Institute of Formal and Applied Linguistics

Understanding legal text is crucial and yet it often poses problems, especially to anyone without legal training. The main culprit is legalese, a style characteristic for legal texts. It includes long sentences and complex syntax, ambiguous, abstract, archaic, or redundant expressions, terminology, and more. We present two systems that try to combat these issues and increase readability of legal texts. The first one is PONK, a rule-based system that highlights problematic segments of text and signals them to the user. The other system is called SPRINT and it combines rules inspired by PONK and artificial intelligence. We also present the results of an experiment that show how large language models (LLMs) work with linguistic rules and how careful prompt engineering can overcome perceived obstacles.