Deductive Question-Answering on Relational Data Bases

The principal concern of this paper is the design of a retrieval system which combines current techniques for query evaluation on relational data bases with a deductive component in such a way that the interface between the two is both clean and natural. The result is an approach to deductive retrieval which appears to be feasible for data bases with very large extensions (i.e. specific facts) and comparatively small intensions (i.e. general facts). More specifically, a suitably designed theorem prover "sweeps through" the intensional data base, extracting all information relevant to a given query. This theorem prover never looks at the extensional data base. The end result of this sweep is a set of queries, each of which is extensionally evaluated. The union of answers returned from each of these queries is the set of answers to the original query.