Electrophysiological Evidence for Dynamic Temporal Priors From Neural Signatures of Temporal Prediction Errors

Increasing evidence shows that time perception can also be described as a form of unconscious Bayesian inference. A controversy in Bayesian modeling of time perception revolves around whether individuals rely on the statistical regularity of input durations as priors (statistical priors) or dynamically update the priors with each estimate (dynamic priors). In a time reproduction task, participants were asked to reproduce a series of sample durations within both short and long time ranges, while the electroencephalography (EEG) was recorded synchronously. Behavioral results showed that within each range, short durations were overestimated while long durations were underestimated. Additionally, the reproduced time increased proportionally with the duration of the preceding trial. Prediction errors (PE) were calculated based on both types of priors. Within each range, the statistical and dynamic priors both predicted the PE decreased as the sample duration of the current trial increased, while only the dynamic prior predicted PE increased with the sample duration of the preceding trial. EEG results showed that the offset of the sample duration elicited a frontocentral distributed P3a. Within each range, the P3a amplitude decreased as the current sample duration increased, and increased as the duration in the previous trial increased. The PE calculated based on dynamic priors more accurately predicted the P3a amplitude compared to the PE calculated using statistical priors. These results suggest that the P3a serves as a neural signature of temporal PE and provide electrophysiological evidence supporting the dynamic prior hypothesis. Taken together, our study demonstrated a link between Bayesian models and predictive coding theory, offering a viable neural implementation of Bayesian predictive coding.