Baylor Neuroscientists: The Anesthetized Brain Still Decodes Language and Predicts Upcoming Words

Even when a person is fully unconscious under general anesthesia, their hippocampus continues to listen, learn and predict the meaning of upcoming words, according to a study published this week in Nature by neuroscientists at Baylor College of Medicine. The finding, drawn from neural recordings in 17 epilepsy patients undergoing brain surgery, is the most direct evidence yet that the human brain performs sophisticated language processing without conscious awareness — and it is forcing a sweeping rethink of what consciousness contributes to cognition.

The Baylor team, led by Dr. Daniel Yoshor and Dr. Brett Foster of the Department of Neurosurgery, took advantage of a rare clinical opportunity. Their patients, who were preparing for surgery to treat drug-resistant epilepsy, had Neuropixels probes — silicon arrays capable of recording from hundreds of individual neurons at once — temporarily inserted into the hippocampus during routine intraoperative mapping. As the patients lay anesthetized with propofol, the surgical team played them sequences of beeps, common words and short sentences in English. The probes captured neural firing rates from more than 5,000 hippocampal neurons across the cohort.

The hippocampus, long known to be central to memory formation, was conventionally thought to fall silent under deep anesthesia. The Baylor recordings showed the opposite. Within ten minutes of the start of the auditory sequences, the anesthetized hippocampus had become measurably better at distinguishing routine tones from rare "oddball" tones — a hallmark of perceptual learning that requires the brain to update its internal model of the auditory environment. "The brain was learning," Foster told reporters in Houston. "Not just reacting, but learning, while the patient could not open their eyes or respond in any way."

More striking still, the firing patterns carried clear signatures of word meaning. Hippocampal neurons fired in similar ways for words from the same semantic category — "cat" was close to "dog" but distant from "pen" — and the team was able to decode upcoming words in heard sentences from the activity that preceded them. "The hippocampus was running a predictive model of what would come next," said co-first author Dr. Andrew Watrous. "That is exactly the kind of computation we used to think required wakeful attention."

The results sharply complicate the long-held textbook view that loss of consciousness suspends complex cognition. They also raise practical questions for surgical anesthesia, intensive-care sedation and end-of-life care, where families and clinicians have long wondered how much an unresponsive patient takes in. The Baylor authors stressed that their findings do not establish that anesthetized patients form lasting memories — earlier behavioral studies suggest most do not — and that the work was confined to a single brain region under one specific anesthetic. "We are not saying you remember surgery," Yoshor said. "We are saying your hippocampus is doing real work while you do not."

Independent commentators called the paper a turning point. Sophie Schwartz, a cognitive neuroscientist at the University of Geneva, wrote in an accompanying News & Views piece in Nature that the study "redraws the boundary between conscious and unconscious processing more decisively than any single experiment in the past 20 years." Liina Pylkkänen, a psycholinguist at New York University, said the predictive coding signature in particular "tells us that meaning is being assembled at a level that does not require awareness." The Baylor team is now extending the protocol to other anesthetic agents and to brain regions outside the hippocampus, work supported by a five-year $11 million grant from the National Institutes of Health.