Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Neural Encoding of Syntactic Structures during Natural Speech Planning and Production
Poster C44 in Poster Session C, Wednesday, October 25, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Also presenting in Lightning Talks C, Wednesday, October 25, 10:00 - 10:15 am CEST, Auditorium
This poster is part of the Sandbox Series.
Piermatteo Morucci1, Mamady Nabé11, Sebastian Sauppe2,3, Martin Meyer2,3, Timothée Proix1, Balthasar Bikel2,3, Anne-Lise Giraud1,4; 1University of Geneva, 2Department of Comparative Language Science, University of Zurich, 3Interdisciplinary Center for the Study of Language Evolution, University of Zurich, 4Institut Pasteur, Université Paris Cité, Inserm
Spoken language is a highly efficient and structured means of conveying information, enabling humans to transform thoughts rapidly and accurately into meaningful messages. Central to this expressive capacity is our ability to manipulate complex syntactic configurations, transforming conceptual meaning into hierarchically organized sequences of words. However, the precise mechanisms by which the brain constructs such syntactic structures during speech planning and production remain largely unsettled. Existing studies on sentence planning and production have primarily used techniques with limited temporal resolutions or employed artificial tasks, leaving a significant gap in our understanding of the neural dynamics supporting the generation of syntactic structures during natural speech. To address this gap, we analyzed intracranial brain recordings obtained from French-speaking patients with intractable epilepsy while they engaged in spontaneous speech production, such as talking about past events or narrating a story. After transcribing and aligning the speech material, we leveraged state-of-the-art natural language processing (NLP) models to extract linguistic features that define the syntactic structure of the patients’ produced sentences. These features were based on constituency (e.g., syntactic classes; the number of opening and closing syntactic nodes, or “top-down” and “bottom-up” chunking schemes) and dependency tree structures (e.g., thematic labels, dependency distances, number of left- and right-hand side dependencies), as well as probabilistic measures such as syntactic conditional probabilities (e.g., syntactic surprisal). To characterize how constituency tree structures are encoded in the neural signal during spontaneous speech planning and production, we first quantified the number of syntactic nodes at each word following both a top-down and bottom-up parsing strategy. We then fit multivariate temporal response function (TRF) models to predict brain responses based on the two parsing schemes. Preliminary findings indicate that decreases in broadband high-frequency activity in the left middle temporal gyrus and middle frontal cortices are predictive of the number of opening syntactic nodes at each word, in line with a top-down chunking scheme. This analysis will be extended to other syntactic dimensions such as syntactic class, dependency structure, and syntactic conditional probabilities, to examine how these features are encoded in cortical and subcortical brain regions during spontaneous speech planning and production.
Topic Areas: Syntax and Combinatorial Semantics, Language Production