Overwriting and rebinding: Why feature-switch detection tasks underestimate the binding capacity of visual working memory

Abstract

In these two experiments, we explored the ability to store bound representations of colour and location information in visual working memory using three different tasks. In the location-cue task, we probed how well colour information could be recalled when observers are given a location cue. In the feature-cue task, we probed how well location information could be recalled when observers are given a colour cue. Finally, in the feature-switch detection task, we tested how well observers could detect a recombination of features (e.g., switching the locations of the red and green items). We hypothesized that these tasks might reveal differences in binding capacity limits between switching and nonswitching tests of visual working memory. We also hoped the tasks could provide an explanation for those differences in terms of the component processes of working memory—do failures occur in the encoding, maintenance, or retrieval stages of the task? Experiment 1 showed that performance in the two cued-recall tasks was equally high, and was significantly better than performance in the feature-switch detection task. Thus, the feature-switch detection task underestimates the number of colour–location bindings that can be remembered, but is a useful task for examining the fragile nature of feature binding in working memory. Experiment 2 explored why feature-switch detection underestimates the binding capacity of visual working memory by examining whether the feature switch errors occur at the level of encoding, maintaining, or retrieving binding information from visual working memory. The results suggest that feature switch errors reflect failures to maintain bound objects in working memory, perhaps due to the automatic rewriting and rebinding of information in the face of new perceptual input.

Keywords:

• Working memory
• Binding
• Overwriting
• Cued recall
• Feature switch

Acknowledgements

GAA was supported by NIH/NEI fellowship no. F32EY016982. TWT was supported by an NDSEG grant from the Office of Naval Research.