A Rule-based Morpho-semantic Analyzer of the Japanese Verb Phrases of Simple Sentences

A Rule-based Morpho-semantic Analyzer of the Japanese Verb Phrases of Simple Sentences * Yukiko Sasaki Alam Hosei University, Department of Digital Media, 3-7-2 Kajino-cho, Koganei, Tokyo, 184-8584 Japan sasaki@hosei.ac.jp Abstract. This paper presents the design and algorithms of a morpho-semantic analyzer to parse the Japanese verb phrases of simple sentences. This parser aims to understand the whole semantics of verb phrases by parsing them into semantic units, and thus differs from existing morphological analyzers that primarily segment sentences into morpho-phonemes, labeled with the classifications. Unlike other statistically aided morphological parsers, the algorithms used are based on rules derived from linguistic analysis. The present system can identify the syntactic category of the head word of a verb phrase, and, if it is a verb, the conjugation group, even when not listed in the dictionaries of the system. This ability of the system enables quick access to the dictionary in the category of the head word. The design is object-oriented, modeling linguistic constructions and the components, and it is thus easy to grasp the structures and algorithms, enhancing scalability, maintainability and portability. The system can be embedded in a larger system, and be used when the larger system starts parsing verb phrases. Keywords: Morpho-semantic analyzer, Japanese verb phrases, Morphological parser, Rule-based, Object-oriented, Parsing 1. Introduction The purpose of the current morpho-semantic analyzer is to understand the semantic structures of Japanese verb phrases, and to contribute to the semantic understandings of sentences. At present the system understands the morpho-semantic structures of the verb phrases of simple sentences. It employs the algorithms based on linguistic analyses and several lists of carefully grouped verbs. Being able to understand verb phrases semantically, it differs from other Japanese morphological analyzers, notably Juman (Kurohashi and Nagao, 2003) and Chasen (Matsumoto et al., 2000), which focus primarily on the segmentation of sentences into morpho- phonemes such as prefixes, suffixes, inflections, Case particles and the components of compound words, and on labeling them with syntactic classifications. Another major difference from other morphological analyzers is that the current system is exclusively rule-based while most analyzers, whether morphological or syntactic, are based on statistics, such as Uchimoto et al. (2003) and Murata et al. (2005), to name a few. The third difference is that the present system is able to identify the syntactic category, such as the adjective and the verb, of the head word of a verb phrase, and, if it is a verb, the conjugation group, even when it is not listed in the dictionaries of the system. The current system, unlike finite state models, is easy to trace the problems and extend itself according to needs, because each step in the algorithms is linguistically accountable and * Copyright 2008 by Yukiko Sasaki Alam 101 22nd Pacific Asia Conference on Language, Information and Computation, pages 101–112 intuitively comprehensible. It parses the Japanese verb phrases of simple sentences, but not those with modal and honorific expressions and those in complex sentences. It can be embedded in a larger parsing system such …

Content truncated to 500 words