Abstract

There is evidence that rule-based category learning is supported by a broad neural network that includes the prefrontal cortex, the anterior cingulate cortex, the head of the caudate nucleus, and medial temporal lobe structures. Although thousands of studies have examined rule-based category learning, only a few have studied the development of automaticity in rule-based tasks. Categorizing by a newly learned rule makes heavy demands on declarative memory, but after thousands of repetitions rule-based categorizations are made with no apparent effort. Thus, it seems likely that the neural systems that mediate automatic rule-based categorization are substantially different from the systems that mediate initial learning. This research aims at identifying the neural systems responsible for early and late rule-based categorization performances. Toward this end, this article reports the results of an experiment in which human participants each practiced a rule-based categorization task for >10,000 trials distributed over 20 separate sessions. Sessions 1, 4, 10, and 20 were performed inside a magnetic resonance imaging scanner. The main findings are as follows: (1) cortical activation remained approximately constant throughout training, (2) subcortical activation increased with practice (i.e., there were more activated voxels in the striatum), and (3) only cortical activation was correlated with accuracy after extensive training. The results suggest an initial subcortical neural system centered around the head of the caudate that is gradually replaced by a cortical system centered around the ventrolateral prefrontal cortex. With extensive practice, the cortical system progressively becomes more caudal and dorsal, and is eventually centered around the premotor cortex.