Dataset Card for "wmt24-de-en"

Curated German–English WMT24 datasets at different scales. Select a config (subset) when loading, similar to language pairs on wmt/wmt14.

Dataset Summary

from datasets import load_dataset

# Quick smoke tests (~1k pairs total)
ds = load_dataset("neelpawar/wmt24-de-en", "toy")

# Curated ~5.18M train pairs + official val/test
ds = load_dataset("neelpawar/wmt24-de-en", "curated_5m")

Config	Description	Status
toy	WMT24 test set, split for fast debugging	Available
curated_5m	~5.18M curated train + WMT23 val + WMT24 test	Available
full	Full WMT24 de–en corpus (~295M pairs)	Planned

toy

Derived from the WMT24 test set (human reference translations). The original 997 sentence pairs were split into train, validation, and test via two train_test_split calls (test_size=256, then test_size=128 on the held-out portion), both with seed=42.

curated_5m

A curated ~5.18M subset of the WMT24 German–English training data, with:

• train: 5,180,244 curated sentence pairs
• validation: 557 pairs from the WMT23 test set (human references)
• test: 997 pairs from the WMT24 test set (human references)

Dataset Structure

Data Instances

{
  "translation": {
    "de": "Irgendwo hab ich welches, nur wo?",
    "en": "Got some somewhere but exactly where?"
  }
}

Data Fields

• translation: a multilingual string variable, with possible languages including de, en.

Data Splits

config	train	validation	test
toy	741	128	128
curated_5m	5,180,244	557	997

Licensing

The sentence pairs are derived from WMT24 and are governed by the WMT24 licensing terms:

The data released for the General MT task can be freely used for research purposes, we ask that you cite the WMT24 shared task overview findings paper, and respect any additional citation requirements on the individual data sets. For other uses of the data, you should consult with original owners of the data sets.

This repository repackages that data into train/validation/test splits for convenience; it does not relicense the underlying text.

Citation

@inproceedings{kocmi-etal-2024-findings,
  title     = {Findings of the {WMT}24 General Machine Translation Shared Task: The {LLM} Era Is Here but {MT} Is Not Solved Yet},
  author    = {Kocmi, Tom and Avramidis, Eleftherios and Bawden, Rachel and Bojar, Ond{\v{r}}ej and Dvorkovich, Anton and Federmann, Christian and Fishel, Mark and Freitag, Markus and Gowda, Thamme and Grundkiewicz, Roman and Haddow, Barry and Karpinska, Marzena and Koehn, Philipp and Marie, Benjamin and Monz, Christof and Murray, Kenton and Nagata, Masaaki and Popel, Martin and Popovi{\'c}, Maja and Shmatova, Mariya and Steingr{\'i}msson, Steinth{\'o}r and Zouhar, Vil{\'e}m},
  booktitle = {Proceedings of the Ninth Conference on Machine Translation},
  month     = {November},
  year      = {2024},
  address   = {Miami, Florida, USA},
  publisher = {Association for Computational Linguistics},
  pages     = {1--46},
  url       = {https://aclanthology.org/2024.wmt-1.1/},
  doi       = {10.18653/v1/2024.wmt-1.1}
}