Category Generation

People exhibit the ability to imagine new category instance s and new categories, with examples ranging from everyday activities like cooking to scientific discovery. This ability , which we call category generation , is not addressed by standard models of category learning, which focus on classifying instances rather than generating them. We develop a probabilistic ac- count of category generation and evaluate it using two behavioral experiments. Our results confirm that people find it natural to generate new category instances and suggest that our model accounts well for this ability.