New GAND resource aims to expose gender bias in machine translation

Researchers have introduced GAND, a new benchmarking resource designed to analyze how machine translation systems handle gender ambiguity. GAND consists of English source sentences that are intentionally gender-neutral, aiming to reveal biases and stereotyping in translations. The resource facilitates interpretability analysis by translating ambiguous sentences into gendered languages and using contrastive attribution methods to identify which source words influence the gender assigned in the translation. AI

IMPACT This resource could lead to more equitable and less biased machine translation systems, improving user experience and reducing harm from stereotyping.

RANK_REASON The cluster contains an academic paper detailing a new resource for evaluating machine translation systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Janiça Hackenbuchner
• Machine translation

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →