Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Quantifying information gain in narrative stimuli using open-source large language models

Aidan Goeschel¹, Rohin A. Palsule¹, Janice Chen², Aaron M Bornstein¹, Ari Khoudary¹; ¹University of California, Irvine, ²Johns Hopkins University

Presenter: Aidan Goeschel

We present an open-source, ensemble-based implementation of the Sequentiality metric and demonstrate the feasibility of using it to approximate an eligibility trace that guides human credit assignment in causally complex environments. To do this, we demonstrate that the original Sequentiality metric—designed to quantify narrative “flow” in transcriptions of episodic recall—can equivalently be shown to quantify contextual information gain. We test this interpretation by applying the metric to stimuli and human behavior from the FilmFest dataset , and show that Sequentiality significantly correlates with the importance and causal centrality of events, but not their semantic centrality. This work expands the range of questions Sequentiality can be used to investigate, and takes an important first step toward understanding the neural and computational processes facilitating credit assignment in humans in naturalistic, multi-causal environments.

Topic Area: Methods, Tools, Theory & Neural Coding