Speaker
Description
The emergence of artificial intelligence represents the latest in a series of technological challenges that have periodically reshaped the relationship between human expertise and technological tools. This paper takes as its point of departure the conviction that AI cannot exist, function, or evolve independently of human intelligence; lexicography, as a discipline situated at the intersection of linguistic knowledge, cultural competence, and technological practice, is particularly well placed to explore how human expertise can guide AI systems rather than be displaced by them. Drawing on the resources of the Lessico dei Beni Culturali (LBC) project, a multilingual lexicographic initiative at the University of Florence covering Italian cultural heritage terminology across Italian and seven further languages, the paper investigates how human-curated reference materials and expert-designed prompts can systematically improve AI performance on culturally embedded art terms. Three representative lemmas (cartone, grazia, and maniera) were tested across three conditions, namely baseline, expert-prompted, and LBC-constrained, using three major AI systems (Claude Sonnet 4.6, Gemini 3.5 Flash, and GPT-5). The results confirm a consistent gradient: corpus evidence not only improves output quality but, in some cases, reveals lexicographic realities that expert prompting actively conceals, most strikingly the absence of a stable specialised equivalent for maniera in Spanish, which the LBC corpus exposes as a structural lexical gap rather than a straightforward translational choice. The study argues that the value of expert lexicographic knowledge does not diminish in AI-assisted workflows but becomes more concentrated, shifting from the direct production of entries to the design of the knowledge infrastructure within which AI systems can operate effectively.