European Parliament Proceedings Parallel Corpus 1996-2009
For a detailed description of this corpus, please read:
Europarl: A Parallel Corpus for Statistical Machine Translation, Philipp Koehn, MT Summit 2005, pdf.
Please cite the paper, if you use this corpus in your work.
See also the extended (but earlier) version of the report
(ps,
pdf).
The Europarl parallel corpus is extracted from the proceedings of the European
Parliament. It includes versions in 11 European languages: Romanic (French,
Italian, Spanish, Portuguese), Germanic (English, Dutch, German,
Danish, Swedish), Greek and Finnish.
The goal of the extraction and processing was to generate sentence aligned text
for statistical machine translation systems. For
this purpose we extracted matching items and labeled them with
corresponding document IDs. Using a preprocessor we identified sentence
boundaries. We sentence aligned the data using a tool based on the
Church
and Gale algorithm.
Release v5
On 20 January 2010 we released a further expanded and improved version of the corpus. Previous versions are available here. The corpus is released as a source release with the document files and a sentence aligner, and parallel corpora of language pairs that include English.
Changes since v3 (v4 was only released partially
for WMT 2009)
- added 11/2007 - 10/2009 data, now up to 55 million words per language
- further refined preprocessing, cleaning
All formats contain document (<CHAPTER id>), speaker (<SPEAKER
id name language>),
and paragraph (<P>)
mark-up on a separate line. The data is stored in one file per day, and in
smaller units for newer data.
Some documents have the SPEAKER tag attribute LANGUAGE which indicates what language the original speaker was using.
To use the parallel corpora with tools like GIZA++, you want to:
- tokenize the text (required)
- lowercase the text (recommended)
- strip empty lines and their correspondences (required)
- remove lines with XML-Tags (starting with "<") (required)
Download
- source release (text files with preprocessing tools and sentence aligner), 1616 MB
- tools (preprocessing tools and sentence aligner only), 8.1 KB
- parallel corpus Danish-English, 163 MB, 04/1996-10/2009
- parallel corpus German-English, 164 MB, 04/1996-10/2009
- parallel corpus Greek-English, 120 MB, 04/1996-10/2009
- parallel corpus Spanish-English, 169 MB, 04/1996-10/2009
- parallel corpus Finnish-English, 162 MB, 01/1997-10/2009
- parallel corpus French-English, 176 MB, 04/1996-10/2009
- parallel corpus Italian-English, 170 MB, 04/1996-10/2009
- parallel corpus Dutch-English, 172 MB, 04/1996-10/2009
- parallel corpus Portuguese-English, 172 MB, 04/1996-10/2009
- parallel corpus Swedish-English, 153 MB, 01/1997-10/2009
Size of the Corpus
Sizes for single-language data after tokenizing and removing XML.
| Language | Sentences | Words |
|---|
| Danish | 2,009,958 | 47,305,502 |
| German | 1,822,735 | 44,688,020 |
| Greek | 1,257,518 | - |
| English | 1,891,918 | 50,978,295 |
| Spanish | 1,871,700 | 52,503,808 |
| Finnish | 1,834,727 | 34,106,317 |
| French | 1,904,613 | 55,088,177 |
| Italian | 1,827,091 | 50,161,729 |
| Dutch | 2,054,417 | 50,926,645 |
| Portuguese | 1,849,973 | 51,294,994 |
| Swedish | 1,936,391 | 43,291,692 |
Sizes for parallel corpora after sentence aligning, tokenizing, and removing XML.
| Parallel Corpus (L1-L2) | Sentences | L1 Words | English Words |
|---|
| Danish-English | 1,684,664 | 43,692,760 | 46,282,519 |
| German-English | 1,581,107 | 41,587,670 | 43,848,958 |
| Greek-English | 960,356 | - | 27,468,389 |
| Spanish-English | 1,689,850 | 48,860,242 | 46,843,295 |
| Finnish-English | 1,646,143 | 32,355,142 | 45,136,552 |
| French-English | 1,723,705 | 51,708,806 | 47,915,991 |
| Italian-English | 1,635,140 | 46,380,851 | 47,236,441 |
| Dutch-English | 1,715,710 | 47,477,378 | 47,166,762 |
| Portuguese-English | 1,681,991 | 47,621,552 | 47,000,805 |
| Swedish-English | 1,570,411 | 38,537,243 | 42,810,628 |
Test Sets
Several test sets have been released for the Europarl corpus.
In general, the Q4/2000 portion of the data (2000-10 to 2000-12) should be reserved for
testing. All released test sets have been selected from this quarter. The shared tasks for the 2006 and 2007 ACL Workshops on
Statistical Machine Translation provide test sets from the Europarl corpus.
The original common test set from the Koehn/Och/Marcu ACL 2003 Paper is available in the archives.
Known Bugs
- Some special HTML entities and noisy characters are not removed from the data.
- Some recent Greek data has only parts of transcripts in the files.
Terms of Use
We are not aware of any copyright restrictions of the material.
If you use this data in your research, please contact
pkoehn@inf.ed.ac.uk.
Please let us know if you find problems with the data
or if you want the data for other language pairs.
We recommend using the last quarter of 2000 for testing
(2000-10 until 2000-12) for consistency in reporting
research results on this data.
Acknowledgments
The work was in part supported by the EuroMatrixPlus project funded by the European Commission (7th Framework Programme).