PhDniX's blog

Proto-Berber Verbs: C1=C2 verbs

Similar to Semitic language, a very large amount of Berber verbs have three root consonants. This made it incredibly easy to integrate the basic Semitic verb stem from Arabic into its own system, which many Berber language have therefore also done.

But besides three-consonantal there are many verbs that have only one or two obviously present root consonants. Inspired by Semitic reconstruction, it has been assumed that these 1- or 2-consonantal roots are 'weak' roots, that have lost an ancient consonant. With recent discoveries, it has been shown that this is certainly true for a large group of these verbs, which originally contain *ʔ or *β, of which reflexes can still be found in some Berber languages. But there are some some roots where this original weak consonant is less obvious.

One of the verbal groups that may have a lost radical is the roots of this first apophonic class (henceforth A1 roots)that we discussed in the last blogpost, which have a long initial consonant, in the place where 'sound' A1 roots have two consonants, compare the paradigms of *ălmŭd 'to learn' (which is a very early loan from Punic, or Hebrew) and *ăkkŭs 'to take away'

aor. *ălmŭd  ~ *ăkkŭs

pf. *ŭlmăd ~ *ŭkkăs

impf. *əlămmăd ~ *ətăkkăs

Already we detect an interesting difference between the two verb types. Rather than the expected **əkăkkăs which you might expect if the long *kk was simply the result of the first and the second root consonant being identical, it seems like this long consonant cannot be broken up. Instead a 'dummy' consonant t is inserted to retain the əCăCCăC pattern.

But the situation becomes even more confusing when we get to the causative forms. From the verbs with an initial long vowel (from an original *h?) we were able to deduce that the causative vocalisation was ës-ïCCëC, compare the following Tuareg forms:

*s-əlməd 'to teach' ~ *s-igəm 'to cause to pull water from a well' ~ *s-ukəs 'to cause to take away'

As we can see, for some reason the long consonant kk has been shortened, and a vowel *u shows up in the place where we would expect a *ï. It has been proposed, and Prasse follows this proposal, that the original initial consonant of these types of roots was a *w, this idea is not unreasonable, but the exact reconstruction is actually quite difficult to determine. Let's assume that this hypothesis is correct, the Proto-Berber forms would then look as follows:

aor. *ăwkŭs

pf. *ŭwkăs

impf. *əwăkkăs

impf. neg. *əwĭkkĭs

caus. *əsĭwkəs

First we must assume assimilatory rules *ăwC > ăCC and *ŭwC > ŭCC. This is reasonable.But note that we must assume that the assimilation did not happen in the sequence *ĭwC, which must have yielded *uC. I find it difficult to justify, phonetically that *ŭwC yielded *ŭCC and *ĭwC yielded *uC. I would sooner expect it the other way around.

There are other problems: now that we reconstruct a third consonant *w, it is actually a lot more difficult to argue why the imperfective formation would use the *t- element.

We must think of some kind of reason why *w was not allowed to be the onset of a syllable, which is hard to argue when you have deictic elements like *wa and *wi which can probably reconstructed for Proto-Berber.

An ad-hoc soundlaw *əwă > *ă may solve this. This would yield **ăkkăs for the imperfective, then perhaps to reinstate the əCăCCăC syllable structure, the *t element was added, to then yield *ətăkkăs, but there is absolutely no basis to assume such a sound law except to explain this one verb form, which is unsatisfying. This explanation, however is attractive on one point. If we would reconstruction impf. *ətăwkăs neg. impf. *ətĭwkĭs we would have serious difficulty explaining why *ətĭwkĭs yields təkkəs in Tuareg and not **tukəs, as we find in in the causative *əsĭwkəs > sukəs.

It also is not particularly plausible that *əwă would lose intervocalic *w while other instances of intervocalic *w would be retained For example *awŭy (= *ăhwŭy?) 'to carry'

While this explanation works out somewhat elegantly, we've mostly moved the problems into a realm of internal reconstruction, which cannot be falsified at all. Moreover, the sound laws that are needed to get these forms are not particularly plausible, phonetically.

The next issue we'll discuss will be the probably with *h initial verbs, and whether we should really reconstruct a radical *h for Proto-Berber, or simply a long vowel.

 

Some thoughts on Proto-Berber short vowels: Verbs

Last time I talked about the short vowels in the noun system, and hopefully showed that I'm of the opinion that three short vowels probably should be reconstructed, but that it can be very difficult to distinguish *ĭ from *ŭ.

The huge Manuel de grammaire Touarègue by Prasse (1972-4) is not only the most comprehensive grammar of a Berber language ever written, it also has an enormous amount of notes and discussion on the Proto-Berber reconstruction. His reconstructions at the time were perhaps a little Touareg-centric (which is something I'm sure he would not deny), but it's one of the only real examinations into Berber verb and noun formation.

I want to spend some time on the vocalisation he reconstructs for the basic triradical verb. For Proto-Berber, Prasse reconstructs a Imperative/Aorist, Perfective, Negative Pefective, Imperfective (In his terms: Intensive Aorist) and a Negative Imperfective. With evidence of the Eastern Berber languages (esp. Ghadames, Sokna and Awjila) it seems like we can also reconstruct a Future, and that the Imperative and Aorist were perhaps separate. These different stems all have different vocalizations.

The basic triradical verb in the Aorist, Perfect and Imperfective (the negative stems can mostly be derived from the positive, but may warrant a separate blogpost at some point) is reconstructed by Prasse as follows:

aor. *ălmĭd

pf.  *ŭlmăd

impf. *ĭlămmăd

(I'm allowing some changes to Prasse's reconstructions, he actually reconstructs ĭlämmäd with an initial long vowel ĭlammăd. This is the 'long imperfective', other dialects also have a 'short imperfective' which has the formed cited abovem and this form is more in line with other Berber languages. The long imperfective (and long/intensif perfective, for that matter) seems to be a Touareg internal development)

The fact that Prasse was able to reconstruct the vocalisms from the aor. and pf. is quite impressive, because Touareg has a complex system of vowel harmony that often obscures this distribution. In Ghadames however, this vocalization is perfectly present: aor. ălməd pf. əlmăd impf. əlămmăd.

But neither Ghadames nor Touareg retain the short vowel contrast between *ï and *ŭ, so how did Prasse arrive at these vocalisms? He arrived at this system by an ingenious system of lost consonants, akin to the Proto-Indo-European laryngeals, which would cause lengthening of the vowel in the same timbre as the short vowels.

For example, the Touareg verb agəm 'to draw water from a well' is attested in Touareg as follows:

aor. agəm

pf. ugăm

impf. əttagăm

Which Prasse reconstruct with an initial root consonant *h

aor. ăhgĭm

pf. ŭhgăm

impf ət(t)ăhgăm

Notice that the imperfective has a morpheme t and no lengthening of the medial vowel. But the vocalic pattern is the same ə-ă-ă. The t morpheme was perhaps introduced to retaint the C-CC-C consonantal pattern, while avoiding a *h in word-intial position. For a similar explanation, see Lameen Souag's post from long ago.

*ăh yields a, *ŭh yields u

From this you'd probably expect that *ĭh yields i, an that indeed seems to be the case. The causative formation, which is derived from the verb with an *s(s)- prefix and a ə-ə-ə vowel pattern in the Aorist looks as follows in the two verbs discussed so far: əssəlməd  and əssigəm

It would seem that these forms then point to *əssĭlməd and *əssĭhgəm respectively.

These rules in the case of *h are Pan-Berber, there is therefore not a single Berber language that retains this consonant. This is therefore somewhat comparable to the situation of Proto-Indo-European before Hittite was discovered.

But there is a problem with this reconstruction (something Prasse seems to have noticed as well as he has changed in reconstructions found in his Bilan sur les laryngales Protoberbère in 2010.) The verb agəm is attested with labialisation in Tashelhiyt agʷm and Kabyle agʷəm: , which would indicate that the second vowel is in fact *ŭ. Aor. *ăhgŭm (and by extension, probably also *ălmŭd). Labialisation is lost in the perfective of both language ugm ~ ugəm which can be either explained by the direct contact with u or because the second vowel in the root is ä. I don't think that the causative is attested in either of these languages.

Seeing this, it may be unclear why Prasse ever reconstructed *ĭ as the vowel if the aorist, rather than *ŭ. This is came to be because at the time he only reconstructed one consonant *h, which we now know should be reconstructed as three separate consonants.

*h, lost everywhere, lengthens vowel

*ʔ, lost everywhere but in Zenaga, but clearly stayed present past the Proto-Berber stages, since other languages show vestiges of this consonant (Most notably Ghadames, but also Tetserrét and Nefusa Berber).

*β, lost everywhere but in Tuareg (where it is an h), Ghadames (where it is a β), Awjila (where it is a v) and Zenaga where it's found as vowel length (vowel length originally cause by *h is lost in Zenaga).

He needed a final *ĭ as wel as a rare final *ŭ to distinguish verbs with a word-final long vowel from each other.

eg. ăls 'to dress' has a variable final vowel in Tuareg which alternates between ø, i and a, while əmdu 'to finish' has a final vowel that alternates between u and a.

aor. ăls(ø/i-) < *ălsĭh

pf. əlsa < *ălsăh

impf. lăss(ø/a-) < *əlăssăh

cf.

aor. əmdu < *ămdŭh

pf. əmda < *ŭmdăh

impf. əməddu !< *əmăddăh (the impf. vocalism does not agree with the general system.

But, with the discovery of the glottal stop, these two apophonic classes no longer need to be reconstructed with the same final consonant. Zenaga clearly has a glottal stop in the final position of the verb 'to dress': ätʸšiʔ 'to dress' (< *ălsəʔ), while the cognate of the second clause is probably: uṃdih 'to be used up' (< *əmdəh).

Interestingly, this second verb does not have the initial *ă in the aorist that we would expect in Zenaga. And this is in fact also the case in Ghadames əmdu 'to finish'. Ghadames clearly shows that this verb in fact belongs to a different apophonic class (Kossmann's (2013) 'second apophonic class'), which has all *ə in every vowel slot, and as a result, explains the unexpected imperfective formation in Tuareg.

Verbs with a final long vowel (or final *h) are in fact very rare, and this is one of the few verbs that can be reconstructed with it in Proto-Berber. The only other verb in this class in Prasse's reconstruction is Touareg əlku 'to despise', which, if one compares it to Mali Touareg turns can be shown to have a final consonat h < *β which is lost in the dialect Prasse described at the time (cf. Zng. äykīh 'to despise'; Ghd. ălkəβ 'to despise').

So now that this opposition is gone from Proto-Berber, we can realistically reconstruct simply *ălsŭʔ for verbs with a variable final long vowel in the basic triconsonantal verb class.

One might be worried about the fact that ŭʔ yields ø- or i-vowels in Berber, for example Ghadames:

3sg.m. y-ăls 3pl.m. ălsi-n

But this is not problematic. As pointed out, the *ʔ was only lost in the Berber languages after the Proto-Berber stage. It is not unrealistic to think that it was in fact only lost in after *ĭ and *ŭ merged into *ə. Then this new high central short vowel in the sequence *əʔ was lengthened and could merge with either of the two high long vowels. It merged with *i eventually.

As it stands, I think it is therefore fairly safe to reconstruct the following apophonic schemes for the First Apophonic Class:

aor. ăCCŭC = imperative

pf. ŭCCăC = future (neg. pf. ŭCCeC)

impf. əCăCCăC (neg. impf. əCĭCCĭC) (If C1 = *h, *ʔ , prefix *t-, and confirm the pattern to the right consonantal scheme)

There's still plenty of complexities in some of the subtypes of this first apophonic class. For example verbs with an initial *ʔ need special attention, and verbs with C1=C2 need special attention. Finally the verbs with only two consonants and no trace of a lost *h/*ʔ  need to be discussed, and particularly, its perfective/negative imperfective.

For brevity I have left out the complete discussion of the reconstruction of the negative perfective and imperfective. Hopefully with the explanation above, you will understand why Ghd. aor. agəm pf. ugăm neg.pf. ugem impf. əttagăm neg.impf. əttigəm points to that reconstruction.

Finally I should close this post with a reference to an article which inspired and influenced much of this post:Maarten Kossmann. 2001. The Origin of the Glottal Stop in Zenaga and its Reflexes in the other Berber Languages. Afrika und Übersee 84.

Some thoughts on Proto-Berber short vowels: Nouns

I've been on a bit of a Proto-Berber reconstruction binge lately. I've tried multiple times to get funding to write a long overdue a Berber Etymological dictionary. I've never received funding though, but as a on-the-side thing, I've been piecing together some basic stuff, and I've been thinking about how to present the data of Proto-Berber.

This also means I need to start making up my mind about 'stuff'. One of those stuffs is the short vowel system.

Most Berber languages either have a non-phonemic schwa, or a marginally phonemic schwa as the only short vowel, but a select group of Berber languages has a more elaborate system. Most importantly are Ghadames and Tuareg which retain both a short vowel ă and a ə. These vowels are essential to both nominal plural marking, and for forming the stems of the verb. In Tuareg the system is obscured somewhat because of vowel harmony, but nevertheless the system is present.

Two other languages which basically retain the short vowel system (but also merge the long vowels with the short vowels) are Zenaga and Tetserrét (although, in the latter the system is more difficult to piece together).

But for Proto-Berber, it seems necessary to even reconstruct a three-vowel contrast in the short vowel system: *ă, *ĭ, *ŭ. In Ghadames and Tuareg (and also Zenaga?) *ĭ, *ŭ merged into ə.

In Essai sur la phonologie du proto-berbère (1999), Kossmann argues, mainly on basis of vowel alternations found in the nominal plural system, that the *ŭ gave rise to labialisation of velars in Tashlhiyt and Kabyle and Midle Atlas Berber.

Internal plural in Berber are formed by the following rules.

1. The vowel of the last syllable is replaced with a.

2. If a vowel a precedes it, it changes to u, if a vowel u or i precedes it, it is unchanged.

In Tuareg and Ghadames, there is a third rule, parallel to the 2nd rule:

3. If a vowel *ă precedes it, it  changes to ə, if a vowel ə precedes it, it is unchanged.

e.g.

Tashlhiyt a-gadir pl. i-gudar 'fortified grain silo' (the first element of the word is an obligatory noun prefix found on (almost) every noun in Berber).

Ghadames ta-miḍăz-t pl. tə-miḍaz 'scissors' (the -t is a feminine singular suffix.)

Ghadames a-măzwar pl. i-məzwar 'a first one'

It just so happens that Tashelhiyt sometimes has velars that labialize in this exact environment where a changes to u.

ti-gmmi pl. ti-gʷmma 'house'

Other nouns never have labialization.

a-gždi pl. i-gžda 'pole'

Finally, some nouns always have labialization.

ta-gʷzzi-t (< *-əy-t) pl. ti-gʷzzay 'herd'

This distribution is perfectly analogous to the situation when a long vowel precedes the a of the plural, and thefore it seems possible to reconstruct short vowels in these case in a scheme that can be presnted as follows:

sg. a - X > pl. u - a | sg. i - X > pl. i - a | sg. u - X > u - a

sg. ă - X > pl. ŭ - a | sg. ĭ - X > pl. ĭ - a | sg. ŭ - X > ŭ - a

This basically means, that in the nominal system, the only time that we can distinguish short ŭ from ĭ is when: a velar directly precedes these short vowels.

But when we either find labialization throughout the singular and plural, or no-labialization throughout the singular and plural, we still cannot be completely certain of the vowel qualities.

There are only very few words that show variation in labialization in any of the languages that retain it, so this is definitely not a regular system in these languages, but rather a relic. It stands to reason to assume that in many cases, either the non-labialized form was levelled through the paradigm or the labialized form was levelled through the paradigm.

The absence of labialization then, can never be used as evidence that there was a short *ĭ in the singular and the plural. The presence of labialization can be used as evidence that there was at least an *ŭ in the plural. And the presence of labialization in only the plural, is evidence that can be used for *ă in the singular, and *ŭ in the plural.

Moreover, if we look at the languages that have ă ~ ə variation between the singular and plural, it seems fairly safe to reconstruct a *ŭ for the plural.

As you can see, the conditions really have to be ideal for us to be able to reconstruct these Proto-Berber shortvowels in the nominal system, and the rest of the time, we are forced to reconstruct a placeholder phoneme *ə which may stand for both *ĭ and *ŭ.

If we start understanding more about how nouns are derived from roots, we might be able to make a few more educated guesses about the vocalization of words. But as it stands, we know very little about Proto-Berber noun formation.

The subject discussed in this post is a basic introduction to the Proto-Berber short vowel system. In the next post I will talk about the vocalization of the Proto-Berber verb stems, and how Prasse's reconstruction of the basic Triradical verb stems, should probably ammended, in light of recent developments of Proto-Berber reconstructions.

 

Archaic decades

Hey everybody, It's been a long time since I last blogged, nevertheless, I feel the itch again to start blogging, so I will.

First of all, a lot has happened since I last blogged. First of all, I received my PhD at the Leiden University for writing a grammar of the Awjila Berber language. This grammar would have never been finished at the speed that I've written it at if it wasn't for the help of commenters on my blog when I first started translating the Awjila texts. The thesis will (hopefully soon) be published in Berber Studies.

And now for something completely different! My native tongue, Dutch.

Different from many speakers of 'standard' Dutch, I retain a voicing distinction in initial /s/ versus /z/ and /f/ versus /v/. These sounds are only marginally contrastive in this position, and only very few minimal pairs can be found. One if fee 'fairy' (From French fée 'id.' ultimately from latin fata '(goddess of) fate' versus vee 'cattle' (Related to English fee and ultimately from PIE *peḱu- 'livestock').

In my speech, the initial voicing of the z or v generally agrees with the Dutch orthography, so zeven is pronounced [ˈze.və] 'seven' and vraag is pronounced [ˈvʀaχ], but there is one specific series of words, where my speech does not agree with the orthography, and that is the decades: veertig, vijftig, zestig, zeventig (40, 50, 60, 70), all of which I pronounce with a voiceless fricative. This is all the more puzzling, because I in fact pronounce vier, vijf, zes and zeven, veertien, vijftien zestien and zeventien with a clearly voiced initial consonant.

This apparent irregularity in my Dutch has always puzzled me, until recently when I ran into Middle Dutch writing of these decades. As it turns out, 60 and 70 originally had a voiceless element t in front of the now written initial consonant. The devoicing of the initial consonant of the decades is atually an archaism going back to the time that a voiceless stop preceded it. The stop has been lost, but voicelessness has been retained. This initial element t is obscure in meaning, but still visible in the Dutch word for eighty: tachtig.

More puzzling is the fact that I also retain a voiceless stop in veertig and vijftig, in Middle Dutch, no element t is attested in front of these words, and they're consistently written with v.

Analogy seems unlikely, as that would require an awareness of the speaker that the difference between z and s is one voice, and that this is the same distinction between v and f. From there they would have to deduce that if zeven [ˈze.və] is to zeventig [ˈse.və.təχ] as vier [ˈviːʀ] is to viertig **[vɪːʀ.təχ] then the initial consonant should be f! This does not stop taaladvies to posit exactly  this explanation.

It seems rather, that the Proto-dialect of my Dutch analogically spread the t- prefix to all decades (where in twintig and dertig it was obviously lost without a trace) when the t- had not yet disappeared. This Proto-dialect of my Dutch appears to be completely unattested in Middle Dutch (but I'd love to have an expert prove me wrong!).

 

Middle Dutch    My Dutch        Orthography    Proto-My Dutch

viertech              [ˈfɪːʀ.təχ]          veertig             *t-viərtəχ

viftech                 [ˈfɛif.təχ]          vijftig                *t-vīftəχ

tsestech            [ˈsɛs.təχ]         zestig              *t-zestəχ

tseventech        [ˈse.və.təχ]      zeventig         *t-zevəntəχ

tachtech            [ˈtɑχ.təχ]           tachtig             *t-aχtəχ