Abstract
As AI-driven language tools enter the workflows of many professions, including specialised translators, questions arise not only about technological performance but also about the social dynamics surrounding their development and evaluation. Against this background, the UniTermGPT (University Terminology in German in the Age of ChatGPT) project investigates how ChatGPT handles university terminology across different German varieties; an area where terminological variation is high and where large language models (LLMs) often fall short. Additionally, UniTermGPT foregrounds the social and ethical implications as well as solidarity aspects of this technological shift. Methodologically, UniTermGPT adopts a community-informed workflow that combines corpus-based research with collaborative analysis of ChatGPT translations, co-developing annotation guidelines, practitioner recommendations and a policy brief with university translators and terminologists. Engaging practitioners as co-creators rather than merely data providers is central to this approach. The project also reflects on solidarity by examining research conducted from a dual and hybrid positionality: as a trained translator embedded in the professional community, and as a researcher analysing the translation practice through a critical lens. When engaging with professional communities, the researcher is sometimes regarded as a peer whose work embodies research solidarity (someone who genuinely seeks to support the community). Yet this proximity can also raise questions about authority and neutrality. Conversely, when the researcher is viewed primarily in her academic role, participants sometimes fear that the project seeks to replace human expertise with LLMs (casting the researcher in the role of an unwanted intermediary or even an adversary). UniTermGPT therefore unfolds within a liminal space where the researcher must negotiate expectations and ethical responsibilities while relying on the time and goodwill of practitioners who themselves face heavy workloads and often precarious working conditions. For future research and practice, UniTermGPT underscores the importance of transparent, cocreated workflows and sensitivity to experts’ concerns in studies in LLMs and translation. The project’s open datasets, practitioner recommendations and policy brief highlight the societal value of language diversity and the need for digital tools that respect the lived realities of language professionals.