Sample-Size Scaling of the African Languages NLI Evaluation
Anuj Tiwari, Oluwapelumi Ogunremu, Terry Oko-odion, Jesujuwon Egbewale, Hannah Nwokocha, Noida Institute of Engineering, Technology, ML Collective [email protected] , [email protected] , [email protected] , [email protected] , [email protected] · 2026
Read Sample-Size Scaling of the African Languages NLI Evaluation by Anuj Tiwari, Oluwapelumi Ogunremu, Terry Oko-odion, Jesujuwon Egbewale, Hannah Nwokocha, Noida Institute of Engineering, Technology, ML Collective [email protected] , [email protected] , [email protected] , [email protected] , [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-03219.txt.
Description
Tiwari and co-authors test how much labeled data African languages need for a language task. African languages have very little labeled data. It is not clear that more labels always help. The team runs a scaling study on natural language inference, using the AfriXNLI benchmark across 16 African languages.
Two multilingual transformer models of about 0.6 billion parameters are tested. Sample sizes run from 50 to 500 labeled cases. Results are averaged over repeat random draws. The trend is not a steady rise. Some languages level off early, and some get worse as data grows. Low-resource settings also show high variance from run to run. The authors conclude that data volume alone does not promise stable gains for African language inference. They call for language-aware dataset building and stronger multilingual models.
Catalog record
| CREATOR | Anuj Tiwari, Oluwapelumi Ogunremu, Terry Oko-odion, Jesujuwon Egbewale, Hannah Nwokocha, Noida Institute of Engineering, Technology, ML Collective [email protected] , [email protected] , [email protected] , [email protected] , [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.03219