StructureMapping

Structure-mapping theory is a cognitive framework for explaining analogy and similarity. It argues that comparison relies on aligning relational structures rather than surface features. Developed by Dedre Gentner in the 1980s, the theory describes how a base domain maps to a target domain through correspondences that preserve relationships. A key idea is that relational structure, not attributes, governs analogical strength. The theory has been extended computationally as the Structure-Mapping Engine (SME) by Falkenhainer, Forbus, and Gentner to model similar mappings.

Core concepts include base and target domains, relata, and relations. A mapping selects correspondences between base and target terms so that the relational structure is preserved. Relations such as cause, means-end, and role-filler mappings are emphasized over surface attributes like color or size. The evaluation of a candidate mapping rests on coherence and systematicity: a good map preserves higher-order relations and yields a coherent, extendable structure. Inferences from the base can be projected into the target, allowing analogical reasoning to generate new hypotheses or explanations.

Structure-Mapping Engine operationalizes these ideas as an algorithm that searches for mappings with maximal structural consistency. It constructs multiple candidate mappings, scores them for relational preservation and overall coherence, and selects the best. The SME can model a wide range of analogies, from everyday comparisons to scientific reasoning, and it has informed research on how people detect parallels and generate explanations.

Applications and impact: SMT has influenced cognitive psychology, education, and artificial intelligence by clarifying how people learn through analogy, how teachers can use analogies effectively, and how machines might reason by structural similarity. Criticisms include debates over the sufficiency of relational structure to explain all analogy, the computational complexity of searching large spaces, and the role of prior knowledge and context in guiding mappings.