Running head: Inductive inferences and exploratory play Word, thought, and deed: The role of object categories in children’s inductive inferences and exploratory play Laura E. Schulz1, Holly R. Standing2, and Elizabeth B. Bonawitz1 1Massachusetts Institute of Technology 2 Cambridge University This research was supported by a McDonnell Foundation Collaborative Initiative Causal Learning grant and a John Templeton Foundation grant to L.S. Thank you to Noah Goodman, and Rebecca Saxe for helpful comments and suggestions. We are also very grateful to the Boston Museum of Science Discovery Center and participating parents and children. Address for correspondence: L. Schulz, 46-4011, MIT Department of Brain and Cognitive Sciences, 77 Massachusetts Avenue, Cambridge, MA 02139. Phone: 617-253-7957. E-mail: [email protected] Previous research (e.g., Gelman & Markman, 1986; Gopnik & Sobel, 2000) suggests that children can use category labels to make inductive inferences about non-obvious causal properties of objects. However, such inductive generalizations can fail to predict objects’ causal properties when A) the property being projected varies within the category; B) the category is arbitrary (e.g., things smaller than a bread box), or C) the property being projected is due to an exogenous intervention rather than intrinsic to the object kind. In four studies, we show that preschoolers (mean: 48 months; range: 42-57 months) are sensitive to these constraints on induction and selectively engage in exploration when evidence about objects’ causal properties conflicts with inductive generalizations from the objects’ kind to their causal powers. This suggests that the exploratory actions children generate in free play could support causal learning. Keywords: inductive inference, preschoolers, causal learning, categorization, exploratory playInductive inferences and exploratory play Page 3 of 34 Word, thought, and deed: The role of object categories in children’s inductive inferences and exploratory play Few ideas in developmental psychology are more widely accepted than the idea that children learn through play. However, despite a broad consensus that children “construct” knowledge (and particularly causal knowledge) by actively exploring their environments (Bruner, Jolly, & Sylva, 1976; Piaget, 1952; Singer, Golinkoff, & Hirsh-Pasek, 2006), relatively little is known about how children’s exploratory play might support accurate learning. The majority of research on children’s play is descriptive rather than experimental and much of the seminal work on exploratory play predates contemporary analyses of children’s causal reasoning (Berlyne, 1954; 1960; Bruner, 1973; Dember, 1957; Henderson & Moore, 1980; Hutt & Bhavnani, 1972; McCall, 1974; Pavlov, 1927; Piaget 1951; 1952). Moreover, there is considerable evidence that children are poor at designing informative experiments (Chen & Klahr, 1999; Inhelder & Piaget, 1958; Koslowski, 1996; Kuhn, 1989; Kuhn, Amsel, & O’Laughlin, 1988; Masnick & Klahr, 2003; Mayer, 2004). Such findings pose a challenge for accounts of cognitive development that emphasize the role of children’s active exploration. If children’s exploratory play is largely unsystematic, how might children generate the type of evidence that could support accurate causal learning? Here we investigate the hypothesis that children’s spontaneous exploratory play is affected by some of the same principles of rational inductive inference that support children’s causal judgments. Specifically, we look at the extent to which children’s exploratory play is sensitive to the relationship between an entity’s category membership and its causal properties. If children selectively engage in exploratory play when evidence about objects’ causal properties conflicts with inductive generalizations from the objects’ kind to their causal powers, then children’s exploratory actions could support causal discovery.Inductive inferences and exploratory play Page 4 of 34 In everyday reasoning, we routinely use information about an object’s kind to make inferences about its unobserved properties. If we think for instance, that a new device is a cell phone, we infer that it can do different things than if we think the new object is a TV remote or a garage door opener. Although both adults and children also rely on perceptual cues, including an entity’s shape, color, and texture, to extend inferences about unobservable internal states and causal powers (Baldwin, Markman, & Melartin, 1993; Gelman & Markman, 1986; Gopnik & Sobel, 2000), contra Piaget (1952), very young children are not limited to making inferences based only on superficial perceptual properties; even preschoolers make inductive inferences about objects’ causal properties on the basis of their kind membership. Indeed, in induction tasks, preschoolers will extend inferences about unobservable properties of objects on the basis of category labels rather than on the basis of perceptual cues, and in categorization tasks, children will use causal, rather than perceptual information to make category judgments (Gelman & Coley, 1990; Gelman & Markman, 1986; Gopnik & Nazzi, 2003; Gopnik & Sobel, 2000; Keil, Smith, Simons, & Levin, 1998; Nazzi & Gopnik, 2000; Simons & Keil, 1995). Children who are taught, for instance, that gold melts and clay burns infer that a substance that looks like clay but is called gold will melt rather than burn (Gelman & Markman, 1986). Similarly, children who know that a red block called a ‘blicket’ lights up a toy, will infer that a blue block called a blicket will light the toy and that a red block which is not a blicket, will not (Gopnik & Sobel, 2000). One of the benefits of inferring causal properties from information about the objects’ kind is that it obviates the need to learn the objects’ causal properties by trial and error. If we know for instance, that blickets activate a toy and that these blocks are blickets, we can infer that these blocks will activate the toy without testing them. However, inductive inferences are only true in probability. If we subsequently observe evidence that these blocks fail to activate the toy -- that is, we get evidence against the inductive generalization – then exploratory learning may be advantageous: perhaps not all blickets activate the toy, perhaps these blocks are not reallyInductive