A more recent version of the text corpus is published here: https://github.com/GermanT5/wikipedia2corpus

This is a German text corpus from Wikipedia. It is cleaned, preprocessed and sentence splitted. Its purpose is to train NLP embeddings like fastText or ELMo Deep contextualized word representations.

The advantage of this text corpus is that it does not only contain the article space of the wiki, but also the comments for a larger text corpus and a more sloppy language. This should improve the quality of downstream tasks when you process conversations like mails, chats, tweets or support tickets.

We used a wikipedia dump as the data source.

• They can be downloaded here: https://dumps.wikimedia.org/
• Mirror here: https://dumps.wikimedia.org/mirrors.html
• dewiki-20181001-pages-meta-current.xml.bz2 was used because this dump also contains the comments

Then the tool WikiExtractor was used to extract the xml dump. To also include the discussion, the WikiExtractor tool has been modified:

def keepPage(ns, page):
    if ns != '0' and ns != '1': # Aritcle and Talk
        print('skipped ns:', ns)
        return False
    # remove disambig pages if desired
    if options.filter_disambig_pages:
        for line in page:
            if filter_disambig_page_pattern.match(line):
                return False
    return True

Now some hand-crafted python tool was used for further processing: https://github.com/PhilipMay/de-wiki-text-corpus-tools/blob/master/process_wiki_files.py

• SoMaJo was used for tokenization and sentence splitting
• spaCy and gensim also have been tested for tokenization and sentence splitting but have not been as good as SoMaJo for German language
• article headlines and some markup was removed

Everything has been shuffled on sentence-level with linux shuf command.

You can download the texts here:

• wiki-all-shuf.tgz.part-00
  • MD5: 9cd27b9a22ee4de391435b4bcbb30428
  • SHA1: 66ccc99ccfeb4b546f9c888af9b23e5fc1a67236
• wiki-all-shuf.tgz.part-01
  • MD5: bf187bdda21ea9f7af1ecdf085ca54d5
  • SHA1: 73749be0285a6e359dead08acf65b33e9a55c9b4
• wiki-all-shuf.tgz.part-02
  • MD5: b887df79f54d7d36d3da22e5e6f8add1
  • SHA1: f9c773821bee112b976ae5247ac55ffdba6e20f7
• wiki-all-shuf.tgz
  • MD5: 51ddcca730dca6e48c29d6339c2059f9
  • SHA1: f1c7ef0245abca47d3be2657ac4c345a3dc8d121

Using these commands, you can unpack the files (Linux and macOS):

cat wiki-all-shuf.tgz.part-* > wiki-all-shuf.tgz
tar xvfz wiki-all-shuf.tgz

As Wikipedia itself, this is published under Creative Commons Attribution-ShareAlike 3.0 Unported license.