The Summarization of Italian Tax-Law Decisions: The Case of the PRODIGIT Project

This paper presents an innovative approach to the automated summarization of tax-law decisions developed within the PRODIGIT project. The work addresses the growing challenge of managing large volumes of judicial rulings, which often hinder access to relevant legal precedents. We introduce a hybrid methodology that combines extractive and abstractive summarization techniques, with a particular emphasis on the application of large language models (LLMs) for abstractive summarization. Our approach includes designing and evaluating various prompt-based configurations, leading to the development of a ‘‘combined summarization’’ method, where distinct summary components are generated and integrated into a cohesive text. Experimental results, validated by tax law experts, demonstrate that this method yields the most comprehensive and accurate summaries. Additionally, we explore the integration of these summaries into semantic search functions, enabling users to retrieve and comprehend relevant case law efficiently. Our findings highlight the potential of AI-driven summarization tools to enhance legal transparency, promote judicial consistency, and support the work of judges, lawyers, and legal scholars.