Do You Bind? The Representation of Visual Features and Objects in Working Memory

Abstract

A central debate in visual working memory research centres on the nature of visual working memory representations. The object account proposes that the fundamental units of visual working memory are integrated representations, such that items are stored in working memory in an all-or-none fashion. The feature-channel account, on the other hand, proposes that visual features are stored independently of each other. According to this account, additional representations are required to code for the associations between visual feature dimensions. Change-detection tasks are a commonly used paradigm to investigate the representational format of visual working memory. Performance in detecting novel visual features (novel changes) can be compared with detecting a recombination of visual features (binding changes) in this paradigm. While presenting a single-item as test display produces equal performance in novel and binding changes, there is disagreement about whether multi-item displays preferentially disrupt performance in binding changes. In Study 1, I showed that while a multi-item test display results in worse performance for binding conditions relative to novel colour and novel letter conditions when novel locations are used for the test display, this effect is alleviated if an informative retroactive cue is presented in the delay period. In Study 2, I used a two-alternative forced choice task to investigate the representational format of visual working memory. Participants were presented with three coloured shapes which they had to maintain in visual working memory. They were then presented with two items and had to select, using the mouse, which item was from the original display (the target). The other item (the lure) was either a recombination of visual features from the original memory display (binding condition) or contained a novel visual feature (novel colour and novel shape conditions). Mouse trajectories showed greater complexity and curvature for the binding condition, supporting the hypothesis that visual working memory stores information in parallel visual-feature channels. Study 3 explored whether the critical process disrupting working memory for binding information in change-detection tasks with multi-item test displays tasks is the perceptual binding of visual features. This was achieved by presenting two versions of a secondary visual search task in the delay period of a change-detection task with a single-item test display. One was a conjunction search which required the binding of visual features; the other was a “pop-out” search. Only the conjunction search task disrupted working memory for binding information more so than maintenance of visual features. Finally, a neural-cognitive model is proposed that highlights the role of posterior parietal regions in coordinating the binding of visual features across the domains of visual perception and visual working memory.