Explaining Promotes Discovery: Evidence from Category Learning

Explaining Promotes Discovery: Evidence from Category Learning Joseph Jay Williams (joseph_williams@berkeley.edu) Tania Lombrozo (lombrozo@berkeley.edu) Department of Psychology, University of California, Berkeley Abstract Research in education and cognitive development suggests that explaining plays a key role in learning and generalization: when learners provide explanations – even to themselves – they learn more effectively and generalize more readily to novel situations. This paper explores a potential mechanism underlying this effect, motivated by philosophical accounts of the structure of explanations: that explaining guides learners to interpret observations in terms of unifying patterns or regularities, which in turn promotes the discovery of broad generalizations. Experiment 1 finds that prompting participants to explain while learning artificial categories promotes the induction of a broad generalization underlying category membership. Experiment 2 suggests that explanation most readily prompts discovery in the presence of anomalies: observations inconsistent with current beliefs. Experiment 1 additionally suggests that explaining might result in reduced memory for details. These findings provide evidence for the proposed mechanism and insight into the potential role of explanation in discovery and generalization. Keywords: explanation; learning; explanation; category learning generalization; self- Seeking explanations is a ubiquitous part of everyday life. Why is this bus always late? Why was my friend so upset yesterday? Why are some people so successful? Young children are notorious for their curiosity and dogged pursuit of explanations, with one “why?” question followed by another. Equally curious scientific researchers might wonder: Why is explaining so important? Psychologists and philosophers have independently proposed that in explaining observations about the past, we uncover underlying structure in the world, acquiring the knowledge to predict and control the future. For example, in explaining a friend’s behavior, you might come to appreciate the extent of her ambition, which informs expectations about her future actions. Research in education and cognitive development confirms that the process of explaining – even to oneself – can foster learning. This phenomenon is known as the self- explanation effect, and has been documented in a broad range of domains: acquiring procedural knowledge about physics problems (Chi et al., 1989), declarative learning from biology texts (Chi et al., 1994), and conceptual change in children’s theory of mind (Amsterlaw & Wellman, 2006), to name only a few. Compared to alternative study strategies like thinking aloud, reading materials twice, or receiving feedback in the absence of explanations (e.g. Chi, 1994; Amsterlaw & Wellman, 2006), self-explanation consistently leads to greater learning, with the greatest benefit for transfer and generalization to problems and inferences that require going beyond the material originally studied. Researchers have made a number of proposals about the mechanisms that underlie explanation’s beneficial effects on learning. These include the metacognitive consequences of engaging in explanation (such as identifying comprehension failures), explanation’s constructive nature, and its role in dynamically repairing learners’ mental models of particular domains (e.g. Chi, 1989; 1994). Given the diversity of the processes which can underlie learning (Nokes & Ohlsson, 2005), it is likely that explanation influences learning via multiple mechanisms. In this paper we explore why explaining plays such an important role in transfer and generalization. We investigate the hypothesis that engaging in explanation will promote the discovery of broad, abstract generalizations that underlie what is being learned. This hypothesis is motivated by work on the structure of explanations. By the structure of explanations, we mean the relationship that must hold between an explanation and what it explains for it to be genuinely explanatory. Little research in psychology has addressed this question directly (see Lombrozo, 2006), but a rich tradition from philosophy provides candidate theories. While there is no consensus, we focus on pattern subsumption theories, which identify good explanations as those that demonstrate how what is being explained is an instance of a general pattern (for discussion see Strevens, 2008; for suggestive empirical evidence see Lombrozo & Carey, 2006; Wellman & Liu, 2006). A subset of these accounts further emphasizes unification: the value of explaining disparate observations by appeal to a single explanatory pattern. For example, in explaining a friend’s current cold by appeal to the contraction of a germ from another person, a specific event (Bob’s cold) is subsumed as an instance of a general pattern (the transmission of germs produces illnesses in people), and this general pattern can account for both this observation and a range of other data. Subsumption and unification accounts of explanation predict the privileged relationship between explanation and generalization demonstrated by the self-explanation effect. If the explanations people construct satisfy the structural demands of subsumption, then successfully engaging in explanation should result in the induction or explicit recognition of generalizations that underlie what is being explained. Generating or explicitly representing such generalizations should in turn facilitate the transfer of what is learned in one context to novel but relevant contexts. We therefore investigate the hypothesis that the structure of explanations contributes to the relationship between explanation and generalization, and that explaining will drive learners to discover broad, abstract generalizations that support effective transfer to novel contexts.