Narrative linking during encoding drives associative inference

Poster Session D - Monday, April 15, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Shuran Tang¹ (rayna@wustl.edu), Zachariah Reagh¹; ¹Washington University in St. Louis

We often make inferences about novel associations based on prior experiences. This ability, associative inference, is thought to be a key process by which memory guides future behavior. Prior studies of associative inference have typically instructed participants to use narratives or mental imagery to aid in memory of associations between objects. However, these studies have not explicitly tested the function of narratives in this process, leading to an important unanswered question: how crucial are narratives in associative inference, and what role might they be playing? We used a modified associative inference paradigm to answer these questions. Importantly, our design factorially combined two factors during associative encoding: use of a narrative, and explicitly linking the two objects. Participants separately learned AB and BC object pairs within ABC triads using four sets of instructions: (1) generate stories to link objects (Story Together), (2) generate stories for objects independently (Story Separate), (3) compare physical properties of two objects (Compare Together), and (4) describe physical properties of objects individually (Describe Separate). We tested participants’ ability to make inferences about AC associations as a function of encoding strategy. We found that active linking conditions have higher AC accuracy. This suggests that both narratives and explicit linking matter for associative inference. Ongoing analyses are using Large Language Models (BERT) to examine the prediction that narratives shape semantic representations to align associative links and promote successful inference.

Topic Area: LONG-TERM MEMORY: Episodic