The abstraction of covariation in conceptual representation
Ken Richardson, T. Carthy · British Journal of Psychology · 1990
Over the last few years considerable doubts have been cast on traditional models of conceptual representation, with more recent models stressing the flexible, rather than the static, structure of concepts, and the relations among feature variables rather than these being represented independently of each other. Largely for methodological reasons, however, the investigation of such relations has so far been confined to the special case of feature co‐occurrences, whereas, we argue, feature relations in the natural world are likely to be less invariant and less linear than this. The present paper considers ‘fuzzy’ covariations among feature variables as the source of property relations in concepts and describes the use of log‐linear models for examining them. It extends previous work (e.g. Richardson, 1987) by focusing on specific covariations (between pairs of variables), rather than looking only at ‘overall effects’ of covariation representation. In two experiments, subjects experienced ‘fuzzy’ covariations between pairs of variables in exemplars of an artificial category. Representation of this experience was assessed in delayed recognition tests. Representation in terms of ‘fuzzy’ covariation abstraction (as predicted by a suitable log‐linear model) was compared with representation expected by other models of conceptual representation. In both experiments, the results were consistent with abstraction of the ‘fuzzy’ covariations. In the general discussion section, some theoretical speculations are offered about the further roles of covariation abstraction in representation and reasoning.