Project

FlemBench: Benchmarking of Flemish Large Language Models

Code
174M03425
Duration
10 November 2025 → 09 November 2026
Promotor-spokesperson
Research disciplines
  • Humanities and the arts
    • Computational linguistics
  • Engineering and technology
    • Audio and speech computing
Keywords
Benchmark Large Language Models (LLMs)
 
Project description

The rise of large language models (LLMs) creates new opportunities for digital innovation while simultaneously raising fundamental questions about cultural representation and language sensitivity. In the development and evaluation of Dutch-language language technology, the Flemish varieties of Dutch remain underrepresented in existing benchmarks, language models, and datasets. FlemBench therefore aims to develop a culture-sensitive benchmark that explicitly incorporates Flemish linguistic and cultural characteristics into the evaluation of language models. Building on international frameworks for culturally inclusive language technology (Adilazuarda et al., 2024), FlemBench operationalizes the Flemish cultural context through demographic and semantic proxies by creating datasets grounded in the Flemish-specific content.  In this way, the project facilitates the development of locally robust and culturally rooted language models for public and private applications in Flanders, in line with the current Flemish AI and media policy.