Single Neurons Reveal How Bilingual Brains Share a Universal Meaning Map
New research shows that similar concepts across languages occupy the same brain region, enabling bilinguals to switch languages effortlessly.
Related ideas across languages appear to occupy the same neural region.
My eighty‑year‑old father‑in‑law, a lifelong football enthusiast, flips effortlessly between English, Spanish and French while cheering for his teams. Watching him switch languages mid‑sentence highlights how the brain can juggle multiple tongues with apparent ease.
Researchers have long probed the neural underpinnings of multilingualism, noting that even dormant languages can be reactivated with minimal exposure, bypassing the need for explicit relearning. Small‑scale investigations have linked bilingualism to slower cognitive aging, reduced dementia incidence and modest improvements in executive control.
Most of these conclusions stem from functional imaging that captures broad patterns of activity but lacks cellular resolution. A new collaborative effort from Baylor College of Medicine and partner institutions has now tracked the firing of individual neurons in four English‑Spanish bilingual volunteers who were already implanted with hippocampal electrodes for epilepsy monitoring.
“This is the first investigation to examine bilingual processing at the single‑neuron level in real time,” said study co‑author Xinyuan Yan in a press release. The team recorded over a hundred neurons per participant while they listened to spoken material, read paired phrases and engaged in spontaneous conversation.
The data reveal a two‑tiered organization. Certain cells displayed a strong bias for one language when presented with equivalent words, yet larger neuronal ensembles formed language‑neutral clusters. These clusters arranged semantically related concepts—such as “dog” and “wolf”—closer together than unrelated items like “fork,” effectively constructing a shared conceptual map.
Remarkably, the English‑derived map could predict the positioning of Spanish equivalents, indicating that both languages draw on a common semantic scaffold. “It’s akin to looking at the same room from different windows; the interior remains unchanged while the viewpoint shifts,” explained co‑author Sameer Sheth.
Connecting Linguistic Realms
Language underpins social interaction, and speakers routinely convey identical ideas across different tongues despite occasional gaps in direct translation. Children raised in multilingual homes often blend lexical items, demonstrating the brain’s capacity to maintain distinct grammatical frameworks while merging meaning.
Before articulation emerges, neural circuits translate thoughts into electrical signatures that become words and sentences. Given the structural diversity among languages—such as verb placement or word order—one might expect each language to possess a unique neural fingerprint.
Recent AI‑driven analyses of fMRI scans from monolingual speakers of 21 languages, however, suggest a common neural architecture for representing meaning. Even constructed languages like Klingon from Star Trek and Na’vi from Avatar appear to tap into the same underlying system (source).
Parallel findings from bilingual participants reinforce this view. An fMRI study showed that Chinese speakers learning English recruited the neural networks originally used for Chinese, while another investigation identified overlapping speech‑related activity that allowed word decoding across languages (source).
Standard imaging methods, however, cannot resolve the rapid, fine‑grained dynamics that occur when bilingual individuals shift between languages. Capturing activity at the single‑cell level is essential for unveiling the precise mechanisms.
Charting Neural Activity
The four volunteers, all fluent in English and Spanish since early childhood, participated in three experimental phases while their hippocampal electrodes recorded neuronal spikes. First, they listened to an hour of YouTube content and the audiobook Eat Pray Love (Come Reza Ama). Next, they read aloud nearly 100 bilingual phrase pairs displayed on a screen. Finally, they engaged in up to 90 minutes of free‑form conversation with native speakers, covering topics from family life to their epilepsy experiences.
Across these tasks, the researchers amassed thousands of spoken words, hundreds of matched phrases and extensive conversational data, providing a rich dataset for analysis.
The Semantic Terrain
Only a minority of recorded neurons behaved truly bilingually, responding similarly to direct translations such as “friends” and “amigos.” To interpret the broader pattern, the team compared the neural recordings with representations from mBERT, Google’s multilingual transformer model that captures word relationships through contextual embeddings.
Both biological and artificial systems demonstrated that meaning emerges at the population level rather than within isolated cells. Neuronal ensembles, like the embeddings in mBERT, organized words into a semantic geometry where conceptually linked items clustered together regardless of language.
This geometry remained stable across English and Spanish, enabling the researchers to predict Spanish clusters based solely on the English map. “The brain encodes word meaning through coordinated activity of neuronal groups, not by translating individual words one‑by‑one,” said Yan.
The study focused on semantics, leaving syntactic processing—governed by grammatical rules—to be explored in future work. A recent single‑cell investigation suggests that frontal brain regions may handle grammar independently of meaning (source). Whether these areas also share a language‑independent map remains an open question.
Looking ahead, the team plans to monitor how these semantic maps evolve as participants acquire a new language, potentially shedding light on the neural integration of novel concepts. Such insights could inform the development of more efficient language models in artificial intelligence (source).
“Our findings indicate that the brain is inherently equipped to support multiple languages,” concluded Benjamin Hayden, another study author.
This article has been fact checked for accuracy, with information verified against reputable sources. Learn more about us and our editorial process.
Last reviewed on .
Article history
- Latest version
Reference(s)
- “Checking your browser - reCAPTCHA.” <https://pubmed.ncbi.nlm.nih.gov/28137833/>.
- <https://www.eurekalert.org/news-releases/1132786>.
- Malik-Moraleda, Saima., et al. “Constructed languages are processed by the same brain mechanisms as natural languages.” Proceedings of the National Academy of Sciences, vol. 122, no. 12, March 17, 2025 National Academy of Sciences, doi: 10.1073/pnas.2313473122. <https://www.pnas.org/doi/10.1073/pnas.2313473122>.
- Silva, Alexander. “A bilingual speech neuroprosthesis driven by cortical articulatory representations shared between languages - Nature Biomedical Engineering.”, vol. 8, no. 8, pp. 977-991. Nature, doi: 10.1038/s41551-024-01207-5. <https://www.nature.com/articles/s41551-024-01207-5>.
- Devlin, Jacob. “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.” arXiv.org <https://arxiv.org/abs/1810.04805>.
- Cai, Jing. “Mapping the neuronal building blocks of human language with language models - Nature.”, June 17, 2026, pp. 1-9. Nature, doi: 10.1038/s41586-026-10691-5. <https://www.nature.com/articles/s41586-026-10691-5>.
Cite this page:
- Posted by David Anderson