Abstract

We introduce a framework for measuring delegation in democratic political institutions using fine-tuned large language models (LLMs). Delegation, the transfer of policymaking authority from principals to agents, is a central but challenging concept to measure across legal systems. Building on past work in rule-based natural language processing (NLP) and machine learning, we develop a unified approach that combines the interpretability of linguistic pattern matching with the adaptability of LLMs. We fine-tune legal-domain LLMs to detect delegation provisions in statutory texts and validate our approach on landmark American legislation. Our results demonstrate that even with limited computational resources, fine-tuned LLMs like LegalBERT can robustly identify delegatory provisions with high accuracy and meaningful alignment to legal language. This work enhances the ability of researchers to systematically study delegation in diverse legal and institutional settings.