Abstract
Most computational and psychophysical work in intuitive physics treats time as continuous. Yet, from infant cognition to language, discrete interactions between objects and external forces dictate how we think about them: objects collide, fall, topple; liquids splash, slosh, pour; cloths flap, billow, drape. Do we also see physical scenes as structured over time during online perception? We propose that the visual system uses a set of intuitive physical rules to segment continuous experience into a sequence of physical events: causal contexts that are sufficient to explain motion patterns of objects within that event. In Experiment 1, we showed participants videos of three-dimensional physical scenes with objects rolling, sliding, colliding, toppling, falling, and entering or exiting containers and occluders. Participants described these physical scenes containing individual events. Using natural language processing, we verified that videos could be distinguished by the event types that compose them. In Experiment 2, we investigated whether we represent temporal boundaries between physical events in perception. Participants viewed combinations of events (e.g., a ball collides with another ball, which then goes behind an occluder) with or without a temporal probe (a 125ms interval in which the video slows down) and pressed a key immediately upon perceiving a probe. Probes occurring at boundaries between events, compared to probes at non-boundaries, were significantly less likely to be detected. Participants were also significantly less confident in their judgments of these probes, and even when these probes were correctly detected, reaction time was slower. Importantly, absolute pixel change was not different between boundary vs. non-boundary probes. These results suggest that the mind spontaneously imposes discrete temporal structure over continuously unfolding state-space of objects and forces, providing support for physical event representations as building blocks of the mind and a tractable “model organism” for studying event cognition in neurological and modelling studies.