Thursday, April 14, 2011

Why *h1 and *h2 were not valid onsets in late proto-Berber

I've been working on my hopefully-forthcoming book about Siwi and thinking more about Berber laryngeals (see also Phoenix's recent post), two tasks that intermesh rather handily. Now Siwi has a wide range of strategies for forming the intensive (ie, in Siwi, the realis imperfective) of verbs, not obviously related to one another. But it is usually possible to predict which will be used from the form of the root. Basically, to recap the relevant page of my thesis, ignoring the fəl verbs discussed in the previous post and some other synchronic irregularities (U=consonant or full vowel; either count as a unit of the root):

Prefix t-:
- to geminate-initial roots
- to roots with the mediopassive prefix ən-
- to vowel-initial roots
- to vowel-medial (CVC) roots
Geminate U2:
- when U1 and U2 are distinct consonants, and U2/U3 is final
Put -a- after consonantal U3, changing any previous full vowels to a:
- when the last two units are distinct consonants (unless geminate-U2 / prefix-t applies), or
- when U2 is a full vowel (in which case prefix-t also applies)
Suffix -u:
- to geminate-final roots

Can we further simplify these conditions? In particular, what do the rather disparate environments to which t- is prefixed have in common?

Well, Siwi, like most Berber languages, shows the so-called “mobile schwa” phenomenon – ie, the position of schwa is mostly predictable solely from the consonants and long vowels of the word. (Basically, you put a schwa between any two adjacent consonants followed by a consonant or word boundary, starting from the left cyclically.) This also means that the coda/onset status of a given consonant in a stem is predictable, and depends on the affixes – for example, the k is a coda in əktər “bring!”, but an onset in kətr-ax “I brought”. However, there are a few exceptions to this principle – clusters that cannot be broken up by schwa, or, equivalently, codas that do not become onsets. These include:
- geminates: geminates cannot be broken up by schwa, and the first element of a geminate is always a coda.
- mediopassive ən-: the cluster ən+C that it forms cannot be broken up by schwa, and the n is always a coda (except before the borrowed voiced pharyngeal ʕ.)

Full vowels are by definition not onsets (semivowels behave quite differently from full vowels in Siwi.) So we can reduce the first three conditions for t- to a single one: t- is used when the first element of the root is not an acceptable onset. The fourth condition seems to be separate.

The use of t-/tt- under two of three of the conditions we have unified is reconstructible for proto-Berber (mediopassive ən-, or at least its syllabic structure, is a borrowing from Arabic), so it would be reasonable to reconstruct the No-Onset condition for proto-Berber too. Geminate-initial roots were clearly already geminate-initial in late proto-Berber (although Prasse, probably correctly, reconstructs them as *w-initial for pre-proto-Berber.) However, vowel-initial roots come from at least two sources: roots with vowel length (pre-proto-Berber h?) and roots with a glottal stop. The distinction is preserved in Zenaga, and t- shows up there in both cases. And, as it happens, Zenaga only allows the glottal stop in coda position. So it seems probable that late proto-Berber too allowed the glottal stop only in coda position.

No comments: