Organisms learn from experience in many ways. One component of learning from experience is recording what has happened in the world when actions are taken, a form of episodic memory, and distilling such experience over time to learn models of phenomena for generating expectations. As further actions are taken, the accuracy of such models can be monitored, to detect surprises and to help identify and prioritize learning goals. This publication-based talk will describe some recent results in exploring the use of analogical generalization over episodic memories in the Companion cognitive architecture to formulate models of the effects of actions in a complex dynamic world. Measures of novelty, surprise and for prioritization of learning goals will be discussed.