This paper aims to address the challenge of using artificial intelligence for empirical legal studies. We introduce a new annotated dataset on French marine environmental law dealing with definitions, bans, sanctions, and controls on living (turtles and seabirds) and non-living (plastic bags and straws) subjects. The annotation has been produced by law students and validated by a legal expert. Based on the developed dataset, we train a CamemBERT-based classifier which accurately predicts the class of a given legal article according to the pre-defined classes within the dataset we have created. The proposed training set and the resulting trained model provide a better interpretation and accessibility of legal texts to specialists and the general public, based on findings from legal studies and on natural language processing techniques.