Abstract
Items maintained in working memory cannot be studied independently; memories are biased by other information, like long-term priors (e.g., color categories: Bae et al. 2014), recent history (e.g., serial dependence, Fischer & Whitney, 2014) and the other contents simultaneously maintained in working memory (e.g., repulsion biases and ensembles; Chunharas et al. 2019). To examine the relative strength of these sources of bias by amplifying them, we implemented a within-subject iterated-reproduction paradigm (the reported color from trial N would be used as the stimulus in a later trial e.g., trial N+20). In Experiment 1 and 2, a single color was remembered and reported on a color wheel. Experiment 1 found that the drift in the reported colors within-subject (t=6.0, p<0.001) was influenced by color categories, but this drift somewhat differed between subjects (t=4.7, p<0.001). In Experiment 2 we manipulated the color in the trial proceeding each iteration and found that serial dependence modulated the drift significantly (t=2.1, p<0.05), but some convergence to categories was still apparent. In Experiment 3 we asked subjects to remember two colors on each trial and manipulated the accompanying color on iteration trials. We find that the repulsion away from the accompanying color is much larger in magnitude than both the serial dependence bias and that the apparent drift in the iterated chains is such that converges to color categories is almost entirely obscured by such repulsion (t=4.4, p<0.001). In addition, biases were modulated by the iterated chain’s initial color, suggesting that long-term priors somewhat constrain the interactions within working memory. Overall, these findings suggest that information maintained simultaneously in working memory has a much larger impact than either recent history or long-term memory priors, but long-term priors still impose constraints on these interactions.