Integrating Multiple Dependency Corpora for Inducing Wide-coverage Japanese CCG Resources

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1042–1051, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Integrating Multiple Dependency Corpora for Inducing Wide-coverage Japanese CCG Resources Sumire Uematsu† uematsu@cks.u-tokyo.ac.jp Takuya Matsuzaki‡ takuya-matsuzaki@nii.ac.jp Hiroki Hanaoka† hanaoka@nii.ac.jp Yusuke Miyao‡ yusuke@nii.ac.jp Hideki Mima† mima@t-adm.t.u-tokyo.ac.jp †The University of Tokyo Hongo 7-3-1, Bunkyo, Tokyo, Japan ‡National Institute of Infomatics Hitotsubashi 2-1-2, Chiyoda, Tokyo, Japan Abstract This paper describes a method of in- ducing wide-coverage CCG resources for Japanese. While deep parsers with corpus- induced grammars have been emerging for some languages, those for Japanese have not been widely studied, mainly be- cause most Japanese syntactic resources are dependency-based. Our method first integrates multiple dependency-based cor- pora into phrase structure trees and then converts the trees into CCG derivations. The method is empirically evaluated in terms of the coverage of the obtained lexi- con and the accuracy of parsing. 1 Introduction Syntactic parsing for Japanese has been domi- nated by a dependency-based pipeline in which chunk-based dependency parsing is applied and then semantic role labeling is performed on the de- pendencies (Sasano and Kurohashi, 2011; Kawa- hara and Kurohashi, 2011; Kudo and Matsumoto, 2002; Iida and Poesio, 2011; Hayashibe et al., 2011). This dominance is mainly because chunk- based dependency analysis looks most appropriate for Japanese syntax due to its morphosyntactic ty- pology, which includes agglutination and scram- bling (Bekki, 2010). However, it is also true that this type of analysis has prevented us from deeper syntactic analysis such as deep parsing (Clark and Curran, 2007) and logical inference (Bos et al., 2004; Bos, 2007), both of which have been sur- passing shallow parsing-based approaches in lan- guages like English. In this paper, we present our work on induc- ing wide-coverage Japanese resources based on combinatory categorial grammar (CCG) (Steed- man, 2001). Our work is basically an extension of a seminal work on CCGbank (Hockenmaier and Steedman, 2007), in which the phrase structure trees of the Penn Treebank (PTB) (Marcus et al., 1993) are converted into CCG derivations and a wide-coverage CCG lexicon is then extracted from these derivations. As CCGbank has enabled a va- riety of outstanding works on wide-coverage deep parsing for English, our resources are expected to significantly contribute to Japanese deep parsing. The application of the CCGbank method to Japanese is not trivial, as resources like PTB are not available in Japanese. The widely used re- sources for parsing research are the Kyoto corpus (Kawahara et al., 2002) and the NAIST text corpus (Iida et al., 2007), both of which are based on the dependency structures of chunks. Moreover, the relation between chunk-based dependency struc- tures and CCG derivations is not obvious. In this work, we propose a method to integrate multiple dependency-based corpora into phrase structure trees augmented with predicate argument relations. We can then convert the phrase structure trees into CCG derivations. In the following, we describe the details of the integration method as well as Japanese-specific issues in the conversion into …

Content truncated to 500 words