Tether Unveils AI Translation Models for Africa's Low-Resource Languages

Tether's AI Research Unit Releases Open-Source Translation Models

Tether, the stablecoin company, has released a set of open-source neural machine translation models for 19 Sub-Saharan African languages. The models, called TranslatePsy-AfriSLM, are tiny and can run directly on a phone without requiring internet connectivity. This is significant because many communities in Africa rely on low-end devices with limited internet access.

The dataset used to train the models contains over 215 million bidirectional training examples paired with English, making it one of the largest datasets for African languages. The models are built to be Bergamot-compatible, allowing them to slot into existing infrastructure without requiring developers to rebuild from scratch.

The QVAC SDK, a software development kit released alongside the translation models, allows developers to bundle various AI models into a unified deployment framework. This includes support for European languages and can extend coverage to dozens or hundreds of languages using approximately 50 language pairs.