Harvard researchers discover specialized brain cells controlling speech

Researchers at Massachusetts General Hospital and Harvard Medical School have discovered that the human brain uses highly specialized individual neurons to produce speech, with each performing only one narrow grammatical function. The study recorded over 10,000 words of natural speech using microelectrodes implanted in the brains of epilepsy patients. The results were published in Nature and could help develop more precise brain-computer interfaces.

Researchers at Massachusetts General Hospital and Harvard Medical School have shown that the human brain does not produce speech in a dispersed manner across an entire region, but instead uses individual highly specialized neurons, each with its own narrow role to play. The study results were published in the prestigious scientific journal Nature.

How the study was conducted

Scientists used a medical procedure in which microscopic electrodes were implanted in the brains of eight epilepsy patients to precisely map neural activity. This gave researchers a rare opportunity to listen directly to the activity of neurons in the frontal and temporal lobes. Patients were asked to speak freely during the observation period and answer everyday questions to avoid repeating memorized texts. In this way, over 10,000 words and thousands of sentences of natural speech were recorded in real time.

The collected data revealed an extremely clear division of labor, concentrated mainly in the left hemisphere of the brain. About one-tenth of the neurons monitored activated specifically in response to parts of speech, such as verbs or adverbs, about one hundred milliseconds before the word was spoken. Other cells tracked whether a word was the subject or object of a sentence, while separate neurons responded only to the end of a clause.

Machines and biology converge

To analyze the deeper structure of language, the researchers enlisted the help of large language models and discovered surprising similarities between artificial intelligence and living cells. Both algorithms and nerve cells must keep track of the broader context of a conversation, with grammar specialists mostly remembering references to previously spoken words to predict the form and meaning of the next word.

It is also interesting that although a single neuron deals narrowly with only one language element, the tissue surrounding it often analyzes quite different aspects of language, such as sentence rhythm. Angela Friederici, a neuropsychologist at Germany's Max Planck Institute, noted that the question remains initially unanswered: how do these strictly separate language building blocks combine to produce flowing and comprehensible speech in a fraction of a second.

What's next?

Neurosurgeon Ziv Williams at Massachusetts General Hospital was optimistic with his colleagues that such detailed brain maps would eventually make it possible to read the complex thoughts of people directly from neurons. "With improved interfaces, doctors could restore communication abilities to patients who have completely lost the ability to speak due to serious illness or physical trauma," the researchers wrote.

The scientists themselves caution against far-reaching conclusions, as the observation period was relatively short. It is not known whether individual cells perform the same tasks throughout a person's lifetime. Previous animal studies suggest that the cerebral cortex undergoes continuous reassignment of specialization among cells over months or years.