Research

Published on November 1, 2026

Africa is home to more than 2,000 languages, yet support for these languages at the point of text input remains limited. Many African orthographies require diacritics and tone marks that standard keyboards do not provide, and emoji search in an African language finds little or nothing because the keyword data in Unicode CLDR remains uneven or absent. We demonstrate TonalKeys, a live input system that closes both gaps with two front-ends sharing one retrieval core. A publicly released Chrome extension works anywhere on the web, including canvas applications such as Google Docs where direct insertion is blocked, and a working Android keyboard prototype serves mobile-first settings. TonalKeys offers curated character entry for 18 languages across five regions and a morphology- and code-mix-aware emoji search that currently covers Swahili, with further languages planned.

Published on August 7, 2026

Multilingual neural machine translation systems exhibit well-documented performance asymmetries across translation directions, but the structural drivers and scale-dependence of these asymmetries remain largely understudied for African languages. Existing work largely attributes asymmetries in directional translation faithfulness to model-level factors such as training data imbalance and decoding strategies, without isolating the role of source language typology or quantifying how training data coverage moderates per-language performance. We present a direction-controlled empirical investigation using 1,000 meaning-equivalent English--Swahili sentence pairs translated into six African target languages by three multilingual NMT systems under a cont…

Published on April 20, 2026

A look back at our 6-week intensive pilot program designed to bridge the gap between African linguistics and AI technology.

Published on February 27, 2026

The creation of language resources for African languages faces significant challenges related to sustainability. As a result, thousands of languages on the continent remain severely low-resource. Although community-led efforts have been impactful, they are expensive and less scalable. The use of synthetic data from large language models may be more scalable, but present the risks of introducing "translationese" and amplifying existing biases.

Published on December 19, 2025

How we grew from a March 2025 idea into a language infrastructure company with 51,000+ validated African sentences.