Orthographic epenthesis and vowel deletion in Welsh

In this paper, we examine the distribution of epenthesis in final clusters and initial syllable deletion in trisyllabic words in Welsh using a corpus of Twitter data (Jones et al. 2015). We show that the generalisations established in Hannahs 2009, Hannahs 2011, and Hannahs 2013 are largely borne out, but there are additional lexical and phonological complications. Specifically, we show that these two processes are subject to lexical frequency effects that go in opposite directions. While this seems at first paradoxical, we go on to show that the frequency effects make sense given what we know about phonological processes generally and what we know about Welsh phonotactics specifically. The organization of this paper is as follows. We first review Hannahs's foot-based account of the facts. We then turn to our Twitter data testing Hannahs's claims, but also considering additional variables. We show that: i) the phenomena are gradient; and ii) that they are subject to lexical frequency effects. We then argue that these effects are, in fact, to be expected and we justify that claim by looking at further data from another corpus. In the first section we review the analysis of final clusters.