Natural Language Inference via Dependency Tree Mapping: An Application to Question Answering

We describe an approach for answer selection in a free form question answering task. In order to go beyond a key-word based matching in selecting answers to questions, one would like to develop a principled way for the answer selection process that incorporates both syntactic and semantic information. We achieve this goal by (1) representing both questions and candidate passages using dependency trees, augmented with semantic information such as named entities, and (2) computing a generalized edit distance between a candidate passage representation and the question representation, a distance which aims to capture some level of meaning similarity. The sentence that best answers a question is determined to be the one that minimizes the generalized edit distance we define, computed via a dynamic programming based approximate tree matching algorithm. We evaluate the approach on question-answer pairs taken from previous TREC Q/A competitions. Preliminary experiments show its potential by significantly outperforming common bag-of-word scoring methods.