Skip to content
DraconiticResearch
Search Sources API
197,768 studies31 sourcesUpdated 14 min ago
Peer-reviewedCC BY-NC-NDNature Neuroscience

Noninvasive decoding of typed sentences from human brain activity

  • Jarod Levy1
  • Mingfang Zhang2,3
  • Svetlana Pinet4,5
  • Jérémy Rapin6
  • Hubert J. Banville6
  • Stéphane d’Ascoli7
  • Jean-Rémi King8
8 affiliations
  1. 1Meta AI, Paris, France. [email protected]
  2. 2Hospital Foundation Adolphe de Rothschild, Paris, France
  3. 3École Normale Supérieure, Université PSL, CNRS, Paris, France
  4. 4Basque Center on Cognition, Brain and Language, San Sebastian, Spain
  5. 5Univ. Lille, CNRS, UMR 9193 -SCALab-Sciences Cognitives et Sciences Affectives, Lille, France
  6. 6Meta AI, Paris, France
  7. 7Meta AI, Paris, France. [email protected]
  8. 8Meta AI, Paris, France. [email protected]
Open at sourcePDFDOICite

Abstract

Modern neuroprostheses can now restore communication in patients who have lost the ability to speak or move. However, implanting these invasive devices comes with risks inherent to neurosurgery. Here we introduce a noninvasive method to decode the production of sentences from brain activity and demonstrate its efficacy in a cohort of 35 healthy volunteers. For this, we present Brain2Qwerty, a new deep learning architecture trained to decode sentences from either electro- or magnetoencephalography, while participants typed briefly memorized sentences on a QWERTY keyboard. With magnetoencephalography, Brain2Qwerty reaches, on average, a character error rate of 29% and substantially outperforms electroencephalography (character error rate: 65%). For the best participants, the model achieves a character error rate of 18%, and can perfectly decode a variety of sentences outside of the training set. Overall, these results narrow the gap between invasive and noninvasive methods and thus open the path for developing safe brain–computer interfaces for noncommunicating patients. Here the authors introduce Brain2Qwerty, a deep learning model that decodes typed sentences from non-invasive brain activity with character error rate down to 18%. This opens a pathway for non-invasive communication neuroprostheses.

Related

  1. Human learning of noninvasive brain–computer interfaces via manifold geometry
    Erica L. Busch, E. Chandra Fincke, Guillaume Lajoie +2
    Nature NeurosciencePeer-reviewedLicence not stated7 citedNeuroscienceOpenAlex
  2. Decoding how RNase E membrane targeting controls RNA degradosome activity in Pseudomonas aeruginosa
    Sandra Amandine Marie Geslain, George Edward Allen, Johan Geiser +2
    ISciencePeer-reviewedCC BYFull textGenetics & genomicsEurope PMCCrossref
  3. A Compendium of 49 Experimental SBS Signatures for Decoding Human Cancer Mutational Processes
    M. Zhivagui, J. N. Au, S. Sharma +5
    bioRxivPreprintCC BYFull textGenetics & genomicsbioRxiv
  4. Real-time intracortical decoding from human motor cortex reveals a task-independent commitment to act that aligns with conscious intent
    C. Verbaarschot, A. de Miguel Gomez, A. Alamri +5
    bioRxivPreprintCC BY-NC-NDFull textNeurosciencebioRxiv
  5. Simulated transcranial ultrasound of human language cortex: verifying heard speech needs far less data and signal than decoding it
    Omer Shalev
    arXivPreprintarXiv licenceFull textNeurosciencearXiv

Cite

BibTeXRIS
BibTeX
@article{levy2026noninvasive,
  title = {{Noninvasive decoding of typed sentences from human brain activity}},
  author = {Jarod Levy and Mingfang Zhang and Svetlana Pinet and Jérémy Rapin and Hubert J. Banville and Stéphane d’Ascoli and Jean-Rémi King},
  journal = {Nature Neuroscience},
  year = {2026},
  month = jun,
  doi = {10.1038/s41593-026-02303-2},
  url = {https://doi.org/10.1038/s41593-026-02303-2},
  note = {CC BY-NC-ND},
}
RIS
TY  - JOUR
TI  - Noninvasive decoding of typed sentences from human brain activity
AU  - Jarod Levy
AU  - Mingfang Zhang
AU  - Svetlana Pinet
AU  - Jérémy Rapin
AU  - Hubert J. Banville
AU  - Stéphane d’Ascoli
AU  - Jean-Rémi King
PY  - 2026
DA  - 2026/06/29
JO  - Nature Neuroscience
DO  - 10.1038/s41593-026-02303-2
UR  - https://doi.org/10.1038/s41593-026-02303-2
AB  - Modern neuroprostheses can now restore communication in patients who have lost the ability to speak or move. However, implanting these invasive devices comes with risks inherent to neurosurgery. Here we introduce a noninvasive method to decode the production of sentences from brain activity and demonstrate its efficacy in a cohort of 35 healthy volunteers. For this, we present Brain2Qwerty, a new deep learning architecture trained to decode sentences from either electro- or magnetoencephalography, while participants typed briefly memorized sentences on a QWERTY keyboard. With magnetoencephalography, Brain2Qwerty reaches, on average, a character error rate of 29% and substantially outperforms electroencephalography (character error rate: 65%). For the best participants, the model achieves a character error rate of 18%, and can perfectly decode a variety of sentences outside of the training set. Overall, these results narrow the gap between invasive and noninvasive methods and thus open the path for developing safe brain–computer interfaces for noncommunicating patients. Here the authors introduce Brain2Qwerty, a deep learning model that decodes typed sentences from non-invasive brain activity with character error rate down to 18%. This opens a pathway for non-invasive communication neuroprostheses.
SN  - 1097-6256
SN  - 1546-1726
ER  -