About
FlemBench is an initiative dedicated to developing a culturally aware and linguistically grounded benchmark for evaluating Large Language Models (LLMs) in Flanders. While most existing benchmarks are built on English‑centric datasets or automatically translated materials, FlemBench focuses explicitly on the unique linguistic, cultural, and societal characteristics of Flemish Dutch.
Our goal is to provide a reliable evaluation framework that captures how well language models understand, interpret, and generate language within the Flemish context—including regional vocabulary, cultural references, pragmatic nuance, and expressive variation. By combining demographic and semantic perspectives with modern NLP evaluation techniques, FlemBench helps ensure that AI systems remain locally relevant, inclusive, and context‑sensitive.
FlemBench supports research, industry, public institutions, and media organizations seeking to develop or deploy language technologies that work accurately and responsibly for users in Flanders. Through high‑quality datasets, transparent evaluation tasks, and a shared testing platform, the project aims to strengthen digital sovereignty, improve AI fairness, and promote the development of trustworthy, culturally aligned AI within the Dutch‑speaking landscape.
Contact
- Veronique Hoste (LT3, Universiteit Gent)
- Walter Daelemans (CliPS, Universiteit Antwerpen)
- Florian Debaene (LT3, Universiteit Gent)