RicardoRei's picture
Update README.md
7ed04dd
metadata
license: apache-2.0
size_categories:
  - 1M<n<10M
language:
  - cs
  - de
  - en
  - hr
  - ja
  - liv
  - ru
  - sah
  - uk
  - zh
tags:
  - mt-evaluation
  - WMT
  - 12-lang-pairs

Dataset Summary

In 2022, several changes were made to the annotation procedure used in the WMT Translation task. In contrast to the standard DA (sliding scale from 0-100) used in previous years, in 2022 annotators performed DA+SQM (Direct Assessment + Scalar Quality Metric). In DA+SQM, the annotators still provide a raw score between 0 and 100, but also are presented with seven labeled tick marks. DA+SQM helps to stabilize scores across annotators (as compared to DA).

The data is organised into 8 columns:

  • lp: language pair
  • src: input text
  • mt: translation
  • ref: reference translation
  • score: direct assessment
  • system: MT engine that produced the mt
  • annotators: number of annotators
  • domain: domain of the input text (e.g. news)
  • year: collection year

You can also find the original data here

Python usage:

from datasets import load_dataset
dataset = load_dataset("RicardoRei/wmt-sqm-human-evaluation", split="train")

There is no standard train/test split for this dataset but you can easily split it according to year, language pair or domain. E.g. :

# split by year
data = dataset.filter(lambda example: example["year"] == 2022)

# split by LP
data = dataset.filter(lambda example: example["lp"] == "en-de")

# split by domain
data = dataset.filter(lambda example: example["domain"] == "news")

Note that, so far, all data is from 2022 General Translation task

Citation Information

If you use this data please cite the WMT findings: