Recognizing Text Entailment Based On FrameNet Relations

Recognizing Textual Entailment is an effective way to process the synonymous shaped phenomenon widespread in natural language. And it is basis and important in the field of Natural Language Processing. Improving the efficiency of Recognizing Textual Entailment plays an important role in Information Retrieval, Information Extraction, Question Answering and Summarization. After analysis the domestic and aboard research, this thesis proposes the way which recognizing the textual entailment by using the Frame and Frame-Relation based on FrameNet.This thesis aims to research how to use the knowledge of the FrameNet for improving the performance of Recognizing Textual Entailment system. This thesis takes frame semantics as the theory basis, delves into the Frame, Frame-Relation and Frame Element in FrameNet, determine the knowledge of the FrameNet which is available to recognizing textual entailment. And then, formalize description of the knowledge, construct the graphs of the Frame entailment relations, map the Frame Element between the two Frames with semantic relations. Based on the lexical resource FrameNet, this thesis use the method, which finds the path between the frames evoked by the lexical units in text T and hypothesis H in the Frame Entailment Relation Graph using depth-first Search Method in order to identify the hyponymy relationships between the frames and compares the content of span which are filled the mapping FE slots, for recognizing textual entailment.The experiment is based on the corpus of the Third Recognizing Textual Entailment in2007. The experimental results show that the precision and the recall are77.06%and56.13%, the precision of the method is only below the best result of the RTE-3. And the experimental results show that the lexical resource FrameNet if helpful to improving the efficiency of Recognizing Textual Entailment.