Teaching AI to Contextualize the Law: AI-Generated Definitions from UK Statutes and Case Law

This paper investigates how Large Language Models (LLMs) can automatically generate concise, context-neutral definitions of key legal terms grounded in both statutory text and judicial interpretation. The work presented here builds on our previous study [11], which combines bottom-up neural inference with top-down, expert-curated LegalDocML data from The National Archives ( https://www.nationalarchives.gov.uk ), the UK Publication Office, to link key legal terms in UK legislation to the case-law paragraphs in which they are interpreted. In this paper, we extend the pipeline introduced in [11] by adding an automated definition-generation step. This extension was motivated by discussions with legal experts and collaborators at The National Archives, who observed that interpreting the extracted term–paragraph pairs requires mentally synthesizing definitions from judicial reasoning, a process that is time-consuming and cognitively demanding. To address this challenge, we propose an LLM-based methodology that integrates ensemble voting and adjudication with chain-of-thought prompting to generate precise, context-neutral definitions that reflect judicial reasoning while minimizing the influence of case-specific factual backgrounds. By automating the synthesis of judicial interpretations into coherent definitions, our approach supports the harmonization of statutory language across legislative and judicial sources, facilitates more consistent application of legislation to case law, and contributes toward a fully integrated LegalTech system that reduces cognitive burden for legal practitioners while leveraging the official UK legislation and case-law datasets published by The National Archives.