Fig. 5. The temporal bias in conversational references drives participants’ asymmetric retrodiction and prediction performance (main experiment).

A Illustration of the annotation approach. We manually annotated references to events in past or future segments in characters' spoken conversations. We matched each such reference with its corresponding storyline event (and its corresponding segment number for onscreen events, or half-step segment number for offscreen events). We then tracked the hit rate separately for referenced versus unreferenced events in participants' uncued retrodictions and predictions. B Reference rate as a function of lag. Across all possible just-watched segments (lag 0), the bar heights denote the average proportions of events referenced in other past or future segments. C Difference in hit rates between all events and unreferenced events. To highlight the effect of characters' references to past and future events on participants' retrodictions and predictions, here we display the difference in across-segment mean hit rates between all events and unreferenced events, as a function of temporal distance (lag) to the just-watched segment. D Hit rates for unreferenced events. The average response hit rates for unreferenced events are displayed as a function of temporal distance to the just-watched segment. Each point represents one segment (paired with a just-watched segment). Panels (B–D): colors are described in the Fig. 4 caption. E Hit rates and counts of referenced and unreferenced events. As a function of temporal distance to the just-watched segment, the sub-panels display the across-segment mean numbers (x-axes) and hit rates (y-axes) of referenced (red) and unreferenced (gray) events that participants hit (darker shading) or missed (lighter shading) in their uncued retrodictions (top sub-panel) and uncued predictions (bottom sub-panel). Intuitively, the widths of the rectangles at each lag denote the total number of events at each possible lag. The darker shading denotes the proportions of events that participants retrodicted or predicted, and the lighter shading denotes the proportions of events that participants “missed” in their responses. For an analogous presentation of results from the replication experiment, see Supplementary Fig. S8.