Serial-dependencies in the perception of orientation, number, faces and bodies

Abstract:

It is well known that the perception of stimuli depends strongly on spatial and temporal context. One clear example of temporal contextual effects is serial dependency, the influence of preceding stimuli on the perception of current stimuli in a sequential display.

We suggest that serial dependence reflects an optimal encoding strategy, combining an internal model of the world derived from past experience with current sensory experience, well modeled by an intelligent, adaptable Kalman filter. The Kalman filter model makes clear predictions of how the magnitude of serial dependence should vary with the sensory precision of the current and past stimuli, and the physical difference between the two stimuli.

We tested these predictions for the perception of orientation, numerosity, faces, facial expression, and bodysize. In all cases, the parameter-free Kalman filter model predicted the results, qualitatively and quantitatively. To relate this work to the previous literature, we also measured reaction times.

Reaction times were fastest when the current stimuli were most similar to the previous, and the magnitude of the advantage of reaction time correlated well with the magnitude of serial dependence, across subjects.

All this data is consistent with the notion that perception depends strongly on internal model of the world, constantly updated from sensory experience.