A voicing alternation in the obstruent class

Root final voicing alternations

In this section, we'll describe and analyze an alternation between voiced and voiceless obstruents that occur in root final position. If we want to understand consonant voicing patterns in a language, it's useful to know what the consonant inventory is, and how the feature [voice] is used in expressing contrasts in the sound system. To review this information, revisit Turkish's inventory of sounds.

Since the problem we're working on in this section involves obstruent voicing, our first question must be whether we find evidence for an underlying voicing contrast in this sound class. In fact, we do find abundant evidence. Remember that in order to demonstrate a phonological contrast between two sounds in a language, we need to show that these sounds can occur in overlapping contexts (that is, in at least some of the same contexts, or if not exactly the same, then as similar as possible). Turkish's consonant inventory has six pairs of obstruents matched for place and manner of articulation, but which differ in voicing (e.g. [p] vs. [b]; [f] vs.[v]; and so on). The table in (1) provides pairs of words that show a contrast between the voiced and voiceless members of each pair in either medial or in word initial position, before a vowel.

(1) The boldfaced obstruents in each row contrast for voicing.
(A period indicates a division between syllables.)

[p] / [b] sapa 'stalk dat.' kaba 'container dat.'
piʃirir 's/he cooks' birisi 'someone'
[t] / [d] ata 'horse dat.' tada 'taste dat.'
tec 'single' deer 'value'
[t͡ʃ] / [d͡ʒ] sat͡ʃa 'hair dat.' ɟyd͡ʒe 'power dat.'
t͡ʃɨɨr 'era' d͡ʒezaa 'punishment'
[k] / [g] kaba 'container, dat.' gurup 'group'
jakut 'emerald' ag 'chest, bosom'
[f] / [v] fihrist 'index' verdi 's/he gave'
ufak 'small' sevap 'good deed'
[s] / [z] sap 'stalk nom.' zam 'price increase, nom.'
havasɨz 'air poss.' kazaa 'accident'
[ʃ]/[ʒ] ʃehir 'city'    
diʃe 'tooth, dat.'

(Note that the palatal stops [ɟ] and [c] are fronted (palatalized) allophones of /g/ and /k/, respectively. Palatal and velar plosive allophones behave alike in voicing alternations. From now on in this section, we will treat [k c] as allophones of /k/ and [g ɟ] as allophones of /g/. The argument for this analysis can be found on the palatalization page.)

Given the voicing contrasts seen in (1), we have good reason for analysing the sounds [p], [b], [f], [v], [t], [d], [ʃ], [ʒ], [t͡ʃ], [d͡ʒ], [k c], and [g ɟ] as allophones of distinct phonemes. This distributional analysis is schematically represented in (2).

(2) Phonemic inventory of obstruents motivated by syllable-initial contrasts.

a. Plosives

| \
[k] [c]
| \
[g] [ɟ]

b. Fricatives


As we discussed earlier, all we need to establish a contrast between a pair of speech sounds is to show that they can occur in like contexts. If the distribution of these sounds is perfectly symmetrical, then they will occur in all of the same contexts, and neither will occur where the other does not. In fact, such perfectly symmetrical distributions rarely (if ever) occur. Instead, it is much more common for two contrasting sounds to be asymmetrically distributed, meaning that they share only a subset of their environments. If we look further, we see that this is the case for the distribution of certain voiced and voiceless consonants in Turkish. To see this, let's study the data below.

Figure (3) shows a set of Turkish nouns in the nominative case. Each noun is a bare unaffixed stem (a free root that can stand on its own as a stem). Notice that the final consonant in each of these words is voiceless.

(3) Turkish nouns, nominative case. ('.' denotes a division between syllables.)

a. sap 'stalk' b. kurt 'worm' c. kap 'container'
at 'horse' ɟeɲt͡ʃ 'young' tat 'taste'
sat͡ʃ 'hair' reɲc 'color' ɟyt͡ʃ 'power'

The forms in figure (4) show the roots in (3) before suffixes beginning with a consonant. In all of these examples, the root final obstruent is voiceless, just as in (3).

(4) Turkish nouns, ablative singular and nominative plural cases.

Abl Nom PL Gloss
saptan saplar 'stalk'
attan atlar 'horse'
sat͡ʃtan sat͡ʃlar 'hair'
kaptan kaplar 'container'
tattan tatlar 'taste'
ɟyt͡ʃten ɟyt͡ʃler 'power'
kurttan kurtlar 'worm'
ɟeɲt͡ʃten ɟeɲt͡ʃler 'young'
reɲcten reɲcler 'color'

The pattern we've seen so far changes with the forms in figure (5), which shows the stems we've already seen in their nominative and dative forms. In (5), the root final obstruents in the first set, on the left, are voiceless, just like their counterparts in (3) and (4). However, the obstruents in the second set (on the right) surface as voiced before the dative and 2nd singular possessive suffixes. If we compare these root final obstruents with their counterparts in (3) and (4), we see evidence for a voicing alternation: these obstruents sometimes appear as voiceless (see (3) and (4)), but as voiced below.

(5) Turkish nouns, dative and 2nd singular possessive cases.

Nonalternating obstruents Alternating obstruents
Dat 2nd Sing. Poss. Gloss Dat 2nd Sing. Poss. Gloss
sapa sapɨn 'stalk' kaba kabɨn 'container'
ata atɨn 'horse' tada tadɨn 'taste'
sat͡ʃa sat͡ʃɨn 'hair' ɟyd͡ʒe ɟyd͡ʒyn 'power'
      kurda kurdun 'worm'
      ɟeɲd͡ʒe ɟeɲd͡ʒin 'young'
      reɲɟe reɲɟin 'color'

Notice that the dative suffix alternates between -a and -e while the 2nd singular possessive has four allomorphs, -in, -yn, -un, and -in.. These differences are due to the effects of vowel harmony, discussed in a different module.

To summarize, we have demonstrated an asymmetry in the distribution of voiced and voiceless plosives in Turkish. This asymmetry is represented in (6). Our generalization so far is that in Turkish words, both voiced and voiceless obstruents can precede a vowel. However, the distribution of voiced obstruents seems to be more restricted than that of voiceless obstruents. There are two contexts in which voiced obstruents may occur, but voiceless obstruents cannot: in word final position, and before a consonant.

(6) Voicing asymmetry

  ____ V ____ C ____ #
Voiced X X

At this point in our analysis, there are two questions we should ask in order to make sure that our generalization is as tight as it can be. First of all, we have stated that our voicing alternation is found in the obstruent class. However, we have seen no data to indicate whether the voicing behaviour of fricatives, a subclass of obstruents, is consistent with this generalization. If additional evidence shows that fricatives do not pattern like the stops and affricates we have seen, we will need to restate our generalization, restricting the voicing alternation to the class of plosives, the set of [- continuant] obstruents (stops and affricates).

In fact, the additional data shown in (7) establishes that fricatives do not undergo devoicing. Rather, we find that voiced and voiceless fricatives contrast in just the environments from which [+voiced] plosives are excluded.

(7) Fricatives do not display a voicing alternation.

Nom Dat 2nd Sing. Poss. Abl Nom. Plural Gloss
masraf masrafa masrafɨn masraftan masraflar 'expense'
ev eve evi evden evler 'house'
elmas elmasa elmasɨn elmastan elmaslar 'diamond'
kɨz kɨza kɨzɨn kɨzdan kɨzlar 'daughter, girl
aɨz aɨza aɨzɨn aɨzdan aɨzlar 'mouth'
deniz denize denizin denizden denizler 'sea'
diʃ diʃe diʃin diʃten diʃler 'tooth'

Note: In the Ablative we see the effects of a second voicing alternation. In these cases, the initial consonant of the ablative suffix alternates between [t] and [d], where [t] occurs after voiceless sounds, and [d] occurs elsewhere. Since the suffix vowel also alternates between [e] and [a], the ablative morpheme has four allomorphs, -ten, -den, -tan, and -dan. (See also the section on vowel harmony.) We don't provide an extended discussion of the [t] ~ [d] alternation. However, given that [t]'s distribution is more restricted (only after voiceless sounds), we will assume that inthis case, [t] is an allophone of underlying /d/.

The second question we need to ask is whether it is possible to simplify our description of where the voiceless stops and affricates occur, in cases where there is a voicing alternation. That is, we've seen that the voiceless stops and affricates are found word finally and before another consonant (see (6)). We could simplify our distributional statement if we could express the contexts __C and __# as a single coherent environment. In fact, it is possible to do this. Turkish places a syllable boundary immediately before a prevocalic consonant. The examples in (8) show how forms we've already seen are parsed into syllables. (Periods are used to indicate divisions between syllables.)

(8) Syllable initial [-voi] / [+voi] and syllable final [-voi]

  2nd Sing. Poss. Dative Nom. Ablative Nom. plural  
a. sa.pɨn sa.pa sap sap.tan sap.lar 'stalk'
sa.t͡ʃɨn sa.t͡ʃa sat͡ʃ sat͡ʃ.tan sat͡ʃ.lar 'hair'
b. ka.bɨn ka.ba kap kap.tan kap.lar 'container'
ɟy.d͡ʒyn ɟy.d͡ʒe ɟyt͡ʃ ɟyt͡ʃ.ten ɟyt͡ʃ.ler 'power'

Once we understand how syllables are formed, we see that the two environments in which we find only voiceless stops, __# and __C, together form a single coherent phonological environment - the end of a syllable. (The end of a word also counts as the end of a syllable.) Thus we find kap and kap.tan, but forms like *kab and *kab.tan do not exist in Turkish. We are now ready to restate our distributional statement of alternating voiced and voiceless plosives in Turkish as in (9).

(9) Turkish plosive voicing alternation.
a. In some Turkish forms, there is an alternation between voiced and voiceless plosives (stops and affricates, which have the feature [-continuant]).
b. In alternating forms, the voiceless plosives occur syllable finally, or following vowels, while their voiced counterparts occur before vowels, or syllable initially.

prev | top | next