A few days ago I got lent a copy of a recent book in Arabic by Othmane Saadi: Dictionary of the Arabic Roots of Amazigh (Berber) Words معجم الجذور العربية للكلمات الأمازيغية (البربرية) (Tripoli: Academy of Arabic Language 2007.) My reaction, in brief, is that it's unscientific jingoistic claptrap. But I happen to have friends (not linguists, of course) who take it seriously; and I am told that the author, a proud member of the Chaoui Berber Nememcha (Nmamša) tribe, genuinely believes his own theory. I will therefore try to explain as simply as possible where the book goes wrong.
His starting point is noting the existence of strong similarities between Arabic and Berber in the vocabulary and grammar (p. C: “90% of Amazigh Berber words are pure or Arabised Arabic, and the grammar of Berber agrees with the grammar of Arabic.”) This is substantially correct, and has been known for a long time (see, for example, Igor Diakonoff's Afrasian Languages, Moscow: Nauka 1988, or at a more basic level one of my first posts), except that 90% is a substantial exaggeration – many of the comparisons he puts forward are at best questionable, as will be seen below. But he claims that the explanation for these similarities is that Berber descends from Arabic. Not just Berber either, as he says on p. B: “The term Arabitic عروبية means the ancient Arabic languages which are wrongly called the Semitic languages and which branched out from the source language Arabic thousands of years ago, such as Babylonian, and Assyrian, and Akkadian, and Phoenician Canaanite, and Aramaic, and Himyaritic, and Sabaean, and Thamudic, and Lihyanite, and Ma'inic, and ancient Egyptian, and Berber, and others.” Linguists subscribe to a rather different explanation for the observed similarities: that Berber and Arabic (and all the other languages he listed, and many he doesn't list such as Hausa and Somali) are all descended from a single language, called for convenience Proto-Afroasiatic (Greenberg 1950), which was different (and probably about equally different) from any of them.
How would you choose between these two hypotheses? Well, if the original language was different from Arabic, then you would expect some original forms to have been lost in Arabic but kept in other languages. Oddly enough, Saadi himself gives evidence for exactly that: he links the Berber ur “not” to Akkadian ul (p. 12), and the Berber -as “to him/her” to Akkadian -šu (p. 12), and the Berber nəkk “I” to Ancient Egyptian ink and Akkadian 'anāku, none of which are attested in Arabic. Unless you believe that Akkadian and Berber each independently invented the same new forms, or that they are more closely related to each other than to Arabic – which Saadi (correctly) does not claim – you have to conclude that the common ancestor of Arabic and Berber included words like ur/ul for “not”, and 'anāku for “I”, and so on, and hence was different from what we know as Arabic, just as it was different from Berber.
So maybe this common ancestor was Arabic in a different sense: Saadi argues that it was originally spoken in Arabia, so Arabic would be the one language that stayed at home, and presumably got less affected by foreign influence. Unfortunately, he doesn't have much of a case. His first argument (p. 1) is frankly risible: “Europe and North Africa were covered with ice before [18000 BC], whereas the Arabian peninsula enjoyed a climate similar to that of southern Europe now. The ice melted in the former and drought hit the latter, so mankind left the Arabian peninsula and settled North Africa and southern Europe.” The quote he cites on this actually says nothing about North Africa, and for good reason: even at the last glacial maximum North Africa was never covered by ice (see map), and was if anything more habitable before 18000 BC than it is now. He also notes (p. 2) that Berber princes have long claimed Yemenite origins. Such claims are questionable for many reasons (the desire for prestige, the originally matrilineal traditions of many Berber tribes, and no pre-Islamic attestations) – but even if true, it would prove nothing about the language: people change their language all the time without changing their ancestry, as any emigrant can tell you. The rest of his argument is a hotchpotch of miscellaneous quotes which at best claim that various early North African peoples or languages or cultures originated in the Middle East; in a particularly ludicrous case, he blithely quotes Bousquet (1957) to the effect that the Berber language “came from Asia Minor” [Turkey!] None of these quotes so much as mention the Arabian peninsula.
In fact, the linguistic evidence means that Proto-Semitic may well have been spoken in Arabia and certainly was spoken in the Middle East, but the common ancestor of Berber, Egyptian, and Semitic was most likely located in Africa. You see, as noted above, these three language families are also quite closely related to Chadic (spoken mainly in Nigeria and Chad) and Cushitic (spoken around the Horn of Africa) – which means that 4 out of 5 branches of this family are native to Africa. It is more likely that one branch left Africa than that 4 branches each separately followed the same narrow path across Sinai or crossed the Red Sea. (For theoretical background, see Campbell 2004.)
In other words: whether the similarities this book gathers between Arabic and Berber are valid or not, they don't do anything to support the author's claim that Berber descends from Arabic. Do they at least have the merit of being valid comparisons? Sometimes, but not with any consistency. Many of his comparisons look rather far-fetched, eg on p. D:
taməṭṭuṯ “woman” < Ar. ṭāmiṯ طامث “menstruator”
argaz “man” < Ar. rakīza(tu l-'usrā) ركيزة الأسرى “pillar (of the family)”
ixəf “head” < Ar. xf' خفأ “appear”, because the head stands out
tadaγt “armpit” < Ar. daγdaγah دغدغة “tickling”
alγəm “camel” < Ar. luγām لغام “the foam that comes out of camels' mouths”
Many others are clearly genuine loanwords, often featuring sounds that cannot be reconstructed for Proto-Berber, though I don't think many of these are original suggestions, eg:
(p. D) axərraz “cobbler” < Ar. xaraza خرز “to sew leather”
(p. H) abrid “road” < Ar. barīd بريد (confirmed by the Tuareg pronunciation of this word, abărid)
(p. 38) ləbṣəl “onion” < Ar. baṣal بصل (Siwi happens to preserve an older word for "onion": afəllu)
(p. 78) taħzamt “belt” < Ar. ħizām حزام
A couple are known Phoenician loanwords:
(p. 57) agadir, ažadir "wall" - Ar. jidār جدار
A few are well-known Afroasiatic cognates, and scattered among them may be other valid cognates:
(p. 250) iləs “tongue” - Ar. lisān لسان
(p. 110) iđammən “blood” - Ar. dam دم
(p. 292) tiqqad “burning” - Ar. wqd وقد
But the book makes no attempt to distinguish between words taken from Arabic comparatively recently and words inherited from the common ancestor of Berber and Arabic, and seems to assume that any word found in both dialectal Arabic (Darja) and Berber must automatically be originally Arabic, rather than possibly being a borrowing from Berber into Arabic. There is a well-known technique for sorting out inherited cognates from loanwords from coincidental similarities: sound correspondences. Sounds don't usually change at random: they change systematically, just as all j's in Egyptian Arabic become g. You establish which Berber sounds normally correspond to which Arabic ones under what circumstances, based on looking at what happens in the clearest cases; that gives you a standard by which to judge the doubtful ones. Saadi has made no effort to do this, and the unfortunate result is that in his comparisons the chaff far outweighs the wheat.
Berber and Arabic both descend from the same language, but that language was neither Berber nor Arabic, and probably didn't come from Arabia - and if you want to know about that common source, then you'll learn more from the works of Diakonoff or Greenberg, or even from more problematic sources like Orel and Stolbova 1999 or Militarev's online database, than from Saadi 2007.