Abstract: The accuracy of event patterns is very important in semi-supervised event extraction.Currently,semi-supervised Chinese event extraction system based on the pairwise pattern (e.g.,Trigger-Argument) suffers much from the issues of polysemy of triggers and sparse patterns.This paper put forward a argument-based mechanism to solve trigger sense disambiguation,and then applied it to pattern filtering to eliminate invalid patterns.In addition,for several special Chinese sentence structures,this paper proposed a pattern conversion method based on syntactic structure to enhance the applicability of the pattern.The experimental results on the ACE 2005 Chinese data show that our methods can effectively improve performance of semi-supervised Chinese event extraction system.

Key words: Event extraction,Pattern filtering,Pattern conversion