Tense and Aspect in English Hans Kamp (Editor) dyana Deliverable R2.3.A January 1990 dyana Dynamic Interpretation of Natural Language esprit Basic Research Action br3175 University of Edinburgh Centre for Cognitive Science Department of Artificial Intelligence Centre for Speech Technology Research Department of Linguistics Department of Computer Science Universiteit van Amsterdam Instituut voor Taal, Logica en Informatie Universit¨at Stuttgart Institut fu¨r maschinelle Sprachverarbeitung Universit¨at Tu¨bingen Seminar fu¨r natu¨rlich-sprachliche Systeme Forcopiesofreports,updatesonprojectactivitiesandotherdyana-relatedinformation, contact dyana Administrator Centre for Cognitive Science University of Edinburgh 2, Buccleuch Place Edinburgh eh8 9lw, UK Copyright (cid:176)c 1990 The Individual Authors No part of this document may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. 1 The Semantics of the Progressive and the Perfect in English Toshiyuki Ogihara University of Stuttgart 1 Introduction The purpose of this paper is to investigate the progressive and the perfect in English, andtoprovideaframeworkwhichaccountsforthesemanticsofthesetwoconstructions. In order to reach this goal, we should isolate each aspectual form and consider its semantic contribution. However, it is also important to realize that the progressive and the perfect can cooccur in a single clause. Thus, one of the desiderata for a semantic analysis of these two constructions is that it correctlypredicts the interaction of the two constructions. We consider the progressive first, and the perfect later. 2 The Progressive As far as I can see, there are two major ways of dealing with the semantics of the progressive. One approach is to try to obtain the times at which a progressive sentence is true from the times at which its non-progressive counterpart is true (henceforth, the traditional approach). The other is to say that the extension of a progressive sentence is related to that of its non-progressive counterpart in some other way (henceforth, the non-traditional approach). 2.1 The Traditional Approach I will not go into a detailed review of how the English progressive has been analyzed in formal semantics, but it should be noted that all the analyses of the progressive in the formal semantics literature up to and including Dowty (1977, 1979) pursue the idea that the times at which a progressive sentence is true are derived systematically from the times at which its non-progressive counterpart is true. More specifically, a (cid:48) (cid:48) progressive sentence is said to be true at a time t iff there is an interval t such that t includes t and the non-progressive counterpart of ϕ is true at t(cid:48). These analyses include Scott (1970), Bennett and Partee (1972). It is well-known that these analyses face the problems associated with the imperfective paradox. For details, the reader is referred to Dowty(1979). Dowty(1977,1979)introducesmodalconcepts,therebytryingtoremove the problems associated with the earlier proposals. According to Dowty’s proposal, the truth conditions of the progressive are described in the following manner: 2 (1) [Prog ϕ ] is true at (cid:104)I,w(cid:105) iff for some interval I(cid:48) such that I ⊂ I(cid:48) and I is not a final subinterval of I(cid:48), and for all w(cid:48) such that w(cid:48) ∈ Inr ((cid:104)I,w(cid:105)), ϕ is true at (cid:104)I(cid:48),w(cid:48)(cid:105). [N.B. Inr is a primitive function which picks out a set of worlds which are exactly like the real world up to the time in question and in which the future course of events after this time develops in ways most compatible with the past course of events.] It is important to note that Dowty’s proposal inherits from the earlier analyses the idea that the time(s) at which ϕ is true “surround(s)” the times at which Prog ϕ is true. However, as other linguists such as Vlach (1981) and Abusch (1985) demonstrate, it is quite difficult to determine the set of inertia worlds. For example, Vlach states that when Max is crossing a street and a bus traveling at 30 miles per hour is an inch away from hitting him, the most natural course of event would be that Max is hit by the bus, andhewillnevercrossthestreet. Yet, (2)isperceivedtobeatruestatementdescribing the situation: (2) John was crossing the street when the bus hit him. Although this is not a conclusive argument against Dowty’s analysis, it shows the dif- ficulty involved in selecting the desired set of inertia worlds for a given progressive sentence. Abusch (1985: 177–8) finds another problem with Dowty’s analysis. She shows that it is not sufficient to have a single inertia function. For example, suppose that John was hit by a car at 12 o’clock. Then, the following sentences are both true at 11:59: (3) a. John was crossing the street. b. John was walking to his death. Thisisimpossibleifwehaveasingleinertiafunction. InallinertiaworldsinwhichJohn dies, he does not cross the street, whereas in all inertia worlds in which John crosses the street, he does not die. Thus, it seems that these two sentences must be evaluated by different inertia functions. In other words, the function must be a variable whose value is fixed in a given context. However, even this modification is not sufficient. Consider the following example (Abusch 1985: 179): (4) The wheel was rolling across the road when it was knocked over by the falling rock. Since the rolling of the wheel and the falling of the rock are both governed by the laws of motion, we cannot pick a notion of inertia that contains one but not the other. Thus, 3 we cannot pick a set of inertia worlds which makes the entire sentence true. This also shows how problematic is the idea that we can (and should) derive the extensions of progressives from the times at which their non-progressive counterparts are true. There is another line of attack on Dowty’s analysis of the English progressive suggested by Vlach (1979). It involves achievement sentences, but it is independent of the imper- fective paradox: (5) a. John was dying at t & John died (cid:54)−→ John died at some t(cid:48) such that t ⊆ t(cid:48) b. John was leaving at home at t & John left (cid:54)−→ John left home at some t(cid:48) such that t ⊆ t(cid:48) Suppose that John was dying at t. Our intuition about death (or any other achieve- ments) is that it takes place instantaneously or almost instantaneously. For instance, if John’s life comes to an end at some time t because of the illness with which he was afflicted at some time before t, John dies at t, not at an extended interval including an earlier interval during which he was dying.1 If so, the time at which John was dying cannot be (at least need not be) included in the interval at which John die is true, and we cannot derive the times at which John be dying is true from the times at which John die is true. Dowty’s proposal has another problem which, to the best of my knowledge, has not been noted in the literature. If we blindly apply Dowty’s semantics, examples such as (6b) and (6d), which are dubious, are predicted to be true at some time t if there is (cid:48) an interval t which properly includes t and at which their non-progressive counterparts are true: (6) a. John was drinking a glass of wine at 10. b. ?John was drinking three glasses of wine and a cup of tea at 10. c. John was reading a book at 10 p.m. on May 5th. d. ?John was reading 500 books at 10 p.m. on May 5th. Suppose that John drank a cup of tea and three glasses of wine yesterday and that he was in the process of drinking a glass of wine at 10 yesterday. Then, Dowty’s theory predicts that (6b) is as good as (6a). The same is true of the pair (6c) and (6d). Suppose that John is an avid reader, and he actually read five hundred books last year. Suppose further that he was reading a book at 10 p.m. on May 5th last year. Then, (6c) is predicted to be just as good as (6d). Contrary to what Dowty’s theory predicts, however, (6b) and (6d) sound very strange given the circumstances described. We might say that the reason has to do with the fact that these drinking or reading events cannot be regarded as constituting one coherent event. This is partly because 1Sometimes it is argued that dying takes time. However, even if it is true, it need not be the case that the time at which John be dying is true is included in the interval at which John die is true. 4 John does not consider the four drinking events or the five hundred reading events to form one unit. The following example due to Hans Kamp (p.c.) shows that there are some criteria (probably influenced by our world-knowledge) which determine whether an amalgamation of some atomic events counts as a collective event. Suppose Jones has a course with 50 students, and he gave an exam today. He has to return the exams to the students tomorrow, so he is now working on them. A student stops by and asks his secretary if he is available. (7) is the answer uttered by the secretary: (7) I am sorry, Professor Jones is busy right now. He is grading 50 exams. In the context under consideration, it is natural to consider the fifty grading events to form one unit. This seems to contribute to the fact that (7) is more acceptable than (6b) or (6d). 2.2 The Non-traditional Approach – The Progressive and Stativity Having found problems with the traditional approach in which the set of times at which a progressive sentence is true is determined on the basis of the set of times at which its non-progressive counterpart is true, I now turn to the non-traditional approach. As mentioned earlier, the general idea behind the non-traditional approach is that the truth conditions of a progressive sentence do not make reference to the times at which its non-progressive counterpart is true. In principle, this leaves us with a lot of possible ways of making a connection between the extension of a non-progressive sentence and that of its progressive counterpart. However, to the best of my knowledge, only one concrete proposal has been presented in the literature. It is a proposal advocated by Vlach(1981)andMoens(1987),anditclaimsthattheprogressiveformisa“stativizer”. The linguistics literature which concerns the progressive and stativity is rather confus- ing. One source of the claim that progressives are stative is Vlach (1981). He contends that the traditional approach, in which one tries to derive the times at which a progres- sive sentence is true from the times at which its non-progressive counterpart is true, is fundamentally misguided. Vlach argues for a proposal in which the notion of “process” plays an important role. For him, the progressive marker in English (i.e. be V-ing) is a stativizer. That is, it combines with a process and produces a stative sentence as a result. The leading idea in his account of the semantics of the progressive is that the progressive operator always combines with a process sentence. When a non-process sentence combines with the progressive operator, the sentence is turned into a process in advance. If the progressive marker always combines with a process, it is clear that the problems associated with the imperfective paradox (Dowty 1977, 1979) never arise there. LetusseehowVlach’sargumentproceeds. Considerfirstthefollowingpairofsentences. (8) a. John ran. 5 b. John was in the room. What is the difference between (8a) and (8b)? The difference between them is brought out clearly when they occur with when-adverbials: (9) a. John ran when I arrived. b. John was in the room when I arrived. It is clear that (9a) is true in cases where the time at which John run is true follows the time at which I arrive is true. On the other hand, (9b) is true iff the time at which John be in the room is true continues at least until the time at which I arrive is true. Vlach assumes that point adverbials like when-clauses and at-adverbials serve as the means by which to distinguish between stative and non-stative sentences. Vlach adopts the following truth conditions for stative sentences: (10) A sentence ϕ is stative if and only if the truth of (Past ϕ) when I arrived requires that ϕ was true for some period leading up to the time of my arrival. Let us now consider progressive sentences such as (11)? (11) John was running when I arrived. Exampleslike(11)leadVlachtoconcludethatprogressivesarestativebecausethetruth of the sentence entails that the process of John’s running must continue at least until the time of my arrival. Note that when the main clause is not progressivized as in (9a), the time of John’s running is understood as following the time of my arrival. The above truth conditions also apply to progressive sentences involving the imperfective paradox: (12) John was building a house when he was struck by the lightning and died. (12) requires that John be building a house be true at an interval extending up to the time of his death. (The possibility that John’s building a house surrounds the time of his death is excluded due to pragmatic reasons.) In order to do justice to these intuitive judgments, Vlach sets up his system in the following way: The progressive is a stativizer. What does a stativizer do? It simply changes the way in which a sentence interacts with point adverbials. The progressive operator combines only with a process sentence, and the resulting sentence has exactly the same truth conditions as the process sentence with which the progressive operator combines. For example, John run and John be running have exactly the same truth 6 conditions in Vlach’s system. The difference between them manifests itself only when they cooccur with point adverbials. A question naturally arises as to how to deal with progressivized telic sentences. Consider the following examples: (13) a. John was building a house. b. John was running. c. John was dying. (cid:48) a. PST PROG PROC [John build a house] (cid:48) b. PST PROG [John run] (cid:48) c. PST PROG PROC [John die] As mentioned above, the progressive operator always combines with a process sentence. In order to cope with progressivized telic sentences, Vlach posits the operator PROC, which serves to turn a telic sentence into a process sentence. Note that for someone subscribing to the traditional approach, the imperfective paradox is construed as the problem of deriving from the times at which a telic sentence ϕ is true the times at which PROG ϕ is true. As the progressive operator never combines directly with a telic sentence in Vlach’s system, the imperfective paradox in its original form never arises. This does not mean, however, that we have found a true solution to the problem. What Vlach has done is to shift the problem associated with the imperfective paradox elsewhere. It is now trivial to account for the meaning of the progressive operator because it always combines with a process sentence and simply changes the sentence withwhichitcombinesintoastativesentence. Buttheimperfectiveparadoxre-emerges as the problem of deriving from the extension of a telic sentence ϕ the extension of PROC ϕ. Vlach argues against the view that the times at which a progressive sentence is true should be predictable from the times at which its non-progressive counterpart is true. Moreover, the truth conditions for PROG PROC ϕ and PROC ϕ (where ϕ is a telic sentence) are exactly the same for Vlach. It follows, therefore, that he is required to derive the extension of PROC ϕ from the extension of ϕ without making reference to the times at which ϕ is true. As far as I can see, Vlach does not offer a proposal which is truly distinct from Dowty’s account of the imperfective paradox. Vlach characterizes the process associated with an accomplishment in the following way (Vlach 1979: 288): (14) If ϕ is an accomplishment sentence, then Proc [ϕ] is that process P that leads to the truth of ϕ, and such that if ϕ is to become true at I, then P starts at the beginning of I and ends at the end of I. Here the definition includes the phrase “the process P that leads to the truth of ϕ”. This phrase is subject to several possible interpretations. One is that it is the process that typically leads to the truth of ϕ. Another interpretation is that it is the process that actually leads to the truth of ϕ if the process continues without being blocked or interrupted. In what follows, I will show that (i) the first interpretation does not work, and that (ii) if we take the latter interpretation, it is not very different from Dowty’s proposal. 7 One way of making Vlach’s analysis truly distinct from Dowty’s analysis is to pursue thefollowinglineofreasoning. Inordertoderivetheextensionofaprogressivesentence from the “meaning” of its non-progressive counterpart, we assume that the process (or the set of processes) associated with an accomplishment is conventionalized, and it is known to the native speaker in advance. It seems prima facie reasonable to assume that there is a series of processes which jointly make up an accomplishment event. For example, an event of John’s building a house can be assumed to consist of the following set of activities: {John hammers in nails, John puts up posts, etc.}. Then the truth condition of the progressivized sentence can be described in the following way: (15) John is building a houseistrueifandonlyifone(ormore)oftheprocessesobtains at the speech time. Unfortunately, this truth condition is wrong on two counts. First, this is not a sufficient condition. SupposethatJohnwashammeringinnails, andsomeoneuttersthefollowing sentence: (16) John is building a house. But it is possible that John was in fact building a kennel for his dog. (Of course, a question arises as to how we can verify this. For our purposes, it is sufficient to assume that John had the intention of building a kennel and he in fact finished building one later.) Ifso, Johncanlegitimatelycomplainthat(16)doesnotaccuratelydescribewhat he was doing at the time. This shows that engaging in one of the activities described abovedoesnotconstituteasufficientconditionforthetruthof(16). Secondly,itisnota necessary condition, either. Due to technological advances, modern house-building does not have to involve all of the sub-events associated with conventional house-building. SupposeJohnisweldingpre-fabricatedpartstogether. Thisisnotpartofaconventional house-building event. However, once we understand this act as something which leads to the completion of a house, it becomes possible to use (16) in order to make reference to this process. That is, knowing the meaning of the lexical items constituting the sentence (16) is not sufficient to predict what counts as a “process of building a house”. The occurrence of any concrete process can qualify as a “process of building a house” if it can be understood in a particular situation as a process “leading to” the completion of a house. Hinrichs (1983) makes the same point by giving the following example: (17) John was making Bill a millionaire. There are many ways in which John can make Bill a millionaire. John may have been dying, leaving a fortune to be inherited by Bill. John may have helped Bill invest his money in a clever way. It is obvious that there is no predetermined set of specific processesthatjointlymakeupaneventof“John’smakingBillamillionaire”,noristhere 8 a set of alternative processes. The above discussion shows that the first interpretation of Vlach’s approach fails. Another way of interpreting Vlach’s condition is to say that the process associated with ϕ is the process that actually “leads to” the truth of ϕ. This interpretation works as long as we restrict ourselves to examples which do not involve the imperfective para- dox. However, once we turn to examples involving the imperfective paradox, it is hard to avoid the conclusion that the progressive operator involves modal concepts just as Dowty’s proposal does. That is, if the process is in fact blocked or interrupted in the real world, how could we check whether a certain process would lead to the truth of ϕ? The only possibility is to consider if the process would lead to the truth of ϕ if it had not been blocked. This clearly involves modal concepts. Presumably, Vlach’s proposal still differs from Dowty’s in that it does not make reference to the times at which non-process sentences are true. However, this makes only a very small difference between Dowty’s approach and Vlach’s approach. With some achievement verbs such as die, Vlach’s approach may have some advantages over Dowty’s. Vlach can say that any achievement event is necessarily preceded and abutted by a process which leads to the event. For example, suppose that John is in the extension of the predicate die at an (cid:48) interval I. Then, there is another interval I which precedes and abuts I and throughout which the process which leads to the truth of John die obtains. This predicts that the times at which John die is true do not surround the times at which John be dying is true. Thus, Vlach is able to avoid the problem pointed out above in connection with (5a) and (5b). However, with accomplishment sentences, I do not see any substantial difference between Dowty’s approach and Vlach’s approach. According to (14), the process associated with an accomplishment sentence is that process P that leads to the truth of ϕ, and such that if ϕ is to become true at I, then P starts at the beginning of I and ends at the end of I. Note that if we confine ourselves to cases in which ϕ is true in the actual world at some interval, say I, it follows from Vlach’s semantics that the timesatwhichProc[ϕ]obtainsarerestrictedtothesubintervalsofI.Thisiscompletely equivalent to Dowty’s semantics for progressives. In other words, Vlach makes reference totimesatwhichϕistrueinordertodefinetheintervalatwhichProc[ϕ]obtains. This is precisely what Vlach argues against. If we turn to accomplishment sentences which trigger the imperfective paradox, we arrive at the same conclusion: Vlach’s proposal is essentially equivalent to Dowty’s proposal. We must conclude that Vlach’s contribution lies in pointing out problems with the existing systems, not in offering something that substantially improves on them. Moens (1987) also claims that the progressive aspect marker in English only combines with process sentences and the resulting sentence is stative. If I understand him cor- rectly, his leading idea is essentially the same as that made by Vlach (1979), and the comments directed at Vlach also apply to Moens’s analysis of the progressive. He de- scribes the transition from culminated processes to processes as something that “strips off” the culmination: (18) A culminated process (like write a novel) can only combine with a progressive if a route is found from culminated process to process first. The most obvious one is labelled “–culmination”, involving the “stripping off” of the culmination 9