WMT-24

The WMT-24++ dataset is a benchmark that contains translations across 54 language pairs and is used to evaluate the literality of human translations and machine translation systems, including their performance in direct translation, iterative self-revision, and post-editing tasks.

Papers using WMT-24 (3)

TranslateGemma Technical Report2026 · 1 citesTesting the Deliteralization Hypothesis in Human and Machine Translation2026Please Translate Again: Two Simple Experiments on Whether Human-Like Reasoning Helps Translation2025