The Typology of Voicing and Devoicing

ing away from the fact that languages may choose a final devoicing rule conditioned by a prosodic category higher than the word, Table 10 contains all possible combinations of the relevant (de)voicing parameters. Because of the implicational relation that holds between syllable-final and word-final devoicing, some languages are predicted not to exist. We think these predictions are correct. One language 19 The facts of Ya:thê will be discussed in §4.3. 20 A quite different approach is taken by Steriade (1997), who argues that OT constraints must make reference to phonetic cues and that prosodic factors do not determine voicing alternations. The clearest case LANGUAGE, VOLUME 77, NUMBER 2 (2001) 226 type that we predict to exist, illustrated in the last row of Table 10, was not encountered in the literature. This language would be like English, but with word-final devoicing. We suppose that its rareness is due to the relative rareness of the no-assimilation languages. Otherwise, all predicted combinations are attested. Notice that for syllablefinal devoicing languages it is structurally impossible to show the three types of [voice] spreading that we have distinguished for assimilation-only languages. This is because syllable-final devoicing reduces the possible clusters that are inputs to assimilation to [ voice][ voice] or [ voice][ voice]. If assimilation applies in a [ voice][ voice] sequence, the effect of [ voice] spreading is invisible, or, put differently, [ voice]-spreading-only remains indistinguishable from no-spreading. Furthermore, when [ voice] spreads in a [ voice] [ voice] cluster, the process is indistinguishable from the spreading of [ voice]. To be able to show the relevance of single feature spreading combined with syllable-final devoicing, we need to find languages that also possess, for example, consonantal prefixes with different lexical voice specifications, hypothetically /f [C . . . / and /z [C . . . /, of which only one, but not the other adapts to the voice value of the root-initial consonant. We have not found such languages. We have shown that the assimilation of voicelessness in the languages considered in §§3.1–3 cannot be explained properly by privative voice as integrated in a theory of devoicing as conceived by Cho and Lombardi. In the next two sections we will directly address the issue of the privativity of the voice feature, and show that [ voice] may function independently of [ voice]. 4. ARGUMENTS AGAINST PRIVATIVE VOICE: POSTLEXICAL VOICELESSNESS. It is commonly agreed that the feature [ voice] is relatively ‘unmarked’ compared to [ voice], or equivalently, that it represents the default value for [voice]. Some phonologists have drawn the conclusion that [voice] is a privative feature. Consequently, one does not expect to find a language where the feature [ voice] is specified phonologically at any level of representation, or participates in phonological processes of any kind, including rules of assimilation and dissimilation. One of the most compelling arguments for the privativity hypothesis is the existence of a language like Ukrainian, which spreads only the positive value of [voice]: compare ča[s] ‘time’ ča/s-d/ijaty N ča[z-d]ijaty ‘time to act’, žyra[f] ‘giraffe’ žyra[v#b]ižyt ‘the giraffe is running’ with xo[bt]y ‘trunk-GEN.SG’, be[Ç-t+]estia ‘dishonor’ (from Danyenko & Vakulenko 1995). If [ voice] does not exist, and since Ukrainian has no syllable-final devoicing, spreading can only apply to the only existing (positive) value, i.e. [ voice], creating clusters that we know that shows the relevance of syllable structure comes from Catalan, where obstruents assimilate in voicing to both obstruent and sonorant consonants. Consider the prefixes des and sub. (i) de[s.p]entinar ‘to uncomb’ su[p.t]ı́tol ‘subtitle’ de[z.γ]laçar ‘to defrost’ su[b.Ç]ènere ‘subgenus’ de[z.r]igar ‘to untie’ su[b.l]ı́mit ‘sublimit’ de[z.r]atizar ‘to clear of rats’ su[b.r]utina ‘subroutine’ The final consonant of a prefix is never syllabified as the onset of otherwise permissible tautosyllabic clusters. In onset clusters, the contrast between voiced and voiceless is systematically maintained. (ii) sem[.pr]e ‘always’ de[s.pl]egar ‘to unfold’ sem[.br]a ‘sows’ de[s.tr]iar ‘to separate’ It is not clear to us whether and, if so, how prosodic factors might interfere with the different phonetic cues distinguished by Steriade in the process of grammaticalization of devoicing and voice assimilation. Further research on this question might show that the phonetic and the prosodic approaches are to a certain extent