Rule-Learning Events in the Acquisition of a Complex Skill: An Evaluation of Cascade

Abstract Acquiring a complex cognitive skill often involves learning principles of the task domain in the midst of solving problems or studying examples. Cascade is a model of such learning. It includes both rule-based reasoning and several kinds of analogical, case-based reasoning. Task domain principles are represented as rules, and Cascade learns new rules at rule-learning events, which are initiated by an impasse and utilize multiple kinds of reasoning. In this article, I evaluate Cascade's model of rule-learning events by analyzing ones gleaned from protocols of physics students solving problems and studying examples. As expected, Cascade's model is overly simple, but it appears feasible to extend it to cover all the observed learning events. The data themselves were surprising in that there are few learning events relative to the number that could have occurred, and those that did occur often involved forms of reasoning that are considerably shallower than expected. The data suggest ways that instruction can be improved to increase both the quantity and depth of learning events.