African-Language Science Translation Is Becoming a Technology Infrastructure Question

Africa’s technology debate is often framed around connectivity, startups, data centres and digital payments. But a quieter infrastructure question is becoming just as important: in what languages will Africans access scientific knowledge?

A new research paper, AfriScience-MT: Towards Decolonizing Science in Africa through Text Translation, introduces a parallel corpus for translating scientific text into six African languages: Amharic, Hausa, Luganda, Northern Sotho, Yorùbá and isiZulu. The paper covers 11 scientific domains and is designed to support machine translation benchmarking for African-language scientific communication. The paper is available on arXiv.

The project matters because scientific access in Africa is still heavily mediated through colonial languages. English, French and Portuguese dominate education, research publication and public science communication across much of the continent. This creates a barrier for millions of people who may speak African languages fluently but encounter science, health information and technical knowledge in languages that are not their strongest.

The researchers behind AfriScience-MT argue that one obstacle is the limited availability of established scientific terminology in many African languages. To address this, professional translators worked with expert science communicators to translate plain-language summaries of scientific papers and create new terms where needed.

This moves the issue beyond translation as a simple software problem. Translating science into African languages requires linguistic judgment, domain expertise, terminology development and cultural context. Machines can help, but the underlying infrastructure still depends on human knowledge work.

The paper also tests machine translation systems and large language models across different settings. It reports that closed-source models performed strongly, while some open systems also showed competitive results after fine-tuning or in-context learning.

For Africa’s digital future, this raises several important policy questions.

First, language should be treated as part of digital inclusion. A person who has internet access but cannot easily access technical knowledge in a familiar language is still partly excluded from the knowledge economy.

Second, African-language AI cannot be built only by importing general-purpose models trained mainly on global data. It requires local datasets, terminologies, benchmarks and evaluation methods that reflect African linguistic realities.

Third, scientific translation has direct development relevance. Public health, climate adaptation, agriculture, education and technology adoption all depend on whether communities can understand technical information in practical and trusted ways.

AfriScience-MT does not solve the wider African-language technology gap on its own. It is a research contribution, not a continent-wide deployment. But it points in the right direction. It shows that language infrastructure can be built deliberately, with translators, scientists and technologists working together.

The next challenge is scale. African governments, universities, publishers, AI labs and development partners will need to decide whether African-language knowledge systems remain niche research projects or become part of mainstream science, education and innovation policy.

The future of African AI will not only be written in code. It will also be written in Amharic, Hausa, Luganda, Northern Sotho, Yorùbá, isiZulu and hundreds of other languages that carry the continent’s ideas, questions and communities.