metadata

license: apache-2.0
language:
  - ca
  - nl
  - fr
  - af
  - ga
  - gd
  - ms
  - oc

Dataset Summary

This is a variation of mMARCO by Bonifacio et al. used for the "Improving Low-Resource Retrieval Effectiveness using Zero-Shot Linguistic Similarity Transfer" ECIR2025 paper.

The source code for the paper can be found here

mMARCO is a multilingual version of the MS MARCO passage ranking dataset. For more information, checkout the original mMARCO papers:

Dataset Structure

queries/ contains both the translated dev queries used for evaluation under /dev_small and the translated train/judged queries used for training under /train/.

The dev queries are in the following languages: Afrikaans, Catalan, Scottish Gaelic, Malay, Occitan, Sicilian, Cantonese. the train queries are in the following languages: Afrikaans, Catalan.

train_with_negs contains the triples (query, positive doc, negative doc) used for training the models our ECIR 2025 IR4GOOD paper.

Licensing Information

This dataset is released under Apache license 2.0.

Citation Information

If you use this dataset, please cite:

@article{DBLP:journals/corr/abs-2108-13897,
  author       = {Luiz Bonifacio and
                  Israel Campiotti and
                  Roberto de Alencar Lotufo and
                  Rodrigo Frassetto Nogueira},
  title        = {mMARCO: {A} Multilingual Version of {MS} {MARCO} Passage Ranking Dataset},
  journal      = {CoRR},
  volume       = {abs/2108.13897},
  year         = {2021},
  url          = {https://arxiv.org/abs/2108.13897},
  eprinttype    = {arXiv},
  eprint       = {2108.13897},
  timestamp    = {Mon, 20 Mar 2023 15:35:34 +0100},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2108-13897.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}