Summarization of tax rulings in the PRODIGIT project
DOI:
https://doi.org/10.6092/issn.1825-1927/19618Keywords:
Tax law, artificial intelligence, Abstractive summarisation, Extractive summarisation, Large Language ModelsAbstract
This contribution illustrates the experimental summarisation of tax-law decisions conducted within the PRODIGIT project. After a brief introduction on the concepts of summarisation and "massimazione" (the authoritative extraction of judicial rulings by the Court of Cassation in the Italian legal system), we present the methodology adopted in the project. This methodology includes the experimentation and comparative evaluation of techniques for extractive and abstractive summarization. We then describe the results obtained with these techniques, focusing on abstractive summarization using large language models, the solution that delivered the best results. The different combinations of commands tested (prompt) and the related results are presented. The evaluation by tax experts shows that the most correct and complete summaries are provided using the method of "combined summarisation", which requires the generation of the different contents of the summary and their combination into a single text. Finally, the use of summaries for semantic search functions is presented.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2024 Giuseppe Pisano, Alessia Fidelangeli, Federico Galli, Andrea Loreggia, Riccardo Rovatti, Piera Santin, Giovanni Sartor
![Creative Commons License](http://i.creativecommons.org/l/by/4.0/88x31.png)
This work is licensed under a Creative Commons Attribution 4.0 International License.