A peculiarity of Welsh and the other Celtic languages is their system of Initial Mutations. These are regular alternations of word-initial phonemes triggered by a variety of lexical and syntactic triggering contexts. This feature of the Celtic languages poses a number of challenges to grammatical description, not least because it requires direct reference to adjacency relations in the linear string. We describe here an approach which covers the full range of mutation processes and their distribution in Welsh using the XLE grammar development environment and the associated finite state and tokenisations tools.