Theoretical and Computational Approaches to Reasoning by Analogy

This paper examines the existing research concerning both formal and computational approaches to analogy. It first reviews the approached taken in analogical reasoning; and discusses whether analogy is best modelled as similarity-based, deductive, or graph isomorphic problem, or whether analogy is best modelled as a cognitive process. Then it reviews major computational models of analogy, including case-based reasoning, derivational analogy, purpose-directed analogy, NLAG, the Structure-Mapping Engine, ACME, and Copycat and Tabletop. it categorizes each of these as a primarily similarity-based approach to analogic comparison, a deductive approach, or a graph-matching theoretic approach. it examines the weaknesses and areas for further research inherent in all of these approaches, focusing on the areas of problem reformulation and search for a suitable source analogy. It then suggests a framework for a new model of analogy, called dynamic analogy, which views analogy as a cognitive process that is utilized in reasoning about complex, process-oriented situations. It uses Situation theory as a framework to address the issues of context and reformulation, and attempts to address the search issue, in part, in the framework of ecological attunements