Dependency Parsing and Its Application using Hierarchical Structure in Japanese Language

Abstract—Conventional Japanese dependency parsing methods are primarily based on the bi-nominal model between phrases, which has a limitation related to the order of phrases. Accord-ingly, the length of a phrase, which depends on the language de-pendency, is limited. In this paper, we propose a novel dependency parsing method for Japanese based on an extended hierarchical language model simulated by the hierarchical Pitman-Yor process in order to overcome the abovementioned limitation. We also eval-uate the accuracy of the proposed dependency parsing method as well as its practicality to detect the topics of each document. Experimental results show that the proposed method can parse dependencies in long, complex sentences and can allocate topics to each document relatively well compared with the conventional method. Consequently, it can be said that the proposed method is feasible in the research fields of both Japanese dependency parsing and topic modeling