Linguistic Overhypotheses in Category Learning: Explaining the Label Advantage Effect

When learning to partition the world into categories, people rely on a set of assumptions (overhypotheses) about possible category structures. Here we propose that the nature of these overhypotheses depends on the presence of a verbal label associated with a given category. We describe a computational model that demonstrates how labels can either accelerate or hinder category learning, depending on whether or not the prior beliefs imposed by their presence align with the true category structure. This account provides an explanation for the phenomena described in prior experimental work (Lupyan, Rakison, & McClelland, 2007; Brojde, Porter, & Colunga, 2011) that have remained unexplained by other models. Based on these results, we argue that the overhypothesis theory of label effects provides a way to formalize and quantify the effect of language on category learning and to develop a more precise delineation between linguistic and non-linguistic thought.