A comparative investigation into the mechanisms that support multidimensional categorization

Some categories are easy to define (e.g., all squares have four equilateral sides joined by four right angles), whereas other categories are not (e.g., most birds fly, but not all do). According to a prominent theoretical framework, these two types of categories should engage different mechanisms – learners should generally start by testing declarative decision rules (“does it fly?”); but, if those rules fail, then learners should shift to associative, similarity-based categorization (“does it look more like a bird than anything else?”) In the experiments reported here, we more closely examined these predictions. By comparing humans, who regularly depend on rule use, to pigeons, which show no evidence of rule use, we found that humans do approach new category structures by exploring simple decision rules. However, when those rules fail, humans may not reliably abandon rule use; instead, they often develop more complex declarative strategies. We also found that, although pigeons did not appear to deploy declarative strategies, they robustly succeeded even at tasks that challenged human learners. These data suggest that associative learning mechanisms merit far greater attention than they commonly receive in the broader category learning literature.