Towards Clear, Simple, and Inclusive Legal Language — Exploring Generative Artificial Intelligence (GAI) models as content rephrasing assistants in multilingual settings

This study explores the potential for GAI language models and Natural Language Processing (NLP) tools to be used as writing aids in the transformation of expert content into clear, simple and inclusive language. In particular, we focus on the feasibility of fully automating the reformulation of legal texts, and the challenges of setting standards for automatically generating content in multilingual settings. While automatic assistants can already produce effective summaries and simplified versions of complex legal texts, we suggest that they lack the functional awareness and metalinguistic reflexivity that are particular to human review and revision. We illustrate this by examining a series of authentic examples. First, we look at tools like ChatGPT which can be configured to summa-rise complex legal documents and generate texts aligned using the guidelines for Plain Language (PL) and Easy-to-Read (E2R). We give examples of linguistic features, outputs from NLP models, and GAI-generated content evaluation. Our initial observations suggest that while automatic assistants can produce greatly simplified texts, they are unable to take into account legal implications and contextual reasoning. Then we look at the issue of prescriptive guidelines in the light of the new paradigm of content management. Com-paring examples of official discourse and their equivalents ‘translated’ into E2R, we find that the reformulations involved reflect the very different functional requirements of such radically simplified texts. This observation leads us to stress the importance of human oversight in auto-generated content, as well as the need for guidelines, integrated revision tools, transparency, user training, and continuous monitoring.