A Semantic Associational Memory Net That Learns and Answers Questions
Stuart C. Shapiro, G. H. Woodmansee, Myron W. Krueger
This paper describes a general semantic memory structure and associated executive functions, which, using the memory, are capable of learning and answering questions. The system is capable of three types of learning: by being explicitly told facts; by deducing facts implied by a number of previously stored facts; by induction from a given set of examples. The memory structure is a net built up of binary relations, a relation being a label on the edge joining two nodes . The relations, however, are also nodes and so can be related to other nodes which may also be used as relations. Most significantly, a relation can be related to one or more alternate definitions in terms of compositions of other relations and restrictions on intermediate nodes. Throughout this work an attempt was made to maintain complete generality and thus allow the system to be used in a wide variety of applications without change. In fact, it could be used for several different purposes simultaneously. The system, as described, is presently programmed in SNOBOL, a string manipulation language, and is running at the University of Wisconsin on a C.D.C. 3600 computer.
Download this report (PDF)
Return to tech report index