A voicing alternation in the obstruent class

An optimality-theoretic analysis

In the file on plosive devoicing, we looked at an alternation between voiced and voiceless obstruents occurring in root final position. We found that stops and affricates are voiceless at the ends of syllables, as we see in (1).

(1) Turkish nouns in the nominative nonplural and ablative nonplural, and the nominative plural cases. (Syllable divisions are indicated using periods.)

Nominative Ablative Nominative Pl. Gloss
sap sap.tan sap.lar 'stalk'
at at.tan at.lar 'horse'
sat͡ʃ sat͡ʃ.tan sat͡ʃ.lar 'hair'
kap kap.tan kap.lar 'container'
tat tat.tan tat.lar 'taste'
gyt͡ʃ gyt͡ʃ.ten gyt͡ʃ.lar 'power'
kurt kurt.tan kurt.lar 'worm'
ɟeɲt͡ʃ ɟeɲt͡ʃ.ten ɟeɲt͡ʃ.ler 'young'
reɲc reɲc.ten reɲc.ler 'color'

This is true regardless of whether a syllable final obstruent is phonemically voiced or voiceless; see (2).

(2) Turkish nouns: dative nonplural and genitive nonplural cases. (UR = underlying or phonemic representation)

Nonalternating Plosives
UR for root Dative Genitive Gloss
/sap/ sa.pa sa.pɨ 'stalk'
/at/ a.ta a.tɨ 'horse'
/sat͡ʃ/ sa.t͡ʃa sa.t͡ʃɨ 'hair'
Alternating Plosives
UR for root Dative Genitive Gloss
/kab/ ka.ba ka.bɨ 'container'
/tad/ ta.da ta.dɨ 'taste'
/gyd͡ʒ/ gy.d͡ʒe gy.d͡ʒy 'power'
/reng/ reɲ.ɟe reɲ.ɟi 'color'

Moreover, we find the pattern of syllable final voicelessness even when a following consonant is voiced, as we see from examples like those in the nominative plural column of (1), where a plosive is followed by a voiced lateral liquid. Finally, it turns out that while plosives (stops and affricates) are required to be voiceless syllable-finally,this restriction does not hold on fricatives, (3).

(3) Fricatives do not display a voicing alternation.

Nominative Ablative Nominative Plural Gloss
mas.raf mas.raf.dan mas.raf.lar 'expense'
diʃ diʃ.den diʃ.ler 'tooth'
el.mas el.mas.den el.mas.lar 'diamond'
ev ev.den ev.ler 'house'
kɨz kɨz.dan kɨz.lar 'daughter, girl'
aɨz aɨz.dan aɨz.lar 'mouth'
de.niz de.niz.den de.niz.ler 'sea'

We argued that this distribution is the result of a phonological rule that devoices [-sonorant,-continuant] consonants (that is - stops and affricates) at the end of any syllable, resulting in the neutralization of the underlying voicing contrast in this environment.

Our job in this section is to sketch a brief OT analysis for the voicing alternation. Remember that OT analyses explain sound behaviours as the result of language-specific rankings assigned to universal constraints. To start with, let's remember that in Turkish, the set of voiced plosives keep their underlying specification for [+voice] in syllable initial position. The relationship between a surface representation and its corresponding phonemic representation is harmonic (natural or, in some sense, preferred) to the extent that it is faithful, or preserves underlying features on the surface. In a form like ka.ba 'container, dative', in which the root /kab/ is followed by the dative suffix, the voicing of /b/ ispreserved in the output. This match satisfies the faithfulness constraint IDENT(Voi) in (4). This constraint requires identity (a match) in the voicing of a sound's input and output states.

(4) IDENT(Voi): Corresponding segments of the input and output have identical valuesfor [voice]. (One * for every input/output segment pair whose values for [voice] do not match.)

(This constraint cares only about voicing. Other constraints in the Ident family could be used in cases where different features are important.)

On the other hand, outputs like kap 'container, nominative (sing.)', in which the final stop is voiceless, violate Ident(Voi). If faithfulness between input and output is harmonic, then why would we find outputs like kap that are non-harmonic with respect to (4)?

The OT view is that the violation of a phonological constraint can only occur to avoid a violation of another, competing constraint. In this case, the relevant constraint is CODACOND(VOI) in (5), which states that a postvocalic consonant in a syllable may not have the features [+voice,-continuant].

(5) CODACOND(VOI): * V C ...
[-cont] SYLL
(One * per [+voice,-cont] consonant in a syllable coda.)

CodaCond(Voi) directly competes with Ident(Voi) in any case in which an underlyingly voiced obstruent comes at the end of a syllable, in which case, it is not possible for a single output to satisfy both constraints. In Turkish, the attested outcome is voicelessness in coda position, which tells us that the requirement stated by CodaCond(Voi) is prioritized more highly in the grammar than the requirement stated by Ident(Voi). An OT analysis expresses this by encoding the constraint ranking CodaCond(Voi) ≫ Ident(Voi) in the grammar. (In an OT analysis, a statement such as A ≫ B means that constraint A outranks constraint B in the hierarchy.) The tableau in (6) shows how this ranking determines the attested outcome for forms like kap.

(6) CodaCond(Voi) ≫ Ident(Voi) (generates the attested result)

Input /kab/ CodaCond(Voi) Ident(Voi)
a. ✓ kap *
b.    *kap *!

The tableau in (7) shows that the inverted ranking produces the wrong result for Turkish.

(7) Ident(Voi) ≫ CodaCond(Voi) (generates the wrong result)

Input /kab/ Ident(Voi) CodaCond(Voi)
a.    kap *!
b. X *kap *

Even though the alternative ranking in (7) cannot be correct for Turkish, it can and does occur in other languages. For example, English has a surface contrast between voiced and voiceless plosives in syllable final position, in pairs such as bet/bed, back/bag, and so on. The ranking Ident(Voi) ≫ CodaCond(Voi) in (7) must be the right ranking for English, otherwise we would have no voiced plosives in syllable final position.

The ranking CodaCond(Voi) ≫ Ident(Voi) that we have inferred for Turkish is consistent with the outcome found for inputs like /sap/ 'stalk', which produce outputs like sap. In cases like these, the outcome is straightforward. Since output [p] is faithful to input /p/, the faithful candidate sap satisfies both CodaCond(Voi) and Ident(Voi) so that in this case, at least, there is no competition. That is, conditions are such that nothing forces the violation of the faithfulness constraint in order to satisfy the constraint on syllable structure.

(8) CodaCond(Voi) ≫ Ident(Voi) (generates the attested result)

Input /sap/ CodaCond(Voi) Ident(Voi)
a. ✓ sap
b.    *sap *! *

Another case in which we find no conflict between constraints is when an underlyingly voiced plosive occurs in syllable initial position. In this case, CodaCond(Voi) does not force a violation of Ident(Voi). It is exactly in these cases that the underlying contrast between voiced and voiceless plosives is allowed to surface. Tableau (9) shows the outcome for the forms sa.pa and ka.ba, in which the roots /sap/ and /kab/ are followed by the dative suffix.

(9) CodaCond(Voi) ≫ Ident(Voi) (generates the attested result)

Input /sap + a/ CodaCond(Voi) Ident(Voi)
a. ✓ sa.pa
b.    *sa.pa *!
Input /kab + a/ CodaCond(Voi) Ident(Voi)
a.    *ka.pa *!
b. ✓ ka.ba

Notice that looking at different kinds of input-output pairs provides us with different information about relevant constraints and their rankings. Cases in which there is no conflict between constraints doesn't tell us much! Direct arguments for ranking constraints depend on cases in which constraints with different requirements compete.

By stating CodaCond(Voi) to prohibit the feature combination [+voice, -continuant] in coda consonants, we have accounted for why devoicing applies to stops and affricates, but not fricatives, (cf. kɨz 'daughter, girl'). Fricatives are [+voice,+continuant]. Affricates have these features, but overlap with stops in also having [-continuant]. A trickier problem arises in the case of coda nasals. Nasals are also characterized by the features [+voice, -continuant], but they are not devoiced in coda position, as predicted by CodaCond(Voi) (e.g. reɲc 'color, nominative sing.', not *reNc, where the voiceless nasal is capitalized). Why wouldn't nasals pattern with plosives in this case?

A plausible explanation for the exclusion of nasals from the devoicing set is that voicelessness is not the most "natural" or harmonic state for nasals, or indeed, for any other sonorant sound. The most harmonic state is for segments with the feature [+sonorant] to be voiced. This requirement is stated as the constraint SON/VOI in (10).

(10) SON/VOI: [+sonorant] implies [+voice]. ('All [+sonorant] segments are [+voice].' One * for any voiceless sonorant.)

That nasals do not devoice syllable finally is accounted for if Turkish's grammar ranks Son/Voi above CodaCond(Voi) in the constraint hierarchy. The tableau in (11) demonstrates this result, with an example based on the input /reng/ 'color'. This case is a good one for testing the relationship between the three constraints introduced so far.

(11) Son/Voi ≫ CodaCond(Voi) ≫ Ident(Voi) (generates the attested result)

/reng/ Son/Voi CodaCond(Voi) Ident(Voi)
a. ✓ reɲc * *
b.    *reɲɟ **!
c.    *reNc *! *

(Capital [N] in (11) denotes a voiceless nasal.)

In a nutshell, ranking Son/Voi above CodaCond(Voi) and Ident(Voi) below CodaCond(Voi) accounts for why plosives devoice syllable finally, whereas nasals do not.

The introduction of Son/Voi in (10) raises yet another issue. Whereas [+voice] is the harmonic state for sonorant sounds, voicelessness is the harmonic state for obstruents. The latter is expressed as the constraint in (12).

(12) OBST/VOI: [-sonorant] implies [-voice]. ('All [-sonorant] segments, are [-voice].' One * for any voiced obstruent.)

If Obst/Voi is a highly ranked constraint, then we might ask why the sound inventory of Turkish (or any other language) contains any voiced obstruents at all. How marked sounds like voiced obstruents come to be in the sound inventory is a larger question that we can explore here. (A short answer is that allophonic rules can create marked allophones that later become phonemes, often as the result of opacity, or the loss of information needed to deduce the context in which the marked allophones were derived.) Once marked sounds are present in a sound inventory as phonemes, however, then their preservation in the inventory is enforced by faithfulness constraints. That any voiced obstruents are possible in Turkish at all is an effect of ranking Obst/Voi in (12) below Ident(Voi). Tableau (13) shows that the input ka.ba from (9) surfaces as ka.ba under this ranking.

(13) Ident(Voi) ≫ Obst/Voi generates the attested result ka.ba

Input /kab + a/ Ident(Voi) Obst/Voi
a. ✓ *ka.ba *
b.    ka.pa *!

Indeed, a language with the inverted ranking Obst/Voi ≫ Ident(Voi) would be a language that does not permit voiced obstruent phonemes.

To conclude, in this short section we've motivated (argued for) a number of crucial constraint rankings in the sound grammar of Turkish. Crucial rankings are cases in which it must be the case that constraint A outranks constraint B. The argument for a crucial constraint rankings is always based on a case in which two constraints compete for different outcomes, but only one constraint can be (and is) satisfied in the attested output. Combining all of our ranking arguments, we can see that we have motivated the fragment of Turkish's constraint hierarchy shown in (14).

(14) Final constraint (sub)hierarchy:Son/Voi ≫ CodaCond(Voi) ≫ Ident(Voi) ≫ Obst/Voi

Remember that [c] and [ɟ] are palatal allophones of the velars. Visit the page on
palatalization for a detailed discussion.

prev | top