Home | Publications | Hof23

Computational Investigations of Derivational Morphology

MCML Authors

Valentin Hofmann

Dr.

* Former Member

→ Group Hinrich Schütze
Computational Linguistics

Abstract

This dissertation revisits the long-standing view that derivational morphology is hard to predict, showing that recent advances in NLP enable much stronger predictive models than previously assumed. By applying modern computational methods to large, socially stratified datasets, it demonstrates that derivational processes are more systematic and predictable. The work further shows that standard tokenization harms morphological prediction and proposes morphology-aware tokenization strategies that improve performance, highlighting the continued importance of linguistic insights in modern NLP. (Shortened.)

phdthesis Hof23