Analogical Reasoning, Models of Development

Relational Knowledge

Usha Goswami has argued that children are able to map relations in a rudimentary manner from early infancy, but their later analogical reasoning skills build on prerequisite content knowledge. Thus, children's analogical reasoning becomes more and more adult-like on a domain-by-domain basis as knowledge develops.

Similarly, Dedre Gentner and colleagues hypothesized a “relational shift” during cognitive development such that, as children build knowledge in a domain, they move from attending to similarity based on object features to relational similarity. These authors postulate this process is not an age-related phenomenon but rather is tied to knowledge acquisition. Robert Morrison and colleagues have alternatively argued that the relational shift can be understood as a deficit in inhibitory control in working memory, one aspect of executive functions.

Executive Functions

Even when young children can demonstrate relational knowledge in a domain, they frequently have difficulty using analogies requiring integration of multiple relations. Graeme Halford has proposed a theory of relational complexity to categorize relations by the number of sources of variation that must be processed in parallel. Halford suggested that on average, children's working-memory capacity is such that after age 2, children can process binary relations, and after age 5, they can process ternary relations. Thus, children of age 2 can perform very simple analogy problems but not problems that require integrating multiple relations.

Building Relational Representations

All successful models of analogical reasoning operate on structured representations in long-term memory; however, until recently, no explanation existed for how these structured representations might arise. The lack of an account of where relational representations come from has led Robert Leech and colleagues to postulate that analogy might not rely on structured representations but rather on simple associations. However, these approaches have failed to provide an explanation of how children can process progressively more relationally complex analogies or exhibit the flexibility in relational thinking characteristic of adults. A more viable option is that humans can learn structured relational representations from unstructured examples. These representations might then be used in traditional symbolic or symbolic-connectionist models to perform analogical reasoning. Leonidas Doumas and colleagues recently described one such approach that uses comparison to bootstrap learning structured relational representations starting with simple distributed representations of objects as feature vectors.

...