The corpus collection of 40 languages

The OPUS2 parallel corpus is a set of text corpora which have aligned sentences so sentences correspond the same sentences in other languages. OPUS project collects 40 languages. On account of this, users can check translation sentence pairs for many languages. 

The parallel corpora available here have been collected, prepared and aligned by Joerg Tiedermann in the OPUS project (see We are most grateful to him for his great work and co-operation. The data was prepared for the Sketch Engine using a range of lemmatisers, part-of-speech taggers and Sketch Grammars.

The OPUS2 corpora are the second version having the alignment m:n, which allows for just one corpus per language.

OPUS an open source parallel corpus allows users to search bilingual and multilingual data in many languages, find concordances, collocations, word list and more.

The OPUS project in Sketch Engine contains 40 languages: Afrikaans, Albanian, Arabic, Bosnian, Bulgarian, Chinese Simplified, Chinese Traditional, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Italian, Japanese, Korean, Latvian, Lithuanian, Macedonian, Norwegian, Persian, Polish, Portuguese, Portuguese, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swedish, Turkish, Ukrainian.

  • ECB – European Central Bank corpus (v.0.1)
  • EMEA – European Medicines Agency documents (v.0.3)
  • EUconst – The European constitution (v.0.1)
  • EUROPARL – European Parliament Proceedings (v.3) [transcripts of spoken language]
  • Opensubs – Open Subtitles corpus (v.2) [transcripts of spoken language]
  • KDE4 – KDE4 localization files (v.2)
  • KDEdoc – KDE manual corpus
  • MultiUN – Translated UN documents
  • OpenOffice – a collection of documents from
  • OpenOffice (v.3) – a collection of documents from
  • OpenSubtitles2011 – Open Subtitles corpus (2011 version) [transcripts of spoken language]
  • RF – Regeringsförklaringen – Declarations of Government Policy by the Swedish Government
  • SETIMES2 – A parallel corpus of the Balkan languages (v.2)
  • SPC – Stockholm Parallel Corpora (v.1)
  • TEP – The Tehran English-Persian subtitle corpus (v.0.1) [transcripts of spoken language]
  • Tatoeba – a collection of translated sentences from Tatoeba
  • TedTalks – transcription and translation of TED talks [transcripts of spoken language]
  • UN – Translated UN documents
  • hrenWaC – Croatian-English Parallel Web Corpus
Language Tools Used Grammar
Arabic Stanford tagger using Faster Arabic model trained on the Penn Arabic Treebank with  Tagset Universal Sketch Grammar with AMIRA Tagset
Bulgarian TreeTagger using UTF-8 Bulgarian parameter file trained on Tagset Yes
Chinese (Traditional, Simplified, mixed) Segmented using Stanford Segmenter modelled on segmentation standards by Chinese Penn Treebank. Tagged using Stanford tagger trained on a combination of Chinese Treebank texts from Chinese and Hong Kong sources with Tagset Universal Sketch Grammar with tags
Dutch TreeTagger using UTF-8 Italian parameter file trained on Tagset Dutch Sketch Grammar (NLWAC tagset) v4.0 by Carole Tiberius
English TreeTagger using UTF-8 English parameter file trained on Tagset English Sketch Grammar v.2.0 (Penn Treebank tagset) by Niels Ott
Estonian TreeTagger using UTF-8 Estonian parameter file trained on Tagset Estonian Sketch Grammar v1.2 by Maria and Jelena
French TreeTagger using UTF-8 French parameter file trained on Tagset French Sketch Grammar v1.0 by Adam Kilgarriff
German TreeTagger using UTF-8 German parameter file trained on Tagset Sketch Grammar for German by Matej Durco v3.3
Italian TreeTagger using UTF-8 Italian parameter file trained on Tagset Sketch Grammar for Italian v1.2 by Marco Baroni
Portuguese TreeTagger using Pablo Gamalo’s parameter file Portuguese Wordsketches (Linguateca parsed data) v1.0 by Adam Kilgarriff & DP
Russian TreeTagger using Serge Sharoff’s Russian parameter file trained on Tagset Russian Wordsketches v1.0 by Maria Khokhlova
Spanish TreeTagger using UTF-8 Spanish parameter file trained on Tagset Spanish Wordsketches v1.0 by Nuria Bel and Hada Ross Salazar (Pompeu Fabra University, Barcelona)

All other languages without tagging tools were just tokenised using Universal tokenizer built by Jan Pomikalek and inspired by Laurent Pointal’s TreeTagger wrapper.

The Montenegrin-English parallel corpus consists of movie subtitles. The total number of tokens is 1.07 million. The original data a more information is available on the original website Preprocessing of data was carried out by Nikola Ljubešić.


Božović, P., Erjavec, T., Tiedemann, J., Ljubešić, N., & Gorjanc. V. (2018). Opus-MontenegrinSubs 1.0: First electronic corpus of the Montenegrin language. In Proceedings of the Conference on Language Technologies & Digital Humanities 2018 (pp. 24-28).

A comparison chart of OPUS parallel corpora

The table shows the number of aligned tokens (in millions) for each pair of languages. Click to enlarge.

Tools to work with the OPUS parallel corpora

A complete set of Sketch Engine tools is available to work with OPUS parallel corpora to generate:

  • word sketch – collocations categorized by grammatical relations (this function requires part-of-speech tagging)
  • thesaurus – synonyms and similar words for every word (this function requires part-of-speech tagging)
  • word lists – lists of nouns, verbs, adjectives etc. organized by frequency
  • n-grams – frequency list of multi-word units
  • concordance – examples in context

Please cite the following article if you use any part of the corpus in your own work:

Jörg Tiedemann, 2012, Parallel Data, Tools and Interfaces in OPUS. [pdf] In Proceedings of the 8th International Conference on Language Resources and Evaluation (LREC’2012).

version 2 (m:n mapping)

  • texts can be aligned differently, it means one token with zero/one/more token(s) as necessary

version 1 (1:1 mapping)

  • strictly alignment one token with one token

Search the OPUS parallel corpus

Sketch Engine offers a range of tools to work with the OPUS parallel corpus.



Create your own multilingual or parallel corpora in Sketch Engine.

See our user guide.

Use Sketch Engine in minutes

Generating collocations, frequency lists, examples in contexts, n-grams or extracting terms is easy with Sketch Engine. Use our Quick Start Guide to learn it in minutes.