News Analysis · Science and culture · African languages · AI
African-language AI: why speakers and evidence both matter
Aiseng's SAYAS recognition brings attention to his research. Comparing its published limits with Masakhane's principles shows what language-inclusive design needs.

Wits media scholar Dr Kealeboga Aiseng has been elected to the South African Young Academy of Science, the university announced on 9 November 2026. His work raises a practical question for anyone using translation, captions or voice search: who decides whether a tool works for your language?
The university's announcement (opens in a new tab) describes his research on African-language and community media, broadcasting and cultural representation.
In a paper published on 4 May 2026 (opens in a new tab), Aiseng proposes that African-language speakers help shape datasets, design decisions and evaluation. The study concerns Setswana, Tshivenda and Xitsonga, using platform audits conducted from June to September 2025 alongside interviews and rural focus groups.
He argues that communities should help decide which meanings and uses count when judging a system. The framework is a proposal for future design and evaluation.
What the evidence can establish
The published text leaves participant totals unresolved. Its methods describe six focus groups with 48 participants. Later passages refer to four groups and 32 participants without explaining how those figures relate.
He limits the findings to three languages and selected rural settings; external audits cannot reveal companies' internal systems. Observations from 2025 cannot establish performance after updates.
A separate primary source offers a useful design comparison. On its public principles page (opens in a new tab), African-language technology community Masakhane says Africans should drive research and decide what data represents their communities and how it is used. It also commits to sharing research code and data so others can reproduce and build on the work.
Those principles describe how research should be conducted. They do not validate the particular measurements in Aiseng's paper.
Our comparison points to two tests for claims about language-inclusive technology: whether speakers can influence its design, and whether the evidence is clear enough for others to check. Readers need both to judge what a tool can offer.
Sources, evidence and corrections
- Wits researchers recognised among South Africa's leading scholars (opens in a new tab) · 9 November 2026.
- Decolonizing Algorithms: AI-Driven Media and the Reclamation of African Linguistic Identities (opens in a new tab) · First published online 4 May 2026.
- Masakhane mission, values and participation (opens in a new tab) · Undated principles page. Checked 13 November 2026.
