Automatic Knowledge Acquisition for Case Alternation between the Passive and Active Voices in Japanese

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1213–1223, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Automatic Knowledge Acquisition for Case Alternation between the Passive and Active Voices in Japanese Ryohei Sasano1 Daisuke Kawahara2 Sadao Kurohashi2 Manabu Okumura1 1 Precision and Intelligence Laboratory, Tokyo Institute of Technology 2 Graduate School of Informatics, Kyoto University {sasano,oku}@pi.titech.ac.jp, {dk,kuro}@i.kyoto-u.ac.jp Abstract We present a method for automatically acquir- ing knowledge for case alternation between the passive and active voices in Japanese. By leveraging several linguistic constraints on al- ternation patterns and lexical case frames ob- tained from a large Web corpus, our method aligns a case frame in the passive voice to a corresponding case frame in the active voice and finds an alignment between their cases. We then apply the acquired knowledge to a case alternation task and prove its usefulness. 1 Introduction Predicate-argument structure analysis is one of the fundamental techniques for many natural language applications such as recognition of textual entail- ment, information retrieval, and machine transla- tion. In Japanese, the relationship between a pred- icate and its argument is usually represented by us- ing case particles1 (Kawahara and Kurohashi, 2006; Taira et al., 2008; Yoshikawa et al., 2011). However, since case particles vary depending on the voices, we have to take case alternation into account to rep- resent predicate-argument structure. There are thus two major types of representations: one uses surface cases, and the other uses normalized-cases for the base form of predicates. For example, while the Ky- oto University Text Corpus (Kawahara et al., 2004), one of the major Japanese corpora that contains an- notations of predicate-argument structures, adopts 1Japanese is a head-final language. Word order does not mark syntactic relations. Instead, postpositional case particles function as case markers. the former representation, the NAIST Text Corpora (Iida et al., 2007), another major Japanese corpus, adopts the latter representation. Examples (1) and (2) describe the same event in the passive and active voices, respectively. When we use surface cases to represent the relationship be- tween the predicate and its argument in Example (1), the case of “女 (woman)” is ga2 and the case of “男 (man)” is ni.2 On the other hand, when we use the normalized-cases for the base form, the case of “女 (woman)” is wo2 and the case of “男 (man)” is ga, which are the same as the surface cases in the active voice as in Example (2). (1) 女が 男に 突き落とされた． woman-ga man-ni was pushed down (A woman was pushed down by a man.) (2) 男が 女を 突き落とした． man-ga woman-wo pushed down (A man pushed down a woman.) Both representations have their own advantages. Surface case analysis is easier than normalized-case analysis, especially when we consider omitted ar- guments, which are also called zero anaphors (Na- gao and Hasida, 1998). In Japanese, zero anaphora frequently occurs, and the omitted unnormalized- case of a zero anaphor is often the same as the surface case of its antecedent …

Content truncated to 500 words