Abstract
Recent research into probabilistic models of mental representation have profited from requiring participants to give multiple responses on a given trial (Vul et al., 2009). This method assumes that accessing a mental representation leaves it intact for subsequent sampling. Here we tested the consequences of sampling (memory access) on the mental representation of an object's velocity. On each trial, observers saw an object move at a constant velocity and were instructed to continue mentally tracking the object's position as it moved behind a virtual occluder. Either one or three visual markers were posted at a range of distances (early, middle, late) along the length of the occluded path. Observers responded by pressing a key when the object was imagined to have reached each marker. The results demonstrate that the accuracy of the velocity representation depends on whether previous responses were given. When three responses were given, the object's velocity was reliably underestimated (even on the first response), and the degree of underestimation increased with position. Velocity was also underestimated in the single-click condition, but the degree of underestimation did not increase as a function of position [the interaction between number-of-responses and position was highly reliable, F(2,30) = 10.779, p<0.001]. In the three-response condition, the increase in underestimation with each response was amplified at higher speeds. A model with a single free parameter (a temporal delay to updating object position information when a response is given) accurately predicts the mean velocity estimate for the middle and late marker position at 3 different speeds (best fit temporal delay = 40 ms per response). These findings suggest that accessing a mental representation to generate a report briefly delays updating of that mental representation, and more generally that the architecture of working memory may prevent simultaneous reading and writing from memory.