Danish Foundation Models – Empowering the Danish Language in the Digital Age

The Danish Foundation Models (DFM) is a cooperative national initiative aimed at developing high-quality, open-access AI foundation models specifically tailored for the Danish language. In collaboration with leading academic and industrial partners, DFM is building tools to support the pre-training, fine-tuning, and evaluation of advanced Danish language models for both text and speech technologies. This initiative addresses the growing risk that smaller languages like Danish may be underrepresented in the global AI landscape, which might negatively impact future application of AI in for example healthcare and the public sector. 

DFM promotes a sustainable, inclusive, and transparent approach to AI development by emphasizing data privacy, ethical use, and open-documentation. It seeks to empower public institutions, researchers, and private companies by ensuring that Danish language technologies meet sector-specific demands and are usable across a wide array of domains from public administration and healthcare to education and commercial enterprises.

Role of Center for Humanities Computing  

The DFM project is co-led by the Center for Humanities Computing in collaboration with the University of Southern Denmark, the University of Copenhagen, and the Alexandra Institute. A core team of infrastructure and AI specialists from these institutions contribute to the project.

CHC is responsible for:

• Developing state-of-the-art Danish language models using ethical training regimes
• Designing and implementing secure pipelines for pre-training and evaluation
• Ensuring robust model documentation through tools such as model cards and datasheets
• Contributing to benchmarks and datasets that support fine-grained evaluation of Danish NLP capabilities
• Building tools and applications that enhance community-driven use-case development and reproducibility