26th Manchester Phonology Meeting, University of Manchester, Manchester, UK.
The dual behaviour of /v/ has been a notorious crux in Russian phonology since at least Jakobson (1956). The basic problem is that /v/ fails to trigger voicing assimilation, like sonorants, but undergoes both assimilation to following obstruents and final devoicing, like obstruents.
The commonest solution in the literature is representational: [v] is treated as a sonorant, either an underlying /w/ (Lightner 1972, Hayes 1984, Kiparsky 1985) or some labial non-glide (Coats & Harshenin 1971, Padgett 2002). Most recently, Reiss (2017) analyses /v/ as /V/, underspecified for [±voice] – patterning with sonorants, but not a sonorant in featural terms – that is devoiced to [f]. He argues that this is an insurmountable challenge to the Contrastivist Hypothesis (Hall 2007), since [v] and [f] are minimally distinct in terms of voicing, but do not pattern phonologically like other similar pairs in Russian.
The argument rests on two premises: Russian /v/ lacks a voicing specification, and is thus unable to trigger assimilation (cf. Kiparsky 1985, Hall 2004), but becomes phonologically voiceless in the course of the derivation. In this paper I argue that the first premise is correct, but there is no evidence that the latter process is phonological; instead, there is every reason to treat all (de)voicing patterns affecting Russian /v/ as belonging to the phonetic component, just as is the case for sonorants.
Under this analysis, the apparent serial interaction of the obstruentization of /v/ and its devoicing follows from the feed-forward architecture of grammar, not from ordering within the phonology (for a related analysis, see Knyazev 2004, 2006). I argue that Russian /v/ is laryngeally underspecified not just in the input to phonology but also in its output; it is in no sense the voiced counterpart of /f/, which joins the better known [t͡s], [t͡ʃʲ], and [x] in lacking such a counterpart (cf. already Kiparsky 1985). If the patterns of /v/ (de)voicing are phonetic, then a whole host of facts cease to be problematic for phonological analysis, including the famous variability in the behaviour of devoiced /v/ as an assimilation trigger ([jazf] ∼ [jasf] for /jazv/ ‘sore.GEN.PL’; Reformatskiĭ 1975); the behaviour of [v] and sonorants in voicing assimilation (Kulikov 2013); the fact that some (de)voicing patterns are specific to [v], such as the minority but robust pattern of progressive assimilation in [v#v] sequences (Vorontsova 2007); and the sensitivity of sonorant and /v/ (de)voicing to prosodic boundary strength.
The dissociation between laryngeal assimilation in obstruents and in /v/ is also expected to follow from the life cycle of phonological processes (Bermúdez-Otero 2015): historically, the former predates the latter by several centuries, and there is ample dialect evidence showing that the devoicing of /v/, even where it did happen, did not always result in [f], [x] being a frequent outcome (e. g. Kasatkin 2005, Galinskaya 2008). The proposed analysis is reminiscent of ‘rule scattering’ (Zsiga 2000, Bermúdez- Otero 2015), with both phonological and phonetic patterns of (de)voicing found in the same language; in fact, I argue that phonetic /v/ (de)voicing coexists with another phonetic pattern of obstruent devoicing and assimilation – the gradient congener of phonological assimilation rules arising via rule scattering.
Ultimately, many aspects of the behaviour of Russian [v] may have a phonetic rationale, at least historically (albeit possibly not synchronically, as argued by Bjorndahl 2015). However, in this paper I argue that the best analysis of the problem of /v/ is a substance-free one that does not encode these explanations in the phonological grammar, contra e. g. Padgett (2002), but also respects the privileged role of contrast in phonological representation, contra Reiss (2017).