Prince Mario-Max Schaumburg-Lippe: ZWAG AI Puts Every Language First in Dubai

Ask a large language model a question in Thai, Tamil, Arabic or Swahili and the answer often looks smooth at first glance. Read it again, slowly, the way a native speaker would. A verb form slips. A formal register collapses into something too casual. A sentence that should bend one way bends the other. That gap, between sounding fluent and actually being right, is exactly what ZWAG AI set out to close this week in Dubai.

On October 26, 2026, at the Dubai AI Festival, the Dubai-based company announced the launch of the Center of Excellence for Linguistic Sovereignty, known as CoE-LS. The idea is plain and, frankly, overdue. The digital future, the company said, belongs to every language and not just a few. Governments, universities and incubators are being invited to help map missing linguistic structures, build full language profiles and give communities a stronger say in how their languages are handled by AI systems.

A correction layer, not another model

The workhorse here is Matr, spelled Mātṛ in the company’s materials. Matr is not a new large language model competing for a leaderboard spot. It is a model-agnostic correction layer that sits alongside an existing model and checks its output as it is produced. Instead of retraining a model at great cost, Matr treats each language as a structured system. It evaluates what the model has written against native grammatical and typological rules, then selects or corrects the output in real time.

That distinction matters for cost and for speed. Retraining is slow, expensive work. A correction layer can be added to systems already in use, which is why ZWAG AI describes Matr as trust and correctness infrastructure. Readers who follow enterprise adoption will recognise the same practical logic at work in DXC putting governed AI agents to work for banks and insurers, where control and checking matter as much as raw capability.

2,400 languages in public beta

The headline number is hard to miss. Matr’s Translation Studio, now in public beta, currently supports more than 2,400 languages. Most translation tools compete on dozens of languages, sometimes a hundred. A count above 2,400 changes the conversation. It points toward communities whose languages rarely get first-class treatment in software at all.

Co-founder and chief executive Priya M. Nair put the point in human terms. “We shouldn’t have to water down our native languages to fit a machine’s logic,” she said. “Our mission is teaching AI to respect our heritage, our values, and our mother tongues; and it starts with getting the language right.”

Nipuna Abeykoon, co-founder and chief innovation officer, was just as direct. “AI does not fail people because it lacks intelligence. It fails them because it misreads the structure of their language,” he said. “Correctness, not fluency, is what earns trust.” Abeykoon leads the research behind Matr and is lead author of the company’s work on typologically informed correction. He speaks English, Sinhala and Hindi, and his background as a trained classical musician is, the company notes, part of how he hears structure in language.

What the new center will actually do

CoE-LS is designed as an institutional framework, not a ribbon-cutting photo. It will partner with universities, cultural organisations and regional governments to host Linguistics Excellence Centres. Academic institutions, professional translators, linguists and native speakers are invited to contribute through a global contributor program, building and refining native language profiles on the Matr platform.

That contributor model is the smart part. Grammar rules can be documented, but register, courtesy and context live with speakers. A profile built only in a lab will miss things. A profile built with teachers, translators and elders stands a better chance of holding up in daily use, in a classroom, a clinic or a government service portal.

A company rooted in Dubai, backed by research

ZWAG AI Ltd is registered in DIFC and operates out of DIFC Innovation One at the Dubai AI Campus. It is a subsidiary of Delaware-incorporated ZWAG INC. Its work is backed, the release notes, by empirical research co-authored with researchers from IIIT Hyderabad, cited as arXiv:2602.01162. Nair, for her part, brings a long record in responsible AI advocacy. She is a Global Ambassador for Responsible AI, author of the 2021 book Formula-G: Scale Your Startup to a Growth Machine, and a recent recipient of the Entrepreneurial Leadership Award at the Women of Substance 2026 ceremony at BITS Pilani Campus in Dubai.

The broader field is moving in a similar direction, toward smaller, checked and combined systems rather than one giant answer for everything. That is the same instinct behind five small Qwen models teaming up to match a frontier giant and behind voice systems that learn more carefully, as in Block’s Dessa learning to talk from voices alone. Accuracy, in each case, is a design choice, not an afterthought.

Why this matters beyond Dubai

Language is where AI meets daily life first. A parent helping with homework. A shop owner writing a product page. A student reading instructions in a second or third language at midnight. If the output quietly breaks the rules of their own language, trust drains away fast, and people stop using the tool for anything that counts.

Get it right, though, and the opposite happens. A correct sentence in a person’s mother tongue carries weight. It says the system was built with them in mind, not as an afterthought added in a later update. That is the bet ZWAG AI is making with CoE-LS, and it is a generous one. 2,400-plus languages is a long list. Working through it, profile by profile, speaker by speaker, will take patience. It is also the kind of patience that tends to pay off.

For governments and universities weighing whether to join, the offer is concrete: help map what is missing, help build the profiles, and help keep each language intact as AI spreads. The festival announcement is done. The contributor work starts now, and that is where this story will be won or lost, one language at a time.

Also published on: