What MasakhaneTool is​

MasakhaneTool is an Apache 2.0, mobile-first, offline-capable Progressive Web App for African-language NLP data collection. We built it because most existing annotation tools assume desktop browsers, fast connectivity, and Latin script — assumptions that don't hold for the African contexts we work in.

In its current shape, the tool supports:

• Sentiment annotation (e.g., AfriSenti-style three-way labelling)
• Named entity recognition with custom entity sets
• LLM-evaluation workflows — score model output for accuracy, fluency, and safety
• Inter-annotator agreement dashboards (Fleiss' κ, Krippendorff's α)
• Virtual keyboards for non-Latin scripts (Ge'ez and others)
• Offline-first capture with background sync when connectivity returns

Pilot deployments are running at Bayero University, Kano and Bahir Dar University ICT4D.

Why we need testers now​

We've validated the architecture and the core annotation flows internally. What we need next is real users on real tasks — annotators using the tool on their own phones and laptops, on their own networks, in their own languages — to surface the gaps we can't see from inside the team.

Specifically:

• Usability — what feels confusing? what slows you down?
• Scripts and keyboards — does typing in your language work the way you expect?
• Offline behaviour — does the sync feel reliable on patchy 3G?
• Mobile vs desktop — which workflows feel right on each?
• Accessibility — anything that breaks for screen readers, low vision, or limited motor control?

Who we're looking for​

You don't need to be a senior researcher. We're especially interested in:

• Annotators doing labelling work for African-language datasets (paid or volunteer)
• Students learning NLP or linguistics
• Researchers running annotation projects (your team becomes a pilot)
• Educators who teach data-collection workflows
• Native speakers of any African language — we want broad script and morphology coverage

If your work touches African-language data in any way, you fit the profile.

What testing looks like​

• Time commitment: 1–3 hours over a 2-week window. You can spread it however you want.
• Tasks: complete a small annotation project we provide (or use the tool on a project of your own).
• Feedback: a 15-minute call OR a written form — your choice.
• No NDA, no signup paperwork.
• All testers are credited in the v1 release notes if they want to be (or stay anonymous).

How to sign up​

Three ways, pick whichever fits:

1. Discord — say hi in #masakhane-tool and a maintainer will reach out
2. GitHub Discussions — open a thread in the project's discussions with a brief intro
3. Email — write to the team via the address on the About page

We'll get back within 48 hours with onboarding details and a tester invite.

If you can't test but know someone who would be great — please forward this. The strongest signal of community-built infrastructure is community-led testing.

Read the paper: The Rise of AfricaNLP: A Survey of Contributions, Contributors, Community Impact, and Bibliometric Analysis

What we did​

We assembled and analysed a dataset of 2.2K NLP papers, 4.9K contributing authors, and 7.8K human-annotated contribution sentences spanning 2005–2025. The goal: examine how the field has grown across:

• Publications and venues — where AfricaNLP work appears, and how that's shifted
• NLP topics and tasks — what's being worked on, and what's missing
• Contributors — authors, institutions, geographic distribution
• Funding and community structures — who funds what, who collaborates with whom

We also built a research explorer tool so the community can query the dataset and track emerging trends without re-doing the bibliometric work.

Why it matters​

A field's reputation is often built on a handful of headline projects. The reality is broader and messier — and the only way to see it is to count carefully.

A few things stood out to us in the data (the paper has the full picture):

• The growth in AfricaNLP publications over the last five years is non-linear, but the contribution is unevenly distributed — both across languages and across institutions. Closing those gaps is the field's next decade of work.
• Community-led collaborations like Masakhane, HausaNLP, EthioNLP, and others account for a meaningful share of the recent growth, especially for low-resource languages that don't have institutional NLP labs of their own.
• Annotation, dataset construction, and evaluation remain disproportionately under-cited relative to the engineering effort they require — a structural issue this survey hopes to nudge.

Why this is on the Masakhane Playbook blog​

Several of the survey's authors — including the lead authors and reviewers — are active in the Masakhane community. The work intersects directly with what the Playbook documents: how annotation, dataset, and benchmark contributions accumulate over time and shape what's possible.

If you're choosing what to work on next, or framing a grant or a thesis around African-language NLP, we'd encourage you to read the survey, query the explorer, and write back to us with what you find missing. The dataset is open; gaps are an invitation, not a complaint.

Read & contribute​

• 📄 Paper: arxiv.org/abs/2509.25477
• 🤝 Discuss the findings: join #research in our Discord
• ✏️ Suggest corrections to the dataset: the explorer tool accepts community input — see the paper for details

The next two decades of AfricaNLP will look very different if we're deliberate about how we build infrastructure, fund work, and credit contribution. This survey is one snapshot — we'd like to keep taking them.

Cite the survey:

Belay, T. D., Hussen, K. Y., Imam, S. H., Ahmad, I. S., Inuwa-Dutse, I., Haile, A. B., Sidorov, G., Vazquez, E. R., Ameer, I., Abdulmumin, I., Gwadabe, T., Marivate, V., Yimam, S. M., & Muhammad, S. H. (2026). The Rise of AfricaNLP: A Survey of Contributions, Contributors, Community Impact, and Bibliometric Analysis. arXiv:2509.25477.

Why a Playbook?​

African languages remain underrepresented in mainstream NLP. The Playbook exists to close that gap by collecting practical, reproducible, and culturally grounded guidance: from data collection and annotation, through model training and evaluation, to deployment and ethics. Rather than a static textbook, it is a living resource maintained by the community it serves.

What we're looking for​

We welcome chapter proposals on (but not limited to):

• Data — curation, licensing, dialectal coverage, low-resource bootstrapping, parallel corpora, speech datasets.
• Annotation — guidelines for African languages, inter-annotator agreement, tooling, working with native-speaker annotators.
• Modeling — machine translation, ASR/TTS, NER, POS, sentiment, multilingual pretraining, adapters, and small/efficient models.
• Evaluation — human evaluation protocols, benchmarks, error analysis specific to African language families.
• Deployment — productionizing models on the continent, edge devices, offline-first applications, language ID, code-switching.
• Ethics & Community — consent, data sovereignty, fair compensation for contributors, language preservation.
• Case Studies — end-to-end stories from real projects (Hausa, Amharic, Swahili, Yoruba, Igbo, Zulu, Wolof, and many more).

Who can contribute?​

Anyone. You do not need to be an academic. We particularly encourage:

• First-time authors paired with mentors from the community
• Native speakers and linguists working alongside ML practitioners
• Students, independent researchers, and industry contributors
• Cross-organizational collaborations (Masakhane, EthioNLP, HausaNLP, Lanfrica, Zindi, and others)

How to propose a chapter​

1. Submit a proposal — a short abstract (200–400 words), an outline, target audience, and the languages/topics covered.
2. Review — proposals are reviewed by the editorial team for fit, scope, and overlap with existing chapters.
3. Draft — accepted authors work in the open on the Playbook GitHub repository, with reviewer support.
4. Publish — accepted chapters are merged into the Playbook with author attribution and DOIs (where applicable).

Key dates​

How to get involved​

• Read the existing Playbook: /playbook
• Open a proposal issue on the GitHub repository
• Join the Masakhane community: masakhane.io
• Questions? Reach out via the GitHub Discussions tab on the repo.

We are building a resource by the community, for the community. If you have a perspective, an experiment, or a story that deserves to be in the Playbook — we want to hear from you.

Ǹbá! Karibu! Welkom! Bienvenue! Bem-vindos!

What you can expect here​

• Calls for contribution — chapters, datasets, annotations, reviewers.
• Release notes — when new versions of the Playbook ship.
• Community spotlights — research, tools, and projects from across the Masakhane network.
• Tutorials & how-tos — short, practical pieces that complement the Playbook.

Subscribe​

The blog publishes an RSS feed at /blog/rss.xml and an Atom feed at /blog/atom.xml. Drop either into your feed reader to follow along.

How to write a post​

Posts live in the blog/ folder of the MasakhanePlaybook repo. Each post is a markdown file (or a folder containing index.md if you have images). The frontmatter looks like this:

---
slug: your-post-slug
title: "Your post title"
authors: [shamsuddeen]
tags: [announcement]
image: /img/blog/your-thumbnail.png
---

Add a <!-- truncate --> marker after the intro paragraph so the blog index shows only that intro on the card.

That's it — open a PR, and we'll publish it.