Similarity Effects in Artificial Grammar Learning

Theresa Johnstone and David R. Shanks
Department of Psychology
University College London

Abstract Two experiments investigated the extent to which classification judgments are influenced by knowledge of abstract rules versus perceptual similarity in artificial grammar learning, using the biconditional grammar and Match and Edit tasks designed by Mathews, Buss, Stanley, Blanchard-Fields, Cho and Druhan (1989, Experiment 4). In both experiments a Match group memorisd letter strings, while an Edit group hypothesis tested in order to actively identify the rules of the grammar. Both groups processed the same training strings and classified the same novel test strings. In Experiment 1 whole item similarity was manipulated independently of grammaticality, whereas in Experiment 2 the similarity of fragments of two letters (bigrams) and three letters (trigrams) was manipulated independently of grammaticality. The Match group showed no effect of grammaticality in either experiment and no effect of whole item similarity in Experiment 1. However they did show an effect of fragment similarity in Experiment 2. In contrast, the Edit group showed effects of grammaticality in both experiments and no effects of similarity in either experiment. These results suggest that whether similarity or rule based classification occurs depends on the specific processing demands of training tasks.