Incorrect Translations

#2
by prokojin - opened

Hello,
it seems something went wrong during crafting of this dataset. Few examples from cs-en:

{'translation': {'cs': "That's a problem of popularity, we found other suitable rooms nearby.", 'en': 'There is a free wi-fi hotspot which is available in a hall.'}}
{'translation': {'cs': 'A breakfast was really good. There is a free wi-fi hotspot which is available in a hall.', 'en': 'All in all, I will visit this hotel again.'}}
{'translation': {'cs': 'Me gustaría destacar su ubicación, por su cercanía al Vaticano y por la excelente conexión con el transporte urbano.Destaca también por la comodidad de sus instalaciones y la tranquilidad de su entorno.', 'en': 'Hospitalidade, da localização geográfica, das condições do quarto e de todo o hotel.'}}
{'translation': {'cs': "nous étions un peu inquiets car il n'y avait aucun commentaires lorsque nous avons réservé! en fait il s'agit d'un appart dans un tres bel immeuble rue agréable, entree par cour intérieure avec palmiers et fontaine, cet appart a été transformé en 6 chambres avec s de bain et une pièce est réservée pr les petits dej la déco est desing tt est neuf.à recommander en passant par votre site, sinon les chambres sont à 150euros la nuit, or ns avons payé 80 euros.Situé à 10mn du Vatican et proche d'une station de métro.", 'en': 'Me gustaría destacar su ubicación, por su cercanía al Vaticano y por la excelente conexión con el transporte urbano.Destaca también por la comodidad de sus instalaciones y la tranquilidad de su entorno.'}}
{'translation': {'cs': 'Het is erg leuk om in een mooi appartementencomplex te verblijven in een van de topwijken van Rome. Elke ochtend kun je een gezellig praatje maken bij het ontbijt waarbij je veel tips krijgt.', 'en': "nous étions un peu inquiets car il n'y avait aucun commentaires lorsque nous avons réservé!"}}

How was this dataset created? I'm interested in using it and I would like to fix it.

WMT: Workshop on Statistical Machine Translation org
edited Dec 2, 2022

Thanks for reporting, @prokojin .

You have the loading scripts in the files:

This dataset uses the following subsets:

  • train:
    • "europarl_v7"
    • "europarl_v8_16"
    • "commoncrawl"
    • "newscommentary_v11"
    • "czeng_16pre"
    • "yandexcorpus"
    • "wikiheadlines_fi"
    • "wikiheadlines_ru"
    • "setimes_2"
  • validation:
    • "newsdev2016"
    • "newstest2015"
  • test:
    • "newstest2016"
    • "newstestB2016"
WMT: Workshop on Statistical Machine Translation org
edited Dec 2, 2022

After investigating this issue, we have discovered that there are issues with the source data file of the Common Crawl corpus: https://www.statmt.org/wmt13/training-parallel-commoncrawl.tgz

We have contacted the WMT organizers.

Also this example, German instead of Romanian:
raw_datasets = load_dataset("wmt16", "ro-en")
raw_datasets["train"][20]

ouput:
{'translation': {'en': '(The sitting was opened at 9.35 a.m.)',
'ro': '(Die Sitzung wird um 9.35 Uhr eröffnet.)'}}

Sign up or log in to comment