On Monday, October 5, 2026, Dr. Esteban Garcés Arias successfully defended his dissertation titled “Neural Text Generation under Uncertainty: Decoding, Evaluation and Low-Resource Applications” at LMU Munich. The dissertation developed adaptive decoding methods to balance coherence and diversity, statistical approaches to evaluating multiple dimensions of text quality, and methods for learning from scarce data. A central theme was the variability of human text production and its implications for generating and evaluating text. His work connects methodological research in statistics and natural language processing with applications spanning open-ended text generation, handwritten text recognition in historical documents, and part-of-speech tagging for medieval Romance languages. He was supervised by Prof. Dr. Christian Heumann and Dr. Matthias Aßenmacher at the Department of Statistics, LMU Munich.
He is currently a postdoctoral researcher at MaiNLP, LMU Munich, under the supervision of Prof. Dr. Barbara Plank, focusing on generative AI for low-resource languages.