Brain2Qwerty v2 Decodes Natural Sentences from MEG Recordings
August 20, 2026
A new model decodes natural language sentences from non-invasive magnetoencephalography (MEG) data with a 39% average word error rate. The system uses character, word, and sentence-level representations, showing that performance scales log-linearly with increased data volume.
HOW THIS AFFECTS YOU
●
researcherData scaling appears to be a viable path for improving non-invasive neural decoding accuracy.
●
healthThis could bridge the gap between invasive implants and non-invasive brain-computer interfaces for communication restoration.