A voicing alternation in the obstruent class

A phonemic analysis

The next step in solving this problem is to perform a phonemic analysis of alternating plosives, and of the non-alternating ones as well. To step back briefly, the data seen earlier in (1) established that we have a contrast between voiced and voiceless obstruents in Turkish. This means that each of these sounds corresponds to a phoneme, as shown in (2). If we reconsider noun sets like those in (8a) and those in the leftmost column of (5), we see that these noun roots have a final obstruent that does not vary with context (for example, [p] in the 'stalk' forms and [t͡ʃ] in the 'hair'). By contrast, the root final obstruents in (8b) and in other cases (compare (3b,c) the rightmost set of forms in (5)), alternate between voiceless and voiced qualities in the environments described in (9).

It makes sense to assume that a single phoneme underlies each of the alternating pairs of sounds in (6b), [p] ~ [b] in 'container' and [t͡ʃ] ~ [d͡ʒ] in 'power'. The question to be decided is whether the phonemes underlying the voicing alternation are voiced or voiceless. That is, for example, should we analyse the alternating [p] and [b] in the 'container' set as /p/ or /b/?

(10) / ? / ( /p/ or /b/? )
  [p] [b]

To decide this, we need to consider not only the alternating cases, but also the cases in which no variation is observed.

In cases like (8a), where a voiceless obstruent is observed in every member of the noun set with no variation at all, the simplest analysis is that these voiceless obstruents are basic allophones of voiceless phonemes. That is, [p] in sap- is the basic allophone of the phoneme /p/. It makes no sense to posit that non-alternating [p]'s are derived from /b/ because there is no direct evidence for this change.

Now, let's consider the alternating cases. If the non-alternating voiceless obstruents are analysed as phonemes of "themselves" (i.e. basic allophones of /p/, /t͡ʃ/, etc.), then to distinguish these cases from the alternating cases, we should explore the idea that the phonemes underlying the voiced-voiceless alternations (e.g. (8b)) are voiced obstruents (i.e. /p/, /d/, etc.). This works well: in the alternating cases, we may hypothesize that voiced obstruent phonemes /b d d͡ʒ g/ surface unchanged, [b d d͡ʒ g ɟ], in syllable-initial position, but are devoiced, surfacing as [p t t͡ʃ k c] when they occur at the end of a syllable. This analysis accurately accounts for the alternating cases, and leaves the nonalternating cases alone. The rule of Syllable Final Devoicing assumed under this analysis is shown in (11a), and the effects of this rule are shown in (11b).

(11) a. Syllable Final Devoicing:
[-sonorant, -continuant] → [-voice] / __ ] SYLL
  b. Assumed phonemic representations: /kab/ → [kap] (devoicing)
/sap/ → [sap] (no effect)

It is worth noting at this point that an approach that treats devoicing as assimilation doesn't account for the pattern we are trying to explain. If devoicing were due to assimilation, we would expect to find voiceless allophones before voiceless sounds, such as [t] in the ablative forms. However, devoicing also occurs before voiced consonants, such as [l] in the nominative plural, and word finally, where no consonant follows. Thus, devoicing in Turkish is linked to the structural environment rather than to any property of an adjacent sound segment.

An alternative to the syllable final devoicing approach outlined above would be an analysis that takes the phonemes underlying the voice alternation to be voiceless (/p/, /t/, etc.). Under this alternative, we would want a rule like the one in (12) that voices the voiceless obstruents whenever they occur in syllable initial position.

(12) a. Syllable Initial Voicing
[-sonorant, -continuant] → [-voice] / __ ] SYLL
  b. Assumed phonemic representation: /kap-a/ → [ka.ba] 'container, dat.'
  c. *sa.bɨ     *sa.ba    'stalk' (from 8a)
*sa.d͡ʒɨ   *sa.d͡ʒa   'hair' (from 8a)

The voicing approach is successful in accounting for the alternating cases, (12b), but it over-generates, as shown in (12c). Recall that non-alternating obstruents (e.g. [p] in sap) are also analysed as voiceless phonemes. The voicing analysis incorrectly predicts that voiceless obstruents in nonalternating roots should voice at the beginning of syllables as well. That is, if the voicing analysis were correct, there should be no voiceless obstruents at all in syllable initial positions on the surface. We could save the voicing analysis by listing the non-alternating roots in the lexicon as exceptions to voicing. However, this would be missing the point - the failure of the nonalternating obstruents to voice in syllable initial position is not accidental, but is part of a systematic sound pattern that should be analysed as the result of a sound process that applies regularly. Moreover, obstruent devoicing in syllable final position is a common pattern crosslinguistically, whereas voicing in this position is extremely rare, if in fact it occurs at all. For these reasons, we will take the devoicing analysis in (11) to be the correct one.

Based on the reasoning set out above, we may now conclude that the phonemic relationships between voiced and voiceless plosive allophones are those shown in (13).

(13) Phonemic analysis of alternating stops and affricates.
  /b/   /d/   /d͡ʒ/   /g/
  [p] [b] [t] [d] [t͡ʃ] [d͡ʒ] [k c] [g ɟ]

The effect of the process of syllable final devoicing we have been discussing is to neutralize (or "nuke") an underlying contrast between voiced and voiceless plosives. That is, we know that Turkish has a contrast between voiced and voiceless stops and affricates based on forms like those in (1). This contrast is apparent at the beginnings of syllables, where we see all of the obstruents in Turkish's consonant inventory. However, in the case of stops and affricates, we don't see this contrast in syllable final position: due to the devoicing rule, no voiceless stops or affricates ever appear at the ends of syllables. An underlying contrast between distinct phonemes is neutralized when some process derives an allophone of another phoneme. That is, in Turkish, voiceless plosive phonemes have voiceless basic allophones, as expected. However, syllable final devoicing derives voiceless allophones of voiced phonemes. That is, for the phoneme /b/, devoicing derives an allophone [p] that is just like the basic allophone of the phoneme /p/. Neutralization is represented schematically in (14).

(14) Phonemic inventory of obstruents motivated by syllable-initial contrasts.
  /p/ /b/ /t/ /d/ /t͡ʃ/ /d͡ʒ/ /k/ /g/
  [p] [b] [t] [d] [t͡ʃ] [d͡ʒ] [k c] [g ɟ]

Due to the neutralizing effect of the devoicing rule, if all we saw were plosives occurring at the ends of syllables, we wouldn't realize that Turkish also has voiced plosives. To know that the voiced plosives exist, we need to see plosives in the other (syllable initial) environment as well.

To summarize, we've seen that in Turkish, an obstruent voicing contrast found in syllable-initial position is neutralized when obstruents occur at the ends of syllables: only voiceless obstruents occur in syllable final position. We have argued that this pattern is best accounted for by an analysis that assumes that underlyingly voiced obstruents are devoiced in syllable final positions.

It is worth noting that there are a few sporadic exceptions to the syllable final devoicing rule, for example Ad, a proper name, and serhad 'frontier'. In fact, that there are some, but very few lexical exceptions can be taken as evidence that the devoicing process we have been discussing is truly a phonological process, and cannot be attributed to a low-level phonetic effect. If this were a phonetic effect, we would expect devoicing to apply everywhere the structural description for the rule is met, with no exceptions. However, phonological rules, which are more categorical in nature, are known to have the odd lexical exception. Of course, if we encountered too many exceptions, we might have to reconsider the analysis we've proposed.

Click on the link below to go to the optimality theoretic analysis of the alternation we have analysed in this section.

prev | top | next