Brain-to-text decoding rides a timing shortcut; removing it cuts word error rate to 36.6%
pnpl · hf · 2026-10-01
Researchers show that reported gains in decoding words from non-invasive brain recordings are largely reproducible without any brain data: windowing brain activity per word leaks word intervals through window overlap, letting networks exploit word-duration statistics. The method hits 22.0% balanced accuracy on synthetic signals with no brain information versus 22.3% on real recordings. Processing each window independently removes the shortcut, and both aggregating multiple responses per word and using a pretrained LLM as a linguistic prior become far more effective. The resulting SimpleB2T recipe achieves 36.6% word error rate with five observations per word, approaching past invasive speech decoding under different conditions.
More from Research
- AI Atlas: Data Coverage as a Potential Fix for World Model Hallucination — rudina11 · 2026-10-01
- BlockSearch: a 0.6B in-context retriever rivals dense retrieval at million-token scale — CShorten30 · 2026-10-01
- Eval-cooperativeness alignment research wins Corrigibility Research Fund prize — dhadfieldmenell · 2026-10-01
- 0.8B model plus 9 LoRA adapters routes agent decisions 38x faster with +8.7 accuracy — Usual_Maximum7673 · 2026-10-01
- The first AI-discovered cancer drug reportedly just worked — theimposingshadow · 2026-10-01
- MedKIT Benchmark at NeurIPS 2026 Shows LLMs Can Recall Updated Facts But Fail to Use Them — zeynepakata · 2026-10-01