Togo Is Teaching Its Local Languages to AI.
Ewe (Eʋegbe in the language itself) is spoken by about 5 million people in West Africa, mainly in Ghana and Togo. It belongs to the Gbe language group, whose other major language is Fon, spoken mostly in Benin. Like many African languages, it is also tonal. (Source: Wikipedia)
There, I said it
I don’t like the accent of AI-generated videos when they speak Ewe. It drives me crazy! :( But let’s be honest: it’s real progress. A few years ago, AI simply didn’t know Ewe at all.
So why does it matter? Because some local languages are at risk of disappearing. Because they are rarely taught, at home or at school. And because access to education, healthcare and information also runs through our languages.
Why AI speaks our languages poorly
According to Right for Education, the main reason is a lack of data. AI learns from huge amounts of text and voice recordings, and for many African languages, there is very little of either. But African researchers and organizations are already building open datasets. Africa is not only a user of AI, it is also a creator.
The first building blocks: Takwimu Lab
It didn’t all start in 2026. On November 10, 2020, the Takwimu Lab team, as part of the AI4D initiative, openly published a dataset of more than 70,000 sentence pairs, annotated for machine translation from French to Ewe and from French to Fongbe. Released under a CC BY 4.0 license, it lets researchers and developers train and test models on our languages.
Four people are behind this work: Momboladji Balogoun (Bolaji), co-founder of Takwimu Lab and originally from Benin, Kevin Degila, Godson Kalipe and Jamiil Touré Ali. Their project reminds us of something essential: building AI for Africa is not only about adapting existing models. It means investing in our languages and in data that looks like us. (Sources: Knowledge 4 All Foundation, Zenodo, AI4D Takwimu Lab Challenge)
Togo takes action
And it’s very recent. On September 28, 2026, in Lomé, Togo AI Lab held a public testing session of Voix du Togo (“Voice of Togo”), a platform designed to give access to public services in local languages. Participants were selected based on the languages they speak, to get a variety of feedback. (Source: Togo First, in French)
The project starts with three pilot languages: Kabyè, Ewe and Moba, with the ambition to extend later to the country’s roughly 50 languages.
This initiative is part of the National Local Languages Challenge, presented in July at the AI for Good summit in Geneva. It is led by the Ministry of Public Service Efficiency and Digital Transformation, with partners including Zindi, Umbaji, CEGA at the University of California, Berkeley, and UNDP. The goal: to build voice and text datasets to train AI models adapted to our linguistic realities.
A contribution platform is meant to collect voice recordings, texts and translations. The models could then serve public administration first, and gradually education, health, justice and agriculture. UNDP also supports the project with a GPU-based computing infrastructure, described as a supercomputer, to train these models in Togo. (Source: UNDP Togo, in French)
Get Afi Emmanuela AGBEKPONOU’s stories in your inbox
Join Medium for free to get updates from this writer.
Concrete results are still to come, but the direction is clear: for minister Cina Lawson, language models are essential public infrastructure for the digital age.
Also worth knowing: Masakhane and Afriklang
- Masakhane: a grassroots organization that aims to strengthen natural language processing research in African languages, for Africans, by Africans. Its name roughly means “we build together” in isiZulu. It starts from a simple observation: about 2,000 of the world’s languages are African, and yet they are barely represented in technology. (masakhane.io)
- Afriklang: an initiative incubated in MEST Africa’s AI Startup program. It builds voice datasets for companies developing AI in African languages, using recordings from native speakers that are transcribed and annotated. To capture natural speech, it notably uses images that speakers describe spontaneously. Its catalog includes an Ewe voice corpus, as well as Wolof, Hausa, Fulfulde, Twi, Yoruba and Luganda. The team includes Agnilonda Pakou (operations and leadership), Khalifa Niamadio Mamadou (business development), Mahougnon Fredy Houndayi (AI and engineering) and Babacar Ndao (linguistics). (afriklang.com)
The link with Frugal AI
Local languages are not just a feature to bolt onto an AI that already exists. They can be at the heart of a real research question: how do we design intelligent systems that work where data, energy, infrastructure and money are limited? That is the idea behind Frugal AI, and it is exactly our reality.
You can contribute today
You don’t need to be an expert (you read my previous article 😉). Pick a project and jump in: translation, reading aloud, describing images, proofreading. You can contribute on these platforms:
What if you did it with your parents?
For me, it’s real quality time with my mom, who is Ewe and writes and speaks it very well. Together, we translate, read and describe. I find it a lot of fun, and it helps me improve my Ewe. I think she even enjoyed it more than I did. I also think it’s a good cognitive exercise for our parents, and I recommend it.
Our parents may not push or commit anything on GitHub, but they can help us with translations. So do it with them, whether it’s Ewe, Kabiyè, Moba, whatever the language: do it.
What I hope for
- That parents teach their children to read, write and speak local languages.
- That the education system teaches them too. I was lucky to learn Ewe at my high school in Togo, and it mattered to me.
Why it’s worth it
- Your language counts in the digital world: every recording and every translation helps models better understand and speak our languages.
- More accessible services: eventually, being able to handle administrative procedures, then access healthcare or education, in your own language.
- Hands-on experience: contributing to an open source project means learning as you go, without waiting to be ready.
- A bond with your family and culture: you help pass your cultural identity on from generation to generation.
