(A)telicity and intentionality (1).

Depraetere, Ilse

Abstract

This article argues for a treatment of telicity that gives due space to intentionality, i.e., that recognizes the role intentionality may play in establishing the mutually manifest inherent or natural endpoint crucial to the definition of telicity. Sentences with numerical NP objects and for adverbials are shown not to be automatically telic. It is only if the constituents in question are contextually given as inherent endpoints that they are [+telic], intentionality being one of the extralinguistic factors that may bring about telicity.

1. Introduction

The concepts of telicity and atelicity are commonly used to refer to two fundamentally different situation types. While the labels are always associated with presence vs. absence of endpoints in some way, the definition of endpoints is not uniform. The following is a selective survey of ways in which telicity has been defined: (2)

(1) Binnick (1991: 192): "The definition of telicness must take into account that what is crucial is not that there be a potential culmination of the actual process described, but rather that the phasic structure of the situation include a culminatory phase distinct from mere termination." (3) [italics are mine]

Dahl (1981: 81): A situation, process, action, etc. or the verb, verb phrase, sentence, etc. expressing this situation etc. has the T property if

(DEFINITION 1, S. G. Andersson 1972) it is directed toward attaining a goal or limit at which the action exhausts itself and passes into something else

(DEFINITION 2, Comrie 1976) it leads up to a well-defined point behind which the process cannot continue.

Declerck (1991: 121): Telic expressions have a "natural endpoint", "a telic expression involves reference to a nonarbitrary point of completion towards which the action tends to proceed and beyond which it cannot continue".

Dowty (1991: 567): Telic predicates have an "incremental theme".

Filip (1999: 16): "On the most general level of classification, two main classes of verbal predicates and sentences are distinguished: events, which are telic or quantized, and states and processes, which are atelic or cumulative".

Garey (1957: 106): Telic verbs are "verbs expressing an action tending towards a goal".

Krifka (1992: 30): "A verbal expression is atelic if its denotation has no set terminal point (e.g., run), and it is telic if it includes a terminal point (e.g., run a mile)".

Lascarides (1991: 423): "Event sentences describe culminations, and some of them are also associated with 'prior' processes that led to the culmination".

Michaelis (1998: 17): Telic situations are "events with goal states".

Smith (1997: 19): "Telic events have a change of state which constitutes the outcome, or goal, of the event. (...) To avoid agentive connotations, I will say that telic events have a natural final endpoint, or intrinsic bound. (...) Atelic events have arbitrary final endpoints."--"The syntactic evidence for atelic event turns on the notion of completion, which involves the interaction of duration and change of state." (1997: 42)

Van Valin and LaPolla (1997: 93): "The feature 'telic' has to do with whether a verb depicts a state of affairs with an inherent terminal point or not".

Although the definitions are indicative of differences in relative weight attributed to semantic and syntactic criteria, there is a consensus among researchers that the property of telicity is determined by the multilayered interaction between the lexical semantics of the verb and different syntactic factors, which has led e.g., Verkuyl (1993) to use the term aspectual composition. Numerous studies have, for instance, been devoted to the influence of object NPs on (a)telicity, a unanimous conclusion being that mass or bare plural NPs establish atelic predicates (John drank beer/ glasses of wine), while count NPs establish telic predicates (John drank a/the glass of wine). Relatively little attention has been paid though to numerical NPs of the kind as in eat 10 cakes, drink 5 bottles, the few analyses explicitly referring to such predicates presenting them as telic propositions (cf. e.g., Filip 1997; Krifka 1992; Ramchand 1997; Rappaport Hovav and Levin 2002: 273; Smith 1997: 29). The compatibility of sentences with either a for adverbial or an in-adverbial also features prominently in the situation types literature as a test to determine the atelic/telic nature of the predicate (John drank a glass of wine in 30 seconds (John drank a glass of wine = telic) vs. John drank wine for 30 seconds (John drank wine = atelic)). However, base cases such as John stayed in the room for 30 minutes, seemingly unproblematic and telic, are virtually absent in previous research. It is my aim to take a closer look at these less commonly discussed cases in this article.

More in general, I would like to look at the effect of intentionality, a pragmatic notion, on (a)telicity. I would like to present some descriptive evidence that shows that extralinguistic factors, such as intentionality, can play a decisive role in determining the telic character of particular predicates and should accordingly be given their due place in the list of factors affecting (a)telicity. While I agree that lexical semantics and syntax are crucial to (a)telicity, (4) it is my contention that pragmatics also needs to be taken into account and is the decisive factor in classifying predicates such as John ate 10 cakes and John stayed in the room for 30 minutes, which I do not believe are unambiguously telic. The hypothesis defended relates to the subclass of telic utterances that contain (a) transitive dynamic verbs (run three rounds, write an essay, etc.) and (b) stative or dynamic (in)transitive verbs combined with a durational for-PP. Telicity is defined as follows: a sentence is telic if the situation (as it is represented in the sentence) is represented as having an inherent (natural or intended), mutually manifest endpoint beyond which the situation (as it is represented in the sentence) cannot continue. (cf. e.g., Depraetere 1995)

I will first give of survey of the different ways in which the influence of pragmatic factors (including the extralinguistic context) features in previous research on aspectuality (cf. e.g., Hay et al. 1999; Olsen 1994), one of the findings being that intentionality, or at least agentivity, is orthogonal to telicity. I will then describe some phenomena that cannot be accounted for unless there is scope for intentionality in (a)telicity: the inherent endpoint associated with telicity in the original and genuine sense of the word (cf. e.g., Garey 1957) needs to contextually given, and in a number of cases, the identification of such an endpoint is dependent on the mutual manifestness (Sperber and Wilson 1995: 39) (5) of an intention to achieve a certain aim by one of the discourse participants. So while I would not want to question in the least the importance of syntax and (lexical) semantics on (a)telicity, I would like to focus on the importance of the pragmatic level, with space for the influence of (mutually manifest) intentions.

2. Pragmatics and aspectuality

In the literature, the possible influence of the (linguistic and extralinguistic) context on aspectual classification is basically centered on three issues: (a) temporal interpretation of discourse; (b) verb classification; (c) agentivity and telicity.

2.1. Contextually established temporal boundaries

In discussions that deal with the temporal interpretation of discourse, it has been pointed out that temporal boundaries may be established by the context and result in a shift of reference time. The following examples illustrate the temporal shift brought about by our knowledge of the world:

(2) John was watching television when he fell asleep. (Dowty 1986: 59)

(3) At around five fifteen the tall boy in the jerkin came out of Geiger's with an umbrella and went after the cream coloured coupe. When he had it in front Geiger came out and the tall boy held the umbrella over Geiger's bare head. He folded it, shook it off and handed it into the car. (Couper-Kuhlen 1987: 24)

(4) James entered the room, shut the door carefully switched off the light. It was pitch dark around him. because the Venetian blinds were closed. (Hinrichs 1986: 68)

It is beyond doubt that it is contextual information (i.e., real-world knowledge about falling asleep [2], folding an umbrella [3], switching off the light [4]) that puts an end to the situation of watching (in [2]) and of holding the umbrella (in [3]), or alternatively makes it clear that a new situation comes about (in [4]), and in this way triggers temporal progression. (cf. also e.g., Hamann 1991: 429)

However, these examples are not directly relevant to the present discussion because, as I have argued in e.g., Depraetere (1995) and (1996a), the temporal endpoint identified in these discussions is of a different nature: it is a factual boundary, which is associated with boundedness rather than telicity. Endpoints may indeed be of (at least) two kinds (cf. e.g., Declerck 1989, 1991, and 2006), which matches a dual conceptual distinction between, on the one hand, telicity vs. atelicity, a distinction that hinges on the presence or absence of potential inherent endpoints ("ontological aspect", "situation-template" Declerck 2006), and, on the other hand, boundedness vs. unboundedness, a distinction based on the presence or absence of factual, arbitrary endpoints ("actualisation aspect" Declerck 2006). (6) In such an approach, the sentence in (5) is bounded and telic, that in (6) is unbounded and telic.

(5) I wrote an article on telicity in 1995. (bounded and telic)

(6) I am writing an article. (unbounded and telic)

In a similar way, the italicized clause in (7) is atelic and unbounded, that in (8) is atelic and bounded.

(7) She was in the attic when I arrived. (unbounded and atelic)

(8) She forgot time while she was filing through her children's photo albums. She was in the attic for three hours. (bounded and atelic)

The etymology of telos, which includes an element of meaning like end, purpose or goal (cf. e.g., Peters 1967; Waanders 1984) is also a telling reminder of the kind of endpoint that is involved in telicity. As will be clear from some of the "aspectual definitions" in (1), telic endpoint is sometimes used rather loosely: it does not necessarily catch the original meaning of the term, and even if it does at first sight, the discussions that follow the definitions very often take a mere factual boundary as a sufficient criterion for telicity. While this may reflect a conscious choice (cf. e.g., Filip 2000), (7) the failure to distinguish clearly between (un)boundedness and (a)telicity implies that one is no longer in a position to explain accurately a number of data, as will be shown below. (cf. also Depraetere 1995, 1996a) As I see it, the defining criterion of telicity is the presence of an inherent (natural or intended), mutually manifest (Sperber and Wilson 1995: 39) endpoint. Once that endpoint has been reached, the situation as it is represented in the sentence cannot continue any more. It is crucial to realize that the endpoints of a telic situation are potential endpoints: no matter whether there is "actualization" or not, i.e., no matter whether the sentence represents the situation as reaching/having reached the inherent or intended endpoint (cf. e.g., [5]) or not (cf. e.g., [6]), it remains a fact that it is mutually manifest and this is what telicity is about. (8) In other words, the endpoint that is characteristic of telic predicates is not a post hoc factual endpoint but one that is mutually manifest as soon as the situation is referred to. (Un)boundedness relates to whether or not the situation is described as having reached a temporal boundary. A sentence is bounded if there are linguistic signals that indicate that the situation has reached a boundary, irrespective of whether that boundary constitutes an inherent (possibly an intended) endpoint or not. (9)

2.2. Volition and intentionality

The effect of the extralinguistic category of intentionality and volition features in discussions of situation types (verb classes) and in discussions on thematic roles and their importance for telicity. (10,11)

From the early "aspectual" (pre-case-grammar) days onwards (cf. e.g., Ryle 1988 [1949]), the less than perfect match between "volition" and Achievement verbs has been referred to. Smith (1997: 31) uses the term "purely lucky Achievements" to refer to Ryle's examples find, win, etc. (1988 [1949]: 125, 144-146), which are not compatible with the adverb deliberately in spite of the fact that Achievements are typically associated with the semantic role of [+"volitional"] Agents. (12)

More in general, the influence of the context on Aktionsart category is also clear from examples like She doesn't know me vs. Apparently, my neighbor isn't knowing us again. (Schopf 1984: 247). Schopf explains that whereas know is a state verb in sentences like the former, it testifies to "willentliches Verhalten" ["intentional behavior"] in the latter case. A lot of stative verbs (I think you're right.) can indeed get, the context permitting, an activity reading (I'm thinking of you night and day.) or a process reading (I'm understanding the problem better and better.). (cf. e.g., Bach 1981; Quirk et al. 1985: 202; Huddleston and Pullum 2002: 167)

While "volitional involvement" (Dowty 1991: 572) or intention (cf. e.g., Lakoff 1977: 248; Haegeman 1994: 49; Nishimura 1993: 518) is readily associated with the thematic role of agent, it has been pointed out that not all agents answer that requirement. (13) For instance, the (animate) subjects of what Levin (2000: 420) calls "verbs of substance emission" such as sneeze, vomit, (14) and other verbs of bodily process like yawn, are readily associated with the semantic role of agent although the agent does not appear to have the property [+"volitional"]. In a related context, Verkuyl uses the apparent tautology "voluntary agency" (Verkuyl 1993: 38). Cruse (1973) also points to the need of setting up subclasses within the category of "doers"; "agentiveness" being but one of four (next to volitive, effective, initiative) features that establish reference to a "doer": volitive "is present when an act of will is stated or implied" (1973: 18), agentive "is present in any sentence referring to an action performed by an object which is regarded as using its own energy in carrying out the action" (1973: 21) (cf. also e.g., Van Valin and LaPolla 1997: 119-120). In Van Valin and Wilkins (1999: 291) it is argued convincingly that the role of effector, "roughly the dynamic participant doing something in an event" is more basic than that of agent, agentivity arising as a result of an agent implicature. Following Holisky (1987), it is shown how the semantic role of effector can be contextually enriched so as to establish agentivity. Larry killed the deer (Van Valin and Wilkins 1999: 309) is said to be neutral with respect to whether the actor is an agent or not. While the default reading is one in which Larry is agent, a reading that is brought out by adding for instance, intentionally, the addition of accidentally cancels the (implicated) default reading. In other words, rather than considering intentionality as a defining feature of agents in general, with the complication that certain agents do not have that characteristic, they opt for an approach in which agents constitute a subclass (dependent on the feature of intentionality) of the more basic role of effector.

Before taking a closer look at the link between agentivity, intentionality, and telicity, I would like to develop in more detail an aspect about telicity that is vital to the argumentation in this article. More in particular, I would like to take up the notion of "inherent endpoint" referred to in 2.1 and apply it to examples with transitive dynamic verbs with an object NP that contains a numerical NP. As pointed out in 2.1, a fact that I believe has been insufficiently stressed in the literature is that in order for a predicate to qualify as telic, the discourse participants need to be able to conceptualize the situation referred to in a particular predicate as having an inherent endpoint: as soon as the situation has started and as soon as the situation is referred to, it is mutually manifest what its potential endpoint is. (15)

In the case of dynamic verbs that are subcategorized for a multiple numerical object NP that is affected, constructed or consumed theme (e.g., eat 3 apples, write 5 books), the most likely way in which the boundary referred to by a quantified NP will be identified as an inherent endpoint is when it is understood to be part of a plan of one of the discourse participants. If it is in the "scope" (16) of the discourse participant's intention, it will be immediately visible as the potential endpoint of the situation.

In order to prove this hypothesis, we need to start by stating the obvious: a number is the result of a counting process. It will be clear that counting can only be completed when the situation is over, so unless a particular number is fixed as a (predetermined) target, a numerical NP is not likely to be considered an inherent endpoint. The difference in "telic potential" of different types of object NP becomes particularly clear when we consider a set of progressive dynamic sentences with an object NP that is affected, constructed or consumed theme and examine how "automatic" the assignment of the function of potential inherent endpoint to a NP with a singular indefinite article, a singular definite article, a singular numerical NP object, a multiple numerical NP is:

(9) a. Look! He's killing a chicken!

b. (context: language class, teacher showing a picture and asking questions about it) A: What is this? B: It's a chicken. A: What's the man doing? B: He's killing the chicken.

c. ?? Look! He's killing one chicken!

d. Look! * He's killing five chickens! (acceptable only if it is clear from the context that exactly five chickens will be killed, for instance, if there is a cage with five chickens to be killed) (17,18)

(10) a. Look! He's smoking a cigarette! (I thought he didn't smoke.)

b. Look! He's smoking the cigarette! (context: teenager has been fiddling with a cigarette that has been left on the table. After a while, he lights it and he starts smoking.)

c. ?? Look! He's smoking one cigarette!

d. Look! * He's smoking five cigarettes! (acceptable only if we envisage a context in which the number is fixed before the situation begins, for instance, a betting context in which one has to smoke five cigarettes in, for instance, a certain amount of time)

Dynamic telic sentences can normally be used in the progressive whose function it is to refer to the "middle of the event" (ongoing event convention, Smith 1997: 89), in which case the beginning and end of the situation are out of focus. (19) The fact that some of the closely related sentences with quantified NPs in (9) and (10) are questionable or acceptable only with a particular reading, implies that they are not (straightforwardly) telic.

These examples show indeed that even when the number of referents involved is the same (i.e., a single one: (a): a, (b): the, and (c): 1), the different determiners affect the likelihood of a sentence being understood as telic. Sentences with a definite or indefinite article object NP appear to be automatically understood as telic; explicit or contextual reference to intentionality is not required. In order to answer the question why it is that sentences with an (in)definite singular NP that is affected, constructed or consumed theme, unlike those with a single numerical NP, are automatically telic, we need to look at differences in degree of mutual manifestness of a potential inherent endpoint. As soon as we have some evidence (for instance, if someone takes out a cigarette and a lighter or (slightly more lugubriously) if someone gets hold of a chicken and a pair of scissors or a hatchet) we will easily conclude that the process about to be started is that of smoking a cigarette or killing a chicken. In other words, the chicken (the cigarette) or a chicken (a cigarette) is contextually given. (20) We may be mistaken (someone may be about to burn a cigarette or clip a chicken's wings), but that is a different problem that is not at stake here. (21) We readily accept a default situation in which an item is created, consumed or affected (even though the evidence on which that concept is based may be limited).

The situation of taking out a lighter and a cigarette will less likely induce us to conclude that someone is about to smoke exactly one or five cigarettes. We need more evidence to conceptualize such a situation and one way in which this evidence may be provided is by indicating that the referents of the NP are in the "scope" (cf. Note 16) of an intention (context a): We will be aware that the situation is that of smoking five cigarettes (or killing five chickens) when it is made clear from the beginning that this number is a target to be reached. John smoked five cigarettes is telic and bounded. John smoked five cigarettes is atelic and bounded in a context like the following (context b): when the situation of smoking (or killing) is over, we can count the cigarette stubs (or dead animals) and conclude: John smoked five cigarettes (John killed five chickens). In this case, five cigarettes is the result of an "after-the-fact" counting process. It constitutes the factual boundary to the situation ("actualization aspect"), but it is not the inherent endpoint ("ontological aspect").

The (un)acceptability of the progressive is dependent on whether we find ourselves in context (a) rather than (b): in the former case, it is possible to use the progressive, in the latter, it is not. There is a factual boundary to both the situations as they are described in the nonprogressive sentence, i.e., they are both bounded, but this does not mean that they are both telic. On reading (b), the sole function of the numerical NP is to establish a boundary to the situation. The addition of the progressive, which represents the situation as ongoing and therefore leaves the beginning and end of the situation out of focus, makes the sentence uninterpretable because of the tension between the [+bounding] operator (i.e., the multiple numerical NP) and the [-bounding] operator (i.e., the progressive). On reading (a), there is no such problem of incompatibility as the numerical NP, apart from bounding the situation, also has the function of indicating the inherent endpoint. It retains the latter (but not the former) function when the progressive is used.

(10) e.

John smoked five cigarettes. Context (a) Context (b)

Factual boundary + +
Inherent endpoint + -
Bounded + +
Telic + -
Progressive acceptable? + -

The progressive can only be used if the numerical NP constitutes an inherent boundary, i.e., if the sentence in which it occurs is telic. In this case, the numerical NP retains its function as inherent endpoint, but it no longer bounds the situation. It is important to point out that I do not claim that the progressive is ruled out on reading (b) because the sentence is atelic, but rather because of the incompatibility between an unbounding operator (the progressive) and a bounding operator (the numerical NP). (22) If the function of the numerical NP is not restricted to bounding the situation, but constitutes a telos (inherent endpoint), there is no such clash between the progressive and the numerical NP: the sentence is telic but the inherent endpoint is not represented as having been reached (i.e., the sentence is unbounded). The line of reasoning developed here is meant to prove that the presence of a numerical NP does not automatically justify the conclusion that the sentence is telic.

The questionable acceptability of the progressive sentences with a single numerical NP (cf. [9c], [10c]) or rather the particular context required to use them felicitously can be explained along similar lines: whether the reference be to one (1) or several items (e.g., 5), unless there is immediately accessible evidence about the number of items that will be involved in the situation, a single numerical NP is not automatically given as inherent endpoint. It is rather more naturally associated with a post-hoc counting process. Once the killing or smoking starts, it is evident that it will stop after exactly one cigarette or chicken only if there is an intention to do so that is mutually manifest. In other words, the sentences in (9c) and (10c) as well are only acceptable if there are contextual clues that show that one chicken or one cigarette is part of a plan in some way.

Comrie's (1976) observations are also revealing in this context: he argues that given an appropriate context, sentences that are typically considered to be atelic can be given a telic interpretation: "Imagine, for instance, a singing class where each of the pupils is required to sing a certain set passage; then the verb sing on its own, in this context, may be taken to mean 'sing the set passage', so that from John is singing it will not follow that John has sung" (Comrie 1976: 46; cf. also Dowty 1979: 61; Smith 1997: 53). Two conclusions may be drawn from this quote: (a) First, it is the context of utterance that determines how a situation referred to by a particular sentence is conceptualized or understood. Comrie's example is even more extreme than our "chicken" and "cigarette" examples because the object NP is understood and not explicitly mentioned. Still, given the appropriate evidence (i.e., context) available, the speaker may be induced to conceptualize He sang as He sang a song. (b) Secondly, it is the clear (i.e., mutually manifest) presence of the intention to perform a particular song or to reach a clearly stipulated endpoint (realized by an elided NP in this case) that establishes the telic reading. (cf. e.g., also Dowty 1979: 61; Olsen 1994: 364)

To sum up, the hypothesis defended is that a sentence with a numerical NP that is affected, consumed or constructed theme is relic only if it is mutually manifest that the numerical NP is an inherent endpoint. This means that it is impossible to determine, out of context, whether a NP like 4 cigarettes or 1 cigarette (that functions as affected, constructed or consumed theme) will establish telicity or not. In other words, we do not take sides with e.g., Krifka (1992), Ramchand (1997), and Smith (1997: 29) when they claim that sentences with numerical NPs are straightforwardly telic.

Verkuyl (1993) also refers to transitive sentences with multiple numerical object NPs, but the question of whether such sentences are telic or not is not raised, because of the way in which "terminative aspect" is defined: it simply implies reference to a boundary (inherent in the verb or realized by a subject or an object NP) and accordingly, the nonprogressive variants of the sentences discussed so far all have terminative aspect. Verkuyl also deals with progressive examples of the type He is drawing three circles (1993: 318-327); the acceptability of such sentences is taken for granted, so it seems. On various occasions (cf. esp. 1993: 36-38), he insists that considerations about agentivity should not be confused with Continuous Tense Criteria to classify situations: "Vendler's criterion Progressive Form (23) turns out to be actually focused on some unclear concept of agentivity" (1993: 39). (24) Applying that remark to what precedes, it might be objected that a similar mistake is made here, as my examples might be said to point to a constraint on the use of the progressive, rather than pertain to a characteristic of telic sentences. However, it should be clear that we do not use the incompatibility with the progressive as a diagnostic to distinguish telic and atelic sentences. We start from the observation that telic dynamic transitive verb sentences are compatible with the progressive (in its ongoing event convention), and that the presence of a multiple numerical NP is usually taken as a sign that a sentence is telic, from which it follows that such sentences should be compatible with the progressive. However, it appears that the progressive cannot be used in certain numerical NP sentences (usually classified as telic sentences), which proves that the boundary the NP establishes is not always an inherent endpoint. In other words, what we refute is that the mere presence of such a NP can be used as a diagnostic test to determine the Aktionsart category to which a sentence belongs. (25) Verkuyl need not explain the constraint on the progressive, as he is not faced with it: he takes it for granted that all examples of the type John ate three cakes are compatible with the progressive. We readily accept Verkuyl's claim that John ate three cakes is terminative and John was eating three cakes is not. What we are trying to account for though is why it is not always possible to add the progressive marker to such sentences in order to refer to the "middle of the situation" and as far as I can see, Verkuyl's account does not provide us with the necessary tools to explain the constraint observed. If one does away with distinction between Aktionsart classes and aspect, and replaces them by aspectuality, it seems to me that one rids oneself of the key that can help us to solve the riddle: it is the presence or absence of an inherent endpoint (and hence telicity) that is at the basis of the constraint observed. It will be clear that the explanatory relevance of the distinction between (un)boundedness and (a)telicity comes again to the fore. (cf. e.g., Depraetere 1995; cf. also Declerck 2006)

In recent research on agentivity and telicity (and unaccusativity), it has been argued that agentivity is orthogonal to telicity (cf. e.g., Dowty 1991; Levin and Rappaport Hovav 1996: 299). Given that intentionality characterizes the prototypical agent, one may wonder whether there is no contradiction between the hypothesis defended here (in which intentionality brings about telicity) and the claim that nonagentive as well as agentive situations may be telic. There is no contradiction involved, and this becomes clear when we pin down the common ground between the topics involved in the discussions:

--Agentive as well as nonagentive situations may be telic. We do not challenge that hypothesis.

--In the semantic role literature, the question whether all agents are [+intentional] or [+volitional] has been dealt with and it has led e.g., Van Valin and Wilkins (1999) to distinguish between agents and effectors. These discussions are focused on the link between the extralinguistic category of intentionality (and control) and a semantic role. We have presented some of the approaches, but we have not challenged any particular claim.

--We have looked at dynamic sentences (with human subjects) with numerical NPs. The presence of such a constituent is commonly thought to be indicative of telicity. We have challenged that point of view and given evidence that shows that in the examples given, it is only if the numerical NP is in the "scope" of an intention that the sentence is telic. Whether that means that the Subject performs the role of Agent more prototypically in such a case is interesting, but ultimately not the key question we aim to address.

In other words, we do not aim to refine the definition of agentivity, we do not generalize in the sense that intentionality is a conditio sine qua non for telicity; rather, we have focused on ways in which the inherent endpoint associated with telicity may be identified. Our discussion is meant to show that certain types of quantified constituents (numerical NPs, and for adverbials (cf. Section 3) in particular contexts) do not automatically result in telicity. In the examples discussed here, the constituents in question will be associated with the function of an inherent endpoint only if they are in the "scope" of a mutually manifest intention. There is therefore no contradiction between the claim that agentivity is orthogonal to telicity and the hypothesis that in certain contexts, intentionality determines telicity.

As pointed out above, the mutual manifestness of an inherent endpoint to the situation (be there reference to a "volitional" agentive, a "nonvolitional" agentive (26) or a nonagentive situation) is crucial to the definition of telicity. This inherent endpoint may be mutually manifest in a number of ways. To indicate just a few:

(a) Sometimes, the lexical semantics of a verb (e.g., find, sneeze, explode) are sufficient to allow the conceptualization of an inherent endpoint.

(b) In other cases, a PP may refer to an inherent boundary (He drove the car into the garage).

(c) To take up Comrie's example again, the contextual knowledge shared by a number of discourse participants may be such that the inherent endpoint is mutually manifest without it being explicitly mentioned: he sang may be understood as he sang a song.

(d) In the case of dynamic transitive verbs subcategorized for an object NP that is affected, constructed or consumed theme, the object NP is likely to assume the function of inherent endpoint fairly automatically, provided there is reference to the default situation in which the definite article NP or indefinite article NP is contextually given as the natural endpoint to the situation (e.g., drink a glass of wine, read the book).

(e) NPs with cardinal numbers by their very nature (number can only be determined once the situation has come to an end) are less likely to be "contextually given", which explains why they can only function as inherent endpoints when they are in the "scope" of a mutually manifest intention.

In other words, rather than challenging the claim that agentivity is orthogonal to telicity, the argumentation developed here serves to underline the importance of the visibility of a mutually manifest endpoint and the part played in this process by extralinguistic factors such as intentionality.

3. Intentionality, telicity and for-adverbials

As pointed out in the introduction, the compatibility of a clause with either a for adverbial or an in adverbial is often used as a diagnostic for (a)telicity/(un)boundedness (John drank a glass of wine in 30 seconds (John drank a glass of wine = telic) vs. John drank wine for 30 seconds (John drank wine = atelic)). (27) However, the question of how to classify base cases with a for adverbial has received less attention. Given that the quantified PP in e.g., He worked in the garden for five hours or in She was in the garden for three hours puts a temporal boundary to the situation of working, it may hypothesized that sentences like these are likely to be considered as telic/bounded by most researchers. Taking into account the previous discussion, it will be clear that we believe such sentences are telic only if the adverbial is in the "scope" of a mutually manifest intention. In other words, these sentences also show that intentionality can be decisive in establishing (a)telicity.

I will again use the "ongoing event progressive" as a test: if all nonprogressive (dynamic) sentences with a for adverbial are telic, it should always be possible to add the progressive marker (given an appropriate context in which the use of the progressive is justified), since dynamic telic sentences are compatible with the progressive. However, in dynamic examples with durational for adverbials, the progressive as a marker that focuses on the middle of the situation can only be added if the adverbial is in the "scope" of the intention. In other words, we can draw a similar conclusion to that above: the for adverbial does not automatically constitute an inherent endpoint and therefore, a sentence with a for adverbial is not automatically telic. (28) Let us take the sentence in (11) as a starting point:

(11) (said by John) I marked exam papers for two hours.

I would like to argue that this sentence is not necessarily telic (or derived telic [Smith 1997]). It is certainly bounded, in the sense that there is a factual endpoint to the situation of correcting, which is marked by the adverbial. Whether the situation is telic or not depends on whether or not the boundary is contextually given as an intended (and therefore inherent) endpoint, which here boils down to determining whether or not it was John's intention to mark papers for two hours. If John has to mark 500 exam papers during the Easter break, it is not difficult to imagine a situation in which he decides to make that task digestible by sacrificing a number of hours every day. On that interpretation, we could envisage the following utterance:

(12) I ended up with a pile of 500 papers to correct. I wanted to get this marking business over and done with as soon as possible so I decided to mark copies for two hours every day during the Easter break. The first day I that was marking exam papers for 2 hours, my son came in and asked whether I wanted to play cards with him. (intentional reading)

In this example, for two hours is the intentional endpoint that has not been reached yet when John's son enters the room. In (13), on the other hand, the adverbial is a temporal boundary but not an inherent endpoint. The presence of, on the one hand, a constituent that establishes a temporal boundary (the adverbial) and, on the other hand, the progressive marker, which typically leaves beginning and endpoints out of focus and therefore has an unbounding effect, results in unacceptability:

(13) I ended up with a pile of 500 copies to correct. I wanted to get this marking business over and done with so sat down immediately. *I was marking exam papers for 10 hours on end when Mary came in and asked whether I wanted to go out for coffee. (nonintentional reading)

A pseudocleft construction brings out the difference between the two interpretations: (29)

(14) telic interpretation (11): What he did was: mark exam papers for two hours.

atelic interpretation (12): What he did for ten hours was: mark exam papers.

These examples strengthen the hypothesis formulated earlier on about the importance of intentionality for telicity: sentences with durational for adverbials only result in telicity if it is mutually manifest that the adverbial is in the "scope" of the intention. In Smith's terminology (1997:113), we might say that the adverbial is an inner adverbial in (11), whereas it is an outer adverbial in (12). (30)

Examples with adverbials are particularly helpful to prove that endpoints should not automatically be seen as inherent endpoints and are therefore not automatically markers of telicity. The reason for this is probably that intuitively, we are more inclined to agree that the duration of a situation can only be determined once the situation is over is posterior to the situation, unless we decide on its duration before the situation starts. From this it follows that out of context we cannot tell whether the presence of an adverbial like for three hours establishes a telic reading or not.

4. Countering counterexamples

In this section, I would still like to mention some examples, which, at first sight, appear to refute the hypothesis that it is impossible to combine a constituent (e.g., a for adverbial or a numerical NP) whose sole function it is to establish a factual boundary with an (ongoing event) progressive marker, which by default establishes an unbounded reading. Let us start with an example with a for adverbial:

(15) A: Why is daddy so cranky?

B: He marked exam gapers for two hours yesterday. The students' bad marks make him feel depressed. (atelic and bounded)

Let us assume that in this example, for two hours is not in the "scope" of the intention and that He marked exam papers for two hours is atelic and bounded. (As a matter of fact, we do not have sufficient contextual information in [15] to make that decision). So far, example (15) does not challenge the hypothesis that a sentence with a for-adverbial is not automatically telic. The fact that this bounded sentence is compatible with a progressive marker is somewhat more problematic:

(16) A: Why is daddy so cranky?

B: He was marking exam papers for two hours yesterday. The students' bad marks make him feel depressed.

Example (16) appears to challenge the other hypothesis we put forward, namely that it is impossible to add a progressive marker to a bounded atelic sentence, i.e., to a sentence with a for adverbial whose sole function it is to bound the situation (cf. [13]), as there will be a clash between the progressive marker, which brings about unboundedness, and the adverbial, which establishes a temporal boundary. However, as (16) shows, the addition of the progressive does not result in unacceptability. Our hypothesis is not falsified: the example in (16) is indeed different (from those in [12] and [13]) because there is reference to the whole period of two hours of correcting. In (12) and (13), the progressive is used to focus on the middle of the situation (of correcting for a number of hours), and the beginning and the end of the situation are "off screen", as Jackendoff (1996) puts it. This is not the case in (16). The progressive is not used in its ongoing event convention (Smith 1997: 89), but rather to stress the fact that for a full two hours, the person in question was marking papers. As the progressive does not have its usual unbounding effect, there is no clash with the for adverbial. I would argue that the sentence in (16) is still bounded, even though the progressive is used. The adverbial makes the situation of marking exam papers bounded and constitutes the variable that is filled by a value (a situation). Put more informally, a kind of frame is established by for two hours, a frame that is filled by the activity of correcting: no matter whether we use the progressive or not, the sentence remains bounded. (cf. Depraetere and Reed 2000:112-113) This brings up the issue of the hierarchy among the factors that contribute towards establishing (un)boundedness and the interaction between them. (31)

The following set of examples also seems to question the hypothesis defended: (32)

(17) Unintentionally John is killing five chickens by putting rat poison round the farmyard.

(18) (context: John has begun to clear out the empty boxes in the building.) John is unknowingly driving away three families of pigeons by his spring cleaning.

(19) (context: John is nervously smoking without noticing what he's doing.) I say, "You are smoking two packs a day, aren't you?"

These examples combine markers that I have argued are incompatible: an adverb that clearly shows there is no intention involved (unintentionally, unknowingly), the progressive (ongoing event convention) and a numerical NP. However, I do not believe they falsify my hypothesis:

(a) The sentence in (17) is not judged as 100% acceptable by the native speakers that I consulted. Indeed, in the context given, it is clear that John intends to kill (i.e., murder) animals by putting poison (admittedly, it is the wrong ones that get killed). The fact of adding unintentionally is therefore likely to result in a semantic contradiction. Another reason why the acceptability is questionable is that in (17), there is insufficient contextual information available that leads to the identification of a mutually manifest inherent endpoint that warrants the assertion that exactly five chickens will be killed. One suggestion by an informant to change the sentence into Unintentionally John is killing these five chickens by putting rat poison round the farmyard to make it acceptable, is rather telling. The addition of these serves to identify a contextually given boundary (i.e., it is no longer information that is retrieved once the killing process is over), in which case the progressive can more easily be added.

(b) A first observation on the example in (18) is that there is not necessarily reference to a sequence of events: the situations of the families of pigeons being driven away are likely to overlap, while in the examples discussed in Section 2, the referents are involved separately (one after the other) in what constitutes a sequence (cf. Note 17). Moreover, as has been pointed out already, while intentionality can be one way of making an endpoint mutually manifest as an inherent endpoint, other kinds of contextual information may also perform that function. For instance, we can envisage a context in which it is mutually manifest to the discourse participants that there live three families of pigeons in the house and that the enthusiasm with which John is cleaning is such that any animal living in the house will be driven out. While "3 families of pigeons" is not in the "scope" of the subject's intention, it is all the same given as an inherent endpoint. (33)

It is inherent in the nature of contextual information that it is difficult to make generalizations about the way in which it may make inherent endpoints mutually manifest. This observation as such does not falsify the claims we have made in this article; the examples discussed definitely show that intentionality is one, but not necessarily the only contextual player when it comes to establishing mutually manifest inherent endpoints. (34) In the following telic nonagentive example, for instance, the inherent endpoint cannot--by the very nature of the subject--be intentional, but it all the same has to be contextually given in order for the progressive to be used felicitously, e.g., it might be uttered while someone is looking at a thermometer:

(20) The lake is cooling four degrees. (Hay et al. 1999: 133)

(c) The sentence in (19) is neither problematic for our hypothesis. In this example, the progressive is not used to refer to the middle stage of the process of smoking two packs on a particular occasion, but rather to the recurrent situation or (temporary) habit of smoking two packs a day. If we change a day into today, i.e., if there is reference to a single situation, it seems that the sentence is only acceptable if it is understood as some kind of ironic remark on the incurable behavior of a cigarette addict to whom the speaker attributes an intention to smoke two packs.

5. Conclusion

One conclusion to the discussion is that it is important to bear in mind the true nature of the telos, i.e., the "inherent" endpoint originally associated with telicity: an endpoint is inherent only if it can be conceptualized as being the "natural" endpoint of the situation referred to by the discourse participants, i.e., when it is mutually manifest. When an affected, constructed or consumed Direct Object theme has a single referent, (35) the identification of this NP as the potential built-in endpoint is more automatic than when there is more than one referent involved in the form of a multiple numerical NP (or when a singular referent is referred to in a single numerical NP). In the latter case, the speaker requires more extensive evidence in order to accept the fact that, as soon as the situation has started, the numerical NP constitutes an inherent endpoint. (A similar remark applies to sentences with a for adverbial.) Explicit reference to an intention may contribute to the identification of a mutually manifest endpoint. In other words, the presence of a numeral NP or a for PP does not automatically imply that the sentence is telic; the sentences will only be recognized as such if there is sufficient contextual information available that identifies the NP or the PP as a mutually manifest inherent endpoint. This approach has explanatory value: it helps to explain why a progressive marker in its ongoing event convention can only be added to a dynamic sentence if the multiple numerical NP or the for adverbial does not only bound the situation but also makes it telic. The discussion confirms the view expressed in Depraetere (1995) that it is necessary to distinguish (a)telicity and (un)boundedness, because they have different roles to play in explaining aspectual and temporal phenomena.

UMR 8163 STL, Lille III University

Received 25 September 2003

Revised version received

31 January 2005

Notes

(1.) I am very grateful to Beth Levin, Christopher Pinon and Susan Reed, for commenting on one of the earlier versions of the text. I also wish to thank the two anonymous referees for their observations. Thanks to H. Borer, S. Engelberg, H. Filip, B. Levin, F. Martin and M. Lemmens for directing my attention to particularly relevant publications. Correspondence address: Lille III, UFR des LEA, 14 Place Bodart Timal, 59058 Roubaix, France. E-mail: ilse.depraetere@univ-lille3.fr.

(2.) Cf. Dahl (1981) and Filip (1999: 53) for a good survey of the labels used to refer to this Aktionsart distinction.

(3.) Anticipating the discussion about (un)boundedness and (a)telicity in Section 2.1 and 2.2, it is interesting to add that Binnick points out one should not confuse "effective (coincidental) boundedness" and "inherent boundedness" (1991: 191).

(4.) I will not attempt to summarize previous research about these aspects: it is vast and it is therefore impossible to be do justice to the findings in a summary. I recognize the important role that is played by semantics and syntax, but I will focus on the pragmatic level in this article.

(5.) "To be manifest, then, is to be perceptible or inferable" (Sperber and Wilson 1995: 39).

(6.) Cf. Depraetere (1995) for further evidence for the necessity of distinguishing between (a)telicity and (un)boundedness. Cf. Depraetere (2000) for a survey of the linguistic markers that affect the representation of a situation as (un)bounded and (a)telic. Declerck (2006) also provides a detailed description of the difference that is at stake.

(7.) Filip (2000: 4) explicitly points out that she uses the term telic to refer to "all verbal predicates that entail some delimitation in their semantic structure, regardless of its nature and regardless whether they have animate or inanimate, human or non-human subjects." [emphasis mine] and in doing so opts for an approach that is different from that in e.g., Garey (1957) and Depraetere (1995), whose definitions are closer to the etymological sense of the word (2000: 3-4).

(8.) It has often been pointed out (cf. e.g., Dahl 1981; Declerck 1989; Smith 1997) that (a)telicity is a notion that applies to representations of situations and not to situations as such. For instance, an extralinguistic situation of writing a letter may be referred to by means of the atelic She's writing as well as by the telic She's writing a letter. For convenience, I will use both the label "telic situation" or "telic predicate" to refer what in more accurate terms is "the telic representation of the situation in a predicate". In other words, I do not follow Vetters (1996), for instance, who argues that situation types should not be looked upon as linguistic categories; they are metaphysical or extralinguistic categories. His conclusion is based on the observation that none of the tests traditionally used to distinguish situation types are fully waterproof: "ces categories sont independantes des tests heuristiques traditionellement proposes" [these categories are independent of the heuristic tests traditionally used]. The fact that syntactic tests do not always provide clear indications for Aktionsart class does not justify assigning them a metaphysical or extralinguistic status. As Gosselin and Francois (1991) point out, "une predication est en effet le correlatif linguistique d'une conceptualisation de proces, ce qui implique qu'il y a une interpretation de le realite exterieure" [a predicate is indeed the linguistic correlate of the conceptualisation of a situation, which implies that there is an interpretation of reality]. Situation types are indeed a categorisation of extralinguistic information, but it is reality moulded into a linguistic form. From that point of view, we are not concerned with a classification of extralinguistic situations as such, but with an interpretation of situations as they are perceived and expressed by means of a linguistic form by a particular speaker.

(9.) Even though it is not the purpose to examine the distinction between (un)boundedness and (a)telicity in detail here, I would still like to add that the use of a past tense does not automatically imply that the situation is bounded. While a past tense sentence that refers to a state (She was in the kitchen a minute ago.) may give rise to an implicature that the situation referred to is no longer the case at the moment of speaking, the semantics of the past tense as such do not explicitly communicate that the situation is bounded. Cf. Depraetere (1996b) for a detailed discussion of this issue.

(10.) Obviously, it is impossible to summarize this field of research (cf. e.g., Van Valin and Wilkins 1999; for an excellent survey of approaches to agentivity, cf. also Nishimura 1993). I will highlight a few findings that are particularly relevant to the topic under discussion, but it will become clear toward the end of Section 2.2 that it is not the category of agentivity as such that needs to be refined in order to account for the data mentioned in this article or to explain why intentionality is important for telicity.

(11.) Note that we understand volition and intentionality in the way they are defined by Van Valin and Wilkins (1999). Volition is a "property of entities which manifest nonconscious basic acts of will (such as a baby crying for milk)", and intention is a property "which requires consciousness of wills, and ability to plan" (1999: 313). This distinction is not always made in the literature: volition is sometimes used as a synonym of control or intention. When this is the case, "volition" has been put in inverted commas.

(12.) Van Valin and LaPolla (1997: 95) point out that, when applying this diagnostic test for Achievement verbs, it is "crucial to avoid adverbs which require a controlling subject (...) In selecting adverbs for this test, it is necessary to test their compatibility with involuntary verbs like shiver and with verbs like shake which can have an inanimate subject".

(13.) Note that the verbs mentioned in Section 2.2 are either intransitive (sneeze, yawn, etc.), transitive (e.g., find) or pseudotransitive (e.g., win).

(14.) Note that Levin's class also includes verbs with inanimate subjects.

(15.) At first sight, this sentence might seem to be suggesting that I am talking about a classification of extralinguistic situations (whose endings are or are not mutually manifest to the discourse participants) and am therefore contradicting the point made in Note 8, i.e., telicity is a notion that applies to linguistic expressions that refer to situations. The contradiction is only apparent. Having, for instance, decided to refer to a situation of singing a song by means of the utterance She's singing (cf. Comrie's example (1976: 46) in Section 2.2), it has to be mutually manifest to the discourse participants that "sing" in this particular context means sing a song and is therefore telic.

(16.) Note that scope is used in the nontechnical sense in this text.

(17.) I will focus on the sequential interpretation of the numerical NP (one after the other) and disregard the other possible reading in which the referents are simultaneously (either literally or not) involved in one situation.

(18.) Cf. Jayez (1999: 151) for a similar point.

(19.) Admittedly, this is a rather informal characterization of the progressive. For the purposes of this article, however, we indeed need only appeal to the characterizing feature of the progressive commonly captured by means of the label ongoingness. For more elaborate treatments of the progressive, cf. e.g., Bruyndonx (2001); Lascarides (1991); Mittwoch (1988); Williams (2002).

(20.) It will be clear that the indefinite article and the definite article are used in different contexts, but that need not concern us here.

(21.) In his 1977 "imperfective paradox" article, Dowty (1977: 59) writes that agents normally have an intention to produce a certain result when they begin their actions. He points out that there is a problem in the sense that agents are not always certain as to their exact goals. His example is that of John starting to draw, being undecided as to whether it is a unicorn or a horse that he is going to draw. For Dowty's analysis of the progressive, this is problematic, as it suggests that both John is drawing a horse and John is drawing a unicorn are true at the same moment. Smith (1997: 26) also touches upon the problem that people may change their mind while they are doing something or that we may misinterpret situations: "In such cases the speaker makes a judgment about the kind of situation that is going on, including the intentions of participants" (1997: 26) (cf. also Smith 1997: 83). However, while admitting that we may misjudge the situation, it remains a fact that we "read people's minds" by attributing particular intentions to them. In doing so we are constrained by what we see but whether the descriptions are ultimately true or false is not important in this context (cf. also e.g., Declerck 1979: 272).

(22.) Cf. e.g., Engelberg (2002) and Naumann and Pinon (1997) for observations on intention and the progressive.

(23.) There is reference to the syntactic test Vendler (1967) uses to distinguish on the one hand, Activities and Accomplishments, which are compatible with the progressive, and States and Achievements, on the other, which are not. Verkuyl (1993:35 38) lists a set of examples that falsify that hypothesis.

(24.) Rather challengingly, he goes on: "Vendler seems to follow here the linguistic tradition which coined the term "Aktionsarten" for the phenomena at issue: the ways actions are conceived of. This term itself suggests agentivity plays an important role. However, important for what? Human actions are extremely important for philosophers. But are they linguistically important as well? The answer should be negative, I think. It is a rather bad habit of linguists to choose mainly sentences pertaining to animate beings such as John and Mary, or Jack and Jill, at the expense of nonanimacy. However, the concept of agentivity may overlap greatly with the concept of processes ongoing in time, because most of the time we speak about human actions. But this does not mean at all that these concepts can be equated" (1993: 39-40).

(25.) The view that a multiple numerical Object NP automatically makes a situation telic seems indeed to be often taken for granted in the aspectual literature. Previous discussions of sentences with such quantifiers tend to be focused on the possible interpretations of e.g., He knocked on the door three times ('one knock on the door on different occasions or three knocks on one single occasion') (cf. e.g., Mourelatos 1978: 205, 209) or of even more complex cases like Three boys lifted four boxes (cf. e.g., Jackendoff 1996: 344; Verkuyl 2001:10-11).

(26.) Van Valin and Wilkins (1999) would qualify the role as that of effector rather than agent if there is no "volition" involved.

(27.) It might be objected that the hypothesis defended is problematic because the widely accepted diagnostic in/for test does not appear to yield adequate results. It is well-known that telic predicates are compatible with in adverbials (I read the book in two days/*for two days), while atelic predicates typically combine with for adverbials (He read for two days/*in two days). Applied to the hypothesis defended here, John smoked 10 cigarettes, on its atelic reading should not be compatible with an in adverbial, which it is all the same: John smoked 10 cigarettes in five minutes is acceptable both on the telic (bounded) reading and on the atelic (bounded) reading. In other words, the in/for test predicts that John smoked 10 cigarettes is a telic predicate on any (i.e., intentional or not) reading. A first observation is that the for/in test applied to progressive telic predicates also fails to produce adequate results: while everybody would agree that eat an apple is telic, this does not appear to follow from ??I am eating an apple in ten minutes. The in/for test is also problematic when applied to punctual predicates (cf. e.g., Moens 1987). Obviously, the examples discussed in this article are not the only cases in which the results of the in/for test are not in line with the unmarked Aktionsart interpretation (cf. e.g., Depraetere [1996c: 17-24] for a survey of diagnostic Aktionsart tests and their limitations). Therefore, it is not unlikely that just as the questionability of ??I am eating an apple in ten minutes does not invalidate the claim that eat an apple is telic, the acceptability of John smoked 10 cigarettes' in five minutes does not necessarily invalidate the hypothesis defended in this article (i.e., John smoked 10 cigarettes is not necessarily telic). More in general, it might also be asked whether the in/for test is not a diagnostic for (un)boundedness ("actualization aspect") rather than (a)telicity, which is an issue that is beyond the scope of this article.

(28.) As regards the effect of for adverbials on Aktionsart classes, Smith (1997) is (to my knowledge) the lonely defender of the idea that depending on their status of inner or outer adverbial, they will or will respectively not result in telicity. Smith distinguishes between natural endpoints (telic situations) and arbitrary endpoints (atelic situations). The Principle of External Override is used to explain why, for instance, the addition of a for adverbial turns an Activity (He played soccer) into a telic sentence (He played soccer for three years) (1997: 23-24). (cf. also Michaelis 1998 for a similar point of view) As I see it, it is a bit of a misnomer to say that He played soccer for three years is telic, because, given no evidence to the contrary, the endpoint is arbitrary and therefore does not automatically answer the criterion of inherent endpoint associated with a telic sentence. Admittedly, Smith would use the term derived Accomplishment to refer to the sentence with the adverbial, which makes it clear that it is not a "straightforward" telic example (which presupposes a change of state in Smith's framework), but it seems to me that two distinct notions are conflated all the same.

(29.) Verkuyl (1993: 276) uses the pseudocleft construction in a different, "events and intervals" context, i.e., to bring out the readings of He hit Miles six times. What he did was hit Miles six times, implies that there "is one event in which there are subevents (most probably six)"; What he did six times was hit Miles means that there are six events. Verkuyl (1993: 14) explicitly states that "in the absence of a sufficiently articulated theory of adverbial modification" he wishes to restrict himself to the composition of inner aspect, i.e., the relationship between a verb and its arguments. Accordingly, sentences with for adverbials that are arguments of the verb (cf. e.g., [11]) are not in the scope of his discussion.

(30.) Smith (1997:114) writes: "'Mary went to Paris for three days for three weeks' may be paraphrased as: Mary went to Paris with the plan or intention of staying for three days, but actually stayed for three weeks. The three-week duration is an intensional property of the idealized situation, independent of its actual unfolding in time. In other words, the inner adverbial specifies a property of the situation type. The outer adverbial gives the temporal location of the situation. (...) On the basis of this (Mary slept for three hours) and many other examples, we conclude that a single adverbial of duration does not indicate the desired or intended length of a situation. Thus single durative adverbials have only the external interpretation. However, they must be compatible with the telic/atelic value of the situation type, which involves its internal structure." I am not quite sure whether this claim is compatible with Smith's suggestion (cf. Note 28) that a sentence like He played soccer for three years is derived telic in the sense that the adverbial turns the Activity into an Accomplishment. This seems hard to match with the stipulation that "single durative adverbials have only the external interpretation" and that the adverbial should be "compatible with the telic/atelic value of the sentence", as the role performed by the adverbial is exactly that of changing the Aktionsart. Smith also points out that Adjectives in Adjectival Phrases of the type Mary took a three-hour walk are of the inherent type as well; they indicate "the planned or intended length of the event" (1997: 121).

(31.) Cf. Depraetere (2000: esp. 29-30) for modest initiatives to that effect. It will be clear that the interaction between the different operators enter into a so-called compositional approach to "aspectuality" (cf. e.g., Verkuyl 1993).

(32.) I am very grateful to Liliane Haegeman, Ruth Huart, John Osborne, Susan Reed, Christopher Williams, for discussing these examples with me. My comments have been largely inspired by their observations, but it is clear that they do not hold any responsibility for possible flaws in the argumentation.

(33.) C. Williams points out that both (17) and (18) become more easily acceptable when couched in a narrative with an omniscient narrator who knows in advance the exact number of chickens or families of pigeons affected by the (completed) action that is still in progress when it is being described. In other words, the so-called counterexamples show that while intentionality may be an important factor that helps to establish inherent endpoints, other kinds of contextual information may also perform that function. The outcome is in any case the same: the numerical NP is understood as the built-in endpoint.

(34.) Hay et al. (1999) point out that real-world knowledge can help us decide whether a verb like lengthen is telic (as in The tailor almost lengthened my pants) or atelic (as in The teacher almost lengthened the exam.) In their article, the authors show that "the telicity of a particular DA [degree achievement] can be, and often is, derived through a process of conversational implicature, indicating that the aspectual properties of a particular predicate often cannot be completely specified in terms of formal (semantic or syntactic) features" (1999: 129). This is one of relatively few texts in which the level of pragmatics is explicitly integrated into the discussion of aspectuality ("context-dependent telicity" [1999: 136]) (cf. also Olsen 1994). Although the authors seem to be using telicity in the sense of what ! would call boundedness and focus on degree achievements, their account seems to point in the same direction: "the discussion of the role of contextual/pragmatic information in determining telicity is particularly important, as it indicates that (a)telicity cannot be fully specified by the linguistic form, either semantically, or, as has more recently been argued, syntactically" (1999: 138).

(35.) That is referred to by means of a singular indefinite or definite (rather than a numerical) NP.

References

Andersson, S. G. (1972). Aktionalitat im Deutschen: Eine Untersuchung unter Vergleich mit dem Russischen Aspektsystem. Uppsala: Acta Universitatis Upsalienis.

Bach, Emmon (1981). On time, tense, and aspect: an essay in English metaphysics. In Radical Pragmatics, P. Cole (ed.), 63-81. New York: Academic Press.

Binnick, Robert I. (1991). Time and the Verb. New York and Oxford: Oxford University Press.

Bruyndonx, Jim (2001). The expanded form in British English: meanings and constraints: a corpus-illustrated description. Unpublished doctoral dissertation, Catholic University of Louvain.

Comrie, Bernard (1976). Aspect. Cambridge: Cambridge University Press.

Couper-Kuhlen, Elisabeth (1987). Foregrounding and temporal relations in narrative discourse. In Essays on Tensing in English, vol. 1 : Reference Time, Tense and Adverbs, Alfred Schopf (ed.), 6-29. Tubingen: Niemeyer.

Cruse, R. A. (1973). Some thoughts on agentivity. Journal of Linguistics 9, 11-23.

Dahl, Osten (1981). On the definition of the telic-atelic (bounded-non-bounded) distinction. In Syntax and Semantics 14: Tense and Aspect, P. J. Tedeschi and A. Zaenen. (eds.), 7990. New York: Academic Press.

Declerck, Renaat (1979). The progressive and the imperfective paradox. Linguistics and Philosophy 3, 267-272.

--(1989). Boundedness and the structure of situations. Leuvense Bijdragen 78, 275 304.

--(1991). Tense in English." Its Structure and Use in Discourse. London and New York: Routledge.

--(2006). The Grammar of the English Verb Phrase, vol. 1: The Grammar of the English Tense System." A Comprehensive Analysis, in collaboration with Susan Reed and Bert Cappelle. Berlin and New York: Mouton de Gruyter.

Depraetere, Ilse (1995). On the necessity of distinguishing between (un)boundedness and (a)telicity. Linguistics and Philosophy 18, 1-19.

--(1996a). Foregrounding in relative clauses. Linguistics 34, 699-731.

--(1996b). Some observations on the likelihood of past tense situations still being the case at the time of speaking. ITL Review of Applied Linguistics 113-114, 335-348.

--(1996c). The Tense System in English Relative Clauses. A Corpus-Based Analysis. Berlin and New York: Mouton de Gruyter.

--(2000). Bornage, temps et aspect. Document de synthese (HDR). Lille III.

--and Reed, Susan (2000). The present perfect progressive: constraints on its use with numerical object NPs. English Language and Linguistics 4(1), 97-114.

Dowty, David R. (1977). Toward a semantic analysis of verb aspect and the English 'imperfective' progressive. Linguistics and Philosophy 1, 45-77.

--(1979). Word Meaning and Montague Grammar." The Semantics of Verbs and Times' in Generative Semantics' and in Montague's PTQ. Dordrecht: Reidel.

--(1986). The effect of aspectual class on the temporal structure of discourse: pragmatics or semantics? Linguistics and Philosophy 9, 37-61.

--(1991). Thematic proto-roles and argument selection. Language 67, 547-619.

Engelberg, Stephan (2002). The semantics of the progressive. In Proceedings of the 2001 Conference of the Australian Linguistic Society, C. Allen (ed.), 1-8. Canberra: Australian National University.

Filip, Hana (1997). Integrating telicity, aspect and NP semantics: the role of thematic structure. In Formal Approaches to Slavic Linguistics, vol. 3: The College Park Meeting 1994, Jindrich Toman (ed.), 61-96. Ann Arbor: Michigan Slavic Publications.

--(1999). Aspect, Eventuality Types and Nominal Reference. New York and London: Garland.

--(2000). The quantization puzzle. In Events as Grammatical Objects, From the Combined Perspectives of Lexical Semantics, Logical Semantics and Syntax, James Pustejovsky and Carol Tenny (eds.), 3-60. Stanford: CSLI.

Garey, Howard (1957). Verbal aspect in French. Language 33, 91-110.

Gosselin, Laurent and Francois, Jacques (1991). Les types de proces: des verbes aux predications. Travaux de linguistique et philologie 29, 19-86.

Haegeman, Liliane (1994). An Introduction to Government and Binding Theory, 2nd ed. Oxford: Blackwell.

Hamann, Cornelia (1991). Semantics and pragmatics--the case of temporal conjunctions. Linguistische Berichte 136, 403-437.

Hay, Jennifer; Kennedy, Christopher; and Levin, Beth (1999). Scalar structure underlies telicity in 'degree achievements'. In SALT IX, Tanya Matthews and Devon Strolovitch (eds.), 127-144. Ithaca, NY: CLC Publications.

Hinrichs, Erhard W. (1986). Temporal anaphora in discourses of English. Linguistics and Philosophy 9, 63-82.

Holisky, Dee Ann (1987). The case of the intransitive subject in Tsova-Tush (Batsbi). Lingua 71, 103-132.

Huddleston, Rodney and Pullum, Geoffrey (2002). The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press.

Jackendoff, Ray (1996). The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14, 305-354.

Jayez, Jacques (1999). Imperfectivity and progressivity: the French imparfait. In SALT IX, Tanya Matthews and Devon Strolovitch (eds.), 145-162. Ithaca, NY: CLC Publications.

Krifka, Manfred (1992). Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, Ivan Sag and Anna Szabolcsi (eds.), 29-53. Stanford: CSLI.

Lakoff, George (1977). Linguistic gestalts. In Papers from the Thirteenth Regional Meeting, Chicago Linguistic Society, W. A. Beach, S. E. Fox, and S. Philosoph (eds.), 236-287. Chicago: Chicago Linguistic Society.

Lascarides, Alex (1991). The progressive and the imperfective paradox. Synthese 87, 401-447.

Levin, Beth (2000). Aspect, lexical semantic representation and argument expression. In Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society, L. J. Conathan, J. Good, D. Kavitskaya, A. B. Wulf, and A. C. L. Yu (eds.), 413-429. Berkeley: University of California, Berkeley.

-- and Rappaport Hovav, Malka (1996). Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge, MA and London: MIT Press.

Michaelis, Laura (1998). Aspectual Grammar and Past-Time Reference. London and New York: Routledge.

Mittwoch, Anita (1988). Aspect of English aspect: on the interaction of perfect, progressive and durational phrases. Linguistics and Philosophy 11,203-254.

Moens, Marc. (1987). Tense, aspect and temporal reference. Unpublished doctoral dissertation, University of Edinburgh.

Mourelatos, Alexander P. D. (1978). Events, processes and states. In Syntax and Semantics, vol. 14: Tense and Aspect, Philip J. Tedeschi and Annie Zaenen (eds.), 191-212. New York: Academic Press.

Naumann, R. and Pinon, C. (1997). Decomposing the progressive. In Proceedings of the 11th Amsterdam Colloquium, Paul Dekker, Martin Stokhof, and Yde Venema (eds.), 241-245. Amsterdam: ILLC/Department of Philosophy.

Nishimura, Yoshiki (1993). Agentivity in cognitive grammar. In Conceptualizations and Mental Processing in Language, Richard A. Geiger and Brygida Rudzka-Ostyn (eds.), 488-530. Berlin and New York: Mouton de Gruyter.

Olsen, Mari Jean Broman (1994). The semantics and pragmatics of lexical aspect features. Studies in the Linguistic Sciences 24, 361-375.

Peters, Francis Edwards (1967). Greek Philosophical Terms. A Historical Lexicon. New York: New York University Press.

Quirk, Randolph; Greenbaum, Sydney; Leech, Geoffrey; and Svartvik, Jan (1985). A Comprehensive Grammar of the English Language. London and New York: Longman.

Ramchand, Gillian C. (1997). Aspect and Predication. Oxford: Clarendon.

Rappaport Hovav, Malka and Levin, Beth (2002). Change of state verbs: implications for theories of argument projection. In Proceedings of the 28th Annual Meeting of the Berkeley Linguistics Society, J. Larson and M. Paster (eds.), 269-280. Berkeley: University of California, Berkeley.

Ryle, Gilbert (1988 [1949]). The Concept of Mind. London: Penguin.

Schopf, Alfred (1984). Das Verzeitungssystem des Englischen und seine Textfunktion. Tubingen: Niemeyer.

Smith, Carlota S. (1997). The Parameter of Aspect, 2nd ed. Dordrecht: Kluwer.

Sperber, Dan and Wilson, Deirdre (1995). Relevance." Communication and Cognition, 2nd ed. Oxford: Blackwell.

Van Valin, Robert D. and LaPolla, Randy J. (1997). Syntax: Structure, Meaning, Function. Cambridge: Cambridge University Press.

-- and Wilkins, David P. (1999). The case for 'effector': case roles, agents, and agency revisited. In Grammatical Constructions, Masayoshi Shibatani and Sandra A. Thompson (eds.), 289-232. Oxford: Clarendon Press.

Vendler, Zeno (1967). Linguistics in Philosophy. Ithaca, NY: Cornell University Press.

Verkuyl, Henk (1993). A Theory of Aspectuality: The Interaction Between Temporal and Atemporal Structure. Cambridge Series in Linguistics 64. Cambridge: Cambridge University Press.

--(2001). Aspectual composition: surveying the ingredients. Available from http://wwwuilots.let.uu.nl/conferences/Perspectives_on_Aspect/ P_o_A_index.html

Vetters, Carl (1996). Le mode d'action: une categorie linguistique ou metaphysique? Le langage et l'homme 31,323-342.

Waanders, Frederik M. J. (1984). History of Telos and Teleo in Ancient Greek. Amsterdam: Gruner.

Williams, Christopher (2002). Non-progressive and progressive aspect in English. Fasana: Schena Editore.