Summary: For bilingual people, switching between languages often feels effortless. Researchers have now traced how the brain accomplishes this at the level of single neurons, revealing that bilingual brains encode meaning in a shared semantic map that spans multiple languages. Individual hippocampal neurons are largely language-specific, yet their coordinated activity creates a unified conceptual map that allows seamless transitions between tongues while keeping each language distinct.
A study published on June 24 in the journal Cell reports that the brain uses language-specific neurons to read from the same multilingual map of concepts. Although most individual neurons responded preferentially to words in a single language, groups of neurons shifted their activity patterns to align equivalent concepts across languages. This population-level coding organizes words by meaning so that conceptually related items—like “dog” and “wolf”—sit near each other on the internal map regardless of whether the word is English or Spanish.
Key Facts
- The shared semantic map: Words are arranged according to meaning within a universal internal map. Words with similar meanings are represented in nearby neural neighborhoods across languages.
- Language-specific neurons: At the single-neuron level in the hippocampus, most cells respond preferentially to one language and do not necessarily fire for the translation of the same concept in another language.
- Predictive mapping across languages: By examining how nearby concepts are arranged around a word in the English map, researchers could predict the corresponding location of its Spanish translation in the neural map.
- Similarity to multilingual AI models: The hippocampal organization mirrors large multilingual language models, which use shared conceptual spaces to map many languages.
- Built-in multilingual capacity: Findings suggest the brain is structurally well suited for learning multiple languages: once a core map of relationships among concepts exists, it can be applied to additional vocabularies.
Study overview
This investigation provides the first real-time, single-neuron view of bilingual language processing in the human hippocampus. The team worked with four early-acquisition English–Spanish bilingual participants who were equally comfortable in both languages. All participants were undergoing clinical monitoring for epilepsy and had intracranial electrodes implanted as part of their medical care, allowing researchers to record high-precision neuronal activity while participants listened, read, and conversed in English and Spanish.
Surprisingly, only a small fraction of hippocampal neurons responded to both a word and its translation (for example, “dog” and “perro”). Instead, most neurons showed language-specific tuning. Yet when considered as populations, these neurons formed consistent geometric relationships: the relative arrangement of concepts was preserved across languages even when individual cells did not directly encode both words. That preserved geometry — the shape and neighborhood structure of the map — allowed the researchers to predict where a translated word would appear on the map of the other language.
“This is the first study to observe bilingual neural encoding at the single-neuron level in real time,” says first author Xinyuan Yan of Baylor College of Medicine. Yan, who is bilingual, notes that the brain appears to maintain an internal model of word meanings that generalizes across languages: once relationships among concepts are established, they can be deployed in any language.
Senior author Sameer Sheth of Baylor College of Medicine summarizes the idea with a visual metaphor: “It’s like looking into a room from different windows. Everything inside is the same, but the perspective shifts.” By reading the shared semantic map from distinct neural readout angles, the brain can use multiple languages without confusion.
The team also compared their neuronal maps to a multilingual artificial intelligence model trained on many languages. The model showed a similar strategy of mapping conceptual relationships in a shared space, suggesting convergent organizational principles between biological and artificial multilingual systems.
Implications
These results indicate that the human hippocampus encodes a language-independent internal model for meaning. Rather than dedicating single “translation” neurons to pair words across languages, the brain relies on distributed population codes: groups of language-tuned neurons jointly form patterns that correspond to concepts. This architecture may help explain why bilinguals can switch languages fluidly while keeping languages separate enough to avoid interference.
Senior author Benjamin Hayden of Baylor College of Medicine notes that this structure makes multilingual learning efficient: once the brain maps relationships among concepts, it can apply that map to a new language without rebuilding the underlying structure. The study therefore supports the idea that human brains are naturally equipped to become bilingual or trilingual.
Funding:
This research was supported by the McNair Foundation, the National Institutes of Health, the SNS Allan Friedman RUNN Research Grant, the National Library of Medicine, the Gordon and Mary Cain Pediatric Neurology Research Foundation, and the National Research Foundation of South Korea.
Key Questions Answered:
A: Because the study involved patients receiving clinical monitoring for epilepsy, researchers were able to record electrical activity from implanted high-precision electrodes. That rare clinical setup made it possible to monitor single-neuron firing in the hippocampus while participants listened, read, and spoke in both English and Spanish.
A: The brain uses distributed population coding: individual neurons tend to be language-specific, but as a group they adjust their activity to produce similar patterns for equivalent words in different languages. That preserves the relationships among concepts across languages without requiring one-to-one neuron translation.
A: The findings suggest the brain is structurally predisposed to multilingualism. Once it forms a core map of conceptual relationships, that map can be projected onto new vocabularies, making the addition of another language efficient and less likely to interfere with existing language representations.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full.
- Additional contextual details were added by staff for clarity.
About this language and neuroscience research news
Author: Julia Grimmett
Source: Cell Press
Contact: Julia Grimmett – Cell Press
Image: The image is credited to Neuroscience News
Original Research: Open access. “Shared neural geometries for bilingual semantic representations in human hippocampal neurons” by Xinyuan Yan et al., published in Cell. DOI: 10.1016/j.cell.2026.05.020
Abstract
Shared neural geometries for bilingual semantic representations in human hippocampal neurons
The human brain can understand and express the same concepts in multiple languages. To investigate how it does this, researchers recorded hippocampal neuron responses during passive listening, directed speaking, and spontaneous conversation in both English and Spanish in a small group of balanced bilinguals. They identified a small number of putative “cross-language neurons” whose responses to equivalent words (for example, “tierra” and “earth”) were correlated. However, most neurons’ semantic tunings differed by language, indicating language-specific neural implementations. Crucially, a preserved geometric organization of neural responses linked the two languages: the same conceptual relationships were encoded across languages even without neuron-level overlap. This geometry was implemented by a common set of neurons along distinct readout axes, a difference in readout that may help prevent cross-language interference. Together, these results suggest the hippocampus encodes a language-independent internal model for meaning.