Although there has been a wealth of research into the decay properties of procedural memories, relatively little of this research stands on firm theoretical footing--in particular, while the issue of training duration has been investigated, the definitions used to distinguish "overtraining" from ordinary training are generally quite vague. On the other hand, the SPEED model of automaticity in category learning constitutes a well-encapsulated paradigm for describing the effects of overtraining in terms of a shift in the neural locus of behavior; this model was fitted to motivating behavioral data, and has since been tested using fMRI. However, the model only specifies learning processes, i.e., there are no characterizations of the decay that presumably affects all human memory systems. The present study is designed to examine the processes of memory decay, from within the framework of the SPEED model, and to elucidate possible decay properties of the neural systems underlying procedural learning and memory in this model.

Using a well-described category-learning task, participants were trained for varying durations chosen to load preferentially on each of the two systems posited by the model. This variable was crossed factorially with testing interval: subjects returned for unsupervised test sessions either monthly, quarterly, or yearly over the one-year duration of the experiment. This allowed both for querying the timecourse of decay at varying granularities, as well as investigating so-called "re-test effects." Results demonstrate a substantial difference in the retention performance of the two training groups, quantitatively and qualitatively. Moreover, although sample size is limited, there is some evidence for differential re-test effects in the two groups, such that frequency of re-test had no measurable impact in one training group but showed a trend in the other group. These results are interpreted in terms of the motivating SPEED model, and suggestions for updating the model on the basis of the present findings are given.