KPWr is a collection of text documents available under a Creative Commons license. Documents have been tagged with the wcrft2 tool and described with different types of information such as syntactic phrases (chunks), relations between syntactic phrases, identification units (including relations between them and lemmatization), unified word meanings, spatial expressions, verbs with a default subject, text keywords, temporal expressions (locally and globally normalized), situations, semantic roles and correlations. Detailed statistics can be found on the page: http://nlp.pwr.wroc.pl/narzedzia-i-zasoby/zasoby/kpwr Each document is saved in three files that contain the following information:
*.xml (CCL file) – contains tokenization, sentence division, morphological analysis of the text, annotations and lemons,
*.rel.xml (CCL-REL file) – contains relations between annotations,
*.ini (INI file) – contains document metadata.
Additionally, the body can be exported in conll, txt and json format.
Samples for the corpus have been taken from sources such as: Wikipedia, Wikinews, information portals with content under Creative Commons license, literary works from the public domain or made available under an open license, etc., which provide legal and free use of the corpus.
The Corpus in its latest published version consists of 449,985 tokens, but is constantly being expanded and developed towards a balanced corpus, containing equally scientific, official, artistic/rethorical, press/publicist and colloquial texts.