Abstract
The visual information retained following a change in gaze is limited and depends on the distribution of attention in the pre-saccadic fixation. It is commonly accepted that objects that have been attended are retained in visual memory in the form of “object files”, representing bound features. The current study suggests that this is too general a description, and the precise nature of the information required for the immediate task determines the information acquired and maintained. In order to approximate the attentional demands of ordinary behavior, Subjects performed a sorting task in a virtual environment with haptic feedback. Each trial required participants to select one of five bricks based on a particular feature (color, width, height or texture) and then to sort the brick onto one of two conveyor belts on the basis of that feature. A change was made to one of the features on about 10% of trials, and Subjects were told to discard the brick into a “waste bin” if they detected a change. The relevant feature was fixed for any one subject, but varied across groups. When color was relevant, color change was detected often (83%) and no changes in any other feature were detected. When width was relevant, color change detection dropped (38%) and width changes were noticed (25%). Similarly, only when height was relevant were height changes detected (42%). Performance was predicted by the relevance of each of the four features in the task in addition to intrinsic salience of the different features. Trials in which subjects did not identify a change nonetheless frequently showed response changes in hand movement. We conclude that working memory representations are not necessarily composed of integrated objects. Rather, visual memory representations depend on the precise information required by the task, and directing attention to an object does not necessarily ensure a coherent representation.