Toward a conceptual grammar and lexicon

Abstract

Computers have made possible the collocation and rearrangement of the multitudinous tokens and relationships of language but as yet we have little in the way of useful methods for such manipulations. Probably the most systematically abstracted body of knowledge about language is in grammars. Many computational linguists, on the assumption that syntax must of necessity reflect human cognitive processes of prime importance, have thrown their energies into developing programs that will automatically identify or utilize grammatical relationships. These programs hope to deal with semantic aspects of language by incorporating data bases which specify the variety of conditions under which particular words may be used in particular grammatical contexts. But there is little reason to believe that grammar as we know it parallels cognitive processes or that it offers a sounder basis for a psychologically relevant analysis of texts than non-grammatical approaches which, for example, employ sentence length, type-token ratio, vocabulary frequency, or any of a host of other measures. Data base, to be adequate to unsystematic and metaphoric word usages, would surely have to exceed in size our largest dictionary, since it must specify all the conditions under which a word may and may not be used, rather than its dictionary meaning. This paper proposes and illustrates a cognitively relevant approach to the analysis of texts, utilizing a concept grammar and a concept lexicon. The concept lexicon is currently employed in a computer program for the automatic analysis of content. The grammatical ideas are at present mere speculation. Two familiar psychological principles are (1) that a fundamental cognitive response to stimuli is to group, classify, or categorize them and (2) that stimuli which occur in close contiguity with each other, even though of different classifications, are associated cognitively. These two principles reiterate Aristotelian ideas (Warren, 1921) about similarity and contiguity as bases for association of stimuli. Contrast was a third feature in the Aristotelian system, but may readily be subsumed under similarity. The two principles suggest that texts might be analyzed and compared for those items that link together conceptually on the basis of some element of similarity, and for those linked by having occurred in the same context.