Constraining Analogical Inference with Memory-based Verification
Diarmuid O’Donoghue · 1999
From, Analogies complete their primary task of generating new knowledge about a problem domain, based on a noted similarity between the problem and some other familiar domain. Analogical inferences are generated as a form of pattern completion occurring in the target domain, but not all comparisons between structurally similar domains generate valid inferences. We highlight the need for a domain independent verification mechanism. We describe an extended model of analogical reasoning that includes an explicit verification mechanism, employing existing memory to verify candidate inferences. We focus on the soundness of first-order predicates by comparing candidate inferences with background memory, with greater similarity leading to quicker and clearer verification.