A Generative Connectionist Model of the Development of Rule Use in Children

The cascade correlation algorithm (CASCOR), a generative connectionist model, was used to simulate age-related changes on the dimensional change card sort (DCCS), which has traditionally been used to evaluate the complexity of children’s rule-use abilities. Like 2.5-year-olds, inexperienced networks behave as if following one rule; slightly more experienced networks (akin to 3-year-old children) behave as if following a pair of rules; and the most experienced networks (akin to 5-year-olds) behave as if following two pairs of rules. Analysis of the networks’ activation levels revealed that mastery of simple rules is a necessary precondition for using higher order rules. The model also generated four novel predictions that can be tested in future research with children.