Bootstrapped Ubini Builds Machine-Readable Corpus for Nigeria's Edo Language
Stephanie Ohumu's bootstrapped Ubini has transcribed 20+ hours of Edo oral history and targets a 250-hour corpus by 2029, with licensing and media IP as revenue lines.
By Amara Osei
4 min read

What's News
- More than 98% of African languages lack support in language models, per a 2026 study of 41 supported languages and 23 public datasets.
- Ubini, founded in 2024 by Stephanie Ohumu, has interviewed about 20 people and transcribed more than 20 hours of Edo recordings.
- The company targets 50 hours of audio by December 2026 and a 250-hour corpus by 2029.
- Ohumu is seeking about $200,000 in non-dilutive or patient capital after funding the startup herself.
- Ubini's documentary 'Mia Mia' won Best Documentary at the RevolutionMe Festival in Brooklyn.
More than 98% of African languages have no support in today's large and small language models, according to a 2026 study by Kedir Hussen and six other researchers published in the Proceedings of Machine Learning Research. Stephanie Ohumu, a journalist and AI expert, founded Ubini in 2024 to fix that gap for one language: Edo, spoken by an estimated 2 million people across seven local government areas in Edo State, southwestern Nigeria.
The company has interviewed about 20 people and transcribed more than 20 hours of recordings. It targets 50 hours of audio by December 2026, 150 hours transcribed and translated by the end of 2027 with 30 transcribers, and a 250-hour corpus by 2029. Ohumu estimates the project needs at least 250 contributors, averaging two hours each.
The origin is personal. Ohumu's grandmother died in 2023, taking with her recipes, translations and unwritten cultural knowledge. "It used to be that if I encountered Edo words in the wild, I'd text my grandma or call her and she would translate for me," Ohumu said. "Not just that, there are meals we cannot make the way we grew up eating them because they were never written down."
"Language is a repository of knowledge. It took my grandmother dying for me to really understand that," she said.
Why can't AI models handle Edo?
Africa has more than 2,000 languages, but few have the digital resources needed for natural language processing. The Hussen study identified 41 African languages supported by language models and 23 publicly available datasets, leaving the overwhelming majority unsupported.
Ohumu tested existing models and found the results dangerous precisely because they sound plausible. "Claude can write a convincing enough cosplay of Edo, but it fills it with a lot of nonsense," she said. "It's really dangerous because if you don't know the language, you wouldn't know that."
In one case, a model translated "download" as "D-A-N-L-O-D", reproducing an English pronunciation instead of an Edo equivalent. Ohumu blames the thin online record. The largest body of Edo material she found comes from Jehovah's Witness translations of the Bible and devotional books, and some existing texts are heavily Anglicised. "It's like seeing Pidgin English being passed off as English," she said.
Edo is tonal, so tone markings change meaning, and models trained on incomplete data risk reproducing errors in translations and generated speech. A verified dataset, Ohumu argues, could power practical tools — she cited financial services that would let market women run banking transactions through audio commands in Edo.
How does Ubini build the archive?
The bootstrapped company focuses on people over 70 who speak Edo as their first language, capturing everyday conversation, ceremonial speech and marginalised knowledge. Its current food collection concentrates on women, who Ohumu said have historically done most of the cooking. Future collections will cover bead-making, traditional guilds and wood carving.
Ubini works in quarterly research sprints through existing community networks around Benin City, where Ohumu grew up. Consent is central. "Sometimes people say no, and you have to leave it at that," she said. Researchers log contributors' ages, occupations and backgrounds.
Transcription is the bottleneck. Three transcribers work with lead researcher Elliot, and every draft gets quality-checked by two University of Benin professors. Automated tools like Whisper and Granola have fallen short. "The transcription is a lot harder than I thought it would be," Ohumu said, noting that even fluent speakers often can't write the language. A state government initiative to train young writers was discontinued, so she plans to train transcribers herself.
Finished material goes into XML files containing speaker turns, Edo transcriptions, English translations and contributor context.
What's the business model?
Ubini has already produced "Mia Mia", a documentary featuring 85-year-old Maria Evbagharu Osazee. The film premiered at the Uncolonial Film Festival in Germany in August 2026, was selected for the Human Screen Festival in Tunisia in November, won Best Documentary at the RevolutionMe Festival in Brooklyn, and screens at the Kano Indigenous Languages of Africa Film Market and Festival later in November.
Ohumu sees licensing Edo data as the fastest revenue line. "I do think that licensing is the lower-hanging fruit," she said. "But the bulk of the sustainable funding will come from media products, whether that be selling a strong enough story to Netflix or Prime or EbonyLife, or developing it ourselves."
Self-funded since 2024, Ubini is hitting a ceiling. "The really expensive layer has been transcription and that QA process," Ohumu said. She is now seeking roughly $200,000 in non-dilutive or patient capital. "There's no other way this works," she said. "Even if this company doesn't make one naira, I'm still going to do this anyway."
Over the next year, she plans to expand the team, improve transcription quality and collect an additional 100 hours of audio toward the 2027 and 2029 targets.
Source: techcabal.com
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,169 articles