Length and voicing in Friulian and Milanese Or why rule-free derivations are needed

This paper claims that phonology should express the relationship between vowel length and obstruent voicing operationally rather than in parallel. The empirical focus in on Friulian and Milanese. The distribution of vowel length in Friulian is predictable from the underlying laryngeal specification of obstruents. Stressed vowels are long before underlyingly voiced word-final obstruents although they devoice. This situation creates opacity. In the light of the interaction between vowel lengthening and final devoicing, this paper argues in favor of Harmonic Serialism, a version of Optimality Theory that combines constraint ranking with serial derivations. I demonstrate that only in Harmonic Serialism does vowel length naturally follow from the independent need to satisfy NoVoicedCoda and FootBinarityμ, instead of assuming that vowel lengthening is the consequence of whether or not obstruents project a mora depending on their laryngeal specification. The facts of Milanese are also considered. Milanese patterns as Friulian except that final devoicing is not categorical. I use Serial Variation (Kimper 2011) to solve the problem of opacity and variation in the realization or not of final devoicing. An alternative approach to Milanese that assumes gradual devoicing is also suggested. This analysis combines Harmonic Serialism and turbid output representations (Goldrick 2001, van Oostendorp 2008), and allows us to express formally the difference between Friulian and Milanese in terms of categorical versus gradual devoicing.