Automatic recognition of habituals: a three-way classification of clausal aspect

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2471–2481, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Automatic recognition of habituals: a three-way classification of clausal aspect Annemarie Friedrich Manfred Pinkal Department of Computational Linguistics Saarland University, Saarbr¨ucken, Germany {afried,pinkal}@coli.uni-saarland.de Abstract This paper provides the first fully au- tomatic approach for classifying clauses with respect to their aspectual properties as habitual, episodic or static. We bring together two strands of previous work, which address only the related tasks of the episodic-habitual and stative-dynamic distinctions, respectively. Our method combines different sources of information found to be useful for these tasks. We are the first to exhaustively classify all clauses of a text, achieving up to 80% accuracy (baseline 58%) for the three-way classifi- cation task, and up to 85% accuracy for related subtasks (baselines 50% and 60%), outperforming previous work. In addi- tion, we provide a new large corpus of Wikipedia texts labeled according to our linguistically motivated guidelines. 1 Introduction In order to understand the function of a clause within a discourse, we need to know the clause’s aspectual properties. The distinction between dy- namic and stative lexical aspect, as in exam- ples (1a) and (1b) respectively, is fundamental (Vendler, 1957). Its automatic prediction has been addressed previously (Siegel and McKeown, 2000; Zarcone and Lenci, 2008; Friedrich and Palmer, 2014). (1) (a) Bill drank a coffee after lunch. (dynamic) (b) Bill likes coffee. (stative) In this work, we focus on habituality as another fundamental aspectual property. While example (1a) is an episodic sentence, i.e., a sentence ex- pressing information about a particular event, the same dynamic verb can be used to characterize the default behavior of an individual or of a kind in a certain type of situation (2). (2) (a) Bill usually drinks coffee after lunch. (habitual) (b) Italians drink coffee after lunch. (habitual) The entailment properties of episodic and ha- bitual (or characterizing) sentences differ substan- tially. Also, they have different functions in dis- course. The episodic event expressed by (1a) is typically embedded in the temporal structure of a narration. The habitual sentence (2a) can be used, e.g., as an explanation (why Bill is still sitting at the table), or in a contrastive context (today, he or- dered a grappa instead). Generic sentences with kind-referring subjects (2b) can also be habitual, generalizing at the same time over typical mem- bers of this kind and over situations in which they typically carry out some action. Habitual sentences do not move narrative time, similar to stative clauses such as (1b). Carlson (2005) considers habituals to be aspectually sta- tive. Since there are clear differences between or- dinary statives such as (1b) and habituals, we ap- ply a three-way distinction for clausal aspect in this work. We classify clauses as one of the three categories habitual, episodic and static.1 Through its impact on entailment properties and temporal discourse structure, the determination of clausal aspect is relevant to various natural lan- guage processing applications …

Content truncated to 500 words