Emmanuel M. Pothos & Todd M. Bailey

Abstract Generalisation to novel stimuli can be seen as being driven by either a similarity-to-old-exemplars mechanism (e.g., Nosofsky, 1989) or a mechanism involving some abstract, rule-based representation of the instances encountered so far (see Chater & Hahn, 1997, for a review). Artificial Grammar Learning (AGL) has provided a suitable experimental framework to address this issue (Reber, 1989, Brooks & Vokey, 1991). In this work, we have tried to investigate the extent to which Nosofsky's Generalised Context Model (GCM; Nosofsky, 1989) of categorisation can account for AGL competence results, improving on previous research whereby the measures of similarity employed to model AGL performance have been criticised as being insensitive (Knowlton & Squire, 1996; Redington, 1997, 1996). Our results provide support for the hypothesis that both adherence to the rule structure of the grammar used to construct grammatical items, and similarity of test items to training items independently motivated subjects' grammaticality endorsements in test.