Learning Verb Inference Rules from Linguistically-Motivated Evidence

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 194–204, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Learning Verb Inference Rules from Linguistically-Motivated Evidence Hila Weisman§, Jonathan Berant†, Idan Szpektor‡, Ido Dagan§ § Computer Science Department, Bar-Ilan University † The Blavatnik School of Computer Science, Tel Aviv University ‡ Yahoo! Research Israel {weismah1,dagan}@cs.biu.ac.il {jonatha6}@post.tau.ac.il {idan}@yahoo-inc.com Abstract Learning inference relations between verbs is at the heart of many semantic applications. However, most prior work on learning such rules focused on a rather narrow set of in- formation sources: mainly distributional sim- ilarity, and to a lesser extent manually con- structed verb co-occurrence patterns. In this paper, we claim that it is imperative to uti- lize information from various textual scopes: verb co-occurrence within a sentence, verb co- occurrence within a document, as well as over- all corpus statistics. To this end, we propose a much richer novel set of linguistically mo- tivated cues for detecting entailment between verbs and combine them as features in a su- pervised classification framework. We empir- ically demonstrate that our model significantly outperforms previous methods and that infor- mation from each textual scope contributes to the verb entailment learning task. 1 Introduction Inference rules are an important building block of many semantic applications, such as Question An- swering (Ravichandran and Hovy, 2002) and In- formation Extraction (Shinyama and Sekine, 2006). For example, given the sentence “Churros are coated with sugar”, one can use the rule ‘coat → cover’ to answer the question “What are Churros covered with?”. Inference rules specify a directional inference relation between two text fragments, and we follow the Textual Entailment modeling of infer- ence (Dagan et al., 2006), which refers to such rules as entailment rules. In this work we focus on one of the most important rule types, namely, lexical en- tailment rules between verbs (verb entailment), e.g., ‘whisper → talk’, ‘win → play’ and ‘buy → own’. The significance of such rules has led to active re- search in automatic learning of entailment rules be- tween verbs or verb-like structures (Zanzotto et al., 2006; Abe et al., 2008; Schoenmackers et al., 2010). Most prior efforts to learn verb entailment rules from large corpora employed distributional similar- ity methods, assuming that verbs are semantically similar if they occur in similar contexts (Lin, 1998; Berant et al., 2012). This led to the automatic ac- quisition of large scale knowledge bases, but with limited precision. Fewer works, such as VerbOcean (Chklovski and Pantel, 2004), focused on identi- fying verb entailment through verb instantiation of manually constructed patterns. For example, the sentence “he scared and even startled me” implies that ‘startle → scare’. This led to more precise rule extraction, but with poor coverage since contrary to nouns, in which patterns are common (Hearst, 1992), verbs do not co-occur often within rigid pat- terns. However, verbs do tend to co-occur in the same document, and also in different clauses of …

Content truncated to 500 words