Proceedings of the Annual Meeting of the Cognitive Science Society

Understanding how statistical regularities result in category learning requires access to the underlying psychological spacesin which these categories are represented. However, uncovering these spaces, especially in developmental settings, posessignificant experimental and methodological challenges what are relevant dimensions on which these spaces are organizedand how can we uncover them without prohibitively long or straining experiments?Here, we propose a novel way of uncovering these spaces. We learn participants implicit similarity functions, instantiatedas a neuronal network, by training on simple groupings of stimuli. In simulations, we show that our method can recovergroup-specific categorical structures. Furthermore, we show that young children quickly understand the grouping task, andspaces can be obtained in short, engaging experiments. Finally, we apply our method to uncover age-related differencesin category representations. In an experiment contrasting 4-5, 6-7 year-olds, and adults, we find that the learned spacesexhibit age-specific feature biases.