Large Language Models for Text Reuse: Identifying and Studying the Success of Legislative Amendments in Spain.

Comparing legislative texts (or version so the same text at different stages) is central to the many political and legal studies. Computational text-reuse methods allow researchers to learn from large amounts of legislative text. Yet, existing text-reuse methods are limited in scope and accuracy. For example, text reuse methods fail at measuring the success of “deletion-only” amendments that state the article or number to be deleted, without specifically mentioning the actual amended text. Recent advancements in Large Language Models (LLMs), however, provide new opportunities to overcome these challenges by 1) improving the text-reuse workflow and 2) improving measurement accuracy. We develop an LLM-based approach to identify and study the success of legislative amendments. We use a novel dataset of all legislative amendments introduced to the Spanish parliament between 1996 and 2019 (N ~ 93,000), including a validation hand-coded dataset of over 5,000 amendments.