Research Of Chinese Textual Entailment Recognization

With the explosive growth of the Internet information,how to make computer understand natural language profoundly, rather than just dealing with the surface of the sentence, is becoming the challenge of the researcher of natural language process. The Internet information is growing, at the same time, the meaningless information is growing, too.How to make computer automatically understand and process the text is becoming the research direction of researchers. In the domain of natural language process, the technology of analyzing the relationship between texts attract researchers' attention, recognition of textual entailment is one of the technology.Recognition of textual entailment is given two sentences, judge one of the sentence whether or not infer another. If it can infer another, then the entailment relationship is entail, or else the relationship is not entail.Recognition of textual entailment is widely used in question answer,machine translate, automatic abstracting, information retrieval and so on.By analyzing the defects of the traditional textual feature on word similarity,this paper proposes a novel Chinese textual entailment recognize method that based on ordered word mover distance, this method compute ordered word mover distance based on word2vec, then use ordered word mover distance feature and traditional textual feature to generate classification module based SVM and so on. With the use of classification module, obtain the entailment result. This article conducted an experiment in the CS data of RITE-VAL evaluation task in 2014, the MacroF1 of the experiment is 0.629, outperform optimal value (0.615, BUPTTeam), which illustrated the effectiveness of the method to lifting the performance of Chinese textual entailment.The main contribution of this paper is as follows,1. Propose a method to the text unification based on synonym The previous method did not unify the synonym, ignoring the diversity of natural language. As a consequence, this paper use synonym to unify the text. can reduce the diversity of language description, and improve the correlation of the text. This is a groundwork for feature extraction and disaggregated model building.2. Propose the method to process the entity recognition result. Most of the previous method did not consider the inconsistent result of entity recognition, which causing interference when the system judging the entailment relationship. For that reason, this paper process with the entity recognition text.3. Propose a novel Chinese textual entailment recognize method that based on ordered word mover distance, previous method ignored the word order and the similarity between words. This method compute ordered word mover distance based on word2vec and word order information. The experiment result illustrated the effectiveness of the method to lifting the performance of Chinese textual entailment.