Researchers have developed the ALTO dataset, a new collection of 3,022 instances of tech-facilitated gender-based violence (TFGBV) in Levantine Arabic, African French, and Swahili. This community-led effort, in collaboration with civil society organizations from Jordan, Francophone West Africa, and Kenya, addresses the lack of culturally and linguistically relevant data for TFGBV detection in non-Global North languages. Benchmarking experiments revealed that existing harmful content detection models perform poorly on this dataset, particularly missing covert harms and showing higher error rates in Swahili, though fine-tuning shows improvement. AI
IMPACT This dataset aims to improve AI's ability to detect and combat online gender-based violence in underrepresented languages and cultural contexts.
RANK_REASON The cluster is about a new academic dataset and paper released on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]
- African French
- Francophone West Africa
- Hugging Face
- Jordan
- Kenya
- Levantine Arabic
- Swahili
- Tianwei (Terry) Zhang
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →