Home Originals Beyond NLP: Where Yoruba-Driven Tech Is Actually Being Built
OriginalsTech

Beyond NLP: Where Yoruba-Driven Tech Is Actually Being Built

Share
Share

In my article on this subject for Techies Node I looked at why Yoruba NLP research barely registers at Nigerian tech and creator events. But narrow the lens to NLP research alone and you miss most of the picture. Once you look past academic workshops and into fintech, healthtech, and edtech, native-language-driven tech is quietly further along than the conference circuit would suggest, it’s just scattered across backend features, accessibility add-ons, and university pilot projects that nobody has bothered to put in the same room.

Fintech: voice and text as the new access layer

Fintech is where native-language tech has the clearest commercial logic, because Nigeria’s financial-inclusion problem is fundamentally a language and literacy problem before it’s a smartphone problem.

EqualyzAI, a Nigerian AI voice company, has been building dialect-specific voice data collection networks and is now expanding into regional variants like Ijebu Yoruba, on the reasoning that people trust and use voice systems more when those systems sound like them, according to its CEO. Elsewhere in the open-source layer, Axiveri’s NaijaVox speech-recognition models now on a second generation, are explicitly positioned for fintech and banking use cases, transcribing Yoruba, Hausa, Igbo, and Pidgin with steadily improving accuracy, trained on a mix of academic and community-contributed datasets.

A smaller but sharper example is owo-parse, an open-source financial-intent parser that takes free-form instructions in English, Pidgin, Yoruba, Hausa, or Igbo, the kind of code-switched, informal phrasing people actually type or say, and converts it into structured data any payment backend can use. And an academic project out of a federal polytechnic, the Lédèe Yorùbá API, combined speech recognition, translation, and a financial glossary engine for USSD and mobile banking, reporting notably strong glossary accuracy in testing.

However, none of these are headline products. They’re infrastructure sitting underneath customer support lines, USSD menus, and chatbot backend, which is exactly why they don’t show up on a fintech demo-day stage, even though they’re doing more practical native-language work than most panels on “AI for financial inclusion.”

Healthtech: the translation problem nobody funds directly

Healthtech’s native-language work follows a similar pattern but for a starker reason: Nigeria’s own patients’ bill of rights states that people are entitled to health information in a language they understand, and researchers have been pointing out for over a decade that this often isn’t happening in practice. That gap produced things like RTCHAT, a rule-based English-to-Yoruba translation system for doctor-patient mobile chat, tested at a Nigerian polytechnic and found to outperform Google Translate on health-specific sentences by a wide margin, a reminder that general-purpose translation tools still struggle badly with domain-specific Yoruba, medical terminology included.

On the deployment side, Honey, a WhatsApp chatbot built by Data Science Nigeria with DKT International, offers family-planning guidance in English, Hausa, and Yoruba, meeting people in the language they’re already comfortable discussing sensitive topics in. A handful of telemedicine platforms now advertise Yoruba-language consultations as a baseline feature rather than a differentiator.

The most functionally advanced example, tellingly, isn’t Nigerian at all. Jacaranda Health, a Kenyan maternal-health organization, expanded its open-source language model to cover five African languages including Yoruba and folded it into an SMS-based digital health assistant already serving underserved communities. A foreign-built tool is currently doing more for Yoruba-language maternal health access than most Nigerian health-tech coverage acknowledges exists.

Edtech: mostly teaching the language, not teaching in it

Edtech is where the pattern breaks in an instructive way. Search for Yoruba-language edtech and what surfaces is almost entirely apps that teach Yoruba as a subject — Masoyinbo, Yoruba101, Akonilede, Lingawa — built primarily for children, learners, and a visibly large diaspora audience reconnecting with heritage. That’s real and valuable work, but it’s a different category from tech that delivers the existing Nigerian curriculum in Yoruba as the medium of instruction. The AI tutoring research aligned to WAEC, NECO, and JAMB curricula that surfaced in NLP workshop proceedings hasn’t yet produced a consumer product doing the same thing in-language. Native-language edtech, in other words, is currently stronger at preserving the language than at using it to deliver everything else.

The actual state of play

Put fintech, healthtech, and edtech side by side and a pattern emerges: native-language tech is being built by small open-source projects, sector startups treating it as an accessibility feature rather than a headline, and academic teams whose work rarely leaves a journal PDF, plus, in at least one case, a non-Nigerian organization doing the most visible deployment. It’s not a story of absence. It’s a story of fragmentation severe enough that founders in fintech have likely never heard of the healthtech translation research solving an adjacent version of their own problem, and an edtech founder building a Yoruba vocabulary app has little reason to know that an open-source financial-intent parser down the road has already solved the harder problem of understanding informal, code-switched Yoruba input.That fragmentation has a direct cost. A fintech team debugging Yoruba misrecognition and a healthtech team debugging Yoruba medical mistranslation are hitting overlapping technical walls, tonal ambiguity, code-switching, missing domain vocabulary, with no shared venue to compare notes. Each sector is quietly re-solving pieces of the same problem in isolation. That’s the gap worth naming before this vertical gets anywhere near a conference stage: not that native-language tech doesn’t exist across Nigerian industries, but that nobody building it across those industries is talking to each other yet.

Share

Leave a comment

Leave a Reply

Your email address will not be published. Required fields are marked *