The GGPONC project aims to provide a freely distributable corpus of German medical text for NLP researchers.
Clinical guidelines are particularly suitable to create such corpora, as they contain no protected health information
(PHI), which distinguishes them from other kinds of medical text.
The second version of the corpus (GGPONC 2.0) consists of 30 German oncology guidelines with 1.87 million tokens.
It has been completely manually annotated on the entity level by 7 medical students using the INCEpTION platform over a
time frame of 6 months in more than 1200 hours of work. This makes GGPONC 2.0 the largest annotated, freely
distributable corpus of German medical text at the moment.
Annotated entities are Findings (Diagnosis / Pathology, Other Finding), Substances (Clinical Drug, Nutrients / Body
Substances, External Substances) and Procedures (Therapeutic, Diagnostic), as well as Specifications for these entities.
In total, annotators have created more than 200000 entity annotations. In addition, fragment relationships have been
annotated to explicitly indicate elliptical coordinated noun phrases, a common phenomenon in German text.
