LAUSR

Human activity requires an ability to generalise beyond the available evidence, but when examples are limited – as they nearly always are – the problem of how to do so becomes particularly acute. In addressing this problem, Shepard (1987) established the importance of representation, and subsequent work explored how representations shift as new data is observed. A different strand of work extending the Bayesian framework of Tenenbaum and Griffiths (2001) established the importance of sampling assumptions in generalisation as well. Here we present evidence to suggest that these two issues should be considered jointly. We report two experiments which reveal replicable qualitative patterns of individual differences in the representation of a single category, while also showing that sampling assumptions interact with these to drive generalisation. Our results demonstrate that how people shift their category representation depends upon their sampling assumptions, and that these representational shifts drive much of the observed learning.