AI | Case studies

AI for social good: Masakhane

by Kasia Kotlarska / July 2026

This article is part of the ongoing AI for Social Good series, where we spotlight social enterprises harnessing artificial intelligence to drive meaningful change. From education to healthcare, each story in this series explores how purpose-driven organisations are navigating the opportunities and challenges of AI adoption in the real world. You can now also read a story of M-Shule, Kabakoo, Wadhwani AI, Rainforest Connection’s Guardian and Social Enterprise Republic of Ireland and learn how they use AI to serve their communities.

Masakhane is a grassroots community-led initiative dedicated to advancing natural language processing (NLP) research for African languages, by Africans, for Africans. Founded on the principle of inclusive collaboration, it addresses the systemic underrepresentation of Africa’s 2,000+ languages in technology. By creating datasets, benchmarks and open-source models, Masakhane fosters both scientific innovation and cultural empowerment. With over 1,000 participants across 30 African countries, the community has published widely, built translation and NER corpora for dozens of African languages and launched flagship projects such as MasakhaneMT: Decolonising Science and MasakhaNER: Know Our Names. Recognised at leading international conferences, Masakhane has demonstrated that African-led participatory research can drive breakthroughs in low-resource NLP, building not only tools but also reclaiming linguistic and cultural ownership in the digital age.

Background

Despite Africa being home to more than 2,000 languages, the overwhelming majority are absent in modern technology. Colonial legacies, low investment in local languages and barriers to data access have left African languages marginalised in NLP. This exclusion undermines access to information, governance, education and healthcare, while reinforcing inequalities in the digital economy.

Masakhane, meaning “We build together” in isiZulu, was established to challenge this status quo by centring African voices, languages and knowledge systems in NLP research. Built around values such as African-centricity, openness, multidisciplinarity and data sovereignty, the movement rejects extractive “parachute research” and instead emphasises participatory, community-driven approaches.

The AI solution

Masakhane develops open-source NLP resources and models tailored for African languages, using neural machine translation (NMT), named entity recognition (NER) and other core tasks as entry points.

Machine Translation: Early efforts trained Transformer-based models on the JW300 dataset for languages such as Edo, Esan, Urhobo and Isoko, achieving competitive BLEU scores despite limited data. These projects proved that community-driven efforts can bootstrap machine translation for previously neglected languages.
MasakhaneMT: Decolonising Science: A flagship project creating multilingual corpora by translating African research (e.g., AfricArxiv papers) into isiZulu, Northern Sotho, Yoruba, Hausa, Luganda and Amharic. This work makes science accessible in Indigenous languages and develops scientific terminology for African contexts.
MasakhaNER: Building high-quality POS and NER datasets for 20 African languages (e.g., Yoruba, Luganda, Wolof, Swahili). This provides critical benchmarks for downstream applications like voice assistants, spell-checkers and chatbots in African contexts.

By focusing on participatory research and open data, Masakhane ensures that Africans own, shape and benefit from the technologies developed.

Implementation and partnership

Masakhane operates as an open community with no entry barriers beyond its Code of Conduct. Its implementation model combines:

Inclusive community building: Slack channels, weekly meetings and Google Colab notebooks for collaborative learning.
Dataset creation: “Data archaeology” to find, curate and annotate resources. All outputs (code, datasets, benchmarks) are shared openly.
Capacity-building: Workshops and mentoring for new NLP practitioners.
Global partnerships: Collaborations with AfricArxiv, ScienceLink, universities (Makerere, Saarland, Brandeis, etc.) and grassroots translators.
Ethical governance: Drafting of the Masakhane Ethical Manifesto to ensure African-centric, responsible dataset creation.

Through these approaches, Masakhane has mobilised a large, diverse community – from linguists and students to software engineers and AI researchers.

Impact and results

Masakhane’s participatory model has yielded concrete outputs and recognition:

Research contributions: Over 49 translation results for 38+ African languages published on GitHub and papers presented at ICLR AfricaNLP, ACL, COLING and EMNLP.
Capacity and community: >1,000 participants from 30 African countries actively collaborating on open NLP projects.
Language resources: Creation of multilingual corpora (e.g., 360,000+ parallel scientific sentences in African languages) and NER/POS datasets for 20 languages.
Visibility and leadership: Keynotes at global NLP workshops (e.g., WMT at EMNLP 2020), positioning African researchers as leaders in the field.
Cultural empowerment: By enabling NLP systems to recognise African names, places and scientific terms, Masakhane addresses issues of dignity, identity and access.

Lessons and next steps

Key lessons learned:

Low-resource is not just a data problem: It is also a social and historical issue requiring participatory solutions.
Community power matters: Grassroots, volunteer-driven collaboration can achieve breakthroughs that institutions alone cannot.
Ethics and sovereignty are central: Open access, data sovereignty and African leadership are essential to avoid repeating exploitative research practices.

Next steps:

Expanding machine translation and NER coverage to more African languages, dialects and domains.
Deepening scientific translation efforts to support STEM education and decolonisation of science.
Strengthening the Masakhane Web platform for open access to trained models and datasets.
Sustaining efforts through the establishment of the Masakhane Legal Entity and long-term community governance.
Exploring speech technologies and audio datasets to build inclusive speech-to-speech systems for Africa’s oral traditions.

References:

https://arxiv.org/pdf/2003.12111

https://arxiv.org/pdf/2003.10704

https://arxiv.org/pdf/2003.11529

https://www.masakhane.io/ongoing-projects/naijanlp-sentiment-lexicon-hate-speech

https://www.masakhane.io/ongoing-projects/makererenlp-text-speech-for-east-africa

https://www.masakhane.io/ongoing-projects/masakhaner-know-our-names

https://www.masakhane.io/ongoing-projects/masakhane-mt-decolonise-science