Classification of German court rulings: Detecting the area of law

This paper investigates on the feasibility of automatically detecting the legal area of court rulings. Hereby, we establish the hypothesis that the allocation to a field of law is often ambiguous and errors occur in that process as a result. A dataset constituting over 9.000 labelled court rulings was used in order to train different machine learning (ML) classifiers. Additionally, we applied rule-based approaches utilizing domain knowledge of legal experts. Our models outperformed the rule-based approaches significantly. Hence, we could show that the performance of ML models are less prone to errors than the manual assignment of legal experts.