On the Limitations of Cross-Lingual Consistency in Multilingual Text-to-image Generation
LingT2I tests multilingual image generation across 10 languages and finds uneven behavior by language.
The paper introduces a 33,000-prompt benchmark for measuring cross-lingual effects in both image content and text rendering. Its analysis reports “linguistic inequality” and language-dependent trade-offs across evaluation dimensions. The authors also say linguistic and cultural context systematically shape model outputs. ArXiv · AI/CL/LG's note
The paper introduces a 33,000-prompt benchmark for measuring cross-lingual effects in both image content and text rendering. Its analysis reports “linguistic inequality” and language-dependent trade-offs across evaluation dimensions. The authors also say linguistic and cultural context systematically shape model outputs. ArXiv · AI/CL/LG's note
score 5