Abstract
Previous studies demonstrated that visual representations in working memory exhibit biases in relation to the categorical structure of the stimulus space. For example, color representations are biased away from the nearest category border and orientation representations are biased away from the nearest cardinal axis. However, a majority of those studies are based on behavioral measures of working memory (e.g., delayed estimation), and thus it is unclear whether the bias is driven by the shift of representation per se or by the processes that translate the memory representations into a behavioral response. Here, the present study attempted to find evidence that the actual content of working memory is categorically biased even before a response is made using a recently developed EEG decoding method. In Experiment 1, EEG was recorded while participants were performing a simple location delayed estimation task. I found that the decoding of location was biased away from the nearest cardinal location, consistent with the biases observed in the behavioral responses. Follow-up analyses showed that the categorical bias started as early as the stimulus encoding stage, indicating that the categorical bias in behavior was originated from the perception of the stimulus. Experiment 2 replicated these findings using a 2AFC orientation working memory task where participants reported the remembered orientation using a button press to indicate which of the two probe items matched the orientation of the original sample object. Because the two probe items were simultaneously presented in locations different from the location of the sample object in the task, this result indicates that the categorical bias in decoding was not merely driven by a location-based response preparation strategy. Together, these results provide neural evidence that working memory representations themselves are categorically biased, imposing important constraints on the computational models of working memory representations.