Jayalath and Parker Jones analyzed the word-aligned brain-to-text framework introduced by d'Ascoli et al.
Decoding words independently beats joint decoding by removing timing shortcuts
Jayalath and Parker Jones show that overlapping word-aligned windows leak word durations, and a simple fix makes aggregation and LLM priors effective, reaching 36.6% word error rate.
Top University
Dulhan Jayalath · Oiwi Parker Jones
Neural Processing Lab (PNPL) · University of Oxford
Research Digest··3 min read
The authors show that reported gains from jointly decoding brain-to-text can be reproduced with synthetic signals containing no brain information, because overlapping windows reveal word durations.
Why this paper
From University of Oxford and Neural Processing Lab (PNPL)
In one line
Most gains in non-invasive brain-to-text decoding come from timing information in overlapping windows, not brain activity.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§