A shared phonology under every language
In the 19 November issue of Nature, a study of intracranial recordings from the human temporal lobe reports that speech perception is not lost in translation: the brain’s phonological machinery combines a shared, language-general core with language-specific tuning, so that hearing a foreign language still engages the same basic sound-processing circuitry.
There is a question at the bottom of every encounter with a foreign language: when the sounds are unfamiliar, is the machinery that hears them also unfamiliar? The brain does not have separate hardware for every language, and yet speakers of different languages clearly hear different things in the same acoustic stream. A paper in the 19 November issue of Nature addresses the question directly with recordings from inside the human brain, and the answer is a middle path with a clear shape: the core phonological machinery is shared, and the tuning around it is language-specific. Speech perception is not lost in translation, but it is also not indifferent to which language is being spoken.
The study, led by researchers including those at Columbia, used intracranial recordings from the temporal lobe of patients undergoing surgery for epilepsy, a method that gives millisecond-level access to the brain’s response to sound. The patients were native speakers of different languages, and the team presented speech sounds in a controlled way to compare how the same acoustic signals were processed across languages. The design is the strength of the work: rather than comparing brain responses to different languages in different people, which confuses the language with the speaker, the study looked for the same sounds processed through different native phonologies.
Two distinct patterns emerged from the neural activity. The first was shared and language-general: a core set of temporal-lobe responses that treated the sounds the same way regardless of the listener’s language, tracking the acoustic structure of the signal itself. The second was language-specific: regions and response patterns that were organised around the listener’s native phonology, the particular set of sound distinctions that their language requires them to make. A sound that is a meaningful contrast in one language and irrelevant in another leaves a different trace, and the trace is measurable in the same brain region that carries the shared response.
The finding matters for a practical reason beyond the neuroscience. It gives a concrete neural picture of why second-language listening is hard and why it improves: the shared core means that any listener’s brain is already built to process speech-like sounds, so the difficulty of a foreign language is not that the machinery is missing but that the native tuning must be supplemented, a new set of distinctions layered onto the common substrate. It also complicates the simple idea, popular in both linguistics and AI, that perception is just a filter set by the first language. The filter is real, but it is an adjustment on shared circuitry, and the circuitry is more universal than the adjustments.
The study is one more step in a long collaboration between neuroscience and phonology, and its result is satisfyingly balanced: the brain neither stores a separate processor for each language nor ignores language altogether. It runs one phonological machine, tuned by experience. Translation is not lost, because the machine is shared; but the tuning is personal, because every language asks its speakers to hear the world in its own way, and the hearing, once trained, is never quite retrained.