Corpora

 

In the following, we briefly describe the textual corpora used at STR and RXRC for the work described here. The STR corpora have been used for tagger and tagset evaluation, the RXRC corpora for text type evaluation.

• STR: manually tagged corpus of 75,000 tokens
  • 50,000 from Frankfurter Rundschau
    This corpus is the ELSNET/EAGLES DE reference corpus, and is composed of news stories about economy (12,500), politics (12,500), culture (12,500), local events (6,250) and sports (6,250).
  • 25,000 from Stuttgarter Zeitung
    This corpus is STR internal.
• RXRC: automatically tagged and manually corrected corpus of 90,000 tokens.
  • 20,000 reference corpus from SfS Tübingen
  • 30,000 news (Spiegel, Welt, taz, spektrum)
  • 26,000 fairy tales (Andersen, Bechstein, Grimm )
  • 14,000 other (reports, guidelines, ...)