-o
pechanko-ni
metal-ACC
flat-ADj;/
tataita
pounded
hetoheto-ni
hasitta
tired-ADjui/
ran
(a)
(b)
(a)
John-ga
eki-ni
tuita/itta
J-NOM
station-LOC
arrived/went
*John-ga
eki-ni
aruita/hasitta
J-NOM
station-LOC
walked/ran
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
tuita
J-NOM
{5 minutes-for/5 minutes-in}
arrived
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
(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
the other side of the river-up to
5 minutes-in
'John swam to the other side of the river in 5 minutes.'
oyoida
swam
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
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)
(b)
[State]
[Activity]
(c)
[Achievement]
(d)
[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)
(d)
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)
(b)
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
210
(30)
Eri Tanaka
(a)
(b)
[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}.
=> 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
(a)
(b)
(35)
(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)
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.
(a)
John-ga
eki-ni
tuita/itta
J-NOM
station-LOC
arrived/went
*John-ga
eki-ni
aruita/hasitta
J-NOM
station-LOC
walked/ran
(a)
John-ga
eki-made
aruita/hasitta
J-NOM
station-up to
walked/ran
John-ga
mukoogisi-made
oyoida
J-NOM
swam
(a)
kodomo-ga genkan-ni
iru
is
213
plural]
kodomo-ga genkan-made
iru
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
be-POLITE-Q
nandan
nandan
ari-masu-ka
be-POLITE-Q
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)
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-ga
kono kawa-o
oyoida
J-NOM
this
swam
river-ACC
The accusative case marked path cannot appear with verbs like arrive:
(45)
*John-ga Midoosuji-o
tuita
J-NOM
arrived
Midoosuji avenue-ACC
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-ga
kono kawa-o
mukoogisi-made
oyoida
J-NOM
this
swam
river-ACC
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-ga
Midoosuji-o
itta/kita
J-NOM
Midoosuji avenue-ACC
went/came
John-ga
Midoosuji-o
(*eki-ni/eki-made)
itta/kita
J-NOM
Midoosuji avenue-ACC
{station-LOC/station-up to}
went/came
(a)
(b)
(c)
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
hanbun
aruita/hasitta
J-NOM
halfway
walked/ran
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
tukue-ni
desk-LOC
oita 14
put
-ni
-made
hanbun 'halfway'
Eat
walk/run/swim
go/come
arrive/put
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
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
Theo
is
niet
blij.
Theo
is
not
happy
(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
(a)
Die
opmerking
that
remark have
heeft
me erg
verdrietig/blij
gemaakt.
me very
sad/happy
made
Dat
medicijn
heeft
me erg
ziek
gemaakt.
that
drug
has
me very
sick
made
(a)
(b)
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-ga
kinzoku-o
J-NOM
metal-ACC very
totemo
{taira-ni/usuku}
nobasita
{flat-ADJin//thin-ADJm/}
flatten
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
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
*mizu'umi-ga
hanbun
kataku
kootta
lake-NOM
halfway
solid
froze
(a)
John-ga
kinzoku-o
J-NOM
metal-ACC halfway
hanbun
nobasita
flatten
mizu'umi-ga
hanbun
kootta
lake-NOM
halfway
froze
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
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
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.'
S.
(2001):
'An
Analysis
of
English
Resultatives
Under
the
Reflexive/Reciprocal se:
(a)
(2)
(3)
(4)
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)
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'
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
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.
(9)
(10)
()
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)
226
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.
227
228
(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.'
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)
[ e ! . . . eg ]
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)
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
*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.'
(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)
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 :
\= 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
= 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)
D
I
Dos
(c)
PRED
I
hombres
Cl
DIE(SUBJ)
SPEC
PRED
TWO
SUBJ
PRED ' M E N '
ADJ
(d)
[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
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)
10
(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.'
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
= Ld[d G A W'[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)
3p[L(e) = B{p)\\
3 p [ L ( e ) = E(p)\]
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)
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
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
242
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.
(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)
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
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.
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)
(duration/a/te/)
{after)
(a)
(b)
(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)
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)
(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)
{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.
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
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)
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
Gradable
dur/grad
dur/grad
X
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.
(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)
{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)
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
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,
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:
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.
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
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)
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
(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)
(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)
( 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
e'
//
e'"
^
///
(b)
e'
e"
e'"
///
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.
{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)
(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)
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
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.
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
SECTION IV
Event Structure and Plurality
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)
(b)
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
4
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)
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)
(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)
(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
274
Angelika Kratzer
275
10
276
Angelika Kratzer
(a)
(b)
Be3x3y [*child(x) & /x/ = 20 & *agent(x)(e) & *pizza(y) & /y/ = 10 &
*eat(y)(e)]
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
Extension of lift
{<ei, Red>, <e2, Red>, <e3, Red>, <e4, Green>,
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>,
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'
>
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)
(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
(14)
(a)
(b)
279
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
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
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)
(b)
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.
(19)
(a)
(b)
* Xxke3y
281
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:
282
Angelika Kratzer
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!
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
(23)
(c)
(d)
(e)
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)
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'
(a)
(b)
have a
285
(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)
(b)
fmmucc(e) = 5
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 ^operators
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?
287
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
[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
289
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)
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.
(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
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)
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
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)
(b)
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
>)
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.
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
(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)
(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)
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
(a)
[Otokonoko
san-nin]-ga
kinoo
isu-o
tukut-ta.
[boy
three-CL]-NOM
yesterday
chair-ACC
make-PAST
[Drei
Jungen]
haben
einen
Stuhl
gebaut.
[three
boys]
have
chair
built
^/collective, ^ d i s t r i b u t i v e
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.
(a)
(b)
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)
ATOM(y)
lift.the.piano(y) ]
(Link 1983:309)
304
Kimiko Nakanishi
FACT:
(Landman 2000:148)
(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)
haben
hirot-ta.
find-PAST
in street-ditch
Vdistributive, Vcollective
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)
Vdistributive, Vcollective
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,
married
Bekannte
haben gestern
acquaintances have yesterday
geheiratet,
zwei
two
married
Vdistributive, Collective
(a)
(b)
[Gakusei
san-nin]-ga
kinoo
Peter-o
korosi-ta.
[student
three-CL]-NOM
yesterday
Peter-ACC
kill-PAST
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
?? 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
(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
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
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 homomorphism of the semilattice S, = <S, "> into the semilattice S 2 = <S2, > is a mapping F: S,
308
(17)
(18)
Kimiko Nakanishi
cz>
'
measured amount
C3
E
measured amount
(20)
(a)
V P V x , y e D s [ [P(x) P(y)]
(b)
(c)
(21)
P(xusy) ]
(a)
(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)
(23)
(a)
XUiyUiz
XL
(b)
J\Z
P(xusy) ]
eiUEe2UEe3
eiU
JEe3
(a)
Davidsonian:
Xx e Ay e Aev. see(x,y,e)
(b)
Neo-Davidsonian:
(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
(26)
e 1 u E e 2 u E e3
eiU E 2
ei
eiU E 3
e2uEe'3
e2
e3
xujyuiz
..S. -Ul.!.'.-.
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
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
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
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.
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
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
(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
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
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
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-wa
[gootoo-ga
sokode
san-nin Mary-o
korosi-tei-ta]-to
itta.
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)
kinoo
san-nin
issyoni
isu-o
tukut-ta.
boy-NOM
yesterday
three-CL
together
chair-ACC
make-PAST
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
11
318
(43)
Kimiko Nakanishi
eiU E e 2 u E e3
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)
319
(a)
(45)
(b)
(a)
(b)
(Gillon 1987)
kaji-o
kesi-ta.
[firefighter 400-CL]-NOM
fire-ACC
put OUt-PAST
Colorado-in
(Landman 2000:125)
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
320
Kimiko Nakanishi
(a)
(b)
san-satu
yon-da.
John-ga
hon-o
kinoo
three-CL
read-PAST
Bcher
hat
Hans
gestern
drei
gelesen,
books
has
Hans
yesterday
three
read
(49)
(a)
John-ga
kakuteru-o
syeikaa-de
san-bai
maze-ta.
John-NOM
cocktail-ACC
shaker-by
three-CL
mix-PAST
Getrnke
drinks
hat
has
Hans
Hans
drei
three
gemixt,
mixed
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
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)
(b)
In this class I will try to combine every semantic theory that has been proposed in the
literature.
(Landman 2000:83)
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)
(b)
(c)
(d)
(e)
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
325
(a)
(b)
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)
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)
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)
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)
"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)
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.
(a)
(b)
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.'
(b)
(12) (a)
(b)
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'
(b)
(15) (a)
(b)
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)
334
Alexis Dimitriadis
(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)
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)
(Greek)
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)
(Greek)
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)
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)
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)
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)
(Greek)
(Hungarian)
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)
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)
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)
*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
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
Bill and Mary (each) thought that they had kissed each other.
Bill and Mary (each) thought that they had kissed.
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
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
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
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.'
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)
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)
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).
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].
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)
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)))
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
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/^english/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.
356
Sheila Glasbey
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
(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)
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.
But of course, 'John' can similarly be in topic position within a similar sentence with 'hate':
(7)
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(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
John(x)
hate-lawyers(x)
1 3 Y(lawyers ( Y) _& ko w(x, _Y))
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)
361
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.
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
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.
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
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'
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.
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)
366
Sheila Glasbey
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
_ J
know(j, 1)
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
Adjectival
(-e)
Psych-ES verbs
(-e)
No existential bare
plural objects
"the rest"
(+e)
Existential bare
plural objects
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)
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
e'
chase(dog, cat)
Existential inference
chase(d, c)
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)
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
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)
374
Sheila Glasbey
we have:
s
hungry(child)
Existential inference
hungry(c)
375
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)
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)
hate(j, lawyer)
Existential inference
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)
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?
One verb 'rue' from this class was prefaced by a question mark in (Levin 1993) and we have
omitted it.
378
Sheila Glasbey
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)
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
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
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
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
(a)
(b)
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
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)
(b)
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)
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:
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)
(b)
This is exactly the same interpretation that is assigned to (2b) (repeated below
as (6a)):
(6)
(a)
(b)
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)
(b)
MOST, [car(x) bought in the 80s(x)] [GEN, [in (x, s)] [blue(x, s)]]
392
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)
(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:
(9)
(a)
(b)
393
[blue(e)]
This means that without further assumptions, the existing situation/event semantics accounts of QVEs cannot explain the observed contrasts either.
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)
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)
(b)
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)
(a)
(b)
in 80s(e')
e' @ ie]
C"(X)]
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)
1.
2.
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.
(20)
(a)
(b)
397
in 80s(e')
e' @ i]
C"(X)]
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)
in BOs(e')
e' @ 80s!
C"(X)]
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)
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)
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)
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)
The past tense demands zfej, i. e. the time of being blue, to end before the
speech time t0\
(26)
(a)
(b)
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)
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)
(b)
To summarize the results of this section, we claim that (2b) is acceptable for
the following reasons:
10
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
(a)
,?
(b)
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)
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)
(b)
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)
(b)
11
(a)
(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)
(b)
MOSTe [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', ) past (e')
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)
404
(39)
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)).
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)
(b)
MOSTe [3x. Arg(e, ) car(x) [Be', buy (e') Theme(e', ) past (e')
405
(43)
A house that was built in the 19 lh century usually has a gabled roof.
(44)
(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
(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)
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
(a)
A house that was built in the 19 lh century usually has a gabled roof,
(b)
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)
(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
Massachusetts, Amherst.
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
Philosophy 14, 39-100.
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
(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
(some)
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
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
(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))
(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)
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)
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
417
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.
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()) -(.())
418
Marko Malink
(21)
A L R E A D Y ( . 3 S ( S , ) ) = tr
zR(z,tr)
3 f V f ( f , <t
3zR(z,t))
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.
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.
Q Q
(40)
alie
einige
nicht alie
Mz
3z
'
V
niemand
3z
/
(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
This m e a n s :
\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))
0 ALREADY V; 0
alle schon
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
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^(,.)=>)
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
14
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
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))
(a)
rmi(z)
3)
(b)
[Tj]()
=3
* P
=df
Q = p V
3z(rel(z,R)A\Q](z)A^=U\)
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
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
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.
436
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.
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
&
r0]
U N T I L ( r i , ( P ' ( s, x,
[ x
&
/>]]] =>
(d)
r]
>
r 0 & [ 2
&
+ ) ) ) a r]
[ r o , , s , I P ' ( s , ,
-) & s a
S I N C E ( r i , (P'(s, x,
-)))ar0]
&
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.
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)
[IP
439
UNTIUr,, P'(s, X, -)) 2 r0 & ALT(sp, r0, [ r, < r0 & SINCE(ri, P'(s, x, +)) 3 r,,])]
(b)
(c)
(d)
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.
(a)
(b)
440
(c)
(d)
(a)
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
(d)
(c)
the
back-
(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.
(a)
(b)
(a)
(b)
443
E:
D:
(11)
E:
D:
(12)
E:
D:
444
(13)
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)
(b)
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.
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
(a)
(b)
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
'We reached the top of the mountain.'
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)
(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)
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)
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
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)
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)
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)
xie
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)
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
(a)
(b)
(14)
(a)
(b)
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)
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)
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)
(a)
(b)
jiu he
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)
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)
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
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
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)
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)
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)
(b)
(37)
(a)
(b)
459
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
(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.'
(a)
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)
(a)
(b)
le.
463
(a)
(b)
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)
(b)
(c)
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)
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
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)
15
We thank Manfred Krifka (personal communication) for pointing out this prediction of our
analysis. See also Lbner (1989).
465
(a)
(b)
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)
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
(a)
(b)
16
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)
(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)
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)
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)
(b)
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)
(b)
(c)
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)
(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.
Comrie (1985:78) treats present perfect differently from past and future perfect.
We thank Jeanette Gundel for discussions.
470
Nurse :
that now.
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)
(b)
(a)
(b)
(a)
(b)
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)
(a)
(b)
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
Philosophy 12, 167-212.
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.
Linguistics and Philosophy 26, 459-510.
473
SECTION VI
Event Structure and Natural Language Ontology
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 ') )
(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)
(6)
479
want to model the grammatical effects of folk notions about the world and
time.
(9)
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 -^ejcz e )
3e Vi( e c e ( )
483
Let
of some equivalence
class in
()/~}
We now need to conjoin these formulae with the elementary theory of E: Let
therefore
Th(E) := { \ ^atomic 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*) )
484
Regine Eckardt
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:
Homogeneous predicates apply to events that consist of P-parts all the way down:
HOM(P)
Ve Ve ' ( -i 3e*(
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)
(6)
Differences: If e' is part of e, then there are non-overlapping e", e'" that add up e' to e:
(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).
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)
(16)
Any e in P{e) is linked to some event such that there is a finite sequence
e=e, e2, e3, ..., e^e
and such that
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) ^
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.
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
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)
(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)
(f)
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
(a)
(b)
(c)
(d)
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)
(b)
(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)
(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
: , y, , . . . (physical objects)
E: e, e', e", . . .
(events)
: t, t', t", ...
(times)
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)]
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)
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)
> n ( s o j ' | ]
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)
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 availability of enables the following two axioms to be stated more succinctly:
(22) (a)
(b)
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
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
(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)
g\a^b] = ^ if there is
500
Christopher Pin
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]
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) ) > ^exist(i,x)]
(x is created in e)
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)
(a)
build
(b)
(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)
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)
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)])
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
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
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", . . .
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,)]]
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
(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 ]
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)
(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
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.
Axiom.
( 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.
( 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,~<,
, 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)
(b)
C c h
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)
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)
(a)
d e s i g n s A x A e [ d e s i g n ( e ) CREATED(e,x)]
(b)
(c)
(d)
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)]]
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)
(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)
(b)
Cummings)]]]
a g e n t ( e , d a n i e l )]
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)
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)
designed
A S A e [ 3 x [ S ( e , x ) VICTORIAN-STYLE-HOUSE(x) A 3e'[design(e')
CREATED^',)
(b)
[(Rebecca)
Aagent(e',sarah)]]]
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 )
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
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)
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
The definition of SSH-4 is given in (67a) and its application to design (see (53a))
is shown in (67b).
(67) (a)
(b)
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]]]
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)]]
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
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
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)])]
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
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
University of Chicago Press.
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.
University of Chicago Press.
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.
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.
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
Williams, Alexander
Postdoctoral fellow, University of Maryland. Research interests: verb semantics, Sino-Tibetan languages.
Index
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
Index
529
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,
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