Abstract
Viewing objects that are semantically inconsistent with the scene they are embedded in — like a toaster on the beach — elicits ERPs with a parietal negativity that peak about 400 ms after scene onset and typically reflect semantic violations. What information of the scene context is sufficient to trigger an N400 response? To answer this question, we created a synthesized texture for each scene with identical summary statistics but without any obvious semantic meaning. We then presented objects on either images of scenes, their texture versions, or a scrambled control background that maintained the colors of the image, but no higher-order statistics, resulting in three types of backgrounds: Real-world scenes, textures, and controls. To create semantic inconsistencies, we paired indoor and outdoor scenes with either a consistent or an inconsistent object thumbnail. The object was centered on the background and both were presented simultaneously. We found a parietal negativity in the N400 time window (350-600 ms) for consistent versus inconsistent objects on real-world scenes. Interestingly, objects on inconsistent texture backgrounds also elicited an N400 response with a similar time-course and topography, though less pronounced. The control condition, however, showed no such response. These results first of all replicate previous findings of N400 responses to semantically inconsistent objects in scenes. More importantly, our data suggest that the mere presentation of the summary statistics of a scene is sufficient to lead to integration costs of objects that are semantically inconsistent with their scene context. These costs are reflected in a less pronounced, yet substantial N400 response compared to real-world scenes. Future electrophysiological and/or behavioral investigations may replicate these preliminary findings and clarify the role of summary statistics for the semantic integration during object and scene processing.
Meeting abstract presented at VSS 2016