Where Generative AI Fails: Explaining International Standards through an Explanatory Dictionary

Lexicography has always been at the forefront of technology whether it be with the introduction of printing or the use of digital technology. It is thus not surprising that generative artificial intelligence (AI) along with its so-called Large Language Models (LLM) have been quickly put to lexicographical use. This paper sets out to demonstrate that the advantages brought by tools as ChatGPT in rapidly building entries and even setting them in a desired XML format has limitations. Here, we look at how AI can help in building an experimental explanatory dictionary on Corporate Social Responsibility in French. This entails explaining usage from a special language, that of ISO standards, to a non-specialised audience. What we show is that whilst AI can explain the terms, a number of difficulties arise in that the standard has been translated leading to cross-cultural misinterpretations and that it cannot adapt to a targeted audience which is a Discourse community that must rely on mediation through an expert in the language norms of standards and business practice. The solution adopted is to use a corpus-driven approach alongside TLex Tools for management of both the source data and the overall dictionary.