CaMMT: Benchmarking Culturally Aware Multimodal Machine Translation
Published in Findings of the Association for Computational Linguistics: EMNLP 2025, 2025
Translating cultural concepts requires grounding linguistic text in rich visual context and cultural knowledge. CaMMT establishes a rigorous multilingual and multimodal evaluation framework to measure how state-of-the-art models handle culturally specific phenomena across diverse global regions.