Finding the commonalities: The role of analogical mapping during similarity judgments

Similarity plays a central role in cognitive theories. Much research has been devoted to understanding what makes a pair of objects similar. This research has demonstrated that the similarity of a pair increases with its commonalities and decreases with its differences. These common and distinctive elements can take the form of parts of objects, relations between parts or properties of whole objects. However, previous work has been unable to reconcile this variety of information within a single framework. We suggest that structural alignment, like that proposed to mediate analogical reasoning, provides a sufficiently powerful process for determining the commonalities and differences of complex representations. In order to make this suggestion more concrete, we present two simulations of the similarity comparison process using the Structure Mapping Engine (Falkenhainer, Forbus and Gentner, 1989), a computer implementation of Gentner's (1983) structure-mapping theory of analogy. The simulations indicate that salient object commonalities yield a preference for mappings based on object similarities, while salient relational commonalities yield a preference for mappings based on relational similarities. We test these predictions in five experiments using the one-shot mapping technique which places local and global similarities in opposition. The results support the general conjecture that similarity comparisons involve structural alignment, as well as the more detailed predictions derived from the simulations. These results pose a difficulty for models using representations that do not encode the binding between relations and their arguments.U of I OnlyETDs are only available to UIUC Users without author permissio