HAT
datasetHallucination Annotation for Translation — the first large-scale machine-translation hallucination benchmark: 350,959 span-level, professionally annotated samples across 38 language pairs. ACL 2026 main-conference long paper (no arXiv preprint). CC-BY-NC-ND-4.0.
Paper
Venue ACL 2026
Dataset
Size 350,959 annotated samples, 38 language pairs
License CC-BY-NC-ND-4.0