What Do We Remember When We’re Wrong: Memory for Unexpected Words and Their surrounding context

Prediction error (PE) is said to support memory encoding (Henson & Gagnepain, 2010; van Kesteren et al., 2012). However, findings are mixed (Hubbard & Federmeier, 2024; Gambi et al., 2024; Höltje & Mecklinger, 2022), and few studies have examined PE’s effects on surrounding context.

The present study comprised two experiments. In each experiment, we employed a two-phase design with either no delay (Experiment 1) between phases or a 24-hour delay (Experiment 2). During the first training phase, participants heard sentences that biased them toward either a strong or weak prediction for a target word. Instead of the predicted word (lure), they encountered a plausible but unexpected alternative that either strongly violated their prediction (high PE) or only weakly violated it (low PE). Memory was later assessed using a word recognition task and a semantic relatedness judgment task.

Experiment 1 showed that participants produced false memories for predicted lure words but showed no recognition advantage for unexpected words. However, participants tended to form associative links between the lures and the unexpected alternatives. Experiment 2 was conducted to examine whether this effect persisted over time, but it did not yield statistically significant results. With increased statistical power, the combined data suggest a pattern consistent with Experiment 1. These findings provide no evidence that prediction error enhances recognition memory for unexpected words but suggest that it may promote associative links between predicted and unexpected words.

Link to session