An error driven method to improve rules for the recognition of Chinese modality “LE”

Abstract

We have a “Trinity” way for the recognition of Chinese modality “LE”, in which dictionary, usage rule base and usage corpora combine as the knowledge base. Handcrafted rules can hardly cover all usages in the real texts. So this paper proposes an error driven method for the automatic rules improvement. Experimental results show that, after the automatic rules improvement, the recognition precision of the modality “LE” improves by over 1.85%.