Next-gen terminology: Transforming terminology work with large language models

Abstract As large language models (LLMs) are being integrated into the field of terminology, they offer promising support across various stages of the terminology workflow. From term extraction to definition generation and concept system creation, LLMs can be integrated into traditional workflows. This raises the question as to whether LLMs will supplement or replace conventional terminology tools. LLMs differ from traditional tools as they do not rely on a rigid set of instructions and predefined functions, but rather generate output based on pattern recognition in large datasets by means of prompts. However, unlike terminology tools, such as term extraction or corpus analysis tools, LLMs may not provide the same level of accuracy and consistency, particularly in emerging domains. This unpredictability of LLMs poses a challenge in terminology-related tasks, such as extracting candidate terms from texts, drafting definitions and exploring equivalencies across languages. While LLMs can be integrated into the terminology workflow in a similar manner to traditional terminology tools, they may also be used as a ‘co-terminologist’ that terminologists can consult to receive feedback or challenge their own suggestions, as a tool for creating engaging terminology training materials or for processing terminology for use in other fields of action.