Search Abstracts | Symposia | Slide Sessions | Poster Sessions
A frontotemporal network processes conceptual relatedness during spoken narrative comprehension
Poster Session E, Sunday, September 14, 11:00 am - 12:30 pm, Field House
Hannah Mechtenberg1, Jamie Reilly2, Jonathan E. Peelle3, Emily B. Myers1; 1University of Connecticut, 2Temple University, 3Northeastern University
INTRODUCTION. The semantic landscape of a narrative necessarily evolves over time. Listeners must, then, integrate content across multiple timescales to iteratively construct meaning. As discourse unfolds, listeners integrate incoming words with preceding context; a process linked to activity in frontotemporal regions. Whereas there is a long history of functional imaging studies of spoken sentence processing, little work has examined how the brain responds to shifting semantic relationships between local and more global elements of spoken discourse processing. Discourse offers the opportunity to use a naturally-evolving speech stream to identify brain networks that are sensitive to the evolution of a narrative across multiple timescales. Leveraging the richness of spoken narratives is not new. However, emerging work has largely focused on measuring word-level and concept-level neural activation, leaving the jump to larger timescales largely unexplored. A useful metric by which to examine how listeners contend with an unfolding discourse is semantic distance (Reilly et al., 2023), wherein the dissimilarity between two or more concepts can be quantified. The current fMRI study modeled semantic shifts across multiple timescales using a novel n-gram measurement index that yielded distance metrics between successively larger chunks of words. Our aim was to elucidate brain regions that are sensitive to fluctuation(s) in semantic distance in naturalistic discourse. We hypothesized that larger jumps in semantic space would require additional neural support—perhaps from frontotemporal regions—not needed for making small hops. METHODS. We examined data from Mechtenberg et al., (2024), wherein a total of 79 participants listened to a 10-minute segment of the podcast Fresh Air. We calculated semantic distance of content words from progressively larger units of preceding context using the ‘SemanticDistance’ R package (Reilly et al., 2025). Subject-level analyses used an amplitude-modulated regressor time-locked to the onset of the second word event. For example, in the word-to-word model, values were time-locked to the onset of the second word. We included lexical frequency and phonological neighborhood density as covariates. The output, at the group-level, allows us to examine neural activation scaled to the semantic distance between gram sizes. Results were corrected for whole-brain statistical significance at the cluster level using permutation testing. RESULTS and DISCUSSION. For the word-to-word model, activity in the bilateral posterior superior temporal gyrus, bilateral fusiform gyrus, right cuneus, and left superior occipital cortex showed a negative relationship with semantic distance, such that as semantic distance increased, activity in these regions decreased. Right angular gyrus, however, showed a positive relationship with semantic distance. We interpret these findings as reflecting semantic integration at a moment-to-moment timescale, which may rely less on regions that are implicated in selective attention (e.g., Yoncheva et al., 2010) and in directing more challenging processes—such as integrating semantic content over larger chunks. The results of the larger n-gram models of semantic distance show different patterns, lending credence to the hypothesis that narrative construction may require listeners to rely on additional brain regions to support long-scale content building.
Topic Areas: Meaning: Discourse and Pragmatics, Meaning: Lexical Semantics