Prediction error is out of context: The dominance of contextual stability in structuring episodic memories.

Abstract

Our everyday experiences unfold continuously, yet we segment them into distinct memory units-a phenomenon known as event segmentation. Although extensively studied, the underlying mechanisms of event segmentation remain controversial. This study addresses this by comparing the two contrasting theories: prediction error and contextual stability. Across four experiments, we manipulated these factors separately to examine their distinct impacts on event segmentation, measured by temporal order and distance tasks. Experiments 1-3 demonstrate that contextual stability leads to more pronounced event segmentation than prediction errors in unstable contexts, underscoring its critical role. Experiment 4 further supported this by providing strong evidence for equally robust event segmentation for predicted and unpredicted transitions across stable contexts. We conclude that contextual stability plays a pivotal role in driving event segmentation, outweighing the effect of prediction errors. This study sheds new light on how our minds encode continuous experiences into coherent and meaningful memory units.