A CASE FOR NONDECOMPOSITION IN CONCEPTUALLY DRIVEN WORD RETRIEVAL

Theories of lexical access in language use of a productive nature such as speaking, writing and verbal thinking differ in whether they assum e that words are retrieved from memory in a conceptually decomposed or nondecomposed manner. Decomposition has been the received view for ce nturies, while nondecomposition is mostly not taken very seriously-und eservedly so, as I demonstrate in this paper. I review several theoret ical objections that have traditionally been raised against nondecompo sition and indicate how a nondecompositional approach can cope with th em. Furthermore, several theoretical arguments in favor of nondecompos ition are given. The issues concern the componential analysis of word meanings, the conceptual primitives, word definitions, the acquisition of word meaning, the conceptual dissection of messages, context depen dence of word meaning, decomposition for syntactic encoding, word-to-p hrase synonymy, hyperonymy, hyponymy, and the locus of decomposition. In addition, the major computational models of conceptually driven wor d retrieval proposed during the last few decades are evaluated both in formally and by computer simulation. The decompositional models are sh own to fail, whereas a specific nondecompositional model is shown to a ccount for the difficulties. It is concluded that there are no evident ial grounds for rejecting nondecomposition. On the contrary, for a the ory of word retrieval there are, instead, good reasons to prefer nonde composition to decomposition. Nondecomposition should be given more se rious consideration in future work in the field.