Semantics Drive Noticing Rates in Realistic Change Blindness Scenarios

Abstract

Change blindness is the inability to perceive visual changes when the transient signals associated with the change are interrupted or absent. Previous research has found that both low-level perceptual features and higher-level semantic information can guide visual attention. However, in immersive environments, their relative contributions to change detection remain unclear. Here we investigate the effects of low-level and semantic similarity on change detection in virtual reality (VR) while controlling for bottom-up salience. Using a between-subjects design, we conducted an experiment in which n = 64 participants performed a visual search task consisting of two phases: (1) a target search phase, and (2) a change detection phase. Changes were induced using a blink-contingent paradigm, where the manipulated objects teleported between two fixed positions during eye blinks. A Cox proportional hazards model was used to analyze detection performance, which revealed that the semantic similarity between the target and the manipulated object was a stronger predictor of detection than low-level similarity. These findings argue that, contrary to many results using simple displays, higher-level semantic relationships play a dominant role in guiding attention during change detection in more realistic environments.