Using Numbers to Structure Space Dedre Gentner (gentner@northwestern.edu) Department of Psychology, Northwestern University 2029 Sheridan Rd., Evanston, IL 60201 Stella Christie (christie@northwestern.edu) Department of Psychology, Northwestern University 2029 Sheridan Rd., Evanston, IL 60201 Abstract We investigated the claim that relational language promotes the development of relational reasoning (Gentner, 2003). Prior research has shown the benefit of spatial relational language (e.g. top, middle, bottom) in preschoolers’ performance in spatial mapping (Loewenstein and Gentner, 2005), suggesting that spatial relational language invites a delineated relational representation. We generalized this conclusion by testing the benefit of using nonspatial relational language in a spatial analogical task. Preschool children were presented with two identical three-tiered boxes, in which they watched an item being hidden in one box and were then asked to search for a similar item in the corresponding location at the second box. Half of the children heard a set of systematic terms conveying monotonic structure (1 2 3), whereas the other half of the children heard non-systematic terms consisting of names of familiar animals. Both sets of terms are familiar to preschoolers and neither directly denotes spatial locations. We found that preschool children who heard the 123 labels performed better than those who heard animal names. The results are evidence of young children’s sensitivity to the relational structure conveyed by language, and to their ability to apply this structure into a different domain. Introduction Relational similarity is a crucial construct in human cognition. The ability to perceive relational similarity underlies a number of fundamental cognitive processes such as analogies (Gentner, 1983, 2003), categorization (Ramscar and Pain, 1996), and inductive inferencing (Holland, Holyoak, Nisbett, and Thagard, 1986). How do children come to develop relational thinking? This question becomes especially relevant in the domain of spatial cognition – where an appreciation of spatial relations is necessary in a broad range of spatial tasks, from navigating to perceiving the configuration of a landmark array to map reading (e.g. Newcombe, 2002; Newcombe and Huttenlocher, 2000). Previous research has shown that the apprehension of relational similarities is not immediate in development: children first rely on overall similarity or on commonalities based on element matching, and then shift to appreciating commonalities based on relations (Gentner & Rattermann, 1991, Rattermann & Gentner, 1998a; Halford, 1993). This relational shift (Gentner, 1988) is also observed in the development of spatial cognition. For example, children understand element-to-element correspondences before they understand spatial relational correspondences (Bluestein & Acredolo, 1979; Presson, 1982). Liben (1998) also described a relational shift pattern in map understanding: chidren understood object-based correspondence before they understood relation-based correspondences. Blades and Cooke (1994) adapted the classic DeLoache task in which children watched a toy being hidden at an object in a model room and were asked to retrieve a similar toy hidden at the same object in a second model room. All objects (toy locations) were distinct except for one pair of identical objects. 3-year-olds succeeded when the toy was hidden at a unique object, but they failed the task when the toy was placed at one of the identical objects, suggesting a reliance on object-matching. Gentner (2003) has proposed that the learning and application of relational language is a route to learning domain relations. Indeed, there is evidence suggesting that relational language might foster attending to and encoding particular relations (Hermer-Vasquez, Spelke, & Katsnelson, 1999; Logan & Sadler, 1996; Regier & Carlson- Radvansky, 2001). For example, in the development of spatial cognition, Hermer-Vasquez, Moffet, & Munkholm (2001) found that children’s performance in a search task was correlated with their ability to use the spatial language relevant to the task. Recently, Loewenstein and Gentner (2005) found evidence from a spatial mapping task for the claim that spatial relational language fosters the development of spatial relational knowledge. Because our study is based on their methods we describe it in some detail. In the more challenging version of their task, using the cross-mapping technique—in which object matches do not correspond to relational matches—they found a semantically specific effect. Children who heard the set of spatial terms top, middle, bottom, performed significantly better than children who heard the set of spatial terms on, in, under. Loewenstein and Gentner conjectured that this difference in performance is due to the advantage of systematicity, that is, a connected system of relations such as top, middle, bottom invites a better relational representation, and this in turn supports relational mapping. The set of terms top, middle, bottom form a connected structure: top > middle > bottom. Thus they form an interconnected system governed by the 267