The 2025 EurAI Artificial Intelligence Dissertation Award is awarded to Jaap Jumelet for his PhD "Finding Structure in Language Models", defended at the University of Amsterdam (The Netherlands).

In the 2025 EurAI Dissertation Award edition, 17 theses from 12 countries were submitted. The jury consisted of 23 EurAI Fellows.

Jumelet's dissertation tackles a foundational question in NLP: whether LLMs truly capture the grammatical structure of language in a (seemingly) human-like way, rather than merely exploiting statistical patterns. By developing new interpretability methods and analysing how linguistic structure emerges inside neural models, the work advances both the theoretical understanding of LLMs and the broader goal of making AI systems more transparent, reliable and linguistically grounded.

This thesis develops several creative new methods, including a technique for testing what models can learn without ever directly seeing certain constructions. Furthermore, the author proposes a rigorous framework for evaluating interpretability methods. The finding that language models show grammatical behaviours remarkably similar to those found in humans is a very important result. Add to this solid methodology, a well-thought-out organisation, cross-disciplinarity and clear writing. The thesis won the 2026 Dissertation Award of the Benelux Association for Artificial Intelligence (BNVKI).


To see other EurAI Dissertation Awardees click here