Giving Sudan’s Languages a Voice in the Age of AI
A community-led research initiative building speech and language data for Sudan’s 450+ tongues — so the next era of AI is built with them, not without them.
The landscape
A global imbalance
AI speaks the world’s well-resourced languages fluently — and barely hears the rest. Africa is the sharpest edge of that divide.
<1% of AI training data meaningfully represents African languages
African Languages Lab, 2025
Illustrative of the high- vs low-resource language divide in AI — not a precise per-country score. Africa is overwhelmingly low-resource: 88% of its languages are underrepresented or ignored (African Languages Lab, 2025).
Hover a country to see its standing — tap on mobile. Then focus on Africa.
This is happening now
The people behind the gap
These aren’t hypotheticals. They are the everyday reality of millions who speak languages AI was never trained to understand.
Now
The graduate who can’t get in.
Ahmed graduated in computer science and writes clean Python. The remote-work platforms that pay people to train and evaluate AI are hiring — but every task assumes you work in English or another well-resourced language. His Sudanese Arabic isn’t an option on any of them. So a capable engineer sits outside a market that was, in theory, built for exactly his skills.
Now
The graduate who can’t get in.
Ahmed graduated in computer science and writes clean Python. The remote-work platforms that pay people to train and evaluate AI are hiring — but every task assumes you work in English or another well-resourced language. His Sudanese Arabic isn’t an option on any of them. So a capable engineer sits outside a market that was, in theory, built for exactly his skills.
Today
Learning blocked by language.
A student wants to use AI the way the world now does — as a patient tutor that explains, quizzes, and corrects. She even wants it to help her improve her English. But when she asks in her own language, the model misunderstands her, answers in ways that don’t fit her context, or simply gets it wrong. The tool that is teaching millions can’t quite teach her, because it was never taught her language first.
Today
Learning blocked by language.
A student wants to use AI the way the world now does — as a patient tutor that explains, quizzes, and corrects. She even wants it to help her improve her English. But when she asks in her own language, the model misunderstands her, answers in ways that don’t fit her context, or simply gets it wrong. The tool that is teaching millions can’t quite teach her, because it was never taught her language first.
Across the world
Fluent at home, invisible abroad.
Across the world, Sudanese and other African families have been displaced. Many speak their mother tongue with full fluency — the kind of native command these platforms desperately need. Yet the work isn’t open to them: their language isn’t listed, and their identity papers don’t fit the form. Must a person wait years for a new citizenship just to be allowed to contribute the one thing they already master — their own language? And even then, will the opportunity be there?
Across the world
Fluent at home, invisible abroad.
Across the world, Sudanese and other African families have been displaced. Many speak their mother tongue with full fluency — the kind of native command these platforms desperately need. Yet the work isn’t open to them: their language isn’t listed, and their identity papers don’t fit the form. Must a person wait years for a new citizenship just to be allowed to contribute the one thing they already master — their own language? And even then, will the opportunity be there?
Right now
The new doors that stay shut.
Today you can build an app without writing code, or design a poster just by describing it — if you can talk to the tools in a language they understand. For someone whose language the models barely grasp, even these “no-code,” “just-describe-it” doors stay shut. The barrier moved from writing code to speaking the right language — and a whole generation is still locked out.
Right now
The new doors that stay shut.
Today you can build an app without writing code, or design a poster just by describing it — if you can talk to the tools in a language they understand. For someone whose language the models barely grasp, even these “no-code,” “just-describe-it” doors stay shut. The barrier moved from writing code to speaking the right language — and a whole generation is still locked out.
The pattern is always the same: when the technology doesn’t speak your language, every door behind it — work, learning, creation — narrows. Birtal Lab exists to open those doors, one language at a time.
The evidence
Grounded in the record
This is not advocacy in a vacuum. Policy institutions and peer-reviewed research independently document the same gap — and point to the same remedy: language data built with the communities who speak these languages.
of African languages are severely underrepresented or ignored by modern AI
Issaka et al., The African Languages Lab (2025)
of AI training data meaningfully represents African languages
The African Languages Lab (2025)
Policy & Industry
Microsoft AI Economy Institute
“Global AI Diffusion: Q1 2026 Trends and Insights” · May 2026
Sudan at 7.6% AI diffusion — among the world’s lowest.
Full citationUNESCO
“The cultural cost of AI in Africa’s education systems” · Oct 2025
Warns of a “digital-territorial coloniality” in education.
Read reportWorld Bank
Education Working Paper No. 21 · Nov 2025
~50% of African graduates lack the technical skills employers need.
Full citationAfrican Union
“AI for Sustainable Youth Development in Africa” · 2024
A continental call to invest in AI skills and inclusion.
Full citationPeer-Reviewed Research
Issaka et al.
“The African Languages Lab” · 2025
88% of African languages are severely underrepresented or ignored — a 20–70× research gap.
Read paperImam et al.
“ASR for African Low-Resource Languages: Challenges & Future Directions” · 2025
Calls for community-driven data collection as the path forward.
Read paperDiarra et al.
“Dealing with the Hard Facts of Low-Resource African NLP” · Nov 2025
612 hours of Bambara speech; the decisive importance of human evaluation.
Read paperSurvey
“Charting the Landscape of African NLP” · 2025
A field-wide map of progress, methods, and the gaps that remain.
Read paperLinks open the primary source in a new tab. Entries without a link are cited in full for verification; we add direct links as canonical URLs are confirmed.
Our research
How we work
In short: we record everyday speech with native speakers, and other speakers check every recording before it is kept. The four research programmes behind that are set out in full on the research page.
Tone matters
In these languages a change in pitch can change the meaning, so reviewers listen for tone, not only for words.
By voice, even offline
People take part by speaking. You don't need to read well or have a strong connection.
Checked by speakers
A recording is kept only when several native speakers agree on it.
Where we are today
This is a pilot, in its first months
We are collecting the first recordings in Sudanese Arabic, Zaghawa and Fur — small on purpose, so the method is right before the corpus is large.
The cost of exclusion
How a language gap becomes an opportunity gap
The harm is not abstract. It compounds, step by step.
Unsupported language
Less digital content
Fewer digital skills
Fewer opportunities
No linguistic data to build on
Weak educational content
Fewer digital job opportunities
Little local research
Weak cultural representation
The AI gap widens each year
What if
What if AI spoke Sudan’s languages?
The same technology that excludes can include. The upside is just as concrete as the cost.
Local digital assistants that understand you
Better, language-fair education
Easier access to government services
Local agriculture & health support
Heritage and oral history preserved
More local digital content
Greater community participation in AI
Opportunities
The work is already paying — elsewhere
A global market already compensates people for AI language work. African contributors are too often left out, because their languages aren’t represented. Birtal Lab exists to bring these opportunities to Sudanese and African language communities.
We don’t just point to the door — we build the bridge: the data, the methods, and the community that make Sudanese and African languages fundable, employable, and heard.