Search Abstracts | Symposia | Slide Sessions | Poster Sessions
On-Demand Assembly of Articulatory Representations in Ventral Sensorimotor Cortex
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
J. Raouf Belkhir1, Arka N. Mallela2, Julien Dirani1, Catherine Liegeous-Chauvel2, Jiahao J. Chen2, Eliza Reedy2, Steven Salazar2, Luke Henry2, Thandar Aung2, Jorge Gonzalez-Martinez2, Bradford Mahon1; 1Carnegie Mellon University, 2University of Pittsburgh Medical Center
Introduction: Distinguishing whether articulatory codes are preassembled during lexical access or instead constructed only when motor initiation is triggered is a central unresolved question in theories of speech production. Speech requires transforming perceptual or lexical-semantic representations into articulatory motor plans implemented in the ventral sensorimotor cortex (vSMC). Although vSMC becomes active shortly after picture onset and well before articulation (1), it remains unclear whether this reflects preassembled word-specific content or a nonspecific preparatory state. Here, we used human stereo-EEG and delayed picture naming to test whether vSMC contains articulatory structure during lexical access or instead assembles articulatory representations only near speech initiation. Methods: Stereo-EEG (SEEG) data were collected from 25 patients performing picture-naming tasks. Fourteen completed a delayed-naming paradigm that dissociates stimulus processing from motor initiation using go-cue delays of 0, 400, and 1000 ms. Eleven additional patients completed an immediate-naming version. Broadband γ (70-150 Hz) activity was extracted from vSMC, IFG (including pars orbitalis, triangularis, opercularis), and early visual and ventral temporal cortices. Nonparametric representational similarity analyses (RSA) were computed across temporal sliding windows, comparing neural representational structure against visual and articulatory feature models. Visual cortices provided benchmarks for early, delay-invariant perceptual encoding, allowing us to test whether articulatory structure in vSMC emerges during lexical access or only near response initiation. Results: Behavioral response-time analyses reveal a double dissociation: lexical-semantic factors dominate in immediate-naming trials, whereas articulatory complexity selectively slows responses in delayed conditions (2). Neural RSA mirrored these behavioral effects. Occipital and basal temporal cortices showed early, delay-invariant representational structure that aligned with deep CNN-derived features (VGG16) (3). In contrast, vSMC showed no articulatory structure in the early post-stimulus period but exhibited a robust Delay × Time interaction, with articulatory representations emerging only after response initiation is cued (~200ms before speech onset). Discussion: These findings suggest that articulatory representations are assembled on demand rather than preassembled during lexical access. Early perceptual encoding in visual cortex provides a benchmark for evaluating when higher-level representations emerge. Together, these findings refine the temporal architecture of speech planning by showing that vSMC articulatory codes are not preassembled during lexical access but are constructed on demand when speech is initiated. 1. Strijkers, K et al. 2017. NeuroImage, 163, 206-219. 2. Janssen, N., et al. 2008. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(1), 249. 3. Simonyan, K. et al. 2014. arXiv preprint arXiv:1409.1556.
Topic Areas: Speech Motor Control, Control, Selection, and Executive Processes