Adaptive Resonance Theory as a model of polysemy and vagueness in the cognitive lexicon

Abstract

AbstractCognitive linguistic analyses of the relationship between ambiguity and vagueness suggest that these categories are gradable. That is, whether a given distinction in lexical meaning amounts to ambiguity can vary across contexts. A computational model of the representation of lexical polysemy was applied to three nouns, and the verb paint. The model, based on a neuropsychologically inspired framework, Adaptive Resonance Theory, was applied to judgements and property ratings from 120 participants. Model results were compared to an independent usage-based analysis, and in the case of paint to an earlier cognitive linguistic analysis. Good correspondence was found, with the model reproducing specific features of those analyses, such as the extraction of similar schemata, as well as the key general feature of the gradability of ambiguity and vagueness. At the same time, the analysis revealed substantial interpersonal variability. Additional modelling demonstrated that the availability of specific aspects of conceptual knowledge could be contextually modified, leading to predicted changes in the pattern of ambiguity. Because Adaptive Resonance Theory uses elementary neural processing, previously used to successfully model cognitive phenomena, the gradability of ambiguity can be explained as a consequence of these general properties of memory and cognition.