Abstract
The ability to recognize other’s emotions is critical for social interactions. Despite constant fluctuations in visual signals, we perceive emotions to be relatively stable over time. Recent studies demonstrate that serial dependence may contribute to this: the perception of facial expressions is systematically biased towards recently seen similar expressions. To the extent that there are physical autocorrelations in the world, this serial dependence could be adaptive. Here, we used natural movies to measure autocorrelations in physical displays of emotion, and we measured serial dependence in perception using the same stimuli. Observers rated the emotions of target characters in 4057 static frames extracted from video clips of Hollywood movies and documentaries. Frames from different videos were shuffled and presented to each observer in a random order, such that visual stimuli in consecutive trials were independent. Observers were asked to report the valence and arousal of the character in each trial. We found that emotion ratings in the current trial were significantly pulled by the emotions seen in previous trials, despite the fact that the sequence of images was random. Further analysis showed that the emotion seen up to 10 or more seconds in the past influenced current emotion judgments. To test whether this effect of serial dependence in perception is associated with the intrinsic autocorrelations in natural stimuli, we quantified the physical autocorrelation of emotion in our videos. We found significant autocorrelations in the displayed emotions in the movies and documentaries, and they remained significant for lags up to 10 or more seconds. This temporal tuning is similar to that of perceptual serial dependence. These results suggest that continuity fields introduce serial dependence in perception that matches the physical autocorrelations in the world, which could facilitate the stability of emotion perception.