Last night, I discussed in some depth the syllabification of the Bedouin-Type Arabic dialects.
I worked mostly from Owens' description, which got me a little caught up in a bind, as I had no way to explain why the feminine ending was behaving like a *-t-, rather than the expected *-at-.
Today I studied Cantineau's excellent descriptions of the Bedouin-Type dialects from 1936 and 1937 "Études sur quelques parlers de nomades arabes d'Orient" which might provide somewhat of a solution to this issue. Turns out that Cantineau has actually intimately studied exactly this question, and even casts it somewhat in a historical light.
Part of the dialects simply behave as if the *a vowel was there (e.g. ʕmūr) whereas others behave like the *a is not there (e.g. Rugga):
(1) *kalbat-ī> kal.bə.ti> kalbiti'my bitch' (ʕmūr)
(2) *kalb-t-ī> kalb.ti > kalibti 'my bitch' (Rugga)
In front of suffixes that start with a consonants, all the dialects are in agreement that the suffix is at (as is the case in Douz; Benghazi points to *-it-, but this is presumably an analogical development)
(3) *kalbat-nā> kal.bat.na> kalbatna'our bitch'
We are then confronted with parts of the dialects syncopating the a vowel, but only in the -at- ending in this context, while other dialects do not.
Both dialect groups agree however on the treatment of the *baqarat- type words, and they differe in this respect from Benghazi: The first and third short vowel are always syncopated:
(4) *baqarat-ī > bə.gə.rə.ti> bgurti 'my cow'
(5) *baqarat-nā > bə.gə.rat.na> bguratna 'our cow'
Words of this type, then, display exactly the kind of alternation that we find in, e.g. the Rwāla and Benghazi dialects in all feminine nouns. Could it be that this allomorphy was expanded to all nouns from this, admittedly small, but not insignificant group of nouns?
We would expect the same vocalism in the verb when suffixed by pronominal suffixes. Here we find that several dialects have developed secondary lengthening of the vowel or consonant to retain the vowel of the -at suffix:
(6) *katabat-ah> kə.tə.bə.tah> ktibta 'she wrote it' (ʕmūr)
(7) ktibāt-a'she wrote it' (Benghazi, Rugga)
(8) kitbátt-a 'she wrote it' (Ḥadīdīn)
These lengthenings have also spread to environments where the *at would not have been regularly lost.
(9) *šāfat-ah > šā.fə.tah> šāfata 'she saw it' (ʕmūr)
(10) šāfāt-a 'she wrote it' (Benghazi, Rugga)
(11) šāfatt-a 'she wrote it' (Ḥadīdīn)
So long story short: It does not seem likely that there was a feminine ending *-t- around. The ʕmūr dialect (along with several others) presumably retain the original situation, which due to syncope rules developed a *-t-/-at- alternation in certain paradigms. Several dialects have expanded this alternation to nouns where it would not regularly occur.
As for the syncope rule that I formulated yesterday. It seems better to start from the rule that we actually find in the peninsula:
*cv.cv.cv. > ccv.cv
*cv.cv.cv.cv. > ccv.ccv (the syllable boundary question is up for debate here of course).
More elegantly formulated: From left to right, for every pair of two short open syllables, syncopate the first vowel.
I'm sure that people who do metrical phonology would have a field day describing this phonologically productive distribution.
To account for the Benghazi situation, we have to assume a somewhat unusual metathesis of the medial vowel:
*ccv.ccv >cvcccv > cvcvccv
There are some other ways to envision the Benghazi situation. Owens (1984), for example envisions that *cv.cv.cv.cv. simply syncopates all vowels, and that the outcome of vowel epenthesis in *cccc is always cv.cvcc. It's rather difficult to formulate testcases outside of these very specific CaCaCat- + vowel initial suffix nouns; making the 'correct' interpretation very difficult to formulate.