On Friday, TEXT actively engaged in the Language Technology Conference, organized by the Center for Language Technology (NorS) at the University of Copenhagen in collaboration with the Danish Agency for Digital Government. The event provided a valuable platform for scholarly exchange and interdisciplinary dialogue on pressing challenges in language technology.

As part of the program, Kasper Fyhn presented a research poster based on his PhD project, supported by Charlotte and Ida during their internship at TEXT. In addition, Ross, Mina, and Laurits participated in a series of thought-provoking presentations addressing cultural and dialectal representation in machine learning systems, with particular emphasis on large language models (LLMs).

The conference discussions highlighted several critical research questions:

• How can cultural dimensions be disentangled from linguistic structures in studies of LLMs?
• Where does cultural representation manifest within the complex architecture of self-learned matrix multiplications that underpin these models?
• What methodological limitations arise when applying widely used algorithms to languages whose syntax and morphology diverge significantly from Indo-European norms?

Ross says: "Denmark’s language technology research is thriving right now, and growing every year. There is both room and need for continued research into how language models handle smaller languages like Danish. It was a nice reminder that research needs to slow, mindful of data security, and interdisciplinary to avoid trivializing complex issues.”