Search Abstracts | Symposia | Slide Sessions | Poster Sessions
A bottom-up neurostimulation model of language across IFG, MFG, MTG, STG, and SMG: 29 elements from 220 papers
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
T R Williamson1,2, Gesa Hartwigsen3,4, Kris Kinsey1,5, Neil U Barua1,2, Naomi Heffer5, Philipp Kuhnke3,4, Sonia Mariotti1,2, Eimear McKnight1,6, Lydia Wiernik1,7,8, Jemma Sedgemond5, Antonia Vogt9, Anna E Piasecki1,2,5; 1Brain, Language, and Behaviour Laboratory, UWE Bristol, UK, 2Southmead Hospital, North Bristol NHS Trust, UK, 3Wilhelm Wundt Institute for Psychology, Leipzig University, Germany, 4Cognition and Plasticity, Max Planck Institute for Human Cognitive and Brain Sciences, Germany, 5School of Social Sciences, UWE Bristol, UK, 6Dept. of Linguistics, Queen Mary University of London, UK, 7SignLab, Dept. of German Philology, University of Göttingen, Germany, 8Neuropsychology, Max Planck Institute for Human Cognitive and Brain Sciences, Germany, 9Dept. of Medicine, University of Cambridge, UK
Introduction. Whether the brain's language regions are best characterised by traditional linguistic categories (semantics, syntax, phonology), by domain-general selectivity (Fedorenko et al., 2024), or by region-specific functional profiles (Genon et al., 2018) remains contested. Causal evidence from neurostimulation — TMS, transcranial electrical stimulation (tES), and direct electrical stimulation (DES) — could in principle adjudicate this question by mapping the specific subprocesses that disruption at a region perturbs. We present a bottom-up region-by-region inventory derived from a preregistered systematic review of the published neurostimulation literature (PROSPERO CRD42024602006). Methods. Searches across PubMed, Scopus, Embase, and PsychInfo returned 12,763 records; 220 papers (1999–2025) met inclusion criteria. The 608 significant stimulation outcomes were qualitatively analysed (Braun & Clarke, 2006) to infer the most specific causally modulated subprocess — an element — and organised within a seven-level hierarchy of process specificity. We then derived per-region element inventories for five canonical language regions by clustering coordinate-resolved outcomes with AI-assisted construct grouping, calibrated to the neuroanatomical hierarchy level of the searched region. Results. Across the five regions, 29 elements were derived from 185 paper-region pairings. IFG (74 papers, N = 2,036): 8 elements — novel form-meaning encoding; semantic relation retrieval; communicative output planning; competing candidate suppression; hierarchical syntactic assembly; lexical retrieval under constraint; sublexical phonological computation; prosody-driven structural interpretation. STG (45, N = 1,161): 5 elements — lexical form access; semantic-conceptual content activation; contextual discourse integration; sub-lexical phonological form computation; verbal sequence maintenance. MTG (26, N = 712): 4 elements — lexical form access; controlled semantic retrieval; automatic conceptual activation; utterance-level regulation. SMG (23, N = 739): 7 elements — sensory-motor feature knowledge retrieval; novel word-form binding; lexical-semantic access; sublexical grapheme-phoneme decoding; phonological sequence retention; phonological pattern matching; syntactic dependency resolution. MFG (17, N = 499): 5 elements — speech output sequencing; memory-guided lexical access; predictive semantic inference; competing response suppression; non-dominant meaning selection. Cross-region patterns are informative: lexical form access recurs at MTG and STG; competing candidate suppression and competing response suppression are realised separately at IFG and MFG; the phonological domain is carved three ways across IFG (sublexical phonological computation), STG (sub-lexical phonological form computation), and SMG (sublexical grapheme-phoneme decoding). Conclusion. No single traditional linguistic construct cleanly delineates any of the five regions. Each region instead supports a heterogeneous bundle of fine-grained operations, recurring across regions in patterns that cross-cut traditional category boundaries. The substantive question for future work is the ontological coherence of these by-region bundles: whether the elements that cluster at each region are unified by shared computational principles, by shared inputs and outputs, or by accident of how the published causal-evidence base has accumulated. Resolving that question is what would license a functional recharacterisation of language regions — relabelling not by inherited linguistic category but by the operations the brain actually performs there (Genon et al., 2018). The full inventory is openly published at language-elements.org with coordinate-resolved access and continuous extension.
Topic Areas: Computational Approaches, Development of Resources, Software, Educational Materials, etc.