Elsevier

Cognitive Science

Volume 27, Issue 4, July–August 2003, Pages 649-668
Cognitive Science

Intra-sentential context effects on the interpretation of logical metonymy

https://doi.org/10.1016/S0364-0213(03)00035-1Get rights and content

Abstract

Verbs such as enjoy in the student enjoyed the book exhibit logical metonymy: enjoy is interpreted as enjoy reading. Theoretical work [Computational Linguistics 17 (4) (1991) 409; The Generative Lexicon, MIT Press, Cambridge, MA, 1995] predicts that this interpretation can be influenced by intra-sentential context, e.g., by the subject of enjoy. In this article, we test this prediction using a completion experiment and find that the interpretation of a metonymic verb is influenced by the semantic role of its subject. We present a Bayesian model that accounts for the interpretation of logical metonymy and achieves a good fit on our experimental data. We show that the parameters of the model can be estimated from completion data or from corpus data.

Introduction

Much work in lexical semantics Bach, 1986, Briscoe et al., 1990, Copestake, 2001, Copestake and Briscoe, 1995, Godard and Jayez, 1993, Pustejovsky, 1991, Pustejovsky, 1995, Pustejovsky and Bouillon, 1995, Vendler, 1968 has been concerned with accounting for the interpretation of verbs like finish and enjoy in constructions such as (1). The verbs in (1) take on different meanings depending on their local syntactic context Pustejovsky, 1991, Pustejovsky, 1995. The meaning of finish varies depending on whether its object is cigarette or beer: (1a) usually means (2a) and (1b) usually means (2b). Along the same lines, one enjoys eating an ice-cream and reading a book (see (1c,d) and (2c,d)). Verbs like finish or enjoy select for an argument that denotes an activity or an event. Such an argument can be realized (for instance) as a VP complement (see (2)). However, finish and enjoy can also occur with an NP complement denoting an artifact (see (1)). In this case, the complement must be type shifted from artifact to activity or event in order to conform to the verb’s semantic restrictions Jackendoff, 1997, Partee, 1992, Pustejovsky, 1991, Pustejovsky, 1995. Pustejovsky (1991) dubs this phenomenon logical metonymy. Roughly, logical metonymy occurs when a subpart of an event or entity ‘stands for’ the event or entity itself. In example (1a) cigarette stands for the event of smoking a cigarette and in (1c) ice-cream stands for the event of eating an ice-cream.

Arguably the most influential account of logical metonymy is the Generative Lexicon Pustejovsky, 1991, Pustejovsky, 1995. In this framework, artifact-denoting nouns are represented as qualia structures specifying key features of the word’s meaning. These features are derivable from world knowledge but are lexicalized and take part in conventionalized interpretation processes. Qualia structures typically include a telic role (i.e., the purpose of the object denoted by the noun), an agentive role (i.e., the event which brought the object into existence), a formal role (i.e., the physical characteristics that distinguish the object within a larger domain) and a constitutive role (i.e., the relation between the object and its constituent parts). For example, the telic role of book is ‘read,’ whereas its agentive role is ‘write.’ The formal role of book is ‘physical object’ (e.g., The book weighs four ounces) and its constitutive role is ‘information’ (e.g., This book is interesting). The telic and agentive roles are central in Pustejovsky’s account of logical metonymy.1 When enjoy combines with an artifact-denoting object noun, a metonymic interpretation is constructed in which missing information is provided by the qualia structure of the noun. More technically, the semantic composition of enjoy with book causes the semantic type of the noun to be coerced into its telic event ‘read’ (or, under specific circumstances, into its agentive event ‘write’), and the semantic relation corresponding to the metonymic verb then predicates over this event. This results in an interpretation of (1d) as (2d).

Qualia representations are intended to capture default interpretations for expressions like (1) in an otherwise neutral context Lascarides and Copestake, 1998, Pustejovsky, 1995. A variety of factors can influence the coercion process and may override the default interpretations. The type of metonymic verb, the sentential subject, and the broader discourse context within which the sentence is embedded may all affect the interpretation being recovered Godard and Jayez, 1993, Lapata and Lascarides, 2003, Lascarides and Copestake, 1998, Pustejovsky, 1995, Verspoor, 1997. The examples in (3) illustrate that the sentential subject can have an influence on the interpretation of metonymic constructions: students typically read books, while authors usually write them. Intuitively, when author is the subject of enjoy the book, an agentive role interpretation of book is preferred, whereas with student as the subject, the intuition is that the telic role interpretation is preferred. Example (4), taken from Lascarides and Copestake (1998), illustrates the effect of discourse context, which in this case triggers the non-conventional interpretation eating the book. It is important to note, however, that examples of contextual influence are exceedingly rare in naturally occurring text. Verspoor (1997) conducted a manual analysis of the verbs begin and finish in the British National Corpus (100 million words) and found that 95% of the logical metonymies of these verbs can be resolved on the basis of information provided by the object of the verb.2

In contrast to the extensive theoretical literature on logical metonymy, experimental work devoted to this topic is still very sparse. The results available so far mainly deal with the cost of coercion (also referred to as enriched composition): McElree, Traxler, Pickering, Seely, and Jackendoff (2001) and Traxler, Pickering, and McElree (2002), for example, investigated the on-line processing of metonymic expressions and found that sentences like (1) lead to higher reading times than sentences that do not require coercion, but yield comparable interpretations. This appears to be in line with Pustejovsky’s (1995) representational account which assumes that coercion involves the computation of additional structure.

However, it still remains to be shown whether nouns such as book, when serving as the object of a metonymic verb, indeed elicit a default interpretation such as reading the book (as argued above), and whether this default interpretation can be overridden by intra-sentential context. The present article addresses these questions by means of a sentence completion study where intra-sentential context is explicitly manipulated (in contrast to McElree et al., 2001, Traxler et al., 2002). We will focus on one aspect of intra-sentential context: the influence of the sentential subject (as in (3)). Note that we will not investigate discourse effects (as in example (4)). On the basis of our experimental data, we will then propose a computational model which is able to account for the interpretation of metonymic verbs.

Section snippets

Experiment

The following experiment investigates the influence of intra-sentential context on the interpretation of logical metonymy. More specifically, we test the hypothesis that the sentential subject can determine how a metonymic construction like enjoy the book is interpreted, i.e., which part of the qualia structure (Agentive vs. Telic) of book is accessed. We will focus solely on the qualia roles that are relevant for logical metonymy (ignoring the constitutive and formal qualia roles). We compare

Model

Lapata and Lascarides (2003) have proposed an account of the interpretation of metonymic expressions based on Bayesian inference. They show that this account successfully generates interpretations for metonymic verbs and adjectives that correlate reliably with behavioral data. In what follows, we will present a model of the experimental data in Section 2 based on Lapata and Lascarides’s (2003) framework.

Conclusions

The experiment presented in this article demonstrated that intra-sentential context has an influence on the interpretation of metonymic verbs like enjoy. We found that a verb–object combination like enjoy the book is interpreted as agentive or a telic, depending on its subject. This is in line with predictions from the theoretical linguistic literature Pustejovsky, 1991, Pustejovsky, 1995.

However, we also found that specific metonymic verb–object combinations differ in their default

Acknowledgements

We would like to thank Matthew Crocker, Gary Dell, Martin Pickering, and two anonymous reviewers for important comments on an earlier version of this paper. We are grateful to Sabine Schulte im Walde for providing access to the verb complement data that she extracted from the HGC.

References (34)

  • Chater, N., & Oaksford, M. (Eds.). (1998). Rational models of cognition. Oxford: Oxford University...
  • Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Lawrence Erlbaum...
  • Copestake, A. (2001). The semi-generative lexicon: Limits on lexical productivity. In: Proceedings of the 1st...
  • A. Copestake et al.

    Semi-productive polysemy and sense extension

    Journal of Semantics

    (1995)
  • M. Corley et al.

    Syntactic priming in English sentence production: Categorical and latency evidence from an Internet-based study

    Psychonomic Bulletin and Review

    (2002)
  • Godard, D., & Jayez, J. (1993). Towards a proper treatment of coercion phenomena. In: Proceedings of the 31st Annual...
  • Howell, D. C. (2002). Statistical methods for psychology (5th ed.). Pacific Grove, CA:...
  • Cited by (32)

    • The production of coerced expressions: Evidence from priming

      2014, Journal of Memory and Language
      Citation Excerpt :

      This is because the meaning of the complement of verbs such as begin, finish, and enjoy depends on both the meaning of the verb and the meaning of the noun phrase. In addition, interpretation of the verb phrase can be affected by the sentence subject (Lapata, Keller, & Scheepers, 2003): The bricklayer began the wall is likely to mean the bricklayer began building the wall, but The decorator began the wall is likely to mean the decorator began painting the wall, since (typically) bricklayers build walls and decorators paint them. More formally, work in lexical semantics (e.g., Briscoe, Copestake, & Boguraev, 1990; Copestake, 2001; Copestake & Briscoe, 1995; Pustejovsky, 1991, 1995) seeks to account for the systematic flexibility of meaning illustrated above.

    • Evidence for serial coercion: A time course analysis using the visual-world paradigm

      2008, Cognitive Psychology
      Citation Excerpt :

      Previous experimental studies have shown that participants experience processing difficulty when reading metonymic constructions such as the artist started the picture. This slowdown in processing has often been attributed to enriched composition, which claims that additional structure has to be constructed when a noun referring to an artifact (such as picture) is coerced into the event representation required by the verb start (Lapata et al., 2003; McElree et al., 2001, 2006; Traxler et al., 2002). However, as discussed in the introduction of Section 4, such differences in reading time could also be due to the fact that readers compute multiple interpretations for metonymic verbs (e.g., the artist started painting/analyzing/framing the picture) which compete with one another and thus decelerate the process of establishing an interpretation that is unambiguous enough for the reader to decide to move on in the text.

    • Complement Coercion Is Not Modulated by Competition: Evidence From Eye Movements

      2008, Journal of Experimental Psychology: Learning Memory and Cognition
    View all citing articles on Scopus

    Most of the work reported here was carried out while all three authors were based at the Department of Computational Linguistics at Saarland University, Germany.

    1

    Tel.: +44-114-222-1832; fax: +44-114-222-1810.

    2

    Tel.: +44-1382-344617; fax: +44-1382-229993.

    View full text