On Deriving Aspectual Sense

Size: px

Start display at page:

Download "On Deriving Aspectual Sense"

Lawrence Ball
6 years ago
Views:

1 COGNrTlVE SCIENCE 2, (1978) On Deriving Aspectual Sense BONNIE LYNN WEBBER University of Pennsylvania and Bolt Beranek and Newrnan. Inc. In his recent article "Verbs, Time and Modality," M. J. Steedman (1977) proposes a recursive scheme for identifying the aspectual character of a proposition. The purpose of this squib is to point out some faults and gaps in that account that may suggest some interesting directions for future research. 1. THE BASIC ACCOUNT Steedman proposes to identify the aspectual character of a proposition by means of a system of syntactic rewrite rules, each with a corresponding semantic rule, as in a Montague (1974) grammar. The given syntactic ruks express how a proposition of one aspectual type can be built up from, or decomposed into, a proposition of another type. The aspectual types are similar to those in Vendler's (1967) earlier aspectual classification of verbs and include accomplishments-happenings which occupy a period of time and have an intrinsic end beyond which the process cannot continue (e.g., "Chip built a chair''); activities-happenings over time which have no intrinsic end (e.g., "Chip was singing"); achievements-instantaneous events (e.g., "Chip reached the head of the trail'?; and actions-events which share some properties with accomplishments and some properties with achievements but which differ in certain important respects from both (e.g., "Chip bounced the squash ball against the wall'?. Steedman provides many examples which serve to motivate his basic observation that propositions of one type can be transformed into propositions of another type by their appearance in particular contexts, such as with a temporal adverb. For example, the proposition 1. The program ran in 2.3 seconds. is seen as describing an accomplishment, one whose intrinsic end is the program's completion. This is seen as deriving from the proposition 2. The program ran. which simply describes an activity with no intrinsic end. In terms of production rules, Steedman expresses ti' 1s as

2 386 WEBBER accomplishment - activity [+ in time] where square brackets indicate optionality. The interpretation of the lefthand side is defined in terms of the interpretation of the right-hand side constituents as activity & achievement-of(activity) [& T(activity) = time & T(achievement-of(activity)) = end-of(time:)] That is, the interpretation of the left-hand side as an accomplishment has the sense of both the activity expressed by the embedded proposition and its conclusion. If an in temporal adverb is present, this interpretation also includes an indication of the time spent on the activity in reaching its conclusion. 2. TEMPORAL EXTENT AS A FACTOR IN INTERPRETATION One interesting fault in Steedman's account lies in his claim that when an accomplishment proposition occurs in the context of a for temporal adverb, the resulting proposition expresses "an activity involving the continual repetition of the basic event" (Steedman, 1977, p. 220). For example, where the proposition 3. Chip played the "~oonlighi Sonata." is regarded as expressing an accomplishment-one whose intrinsic end corresponds to the conclusion of the piece, the proposition 4. Chip played the "Moonlight Sonata" for three days. is regarded as expressing an activity involving the continual repetition of the piece. Steedman presents this claim in the form of the syntactic rewrite rule activity - action [+for time] where the interpretation of the left-hand side derives from the interpretation of the right-hand constituents as follows: continually-repeated(action) [& T(continual1y-repeated(action)) = time] However, consider the similarly structured proposition 5. Chip played the "Moonlight Sonata" for 13 seconds. No one to whom I presented this sentence interpreted it as expressing continual repetition of an event: it only seemed appropriate if either Chip had announced that he was going to play the "Moonlight Sonata" and then, had only done so for 13 seconds, or if the speaker had in some other way recognized that the music that Chip had played for 13 seconds was the

3 ON DERIVING ASPECTUAL SENSE 387 "Moonlight Sonata." This suggests that a repetitive sense only appears if the temporal extent denoted by the for adverb is greater than the extent of the event. That is, the aspectual character of a proposition cannot be computed without taking account of the temporal extent ascribed to an event. However, in questioning my informants I found that only one of them actually knew how long a typical performance of the "Moonlight Sonata" would take. The others merely felt that 13 seconds was too short a time, whereas 3 days was too long. This suggests that when the temporal extent.of an event is not known, people will ascribe it a plausible one, probably using reasoning strategies of the type described in Collins (1978), simply in order to work out the semantics of the sentence.' So two factors missing from Steedman's account are the typical extent of an event and people's tendency to compute a plausible typical extent when the actual one is not known. 3. MULTIPLE ASPECTUAL SENSES In Steedman's system of rules, there are three terminal symbols: Saclivity, Sachievemen,, and Saction. These correspond to propositions whose verb's basic meaning is identified in the dictionary as being of type activity, achievement, or action, respectively. (Steedman argues at length for accomplishment being a derived category: accomplishment = activity + achievement.) Although Steedman notes that some verbs seem to have several aspectual senses, he does not discuss disambiguation. What I want to point out is that, not surprisingly, any aspectual analysis system designed to apply Steedman's rules must be able to access the entire discourse context in disambiguating what rules have been applied to what basic propositions. As an example of this, consider the following sentence pairs. 6a. The patient was told to exhale slowly. 6b. However she exhaled in 5 seconds. 7a. The patient was told to hold her breath. 7b. However she exhaled in 5 seconds. In the first pair, I am using sentence 6b to convey that for those 5 seconds the patient was exhaling, but that I do not consider that to beexhalingslowly. In example 7, I am using that same sentence (7b) to convey that the patient was able to hold her breath for only 5 seconds and that after that, she exhaled. 'Notice that even when an event does not seem to have an intrinsicend, people willstill ascribe it one when necessary to make sense of the subsequent discourse. For example, Chip played the piano. Then he danced a fast fandango. while Kathy stared on in amazement. The deictic term "then" points to some time after Chip finished playing.

4 388 WEBBER Notice that Steedman's rules can be used to generate these alternative readings as follows. Accomplishment propositions containing an in temporal adverb (such as 6b and 7b above) can be formed from either a simpler activity proposition or a simpler achievement. In terms of Steedman's syntactic rewrite rules, a. accomplishment - activity [+ in time] b. accomplishment -- achievement [+ in time] The interpretation of (a) as an accomplishment can be expressed in terms of the interpretations of its constituents as activity & achievement-of(activity) [& T(activity) = time & T(achievement-of(activity)) = end-of(time)] while the interpretation of (b) can be expressed as achievement & activity-for(achievement) [& T(activity-for(achievement)) = time & T(achievement) = end-of(time)] To this pair of rules can be attributed the pair of senses associated with the proposition, "However, she exhaled in 5 seconds;" Either the Sacliv,v sense of "exhale" is involved or the Sachievemenr sense is. In the former case, the interpretation rule (a) assigns to the accomplishment' the sense that the activity of exhaling has occurred during and has been achieved by the end of 5 seconds, i.e., T (exhaling) = 5 seconds & T(achievemest-of(exha1ing)) = end-of(5 seconds) In the latter case, interpretation rule (b) assigns to the proposition the sense that some activity associated with exhaling (here, breath-holding) has occurred for 5 seconds, after which the instantaneous event of exhaling occurs. This exemplifies the obvious point I want to make: that a system which is trying to use Steedman's rules to identify the aspectual sense of a sentence will also have to take into account the whole discourse context in making that decision. 4. CONJUNCTION Another interesting gap in Steedman's account is the absence of any discussion of conjunction. My point here is to show that conjunction can affect the aspectual character of a proposition and thus why such a discussion is necessary. Consider the following sentences 8. The patient inhaled for 20 seconds. 9. The patient exhaled for 20 seconds. 10. The patient inhaled and exhaled for20 seconds.

5 ON DERIVING ASPECTUAL SENSE 389 All three sentences express activities. However, while sentences 8 and 9 convey that for 20 seconds the patient engaged in one single continuous activity (inhaling and exhaling, respectively), sentence 10 conveys the continual repetition for that time of the accomplishment of an inhalelexhale cycle. One question to be raised is when will the conjunction of activities convey a sense of repetition of an accomplishment. Notice that in sentence 11, there is no such sense. 11. The patient wept and hit the doctor for 20 seconds. That is, weeping and hitting thedoctor are still construed as separateactivities rather than forming a single accomplishment. One suggestion is that accomplishments only result from conjunctions whose conjuncts are mutually exclusive and can be construed as the parts of some cycle (like breathing). There are certainly other questions to be raised about theeffect of conjunction on the aspectual character of propositions, including whether disjunction can produce similar effects and how temporal extent comes in (cf. Section 2). 5. ACHIEVEMENT RELATED ACTIVITIES One final gap in Steedman's account involves the second component of interpretation rule (b) above-"activity-for(achievement)." This component is meant by Steedman to represent the activity leading up to the achievement described by the given proposition. The gap in his account is the lack of discussion of how to identify that activity. That is, can anactivity be paired a priori with each achievement verb, or is the appropriate activity only to be found by analyzing the situation underlying the discourse? Some achievement verbs do seem to have a particular activity strongly associated with them. In fact, the verb that Steedman uses for illustration ("arrive") seems to be one of them: its associated activity is something like "traveling." For example, given the sentence 12. John arrived in London in an hour. a correct answer to the question "What was John doing for that hour?"iswhe was traveling to London." "Exhale" on the other hand does not seem to have a particular activity strongly associated with it: that activity must be derived insome way from the discourse. For example, given sentence 7 (repeated below), a correct answer to the question "What was the patient doing for those 5 seconds?"isushe was holding her breath." 7a. The patient was told to hold her breath. 7b. However she exhaled in 5 seconds. However given sentence 13, a correct answer is "She was inhaling."

6 390 WEBBER 13a. The patient was told to inhale until her lungs were full. 13b. However she exhaled in 5 seconds. As far as I am aware, this problem of how to identify the activityassociated with a given achievement has not been treated. ACKNOWLEDGMENT The author would like to thank Dr. Aravind Joshi for his comments on earlier drafts of this squib. REFERENCES Collins, A. Fragments of a. theory of human plausibility reasoning. In D. Waltz (Ed.), Proceedings of TINLA P-2. University of Illinois. Urbana IL, July, (Available from the ACM.) Montague, R. Universal Grammar. In R. H. Thomason (Ed.), Forrnalphilosophy: Selected papers of Richard Montague. New Haven, Yale Univ. Press, Steedman, M. J. Verbs, time and modality. Cognitive Science. 1977, 1, Vendler, 2. Linguistics in Philosophv. Ithaca, N.Y.: Comell Univ. Press, 1967.

Logical Forms. Prof. Sameer Singh CS 295: STATISTICAL NLP WINTER February 16, 2017

Logical Forms. Prof. Sameer Singh CS 295: STATISTICAL NLP WINTER February 16, 2017 Logical Forms Prof. Sameer Singh CS 295: STATISTICAL NLP WINTER 2017 February 16, 2017 Based on slides from Noah Smith, Dan Klein, Tom Kwiatkowski, and everyone else they copied from. Outline Logical Semantics