The unreliability of Afroasiatic etymologies

The fact that Semitic, Egyptian, Berber, Cushitic, and Chadic all belong to a single family - Afroasiatic - is fairly secure, based on striking correspondences in basic morphology. However, it is often not appreciated just how difficult it is to find reliable lexical comparisons between these families, and just how primitive the current state of AA reconstruction is. The easiest source of AA etymologies online is Militarev's database on Starling, so I'm going to pick on it for this post (Orel & Stolbova and Ehret reveal similar issues, but the latter doesn't even include Berber, and I'm focusing mainly on Berber entries here for convenience.)

Suspiciously many entries are listed as having a cognate in only one Berber language (eg earth, hide, skin, run away); given the general closeness of different Berber varieties, you would expect valid proto-Berber terms to be reflected in more than one place. However, these could always be right. Other issues are more serious.

In several cases, a single proto-Berber root is split across several AA ones, due to mistaken sound correspondences. For example:

• Proto-Berber *i-qăs "bone, (fruit) pit" is split between PAA *ʔayš/ʔawš- "ripened grain, corn" with Zenaga iʔssi (quoted without the glottal stop) "os; grain, graine, baie; comprimé, pilule, cachet, pastille; perle" (Taine-Cheikh), and *ḳ(ʷ)as "bone", with all other reflexes of *iqăs, even though Berber γ (<*q) commonly corresponds to Zenaga ʔ.
• Proto-Berber *ta-Hăli (> *ti-Həli) "sheep" is split between pAA *ʔayl "ram" and *bawil "ram", although Ghadames-Awjila v corresponds regularly to Tuareg h and other Berber Ø. (A couple of forms, like Figuig tili mistakenly glossed as "ram", have even somehow found their way into a third etymon, "proto-Berber" *laH!) The issue is alluded to in a cryptic comment under the Berber section of PAA *waʔil "wild goat/ram; antelope": "Pr. H No. 220 (and Kössm. 193): Ghdm., Audj. Hgr etc. te-hele < *tiHeli, which, on the contrary, is to be connected with *ʔayl- 'ram' 3061 (together with Brb. forms of the t-ili type), as *ʔ > h in Hgr, while *ʕ > Hgr 0".
• 
  Most reflexes of pan-Berber ikərri / akrar "ram" are assigned to PAA *kar(w)- "ram, goat; lamb; kid". (The Semitic parallels listed for this word are rather interesting.) But Zenaga ǝgrǝrh, pl. gurănh 'bélier' (Nic. 156), on its own, is given a supposed proto-Berber form *gur- "ram", corresponding to an AA form *(ʔa-)gʷar "kind of antelope; ram; goat". In fact, however, there is a common correspondence of Zenaga g followed by a sonorant to proto-Berber k (eg ägärgur "chest" = Siwi ikərkər, əməgyih "dine" = Kabyle iməkli etc), and this word is obviously related to the other Berber forms.

Another case is listed as doubtful, eg:

• Most reflexes of Proto-Berber *a-lăqŭm "camel" are under PAA *ʕalVḳ/g- ˜ *lVḳ/gum- ˜ *ḳalVm- "camel"; but the Zenaga one äyiʔm, with regular *l > y (in his source's transcription ǯ) and common *γ > ʔ as seen previously, ends up as PAA *gam-al- (?).
  

Similarly, unrelated forms may be grouped together due to accidental similarity, eg:

• Under PAA *kʷay(-t)- "hen; partridge; dove; chick" is listed a "proto-Berber" form *i-kaHi; but the Ahaggar form listed corresponds regularly to Niger Tuareg tekažit, Mali Tuareg tekazzit, Awjila təkažit "hen" (see Kossmann 2005:60), and as such is unrelated to the Ayr and Tawllemmet forms takəyya quoted.

Another problem is undetected loans; this applies especially in sub-Saharan Africa, where little work has been done on their impact. PAA *ʔa/iw / *waʔ "bull, cow" is supported by Tawellemmet hawu "cow", isolated in Berber and obviously borrowed from Songhay, cp. Zarma haw, Tadaksahak hawú; removing this from the etymology leaves only pan-Tuareg iwan "cows", with no evidence for the desired *H. PAA *bar "cereal, corn" is supported by Zenaga būru "bread"; but this word is isolated in Berber and widespread in West Africa (eg Wolof mbuuru, Soninke buuru, Bambara nbuuru, Peul mbuuru, Zarma buuru), and is more likely a loan from Wolof or Pulaar.

Interestingly, most of the problem cases I've noticed in this quick skim are related to agricultural terminology. I wonder if that has anything to do with the particular interest of such terms for archeologists motivating a more intense search for cognates.