Spatial constraints on probability learning in visual working memory

Abstract

ABSTRACTVisual working memory is limited in capacity so it is essential to use it efficiently. Previous work has shown that statistical learning can help boost working memory efficiency by prioritizing the encoding and/or maintenance of objects most likely to be tested. In this study, we considered that the potential benefits of statistical learning could be limited by spatial constraints. Across three experiments, we found that statistical learning prioritizes working memory allocation to items based on their likelihood of being tested, but this prioritization is greatly modulated by spatial constraints. In particular, when two locations each had a high probability of being tested, we primarily observed performance benefits over low probable locations when these two locations were horizontally adjacent to one another. Vertically adjacent and diagonally arranged high probable locations produced no accuracy benefit over low probable locations and a modest response time benefit. These findings contrast with previously observed hemifield-independent effects (i.e., a “bilateral field advantage”) and reveal surprising limitations on the potential benefits of statistical learning.