metadata

configs:
  - config_name: default
    data_files:
      - split: sentences
        path: data/sentences-*
extra_gated_heading: Protecting the integrity of FLORES-250 for evaluation
extra_gated_fields:
  I agree not to re-host FLORES-250 in places where it could be picked up by web crawlers: checkbox
  If I evaluate using FLORES-250, I will ensure that its contents are not in the training data: checkbox
dataset_info:
  features:
    - name: rus
      dtype: string
    - name: udm
      dtype: string
  splits:
    - name: sentences
      num_bytes: 129728
      num_examples: 250
  download_size: 72479
  dataset_size: 129728
language:
  - udm

FLORES-250, Russian and Udmurt sentences

Compared to the original FLORES-250, in the Russian version the sentence

Вечер начал певец Санджу Шарма, за ним выступил Джай Шанкар Чаудхари. esented the chhappan bhog bhajan также. Ему аккомпанировал певец Раджу Кханделвал

changed to

Вечер начал певец Санджу Шарма, за ним выступил Джай Шанкар Чаудхари. Ему аккомпанировал певец Раджу Кханделвал.

Usage

from datasets import load_dataset

dataset = load_dataset("udmurtNLP/flores-250-rus-udm")

Citation

@inproceedings{yankovskaya-etal-2023-machine,
    title = "Machine Translation for Low-resource {F}inno-{U}gric Languages",
    author = {Yankovskaya, Lisa  and
      Tars, Maali  and
      T{\"a}ttar, Andre  and
      Fishel, Mark},
    booktitle = "Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa)",
    month = may,
    year = "2023",
    address = "T{\'o}rshavn, Faroe Islands",
    publisher = "University of Tartu Library",
    url = "https://aclanthology.org/2023.nodalida-1.77",
    pages = "762--771",
    abstract = "This paper focuses on neural machine translation (NMT) for low-resource Finno-Ugric languages. Our contributions are three-fold: (1) we extend existing and collect new parallel and monolingual corpora for 20 languages, (2) we expand the 200-language translation benchmark FLORES-200 with manual translations into nine new languages, and (3) we present experiments using the collected data to create NMT systems for the included languages and investigate the impact of back-translation data on the NMT performance for low-resource languages. Experimental results show that carefully selected limited amounts of back-translation directions yield the best results in terms of translation scores, for both high-resource and low-resource output languages.",
}