Study finds only a fraction of African languages supported by major AI models
Of the top 34 languages used online worldwide, the World Economic Forum reported, none was African, and most AI systems are trained on only around 100 of the world's 7,000-plus languages.
- Culture & impact
- Minor
The World Economic Forum reported on a widening gap between the linguistic diversity of Africa and the languages large AI models are trained on, drawing on discussion at its 2024 Sustainable Development Impact Meetings. Africa is home to roughly 2,000 languages, the piece noted, yet of the top 34 languages used on the internet globally, not one is African.
The scale of the underlying data gap was the article’s central point: most chatbots are trained on only around 100 of the world’s more than 7,000 languages, and the great majority of African languages have little to no digitised text or annotated data for the natural-language-processing work that model training depends on. Less than a fifth of the world’s population speaks English, the piece observed, yet English-language data continues to dominate model training — a mismatch that leaves speakers of under-resourced languages able to access AI tools only in a second or third language, if at all.
The report cited efforts working against that gap, including Rwanda’s use of multilingual AI in community health work and the pan-African Masakhane research collective, which builds NLP tools for African languages through open collaboration rather than waiting for commercial labs to prioritise them. The broader concern raised was distributional: without deliberate investment in low-resource languages, the populations already facing the weakest internet access and computing infrastructure risked falling further behind as AI tools became more central to education, government services and the economy — with Africa’s under-35 population set to make up a growing share of the world’s youth in the coming decades.