Multiple systems of category learning

Abstract

We review neuropsychological and neuroimaging evidence for the existence of three qualitatively different categorization systems. These categorization systems are themselves based on three distinct memory systems: working memory (WM), explicit long-term memory (explicit LTM), and implicit long-term memory (implicit LTM). We first contrast categorization based on WM with that based on explicit LTM, where the former typically involves applying rules to a test item and the latter involves determining the similarity between stored exemplars or prototypes and a test item. Neuroimaging studies show differences between brain activity in normal participants as a function of whether they are instructed to categorize novel test items by rule or by similarity to known category members. Rule instructions typically lead to more activation in frontal or parietal areas, associated with WM and selective attention, whereas similarity instructions may activate parietal areas associated with the integration of perceptual features. Studies with neurological patients in the same paradigms provide converging evidence, e.g., patients with Alzheimer's disease, who have damage in prefrontal regions, are more impaired with rule than similarity instructions. Our second contrast is between categorization based on explicit LTM with that based on implicit LTM. Neuropsychological studies with patients with medial-temporal lobe damage show that patients are impaired on tasks requiring explicit LTM, but perform relatively normally on an implicit categorization task. Neuroimaging studies provide converging evidence: whereas explicit categorization is mediated by activation in numerous frontal and parietal areas, implicit categorization is mediated by a deactivation in posterior cortex.

Figure 1. Koenig et al. (2005) experiment

Illustration of arrays seen during the training session. Panel A: rule-based training, showing the four criterial features in the upper half of the screen and two CRUTTER in the lower half (Member on left). Panel B: Similarity-based training, showing a prototype in the upper half of the screen and two CRUTTER in the lower half (Member on left).

Figure 2. Koenig et al. (2005) experiment, Behavioral results

Category endorsement functions, which show the percentage of time that participants endorsed a test item as a function of the number of critical features it contained, separately for Rule and Similarity groups.

Figure 3. Koenig et al. (2005) experiment, Imaging results

Right panel: Rule- minus Similarity-based categorization for clear-cut Members and Non-members. Panel B: Similarity- minus Rule-based categorization for clear-cut Members and Non-members.

Figure 4. Koenig et al. (in press) experiment

Category endorsement functions for Rule and Similarity groups: Panel A: Controls; Panel B: AD patients; and Panel C: CBD patients.

Figure 5. Koenig et al. (in press) experiment

Number of training trials needed to reach learning criteria as a function of type of participant, separately for Rule and Similarity groups.

Figure 6. Koenig et al. (2007) experiment, Imaging results

Panel A: Hippocampal activation during an implicit categorization task; Panel B: Posterior deactivation during the same implicit categorization task.