Serial Dependence in Temporal Perception Reveals the Dynamics of Constructing an Internal Reference Frame

Temporal perception is crucial to cognitive functions. To better estimate temporal durations, the observers need to construct an internal reference frame based on past experience and apply it to guide future perception. However, how this internal reference frame is constructed remains largely unclear. Here we showed the dynamics of the internal reference construction from the perspective of serial dependence in temporal reproduction tasks. We found the current duration estimation is biased towards both perceived and reproduced durations in previous trials. Moreover, this effect is regulated by the variability of sample durations. The influence of previous trials was stronger when the observers were exposed to context with more variable durations, which is inconsistent with previous theories that the similarity between successive stimuli induces serial dependence. We proposed a Bayesian model with an adaptive reference updated continuously after each observation, which can better explain the serial dependence observed in temporal perception.