Explaining to Others Prompts Children to Favor Inductively Rich Properties

Explaining to Others Prompts Children to Favor Inductively Rich Properties Caren M. Walker 1 , Tania Lombrozo 1 , Cristine H. Legare 2 & Alison Gopnik 1 (caren.walker@berkeley.edu, lombrozo@berkeley.edu, legare@psy.utexas.edu, gopnik@berkeley.edu) Department of Psychology, University of California, Berkeley, 3210 Tolman Hall, Berkeley, CA 94720 USA Department of Psychology, University of Texas at Austin, 1 University Station #A8000, Austin, TX 78712 USA Abstract Three experiments test the hypothesis that engaging in explanation prompts children to favor inductively rich properties when generalizing to novel cases. In Experiment 1, preschoolers prompted to explain during a causal learning task were more likely to override a tendency to generalize according to perceptual similarity and instead extend an internal feature to an object that shared a causal property. In Experiment 2, we replicated this effect of explanation in a case of label extension. Experiment 3 demonstrated that explanation improves memory for internal features and labels, but impairs memory for superficial features. We conclude that explaining can influence learning by prompting children to favor inductively rich properties over surface similarity. Keywords: Explanation; causal learning; category labels; non-obvious properties; inductive inference Introduction The world has a complex structure, and the challenge of causal learning is to discover the nature of this structure to facilitate prediction and action. This is not a trivial task; it is sometimes impossible to predict how an object will behave based on its appearance. In fact, perceptually similar objects can be endowed with very different causal properties: Poison hemlock may look identical to wild carrot, but it is certainly not good to eat. Learning to override perceptual features in favor of non-obvious but inductively rich properties is thus an important achievement. Previous research has examined the role of obvious (perceptual) properties versus non-obvious (internal or abstract) properties in children’s inferences. Young children can use both perceptual and non-perceptual properties in categorizing objects (e.g., Gelman & Markman, 1987; Gopnik & Sobel, 2000), but adults often group objects according to common internal properties, labels, and causal affordances (regardless of perceptual similarity) in cases where young children tend to group objects based on salient perceptual similarity. To illustrate this shift, consider the findings from Nazzi and Gopnik (2000). Children observed four objects placed on a toy, one at a time. Two of these objects were shown to be causally effective – they made the toy play music – and two were inert. One of the causal objects was then held up and labeled (e.g., “This is a Tib.”), and children were asked to give the experimenter the other “Tib.” In no-conflict trials, perceptual and causal properties were always correlated. However, in conflict trials, the same perceptual properties appeared across causal and inert objects. All children were more likely to choose the causal object in the no-conflict trials than in the conflict trials, but analyses of conflict trials revealed a developmental shift: when generalizing the novel label, 3.5-year-olds relied on perceptual cues over causal cues, while 4.5-year-olds relied on causal cues over perceptual cues. Subsequent work has demonstrated a comparable shift in generalizing internal parts (as opposed to a category label). Sobel et al. (2007) used a similar procedure to demonstrate that 4-year-olds, but not 3-year-olds, are more likely to infer that objects have shared internal parts when they share causal properties than when they share external appearance. These examples – and many others (see Keil, 1989; Gelman, 2003) – demonstrate that by 5 years of age, children begin to favor inductively rich but subtle cues, such as category membership and internal parts, over perceptual similarity when generalizing from known to unknown cases. But how is this transition achieved? Here we explore the hypothesis that the process of generating explanations is an important mechanism in scaffolding this transition. Explanation and Causal Learning Accounts of explanation from both philosophy and psychology suggest an important relationship between explanation and causal learning: By explaining past observations we uncover information likely to support future judgments and interventions (e.g., Lombrozo, 2012; Walker, Williams, Lombrozo, & Gopnik, 2012). Consistent with this idea, research with adults finds that prompts to explain can improve learning (e.g., Chi et al., 1994) and promote the discovery and application of broad generalizations underlying what is being explained (e.g., Williams & Lombrozo, 2010). Prior research also suggests that even young children’s explanations have characteristics that make them well suited to highlighting inductively rich properties: they often invoke broad generalizations (Walker et al., 2012) and go beyond appearances (Legare, 2012). For example, Walker et al. (2012) found that prompting preschool-aged children to explain causal events made them more likely to favor broad patterns in generalizing causal properties to novel objects. In the first of these studies, children were presented with evidence that was consistent with two candidate causes (e.g., “green objects make the toy go” versus “yellow objects make the toy go”), where one accounted for more observations. Children who were prompted to explain were more likely than controls to generalize according to the candidate cause that accounted for more of the data. In the second study, the cause that accounted for more of the data was contrasted with an alternative cause that was more consistent with children’s