Japanese Parser on the basis of the Lexical-Functional Grammar Formalism and its Evaluation

Japanese Parser on the basis of the Lexical-Functional Grammar Formalism and its Evaluation Hiroshi MASUICHI Corporate Research Group, Fuji Xerox Co., Ltd. 430 Sakai, Nakai-machi, Ashigarakami-gun, Kanagawa, Japan hiroshi.masuichi@fujixerox.co.jp Hiroki YOSHIMURA Corporate Research Group, Fuji Xerox Co., Ltd. 430 Sakai, Nakai-machi, Ashigarakami-gun, Kanagawa, Japan hiroki.yoshimura@fujixerox.co.jp Tom oko OHKUMA Corporate Research Group, Fuji Xerox Co., Ltd. 430 Sakai, Nakai-machi, Ashigarakami-gun, Kanagawa, Japan ohkuma.tomoko@fujixerox.co.jp Yasunari HARADA School of Law, Waseda University 1-6-1 Nishi-waseda, Shinjuku, Tokyo, Japan harada@wasedajp Abstract We report a Japanese parsing system with a linguistically fine-grained grammar based on the Lexical-Functional Grammar (LFG) formalism. The system is the first Japanese LFG parser with over 97% coverage for real-world text. We evaluated the accuracy of the system comparing it with standard Japanese dependency parsers. The LFG parser shows roughly equivalent performance on the dependency accuracy to the standard parsers. It also provides reasonably accurate results of case detection. 1� Introduction Deep grammatical analyses of input sentences based on theoretically sound grammar formalisms are essential for the further development of such NLP applications as machine translation, dialogue understanding, message extraction, etc. In this paper we report development and performance of a parser for the Japanese language based on the Lexical-Functional Grammar (LFG) formalism (Kaplan and Bresnan, 1982; Dalrymple, 2001), The Japanese LFG grammar used for this parser is being developed in relation to the Parallel Grammar (ParGram) project (Butt et al., 1999, 2002). In this project, grammars for six languages, English, French, German, Japanese, Norwegian and Urdu', are under way, sharing various design decisions within the LFG formalism. LFG assumes two levels of syntactic representation for a sentence: a c(onstituent)-structure (a tree) and an functional)-structure (attribute value matrices: AVMs). Within LFG, an f-structure is meant to encode a language universal level of analysis, allowing for cross-linguistic parallel ism.2 Our research goal is to construct a practical Japanese LFG parsing system with broad coverage and deep analysis for real-world text. In this paper we describe the details of the system, and show its coverage and accuracy. For the accuracy evaluation, we compared outputs of the Japanese LFG parsing system with outputs of standard bunsetsu3 dependency parsers for Japanese. I Korean has recently been added to this list. The Japanese LFG grammar is used as the basis for a grammar of Korean (Kim et al., 2003). 2 Butt et al. (2002) reported how far the parallelism of f-structures can be maintained across languages in the ParGram project. Frank (1999) reported on a machine translation system which took advantage of the f-structure parallelism. 3 A bunsetsu is a widely accepted unit of syntactic and phonological phrase structures in Japanese. It consists of one content word plus optionally following function words. 298 V XLE input Sentence Morphological Analyzer Morpheme Sequence� Generator�Lexical Entry Generator Morpheme Sequence�Sentence for Input Sentence� Lexical Entries Core Lexical Entries Verb Lexical Entries Adjective Lexical Entries Adjectival Noun Lexical Entries Grammar Rules c-structure� f-structure Figure 1: Diagram of the Japanese LFG system This paper is organized as follows. Section 2 describes the architecture …

Content truncated to 500 words