The Unconscious Brain Can Process Language and Predict Words

Researchers at Baylor recorded hundreds of hippocampal neurons with Neuropixels probes in patients under general anesthesia and found that the unconscious brain continues to perform sophisticated auditory and language processing. In an oddball tone task neurons reliably distinguished unexpected sounds and showed experience-dependent changes consistent with plasticity. When short stories were played, distinct neural patterns encoded lexical categories such as nouns, verbs and adjectives, and activity predicted upcoming words in sentences - evidence of predictive coding operating without subjective awareness. The work, published in Nature, demonstrates that hippocampal circuits can analyze meaning-bearing features of language and anticipate continuations even when patients are anesthetized.

These results challenge the idea that consciousness is required for complex information processing and suggest awareness depends on how activity is coordinated across widespread networks rather than on isolated regional processing. The findings draw a parallel with large language models, which generate text by statistical prediction, and point toward practical uses such as decoding language signals for speech prosthetics and brain-computer interfaces. Important limits remain: recordings were confined to the hippocampus during one anesthetic regime in epilepsy surgery patients, so generalization to other anesthetics, sleep, coma, or interactions with other brain regions is not established.