Daryl Fougnie¹ (), Ying Zhou^1,2, Anek Rajbhandari¹; ¹New York University Abu Dhabi, ²New York University

Only a limited amount of information can be maintained in working memory (WM). We have control over how this capacity is allocated–retrocue paradigms demonstrate that participants can prioritize a subset of information. However, the other side of this coin is less clear–can participants deprioritize or drop cued items from memory and does doing so improve memory for the remaining items? Previous studies have observed benefits of forgetting, but critically, in these paradigms deprioritized information is accompanied by increased priority in the other items, raising the possibility that participants are not benefiting from forgetting but translating forgetting cues into ‘prioritize other’ cues. For example, in Williams and Woodman (2012), participants were instructed to retain or forget half of the items. Both conditions improved performance, but the forget instruction took many trials to yield effects, consistent with participants learning to translate ‘forget’ into ‘prioritize other’. We developed a novel paradigm that disentangles prioritize and forget cues. Specifically, subjects had to remember three orientations and report one of them after a brief delay. In each trial, they were either provided with only a prioritize cue indicating an item more likely to be tested, or both prioritize and forget cues (the latter indicating an item that would never be tested). Unlike previous attempts, forget cues did not increase the probability of other items being tested, due to the inclusion of no-report trials equal to the proportion that the dropped item would have been tested. In two experiments (differing only if cues were blocked or intermixed) we observed benefits of forgetting, but curiously these benefits were found for the neutral and not the prioritized item. These results are consistent with the idea that forgetting can benefit the remaining items in WM, but tentatively suggest that such benefits are not spread evenly among items.

Acknowledgements: NYUAD Research Institute grant CG012