Basque AI Gender Bias Study Shows Male Defaults in Professional Translations

Researchers from HiTZ at the University of the Basque Country presented new evaluation resources at the LREC 2026 conference to measure gender bias in artificial‑intelligence translation systems. The tools – WinoMTeus and FLORES+Gender – focus on the Basque language, which lacks grammatical gender, revealing that current AI models often default to masculine forms even for occupations predominantly held by women.

The study highlights persistent gender stereotypes in AI outputs and provides the first metrics tailored to languages beyond English. The resources are now publicly available for the scientific community to improve the fairness and inclusivity of multilingual AI systems.