Meaning maps predict reaction time in change detection

To detect changes in our visual environments, the visual system compares pre-and post-change representations maintained in active working memory. Previous research has suggested that change detection is primarily informed by high-level semantics in naturalistic scenes. Here, across two experiments, we used meaning maps - a data driven method to measure the visual semantic information in naturalistic scenes - to investigate whether semantic features predicted visual change detection in a flicker paradigm. Experiment 1 showed that changes in highly meaningful regions were more easily detected than changes in non-meaningful regions despite controlling for low-level visual saliency. Experiment 2 found that the meaning-driven advantage was significantly reduced by scene inversion, further supporting the role of semantics in change detection. Together, these results demonstrate that the visual system relies on semantic features during change detection.