Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
Distinct prefrontal networks for semantic integration and articulatory planning
Poster B72 in Poster Session B and Reception, Thursday, October 6, 6:30 - 8:30 pm EDT, Millennium Hall
Also presenting in Poster Slam B, Thursday, October 6, 6:15 - 6:30 pm EDT, Regency Ballroom
Leyao Yu1, Nikolai Chapochnikov2, Werner Doyle2, Orrin Devinsky2, Adeen Flinker1,2; 1New York University School of Engineering, 2New York University School of Medicine
The spatiotemporal neural dynamics in frontal cortex underlying speech production and word retrieval remain poorly understood. Traditionally, the inferior frontal gyrus (IFG) has been implicated in various aspects of language processing including articulatory, syntactic, and semantic processes (Frederici, 2012; Hickok & Poeppel, 2007). Growing evidence has implicated the posterior MFG (area 55b) and dorsal precentral gyrus as critical for language (Chang, 2020; Ozker, 2022). In order to investigate the spatiotemporal dynamics of speech articulation and semantic integration, we leveraged a battery of language production tasks within a cohort of 24 neurosurgical patients undergoing Electrocorticographic monitoring. We hypothesized that articulatory preparation, lexical retrieval, and semantic integration may not be localized to one region but rather may be distributed in nature. All participants underwent the same tasks including picture naming, word reading, auditory naming, auditory word repetition, and auditory sentence completion, designed to produce the same unique 50 words in each task but with a distinct route of retrieval (randomly interspersed within the block). Focusing on high gamma broadband (70-150 Hz), we reported the spatiotemporal dynamics across peri-sylvian cortices locked to language perception and articulation. A region of interest analysis provided unique profiles of task-related neural recruitment per region during perception and production across all significant electrodes (sustained activity for over 100ms compared to baseline, t-test p<0.001). The neural profiles replicated previous temporal dynamics for sensory perception (STG and occipital cortices), speech planning (IFG), and motor execution (pre- and post- central gyri). However, we found specific enhancement in IFG and MFG related to semantic load. In order to test how distributed these networks were, we employed an unsupervised clustering approach to uncover dynamics in a data driven manner without an anatomical bias. This approach revealed two new networks in frontal cortex with differentiated articulatory and semantic specificity. One cluster of electrodes, centered in IFG and precentral cortices, showed comparable activity prior to speech onset across all tasks. The second cluster, centered on the border of IFG and MFG, showed strong task-selectivity prior to articulation with significantly greater (t-test, p<0.001) activity for tasks requiring semantic access. In order to investigate the nature of these two clusters, we performed representational similarity analysis across the auditory naming and sentence completion tasks. Neural covariance across time locked to perception in the semantic cluster was highly correlated with the semantic embeddings of the changing auditory stimuli (based on the last 3 layers of a deep neural network GPT-2 model, spearman correlation p<0.001 rho= 0.53 and 0.27, respectively for each task) which quickly diminished as speech production onset approached. The articulatory cluster, however, was strongly correlated with phonetic information rather than semantic embeddings prior to speech onset across tasks. Our results suggest two distinct language production components distributed across frontal cortices, a preparatory motor-related component agnostic to task, and a component recruited specifically as semantic integration is required.
Topic Areas: Language Production, Speech Perception