A Structure-Mapping Model of Raven's Progressive Matrices

Abstract

We present a computational model for solving Raven’s Progressive Matrices. This model combines qualitative spatial representations with analogical comparison via structuremapping. All representations are automatically computed by the model. We show that it achieves a level of performance on the Standard Progressive Matrices that is above that of most adults, and that the problems it fails on are also the hardest for people. Keywords: Analogy, Spatial Cognition, Problem-Solving