Finding information is a problem shared by people and intelligent systems. This paper describes an experiment combining both human and machine aspects in a knowledgebased system to help people find information in text. Unlike many previous attempts, this system demonstrates a substantial improvement in search effectiveness by using linguistic and world knowledge and exploiting sophisticated knowledge representation techniques. It is also an example of practical subsumption technology on a large scale and with domainindependent knowledge. Results from this experiment are relevant to general problems of knowledge-based reasoning with large-scale knowledge bases.
William A. Woods