zouhar/wmt-human-all

This dataset is continuously updated and contains a compilation of human translation quality assessment from past WMT campaigns. Specifically, this dataset merges all annotation protocols (DA, MQM, ESA) on a semi-unified scale (0 to 100). The current version of the dataset includes human scores up to WMT 2025 (inclusive) and has been created with the following script: import subset2evaluate # version 1.0.20 import json import statistics data =… See the full description on the dataset page: https://huggingface.co/datasets/zouhar/wmt-human-all.