Dieuwke Hupkes

Dieuwke Hupkes is a Researcher at Meta AI Research and an ELLIS Scholar, focusing on understanding neural networks' ability to process hierarchical structure and generalize compositional patterns. Her work bridges computational linguistics, cognitive science, and artificial intelligence, emphasizing interpretability of language models. She holds a PhD from the University of Amsterdam (ILLC) and has supervised numerous student theses on neural network analysis and language processing. Notable contributions include the Llama 3 models, GenBench generalization taxonomy, and studies on compositional learning in recurrent networks.

Education: PhD in Logic (University of Amsterdam, 2020), MSc in Artificial Intelligence, and earlier studies in Physics. Awards include ELLIS Scholar (2024), CoNLL 2019 Honourable Mention, and MLRC 2022 Best Paper Award. She has taught courses on natural language processing and co-organized workshops on neural network interpretability.

Research interests span generalization in NLP, emergent language structures, and alignment of large language models. Her work frequently employs diagnostic classifiers and benchmarking to probe model behavior, with applications to both theoretical understanding and practical model evaluation.