BCIwiki (bciwiki.com) — University of Essex researchers decoded semantic categories from EEG, with mean accuracies of 68.2% to 73.4% across three pairwise tasks, all above chance. The paper appeared in the Journal of Neural Engineering on October 21, 2026.
The Graded Inventory of Semantic Triggers (GIST) task asked 14 participants to retrieve concepts from increasingly specific verbal clues. Targets concerned food, musical instruments or body parts. Participants pressed the spacebar when they recalled a concept, allowing EEG analysis to be aligned to that moment and reducing interference from perception of the clues.
An encoder first learned shared features from other participants. It was then frozen and transferred to a participant excluded from that stage. A support vector machine was calibrated using that person’s labelled items under nested leave-one-item-out validation. Individual calibration remained part of the procedure.
Mean accuracies were 73.4% for body parts versus food, 70.7% for body parts versus instruments and 68.2% for food versus instruments. Participant-level permutation tests reached significance in 9/11 people for the first task and 8/11 for each of the other two (p < 0.05). The denominator of 11 differs from the 14 people whose EEG was recorded; the public abstract does not explain the difference.
The interval 1.5 to 0.5 seconds before the key press was most discriminative. These findings concern semantic categories in a controlled task; arbitrary thought decoding and practical communication for people with severe motor impairments were not tested.