Active Semantic Networks as a Model of Human Memory

David E. Rumelhart Donald A. Norman Department of Psychology University of California, San Diego La Jolla, California 92037 Abstract A general system to simulate human cognitive processes is described. The four-part system comprises a nodespace to store the network structure; a supervisor; a transition network parser; and an interpreter. The method by which noun phrases operate and the process for the determiner "the" is presented. An analysis of verb structures illustrates how network structures can be constructed from primitive verb definitions that get at the underlying structures of particular verbs. The paper concludes with an illustration of a problem in question-asking. A Model of Human Memory We have constructed a large general simulation of human language and long-term memory on the premise that the study of the interrelationships among psychological processes will lead to more insight into human cognition and memory. The general implementation is basically complete, and a variety of users are starting to study specific psychological tasks (language understanding; children's development of language; primitive verb structure; reading; inference; game playing--Go and Gomoku; visual representation and memory; learning; and question answering). It is still too early to report on the results of the psychological investigation.. Therefore, this paper is a progress report on the system and the underlying psychological principles. The major guidelines have come from our attempts to represent long-term memory structures. We know that people rapidly forget the details about the surface structure of an experience but retain the meaning or interpretation of that experience indefinitely.