It now seems uncontroversial to assert that "prior knowledge" or naïve theories play a key role in adult categorization. Twenty years ago, Murphy and Medin (1985) published a seminal paper arguing that most concepts are acquired in the context of larger "theory-like" knowledge structures. Since then a wealth of data has accumulated documenting the effects of prior knowledge on adult categorization (Ahn & Luhmann, 2005; Heit, 1997; Murphy, 2002), and a variety of formal models have been developed to explain these effects (Heit, Briggs, & Bott, 2004; Rehder & Murphy, 2003). In the case of category learning by infants and young children, however, there remains intense debate about how background knowledge may (or may not) affect concept acquisition and development. At one extreme, some view children's concepts as informed by broader knowledge structures or theories from an early point in development (Gelman, 2003; Keil, Smith, Simons, & Levins, 1998). Background knowledge or naïve theories are seen as providing important constraints on the way that children encode new exemplar information and make decisions about category membership. Age-related changes in children's knowledge or theories therefore affect the way that they categorize (Carey, 1985; Vosniadou & Brewer, 1992). This will be referred to as the "knowledge-based" approach to early category learning. Many, however, remain skeptical about the impact of such theories on early categories, insisting that these are acquired from the "bottom-up," through the learning of the perceptual and statistical regularities present in category exemplars (Jones & Smith, 1993; Quinn, 2002; Rakison & Hahn, 2004; Sloutsky & Fisher, 2004). Such approaches include applications of seasoned theories of adult category acquisition involving either the abstraction of prototypes from experience with category members (Hayes & Taplin, 1993; Younger, 1990) or the encoding and retrieval of specific exemplars (Boswell & Green, 1982; Hayne, 1996). According to this "similarity-based" approach, naïve theories are the products of rather than contributors to conceptual development. One reason why this debate has proven difficult to resolve is a lack of consensus as to just what constitutes a "theory" or "knowledge" in early categorization. A second problem is that proponents of the knowledge-based approach have often been vague about how knowledge interacts with more fundamental categorization processes such as exemplar encoding or prototype abstraction. In this chapter, the aim is to deal with these issues by proposing a rapprochement between knowledge-based and similarity-based views of children's categorization. The central argument is that from an early point in development, prior knowledge and statistical learning processes are closely integrated and interdependent. Prior knowledge suggests which features are most relevant during category learning and influences how similarity is computed during categorization. However, the effects of prior knowledge are not static. Increasing experience with the members of a new category can lead children to incorporate into their concept representation features that were not predicted by prior knowledge. Experience with instances that violate expectations can also lead children to revise their general beliefs about a category domain. This analysis suggests that there are three routes to developmental change in children's concepts. The first route is via domain-general changes in the efficiency with which children process the regularities in observed exemplars. The second involves age-related changes in the accrual and reorganization of domain knowledge. Finally, there may be developmental change in the interaction between these two sources. With age we may see changes in how children integrate their new experiences with their knowledge to produce conceptual representations. In this chapter, I will review evidence bearing on the first two of these routes, suggesting that each has a role to play in shaping children's concepts. I will also present data suggesting that the third route is particularly important for understanding conceptual development. To make this case I will first review the arguments for a role of background knowledge in children's categorization and examine the various ways that "knowledge" has been defined. I will then review the empirical evidence illustrating the many different ways that background knowledge can affect children's encoding of new exemplars, decisions about category membership, memory for instances, and category-based inferences. I will then describe an explicit theory of how prior knowledge and exemplar information are integrated during category learning and examine how this theory can be applied to children's concept learning and conceptual development. In the final sections, I will examine the challenges that remain for theories that assume that early categorization is driven by an interaction between knowledge and experience with novel category exemplars. © 2006 Elsevier Inc. All rights reserved.
