Anesthetized Hippocampus Predicts Words in Stories, Nature Study Finds

Neurons in the human hippocampus distinguished word classes and predicted upcoming words in spoken stories while patients lay fully unconscious under general anesthesia, researchers at Baylor College of Medicine report in Nature (2026, volume 654, pages 714 onward; DOI 10.1038/s41586-026-10448-0). The single-neuron recordings, made with Neuropixels probes in epilepsy surgery patients, showed signs of neural plasticity — and they come from a brain structure that this technology had never before been used to probe in this way.

The study reframes a question that reading rooms and monitoring units rarely get to test directly: what the unconscious brain still computes. Its answer, for the hippocampus under one anesthetic regimen, is considerably more than the field assumed.

How the recordings were made

Dr. Sameer Sheth, professor and Cullen Foundation Endowed Chair of neurosurgery at Baylor and a neurosurgeon at Baylor St. Luke's Medical Center, led the team. The patients were undergoing epilepsy surgery under general anesthesia, and the surgical exposure gave the researchers direct access to the hippocampus, the temporal-lobe structure central to forming new memories.

The investigators used Neuropixels probes — implantable devices dense enough to capture electrical activity from hundreds of individual neurons simultaneously. This is intracranial single-unit recording, not scalp EEG; it yields spike-level resolution from named cell populations, a methodological step the researchers say had not previously been applied to this region in this setting.

What did the anesthetized hippocampus do?

Two experiments framed the findings.

• In the first, the team played repeating tones, occasionally interrupted by a deviant sound. Hippocampal neurons responded differently to the unexpected tones, and their discrimination improved over the course of the session — evidence of neural plasticity (experience-driven change in neural responses) operating without awareness.
• In the second, the researchers played short stories while recording the same neurons. Distinct activity patterns carried information about word categories — nouns, verbs, adjectives — indicating the hippocampus tracked meaningful linguistic features, not merely the presence of sound. The same signals contained information that could be used to predict which word would come next in a sentence.

"Our findings show that the brain is far more active and capable during unconsciousness than previously thought," Sheth said. "Even when patients are fully anesthetized, their brains continue to analyze the world around them."

Why the word-prediction result matters

Anticipating the next word reflects predictive coding — the brain's continual use of prior patterns to forecast incoming information. Scientists have typically documented this kind of prediction in awake, attentive listeners. Finding it under general anesthesia suggests at least some of the underlying processes run independently of consciousness.

"This kind of predictive coding is something we associate with being awake and attentive, yet it's happening here in an unconscious state," said Dr. Benjamin Hayden, professor of neurosurgery at Baylor.

The researchers are careful about interpretation. The results do not show that patients consciously understood the stories. They show that complex language analysis can proceed when subjective awareness is absent. That dissociation bears on how consciousness is theorized to arise: perhaps awareness depends less on whether individual regions stay active and more on how information is coordinated across larger networks — networks that anesthesia may disrupt even as the hippocampus keeps working.

The parallel the authors draw to large language models is explicit: both systems exploit statistical patterns in language to predict what comes next. The similarity is functional, not mechanistic, but the team suggests studying it may clarify prediction in both natural and artificial systems.

Implications for speech prosthetics

For brain-computer interface researchers, the most concrete implication is signal viability. If meaningful language signals can be reliably identified in hippocampal recordings — here, single-unit data from implanted probes — those signals could in principle support speech prosthetics (communication devices driven by recorded brain activity) for patients who have lost speech after stroke or injury.

"Can we use these signals to deploy and run a speech prosthetic for some of the parts of the brain that are damaged by stroke or injury? These are questions that we can now consider in relation to this part of the brain," said Dr. Vigi Katlowitz, first author and a neurosurgery resident at Baylor.

What the study does not establish

Three limitations bound the claims. All experiments ran under one particular general anesthetic regimen, so the findings do not automatically extend to other anesthetics or to states such as sleep or coma. The recordings covered the hippocampus only; whether other regions retain comparable abilities remains untested. And the work is investigational — the probes and analyses are research tools, not FDA-cleared clinical monitors, and no bedside application is claimed.

Even so, the evidence is direct: unconsciousness in this setting did not stop the hippocampus from detecting anomalies, improving with experience, parsing word categories, and anticipating upcoming words. "This work pushes us to rethink what it means to be conscious," Sheth said. "The brain is doing much more behind the scenes than we fully understand." The next studies, the team indicates, will need to test other brain regions, other anesthetic states, and ultimately whether these signals can drive a working speech prosthetic.

By Priya Raman