Competition in learning language-based categories

Abstract

Competition is a property of cognitive processing which results from learning that is characterized by the indeterminate encoding of instances and from processing that is characterized by the mutual influence of all activated representations on one another. In the present study, non-French participants learned gender-appropriate adjectives (petit or petite) for a set of 24 French nouns. We found that learning the same set of feminine French nouns could be made more or less difficult when the nouns in the masculine category created more or less competition. One measure from the Competition Model of MacWhinney and Bates (Bates & MacWhinney, 1987, 1989; MacWhinney, 1985, 1987) – cue reliability – predicted these competition effects. We tested an alternative measure based on the encoding of the nouns in memory (termed exemplars) and found that it predicted participants' mean learning performance somewhat more accurately. In the final section of this article, we extend our exemplar-based measure to a connectionist network in order to account for competition and the time course of learning. The network provided a superior fit to the data, with an average R2= .91.