CONSTRUCTING A PHRASE STRUCTURE GRAMMAR BY INCORPORATING LINGUISTIC KNOWLEDGE AND STATISTICAL LOG-LIKELIHOOD RATIO

Phrase structure grammar is one of the most important components in a syntax-oriented parsing system. However, constructing an adequate PSG is an arduous task. Either the traditional linguistic approach or the fully automatic inference approach has encountered several difficulties. Thus, a human-machine cooperative method is suggested in this paper as a better approach. A statistical tool, Log-Likelihood Ratio, is proposed to enhance the productivity of human grammar writers. The Log-Likelihood Ratio of co-occurring tags is automatically computed by the computer to indicate the strength of linear association. The task of linguists is then to verify the relevance of groupings based on their linguistic knowledge. The advantages of this approach over other methods are pointed out, and the actual procedures are illustrated by a pilot experiment of constructing a Mandarin PSG. The experimental result shows the feasibility of the proposed approach.