The University of Sussex

Deriving answers to logical queries by answer composition: a new approach to deductive question answering

R.J. Gaizauskas

This thesis presents a new approach to deductive question answering based on a theoretical analysis of the compositional properties of logical queries. We review the automated theorem proving, logic programming, and deductive database traditions from the perspective of deductive question answering and from this review evolve a view of how queries, answers, and logical databases should be defined, both for definite clause and full first order deductive question answering systems. Given this view we address the question of how answers to a complex query are determined by answers to the components of the complex query, themselves viewed as queries. Investigation of this question leads to a series of compositionality results, again both for definite clause and first order answering systems. A notion of answer derivation, based on the first order compositionality results, is introduced and is shown to be sound and complete. This notion of answer derivation provides the formal logical basis for a new approach to automated deductive question answering, one which naturally lends itself to parallel computation. A serial algorithm for computing answer derivations is described and the results of running an implementation of it on selected examples are discussed, demonstrating that the theoretical approach to answer derivation is computationally plausible. Some remarks are made about developing a parallel algorithm.

This paper is not available online