A text inference based answer extraction for Chinese question answering

Abstract

Answer Extraction is the core problem of question answering and determines the performance of the Question Answering to a large extent. In this paper, we propose a text inference based answer extraction approach that can recognize the inferable relations automatically. In the inference model, lexical, syntactic and semantic inference relations are taken into account. We also build a voting system of inference to make the judgment if the question can be inferred from the answer sentence. Experimental results show that the text inference-based method for answer extraction achieve an increasing 8.5% accuracy and 9.6% MRR performance of the QA system for only supported documents.