UC Berkeley

Constructing explanations in response to why-questions plays an important role in learning and generalization - in young children, students, adults and scientists - but much remains to be understood about the underlying mechanisms and their implications for practical learning and teaching. The first chapter of this dissertation presents theSubsumptive Constraints account, which proposes that explaining why a fact or observation is true drives learners to understand how it can be subsumed as an instance of a pattern or generalization. Chapters two through four present nine experiments which test whether explaining exerts this selective effect, rather than a general boost for learning, examining contexts ranging from learning artificial categories to learning to predict other people's behavior. In these nine experiments, three specific predictions of the Subsumptive Constraints account were supported. The second chapter presents the first three experiments, which provide evidence that explaining category membership promoted the discovery of broad underlying patterns that supported generalization. Four experiments reported in the third chapter further revealed that seeking explanations increased learners' consultation of existing knowledge that could be used to discover patterns in their observations. Explaining also interacted with existing knowledge to influence which patterns learners generalized to novel contexts. The final two experiments in chapter four confirmed the Subsumptive Constraints account's counterintuitive prediction that seeking explanations can impair learning, by revealing that learning was driven by an interaction between explanation and the reliability of underlying patterns. Explaining drove people to ignore specific observations that were exceptions to patterns, so that explaining fostered overgeneralization at the expense of accurate learning. The fifth chapter concludes this work. By drawing together theory and methodology from cognitive psychology, education, and philosophy, this dissertation provides novel theoretical, empirical, and practical insight into how generating explanations influences and improves learning. Explaining does not simply promote an 'all-purpose' boost to engagement and metacognition - and is therefore not always practically beneficial or optimal. By characterizing the particular nature of the processing constraints promoted by explaining - as selectively constraining the search for and evaluation of patterns that underlie specific observations - this novel Subsumptive Constraints account explains why explaining especially impacts learners' use of existing knowledge in transfer and generalization, and provides insight into when to expect and how to maximize explanation's practical benefits.