Abstract
We propose that fixed imprecision in representations of location produces correspondence problems that constrain working memory performance. To test this hypothesis, we replicated Bays & Husain (2008). Participants briefly saw 1-6 colored squares followed by a single probe horizontally offset. Reports on the direction of an offset varied probabilistically with its magnitude, and response functions were fit with a cumulative Gaussian distribution. As in previous work, the reciprocal of the estimated response function SD declined with increases in memory load. We suggest that this did not reflect a decline in precision, but an increasing probability that the location of the probe item was compared to the wrong memory representation – a correspondence problem failure. Consistent with this hypothesis, fixing the total number of displayed items at 12, while varying the subset to-be-remembered, depressed precision and rendered it invariant by load. Similarly, we found unchanging precision for a colored singleton among an increasing set of uniform items, though the singleton was no more likely to be probed than any other item. Critically, in another experiment, a preview display appeared quickly just before test, consisting of all items except the to-be-probed one. This manipulation, which solved any correspondence problem, resulted in unchanging precision as a function of load. Further experiments found unvarying precision at larger loads (6–8) when a preview was made available; that the gains accrued by previewing items depended on their distances from the probe; and that the precision of orientation memory likewise remained fixed when location previews were available. Collectively, these results imply that measured limits on working memory performance reflect failures to compare current states of the world with the right pieces of stored information. Such correspondence failures emerge because memory is imprecise, but the degree of imprecision is fixed by inherent limitations on visual processing.