Procedural learning: I. Locus of practice effects in speeded choice tasks.

When subjects perform 750 trials of a speeded choice task mapping different categories of symbols (e.g., letters, digits) onto different responses, excellent transfer is observed to new items in the trained categories (Experiment 1). However, when arbitrary sets of stimuli are mapped onto each response, introducing new stimuli substantially retards performance (Experiment 2), even when the size of the potential stimulus set remains constant (Experiment 3). Surprisingly, responses to already trained items are as slow as responses to new items in these transfer tests. When the mapping is categorical, shuffling the assignment of stimuli to responses drastically slows responses (Experiment 4). However, changing to a spatially homologous mapping with responses on the other hand produces excellent transfer (Experiment 5). Together, these results indicate that practice in speeded choice tasks affects primarily the response selection stage, rather than perceptual processing or motor responses. These data suggest that practice primarily strengthens links between category representations and spatially defined responses. Furthermore, when an arbitrary collection of symbols is mapped onto a given response, practice produces an ad hoc category representation and strengthens links between individual items and the category, as well as links between the category and the response. With practice, people become dramatically faster and more accurate in sensorimotor skills. Characterizing these changes is of major importance, for both applied and theoretical purposes. In applied situations, it would be useful to know how conditions of skill acquisition can be structured so that learning is most efficient and to know when one form of training will produce useful transfer to other, related tasks. For theoretical purposes, understanding the functional character of skill learning should provide a basis for developing detailed models of the changes that underlie such learning (such as neural net models). The present article focuses on the changes in performance that occur with modest amounts of practice in very simple sensorimotor tasks, namely those requiring speeded choice responses. In speeded choice tasks, subjects are presented with a single stimulus and must rapidly select and execute a response depending on the identity of the stimulus. This is one of the simplest paradigms in which one can examine procedural skills and their development, and it provides a useful laboratory model for the numerous human skills that involve retrieving and producing arbitrary responses to stimuli. Performance in speeded choice tasks is subject to various kinds of change as subjects perform repeated trials. Subjects' responses become faster (and often more accurate)—the prac