Research On Chinese Textual Entailment Recognition

As one of the hottest area in current research, Textual Entailment Recognition plays a significant vital role in different Natural Language Processing(NLP) tasks. In this paper, we proposed three methods to enhance the ability in Chinese textual entailment.1. Chinese textual entailment recognition based on features fusion.This method employs Na?ve Bayes and Support Vector Machine as classifier. With statistical features, lexical semantic features and amending module, System gets the final classification results. In the feature selection process, the non-structural features which are beneficial to the identification of the implication relation and the semantic features obtained from the semantic resources are used as antecedent features.2. Chinese textual entailment recognition based on syntactic trees clipping.This feature make up the traditional features’ lack of syntactic information. By aggregating the nodes of parsing trees, useless node will be deleted. Two minimal subtree generates and their similarity will calculate. Compared with original syntactic trees, the minimum information trees with less nodes without losing syntactic information about entailment relationship.3. Chinese textual entailment recognition based on word embedding.Due to the analogy of word embedding can effectively identify the entailment relationship between words, with Word2 vec training, a new word embedding can be used to judge whether there is an entailment relationship between words. According to the known relationship between words, the same relationship will be calculated under our new algorithm.In addition, during our research, we participated an international evaluation of textual entailment recognition. Systems that achieved with methods features fusion and syntactic tree clipping got a 59.71% in F-measure.