日本語新聞記事解析における構文情報および意味情報の抽出法

抄録

A natural language understanding system is described which extracts syntactical and semantic information from Japanese newspaper articles. The understanding system parses and understands the input sentences and constructs an intermediate representation in CRS(Contextual Representation Structure)which is independent from the specific languages. The system employes a syntactic analyzer and semantic analyzer. These analyzers operate in a serially lntegrated fashion. The syntactic analyzer(context free grammar parser)analyzes the input sentences and obtains the parsing trees. Since the phrase order in a clause is grammatically unrestricted in Japanese sentences, and since, unlike in English, positional information regarding phrases is usually no help in establishing their relation, the Japanese analyzer is unable to determine phrase relations. So, this Japanese analyzer extracts an rough sketch of a structure of the sentences and the semantic relations between words are determined in the next semantic process. The syntactic structure adopts the right recursive tree structure to represent this Japanese unrestricted construction. This tree structure contains the all ambiguity in the syntactic analysis. From an analysis of these, in turn, the semantic analyzer obtains word-level semantic structures. The semantic analyzer constructs a semantic structure for each phrase. Each word meaning shares a suitable position in the hierarchy of concepts. In this system, a word meaning written in the lexical entry plays an important role in semantic analysis. The interaction between the word meanings enables the check if a word can be embedded in the slot in the other word. If the check is satisfied, the analyzer embeds the word in a suitable slot in the other word. The modifying ralations inside noun phrases and the case realtions among verbs and noun phrases are determined in the word-level semantic structure.