Level 2 perspective-taking distinguishes automatic and non-automatic belief-tracking

Little is known about whether human beings' automatic mindreading is computationally restricted to processing a limited kind of content, and what exactly the nature of that signature limit might be. We developed a novel object-detection paradigm to test adults' automatic processing in a Level 1 perspective-taking (L1PT) context (where an agent's belief, but not his visuospatial perspective, is relevantly different) and in a Level 2 perspective-taking (L2PT) context (where both the agent's belief and visuospatial perspective are relevantly different). Experiment 1 uncovered that adults' reaction times in the L1PT task were helpfully speeded by a bystander's irrelevant belief when tracking two homogenous objects but not in the L2PT task when tracking a single heterogeneous object. The limitation is especially striking given that the heterogeneous nature of the single object was fully revealed to participants as well as the bystander. The results were replicated in two further experiments, which confirmed that the selective modulation of adults' reaction times was maintained when tracking the location of a single object (Experiment 2) and when attention checks were removed (Experiment 3). Our findings suggest that automatic mindreading draws upon a distinctively minimalist model of the mental that underspecifies representation of differences in perspective relative to an agent's position in space.