A REPRESENTATION SHIFT MODEL OF CLASSIFICATION AND ABSOLUTE IDENTIFICA TION

Abstract

We studied shifts of representation using production trials to measure the representation following classification and absolute identification. Identification induced a shift of representation away from the preceding exemplar. Shifts in classification depended on the category of the preceding trial. When the exemplars were from different categories, the category representation was pushed away from the preceding category. When the exemplars were from the same category, the representation of the current category was pulled from its centre in the direction of the preceding exemplar.

We propose a representation shift model to account for trial-to-trial errors in both tasks. The model treats the subject’ s representation of category and item structure as points in a space. The points shift dynamically from trial to trial as a function of the properties of the preceding exemplar.

The RSM captures overall accuracy in both classification and absolute identification, replicates benchmark characteristics of performance in both absolute identification, and classification.