Beyond Swahili: Designing Inclusive AI for Bantu Languages

Alfred Malingo discusses the unique position of Swahili in AI, arguing that its structural similarities to other Bantu languages make it a far more effective pivot language than English for developing inclusive and accurate models. He deconstructs the failures of typologically mismatched transfer from Indo-European languages and presents a case study, AfriMT-a, to demonstrate how Swahili can serve as a technical bridge for machine translation and representation learning across the Bantu language family.