Explorations with a discrimination net model for paired-associate learning

Abstract

A computer simulation model for paired-associate learning is presented. The model emphasizes stimulus discrimination learning, and is based on an EPAM-type discrimination net which grows according to stochastic processes. Group data are simulated for comparison with human data. Three versions of the model are presented. The simplest, SAL I, was first run in several single-list experiments and evaluated on such topics as: intralist similarity, number of response alternatives, probability learning, spacing of item repetitions, whole vs part learning, and the effects of list length. SAL II, a slightly modified version of the model which allows overlearning, was run in retroactive interference experiments varying interlist similarity and degrees of original and interpolated learning, and is discussed in relation to studies of predifferentiation and transfer. Finally, SAL III, which stores multiple associates, was run in experiments on recognition vs recall, second and third guesses, unlearning and modified free recall, retroactive vs proactive interference, distributed practice, and response latencies.