Learning overhypotheses

Abstract

Inductive learning is impossible without overhypothe-ses, or constraints on the hypotheses considered by the learner. Some of these overhypotheses must be innate, but we suggest that hierarchical Bayesian models help explain how the rest can be acquired. The hierarchi-cal approach also addresses a common question about Bayesian models of cognition: where do the priors come from? To illustrate our claims, we consider two specific kinds of overhypotheses — overhypotheses about fea-ture variability (e.g. the shape bias in word learning) and overhypotheses about the grouping of categories into on-tological kinds like objects and substances.