Multiple learning modes in the development of performance on a rule-based category-learning task