Dialectal variation of Italian s-voicing as constraint interaction

Many Italian dialects, such as Lombardian for instance, display intervocalic voicing of the fricative s. In some very restricted environments, intervocalic s-voicing does not apply. When prefixes ending in a vowel are attached to roots starting with an s, this s surfaces as voiceless (a-[s]ociale 'asocial'). However, the final s of prefixes, such as dis-, surfaces as voiced in combination with roots starting in a vowel (as in di[z]-onesto 'dishonest'). At word margins, s is usually voiceless in all varieties of Italian. The Tuscan variety of Italian does not display word-internal intervocalic s-voicing. This variety has contrasts such as ca[s]a 'house' vs. ca[ss]a 'cashier', which are ca[z]a / ca[ss]a, respectively, in most other varieties. Nevertheless, Tuscan has intervocalic s-voicing when prefixes with a final s, as dis-, bis-and others, are attached to vowel-initial stems (Bertinetto 1999). In previous analyses of intervocalic s-voicing (Nespor & Vogel 1986, Kenstowicz 1995, and others) it was claimed that this voicing of prefix-final s is an effect of the same rule or constraint that triggers inter-vocalic s-voicing in other contexts as well. In particular, Kenstowicz (1995) assumes in his Correspondence-theoretic approach that the lack of voicing of root-initial s in combination with a prefix ending in a vowel must be attributed to Output-Output faithfulness of this prefix plus root with the simplex form without the prefix. In this simplex form, the root-initial s is at the word margin and accordingly voiceless. Output-Output Faithfulness enforces this voicelessness in the prefixed form as well, while it does not promote voicelessness in prefix-final s since prefixes do not have independent forms, which can serve as bases for Output-Output faithfulness. The most straightforward way to account for the micro-variation of absence versus presence of inter-vocalic s-voicing in Italian dialects is to assume a different ranking of the intervocalic s-voicing constraint (*VsV) in the grammars of the different dialects. In Lombardian, the constraint *VsV should rank higher in the hierarchy than the markedness constraint against voiced obstruents (*VObs). If we assume the constraint *VsV to be below *VObs in Tuscan, Kenstowicz' analysis makes the wrong prediction: This ranking also rules out the voicing of prefix-final s in forms like di[z]onesto 'dishonest', bi[z]avolo 'great-grandfather' (see tableau ii). I will argue in this paper that the voicing of prefix-final s in Italian is triggered by a different part of the grammar than the constraint against intervocalic voiceless s. This instance of voicing …