Decoding silent reading from non-invasive EEG
Open-vocabulary word information was recoverable from dry-electrode EEG during silent reading.
The paper uses silent reading as a scalable proxy for inner speech, with about 240,000 word presentations from one densely sampled participant over roughly 49 hours. A contrastive decoder aligned short EEG windows with language-model word embeddings and beat permutation baselines on top-10 retrieval. Performance extended beyond frequent words, improved with more data, and showed no saturation in the reported scaling. Removing visual-region electrodes cut word-level gain by about a third, while context tracking remained. ArXiv · AI/CL/LG's note
The paper uses silent reading as a scalable proxy for inner speech, with about 240,000 word presentations from one densely sampled participant over roughly 49 hours. A contrastive decoder aligned short EEG windows with language-model word embeddings and beat permutation baselines on top-10 retrieval. Performance extended beyond frequent words, improved with more data, and showed no saturation in the reported scaling. Removing visual-region electrodes cut word-level gain by about a third, while context tracking remained. ArXiv · AI/CL/LG's note
score 5