Theories of the structure of the subjective lexicon have been based almost exclusively on facts about the relations that structure various semantic domains. The basic fact of polysemy has not been considered in the formulation of models of a mental dictionary. In this paper we argue that if polysemy is taken into account, as it must be, a certain class of theories of the structure of the subjective lexicon are shown to be inadequate. We also present data from several experiments on polysemy that are consistent with a'process' model of meaning representation. In this model it is proposed that the specific surface sense of a word is determined not by finding a stored representation of it in memory, but by constructing it from a core meaning. This general and abstract core meaning representation is presumed to underlie all uses of a polysemous word. Sets of instruction rules operate on core meanings to produce the various surface senses. The application of an instruction rule is constrained by the relation that the final semantic representation will have to factual information a person has stored about the world.

Only in recent years has the issue of the representation of meaning become a respectable research problem in psychology. Under the general rubric'semantic memory', a term coined by Collins and Quillian (1969), a growing stream of reports have been published that implicitly or explicitly address the question of the organization and retrieval of semantic information. Already, the relative merits of several theories of semantic memory are the focus of a heated debate (Collins and Loftus 1975; Holyoak and Glass 1975; Smith,