Chinese Subcategorization Annotation Based on Machine Learning

Abstract

There have been a lot of researches focusing on large-scaled automatic acquisition of subcategorization frames, and many achievements have been made for lexicon building in quite a few languages, but subcategorization annotation for individual sentences still remains in a rarely touched field. This paper proposed to annotate Chinese subcategorization as a classification task by means of sequence kernel methods, which exploited the potential relations among the respective sentential constituents. Our final classification with word sequence kernel congregation and POS sequence kernel C-SVM achieved a very promising accuracy ratio of 92.36% on the testing set, which is 13.51% higher than the baseline performance of the existing Chinese SCF hypothesis generator.