Learning categories via rules and similarity: comparing adults and children

Two experiments explored the different strategies used by children and adults when learning new perceptual categories. Participants were asked to learn a set of categories for which both a single-feature rule and overall similarity would allow for perfect performance. Other rules allowed for suboptimal performance. Transfer stimuli (Experiments 1 and 2) and single features (Experiment 2) were presented after training to help determine how the categories were learned. In both experiments, we found that adults made significantly more optimal rule-based responses to the test stimuli than children. Children showed a variety of categorization styles, with a few relying on the optimal rules, many relying on suboptimal single-feature rules, and only a few relying on overall family resemblance. We interpret these results within a multiple systems framework, and we argue that children show the pattern they do because they lack the necessary cognitive resources to fully engage in hypothesis testing, rule selection, and verbally mediated category learning.