Three-Word Dependency Relations and Their Application to Structural Ambiguity Resolution

摘要

Resolving syntactic ambiguities is an important issue in natural language processing. By measuring the strengths of association that hold among the words in co-occurrence relations, one may be able to determine the correct syntactic structure for an ambiguous construction. In this paper, we propose a syntactic disambiguation method that uses lexical preferences estimated in three-word dependency relations and, at the same time, conduct a disambiguation experiment in Japanese noun phrases containing the particle no. The result shows a better performance than those of other methods, and indicates the applicability of the method to resolving other syntactic ambiguities that appear in coordinated constructions, prepositional phrase attachments, and the like.