Abstract
One of the primary goals of the visual system is to make predictions about upcoming sensory events, which requires extracting and learning regularities from the environment. Research on visual statistical learning demonstrated that humans can not only learn these regularities from very brief exposure (Fisher & Aslin, 2001), but that this learning can occur at the categorical level; when images are always different, but regularities are present across categories (Brady & Oliva, 2008). In the present set of studies, we ask whether this category-level generalization of learning occurs even with exposure to repeated items. Additionally, we ask whether children learn categorical-level regularities like adults. Given evidence that children are more sensitive to features of individual items than adults (Sloutsky & Fisher, 2004) they may or may not show learning of regularities at the category level. We tested this question by performing several statistical learning experiments with adults (18-22 yo) and children (6-9 yo). Participants were exposed to a stream of animal images, which consisted of four sets of triplets, randomly distributed. Each triplet consisted of three animals appearing in the same order. Critically, the images were always novel exemplars from a given category. Observers were tested using a 2AFC Familiarity Judgment task between triplets of images, which either maintained or violated the temporal predictability from the exposure phase. We observed that adults and children showed familiarity to triplets in the exposed sequence, which suggests that they learned the regularities at the category level. To determine whether this categorical abstraction is automatic, we exposed children and adults to an image stream with regularities at the item level and tested with multiple novel exemplars. We found that both adults and children could still learn the statistical regularities at the category level even with single item exposure, suggesting that they can generalize item-level learning to categories.
Meeting abstract presented at VSS 2018