Download, parse, and filter the European Parliament Proceedings, data-ready for The-Pile.
The Europarl parallel corpus is extracted from the proceedings of the European Parliament. It includes versions in 21 European languages: Romanic (French, Italian, Spanish, Portuguese, Romanian), Germanic (English, Dutch, German, Danish, Swedish), Slavik (Bulgarian, Czech, Polish, Slovak, Slovene), Finni-Ugric (Finnish, Hungarian, Estonian), Baltic (Latvian, Lithuanian), and Greek.
To use this parser, first download the source file
http://www.statmt.org/europarl/v7/europarl.tgz
and unpack it to the directory. The parser will look for all file within the txt
subdirectory. Note that the download is slow and make take 12 or more hours.
The parser removes all basic tag information and only retains the name. The tag
<SPEAKER ID=77 LANGUAGE="NL" NAME="Pronk">
Is reduced to
Pronk
Extremely short files (<200 chracters) are removed as they did not contain useful language modeling text. A single file txt/pl/ep-09-10-22-009.txt
fails to open with UTF-8 encoding and is skipped. No other filtering was done.
Data souce temporary hosted at https://drive.google.com/file/d/12Q23Y7IKQyjF28xH0Aw6yZaYEx2YIOiB/view?usp=sharing