Summary: For people who speak two languages, switching between them often feels effortless. A new study now explains how the brain achieves that seamless transition at the level of single neurons. The research finds that bilingual brains encode meaning in a shared semantic map that spans languages, while individual hippocampal neurons remain largely language-specific and operate together to represent equivalent concepts across tongues.
Published June 24 in the journal Cell, this study reveals that the hippocampus organizes words by meaning into a unified, multilingual map. Individual neurons are mostly tuned to one language, but groups of these cells change their activity patterns to align representations of equivalent concepts — enabling fluent switching without confusion.
Key Facts
- Shared semantic mapping: The brain places words into a meaning-based map where conceptually related words (for example, “dog” and “wolf”) sit close together regardless of language.
- Language-specific neurons: At the single-neuron level in the hippocampus, many cells respond preferentially to one language and do not automatically fire for the same concept in another language (for instance, “dog” versus “perro”).
- Predictive cross-language alignment: By examining the neighborhood around an English word in the neural map, researchers could predict the corresponding location of the Spanish translation with high accuracy.
- Resemblance to multilingual AI models: The hippocampal organization parallels large multilingual language models such as mBERT, which also map many languages into a shared conceptual space.
- Built-in multilingual capacity: The findings suggest the brain is structurally suited to learn multiple languages: once a semantic framework of relationships is formed, it can be applied to new vocabularies.
Source: Cell Press
Background: Many bilingual people report that switching between languages feels natural. To investigate how the brain accomplishes this, researchers recorded activity from individual neurons in bilingual participants and analyzed how words and meanings are organized across languages.
This experiment involved four English–Spanish bilingual adults who had been fluent in both languages from an early age and used them interchangeably. Because these participants were undergoing clinical monitoring for epilepsy, they had high-precision electrodes implanted in the hippocampus. This rare clinical setting allowed researchers to record single-neuron activity while participants listened, read, and spoke in both English and Spanish.
Surprisingly, only a small number of neurons responded equally to the same concept across languages. For most hippocampal neurons, responses were language-specific. Rather than relying on individual “translation” neurons, the brain represents meaning through a distributed population code: groups of neurons form a semantic map and adjust their combined activity to represent equivalent words in different languages.
Within this semantic map, related concepts cluster together: for example, “dog” and “wolf” occupy neighboring positions, while an unrelated word like “fork” sits farther away. This geometric organization remained consistent across languages, meaning the relative layout of concepts is preserved even when read out through language-specific neural activity.
To test the depth of this shared mapping, the team used the English map to predict where a Spanish word would appear in the neural representation. By analyzing the arrangement of neighboring concepts around an English term, they correctly inferred the corresponding Spanish location on the hippocampal map. As one investigator described it, the two language views are like looking at the same room from different windows: the perspective changes, but the room’s layout remains identical.
The researchers also compared these neural maps with mBERT, a large multilingual language model trained on more than 100 languages. They found that mBERT produces a similar semantic mapping across languages, reinforcing the idea that both biological and artificial systems can map multiple languages onto a shared conceptual geometry.
Lead and senior authors emphasize that these results point to an inherent neural architecture for multilingual learning: once the brain establishes a map of concept relationships, it can overlay new languages onto that scaffold without rebuilding the network from scratch. In other words, humans have the structural capacity 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: The team recorded single-neuron activity from four early bilingual participants who had clinical electrode implants for epilepsy monitoring. This clinical configuration permitted high-resolution tracking of hippocampal neurons while participants listened, read, and spoke in English and Spanish.
A: The brain uses distributed population coding. Words are placed on a meaning-based semantic map. Although individual neurons often prefer one language, the population as a whole shifts its activity pattern so that equivalent words across languages occupy matching positions in the shared conceptual geometry.
A: The study suggests the human brain is predisposed to learn multiple languages. Once it forms a stable map of relationships among concepts, it can project that structure onto new vocabularies, making additional languages easier to acquire without reconstructing the underlying semantic geometry.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full.
- Additional context was provided by editorial staff.
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., Cell. DOI: 10.1016/j.cell.2026.05.020
Abstract
Shared neural geometries for bilingual semantic representations in human hippocampal neurons
Humans can comprehend and express similar concepts across different languages. To investigate how the brain supports this ability, researchers recorded hippocampal neuron responses during passive listening, directed speaking, and spontaneous conversation in English and Spanish among balanced bilingual participants. 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 showed language-dependent tuning, indicating language-specific implementations at the single-cell level. Crucially, translation depended on a preserved geometric organization of neural responses across languages rather than strict neuron-by-neuron overlap. This shared geometry was realized by the same neural population read out along distinct axes, a mechanism that may reduce interference between languages.
Overall, the results indicate that the hippocampus encodes a language-independent internal model of meaning that supports bilingual communication.