Encoding conditions shape temporal memory precision by modulating temporal uncertainty and temporal bias

Temporal memory about when in the past something happened is suggested to be reconstructed rather than recalled. Participants usually show a degree of mismatch between remembered and actual temporal position of an event. However, recent studies showed markedly different results, including both relatively low temporal precision, for visual objects presented earlier in a series, and relatively high temporal precision, for movie scenes shown at the start of a movie. One explanation could be the use of different stimulus materials, for which participants would employ different cognitive encoding or mnemonic strategies. However, a more parsimonious explanation would be that temporal judgments arise from a common mechanism, regardless of stimulus material or context. In the current manuscript, we reanalysed the results of two previously published experiments and investigated the effect of boundary segmentation and semantic relatedness during encoding in two new experiments. We found that participants showed more temporal uncertainty when the encoded visual objects and contexts were unrelated. Further, we found that increasing the semantic associations during encoding diminished temporal uncertainty but increased temporal underestimation bias, which we interpret as an indication of temporal compression. A simple computational model in which temporal judgment is based on a Gaussian process defined by temporal uncertainty (dispersion) and temporal bias (location) replicated the empirical data of all four experiments, suggesting that patterns of temporal errors observed in different experiments arise from a common mechanism. The model further underscores that semantic relatedness between items decreases temporal uncertainty but enhances temporal compression. These findings have important ramifications for how we memorize the temporal structure of events.