The Typology of Voicing and Devoicing

This article provides empirical evidence against the claims that [voice] is a privative feature and that word-internal devoicing can occur in a language without word-final devoicing. The study of voice patterns in a number of languages shows that the feature value [− voice] although it is the unmarked value of the laryngeal feature [voice], can be active phonologically in a fashion parallel to the marked value [+ voice]. Across languages, voice assimilation may occur independently of devoicing and, although it normally affects both [+ voice] and [− voice], it may affect only one value in some languages.*

Final devoicing and voicing assimilation are two phenomena that have received broad attention in the literature. In this article we re-examine these phenomena, focusing on the representation of [± voice] and on the proper formulation of the mechanisms responsible for surface (de)voicing effects. Our objective is to argue against a number of assumptions on (de)voicing that underlie recent discussions of these phenomena. One assumption is that there are languages—Yiddish, Serbo-Croatian, and Rumanian are claimed to belong to this class—in which word-final coda consonants constitute exceptions to syllable-final devoicing. We will argue, instead, that these languages have no syllable-final devoicing. We will also show that there is no empirical evidence in favor of the claim that devoicing can affect only part of a cluster of voiced consonants without being prosodically conditioned. It furthermore appears that languages that apply devoicing to a class of segments at the end of a prosodic category n, devoice the same set of segments at the end of all prosodic categories that contain n, i.e. syllable-final devoicing implies word-final devoicing, etc. Most importantly, we will empirically falsify the claim that [− voice] does not belong to the universal set of phonological features by illustrating the assimilation of [− voice] in a variety of languages, postlexically as well as lexically. Our conclusion is that [voice] is a binary feature.

1. A Preliminary typology of voice assimilation and devoicing

In some languages voice neutralization (devoicing) occurs at the end of the syllable. In the same language, voice assimilation may or may not occur. Whereas Dutch has voice assimilation, German does not, as is illustrated in Table 1. For the sake of comparison, we add Yiddish,¹ which has only assimilation, and Berber, which shows neither devoicing nor generalized assimilation.²

Table 1.

A typology of voicing assimilation and devoicing.

Standard autosegmental analyses derive surface forms from underlying voiced/voiceless distinctions with rules that change the voicing value through delinking (devoicing) and delinking cum spreading (assimilation). In Table 2 the upper set of structures illustrates the effect of delinking cum spreading, the lower set illustrates delinking in final position (C) and before sonorant consonants [− voc(alic), + son(orant)].³ The representations in the first column of Table 2 show the initial structures, which also correspond to the surface structures in languages like English, which do not show general devoicing or assimilatory effects. In column 2, coda delinking results in devoicing, both in potential assimilatory environments and before sonorants, as in German. If a language has both delinking and spreading, the resulting structures are those in column 3, the Dutch case. Yiddish, which has no delinking in nonassimilatory environments, but where assimilatory environments trigger delinking cum spreading, exemplifies the fourth possibility.⁴

Table 2.

Voice effects.

In many languages devoicing and voicing assimilation are not controlled by syllable structure. In such languages devoicing may occur, but is restricted to a word- or phrasefinal (sequence of) obstruent(s). Word-internally before sonorant consonants, a voice contrast is maintained. If assimilation occurs, the last obstruent in a sequence determines the voice value of the complete cluster. The basic typology for non-syllable-final devoicing languages, i.e. types II and IV in Table 1, is shown in Table 3, where Serbo-Croatian is in all relevant respects like Yiddish. Berber again exemplifies a no-effect language.⁵Table 2. Voice effects.