Statistical learning of task relevance, rather than stimulus prevalence, improves visual working memory recall

Our visual environment often overwhelms us with more information than we can process. We can, however, enhance our processing efficiency by leveraging inherent regularities in the environment, such as prioritizing likely target objects over unlikely ones. Although there is evidence that statistical learning improves visual working memory performance, it is unclear whether this improvement occurs through stimulus prevalence or task relevance. To distinguish between these hypotheses, we examined whether frequent appearance or frequent probing of stimuli (or both) contributes to improvements in visual working memory recall. Participants were asked to recall and replicate the orientation of one of two previously presented Gabors as precisely as possible. In two experiments, we manipulated (1) stimulus prevalence by presenting Gabors more frequently on one side (either left or right), and (2) task relevance by probing Gabors more frequently on one side (either left or right). We found comparable orientation recall performance for to-be-remembered stimuli appearing at probable versus improbable locations, which suggests that regularities in stimulus prevalence did not improve memory recall. Contrastingly, we found better recall for stimuli appearing at locations that were more versus less likely to be probed. Specifically, task relevance enhanced working memory recall, both by reducing the number of categorical errors (e.g., misremembering a vertical-oriented Gabor as a horizontal one) and by increasing fine-grained recall precision (i.e., reducing minor variances in recall for memorized items). These findings highlight that statistical learning of task relevance, but not of stimulus prevalence, enhances visual working memory recall performance.