Arguments analysis of Chinese verb subcategorization based on weighted gap subsequence kernel function

The paper proposes a new arguments analysis method for Chinese verb subcategorization to improve the present performance of analyzing Chinese verb subcategorization frames (SCFs). The method introduces the weighted gap subsequence kernel function into the analysis, and treats the left part and the right part of the traditional rules as training samples and related categories respectively, transforming the originally rule-derived problem into the machine-learning problem. This new kernel can take more cross-distance features. Compared with the rule based methods, the weighted gap subsequence kernel based method improves the precision of argument type analysis from 55.16% to 93.43% on syntactic noisy data. The analyzing performance of whole sentence is also much improved.