A Hybrid Approach to Textual Entailment Recognition

The task of textual entailment recognition is to determine whether a text entails a hypothesis. This paper proposes a hybrid technique to identify the entailment relation between texts and hypothesis. This technique includes an approach based on lexical similarities and an approach based on the classifier of support vector machine. The approach based on lexical similarities is to use the similarities between a set of words within a text and a set of words within a hypothesis. The approach based on the classifier means to treat this task as a classification problem. We propose two kinds of classification features which include features based on semantic roles, and ones based on dependency relations and WordNet. We use our hybrid technique to integrate the two sets of experimental results by the lexical similarities-based approach and the SVM classifier-based approach. The experimental results demonstrate that our technique is effective to solve the problem of textual entailment recognition.