Acquisition of Verb Entailment from Text

Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL, pages 49–56, New York, June 2006. c⃝2006 Association for Computational Linguistics Acquisition of Verb Entailment from Text Viktor Pekar Computational Linguistics Group University of Wolverhampton MB109 Stafford Street Wolverhampton WV1 1SB, UK v.pekar@wlv.ac.uk Abstract The study addresses the problem of auto- matic acquisition of entailment relations between verbs. While this task has much in common with paraphrases acquisition which aims to discover semantic equiva- lence between verbs, the main challenge of entailment acquisition is to capture asymmetric, or directional, relations. Mo- tivated by the intuition that it often under- lies the local structure of coherent text, we develop a method that discovers verb en- tailment using evidence about discourse relations between clauses available in a parsed corpus. In comparison with earlier work, the proposed method covers a much wider range of verb entailment types and learns the mapping between verbs with highly varied argument structures. 1 Introduction The entailment relations between verbs are a natural language counterpart of the commonsense knowl- edge that certain events and states give rise to other events and states. For example, there is an entail- ment relation between the verbs buy and belong, which reflects the commonsense notion that if some- one has bought an object, this object belongs to that person. A lexical resource encoding entailment can serve as a useful tool in many tasks where automatic in- ferencing over natural language text is required. In Question Answering, it has been used to establish that a certain sentence found in the corpus can serve as a suitable, albeit implicit answer to a query (Cur- tis et al., 2005), (Girju, 2003), (Moldovan and Rus, 2001). In Information Extraction, it can similarly help to recognize relations between named entities in cases when the entities in the text are linked by a linguistic construction that entails a known extrac- tion pattern, but not by the pattern itself. A lexical entailment resource can contribute to information re- trieval tasks via integration into a textual entailment system that aims to recognize entailment between two larger text fragments (Dagan et al., 2005). Since entailment is known to systematically inter- act with the discourse organization of text (Hobbs, 1985), an entailment resource can be of interest to tasks that deal with structuring a set of individual facts into coherent text. In Natural Language Gener- ation (Reiter and Dale, 2000) and Multi-Document Summarization (Barzilay et al., 2002) it can be used to order sentences coming from multiple, possibly unrelated sources to produce a coherent document. The knowledge is essential for compiling answers for procedural questions in a QA system, when sen- tences containing relevant information are spread across the corpus (Curtis et al., 2005). The present paper is concerned with the prob- lem of automatic acquisition of verb entailment from text. In the next section we set the background for the study by describing previous work. We then define the goal of the study and describe our …

Content truncated to 500 words