Abstract
Democratic participation and social movement engagement in sustainability transitions depend on how well diverse actors can access, understand and mobilize knowledge. However, terminology, especially when marked by regional variation, often creates barriers for communities, activists and civil society organizations seeking to engage with policy debates or academic frameworks. At the same time, emerging large language models (LLMs) such as ChatGPT shape how knowledge is translated and circulated, raising questions about their role in inclusive communication. Using the example of German-language education terminology, the UniTermGPT project examines how LLM-supported translation can be leveraged to improve knowledge accessibility and support the co-production of transformative knowledge between researchers and practitioners.
A corpus of Austrian, German, Swiss and South Tyrolean university texts in German and English from at least 15 universities will be compiled to identify region-specific terminology relevant for navigating institutional and governance contexts. The terminology will be compared with existing resources and then used to design prompts for LLM-assisted translation experiments in German and English. These translations will be annotated by practitioners to assess accuracy and sensitivity to linguistic diversity. Qualitative analysis will explore how these outputs may support or hinder democratic knowledge practices and collaborative engagement with societal actors.
Preliminary findings show that large language models tend to homogenize linguistic variation, potentially obscuring distinctions that matter for democratic participation and for communities interacting with institutional systems. In future steps, targeted prompt design and retrieval-augmented generation are likely to improve LLM-generated translations. While not direct outcomes, the results are anticipated to shed light on factors that may enable LLMs to facilitate knowledge co-production, improve communication across groups and support actors in navigating policy and governance arenas.
The project’s recommendations for specialized translation and a policy brief (drawing on the results of the qualitative analysis) serve to illuminate the societal relevance of terminology in an era marked by expanding LLM use. The insights gained through UniTermGPT are informative for other contexts where region-specific terminology shapes communication and translation. The conclusions will outline potential implications and considerations for activists, policymakers and researchers integrating AI-assisted translation into strategies for more inclusive and just sustainability transitions.