A more detailed look at the data and their implications

A problem with the generalization stated in (4) is that the statement of context provided is imprecise. This is also a failing of the rules in (7). "X co-occurs with y", or "x occurs in the context of y" can mean many different things. Without a precise understanding of the context that conditions the velar-palatal alternation, it will not be possible to determine the exact nature of the sound process. Our task now is to describe the conditioning context in precise terms, and to refine our initial generalization in this and other ways.

Looking back, we saw in (2) and (3) that front vowels condition palatal dorsals, whether the conditioning vowel precedes (e.g. cir dirt', kol 'arm') or follows (e.g. c 'pour', ak 'white') the dorsal. In every case we've seen, the dorsal and conditioning vowel are immediately adjacent, and a precise generalization should reflect this. There is another way that we can modify our generalization to reflect a connection missed by the first version in (4). Recall from the feature chart in (1) that within the dorsal class, velars are [+back] and palatals are [-back]. Thus, the connection between the backness of a dorsal stop and of a vowel adjacent to it would seem to be no accident: this is a form of agreement. Bearing these factors in mind, we can restate our preliminary hypothesis as in (8).

(8) Interim generalization 2: Within the class of dorsal stops, velars and palatals are in complementary distribution. A dorsal stop agrees with the backness of an adjacent vowel, such that the [+back] velars occur only adjacent to back vowels, while the [-back] palatals only occur adjacent to front vowels.

While more precise, this generalization is based on a limited set of data (dorsals in word initial and word final position), and should be tested by looking at a broader sample. For example, we might want to see whether the behaviour of dorsal stops in medial positions is consistent with our generalization. Looking further, we find examples that seem to confirm our hypothesis. For instance, the quality of the intervocalic dorsal stops in the examples in (9) is exactly what we expect - palatals between front vowels and velars between back vowels.


a. Palatals
harecet 'movement' ficir 'idea'
seciz 'eight' spicer 'speaker'
hycym 'judgment'
b. Velars
oku 'read' ka 'wash'
istakoz 'lobster' akɨl 'intelligence,
ag 'bosom' jakut 'emerald'

The examples in (10) are also consistent with the generalization in (8). Each form in (10a) has a dorsal stop as the first of two consonants in a postvocalic cluster, while in (10b), a dorsal occurs as the second consonant in a prevocalic cluster. In every case, the dorsal agrees in backness with the vowel adjacent to it, exactly as predicted by our generalization.


a. Palatals Velars
nectar 'nectar' boksitj 'bauxite'
icram 'offer' ukte 'ganglion'
mectup 'letter' daktilo 'typist'
reclam 'advertisement'
isticlal 'independence'
b. Palatals Velars
haɲɟi 'which' t͡ʃiŋko 'zink'
orcinos 'tunny fish' teleskop 'telescope'
gal 'occupation'

Even though they are consistent with our generalization, the forms in (9) and (10) still supply us with information we didn't have before. We saw earlier that dorsals at the edges of words agree in backness with an adjacent vowel. Figures (9) and (10) show that the agreement phenomenon is not limited to word edges. In particular, (10) confirms that medial dorsals behave like initial and final ones in that for agreement purposes, adjacency between the dorsal and a vowel on only one side (preceding or following) is sufficient. A medial dorsal doesn't need to be surrounded by vowels of the same backness in order to agree, even though this happens to be the case in (9).

The conclusion that simple adjacency between a dorsal and a single conditioning vowel is all that is required for agreement raises another question that can't be answered by consulting the data we have seen so far. When only one vowel occurs adjacent to a dorsal stop, there is clearly no problem in identifying the vowel that conditions the quality of the stop. The question would then be: in cases where a dorsal stop occurs intervocalically, which vowel will the dorsal agree with? The forms in (9) are not informative in this respect, because in each case, the dorsal is flanked by vowels of the same backness. By adding the words in (11) to our sample however, we can see that when a dorsal stop occurs between vowels of different backness, the dorsal agrees with the following, not the preceding vowel.


Palatals Velars
acis 'echo' iikaz 'warning'
saacin 'calm' dekor 'stage, scenery'
bucet 'bouquet' ikon 'icon'
facir 'poor' zigurat 'ziggurat'

Our observations so far have been based purely on the surface distributions of the sounds we are examining. That is, we have called into evidence systematic gaps in the distribution of palatal and velar stops, namely, that they co-occur with different sets of vowels. As it turns out, a number of alternations in Turkish provide additional support for the conclusions we have reached. In the first column of (12), a velar occurs finally in the unaffixed nominative stem after a back vowel as expected. However, in the second column, where the velar precedes the accusative suffix -i, we find a palatal instead.


Nominative Sing. Accusative Sing.
ittjifak ittjifaac-i 'alliance'
istimlak istimlaac-i 'expropriation'

(Accusative is the case a noun phrase has when it occurs as the object of a verb.)

These alternations support our hypothesis that when there is a choice of adjacent vowels, it is the following vowel that determines whether a dorsal will be velar or palatal. This pattern is not predicted by the generalization in (8), but can be accommodated by making the addition in (13).

(13) Addendum to interim generalization 2: When a dorsal stop is both preceded and followed by a vowel, the stop agrees in backness with the following vowel.

We have just shown that specifying mere adjacency in our generalization is not specific enough without adding (13) to cover intervocalic dorsals. However, in another way, a claim of adjacency is too specific. In (14) we see that a word final dorsal agrees in backness with a preceding vowel even though they are separated by an intervening consonant. It is also interesting that the optional intervening consonant allowed for in the second context for Postvocalic Velar Fronting is a sonorant.


Palatals Velars
sirc 'circus' burk 'sprain'
iljc 'first' kɨrk 'forty
reɲc 'color' zamk 'glue'
deɲc 'equal' fark 'difference'
cyrc 'fur'

In a sense, the pattern in (12) is surprising, as we saw that a dorsal did not agree with a preceding vowel across a consonant in (10b) (e.g. orcinos 'tunny fish', not *orkinos). The difference is that whereas in (10a), the dorsal can agree with a following vowel, there is no following vowel to agree with in (12). This seems to be an important difference: looking over all the data we've seen, whenever a dorsal is immediately followed by a vowel, the dorsal agrees with that vowel. When a dorsal is not immediately followed by a vowel, then the dorsal agrees with the nearest preceding vowel.

Now we've seen all the data we'll need to consider for the primary points we want to make in this section. Taking all of the observations we've made throughout into account, we can revise our generalization as in (15).

(15) Interim generalization 3: Within the class of dorsal stops, [+back] velars and [-back] palatals are in complementary distribution. The dorsal stop agrees in backness with a vowel in its environment:
a. When a dorsal stop is immediately followed by a vowel, the two have the same value for the feature [back].
b. When there is no vowel immediately following a dorsal stop, the dorsal and the preceding vowel agree for backness, and the two may optionally be separated by a [+sonorant] consonant.

prev | top | next