The contribution of low-level sensory features versus knowledge-based factors for encoding of objects in working memory

Brief presentation of a complex scene entails that only a few objects can be fully processed and stored in memory. Both low-level sensory salience and high-level semantic (e.g. conceptual mismatch between an object and the scene) contribute to this selection process. We used fMRI to investigate these factors during encoding in WM. Subjects were presented with complex scenes that on some trials included also an out-of-context object. Following a retention interval, participants judged the location of a target-object extracted from the initial scene. The target was selected according to its high/low salience and its in-/congruence with the scene. Retrieval performance increased with increasing object salience, particularly so in conditions entailing a greater target uncertainty (context-congruent trials). Encoding-related fMRI results showed that target saliency (high vs. low) activated the superior parietal cortex specifically in context-congruent trials. By contrast, the encoding of incongruent trials recruited the inferior parietal cortex, suggesting knowledge-based focussing of attention toward the out-of-context object. A control condition with pre-cued targets indicated that this effect was not entirely due to high certainty about the target identity during encoding. We conclude that activity in parietal cortex reflects perceptual and semantic factors for the selection of objects for encoding in WM.