Is this the best we can do?


In phonological analysis, simpler explanations are valued over more complex ones, other things being equal. This means that given a choice between two analyses that account for the same data, the simpler explanation is generally preferred. The analysis we've proposed for palatalization in the previous section requires five rules (Postvocalic Velar Fronting counts for four rules) to handle what intuitively seems to be a fairly straightforward sound process. This being the case, we should wonder if we have missed something in developing both our generalization and our analysis.

What we have is a process of palatalization that targets a dorsal either immediately before a vowel, or up to two consonants after a vowel (in word-final or pre-consonantal position) - as long as the post-vocalic dorsal is not immediately followed by a vowel. Why should the presence of a vowel immediately following the dorsal be such an important influence on the sound process? Our account of palatalization so far has nothing to say about this, other than to notice the pattern.

If we look at other sound phenomena in Turkish, we see that interestingly, similar restrictions hold elsewhere. For example, consider again the discussion of obstruent devoicing discussed on a different page in this module. Obstruent devoicing is a different kind of sound process: it is not a form of assimilation, in which a property of one sound is transferred to another. Rather, obstruent devoicing affects members of a class of sounds that stand in certain positional relationships to other sounds. A basic description of the context in which obstruent devoicing applies in Turkish is given in (20).

(20) Turkish obstruent devoicing
a. Never targets a consonant occurring immediately before a vowel.
b. Does target a postvocalic voiced obstruent, VC, where a [+sonorant] consonant can intervene between the vowel and the target voiced obstruent - as long as the target obstruent is not immediately followed by a vowel.

These conditions are quite similar to the conditions on palatalization: as (20) makes clear, the presence of a vowel following a voiced obstruent is important because it defines a negative environment, or environment that is "off limits" to the rule. The conditions in (20b) replicate exactly the conditions on postvocalic palatalization.

Palatalization differs from obstruent devoicing in that palatalization involves an interaction between sounds (assimilation) that is not a component of the obstruent devoicing rule. More than this, however, palatalization also seems to be conditioned by structural factors (i.e. as a first pass, how sounds are ordered linearly), and the limits on where palatalization can apply bear a remarkable similarity to the conditions on obstruent devoicing. (Yet another example in which similar structural conditions apply on a sound process in Turkish is discussed in the section devoted to vowel length alternations.) It seems inefficient to replicate a particular set of conditions that hold on a disparate set of sound rules. This should lead us to wonder whether the set of conditions repeated from rule to rule should really be stated independently, as part of a separate generalization. We would argue that this is exactly right, and given that some conditions on both rules are structural in nature, we can isolate these conditions and treat them as a separate aspect of the sound system.


Phonological structures familiar to us include sound segments and words. Segments organize sets of features that determine the character of a speech sound. A word could be defined as a unit of organization for segments. A problem with such a definition is that it does not take into account the fact that segments are not randomly ordered in a word. No language has words with the structure VVVVCCC (to pick an arbitrary example) in which a string of several vowels is followed by a lengthy array of consonants. Rather, consonants and vowels appear in roughly alternating positions in words and there are conditions (which may vary from language to language) on how many consonants may intervene between vowels, as well as on what consonants these may be. One way of understanding this universal patterning of sounds is to view words in Turkish (or any other language) as being segmented into groups of sounds, each organized around a single vowel. What we have just provided is an informal definition of the syllable - a unit of organization larger than the segment, but smaller than the word. Some words, or course, may consist of just one syllable (and Turkish has many of these). Examples of Turkish monosyllabic words are given in (21).

(21) Turkish words containing a single vowel

o 'that (yonder, over there)'
bu 'this'
at 'horse nom.'
iljc 'first'
kap 'container, nom.'
reɲc 'color, nom.'

Notice that these monosyllabic words come in several "shapes". Some begin with a single C (an unmarked, or preferred property of syllables across languages). Some have no initial C. Some of these syllables end in one or two Cs. If a syllable ends in two Cs, the first is [+sonorant]. All of these syllable shapes are generally allowed in Turkish.

In fact, we can use the notion of the syllable to help us understand conditions on rules, and also why some rules are subject to similar conditions. With respect to devoicing, notice that devoicing applies to voiced obstruents at the ends of the unaffixed monosyllabic words (or noun roots) in (22), whereas devoicing does not apply when the root is followed by a vowel (the dative forms). If the only evidence we had available were provided by the nominative and dative forms, we might analyze devoicing as a rule that applies in word final position. That this is not the case is shown by the nominative plural forms, where devoicing applies before the lateral consonant [l].


a. UR (modified) b. Syllables
/kab/ [kap] σ 'container, nom.'
/kab-a/ [ka] σ [ba] σ 'container, dat.'
/kab-lar/ [kap] σ [lar] σ 'container, nom. plural'
/reng/ [reɲc] σL 'color, nom.'
/reng-e/ [reɲ] σ [ge] σ 'color, dat.'
/reng-ler/ [reɲc] σ [ljer] σ 'color, nom. plural'

The representations in (22a) are not quite underlying representations because they incorporate the results of a vowel alternation in the dative suffix -a/-e and plural suffix -lar/-ler. For an extended discussion. In the analysis Turkish vowel harmony developed on a different page in this module, we propose that the vowel in these suffixes is specified only for the features [-high] and [-round] in these cases.

Even so, we could write devoicing as a rule that applies to an underlyingly voiced obstruent in two environments: word finally, __ #, and before a consonant ___ C. Alternatively, an analysis that connects these environments would be to write a rule that captures the generalization that both word-final consonants and consonants before another C have in common that they occur at the end of a syllable, as shown in the second column of (22).

In fact, we can use the notion of the syllable in our analysis of palatalization. A key observation in this section has been that for purposes of palatalization, the quality of a dorsal is controlled by an immediately following vowel, if there is one. We may infer from this that a consonant immediately preceding any vowel is drawn into the syllable unit containing that vowel. Consonants not immediately preceding a vowel are drawn into the syllable defined by the preceding vowel. If we adopt these principles of segmentation (or syllabification) for Turkish, then the (modified version of) underlying representations in (22a) are assigned the syllable divisions in (22b). (Each example in (22) consists of a root or a root plus affixes, where a hyphen marks a morpheme boundary. Syllable units are enclosed between square brackets.)

New hypothesis and generalization:

(23) Syllabification in Turkish:
  • A prevocalic consonant is organized into the same syllable as the vowel following it.
  • In an intervocalic cluster of two consonants, a syllable boundary divides the consonants, VC.CV
  • In an intervocalic cluster of three consonants, VCCCV, the medial consonant is always [+sonorant], and a syllable boundary divides the consonants, VCC.CV
  • A cluster at either edge of the word is parsed tautosyllabically (i.e. into the same syllable).

(24) Generalization (final version): Within the class of dorsal stops, [+back] velars and [-back] palatals are in complementary distribution. The backness of a dorsal is determined by the backness of the vowel in the syllable that contains both.

Having arrived at a satisfactory generalization, we can now write a rule of palatalization for Turkish that focuses on the interaction between sounds ([+back] assimilation) and domain (syllable-internal) without having to include in the rule independent constraints on what syllables in Turkish can be. The final version of our palatalization rule, Velar Fronting, is given in (25). The square brackets in the structural description (context) for this rule indicates the boundaries of the syllable unit. The variable 'α' denotes irrelevant material intervening between the interacting vowel and dorsal.

(25) Turkish Velar Fronting (final version)
a. /velar/ → [palatal] / [ ___ [front vowel] α ___ ] SYLL
b. /-son,-cont,Dor,+back/ → [-back] / [ ___ [+syllabic, -back] α ___ ] SYLL

The rule in (25) is preferable to the set of rules given earlier because it is simpler; explains why assimilation is bidirectional, and formalizes the connection between the regressive and progressive forms of palatalization; and finally, (25) does not recapitulate conditions on syllable structure in Turkish that hold on a number of rules, and are best stated independently.

prev | top | next