Chinese Textual Entailment recognition model based on lexical and semantic matching

Recognizing Textual Entailment (RTE) is one of the fundamental problems in many natural language processing applications. In this paper, a novel feature extraction method is proposed for the Chinese textual entailment recognition task in NTCIR. Firstly, we extract lexical and semantic features. Two lexical semantic dictionaries are used, and a novel semantic feature extraction method based on Stanford Parser and part-of-speech tagging is proposed. Furthermore, three different classification algorithms are applied and compared, rule based algorithm, SVM and C4.5. Experiment results show that C4.5 gives the best performance. Evaluation of the proposed approach on NTICR-9 RITE development data set shows promising precisions of 70.1% in binary classification and 55.9% in 5-class classification.