Exploiting syntactic relations for question answering

Abstract

Recently there has been a resurgent interest in syntax-based approaches to information access, as a means of overcoming the limitations of keyword-based approaches. So far attempts to use syntax have been ad hoc, choosing to use some syntactic information but still ignoring most of the tree structure. This thesis describes the design and implementation of SMARTQA, a proof-of-concept question answering system that compares syntactic trees in a principled manner. Specifically, SMARTQA uses a tree edit-distance algorithm to calculate the similarity between unordered, unrooted syntactic trees. The general case of this problem is NP-complete; in practice, SMARTQA demonstrates that an optimized implementation of the algorithm can be feasibly used for question answering applications.