Automatic Recognition of Chinese Separable Words Based on CRFs

摘要

Currently, most of the automatic recognition tasks of separable words adopt a rule-based method, which relies on automatic word segmentation results and lexical patterns generated from common inserted constituents. However, they suffer from incorrect word segmentation results and inaccurate and limited rules. Moreover, they ignore the rich information contained in the context. To address these issues, this paper proposes a CRFs-based method which employs nine features, such as character, POS tag, punctuation, word boundary, keyword and POS sequential rule. Experimental results on real-world datasets show that our approach can make full use of rich information and achieve significant improvements on recognition efficiency compared to all the baselines.