Extracting Computational Logic from Legal Text: A Decision Support Approach for Public Sector Automation
Simon Price, Markus Bertl · 2024
This research presents first steps towards a generic approach to automatically translating legal text into machineexecutable computational logic.We demonstrate how this approach can be used to automate public sector processes.Since automation of legal processes is a high-risk application of AI, we use explainable AI based on natural language processing using scope-restricted pattern matching and grammatical parsing.Our approach consists of document structure inference from the raw legal text, semantically neutral pre-processing, recognition of internal and external references, target resolution for internal references, paragraph contextualization and, finally, rule extraction.Extracted rules are converted to Prolog predicates and visualized as textual lists and graphical decision trees.Our developed Law as Code prototype has been evaluated as a proof-of-concept at the Austrian Ministry of Finance and successfully demonstrated the automatic extraction of explainable rules from the Austrian Study Funding Act.This validates our approach and suggests promising future research directions, most notably the prospect of integrating GenAI Large Language Models (LLMs) into the rule extraction process, while retaining provenance and explainability.