Final devoicing and vowel lengthening in the north of Italy: a representational approach

In this paper I present a novel approach to the phenomenon of length-related “final devoicing” in the Romance varieties of northern Italy. I propose a representational account of the facts, deriving vowel lengthening from the interplay of constraints regulating word-final delaryngealization and weight-by-position constraints targeting only laryngeally specified obstruents. As Repetti (1992) shows, the facts are broadly similar for a range of Romance varieties in the north of Italy. I concentrate on (Central) Friulian here, since it seems to be the best described variety (Francescato 1966; Vanelli 1979; Hualde 1990; Baroni & Vanelli 2000; Finco 2009). In Friulian, stressed vowels are lengthened in final syllables before underlyingly voiced consonants (2), despite the general absence of stress-to-weight effects. Importantly, Friulian has an independent quantity contrast (as in [mil] ‘thousand’∼ [mi:l] ‘honey’), so this is clearly a phonological phenomenon rather than the much-discussed phonetic recoverability effect (cf. van Oostendorp 2008 for an overview). (1) a. ["lat] ‘milk’ b. [la"ta] ‘to breastfeed’ (2) a. ["la:t] ‘gone (masc. sg.)’ b. ["lade] ‘gone (fem. sg.)’ At first blush, this seems to be a classic instance of counterbleeding opacity leading to overapplication of vowel lengthening, and thus should present problems for a (parallel) Optimality Theory account. Within a derivational theory, on the other hand, a relatively straightforward account is possible: coda voiced obstruents become moraic at an intermediate level of representation; the mora is delinked following final devoicing, but then reassociated to the preceding vowel, giving rise to lengthening (Hualde 1990; for a different approach see Repetti 1992). However, Baroni & Vanelli (2000) show that “final devoicing” does not lead to complete neutralization, at least on the phonetic level: among many other things, devoiced obstruents are shorter than true voiceless obstruents. Devoiced obstruents also differ from voiced obstruents in terms of both their own duration and of the degree of lengthening they impose on preceding vowels. I conclude, therefore, that rather than being a textbook [+voice]→ [−voice] change, Friulian final devoicing involves a phonological ternary contrast, similarly to at least one other language where final devoicing interacts with vowel length, namely (Trégorrois) Breton (Jackson 1960). I propose a representational account of these facts. Specifically, I argue that final devoicing in Friulian is a case of delaryngealization, i. e. deletion of the Lar node (Mascaró 1987; Steriade 1997). I adopt the following (broadly familiar) specifications: