11 Theorem-Proving by Resolution as a Basis for Question-Answering Systems Cordell Green

ABSTRACT This paper shows how a question-answering system can be constructed using first-order logic as its language and a resolution-type theorem-prover as its deductive mechanism. A working computer-program, Q A3, based on these ideas is described. The performance of the program compares favorably with several other general question-answering systems. The type of questionanswering system considered in this paper is ideally one having the following features: 1. A language general enough to describe any reasonable questionanswering subjects and express desired questions and answers. This paper argues the case for formal methods to achieve such a system and presents one particular approach in detail. The name'question-answering system' requires clarification. The system described above might be named an'advice taker' or a'multi-purpose problem-solving system' or'general problem-solving system'. McCarthy (1958) proposed using formal languages and deduction to construct such a system, and suggested allowing the user to give hints or advice on how to answer a question; he referred to the proposed system as an'advice taker'. Research on'multi-purpose' or'general problem-solving' tends to differ from questionanswering as described above by placing more emphasis on solving deeper, more difficult problems and less emphasis on user interaction, formality, and efficient retrieval of relevant facts from a large data base. The situation is further confused by the use of'question-answering' to refer sometimes to natural language systems, sometimes to information retrieval systems having little deductive ability, and sometimes to systems with deductive ability limited to the propositional calculus. It is important to emphasize the distinction between general versus specialpurpose question-answering. If the class of questions asked of a system is small, completely specified in advance, and concerned with a particular subject area, such as the question-answering system of Green, Wolf, Chomsky, and Laughery (1963) concerned with baseball, or that of Lindsay (1963) concerned with family relations, then we shall call such a system'specialpurpose'. Frequently the goal in designing a special-purpose system is to achieve good performance, measured in terms of running speed and memory utilization. In this case the best approach may be first to construct a special data base or memory that is optimized for that subject area and question class, and then to write special question-answering subroutines that are optimized for the particular data base and question class.