When we listen to spoken language, we recognise words extremely quickly, even though speech is a continuous and constantly changing sound signal. This thesis investigates how people achieve this, with a particular focus on words that consist of multiple meaningful parts, such as un-happi-ness or re-view. While it is well established that such internal word structure plays a role in reading, it has long remained unclear whether and how it also influences the recognition of spoken words. Using large-scale experimental data and computational models, this thesis examines how listeners make use of the internal structure of spoken words. The results suggest that word recognition is neither driven solely by whole words nor exclusively by their component parts. Instead, recognition reflects a gradual interaction in which both the full word and its parts can contribute. In addition, words are recognised more quickly when they belong to a larger “word family” of related words, especially when these related words are similar in both sound and meaning. These findings suggest that spoken word recognition is a dynamic and probabilistic process, in which word structure subtly shapes how evidence accumulates. The thesis also demonstrates how the method of computational modelling can help clarify and test theories about how the human brain recognises spoken language.
Hanno Müller is an AI Engineer at the AI Service Centre Berlin-Brandenburg of the Hasso-Plattner Institute, where he develops open-source, on-premise AI infrastructures, with a focus on natural language processing and speech technologies for public-sector, industry, and research contexts. He collaborates closely with public administrations, start-ups, companies, and academic partners, translating research insights into robust and reproducible systems. He conducted his PhD research in computational linguistics and psycholinguistics at Radboud University Nijmegen and Heinrich-Heine-Universität Düsseldorf. His research investigates spoken word recognition and morphological processing, combining large-scale behavioural experiments with computational modelling. Prior to his doctoral research, Hanno completed Master’s degrees in Cognitive Science (MSc) and German Linguistics (MA). He is actively involved in open science and digital humanities initiatives, regularly contributing to workshops, conference papers, and community projects. His work emphasises digital sovereignty and the sustainable deployment of AI in society.