Elevating Legal Understanding: A Dedicated Instruction Dataset and an Optimized LLM for Legal Information Extraction to Support the Juridical Interpretation of Marine Environmental Law

In light of the escalating environmental threats, recent actions have emphasized marine environmental law, as a powerful tool to protect the health of marine ecosystems, preserve biodiversity, and the well-being of present and future generations. The increasing availability of legal documents online, opens up the way for innovative Natural Language Processing solutions dedicated to legal text processing and analysis. While there has been a proliferation of law oriented solutions based on large language models recently, there are still gaps in addressing information extraction from French marine environmental law documents. This paper aims to bridge this gap, by proposing an approach that uses prompt based language models to build a low-cost instruction dataset on French marine environmental law which will be used to fine tune two open-source models, LLaMA 3 and Vigogne. Results show that fine-tuned models significantly outperform their competitors such as LLaMA 3. These findings emphasize how our dataset enhances French marine environmental law models, fostering progress in this field.