Generating event boundaries in memory without prediction error

Continuous experiences are segmented into discrete long-term memories through the generation of event boundaries. A leading theory of event segmentation proposes that event boundaries are triggered by prediction errors caused by unexpected stimuli. However, recent studies have raised doubts about whether prediction error is necessary for event segmentation. In this study, we tested an alternative account: that event boundaries in memory reflect the temporal structure of working memory during perception and can occur even in the absence of prediction error. In experiment 1, participants were asked to detect repeats within sequences of random images. The switch to a new sequence could be predictable, with a continuous display of the number of images remaining in each sequence, or unpredictable, with no prior indication to participants that a sequence was about to end. We found boundary-related effects on temporal order memory in both cases, with higher accuracy for the within-sequence comparisons even when the event boundary between sequences was predictable. In experiments 2a and 2b, event boundaries were always predictable, and participants performed either the (event-related) repeat-detection task from experiment 1 or a (non-event-related) 2-back task. We observed event-boundary effects on order memory only when the working memory task was event-related. Both of these experiments support an alternative theory of event segmentation, in which boundaries are critically related to working memory dynamics rather than prediction error.