Recording hundreds of individual neurons directly from the hippocampus of patients undergoing epilepsy surgery, Baylor College of Medicine researchers found the brain continues to process language at a surprisingly sophisticated level even under general anesthesia — distinguishing between nouns, verbs, and adjectives, and generating neural signals that could be used to predict a word before it was spoken. The study, led by Dr. Sameer Sheth, Professor and Cullen Foundation Endowed Chair of Neurosurgery at Baylor, was published June 2026 in Nature (Katlowitz et al., “Plasticity and language in the anaesthetised human hippocampus,” Nature, 654: 714, DOI: 10.1038/s41586-026-10448-0).
A rare, direct recording opportunity
This kind of single-neuron recording is only possible in a narrow clinical circumstance: some patients with drug-resistant epilepsy have electrodes surgically implanted to map seizure activity before a separate surgical procedure, and that implantation creates a brief, medically necessary window where researchers can record neural activity with a precision no non-invasive method (EEG, fMRI) approaches. The Baylor team used Neuropixels probes — deployed in the human hippocampus in this context for the first time — while patients were under general anesthesia for surgery, playing them repeating tones and short spoken stories and recording how individual hippocampal neurons responded.
The hippocampal neurons detected unusual or novel tones and improved at recognizing patterns over time, evidence that learning and neural plasticity continued despite the patients being fully unconscious. More strikingly for language specifically, the recorded neural patterns could distinguish grammatical categories — nouns, verbs, and adjectives — within the spoken stories, and neural signals showed measurable anticipation, predicting upcoming words before they were spoken. That is a marker of active, structured language processing, not passive sound registration, occurring in patients with no behavioral signs of consciousness.
The consent problem this research design has to solve before any data is collected
This study could only happen with a specific, front-loaded consent process, and that is worth naming directly rather than treating as a procedural footnote: research on data collected while a patient is under general anesthesia is, definitionally, research where the participant cannot consent in the moment the data is actually gathered. All consent — to have research electrodes placed, to have neural activity recorded during the already-scheduled surgery, and to have that recording used for this specific research purpose — has to be obtained from the patient beforehand, while they are still able to understand and agree to what will happen while they are unconscious. That is a materially different ethics posture than most human-subjects research, where a participant can, at least in principle, withdraw or object in real time as data collection happens. Combining a pre-existing clinical procedure (epilepsy electrode implantation) with an added research purpose also raises a related question institutional review boards have to resolve explicitly: separating what is being recorded for the patient’s own clinical care from what is being recorded and used purely for research, and making sure a patient’s consent to one is never treated as implied consent to the other.
What this does and doesn’t establish about consciousness
The finding is genuinely surprising against a common assumption that general anesthesia largely shuts down higher-order cognitive processing, including complex language comprehension. What it shows is that at least some structured, grammatically sensitive language processing continues at the neural level during anesthesia, in a hippocampal circuit associated with memory and learning. It does not show that anesthetized patients are consciously aware of, remember, or can act on that processing — the patients showed no behavioral signs of consciousness, and there’s no evidence here that this neural activity is accessible to conscious recall afterward. The result reopens questions about where exactly the line falls between “the brain is processing information” and “the person is having a conscious experience,” rather than closing them.
Frequently asked questions
Does this mean patients under general anesthesia can hear and remember what’s said around them?
Not established by this study. The recordings show neural evidence of language processing at the level of individual hippocampal neurons, but there is no evidence patients were consciously aware of this processing or could later recall it. The study measured neural activity, not conscious perception or memory.
How did researchers get access to record from individual neurons in a living human brain?
Through a rare clinical opportunity: some epilepsy patients have electrodes surgically implanted to map seizure activity ahead of a separate procedure, creating a brief window where researchers can record single-neuron activity with a precision non-invasive methods can’t match — here, using Neuropixels probes in the hippocampus for the first time in this context.
Why does this study’s consent process differ from typical human-subjects research?
Because the data collection happens while the patient is unconscious under general anesthesia, all consent — to electrode placement, to recording during the already-scheduled surgery, and to research use of that recording — must be obtained beforehand, while the patient can still understand and agree. Unlike most studies, participants cannot object or withdraw in real time as the data is actually gathered.
Sources
Primary source: Katlowitz, K. et al., “Plasticity and language in the anaesthetised human hippocampus,” Nature, 654: 714, June 2026. DOI: 10.1038/s41586-026-10448-0. Secondary coverage: ScienceDaily, June 28, 2026. This article was last checked against the cited sources on August 7, 2026.
