Abstract
Current models of visual working memory (VWM) have very different views about the structure of VWM, while they all appear to share assumptions about its contents. Specifically, current models assume (1) that the contents of memory are best characterized as values within continuous spaces (e.g., a value on a hue circle), and (2) that memory operates identically on all contents (e.g., experimental conditions being equal, red is remembered the same way as green). Here, we propose a revised model for accurately characterizing the contents of working memory, a model that can then be generalized to study the limits of color working memory. Categorization is an intuitive and universal aspect of color perception, and one that likely involves early visual processing. We have recently shown that category structure has large influences on behavioral responses in the widely used delayed estimation task, and also in an undelayed version of the task. We demonstrate that these influences can be parsimoniously explained by a dual content model that integrates a continuous value with a noisy category assignment. The model parameters are not fit; instead, we derive estimates of category centers and boundaries from a separate set of behavioral experiments with different observers. Finally, we demonstrate the applicability of the model to other stimulus features (e.g. orientation), and how existing models of VWM can incorporate dual contents to test hypotheses about memory limits. Currently, inferences about memory limits rest on differences in behavioral performance resulting from experimental manipulations (e.g. set size). But without accurate characterization of memory contents, stimulus-specific variations in performance are confounded with effects of experimental manipulation. The dual content model more directly situates working memory in its appropriate place as an intermediary between perceptual inputs and behavioral responses, rather than a stand-alone system.
Meeting abstract presented at VSS 2015