Inferring the hypothesis spaces underlying inductive generalization

Sean TauberUniversity of Adelaide
Daniel NavarroUniversity of Adelaide
Amy PerforsUniversity of Adelaide
Michael LeeUniversity of California, Irvine

Abstract

Although there has been much interest in inferring mental representations from similarity data, there have been no attempts at inferring representations directly from generalization data. We develop an approach in which a hypothesis space can be inferred from human generalization data. By defining the likelihood function relating human generalization data to a Bayesian generalization model, we are able to infer the most likely hypothesis space(s) humans used to produce the generalization data. One of the advantages is that, unlike with similarity based approaches, we can explore the effect of semantic context on the hypothesis spaces people use when generalizing.

Files

Inferring the hypothesis spaces underlying inductive generalization (1 KB)



Back to Table of Contents