Prediction error is out of context: The dominance of contextual stability in segmenting episodic events

Abstract

Our everyday experiences unfold continuously, yet we naturally segment them into distinct memory units—a phenomenon known as event segmentation. While event segmentation is a well-explored topic, its underlying mechanisms have remained a subject of debate. In this study, we address this debate by comparing the two contrasting theories of event segmentation: prediction error and contextual stability. In two experiments, we manipulated contextual stability while keeping prediction error constant. Event segmentation was more prominent when there were stable contexts (Experiment 2) compared to prediction errors among unstable contexts (Experiment 1). This stronger segmentation took place despite signs of a smaller prediction error in Experiment 2. We conclude that contextual stability plays a pivotal role in driving event segmentation, highlighting its dominance over the role of prediction errors. Our study sheds new light on the mechanisms behind how our minds divide continuous experiences into meaningful memory units.

Effects of context changes on memory reactivation

“…We experience the world continuously but perceive and store events that are segmented into meaningful parts (Zacks & Swallow, 2007). This segmentation in episodic memory reflects changes in contextual features such as background color, sound, location, or task rules (Clewett et al, 2017; Heusser et al, 2018; Horner et al, 2016; Raccah et al, 2022; Wang et al, 2022; Güler et al, 2023; Güler et al, 2024; Nolden et al, 2024). These findings suggest that contextual changes influence how an item is preserved in LTM.…”