What we're working on

Overview

KenBench provides plug-and-play evaluation via lm-evaluation-harness, a public leaderboard, and baselines for open- and closed-source models across 10 Kenyan languages. Target publication: AfricaNLP 2026 at EACL. Synthesis workshop planned for April 2026.

Target languages

Phases

How to contribute

We need help with dataset integration, model evaluation, human annotation, and paper writing. All skill levels welcome - Kenyan language speakers especially valued.

Overview

One language revival project per cycle: collect speech and text data, run evaluation, and train a small model for an endangered or under-resourced Kenyan language. Starting candidates include Ogiek, Kipsigis, Nandi, and Kisii.

How to contribute

Native speakers, field linguists, data engineers, and ASR researchers are all needed. This project begins after KenBench Phase 1 is complete. Express interest now to help shape it.

Overview

Following KenBench, KenyaNLP will run a focused project training small language and automatic speech recognition models optimised for Kenyan language characteristics and data constraints.

How to contribute

ML engineers, speech researchers, and data annotators will be needed. Register your interest now and we'll reach out when the project launches.