ABSTRACT
Researchers in natural language processing have long been concerned with the problem of making the inferences necessary for text understanding. Much of the initial research on inference for natural language experimentally addressed the question of how to apply particular kinds of knowledge to the understanding task. For example, Schank (1975) discussed a mechanism for making those inferences that could be organized around language primitives; Cullingford (1978) described a mechanism to process stories involving scripts (Schank & Abelson, 1977); Wilensky (1978) dealt with texts involving plans and goals. Charniak (1972) proposed a demon-based approach to knowledge application, and Charniak (1978), a frame-based one.
