Anticipating changes: adaptation and extrapolation in category learning

Abstract

Our world is a dynamic one: many kinds of objects and events change markedly over time. Despite this, most theories about concepts and categories are either insensitive to time-based variation, or treat people’s sensitivity to change as a result of process-level characteristics (like memory limits, captured by weighting more recent items more highly) that produce irrational order effects during learning. In this paper we use two experiments and nine computational models to explore how people learn in a changing environment. We find, first, that people adapt to change during a category learning task; and, second, that this adaptation stems not only from weighting more recent items more highly, but also from forming sensible anticipations about the nature of the change.