Abstract
Working memory, the ability to retain task-relevant information in an accessible state over a brief span of time, is strikingly limited. Models of working memory explain these limitations by postulating a finite resource that is divided among stored items in a continuous or quantized manner, and they assume that the quality of memory representations is determined solely by the number of items that the resource is spread among, being otherwise fixed for each individual. Here, we consider the possibility that the precision of a memory is not fixed within an individual, but varies across reports. We model performance on a standard working memory task where participants are asked to remember the colors of a set of colorful dots, and then after some delay, to report the color of a dot selected at random (e.g., Zhang & Luck, 2008). In the fixed-precision model, the participant either remembers something of the probed dot, in which case errors are normally distributed with fixed precision centered on the true color, or remembers nothing about the item and guesses blindly, in which case errors are distributed uniformly. In our variable-precision model, precision itself varies across reports and is normally distributed. We found that the variable-precision model produced a significantly better fit than the fixed-precision model for each participant. Additional experiments found that this variability cannot be explained by variation in state-based fluctuations in attention or arousal, by uneven allocation of a finite resource, or by differences in the color or position of tested items. Instead, there appears to be substantial variability in the precision of representations, which is inconsistent with the assumptions of existing models of working memory capacity, including both slot- and resource-based models. We propose a new framework in which representational variability arises from stochastic processes that play out independently across representations.
Meeting abstract presented at VSS 2012