CICERO: A GPT2-Based Writing Assistant to Investigate the Effectiveness of Specialized LLMs’ Applications in e-Justice

Does it make sense to develop specialized writing assistants in the era of LLMs - Large Language Models? In this paper, we present CICERO, a specialized writing assistant we have developed for writing (pieces of) sentences in the Italian legal system. Our proposed solution involves fine-tuning a transformer on a pre-processed corpus of Italian civil judgments, resulting in a novel language model that can be deployed as a writing assistant for legal users to improve text writing efficiency. The model can also be further fine-tuned for use in other law-related natural language processing tasks. The experimental validation of CICERO allows us also to draw interesting insights on the meaningful, if any, of developing specialized tools for assisting specific classes of users and knowledge workers, in an era in which we witness the widespread adoption of LLMs.