metadata
license: apache-2.0
This dataset contains all MQM human annotations from previous WMT Metrics shared tasks and the MQM annotations from Experts, Errors, and Context.
The data is organised into 8 columns:
- lp: language pair
- src: input text
- mt: translation
- ref: reference translation
- score: MQM score
- system: MT Engine that produced the translation
- annotators: number of annotators
- domain: domain of the input text (e.g. news)
You can also find the original data here. We recommend using the original repo if you are interested in annotation spans and not just the final score.