UT

The breadth problem

The analysis we have developed for the deletion of the consonants /h/, yod, and /v/, and the vowel lengthening that accompanies their deletion in some of those cases, is descriptively adequate in that the analysis accounts for the results we find, and does not produce unattested forms. However, in a number of ways, the analysis we have provided is not satisfying.

Linguists prize efficiency in analyses. In efficient analyses:

  • as much work as possible is accomplished using the least possible amount of formal machinery.
  • different tools have non-overlapping functions; no element of the analysis should repeat work done by another element.

Another important criterion is generalizability: we want our analyses to have the breadth required to make appropriate connections between similar phenomena, without predicting non-occurring outcomes.

Our analysis of CL is inefficient because it fails both of the criteria we have just given. Let's look first at the rules of compensatory lengthening, repeated below in (20).

(20) Consonant deletion with vowel lengthening
a. Yod deletion with CL: /V[-back]j/ → V:[-back]Ø / ___ C[+son]
b. /v/ deletion with CL: /V v/ → V:Ø / ___ C
(Conditions: V (preceding /v/) is labial, or C is labial)
c. /h/ deletion with CL: /Vh/ → V:Ø / ___ C
(Conditions: C is either [+sonorant] or [-sonorant] and not [-continuant].)

A number of elements are repeated in the CL rules in (20). As we've said before, each involves the deletion of a consonant, and in each case, the vowel before the deleting consonant is lengthened. At this point, we'd like to draw attention to the fact that in all of the CL rules, the C being deleted (and the V being lengthened) stand before a consonant of some type. An analysis in which highly similar phonological processes were not stipulated by each of several distinct rules would be much more efficient than the analysis we have proposed. Despite the evident similarities, however, it was not possible to write a single rule of CL because conditions on the identity of the consonant following the deleting C can be are different from rule to rule:

  • The yod rule requires that the vowel preceding yod be [-back] and that the consonant after yod be [+sonorant].
  • For the /v/ rule, the C following /v/ may be labial, but doesn't have to be, if the preceding vowel is labial.
  • For the /h/ rule, the C following /h/ is either [+sonorant], or it is a fricative.

Now, we have also proposed consonant-specific sets of deletion rules. Each set of deletion rules, one set for yod, one for /v/, and the third set for /h/, has one (or more) vowel lengthening rule(s), and one (or more) rule(s) with no vowel lengthening. The pairs of rules we have proposed for yod and /v/ are repeated in (21) and (22).

(21) Yod Deletion
a. No lengthening: /j/ → Ø / V[-back] ___ X[+son]
b. With lengthening: /V[-back]j/ → V:[-back]Ø / ___ C[+son]

(22) /v/ Deletion
a. No lengthening: / v / → Ø / V ___ X
(Conditions: Either V or X is labial; X may be C or V)
b. With lengthening: /V v/ → V:Ø / ___ C
(Conditions: V (preceding /v/) is labial, or C is labial)

The conditions required for yod deletion are identical, and the conditions on /v/ deletion are identical. Because the same work (in some respects) is done by both rules in each, we miss a generalization: having two rules in each case ignores the obvious similarities.

The cases of /h/ deletion is more complicated, but to some extent, the same general point can be made. We need several rules of /h/ deletion, because it isn't possible to express all of the environments for deletion as a single, coherent environment. The CL rule for /h/ in (20c) should really be split into the two rules in (23).

(23) /h/ Deletion with CL
a. /h/ deletion with CL: /Vh/ → V:Ø / ___ C[+sonorant]
b. /v/ deletion with CL: /Vh/ → V:Ø / ___ C[-sonorant] & C is not [-continuant]

Notice that with /h/ deletion, as with yod and /v/ deletion, vowel lengthening occurs only when /h/ precedes another C. The non-lengthening rules for /h/ deletion we proposed earlier are repeated in (24).

(24) /h/ Deletion with no V lengthening
a. /h/ → Ø / (i) V ___ V
b. /h/ → Ø / C[-sonorant,-voice] ___ V

We will need the /h/ deletion rules in (23b) and (24b) in any event. However, notice that the rules in (23a) and (24a) are very similar: once we realize that vowels are [+sonorant], we can restate (24b) as (25), which states that /h/ can be deleted before any sonorant segment.

(25) /h/ Deletion occurs after V and before any sonorant segment, C or V (subsumes (23a) and (24a)).
/h/ → Ø / V ___ X[+sonorant]

In summary, our analysis is quite inefficient because as we have seen, (i) We have several rules that repeat the conditions and effects of consonant deletion and compensatory lengthening, and (ii) we also have pairs of rules that delete consonants in similar environments. One rule in each pair produced a long vowel, while the other does not.

At first glance, it seems that we might achieve greater generality and efficiency by separating the consonant deletion and vowel lengthening functions of the compensatory lengthening rules we have proposed. There would be several advantages to this approach. One is that we could cut the number of deletion rules we'd need almost in half. In fact, we would need only the general rules in (21a) and (22a) to account for all cases ofyod and /v/ deletion. The more complicated case, /h/ deletion would require the rules in (24b) and (25). The CL rule in (23b) would be replaced with (26).

(26) /h/ → Ø / ___ C[-sonorant] & C is not [-continuant]

In the new analysis we are considering, the general rule of Compensatory Lengthening in (27) could be adopted to cover all the cases of vowel lengthening we have seen.

(27) Compensatory Lengthening
V → V: / ___ C

An important advantage of the system of rules we are now considering is that it treats lengthening as independent of consonant deletion so that lengthening can be expressed as a single phenomenon that applies in a number of cases which share structural similarities. One of the most important is that CL applies only before a consonant, and not a vowel. This condition is now written into the CL rule, and does not need to be specified as a special condition on deletion, in most cases. Thus, the new analysis appropriately connects similar phenomena in a general way, and is efficient in that we no longer have sets of repetitive rules.

While the alternative analysis we have just constructed has undeniable advantages, it encounters an insurmountable problem. The CL rule in (27) lifts vowel lengthening out of the consonant deletion context, a move that intuitively has merit. The analysis is successful in that it will prevent vowel lengthening from occurring in words like de.il, from underlying /dejil/.

(28) A derivation: CL follows consonant deletion
Phonemic representation / ejlyl / / dejil /
Syllable structure ej.lyl / de.jil
Yod deletion e.lyl de.il
Comp. Lengthening e:.lyl n/a
Surface representation [ e:.lyl ] [ de.il ]

Note that this analysis works only if the deletion rule precedes Compensatory Lengthening. If the lengthening rule applied first, then we would find long vowels not only in cases where it should apply, but also in /dejil/. If deletion has not already applied, then the vowel /e/ should be lengthened, as it stands before a consonant, /j/, incorrectly yielding *de:.il.

Therein lies the fatal flaw of the new analysis: the rule of compensatory lengthening in (27) predicts that any vowel occurring before any consonant should surface as long. Clearly, this massively over-generates, as it predicts non-occurring outcomes on a systematic basis. Under this analysis, quite simply, the only short vowels in Turkish should occur in word-final position. This is clearly not the case.

In the next section, we will construct an analysis that builds in the advantages of the alternative we have just considered, while avoiding its major flaws.


prev | top | next