&WJEFODFÛBOEÛ$PVOUFS&WJEFODF

Û'FTUTDISJGUÛ'SFEFSJLÛ,PSUMBOEU Û7PMVNFÛ
44(-Û Û"NTUFSEBNÛÛ/FXÛ:PSLÛ3PEPQJ Û Û

THREE IRREGULAR BERBER VERBS:

‘EAT’, ‘DRINK’, ‘BE COOKED, RIPEN’

MAARTEN KOSSMANN

Berber verbal morphology and apophony are well-known for their com-
plex and to some degree irregular behavior. However, while the choice of
apophonic pattern is not always predictable, the basic consonantal scheme
of verbs remains the same. Exceptions to this are mostly the result of low-
level phonological rules or phonological innovations, such as the devel-
opment of w to u under some circumstances, and the loss of intervocalic
*b in most Berber languages (Kossmann 1999: 120-125).
Reconstructible Proto-Berber verbs in which the consonantal shape
changes under apophony are very rare. In this article, I will study three
such verbs, which may represent an old pattern in the language, viz. 
‘eat’, sw ‘drink’, and nw ‘be cooked, ripen’. These three verbs have
irregular forms in the Imperfective 1 and in the verbal noun. This can be
illustrated by the following forms from the Beni Iznasen dialect (Eastern
Rian, Northern Morocco, field notes by the author):

Aorist Imperfective Verbal Noun

 tttt (maa) ‘eat’
sw sss ti-ssi ‘drink’
nw tnnna ta-nnni-t ‘be cooked, ripen’

An example of consonantal irregularities in the verbal noun of ‘eat’ is

found in Mali Tuareg te-tte.
The three verbs belong to the same formal verb class, which has a
final vowel whose quality is dierent with dierent persons in dierent
apophonic stems (for an overview, see Kossmann 1994). This class was
baptized the class of “thèmes à alternance post-radicale (groupe A1)”,

1 I use the term “Imperfective” here as a translation of Galand’s “inaccompli”. In

Berberology, this aspectual stem is known under many names, among others “habitatif”,
“aoriste intensif”, “imparfait intensif”, “intensif”, “extensif”, “cursive”, “inaccompli”,
and “thème II”.
 .""35&/Û,044."//

types 65-71, by André Basset (1929: 58.), while the bulk of these verbs
was classified as class IA7 (also IA9, IA10, IA11) by Karl-G. Prasse (1972-
4). It has long been recognized as a verb class with a lost final radical
(esp. by Prasse 1972-4); recently analyzed data from Mauritanian Zenaga
have shown beyond doubt that the “alternating vowel class” in non-
Mauritanian Berber corresponds to the class of verbs with a final glottal
stop in Zenaga (Kossmann 2001; Taine-Cheikh 2004). Therefore, the three
verbs under investigation can be reconstructed with a final glottal stop in
proto-Berber, i.e.  ‘eat’, sw ‘drink’, and nw ‘be cooked, ripen’.
Before focusing on the irregular forms of the Imperfective and the verbal
noun, a brief look at the history and development of the phonetic shape of
the Aorist of these verbs in the Berber varieties is necessary. A number of
elements need to be noted. In the first place, due to regular vocalization
procedures, in many Berber dialects the semi-vowel w is pronounced u
under some circumstances. For example, in the Imperative singular this
vocalization is extremely common.
The cluster *nw has a tendancy to produce assimilated forms. These
assimilations take dierent directions in dierent dialects:

nw >  w >  ( w in Rian,  in Tuareg 2)

nw > m
.m .
w
(Ouargla)
nw > nn (Ghadames, Gourara)
nw > ww > bb w (ww attested in many Algerian dialects, bb w in
Great-
er Kabylia)

These assimilations changed the verb *nw to a verb with a mono-

consonantal geminate. As this represents a dierent verb type with a dif-
ferent apophonic template in the Imperfective, this has lead to a rearrange-
ment of the Imperfective morphology in a number of dialects.

2 Foucauld (1951-2: 1282) has <eñ> in Ahaggar Tuareg. In Foucauld’s transcription

<ñ> normally stands for a palatal nasal rather than velar nasal, which is transcribed
<ṅ>. It is not clear to what extent Foucauld’s transcription represents a mishearing.
One other word that in other Tuareg languages has  is transcribed <ñ> by Foucauld,
<ña> ‘brother’ (1951-2: 1445). Both cases probably go back to *nw. Therefore, whatever
Foucauld’s notation may represent, there is no need to posit a root ny rather than nw,
as done by Prasse (1972-4, Vol. III: 125).
 5)3&&Û*33&(6-"3Û#&3#&3Û7&3#4 

More dicult is the case of  ‘eat’. Most dialects have forms with , 3
which in most cases behaves like a geminate. By regular developments,
some varieties have changed  to . Some Tuareg dialects (Ahaggar and
Mali Tuareg) have a dierent form, k. Because of its restricted geograph-
ical distribution, and because of the irregular Imperfectives and verbal
nouns (see below), I consider these forms a dissimilation from earlier t
(or , which would represent the same pronunciation) (for a dierent
point of view, v. Prasse 1972-4, Vol III: 110). One may note that mod-
ern Berber varieties have no objection to words with the cluster k, cf.
Tashelhiyt km ‘enter!’.
The phoneme  is rare in most Berber languages, 4 and the verb ‘eat’ may
be the only case which can be reconstructed for Proto-Berber. Elsewhere
I have proposed an analysis of the equally rare cases of other geminated
palatals that can be reconstructed into Proto-Berber. According to this
analysis  < *yy,  < *zy, while no reconstruction of  is provided
(Kossmann 1999: 225; 235). One may note that there is no counter-evidence
against considering  as being derived from *sy; on the other hand, there
is no positive evidence for such a derivation, either.
It is tempting to consider * as derived from a consonant cluster as well.
One could imagine, for example, a pre-Proto-Berber reconstruction *ty.
While there is no clear evidence for this, there is no counter-evidence to it
either: words with consonantal clusters featuring y as the second element
are conspicuously rare in the reconstructible part of the Berber lexicon.
The first verb under consideration is  ‘eat’. In the great majority of
Berber varieties, this verb has an Imperfective form, which lacks the palatal
element, *ttt. Reflexes of this form take the following shapes:
tatta() Zenaga (Nicolas 1953: 423 ttta)
(t)ttt Ghadames

3 In some varieties,  has regularly become . In Berber, there is no phonetic opposition

between mono-phonemic // and bi-phonemic /t/, as, for example, in Polish czy []
‘question particle’ vs. trzy [t] ‘three’. In most Berber varieties, [] cannot be dislocated by
schwa, which is a good reason to consider it a mono-phonemic or a geminate consonant.
As far as I know, Figuig Berber is the only Berber variety where [] is regularly dislocated
by schwa, e.g. ad i-t ‘he will eat’, which proves it is bi-phonemic in this dialect.
4 An important exception are the Zenatic forms with () which correspond to forms with

kk in other dialects (Kossmann 1999: 177-180). The verb ‘eat’, however, has () (> ) in
virtually all Berber varieties, and therefore has nothing to do with this set of cognates.
 .""35&/Û,044."//

> (t)ttt Mzab, Ouargla, Sened, Djebel

Nefusa, Rian, Figuig, Western
Algerian dialects, Kabyle, Aures
> tet Tetserrét
(t)ttta Augila, Middle Atlas
tatt(u) Tuareg

In a number of dialects, the form has been regularized, and the second
consonant is palatal. In such forms, the first (t)t- can be analyzed as the
Imperfective prefix, which is regularly used with verbs only consisting of
one geminated consonant (and a lost final glottal stop):
tkk y Elfoqaha 5
t Sokna, Siwa, Ayt Seghrouchen

A special form is Tashelhiyt tta (Aorist ), which seems to be the result
of a metathesis t > tt.
The Proto-Berber form *ttt can be interpreted in two ways: either the
initial t is interpreted as the Imperfective prefix (Prasse 1972-4, Vol. III:
119), or the form is considered the result of reduplication. In both cases,
the basic  is depalatalized. The reduplication solution is enhanced by
some verbal nouns:
t-ddi (< *t-tt) Zenaga
a-ttti Ghadames
te-tte, te-tte Mali Tuareg, Ahaggar Tuareg
> t-teti Tetserrét

Probably the following forms should be added:

t-uttu-t (< *t-utt(t)u-t ?) Gourara
te-t.t.e (< *te-tte) Ayer Tuareg

The second verb under consideration is sw ‘drink’. The Imperfective and
the verbal noun of this verb often show forms which lack any trace of the

5 Elfoqaha has the unusual Aorist form kk( y), which, I suppose, goes back to earlier *t

by a change of apical to dorsal mode of articulation.

 5)3&&Û*33&(6-"3Û#&3#&3Û7&3#4 

semi-vowel w. There are two major types of w-less forms in the Imperfec-
tive, something which would be represented in Proto-Berber as*sss and
something which would be represented as*t-ss with the Imperfective
prefix t-. The dialectal distribution is as follows:
*sss > sass Zenaga (Nicolas 1953: 368 sss)
> sss Ghadames
> (s)ss Bousemghoun (Sud oranais),
Mzab, Ngoussa (near Ouargla),
Sened, Western Algerian dialects
> ss Ayt Seghrouchen
> ssa Middle Atlas, Tashelhiyt
> sass(u) Tuareg
> e Tetserrét
*t-ss > (t)tss Ouargla, Figuig (dialectal),
Greater Kabylia
> dss Figuig (dialectal) 6

A combination of the two processes is found in Elfoqaha, which has tsäss.

In a number of dialects, forms with the semi-vowel have been intro-
duced. Forms of this type follow dierent strategies, such as employing
regular devices for the formation of the Imperfective of biradical verbs
(either following Prasse’s classes IA7 or IA5), or the prefixation of the
Imperfective marker (t)t-, sometimes combined with irregular gemination
of the first radical. Such forms include Augila uwwa, Gourara, Djebel
Nefousa ssaw, Sokna tssaw, Siwa t(i)su, Eastern Kabylian tssu.
Forms without the semi-vowel are also found in the verbal noun, where
one regularly finds forms derived from a reduplicated basis:
t-i (< *t-) Zenaga
a-sssi Ghadames, Beni Salah (Western
Algeria)
te-sse, te-sse Mali Tuareg, Ahaggar Tuareg
> tes.s.e Niger Tuareg

6 The prefix d- is uniquely found in some variants of Figuig Berber (esp. the village

Zenaga). It is also found in the verb ‘do’: Aorist yy (< *g), Imperfective dkk (< *dgg).
The reasons behind its voiced pronunciation are unknown.
 .""35&/Û,044."//

> tei Tetserrét

ti-sssi Rian, Western Algerian dialects
> tssi Djebel Nefousa
> tissi Figuig, Gourara, Aures, Middle
Atlas, Ayt Seghrouchen,
Tashelhiyt (also tissa)
ta-sssi-t Beni Iznasen
ti-ssi-t Greater Kabylia

The third verb under consideration is nw ‘be cooked, ripen’. The case
for irregularity is less strong in this verb than with the other two verbs,
as most Berber varieties have regular forms of the Imperfective and the
verbal noun. Still, there exist a number of dialects which have forms which
lack the labial element present in the Aorist. The irregular Imperfective
forms of this type are the following:
nnna Augila
tnnn Igli (Sud oranais)
tnnna Beni Iznasen (Eastern Rian), Beni Snous (Western Algeria),
Beni Salah (Western Algeria)
tnanna Temsamane (Central Rian)
tnana Matmata (Western Algeria)
ttnan Aures

Most of these instances come from the Zenatic dialects of Northern

Morocco and Northern Algeria (on Zenatic as a sub-group of Berber,
see Kossmann 1999: 30-32); however, their presence in Augila (Eastern
Libya) shows that they are not restricted to this area. The Zenatic forms
all show a combination of reduplication and the prefixation of t-. The
combination of the Imperfective prefix t- with consonantal gemination is
not uncommon in these dialects, e.g. Beni Iznasen r.wl ‘flee’, Imperfec-
tive: tr.akk wal. Tuareg dialects show an interesting half-way regularization
in the Imperfective of this verb. In some Tuareg dialects, the cluster *nw
has evolved in , and the Aorist of the verb ‘be cooked’ has become .
In these dialects, one finds Imperfective forms such as Iwellemmeden
nau and Mali Tuareg na (similarly Ahaggar <nâññ>). These forms
are probably the result of the introduction of the velar nasal into an
older reduplicated form *nann. Instead of being entirely reformed into
 5)3&&Û*33&(6-"3Û#&3#&3Û7&3#4 

a monoliteral geminated verb, the Imperfective form has preserved the

initial n.
Verbal nouns with loss of the labial element of this verb are attested in
roughly the same dialects as similar forms of the Imperfective. Note that
some of the items given are derivations from the causative stem (‘cooking’
being derived from the transitive verb ‘cook’).
t-nnna-t Augila
ta-nnni-t Beni Iznasen (Eastern Rian)
t-nnni Beni Snous (Western Algeria)
ti-nni (< *ti-nnni ?) Figuig
a-snanni (< Causative) Temsamane (Central Rian)
a-snnni (< Causative) Beni Salah (Western Algeria)

Tuareg dialects with *nw >  present a similar situation in the verbal
noun as in the Imperfective, viz. the introduction of  without loss of
the first n of the reduplicated form: Iwellemmeden, Mali Tuareg, te-ne,
te-ne (similarly Ahaggar <téneññé>).

To what extent and in which form can the items studied above be recon-
structed in Proto-Berber? 7 The Imperfective form of ‘eat’ is the least prob-
lematic, as it is attested almost everywhere in Berber, and as all reflexes
point to *ttt. The verbal noun is less generally attested, but its existence
in Zenaga, Ghadames, and Tuareg point to an early origin. The exact
reconstruction must remain uncertain, but something like *-tt(t)e would
match most forms.
The Imperfective of ‘drink’ has two dierent irregular shapes, one with
reduplication, one with the Imperfective prefix (t)t-. The form with redu-
plication is attested all over the Berber speaking territory, while the form
with t- only appears in the Northern Sahara (Figuig, Ouargla) and in
Kabylia. Therefore, there is no doubt that the reduplicated form is Proto-
Berber. The introduction of (t)t- in Figuig, Ouargla and Kabylia may have
been favored by the fact that s- is normally the prefix of the causative, so

7 The reconstructed forms basically follow the reconstructions in Kossmann (1999, 2001).

I have chosen to represent * and * as <>, as in practice their reconstruction is often

problematic. Moreover, I am less confident than before about the possibility of recon-
structing a ternary vowel contrast in the short series for Proto-Berber rather than the
binary contrast attested in Ghadames, Tuareg, and, it seems, Zenaga (Catherine Taine-
Cheikh, personal communication). The status of *e in the verbal nouns is unclear.
 .""35&/Û,044."//

that forms with reduplication present the danger of erroneous interpre-

tation as a causative verb. The form can be reconstructed as *sss. The
attested forms of the verbal noun point to something like *ss(s)e.
The reconstruction of irregular forms of the verb ‘be cooked’ is prob-
lematic, as regular forms abound in Berber, while the irregular forms
are mainly restricted to one specific area. However, they also appear in
Augila (Eastern Libya), which is in many points among the most conserv-
ative dialects of Berber, and which has no special connections to Zenatic.
Moreover, Tuareg has forms, which, although dierent from the Zenatic
/ Augila forms, point to an earlier, more similar, formation. As it is much
easier to regularize an irregular paradigm by analogy than it would be to
irregularize a regular verb, one can be confident that reduplicated forms
represent the older situation. The reconstruction of the Imperfective form
is complicated, as the Zenatic forms all have a combination of reduplica-
tion and the prefix (t)t-. The adjunction of the prefix (t)t- to Imperfectives
with reduplication is found with a restricted set of verbs in the Zenatic
dialects involved. It is extremely uncommon elsewhere, and probably con-
stitutes a Northern Zenatic innovation (Rian, Western Algerian Berber,
Aures). Therefore, the presence of t- in the verb ‘be cooked’ can hardly
be ancient, and t-less forms, such as found in Augila and reconstructible
Tuareg, must be old. I would therefore reconstruct *nnn. The verbal
noun could go back to something like *nn(n)e.
Summarizing, we have evidence for very similar formations in these
three verbs in Proto-Berber:
Aorist Imperfective Verbal Noun
* *ttt *-tt(t)e ‘eat’
*sw *sss *-ss(s)e ‘drink’
*nw *nnn *-nn(n)e ‘be cooked’
*C 1(C 2) *C 1C 1C 1 *-C 1C 1(C 1)e

It is not clear whether the verbal noun originally had a geminate or not.
Zenaga and Tuareg point to forms without gemination, while Zenatic and
Libyan Berber suggest that the second consonant was a geminate.

How can these formations be explained? At this point, it is useful to

remember the suggestion, made above, that  should be interpreted as
the result of a pre-Proto-Berber assimilation from *ty. If one accepts this
reconstruction, one notes that the three verbs under investigation not only
 5)3&&Û*33&(6-"3Û#&3#&3Û7&3#4 

share the last glottal stop in the shape of their roots, but also share a semi-
vowel as a second consonant. Thus, one could consider the morphology
of the Imperfective and the verbal noun as a formation specially used with
verbs of the shape Cw/y.
This brings us to a possible explanation of the reduplication pattern
found in these three verbs. There are a number of clues that indicate
that in Proto-Berber (or in an earlier stage of the language), there were
restrictions on the set of consonants which could be geminated by mor-
phological processes. The clearest example is provided by verbs which
have ancient *b (> h in Tuareg) 8 as one of their root consonants. In those
morphological formations where one would expect this consonant to be
geminated, one sometimes finds metathesis which leads to gemination of
another consonant (Prasse 1972-4, Vol. I: 75; Kossmann 1999: 78-79; Mali
Tuareg forms):
Aorist Imperfective
lh hall ‘weep’
nhy hanny ‘see’

One has the impression that the reason behind this metathesis is to pre-
vent these forms from being **lahh and **nahhy. In modern Tuareg, this
prohibition no more obtains. It probably goes back to (pre-)Proto-Berber.
A more complicated example is provided by the verbs of Prasse’s class
IA5. This class consists of bi-radical verbs which do not have any irregular
vowel alternations, and therefore are radically dierent from the verbs
which have an original final glottal stop. This verb class has an unexpected
formation of the Imperfective, C 1C 1aC 2 with gemination of the first radical
and insertion of a full vowel a. In all other verb types with gemination, it is
the second radical which is geminated, and they normally have short  as
the apophonic vowel. Scholars have presented dierent explanations for
this situation. One explanation, which seems to fit the facts rather well, has
been proposed by Louali & Philippson (2004: 87-89), who base themselves
loosely on Prasse’s analyses. Their synchronic account can be rephrased
into the following historical scenario: verbs of class IA5 originally had
a glottal stop as their second radical, which was lost in this position in
all dialects, including Mauritanian Zenaga. In an earlier stage of Berber
language history, it was impossible to have geminated glottal stops, and

8 On the reconstruction of this consonant, which is still preserved in Ghadames, in

Augila, and (pronounced h) in Tuareg, see Prasse 1969, Kossmann 1999: 61.
 .""35&/Û,044."//

therefore the gemination was dislocated to the first radical. This is shown
by the following example (forms reconstructed by the present author); the
full a in the modern Berber forms would be a reflex of the sequence *.:
Aorist Imperfective
*gn > gn **gn > *ggn > ggan ‘lie down’

While dierent from the case of *b, which has metathesis in order to
prevent gemination, 9 such a scenario would also point to the prohibition
of certain geminates.
Following this line of thought, one could venture the idea that at some
stage in pre-Proto-Berber there also existed a prohibition on the gemina-
tion of the semi-vowels y and w. 10 In order to prevent the formation of
forms with geminate semi-vowels, our three verbs would have undergone
reduplication, in order to fit the basic Imperfective template, thus:
Aorist Imperfective
*sw **sww > *sss ‘drink’

The same explanation could be used with verbal nouns, provided that one
reconstructs forms with a geminate, i.e. **-C 1C 2C 2e > *-C 1C 1C 1e
One has to stress, however, that no prohibition of this type exists in
any modern Berber dialect, at least as far as w is concerned. 11 In fact,

9 Prasse (1972-4, Vol. III: 75) interprets the Imperfective of this verbal class as the result of

consonantal metathesis and subsequent vowel lengthening for compensation: **gn >
*ggn > ggan (forms as would be reconstructed by the present author). In this analysis,
the development would be identical to that of *b in *lb (> Mali Tuareg lh). This
is more satisfactory from the point of view of similar problems being solved by similar
processes, and it would confirm Prasse’s idea that *b and * (my reconstruction) behave in
an identical way. However, I think the lengthening of the second * in order to compensate
for the loss of the initial syllable, which is proposed by Prasse, is not convincing. One
would either expect unlengthened *, i.e. **ggn, or the introduction of the Imperfective
prefix (t)t-, as is found with verbs with an initial glottal stop: Aorist *kr, Imperfective
*tt-kr (instead of **kkr > **kkar) ‘steal’. Therefore, I prefer Louali & Philippson’s
solution of a transposition of gemination to Prasse’s consonantal metathesis.
10 A similar line of thought is followed by Prasse (1972-4, Vol. III: 109), when he explains

the reduplicative Imperfective of ‘drink’: “ss gém[iné] sert peut-être tout simplement à
éviter ww gém[iné] instable, devenu normalement gg w > gg”. Prasse does not consider
the forms of ‘eat’ and ‘be cooked’ the results of a similar process. In his analysis,*ttt is
based on a root wth with the Imperfective prefix (t)t-, which would belong to a dierent
root from , k. He does not comment on the Imperfective of ‘be cooked’.
11 Medial y is extremely rare in reconstructible Berber verbs.
 5)3&&Û*33&(6-"3Û#&3#&3Û7&3#4 

there are many verbs which have a medial w, which is geminated in the
Imperfective, e.g. Aorist dwl Imperfective dkk wl ‘become’, r.wl – r.kk wl
‘flee’, z.wa – z.kk wa ‘go away’ (Figuig examples). This, however, does not
rule out the possibility that in an earlier stage of Berber dierent conditions
applied, and that the irregular forms of ‘eat’, ‘drink’ and ‘be cooked’ are
the last remnant of this lost prohibition.

Leiden University

REFERENCES 12

Basset, André
1929 Le verbe berbère. Étude des thèmes. Paris: Leroux.
Foucauld, Charles de
1951-1952 Dictionnaire touareg-français. Dialecte de l’Ahaggar. Paris: Imprimerie
Nationale de France.
Heath, Jerey
fc. Dictionnaire tamachek (Mali) - anglais - français. Paris: Karthala.
Khamed Attayoub, Abdoulmohamine
2001 La ttsrret des Ayttawari Seslem. Identification socio-linguistique d’un parler
berbère non documenté chez les Touaregs de l’Azawagh (Niger). Mémoire de
Maîtrise en études berbères, INALCO (Paris).
Kossmann, Maarten
1994 “La conjugaison des verbes CC à voyelle alternante en berbère“. Études
et Documents Berbères 12, 11-19.
1999 Essai sur la phonologie du proto-berbère. Köln: Rüdiger Köppe Verlag.
2001 “The origin of the glottal Stop in Zenaga and its reflexes in the other
Berber languages”. Afrika und Übersee 84/1, 61-100.
Louali, Naïma & Gérard Philippson
2004 “Le thème de l’aoriste intensif : Formes multiples, contenu unique”. In:
Kamal Naït-Zerrad, Rainer Vossen & Dymitr Ibriszimow (eds.) Nouvelles
études berbères. Le verbe et autres articles, 79-94. Köln: Rüdiger Köppe
Verlag.
Nicolas, Francis
1953 La langue berbère de Mauritanie. Dakar: I.F.A.N.
Prasse, Karl-G.
1969 “A propos de l’origine de h touarègue (tahaggart)”. København: Munks-
gaard.
1972-1974 Manuel de grammaire touarègue (thggart). 3 vols. Copenhague: Akade-
misk Forlag.

12 Most forms in the article have been cited from the sources enumerated in Kossmann

(1999: 26-29). Only literature cited explicitly in the article and sources not given in
Kossmann (1999) will be enumerated below.
 .""35&/Û,044."//

Rabdi, Larbi
2004 Le parler d’Ihbachen (Kabylie orientale - Algérie). Esquisse phonologique et
morphologique. Köln: Rüdiger Köppe Verlag.
Taine-Cheikh, Catherine
2004 “Les verbes à finale laryngale en zénaga”. In: Kamal Naït-Zerrad, Rainer
Vossen & Dymitr Ibriszimow (eds.), Nouvelles études berbères. Le verbe et
autres articles, 171-190. Köln: Rüdiger Köppe Verlag.