Abstract
To adapt to the dynamic world, visual working memory(VWM) allows us to store representations of objects and manipulate them in the face of new information. These processes, however, are not infallible. Computational models have revealed three main sources of storage errors: reduction in the precision of representations, forgetting, and object-location misbindings (Bays, Catalao, & Husain, 2009). Conversely, the source of manipulation failures remains unknown. Here, we address this issue by investigating the extent to which the aforementioned errors increase as a function of manipulation vs. storage (Exp1), manipulation load (Exp2), and interference from initial-stored representations (Exp3). Participants saw a display of empty placeholders that were briefly filled with colors (chosen from continuous color space) which then disappeared, requiring participants to store them in VWM. Placeholders either remained stationary or a varying number proceeded to swap positions, requiring the manipulation of color-location bindings. Participants reported the color of a cued item by clicking on a color-wheel. Response errors were separated into the aforementioned categories using the Swap Model. In Exp1, participants stored or manipulated 1 or 2 colors. No differences were observed in forgetting or misbinding rates, though precision decreased as a function of set size, but not manipulation. In Exp2, participants manipulated 2 items, such that two previously-color-occupied placeholders swapped positions (load=2) or one swapped with an empty placeholder (load=1). Precision and forgetting rates were constant across conditions, though misbinding increased as function of manipulation load. In Exp3, we confirmed that this increase in misbindings resulted from an item-limited manipulation resource (increased misbinding when manipulating 3 vs. 2 items) and not interference from a lingering representation (equal misbinding when placeholder moved to a previously-unoccupied-location vs. previously-color-occupied-location). These findings show that the source of errors in VWM manipulation reflect primarily an increased probability of misbinding as the number of manipulations increases.
Meeting abstract presented at VSS 2018