Is a 204 cm Man Tall or Small ? Acquisition of Numerical Common Sense from the Web

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 382–391, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Is a 204 cm Man Tall or Small ? Acquisition of Numerical Common Sense from the Web Katsuma Narisawa1 Yotaro Watanabe1 Junta Mizuno2 Naoaki Okazaki1,3 Kentaro Inui1 1Graduate School of Information Sciences, Tohoku University 2National Institute of Information and Communications Technology (NICT) 3Japan Science and Technology Agency (JST) {katsuma,yotaro-w,junta-m,okazaki,inui}@ecei.tohoku.ac.jp Abstract This paper presents novel methods for modeling numerical common sense: the ability to infer whether a given number (e.g., three billion) is large, small, or nor- mal for a given context (e.g., number of people facing a water shortage). We first discuss the necessity of numerical com- mon sense in solving textual entailment problems. We explore two approaches for acquiring numerical common sense. Both approaches start with extracting numeri- cal expressions and their context from the Web. One approach estimates the distribu- tion of numbers co-occurring within a con- text and examines whether a given value is large, small, or normal, based on the distri- bution. Another approach utilizes textual patterns with which speakers explicitly ex- presses their judgment about the value of a numerical expression. Experimental re- sults demonstrate the effectiveness of both approaches. 1 Introduction Textual entailment recognition (RTE) involves a wide range of semantic inferences to determine whether the meaning of a hypothesis sentence (h) can be inferred from another text (t) (Dagan et al., 2006). Although several evaluation campaigns (e.g., PASCAL/TAC RTE challenges) have made significant progress, the RTE community recog- nizes the necessity of a deeper understanding of the core phenomena involved in textual inference. Such recognition comes from the ideas that cru- cial progress may derive from decomposing the complex RTE task into basic phenomena and from solving each basic phenomenon separately (Ben- tivogli et al., 2010; Sammons et al., 2010; Cabrio and Magnini, 2011; Toledo et al., 2012). Given this background, we focus on solving one of the basic phenomena in RTE: semantic infer- ence related to numerical expressions. The spe- cific problem we address is acquisition of numeri- cal common sense. For example, (1) t : Before long, 3b people will face a water shortage in the world. h : Before long, a serious water shortage will occur in the world. Although recognizing the entailment relation be- tween t and h is frustratingly difficult, we assume this inference is decomposable into three phases: 3b people face a water shortage. ⇔ 3,000,000,000 people face a water shortage. |= many people face a water shortage. |= a serious water shortage. In the first phase, it is necessary to recognize 3b as a numerical expression and to resolve the ex- pression 3b into the exact amount 3,000,000,000. The second phase is much more difficult because we need subjective but common-sense knowledge that 3,000,000,000 people is a large number. In this paper, we address the first and sec- ond phases of inference as an initial step towards semantic processing with numerical expressions. …

Content truncated to 500 words