Unsupervised Labelling of Mutation Triggers in Welsh

Initial consonant mutation is a key morphological feature of Welsh, posing significant challenges for language learners and NLP systems alike. We introduce mutation trigger labelling as a novel computational task: automatically identifying the grammatical factors responsible for triggering consonant mutations in Welsh text. We evaluate rule-based systems and large language models on this task, providing the first systematic analysis of computational approaches to Welsh mutation triggers. Our work contributes to the broader goal of building more robust NLP tools for Welsh and other Celtic languages.

LREC 2026