Intermediate Features and Informational-level Constraint on Analogical Retrieval

Two different sorts of retrieval in analogical tasks, novice-like and expert-like, have been demonstrated in psychological experiments. With recent computational work in object recognition as an inspiration, we propose the computation of intermediate features, and their use as triggers for retrieval, as the relevant constraint at the informational level of characterization of the retrieval process as seen in experts. We conduct computational experiments which show that, in conjunction with a feature-matching retrieval mechanism, features of an intermediate size and complexity give the strongest ana-