This was a corpuslingusitic comparison study of coreference resolution for German and Dutch across the news and literary genres. The systems tested are sieved-based for Dutch (Cranenburgh, 2019) and mention-ranking CorZu for German (Klenner et al., 2010). The sieved-based system results more robust across genres, while the German one is unsurprisingly biased towards the genre used for its training. The full report can be read in the CoreferenceResolution_paper.pdf of this repository. Since this is a purely Corpuslinguistic study based on a manual analysis, it doesn't contain any code but the Excel and final report. It was for the course Coreference Resolution, a Project Module (12 ECTS) at the University of Potsdam, supervised by Dr. Sharid Loáiciga.
- Mention-ranking "CorZu" for German: https://github.com/dtuggener/CorZu
- Sieve-based "dutchcoref" for Dutch: https://github.com/andreasvc/dutchcoref