Abstract
It is tempting to think that visual long-term memory (VLTM) is composed of the visual features of objects we encounter. However, there is a long tradition in the memory literature of thinking of memories as inherently contextual. VLTM for objects may thus not only be based on visual features but also on non-visual features of the encoding context, such as the current task. We ran an fMRI study to examine this possibility and address two questions: (1) Are tasks previously performed on an object retrieved during subsequent object recognition? (2) How are such associations updated when we have to perform a new task on the object? In a first phase, observers were exposed to a sequence of objects, which were randomly assigned to one of two tasks: How easy would it be to draw the object? (artist task); How useful is the object? (function task). Some of these objects were presented again in a second phase with a different task: How natural is the object? (organic task). To address the first question above, we applied multivariate pattern analysis and found that, despite being irrelevant, the artist or function task initially performed on each object was automatically reactivated in the brain. Now, however, two tasks had been performed on the object, raising the question of which task(s) persisted in memory. We hypothesized that the reactivation of the first task during the organic task would trigger competition between these representations and result in the first task being weakened. Indeed, greater classifier evidence for the first task during the second phase was linked to an increased likelihood of later forgetting that this task had been performed on the object. These findings emphasize the contextual nature of visual memory and the role of competition-dependent learning in updating its contents.
Meeting abstract presented at VSS 2015