Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Quantifying information dynamics in speech perception
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Chantal Oderbolz1, Joan Orpella1; 1Georgetown University Medical Center, Washington, D.C.
During speech perception, the brain transforms acoustic input (the speech signal) into abstract linguistic representations. Previous work indicates that distinct subregions of the left superior temporal gyrus (STG) encode representations of units such as phonemes, syllables, and words (Chang et al., 2010; Damera et al., 2023; DeWitt & Rauschecker, 2012) whereas the right STG hosts representations of prosodic units (Oderbolz et al., 2026). These representations are organized hierarchically in ventral processing streams, becoming increasingly abstract and complex in a posterior to anterior gradient along the STG (pSTG → aSTG). What remains unclear is how these representations emerge, i.e. what information is communicated within a stream and how that information contributes to computation. Here, we address these questions using a novel information-theoretic approach to characterize and quantify the information dynamics within the ventral language streams. We first validated this information-theoretic approach using fMRI data acquired during a binaural integration task (Preisig et al., 2021). In this task, an ambiguous consonant-vowel (CV) syllable (midway between /da/ and /ga/) is presented to the right ear, while a disambiguating cue (a high vs. low chirp within the third formant frequency range) is delivered to the left ear. Because successful identification of the CV syllable requires the integration of complementary acoustic features distributed across ears/hemispheres, this task allowed us to probe information dynamics both within and across ventral streams. Our analyses show that the disambiguating feature is initially processed intra-hemispherically in the right hemisphere, propagating from Heschl’s gyrus to the pSTG. Unique information carried by left and right pSTG is then integrated at the level of the right aSTG, consistent with the emergence of a unified speech percept. This integrated information is subsequently transmitted to the left anterior insula. Critically, these information dynamics are strongly predictive of listeners’ ability to correctly identify the CV syllable. Next, we quantified the information dynamics underlying hierarchical processing in the right ventral stream. In a previous study (Oderbolz et al., 2026), we showed that the processing of fundamental frequency (f0) information follows a spatiotemporal progression along the right STG. At early processing stages, acoustic representations of f0 were associated with more posterior regions whereas abstract representations emerged more anteriorly at later processing stages. Listeners relied on these abstract representations to classify stimuli according to their sentence type (question vs. statement). Using our information-theoretic approach, we show that speech perception re-organizes information dynamics between the right pSTG and aSTG relative to resting state. We observe synergistic interactions at the level of the right aSTG, suggesting a role for this region in integrating information, leading to the emergence of abstract representations. Indeed, synergistic interactions increase along the posterior-to-anterior axis of the right STG, underscoring the hierarchical organization of the right ventral stream. Furthermore, these information dynamics are highly stable across participants and differ strongly between correct and incorrect trials, suggesting that they are critical for behavior. Conceptualizing information as a dynamic and decomposable entity provides the necessary level of nuance to better understand how complex cognition and behavior arise.
Topic Areas: Speech Perception,