The summarization of legal texts: an experimentation with GPT-3

Authors

  • Manola Cherubini
  • Francesco Romano
  • Andrea Bolioli
  • Nazareno De Francesco
  • Irene Benedetto

DOI:

https://doi.org/10.32091/RIID0103

Keywords:

Legal informatics, Artificial intelligence, GPT-3, Summarization, Institutional texts

Abstract

The purpose of this paper is to provide a brief overview of the BERT and GPT-3 linguistic models, and to explore the potential of using artificial intelligence to enhance the ability to communicate legal information, specifically information from legislators and officials responsible for drafting administrative texts. Further, the GPT-3 model was tested on a diverse set of administrative acts and regulations at various levels, in order to assess its ability to summarize the content of these documents. The authors, who are experts in computational linguistics and jurists, present a detailed description of the use case that was designed, the system outputs, and the evaluations conducted.

The texts used for the experimentation are available in a file downloadable from the Journal's website.

Author Biographies

  • Manola Cherubini

    Researcher at the Institute of Legal Informatics and Judicial Systems of the Italian National Research Council (IGSG/CNR)

  • Francesco Romano

    Senior Researcher at the Institute of Legal Informatics and Judicial Systems of the Italian National Research Council (IGSG/CNR)

  • Andrea Bolioli

    Researcher at MAIZE srl

  • Nazareno De Francesco

    Data scientist at MAIZE srl

  • Irene Benedetto

    PhD student at the Politecnico di Torino

Downloads

Published

2023-03-24

Issue

Volume 5 , Issue 1 (2023)

Section

Systems and applications

How to Cite

[1]
Cherubini, M. et al. 2023. The summarization of legal texts: an experimentation with GPT-3. Rivista italiana di informatica e diritto. 5, 1 (Mar. 2023), 191–204. DOI:https://doi.org/10.32091/RIID0103.

Most read articles by the same author(s)