Differences in the Development of Analogy Across Cultures: A Computational Account

Theories of the development of analogical reasoning emphasize either the centrality of relational knowledge accretion or changes in information processing. Recent cross-cultural data collected from children in the United States and China (Richland, Chan, Morrison, & Au, 2010) provides a unique way to test these theories. Here we use simulations in LISA/DORA (Doumas, Hummel, & Sandhofer, 2008; Hummel & Holyoak, 1997, 2003), a neurally-plausible computer model of relational learning and analogical reasoning, to argue that the development of analogical reasoning in children may best be conceptualized as an equilibrium between knowledge accretion and progressive improvement in information processing capability. Thus, improvements in inhibitory control in working memory as children mature enable them to process more relationally complex analogies. At the same time, however, children produce more complex and more accurate analogies in domains in which they have learned richer and more refined representations of relational concepts. Relational thinking—i.e., thinking based on the roles that objects play rather than the literal features of those things—is a cornerstone of human cognition. It underlies, among many other things, our ability to make analogies, or to appreciate correspondences between domains (e.g., Holyoak & Thagard, 1995). As with many cognitive processes, our ability to make analogies changes with development. While there is considerable agreement that analogy is a very important process in cognitive development (e.g., Gentner, 2003), there is considerable disagreement as to how the ability to reason analogically develops. Theories of the Development of Analogical Reasoning Three primary hypotheses have been put forward to explain age-related differences in analogical reasoning: changes in domain knowledge, a relational shift from object similarity to relational similarity, and increased processing or working memory (WM) capacity. Goswami and colleagues (Goswami, 1992, 2001; Goswami & Brown, 1989) proposed that the ability to make analogies is present even in early infancy. However, children can only evidence this ability with age and increased knowledge. In other words, the change in children’s ability to make analogies is not a function of a developing mechanism, but rather knowledge accretion. Alternately, Gentner and Rattermann (1991; Rattermann & Gentner, 1998) argued that a domainspecific “relational shift” is responsible for changes in children’s analogical abilities. Gentner and Rattermann suggest that as children build knowledge in a particular domain they progress from reasoning about that domain in terms of the perceptual features of objects, to the relations between those objects. For example, 3 year-old children will categorize objects based on overall featural similarity (e.g., they will match apples to red balls rather than bananas), however by age 4 or 5, children will categorize objects based on relational similarity (e.g., matching apples to bananas even in the presence of featural distracters like red balls; Gentner & Namy, 1999). The ability to make analogies based on relational commonalities between domains, therefore, progresses on a domain-by-domain basis with more complex analogies produced in domains in which knowledge is richer. In contrast to accounts of analogy development based on increases in knowledge, the relational complexity hypothesis of Halford (1993; Andrews & Halford, 2002; Halford et al., 2002) holds that limits in children’s WM capacity affects their ability to process relations simultaneously, and therefore their ability to make analogies. According to Halford and colleagues, children can process only specific levels of relational complexity, defined as the number of sources of variation that are related and must be processed together. The simplest level of relational complexity is a binary relation, where only two arguments are sources of variation. The relation, chase (dog, cat), for instance, specifies a single relation (chase) between two objects (dog, cat). To reason about this relation, a one must keep only the two objects and their relation in mind. A ternary relation (e.g.,