EducationApril 25, 2026

Pronunciation Is a Body Practice — and AI Can Finally Teach It

In yoga we say alignment is felt before it is seen. The same is true of language: the difference between a nasal vowel and a flat one lives in the body, not the grammar book. Here is how we built an AI tutor that understands this.

By Andrea Borghi
Pronunciation Is a Body Practice — and AI Can Finally Teach It

Pronunciation Is a Body Practice — and AI Can Finally Teach It

In yoga, we spend years learning to feel the difference between a hip that is open and one that merely appears to be. The subtle proprioceptive signal — the internal sense of where the body actually is — is not something a diagram teaches. It is something a good teacher returns to, patiently, until the student's nervous system makes it automatic.

Language pronunciation works the same way.

The difference between the French nasal vowel /ɑ̃/ and a flat English approximation is not a matter of knowing that the sound exists. Every student of French knows it exists. The difficulty is feeling where the soft palate rises, noticing when the tongue is too forward, catching the moment the sound collapses. It is a body practice. And for thirty students in a fifty-minute class, it is a body practice their teacher has no time to address individually.

The gap no tool had filled

When we started looking at what AI could offer language learners, the tools we found fell into two categories: gamified apps that treated pronunciation as a score and moved on, or institutional software that required weeks of IT setup and priced out the individual teacher entirely.

Neither category understood that pronunciation is iterative. That a student needs to attempt the same sound eight times in a week — not once, not in a test — and receive honest feedback each time. That the feedback needs to be phoneme-level, not word-level. That "your R is slightly uvular" is more useful than "79%."

What we built

Lingua·Lab is an AI language tutor we built from this frustration. It runs on AWS, uses a faster-whisper STT model for accurate speech capture, and sends each attempt through a GPT-5.5 pronunciation coach that knows the specific phonological priorities of Tuscan Italian, Parisian French, and Castilian Spanish.

The tutor does not tell students they are "close." It tells them their nasal vowel collapsed because they released the velum too early. It gives them the IPA of what they said versus what they meant. It plays back the correct version in the right accent so they can compare by ear.

And it does this for every student, every session, whether there are three or thirty of them.

For teachers

The teacher dashboard shows, each Monday morning, who practised and for how long. Which students are improving on liaisons. Which are still flattening the French R. Three minutes to know what would otherwise take three lessons to discover.

The weekly email report is the feature teachers mention first. Not the phoneme scoring. The time it saves.

The yoga parallel

In class we often say: the pose is not the goal, the awareness is the goal. The same is true here. Lingua·Lab is not trying to replace the language teacher — the cultural context, the improvisation, the human relationship. It is trying to give every student the ten minutes of individual attention that no classroom hour can contain.

Accurate pronunciation, like correct alignment, is felt before it is seen. We built a tool that helps students feel it.


Lingua·Lab supports Italian, French, and Spanish. 14-day free trial, no card required: lingua-lab.org

#language learning#AI#pronunciation#education technology#lingua-lab