Aktionsart, L-syntax and Selection Gillian C.

Ramchand
University of Oxford

1.

The Lexical Determination of Aspect

Vendlers 1967 article presenting the Aristotelian classication of event types and relating it to classes of predicate in natural language is the source of much stimulating work on the aspectual or event structure classication of verbs (Dowty 1979, Taylor 1977, Kenny 1963). While it is now understood that the original division into states, activities, achievements and accomplishments cannot correspond directly to what is specied in the lexicon, many theories attempt to use lower level aspectual features that are derived from these larger natural classes. In particular, notions such as telicity/boundedness, dynamicity or durativity have played an important role in subsequent theories of event structure decomposition and lexical classication. In general, many researchers have attempted to classify verbs by means of their inherent aspectual properties (Grimshaw 1991 Hoekstra 1984 Tenny 1987 Levin and Rappaport 1995) as a way of capturing important linguistic generalisations. Any analysis of the lexical contribution to aspectual classication is complicated by the well-known facts that the addition of PPs to a Verb Phrase can signicantly modify the aspectual character of that VP. Specically, goal phrases create accomplishments from activities (in the Vendler 1967 sense). Perhaps even more difcult to deal with is the notorious fact that the nature of the object (in terms of quantized-ness or some equivalent semantic notion) of a particular verb can have a direct effect on the telicity of the result, as the standard tests show (Verkuyl 1972, Dowty 1979, Tenny 1987, Krifka 1989). The question which naturally emerges from an appreciation of these semantic facts, is the extent to which it is possible to preserve a lexical distinction among verb types while still maintaining a compositional aspectual semantics within the VP. Clearly, the feature contributed by the lexical verb simpliciter is of a slightly different order from the Vendlerian class labels or even from the lower level notions of telicity and durativity. The present paper continues the quest for a set of lexical determinants for aspectual composition. The theory of lexical features that I will present here is in many ways a grateful exploitation of the work and insights that have come before. It owes a lot to, and has much in common with that work. However, it also constitutes a very particular set of axiomatic assumptions and very particular kind of implementation of the way in which syntax and semantics are thought to interact.
I would like to thank David Adger, Miriam Butt, Raffaella Folli and Peter Svenonius for many comments and discussions on the issues found in this paper.

1.1

Classifying Argument Roles

To deal with the problem of aspectual determination, linguists have attempted an explicit aspectual classication of thematic roles and relations. Most notable in this class are the proposals introducing lower level features such as +ADD-TO, +SQA (Verkuyl 1989, Verkuyl 1993) or QUA and Mapping-to-Objects (Krifka 1989) which can combine with the processual nature of the lexical item to give telicity under certain conditions. The aspectual thematic role in this sense is dened by the entailments about aspectual structure that it gives rise to (see also Ramchand 1993, Ramchand 1997). The argument structure of a predicate is one of the things that linguists have felt the need to specify anyway in the lexicon to capture certain linguistic generalisations (Baker 1988, (Grimshaw 1991)).1 The intuition behind the aspectual theta role theories is that the lexical determination of aspect follows directly from the lexically specied (aspectual) argument structure of the predicate. Of course, the ip side of the lexical specication problem is the fact that lexical predicates typically demonstrate argument structure exibility. In other words, a particular verb can appear in one or two or more argument frames. (1) (a) John broke the vase. (b) The vase broke. (2) (a) John walked. (b) John walked Mary to the store. We need to be able to classify the argument types specied by different lexical items while still allowing them the exibility we nd. At the same time, this exibility is not completely general, otherwise the problem would be considerably simplied. (3) (a) John arrived. (b) *Bill arrived John.

1.2

Decomposing the Event

Another very common technique of verbal decomposition over the years is the one that started with the generative semantics tradition (McCawley 1968) and was pursued in various forms over the years by linguists such as (Dowty 1979) and (Parsons 1990), and which nds its echoes today in more minimalist frameworks (Hale and Keyser 1993). In this sort of framework, the verbal meaning is composed of subevents which are independently conceptualised using primitives such as C AUSE, D O and B ECOME . At the same time, in the event semantics tradition, complex events of the accomplishment kind are explicitly decomposed into subevents with primitive modes of event composition available (Parsons 1990,Pustejovsky 1991) The event analogue of the exibility noted in the previous section is what one might call template augmentation (Levin and Rappaport 1998) or event type-shifting (van Hout 2000, 2001). One example might be the adding of a causative subevent to an already possible event structure (as in (4)), or the adding of a telos to a process verb (as in (5)) 2 (4) John jumped the horse over the fence.
1 2

Assuming we are not embracing the constructional radicalism of Borer 1998. Here I am assuming that a telos can be added by a PP, adjectival resultatives, and particles.

(5) John ate the porridge up. Once again, these processes are not completely general since some verbs seem to resist causativisation (6a), and others resist telic augmentation (6b) (6) (a) *John slept the baby. (b) *John watched Mary bored/to boredom. Given the number of different ways of approaching the problem, some natural questions arise. I Which primitives are minimally necessary to express (a) the semantic elements of aspectual composition? (b) the lexical specicity of vocabulary items in semantic and syntactic environments? (c) the featural elements responsible for driving the corresponding syntactic representations? II Are these three separate questions, or just one? In other words: (a) Do we need both (lexical) argument role classifying primitives as well as (semantic) event structure primitives or does one follow from the other? (b) How is the primitive classication (in whatever terms) cashed out and implemented within a syntactic derivation? In this paper, I pursue the intuition that lexical, semantic and syntactic generalisations are all driven by the same set of abstract primitives. This paper is divided into two parts. First, I try to justify the minimal elements required by any theory, based on well known natural language patterns and previous work in the area. In the subsequent sections, I offer an explicit proposal of lexical specication, where the same set of features is the basis for both the semantic/event structure composition and for the construction of syntactic representations. 2. The Relevance of Causation

Ever since the unaccusative hypothesis of Perlmutter 1978, the existence of an external argument or agent has been cited as criterial of a major division in (intransitive) verb types. However, Rappaport-Hovav and Levin 2000 show convincingly that it is not agency per se that determines class membership as either unaccusative or unergative. The following intransitives cited by them pass the diagnostics for unergativity in Italian, Dutch and Basque even though they are profoundly non-agentive. (7) glow, stink, spew Even in English, the fact that these verbs possess an external argument can be demonstrated by their ability to take Xs way objects under certain circumstances (examples 8) and also show an inability to causativise (examples 9)3. (8) (a) He stank his smelly way home. (b) The water spewed its way along the corridor. (c) John ran his way into history.
3

These examples are taken from Rappaport-Hovav and Levin 2000

They also show an inability to causativise. (9) (a) *Michael glowed Karenas face. (b) *We spewed the water out of the sink. (c) *We stank the dog by throwing him in the cesspit. (d) *John ran Mary by scaring her with a live mouse. I accept the general intuition that there is an important primitive underlying the distinction between internal and external arguments, but I will assume (with Rappaport-Hovav and Levin 2000 and many others) that the relevant abstract category is that of causer or initiator. Among transitive verbs as well, external arguments can be volitional agents (10a, 10b), instrumentals (10c), abstract causes/sources (10d 10e, 10f). (10) (a) John broke the window. (b) John built that house. (c) The hammer broke the window. (d) The videotape from the secret camera demonstrated the truth of the matter. (e) The storm broke the window. (f) Johns money built that house. Im going to assume, therefore, that even though agency might be relevant for the felicity in certain circumstances, it does not directly determine syntactically relevant class membership.

3.

The Relevance of Telicity

There seems to be a fair amount of evidence which argues that certain argument positions are more implicated in the construction of telic effects within VPs than others (classically, the internal argument vs. the external one). Here I will argue that there is denitely a relationship, but that there is no straightforward one to one correspondence between internal arguments and the semantic feature [+telic], even when the internal argument in question is quantized (in the sense of Krifka 1989) or [+SQA] (in the sense of Verkuyl 1993). Specically, I will argue that there are two distinct kinds of aspectually sensitive internal arguments, and that quantization is only relevant for a subtype of one of these. The arguments for the lack of a simple relationship between the feature [+telic] and the internal argument go in both directions. First of all, the existence of telicity does not actually imply the existence of an internal argument. (11) John stood up in a second. Conversely, the existence of an internal argument does not imply telicity (not even when it is quantized). (12) John pushed the cart for hours. Indeed, going back to diagnostics for unaccusativity, it seems that verbs which describe a change of state on the part of an argument do not necessarily entail the attainment of a nal state. (13) widen, harden, descend, rise, fall

A gap can widen but is doesnt necessarily become wide; a balloon can rise without ever hitting the top of the sky. (14) (a) The gap widened for three minutes (and then it began to close again). (b) The balloon rose for 3 minutes. (c) The ball rolled for 30 seconds. Moreover, these verbs display unaccusative behaviour in the languages where the diagnostics are clear, indicating that they actually have internal arguments in the relevant sense. Correspondingly in English, as Rappaport-Hovav and Levin 2000 note, these verbs do not occur in the Xs way construction; and many of them do causativise. It seems that what is crucial here is the notion of the argument undergoing some sort of identiable change/transition, whether it is with respect to its location (15), its state (16), or its material extent (17). (15) (a) The ball rolled down the hill. (b) The mangoes ripened in the sun. (c) The bucket lled with rain. In the case of transitive verbs, we nd direct objects that full this condition of undergoing change as well: you can make a good object if the change that of location (16), state (17) or material extent (18) (see Ramchand 1997 and Hay, Kennedy, and Levin 1999). (16) (a) John pushed the cart. (b) Mary dried the cocoa beans. (c) Michael gobbled the mangoes. The broad notion of U NDERGOER (after Van Valin 1990) seems to be the one responsible for class membership here, and includes objects of verbs of creation/consumption like eat, as well as objects of verbs of translational motion like push and drive. In other words, the existence of an U NDERGOER does not necessarily imply telicity, even when it is quantized (however we choose to dene that). (17) (a) Michael gobbled the mango in an hour/for an hour. (b) John pushed the cart for an hour. (c) Mary dried the cocoa beans in the sun for an hour. These objects are merely undergoers of transitional states. If the transitions are related to the objects material extent, then quantizedness will produce a telic entailment as in (15). But if the transitions are related to the objects change of location, then only the specication of a nal location will create telicity (18). (18) John pushed the cart to the end of the garden. If the transitions are related to the objects change of state, then only the specication of the nal relevant state will create telicity (19) (see Hay, Kennedy, and Levin 1999 for a detailed discussion of telicity effects with this type of verb.) (19) Mary dried the cocoa beans bone dry in only 12 hours.

Thus, we really need to distinguish between U NDERGOER and the purely semantic notion of telicity. None of these verbs is obligatorily telic; they can be interpreted as telic as a result of entailments triggered by nature of the direct object or the existence of a nal state achieved. I take the telicity effects in this class of verbs to be purely semantic, and not encoded in the lexical determination of the verb or its syntactic reexes. However, there are certain verbs that are obligatorily telic. They systematically reject the for an hour test, in contrast to the group of verbs above where it is always possible to get an atelic reading. (20) (a) John broke the stick in a second/*for seconds. (b) Mary arrived in two minutes/*for two minutes. (c) Michael gobbled the mangoes up in just 10 minutes/*for ten minutes. Clearly, the telicity of this class of verbs needs to be represented differently from the telicity that sometimes arises from the semantic combination of an U NDERGOER verb and a quantized object. In other words, these verbs resist the atelicity test because their objects are already dened as holders of a nal state. They arent merely U NDERGOERS ; I will call them R ESULTEES . I have stated the generalisations primarily over argument type, but also implied that they corresponded to a generalisation about the relevant subevent. The question is whether these generalisations are indeed actually equivalent, and whether we can get away with specifying only one and not the other. In earlier work, van Hout 1996 claimed that all that was necessary in the theory was a statement of lexical event type. Argument structure could then be derived from this by systematic principles which allow the subevent to be identied (cf. Grimshaw 1991, Grimshaw and Vikner 1993). However, van Hout 2000 retreats from this position, showing that different verbs with the same event type nevertheless can specify different numbers of arguments. In the next section I will try to develop a proposal which goes back to the strong position found in the earlier van Hout, but with a ner grained analysis of subevents.

4.

An L-syntax Decomposition

The challenge now is to understand how the specication of lexical items is stated so as to determine (but not overdetermine) the sorts of syntactic representations they are compatible with. In addition, the lexical specication needs to provide the features that will be the input to a compositional aspectual semantics in the determination of the meaning. The proposal I will make here is that the lexical features allow the licensing of syntactic structure (seen as an l-syntax in the sense of Hale and Keyser 1993) and that this l-syntax is then interpreted by a strict event-compositional semantic component which feeds off the syntactic representation so constructed. In this way, one set of features will be responsible both for the argument structure syntax and the event structure interpretation. According to the discussion in the previous section, I will assume that the lexical item corresponding to a dynamic verb needs to specify (i) the presence or absence of a cause subevent, and (ii) the presence of absence of a result subevent.4
I am assuming here that there is a process portion that is the necessary component of any dynamic (i.e. nonstative) verb. The other two components may or may not be present.
4

Thus, the full l-syntactic structure that could be expressed by a lexical dynamic verb is shown below5. vP

VP

R P
    

R
 

XP

I claim that the different patterns of l-syntactic projection correspond intuitively to what many linguists have tried to capture with the notion of aktionsart category (as opposed to sentential aspect). For the semantic interpretation of this structure, I adopt a Post-Davidsonian 6. semantics which interprets the verbal heads of the l-syntax in a regular and systematic way. There are two primitive modes of sub-event composition to create complex events : Event Composition Rule I e=e e : e consists of two subevents, e , e such that e leads to or causes e (see Hale and Keyser 1993) Event Composition Rule II e= e,e : e consists of two subevents, e , e such that e and e form an accomplishment event structure where e is the process portion and e is a state interpreted as the result state of the process. (see Parsons 1990, Higginbotham 1999, Pustejovsky 1991)
         !      

There are a number of general primitive predicates of events as follows: State(e) : e is a state Process(e): e is a process Causing(e): e is a causing event Further, the objects of particular event types are interpreted according to the primitive role types dened below. Object (x, e) and Causing(e) entails that x is the I NITIATOR of e. Object(x, e) and Process(e) entails that x is the U NDERGOER of the process. Object(x, e) and State(e) entails that x is the H OLDER of the state. The R head in the l-syntax is interpreted as building a state description that has particular holder in its specier position. Its semantic interpretation is given below:


(21)
"

"

Rv =
# # $

P x e [P(e) & State(e) & Object(x,e)]

$ $

The central verbal head, V, is interpreted as a process which has an U NDERGOER in its specier position. If it combines with a state description projection as its complement, that state is interpreted uniformly as the result state emerging from the process. The interpretation of such a head is given below in (22)
See Borer 1998, Ritter and Rosen 1998 and Hale and Keyser 1993 for some recent proposals for event structure decomposition within syntax. 6 I use the term post-Davidsonian to describe the theoretical position whereby verbs do not have events in their argument structures, but where events and subevents corresponding to the interpretation verbal heads are utilised in the expression of compositional semantic meaning.
5

(22)
%

V = R x e e , e [R(e ) & V (e ) & Process(e ) & e =

& & ' ' ' ( ) 0 0 1 ) ) 2

e ,e
) 0 3

& Object(x, e )]
)

Any argument which is the holder of the result state is now the RESULTEE in my terms. Finally, the highest verbal head v, is interpreted as an initiating event which leads to the (possibly complex) event constructed by the lower structure that it combines with. The specier position of this projection is interpreted as the causer or INITIATOR of the subevent. The interpretation of v is given in (23). (23)
% %

v = R x e e , e [R(e ) & v (e ) & Causing(e ) & e = e

& & ' ' ' ( 4 5 5 1 4 4 4

e & Object(x, e ) ]
5 4

Given the semantics of these various heads, if the heads are not built up in the correct order, the derivation will at best converge as gibberish. Even within this broad constraint, it is clear that there are a number of different l-syntaxes that can be built using this basic inventory of functional heads. Thus, the next step involves implementing a traditional notion of selection. Over the past forty years, linguists have used the notion of selection (c-selection or s-selection) to describe a wide variety of different data and phenomena across languages. Within the newest syntactic frameworks, the question of how to implement those generalisations naturally arises. In a system like the Minimalist Program (Chomsky 1995, Chomsky 1999, Chomsky 1998), where syntactic structures are built up by Merge and Move derivationally, subject only to principles of full interpretation at the interfaces, all selection needs to be implemented in terms of feature checking. Further constraints are put on the process if we assume together with Pesetsky and Torrego 2000 that features come only in interpretable/uninterpretable pairs, and that an interpretable feature on a probe checks and values an uninterpretable feature on a goal in its domain within the phase (as in Chomsky 1998). Implementing selection in this kind of system crucially depends on the types of features available in syntactic computations. It would be nice if c-selection could reduce to the checking of the categorial features of heads (as in Svenonius 1994), with no subclass of selectional features involved.7 I will make the assumption that the only features that are allowed are ones that are interpretable somewhere. The features we minimally need for the l-syntax proposed above are the category features of the three eventive functional heads. I will further assume that they are interpretable and that they are the features that trigger the semantics of event composition. Since we minimally need [+v, +V and +Rv] to get the semantics of the event composition to come out right, we should assume that this is all we need. First of all, the l-syntax built according to the template above, can be freely built up by Merge but not all three heads will always be licensed. I implement this by making a distinction between the merging of purely syntactic elements bearing formal features vs. the substitution of lexical items into those syntactic nodes of the structure. To make selection work in these terms, purely syntactic nodes must bear only uninterpretable category features; while only actual lexical items have interpretable category features (recall that an unchecked uninterpretable feature at the end of a phase causes the derivation to crash (Chomsky 1999, Chomsky 1998)).
As briey mentioned before, there are two sides to the selectional problems in this domain: on the one hand verbs in many languages seem to exhibit event structure exibility and argument structure changing in a general and systematic way (a fact emphasised by Borer 1998); on the other hand, verbs rarely, if ever, show the complete exibility one would expect if roots carried no syntactic or selectional features (facts emphasised by the lexicalists, eg. Levin and Rappaport 1998). The system I will propose involves a constrained departure from recent hypotheses of Marantz 1997 and Borer 1998 concerning the impoverished nature of roots/vocabulary items in relation to their syntactic environment.
7

A. Licensing and Identifying Event Structure (cf. van Hout 2000) (i) Syntactic Licensing: Build syntactic structure freely with no lexical terminals. The heads of such structure contain only uninterpretable category features. Structure cannot be licensed unless there is an interpretable category feature in the same phase (on the lexical item) which will check it via AGREE. (ii) Semantic Licensing/Identication: The eventive content of each head needs to be semantically identied. This can be done via overt lexical content in the head, or by being in a chain with something with overt lexical content. Thus, if a lexical item is specied as having the interpretable category feature the structure will be licensed, and not otherwise. In addition to the building of l-syntactic structure, we also need to have some principles which regulate the appearance and interpretation of nominal arguments. I will make the assumption that all the heads in the l-syntax require a lled specier (i.e. they all have an [+EPP] feature). As we have seen, these specier positions are interpreted systematically by the general semantic component as: I NITIATOR, U NDERGOER and R ESULTEE respectively. There are thus no thematic roles, only three universal semantic rules triggered by syntactic structure. One major departure I make from other systems is that I propose that these specier positions are not mutually exclusive. In other words, it is possible for a single argument to be in more than one of these positions simultaneously (or have them linked together in an A-chain). This means that I am assuming that there is no -Criterion, and that the semantic interpretation of the positions so linked get unied. In principle, there is no incompatibility between the semantics of I NITIATOR , U NDERGOER , R ESULTEE and so no violations will occur purely because of unication. To capture the facts about argument structure which underlie much of the work on lexical semantics, we need to state a relationship between the lling of the specier positions of an l-syntactic position and the verbs feature specication. In short, one further condition lexical items appear to impose on their syntactic representation is whether the specier positions of the l-syntactic projections they license are lled by separate nominals or by the same nominal. I implement this with the feature [+Move]. Assuming that a specier position is lled by means of M ERGE in the default case, if the specier is lled by raising an already present argument from below, then it needs to explicitly require M OVE 8 . (See McCloskey 2000 for an identical feature.)
7

B. The Lexical Items The lexical items in this system come specied only with the following: (i).A number of interpretable category features (ii) Information on whether/if any of the category features are [+Move]. C. Basic Principles of Argument Realisation (i) Each head in the l-syntax requires a lled specier position (i.e. comes with an [+EPP] feature). (ii) A single argument can be in more than one Spec position. In this system, no specic rules of augmentation will be required. Structure building will be
If M OVE actually turns out to be conceptually more natural than M ERGE, we could implement the intuition by specifying the heads that require M ERGE specically instead.
8

free, modulo syntactic and semantic wellformedness.9 5. Deriving Verb Classes

To give some concrete examples, if a lexical verb is specied as [+V, +Rv], it can license a Process-Result structure. The verb would Merge and project as Rv, and then the V remaining V feature would check the higher uninterpretable head, presumably also triggering head to head movement in that case to identify the relevant subeventive structure. (24) Process-Result: x broke
9

VP
8 8

8 8

9 9

U NDERGOER
B B B B

V
@ A A A A

V
B B B

RvP
B A A A A

R ESULTEE
D D

Rv
@ C

Rv break

XP

If the lexical verb is specied as [+v, +V, +Rv], then it can license a full Cause-ProcessResult structure, with the verb Merging as little Rv, and then raising to check and identify the features on the V and v heads. (25) Cause-Process-Result:x broke y
9

vP
8 8

8 9

8 8

9 9

I NITIATOR
9 9 9 9 9

v
8 @

8 8

v
9 9 9 9 9

VP
8 8

8 8

U NDERGOER
B B B B

V
@ A A A A

V
B B B

RvP
B A A A A

R ESULTEE
D D

Rv
@ C

Rv break

XP

Thus, the verb break comes with a full complement of category features, but with the v feature optional. Moreover, the R ESULTEE and U NDERGOER are always identied with this verb, indicating that the V feature must be specied as [+Move]10.
As we will see in the next section, some exibility in a verbs lexical entry can also be introduced by the optionality of some of the category features in its specication. 10 Interestingly, there seems to be an empirical generalisation here that in English the R ESULTEE and the U N DERGOER are always identied with each other when they both exist. Identication of the adjacent Specier
9

(26) Lexical entry for break: (v), V

E

, Rv

Unergative motion verbs are distinguished in this system because they effect a different kind of identity that between the I NITIATOR and the U NDERGOER. (27) Cause-Process:x walk
S

vP
R R

S R

S R

I NITIATOR
S S S S S

v
R T R

v
S S S S

VP
R R R R V

U NDERGOER
V

V
T U U

V walk

XP

(28) Lexical entry for walk: v

E

,V

This identity is not logically necessary of course, since both arguments can be separately represented in other dynamic non-R verbs such as push and eat.
I

(29) Cause-Process: x pushed/ate y

X

vP
W W

W W

X X

I NITIATOR
X X X X X

v
W T

W W

v
X X X X X

VP
W W

W W

U NDERGOER
` `

V
T Y Y

V push/ate

XP

(30) Lexical entry for push: v, V (31) Lexical entry for eat: v, V Recall that the verb eat is a directed change verb in this system, not a result verb. Since it is a creation/consumption verb, if the object is quantized, it will give rise to a telic interpretation of the VP (32a). This is different from being a Result verb which is telic regardless of the nature of the object (32c)11
positions may be a prerequisite for accomplishment formation in the interpretation of the l-syntax. I leave further generalisations of this type to further research. 11 This reading forces a specic indenite reading of apples in (32c).)

(32) (a) John ate the apple in an hour. (b) John ate apples *in an hour. (c) John ate apples up in an hour. Pure process verbs are also possible in this system (unaccusatives which express a directed change, with no necessary nal end state). In the case of the verb roll below, it is also possible to nd this verb in a transitive version (although this is not found for all pure process verbs in general). (33) Process: x rolled
b

VP
a a a a e

U NDERGOER
e

V
c d d f g

V rolled

(34) Lexical entry for roll: (v), V There also seem to be systematic ways of of building RvPs within the l-syntax of English to augment process verbs. Recall that structure can be built freely, but that if there is no lexical item to check the structure it will not converge. Under this system, PPs, particles and adjectival resultatives are lexical items which allow the identication and licensing of a result state in English and can therefore be freely added to any process verb, and will be felicitous provided the state so identied is compatible with the conceptual structure of the verb. (35) John ate the apples up. (36) John wiped the table clean. (37) Cause-Process-Result:x wipe y clean
i

vP
h h

h h

i i

I NITIATOR
i i i i i

v
h c

v
i i i i i

VP
h h

U NDERGOER
b b b b b

V
a c a a a a

V
q q q

RvP
p q p p p

wipe

R ESULTEE
s s

Rv
c r r

Rv

XP clean

(38) Lexical entry for wipe: v, V .

6.

Conclusion

In this paper I have argued for a particular set of features which constitute the lexical determinants, or lexical inputs, to an aspectual calculus. I have argued that we can get away with just specifying which of the features v, V and Rv are present in interpretable form on the lexical item, plus information on whether the obligatory speciers are lled by Merge or Move. This system is thus an l-syntax version of the intuition in van Hout 1996 that the lexical item determines event structure directly and only indirectly the argument structure. I also argued that a constructional view of the l-syntax and its universal semantics allows for argument structure exibility within the limits of the verbs lexical specication.

References
Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago and London: University of Chicago Press. Borer, Hagit. 1998. Deriving passive without theta roles. In Diane K. Brentari Stephen G. Lapointe and Patrick M. Farrell, eds., Morphology and its Relation to Phonology and Syntax, 6099, Stanford, California: CSLI Publications. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 1998. Minimalist Inquiries. MITOPL 15. Chomsky, Noam. 1999. Derivation by Phase. MITOPL 18. Dowty, David. 1979. Word Meaning in Montague Grammar. Dordrecht: Reidel. Grimshaw, Jane. 1991. Argument Structure. Cambridge, MA: MIT Press. Grimshaw, Jane and Vikner, Sten. 1993. Obligatory adjuncts and the structure of events. In E. Reuland and W. Abraham, eds., Knowledge and Language II: Lexical and Conceptual Structure, 143155, Dordrecht: Kluwer. Hale, Ken and Keyser, Samuel J. 1993. On argument structure and the lexical expression of syntactic relations. In K. Hale and S.J. Keyser, eds., The View from Building 20: A Festschrift for Sylvain Bromberger, Cambridge, Massachusetts: MIT Press. Hay, Jennifer, Kennedy, Christopher, and Levin, Beth. 1999. Scalar structure underlies telicity in degree achievements. In Proceedings of SALT IX , 127144. Higginbotham, James. 1999. Accomplishments, ms.paper presented at GLOW, Japan. Hoekstra, Teun. 1984. Transitivity: Grammatical Relations in Government and Binding Theory. Dordrecht: Foris. Kenny, A. 1963. Action, Emotion and Will. London: Routledge and Kegan Paul. Krifka, Manfred. 1989. Nominal reference, temporal constitution and quantication in event semantics. In R. Bartsch, J. van Benthem, and P. van Emde-Boas, eds., Semantics and Contextual Expressions, 75115, Dordrecht: Foris. Levin, Beth and Rappaport, Malka. 1995. Unaccusativity: At the syntax-lexical semantics interface. Cambridge, MA: MIT Press. Levin, Beth and Rappaport, Malka. 1998. Building verb meanings. In Miriam Butt and Wilhelm Geuder, eds., The Projection of Arguments: Lexical and Compositional Factors, 97134, CSLI publications.

Marantz, Alec. 1997. No escape from syntax: Dont try morphological analysis in the privacy of your own lexicon. In L. Siegel et. al A. Dimitriadis, ed., UPenn Working papers in Linguistics, volume 4.2, 201225, University of Pennsylvania. McCawley, James. 1968. Lexical insertion in a transformational grammar without deep structure. In Proceedings of the Chicago Linguistic Society, volume 4, 7180. McCloskey, James. 2000. Resumption, successive cyclicity, and the locality of operations, ms. UCSC, Santa Cruz, CA. Parsons, Terence. 1990. Events in the Semantics of English. Cambridge, MA.: MIT Press. Perlmutter, David. 1978. Impersonal passives and the unaccusative hypothesis. In Proceedings of the Berkeley Linguistic Society, volume 4, 157189. Pesetsky, David and Torrego, Esther. 2000. T to C movement: causes and consequences, ms., MIT, Cambridge, MA. Pustejovsky, James. 1991. The syntax of event structure. Cognition 41:4781. Ramchand, Gillian. 1993. Aspect and argument structure in modern scottish gaelic. Ph.D. thesis, Stanford University, Stanford, CA. Ramchand, Gillian. 1997. Aspect and Predication:The Semantics of Argument Structure. Oxford: Oxford University Press. Rappaport-Hovav, Malka and Levin, Beth. 2000. Classifying single argument verbs. In M. Everaert P. Coopmans and J. Grimshaw, eds., Lexical Specication and Insertion, 269304, John Benjamins. Ritter, Elizabeth and Rosen, Sara T. 1998. Delimiting events in syntax. In M. Butt and W. Geuder, eds., The Projection of Arguments, Stanford, California: CSLI Publications. Svenonius, Peter. 1994. C-selection as feature checking. Studia Linguistica 48:133155. Taylor, B. 1977. Tense and continuity. Linguistics and Philosophy 1.2:199220. Tenny, Carol. 1987. Grammaticalizing aspect and affectedness. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA. van Hout, Angeliek. 1996. Event semantics of verb frame alternations: A case study of dutch and its acquisition. Ph.D. thesis, Tilburg University, published in 1998 by Garland Publishing, New York. van Hout, Angeliek. 2000. Projection based on event structure. In M. Everaert P. Coopmans and J. Grimshaw, eds., Lexical Specication and Insertion, 397422, John Benjamins. Van Valin, Robert D. 1990. Semantic parameters of split intransitivity. Language 66:221 260. Verkuyl, Henk J. 1972. On the Compositional Nature of the Aspects. Dordrecht: D. Reidel. Verkuyl, Henk J. 1989. Aspectual classes and aspectual composition. Linguistics and Philosophy 12:3994. Verkuyl, Henk J. 1993. A Theory of Aspectuality. Cambridge: Cambridge University Press.