Speaker
Description
The increasing integration of large language models (LLMs) into (multilingual) specialized communication is challenging traditional distinctions between dictionaries, glossaries and terminology databases. While these resources differ in structure, purpose, theoretical foundations and user groups, LLMs increasingly access their underlying data as machine-readable ‘knowledge’. This paper examines how approaches ranging from prompt-based glossary integration and retrieval-augmented generation (RAG) to terminology-augmented generation (TAG), API-based access and knowledge-graph-based systems are affecting the relationship between language resources and AI. It argues that reducing rich lexicographic and terminological resources to simple equivalence lists risks losing contextual and conceptual information, whereas structured access can preserve and exploit their descriptive richness. At the same time, resource typologies remain relevant: resource type and metadata provide information about a resource’s theoretical basis, purpose, intended users and epistemic status, particularly when LLMs combine multiple resources that may provide competing information. Therefore, resource type, underlying data and resource function should be considered together. This perspective highlights the continuing importance of lexicographic and terminological resources in AI environments, suggesting an increasing role for lexicographers and terminologists in creating language data and organizing knowledge for users and machines alike.