Corpus-Based Analysis of Japanese Relative Clause Constructions

Corpus-Based Analysis of Japanese Relative Clause Constructions Takeshi Abekawa1 and Manabu Okumura2 1 Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology, Japan abekawa@lr.pi.titech.ac.jp 2 Precision and Intelligence Laboratory, Tokyo Institute of Technology, Japan oku@pi.titech.ac.jp Abstract. Japanese relative clause constructions (RCC’s) are defined as being the NP’s of structure ‘S NP’, noting the lack of a relative pronoun or any other explicit form of noun-clause demarcation. Japanese relative clause modification should be classified into at least two major semantic types: case-slot gapping and head restrictive. However, these types for relative clause modification cannot apparently be distinguished. In this paper we propose a method of identifying a RCC’s type with a machine learning technique. The features used in our approach are not only rep- resenting RCC’s characteristics, but also automatically obtained from large corpora. The results we obtained from evaluation revealed that our method outperformed the traditional case frame-based method, and the features that we presented were effective in identifying RCC’s types. 1 Introduction Japanese relative clause constructions (RCC’s) are defined as being the NP’s of structure ‘S NP’, noting the lack of a relative pronoun or any other explicit form of noun-clause demarcation[1]. Japanese relative clause constructions should be classified into at least two major semantic types: case-slot gapping and head restrictive. However, these types for relative clause constructions cannot appar- ently be distinguished. Given the types of Japanese relative clause constructions and a corpus of Japanese relative clause construction instances, we present a machine learning based approach to classifying RCC’s. We present a set of lexical and semantic features that characterize RCC’s, and integrate them as a classifier to determine RCC types. We use decision tree learning as the machine learning algorithm. Distinguishing case-slot gapping and head restrictive relative clauses, or re- solving the semantic relationship between the relative clause and its head noun has several application domains, such as machine translation from Japanese[5]. It also has a place in text understanding tasks, such as splitting a long sentence into multiple shorter sentences, and removing less important clauses to shorten a sentence[6]. R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 46–57, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 Corpus-Based Analysis of Japanese Relative Clause Constructions 47 Previously, relative clauses had been analyzed with rule-based methods that utilized case frames[5,2]. Using hand-crafted rules and knowledge creates several problems: the high cost of constructing them, and lower scalability and coverage. Recently, due to the availability of very large corpora, corpus-based and ma- chine learning-based approaches have been actively investigated[7]. Cooccurrence information between nouns and verbs can be calculated from the syntactically parsed corpus, and this information can be used preferentially instead of hand- crafted case frames to determine whether a noun can be the filler of a case-slot of a verb[7,11]. However, merely using the cooccurrence information between nouns and verbs instead of case frames cannot provide a good solution to the analysis of Japanese relative clauses. Clauses with high occurrence probability of the main verb and …

Content truncated to 500 words