Exploring the use of linguistic analysis for answering why- questions

In the current project, we aim at developing an approach for automatically answering whyquestions (why-QA). In the present paper, we investigate the relevance of linguistic analysis for why-QA. We focus on two tasks: the use of syntactic information for answer type determination and the use of discourse structure for the extraction of possible answers from retrieved documents. For answer type determination, syntactic analysis appears to be of significance: we obtain 77.5% performance using a method based on syntactic parses by the TOSCA parser— compared to 58.1% using a comparable approach without syntactic analysis. Discourse analysis appears to be very relevant for extraction of potential answers to why-questions. We performed a manual analysis of 336 question-answer pairs and the corresponding RST annotated texts. We found that for 58.9% of why-questions, the RST analysis of the source text can lead to a correct answer to the question.