Toward Future Scenario Generation: Extracting Event Causality Exploiting Semantic Relation, Context, and Association Features

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 987–997, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Toward Future Scenario Generation: Extracting Event Causality Exploiting Semantic Relation, Context, and Association Features Chikara Hashimoto∗ Kentaro Torisawa† Julien Kloetzer‡ Motoki Sano§ Istv´an Varga¶ Jong-Hoon Oh∥ Yutaka Kidawara∗∗ ∗ † ‡ § ∥ ∗∗National Institute of Information and Communications Technology, Kyoto, 619-0289, Japan ¶NEC Knowledge Discovery Research Laboratories, Nara, 630-0101, Japan {∗ ch, † torisawa, ‡ julien, § msano, ∥ rovellia, ∗∗kidawara}@nict.go.jp Abstract We propose a supervised method of extracting event causalities like conduct slash-and-burn agriculture→exacerbate desertification from the web using se- mantic relation (between nouns), context, and association features. Experiments show that our method outperforms base- lines that are based on state-of-the-art methods. We also propose methods of generating future scenarios like conduct slash-and-burn agriculture→exacerbate desertification→increase Asian dust (from China)→asthma gets worse. Experi- ments show that we can generate 50,000 scenarios with 68% precision. We also generated a scenario deforestation con- tinues→global warming worsens→sea temperatures rise→vibrio parahaemolyti- cus fouls (water), which is written in no document in our input web corpus crawled in 2007. But the vibrio risk due to global warming was observed in Baker-Austin et al. (2013). Thus, we “predicted” the future event sequence in a sense. 1 Introduction The world can be seen as a network of causal- ity where people, organizations, and other kinds of entities causally depend on each other. This network is so huge and complex that it is almost impossible for humans to exhaustively predict the consequences of a given event. Indeed, after the Great East Japan Earthquake in 2011, few ex- pected that it would lead to an enormous trade deficit in Japan due to a sharp increase in en- ergy imports. For effective decision making that carefully considers any form of future risks and chances, we need a system that helps humans do scenario planning (Schwartz, 1991), which is a decision-making scheme that examines possible future events and assesses their potential chances and risks. Our ultimate goal is to develop a system that supports scenario planning through generat- ing possible future events using big data, which would contain what Donald Rumsfeld called “un- known unknowns”1 (Torisawa et al., 2010). To this end, we propose a supervised method of extracting such event causality as conduct slash-and-burn agriculture→exacerbate desertifi- cation and use its output to generate future sce- narios (scenarios), which are chains of causal- ity that have been or might be observed in this world like conduct slash-and-burn agricul- ture→exacerbate desertification→increase Asian dust (from China)→asthma gets worse. Note that, in this paper, A→B denotes that A causes B, which means that “if A happens, the probability of B in- creases.” Our notion of causality should be inter- preted probabilistically rather than logically. Our method extracts event causality based on three assumptions that are embodied as features of our classifier. First, we assume that two nouns (e.g. slash-and-burn agriculture and desertifica- tion) that take some specific binary …

Content truncated to 500 words