Research
Research that ships.
African Languages Lab studies how to make AI work in African languages, from collecting and validating data to models, speech and evaluation. What we learn goes into products people use every day.
Our core finding
Featured paper, ACL 2026
Data, not size, closes the gap.
Our ACL 2026 paper, The African Languages Lab: A Collaborative Approach to Advancing Low-Resource African NLP, shows that curated, expert-validated data drives the biggest gains for African languages.
African languages in the study
40
Tokens of text data
19B
Hours of speech data
12,628
ChrF++ gain across 31 languages
+23.69
Presented at ACL, San Diego
2026
Our approach
It starts with the data
Most AI reasons in English and translates at the edges. Our research starts earlier, with how language data is gathered, checked and used.
For over a decade we have gathered expert-validated text and speech with native speakers across the continent. Our models are built on 100B+ curated tokens and 13,000+ hours of expert-validated speech.
We share what we learn. African Languages Lab has contributed 62 African languages to Common Crawl, and MansaLLM was named 2025 Innovator of the Year at LocWorld54.
Focus areas
Three areas we research
01
Data and validation
How to collect and check text and speech with native-speaking experts.
02
Speech and translation
Transcription, speech synthesis and translation across 30+ African languages.
03
Evaluation
Measuring how well AI performs in African languages, with expert reviewers through Swarm Intelligence Hub.
Publications
Our published papers
Sep 2026
A Systematic Review of NLP for Ghanaian Languages: Datasets, Models, and a Research Roadmap
The first systematic review of Ghanaian NLP, synthesizing 36 core studies from over 17,000 screened publications across datasets, models, and evaluations.
Jan 2026
Translation as a Scalable Proxy for Multilingual Evaluation
Tested 14 models on 9 benchmarks and found that translation quality is a strong, low-cost signal of multilingual ability.
Oct 2025 · ACL 2026
The African Languages Lab: A Collaborative Approach to Advancing Low-Resource African NLP
How we work with researchers and communities on African NLP, with 19B tokens and 12,628 hours of speech across 40 languages, and gains of +23.69 ChrF++ and +15.34 BLEU across 31 languages.
FAQ
Common questions
Answers to common questions about working on research with African Languages Lab.
Can we collaborate on research?
Can I access your datasets?
Where can I follow your research news?
Collaborate
How a research partnership works
01
Share the question
Tell us the languages, the problem and what you want to learn.
02
Plan data and evaluation
We agree the data, the methods and how results will be measured.
03
Run, review and share
We run the work together, review results with language experts and share what we find.
