Extracting structured knowledge from Dutch legal texts: a rule-based approach

Abstract

Legal texts are difficult to interpret, and its interpretation depends on the knowledge and experience ofthe legal expert. Formalising interpretations can improve transparency. However, creating formalisationsof legal texts is labour-intensive, and automatically creating them is still a challenge. Previous workshowed that rule-based systems have mixed success on Dutch legal texts. They use complex rule systemsfor specific cases, making them hard to compare. Because of the lack of analysis, the success of thesemethods is also unclear. In this paper, we propose a new rule-based architecture for detecting thedifferent roles of Flint frames, a knowledge representation language which aims to be a generic and lesstask-dependent language. The rules in this architecture are based on Part-of-Speech tags and universaldependency tags. Our analysis shows that this combination yields more precise extraction of the roles ofFlint frames than previous methods, and the use of universal dependency tags allows this method to alsobe applied to other languages. For further improvement we suggest extending the rules for extractingthe recipient role, add rules for recognising complex relative clauses, and testing this framework onEnglish legal texts.