5

I would like to know of parallel corpora for English-Spanish (that is, texts or language originally in English translated into Spanish and aligned by excerpts of texts) other than those on the Opus list?

I am particularly interested in finding a corpus of conversation, interviews or other spoken English translated into Spanish.

Here is the Opus list of parallel corpora:

Corpora I am using already:

  • Europarl
  • Global Voices (an online news site)
  • News Commentary (a collection of opinion articles)
  • Tatoeba (edited my manythings.org)
  • Ted Talks
  • Ubuntu
  • UN
  • Wikipedia

Corpora I have rejected because of poor alignment or other issue:

  • Open Subtitles (many erroneous translations)
  • Books corpus (poorly aligned; many erroneous translation pairings)
  • Gnome
  • KDE4 (too much bad English)
  • ECB (too many errors)
  • DGT (language too technical, too many obscure words)
  • eu-bookshop (poorly aligned; many erroneous translation pairings)

Are there any others? (Please include links.)

4
  • 1
    I will update with comments as I find more corpora. The Crater 2 corpus from Lancaster University, which is unique in tagging each individual word by PoS, but I can't figure out where the data come from: catalog.elra.info/product_info.php?products_id=636
    – Joe Walsh
    Jul 2, 2017 at 4:59
  • 1
    I have heard of a 100-language bible corpus. If it isn't already parallel, it should be easily parallelisable using chapter and verse numbers. I don't lknow about the availability conditions of the corpus. Jul 3, 2017 at 16:29
  • Are you using Wikipedia titles or the full corpus? I am surprised to hear the alignment is better than with subtitles given that the latter at least aim to be aligned. Jul 3, 2017 at 18:22
  • 1
    I am using the full Wikipedia corpus from Opus. The issue with subtitles is 1) the translations can be very creative and loose; 2) the subtitles aren't all aligned naturally since the timestamps are different on each set of subtitles and also the transcript of the film. Someone would have to go back and align them and apparently no one has.
    – Joe Walsh
    Jul 6, 2017 at 1:59

1 Answer 1

1

Fitting the requirements of the question (English originals, Spanish translations) is the JW 300 corpus consisting of writings of Jehova's Witnesses. Among the 300+ languages with translations, there is Spanish. The corpus is described in this paper by Agić and Vulić.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.