Excitatory or Inhibitory: A New Semantic Orientation Extracts Contradiction and Causality from the Web

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 619–630, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Excitatory or Inhibitory: A New Semantic Orientation Extracts Contradiction and Causality from the Web Chikara Hashimoto∗ Kentaro Torisawa† Stijn De Saeger‡ Jong-Hoon Oh§ Jun’ichi Kazama¶ National Institute of Information and Communications Technology Kyoto, 619-0289, JAPAN {∗ ch, † torisawa, ‡ stijn, § rovellia, ¶ kazama}@nict.go.jp Abstract We propose a new semantic orientation, Ex- citation, and its automatic acquisition method. Excitation is a semantic property of predicates that classifies them into excitatory, inhibitory and neutral. We show that Excitation is useful for extracting contradiction pairs (e.g., destroy cancer ⊥ develop cancer) and causality pairs (e.g., increase in crime ⇒ heighten anxiety). Our experiments show that with automatically acquired Excitation knowledge we can extract one million contradiction pairs and 500,000 causality pairs with about 70% precision from a 600 million page Web corpus. Furthermore, by combining these extracted causality and contradiction pairs, we can generate one mil- lion plausible causality hypotheses that are not written in any single sentence in our corpus with reasonable precision. 1 Introduction Recognizing semantic relations between events in texts is crucial for such NLP tasks as question an- swering (QA). For example, to answer the question “What ruined the crops in Japan?” a QA system must recognize that the sentence “the Fukushima nuclear power plant caused radioactive pollution and contaminated the crops in Japan” contains a causal relation and that contaminate crops entails ruin crops but contradicts preserve crops. To facilitate the acquisition of causality, contra- diction, paraphrase and entailment relations between events we propose a new semantic orientation, Ex- citation, that classifies unary predicates (templates, hereafter) into excitatory, inhibitory and neutral. An excitatory template entails that the main function or effect of the referent of its argument is activated or enhanced (e.g., cause X, preserve X), while an in- hibitory template entails that it is deactivated or sup- pressed (e.g., ruin X, contaminate X, prevent X). Excitation is useful for extracting contradiction; if two templates with similar distributional profiles have opposite Excitation polarities, they tend to be contradictions (e.g., contaminate crops and preserve crops). With extracted contradictions we can distin- guish paraphrases from contradictions among distri- butionally similar phrases. Furthermore, contradic- tion in itself is important knowledge for Recogniz- ing Textual Entailment (RTE) (Voorhees, 2008). ity.Excitation is also a powerful indicator of causal- In the physical world, the activation or de- activation of one thing often causes the activation or deactivation of another. Two excitatory or in- hibitory templates that co-occur in some temporal or logical order in the same narrative often describe a causal chain of events, like “the Fukushima nu- clear power plant caused radioactive pollution and contaminated crops in Japan”. In this paper we propose both the concept of Ex- citation and an automatic method for its acquisition. Our method acquires Excitation templates based on certain natural, language independent constraints on narrative …

Content truncated to 500 words