Schedule of Events | Symposia

Sketchpad Series

Adaptive control of episodic memory retrieval during story listening

Poster Session C - Sunday, March 30, 2025, 5:00 – 7:00 pm EDT, Back Bay Ballroom/Republic Ballroom

Cody V Dong¹ (codydong@princeton.edu), Samuel A Nastase¹, Kenneth A Norman¹; ¹Princeton University

Prior work has shown that people use episodic memory to predict upcoming words when listening to familiar stories. Can people adaptively stop making memory-based predictions in environments where memory is misleading? In the first fMRI session, participants listen to 78 short stories. In the second session, we manipulate the predictability of the stories between subjects: In the high predictability condition, participants hear the exact same stories from the first session. In the low predictability condition, the stories start the same but always branch off to different endings, so participants who use episodic memory to predict the ending will predict incorrectly; to avoid these prediction errors, we hypothesize that participants in this condition will learn to suppress episodic memory retrieval during the repeated, pre-branch part of the story. To measure episodic memory retrieval, we first train subject-specific encoding models to predict brain activity from story embeddings. Then, we compare encoding performance during the second session using embeddings that are informed by memory for the complete original story, compared to regular embeddings that only reflect current information. We operationalize neural recall as the difference in encoding performance between memory-informed embeddings vs. regular embeddings. In keeping with our hypothesis, in a pilot sample of 4 participants, we find preliminary evidence that participants in the low predictability condition show a lower degree of neural recall compared to high predictability participants, suggesting that people modulate memory retrieval based on the predictability of the environment.

Topic Area: LONG-TERM MEMORY: Episodic