Michael Hahn

Michael Hahn is a Tenure-Track Professor (W2) at Saarland University, based at the Saarland Informatics Campus. He directs the Language, Computation, and Cognition Lab (LaCoCo) and holds joint affiliations with the Department of Language Science and Technology and the Department of Computer Science. He earned his PhD from Stanford University in 2022 under Judith Degen and Dan Jurafsky.

His research lies at the intersection of machine learning, cognitive science, and linguistics. Key areas include the expressive power and learning biases of transformer models and state space models, mechanistic interpretability of neural networks, and computational modeling of human cognition and perception. He investigates how information processing constraints shape both artificial and biological systems.

The recent publications demonstrate a consistent focus on understanding the theoretical and cognitive foundations of language models, particularly transformers. Themes include length generalization, reasoning limitations, in-context learning mechanisms, and the cognitive plausibility of model behaviors. His work spans from formal mathematical analysis to cognitive modeling and large-scale empirical evaluation.

His scientific recognition includes:

• Best Paper Award at ACL 2024
• Second Prize at ICML 2024 Mechanistic Interpretability Workshop
• Outstanding Area Chair at EMNLP 2024
• ELLIS Member
• ELIZA Fellow

Michael Hahn actively advises a team of PhD students and researchers, including Xinting Huang, Yana Veitsman, and Mark Rofin. He is involved in organizing major academic events, such as a NeurIPS 2025 workshop on transformer capabilities. His lab receives significant attention, as evidenced by organizing the ELLIS pre-NeurIPS poster session with 17 posters from across campus.

The Language, Computation, and Cognition Lab (LaCoCo) is a vibrant research group investigating machine learning, language, and the human mind. The team includes postdoctoral researchers, PhD students, and research assistants working on interpretability, cognitive modeling, and NLP theory.