Categories, concepts and causality: A reply to Samuelson and Perone

Abstract

Listen

Translate article

In the research towhich Samuelson and Perone (S& P) direct their commentary,we took two sets of previous findings as our starting point: (a) function facilitates early word learning and categorization (Booth & Waxman, 2002; Gathercole & Whitfield, 2001; Kemler Nelson, Russell, Duke, & Jones, 2000) and (b) the emergence of young children’s ‘shape bias’ is encouraged by their growing understanding of kinds (Diesendruck & Bloom, 2003). We hypothesized that there might be a link between these two sets of findings. Specifically, we considered the possibility that one way in which function might facilitate categorization is by highlighting property dimensions, such as shape, that are particularly functionally relevant, and by extension, useful in distinguishing between kinds of artifacts. Repeated exposure to the causal link between shape and function could therefore lead children to develop a general ‘shape bias’ in categorization. To test this hypothesis, we used a longitudinal training procedure modeled closely on previous research on the origins of such a bias (Samuelson, 2002; Smith, Jones, Landau, Gershkoff-Stowe, & Samuelson, 2002). Seventeen-month-olds underwent six weeks of training with novel categories, each consisting of similarly-shaped objects. The training differed across three groups: the function group (Study 1) learned that the objects within each category could perform a function that was causally-related to their shapes; the non-causal function group also learned about functions, but they were not causally-dependent on the objects’ shapes (Study 2); the no-function control group did not learn about functions (Study 1). In week seven, toddlers in all conditions used shape reliably to extend these now familiar training categories. The critical test, however, occurred in week eight, when we evaluated whether participants would also preferentially extend completely novel categories on the basis of shape. Toddlers in the (causal) function condition were more likely to use shape than were those in the non-causal and no-function conditions and were the only ones to do so at a rate that exceeded chance. Thus, merely learning about correlations between shape and categories or between shape and functionswas not sufficient for promoting a general shape bias. Instead, onlywhen toddlers learned that shape is causally, and thus meaningfully, tied to functions, did they come to view shape