Research Overview

Brief summary

Asking “Why?” is ubiquitous in the lives of adults, children, and scientists. Generating explanations promotes student learning, drives children’s acquisition of knowledge, and influences how people acquire and use categories, reason about causation, and generalize properties. Why does explaining have these effects?

Explaining is often interpreted as boosting processing and engagement (increasing attention, motivation, or metacognition), but I have explored a subsumptive constraints account, which proposes that explaining "Why?" guides people to seek underlying patterns or generalizations. The goal of constructing an explanation for why a fact or observation is true constrains people to understand how it could be anticipated as an instance of a generalization – such as rules, regularities and principles – that subsume or account for the facts and observations they are learning about. The experiments I’ve done so far have tested three predictions of this account: primarily in the context of learning categories from examples, but also in the context of learning to predict people’s behavior.

1. A paper in Cognitive Science explains the subsumptive constraints account and presents experiments on category learning which show that explaining promotes the discovery of patterns and their use in generalization, although it has either no effect or an impairment on memory for observations (Williams & Lombrozo, 2010a). This advantage is observed relative to control conditions in which participants engage in description, thinking aloud, or free study.

2. Two experiments provide evidence that explaining does not only boost overall pattern discovery, but drives learners to consult their prior knowledge to privileges patterns that prior knowledge suggests are likely to be broadly applicable beyond particular observations (Williams & Lombrozo, 2010b; in prep).

3. The third line of work provides evidence for a counterintuitive prediction of this account: While explanation’s subsumptive constraint promotes the discovery of generalizations when reliable patterns are present, it can also impair learning if learners infer spurious or misleading patterns (Williams, Lombrozo & Rehder, 2010; 2011). In the context of learning categories and predicting people’s behavior, explaining promoted learning when a reliable pattern was present, but impaired learning when the pattern was misleading.

Extended Discussion

Trying to explain the world around us is a ubiquitous part of human activity, apparent in the thinking of adults, children, and scientists. Engaging in explanation has a powerful impact on learning and generalization, an effect demonstrated by education researchers in students (Chi et al, 1989; 1994) and by work in cognitive development on children’s conceptual change (Amsterlaw & Wellman, 2006; Siegler, 2002).

1. Williams, J. J., & Lombrozo, T. (2010a). The role of explanation in discovery and generalization: evidence from category learning. Cognitive Science.

The Subsumptive Constraints account (Williams & Lombrozo, 2010a) accounts for explanation’s enhancement of learning and generalization by proposing that asking “Why?” selectively constrains people’s learning, driving them to find patterns and regularities that underlie what they are explaining. This suggests explaining does not benefit learning just through an all-purpose boost to processing, but through a selective benefit to discovering and constructing patterns.

This hypothesis is motivated by work in philosophy of science which has attempted to characterize what good explanations are. Subsumption and Unification theories (Kitcher, 1981; 1989) propose that the key property of an explanation is that it shows how what is being explained is an instance of (subsumed under) a general pattern or regularity. If most good explanations satisfy this property, then attempting to generate an explanation will constrain the process of learning in a selective way- explaining will guide people’s reasoning and learning towards the goal of pattern discovery. Prompting people to explain will therefore drive them to discover underlying regularities, and to a greater degree than other learning strategies or activities.

Three experiments used a category learning task to test this prediction directly. The effect of generating explanations was compared to describing, thinking aloud, or free study. Engaging in explanation promoted the discovery of a subtle regularity underlying category membership to a greater degree than any of the control conditions, which variously matched explanation in time, attention, and verbalization. This occurred even though describing resulted in better memory for item materials.

2. Williams, J. J., & Lombrozo, T. (2010b). Explanation constrains learning, and prior knowledge constrains explanation. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society. Austin, TX: Cognitive Science Society.

Another experiment takes a first step towards examining the relationship between explanation and prior knowledge in learning (Williams & Lombrozo, 2010b). By manipulating the prior knowledge provided in category labels, the experiment demonstrated a superadditive interaction between engaging in explanation and possessing prior knowledge. If the effect of explaining is to constrain learners to discover underlying patterns, prior knowledge may further constrain which patterns people discover through explanation.

3. Williams, J. J., Lombrozo, T., & Rehder, B. (2010). Why does explaining help learning? Insight from an explanation impairment effect. In S. Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual Conference of the Cognitive Science Society. Austin, TX: Cognitive Science Society.

A further experiment (Williams, Lombrozo, & Rehder, 2010) attempted to discriminate the Subsumptive Constraints account from a an account of explanation and learning in terms of learning engagement: whereby explaining increases engagement with the learning task, such as by boosting attention, motivation, or time spent on the task. Participants learned about novel categories by classifying and receiving feedback, and a pattern underlying category membership was either reliable or misleading. If explaining promotes learning through increasing task engagement, it would be unlikely to impair learning. If explaining promotes learning through constraining people to find underlying patterns, its effects will depend on whether there are useful regularities present. The experiment found an interaction between engaging in explanation and the reliability of the pattern: explaining tended to help learning when the pattern was reliable, but impaired learning when a misleading pattern was present, producing an explanation impairment effect. The experiment provided evidence for the importance of subsumptive constraints in explanation’s effects on learning, suggesting that explaining does not simply increase engagement with the learning task.