Adaptive Data Collection for Latin-American Community-sourced Evaluation of Stereotypes (LACES)
Published in Findings of the Association for Computational Linguistics: ACL 2026, 2026
The evaluation of societal biases in NLP is critically hindered by geo-cultural gaps that leave Latin America under-represented. LACES introduces an adaptive community-sourced data collection methodology that iteratively surfaces culturally grounded stereotypes, validation protocols, and comprehensive bias benchmarks.