Fig. 4. Retrodictions and predictions of events at different temporal distances.

A Illustration of the annotation approach. For each uncued retrodiction and prediction response, we calculated the number of (retrodicted or predicted) events as a function of temporal distance from the target segment, or lag. Onscreen (explicit) events are tagged using integer-valued lags, whereas offscreen (implicit) events are tagged using half-step lags (± 0.5,  ± 1.5, etc.). B Number of events hit in participants' (n = 36) uncued retrodictions and predictions for each event type (main experiment). Here we separated events we identified in participants' responses according to whether they occurred in the target segment (lags of  ± 1), during the interval between the target segment and the just-watched segment (lags of  ± 0.5), at longer temporal distances (∣lag∣ > 1), or were incorrect (unmatched with any past or future events in the narrative). The counts displayed in the panel are averaged across just-watched segments. C Number of events hit as a function of temporal distance (main experiment). Here the (across-segment) mean numbers of events hit in participants' uncued retrodictions (orange) and predictions (blue) are displayed as a function of temporal distance to the just-watched segment (lag). Each point represents one segment (paired with a just-watched segment). D Number of events hit in participants' (n = 37) uncued retrodictions and predictions for each event type (replication experiment). Same format as Panel (B). E Number of events hit as a function of temporal distance (replication experiment). Same format as Panel (C). Colors denote temporal direction (orange: past; blue: future) and distance (darkest shading: target events; darker shading: onscreen non-target events; lighter shading: offscreen events).