Selective Attention, Category Structure, and Supervision in Category Learning: A Comparative Approach

We investigated the interactions between category structure, supervision, and the ability to selectively attend in category learning. To examine selective attention, we compared human adults and pigeons in a category learning task, where we manipulated the category structure (dense vs. sparse) and the level of supervision (low vs. high). Humans, who have a more advanced ability to attend selectively, showed faster learning and better generalization overall. Both species showed a benefit for the category structures that had multiple category-relevant dimensions (i.e., dense categories). In addition, there was a benefit of supervision that was particularly strong in learning the sparse categories. Finally, attention was optimized to the category-relevant dimension in the sparse category only in human adults. Subsequent computational simulation of the data indicates that these patterns are well explained by a parameter that controls the ability to flexibly switch attention to category-relevant dimensions.