Decoding silent reading from EEG: open-vocabulary word retrieval from non-invasive brain signals
Decoding silent reading from non-invasive EEG
A new study demonstrates that open-vocabulary word-level information can be decoded from non-invasive EEG during silent reading. Using a contrastive decoder trained on about 240,000 word presentations from a single participant over 49 hours, the model achieved above-chance top-10 retrieval for words, including mid-frequency and rare ones, with performance scaling log-linearly with data volume. Removing occipital electrodes reduced word decoding but not context tracking, and control analyses separated lexical decoding from narrative context and positional priors.
These results establish that open-vocabulary word-level information is recoverable from EEG during silent reading, and that decoding is data-limited rather than saturated.
- dustfinger
It would be nice to add a "right to private thought" to the Bill of Rights and the Canadian Charter of Rights and Freedoms, as well as something similar to other countries' rights and freedoms laws wherever possible.
- PMunch
Having proper sub-vocalization interfaces would be so cool! A 19-channel EEG does seem a bit invasive though (from a social aspect, I understand that it's used here in a medical context).
- SubiculumCode
Am I reading it right that they are claiming about ~%15 word prediction accuracy?
- b00ty4breakfast
The Torment-Nexus industrial complex continues to grow. Would really love to have a conversation with researchers who work on these kinds of things. I'm sure there are peripheral avenues that this research will open up, but how do you just ignore the very obvious dystopia that you are working to bring to fruition?
Deemphasizing the humanities in STEM education is turning into a disaster.
- tbenst
Put another way, EEG moves top-10 retrieval out of 512 candidate words from 20.3% -> 27.7%. Hardly "open-vocabulary word decoding"
The paper is also missing a critical control of feeding Llama the preceding ground-truth text and asking how well it ranks the current word among the same candidate set.