SYLLABLES IN TASHLHIYT BERBER

AND IN MOROCCAN ARABIC

Kluwer International Handbooks of Linguistics
VOLUME 2

The titles published in this series are listed at the end 0/ this volume.
Syllables in Tashlhiyt Berber
and in Moroccan Arabic

by

FRANC;OIS DELL
EHESS-CNRS,
Paris, France

and

MOHAMED ELMEDLAOUI
Faculte des Lettres,
Oujda, Morocco

"
~.

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN 978-1-4020-1077-4 ISBN 978-94-010-0279-0 (eBook)

DOI 10.1007/978-94-010-0279-0

Printed on acidjree paper

All Rights Reserved

© 2002 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2002
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permis sion from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
TO MORRIS HALLE
 TABLE OF CONTENTS

Symbols and abbreviations xi-xm

Preface xv-xvi

Chapter 1. Introduction 1-11

1.1. Goals and general outlook 1
1.2. The Berber 1anguages 5
1.3. Berber in Morocco 6
1.4. Tash1hiyt 7
1.5. Tashlyiyt and Moroccan Arabic in contact 8
1.6. Imdlawn Tashlhiyt 10

Chapter 2. Syntax and morphology, an overview 13-37

2.1. Sound system 13
2.2. Notational conventions 14
2.3. Syntax 17
2.3.1. Basic sentence structure 17
2.3.2. Yerba1 clitics 18
2.3 .3. Relative clauses 21
2.4. Yerbal morphology 23
2.5. Nominal morphology 26
2.5.1. Vowel-initial nouns, the basic facts 28
2.5.2. Alternations involving the augment 31
2.5.3. Consonant-initial nouns 34
2.5.3.1. 1C-initial nouns 34
2.5.3.2. Other consonant-initial nouns 37

Chapter 3. Phonological backdrop 39-69

3.1. Preliminaries on gemination 39
3.2. The long segment as a sequence of two prosodic
positions 41
3.2.1. Heteromorphemic geminates 42
3.2.1.1. Fusion of adjacent short consonants
into a long one 42
3.2.1.2. The genitive preposition 46
3.2.1.3 . (R)AD's final consonant 48
3.2.2. Syllable structure 49
3.2.3 . Templatic morphology I 50

vii
Vlll T ABLE OF CONTENTS

3.3. The long segment as a single melodie unit 53

3.3.1. Templatie morphology II 53
3.3.2. Feature changes in long consonants 55
3.4. "Tension" 56
3.5. Conclusion on the geminates 58
3.6. Dorsopharyngealization 58
3.6.1. Auditory properties 59
3.6.2. The distribution of emphasis 61
3.6.2.1. In the lexicon 61
3.6.2.2. At the phonetic level 63
3.7. The voieed pharyngeal consonant 65
3.8. tu! fronting 68

Chapter 4. Tashlhiyt syllables I 71-114

4.1. Syllabie consonants 71
4.2. Tashlhiyt verse and singing 79
4.3. Singing words to a tune 81
4.4. Parsing Tashlhiyt verse: preliminaries 84
4.5. Pattern satisfaction 85
4.6. Generalizations on orthometric syllables 89
4.7. The role of sonority 97
4.8. Geminates in complex codas 103
4.9. Alternative parses meeting all the constraints 108
4.9.1. Sonority plateaux in complex obstruent rimes 109
4.9.2. Sequences of high vowels 112
4.9.3. Alternative licit parses not due to DETACH 113
4.10. Summary 114

Chapter 5. Tashlhiyt syllables II 115-134

5.1. The syllabifieation of word sequences outside of poetry 115
5.2. Imperfective gemination: the basie generalization 117
5.3. Imperfective gemination and the weight of hollow
syllabIes 122
5.4. Length alternations in the causative prefix 124
5.4.1. Monosyllabic bases beginning with an onset 125
5.4.2. Other bases 127
5.5. Conclusion 134

Chapter 6. Vowelless syllables 135-187

6.1. Vowels vs. transitional vocoids 135
6.2. VTVs are releases with voicing 137
6.3. The distribution of VTVs 139
6.3.1. Two generalizations 140
6.3.2. Release in heterorganie clusters 142
 T ABLE OF CONTENTS ix

6.3.3. Release before a sibling consonant 146

6.3.3.1. SIBLING-RELEASE 146
6.3.3.2. The Fusion rule 149
6.3.3.3. Restrietions on fusion 154
6.4. The only surface vowels are a, i and u, two phonolog-
ical arguments 157
6.4.1. Morphemes with adjacent identical consonants 158
6.4.2. Regressive devoicing 160
6.5. Epenthetic vowels in Rifian Berber 163
6.5.1. The basic pattern for vowel epenthesis 164
6.5.2. e devoicing and e absorption 166
6.5.3. Final CC clusters 170
6.5.4. An outstanding issue: syllabification in kernels 173
6.6. Short vocoids in other works on Tashlhiyt 175

Chapter 7. The syllabification of vocoids 189-226

7.1. Vocoid sequences not containing underlying glides 189
7.1.1 . Sequences a+H 190
7.1.2. Sequences H+a 191
7.1.3. Sequences of potential hvs 192
7.1.4. Sequences a+a 193
7.1.5. Sequences involving aa 195
7.2. The need for underlying glides 196
7.3. Glides which are sonority peaks in the underlying
representations 202
7.3.1. Surface glides (onsets) which are sonority peaks 202
7.3.2. Glide gemination 204
7.3.2.1. Feminine bound state forms 205
7.3.2.2. Verbs and masculine bound state forms 208
7.3.2.3. Other stern-initial glides 212
7.3.3. Surface glides (codas) which are sonority peaks 215
7.4. Geminate glides 218
7.5. Conclusion 224

Chapter 8. Syllable structure in Moroccan Arabic 227-290

8.1. Introduction 227
8.2. Standard transcriptions 230
8.2.1. The distribution of 'e' in standard transcriptions 230
8.2.2. Uncontroversial schwas vs. putative ones 235
8.3. The structure of syllables in MA 241
8.3.1. Hinge syllables; syllable-final schwas 242
8.3.2. Inventory of syllable types 249
8.3.3. Complex nuclei; evidence from syllable weight 257
8.4. Violations of SonPeak in MA 261
x T ABLE OF CONTENTS

8.5. The syllable structure of words 267

8.5.1. A constraint-based analogue of right-to-left scan 269
8.5.2. Kerneis ending in eCC; FinH 273
8.5.3. Sonority in rimes; NoRR 276
8.5.4. Favoring sonority peaks as nuclei; SonPeak 282
8.5.5. Kerneis ending in eC; FinL 284
8.5.6. Free variants in which SonPeak overrides FinL 288
8.6. Summary 290

Chapter 9. Vowelless syllables in Moroccan Arabic 291-334

9.1. The new analysis is simpler 292
9.2. Expanded hollow syllables 296
9.3. Comparing Tashlhiyt and MA 303
9.3.1. Well-formed sequences of syllables in MA and
in Tashlhiyt 303
9.3.2 . Strings pronounced alike in MA and in Tashlhiyt 306
9.3.3. Glides which are sonority peaks 308
9.3.3.1. Vocalized glides 309
9.3.3.2. @w diphthongs; NoRR violations 313
9.4. Releases in sequences of sibling consonants 317
9.4.1. Fusion and NO-TREBLE 318
9.4.2. Earlier views on releases in sibling sequences 320
9.4.3. Releases between short sibling stops 324
9.5. Stable schwas 328
9.6. Summary of Chapter 9 and issues for further research 332

Appendix I. Preliminaries to Appendices Hand III 335-341

Appendix H. Song 343-347

Appendix III. Oratorical encounter 349-357

Appendix IV. Five Ashlhiy tunes 359-361

Appendix V. List of verbs with imperfective gemination 363-366

References 367-377

Index 379-385
 SYMBOLS AND ABBREVIATIONS

y IPA [j]
s IPA [f]
i. IPA [3]
x IPA [X]
y IPA [1$]
h IPA [6]
A 1. sy lla ble nucleu s; 2. (only in §3.6) dorsoparyn gealized
('emphatic' )
<A> 1. extrametric (Rifian); 2. ex ample in Chap 8
2
stop release
11 pau se
» ranked higher than (constraints)
+ cover sy mbol ranging over - , =, #
word-internal morpheme boundary
= boundary between cli tic and host
# word boundary other than =
(perio d) syllable boundary
(tilde) the preceding sy mbol represents the fir st half of a
geminate
+ occurs between two identical letters to indicate that they
do not represent an underlying geminate
* (asterisk) precedes an ill-formed item
AD lad! (co mpleme ntizer)
aor aori st
AR larl (imperfective preverb )
aug augmentative
aux auxiliary
b bound state
C any segment whi ch is not a V
C: geminated C
CA Classical Arabic
D coda
cau causative
col collective
dat dative
DE DelI and Elmedlaou i
def definite
dem demonstr ative

Xl
Xll SYMBOLS AND ABBREVIATIONS

det determiner
dirn diminutive
dir direetional
do direet objeet
DT Dell and Tangi
f feminine
FD Franccis Dell
fut future
G glide
gen genitive
H 1. high voeoid; 2. heavy syllable
hv (potential) high vowel
id idem
IFDQ inferred from direet questioning
imper imperative
impf imperfeetive
indiv individuative
int interrogative
InV initial vowel
IP Intonational Phrase
ITB Imdlawn Tashlhiyt Berber
IYT elitie iyt
ko kind of
L light syllable
lit literally
loe loeative
m maseuline
MA Moroeean Arabie
ME Mohamed Elmedlaoui
n noun
N nucleus
name proper name
o onset
OT Optimality Theory
p plural
PNG person-number-gender affix
prep preposition
prt participle
Pword word+clitics
R 1. rime; 2. eonsonantal sonorant (Resonant)
RAD future preverb
rep reeiproeal
(R)AD AD or RAD
RIPI representation whieh is an input to phonetie implementation
 SYMBOLS AND AB B R E V I A TI ON S Xlll

Rt feature-geometrie Root node

s singular
0' syllable
SVV short voieed voeoid
SWF Syllable Well-Formedness
u free state ('u' for 'unbound')
V vowel (syllabie voeoid)
VjGj homorganie vowel-glide sequenee (iy, uw)
VTV voiced transitional vocoid
WH wh-word
X prosodie position (skeletal slot)
 PREFACE

This book is intended primarily as an original contribution to the investi-

gation of the phonology of the two main languages spoken in Morocco.
Its central topic is syllable structure. Our theoretical outlook is that of
generative phonology.
Most of the book deals with Tashlhiyt Berber. This language has a
syllable structure with properties which are highly unusual, as seen from
the vantage point of better-studied languages on which most theorizing about
syllabification is based. On the one hand, complex consonant sequences
are a common occurrence in the surface representations. On the other
hand, syllable structure is very simple: only one distinctive feature bundle
(phoneme) may occur in the onset, the nucleus or the coda. The way these
two conflicting demands are reconciled is by allowing vowelless sylla-
bies . Any consonant may act as a syllable nucleus. When astring is
syllabified, nuclear status is preferentially assigned to the segments with
a higher degree of sonority than their neighbours. Consider for instance
the expression below, which is a complete sentence meaning 'remove it
(m) and eat it (m)':
/kks=t t-ss-t=t/ [k.st.s .t:"] .k.k~t.t.s. .slt.

The sentence must be pronounced voiceless throughout, as indicated by

the IPA transcription between square brackets ; the syllabic parse given after
the IPA transcription indicates that the sentence comprises four syllables
(syllable nuclei are underlined).
The differences between the dialects of Berber have to do primarily
with the phonology and the lexicon. Tashlhiyt appears to be the dialect in
which the avoidance of vowel epenthesis is pushed to the greatest extremes.
In Chapter 6 we briefly compare Tashlhiyt with Rifian Berber, which resorts
to vowel epenthesis. Rifian is similar in this respect to Tamazight Berber,
which was the main source of the Berber data used in phonological theo-
rizing during the Seventies and the Eighties.
In the last two chapters we take a fresh look at syllable structure in
Moroccan Arabic and we argue that our conclusions about Tashlhiyt Berber
carry over to a certain extent to Moroccan Arabic. The inventories of syllable
types of the two languages are very similar. Unlike Tashlhiyt, Moroccan
Arabic has an epenthetic vowel; but the range of contexts in which vowel
epenthesis occurs is much more restricted than is suggested by the standard
transcriptions. As in Tashlhiyt, any consonant can act as a syllable nucleus.
When they occur at the end of an Intonational Phrase, syllables in which

xv
XVI PREFACE

the nucleus is an obstruent are subject to epenthesis. The inserted vowel

combines with the obstruent to form a complex nucleus. Consider for
instance the word in (i) below, which is the 3rd fern sg imperfective form
of /zbd/ 'pull' . This word is usuaIly transcribed as (ii):
(i) (ii) (iii) (iv)
/t-zbd/ t@zb@d .tt.b@d. .tt.bg .
In our account, the surface representation is (iii) when this word occurs
at the end of an Intonational Phrase, and it is (iv) elsewhere. While in variant
(iv) the nucleus of either syllable is a bare consonant, (iii) has a final syllable
with a complex nucleus @d.
This book is organized as follows . Chapters 1 to 3 set the stage for the
discussion in Chapters 4, 5, 6 and 7, which provide a detailed analysis of
the syllable structure of Tashlhiyt. Chapters 8 and 9 develop an account
of Moroccan Arabic syIlabification that builds on the results of our dis-
cussion of Tashlhiyt.
Our deepest gratitude goes to Michael Kenstowicz and to Jean
Lowenstamm. Their moral support came at two decisive moments during
the writing of this book. They read a complete draft in painstaking detail
and suggested many improvements. We are also deeply grateful to two
anonymous reviewers for Kluwer for their comments, as weIl as to an anony-
mous reviewer for Cambridge University Press, who commented in 1999
on an earlier draft which dealt only with Berber.
We owe a special debt to Nick Clements , Morris Halle and Lisa Selkirk ,
who have discussed with us much of our earlier work on Berber. Talking
with them and reading their work has been an important source of inspi -
ration. We have also learned much from Lionel Galand, and ME wishes
to thank hirn for his support and ever-readiness to help during the initial
stages of his formation in Berber linguistics.
Vast thanks are due to Oufae Tangi and Fouad Saa for spending many
hours answering questions about their native languages.
We would also like to thank the Faculte des Lettres in Oujda in the person
of its dean, Mohamed Laamiri, without whose understanding and assistance
this book could not have been written, as well as the following persons, who
have helped us in various ways: Claude Brenier-Estrine, Salem Chaker,
Redouane Djamouri , Paulette Galand-Pernet, Mohamed Lahrouchi, Alain
Peyraube, Rachid Ridouane, Miriam Rovsing-Olsen, Chakir Zeroual and
also Jacqueline Bergsma, Iris Klug and the other persons at Kluwer who
turned our manuscript into a book.
We are also deeply grateful to Najat, Tarik and Sara Elmedlaoui for
sharing some of the constraints imposed by the preparation of this book.
This work was supported in part by funding from the CNRS (Centre
National de la Recherche Scientifique) and the AUPELF (Association des
Universires Partiellement ou Entierement de Langue Francaise),
 CHAPTER ONE

INTRODUCTION

In this chapter we first state our goals and general outlook, and indicate how
the content of this book relates to our earlier work on Berber. We then
present background information on our subject matter and on the relevant
literature .

1.1. GOALS AND GENERAL OUTLOOK

During the last twenty years, syllabification in Berber has been used on
several occasions as a source of evidence in favor of important theoret-
ical innovations in phonology, notably in Vergnaud, Halle et al. (1979),
Hyman (1985) and Prince and Smolensky (1993).1 By contrast , the main
thrust of this book is descriptive.
Our aim is to describe syllabification in one Berber dialect, viz Imdlawn
Tashlhiyt, and at the same time to provide a point of reference for further
work on closely-related dialects. We intend our work to be of use to
berberists and to linguists without any previous knowledge of Berber.
Besides the information directly relevant to syllabification in Tashlhiyt,
we will endeavor to present background information to help non-berberists
in forming a coherent picture of the overall phonological and morpholog-
ical make-up of the specific dialect under scrutiny. Since the Berber dialects
of Morocco are relatively uniform in their morphological structure, we hope
that the information given in this book will provide a useful point of ref-
erence for researchers working on other dialects.
Despite its relative celebrity among theoretical phonologists, Berber is
not a very well studied language. Searching the literature for useable infor-
mation about it may be a frustrating experience for non-berberists. The
language is divided into many dialects. No varieties of Berber have more
prestige than others. There are very few dialects for which one can piece
together a comprehensive picture of the morphology and phonology. The
transcriptions used by most authors are bare phonemic transcriptions, with
!ittle additional information supp!ied about actual pronunciations. The

I Vergnaud, Halle et al. proposed templatic syllabification as a means of locating the sites
for vowel epenthesi s. Hyman' s ' weight units' were the direct precur sors of moras as the
term is understood in much current work. Prince and Smolensky's monograph inaugurated
Optimality Theory, which has since become the chief competitor to the rule-based, deriva-
tional model of SPE (Chomsky and Halle 1968).

1
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
2 CHAPTER ONE

similarity of these transcriptions across dialects is misleading; it obscures

differences which are crucial for the unravelling of syllable structure.
The syllable structure of Imdlawn Tashlhiyt Berber was discussed in
Elmedlaoui (1985), an unpublished doctoral dissertation, and in DE (1985,
1988). What mainly caught the attention of most readers of these works was
the challenge posed by Tashlhiyt for any syllabification procedure which
first identifies certain segments in astring as syllable nuclei, and then
apportions the remaining segments between onsets and codas . Our discus-
sion in DE (1985, 1988) gave little sense of how our analysis of
syllabification fitted into the larger scheme of Tashlhiyt phonology and
morphology; nor did it deal with the differences between the analysis of
syllable structure we presented for Tashlhiyt and those proposed in the
literature for other dialects of Berber, or for Moroccan Arabic, which has
been in intimate contact with Berber for centuries and resembles it in
many respects. Another area of concern was the nature of our data about
syllable structure. Most of these data consisted of ME 's native speaker
judgements about syllable count. Although most judgements elicited from
ME remained consistent over time, they were not always consistent with
those elicited from other Tashlhiyt speakers and there seems to be little
chance that the divergences could be due to dialectal variation.
Work done subsequently puts us now in a position to address these
concerns. First, a fairly detailed picture of the morphology of Imdlawn
Tashlhiyt and of its morphologically-governed alternations is now avail-
able through various works.' It is out of question to present a systematic
summary of these works in the present book, but we will freely draw on
their results when relevant.
Second, instead of using direct judgements about syllable count as a
source of data, as we did in our 1985 works , we will continue the practice
which we began in DE (1997a) of concentrating on syllabification in singing.
The advantages of this shift are explained in § 4.1.
Third, we can now reap the benefits of first-hand experience we have
gained with other languages spoken in Morocco. One of us has supervised
doctoral dissertations on the phonology and morphology of two Berber
dialects of Eastern Morocco (Tangi 1991, Saa 1995) and has co-authored

2 The verbal morphology of Imdlawn Tashlhiyt was described in a systematic fashion in a

monograph-sized work published in two parts (DE 1989, 1991). This work also surveys various
idiosyncratic altemations which occur when preverbs , clitics and verbs are strung together.
Several major templatic processes for deriving nouns and adjectives were discussed and
abundantly exemplified in DE (1992). The inflectional morphology of nouns is discussed
in detail in Jebbour (1988) and DelI and Jebbour (1991, 1995). These three works describe
Tiznit Tashlhiyt, a dialect which presents a few phonological differences with Imdlawn
Tashlhiyt as weil as numerous lexicaiones. But most of the regularities stated in these
works are also valid for Imdlawn .
 INTRODUCTION 3

further work on one of these dialects. We will present facts which point
to important differences between the syllable structure of Tashlhiyt Berber
and that of Rifian Berber. Furthermore, we have turned our attention to
Moroccan Arabic.
Our starting point for Moroccan Arabic was syllabification in nursery
rhymes and in melhun, a verse genre which has practitioners in Morocco
and in Algeria. Tahar (1975) discovered the basic principles of its versifi-
cation and Kouloughli (1978) pointed out the relevance of some of Tahar 's
findings for the syllable structure of Algerian Arabic . The syllabification
of Arabic required for singing in Moroccan Arabic resembles very much
that required in Tashlhiyt singing. This resemblance could suggest at first
that the two languages have basically the same syllable structure. However
a closer look reveals significant differences between the two languages,
as will be shown in the last chapters of this book . Although our discus-
sion of Moroccan Arabic is informed by that of Berber in the previous
chapters .and cannot be read separately, the last chapters are meant as a
contribution to the study of Moroccan Arabic in its own right. Including
it at the end of a book on Berber is justified by the light which each language
sheds onto the other.
We stated at the outset that our primary goal is descriptive. We do not
mean to imply that theoretical concerns are absent from this work.
Obviously, our reason for dwelling at such length on the syllable struc -
ture of Tashlhiyt is its relevance for linguistic theory. What we meant, rather,
was that we will not deal head-on with issues which are directly relevant
for an adequate characterization of the human language faculty, e.g. we
will not draw on evidence from Tashlhiyt Berber to assess the merits of
sequential derivations or of moraic representations. We want to impart to
our readers the main aspects of the syllable structure of Tashlhiyt, as well
as its basic resemblances to and differences from that of Moroccan Arabic.
This cannot be done in a theoretical vacuum. We adopt the general orien-
tation and the basic assumptions of generative phonology, on which see
Kenstowicz (1994a) and the essays gathered in Goldsmith (1995) . However,
we want to write a book which can still be read with profit by non-berberists
long after the theoretical interests of 1inguists have moved away from the
issues which are currently at the center of attention. Since our primary
concern is empirical coverage, we will allow ourselves to take a prag-
matic view of devices such as rules and constraints, which are the
stock-in-trade of generative phonologists, and to use them as mere descrip-
tive tools.
This descriptive stance is our main reason for adopting a constraint-based
approach in our discussion of syllabification in Tashlhiyt and in Moroccan
Arabic. Like most other work done at the time, our work of 1985 and
1988 relied on sequentia1 derivations in which syllable structure was built
in a stepwise fashion. Unlike in some other languages, in Tashlhiyt Berber
4 CHAPTER ONE

the basic syllable shapes are very simple and can be characterized almost
entirely by generalizations which are surface-true. Consequently step-by-
step derivations do not seem to be necessary. Instead we will use constraints
which are familiar from the literature on syllabification or are reminiscent
of them. We will not attempt strictly to adhere to a constraint-based
approach , however ; we will depart from it whenever we feel that the moves
required to maintain theoretical uniformity would not be rewarded by
proportionate gains in insight. For instance, we will revert to sequentiaIly-
ordered phonological rules when this seems to us the most convenient
way to characterize the phenomena under discussion. In such cases we
will not explore alternative constraint-based accounts. At times we will even
be content with stating generalizations in plain English.
Most of the data presented in this book come from one of the authors
(ME), who grew up speaking both Tashlhiyt Berber and Arabic in the
countryside in western Morocco (see below § 1.6). The languages which
are the object of our inquiry will be referred to as 'Imdlawn Tashlhiyt
Berber' and as ' Lmnabha Moroccan Arabic ', labels based on the places
where these languages are spoken. A few remarks are appropriate here about
how we use these denominations.
Linguists who are engaged in the description of specific languages or
dialects usually characterize what they are describing by using such expres-
sions as 'Eastern Massachusetts English ', 'French as currently spoken in
academic circles in Paris', and so on. Such designations are useful, but
they are misleading if they are taken to refer to a collective entity shared
by many speakers. From our point of view, the only meaningful objects
of inquiry are the grammars stored in the brains of individual speakers on
the one hand, and on the other hand Universal Grammar (UG), which is
common to the entire human species.' Groupings such as 'the speakers of
Russian', 'the speakers of Harrari', etc. may be of great importance from
a sociological or historical point of view, but they have no direct rele-
vance for the line of inquiry to which this book belongs, the ultimate aim
of which is a theory of the language faculty.
We are not denying the obvious practical importance of labels like
'Russian' or ' Harrari': such labels indicate where one is most likely to
find persons whose language has certain properties. We are simply denying
that these labels can be made to refer to well-defined empirical objects which
would be relevant for our line of research.
In our discus sion of Tashlhiyt Berber as weIl as in that of Moroccan
Arabic, we will often refer the readers to places in the literature which
are evidence that a particular feature which we have observed in ME's
language is independently attested in other speakers . In view of what we

3 We adopt the general outlook proposed in Chomsky (1986: 15-50) .

 INTRODUCTION 5

have said in the preceding paragraphs, the point of our references to the
literature in such cases is not to buttress our own factual observations; it
is rather to give an idea of the extent to which a trait of ME's language is
widespread, or to give credit to other researchers who earlier observed
the regularity in question or a similar one.

1.2. THE BERBER LANGUAGES

Berber is a family of closely related languages or dialects spoken over an

area of North Africa which stretches from the Atlantic to the oasis of Siwa
in western Egypt. Berber belongs to the Afroasiatic family. Before the
Arab conquest, the language was presumably spoken over a vast continuous
tract of North Africa. The present-day linguistic map shows an archipelago
of Berber islands in an Arabic-speaking sea. These islands are mostly moun-
taineous or desert areas .
Berber languages are spoken by significant numbers of people in eight
African states, viz Morocco, Algeria, Tunisia, Libya, Egypt, Mauritania,
Mali and Niger. There are also sizeable Berber-speaking communities in
Europe, e.g. in France, Belgium and the Netherlands. Galand (1988: 209)
gives a conservative estimate of 15 million for the total number of Berber
speakers in 1983. Of these, 9 million lived in Morocco and 5 in Algeria."
Until some 25 years ago the Berbers of different areas had no sense of
all being related to one another. The relevance of the term 'Berber' was
strictly historical; it had no sociological implications. Berber dialects do not
have terms grouping together the various Berber dialects or their speakers.
Only the local dialects and their speakers have names, e.g. Tashlhiyt for
the dialects of southwest Morocco, Tamazight for those of central Morocco,
etc. As pointed out by Galand (1988: 209) the use of 'Tamazight' as a
blanket term für all varieties of Berber is an innovation which has arisen
recently in certain intellectual circles. Here are the names of some impor-
tant dialects outside of Morocco (on Berber in Morocco, v. below): Taqbaylit
(Kabylie region, Algeria) , Tashawit (Aures region, Algeria), Tahaggart or
Tamasheq, the language of the Tuaregs (Sahara). Among the Berber dialects
there is none which is considered as a standard, or even one which has more
prestige than the others .'
Except for Tuareg, none of the present-day Berber languages has an
instituted writing system ." The literary tradition is primarily an oral one.

4 According to Galand the total population of either country was approximately 25 million
in 1981.
5 This is true in particular of Tamazight, whose name is now also used by some people to
refer to any variety of Berber (v. above). On the history of 'Amazigh' and "Tamazight' , v.
Galand (1985: 179-180) .
6 On writing Berber, v. Galand (1989: 344-346).
6 CHAPTER ON E

For overviews of Berber, v. Basset (1952), Applegate (1970) , Galand

(1975, 1988), Mitchell (1993 : 5, n. 2). General maps of Berber dialects
are found at the end of the volume which contains Galand (1988) and in
Camps (1984 : 10-11). Galand (1979) gathers in book form (with indexes
at the end) yearly surveys of Berber linguistics which the author origi-
nally published from 1965 to 1975 in the Annuaire de l'Afrique du Nord.
Besides their extensive bibliographical coverage, with short comments, these
surveys are a precious source of information about meetings and about
institutions and individuals active in the field . Chaker (1992) is a collec-
tion of similar surveys for later years. For other sources of references, see
the Encyclopedie berbere edited by G. Camps , Brenier-Estrine (1994-97),
Chaker and Bounfour (1996), Bougchiche (1997).

1.3. BERBER IN MOROCCO

The majority of the people who speak a Berber language live in Morocco,
and more than one third of the Moroccan population speaks a Berber
language.' The three main dialect groups are Tashlhiyt and Tamazight, which
were mentioned above, and Tarifit (spoken in the Rif mountains , in northem
Morocco) ." Tamazight is sometimes also referred to as Biejraber," The
differences between the three groups have to do primarily with the sound
structure and the lexicon . lntercomprehension between the three groups is
limited, witness the following two facts. First , Berber speakers from dif-
ferent areas generally use Moroccan Arabic to communicate. Second, on
television, each day, versions of the same news bulletin in the three dialects
are delivered in succession by three speakers .
Like almost all Moroccans today, Berber speakers are Muslims."
The aggregation of the myriad local dialects into entities such as Tashlhiyt
or Tarifit reflects the speakers' own usage, not adecision made by scholars.
Imagine you are travelling through Morocco to do a survey of Berber. At
every stop on your way you ask the local people how they call the language
they speak . In some areas of Morocco the answer will be 'Tashlhiyt', in
others 'Tarifit' and so on. When researchers working on Berber write that
a dialect spoken in a certain place belongs to Tashlhiyt or to Tarifit, as
the case may be, they do nothing more than restate a local naming practice.
The practice in question carries hardly any information about the similar-

7 A linguistic map of Morocco is found at the end of Boukous (1995).

8 On Tarifit, see e.g. Chtatou (1982), Tangi (1991), Deli and Tangi (1992) and references
therein.
9 On Tamaz ight, see e.g. Penchoen (1973) , Saib (1976 , 1978), Guerssel (1976) , Willms
(1991) and references therein.
10 There used to be a small minority of Berber-speaking Jews, some of them living in wholly
Jewish villages, see e.g. Galand-Pernet and Zafrani (1974).
 INTRODUCTION 7

ities and differenees between the strueture of the dialeet and that of other
Berber dialeets, for it does not require eomparing one variety of Berber with
another: speakers may know that their native tongue is ealled Tashlhiyt even
if they have never heard any other variety of Berber. All this is to say
that while we often use the tradition al nomenclature of Berber dialeets in
this book, we never take it as more than what it is: a folktaxonomy, and
one of the erudest sort at that.

1.4. TASHLHIYT

Most Berbers in Moroeeo speak Tashlhiyt. II This language is spoken over

a eontinuous area with roughly the shape of a four-sided polygon. This
polygon is bounded by the Atlantic Oeean to the west, by the northern slopes
of the Haut Atlas mountains to the north and by the southern slopes of
the Anti-Atlas mountains to the south. The eastern side of the polygon
lies to the east of the city of Ouarzazate. On its northeastern boundary,
Tashlhiyt gradually shades into Tamazight. One part of the area just delim-
ited is traditionally known as the Sous (sus) region. The Sous was the
eraddle of the Berber Almohad dynasty, whieh ruled North Africa and Spain
from 1130 to 1269. In parts of Moroeeo lying outside of the domain of
Tashlhiyt, the speakers of Tashlhiyt are often indiseriminately referred to
as Soussis, i.e. people from the Sous, even though the Sous only forms
one part of the Tashlhiyt-speaking area. See Sehuyler (1979: 9-18) for a
eoncise introduetion to the Tashlhiyt-speaking region and its people .
The form used to refer eolleetively to the people who speak Tashlhiyt
is i-sl1i-iy-n, the maseuline plural of the word a-sl1i-iy. a-sl1i-iy is both an
adjeetive and a noun." The noun designates a male speaker of Tashlhiyt.
Its feminine singular form t-a-slli-iy-t has two meanings." It may desig-
nate a female speaker of Tashlhiyt, or it may designate the language itself."
In this book we will refer to the language as Tashlhiyt. Otherwise we will
uniformly use the maseuline singular form. We will write about 'the
Ashlhiys' or about 'Ashlhiy musie'. In Moroeean Arabie the Ashlhiys are
eolleetively referred to as sluh, a maseuline plural noun," and their language
is ealled slh-a (fern sg). .

II The name of the language is often given with a short i , e.g. 'Tashlhit' , 'Tachelhit' . This
is due to the fact that iy has shortened to i in some dialects of Tashlhiyt.
12 The suffix -iy is used to form occupational nouns and adjectives indicating an origin,
e.g. a-rudan-iy ' frorn the city of Taroudant' (t-a-rudan-t) .
13 On feminine singulars with the shape It- .. . -tl, v. § 2.5.
14 Whereas French, for instance, uses masculine singular forms of adjectives to designate
languages (le provencal, le portugais) , Berber uses feminine singular forms, e.g.
t-a-brtqqis-t 'the Portuguese language' or 'Portuguese female', cf. a-brtqqiz ' Portuguese
male' .
15 Whence the French adjective chleuh (pronounced [sl öj). The singular of sluh is s@lh.
8 CHAPTER ONE

Native speakers of Tashlhiyt live in great numbers outside of the

Tashlhiyt-speaking region we have just delimited. They form a diaspora
speeialized in eommeree, with its own social and eeonomie network
throughout the eountry. They are most visible in retail trade. In most large
eities of Moroeeo, for instanee, groeery stores are typically run by Ashlhiys.
The Ashlhiy diaspora has grown sueeessful offshoots in Europe.
Dialeetal variation is pervasive aeross the Tashlhiyt domain but it does
not hin der intereomprehension. Aeeording to Stroomer (1998 : 38) dialeet
differentiation is less eonspieuous in Tashlhiyt than in Tamazight or in
Tarifit.
For referenees to work on Tashlhiyt, see e.g . Chaker (1994). Aspinion
(1953) is a useful grammar of Tashlhiyt. Roux (1955) eontains a good
eolleetion of Tashlhiyt texts. Stroomer (1998) presents an overview of some
aspeets of dialeet differentiation in Tashlhiyt and a list of published sourees
on various Tashlhiyt dialeets. Galand-Pernet (1967) diseusses the impliea-
tions of dialeet variation for the eommon poetie language of the Ashlhiys.
On the Ashlhiy literary tradition reeorded in the Arabie seript, see e.g.
Stroomer (1992) and Boogert (1997, 1998).

1.5. TASHLHIYT AND MOROCCAN ARABIC IN CONTACT

In many rural areas of the Tashlhiyt domain, Tashlhiyt is still the sole
language used for all purposes, exeept in the recitation of Koranie texts .
Even today young speakers of Tashlhiyt who do not understand Moroeean
Arabie (heneeforth MA) are a eommon oeeurrenee but they ean only be
found in remote plaees and MA is steadily gaining ground in its eompeti-
tion with Tashlhiyt. The main faetors in this evolution are urbanization
and the ever-inereasing availability of the modern media. 16 Children raised
in Ashlhiy families which have migrated to eities tend to be bilingual with
the loeal variety of MA as their dominant language. As the use of Tashlhiyt
tends to beeome eonfined to the horne, the eommand of the language tends
to deteriorate and adaptations from the dominant language are inereas-
ingly resorted to." This proeess of attrition is even more advaneed in
ehildren from Ashlhiy families residing outside of Moroeeo.
We give below four examples to illustrate the phenomenon. Eaeh example
eonsists of three sentenees with the same meaning . Sentenee (a) is one which
would normally be used by a monolingual speaker of Tashlhiyt, or one living
in an area where most people are native speakers of Tashlhiyt. Sentenee (b),

16 On the competition between Berber and MA, see Boukous (1995: 90-93, 102-104).
17 See Boukous (1995: 108-111).
 INTRODUCTION 9

also in Tashlhiyt, could be uttered by a young Ashlhiy raised in an Arabic-

speaking city. Sentence (c) is the MA equivalent of (a).18

(1) a. !krad w-ulaw-n dekilu net-asa

three b-heart.p-p and=kilo genef-liver
'three hearts and one kilogram of liver' 19
b. tlt qluba dekilu n=l-kbd-a
c. tlt qluba uekilu d=l=kbd-a
three heart :p andekilo genedefeliver-f

(2) a. ured a-hanweann a ra y-I[tts]

not=be u-roomedem AD fut 3ms-search
'it is not that room that he will search (but this one)'
b. masi l-biteann ad ra y-f[tt@s]
c. masi hadak l=bit 11i -yadi y-f[tt@s]
not dem deferoom WH fut 3ms :impf-search

(3) a. i-lla=gigi l-f[tq]

3ms-be=in: l s l-hernia
'I have a hernia'
b. i-llaegigi l-f[t@q]
c. kayn f=iya l=f[t@q]
prt:be ine l s defehemia

(4) a. [qu:d]
'go to the devil!' (lit 'pimp!')
b. [quw@d]
c. [quw@d]

The first two examples illustrate the influence of bilingualism on the choice
of words . The linguistic situation in cities also impinges on the phonology.
The speakers try to divest themselves of those features of their pronunci-
ation which are telltale signs of a rural origin . On the one hand, speakers
of dialects with characteristic features which set them apart from the other
Tashlhiyt dialects try to suppress those features, e.g. speakers of Haha
Tashlhiyt, who normally spirantize the noncoronal obstruents, endeavour

18 Like all the abbreviations used in this book, those in the glosses are listed after the
Table of Contents. Their meaning will become clear in the next chapter.
19 Uttered by someone buying meat.
10 CHAPTER ONE

not to do SO .20 On the other hand , in their efforts to rid their pronuncia-
tion of MA of anything that may give away their Ashlhiy background, some
speakers end up 'arabicizing' their pronunciation of Tashlhiyt. While
speaking Tashlhiyt they may for instance break up consonant clusters with
schwas which are not acceptable for mono lingual speakers." This phe-
nomenon is illustrated by the pronunciation of the last word in examples
(2)b and (3)b . Jtts ' search' and {-Jtq 'hernia' originate from Arabic , but
nowadays they have become fully integrated into the lexicon of Tashlhiyt;
even monolingual speakers of Tashlhiyt use them. In the pronunciation of
these speakers the voiceless sequences Ifttsl and Iftql must not be broken
up by a voiced vocoid , see (2)a and (3)a. On the other hand a schwa must
be pronounced before the last consonant in the MA words Jtt@ sand
{-Jt@q when these appear before a pause, see (2)c and (3)c. The Tashlhiyt
pronunciations in (2)b and (3)b by young bilingual speakers mimic the
MA pronunciation. The last example illustrates an analogou s phenomenon
involving the realization of geminate glides . Tashlhiyt and MA being each
analyzed independently ofthe other, the underlying form ofthe word in (4),
an imperative verb, is Iqwwdl in either language. In Tashlhiyt, as exern-
plified in (4)a, a geminate glide which is followed by a prepausal obstruent
can only be realized as the corresponding long vowel, hence the steady-state
[u:] in (4)a .22 In MA, on the other hand , Iqwwdl is realized with a final
[C@C] syllable, like the other triliteral verbs with a medial geminate.
The state of affairs just described has obvious implications for the
students of Tashlhiyt who wish to collect data on the language without doing
fieldwork in a rural area where it is spoken. In particular, it is weIl to
keep in mind that as a rule university students who are native speakers of
Tashlhiyt grew up in a city.

1.6. IMDLAWN TASHLHIYT

Most of the data about Tashlhiyt presented in this book was provided by
one of us, Mohamed Elmedlaoui, henceforth ME. ME was born in 1949
in Imdlawn." His father is a monolingual speaker of Imdlawn Tashlhiyt,

20 P.C. from R. Ridouane , a native speaker of Haha Tashlhiyt.

21 Boukous (2000: 46) has observed MA-like schwas in the speech of Ashlhiy children living
in the city of Agadir. Boukou s compared the command of Tashlhiyt by children raised in
the country side and children raised in cities. His article contains numerous telling example s
of language decay in the latter.
22 The form is nonethele ss dissyllabic: .qu.wd. On geminate glides in Tashlhiyt, see § 7.4.
23 i-mdlaw-n is the plural form of a-mdlaw 'man from the Imdlawn valley' , v. also
t-a-mdlaw-t ' woman from the Imdlawn valley' (plural t-i-mdlaw-in). The Imdlawn valley
lies in the south corner of the map entitled 'Le cadre orographique des Seksawa' which
occupies pp . 8 and 9 in Berque (1955/1978), i.e. the right-hand side corner at the top of
page 9. The Imdlawn are very similar to the Isks awn, the people descr ibed in Berque' s
monograph.
 INTRODUCTION 11

while his mother also spoke Lmnabha Moroccan Arabic . After his birth
his family settled down in Igudar (Lmnabha), 50 kilometers to the south
of Imdlawn, but up to the present day they maintain close ties, both social
and economic, with their horne town in Imdlawn. ME has lived in Lmnabha
until he was thirteen.
Up to this day the only language spoken in the Imdlawn valley is
Tashlhiyt. The Lmnabha area, on the other hand, is an Arabic-speaking
enclave whose inhabitants live in intimate contact with the Ashlhiys .
Tashlhiyt and Arabic were both spoken in the hamlet in Lmnabha where
ME 's family lived . ME's mother tongue was Tashlhiyt, but he has been
bilingual as far back as he can remember. He spoke Lmnabha Moroccan
Arabic in school and Tashlhiyt and Arabic at horne. Between 13 and 19
he was a boarder at the Islamic Institute, fir st in Taliwin, a Tashlhiyt-
speaking city, and later in Taroudant, a city where the prevailing language
is Arabic. In both cities the language spoken in the Institute was almost
exclusively Tashlhiyt, the students, staff and faculty all being native speakers
of Tashlhiyt. At 19 ME left the Sous region and was for four years a uni-
versity student in Fes, where he studied Arabic Language and Literature.
At 24 he went to Oujda, in Eastern Morocco, where he has resided ever
since . He first taught Arabic Philology at Mohammed I University in Oujda;
he now teaches Biblical Hebrew.
At the time of writing this book, ME speaks both Tashlhiyt and Arabic
to the full satisfaction of monolingual speakers of either language."
Completely bilingual speakers like ME are not an unfrequent occurrence
in the Sous area, as already noted by Destaing (1937) in his foreword.

24 But Arabic speakers of eastern Morocco can hear that he comes from the western part
of the country.
 CHAPTER TWO

SYNTAX ANO MORPHOLOGY,

AN OVERVIEW

Our aim in this chapter is to present an overall view in which to fit the more
local facts which will be discussed in the following chapters.

2.1. SOUND SYSTEM

Table (1) lists the phonemes of Imdlawn Tashlhiyt, abstracting away from
length distinctions.
(1) A B C 0 E F G
t !t k kW q qW
b d !d g gW
m n !n
f s !s 8 IV
.S X X
W 11
z !z z IV
.Z "I 'Y
w
~ h
w !l r !r y
u a
= = =
Key : A labial; B dental/alveolar; C palato-alveolar; 0 =
= =
velar/palatal; E uvular; F pharyngeal; G glottal. =
Unless specified otherwise, the phonetic symbols in this book have their
IPA values. The following equivalences are valid throughout this book, even
for languages other than Berber. '8' and 'z' represent palato-alveolar frica-
tives (IPA [I] and [3]). 'x' and '''I', which represent velar fricatives in the
IPA, will always stand for uvular fricatives here (IPA [X] and [ff]). 'h' stands
for a murmured glottal fricative ('voiced h'). 'y' stands for an unrounded
palatal glide (IPA u]). 'r ' represents a voiced alveolar flap or trill depending
on the context. Exclamation marks indicate 'emphatic' (i.e . dorsopharyn-
gealized) consonants. When a word contains an 'emphatic' phoneme,
dorsopharyngealization is spread over the whole word at the phonetic level
(v. § 3.6). 'I' indicates that all the segments in the following morpheme
or word are emphatic, i.e. dorsopharyngealized.
The voice contrast among the obstruents of Tashlhiyt sounds like that
of Standard French. Prevocalic voiceless stops are unaspirated.
Unlike C+w sequences, the labialized consonants r, s". qw, xW and y"
are single segments, as shown by their behaviour in syllabification (v.
Chapter 4) and in templatic morphology (v. OE 1992).
Tashlhiyt has a lexical contrast between simple and geminate consonants,

13
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
14 CHAPTER TWO

e.g. i-mi 'mouth' vs. immi 'Mom' . Exeept for those in the last row, eaeh
symbol in (1) stands for a pair of segments whieh are distinguished by
gemination (length) in the underlying representations. In this book we use
the pairs 'geminate/simple' and 'long/short' interehangeably. Tashlhiyt has
eontrastive vowellength in surfaee representations, but it does not have long
vowels in its underlying representations. [a:] will be notated throughout
as aa . Some instanees of aa derive from a sequenee /a+a/ while others
are realizations of f)/ , /Ila! or /all/ (v. § 3.7). The long vowels [i:] and [u:]
are the standard manifestations of tautosyllabie iy and uw sequenees. iy
sounds like [i:] in niys ([nd]) 'aim at' , where it is tautosyllabie, but like
[iy] in the eorresponding imperfeetive tt-niyas, where it is not.
Stress and intonation in Tashlhiyt are still terrae incognitae, as far as
we know.' Whereas in an English or Italian sentenee every polysyllabic word
has its own prominenee peak , it is highly dubious that the same obtains
in Imdlawn Tashlhiyt. If Imdlawn Tashlhiyt has a phenomenon that eould
be ealled stress or aeeent, it is likely that it is a property of units larger
than words. Our preliminary observations suggest that in general, the main
piteh event in an intonational phrase oeeurs near its end, viz on the last
or next-but-last syllable nucleus whieh is a sonorant.

2.2. NOTATION AL CONVENTIONS

Let us first indieate how we use eertain terms . When referring to segments
in phonological representations, we use 'vocoid' to denote any member
of the set {a, i, u, y, w}, and 'contoid' to denote all the other segments of
ITB. Note that aeeording to this definition h is a contoid.' A syllabie segment
is any segment associated with a syllabic nucleus. A vowel is any syllabie
voeoid; there are only three vowels in ITB, ja, i, u/. A eonsonant is any
segment whieh is not a vowel. The set of eonsonants eontains the nonsyl-
labie voeoids (aka glides) y, wand the eontoids, syllabie or not, e.g. syllabie
n and syllabie h are eonsonants. Throughout this book, V stands for 'vowel',
C stands for 'consonant' in the sense just defined, H stands for a high
voeoid, i.e. for a member of the set {i, y, u, w}, and G stands for a glide .
The terminology and eonventions we are introducing in this seetion
will be followed strietly throughout this book. In particular, we beg readers
to keep in mind the following point. While it is often harmless to use the
term 'vowel' loosely to mean ' vocoid' or 'vowel-like sound', in this book
nothing is referred to as a vowel unless it is a voeoid meeting the fol-
lowing eonditions: (i) it is a segment , i.e. it has its own distinetive features
or prosodie position, and (ii) it is assoeiated with a syllable nucleus.

1 See DE (1985) for some observations, especi ally pp. 119-120 on certain prepausal syl-
lables.
2 On transitional vocoids , which are not segrnents, see below .
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 15

In our transcriptions hyphens indicate morphological boundaries inside

words ; boundaries between clitics and their hosts are marked by equal signs
'= ' ; other word boundaries are indicated sometimes by '#' and sometimes
by a blank. Our choice between '#' and a blank has no significance; it is
only a matter of typographical convenience. Finally, the plus sign '+' is a
cover symbol ranging over '-' , '=' and '#' .
Unless specified otherwise the term ' word' is used to refer to syntactic
words, i.e. to maximal strings dominated by an XO node.' We take each clitic
in a sentence to be a word on its own. For instance sentence (2) contains
five words:
(2) hra=tn=d !zra-nt lt-mvar-in
justedoßmpedir see-3fp bf-woman-fp
'the women just saw them'
By a Pword we mean a sequence which comprises a word together with
all the clitics attached to it, if there are any. A word which has no depen-
dent clitics is a Pword all by itself. Sentence (2) is comprised of three
Pwords . 'Morpheme boundary' ('-') designates a morpheme edge which
is not also a word edge; 'clitic boundary' ('=') designates a clitic edge which
is not also the edge of a Pword. 'Pword boundary' ('#' or aspace) desig-
nates the edge of a Pword.
As a rule every word is given with its ultimate morphemic breakdown
the first time it appears in an example. In a few instances we depart from
that policy to avoid excessive cluttering. All those instances involve
sequences of grammatical morphemes, e.g . the demonstrative modifiers
meaning 'this' and 'that', which we simply note =ad and =ann, are actually
combinations of an independently attested =a 'this' ti-fullus-nea 'these
chickens') with the directional morphemes =d 'hither' and =nn ' thither' ,
which are also verbal enclitics (they can occur in position 3 in (14) below).
If it is necessary to indicate that two morphemes are components of a
portmanteau, their glosses are separated by a colon. For instance the English
words boys and men would be glossed as 'boy-p' and 'rnan.p'.
In this book all transcriptions not enclosed between slanted line s are
phonetic representations given in a ' broad' transcription. This tran scrip-
tion is akin to a phonemic transcription in the structuralist sense. It abstracts
over the variations in vowel quality which are due to neighbouring con-
sonants. It also glosses over the predictable transitional vocoids which are
heard between consonants in certain clusters but do not play any role in
the morphology and phonology of Imdlawn Tashlhiyt, and about which
we must now say a few words .
Throughout this book we use the term vocoid to refer to any stretch of

3 See e.g. Jackendoff (1997: 29, 113, 120).

16 CHAPTER TWO

time, however short, not occupied by a glottal consonant or by an articu-

lation which is consonantal in the sense of Chomsky and Halle (1968 :
302). Note that thi s definition does not say anything about voicing.
The voiced vocoids one hears in Imdlawn Tashlhiyt utterances fall into
three categories, (i) the glides y and w, (ii) the full vowels, i.e . those vocoids
which are uncontroversial allophones of la, i, ul and (iii) short vowel-like
sounds which will be referred to as voiced transitional vocoids, hence-
forth VTVs. VTVs often give the impression of having a shorter duration
than full vowels. Whereas full vowels are manifestations of segments
already present in the lexicon or introduced by word formation processes
(e.g . i in kriz; the negative stern of kr; 'plough'), VTVs' location and
vowel quality are entirely predictable from the phonetic environment and
the location of certain morphological boundaries. VTVs are not repre-
sented in the broad phonetic tran scriptions used throughout this book. The
various transcriptions we will use are illustrated in (3) below. For each
example in (3) we give (I) its underlying representation, (11) its broad
phonetic transcription, (11I) a narrower phonetic transcription, and (IV) a
phonetic transcription of an intermediate kind, in which all VTVs are uni -
formly represented by '@'.4.5
(3) I 11 11I IV
a. It-a- lznk'td-t/ ltaznkrtt [ltcznokqt:"] [!tazn@kqt h]
b. /i-dlh/ idlh [idleh] [idl@h]
c. li-snnq=tt/ isnnqtt [isn:Aqhth] [isn:@qhth]
d. It-ssutl-m/ tssutlm [ts.ötlom] [ts:ötl@m]
In each of the examples in (3) the first vocoid is a full vowel and the second
is a VTY. As the examples illustrate, VTVs come in many varieties, IPA
[a] among them. In most contexts they are central vocoids with an aperture
ranging between high and open-mid. Unless indicated otherwise, we shall
use the symbol '@' to represent any VTV regardless of its vowel color,
see column IV in (3) . For more on VTVs, see Chapter 6.
Unless indicated otherwise, the phonetic transcription of an expression
(i.e. of a word or a sequence of words) represents that expression as
pronounced in isolation. This point deserves special emphasis. As we shall
see later, an important constraint on the phonetic representations of Imdlawn
Tashlhiyt is the prohibition of hiatus: adjacent vowels are disallowed. A
common way of avoiding hiatus is for lil and lul to be realized as the
corresponding glides next to another vowel. For instance the noun /i-kru/

4 The morpheme-by-morpheme glosses of the examples in (3) and their meanings are
the following : (a) /f-u-gazelle-fs/ 'female gazelle ' ; (b) /3ms-collide/ 'he collided ': (c) '
/3ms-wring:neck=do3fs/ 'he wrung her neck' ; (d) l2-surround-2mp/ 'you surrounded' .
5 Here and elsewhere, a raised 'h' represents an audible explosion burst. In 3(a), the voice-
less [ul which occurs between k and t represents the explosion burst of /kw/.
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 17

'kid' has four contextual variants, ikru, ikrw, ykru and ykrw. The variants
with a glide on one end only occur adjacent to a vowel and those with a
vowel on one end only occur adjacent to a pause or a consonant, e.g. ikru
(pronounced in isolation), ikrw a-mzwaru 'the first kid', lt-rra ykru 'she
saw the kid' and azu ykrw=ann 'skin that kid!' . Throughout this book,
whenever we write that a given unit, word or morpheme, begins (resp. ends)
in a high vowel at the phonetic level that assertion is meant to charac-
terize the pronunciation of the unit in question in contexts in which it follows
(resp. precedes) a pause or a consonant. The point is of special impor-
tance because Tashlhiyt has an underlying contrast between high vowels
and glides and that contrast is neutralized next to ,a vowel.

2.3. SYNTAX

2.3.1. Basic senten ce structure

In basic assertive sentences the verb precedes all its arguments and the
subject precedes all other arguments:

(4) t-ga t-frux-t i-fullus-n -y=t-gmmi

3fs-put bf-child-fs up-chicken-rnp inebf-house
'the girl put the chickens into the house'

The word order in (4) is not the only possible one . The subject may precede
the verb, e.g.

(5) tafruxt tga ifuIlusn -y tgmmi

girl.u put chickens in house
'the girl , she put the ckickens into the house'

In (5) there must be an intonational break between the verb and the pre-
ceding phrase. When the subject appears to the left of the verb, as in (5),
it must have its free state form (tafruxt) , whereas it must be in the bound
state (tfruxt) when it follows the verb (on state, v. § 2.5).
Nouns are inflected for gender (masculine vs. feminine), number (singular
vs . plural) and state (free vs. bound) . Verbs agree with their subjects in
person, number and gender. A list of the agreement markers, which will
henceforth be called PNG (from 'person, number, gender'), can be found
below in the first column in (13).
In a noun phrase the complements follow the head noun :

(6) a-mgala ngr=t-mzgida dew-anu

u-distance betweenebf-rnosque andeb-well
'the distance between the mosque and the well'
18 CHAPTER TWO

(7) !krad id-lxmis i-zwar-n new-ayyur

three p-thursday prt-first-prt geneb-month
'the first three thursdays of the month'

(8) sna-t t-srdaneadennk lt-i-bukad-in

two-f bf-mule:p=dem=gen2ms f-p-blind-fp
'these two blind mules of yours'

The last word in example (8) is an adjective. Adjectives have the same
morphology as nouns; they agree in gender and number with the head
noun.

2.3.2. Verbal clitics

The word order exemplified in (4) is that which obtains when the arguments
of the verb all contain lexemes. Certain items must be cliticized onto the
verb or onto some grammatical morpheme which precedes it. For instance
(4) becomes (9) if 'the chickens' (mp) and 'the house' (fs) are replaced
by the corresponding personal pronouns:

(9) t-ga =tn =gi-s t-frux-t

3fs-put 3mp in-3s bf-child-fs
'the girl put them (m) into it (f)'

Some of the morphemes which always precede the verb always attract the
clitics to them. An example is is, a complemetizer used for making yes/no
questions, among other things. (10) and (11) are the interrogative coun-
terparts of (4) and (9) respectively:

(10) is tga tfruxt ifullusn "Y tgmmi

int put girl chickens in house
'did the girl put the chickens into the house?'
(11) is =tn =gis tga tfruxt
int them in-3s put girl
'did the girl put them (m) into it?'

Morphemes such as is, which have the property of attracting after them
items which otherwise follow the verb, will be called preverbs. Aside from
ar (imperfective), rad (future), ur (negation) and lira (immediate past) all
the other preverbs are complementizers or subordinating conjunctions, e.g.
ad 'that', iy 'if', mav 'until' (v. DE (1989: 171) for others). Cliticization
is a process which is internal to clauses. For instance sentence (12) contains
two clauses each enclosed in a pair of brackets:
 SYNTAX AND MORPHOLOGY , AN OVERVIEW 19

(12) [is =a-m i-nna [mas =tt i-hubbaj]

int dat-2fs 3ms-say that doßfs 3ms-Iove
'did he tell you (f) that he loved her?'
The higher clause begins with the complementizer is, and the sentential com-
plement of inna 'he said ' begins with the complementizer mas. Both
complementizers are preverbs, and each triggers cliticization within the
limits of its own clause, as we can see from the fact that tt, the 3fs direct
object pronoun of the embedded clause, must be located to the right of
mas.
The verbal clitics can be divided in four classes: (A) the pronominal
clitics, (B) the directional clitics d and nn, (C) the adverbial clitics, and
(D) iyt and the imperative clitics. We briefly review the four classes below.
The pronominal clitics are the object clitics, i.e. clitic pronouns which
are direct objects of verbs, and the clitic prepositional phrases. When
governed by apreposition, personal pronouns have special forms which
are enclitic to the preposition. Moreover, for certain prepositions, the
prepositional phrase as a whole behaves as a clitic if the preposition governs
a pronoun. An example is gi-s 'into it', in (9). Prepositions such as y, which
form clitic prepositional phrases, will be called cliticizable prepositions."
Along with the paradigm of the PNGs , (13) gives the complete paradigms
for the object pronouns and for clitie prepositional phrases headed by i
(dative), dar and y.
(13) PNG do dar 'Y
ls -x iyi iyi dar-i gig-i
2ms t- -t k a-k dar-k gi-k
2fs idem km a-m dar-rn gi-m
3ms 1- t a-s dar-s gi-s
3fs t- tt idem idem idem
1p n- ax ax dar-nx gi-(t)-nx
2mp t- -m kWn a-wn dar-un gi-wn / gi-t-un
2fp t- -mt kWnt a-wnt dar-unt gi-wnt / gi-t-unt
3mp -n tn a-sn dar-sn gi-(t)-sn
3fp -nt tnt a-snt dar-snt gi-(t)-snt
In order to avoid clutter, the morphological breakdown indicated by the
hyphens in (13) is only a partial one. The final t in the 2fp and 3fp forms
is a suffix on its own, as can be seen by comparing them with the corre-
sponding masculine forms. Similarly, comparing the 2p pronominal forms

6 There are seven clilicizable preposilions: i dative, 'Y locative, d 'and' or 'with ' (comita-
live), f 'upon', S 'toward ' or 'with' (instrumental), SS'Y 'from' , and dar 'at X's place' (cf.
French chez).
20 CHAPTER TWO

with their 3p counterparts shows that in all these forms n and the pre-
ceding segment belong to different morphemes .
Depending on the person, number and gender of the subject, the PNG
marker on the verb may be aprefix, a suffix or a combination of both:
n-ut 'we struck', ut-n 'they (m) struck', t-ut-m 'you (mp) struck'. The object
clitics are the only ones which distinguish gender in the 3s. Except in the
first person singular,' the clitic prepositional phrases all make use of the
same set of pronominal forms, those after gi- in the last column of (13).
These forms are used as objects of noncliticizable prepositions as well."
The directional clitics are d 'hither ' and nn 'thither' , which are realized
as id and inn in certain contexts . The facts concerning their meanings and
uses in Imdlawn Tashlhiyt are similar to those described in Bentolila (1969).
Here are a few clitic adverbs : ak'k" 'completely', ka 'only', sul 'still,
finaIly', zzu'never:past'. V. DE (1989: 170) for a more complete list.
When a clause contains several clitics they are always adjacent to one
another. Nonadverbial clitics occur in a fixed order. Dative pronouns (1)
always precede object pronouns (2), which always precede directionals
(3), which always precede prepositional phrases. Moreover items 1 to 3 must
form an uninterrupted sequence:? no clitic adverb is allowed between them.
Clitic adverbs may however occur before clitic prepositional phrases. This
is summarized in (14).
(14) CL: datives object directional adverbs prep.phrases
1 2 3 4 5
The sequence CL in (14) may contain two dative pronouns in a row, but
only one object pronoun and one directional. It can also contain more than
one adverb and more than one prepositional phrase. In a clause with no
preverbs the sequence CL immediately follows the verb. asku 'because'
does not belong to the class of preverbs, and consequently the sequence
CL (in boldface) must follow the verb in the clause below:
(15) asku t-srs =t =inn =sul =gi-s t-frux-t
because 3fs-put do3ms dir finally in-3s bf-child-fs
'because the girl finally put it (m) into it'

7 Here are the 1s clitic prepositional phrases, listed in the order corresponding to that of
the prepositions in note 6: iyi, gigi, didi, flli or flla (free variants), sri, ssgigi and dari. With
the other prepositions the form of the Is pronoun is iyi, e.g. gr-iyi 'between-l s' , nnig-iyi
'beyond- Is' .
8 The optional t in the plural is allowed by all the prepositions of note 6 except i and dar.
It is also allowed by gr 'between' . All the cliticizable prepositions except dar have a special
form when they appear in a c1itic prepositional phrase. These forms are (in the order of
note 6) a, gi, di, flla , sr and ssgi. They do not seem to be deriveable by any regular phono-
logical processes of Tashlhiyt.
9 Setting aside the behavior of iyt, on which v. below.
 SYNTAX ANO MORPHOLOGY, AN OVERVIEW 21

In a clause which contains one or more preverbs, CL immediately follows

the rightmost of these: 10
(16) is a =t =inn =sul =gi.s t-srus tfruxt
int AR doßms dir finally in-3s 3fs-putimpf girl
'does the girl finally put it (m) into it?'
The clitic iyt can be ordered freely with respect to the other clitics. When
using iyt speakers imply that the truth of the proposition expressed in the
utterance is indifferent to them. The two sentences below are variants of
one another,"
(17) i-g =iyt =a-sn =t =inn =gi-s
3ms-putaor IYT dat-3mp doßms dir loc-3s
'let hirn put it (m) in it for them (m) (it does not matter to me)'
(18) i-g =asn =t =iyt =inn =gis
On the clitics at (2mp) and amt (2fp), which are the second person plural
imperative markers, v. DE (1989 : 178-179).
We have just seen how a clitic sequence can follow a verb or apreverb.
The same clitic sequences can also occur immediately after a noun or a bare
prepo sition. This happens with relative clauses, as we shall now see.

2.3.3 . Relative clauses

Relative clauses follow the nouns that they modify . Some relative clauses
do not begin with a relative pronoun or a complementizer. Such is, for
instance, the case in the sentence below, where the edges of the relative
clause are indicated by brackets.
(19) manza tasrdunt [=ax i-fka babaek]
where mule [=datlp 3ms-give father=2ms]
' where is the mule that your father gave us?'
If the relative clause in (19) stands on its own as an independent sentence,
one gets the following sentence:
(20) i-fkaeax babaek tasrdunt
3ms-give=datl p father=2ms mule
'your father gave us a mule'
In (20) the placement of the dative clitic accords with the regularities

10 When the rightmost preverb is the negat ion ur there are additional complications, on which
v. OE (1989: 173).
\1 V. OE (1989 : 180) for other variants.
22 CHAPTER TWO

stated earlier: since the sentence does not contain any preverb, the clitic
follows the verb. As argued in detail in DE (1989: 180-187), the left edge
of relative clauses acts as a 'silent' preverb, i.e. as one whose phonolog-
ical representation is empty; in particular it attracts clitics. This is the reason
why in (19) the clitic ax precedes its verb even though no overt preverb
is present at the beginning of the relative clause.
When the relativized noun phrase is the object of a cliticizable prepo-
sition, the preposition appears at the beginning of the relative clause without
any overt object behind it. If the relative clause contains any clitics, these
immediately follow the preposition. This happens in the following sentence,
in which the relativized noun is the object of the preposition d ' with' (the
clitic sequence is in boldface):
(2 I) sqqsa-x -y=!imhdam [d =km =nn=sr·s
ask-ls locepupils [withedozfsediretoward-Ss
y-iwi]
3ms-bring]
'I asked about the pupils with whom he took you there'
In the sentence below, in which the relative clause in (21) is a sentence
all by itself, the clitic sequence follows the verb since there is no preverb:
(22) y-IWI =km =inn=sr-s d=!imhdam
3ms-bring=d02fs=dir=toward-3s withepupils
'he took you there with the pupils'
When the relativized noun phrase is a subject the verb of the relative
clause takes on a special form that some students of Berber call a participle.
Sentences (24) and (25) below both contain a relative clause which is related
to sentence (23). In (24) the relativised noun phrase is the object, and the
verb in the relative clause assumes the same form as in (23): it agrees
with its subject (!tmyart). In (25), on the other hand, the relativised noun
phrase is the subject, and the verb in the relative clause is a participle.
(23) t_-ywi ltmvart amakr
3fs-seize woman,b thief
'the woman caught the thief'
(24) is t-ssn-t amakr [t--ywi ltmvart ]
int 2-know-2s thief [3fs-seize woman,b]
'do you know the thief whom the woman caught?'
(25) is t-ssn-t ltamvart [i--ywi-n amakr]
int 2-know-2s woman,u [prt-seize-prt thief]
'do you know the woman who caught the thief?'
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 23

Participles are formed with the prefix lil and the suffix In!. They do not
agree in number nor in gender with their antecedents, except in a small class
of residual cases, on which v. DE (1989: note 34 p. 182). The so-called
participles should probably be considered as impersonal forms," and their
initial li-I as adefault 3ms PNG. The phonological behaviour of the initial
lil in the participle is identical with that of the 3ms PNG, v. DE (1989 :
182-183). All factual assertions which we will make in this book about
the 3ms PNG are meant to include the prefixal part of the participial marker
as a special case. Nonetheless, for the sake of convenience, we will continue
to use the term 'participle' and we will treat the participle marker
li- . . . -nI as a compound PNG, alongside those in the first column in
(13).
Let us end this syntactic sketch with a few words on the conjoined con-
struction.P in which two clauses follow each other without any connective
word. When the verb in the second clause has a perfective meaning, it
must have an aorist stern, v. DE (1989: 174ff). The two clauses describe
two successive events, as in (26), or the first clause expresses some con-
dition and the second an injunction, v. (27):
(26) t-fraet t-ss=t (ßfs-payedoßms 3fs-eat:aor=d03ms)
'she payed for it and ate it'
(27) i-y=t t-ssa t-fru=t (ifedoßms 3fs-eat 3fs-pay:aor=d03ms)
'if she ate it, let her pay for it
Aorist verbs followed by enclitics are a useful source of consonant clusters,
because there are a number of common verbs in which the aorist stern is
the only stern which is vowelless. The conjoined construction is one of
the rare environments in which clitics can be seen to occur after an aorist
verb, for as a rule verbs in the aorist are introduced by apreverb, in which
case the clitics precede the verb, as we have seen. The conjoined con-
struction occurs in many examples in this book.

2.4. VERBAL MORPHOLOGY

The bulk of the morphology of Tashlhiyt is found in verbs and nouns .

Tashlhiyt has both derivational and inflectional morphology. In the
remainder of this chapter we present the overall structure of the inflec-
tional paradigms of verbs and nouns and the affixal part of these paradigms.
Since we have described the morphology of verbs in detail elsewhere,"
we will devote little space to it here . We will dweIl at greater length on

12 V. Basset (1949: 35).

13 V. Leguil (1981) .
14 DE (1989, 1991).
24 CHAPTER TWO

nouns. A clear picture of their prefixal morphology is aprerequisite for

our discussion of the syllabification of high vocoids in Chapter 7.
Verbs are inflected for the person, number and gender of their subject.
This is done with aprefix, a suffix or a combination of both, see (28).
All the verbs in the language use the same set of PNGs.
(28) 'dweil' 'remember' 'remember, rcp'
a. pf 3ms i-zdv i-kti i-mm-kti
a'. ls ZdX-X I5 kti-x mm-kti-x
a" . 2fp t-zdv-mt t-kti-rnt t-mm-kti-mt
b. impf 3ms i-zddv i-ktti i-tt-mm-ktay
b'. ls zddx-x ktti-x tt-mm-ktay-x
b". 2fp t-zddv-mt t-ktti-mt tt-rnm-ktay-rnt"
Each line in (28) displays the same PNG in three verbs. The verbs in the
top block are perfective verbs while those in the bottom block are their
imperfective counterparts. A list of PNGs has already been given in the first
column in (13) . Two items should be added to that list for it to be complete.
One is li- ... -nI, the PNG whieh marks partieiples. The other is zero:
bare aorist and imperfeetive sterns are used to express 2s perfeetive and
imperfeetive imperatives, e.g. !amz 'seize! pf' (cf. (29)a), ltt-am: 'seize!
impf'.
PNGs are the outer layer in the morphology of verbs. A verbal stern is
the string which remains when a verb is stripped of its PNG. In Tashlhiyt
every verb has four sterns, viz . perfeetive affirmative, perfeetive negative,
aorist and imperfective. The name of the perfective affirmative will
henceforth be shortened to 'perfective', and that of the perfective negative,
to 'negative' . The four sterns are exemplified in (29) with verbs in the
3mp (the 3mp PNG is I-ni).
(29)
pf neg aor impf
a. !umz-n !um(i)z-n !amz-n !tt-amz-n 'seize'
,
a . ln-yamaz-n !n-yamaz-n !n-yamaz-n !tt-n-yamaz-n 'seize rep'
b. usi-n usi-n asi-n tt-asi-n 'earry'
b'. ss-usi-n ss-usi-n ss-asi-n ss-asay-n 'earry eau'
e. krz-n kr(i)z-n krz-n kkrz-n 'plough'
d. mmuylt-n mmuyl(i)t-n mmuylt-n tt-muylut-n 'feel siek'

15 Due to regressive devoicing , I'V-xl is realized as xx. On regressive devoicing, see DE

(1996a).
16 It-tt-I is realized as tt. This happens on1y in those sequences in which It-I is a PNG and
Itt-I is the impf prefix, v. DE (1989: 193).
 SYNTAX ANO MORPHOLOGY , AN OVERVIEW 25

The means of marking the differences between the four sterns vary from
one verb to the next, but they are to a large extent predictable from the
phonological make up of the verbs in the lexicon . Except for the imper-
fective prefix Itt-I, the phonological processes involved are nonconcatena-
tive . These fall into three classes: (1) alternations between vowel s, e.g.
u-a in lines a, b, (ii) alternations between zero and a vowel, v. for instance
the optional i which occurs in some negative sterns and the vowel which
appears before the last segment in the imperfective sterns in lines b', d,
and (iii) consonant gemination , as in the imperfective stern in line c, and
also in the imperfective sterns of 'dwell' and 'remember' in (28).
Let us use the term 'base' to refer to the unit which underlies the four
sterns belonging to the same line in (29). For instance, we shall say that
the four forms in (29)a all belong to the base !umz 'seize' (bases are named
after their pf stern) and that the forms in line (29)a' all belong to the base
!n-yamaz 'seize one another '.
Here are for instance sentences each containing one of the four sterns
of the base !umz 'seize' .
(30) a. !y-umz=tt 'he seized her'
b. ur=tt !y-um(i)z 'he did not seize her'
c. !y-amz=iyt=stt 'let hirn seize her (I don't mind)'
d. ar=tt=sul !i-tt-amz 'he still seizes her'
When the category of the stern of a verb in an example is not indicated
in the gloss it is the perfective, unless the verb is in the (perfective)
imperative, in which case it has an aorist stern.
Two different verbal bases may share the same root, as is the case for
!umz 'seize' and !n-yamaz 'seize one another' (v. (29)a, a'), or with usi
'carry' and ss-usi 'cause to carry' (v. (29)b, b') . Whereas the four sterns
of a given base share the same argument structure, two different bases
sharing the same verbal root have different argument structures. Verbal bases
fall into two categories, primary and secondary. Secondary bases are
causative , reciprocal and passive. All other bases are primary. Like stern
formation, base formation resorts to prefixation, to nonconcatenative
processes or to a combination of both, as one can see by comparing the third
column with the second in (28) or line a' with line a and line b' with line
b in (29). The starting point of a secondary base may be another secondary
base. For instance the causative base s-m-bada (cau-rcp-next) 'put next to
one another' is derived from the reciprocal base m-Iuula 'be next to one
another' (from liada 'be next to')."
A verbal kernel is the string which remains after a verb has been stripped

17 V. OE (1989: 78-79) on the various possible combination s of the causative, reciprocal

and passive morphemes whithin a single base. V. Guerssel (1992) on similar combinations
in Ai! Seghrouchen Tamazight.
26 CHAPTER TWO

of all its (concatenative) affixes. The kernel of Iumz-n (seize-3mp) 'they

seized' is lumz: that of ltt-amz-n (impf-seize :impf-3mp) is lamz; and that
of Itt-n-yamaz-n (impf-rcp-seize:rcp-3mp) 'they seize one another' is
lyamaz: In most of this book we shall take kerneis as given, without
attempting to elucidate the mechanisms whereby kerneis sharing the same
root are related to one another.
To sum up, the successive morphological elements in any verbal form
of Tashlhiyt can fit into the sequence of slots represented in (31):
(31) PNG tt PFX* kernel PNG
All the slots may be empty, except that for the kernel. There are ten PNGs,
the nine listed in the first column in (13), plus the participle marker (on
which v. above in § 2.3.3). tt is the imperfective prefix and PFX* stands
for a sequence of one or more of the prefixes which are used to form
secondary bases . Setting aside defective paradigms and particular cases
of suppletion, as well as limitations due to the syntactic and semantic
properties of some verbs, a verbal base gives rise to 42 inflected forms: each
of its four sterns can be combined with 10 PNGs, and to the 40 resulting
forms one must add the two naked sterns (aor and impf) which serve as
imperative 2s forms.

2.5. NOMINAL MORPHOLOGY

In Tashlhiyt as in the other dialects of Berber nouns are marked for number
(singular or plural), for gender (masculine and feminine) and for state
(free or bound). When the number and state of a noun in an example are
not indicated in the gloss, it is the singular and in the free state.
The distribution of gender in the lexicon is roughly the same as in French.
Grammatical gender correlates with biological gender for nouns denoting
human beings and certain animals, e.g. a-frux (m) 'boy' , t-a-frux-t (f) 'girl' .
It is idiosyncratic for the other nouns, e.g. ayyur ' moon' is masculine
whereas t-afuk-t 'sun' is feminine . Pairs of nouns distinguished only by
gender are not limited to a subset of the animate nouns, however, as
Tashlhiyt also uses gender marking as a process of derivational morphology,
for instance to create diminutive and augmentative nouns. The feminine
form of many nouns which are lexically masculine is a diminutive. For
instance, alongside udm 'face', which is masculine , we find the feminine
t-udm-t 'little face'. Similarly, the masculine form of many nouns which are
lexically feminine is an augmentative, e.g. the masculine noun a-ryal 'large
basket' is derived from the feminine t-aryal-t 'basket'. Feminine forms
are put to various uses besides the formation of diminutives . They are for
instance used to derive countable nouns from nouns with a collective
meaning, e.g. tt-a-zalim-t '(bulb of) onion' from la-zalim 'onion(s)',
!t-a-qaymrun-t ' shrimp, indiv' from !qaymrun 'shrimp, col' . They also
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 27

serve to derive names denoting actions or occupations from agentives,

e.g . lt-a-sffar-t ' theft ' , from !a-sffar 'thief', or language names, e.g.
t-a-brtqqis-t 'Portuguese (the language)', from brtqqi: 'Portugal'.
Let us say that gender is secondary in those nouns in which it comes
from derivational morphology and that it is primary in other nouns. We shall
thus say that the feminine gender of t-aryal-t 'basket' is primary whereas
that of lt-a-zalim-t ' (bulb of) onion' is secondary.
State is a category akin to case. A noun takes the bound state form
when it is governed by apreposition, when it is a subject and follows its
verb, or when it follows one of the cardinal numbers from one to ten or
mnnaw 'how many, several'; otherwise it is in the free state.
Setting aside various gaps, accidental or systematic, the paradigm of a
noun is comprised of eight forms. Here is for instance the paradigm of
a-fru x (m) ' boy' , t-a-frux-t (f) 'girl' .
(32) I 11 III IV
m, free m, bound f, free f, bound
a. s a-frux u-frux t-a-frux-t t-frux-t
b. p i-frxa-n i-frxa-n t-i-frx-in t-frx-in
In a nominal form the stern is what remains once one has taken away
the prefixes and suffixes marking gender, number and state. Here are a
few examples (singular forms are given for comparison at the end of lines):
(33) noun stem
a. t-i-funas-in funas 'cows' (s t-a-funas-t)
b. a-m-kraz m-kraz ' plowman' (p i-m-kraz-n)
c. una una ' wells' (s anu)
d. i-fass-n fass 'hands' (s a-fus)
Certain sterns contain derivational affixes , as is the case in (b), a noun
derived from the verb kr; 'plow'. Certain nominal forms are naked sterns,
v. (c).
The inflectional morphology of nouns resorts both to concatenative
processes (prefixation and suffixation) and nonconcatenative ones (vowel
alternations and the like). The concatenative processes operate in a uniform
fashion for all nouns, whereas the nonconcatenative processes are lexi-
cally governed to some extent.
The marking of gender and state only makes use of prefixes and suffixes.
Number marking uses both concatenative and nonconcatenative processes.
On the nonconcatenative processes involved in the formation of plural
nouns, cf. Jebbour (1988).18

18 Jebbour 's work deals with the variety of Tashlhiyt in use in Tiznit, whose morphology
resembles very much that of Imdlawn Tashlhiyt.
28 CHAPTER TWO

As a rule free state masculine nouns which are not loans begin with a
vowel, and so do free state feminine nouns if one sets aside their initial
t-, which is the feminine marker, v. e.g . a-frux (m) 'boy', t-a-frux-t (f) 'girl' .
We will deal later with those nouns which do not have an initial vowel.
Let us simply give an example for the time being."
(34) I II III IV
m,free m,bound f,free f,bound
a. s XIXXl XIXXl t-i-xixxi-t t-xixxi-t
b. P id xixxi id xixxi t-i-xixxit-in t-xixxit-in 20
We shall use the expression 'initial vowel' (henceforth InV) to denote the
leftmost vowel in a free state noun, provided that if that vowel is preceded
by a consonant, the consonant should not belong to the stern. The InV in
ussn 'jackal' is u and that in t-i-xixxi-t «34)III-a) is i, but the noun xixxi
«34)I-a) has no InV, for x, the consonant whieh precedes its leftmost vowel,
does belong to the stern. We shall use the expression 'vowel-initial noun'
to denote nouns which contain an InV.

2.5.1. Vowel-initial nouns, the basic facts

Let us begin with the morphology of the vowel-initial nouns, dealing first
with the alternations which affect the InV. The analysis we will outline
was originally proposed by Basset (1932, 1945) as a historieal reconstruc-
tion. Starting with Guerssel (1983) it has since been adopted as a synchronie
account for various dialects of Berber, v. e.g. Bader and Kenstowicz (1987),
Dell and Jebbour (1991), Tangi (1991).
Whereas in certain nouns (the minority) the InV belongs to the stern,
in the others it is aprefix. Let us use the term 'augment' to refer to those
initial vowels which are prefixes. Tables (35), (36), (37) and (38) below
summarize the facts concerning all those nouns which have an InV. In
(35) we give the complete paradigm of aylal 'bird ' , whose stern begins with
JaJ,21 and we give in (36) that of adrar 'mountain', whose initial JaJ is an
augment. In the phonological representations of the bound state forms the
vowels subject to deletion are marked with a slash for the sake of con-
spicuousness (on deletion, v. below).

19 xixxi, tixixxit 'person full of shit' .

20 The t which appears at the end of the stern before the plural suffixes in (34) III-b and
(34) IV-b has nothing to do with the fs suffix -t; it occurs in masculine nouns as weil. It is
an epenthetic consonant which is inserted at the end of vowel-final sterns when they precede
a plural suffix, e.g. afat-n 'summits' (s afa), i-sqsit-n 'questions' (s a-sqsi) .
21 t-aylal-t is a diminutive .
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 29

(35) I (free) 11 (bound)

a. ms aylal laylall waylal /u-aylal/
b. mp aylaln laylal-n! waylaln /u-aylal-n/
c. fs taylalt /t-aylal-t/ taylalt It-aylal-t/
d . fp taylalin /t-aylal-in/ taylalin /t -ay lal- in!

(36) I (free) 11 (bound)

a. ms adrar /a-drar/ udrar /u-a-drar/
b. mp idrarn /i-drar-n/ idrarn /u-i-drar-n/
c. fs tadrart /t-a-drar-t/ tdrart /t-a-drar-t/
d. fp tidrarin /t-i-drar-in/ tdrarin It-I-drar-in!

These tables illustrate the following regularities, which hold for all
vowel-initial nouns: (i) in those nouns in which the plural is marked by a
suffix, that suffix is I-ni in the masculine and I-inl in the feminine; (ii)
feminine nouns begin with prefix It_I;22 (iii) unless their gender is primary
and their stern ends in a vowel, feminine singular nouns end with suffix
l_t/.23
Table (35) summarizes the facts for those nouns in which the initial a
belongs to the stern, and (36) for those in which it is an augment. The
difference between an augment and an InV which belongs to the stern can
be seen in bound state forms and in plural forms. Augments drop as a
rule in bound state forms (v. (36)11), which never happens to nonprefixal
InVs (v. (35)11). In the plural the augment is i in all nouns, whereas a
nonprefixal InV is subject to the various ablaut processes which affect the
leftmost vowel in plural sterns. That vowel alternates in some nouns but
not in others. For instance the initial a in anu 'well' , which belongs to
the stern, alternates with u in the plural (una) , in the same manner as the
stem's leftmost vowel in a-safu 'torch' (p i-sufa). In aylal, on the other hand,
the initial a remains unchanged in the plural (v. (35)b,d), in the same way
as the stern's leftmost vowel in a-baddaz 'rnaize couscous' (p i-baddaz-n) .
Here is how one derives the bound state form of a vowel- initial noun
from the corresponding free state form . First, one prefixes lul to the free
state form (v. (35)a,b and (36)a,b), provided that the free state form begins
with a vowel." Second, if there is an augment, it drops (v. (36)11). The

22 There are only a handful of exceptions, e.g. immi 'Mom' , illi 'daughter', issa, littu (proper
name s for women) . These are also exceptional in that the prefixation of u- (see below) is
only optional in the bound state .
23 There are a few exceptions, e.g. t-ass 'Ill-fated, f ", !t-a-mubil ' automobile'. There is no
suffix at the end of t-a-rga 'ditch' , for instance, because the feminine gender is primary
and the stern ends in a voweJ.
24 This restrictions accounts for the fact that lul does not show up before the initial t- in
the feminine.
30 CHAPTER TWO

combination of these two operations, u-prefixation and augment deletion,

makes it look in some cases as though the augment had simply changed
its vowel colour, v. e.g. (36)a, where free state adrar (la-drar/) alternates
with bound state udrar (lu-a-drar/).
The bound state prefix lul materializes in some cases as a vowel u or i
and in others as a glide w or y. The surface backness of the bound state
prefix depends on the following vowel. The prefix surface s as a back vocoid
when the following vowel is lai (v. (35)Il-a,b and (36)Il-a) or lul;25 it
surfaces as nonback before lil (v. (36)I1-b and (37)Il-a below). The rule
which accounts for the backness assimilation of the prefix to a following
lil must be ordered before that which deletes the augment in the bound state:
in /u-i-drar-n/ (v. (36)II-b) the bound state prefix first assimilates its backness
to that of the plural augment (hence /i-i-drar-n/), then the augment deletes,
hence /i-drar-n/. The free state forms and the bound state forms of the nouns
with an i augment are homophonous, and yet they have different underlying
representations. Whereas the initial i in (36)I-b is a reflex of the plural
augment, the initial i in (36)I1-b derives from prefix lul.
Whether the bound state prefix surfaces as a vowel or a glide is deter-
mined by syllable structure, see Chapter 7. It is a glide when it precedes
a vowel in the surface form (v. (35)II-a,b), and a vowel otherwise
(v. (36)II-a,b). Owing to their syntactic properties, bound state nouns cannot
occur after a pause ; everywhere in this book transcriptions and statements
about the pronunciation of bound state forms are intended to cover the
pronunciation of these forms when the preceding word ends in a con so-
nant, e.g. y-aggug u-drar 'the mountain is far away'; when the preceding
word ends in a vowel the bound state prefix is realized as a glide to avoid
hiatus , e.g. i-buvla w-drar 'the mountain is full of greenery'
The singular augment is Ia! in certain nouns and lil in others. There is
no augment lul; when lul is a InV, it always belongs to the stern, and Table
(35) above also provides an adequate summary of the facts concerning those
nouns with u as an InV, e.g. the bound state of uday 'Jew' is w-uday and
that of its plural uday-n is w-uday-n; in the feminine the bound state
forms are identical with the corresponding free state forms (s, t-uday-t;
p, t-uday-ini. Let us now exemplify the difference between those InVs which
belong to the stern and those which are augments in nouns beginning with
li/. Our examples are isk 'horn' and i-nrfl 'tail (of a shirt)' . 26

25 I.e . the bound state form of ul 'heart ' is w-ul and that of ussan 'days' (s ass ) is
w-ussan .
26 In isk the initial tit changes colour in the plural, but there also exist sterns in which it
remains unchanged, e.g . ird 'grain of wheat' (b y-ird), plural ird-n (b y-ird-n).
 S YNTAX AND MORPHOLOGY , A N OVERVI EW 31

(37) I (free) 11 (bound)

a. ms isk liskl yisk lu-iskl
b. mp askiwn laskiw-nl waskiwn lu-askiw-n/
c. fs tiskt It-isk-t/ tiskt It-isk-t/
d. fp taskiwin It-askiw-in/ taskiw in It-askiw-in/

(38) I (free) 11 (bound)

a. ms inrfl /i-nrfl/ inrfl /u-i-nrfl/
b. mp inrfln li-nrfl-n/ inrfln /u-i-nrfl-n/
c. fs tinrflt It-i-nrfl-t/ tnrflt /t-i-nrfl-t/
d. fp tinrflin It-i-nrfl-in/ tnrflin /t-i-nrfl-in/

Setting aside various complications." one way to tell apart those nouns
in which the InV belongs to the stern from tho se in whieh it is an augment,
is by the syllabie shape of their bound state form s at the phonetie level:
in the bound state the former nouns beg in with CV, but not the latter,
compare waylal ((35)II-a), which begins with CV, and udra r ((36)II-a),
whieh doe s not.

2.5.2. Alternations involving the augment

To aeeount for the prefixal morphophonemics of vowel-initial nouns a

grammar of Imdl awn Tashlhiyt must eontain the following rule s. The first
is a morpholog ical rule ; the others are phonologieal rule s and they must
apply in the order given.

(39) PIAug : in the plural the augment is lil

(40) WI/YI: N
I
X [ X
r----_
[bound] - - - _
j
Vplace
I
[eor]

(41) AugDel: X~01 X [_[

I
[bound]

27 On the exceptions to the rule of augment deletion, v. below.

32 CHAPTER TWO

The rule WI/YI states that when abound state prefix is followed by i, the
timbre of i spreads onto the prefix . The rule operates in a transparent fashion
in y-isk (lu-iskl), v. (37)II-a . The front vocoid which follows the bound state
prefix must be syllabic, for y does not spread, witness the bound state
form of a-ynnri 'medicinal plant with a comestible root', which is u-ynnri,
not *i-ynnri and that of a-ydi 'dog' , which is w-iydi, not *y_iydi. 28 Rule
(40) is obligatory and does not have any exceptions. Aside from forms
such as w-iydi, in which iy is a realization of an underlying y, there are
no bound state forms beginning with wi.
The sound structure of Imdlawn Tashlhiyt does not in general disfavor
the sequences wi and uy, either morpheme-internally or across morpheme
or word boundaries, which is why the operation of rule WIIYI must be
confined to sequences with a specific morphological make up. Imdlawn
Tashlhiyt has another assimilatory phenomenon similar to WI/YI but it is
doubtful that both can be subsumed under a single process in a synchronic
analysis. The phenomenon in question involves the masculine empty noun
bu and its feminine counterpart mmu. These empty nouns are used pro-
ductively as heads of phrases meaning 'the one with X' or 'the one which
has X', where X is the meaning of an immediately following bound state
noun, e.g. busst-Itanu-t 'the one (m) with the shop' , mmuet-liuna 'the one
(f) with the shops' .29 bu and mmu are optionally realized as bi and mmi
when they precede a noun whose first segment is a front vocoid, e.g.
mmueyd-xixxi or mmi=yd-xixxi 'the one (f) with those (m) full of shit',
bu=y-zakar-n or biey-zakar-n /bueu-i-zakar-n/ 'the one (m) with the ropes '.
In biyzakarn the segment which triggers the assimilation has itself been
subject to rule WI/YI, for according to the analysis presented here, in the
bound state form i-zakar-n (lu-i-zakar-n/) the initial i is a reflex of the bound
state prefix lu-I. The fronting of the vowel in bu and mmu is triggered by
the surface front glide which is a realization of the bound state prefix
lu-I, but not by underlying Iy/, e.g. fronting can occur in /bueu-iza-n/, 'the
one (m) surrounded by flies', which can be pronounced bue y-iza-n or
bi=y-iza-n, but it may not occur in /istt-mmueyumayn/ 'the two day ones
(f)' ,30 which can only be pronounced istt-mmu=yumayn .
Rule AugDel (41) deletes the augment in bound state nouns. It is stated
so as to delete a one slot morpheme which immediately follows abound
state prefix . The rule operates in a transparent fashion in feminine nouns,
v. (36)c,d and (38)c,d .
The analysis presented here predicts that only those nouns whose InV
belongs to the stern have bound state forms which begin with a CV sequence

28 On the bound state forms of nouns whose sterns begin with glides, v. Chapter 7.
29 The free state forms for 'shop ' are t-a-hanu-t (s) and t-i-buna (p).
30 E.g. those women who have worked two days, to contrast them with others who have
worked for another length of time. yumayn 'two days' .
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 33

at the phonetic level. This is slightly at variance with the facts, however,
because some nouns are exceptions to rule AugDel. These exceptions fall
into three classes. First, in some nouns the augment does not drop in any
bound state form . Consider for instance tallunt (p tilluna) 'tambourine'.
We know the InV in this noun is an augment because there are no nouns
whose stern-initial vowels alternate between a in the singular an i in the
plural, and yet the augment in question does not drop, i.e. the bound state
forms of this noun are homophonous with its free state forms (*tllunt,
*tlluna). Other such nouns are tawwunt (p tiwwuna) 'stone' and tazzrt
(p tizzar) 'kind of pitchfork' . In most nouns of that type the stern begins
with a geminate, but stern-initial geminates do not always block augment
deletion. Second, there exist a handful of nouns whose augment regularly
drops in plural bound state forms but not in singular ones, e.g. afud 'knee',
b ufud or wafud, p ifaddn , b ifaddn (*yifaddn). In some of these, as in the
example just given, augment deletion in the singular is only optional,
while in others it is prohibited, e.g . the bound state form of asif 'river'
can only be wasif (*usif), whereas its plural form is isaffn (*yisaffn) . Third
and last there are nouns in which the deletion of the augment is optional
both in the singular and in the plural, e.g. ifr 'wing' (p ifrawn), ixss 'bone'
(p ixsan). Nouns of the latter type behave as though their InVs were at times
analyzed as augments and at others as being stern-initial.
In the formulation in (40) the label [bound] is intended to prevent
spreading in masculine pronouns of the form w-i . . . , e.g. w-i=nn-k
'yours' (m-/i/=gen-2ms), i.e, 'that/those (m) which belong(s) to you (ms)' .
These pronouns belong to a set of forms at the beginning of which there
is a systematic contrast between wand t which signals the distinction
between masculine and feminine. These forms comprise the following,
among others: (i) a paradigm of 'demonstrative' pronouns whose heads
are w-a (ms), t-a (fs), w-i (mp) and t-i (fp), e.g. man=t-a (whichef-s) 'which
one (f)?', w-i=nna (m-p=WH:ever) 'whichever ones (m)'. (ii) A paradigm
of possessive pronouns, i.e. pronouns followed by a genitive complement;
their heads may be w-i (m) or t-i (f), e.g. w-i n=!brahim 'the one (m) I those
(m) which belong(s) to Brahim', t-i=nn-un 'the one (f) I those (f) which
belong(s) to you (mp)"." (iii) Ordinal numerals formed on numbers greater
than one, e.g . w-i-ss-mraw 'tenth, m' , t-i-ss-mraw 'tenth, f'. (iv) a handful
of nouns such as w-ass 'ill-fated, m', whose feminine counterpart is
t-ass. Like the consonant-initial nouns (v. below) , these items have the same
form in the bound state and in the free state. We have assumed that the
prefix lu-I at the beginning of masculine nouns in the bound state is a marker
of the bound state. The forms in (i)-(iv) might instead be taken to suggest
that it is a marker of the masculine which occupies the same morpholog-
ical slot as the feminine prefix It-I. On such an analysis the underlying

31 Words of types (ii) and (iii) do not distinguish between singular and plural fonns .
34 CHAPTER TWO

form of free state masculine nouns would begin with the masculine marker
lu-I, e.g. the underlying form of a-drar ' mountain' (v. (36)I-a) would be
lu-a-drar/. In order to account for the lack of any surface reflex of lu-I at
the beginning of masculine nouns in the free state one could posit a special
rule deleting the masculine marker which would somehow be blocked in
bound state forms as well as in the pronominal forms above.
It may make good sense, from a historical perspective, to argue that
the prefix at the beginning of the masculine form s in (i)-(iv) above and
that at the beginning of the masculine bound state nouns in tables (35) to
(38) are descendents from the same morpheme, whieh was a masculine
marker in earlier times, but we doubt that this analysis is still tenable on
synchronie grounds . One fact suggesting that the two prefixes are no longer
instances of the same morpheme is that rule WI/YI does not affect pronouns
of the form w-i . . . , e.g. w-i=lli 'the aforementioned ones (m)' (m-peafore-
mentioned) cannot be pronounced as *yilli.32

2.5.3. Consonant-initial nouns

Let us now turn to the nouns which do not have an InV; we call them
consonant-initial nouns. Unlike vowel-initial nouns, consonant-initial nouns
do not distinguish between bound state and free state, and they do not
begin with prefix It-I when they are feminine. We interpret both proper-
ties as consequences of the absence of an InV, for we assurne that the bound
state marker lu-I and feminine It-I can only be prefixed to bases which begin
with a vowel. This assumption accounts for the absence of lu-I from those
vowel-initial nouns which are feminine (v. e.g. (35)II-c,d): in the nouns
in question the base to whieh lu-I would have to be prefixed begins with
a consonant, viz. It-I. Consonant-initial nouns are of two kinds, whieh we
will examine in turn.

2.5.3.1. IC-initial nouns

We dub those of the first kind the IC-initial noun s. They begin with an
IICI cluster (e.g. lbanan 'bananas, col'), or else they begin with a geminate
coronal (e.g. !rrgg 'ground'). Most ofthem are recent loanwords . They truly
belong to Imdlawn Tashlhiyt, because they are used even by those speakers
of Imdlawn Tashlhiyt who do not know Arabie. We assurne that noun-initial
geminate coronals are derived from underlying IICI clusters where ICI is
a coronal: at the beginning of nouns /11 completely assimilates to a following

32 On the other hand the genitive preposit ion Inl does assimilate to the initial glide of a
w-i . . . pronoun as it does to the initial vocoid of a bound state noun (on the assimilation
of genitive InI, see § 3.2.1.2). li -xf new-aenna/ 'the head of anyone (m)' can be pronounced
ixfn wanna or ixfuwanna.
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 35

coronal; !r-rgg derives from /l-Irgg/." The evidence in favor ofthis analysis
is twofold.
First there are alternations: the initial /l/ often surfaces in the plural, where
the initial cluster is broken up by a vowel. Here are a few examples:
(42) SG PL
lbrrad labrarid 'teapot'
!lhud !lahwad 'watering-trough'
ssuq laswaq 'market'
!ttrf !ladruf 'edge'?"
Hnb laznub 'pocket'
!ssrd llasrud 'condition'
Complementary gaps in the distribution of initial clusters and gemi-
nates also support our analysis. All geminate consonants which are found
to occur at the beginning of nouns are coronals, and aside from the
exceptions mentioned in note 33 there are no nouns which begin with an
lC cluster in which C is coronal." The proposed analysis provides a simple
explanation for that fact, if one assurnes that all geminate consonants are
forbidden word-initially in the underlying representations of nouns and
that the only source of geminates at the beginning of nouns is the complete
assimilation of IV to a following coronal.
Initial 111 is the reflex of the definite article 11/ of Arabic, where it is
subject to the well-known rule whereby it completely assimilates to a
following coronal: MA l-brrad 'the teapot' vs. brrad 'a teapot', ! i -irda 'the
garden' vs. !irda 'a garden'. Due to the massive borrowing of Arabic nouns
by Berber the assimilation rule has become part of the grammar of Berber.
But in Berber 11/ is no longer an autonomous syntactic unit ; it cannot be
omitted: *brrad, *!i rda. It is aprefix which is part of the lexical repre-
sentations of those nouns in which it appears. We consider it aprefix
rather than part of the kernel, because it is regularly omited by certain
processes of derivational morphology, which does not happen to consonants
which belong to the kernel, e.g. I is absent from a-kssab 'cattle farmer'
and a-ksasb-iy 'idem', which are derived from l-ksib-t 'cattle' .
As a rule lC-initial nouns which are feminine singular end with I-tl,
e.g. s-sbk-t 'fishing net', !l-blas-t 'place' . This suffix corresponds to -a,
the suffix which ends feminine singular nouns in MA and has the variant
-t in the construct state, e.g. before pos sessives, as in MA sbk-a 'a net',

33 There are exceptions, e.g. I-md 'ancestor' , l-inn-t 'paradise', l-zamaa ' Friday' .
34 In the singular /I-!drf/ yields /!ddrf/, whence lttrf. As a rule the geminate reflex of ld is
tu.
35 Nb: this generalization concems words, not kerneIs. Noncoronal geminates can be found
at the beginning of noun kerneIs which are not word-initial, as in a_gWgWrdi (p i-gWgWrda-n)
'flea ' , la-hhram (p ti-hhram-n) 'bastard' , a-bbankik (p i-bbankik-n) 'big stone'.
36 CHAPTER TWO

sbk-t=i 'my net', !blas-a 'a place' , lblas-t>i 'my place'. In a synchronie
account of Imdlawn Tashlhiyt there is no reason not to consider this I-tl
as the same suffix as that which occurs at the end of feminine vowel-
initial nouns.
In general the plural forms of the IC-initial nouns mimie those of their
counterparts in MA ; they resort to suffixes and nonconcatenative processes
akin to those of MA, e.g . lgns, p lgnus 'nation' (MA gns, gnus), lksibt,
p laksayb 'cattle' (MA ksiba, ksayb), lbrrad, p labrarid 'teapot' (MA brrad,
brard), !ssff, p Ilasfuf 'row, line' (MA !s@ff, !sfuf).
Like the vowel-initial nouns, IC-initial nouns can give rise to secondary
masculine nouns (augmentatives) and secondary feminine nouns (e.g.
diminutives, individuatives). These secondary nouns all have sterns formed
by prefixing a to the stern of the source noun . For instance lbanan 'bananas,
col' yields albanan 'big banana' and talbanant 'banana, indiv', and ssbkt
'net' yields assbk 'big net' and tassbkt 'small net'. Secondary nouns derived
from IC-initial nouns inflect like nouns whose In V belongs to the stern, v.
(35); their initial vowel is retained in bound state forms and it does not
become i in the plural, e.g . the bound state form of albanan 'big banana'
is walbanan (*ulbanan) and the plural of talbanant 'banana, indiv' is
talbananin (*tilbananin), b talbananin (*tlbananin).36 We noted earlier
that in the plural the source nouns often resort to nonconcatenative processes
and suffixes akin to those of MA. On the other hand the secondary nouns
only make use of the regular plural suffixes of Berber, viz I-ni in the
masculine and I-in! in the feminine . Whereas the plural of s-sbk -t 'riet' is
s-sbayk (v. MA sbk-a, p sbayk or sbk-at), the plural form of the augmen-
tative assbk is assbk-n (b w-assbk-n) and that of the diminutive t-assbk-t
is t-assbk-in.
There exist nouns which are phonologically and morphologically similar
to the It-a-I-CZI derivatives, and we shall analyze them as such, although
ICZI does not exist as an independent noun. Such is for instance the case
of aljluk ' small boat' (b w-aljluk, p aljluk-n); there is no *ljluk in Imdlawn
Tashlhiyt. Given the existence of the I-assimilation rule we can also analyze
as la-I-CZI a number of nouns of the form laC:ZI CC:' a coronal geminate)
whose initial a is part of the stern. Such is for instance the case of arrfad
'plot of land' (b w-arrfad, p arrfad-n), assbar 'barricade' (b w-assoar,
p assbar-n), whose phonological representations we shall take to be
/a-l-rfad/ and /a-l-sbar/.

36 In the Tiznit dialect, on the other hand, the initial a in those nouns behaves as an augment,
v. Deli and Jebbour 1995.
 SYNTAX AND MORPHOLOGY, AN OVERVIEW 37

2.5.3.2. Other consonant-initial nouns

Let us now turn to the remainder of the consonant-initial nouns . What

these nouns have in common is the fact that they do not resort to suffixa-
tion and/or nonconcatenative processes in the plural; they prefix id- (m)
or istt- (f) instead." They belong to the Berber stratum of the lexicon, or
else they are loans from languages other than Arabic. For the sake of
expository convenience we divide them in three classes.
The first dass comprises a few nouns which begin with w- if they are
masculine, and with t- if they are feminine." One of these is w-ass 'unlucky
person, ms ', fs t-a ss, mp id-w-ass, fp istt-t-ass; most of these nouns are
war-Z compounds meaning 'Z-less', e.g. war-laman 'undependable person,
ms', from laman 'confidence', fs tar-laman, mp id-war-laman, fp
istt-tar-laman.
The second dass comprises monomorphemic nouns such as millus
'slovenly brat', furi sti 'forester' , sbbalyun ' Spain' , and bu (m) and its
feminine counterpart mmu, which have already been mentioned in con-
nection with rule (40). Besides acting as heads in noun phrases of the
form bu+N and mmu-N whose meanings are fully compositional, bu and
mmu occur as first terms in a whole gamut of compounds which are
transparent to various degrees, e.g. !buykurayn or !biykurayn 'the one with
the sticks' (v. !i-kuray-n, 'sticks'), butagant 'wild boar' (v. t-agan-t 'forest'),
bufqqus 'kind of date'. Imdlawn Tashlhiyt does not have any other words
built on -fqqus , but the composite nature of bufqqus is apparent in the fact
that it contains two labial consonants, which is otherwise prohibited in
Imdlawn Tashlhiyt words.
The third dass comprises compounds in which both terms are 'full'
lexemes, e.g. Itri-bri 'kind of soup ', from liri 'scrape off' and bri 'grind',
frd-xxi 'u seless person', from frd 'graze' and xxi ' shit', slm-agrgr m 'kind
of insect', from slm 'eat (something powdery)' and a-g"gWrn 'flour' .
Like the IC-initial nouns, the other consonant-initial nouns give rise to
secondary masculine and feminine forms which begin with a-, but for the
consonant-initial nouns which are not IC-initial, the prefixed vowel is an
augment, i.e. it becomes i- in the plural and it drops in bound state forms .
Columns III and IV in (34) give the complete set of secondary feminine
forms corresponding to xixxi. Here are two other examples. sbbalyun
'Spain' , a-sbbalyun (b u-sbbalyun) 'Spaniard, m', i-sbbalyun-n 'id ., mp ',
t-a-sbbalyun-t (b t-sbbalyun-ti 'Spaniard, fs', t-i-sbbalyun-in (b
t-sbbalyun-in) 'Spaniard, fp'; the diminutive of bri-bri 'kind of soup' is
t-a-nribri-t (b t-hribri-ti, p t-i-luibrit-in (b t- hribrit-ini.

37 A few lC-initial nouns have plural forms beginning with id-, e.g. rribab (p id-rribab)
'single-stringed violin' , IkWmmiyt (p id-lk' mmiyti 'ceremonial dagger' . Some of these are
feminine, v. the second example . lC-initial nouns never use istt in the plural.
38 V. our discussion ofthe w-It- constrast at the end of our survey of the vowel-initial nouns .
 CHAPTER THREE

PHONOLOGICAL BACKDROP

In the first half of this chapter (sections 3.1 to 3.5) we review the proper-
ties of the geminates. In the remaining sections we present information about
several phonological phenomena of Imdlawn Tashlhiyt. These phenomena
are not central to the concems of this book but their effects are seen in many
examples and they come into play at one point or another of our discus-
sion.

3.1. PRELIMINARIES ON GEMINATION

Like the other dialects of Berber, Imdlawn Tashlhiyt has a lexical contrast
between two sets of consonants. The consonants in one set have greater
duration than their counterparts in the other set, and they are articulated with
greater energy. In this chapter we review the evidence that the underlying
contrast in question is indeed one of length, not of tenseness. The evidence
is overwhelming. Unless proof to the contrary is given, the same analy sis
must be assumed for the other dialects of Berber, as we shall argue .
Here are some minimal pairs illustrating the lexical contrast in question
in Imdlawn Tashlhiyt.
(1) a. i-mi 'mouth' immi 'Mom'
b. t-ut 'she struck' ttu=t 'forget hirn!'
c. kks=t 'remove hirn!' kks=tt 'remove her !'
d. i-grra 'he picked up' i_gWgWra 'he is last'
e. t-a-rnda 'pond' t-a-mdda 'brown buzzard'
f. y-ukr 'he stole' y-ukrr 'he dragged'
As illustrated in (1)b ,c, long and short consonants contrast after or before
a pause. The contrast in these contexts shows no tendency towards neu-
tralization, unlike in Ait Iraten Kabyle (Chaker 1984: 86).
Except in certain cases where it serves a morphological purpose, length
cannot be predicted from other properties of the words in which it occurs.
In general the presence of length in a morpheme and its location within
that morpheme remain invariant throughout all occurrences of that
morpheme. Like voicing or labiality, length must already be present in
the lexical entries of morphemes. Geminates are found in a wide range of
environments. We give a sample of these environments below. The examples
are grouped in quasi-minimal pairs to illustrate the relative freedom of
occurrence of length.

39
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
40 CHAPTER THREE

(2) a. a-brrkal 'old age (horse)' i-frkki 'bark'

b. a-hhram 'bastard' t-a-frruy-t 'fresh honeycomb'
c. la-ddzar 'neighbour' !a-dgWgWal 'in-law'
d. gWgW z 'go down' !gzz 'crunch'
e. !kkd 'poke (an eye)' bdd 'stand up'
f. zzf 'unveil' fss 'be silent'
g. skubbr 'crouch' !zzukrr 'drag'
h. a-zzbir 'kind of bag' a-zbbud 'act of pulling'
The stricture of long consonants is steady-state, without any momen-
tary relaxation in its time course. Although they are made of successive
parts the doubled symbols (mm, gWgW) should be taken as unitary. Like the
symbol gW in i-grra 'he pieked up ', the sequence gWgW in i_gWgWra 'he is last'
represents a single uninterrupted period of velar closure with labializa-
tion. mm and gWgW represent IPA [m:] and [gw:].
Besides sounding longer than their short counterparts the long consonants
of Imdlawn Tashlhiyt also give the impression of being articulated with
greater energy.
At present there is no consensus on the phonologieal nature of the contrast
between long and short consonants in Berber. Some authors see it as a length
distinction (v. below) while others take it to be primarily a tenseness dis-
tinction. Lionel Galand was the first to suggest the latter analysis and he
has recently written a whole article (1997) in support of it. We will argue
in favor of the former analysis and show that the long consonants of
Imdlawn Tashlhiyt are geminates.
Following a commonly accepted view, I we assume that a geminate is a
single melodie unit (i.e. a single feature bundle) associated with two
prosodie positions. Here are for instance the representations of (a) a simple
t, (b) a geminate t (i.e. tt), and (c) a sequence of two simple ts.

(3) a. b. tt c. t+t
x X X X X
I V
t
It It
t

In the diagrams in (3) each occurrence of X represents a prosodie position

and the letter t stands for the bundle of distinctive features whieh defines
[tl . More precisely, it stands for the Root node of the feature tree which

I V. Leben (1980) and McCarthy (1979, 1981). For recent overviews, v. Inkelas and Cho
(1993), Kenstowicz (l994a), Broselow (1995) and Perlmutter (1995).
 PHONOLOGICAL BACKDROP 41

defines [t].2 Throughout this book the term 'geminate' is used to refer to
doubly associated feature bundles such as (3)b.
We shall see that in Imdlawn Tashlhiyt almost all the properties of long
consonants accord with representation (3)b. Since in what follows almost
all the data is drawn from Imdlawn Tashlhiyt, our conclusions are only valid
for Imdlawn Tashlhiyt: each dialect of Berber must be described on its
own terms. Our discussion should nonetheless be benefieial to the analysis
of some other dialects of Berber, inasmuch as these show facts similar to
those presented below.
We now compare the merits of two conceptions of consonant length in
Imdlawn Tashlhiyt. One is what we will call the 'configurational' con-
ception ; it holds that in Imdlawn Tashlhiyt long segments differ from short
ones in the number of prosodie units that they are comprised of, v. (3).
According to the other conception , the contrast between the long and short
consonants of Imdlawn Tashlhiyt is one between two values of a distinc-
tive feature , say [tense] : ttJ and tttJ are identical in all respects but one:
ttJ is [-tense] whereas tttJ is [Hense] . Let us dub this conception of con-
sonant length in Imdlawn Tashlhiyt the 'featural' conception .
The term 'tense' (or 'fortis', 'strong', etc.) has been used in different
ways for different languages and to this day it lacks a commonly accepted
meaning, v. Catford (1977: 199-204) and Ladefoged and Maddieson (1996:
95-98) for discussions . We use the term with the second sense mentioned
by these authors . In that sense, tense segments require more artieulatory
energy than their lax counterparts. Actually the precise phonetic charac-
terization of [tense] will not matter for our arguments against the featural
view of length in Imdlawn Tashlhiyt. All that will matter is that according
to that view the only difference between the members of pairs such as tltt
resides in the specifieation of some binary feature F.

3.2. THE LONG SEGMENT AS A SEQUENCE OF TWO PROSODIC POSITIONS

The geminate consonant represented in (3)b is comprised of two X slots,

like the cluster in (3)c, but it is comprised of one feature bundle only, like
the simple consonant in (3)a. Such a representation leads one to expect
that the behaviour of the geminate in (3)b will in some respects be like
that of a consonant cluster, and, in others, like that of a single consonant.
The long segments of Imdlawn Tashlhiyt do indeed exhibit such a dual
behaviour, as we shall now see. In this section we review some circum-
stances in whieh long segments behave like sequences of two units.

2 Concerning the internal structure of segments we shall adopt the version of 'feature
geometry' advocated in Clements and Hume (1995 ), in which all the features defining a
segment are organized in a tree structure. The Root node is the node dominating the whole
tree.
42 CHAPTER THREE

Saib (1976) and Guerssel (1977) were the first to argue in detail that
the long consonants of Berber are sequences. They both discussed vari-
eties of Tamazight.' The facts they used to support their conclusions have
analogues in Imdlawn Tashlhiyt or in Ath Sidhar Rifian. These facts are
(a) fusion (v. § 3.2.1.1), (b) the fission of some final geminates in the imper-
fective, as when pf bdd 'to stand' yields impf tt-bdad.' and (c) similarities
between geminates and clusters in some processes involving schwa.'

3.2.1. Heteromorphemic geminates

The merger of two adjacent short consonants into a long one is a perva-
sive phenomenon in Berber. We present three such processes which are
found in Imdlawn Tashlhiyt. These will be seen at work in many of the
examples cited in this book. The purview of all three processes is limited
to sequences of short consonants. This shared property will provide us
with our first reason to prefer the configurational analysis.

3.2.1.1. Fusion 0/ adjacent short consonants into a long one

Morphology or syntax can create sequences of two identical short conso-
nants, as in /t-zri-t=t/ ' you overtook hirn', where the 2s PNG /t/ is
immediately followed by the doßrns pronoun /t/. In Imdlawn Tashlhiyt
such sequences of identical short consonants are in general" homophonous
with the corresponding long consonant. /t-zri-tet/ is homophonous with
/t-zri=tt/ 'she overtook her', where /u/ is the doßfs pronoun. Similarly
/gn-n/ 'they slept' (sleep-3mp) is homophonous with /g=nn/ 'put yonder!'
(putaor=dir). Similar mergers do not occur in analogous sequences in which
one consonant is long,? e.g. the sequence /t+tt/ is phonetically distinct
from /tt/ and /t/, and so are the sequences /tHt/ and /tHtt/.
In our view the homophonous forms in the last example have at the under-
lying level the representations in (4).

3 Willms (1962) had already reached the same conclusion for Kabyle, basing hirnself on
evidence about the phonotactics of Kabyle syllabIes.
4 Verb-final geminates are split by the 'chameleon', a vowel which is inserted to form certain
imperfective sterns, v. DE (1991: 96) and § 5.2 in this book.
5 On schwa epenthesis in Rifian Berber, v. § 6.5.

6 'In general', that is: unless the first consonant is released, which happens only in certain
contexts, about which v. § 6.3.3.
7 There are a few exceptions to this statement. After averb, for instance, a suffix or clitic
comprised of one short consonant which immediately follows its long counterpart merges with
it, e.g. Idd=d/ must be pronounced as though it were Idd/ in Is-bidd=d a-gzdi/ (cau-standedir
u(beam) 'raise the beam on this side !' .
 PHONOLOGICAL BACKDROP 43

(4) a. Ign-nl b. Ig=nnl

X X X X X X
I I I I V
g n n g n

In (4)b a single occurrence of the feature complex defining n is associ-

ated with units whieh correspond to two succe ssive points in time . The
relationship between (4)b and the corresponding realization ([gn:]) is a rather
direct one. In order to account for the homophony of Ign-nl with Ig=nnl
one must confront two problems, whieh we dub the problem of articula-
tory fusion and the problem of duration . Let us consider these problems
in turn and see how the two competing analyses of length in Berber fare
in dealing with them.
The underlying representation in (4)a contains two occurrences of the
feature bundle defining n. If the first n were realized as a full-fledged
consonant ending with the release of its oral constrietion, the realization
of Ign-nI would sound something like [gnän], whieh is not homophonous
with [gn:]. In order to account for the homophony of Ign-nI with Ig=nnl,
the two successive n sounds allowed by (4)a must be merged into a seamless
whole, or more precisely one must derive from (4)a a surface representa-
tion whose phonetie implementation may only comprise a single
uninterrupted coronal closure. This is what we call 'the problem of artic-
ulatory fusion' . Let us turn to the problem of duration. Having ensured
that the sequence In-ni in (4)a is realized as an uninterrupted coronal closure,
one must in addition make sure that that closure has the same duration as
the closure of Innl in the realization of Ig=nnl.
Simplifying somewhat," here is how we propose to solve these problems.
We posit a fusion process which merges into one two identieal feature
bundles which are adjacent , preserving at the same time their associations
with the prosodie positions .
(5) x X X X
I aI ---7 V
a o.
By (5) representation (4)a is tumed into a representation identieal with (4)b,
hence the homophony.
Let us now see how the featural analysis would account for the
homophony in question . (6)a and (6)b are the underlying representations
which such an analysis would assign to gn-n and g=nn .

8 v. § 6.3.3.2 für details .

44 C HA PT E R THREE

(6) a. gn-n b. g=nn

g n n g n
[-T] [-T] [-T] [-T] [+T]

As in (4), each letter repre sent s a bundle of distin ctive features, but for
the sake of con spicuousness we have represented on aseparate line the
values of the feature [tense] contained in each bundle. Given the repre-
sentations in (6), one could posit the fusion process (7 ), which is the
analogue of (5) within the featural analysi s.

(7) a a a
[-T] [-T] [+T]

(7) operates on a sequence of two short segments which are identical and
changes it into one occurrence of the corresponding long segment."
(5) and (7) solve the problem of articulatory fusion by identical means
but they differ in how they deal with the problem of duration." The oper-
ation specified in (7) is a peculiar one. It takes two identical segments which
are unmarked for a certain feature and blend s them into their marked coun-
terpart. The languages of the world do not possess parallel operations
involving other feature s. For instance a parallel operation involving voicing
would turn a sequence of two identi cal voiceless consonants into one occur-
renc e of their voiced counterpart , e.g. t+t > d, s+s > Z, etc. Rather, the
oppo site is common, e.g. d-vd > t.
Let us restate our point in a slightly different manner. Leaving out artic-
ulatory fusion, which (5) and (7 ) achieve in the same way, let us focus on
how they handle duration:
(8) a. X X --7 X X b. [-T] [-T] --7 [+T]
As shown in (8)a, (5) simply says that fusion is a process which leaves
the prosodie structure unch anged; it only changes how that structure is
associated with melodie units. On the other hand (8)b restates the fact that
the process in (7) specifies three values for the feature [tense], viz. 'minus'
for the first segment in the input, ' minus' for the second, and 'plus' for
the result of the fusion. The featural analysis of length does not explain why
the formulation in (7) requires precisely that combination of plusses and
minu sses rather than any other among the eight possible such combinations.
Wh y not, for instance, a pro cess which would change a short segment

9 Abdel-Massih (1968: 127ft) posits various instantiations of (7) for particular consonant
sequences in Ait Ayache Tamazight.
10 Evidently, someone upholding the featural analysis would rather talk about a problem
of tenseness.
 PHONOLOGICAL BACKDROP 45

followed by the eorresponding long one into their short counterpart

(Htt > r)?"
The proponents of the featural analysis might retort that it is more restrie-
tive than the eonfigurational analysis. If length in Imdlawn Tashlhiyt is
eharaeterized by a binary distinetive feature , the length eontrast ean only
be a binary one. But if length depends on the number of prosodie units
eontained in a segment, one ean eoneeive of segments eomprising more than
two prosodie units. One eould for instanee entertain treble segments like
that in (9).

(9) X X X
~
t

If Fusion is formulated as in (5), what prevents it from operating on

sequenees whieh eontain long eonsonants, ehanging for instanee It+tt/ or
Itt+t/ into the treble eonsonant represented in (9)?
This brings us to the other half of our argument. Imdlawn Tashlhiyt
possesses several rules of total assimilation. Like the rule of fusion pre-
sented above, these rules only operate on sequenees of short segments.
The eonfigurational analysis allows us to eonsider this eommon restric-
tion as a eonsequenee of aseparate eomponent in the grammar, as we
shall see. Under the featural analysis, on the other hand , the similarity
between fusion and the rules of total assimilation eannot be faetored out.
Treble segments like (9) are not found at any level in the derivations
in Imdlawn Tashlhiyt. All representations in Imdlawn Tashlhiyt meet the
following eondition: 12
(10) NO-TREBLE: A melodie unit (i.e. a Root node) may not be
assoeiated with three prosodie positions which
are adjaeent.
In the text above (4) we stated that fusion only affeets adjaeent eonso-
nants whieh are both short, e.g . It+ttl remains unaffeeted by rule (5). If
(5) did merge the two eonsonants in It+tt/, the merger would yield (9), in
violation of NO-TREBLE.
NO-TREBLE eonstrains all the representations in the phonological eom-
ponent. In partieular it eonstrains those ereated by assimilation rules. In
addition to the i-assimilation rule whieh operates in Arabic loans (v.
§ 2.5.3 .1) Imdlawn Tashlhiyt has two rules whieh eompletely assimilate a

11 The special fusion exemplified in note 7 does not weaken our argument against (7). In
order to account for this fusion , (5) and (7) must be supplemented with devices of com-
parable complexity.
12 The formulation in (10) is only a first approximation, v. § 6.3.3.
46 CHAPTER THREE

segment to the one which follows it. Both rules involve grammatical mor-
phemes which occur very frequently and their effect will be seen in many
places in this book.

3.2.1.2. The genitive preposition

The genitive preposition In! completely assimilates to the initial segment
of the following word if that segment is a sonorant other than a. This assim-
ilation is optional. It is exemplified in (11), where a slanted line separates
two pronunciations which are both acceptable. Assimilation does not take
place in the first example because the word which follows the preposition
does not begin with a sonorant.
(11) a. a-qssab n dadda 'rny older brother's smock'
b. a-ydi n msaawd I aydi m msaawd 'Messaoud's dog '
c. a-yda n l-qqayd I ayda 1 lqqayd 'the chief's estate'
d. t-i-gira n !rmdan I tigira r !rmdan 'the end of the Ramadan'
e. i-xf n u-vyul I ixf u wvyul 'the donkey 's head'
f. i-xf n i-zikr I ixf i yzikr 'the end of the rope'
g. i-xf n w-aassas I ixf u waassas 'the watchman's head'
h. ifr n !wizugn I if!r u wizugn 'the cicada's wing '
1. lgdr n yumayn I lgdr i yumayn 'two days'time'
Before we proceed, a word of explanation is in order for lines e and f. In
line e, u-yyul is the bound state form of a-vyul. On the syllabifieation of
the bound state prefix, v. § 7.4. The assimilation of In! to the bound state
prefix creates a geminate high vocoid, which is syllabified as a vowel-
glide sequence since it occurs between two consonants." Similarly, in line
f, i-zikr is the bound state form of i-rikr. Note also that lfi.l does not trigger
the assimilation, as one would expect if f were a sonorant, as has some-
times been suggested: a-fus n !fumar 'Omar's hand' (*afus !ffumar).
The assimilation rule is given in (12) .

(12) N-ASSIM (OPT): x ]p N

f- -------J
n [+son]
-low

Rule (12) says that if n is located at the end of apreposition and is followed
by a nonlow sonorant, that sonorant spreads onto the prosodie position
occupied by n (i.e, it becomes associated with that position) and the link

13 After a vowel the geminated high vocoid created by assimilation is pronounced as a

geminate glide, e.g. t-a-wada n u-vyul I tawada w w-vyul 'the donkey's gait' .
 PHONOLOGICAL BACKDROP 47

between n and that position is deleted. The unassociated n will subsequently

be deleted, in accordance with the Stray Erasure Convention , which deletes
all unlinked material." As a result of the rule, the sequence Inrn/ in (l1)b
is for instance changed into a geminate m.
Assimilation does not occur when the sonorant following Inl is a
geminate . The prepositional phrases in (13) only have one acceptable pro-
nunciation.
(13) a. n lluz I *llluz 'of the almonds'
b. n lrrg I *Ir rrg 'of the turban'
c. n mmu t-wnza I *m mmu twnza 'of the one (f) with the fringe'
NO-TREBLE (10) accounts for the fact that long consonants cannot
assimilate In! . There is nothing in rule (12) itself which would prevent it
from applying in In lluzl (13)a , for instance, but if it were to do so, it
would create a treble I, in violation of NO-TREBLE.
Let us now see how these facts would be accounted for under the featural
analysis of length in Imdlawn Tashlhiyt. We give in (14) an alternative
formulation of the assimilation rule which is compatible with that analysis.
To make things simpler we have left out the condition which requires In!
to be the last segment of apreposition.
(14) n o [+tense]
[ ~:~n:e]
-low
2 1 2
Rule (14) takes a sequence comprised of n and a short nonlow sonorant
and changes it into the corresponding long sonorant. The rule calls for
two comments.
The first comment is analogous to one we made in the preceding section
concerning the fusion rule. It has to do with the fact that the assimilation
gives rise to a long segment. If one compares how rules (12) and (14)
deal with length, one finds again the changes portrayed in (8)a and (8)b,
and the remarks we made earlier about these are also valid here.
Our second comment concerns the specification [-tense] on the left-
hand side of the arrow in (14), whose presence is necessary to prevent
the rule from applying when the noun begins with a long segment. If it
did, this would amount to simply deleting In!, which is incorrect, e.g. the
rule would predict, incorrectl y, that /nelluz/ 'of the almond ' can be pro-
nounced homophonous with Illuzl 'almond' . Within the featural analysis,
then, one must incorporate an ad hoc specification into the assimilation
rule in order to prevent it from applying when the noun begins with a

14 On Stray Erasure v. e.g. Marantz (1982: 446) .

48 CHAPTER THR EE

geminate. No ad hoc restriction on the rule is necessary within the con-

figurational analysis , on the other hand, because assimilation by geminates
is precluded by an independent constraint, viz. NO-TREBLE (10).
To sum up: in Imdlawn Tashlhiyt one property which fusion and complete
assimilation have in common is that they only affect sequences of short
consonants. The configurational conception of length allows us to ascribe
that property to a component of the grammar (NO- TREBLE) which is
independent of the fusion and assimilation rules. Furthermore that com-
ponent may very weIl belong to UG; we know of no counterexample to
NO-TREBLE in other languages. The featural conception of length does
not allow us to factor out that similarity between fusion and complete
assimilation, thus forcing us to treat it as a coincidence.

3.2.1.3. (R)AD's final consonant

When the complementizer ladl and the future marker Irad/ precede certain
grammatical morphemes, the final Id/ drops in certain cases, and completely
assimilates to the following segment in others . Complete assimilation is
obligatory in certain contexts and optional in others. For a review of the
various morphosyntactic environments with systematic exemplification, v.
DE (1989: 188-190). A few examples will suffice here.

(15) a. /radezwar i-ssl ra[dz]war - ra[zz]war

fut=first 3ms-eataor
'he will first eat'
b. Irad=fll-i i-zril ra[ff]lli
futeon-Is 3ms-pass
'he will pick me up on the way'
a'. lur rad=zzu i-Ikrn! ra[zz]u
neg fut=never 3ms-arrive
'he will never arrive'

b' . Irad=ssgi-s i-ffil ra[ss]gis

fut=from-3s 3ms-pour
'he will pour from it'

Assimilation is optional before an adverb (v. (15)a) and obligatory before

a pronoun (v. (15)b), but d fails to assimilate when it precedes a long con-
sonant, and it deletes instead, v. (15)a',b'. (R)AD 's behaviour thus
strengthens our case in favor of constraint NO-TREBLE and of the con-
figurational analysis of length in Imdlawn Tashlhiyt.
 PHONOLOGICAL BACKDROP 49

3.2.2. Syllable structure

This subseetion antieipates the results of our diseussion of syllable strue-

ture in Chapter 4.
In Imdlawn Tashlhiyt a geminate eonsonant ean belong to two sylla-
bles at onee, whereas a simple eonsonant eannot. Consider the following
examples.
(16) a. imlul 'be white aor' (im.lul)
b. i-mIlul 'be white pf 3ms' (i.ml.lul)
e. a-mlhaf 'haik' (a.ml.haf)
Aeeording to native judgments the first word is disyllabie and the other
two are trisyllabie, and furthermore in the trisyllables the peak of the seeond
syllable eoineides with I. In our view, these judgments are a refleetion of
the syllable struetures that the words in question have in the terminal rep-
resentations of the phonologieal eomponent. We take these syllable
struetures to be as given in parentheses after the glosses in (16) (a dot stands
for the edge of a syllable). The syllabie parsings in (16) are nothing but
eonvenient stand-ins for the representations in (17).

(17) a. im.lul b. i.ml.lul

o o

I1 If\
o o o

f\ If\
I
N
I
X X X X X X X X X X X
I
m
I I
I
I
u
I
I
I
m
I <::
I
I
u
I
I
c. a.ml.baf

I1 If\
o o o
I
N
I
X X X X X X
I I I I I I
a m I li a f
The units whieh are grouped into syllables are not the melodie units but
the prosodie positions. In (17)b the first position in 111/ is the nucleus of
the second syllable; the second position is the onset of the third syllable.
Through the mediation of the skeleton, then, the feature bundle /1/ in imllul
belongs at once to two syllables,
The difference between a short eonsonant and a long one is in some cases
sufficient to ereate a differenee in syllable count, as in the contrast between
50 CHAPTER THRE E

imlul and imllul. It can also create a difference in syllable weight in poetry,
as we now explain.
The versification of Tashlhiyt distinguishes between light and heavy
syllables . Let us say as a first approximation that a syllable is light if its
nucleus is syllable-final and heavy otherwise. In (17) the light syllables
are i and ml (v. (17)b), and a and ml (v. (17)c); all the others are heavy.
The relevant fact for the present discussion is this: a syllable is heavy if
its nucleus belongs to a tautosyllabic geminate. Consider the words gn
'sleep!', g=nn (put.aoredir) 'put yonder!' and g-nt (putaor-3fp) 'let them
pul' . When they precede a word beginning with CV, their boundaries are
also syllable boundarie s; gn is a light syllable, whereas gnn is heavy, like
gnt. The three syllables are represented in (18).

(18) a. gn b. gnn c. gnt

er er er

X
;1 X X
ffi X X X
ffi X X

g
I I
n g
I Vn g
I I
n
I
t
Syllable (18)b is heavy: like that of syllable (18)c, its nucleus is not syllable-
final. The representations in (17) and (18) are examples of how length
and syllable structure mesh if length is represented configurationally. The
proponents of the featural analys is of length have yet to indicate what
their conception of syllable structure is and how the representation of length
and that of syllable structure are interrelated."
Consonants can be syllable nuclei in Imdlawn Tashlhiyt and the above
discussion relies crucially on this fact. The bipositional nature of long
consonants is no less evident in the syllabification of those dialects of Berber
which allow only vocoids as syllable nuclei, v. § 6.5.

3.2.3. Templatic morphology I

Templates in Imdlawn Tashlhiyt are discussed in DE (1992). In this section

we limit ourselves to the facts relevant for the analysis of length.
We will draw our evidence from a class of nouns and adjectives which
have the form uCCiC and are derived from verbs. A few examples are given

15 Ouakrim (1995 : 102) claims that in Ihahan Tashlhiyt tense consonant s (what we call
geminates) cannot be ambisyllabic. Take t-nna ' she said ' , which is disyllabic in Imdlawn
Tashlhiyt (tn.na). If it is also disyllabic in Ihahan, the only syllabifications compatible with
Ouakrim's claim are t.nna and tnn.a.
 PHONOLO GI C AL BA CKDROP 51

below. Eaeh line eontains, in this order, a derived ward, the verb it is deri ved
from and their respeetive meanings.
(19) a. ukris krs 'trousseau / tie in a bundle'
b. urkim rkm 'rotten / rot'
e. !uttid !ttd 'eoagulated blood / eoagulate '
d. ukkim" kWkwm ' a blow / strike'
e. !ugziz !gzz 'mouthful / eruneh'
f. t-ugmim-t grrnm 'mouthful / hold (liquid) in one 's mouth'
We will eall derived words sueh as these 'UKRIS words ' , after the example
(19)a. How are the UKRIS word s derived from the eorresponding souree
verbs? For the purposes of the pre sent di seus sion we will eontent our-
selve s with a simplified view of templatie word-formation, v. DE (1992) for
a fuller view . Let us assurne that the UKRIS word s result from mapping
the souree verb s onto the template displayed in (20)a .

(20) a. uCCiC template b. urkim

u i u i
I I I I
V c C V C V C C V C
e. !uttid I I I
r k m
u i
I I
V c C V C
V I
!t d
In (20) a, C repre sent s a prosodie position (an X slot) whieh ean only be
associ ated with a eon sonant, and V, a prosodie position whieh ean only
be assoeiated with a vowel. In the template in (20)a - and arguably in all
the other templates in the derivational morphology of Imdlawn Tashlhiyt
- all the V positions are already provided with melod ie unit s. To form an
UKRIS word from a verb , one assoei ate s one eonsonant (and only one)
in that verb with eaeh C position in template (20)a, e.g. mapp ing /rkm/
onto the template yields the repre sentation in (20)b , i.e. urkim (v. (19)b) .
Let us now see how mapping onto templates deal s with eonsonant length.
Note first that a single C position in the template is never oeeupied by a
long eonsonant; form s like ummli s or umllis are not possible UKRIS word s.
Thi s faet follow s direetly from the template (20)a. The template ean only

16 In free variation with ukwkwim. In Imdlawn Tashlh iyt labiali zed co nsonants optionally
del abiali ze when they occur to the right of a rounded vowel belong ing to the same word.
We only give the delabialized varia nts in order to avoid c1utteri ng.
52 CHAPTER THRE E

give rise to words of the form VCCVC, whereas the shape of ummlis and
umllis is VCCCVC under the configurational analysis of length. Under
the featural analysis of length, on the other hand, the canonieal shape of
ummlis and umllis is VCCVc. Within that analysis, templatic mappings must
be regulated by the following convention.
(21) Individual C positions in templates must be allowed to asso-
ciate only with [-tense] consonants.
The behaviour of long consonants in templatic mappings is illustrated
by the examples in (l9)c-f. Examples (l9)c,d illustrate the fact that a ce
sequence in the template can be occupied by a long consonant. Under the
configurational analysis of length this fact follows directly from template
(20)a, e.g. one can see from the representation of !uttid «(l9)c), which is
given in (20)c above , that its canonical shape is indeed VCCVc. Under
the featural analysis of length, on the other hand, convention (21) must
be supplemented with a codieil allowing [+tense] consonants in derived
words to count as ce sequences in the template. This codieil is yet another
manifestation of the equivalence between a long consonant and a sequence
of two short ones. The configurational theory of length acknowledges the
central position of this equivalence: it takes it as a primitive and incorpo-
rates it into the phonologieal representations, v. (3) . Under the featural
analysis of length, on the other hand, the equivalence in question goes unrec-
ognized, and it is forced to manifest itself in different guises: in the
conditions of phonologieal rules (v. (8)b) or in the conventions on how to
interpret templates .
The proposed emendation to convention (21) allows a long consonant
to be equivalent to a sequence of two short ones in the case a derived
word is examined in order to determine whether it matches the template.
Even emendated thus, convention (21) is still inadequate, however, for it
has nothing to say about examples (l9)e,f. The purpose of these examples
is to show that a long consonant in the source word can give rise to two
short ones in the derived word. A detailed account of such cases can be
found in our general discussion of templates in Imdlawn Tashlhiyt in DE
(1992). Our analysis there relies crucially on the fact that the long conso-
nant in the source verb, e.g. in !gzz in (l9)e, contains two prosodie positions.
All the data in the discussion above are drawn from the UKRIS forms,
but Imdlawn Tashlhiyt possesses other classes of templatic words which
present the same problem for the featural analysis of length.'?
Templatic forms will again be brought in in the next section, where
they will provide evidence on another aspect of the structure of long
segments.

17 v. DE (1991: 96-99), DE (1992).

 PHONOLOGICAL BACKDROP 53

3.3. THE LONG SEGMENT AS A SINGLE MELODIC UNIT

Let us now review the Imdlawn Tashlhiyt evidence that geminates are
comprised of a single melodie unit (a single bundle of distinctive features),
as implied by representation (3)b.

3.3.1. Templatic morphology I1

An important factual generalization about the templatic morphology of

Imdlawn Tashlhiyt is the following: whereas derived forms do not retain
any trace of the vowels of their source forms, the consonants of the source
forms are retained in the derived forms, where they all appear unchanged
except for length . We will refer to this property as 'consonantal invari-
ance'. Consider the following UKRIS forms:
(22) a. utlif tlf 'lost soul/be confused'
b. ubbiz bbz 'a punch / to punch'
c. !ugziz !gzz 'mouthful / crunch'
d. ulmis lmmus 'something bland / be bland'
e. uqsif qssf 'squat person / be narrow'
f. udmiv dmmv 'mentally retarded person / be retarded'
Forms a to c are only given for the sake of comparison. They are similar
to examples given earlier in (19). Forms d to f illustrate one aspect of the
special status of length in consonantal invariance : in certain circumstances
a consonant which is long in the source form has a short reflex in the derived
form . Consider for instance the fact that mapping qssf onto the uCCiC
template yields uqsif (v. (22)e). (23)a is template (20)a, repeated here for
convenience, (23)b is the phonological representation of uqsif and (23)c
is that of qssf.

(23) a. uCCiC b. uqsif

u i u 1

I I I I
V C C V C V C C V C
c. qssf I I I
q s f

C C C C
I V I
q s f

The template contains three C positions; it can accomodate either three short
consonants , as in utlif (22)a, or a long consonant and a short one, as in ubbi:
54 CHAPTER THRE E

(22)b. In going from qssj (23)c to uqsif (23)b the feature content of the
source is preserved in its entirety : both forms contain the same sequence
of consonantal melodies: q, s, f Cases where a consonantal melody gets
lost in a templatie mapping are only a handful in Imdlawn Tashlhiyt, e.g.
t-i-vzi 'length' , whieh results from mapping yzzif 'be long' onto the template
(C)CCi (conformity with consonantal invariance would have yielded t-i-yzJi
instead) .
Consonantal invariance has to do with melodie units (feature bundles),
not with prosodie positions (C positions), and it is one of the merits of
the theoretical framework adopted here that it provides a plane of repre-
sentation on whieh a long consonant is equivalent to its short counterpart:
on the melodie plane Issl consists of one feature bundle, like Is/.
The fact that long consonants sometimes become short in templatic
mappings certainly does not make life easier for the proponents of a featural
analysis of length; but above all it poses a direct challenge to another
view of length, the strictly sequential view, according to which a long
segment is simply a sequence of two identical consonants.
Imagine that we adopt unilinear representations in the manner of
Chomsky and Halle (1968), and that instead of being a single feature bundle
s associated to two positions (v. (23)c), the long consonant in qssj is rep-
resented as two identical s feature bundles standing side by side. The
representation of qssjis a sequence of four feature bundles , but the template
has only room for three of them, so one of the four must fail to transfer
to the derived form. How does one explain the fact that the feature bundle
left out in the transfer is one of the s bundles , rather than q or f, i.e. why
is the derived form not ussif or uqsis, instead of uqsif? Of course, uni-
linear representations could be supplemented with conventions which would
allow one of two adjacent identical segments to be disregarded for certain
purposes. But then it would have to be shown that the resulting frame-
work would not be a notation al variant of that adopted here.
One might object that our argument is without force because it relies
on assumptions about the templatic morphology of Imdlawn Tashlhiyt which
are erroneous. We have been tacitly assuming that in derivations with tem-
plates, what is mapped onto a template is the source word, but what is
mapped instead, our opponent might claim, is only the source word's con-
sonantal root, as in root -and-pattern descriptions of Arabic and Hebrew.
While we are assuming that it is the form qssj itself whieh is mapped onto
the uCCiC template to derive uqsij, an advocate of the strietly sequential
conception of length might claim that (i) the source word qssj is itself the
result of mapping a consonantal root Iqsfl onto a template CC;C;C and
that (ii) it is the root Iqsfl rather than the word Iqssfl which is mapped
onto the uCCiC template to derive uqsif. Regardless of the validity of (i),
a question whieh we leave open for the time being, (ii) is untenable.
According to our opponent's assumptions, in the derivation of ubbi; from
 PHONOLOGICAL BACKDROP 55

bb: (v. (22)b) and of lug zi: from !gzz (v. (22)c), the objects mapped onto
the uCCiC template would be the roots Ihzl and I!gz/, but then there would
be no explanation for the fact that in Imdlawn Tashlhiyt the templatically
derived forms preserve as much as possible the length of the consonants
in the source words; bb: does not yield ubzii. nor does !gzz yield luggiz:

3.3.2. Feature changes in long consonants

One prediction implicit in the assumption that long segments contain a single
feature bundle is that an alternation cannot involve one 'half' of a long
segment while leaving the other 'half' invariant. Imagine a language with
a length contrast and a rule which rewrites Itl as s before i. Ittl would not
be realized as ts in that context: since Ittl contains a single feature bundle,
it contains in particular a single occurrence of the specification [-cont]
that could be changed by the rule ; Ittl can either become Isst or remain
Itt/. Imdlawn Tashlhiyt has a number of processes which affect long
segments and which agree with that prediction, e.g . the realization of
emphatic Iddl as [!tt], the occlusivization of hryl and l"{w"{w/, which surface
as qq and qwqw,18 sibilant harmony (v. § 5.4) and labial dissimilation." lt
also has two processes which contradict the prediction in question.
The vowel i is optionally lowered to [e] after h, e.g.rtthi-n/ 'they jostled'
may be realized as [tthen]. When it occurs between two consonants, Iyyl
is normally realized as [i:], e.g./nyysl niys [ni.s] ' aim' . When the preceding
consonant is h, however, Iyyl is realized as [ey], that is, only Iyyl's first half
undergoes lowering. l1iyydl 'step back', which we transcribe as hiyd, is
actually pronounced as [heyd] . This is an impossible outcome in the frame-
work adopted here. Given our assumptions about the representation of
length, Iyyl can only surface as [i:] or as [e:] in the context in question.
The other problematic case also has to do with the pronunciation of
geminate glides. When a geminate glide occurs between a consonant and
a vowel it is realized as a high vowel followed by a glide (v. § 7.4) , e.g.
Ihyyal 'be magnificent' is pronounced hiya. The surface representation of
Ihyyal is given in (24)a.

o o

11
(24) a. o o b.

;1 ;1
X X X X X
;1 X X X
I ~I I ~ I
h I a h I I

18 /ww/ is also reali zed as gWgW in certain morphological environments, v. § 7.4 .

19 On labial dissimilation, v. Elmedlaoui (l995a: 43-78).
56 CHAPTER THREE

As will be seen in § 4.6, we assume that a glide and the eorresponding

high vowel share the same melodie unit, e.g. i and y are eomprised of the
same feature bundle, whieh is represented as 'I' in (24) . The symbol ' i'
stands for a syllabie I, and the symbol 'y', for a nonsyllabic one, henee
the transeription 'hiya' for the objeet depicted in (24)a. Consider next the
neg 3ms form of hiya, which is i-hiyi. The representation of its kerneI is
given in (24)b. In that word the sequenee we transeribe as iyi sounds like
that in the Freneh word Chantilly [sätiyi]: the glide gives the impression
of having a greater degree of eonstrietion than the adjaeent vowels.
Strengthening of glides in onset position is a eommonplaee phenomenon,
V., e.g., Harris and Kaisse (1999) for Spanish, Rubaeh (1993) for Slovak,
Booij (1995) for Duteh. If, as in the preeeding referenees, the strength-
ening of onset glides in Imdlawn Tashlhiyt is a phonologieal phenomenon,
i.e. if it must be represented by the feature contents of the terminal repre-
sentations of the phonological eomponent of the grammar, it presents a
serious problem for the view of length adopted here, sinee in /hyya/ and
/i-hyy-i/ the feature specification representing the strengthening must only
be assoeiated with the seeond half of the long glide, which is impossible,
given the representations in (24).
The two facts just diseussed are problematic for the eoneeption of length
we have adopted," but we will retain that eoneeption here , as we are
unable to propose a better one. Note that these two facts ean give no eomfort
to the advoeates of the featural analysis of length in Berber, sinee what these
two facts challenge is the assumption that long segments are eomprised
of a single feature bundle, an assumption whieh is also part of the featural
analysis.

3.4. "TENSION"

The previous literature eoneerning Berber geminates has dealt with two
issues:
(i) Is a geminate a sequenee or a single segment?
(ii) What are the phonetic eorrelates of the simple vs. geminate eontrast,
and what is their phonologieal relevanee?
Sinee Saib (1976)21 and Guerssel (1977), who were the first to diseuss
question (i) in any detail, all the authors working on Berber within gener-
ative grammar have adopted the eonfigurational analysis in one form or
another, depending on their particular views about the strueture of phono-
logical representations, v. e.g. Elmedlaoui (1985, 1988), DE (1985, 1992,

20 Hayes (1990) and Selkirk (1990) discuss other problems with that conception.
2\ In chapter three, published as Saib (1977).
 PHONOLOGICAL BACKDROP 57

1996a), Bader and Kenstowicz (1987), Boukous (1987a, b), Dell and Tangi
(1992, 1993), Jebbour (1995, 1996), Bendjaballah (1995). Outside of gen-
erative grammar, on the other hand, the discussion of question (i) has been
summary at best," and a featural analysis of length has in general been
accepted, v., e.g., Chaker (1984), Galand (1988, 1997).
Answering question (ii) has been a more pressing concern for the
advocates of the featural analysis of length than for those of the configu-
rational analysis. One reason for this is easy enough to see. If one holds that
geminates and simple segments differ only in their distinctive features,
one cannot give a complete distinctive feature analysis of the underlying
inventory unless one is able to say what the distinctive feature involved
in the length contrast ls. On the other hand, if one adopts a configura-
tional conception of length, the phonetic correlates of the length contrast
may be considered irrelevant to the functioning of the phonological com-
ponent.
Most proponents of the featural analysis accept Galand's view that the
distinctive feature involved in the length contrast in Berber is tension:
what we call geminates are tense whereas their simple counterparts are
not.
To its advocates, tension seems best suited to provide a common source
to the following four phenomena :
(A) the greater duration of the closure period in long consonants, V.
Applegate (1958: 13) on Ifni Tashlhiyt, Chaker (1975) on Ait Iraten Kabyle,
Ouakrim (1993)23 on Haha Tashlhiyt and Louali and Puech (1994) on
Tiznit Tashlhiyt.
(B) Long consonants seem to require greature articulatory energy. This
is suggested by the palatograms in Mitchell (1957) on the dialect of Zuara
(Libya) , by Louali and Puech 's air pressure tracings and by the greater
acoustic energy in stop bursts in the spectrograms in the works by Ouakrim
and by Louali and Puech.
(C) Whenever a simple consonant and its geminate counterpart differ
in continuancy, the long consonant is a stop whereas the short consonant
is a fricative, and not the other way around," V. the pairs ylqq and wlgWgW
in Imdlawn Tashlhiyt. In some dialects all the simple stops in the native
vocabulary have become fricatives while their geminate counterparts have
remained stops, v., e.g., Saib (1974) on Ayt Ndhir Tamazight and Chaker
(1984) on Ayt Iraten Kabyle.
(D) Whenever a simple consonant and its geminate counterpart differ
in voicing, the short consonant is voiced and the long one is voiceless,

22 E.g. Chaker (1984: 84-85). Kossmann (1994: 59-60) poses problem (i) and discusses
various relevant facts in Figuig, but in the end he leaves the question unanswered.
23 Published with slight revisions as Ouakrim (1995).
24 V. Elmedlaoui (1993) for some discussion.
58 CHAPTER THREE

and not the other way around, V., e.g. , the realization of I!dd/ as [!tt], a
pan-berber fact, and the devoicing of geminate /d/ and /b/ in Zenaga Figuig
(Saa 1995).
In DE (l997b) we review the various arguments which have been
adduced in favor of a featural analysis of length in Berber. We refer the
reader to that article , where it is argued that facts (A) to (D) are no reason
for preferring a featural analysis."

3.5. CONCLUSION ON THE GEMINATES

In view of the above discussion, the nature of the length contrast in Berber
dialects should not be a controversial matter. The burden of proof clearly
lies with the proponents of a featural analysis of length.

3.6. DORSOPHARYNGEALIZATION

This section deals with emphasis (i.e. dorsopharyngealization) and its dis-
tribution in phonological representations and at the surface level." In a
nutshell: In phonological representations, emphasis is a property of indi-
vidual segments; Imdlawn Tashlhiyt has a set of emphatic phonemes, all
coronal. At the surface level, on the other hand, emphasis can affect all
the segments comprising a word or even a sequence of words. For instance
/t-i-grmmad-in/ '(river) banks, dirn' only contains one emphatic segment,
/Q/.27 The pronunciation of this word sounds something like [turg''m.udurn] ,
with emphasis spread over the entire word.
The articulation of dorsopharyngealized segments in Imdlawn Tashlhiyt
and their featural representation are discussed in Elmedlaoui (l995a: 161ss).
As is the case in other dialects of Berber the presence of emphasis in a word
cannot be predicted on the basis of other phonological properties in that
word, witness the existence of minimal pairs in which the only distin-
guishing feature is emphasis :
(25) a. a-fud 'knee' !afud ' leave!'
b. lZl 'fly' !izi 'gall'
c. ndr-n 'they moaned' !ndr-n 'they flew off'
d. frd 'graze!' !frd 'clear (of undergrowth)!'
e. smsr 'pursue!' !smsr 'trade! '

25 In that article we also examine an argument by Ouakrim (1995) based on duration

measurement s in Ihahan Tashlhiyt. We suggest that the author's conclusions probably rest
on a faulty phonological analysis of the data on which the measurements were made.
26 On emphasi s in Tashlhiyt, v. Elmedlaou i (1985), Boukous (l987a) and the references
cited in these works.
27 In §3.6, emphasis is indicated by underlining or by ' !' . The two are equivalent .
 PHONOLOGICAL BACKDROP 59

When examining an expression (a word or a group of words) there is

never any doubt whether it contains emphatic segments, and in an expres-
sion in the surface forms of which there are both emphatic and nonemphatic
segments it is often possible to obtain clearcut judgements about the exact
location of the endpoints of an emphasis span, even if they fall in the middle
of a consonant cluster.

3.6.1. Auditory properties

We present below some of the auditory impressions to which emphasis gives

rise in Imdlawn Tashlhiyt. We deal first with the vowels. Setting aside
certain special cases, lu, i, al are realized as [u, i, re] in nonemphatic
contexts. In emphatic contexts, on the other hand, lul sounds like [0], lil
sounds like a lowered [W]28 and Ia! is close to [0].29 Recall that at the
phonetic level Imdlawn Tashlhiyt also has three long vowels: [u:] for
tautosyllabic UW, [i:] for tautosyllabic iy and [re:] for aa. In emphatic.
contexts these long vowels are pronounced with the same color as their short
counterparts, but longer. Here are words exemplifying the emphatic vowels:
I!bdul 'share!', I!mdil 'taste!', l!bda-nI 'they shared', /i-Iduws/ 'it squirted' ,
li-!siyfl 'he spent the summer', i-!daaJ'he is lean'. In this book the emphatic
variants of lu, i, a! and their long counterparts are most of the time notated
simply as ' u' , 'i' and 'a' to avoid the proliferation of phonetic symbols.
For a nonnative speaker becoming acquainted with Imdlawn Tashlhiyt,
the auditory differences between the emphatic realizations of the vowels
and their nonemphatic realizations are easy to perceive from the start.
Only a little training is needed to learn to detect the corresponding differ-
ences in consonants. Whether a given consonant is plain or emphatic is often
clearly audible even in the absence of any neighbouring vowel. The con-
striction phase of emphatic sonorants and fricatives gives rise to an auditory
impression which is different from that of their nonemphatic counterparts.
In noncontinuants the explosion burst sounds different in emphatics and

28 This sound is identical with the emphatic variant of lil in Moroccan Arabic, which various
authors describe as [eI, erroneously in our opinion.
29 Ia! is furthermore subject to a phenomenon of prepausal backing. Immediately before a
pause it is realized more posterior than its nonprepausal counterpart , e.g. whereas Isala-n/
'they finished ' is realized as [srelzen], before a pause Isala! ' finish!' sounds more or less
Iike [salu] (v. Heath (1987: 23) for a similar phenomenon in Moroccan Arabic). Prepausal
backing also occurs in emphatic contexts, e.g. Ia! in prepausal It-!bda! 'she shared' is even
more posterior than Ia! in I!bda-n/ 'they shared' . Prepausal backing does not obliterate the
difference between emphatic a and nonemphatic a; the vowel in It-!bda! ' she shared' and
that in It-bda! 's he began' are both back a's, but that in the first word sounds more back
than that in the second.
Prepausal backing and emphatisation have the same influence on aa as on a. The only
audible difference between a and aa in the various contexts where they contrast is one in
duration. On aa, see the next section.
60 CHAPTER THREE

nonemphatics. These differences are enhanced by concomitant glottal vibra-

tions but they are easily perceived even in the absence of voicing. They give
rise to surface contrasts such as those in the minimal pairs below, where
the broad phonetic transcription used throughout this book is followed by
a narrower transcription.
(26) a. fttstt!sstt [ftstS:t;hfo Iftts=t t-!ssd-t/
examineedoßms 2-slide:aor-2s
'examine it and slide!'
b. fttsttsstt [ftSts:t h] Iftts=t t-ss-t=t/
examineedoßrns 2-eat:aor-2s=d03ms
'examine it and eat it!'
(27) a. !tkktt [thk:ht h] /t-!kkd-t/
'you poked (eye)' 2-poke-2s
b. tkktt [thk:ht h] /t-kk-tet/
'(and then) you went through it' 2-pass:aor-2s=d03ms
The hushing sound of the sibilant in (26)a clearly has a lower pitch than
that in (26)b. The explosion bursts in (27)a have a lower pitch than those
in (27)b. Exploded plain t is furthermore slightly affricated, which is never
the case with [!t] . The lowered pitch of the friction noise due to emphasis
is more easy to perceive in some voiceless fricatives than in others. It is
quite salient in [!s] and [!li]31 while it is not easy to make out in [!f] and
[!S].32 The auditory differences due to emphasis in fricatives are analo-
gous with - but not identical to - those due to lip-rounding, v. for instance
[s] in Eng. shoot, which has a lower pitch than in sheet.
Here are now examples involving voiced segments.
(28) a. ss !tsttlt [s :tst :lt h] Iss t-!sttl-t/
eat .aor 2-shave:head:aor-2s
'eat and shave your head!'
b. sststtlt [s:tstlth] Iss=t s=l-tlt/
eat.aoredoßms withe/lz-third
'swindle hirn by proposing hirn one third!'33

30 We are unable to determine whether the long t which immediately precede s fIs :] is
emphatic or not.
31 The existence of a marked audible difference between the plain variant of the pharyn-
geal fricative fh/ and its emphatic variant is evidence that the secondary articulation of
Irndlawn Tashlhiyt which we are calling ernphasis is not rnere pharyngealization.
32 On the other hand the difference between [z] and [!z] is easy to perceive , even in the
absence of an ajacent vowel.
33 Literally 'eat hirn with a third' .
 PHONOLOGICAL BACKDROP 61

(29) a. lngddrakrk" [n;:lgd:rak:q1 In- !gddr=akwkwI

'we are all fat' lp-fat=all
b. ngddmakrk" [nrgd .mak.u] /n-gddrneak 'tk" I
'we all rushed ' 1p-rusheall
Apart from a difference in the releases of the final stops (v. above) a
salient difference between the two expressions in (28) lies in the coloration
of the liquid, which sounds more or less like the 'dark l' of English bottled
in (28)a but not in (28)b. In (29) one hears a short voiced vocoid between
the release of the coronal closure of Inl and the onset of the velar closure
of Ig/. The vocoid in question sounds like a short lax i in (29)b, whereas
it is lower and more central in (29)a.

3.6.2. The distribution 0/ emphasis

At present our knowledge of the regularities which govern the distribu-
tion of emphasis in words and sequences of words in Imdlawn Tashlhiyt
is rather patchy, but it is sufficient to allow us to map out the distribution
of emphasis in lexical representations, as we shall now see.

3.6.2.1. In the lexicon

The basic facts are contained in the following two statements about the
lexicon of Imdlawn Tashlhiyt: (i) there are two types of roots , emphatic
and nonemphatic, and (ii) all emphatic roots contain a coronal consonant.
These facts can be explained by assuming that in the lexicon of Imdlawn
Tashlhiyt, emphasis is a property of certain coronal segments, and that
the emphasis of an underlying consonant propagates to the neighbouring
segments. Let us dweIl brieflyon the two generalizations we have just stated.
Starting with words pronounced in isolation , one can divide them into
two classes, the emphatic words, which contain emphatic segments, and the
nonemphatic words, which do not. !t-i-bggar-in (tibggarin), 'women who deal
in cattle', from /t-i-bggar-in/, is an emphatic word, while i-gzzar-n 'butchers'
is a nonemphatic word. In sentences the emphasis spans overlap with
emphatic words but often their edges do not coincide with word boundaries.
Whether a word is emphatic or not only depends on the phonological
properties of its kernel: the words built on a given kernel are either all
emphatic or all nonemphatic. Similarly all the kernels which are morpho-
logically related agree with respect to emphasis, a generalization which
has very few exceptions in Imdlawn Tashlhiyt. 34 For instance the following

34 A few nouns have singular and plural kemels which do not agree with respect to emphasis,
e.g. a-ydi 'dog' vs. li-yda-n ' dogs'. There are also a few nouns in which emphasis and the
lack of it are in free variation, e.g. a-frux or !a-f rux 'boy' , t-i-frx-in or !t-i-frx-in 'girls' .
62 CHAPTER THREE

words are emphatic, and so are all the other words built on the same kernels:
Iwrrv-n 'they are yellow', Ia-wrav 'yellow (ms adjective)', It-i-wrvi 'yellow
color', l i-wriv 'vomit of gall', Is-awrav 'jaundice' . On the other hand the
following words are nonemphatic, and so are all the other words built on
the same kerneis: gzzul-n 'they are short', a-gWzzal 'short (ms adjective)' ,
a-gzzaylu 'short (ms adjective)', i-gzzuyla, a plural form of the same.
Let us say that the words in each set share a common root, e.g. all the
words in the first set are built on the root {lw-r-v, 'yellow'}, and more
generally let us use the term 'root' as an informal label for the set of phono-
logical and semantic properties which are shared by all the words in a family
of morphologically related words." As a consequence of the generalizations
at the beginning of the preceding paragraph, all the roots in Imdlawn
Tashlhiyt fall into two classes, emphatic and nonemphatic.
All the emphatic roots without exception contain a coronal consonant.
One can account for this fact by assuming, as is usually done by authors
writing on Berber, that Imdlawn Tashlhiyt has a set of emphatic phonemes,
all coronal. The reason there are no emphatic lexemes containing only
vowels and/or noncoronal consonants is that Imdlawn Tashlhiyt does not
have emphatic phonemes which are vocoids or noncoronal consonants.
Does one need to include in the phoneme inventory of Imdlawn Tashlhiyt
an emphatic counterpart for each of the plain coronal phonemes? Our answer
is Yes: for any one of the coronal consonants, one can find emphatic kernels
in which the consonant in question is the only coronal, e.g. li-wtfa-n 'ants'
for I!t!, lidgam 'yesterday' for I!d/, !uskay 'greyhound' for I!s/, !a-ggaz
'afternoon snack' for I!z/, !mu ss 'cat' for I!s/, !bbii ' knead !' for I!z/,
!t-mnnk 'she stayed away (from school)' for I!n!, It-rmi 'she is tired' for
I!r/, !yallah 'let us go' for I!l/. Roots in which the only coronal is I!s/,
I!n! or I!ll amount to just a handful.
In emphatic kernels which contain several coronals we do not know of
any reason for preferring one coronal as the underlying source of emphasis,
to the exclusion of the others. There is for instance no way to determine
whether in li-smmid 'it is cool' the underlying representation of the kernel
is I.s.mmid/, /smmid/ or l.s.mmiQ/. If at the phonetic level dorsopharyngeal-
ization were a matter of degree rather than a categorial property, one could
expect segments closer to the underlying emphatic consonant to have a
greater degree of emphasis than those further away. This does not seem
to be the case. In an emphasis span the degree of emphasis does not
vary in an audible manner. Consider It-i-bukad-in 'blind women ' and
lt-i-zukat-in 'juniper trees'. Under our assumption that in Imdlawn Tashlhiyt
the emphatic phonemes are all coronals , the kernels of these words must

35 We use the term 'root' merely as an expository convenience.

 PHONOLOGICAL BACKDROP 63

be /bukad/ and huka! at the phonological level." We fail to hear any dif-
ference between the initial ti sequences in these words, although one is
adjacent to the underlying emphatic consonant while the other is located
two syllables away.
Emphasis is an important stylistic cue; in more formal speech styles
emphasis spreads over shorter spans. /t-a-!ddinga! 'wave', which is norrnally
pronounced taddinga, can also be pronounced taddinga, a pronunciation
appropriate only in a speech style used in certain forms of public address
by ambulant preachers and by minstrels (!rrways). To take another example,
/i-kigJ 'eye ailment', which is normally pronounced ikid, can be pronounced
ikiq in that same elevated style." As suggested by taddinga above, data
about this speech style make it possible, at least in some cases, to pinpoint
the underlying emphatic coronal(s) in emphatic kernels with several
coronals . This will have to await further research. In the remainder of this
discussion we limit ourselves to manners of speaking which do not have the
distinctly highbrow flavor conveyed by pronunciations like taddinga and
ikiq.
Since affixes, clitics and prepositions are emphatic only when they belong
to words or phrases which contain a lexical morpheme with an emphatic
kernel, one must assume that they are all nonemphatic in their underlying
representations.

3.6.2.2. At the phonetic level

Let us now turn to the distribution of emphasis at the phonetic level.

Borrowing the expression 'emphasis span' from Heath (1987: 311) let us
say that the emphatic span of an emphatic phoneme or morpheme in a
terminal representation is the maximal emphatic string which contains the
surface reflex of that phoneme or morpheme. To illustrate this definition,
consider the following sentence:

36 In !tizukatin the stern-final t is epenthetic , v. § 2.5, note 20. The corresponding ms noun
is la-zuka.
37 In ikid, the degree of emphasis seems to increase gradually during the time course of
the second vowel, which sounds Iike emphatic [i] near the end. The vowel gives the impres-
sion of beginning as a plain i, an impression which may in part be due to the fact that the
preceding consonant is palatalized ([k Y]) . While fkI and Igl regularly palatalize before lil
and lat in nonemphatic context s, their palatalized variants never occur inside an emphasis
span. For the speakers' categorial perception, the semi-emphatic i in ikid counts as a plain
vowel, Iike the plain i in the last syllable in i-zri=k=id (3ms-overtake=do2ms=dir) 'he overtook
you (coming hither)' .
64 CHAPTER THREE

(30) /ur y-umz u-zgvg rar a-bukad/ uryumzuzg Wg Warabukad 38

neg 3ms-cling b-jujube u-blind
'the jujube did not cling to the blind man '
In the pronunciation represented in (30) the emphatic spans of / !z/ and
/!d/ are the strings uryumzu and bukad. Mapping out the facts about
emphasis spread in Imdlawn Tashlhiyt is not an easy task, owing to the exis-
tence of extensive free variation as well as to frequent uncertainties in
acceptability judgements. Among the expressions which contain emphatic
segments, short words pronounced in isolation must be emphatic from one
end to the other, but longer expressions usually have more than one accept-
able pronunciation, as far as emphasis is concerned. The following
generalizations seem to be exceptionless:
(31) a. The segments in a CV sequence must be both plain or both
emphatic; this requirement must be met regardless of the
morpho-syntactic relationship between C and V.
b. All the coronals in an emphatic kernel must have the same
emphatic span.
As an illustration of these generalizations, consider the sentence /i-y i-!smmid
asi=t/ (if 3ms-cool take .aoredoßms) 'if it has cooled down, take it' , which
must be pronounced iyismmidasit.39 (3l)b requires that the surface reflexes
of sand d in /!smmid/ belong to the same uninterrupted string of emphatic
segments and (31)a requires /a! to be emphatic like the preceding d.
The above generalizations are valid for all styles of speech. The other
factual observations one can make about emphasis spans must include qual-
ifications about factors such as speed of delivery and the degree of formality.
In (30) above /a-bukad/ can be pronounced abukad, but in isolation this
word can only be pronounced abukad. In the isolation form of an emphatic
word, emphasis always starts at the beginning of the word, but in some
words it can stop before the end, e.g. one acceptable pronunciation of
/t-i-guka-in/ 'j uniper trees' is tizukatin'" (the other is tizukatin).
Concerning the maximal size of emphasis spans, the following gener-
alization is only meant to convey the general flavor of the facts .
(32) Let E be an emphatic unit (phoneme or morpheme) and let P
be the smallest X'?" containing E. The emphatic span of E does
not include segments not contained in P.

38 The sentence has another acceptable pronunciation , in which the second emphatic span
is rabukad .
39 Pronouncing iyismmidasit is not altogether inacceptable, but feels far less natural.
40 t before the suffix is epenthetic (v. note 36), which explains why it can be nonemphatic ,
in apparent contradiction with (31)b.
 PHONOLOGICAL BACKDROP 65

We illustrate this generalization with the following sentence:

(33) i-lbdaeyax" ird-n n=t-!addzar-in=ad
3ms-share=dat Ip wheat-p genef-neighbor-fpedem
n-awietn
Ip-take.aoredoßmp
'he shared among us the wheat of these neighbors (f) and we
took it away'
Sentence (33) has four acceptable pronunciations; it contains two emphatic
morphemes which each give rise to two emphatic spans in free variation.
These are listed in (34):
(34) ibda I ibdayax taddzari I taddzarina
Assuming that c1itics belong to the same x max as their host, we see that in
(33) the span on the left remains within the bounds of the sentence's initial
vmax while the span on the right is confined within the limits of the prepo-
sitional phrase n=!taddiarin=ad, which is the smallest X'?" containing
/laddzar/, given the assumption we have just made about the c1itics.
In the present state of our knowledge about emphasis spans, general -
ization (32) is obviously little more than a tool for finding out more about
the facts . In the absence of a discussion of phrase structure in Imdlawn
Tashlhiyt, it is in many cases unc1ear what is a genuine counter-example
to (32) .
All the uncontroversial counter-examples to (32) which we have encoun-
tered occur in expressions in which an onset-nucleus sequence straddles
an X?" boundary. In h#U1 in (30) and in Id.#a/ in the example in the text
immediately below (31) the nuc1eus in the onset-nucleus sequence is a
vowel, and the violations of (32) may be due to the overriding necessity
of meeting (31) . But there are also analogous counter-examples in which
the nuc1eus involved is a consonant, as is the case of Inl in ummaz
l1=dadda=s (fist gen=elder:brother=3s) 'his elder brother's fist' . The accept-
ability judgements for Onset-Nucleus sequences straddling an X'?" boundary
are often fuzzy or inconsistent over time.

3.7. THE VOICED PHARYNGEAL CONSONANT

In this seetion we show that all the surface occurrences of tautomorphemic

aa derive from l'i.l.
1'11 and tautomorphemic aa only occur in Arabic loans or in words derived

41 From more abstract I!bda=ax/, with a hiatus-breaking yod.

66 CHAPTER THRE E

from such loans." Compared to that of the other consonants, the distribu-
tion of f in the surface forms of Imdlawn Tashlhiyt is very limited. Let
us begin with nongeminated r.
All the occurrences of nongeminated f which are found in surface forms
occur next to a high vowel which belongs to the same word, e.g. Yum
'swim' , riss 'up yours! (interjection)' , larrbii 'sugar box, aug'," fisa
'Aysha', labyuf 'sale, p'. We have furthermore found a few cases where
nongeminated f alternates with aa. Finally, the morphemes which contain
an invariant aa throughout all their occurrences behave morphologically
as though aa were the surface reflex of a consonant, not a vowel (examples
will be given below). The following rule accounts for the distribution of
tautomorphemic aa and for the altemations involving f:
(35) 'l-TO-aa: 'l ~ aa I except when adjacent to a high vowel
Here are alternations involving f:
(36) a. II-!sn'l-tI Is-snaa-t 'occupation, vocation':"
a'. la-!snay'l-iyl la-snay '[-iy 'craftsman'
b. /l-sl'i-t/ s-slaa-t ' wares'
b'. /a-slay'i-iy/ a-slay f-iy ' wholesaler '
c. la-'lzr-iyl aazr-iy 'bachelor'
c' . It-i-'l'luzra! t-i- ffuzra 'bachelorhood'
d. Id'i'a-nI daa-n 'pray 3mp'
d'. Id'i'u-nI dfu-n 'pray aor 3mp'
e. la-byya'll a-biyaa 'informer' (p i-biyaa-n)
e'. /t-i-byya'i-in/ t-i-biya f-in 'informer fp'
The second item in each pair contains an occurrence of f which is adjacent
to a high vowel. If one posits the underlying forms given at the begin-
ning of each line, the morphology in the examples above is perfectly regular.
The nouns in (36)a', (36)b' and (36)c' are templatic nouns of the
ABNAKLlY type and of the TIRRUGZA type, in the terminology of DE
(1992), where these templatic nouns are discussed at length. The verb in
(36)d conjugates like any CCa verb, e.g. bda 'begin' , aor bdu (v. DE 1991:
80-82). Item (36)e is an occupational noun with the shape la-CC:aCI, a very
productive type . As exemplified by (36)d ,e sequences, l'la! and la'i'l surface
as aa, instead of aaa as rule (35) would lead one to expect. The sequence
aaa, tautosyllabic or heterosyllabic, is never found in Imdlawn Tashlhiyt.

42 A possible exception is a-raam 'camel' (p i-raaman); this word has no close phonetic
analogue in Arabic.
43 The feminine form (non augmentative) is lt-arrbih-t, with /'i/ devoiced to Ii before /t/.
44 !lasnay'i, the plural of Issnaat, is the only exception to rule (35) we have encountered .
 PHONOLOGICAL BACKDROP 67

Rule (35) also enables us to make sense of the morphological altema-

tions involving morphemes with an invariant aa. The realization of /~/ as
aa results in the appearance of canonical forms not attested otherwise in
Imdlawn Tashlhiyt, and the words with these canonical forms show various
apparent irregularities in their morphology. Positing underlying represen-
tations with f)/ enables one to ascribe the words in question to morpho-
logical types which are otherwise weIl-attested.
For instance nouns with singular forms ending in a short a never have
identical kernels in both numbers , e.g. a-saka 'ford', p i-sakat-n, a-ynia
'ladle' , p i-vn iaw-n; on the other hand, in nouns ending in aa which have
Berber plural forrns," these forms always end in aa-n, as is expected if these
nouns end in a consonant (I~/) at the level of representation which is relevant
for plural formation, e.g. aidaa /aidf/ 'colt' , p aidaa-n.
Or consider the verb /1%/ laab (impf tt-laab /tt-l'iab/) 'play (in sports)'.
The corresponding ABNAKLlY noun is a-laayb-iy 'player', for which we
posit /a-l'iayb-iy/, which has the form /a-CCayC-iy/, the normal shape of
ABNAKLlY nouns derived from a kernel which contains three consonants
(cf. (36)a' and (36)b' and v. DE (1992)). If laab were derived from /laab/
one would expect the corresponding ABNAKLlY noun to be "a-lwayb-iy,
i.e. it would have the shape /a-CwayC-iy/, like the other ABNAKLlY nouns
derived from kernels which contain two consonants.
Positing /~/ in the underlying form of every morpheme with a surface
aa makes the underlying distribution of f).! more similar to that of the
other consonants; it also explains why long aa, which cooccurs rather freely
with the various consonants, cannot occur in the same morpheme as Ih/.
Short a is not subject to a similar restriction, v., e.g., sblassm 'get away
stealthily' . In Tashlhiyt as well as in Arabic, consonants which differ only
with respect to voicing cannot cooccur within the same morpheme," and
the prohibition of tautomorphemic aa and Ih/ is simply an instance of that
restriction: morphemes which contain both /hI and /~/ are prohibited. Arabic,
the language which is the source of almost all the lexical items in Imdlawn
Tashlhiyt which contain aa or f, is subject to the same cooccurrence restric-
tion.
We have encountered only three lexical items where geminated f occurs ,
ffad 'yes (French si)', sgWaffi (impf sgWaffay) 'beleh' and t-i-ffuzra
'bachelorhood' ((36)c'). ffad and the imperfective stern sgWaffay suggest
that rule (35) only applies to nongeminate I'il,
Rule (35) is blocked by an adjacent high vocoid if it is a vowel but not

45 On Berber and Arabic plurals, v. § 2.5.

46 Violations of this restriction are extremely rare , e.g . t-i-mtd-in 'lo in (cut of meat)' .
Some exceptions are only apparent, e.g. tuä 'coagulate' , whose underlying representation
can be assumed to be I!dd+d/, on account of the fact that I!ddl regularly surfaces as ltt , On
'+', v. § 6.4.1.
68 CHAPTER THREE

if it is a glide, for aa can occur next to a glide, V. , e.g., la -maayud

(p !i-maayad) 'lark', laawin (p laawayn) 'supplies', waar (impf tt-waar)
'be difficult' . There is a tendency for Imdlawn Tashlhiyt speakers of the
younger generations to reinstate 5', e.g. whereas older speakers only say aali
'Ali' and i-sbaa 'he has enough' , younger speakers also have the free
variants ru and isb5'a.

3.8. /u/ FRONTING

The full vowel /u/ is fron ted and lowered when it occurs between two
consonants which both belong to the set [coronal OR pharyngeal OR glottal] .
In that context the realizations of /u/ range between [öl and [re]. We use
the symbol ' ö' to represent them. This rule is obligatory and exception-
less. In the examples below, /u/ is in the fronting context in the forms on
the right but not in those on the left.
(37) I 11
a. ttu ttö=t 'forget'
b. sshu ssh öet 'hypnotize'
c. i-flu t-i-flö-t ' door '
d. a-hanu t-a-han ö-t 'shop'
e. uyyl-n y-öyyl ' fl y away'
f. uhl-n y- öhl 'be stuck, cornered'
The forms in lines a-b are 2s imperatives, with a doßms clitic pronoun added
in column 11. In line s c-d the forms in column 11 are feminine nouns and
those in column I are the corresponding augmentatives. The forms in lines
e-f are perfective verbs, 3mp in column land 3ms in column 11.
/u/ does not undergo fronting when it belongs to an emphatic span, e.g .
whereas /u/ is fronted in /ssnu=tI 'rninimize its weight !' , it has its back
(dorsopharyngealized) realization in /!ssnu=tI ' strengthen it!' . The steady-
state [u:] which is the realization of /ww/ does not undergo fronting."
Whereas /sus/ 'Sous ' (a toponym) must be pronounced [sös], /swws/
'disturb!' must be pronounced [su:s], not [Sö:s].
Fronting occurs even when the triggering consonants do not belong to
the same ward, e.g . the occurrence of /u/ indicated by angled brackets has
the same fronted variant in all three expressions in (38):

(38) a. i-tt-cu>t ar alliv

3ms-pass-rot up:to bottom
'it is rotten to the core'

47 On geminate glides, see § 7.4.

 PHONOLOGICAL BACKDROP 69

b. i-tt<u> t-a-qssab-t
3ms-forget smock
'he forgot the smock'
c. t-a-funas-t <u>t-nt t-hsmiy-in
cow hit-3fp girls
'the cow that the girls struck'

u fronting also occurs in some varieties of Moroccan Arabic, v. Al-Ghadi

(1990: 15). For further discussion of this phenomenon, see Elmedlaoui
(1995a: 222 ff). To avoid unnecessary clutter, the effects of u fronting are
normally not represented in our phonetic transcriptions in this book, e.g.
our phonetic transcriptions for the forms in column 11 in (37)a and (37)b
would simply be ttu=t and sshu=t.
 CHAPTER FOUR

TASHLHIYT SYLLABLES I

One of the remarkable features of the phonology of Imdlawn Tashlhiyt is

its syllable structure, which allows any segment to be a syllable nucleus.
This feature is not without precedent in the phonological literature, but it
is sufficiently rare to require a detailed justification. Among the languages
which have been claimed to have syllabic obstruents, those which have
received most attention from a theoretical point of view are languages
spoken on the Northwest Pacific coast of North America , e.g. Bella Coola,
and certain Mon-Khmer languages spoken in Laos and the Malay peninsula,
see Bagemihl (1991) and Shaw (1993) and references therein.' In this
chapter and the next we determine the basic inventory of syllable types in
Tashlhiyt and we discuss the role of sonority in assigning syllable struc-
ture to sequences of segments. Syllabic consonants will be the focus of
Chapter 6.
The first section of the present chapter introduces two mutually inde-
pendent claims around which our analysis revolves, the Sonority-Driven
Syllabification thesis and the Licit Consonantal Nuclei thesis. The rest of
this chapter is devoted to an inventory of the syllable types and to the
Sonority-Driven Syllabification thesis, using text-to-tune alignment in
singing as evidence.

4.1. SYLLABIC CONSONANTS

We have argued in our earlier work.? and we shall again argue in this
book, that the only vocoids in the underlying representations of Imdlawn
Tashlhiyt are laI, li/, lul, Iyl and Iw/, and that syllables which do not
contain an occurrence of one of these segments have a consonant as their
nucleus. Any consonant of Imdlawn Tashlhiyt may act as a syllable nucleus.'
Long strings entirely devoid of vocoids are commonplace at the under-
lying level because morphemes which lack underlying vocoids are
numerous, especially among the grammatical morphemes. Most of the

I Closely related issues are raised by certain members of the Northwest Caucasian and Yuman
families, see Anderson (1978).
2 See Elmedlaoui (1985) and DE (1985, 1988, 1996a, b).
3 This claim has already been made about other dialects, see Applegate (1958: 13) on Ifni
Tashlhiyt : 'all consonants in.certain environments have syllabic allophones ' , and Mitchell
(1957: 198) about the dialect spoken in Zuara (Libya): 'If the statement is at all meaningful
at the phonetic level, any "consonant" may be syllabic in Berber' .

71
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
72 CHAPTER FOUR

affixes lack vowels in their underlying representations, and so do a number

of prepositions and clitic pronouns. Complex consonant clusters are a
common occurrence at the phonetic level in Imdlawn Tashlhiyt. Indeed, it
is not difficult to find utterances which do not contain a single vowel, or
even a single stretch of speech with glottal vibration. Examples have already
been given in § 3.6.1. Here are some more. (1) must be pronounced with
an uninterrupted midsagittal closure from one end to the other. No voiced
segment should be heard in (2) and (3).4
(1) [tntlt.nt''] tntlttnt /t-ntl-tetnt/
Zs-hide-Zsedoßfp
'you hid them (f)'
(2) [k.st .s .t:"] kksttsstt /kks=t t-ss-t=t/
rernove.aoredoßms Zs-eat.aor-Zsedoßms
'rernove it (m) and eat it (m)'
(3) [tft.sts.k.xr'k.sts .k:"] tfttstsskkxtkkstsskk
/t-ftts-t l-skk 'Y t-kks-t l-skk/
2-examine-2s /l/-currency in 2-remove-2s /l/-doubt 5
'you examined the currency which you checked out'
In consonant sequences such as those in the preceding examples the broad
phonetic transcriptions used in this book provide a faithful enough picture
of what is actually heard when the forms in question are uttered. However
there are also many sequences which cannot be pronounced without inter-
vening short voiced vocoids that are systernatically omitted in our
transcription. For instance our broad phonetic transcription for /t-!bttn/
'she put a lining' is !tbttn, but one actually hears [ltabt:n]. These short
vocoids were already mentioned in § 2.2, where they were dubbed VTVs
(voiced transitional vocoids). In the following paragraphs we summarize
our position concerning the VTVs. Further discussion will have to await
Chapter 6.
Given that the VTVs' location and vowel quality are entirely predictable
from the phonetic environment and the location of certain morphologieal
boundaries, one expects them to be epenthetic vowels, i.e. segments inserted
by the phonology to act as syllable nuclei. However it is our contention
that VTV s do not correspond to separate segments in the representations
which are inputs to phonetic implementation.
Consider for instance /t-zgzaw/ ' it (f) is green', which sounds like
[tzrgzaw]. This word is disyllabic and according to the analysis to be pre-

4 For each example we give both the broad phonetic trans cription used throughout this
book and an IPA transcription.
5 The prefix 11-1 is the reflex of the definite article of Arabic, v. § 2.5.3.1.
 TASHLHlYT SYLLABLES I 73

sented below, its terminal representation is tgg.zgw (periods indicate syllable

boundaries; underlyings indicate syllable nuclei). The first syllable has a
simple onset (r), not a eomplex one (rz): its nucleus is a eontoid (z), not a
voeoid (I) . Transitional voeoids do not fall within the purview of the phono-
logieal eomponent of Imdlawn Tashlhiyt. They ean indeed be heard, and
they may sound like eertain vowels in other languages or even in other
dialeets of Berber, but they are not units manipulated by the phonology.
When we give the narrow phonetie transeription of /tzgzaw/ as the seven
symbol sequenee 'tzrgzaw' we only mean to eonvey what is heard when the
word in question is uttered; we do not intend to make any claim about
the phonologieal eomponent of Imdlawn Tashlhiyt. As far as phonology
is eoneerned, the representation of /tzgzaw/ does not eomprise more than
six segments at any stage of the derivations.
Our diseussion of the surfaee syllabification of words and sequenees
of words in Imdlawn Tashlhiyt will revolve around the following two claims:
(4) Lieit Consonantal Nuclei :
The only surfaee vowels are a, i and u; any eonsonant of
Imdlawn Tashlhiyt ean be a syllable nueleus.
(5) Sonority-Driven Syllabifieation:
a) All the information relevant for predicting the syllable
strueture of an expression resides in the eonsonants and the
full vowels eontained in that expression.
b) In the eompetition for the status of syllable nucleus, more
sonorous segments are favored over their less sonorous
neighbours.
Let us give examples illustrating these two claims. Justifications will
be given later.
The Lieit Consonantal Nuclei thesis (4) is illustrated in the mid eolumn
in (6) below. Eaeh line in (6) eontains, in this order, the underlying repre-
sentation of an expression (a word or a sequenee of words), the terminal
representation of that expression under our analysis , and a narrow phonetie
transeription indieating what one aetually hears when the expression is
uttered in isolation. In the terminal representations, syllable edges not
adjaeent to pauses are indieated by periods and syllable nuclei are under-
lined ."

6 Here is the meaning and the morphemic analysis for each expres sion in (6): (a) 3ms-collide
'he collided' , (b) 3fs-stingy=even ' she is even stingy' , (c) f-u-gazelle-fs ' fernale gazelle ',
(d) 2-surround-2mp ' you (mp) surrounded' , (e) pick=do3ms 2s-eat :aor-2s=do3ms 'pick it
and eat it'.
74 CHAPTER FOUR

(6) underlying terminal narrow

representation representation phonetic tr.
a. /i-dlh/ i.dIh [idlch]
b. /t-bxleakk'? 1.bx.li!kk w [tobxlak.u]
c. It-a-lznk rd-t/ ti!.z!.l.kW1t [ltuznuku t h]
d. It-ssutl -m/ ts.sö.tlm [ts. ötlom]
e. /kks=t t-ss-t=t/ k.k~t.tis1t [k.st.s .t:"]
According to the terminal representations in the mid column, the nucleus
of the second syllable in (6)a is l, the nuclei in the first two syllables in
(6)b are t and x, those of the last two syllables in (6)c are n and t, and so
on.' These, then, are instances illustrating the Licit Consonantal Nuclei thesis
(4). In our analysis this thesis goes hand in hand with another claim:
Imdlawn Tashlhiyt does not have any vowel epenthesis. As already stated
above, the only nonconsonantal segments present in the terminal repre-
sentations are the glides y, wand the vowels a, i and U or their contextual
variants." Some of the vocoids recorded in the narrow phonetic tran scrip-
tions in (6) do not have any reflex in the terminal representations, because
we claim that these sounds are not segments in their own right, but mere
tran sitions between segments. Some of these transitional vocoids are voice-
less , e.g. the short voiceless u which is heard in (6)c when the articulation
moves from [k"] to [tl; others are voiced, e.g. [c] in (6)a. A consonant which
is immediately followed by a transitional vocoid may occupy any position
in syllable structure, i.e . it can be an onset, a coda or a nucleus. Instances
of the latter situation are found for example in (6)a, where [c] occurs
between a nucleus (I) and the following coda (11), and in (6)b, where [~]
occurs between a nucleus (t) and the onset of the following syllable (b).
If these vocoids were manifestations of surface vowels, that is, of segments
syllabified as nuclei in the terminal representations, the consonants occur-
ring immediately before them would be expected to be onsets, contrary to
the evidence which will be presented below about surface syllabification
in Tashlhiyt. The transitional vocoids will be discussed in Chapter 6.
We now turn to examples illustrating the thesis of Sonority-Driven
Syllabification (5). We repeat that thesis under (7):
(7) Sonority-Driven Syllabification:
(a) All the information relevant for predicting the syllable
structure of an expression resides in the consonants and the
full vowels contained in that expression.

7 At the end of (6)c the underl ying sequence td-tJ surfaces as the geminate u, in which
the first skeletal position is a nucleus.
8 E.g. [öl in (6)d, which is a fronted variant of tut. On u fronting, v. § 3.8.
 TASHLHIYT SYLLABLES I 75

(b) In the competition for the status of syllable nucleus, more

sonorous segments are favored over their less sonorous
neighbours.
Let us consider each branch of this thesis in turn. The treatment of Arabic
loans in Imdlawn Tashlhiyt provides a simple demonstration of the import
of branch (a) . Moroccan Arabic (MA) has three full vowels a, i, u, like
Imdlawn Tashlhiyt, but unlike Imdlawn Tashlhiyt it has in addition a vowel
@, the location of which can be the sole distinguishing feature between
words. For instance MA has a contrast between words of the shapes CC@C
and C@CC. This contrast is illustrated below in columns land 11 of (8).
The forms in column I are 3ms perfective verbs ('he jailed', etc .) and
those in column 11 are deverbal nouns corresponding to these verbs ('impris-
onment', etc.). Columns I' and 11' will be explained below."

(8) Moroccan Arabic Imdlawn Tashlhiyt

I 11 I' 11'
verb noun verb noun
a. hb@s h@bs hbs l-hbs'"
b. !sx@t !s@xt !sxd !s-sxd
c. sb@t s@bt sbt s-sbt
d. !rx@s !r@xs !rxs lr-rxs
e. rz@q r@zq rzq r-rzq
f. zh@d z@hd zhd z-zhd''
Contrasts such as that between columns land 11 show that branch (a) of
the thesis of Sonority-Driven Syllabification does not hold in MA. !ib@s
and !i@bs do not contain any full vowel and they share the consonant
sequence Ihbs/. If there exist representations from which the different
syllable structures of hb@s and h@bs can be derived, these representa-
tions must contain more information than the mere sequence /hbs/. We
will see in Chapter 8 that the contrast between land 11 is due to tem-
plates, i.e. to requirements on syllable structure which are associated with
certain morphological categories or specified in the lexicon. Templates
fall outside the bounds of (7)a.
The forms in columns I' and 11' in (8) are bona fide words of Imdlawn

9 Here are the meanings of the verbs in column I: (a) 'jaiI', (b) 'curse', (c) 'keep the Sabbath' ,
(d) 'become cheap', (e) 'grant' (the subject is usually God), (f) 'be strong' . The Imdlawn
Tashlhiyt words on the right-hand side have meanings identical with those of their MA
counterparts.
10 In lines a, c and e, the medial consonant of the Berber words optionally devoices, see
§ 6.4.2 on regressive devoicing in Imdlawn Tashlhiyt.
11 In free variation with lihd.
76 CHAPTER FOUR

Tashlhiyt." Our examples in (8) come from cases in which both a verb
and its deverbal noun have been borrowed from MA. As is the rule for recent
loanwords, the nouns in (8)-11' all begin with the prefix 11-1, which assim-
ilates to a following coronal," but otherwise they are homophonous with
the corresponding verbs. This homophony is an instance of branch (a) of
the Sonority-Driven Syllabification thesis. The Imdlawn Tashlhiyt words
bbs 'jaiI!' and l-bbs 'imprisonment' have as their kerneI the same sequence
/hbs/. Since all the information necessary to predict the surface forms is
contained in that sequence, the sequence has the same phonetic represen-
tation in both words. In Imdlawn Tashlhiyt, Arabic loans are completely
oblivious of the vowel @ of MA. This obliviousness is also typical of the
pronunciation of MA by the Ashlhiy speakers who speak it 'with an
accent' .14
Some dialects of Berber spoken in Morocco are like MA in that branch
(a) of the Sonority-Driven Syllabification thesis is not valid for them. Rifian
Berber is one of them, and it will be seen in Chapter 6 that in Rifian, Arabic
loans take into account the schwas of MA.
We now turn to branch (b) of the Sonority -Driven Syllabification thesis
(5). We are assuming the following sonority scale, in order of decreasing
sonority :
(9) Sonority Scale: a, high vocoids, liquids, nasals, fricatives, stops
Each line in (10) below displays two words which contain similar con-
sonants but are syllabified differentlyon account of their different sonority
contours."
(10) one syllable two syllables
a. Ikrm! krm a'.lgwmrl g:.mr
b. Ismdl smd b'. Izdml z.dm
c. Ikrzl krz c'. Irksl r.ks
d. Ixngl xng d' . I!ngdl n.gg
The forms on the left-hand side are each comprised of one heavy syllable,
i.e, a syllable with a coda; those on the right-hand side are sequences of two
light syllabIes, the first lacking an onset. In (10) all the segments which
are local maxima of sonority are syllabified as nuclei, which is in accord

12 The verbs are in the 2s pf imperative.

13 V. § 2.5.3.1.
14 The mishandling of schwas occurs prominently in the skits of Abdeljabbar Luzir and
Ahmad Belqas , a comic duo well-known during the Sixties and the Seventies who had made
a specialty of imitating the broken MA of Ashlhiy shopkeepers.
15 The words in (10) are 2s imperative verbs , i.e. bare aorist sterns. Here are the meanings
of the verbs in (10): (a) 'be dried out' , (a') 'hunt' , (b) ' add' , (b') 'gather firewood ', (c) 'plough' ,
(c') 'hide' , (d) 'choke', (d') 'drown' .
 TASHLHIYT SYLLABLES I 77

with the thesis of Sonority-Driven Syllabifieation (5). Segments which are

loeal maxima of sonority are indicated by bold type in the underlying rep-
resentations in (10) . A segment is a loeal sonority maximum if it is higher
on the sonority seale than the abutting segments. r is for instanee a loeal
maximum of sonority in krm «(10)a) beeause the adjaeent segments k and
mare lower on the sonority seale; r is also a loeal maximum of sonority
in gWmr «(10)a') beeause it oeeurs in final position and the preeeding segment
(m) is less sonorous. Whereas the Sonority-Driven Syllabifieation thesis
(5) claims that segments whieh are loeal sonority maxima are nuclei, it does
not make the eonverse claim; not all syllable nuclei are loeal sonority
maxima, e.g. gW is a nucleu s in gWmr «(10)a') although it is not a loeal
maximum of sonority, and the same is true of d in !ngd «lO)e').
The Lieit Consonantal Nuclei thesis (4) and the Sonority-Driven
Syllabifieation thesis (5) are closely eonneeted in our analysis, and yet in
essenee they are logieally independent of one another. It is possible to
deny that eonsonants ean ever be syllable nuclei in Tashlhiyt while retaining
the gist of Sonority-Driven Syllabification. One eould for instanee claim
that in the vowelless syllables in our syllabic parses in (6) and (10) the
eonsonants whieh we take to be syllable nuclei are in faet preeeded by an
epenthetie vowel whieh is obseured in various ways by phonetic imple-
mentation. In (11) below we give side by side our terminal representations
in (6) and the eorresponding ones in an alternative analysis developed along
these lines Ce ' stands for the epenthetic vowel):
(11) terminal representations alternative terminal
in (6) representations
a. i.dlh j.delh
b. !.bx.lflkk w ~t.b~x.lflkkw
e. tfl.zn.k Wlt tfl.z~n.kW~tt
d. t~.sQ.tIm t~s.sQ.t~lm
e. k.k~t.tlslt ~k.k~st.t~s.s~tt

The Ioeation of syllable edge s is the same under both analyses. The alter-
native analysis esehews syllabic eonsonants but it retains the basic idea
behind Sonority-Driven Syllabifieation: more sonorous eonsonants are
favored over less sonorous ones in determining the Ioeation of epenthesis
sites." More preeisely, every eonsonant which is a Ioeal maximum of
sonority in the underlying form has an epenthetic vowel inserted before
it. As noted earlier the eonverse is not true: eonsiderations of sonority
alone are not suffieient to determine the parses in (11); other faetors eome
into play, sueh as hiatus avoidanee, a prohibition against eomplex onsets,
ete.

16 Armenian is a language in which vowel epenthesis is motivated by sonority-driven syl-

labification, see Vaux (1998).
78 CHAPTER FOUR

This chapter deals with those aspects of the syllable structure of Imdlawn
Tashlhiyt which are common to both analyses in (11), i.e. with all those
aspects which are not tightly connected with the thesis of Licit Consonantal
Nuclei, Although our main evidence in favor of the Licit Consonantal Nuclei
thesis will only be presented in Chapter 6, the analysis which will be
developed in the remainder of the present chapter already takes that thesis
for granted . This is merely an expository convenience; the basic insights
and regularities presented in the discussion below can be reformulated so
as to be made compatible with alternative analyses such as that illustrated
in the right-hand side of (11).
Our richest source of data about surface syllabification in Imdlawn
Tashlhiyt consists in the native speakers ' judgements about syllable count.
These judgements are of two types . Some are answers to questions like
'How many syllables do you think there are in word (or word sequence)
such and such, and where do you think the peak of each syllable is located ?'
Others are only implicit and can be inferred from judgements about what
constitutes a well-formed line of verse. Our earlier work on syllable struc-
ture in Imdlawn Tashlhiyt drew most of its evidence from judgements of
the first type, v. Elmedlaoui (1985) and DE (1985). In this book, on the
other hand, we will forsake the direct questioning of speakers about syllable
count and rely instead on well-formedness judgements about lines of verse.
We feel that the data derived from judgements about well-formedness in
versification lies on firmer methodological ground than data from answers
to questions about syllable count. Tashlhiyt poetry is normally sung. Like
French or English speakers, Tashlhiyt speakers have clearcut intuitions about
which words can be fitted onto a given tune . One could say that singing
provides the native speakers with a yardstick for assessing the well-formed-
ness of verse. Like language, this yardstick is acquired in early childhood
without any explicit instruction. Making acceptability judgements about text-
to-tune alignment is a rather well-defined task for the speakers who are
asked to perform it. It is debatable whether the same can be said of counting
syllables and identifying syllable nuclei in nonpoetic language ."
Concentrating on poetic syllabification has another advantage: the
syllabic parses which are relevant for the syllable-based regularities found
in the morphology of Tashlhiyt are those required by poetry, as we shall
see in the next chapter."
Usually, in the phonologicalliterature, syllabification in word sequences
is only discussed after the syllable structure of isolated words has been con-
sidered. The nature of our evidence about surface syllabification in Tashlhiyt

17 Louali and Puech (1999) and Puech and Louali (1999) report on an experiment in which
Tashlhiyt speakers were presented with pairs of Tashlhiyt words and asked to judge whether
the words were rhythmically alike or different. The interpretation of their results is unclear.
18 This fact was already pointed out in DE (1988).
 TASHLHlYT SYLLABLES I 79

leads us to proceed differently. The stretches of linguistic material involved

injudgements about text-to-tune alignment are not isolated words, but word
sequences cotenninous with lines of verse . In view of this we will discuss
word sequences directly .

4.2. TASHLHIYT VERSE AND SINGING

Let us begin with a caveat. The relevance of poetic meter for our under-
standing of the phonology of Tashlhiyt in its colloquial uses is apt to be
obscured by associations which terms such as 'poetry' and 'verse' carry
in literate cultures with a written corpus of poetry which has accumulated
over several centuries. In France, for instance, these terms bring to mind
activities which involve conscious effort, rote learning and specialized skills
acquired through formal teaching. However there are occasions when native
speakers of French engage in activities which are closer, for the purposes
of our discussion, to Tashlhiyt poetry: as children, many have had the
experience of making up new words and singing them to the tune of a
well-known jingle or nursery rhyme.
In Imdlawn, poetry is inseparable from music. Except in one situation, 19
lines of verse are never heard unless sung to a musical tune. One cannot
lay too much emphasis on the importance of the connection between poetry
and singing in Ashlhiy culture . To be sure, the metrical structure of a line
of verse is the same regardless of whether that line is sung or not. But
singing brings the meter of verse in sharp focus , for the metrical structure
of a sequence of words is a key element in the mental computations which
enable the singer to keep the words in step with the tune.
In France virtually all singing involves pieces in which the words have
been memorized. The people in Imdlawn have a repertory of such pieces,
e.g. lullabies (a-snuhnnu) and songs sung while preparing the bride
(a-sallaw), but in many occasions they engage in a singing of a different
kind, in which a familiar tune is combined with newly coined words . This
happens for instance in oratorical contests such as the one transcribed in
Appendix III at the end of this book. It also happens in a genre called
t-i'uza, in which a group of people sing together to a fixed tune verse impro-
vised on the spot by one of them ." Only certain people in Imdlawn are
able to improvise verse worth listening to, but anyone there has the ability
to sing to a tune words which they have never heard sung to that tune before.
There is no explicit teaching of the conventions which regulate the struc-

19 During a conversation a weIl-known line may be quoted on account of its content, as

one quotes proverbs .
20 V. also the description of a-ltwas in Basset (I95211987: 87). On ahwas, a rural party with
dances, see Schuyler (I979: 49-52).
80 CHAPTER FOUR

ture of verse or its relation to music. The children of Imdlawn simply acquire
them through repeated listening, as French children do.
The conditions which ME has experienced in Imdlawn presumably hold
throughout the Ashlhiy area, or at least they did until very recently. For
discussions of poetry and music in Ashlhiy society, v. Galand-Pernet (1972)
and Lortat-Jacob (1980). The relative linguistic homogeneity of the Ashlhiy
area has allowed it to evolve a common poetic tradition. This tradition is
embodied in the lrrays." a travelling singer and musician who makes a
living by giving performances. Some !rrways are also poets who perform
their own works . Nowadays cassette recordings and the radio contribute
powerfully to the diffusion of these works, not only in Morocco, but also
among the expatriates abroad.
During the last eenturies tashlhiyt poems have occasionally been written
down in Arabic script." Since the Thirties many recordings have been made
of performances by !rrways. Some of these have been put down in writing
and published in anthologies aimed at Ashlhiy people who are literate in
Arabic , e.g. Amarir (1975).
In recent years some intellectuals have started a new trend of deliv-
ering their compositions without singing. The people in the audience are
still able to spot ill-formed lines of verse, although there is no text -to-
tune correspondence to help them in their judgements.
In what follows we will diseuss certain basic properties of the struc-
ture of verse in Tashlhiyt poetry. Ultimately, the evidence we will make
use of in our discussion is native speakers' judgements about the well-
formedness of lines of verse. One way such judgements are manifested is
through text-to-tune alignment in singing. Aligning a text with a tune is
something that people do 'autornatically', i.e. without deliberation. As
already stated, all the speakers in Imdlawn can do it, and they do not have
to be taught how to do it; they just pick it up during their early years .
This suggests that while performing the mental operations which are needed
to compute the metrical structure of lines of verse the Ashlhiys rely to a
great extent on abilities which they possess anyway as speakers of their
language, independently of their musical experience, e.g. they presumably
break down the poetic text into phonologie al units which are more or less
those used in processing nonpoetic language.
Let us first diseuss briefly the relationship between words and music
in singing.

2\ From underlying /l-!rays/. The plural is lrrways and the feminine singular !tarrayst. On
the lrrways, their music and the social and economic background of their activity , see
Schuyler 's (1979) outstanding work.
22 See Stroomer (1992) and Boogert (1997).
 TASHLHIYT SYLLABLES 1 81

4.3. SINGlNG WORDS TO A TUNE

Among the facts about how Ashlhiys put words to music, those which are
relevant to our discussion of syllabification are quite straightforward. Let
us illustrate the kind of facts we have in mind with an example from French.
The facts of French are analogous to those of Berber but the data may be
more familiar to some readers ."
Let us consider three lines in the first stanza of Au clair de la lune, a
nursery rhyme widely known in the French-speaking world."
(12) a. Au c1air de la lune, mon ami Pierrot.
b. Prete-moi ta plume pour ecrire un mot.
c. Ouvre-moi ta porte pour l'amour de Dieu.
Each of the lines in (12) is sung to the same eleven note tune ; in that
tune the duration of the fifth, sixth and eleventh notes is twice that of that
of any of the others . Limiting ourselves to durations, the tune can be rep-
resented as in (13), where '* ' and '0' respectively represent an eighth note
and a quarter note.
(13) 2 3 4 5 6 7 8 9 10 11
o o o
* * * * * * * *
To be able to sing the song, it is not enough to know separately the text
and the tune . One must furthermore know how to align them together.
The mental computations involved in text-to-tune alignment presumably
require that certain features in the text be taken as landmarks and be matched
with landmarks in the tune . In French singing it is syllable nuc1ei which
provide the relevant landmarks for the mapping of texts onto tunes. One can
infer certain aspects of the syllable structure of a sequence of words from
the way it can be sung to a tune. Let us imagine that the phonology of French
had yet been little studied and that we were trying to discover more about
it by doing fieldwork on singing. Let us pretend in particular that we knew
next to nothing about syllabification in French. What kind of evidence could
we gather from singing? We give below a broad phonetic transcription which
represents the words in (12) as they are pronounced when sung in that
particular song. The blanks at word boundaries are for the readers' con-
venience. Square brackets indicate vowe1s which do not occur when the text
is delivered by a speaker of Standard French with a pronunciation appro-
priate in everyday conversation.

23 See the appendices at the end of this book for musical scores of Ashlhiy tunes and
words sung to these tunes.
24 The music and words of the song can be found in Davenson (1955: 581).
82 CHAPTER FOUR

(14) a. 0 kler dce la lünjce] rnön ami pyero

b. pretlce] mwa ta pl ümlee] pur ekrir t mo
c. uvroe mwa ta portce pur lamur dce dyö
If we were to ask our informants to sing the song several times we would
notice that the manner in which the speech sounds associate with the notes
in the tune remains more or less the same in every repetition. In partic-
ular, certain vocoids always align with certain notes . The alignments one
would observe are represented below. Each number in the top row stands
for a note in (13) .25
(15) 1 2 3 4 5 6 7 8 9 10 11
I I I I I I I I I I I
a. o kl erd rel a 1 ün ce m n am i py er 0 ö

b. pr C t ce mw at a pI üm ce pur ekr i r tm 0
c. u vr ce mw at ap ~rt ce p ur! am u rd ce dy Ö
Although these alignments remain constant from one performance to the
next, they need not be memorized by the singers. We would come to
realize this when we discovered that all our informants have the ability to
use a tune they already know as a carrier for words which they have never
heard before, e.g. they can use tune (13) to sing j'ai perdu ma montre
dans un autobus 'I lost my watch in a bus':
(16) j'ai perdu ma montre dans un autobus
1 2 3 4 5 6 7 8 9 10 11
I I I I I I I I I I I
ze p C rd ü m a m ö tr ce d ä z t n o t o b ü s
Most speakers can do this without rehearsal and with no apparent effort.
The publishers of song anthologies assume that the readers have such an
ability, witness the fact that in each song the text-to-tune alignment is
indicated only for the first stanza.
The speakers can also determine whether a sequence of words and a
tune fit together. Take j'ai perdu ma montre dans un autobus and replace
perdu (two syllabIes: /per.dü/) with oublie ' forgotten' (three syllabIes:
/u.bli.ye/), and the resulting text does not fit anymore with tune (13). The
speakers know this by trying to sing the text to the tune and seeing whether
they can reach the end of the text without 'getting stuck'. One 'gets stuck'
when one stops singing because one loses all hope of achieving a legiti-
mate association between the words and the tune. The reason the sequence

25 Actually this is a simplification. It would be more accurate to say that each number
represents a point in time, i.e. the onset of a note (v. Cornulier 1995: 116-120 ,280) or a
musical beat (v. Hayes and Kaun 1996). Text-to-tune alignments are effected primarily by
pairing up the metrical structure of the text with the rhythmical structure of the tune.
 TASHLHlYT SYLLABLES I 83

j'ai oublie ma montre dans un autobus cannot be sung to tune (13) is that
that sequence has one syllable too many. A basic regularity which governs
text-to-tune alignment in French - and in Berber - is that two successive
syllables in the text cannot be associated with the same occurrence of one
note in the tune.
Many speakers can also make judgments of a finer nature. They find
certain fits between text and tune better than others. If ma montre is replaced
with mon styLo (lmö.sti.lo/ 'my fountainpen') in (16), the resulting text
can still be sung on tune (13), but the speakers have the feeling that the
fit between the text and the tune is not as good as in (16).26
The phenomenon which is recorded in (15) is an alignment between
certain vocoids and certain notes. But such an alignment is only an observ-
able consequence of a more abstract alignment which involves phonological
constituents (syllables), not individual segments.
By now it should be obvious how text-to-tune alignment in singing can
be used as a source of data on the syllable structure of French. Consider
eblouira 'will dazzle' and debrouillera 'will disentangle', which are pro-
nounced [e.blu.i:ya] and [de.bruy:ya] in Standard French. As indicated by
our tran scriptions the first word is comprised of four syllables, and the
second of three. When the words are embedded in a sentence spoken at
normal speed the difference between [i] and [y] may not be easy to grasp
to an ear unattuned to French. But have French speakers try to sing the
two words to a particular tune, and the difference will no longer be in doubt.
If tune (13) is used as a carrier for the sentence Au clair de La lune, Lue
s'y debrouillera ([. . . lüksidebruyra]), the words flow effortlessly, with
one vowel aligned with each of the notes in the tune. With sentence Au clair
de La Lune, Sara t'eblouira ([... saratebluira]), on the other hand, speakers
will report that the words do not fit very well. They may be able to patch
things up, for instance by cramming the two syllables of Sara into position
7 in (13), but this amounts to singing to a different tune, a tune derived from
(13) by changing the eighth note in position 7 into two sixteenth notes.
Text-to-tune alignments such as (15) provide direct evidence about the
location of syllable nuclei. It may be the case that they could also give direct
evidence about syllable boundaries. We have not looked into that question .
In our work on Tashlhiyt, verse is used as a means of observing where
syllable nuclei lie in astring. Gur claims concerning syllable boundaries
were arrived at indirectly, through inferences based on our observations
about the location of nuclei.
French was brought in only as an expository device, and there is no point

26 The implicit conventions which govern text-to-tune alignment in French nursery rhyme s
favor a 'fem inine schwa' on the sixth note of tune (13). montre may be pronounced with
such a vowel in its final syllable, whereas stylo cannot.
84 CHAPTER FOUR

here in pursuing the question of exactly how much one can infer about
the syllable structure of French from a systematic study of alignments
such as those in (15). We now turn to the same question for Tashlhiyt.

4.4. PARSING TASHLHIYT VERSE: PRELIMINARIES

The starting point of our work on poetic scansion is Hassan Jouad's

pioneering work on versification in Tashlhiyt and Tamazight, for recapit-
ulations of which v. Jouad (1990, 1995). Jouad's corpus contains both
Tashlhiyt and Tamazight poems and he does not distinguish between the
two dialect groups in his discussion of the data. Since we want to be able
to characterize exactly the relationship between phonological structure and
poetic convention we have found it wise to begin by limiting ourselves to
Tashlhiyt, for which we have detailed firsthand knowledge of the phonology.
We have examined close to one thousand lines of Tashlhiyt verse. The pieces
in our corpus fall in two classes.
A first dass is comprised of poems taken down in Arabic script by
Ashlhiy speakers and published in various books. ME retranscribed these
pieces according to his own pronunciation. This is not always a straight-
forward matter because of some ambiguities in the Arabic transcriptions.
To retranscribe a piece ME sang it line after line to the tune to which it
was sung when first taken down; when he did not know the tune in question
he used another to which he feIt the lines sang naturally. Two pieces
belonging to this first dass can be found in Appendices II and III at the
end of this book.
This part of our corpus contains the following pieces: two songs by !rrays
EI-hajj Belaid , in Mestaoui (1996: 24, 38-39); one by !rrays Hmad
Biyzmawn, one by !tarrayst Rqiya Tandmsirt and one by !rrays Mohmmad
Andmsir, in Amarir (1975: 139-143, 147-150, 132-138); two improvised
oratorical encounters between Asid and Lachgar, transcribed from tape in
Asid and Lachgar (1996: 23-29, 83-86). The total number of lines in these
pieces is 380.
The other part of our corpus is comprised of poems presented by Jouad
and Bounfour in their studies of Berber versification. Either author gives
each line in a broad phonetic transcription, together with a French trans-
lation and a parsing which is consonant with his particular views about
Berber versification. In addition to the whole corpus of Bounfour (1984),
i.e. 322 lines in Igliwa Tashlhiyt, we have examined 276 lines in various
Tashlhiyt dialects published by Jouad, viz. 8 lines in poem II in (1990 :
284) and the 268 lines transcribed in pp. 94-116 and 134-141 of his 1995
book; these lines are parsed in pp. 178-201 and 216-223 in that book.
In poetic scansion, a Berber text is broken into small successive chunks.
Borrowing a term from Malone (1996), we shall call these chunks ortho -
metric syllables, to distinguish them from the syllables of the colloquial
 TASHLHIYT SYLLABLES I 85

language." For the sake of brevity, in what follows we shall refer to ortho-
metric syllables as syllables tout court when there is no risk of ambiguity.
Jouad was the first to realize that Tashlhiyt and Tamazight versifica-
tion distinguishes between heavy and light syllables and that the syllables
in a poem can often be arrayed to form a table such as (19) below, where
all the syllables in the same column have the same weight. Tahar (1975)
had earlier made the same discovery for melliun , a verse genre in Arabic."
In all of Jouad 's work orthometric syllables are set up for reasons which
are independent of any particular views conceming the phonology of Berber.
In DE (1988) we pointed out the connection between orthometric syllabi-
fication and syllabification outside of poetry.
The analysis we will present here is essentially the same as that in DE
(1997a). It differs in some respects with OUf predecessors' views about
poetic scansion , for which v. Bounfour (1984), Jouad (1983, 1986, 1987,
1990,1995), Jouad and Lortat-Jacob (1982). A comparison of our analysis
with theirs can be found in DE (1997a). Let us simply say here that the main
difference between OUf analysis and theirs concems the distribution of
syllable nuclei. As we shall show below the location of syllable nuclei in
astring is to a great extent predictable from the segments and their order
in the string. Jouad in effect takes syllable nuclei as already given in the
strings which are inputs to his parsing operations. Unlike Jouad, Bounfour
attempts to predict the location of nuclei in consonant sequences.
Unfortunately his whole approach is predicated on the assumption that the
syllable structure of the Tashlhiyt dialect of his corpus (Igliwa Tashlhiyt)
is the same as that of Ayt Ndhir Berber, a dialect of the Tamazight group
whose syllable structure is discussed in Saib (1978). This assumption, which
we feel is wholly unwarranted, leads to serious problems, about which
see DE (1997a).

4.5. PATTERN SATISFACTION

In Berber poetry it is common for all the lines of a piece to be sung to

the same tune, e.g. the music of a piece of thirty lines is a sequence of thirty
repetitions of the same tune. As a consequence of this, all the lines of the
piece share the same meter. A meter is characterized, among other things,
by what we shall call a metrical pattern. A metrical pattern is a certain
sequencing of light (L) and heavy (H) syllables. In the lines given below,
for instance , the metrical pattern is LLHLLLHLLLLH, that is, a line must
be comprised of twelve orthometric syllables, of which the third, the seventh
and the twelfth must be heavy while the others must be light. Let us first

27 Malone (1996: 124) uses the term 'orthometric ' to 'denote the set of systematic euphonie
patterns deployed by a given language-cum-tradition in the verbal arts.'
28 On the versification of melhun, see Chapter 8.
86 CHAPTER FOUR

give the phonological representation of these lines, where certain segments

are highlighted for the sake of conspicuity."
(17) a. i-lla=nn 1- !da n=u- !madun )'=imikk net-ammn-t
b. ureas i-!zdar ad=tnt=d i-kks ula adetut Ikrn-n
c. i-ddu dar bab n=t-addar-t ad=fll=as alla-n
The pronunciation used in singing differs little from that used elsewhere."
In /l-rl in line a, 111 assimilates to the following IrI, whence rr (v. § 2.5.3.1) .
In the three occurrences of lad=1 in lines band c the final Idl assimilates
with the initial consonant of the following clitic (v. § 3.2.1.3), hence the
surface geminates tt (line b) and ff (line c); similarly It=dl yields dd in
line b; in lula adl in line b, finally, the sequence la a/ would be pronounced
as a long (tautosyllabic) vowel in normal speech. In singing two adjacent
identical vocoids may be reduced to a single one. We give in (18) the strings
of segments which result from these processes.
(18) a. i-Ilaenn r-Irza n=u-!madun )'=imikk net-ammn-t
b. ur=as i-!zdar at=tnd=d i-kks ul at=tnt lkm-n
c. i-ddu dar bab n=t-addar-t af=fll-as alla-n
Morphosyntactic boundaries do not play any role in orthometric syllabifi-
cation . We give them only for the readers' convenience. Table (19) indicates
how the material in (18) is parsed. The exclamation points indicating
emphasis have been omitted to make the vertical alignments more per-
spicuous. Geminates which straddle successive syllables are indicated by
a tilde. We will seethat such geminates play a special role in the assign-
ment of syllable weight.
(19)
1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a. n- lan- nrr Za nu ma dun )'1 mik- kn tarn- mnt
b. u ra siz da rat- tnd- dikk su lat- tn tl kmn
c. id- du dar ba bn tad- dar taf- fl- la sal- lan
Table (19) displays at once two kinds of information which, although closely
related, must carefully be kept apart. On the one hand it presents data as
to how the material in (18) is sung. Each successive line in (18) is sung

29 These are the first lines of a poem by Mohmmad Andmsir in Amarir (1975: I32ff). 'A
siek man needed a \ittle honey / He eould not gather any (from a hive) nor reaeh any / All
in tears he went to see the owner of the hive'. Our translations do not seek eleganee and
they are sometimes rather approximative . We give them only to allow those who know Berber
to identify morphemes and syntaetic struetures .
30 V. below on the phonological differenees between the forms of language used in speaking
and in singing.
 TASHLHIYT SYLLABLES I 87

to the same tune, and each column in (19) corresponds to a particular

point in that tune, so that two chunks of the text which belong to the same
column are sung on the same portion of the tune." On the other hand,
table (19) displays an analysis of the metrical structure of the material in
(18). According to that analysis the material in question is parsed so as to
satisfy the metrical pattern LLHLLLHLLLLH. Analyses of this kind are the
main topic of this section and those that follow.
In (19) certain syllables do not contain any vowel, e.g. a3, al O, a12 .
Syllables can even lack any voiced segment, as is the case for syllables
a5, b5 and c9 in the following three lines, which have the same metrical
pattern as those in (19):32
(20) I 2 3 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a. i ga zun dl fq- qi hitt fa ru ka nl liaqq
b. r wa hag" ma st ma zir tn na "Y ak - ka W1"Y
c. i ga sak w fa fn ti bin sr tt lu lia man
These voiceless orthometric syllables must be pronounced in singing in
the same way as they are in speaking, viz. without any glottal vibrations.
The occurrence of a voiceless syllable does not cause any perturbation in
the alignment of the neighboring syllables with the tune. The note aligned
with the voiceless syllable is simply skipped in singing.
Besides satisfying a metrical pattern such as LLHLLLHLLLLH, there
are other conditions which a stretch of text must meet if it is to count as
a well-formed line of verse . These conditions restriet the distribution of
consonant types in a line, v., e.g., Jouad (1990: 302ff). We will not take
these restrictions into account in what follows. Also, there is more to the
metrical organisation of a line than a mere sequencing of Hand L sylla-
bies. Jouad (e.g. 1995: 237ff) has proposed that orthometric syllables are
grouped into foot-like constituents, and as a consequence of this organi-
sation Hand L do not combine freely to form metrical patterns. We will
not take into account constituent structure above the level of syllables either,
because we feel we do not understand it weil enough. We shall focus our
discussion on the grouping of segments into orthometric syllables.

31 Actually this is a simplification. The music on which ME sang this particular piece in
order to retranscribe it requires Iines to be grouped in couplets. The tune for the first line
of a couplet is different from that for the second line, but both tunes have the same rhyth-
mical structure. Each column in (19) thus corresponds to a given point in that rhythmical
structure. The score of the tune in question is given at the end of this book, see Tune I in
Appendix IV.
32 These lines are respectively line 69 in the preceding piece and Iines 33 and 46 in that
of Appendix 11. Here is the text of the first line: i-ga zund l-fqqih i-Ittfar ukan l-haqq ' like
the man of leaming, he is indeed entitled' .
88 CHAPTER FOUR

Let us say that an analysis of a line into a succession of orthometric

syllables is a (syllabic) parse of that line, e.g. (19)a is a syllabic parse of
(18)a. One question one may ask is: What are the features in the phono-
logical make up of astring such as (18)a, considered independently of its
possible uses in poetry, which allow that string to be parsed as in (19)a?
Consider for instance the string /tntlkmn/, which is parsed as LLH
(tn.tl.kmn) at the end of line b in (19). Could that string be parsed instead
as HLL (tnt.lk.mn) or as LHL (tn.tlk.mn) to allow it to appear in lines of
verse with metrical patterns requiring HLL or LHL? In order to be able
to answer questions like these, we shall try to answer the question 'How
does one parse a sequence of words into a succession of orthometric
syllables ?' :" We tackled that question previously in our 1988 article, where
we argued that orthometric syllabification is determined in part by the
sonority relations between adjacent segments, as is syllabification outside
of poetry. In DE (1997a) we supplemented our earlier idea about the role
of sonority with a detailed account of the role of geminates in orthome-
tric syllabification. In what follows we draw heavily on DE (1997a) .
Our goal in the following discussion is to make it clear what we mean
exactly when we say that astring satisfies a metrical pattern. This problem
has two subparts : parsing astring, and determining whether the syllabic
parse(s) match the metrical pattern. We assume that parsing astring into
a sequence of orthometric syllables is something which can be done inde-
pendently of any particular pre-determined metrical pattern. Here is how we
propose to formulate the problem of metrical pattern satisfaction.
(21) PATTERN SATISFACTION
Let S be astring of segments in Tashlhiyt and P a metrical
pattern. In order to determine whether string S satisfies metrical
pattern P one goes through two stages:
a. SYLLABIFICATION: one parses S, that is, one lists the set
SYL = {LI' . .. , Ln} of all the (well-formed) syllabic parses
of S; these parses are determined independently of P;
b. EVALUATION : one searches SYL for a parse which
satisfies P;
String S satisfies P if one member of SYL satisfies P.

JJ Nowhere does one find an answer to that question in Jouad's work. In his various pub-
lications, Jouad tries to answer the following questions , which are different: Ci) Given a
metrical pattern and a line of verse which matches that pattern, how does one parse the
line into orthometric syllabies? (ii) Given a sequence of lines which all match the same
metrical pattern, how does one discover what that metrical pattern is? Answering our question
also answers Jouad's, but the converse is not true.
 TASHLHIYT SYLLABLES I 89

Take any text which can be viewed by Ashlhiy speakers as a succes-

sion of well-formed lines with the same metrical pattern. In order to
complete the program outlined in (21) we must devise a definition of the
notion 'well-formed parse' which would enable us to parse such a text so
that the resulting syllables can be arrayed in a table with a layout analo-
gous to that of (19). The data we will examine are native speakers'
acceptability judgements. The set of pieces in our corpus can be consid-
ered as a repository of acceptability judgements made in the past by the
people who composed those pieces and by those who performed them when
they were taken down. We will supplement the acceptability judgements
frozen in our corpus by others whieh ME has elicited from himself by trying
out various text-to-tune alignements. Let us give an example.
At the beginning of (19)a the final n in illann is an onset to rr. Imagine
instead that we make one syllable with the two skeletal slots in nn, leaving
everything else unchanged. Instead of (19)a, line (18)a would have the
following parse:
(22) 12345678 9 10 11 12 13
il- la nn rr za nu ma dun 'Vi mik- kn tarn- mnt
A native speaker who tries to sing line (18)a in accordance with this parse
runs into several problems. Let us limit ourselves to the easiest one to
state: the parse has thirteen syllables, one too many to fit the tune of the
song. In order to sing line (18)a in accordance with the new parse, rr must
move over to the fourth position in the tune as shown in (22), ia to the fifth,
and so on, so that one runs out of notes when one has reached tam-, Two
successive syllables in the text cannot be associated with the same point
in the tune. If a tune is comprised of n successive notes, a text with more
than n syllables cannot be sung to it.
As will become clear later, we do not have a complete solution to problem
(21); but framing our discussion in terms of (21) will at least enable us to
state preci sely how much empirical ground we have been able to cover
and what areas require further study.

4.6. GENERALIZATIONS ON ORTHOMETRIC SYLLABLES

In this section we will present various generalizations on the form of

orthometric syllabl es in Tashlhiyt. Let us first state our general assump-
tions about syllable structure.
The units which are grouped into syllables are prosodie positions, i.e.
skeletal slots ('X slots'). Strings must be exhaustively parsed, i.e. every
X in astring must belong to a syllable." Syllables may not overlap, in

34 See Selkirk (1981).

90 CHAPTER FOUR

other words an X may not belong to more than one syllable. We take
syllables to be arborescent structures analogous to those which represent
constituent structure in syntax. Adopting well-known ideas and terminology
(v. e.g . Harris (1983), Levin (1985» we take a syllable (o) to be composed
of an onset (0) and a rime (R); the rime is composed of a nucleus (N)
and a coda (D). Every syllable contains a rime, and every rime contains a
nucleus. A syllabic segment is a skeletal slot which is dominated by N.
Skeletal positions not dominated by a nucleus are called margins. (23)
displays the first syllables in three French words, briscard [bviskav]
'veteran', yougoslave [yugoslav] and ouvrage [uvvaz] 'work'.

(23) a. bvis b. yu c. u

.r-:
o o o
~ I
0 R 0 R R
r-.
;1
X X
N
I
X
D
I
X X
I
N
I
X
N

X
I
I
I I I I I I I
b )' I s I U U

Following various authors," we assurne that glides have the same feature
composition as the corresponding high vowels. The only difference between
i and y is that i is syllabic while y is not, and similarly for u and w. Let
us use 'U' and 'I' as stand-ins for the feature bundles of (u , w) and Ci, y)
respectively. The same feature bundle is a nucleus in (23)a and an onset
in (23)b .
Constituents are said to be complex if they contain more than one skeletal
slot. (23)a has a complex onset and a complex rime . A geminate is com-
prised of two skeletal slots ; in Tashlhiyt these two slots may or may not
belong to the same syllable, v. § 3.2.2. In (24) we represent syllables I to
3 in (19)a and syllables 7 to 10 in (19)c.

35 See Kaye and Lowenstamm (1984). For more recent references see e.g. Harris and
Kaisse (1999).
 TASHLHIYT SYLLABLES I 91

(24) a. it-Jan-inrr
o o o

R
I .>R
0
~
0 R

N
r-.D r-.
N D
r-,
N D
I I I I I I
x x x x x x x x
I
I
<:
I
I
a
<:
n
Vr
b. dar. ta.fl-,la
o o o o
~ ~ /1 ~
0 R 0 R 0 R 0 R
r-.D
N
I
N N
I
N
I
I I I I I
x x x x x x x x x
I
d
I I
a r
I
t
I I
a f
<>
I
I
a

At thi s point, let us pau se briefly to comment on one aspect of the

notation used to represent the inputs to orthometric syllabification. Thi s
notation is very close to the broad phoneti c transcr iption used elsewhere
in this book . This raises no question for most segments, since in general our
transcription does not represent syllable structure, e.g. an n which is an onset
and one which is a nucleu s are both repre sented simply as 'n'. The only
segments who se syllabic status is recorded in our phonetic transcription
are the high vocoid s, e.g. that transcription uses 'u' to represent U when
it is a nucleu s, and 'w' to represent it when it is a margin , and one might
think that the input to orth ometric syllabification should not have a dis-
tinction between ' u' and 'w ' , since the distinction is the result of syllabifying
that input. In fact, U and I are different from the otherfeature bundle s of
Tashlhiyt, in that their syllabicity cannot alway s be predicted from the envi-
ronm ent. As we shall see later, Tashlhiyt has an underlying distin ction
between two kinds of high vocoids . Some high vocoids alternate between
vowel s and glide s, while other s are marked in the lexical representations
so as to guarantee that they (nearl y) always surface as margins. We use
the expres sion 'underlying glides' to refer to the high vocoid s of the latter
type . Thi s chapter and the next focu s primarily on the syllabification of
segments other than the underly ing glides. The syllabification of the under -
lying glide s raises speci al problems whose discussion must await a later
chapter. In the meanwhile, in our notation of the inputs to orthometric
92 CHAPTER FOUR

syllabification, we follow the convention introduced in § 2.2: every word

in a line is transcribed as it is pronounced when it is neither preceded nor
followed by a vowel, e.g. when it is uttered in isolation.
We now start with two basic observations about orthometric syllables
in Tashlhiyt.
(25) NoHiatus: a syllable which is not line-initial has an onset.
(26) RimeSize: a rime contains at most three slots; in three-slot rimes
the last two constitute a geminate .
These generalizations can be illustrated with the lines in (19) and (20).
The only onsetless syllables are those at the beginning of lines. In all the
other syllables the first segment is an onset. Syllables with a single slot
can only occur at the beginning of a line since they are onsetless : the slot
in question must be a nucleus (v. (23)c). After a syllable has been stripped
of its initial segment , the remainder (the rime) never contains more than
three slots, and all the three-slot rimes have codas consisting of a geminate,
v. b7 in (19) and a7, 12 in (20).
Let us now turn to syllable weight. If one sets aside those rimes in
which a geminate is involved, the situation is a simple one: one-slot rimes
are light and the others are heavy. When the geminates are brought into
the picture the generalization in the preceding sentence remains true in most
circumstances: it is still the case that one-slot rimes are all light (v. c9 in
(19», and furthermore rimes which contain both slots of a geminate are
all heavy, v. e.g. syllables nrr (a3 in (19» and dikk (b7 in (19». The added
complexity comes from codas consisting of a single slot which is the first
half of a geminate, as in it- (al) (and all other two-slot rimes in (19)
which are followed by a tilde). All such occurrences involve a hetero-
syllabic geminate with a first half which is not a nucleus, v. the represen-
tations of the first two syllables in (24)a. In cases of that kind we dub the
first slot of the geminate a hinged coda. Syllables with hinged codas are
extremely common in OUf corpus, and they occupy L positions in the over-
whelming majority of cases; the examples in (19) are typical in that respect.
Our corpus also contains occasional occurrences in which a syllable with
a hinged coda occupies a H position, v. e.g. the third syllable in lines 8,
11 and 39 in Appendix 11. Since the proportion of such occurrences is
quite small it is very tempting to consider them as violations of the con-
straints on syllable weight in Tashlhiyt versification. However ME feels that
the lines containing such occurrences flow as naturally as others whose well-
formedness is not in doubt, which compels us to consider them well-formed.
Here, then, are the generalizations one can formulate about syllable weight:
 TASHLHlYT SYLLABLES I 93

(27) Weight:
a. rimes with one slot are light;
b. rimes with a hinged coda may be either light or heavy;
c. other rimes are heavy.
We have just seen that geminates have two special properties which set
them apart from other XX sequences: they are the only complex codas
allowed (v. (26)), and the only XX rimes which may occupy aL position
are those with a hinged coda (v. (27)b).
The ambiguity of hinged codas (v. (27)b) is a phenomenon which
concerns only the weight of syllables, not the apportionment of X slots
between syllables . As used in (21) and in the discussion below, the expres-
sion 'syllabic parse ' is meant to refer to the distribution of syllabic nuclei
and syllabic margins in astring, not to the associated distribution of syllabic
weights. Consider a hypothetical string Irtta/ occurring at the beginning
of a line. Ir.ttg!, Irt- .t~1 and 1r1-.tg! are three different syllabic par ses in
the intended sense (the underlyings indicate syllable nuclei), but counting
the first syllable in Irt-.tg! as H or as L does not result in two different
syllabic parses. Let us use the expression ' weighed parse' to refer to a parse
with its associated syllable weights. It will emerge from our discussion
that Irtta/ has only one licit syllabic parse in Tashlhiyt verse, viz Irt-.tg!,
and that this parse allows two weighed parses, (lrt- .t~/, LL) and
(lrt-.tg!, HL).
In DE (1988) we followed Jouad and Bounfour in assuming that in poetry
geminates may in general count either as one segment or as two to suit
the needs of the poets. We have shown in DE (1997a) that this assump-
tion is incorrect; it allows many syllabic parses which are never found to
occur. Note that in singing , long consonants are pronounced distinct from
their short counterparts, as they are in the colloquial language . This is in
particular true in the case of rimes with a hinged coda which occupy a L
position in the metrical pattern. When singing (19)c one does not pronounce
the words taddart and allan as though they were tadart and alan.
Certain grammatical morphemes which must be pronounced with a long
consonant in the colloquial language have a poetic variant in which the long
consonant is replaced by its short counterpart. The colloquial form also
occurs in poetry. For instance walaynni 'however' can also be pronounced
walayni in poetry (v. line 43 in Appendix III); as we shall see, the former
variant has four syllables (wa.la.yn-.ni) whereas the latter has three
(wa.lay.ni) . Other items which can be degeminated in poetry are the initial
Innl in certain possessive determiners (v. line 12 in Appendix II) and the
geminate which results from totally assimilating the genitive preposition In!
to a following high vocoid (v. line 25 in Appendix III). The morphemes
involved are all grammatical morphemes, e.g. a-snnan 'thorn' cannot be
pronounced asnan . Furthermore not all grammatical morphemes can undergo
94 CHAPTER FOUR

the degemination in question ; the directional clitic /nn/, for instance, cannot
be pronounced short.
Like the weight ambiguity of the hinged codas, the degemination above
provides poets with a wild card to help them meet the demands of the meter,
but this should not obscure the differences between the two phenomena.
Weight ambiguity provides a choice between two different weight assign-
ments for the same segment sequence. Degemination, on the other hand,
provides a choice between two different segment sequences as exponents
of a particular morpheme .
Table (28) lists the syllable types allowed by generalization (26) together
with their weights according to generalization (27).

(28) types N= [-cons] N= [+cons] weight

a. (0) N nu (a5) kn (alO) L (light)

b. (0) N- ri- 36 fl- (c9)
-------------- -- ---------------- -------------- -------- --
c. (0) N D- rnik- (a9) tnd- (b6) L or H
---------------- - ------------- -- ---- -- -------- ----------
d. (0) N D siz (b3) kmn (b12) H (heavy)
e. (0) N-D biy" nrr (a3)
f. (0) N D-D dikk (b7) bndd"
In the first column each occurrence of a capital letter represents a skeletal
position. Those skeletal positions which are the first half of a geminate
are indicated by a following tilde." The parenthesized Os are areminder
that line-initial syllables may lack an onset. In the examples in the second
column the nucleus is a vocoid, and in those in the third it is a contoid.
Some examples occur in (19). Their location there is indicated by the paren-
thesized letters and numbers. The other examples occur in the pieces in
Appendices II and III.
In the third column of the table the consonant which is a nucleus is a
sonorant in all our examples. This is because we have tried as much as
possible to draw the examples in the table from the three lines in (19),
and all the nuclei in these lines happen to be sonorants. Although sylla-
bles with an obstruent as a nucleus are less common than those with a
sonorant nucleus - why it is so will become clear in the next seetions - they
are every bit as well-formed and they divide likewise into the six cate -
gories a-f of table (28). Having gone over various pieces of poetry recorded
by Jouad, Shaw (1996) pointed out that they did not contain closed sylla-

36 Appendix In, syllabIe 4 in line 64.

37 Appendix In, syllabIe 8 in line 78.
38 Appendix H, syllable 7 in line 49.
39 In the first coIumns of tabIes (28) and (47) all geminates are indicated by tildes. EIsewhere
onIy hetero syllabic geminates are indicated by tiIdes.
 TASHLHIYT SYLLABLES I 95

bles with obstruent nuclei and she proposed that such syllables be excluded
by Universal Grammar. However such syllables are not judged ill-formed,
as the following examples will illustrate. The first example is a ditty about
a bird, the hoopoe. Each line of the song's text is followed by its parse.
(29) stu tutut s tu tu tut
stu tutut s tu tu tut
t-"{la t-isn-t t"{ la ti snt
t-sqqa t-aka-t" tsq- qa ta kat
All four lines share the metrical pattern LLLH. The words in the first two
lines are onomatopeias imitating the hoopoe's song. The first syllable in
the last line is a closed syllable with s as a nucleus.
Our other examples are lines composed by ME. The lines below are sung
to the tune of the song by Hmad Biyzmawn recorded in Amarir (1975 :
139-143). The tenth syllable in each line has a heavy rime with an obstruent
nucleus."
(30) 1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L L L L H L H
a. 1 "{a sur gu mnt- tf ra Wl na tzd mi "{aln42
b. al- la huk ba rd- du ni ti nu tst nu linw
Like Shaw we did not find any closed syllable with an obstruent nucleus
in our corpus. This absence may have to do with preferences in the matehing
of texts with tunes: there seems to be a preference for aligning H posi-
tions in the meter with portions of the tune which are musically prominent,
and obstruent nuclei are not ideal carriers of musical notes. Consider the
following lines, which sound quite natural to ME despite the fact that the
third or the seventh syllable has a complex rime with an obstruent nucleus."
(31) 1 2 3 4 5 6 7 8 9 10 11 12
a. "{U rir si lh na tfk ta da ri lu tan
b. i xa rinn "{a ki ba n55 fa ts- su ta kal
c. ta nq- qzb rut- ta da xas ts "{a lal- la tnx

40 'Stu tutut / Stu tutut / SaIt is expensive / Making ends meet is difficuIt' (Iiterally 'the
hearth is difficult').
41 (a) i'(=as ur gum-nt t-fraw-inea t-zdrn i-val-n, 'if these twigs are not enough, let her
chop armfuls '; (b) !allahukbar d-duni -teinu t-Istn uleinu, ' Allahoakbar! Life here below
troubles my heart' .
42 This meter requires final syllables with 'compound rimes' . On syllables with compound
rimes, see the end of this section .
43 (a) i'( ur i-lrsi l-hna t-fk-t a-ldar i=!Iutan, 'if peace does not settIe, begin your joumey'
(Iiterally : ' give the foot to the lands'); (b) ixar inn '(=ak i-ban s-!sfa t-ssu-t a-kal, 'where
serenity comes to you, there you should take the ground as your carpet ' : (c) t-a-!nqqzbrud-t
adeaxeas t-sva lallaetnx, 'it is a small poncho which our mistress has bought hirn' .
96 CHAPTER FOUR

These lines have the same metrical pattern as those in (19) and in Appendix
11. They can be sung without a hitch to certain tunes which are compat-
ible with that metrical pattern, but not with all. In particular they do not
fit with the tune of the song in Appendix II, a tune in which the notes
associated with third syllable and the seventh are musically prominent.
Besides syllables with an obstruent nucleus, there is another syllable type
which is implicit in table (28) and whose existence deserves explicit recog-
nition : syllables in which the only X slot is the first half of a geminate,
i.e. onsetless syllables of type (28)b . Such syllabl es appear at the begin-
ning of the second and third line below. The first line is given to illustrate
the metrical pattern shared by the other twO. 44
(32) a. a man a-drar ur nkki n-stara i-Izavar-n
b. lss-rmi-v kullu t-!tlba n-ss-!rmi i-grrram-n
c. kki-v l-Ibhur stara-v i-gnwa-n d=i-kal-n
(33)
1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L L L L H L H
a. a ma nad ra ru rnk- ki ns ta ray za vam
b. s- sr miv kul- lut- tl ba ns- sr miy gWr_ ramn
c. k- ki )'lb ou rs ta ra )'i gn wan di kaln
The initial a in the first line is a stop-gap vowel, a common device used
by Ashlhiy poets to add an extra syllable at the beginning of a line ." Any
geminate consonant can be a nucleus-onset sequence at the beginning of
a line. We have chosen the above examples as further illustrations of OUf
claim that even a voiceless consonant can play the role of a nucleus. Note
in particular that the initial syllable in (33)c is completely silent."
In our 1997a article, which the present discussion follows closely, we
allowed for orthometric syllables of a special kind which were dubbed
'compound' syllables. The strings we considered as compound syllables are
found only at the end of lines. Certain meters allow or even demand a
compound syllable at the end of the line, e.g. the meter of the song from
which the lines in (33) are excerpted. In these lines the twelfth position is
occupied by strings vam, ramn and kaln, which cannot be parsed as single

44 Lines 19,20 and 69 of the poem by Hmad Biyzmawn in Amarir (1975: 139-143). (a)
Ah! Which mountain did I not comb, which plains! ; (b) I have badgered the scholars and
the marabouts ; (c) I've crossed the seas and I've been all over the skies and the lands .
45 V. Galand-Pernet (1969).
46 In lines 12 and 56 of the poem in Appendix III (but not in line 59) the first half of the
initial geminate is 'left out' of the parse. ME finds such violations worse when the geminate
involved is a sonorant than when it is an obstruent.
 TASHLHIYT SYLLABLES I 97

syllables which would fit into table (28).47In 1997 we considered each such
string as a syllable resulting from the contraction of two light syllabIes
into one: va.m, ra.mn, etc. We now believe that these strings are not
syllables of a special sort. They must be analyzed instead as sequences of
a heavy syllable followed by an onset, viz varn-, ram.n-, In the few pieces
known to us in which lines end in such strings, the supernumerary segment
at the end of each line is actually sung as an onset to a following vowel.
In some pieces that vowel is a stopgap i or a which serves as a carrier to
the final note(s) of the tune, while in others it is the first segment of a refrain
which is repeated after each line.

4.7. THE ROLE OF SONORITY

There are many ways to parse a given sequence of segments into succes-
sive chunks which all belong to one of the categories listed in table (28).
Consider for instance sequence llattntlkmnJ, which is parsed as lat-itn.tl.kmn
at the end of (19)b. If orthometric syllabification only required that each
syllable fit into table (28), latt.nt.lk.mn would also be a licit parse, coun-
trary to fact.
In order to fulfill the general program outlined in (21) we should produce
a device capable of enumerating all the well-formed parses of any string
in Tashlhiyt. Because more empirie al work is needed on the syllabifica-
tion of certain types of consonant clusters (v. § 4.9) we cannot present
such a device . What we will present instead is a set of conditions which any
parse must meet if it is to be well-forrned. These considerably narrow
down the set of parses associated with any string; in many cases in fact they
reduce it to a single parse.
Gur analysis gives a central role to the sonority relationships between
adjacent segments . As far as sonority is concerned, the empirieal general-
izations which our analysis must account for are those stated in Elmedlaoui
(1985) and DE (1985): in a nutshell, syllable nuclei must have the highest
degree of sonority compatible with other requirements such as the prohi-
bition of hiatus . In our works of 1985 and 1988 we operated within a
rule-and-constraint framework. Starting from representations devoid of
any syllabic structure, syllabic trees were built in a stepwise fashion through
the operation of sequentially-ordered rules; the rules failed to apply when
their operation would have created adjacent nuclei (a hiatus).
The data in our articles of 1985 and 1988 have been used as a testing
ground for various theoretical propo sals." In particular, Prince and
Smolensky (1993) proposed an account of Tashlhiyt syllable structure within

47 For other examples, see for instance the piece recorded in Jouad (1995: 193).
48 Goldsmith and Larson (1992), Prince and Smolensky (1993), Scobbie (1993), Zec (1995),
Shaw (1996), Clements (1997), Frampton (1999).
98 CHAPTER FOUR

Optimality Theory (heneeforth OT). Clements (1997) has raised various

objeetions to that aeeount and set forth his own analysis, also in the OT
framework. The analysis we will present here revolves around SonPeak (39)
and NoRR (42), two eonstrainsts on sonority sequeneing which are borrowed
from Clements.
Let us first reeapitulate the generalizations we have seen thus far.
(34) a. every skeletal slot belongs to one syllable and only one ;
b. NoHiatus (25) ;
e. RimeSize (26);
d. eomplex onsets are disallowed."
The propositions in (34) are deseriptive statements whieh are true of all
the lieit parses. Let us eonstrue them also as eomponents of the grammar
of Tashlhiyt, i.e. as well-formedness eonditions on parses, and more specif-
ieally as eonstraints in the sense of OT. Exeept for the last, maybe, the items
in (34) are not eonstraints in the striet sense but rather eomposites of several
eonstraints. We will nonetheless treat them as unanalyzable wholes. We
assume that (34 )a-d are all undominated eonstraints, i.e, in ease of eonfliet
with other eonstraints those in (34)a-d always take preeedenee. In the
diseussion below, unless stated otherwise, we only take into eonsideration
parses which abide by the eonstraints in (34)a-d.
We use the following sonority scale."
(35) Sonority seale:
(a) low voeoids: a
(b) high voeoids: i,y,u,w
(e) liquids: r, I
(d) nasals: m, n
(e) frieatives: s, s,
x, x", z, z,
-y, -yw, h , fl, h
(f) stops: t, k, k", q, qW, b, d, g, g"
The objeets ranked on this seale are melodie units , i.e. bundles of feature
speeifieations dominated by a Root node, feature bundles for short. The
higher a feature bundle is on the seale, the more sonorous it is said to be,
or equivalently, the higher its degree of sonority. i is more sonorous than
n, whieh is more sonorous than d, whieh is as sonorous as t.51 An assump-
tion implieit in (35) is that the degree of sonority of a feature bundle is
the same in all its oeeurrenees; in particular is not affeeted by syllable struc-
ture. For the sake of explieitness the rung of the seale for high voeoids

49 Implicit in table (28).

50 On sonority and its role in syllabification, see Clements (1990) for a careful discussion
and references.
5\ On the differences between the sonority scale in (35) and that used in our earlier work,
see below the end of § 4.9.1.
 TASHLHIYT SYLLABLES I 99

eontains four symbols but it should eontain only two sinee 'i' and 'y' are
different labels for the same feature bundle, as are also 'u' and 'w' .52
Let us return to the question we asked at the beginning of this seetion
about Ilattntlkmnl. We said that the lieit parse is 1~t.tn.t1.kmn (unless indi-
eated otherwise, underlinings indieate nuclei) and we asked what excludes
lgtt.nt.lk.mn. Aeeording to Clements (1997) the answer is that Igtt.nj.lk.mj;
violates the constraint SonPeak whereas 1~t.tn.t1.kmn does not violate the
eonstraint:
(36) SonPeak (a first approximation):
Every segment whieh is more sonorous than its immediate
neighbours must be a syllable nucleus.
We repeat below the parses under eonsideration, together with the input
string, in whieh the segments whieh are more sonorous than their imme-
diate neighbours are in bold type for the sake of eonspieuity:
(37) input I a t t n t I k m n
a. I a t. t n. t 1. k m n
b. * l ~ t t. n l.l k. m n

Whereas the constraint does not ineur any violation in parse (37)a, it is
violated twiee in (37)b : n and I are sonority peaks, i.e . they are more
sonorous than their immediate neighbours, and yet they are not syllable
nuclei in (37)b.
It is important to note the asymmetry in the eonstraint: while (36) requires
sonority peaks to be nuclei, it does not require nuclei to be sonority peaks.
Consider for instanee the syllables 7 to 9 in (33)a, whieh are ki.n~.t~. The
string kinsta only eontains two sonority peaks, viz i and a, and yet there
are three syllable nuclei in the lieit parse ki .ns.ta. s is a nucleus, but it is
not a sonority peak, sinee it is adjaeent to n, whieh is more sonorous."
In the input string in (37), neither eonsonant in the final sequenee mn
is a sonority peak, sinee either is adjaeent to a segment of equal sonority,
but the sequenee taken as a whole is more sonorous than its immediate
surroundings, and it is useful to have a notion of sonority peak general
enough to be applieable not only to single X-slots, but also to sequenees
of X-slots. In the sense of ' sonority peak' whieh is relevant for the con-
straint in its final formulation (see below in (39)), the final sequenee mn
in (37) is a sonority peak, and eonsequently that sequenee must eontain a
syllable nucleus. Let us say what we mean exaetly by a sonority peak.
Following Clements (1997), let use the expression 'sonority peak' to refer

52 V. text under (23) in § 4.6.

53 n is not a sonority peak either since it is adjacent to i. The fact that i does not belong
to the same syllable as n is irrelevant, as will be explained below.
100 CHAPT ER FO UR

to any sequence of segments which is a local maximum of sonority. A

local maximum of sonority is a sonority plateau which is higher than its
immediate neighborhood, or more precisely,
(38) Sonority peak (definition) :
A sequence Q is a sonority peak within sequence Riff the
following conditions obtain:
(i) Q is contained in R;
(ii) for any two adjacent segments q and r, with q contained
in Q and r not contained in Q, q has a higher degree of
sonority than r;
(iii) all the segments in Q are of equal sonority.
To take an example, consider the the phrase i=y-gnwa-n 'to the skies ' , which
is composed of the dative preposition i followed by the bound state noun
ignwan. iygnwan contains two sonority peaks, iy and a. The one-segment
sequences i and y are not sonority peaks because they do not meet condi-
tion (ii). an and nwa are not sonority peaks because they do not meet
condition (iii) .
We now give the definitive formulation of the constraint.
(39) SonPeak: A sequence which is a sonority peak within the
syllabification domain contains a syllable nucleus.54
In (39) the expression 'syllabification domain' refers to the unit which is
coextensive with the strings taken as inputs to syllabification. In this chapter
the syllabification domain is the Phonological Utterance, which is coter-
minous with the line in Tashlhiyt singing; in the next chapter the
syllabification domain will be the inflectional stern .
Returning to the example in (37) , the readers may check for themselves
that the input lattntlkmn contains four sonority peaks as defined in (38),
viz a, n, land mn, and that each sonority peak contains a nucleus in the
licit parse Igt.tn.tl.kmn «37)a). In lgtt.nt.lk.mg, on the other hand (see (37)b),
SonPeak is violated twice, for the sonority peaks n and I do not contain
nuclei.
Now what about parse Igtt.gt.lk.mg, which has syllable boundaries located
as in parse (37)b, but different nuclei in some syllabIes? This parse does
not violate SonPeak, but it is excluded because syllables nr and [k violate
NoHiatus (25) : they are not line-initial and yet they lack an onset.
Since, as stated above, we will restriet our attention to parses which abide
by NoHiatus and the other undominated constraints in (34), omitting from
the parses the underlinings which indicate syllable nuclei will not result
in any ambiguity except in line-initial position. For instance if sequence

54 SonPeak is Clement ' s Sonority Peak Principle (p. 303), slightly reworded .
 TASHLHlYT SYLLABLES I 101

tlkmn does not occur at the beginning of a line, the notation 'tl.kmn' can
on1y stand for tl.kmn; parses l1.kmn, l1.kmn and jl.kmn wou1d all violate
NoHiatus, and the latter parse would furthermore vio1ate the prohibition
of comp1ex onsets, as wou1d the parse tl.kmg.
Let us now consider the four lines in (40) and their scansions in (41).55
(40) a. yan i-grmr-n ar lid i-y=d ur umz-n yat"
b. ludn-v yaw w-Iattan" i-!dbib-n zla-n=a-y
c. alli-y=d n-wafaq-n f=l-ma)na nna-n
d. a l-mskin !attan 1=I-tmbb58 afefllak"
(41) 1 2 3 4 5 6 7 8 9 10
L H L L L H L H L H
a. ya nig" mr na ri di-y du rum zn yat
b. u dn-y yaw- wat- ta nid bi bnz 1a na-y
c. al- li-y dn wa fa qn fl ma? nan- nan'"
d. a lms ki nat- ta nll hub- baf- fl- lak
The first line is given for the sole purpose of exemplifying the metrical
pattern of the song . There is an interesting difference between sequence
ludn-yyal in line b, which is parsed as u.dnv.ya, and sequence li-ydnwal in
line c, which is parsed as iv.dn.wa. The form ofboth sequences is VCCCCV,
in which the last C must be an onset to the following vowel (that vowel
is a sonority peak) . In u.dnv.ya, it is SonPeak (39) which is responsible
for the fact that Inl is syllabic. If /udnvya/ were parsed as ud.ny.ya, the
sonority peak n would not contain a nuc1eus (it would contain an onset)."
In line c, on the other hand, if the VCCCCV sequence /ivdnwa/ were parsed
in the same fashion as u.dnv.ya, i.e. as V.CCC.CV, constraint SonPeak would
not incur any violation in the resulting parse i.vdn .wa, because there is
no sonority peak between /i/ and /a! in /ivdnwa/. Yet this is the wrang result.
Therefore we need to call upon another constraint. After Clements (1997 :

55 (40)a,b,c,d are respectively lines I, 18,20 and 21 in the song by EI-hajj Belaid in Mestaoui
(1996: 38ss). Here are translations. ' He who hunts till nightfall without catching anything /
I am ill and physicians have lead me astray / Having agreed on the symptoms, they
declare / Unfortunate! Your illness is that of love'.
56 Every line in this song ends with a stop-gap vowel i, a common occurrence in Ashlhiy
singing. The final i has been left out from our transcriptions .
57 From /yan w-lattan/ .
58 From /n=l-nubba/. /a#a/ reduces to a short a, see the text below (17).
59 From /ad=fllak/, see the text under (17).
60 This line is ill-formed : its sixth syllable should be heavy.
61 The same situation obtains later on in the same line, when /ibnzla' is parsed as i.bnhla
rather than as *ib.nz.la.
102 CHAPTER FOUR

303) we assurne that parses such as i.-yQn.wa are ruled out by the fol-
lowing constraint.f
(42) NoRR (No Rising Rimes):
The coda does not have a higher sonority than the nucleus.
If /ivdnwa/ were parsed as i.-yQn.wa, coda n would exceed nucleus d in
sonority.
Let us c1ear up a possible misunderstanding concerning (42) and its
resemblance with SonPeak (39). The parse i.-yQn.wa could seem to violate
Sonf'eak: n is a sonority peak inside syllable vdn, and yet it does not contain
any nucleus. However let us go back to the formulation of SonPeak in
(39): 'a sequence which is a sonority peak within the syllabification domain
contains a syllable nucleus.' n is not a sonority peak within sequence
/i-ydnwa/, since it is adjacent to w, which is more sonorous. An important
difference between constraints NoRR (42) and SonPeak (39) is that the latter
does not legislate over sonority relationships within the syllable. SonPeak
does not rule against syllables in which the most sonorous segment is not
the nucleus. It is not violated by syllables in which the most sonorous
segment is the onset, which are a common occurrence in our corpus, see
for instance lms in (4l)d2 or 1.:S in (33)c5.
NoRR excludes heavy syllables in sequences with a rising sonority.
Consider the (invented) sequence /ksmrua/, in which each segment is more
sonorous than the preceding one. Because of NoRR, this must be parsed
as ks.mr.wa. Constraint NoRR is violated only by certain rimes in which
r is the nucleus and w is the coda. These will be discussed in Chapter 7,
and for the purposes of the present chapter we consider NoRR as an undom-
inated constraint.
The constraints introduced up to this point make predictions which are
compatible with those of the syllabification procedure CS proposed in our
works of 1985 and 1988. Let us recapitulate these constraints here.
(43) a. Every skeletal slot belongs to one syllable and only one
b. Complex onsets are prohibited «34)d)
c. Condition (26) on well-formed rimes
d. NoRR (No Rising Rimes) (42)
e. NoHiatus (25)
f. SonPeak (39)
Setting aside (43)a, which is a general condition on well-formed parses,
the other constraints fall into two categories . The constraints (43)b,c,d
are conditions on syllable shape which must be met in any context. The

62 In Deli and Tangi (1993) the same constraint is posited to prevent Irl from being tumed
into a in certain contexts .
 TASHLHlYT SYLLABLES I 103

constraints (43)e,f further restriet the shape of syllables in certain envi-

ronments.

4.8. GEMINATES IN COMPLEX CODAS

As far as we know, any string of segments of Imdlawn Tashlhiyt can be

parsed exhaustively as a sequence of orthometric syllables, i.e, any such
string has at least one licit parse .f Are there sequences which have more
than one lieit parse? We know of two classes of such sequences. In this
section we deal with one case, which involves complex codas. The other
case will be taken up in the next section.
Consider the following song, which Imdlawn people sing in unison while
winnowing on the threshing floor. The song has only one line, which is
repeated over and over again, always to the same tune."
(44) bab neu-wwtif a65 i-g !rbbi l-baraka "(=u-nnrar
What is the metrical pattern of this song? One cannot compare how suc-
cessive lines align with the tune, since the song has only one line. The
following parse is compatible with everything we have said up to this point.
(45) 1 2 3 4 5 6 7 8 9 10 11
H H L H L H L L L H H
bab nuww ti fay grb- bil ba ra ka "(unn rar
But is this indeed the parsing used in the winnowing song? Here is why
we think it is not. ME has replaced the actual words in (44) by others and
tried to sing the resulting sequence to the original tune." His acceptability
judgements provide us with independent evidence as to the metrical pattern
of the song. If u-wwtif ss replaced by u-varas, u-grtil or u-iddig the resulting
sequence still sings naturally to the tune . These words are trisyllabic; they
begin with aLL sequence (u.va.ras, u.gr.til, u.Zd.dig), not with a H syllable,
which is what is assumed about uwwtifin (45). There is yet other data which
confirm that in the winnowing song uwwtif does not begin with a H syllable:
the line becomes lame (i.e. one gets stuck when one tries to sing it) when
uwwtif is replaced by u-wtil or by u-rgaz; which begin with a H syllable
(uw.til, ur.gaz). Instead of the H syllable in second position in (45), there
should be two L syllables:

63 It is not obvious that it should be so. For instance if closed syllables with an obstruent
nucleus were universally excluded, as proposed by Shaw (1996), and if SonPeak were invi-
olable, such a sequence as nakzdma could only be parsed with d left unsyllabified.
64 'Owner of the threshed grain, let God bless the threshing f1oor'.
65 lad! underlyingly . On the consonant in lad! and Irad! see DE (1989: 188-189).
66 For observations on French songs which were made using the same method, see Deli
(1989).
104 CHAPTER FOUR

(46) I 2 3 4 5 6 7 8 9 10 11 12
H L L L H L H L L L H H
bab nu ww ti fay grb- bil ba ra ka vunn rar

The line ends with the word unnrar, which has the same structure as uwwtif,
but in (46) the two words syllabify differently: u.ww.ti . .. vs. unn. ra . ..
Could it be that in the winnowing song both words have parallel parses, and
that the final word should be parsed as u.nn.rar rather than as unn.rar?
No, as the following data show. Replacing unnrar by u-wtil or u-rga: does
not impair the text-to-tune alignment, while the line becomes lame when
unnrar is replaced by u-varas, u-grtil or u-zddig,
The preceding data suggest that the same sequence VC-CCV may have
two licit parses, VC-C.CV (unn.ra) and V.C-C.CV (u.ww.ti) . That this is
indeed the case is concIusively shown by the fact that when one permutes
uwwtif and unnrar in (44) the resulting sequence sings naturally to the same
tune as (44) : placed at another location in the line, unnra . . . can be syl-
labified as u.nn.ra .. . and uwwti . . . can be syllabified as uww.ti .. ..
In (44), sequence uww in uwwtif and sequence unn in unnrar are what
we call 'EF-G' sequences. An EF-G sequence is a sequence of three X
slots the last two of which are a geminate, and which furthermore has the
following relationship with the surrounding string: (i) E belongs to a sonority
peak within the syllabification domain, and (ii) G does not immediately
precede a sonority peak. An EF-G sequence has two well-formed parses.
It may form a rime with a complex coda (type (28)f) or it may straddle
two open syllables whose nucIei are E and G. In the latter case the second
syllable is comprised of the two skeletal slots of a geminate. Syllable (F-G)
belongs to category (a) in table (28). Table (28) did not take into account
the possibility that an onset and the following nucIeus might belong to
the same geminate. Assuming that that possibility can combine freely with
those listed in table (28), one can derive from (28) a new table in which
each line represents the particular case in which the onset and the nucIeus
belong to the same geminate, see below table (47). Examples b, c, d, e
and f in (47) are invented. An underiining indicates a nucIeus . Otherwise
our conventions are the same as in table (28) .

(47) types examples weight

a. 0- N nn
L (light)
b. 0- N- nn-
------------------ ------------ ------------
c. 0- N D- nnk- L or H
------------------ ------------ ------------
d. 0- N D nnk
e. 0- N-D nnn H (heavy)
f. 0- N D-D nnkk
 TASHLHIYT SYLLABLES I 105

Syllabes oftype (47)b and (47)e are excluded beeause in Tashlhiyt the same
melodie unit eannot be linked to three X slots in a row, v. § 3.2.1.1. As
for the eases c.d and f of (47), we have not eneountered any example in
our corpus." Let us add that in all the instanees of type (47)a whieh we
have eneountered the syllabe in question is the F-G part of an EF-G
sequenee. Unless it belongs to an EF-G sequenee, the first skeletal slot
in a geminate eannot be syllabified as an onset.
Let us posit the following eonstraint, whieh forbids the first half of a
geminate to be an onset:"

(48) NoOns-: prohibit o =

(0 onset
I Rt = Root node)
X X
V
Rt

We now need to specify the way the eonstraints interaet so as to guar-

antee that the only violations of (48) whieh ean result in a lieit parse are
those oeeurring in EF-G sequenees. We have not been able to solve this
problem in a manner eompatible with the spirit of OT. Let us nonetheless
propose the following deviee, for the sake of explicitness.
Let us assurne that whereas all the parses whieh are optimal with respeet
to the eonstraints are lieit, there exist lieit parses whieh are not optimal.
These are derived from optimal parses through the operation of optional
rules. Let us assurne that eonstraint (48) is undominated; no optimal parse
violates it. To derive lieit parses of type (47)a, whieh violate (48), we
posit a late optional rule whieh takes optimal parses as its inputs. The rule
turns a eomplex eoda into a syllable:
(49) DETACH : (X X )D~ (X x i,
(OPT)
To illustrate the role played by DETACH, let us present an analysis of
two examples. In our first example we will reeapitulate how (44) gets to
be parsed as (46) . Let us go baek to the general seheme outlined in (21).
We start from the input sequenee (44) and the metrieal pattern required
by the tune of the winnowing song , which we repeat below in (50).
(50) H L L L H L HL L L H H

67 In Jouad (1995) one finds mu.nnt (Iine 4 p. 183 and line 4 p. 196) and li.mmk (Iine 8
p. 220) . The se parses do not contradict our assertion since in the sequences in question
'nn' et 'mm' are not geminates, see later in this section. On the first form v. DE (1985 :
128 note 46) .
68 This constraint was already discussed in our 1985 work and in Deli and Tangi (1992 :
132-133).
106 CHAPTER FOUR

As preseribed in (21)a, one first enumerates the set of lieit parses of

sequenee (44). (44) has four lieit parses, whieh are listed in (51).
(51) a. bab nuww ti fay grb- bil ba ra ka vunn rar
b. nu ww 'Yunn
e. nuww 'Yu nn
d. nu ww 'Yu nn
Here is how the parses in (51) are arrived at. The lieit parses of a sequenee
are obtained by applying rule DETACH (49) to eaeh of the parses whieh
are optimal with respeet to the eonstraints. In the ease of sequenee (44),
the various eonstraints diseussed above define a single optimal parse, viz
(51)a. Sinee rule DETACH is optional and its eonditions are met in two
plaees in (51)a, the operation of DETACH in (51)a yields four parses, i.e.
(51)a-d, whieh constitute the set SYL mentioned in (21)a.
Let us now move to the seeond step in (21). Only at this stage does
metrieal pattern (50) eome into play. The metrieal pattern requires twelve
syllabies, whieh eliminates parse (51)a, whieh only has eleven, and parse
(51)d, whieh has thirteen . Moreover the pattern requires that the penulti-
mate syllable be heavy, whieh rules out (51)e. We are then left with (51)b,
whieh satisfies pattern (50) on the eondition that the sixth syllable (grb-)
be eonstrued as light, whieh is allowed by (27)b.
The various eomponents of our analysis and the way they interaet are
pietured in diagram (52).

(52) sequenee S

'"
eonstraints
SYLLABIFICATION
'"
optimal parse (2Ia)

'"
ruleDETACH

'"
lieit parses

'"
metrieal pattern P -) I EVALUATION (21)b

'"
yes / no
(does S satisfy P?)
 TASHLHIYT SYLLABLES I 107

For our seeond example, eonsider the beginning of (l8)a. As indicated

in (l9)a, sequenee illannrria parses as shown below in (53)a, but other
parses are eoneeivable. In particular, how is (53)b excluded?
(53) a. il- lan- nrr za
b. il- la nn rr Za
e. il- lann rr za
d. il- la nnrr za
Note first that (53)a is optimal. In partieular it does not violate SonPeak
(39) sinee the four sonority peaks in the sequenee, viz i, a, rr and a, eaeh
eontain a nucleus. NoOns- (48) is not violated either in (53)a. On the
other hand it ineurs two violations in (53)b. On the analysis presented above,
a violation of NoOns- (48) ean oeeur in a lieit parse only if it ean be
traeed to the applieation of rule DETACH in an optimal parse . For (53)b
to result from the operation of DETACH, the input to the rule must have
been (53)e or (53)d. But it is impossible for either representation to have
undergone DETACH, sinee they are not optimal parses : in (53)e, rr violates
NoOns- and in (53)d both SonPeak (39) and NoRR (42) are violated in
the third syllable.
Let us end this seetion with two remarks whieh will clarify the empir-
ical import of eonstraint NoOns- (48) .
The first remark eoneerns the differenee between geminates and
sequenees of identieal eonsonants. Consider the following line (a) and its
seansion (b ):69
(54) a. is t-ttu-m r-ribab ula n-naqus dlhin
1 2 3 4 5 6 7 8 9 10 11 12
H L L L L L L H L H L H
b. is tt- tu mr- ri ba bu lan- na qus dl hin
The seeond syllable in (54)b does not violate NoOns-. Both skeletal slots
in that syllable are linked to a feature bundle t, but they are not linked to
the same occurrence of t. The onset is the second person prefix It-I while
the nucleus is the first half of the initial geminate of the kernel Ittul 'forget'.
Our second remark concerns the fact that as formulated in (48), constraint
Noöns- prevents the first half of a geminate from being an onset. Why
not instead prevent it from being syllable-initial? This alternative formu-
lation would predict that a geminate cannot occur as an onsetless H syllable

69 Line 151 in a song composed by ME, to appear. ' Have you already forgotten the one-
stringed violin and the tuning fork?' .
108 CHAPTER FOUR

at the beginning of a line, an incorrect prediction, witness the second line

in the following example."
(55) a. fulki s- lsbr" ma=)' ur i-g a-smun i=t-daIli-t
b. s-!sbr del-yaqin a=f 72 i-bna i-gmmi d=t-aka-t
(56) a. ful ki ss br ma )'U ri gas mu nit dal- lit
b. ss br dl ya qi na fib" nay gm- mid ta kat
In order for parse (56)b to be optimal the initial geminate must be a nucleus-
coda sequence, i.e. a syllable of type (28)d lacking an onset. If the geminate
were an onset-nucleus sequence it would violate constraint NoOns- and
the parse would not be an optimal one , since there exists another one
which does not violate the constraint.
Note that the word !ssbr also occurs in (55)a. This line has two lieit
parses. One is that of (56)a, which has a violation of NoOns- resulting from
the operation of rule DETACH. The other is optimal and is the input to
DETACH . In that parse the sequence kiss is one heavy syllable.

4.9 . ALTERNATIVE PARSES MEETING ALL THE CONSTRAINTS

This section deals with sequences for which there is more than one parse
meeting all the requirements presented until now. These sequences have
sonority contours which are even or have faIling slopes.
Let us first list all the constraints which have been called upon in the
preceding sections:
(57) a. Every skeletal slot belongs to one syllable and only one
b. Complex onsets are prohibited ((34)d)
c. Condition (26) on well-formed rimes
d. NoRR (No Rising Rimes) (42)
e. NoHiatus (25)
f. SonPeak (39)
g. NoOns- (48)
Let us assurne that these constraints are all undominated. We say that
a parse is '(57)-compliant' if it does not violate any of the constraints in

70 Lines 217 and 218 in the song cited in the preceding note. 'Patience is a fine virtue as
long as it does not go together with humility I Patience and faith are the foundations on
which houses and hearths stand' .
71 Underlyingly /l-Isbr/, The total assimilation of 11-1 to the following coronal gives rise
to a geminate, v. § 2.5.3 .1.
72 lad=fl in the underlying representation.
73 The metrical pattern required by the tune allows the seventh syllable of a line to be
either L or H.
 TASHLHIYT SYLLABLES I 109

(57). Setting aside the violations of NoOns- «57)g) due to rule DETACH
(49) (see last seetion), by and large alllieit parses are (57)-eompliant, 74
Certain segment sequenees only have one (57)-eompliant parse while
others have several. Examples of sequenees of the first type are (40)a,b,e,
(44) and (55)a,b . For sueh sequenees, if the (57)-eompliant parse does not
meet the eonditions of rule DETACH, it is the unique licit parse ; other-
wise there are several lieit parses, all derived through the applieation of
DETACH to the (57)-eompliant parse.
An example of a sequenee which has several parses which are (57)-
eompliant is (40)d. We repeat it here with its parse in (41):
(58) a l-mskin !attan l=l-nubb afefllak
(59) 1 2 3 4 5 6 7 8 9 10
LH LL LHL H LH
a lms ki nat- ta nll hub- baf- fl- lak
The parse in (59) does not violate any constraint in (57), but (58) has yet
another (57)-eompliant parse, whieh only differs from (59) in that /almski/
is syllabified as al.ms.ki (HLL) instead of a.lms.ki.
When a sequenee has several parses which are (57)-eompliant, whieh
among these are lieit parses? More researeh is needed before we ean answer
this question. All we will do below is to point out two regularities which
exclude eertain (57)-eompliant parses from the set of lieit parses , and to
show that in a sequenee with several (57)-eompliant parses, more than
one may be a lieit parse.

4.9.1. Sonority plateaux in complex obstruent rimes

Our first regularity has to do with complex rimes in whieh the nucleus is
an obstruent, whieh we eall 'eomplex obstruent rimes' for short. Examples
of eomp1ex obstruent rimes were given in (30) , (31) and (56)b . Let us
give one more examp1e. (60)a satisfies the same metriea1 pattern as (54) and
(56), as shown in (60)b, where the rime in the eighth syllab1e is a geminate
stop.
(60) a. w-aenna y-ugl-n a-zddig ggru-n=as=d i-zan"
b. 1 2 3 4 5 6 7 8 9 10 11 12
H L L L L L L H L H L H
wan- na yu gl na zd- di ggg ru nas di zan

74 'By and large' : see Chapter 7 on the violations of SonPeak and NoRR which are due
to underlying glides.
75 'Whoever carries flowers and is followed by flies' . This line was coined by ME. From
now, on lines for which no reference is given are lines invented by ME.
110 CHAPTER FOUR

If a complex obstruent rime is to comply with NoRR, its coda must be

an obstruent. The rimes in examples (30), (31), (56)b and (60) are either
fricative-stop sequences or geminates , and this is no accident since these
are the only complex obstruent rimes which are possible , as it turns out.
A rime cannot contain two stops or two fricatives. Consider for instance
the sequence in (61)a. The sequence /gdb/ after tut cannot form a syllable,
witness the fact that it cannot be made to occupy a H position in the metrical
pattern of (19). Parse (61)b is (57)-compliant and yet it is unacceptable:
(61) a. i-!sug=d bda t-i-sita-n d=l-mal i-hmma-n"
b. I 2 3 4 5 6 7 8 9 10 II 12
LLH LLLH LL LL H
su *gdb da ti si tan dl ma li hm- man
Only the other (57)-compliant parse is well-formed, the one in which
tugdbda/ is syllabified as ug.db .da . Here is a well-formed line with a parse
which is identical in all relevant respects (syllables 3 to 5).
(62) a. ahh i-zuged !brahim i-rwl=d ar=akWk Walla-n"
b. I 2 3 4 5 6 7 8 9 10 11 12
L LH LLLH LLL L H
an- ni zug db ra hi mir wl da rak't- kWal- lan
How are we to account for the limited range of permissible complex
obstruent rimes? One could adopt Shaw (1996)'s claim that complex
obstruent rimes are altogether prohibited by Universal Grammar and make
it into a violable constraint, call it NCOR (No Complex Obstruent Rimes) .
That constraint could be ranked below NoOns- and below SonPeak. Let
us examine each ranking in turn.
The ranking of NoOns- above NCOR would account for the well-formed-
ness of complex rimes which are comprised of an obstruent geminate , as
in (31)b and (60). Consider again sequence /banssfa/ in (31)b, which must
be parsed as ba.nss.fa , in violation of NCOR . ban .ss.fa complies with
NCOR, but it is excluded because it is not even (57)-compliant, since it
violates Noöns-.
Ranking SonPeak above NCOR would allow fricative-stop rimes such as
those in the syllabies tzd and tst in (30)a,b and in syllable tfk in (31)a: in
these lines the fricative nucleus is indeed a sonority peak. But this ranking
would fail to account for the fact that a fricative-stop rime can be well-
formed even if the fricative is not a sonority peak, as in the eighth syllable
of the following line, which satisfies the same metrical pattern as (60):

76 He has always led herds and great riches'.

•

77 • Alas!
Brahim has exiled hirnself here, he has fled here all in tears' . Note that in syllable
8 the glide w violates SonPeak. See Chapter 7 about such violations.
 TASHLHIYT SYLLABLES I 111

(63) a. w-assenna t-srufa t-a-x'rs-t sar ur hnna-n"

b. 1 2 3 4 5 6 7 8 9 10 11 12
H L L L L L L H L H L H
was- sn- na ts ru fa ta xWst sa rur hn- nan
At present we are unable to devise an analysis which would make use of
NCOR and at the same time allow parses such as (63)b. The appropriate
empirical generalization is that an obstruent nuc1eus cannot be followed
by a coda of equal sonority, unless both belong to the same geminate.
To see what this generalization implies, consider for instance the various
possibilities when the nuc1eus is a fricative. These are displayed below in
(64) , where sand f can be replaced at will by any other fricatives , and t
by any other stop.

(64) a. R b. R
~ r-;
X X X X
I
s
I
t
Vs
*c. R *d. R
~ ~
X X X X
I
s
I
f
I
t
I
s
Cases (64)a,b,c are all (57)-compliant but we must find a way to exc1ude
(64)c, which is ill-formed, as implied by the above generalization. Case
(64)d has been added for the sake of completeness; it is exc1uded by NoRR
(42). The above generalization allows (64)b but not (64)c. Let us posit
the constraint formulated below in (65) and assurne that only NoOns- is
ranked above that constraint.
(65) NoPICOR: 79
An obstruent nuc1eus cannot be followed by a coda of equal
sonority.
(63)b does not violate NoPICOR: the syllable xWst does indeed contain
adjacent obstruents of equal sonority, but these do not both occur inside
the rime .
Whereas the constraints NoRR and SonPeak only take into account
sonority differences between adjacent segments, NoPICOR furthermore
makes reference to a specific rung of the sonority scale, the obstruents.

78 "The unfortunate whose tooth gives hirn a rough time will never find relief' .
79 NoPICOR: No Plateaux In Complex Obstruent Rimes.
112 CHAPTER FOUR

Introducing NoPICOR, which is justified by the need to exclude sylla-

b1es such as *gdb in (61), has allowed us to assurne that voiced obstruents
and their voice1ess counterparts are of equal sonority. The sonority scale
which we invoked in our earlier work was more differentiated than that
given above in (35). Whereas the scale in (35) only divides the obstru-
ents into two classes, the fricatives and the stops, our previous work divided
them into four. These were, in an order of decreasing sonority: the voiced
fricatives, the voiceless fricatives, the voiced stops and the voice1ess stops."
Consider the word t-xzne-as ' she hoarded for him ' (3fs-hoard=dat3s), which
can only be parsed as t.xz.nas (LLH). In our earlier work, what made this
parse preferable to txz.nas (HH), was the fact that voiced z was more
sonorous than voiceless x ; as a result of this our step-by-step syllabifica-
tion procedure parsed z as a nucleus before its less sonorous neighbour x
could be entertained as a candidate to nucleushood. In the analysis presented
here, on the other hand, the sequence xz is a sonority plateau, and what
makes t.xz.nas preferable to txz.nas is that NoPICOR incurs a violation in
the latter parse but not in the former.
It is only at a late stage in the preparation of this book that we have come
to consider seriously the possibility that positing NoPICOR might render
superfluous our earlier sonority distinction between voiced and voiceless
obstruents. The deci sion of invoking the less differentiated sonority scale
in (35) has ramifications which have yet to be explored. The less differ-
entiated the sonority scale, the more sequences are sonority plateaux, and
constraints SonPeak and NoRR have nothing to say about sonority plateaux.
Clearly, more work is needed on the syllabification of sequences con-
taining only fricatives or only stops.

4.9.2. Sequences of high vowels

Our second regularity concerns certain sequences of high vocoids. It is illus-

trated by (66) below, which is the fourth line of the poem the beginning
of which was parsed in (19). That line has two (57)-compliant parses, but
only one of them is well-formed:
(66) a mass i-s-kufret !rbbi ur a=t=akWkWi-tt-annay"
1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a mas- sis ku fr trb- biw ra tak'r- kWit- tan- nay
In the scansion in (66) /trbbiu/ is syllabified as trb-ibiw. Another (57)-com-
pliant parse is trbb.yu , which is not acceptable. When two high vowels

80 See Elmedlaoui (1985), OE (1985, 1988).

81 'But the owner was a callous man; he did not deign look at hirn.'
 TASHLHlYT SYLLABLES I 113

are separated by a word boundary and one of them glides to avoid hiatus,
it must be the second." This regularity is valid for the eoIloquial language
as weIl as for poetry, as will be seen in § 7.1.3 .

4.9.3. Alternative licit parses not due to DETACH

Let us now turn to situations in which a sequenee has two (57)-eompliant

parses, both of them lieit. Sequenee Iyukrlka/ has two (57)-eompliant parses,
yu.krl.ka and yuk.rl.ka. These are both well-formed, witness the aeeeptability
of the lines below, whieh ean be sung to the tune of (19).83
(67) 1 2 3 4 5 6 7 8 9
10 11 12
L L H L L L H L L
L L H
a. a yu krl ka s"{ tuz- zum tn ma ta ha nut
b. bi hi yuk rl ka su ras gi sn ta ya fut
In Iyukrlkasl SonPeak requires a syIlable nueleus to be loeated on one of
the two segments in the sonority peak Ir1l, but the other eonstraints do not
prefer one loeation over the other.
On the basis of the limited data given in DE (1997a), Billerey (1999)
eonjeetured that alternate lieit parses exist only when the alternate nucIei
have the same degree of sonority, as is the ease in (67) above, where there
is no sonority differenee between Ir/, whieh is a nucleus in (67)a, and 11/,
whieh is a nucIeus in (67)b . However, we have found strings whieh aIlow
alternate parses although the alternate nucIei do not form a sonority plateau.
Sequenee /taxrstsa/ in (63) is a ease in point. It is parsed as ta.x'st.sa in
(63)b, and its other (57)-eompliant parse , viz tax'.st.sa, is also well-formed,
see the syIlables 8-10 in (68)b below : (68)a is aeeeptable as a line with
the same metrieal pattern as (63)b, and (68)b is the parse of (68)a whieh
satisfies that pattern. Parses ta.x''st.sa and tax r.st .sa are both licit, despite
the fact that Ist! is not a sonority plateau .
(68) a. ayhayya w-aenna t-nva t-a-xrs-t sul huwl-n"
b. I 2 3 4 5 6 7 8 9 10 11 12
HL LL LLLH LHL H
ay hay- ya wan- na tn "{a tax" st sul hu- wln
In laxWstsa/ the sonority peaks (the vowels) are separated by a sequenee with
falling sonority, about whieh eonstraints SonPeak and NoRR have nothing

82 See also syllable 10 in (33)b.

83 (a) a y-ukr l-kas -y=t-!uzzurn-t nrn a t-a-hanu-t, 'he stole the glass frorn the back of the
shop' : (b) bihi y-ukr l-kas ur=as=gisn t-a-yafu-t, ' Bihi stole the glass ;' it is of no use to
hirn'.
84 'Ouch! the one whose tooth aches and who, on top of that, is in trouble' .
114 CH APT ER FOUR

to say. Another example is sequence /almski/, which is parsed as a.lms.ki

in (59) . The parse al.ms.ki is also well-formed," witness the acceptability
of the following line, which can be sung to the tune of (19).86
(69) 1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a ym- mal ms ki nu rakwk w yu fi ta gl- lat
More empirical work is needed on syllabification in expressions which
contain sonority plateaux and sequences with falling sonority ramps .

4.10. SUMMARY

We recapitulate the devices which have been called upon in this chapter.
(70) a. Every skeletal slot belongs to one syllable and only one
b. Complex onsets are prohibited ((34)d)
c. Condition (26) on well-formed rimes
d. NoRR (No Rising Rimes) (42)
e. NoHiatus (25)
f. SonPeak (39)---=:=J
g. NoOns- (48)
h. NoPICOR (65)
i. DETACH (49)
The last item in (70) is a phonological rule and all the others are con -
straints. The rule, which is optional, operates on the output of the constraints,
see the box diagram in (52). Except for NoPICOR, which is dominated
by NoOns-, all the constraints are undominated.

85 This corrects DE (1997a: 46), who stated incorrectly that al.ms.ki is unacceptable .
86 a ymma I-mskin ure ak"k" y-ufi t-a-glla-t 'Ah ! while the POOf man does not even find a
tum ip' .
 CHAPTER FIVE

TASHLHIYT SYLLABLES II

The last chapter dealt with syllabification in word sequences sung to a

tune. In this chapter we discuss two morphologically-governed alterna-
tions which are evidence that our analysis of syllable structure in Tashlhiyt
is also valid inside sterns: imperfective gemination and length alternations
in the causative prefix . Before we do this, however, let us consider briefly
syllabification in word sequences outside of poetry.

5.1. THE SYLLABIFICATION OF WORD SEQUENCES OUTSIDE OF POETRY

In the preceding chapter all our evidence about syllabification in Tashlhiyt

Berber was drawn from versification. The evidence on which we relied in
Elmedlaoui (1985) and in DE (1985) was of a different kind : ME uttered
various expressions (words or word sequences) as he would in normal
speech, and for each expression we noted (A) how many syllables ME
felt the expression contained and (B) where he feit the syllabic peaks were
located. ME was for instance asked questions A and B about the expres-
sions /rad t-ksm/ 'she will enter' and /i-!sbukd/ 'he poked (sorneone) in
the eye', which are pronounced ratksm and !isbukd. He answered that he
felt that ratksm was comprised of three syllables whose nuclei were a, k
and m; and that !isbukd was comprised of two, whose nuclei were i and
u. From these answers we inferred the parses rf!.tk.sm and [s.bgkd.' Let
us use the phrase 'inferred from direct questioning' (IFDQ) to distinguish
parses like these from those involved in orthometric syllabification, i.e.
syllabification in poetry.
As noted in DE (1988), orthometric syllabification and IFDQ syllabifi-
cation differ in certain respects . One systematic difference involves obstruent
nuclei, which never occur next to a pause in IFDQ syllabic parses . Whereas
the already cited !isbukd is feit to contain three syllables in li-sbukd baba=s
'he poked his father's eye out' (is.bu.kd.ba.bas) it is felt to contain only two
syllables when uttered in isolation (is.bukd). Another example is the initial
consonant in the stern /skr/, which is felt to be a nucleus in t-skr 'she did'
(IFDQ ts.kr.) but not in skr 'do!'. The latter form is felt to contain only
one syllable (lFDQ .skr.), The IFDQ parse and the orthometric parse are
identical for t-skr, in which the stem-initial /s/ is not adjacent to a pause ,

I See DE (1985: 120). We say that the parses were 'inferred' , rather than simp1y reeorded,
beeause no questions were asked about the loeation of syllable boundaries .

115
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
116 CHAPTER FIVE

while they are different for skr, which can count as two syllables at the
beginning of a line in singing (s.kr).
The analysis proposed in our works of 1985 and 1988 proceeded in
two stages. During the first stage syllable structure was buiIt in a stepwise
fashion over the input string by a simple procedure called Core Syllabifica-
tion. Like the constraint-based analysis developed in the last chapter, Core
Syllabification yielded parses which are those needed for versification.
During the second stage the syllable structure built by Core Syllabification
was readjusted in various ways to fit the IFDQ parses. Take again skr
'do!'. Core Syllabification would first parse /skr/ as dissyllabic ~.kr. In
a later stage of the derivation, in case the form occurred immediately after
a pause, it would be readjusted to monosyllabic skr, with a complex
onset.
We do not make use of IFDQ syllabification in this book because we
think singing is a more reliable source of data, as explained in § 4.1. What,
then, about the syllable structure of Tashlhiyt Berber outside of poetry?
It is a common occurrence across cultures for singing to make use of
pronunciations and of syllabic parses which are not acceptable in the spoken
language. Take for instance the variety of French spoken in Paris. etudiez
'study !' can only have three syllables in the spoken language ([e.tü.dje])
while in singing it can have three or four ([e.tü .di.je]), depending on what
the tune requires; tune 'moon' can only be pronounced with a final con-
sonant in the spoken language ([lün]) while it can take a final vowel in
singing (Ilü.neej). For Imdlawn Tashlhiyt Berber, we have already mentioned
the existence of variant pronunciations which are acceptable only in singing;'
and one cannot rule out the possibility that certain syllabic parses could
be avaiIable in singing but not in the spoken 1anguage, or the other way
around.' We must leave this question open for further research . Whatever
the discrepancies between them, however, syllabification in singing and that
in the spoken language cannot be very different. They are rooted in the same
components of the native speakers' grammars, or at least in components
which overlap to a considerab1e extent. While native speakers of French
sing the French word trois 'three' as one syllable, those of Tashlhiyt Berber
sing it as two (tr.wa) . Clearly, such divergences reflect differences between
the sound patterns of the two languages, not just different conventions for
setting text to music .
Up to this point we have only considered syllabification in word

2 See § 4.6.
3 We have a hunch that only singing allows violations of constraint NoOns-, which forbids
the first half of a geminate to be an onset. Such violations are never found in IFDQ parses .
Jebbour (\999: I09ff) argue s that NoOn s- is never violated in the syllabic parse s which
are relevant for the templatic morphology of Tashlhiyt.
 TASHLHIYT SYLLABLES 11 117

sequences. We now turn to two phenomena which shed light on syllabifi-

cation in units smaller than words, viz in sterns.

5.2. IMPERFECTIVE GEMINATION: THE BASIC GENERALIZATION

Certain verbs of Tashlhiyt form their irnperfective stern by geminating one

consonant in their basic stern. Before we discuss the details of imperfec-
tive gemination, let us set the scene by explaining in broad outline how
irnperfective sterns are formed." Some aspects of this outline will become
relevant later in this chapter, when we discuss Jebbour's claims about
syllable weight.
Details aside, three processes are resorted to in the formation of imper-
fective sterns:
(1) A i. the gemination of a consonant;
ii. the prefixation of tt-;
B. the insertion of a 'chameleon vowel'.
Which of these processes operate for a given verb is to a great extent
determined by the phonological make-up of the basic stern, i.e. the per-
fective stem.' These processes may operate jointly, except that gemination
and the prefix tt- are incompatible, which is why in (1) they have been
grouped to form the supercategory A. The examples in the table (2) below
cover some common stern-types. In (2) the first two columns each corre-
spond to one branch in (1) . The plus and the minus signs indicate whether
the processes just mentioned are involved in the formation of the imper-
fective stern.
(2) A B pf impf
a. + lkm lkkm 'reach'
b. + zla zllu 'lose'
c. + Isa lssa 'wear'
d. + ufa tt-afa 'find'
e. + kkil tt-kkil 'curdle'
f. + + skkiws tt-skkiwis 'sit'
g. + + fruri tt-fruruy 'crumble'
h. + + dl ddal 'cover'
i. + + xtrn tt-xtam 'complete'
J. + + fssr tt-fssar 'explain'
k. - + ssumm ssurnum 'suck'
1. + sni snay 'sign'
m. - + ss-lkm ss-lkam 'cause to reach' (cf. a)

4 See DE (1991: 84-96) for a detailed survey of irnperfective stern formation .

5 See below on the choice of the basic stern.
118 CHAPTER FIVE

(2)f and all the verbs below it insert a chameleon vowel in the imperfec-
tive . The chameleon vowel is inserted before the last segment of the
basic stern. It is a copy of the preceding vowel if there is one; otherwise
it is a.
Our argument in this section is concerned with the verbs which simply
geminate one consonant in the imperfective, see (2)a,b,c. 6 In the examples
below we give for each verb (I) its perfective stern, (II) its imperfective stern
and (III) the syllabic parse of its perfective stern, which will become relevant
below.'
(3) I II III I II III
pf impf syll pf impf syll
of pf of pf
krz kkrz .krz, rks rkks r.ks
nlb nnlb .nIb. !Ibz !lbbz I.bi.
!zlm !zzlm .zIrn. vml vmml x·ml
mrz mmrz .mrz. rsq rssq r.SQ
xng xxng .xng . zbd zbbd i .bg
frn ffrn .fm, xsi xssi x.si
The verbs on the left-hand side of (3) geminate their first consonant in
the imperfective while those on the right-hand side geminate their second
consonant. The boldface letters in the perfective sterns (I) will be explained
later.
In all the verbs which form their imperfective sterns simply by gemi-
nating one consonant in the basic stern, the following two conditions are
mer:S
(4) (a) the basic stern contains three segments none of which is a
geminate;
(b) if the basic stern contains a vowel , that vowel must be the
last segment.
Let us refer to the verbs which have these properties as the 'gerninable
verbs'. Not all geminable verbs resort to gemination in the imperfective, see
e.g. xtm «2)i) and sni «2)1), which meet the above conditions. Let us say
that a geminable verb which does resort to imperfective gemination is a
'geminating verb' . The distribution of the geminating verbs among the

6 The ablaut alu in (2)b also occurs in verbs which do not use gemination in the imperfective
(see OE (1991) . It is not directly relevant in this discussion, which is only concerned with
alternations in which syllable structure is involved .
7 Except for xsi, which means 'go out (fire)', the meanings of all the verbs in (3) are given
in Appendix V at the end of this book.
8 See OE (1991: 85).
 TASHLHlYT SYLLABLES II 119

geminable verbs seems to be a matter of lexical idiosyncracy." On the

other hand the segment which undergoes gemination is predictable. It is
either the first or the second, never the third. One can state the following
generalization:
(5) The segment which is geminated in the imperfective stern is
that segment which is an onset in the basic stern.
In the verbs on the left-hand side of (3) it is the first consonant which is
an onset in the perfective stern, while in those on the right-hand side it is
the second consonant. It is easy to see that the syllabic parses in (3) are
indeed the only ones compatible with the constraints proposed in the pre-
ceding chapter. The sonority peaks in the perfective sterns in (3) are in
boldface . Syllabic parses with nuclei other than those in (3) would violate
constraint SonPeak.
Generalization (5) reflects a property which is specific, at least in part,
to the processes responsible for the formation of imperfective sterns. It does
not simply follow from generallimitations on the distribution of geminate
segments in all sterns, witness the existence of basic sterns such as mmzr
'scatter', nnvl 'pour out', sstl 'weigh in one' s hand' ,frrd ' give the change',
!znng 'be congested', drrm 'trample underfoot' .
We have found only four exceptions to generalization (5) among all
the geminating verbs of Imdlawn Tashlhiyt known to us." Commenting
on DE (1988), the article in which generalization (5) was first proposed,
Shaw (1996) notes that geminating verbs like those on the left in (3),
which geminate their first consonant in the imperfective, all have a sonorant
as their second segment. We have since found one with an obstruent in
second position: fsd 'be spoiled'." That we have not found more is not
too surprising, for there are not many CCC verbs with only obstruents.
Imperfective gemination is strong evidence in favor of the thesis of
Sonority-Driven Syllabification ." As is implied in the preceding discussion,

9 Appendix V at the end of this book lists all the geminating verbs ending in a consonant
which we have found in Imdlawn Tashlhiyt.
10 They are l vra 'read' , yWma 'co at', yli 'go up' and bzg ' swell' , whose imperfective
sterns are not l vrra , yWmma, vlli and bbzg, as predicted by (5), but !aqqra, aqqrma , aqqlay
and azzg, Besides being exceptions to (5), the first three forms have an idiosyncratic initial
a. The third has a further irregularity: although gemination and vowel insertion are both
regularly used to form imperfective sterns, as a rule both processes do not cooccur in the same
stern if it is triliteral, v. DE (1991: 84ff) . Besides bzg; there are a number of other gemi-
nating verbs in which an initial labial is replaced by a in the imperfective.
11 Like a number of other geminable verbs, fs d has a geminating form and a nongemi -
nating one in free variation in the imperfe ctive (ffsd / tt-fsad) . tt-CCaC is arguably the
unmarked case in the formation of imperfective sterns for the CCC verbs. This case is exem-
plified in (2)i.
12 On the thesis of Sonority-Driven Syllabification, see § 4.1.
120 CHAPT ER FIV E

the units whose syllabic structure is relevant for imperfective gemination

are bare sterns; inflectional affixes are irrelevant. /kI is an on set in the
bare perfective stern /rks/ 'hide' , for instance (r.ks) , while it is not in
/rks-n/ 'they hid' (jk.sn), but the imperfective stern of /rks/ is /rkks/ regard-
less of the surrounding morphemes: ar t-rkks ' she is hiding', ar rkks-n 'they
are hiding'. Imperfective gemination indicates that the same constraints
which shape syllable structure in word sequences also operate in bare
sterns .
In view of recent attempts to characterize 'cyclic effects' in terms of
relations between full words,!' one may ask whether the generalization
(5) could be reformulated so as to capture a relation between actual word
forms in the conjugational paradigm of Tashlhiyt, rather than between sterns.
Can the imperfective stern and the basic stern referred to in (5) stand on
their own as words?
In the conjugation of Tashlhiyt verbs the only sterns which can occur
without any overt inflectional affixes are the imperfective stern and the aorist
stern. These are used as imperative 2s forms . The bare imperfective stern
is used in the imperfective imperative, while the bare aorist stern is used
in the perfective imperative, see DE (1991: 178-179).
Can the 'basic stern' referred to in (4) and (5) be equated with the aorist
stern? We are unable to give this question adefinite answer, for such an
answer would presuppose an overall analysis of the verbal morphology of
Tashlhiyt, but we can at least give an idea of the kind of facts involved.
Knowledge of these facts is in any case necessary in order to see the exact
empirical import of generalization (5).
The table in (6) below exemplifies the various types of verbs which resort
to gemination in the imperfective."
(6) pf aor impf
a. lkm lkm lkkm 'reach'
b. zla zlu zllu 'lose'
c. lsa ls lssa 'wear'
d. mla ml mmal ' show'
e. dl dl ddal 'cover'
Table (6) illustrates several regularities. First, the verbs with shapes CCC
and CCi have homophonous sterns in the perfective and in the aorist; either
stern can be used as a starting point for deriving the imperfective stern
(line a). Second, the vowel of most CCa verbs becomes u in the aorist
and in the imperfective (line b), which would suggest that it is the aorist

13 See Kenstowicz (1996), Kiparsky (2000) and references therein.

14 The verbs in lines a, band e also appear in (2)a,b,h.
 TASHLHIYT SYLLABLES II 121

stern which should be taken as basic, as is in fact done in most discus-

sions of Berber verbs . Third, a sizeable minority of CCa verbs have CC
aorist sterns (lines c and d). For most of these the shape of the imperfec-
tive stern is CC:a (line c), while it is C:aC for the others (line d), Chosing
the perfective stern as a starting point for deriving the imperfective stern
allows us to include in our discussion the class of verbs exemplified in (6)c.
If the aori st stern were used instead as the basic stern, various details
would have to be changed in our formulations, but the changes would not
impinge on the main point of this section: the generalization in (5) is
evidence that the account of syllable structure which we set forth in the
last chapter is also valid for sterns.
The support which the facts of imperfective gemination lend to our
analysis of syllable structure in Tashlhiyt, and in particular to the Sonority-
Driven Syllabification thesis, is reinforced when Tashlhiyt is compared with
other dialects of Berber. Consider for instance Ath Sidhar Rifian. 15 In
addition to the underlying vowels la, i, u/, Ath Sidhar Rifian has an
epenthetic vowel @ which is inserted to syllabify consonant clusters, and
the sonority of consonants only plays a marginal role in syllabification in
that dialect. ICCCI verbs are also very common in the lexicon of Ath Sidhar
Rifian. The appendix in Tangi (1991: 313-337) lists close to 600 verbs.
One-fifth of these are ICCCI verbs whose imperfective stern only differs
from the perfective stern by the gemination of a consonant. In all such verbs
it is the second consonant which geminates in the imperfective, regard-
less of the sonority contour of the underlying ICCCI sequence. The
imperfective stern of sm@z 'scratch' is s@mm@ z and that of ! ms@o'comb'
is !m@ [email protected] When pronounced in isolation, e.g. as 2s imperative forms,
all the ICCCI sterns are pronounced as CC@C, with the middle C acting
as an onset. Generalization (5) is also valid in Ath Sidhar Rifian, then.
The same state of affairs prevails in the dialect of Figuig (Kossmann
1994, Saa 1995), in Kabyle (Basset and Picard 1948, Bendjaballah 1995)
and in Tamazight (Abdel-Massih 1968, Penchoen 1973). All these dialects
are on record with syllable structures resembling that of Ath Sidhar Rifian,
and in all the middle consonant is the only one which can geminate to
form imperfective sterns for ICCCI verbs. 17

15 On Ath Sidhar Rifian, see § 6.5.

16 The cognate forms in Tahshliyt are krm: (impf kWkWmz) and lmsd (impf !mssd).
17 According to Basset (1929: 155), among the Berber dialects for which data were avail-
able at the time, Tashlhiyt was the only one regularly to geminate the first consonant in the
imperfective of certain categories of ICCCI verbs. Basset's text also implies that gemina-
tion of the first C in the imperfcctive of ICCCI verbs also occurs, although less regularly,
in the Ntifa dialect, in Central Morocco.
122 CHAPTER FIVE

5.3. IMPERFECTIVE GEMINATION AND THE WEIGHT OF HOLLOW SYLLABLES

From now on we use the expression 'hollow syllable' to refer to any syllable
whose nucleus is not a, i or u. kz and kr; are hollow syllabIes, while ka
and kin are not.
While the views of Jebbour (1999) about the syllable structure of
Tashlhiyt are otherwise in accord with ours, this author disputes our con-
ception of syllable weight in that language. Jebbour claims that hollow
syllables are all light. According to hirn, kr: is a light syllable, on a par with
kz and ka. 18
The author gives two arguments in favor of his position. One argument
concerns imperfective gemination and the other, length alternations in the
causative prefix. These arguments lose much of their appeal under close
scrutiny, as we will now try to show.
We stated earlier that if a verb containing a vowel is to undergo imper-
fective gemination, the vowel must be its last segment, see (4)b. While some
verbs with the shapes eee and eev undergo gemination in the imper-
fective, verbs with the shapes vee, eve and vev never do. This state
of affairs is summarized below in table (7), where the canonical forms
just mentioned are all instantiated. The perfective sterns with shape eee
are instantiated on two lines, one for monosyllabic ece and one for dis-
syllabic C.eh.
(7) pf impf
a. ccc. krz kkrz 'plough'
a' . cyc. nuz *nnuz (tt-huz) 'corner'
b. ccc rks rkks 'hide'
b'. xcc udr *uddr (tt-adr) 'pin down'
c. cey kla klla 'spend the day'
c' . yey uli *ulli (tt-ali) 'cluster'
The column for imperfective sterns displays the form which results from
geminating the onset consonant in the perfective stern. When the gemi-
nated form is ungrammatical, the attested form is indicated in parentheses.
Jebbour seeks an analysis which would not only predict which consonant
geminates in the imperfective in (7)a-c, but would also account for the
fact that gemination is an option in lines a, band c, but not in a', b' and
c' . According to hirn, the reason why gemination is licit only with sterns
in lines a, band c, is that the grammar of Tashlhiyt imposes a certain con-
straint on the shape of imperfective kerneIs, which geminated forms in lines
a', b' and c' fail to meet. He proposes that imperfective gemination is subject
to the following restriction:

18 Jebbour 's article restates the central claim of Jebbour (1996).

 TASHLHlYT SYLLABL ES II 123

(8) Geminate the onset in the basic stern if the resulting form is
LL (a sequence of two light syllables).
In (9) below we give again the content of the 'pf' and 'impf' columns in
table (7). Each geminated form is syllabified and is followed by the syllable
weight which Jebbour assigns in accordance with his claim that in Tashlhiyt
all vowelless syllables are light.
(9) pf impf
a. krz k.krz L L
a'. huz h.huz * L H
b. rks rk.ks L L
b' . udr ud.dr * H L
c. kla kl.la L L
c' , uli ul.li * H L
The LL requirement on the output of gemination is met in lines a, band
c, but not in lines a' , b' and c', and consequently gemination is possible
in the former but not in the latter, as indicated by the asterisks. Note in
particular the contrast between lines a and a', which is consistent with
Jebbour's claim that CCC syllables are light.
At first one may take the LL requirement in (8) to be a constraint on
imperfective kernels in general, and we believe that therein lies much of
its attractiveness. In fact , what is the scope of the LL requirement? Due
to chameleon insertion, among other thing s, imperfective sterns ending in
CVC syllables are commonplace in Tashlhiyt, see line fand all the others
below it in (2). Consequently the LL requirement cannot be construed as
a constraint on imperfective kernels in general.
It cannot even be construed as a constraint on the imperfective kemels
in which gemination operates, for Tashlhiyt has imperfective sterns such
as ddal, from dl «2)h). Until now the discussion has focussed on the three-
segment verbs, but these are not the only ones which can use gemination
to form their imperfective sterns. Most of the two-segment verbs have imper-
fective sterns of the form C:aC. Examples are given below in (10) .
(10) pf impf
a. !dr !ttar 'fall'
b. fl ffal 'leave behind'
c. gn ggan 'sleep'
d. I'wi qWqW ay ' seize '
e. !di !ttay 'take out'
The upshot is that Jebbour's LL requirement is concerned only with the
output of a specific process (gemination), and that its jurisdiction is limited
to the three-segment verbs . The only property of the class of geminable
124 CHAPTER FIVE

verbs which is explained by Jebbour's LL constraint is the fact stated earlier

in (4)b: if a vowel occurs in a (three-segment) geminable verb, that vowel
must be the last segment. The LL requirement is indeed an improvement
over merely stating the fact in question as we did in (4)b, but this improve-
ment is offset by the conundrum which we are then faced with: we cannot
simply discount the fact that versification systematically treats hollow syl-
lables with a coda as heavy. Further evidence is needed before we would
feeI compelled to accept that Tashlhiyt morphology and versification use
different representations to compute syllable weight. However Jebbour's
second argument is even less compelling than the first, as we will see now."

5.4. LENGTH ALTERNATIONS IN THE CAUSATIVE PREFIX

To show that hollow syllables with codas are light, Jebbour (1999) also uses
data pertaining to the length of the causative prefix. In this section we
build on Jebbour's work and present our own characterization of length
alternations in the causative prefix, which is preferable for conceptual
reasons as weIl as for empirical ones. That characterization is compatible
with our claim that in Tashlhiyt the weight of syllables does not depend
on the feature content of their nuclei. The length alternations in the causative
prefix are our second item of evidence which shows that the structure of
syllables inside sterns is the same as that in word sequences .
The morpheme /s-/ is prefixed to verbs to derive causative verbs. Here
are some examples."
(11) a. rgigi 'tremble' s-rgigi 'cause to tremble'
b. nda 'be chumed' ss-nda 'churn'
c. nza 'be sold' zz-nza 'seIl'
d. 'lis 'survive' ss-'lis 'cause to survive'
e. nzm 'be unharmed' zz-nzm 'rescue'
In what folIows, we use the term 'base' to refer to the verb from which a
causative verb is derived by prefixation of /s-/.
Causative /s-/ shows two types of alternations which are independent

19 A questionable aspect of Jebbour 's proposal should be noted in passing, which has to
do with the representation of geminates . Formulated within Hayes's (1989) version of moraic
theory, which is the framework he adopts, Jebbour's basic claim is that syllables which
have a consonant in their nucleus are all comprised of a single mora. It is difficult to see
how this claim can be reconciled with the existence of tautosyllabic geminates in vowel-
less syllabies, e.g. !gzz 'crunch! ' (ggz), dl=tt 'cover her!' (dItt). dl=tt and similar cases
are especially worrysome . Since the author explicitly prohibits branching codas in vowel-
less syllables (note 5 p. 98), sequences such as /dl=tt/ do not have any licit parse in his
analysis .
20 DE (1991: 96-99) present an overview of the causative verbs of Imdlawn Tashlhiyt
and show how their conjugation relates to that of the other verbs.
 TASHLHlYT SYLLABLES II 125

of one another. On the one hand, it is subject to sibilant harrnony, a cover

term for two independent processes of assimilation-at-a-distance: when
the base contains a fricative coronal the causative prefix must agree with
it in voicing and in anteriority; compare for instance ss-nda, zz-nza,
ss- 'liS and zz-nzm
in (11) above. Sibilant harmony has no lexical excep-
tions in Imdlawn Tashlhiyt."
Besides harmonizing with a sibilant in the base, the causative prefix is
also subject to quantity alternations. It is realized as a simplex consonant
in some verbs and as a geminate in others, and these alternations are the
subject of the present section.

5.4.1. Monosyllabic bases beginning with an onset

Setting aside various lexical idiosyncrasies, the length of the causative prefix
is predictable from the canonical form of the base to which it is attached.
The facts which are central in Jebbour's argument are those about causatives
derived from monosyllabic bases in which the initial segment is an onset.
The author's factual claims about such verbs are summarized in (12).
(12) Jebbour's generalizations about onset-initial monosyllabic bases
base pfx example
a. C{;. long !ss-dr 'cause to fall'
b. CV long ss-ni 'cause to ride'
c. C{;.C long ss-frs 'sharpen'
d. CVC short s-mun 'gather'
According to (12), .CCc. bases call for the same variant of the causative
prefix as CV and CC bases, which are light syllables; CVC bases call for
a different variant of the causative prefix.f Considering .CCC. syllables
as unimoraic allows Jebbour to subsume all the facts of table (12) under
a single generalization: the causative prefix is long before unimoraic bases,
and short before multimoraic bases.
We wish to take issue with the factual basis of the generalizations
summarized in (12). The table only covers part of the data. Except for the
.CCc. bases, which always call for a geminate prefix, as claimed by table
(12), the other bases are found both with a simplex prefix and with a
geminate one. Pending a detailed study, our preliminary explorations suggest

2\ See Elmedlaoui (\ 995a : 11-42) for a detailed discussion illustrated with abundant
examples .
22 Since we are only taking into consideration base verbs which begin with an onset, the
notations .Ce. and .CCC. unambiguously refer to bases in which the second consonant is a
nucleus. To avoid c1uttering, we will often dispense with the underscores indicating nuclear
consonants. Anyway, Tashlhiyt has no verb whose perfective stern is an onsetless hollow
syllable, i.e. a syllable .ke. or a syllable .k Ce. whose final CC is a geminate.
126 CHAPTER FrVE

the following broad outline for the causative verbs derived from onset-initial
monosyllabic bases.
Some such verbs allow both variants of the prefix in free variation, e.g.
(s)s-yal 'cause to imagine', !(s)s-run 'cause to reach', while others allow
only one. For instance the verbs in (13) only allow the geminate variant."
(13) a. ss-Iil 'rinse' lil 'be rinsed'
b. lzz-riz 'pollute' !riz 'be polluted'
c. ss-vab 'cause to disappear' )'ab 'disappear'
d. ss-tub 'demand repentance' tub 'repent'
e. ss-hul 'cause to worry' hul 'be worried'
f. ss-dux 'cause to lose one's head' dux 'lose one's h'
Note that the causative verbs in (13) are all counter-examples to general-
ization (12)d, according to which CVC bases call for a simplex prefix;
that generalization is crucial for Jebbour's claim that CVC and CCC syl-
lables have different weights.
In the causative verbs with onset-initial monosyllabic bases and no free
variation in the length of the prefix, the geminate variant is found as a
rule in verbs whose syntactic and semantic relationship with the base is
more regular, while the simplex variant tends to occur in verbs whose
relationship with the base is more idiosyncratic. Some bases give rise to two
different causative verbs, one with either variant. Here are examples.
(14) !dr 'fall' !ss-dr 'cause to fall'
!s-dr 'lay (egg)'
faw 'be clear' ss-faw 'light up'
s-faw 'be able to see'
gn 'sleep, go to bed' ss-gn 'put to sleep, to bed'
s-gn 'lay on its side (e.g. a tree)'
dus 'be tough' ss-dus 'make strong'24
s-dus 'fortify (e.g. a town)'
'lum 'swim' (sjs-Yum 'make swim'
s-Yum 'flood'
The existence of free variation in some verbs, which was mentioned
above, together with doublets like those in (14), suggest that we are dealing
with an area of the morphology which is at present in astate of flux, and
that in the new order which is emerging all onset-initial monosyllabic
bases behave alike, regardless of the content of their rime, and call for

23 (l3)e and (13)f are not to be confused with their free variants s-huwl and s-duwx, which
have a short prefix, as is to be expected for reasons discussed later.
24 As in l-lidid a y-ss-dus-n l-bni 'it is the iron which makes the building strong' (I-iron
AD prt-cau-tough-prt l-building) .
 TASHLHlYT SYLLABLES II 127

the geminate variant of the prefix. This is the assumption which we will
make below when we propose a single generalization eovering all eausative
verbs, no matter the shape of their bases. We will assurne that all eausative
verbs with a short prefix before an onset-initial monosyllabic base are lexical
idiosyncrasies.
Among the onset-initial monosyllabie bases , the .CCc. bases are the only
ones never to oecur with a simplex prefix. This fact has no synchronic
explanation, but its historical antecedents are easy to surmise: the change
now in progress is instating the geminate as the regular variant of the
prefix before base shapes which used to eall for the simplex variant. If
we assurne that .CCc. bases already called for the geminate variant in earlier
stages of Tashlhiyt, the ongoing change leaves unaffected the causative verbs
with .CCc. bases.

5.4.2. Other bases

Let us now turn to the causative verbs in which the base is polysyllabie
or does not begin with an onset. We will see that Jebbour's generaliza-
tions ean be improved upon without adopting his views about the weight
of hollow syllables with codas .
Bases whose first segment is a coronal fricative are excluded from the
upcoming discussion. As a rule the causative prefix does not undergo gem-
ination before sibilant-initial bases . Consider for instance the sibilant-initial
bases with the shape CCV. CCV bases normally call for a geminate prefix ,
e.g. ss-fta 'cause to walk ', ss-rsa 'cause to decay' , and yet the causative
verbs derived from sda 'lean on' or swa 'be good ', for instance, are
s-sda and s-Swa, not *ss-sda and *ss-swa. When the causative prefix imme-
diately precedes a sibilant, the proces s of sibilant harmony mentioned earlier
partially merges its feature-geometric tree with that of the following sibilant.
The phonological objects represented by ' sss' and 'sss' at the beginning
of the ill-formed ss-sda and ss-swa would violate constraint NO-TREBLE,
which prohibits melodic units associated with three X-positions in a roW. 25
The interaction between sibilant harmony and NO-TREBLE in causative
prefixes is discussed in detail in DE (1996a: 381-385).
In (15) below we give the complete array of generalizations set forth
in Jebbour (1999) to cover all the causative verb s. These generalizations
have been rearranged and rephrased slightly for the sake of this discus-
sion, but they have the same empirical import as the original ones. Examples
of the various cases in (15) can be found below, when we give data in
support of our own characterization of the facts.

2S NO-TREBLE was introduced in § 3.2.1.1.

128 CHAPTER FIVE

(15) i. if (a) the base contains a geminate or (b) if it is trisyllabic,

the prefix is short;
ii. otherwise, if the first segment of the base is not an onset,
the prefix is long;
lll. otherwise, if the base is (a) dissyllabic or (b) CVC, the prefix
is short;
iv. otherwise, the prefix is long.
Case iii-b of (15) is the same thing as the generalization stated earlier in
(12)d; case iv of (15) is the same thing as the conjunction of a, band c
in (12).
One insight expressed in (15) is that the syllable count is relevant in
determining the quantity of the prefix. This insight is correct, in our view,
and we shall argue that if it is suitably formulated, the other contextual
factors invoked in (15) become superfluous .
We submit that (16) below must be preferred to (15) as a characteriza-
tion of the quantity altemations in the causative prefix. The idea is to choose
the geminate variant only when a causative stern containing the short variant
would contain less than three syllables :
(16) CausLength:
The length of the prefix only depends on the number of
syllables in the causative stern taken as a whole:
a. if prefixing the simplex variant to the base yields a string
which contains less than three syllabies, the prefix is
geminate ;
b. otherwise, the prefix is simplex .
The causative prefix is a simplex consonant in the underlying representa-
tions, and geminates only under the circumstances described in (16)a. To
illustrate: prefixing s- to aywul yields say. wul , which only contains two
syllables; hence by (16)a, the causative of aywul ' cause to be tall' is
ss-aywul. Similarly, since prefixing s- to frs yields s.frs, with only two
syllables, the causative of frs 'be sharp' is ss-frs. On the other hand, pre-
fixing s- to liada results in a trisyllabic string (s.ha.da); consequently, by
(16)b, the causative of hada 'be next to ' is s-liada. In case (16)a, pre-
fixing the geminate variant achieves trisyllabicity in the causative stern in
some instances , e.g. s.say.wul, but not in others, e.g. ss.frs.
The differences between (16) and (15) are conceptual as well as empir-
ical, and (16) is superior on either count, as we shall now argue. Consider
first the empirical differences.
Setting aside the onset-initial monosyllabic bases, which have already
been dwelt upon, a careful comparison of the two proposals reveals that
the only case for which they make different predictions is when the base
contains a geminate. According to Jebbour's proposal , all causative verbs
with a geminate in the base should begin with a short prefix, see (15)i-a.
 TASHLHIYT SYLLABLES II 129

On our proposal, on the other hand, bases containing a geminate should

not behave differently from the others in their selection of the prefix. Like
the others, they should select the long prefix if selecting the short prefix
does not yield a causative stern which is long enough . This prediction is
borne out by the data. The examples in (17) below all have a long prefix
and a geminate in the base, contrary to Jebbour's claim (15)i-a. In (17)
the transcription of each attested stern (I) is followed by its syllabic parse
(Il) and by that of the ill-formed stern which would result from choosing
the short variant of the prefix (III).
(17) I Il III
parse of I parse with
short pfx
ss-attuy s.sat.tuy sat.tuy 'cause to be high'
ss-gWgW ra s.sgWgw. ra sgWgw. ra 'put last'
!ss-nnra s.snn.ra snn.ra 'cause to win'
ss-alla s.sal.la sal.la 'cause to weep'
ss-xxi s.sx.xi sx.xi 'cause to defecate '
ss-ttu s.st.tu sttu 'cause to forget'
ss-nna s.sn.na sn.na 'cause to say'
ss-ukrr s.su.krr su.krr 'drag'
ss-uff s.suff suff ' make angry'
ss-iff s.siff siff 'sieve'
The strings parsed in III all contain less than three syllabies.
By presenting the forms in (17) as counter-examples to Jebbour's gen-
eralization (15)i-a, we are implying that Jebbour's generalizations (15)
and ours (16) are claims about the same object. However, alternations in the
causative prefix is one area of Berber phonology where dialectal varia-
tion is rife , even within the confines of Tashlhiyt." Could it not be the
case that the differences between (15) and (16) simply reflect differences
between the Tashlhiyt dialect spoken in Tiznit, from which Jebbour 's data
is drawn, and the Imdlawn dialect, which is that spoken by ME? The data
in Jebbour 's work is too limited to enable one to answer that question."
Even if subsequent research on Tizinit Tashlhiyt shows that the length
alternation s in that dialect differ in systematic ways from those described

26 On alternation s in the causati ve prefix in various Berber dialects of Morocco, see Saa
(I995 : 230-259) for a partial survey of the recent literature. This author presents in great
detail the alternations in the causative prefix in Zenaga (Figuig), in Eastem Morocco.
27 Guerssel (I 992) compares the realizations of the causative, reciprocal and passive prefixes
in two Tamazight dialects and a Taqbaylit dialecl. Although the author intends to provide
an overall account of the phonology of these prefix es in the Berber dialects of Morocco
and Algeria, it is not clear how the general scheme he proposes can accommodate the lmdlawn
Tashlhiyt data.
130 CHAPT ER FIVE

here , the differences must presumably be specific to 1ength selection in

the causative prefix, rather than following from a general difference in
the way syllable weight is computed in the two dialects. For if syllable
weight distinctions in Tiznit accorded with Jebbour 's theses, the native
speakers of the Tiznit dialect, who share the same singing tradition with
the other Ashlhiys, would experience special difficulties in acquiring the
H/L distinction employed in verse by that tradition. However, Tiznit
speakers do not experience such difficulties."
We now turn to the conceptual differences between the two proposals.
Dur proposal is preferable to Jebbour's because of its parsimony. According
to CausLength (16), syllable count is the only factor which has an influ-
ence on the quantity of the prefix. (15), on the other hand, lists three different
factors, viz
- the 1ength of the base, measured in syllables (cases i-b and iii-a) , or in
moras (cases iii-b and iv);
- the presence of a geminate in the base (case i-a);
- whether the base begins with an onset or a nuc1eus.
What makes the parsimony of CausLength possible is another concep-
tual difference between the two proposals; that difference has to do with the
morphological unit whose phonological properties are taken into account
to predict the quantity of the prefix. In Jebbour's proposal the unit in
question is the base which is an input to the prefixation process, whereas
in our proposal, it is the causative stern as a whole , i.e. the output of the
prefixation process. Dur generalization establishes a tight relationship
between the phonological properties of the prefixal variants and those of
the bases which se1ect them: since the alternation aims at producing
causative sterns as long as possible up to three syllabIes , it is no wonder
that short bases should select the long prefix, and longer bases, the short
prefix . In Jebbour 's proposal, on the other hand, there is no intrinsic con-
nection between each context listed in (15) and the variant which is called
for by that context. The generalization we are proposing, then, points to a
formal account along the lines of templatic morphology.
Let us now give examples of the various cases covered by our general -
ization CausLength (16) . Table (18) below illustrates case (16)a, in which
the geminate prefix is required. The canonical form at the beginning of each
line is that of the base considered in isolation. Otherwise the layout of
the table is the same as in (17).29

28 According to Amarir (1975 : 114, 151), Lnazz bl'lid and Lhu sayn !zanti , two well-known
!rways whose recordings are among the earlie st still extant , came from the Tiznit area.
29 The verbs in (18) have the following meaning s: (a) bring c1oser, (b) bring down (fruit),
(c) cau se to be heavy , (d) remind, (e) keep awake, (f) cause to stand. up, (g) cause to be
delicious, (h) cause to be right , (i) drag , (j) cause to be lukewarm, (k) put last, (1) strain,
(m) cause to be red.
 TASHLHlYT SYLLABLES II 131

(18) I 11 III
parse of I parse with short pfx
a. vc: !zz-az z.zaz .zaz.
b. .VCc. ss-uss s.suss .suss.
c. V.CV !zz-uzi z.zu.zi zu.zi
d. C.CV ss-kti s.sk.ti sk.ti
e. V.CC zz-iwz z.zi.wz zi.wz
f. C.CC ss-nkr s.sn.kr sn.kr
g. V.CVC ss-imim s.si.mim si.mim
h. C.CVC lzz-vzan z.zv.zan zv.zan
1. V.CCC ss-ukrr s.su.krr su.krr
J. VC.CV ss-ulba s.su1.ba su1.ba
k. ec.CV ss-gWgWra s.sgWgw. ra sgWgw.ra
1. VC.CC ss-uddm s.sud.dm sud .dm
m. VC.CVC zz-iwziv z.ziw.ziv ziw.ziv
When they are syllabified in isolation, the bases in table (18) all contain
less than three syllables and they all begin with a nuc1eus. As can be seen
in column III , adding the simplex variant of the prefix usually gives an onset
to the initial syllable of the base, but it never creates an extra syllable. If
the geminate variant is used, on the other hand, the causative stern begins
with an onsetless syllable whose nuc1eus is the first half of the geminate
prefix; that stern contains one more syllable than the base.
Consider next the on set-initial dissyllabic bases. Prefixing the short
variant is enough to create a third syllable, and consequently the short
variant is selected, in conformity with (l6)b.
(19) I 11
a. CV.CV s-gula s.gu.la 'cause to hurry'
b. Ce.CV s-frsi s.fr.si 'split'
c. CV.CC s-nufl s.nu.fl 'cause to lose patience'
d. Ce.CC ls-brbr s.br.br 'boil'
e. CV.CCC s-muylt s.mu.ylt 'cause to be nauseous'
f. CVC.CV s-gusma s.gus.ma 'cause indigestion'
g. CVC.CC s-nussg s.nus.sg 'cause to be agitated'
Finally, here are examples illustrating the fact that the bases which
contain more than two syllables select the short variant of the prefix. To
highlight the fact that the short prefix is chosen even when its adjunction
does not increase the number of syllables, in all OUf examples but the first,
the base begins with a nuc1eus when it is syllabified in isolation.
132 CHAPTER FIV E

(20) I 11
a. CC.CC.CV s-brks sa s.br.ks.sa 'granulate'
b. V.CC.CV s-asstwa sas.st.wa 'level'
c. C.CV.CV s-fruri sf.ru.ri ' cause to crumble'
d. C.CV.CC !s-knawd sk.na.wd 'cause to roll about '
e. C.CVC.CV s-frussa sf.rus.sa 'clear up (weather)'
f. C.CVC.CC s-hlullf sh.lul.lf 'smooth'
We do not know of many counter-examples to CausLength among the
causative sterns in which the base is polysyllabic or does not begin with
an onset. Most of these counter-examples have the short variant of the prefix,
and there are independent reasons to believe that they are lexical excep-
tions. Consider for instance xsi, which means (a) 'go out (fire)' or (b) 'be
asphyxiated'. Two causative verbs are derived from this verb : ss-xsi and
s-xsi. ss-xsi (s.sx.si), which conforms to CausLength, means (a) 'extinguish'
or (b) ' asphyxiate' . On the other hand s-xsi can only mean 'extinguish'. This
gap suggests that ss-xsi represents the productive case while s-xsi is listed
in the Iexicon.
CausLength (16) states that in a causative form, the unit whose proper-
ties play a role in determining the length of the causative prefix is the
stern: inflectional affixes are irrelevant. Consider for instance the causative
stern ss-frs 'sharpen' . According to CausLength, the prefix is geminate
because the adjunction of a simplex s to frs yields a dissyllabic string (~.frs) .
But consider now ss-frs-n 'they sharpened'. Here it is crucial that CausLength
take sterns into consideration, rather than whole words: once the 3mp prefix
-n is taken into consideration, the string which follows the causative prefix
is dissyllabic (fr.sn) , and CausLength would incorrectly select the short
variant of the prefix.
Like that of the other verbs, the conjugation of the causative verbs
involves four sterns, viz perfective, negative , aorist and imperfective (see
§ 2.4). All the causative sterns cited in our examples up to this point are
perfective sterns, but CausLength is meant to be valid for the other sterns
as well . Consider for instance the four sterns of !ss-udn 'cause to be ill'
(those of !udn 'be ill' are given underneath for the sake of comparison) :
(21) pf neg aor impf
a. !ss-udn !ss-ud(i)n !ss-adn !ss-adan
b. !udn !ud(i)n !adn !tt-adn
In (2l)a the quantity of the prefix in each stern is determined indepen-
dently of that in the other three, and it only depends on the nature of the
phonological string which follows the prefix in the stern in question , e.g.
when CausLength applies in the perfective, the string whose syllable count
matters is su.dn, and when CausLength applies in the imperfective, that
string is sa.dan. Since both strings are dissyllabic , the long variant of the
 TASHLHlYT SYLLABLES II 133

prefix is independently called for in either case . What the preceding sen-
tences imply is that when one wants to predict the quantity of the prefix
of a causative verb for a given stern, there is no need to refer to the
quantity of the prefix in one of the three other sterns, or to any property
of the verb from which the causative verb is derived ."
At the beginning of this section we agreed to use the term 'base' to
refer to ' the verb from which a causative verb is derived ' . We can now
define that term more precisely. In (16) , 'base' is intended to refer to the
string which remains when a causative stern is stripped of its causative
prefix . The base is !udn in the perfective !ss-udn, while it is !adan in the
imperfective !ss-adan. The phonological differences between !udn and !adan
follow from various regularities which are not specific to the conjugation
of causative verbs , see DE (1991).
Shifting from one stern to the other in the conjugation of verbs some-
times impinges on syllable structure, e.g . !udn ends in a light syllable
(dn), and !adan ends in a heavy syllable (dan). One may expect to find
causative verbs in which the length of the prefix changes when one shifts
from one stern to the other. Here is one such case . The imperfective stern
of ls-wrrv 'cause to be yellow' has two forms in free variation: lss-iwriv
and lswrrav." The base liwriv calls for the lang variant of the prefix, Iike
the other VCCVC bases, and the base !wrray calls for the short variant, Iike
the other CC.CVC bases . Such instances are not numerous, however,
because as a rule the various bases of a verb all fall under the same branch
of CausLength (16); in (21)a, for instance, the bases !udn, !udin, !adn
and !adan all fall under branch a of CausLength. Note in particular that
the geminating verbs of section § 5.2 do not give rise to causative verbs
with length alternations in their causative prefixes. Consider for instance
kti , a geminating verb meaning ' remember' , and the causative ss-kti
'remind'. The perfective and imperfective sterns of each are given below
in (22)a,b:
(22) perfective imperfective
a. kti ktti
b. ss-kti (s.sk.ti) ss-ktay (s.sk.tay)
c. * s-ktti (s.kt .ti)
If imperfective gemination operated in the causative, the resulting stern
would be s-ktti «22)c), with a short prefix in alternation with the long prefix

30 Only the aorist stern and the irnperfective stern can occur without any affixes (see above
in § 5.2). Consequently , CausLength cannot in general be construed as capturing a relation
between units which can stand on their own as words.
31 This verb is derived frorn !wrry 'be yellow' , whose irnperfective stern is ltt-iwriv.
134 CHAPTER FrVE

in the perfective stern ss-kti «22)b). s-ktti is ill-formed, however. This ill-
formedness is the consequence of the following two claims r?
(i) The morphological structure of the imperfective stern of a causative
verb derived from a verb Z is [impf[cau[Z]]], e.g. the structure of the
imperfective stern ss-ktay ((22)b) is [impf[cau[kti]]]. Taking Ikti/ as a
starting point, one first derives the causative verb /s-kti/, which is then
inflected for the imperfective.
(ii) Imperfective gemination can only operate on inputs which are kerneIs.
As a result of (i), the input to the morphological processes which derive
the imperfective stern of ss-kti contains a causative prefix, i.e. that input
is not a kerneI, and consequently imperfective gemination cannot apply to
it.

5.5. CONCLUSION

In § 4.1, we introduced two mutually independent claims which are central

to our account of Tashlhiyt syllable structure, the Sonority-Driven
Syllabification thesis and the Licit Consonantal Nuclei thesis. While the two
morphologically-governed alternations discussed in this chapter are evidence
in favor of the Sonority-Driven Syllabification thesis, they are neutral with
respect to the Licit Consonantal Nuclei thesis. Consider for instance imper-
fective gemination, taking as examples the verbs kr; 'plough' and rks 'hide' ,
whose imperfective sterns are kkrz and rkks (v. (3». In accordance with
the Licit Consonantal Nuclei thesis , we are claiming that in the basic sterns
of these verbs, all syllable nuclei are consonants. Now suppose that we
discard the Licit Consonantal Nuclei thesis and analyze the syllabic con-
sonants of Tashlhiyt as phonetic realizations of eC sequences, where 'e'
stands for an epenthetic vowel." As a result of sonority-driven vowel
epenthesis, the basic sterns of krz and rks would now syllabify as .kerz:
and er.kes. Under this alternative analysis, it is still the case that the segment
which undergoes gemination in the imperfective is an onset in the basic
stern, as stated in generalization (5). Since the facts of imperfective gem-
ination are equally consistent with an analysis which does not incorporate
the Licit Consonantal Nuclei thesis, they do not constitute evidence in its
favor. Evidence for the Licit Consonantal Nuclei thesis will be presented
in the next chapter.

32 Ample evidence for the validity of these claims can be found in DE (1991).
33 Such an analysis was entertained briefly in § 4.1, v. (11) and the surrounding text.
 CHAPTER SIX

VOWELLESS SYLLABLES

As explained in § 4.1, our analysis of the syllable structure of Imdlawn

Tashlhiyt revolves around two theses , the Licit Consonantal Nuclei thesis
and the Sonority-Driven Syllabification thesis. The evidence in favor of
the Sonority-Driven Syllabification thesis was presented in the last two
chapters. In this chapter we present our evidence in favor of the Licit
Consonantal Nuclei thesis. According to this thesis, Imdlawn Tashlhiyt
has no epenthetic vowel s; if the nucleus in a syllable is not a full vowel
it is a consonant.
We first discuss the distribution of the short voiced vocoids (sections
6.1 to 6.3). That distribution lends strong plausibility to our contention
that the vocoids in question are not vowels, but mere transitions between
consonants. We then present two phonological phenomena which are natural
consequences of our claim that there are no epenthetic vowels in Imdlawn
Tashlhiyt. One is a gap in consonant combinations inside morphemes
(§ 6.4 .1); the other involves regressive devoicing (§ 6.4 .2) . In the next
section we contrast Imdlawn Tashlhiyt with Ath Sidhar Rifian, a Berber
dialect in which some short vocoids are genuine epenthetic vowels . The last
section is devoted to considerations about the short vocoids in other works
on Tashlhiyt.

6.1. VOWELS VS. TRANSITIONAL VOCOIDS

Transitional vocoids and their distribution is a topic which clearly requires

further research, but the knowledge we have already acquired about it is
sufficient to dispel suspicions that transitional vocoids are short epenthetic
vowels or surface reflexes of underlying segments.
Transitional vocoids were briefly introduced in § 2.2. I The reader may
recall that in that section a vocoid was defined as any stretch of time,
however short, not occupied by a glottal consonant or by an articulation
which is consonantal in the sense of Chomsky and Halle (1968). A voiced
transitional vocoid (VTV for short) was defined as any voiced vocoid which
is not an occurrence of one of the uncontroversial segments a, i, U , y, w.
By 'uncontroversial segments' we mean those segments which are already
present in lexical representations or which are introduced by uncontrover-
sial morphological or phonological processes. Compare for instance the

I See also § 4.1.

135
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
136 CHAPTER SIX

imperfective participle li-tt-!bukud-nl (prt-impf-blind:impf-prt) ' who

becomes blind' !ittbukudn and its perfective counterpart /i-!bbukd-n/ (prt-
blind-prt) !ibbukdn. One hears a voiced vocoid between k and d in the
pronunciation of both forms , which sound respectively as [Iit r'bukudn]
and [!ib:uk@dn]. The intervening vocoid in the first form is an instance
of u, i.e . a copy of the preceding vowel which is regularly in serted to
form certain imperfective sterns (v. § 5.2) . That in the second form is a
transitional vocoid. As another example, compare the pronunciations of
the free state noun It-i-gnil tigni 'sewing' and its bound state form It-gnil
tgni when these words occur in Isala-n tignil (finish-3mp ...) 'they finished
sewing' and in Isala n=tgnil (contest of= . . .) 'sewing contest' . t and g
are separated by a voiced vocoid in both forms. In tigni the invervening
vocoid is the nominal prefix discussed in § 2.5, which drops in the bound
state, while in tgni it is a transitional vocoid, a short [I] glossed over by
our broad transcription.
VTVs never occur adjacent to vowels. As already stated in § 4.1, we
believe that the location and vowel quality of VTVs are entirely predictable
on the basis of the neighbouring segments and of certain morphological
boundaries.
We will use the cover symbol '@' when we want to represent a VIV
without specifying its exact vowel quality. Although the expression 'tran-
sitional vocoid' implies our commitment to a certain analysis, it should
be clear that statements that a VTV occurs at a certain location in the pro-
nunciation of a given sentence have empirical content. To falsify a statement
of that kind one must inspect token s of the sentence in question to deter-
mine whether a vocoid occurs at the said location. If one occurs one must
check whether it is an 'uncontroversial segment' according to our analysis.
Setting aside dialect differences, our 'uncontroversial segments' are pretty
much in agreement with the full vowels and the semivowels in the tran-
scriptions of other authors who have written on Tashlhiyt, v. e.g. those in
Destaing (1920) or in Aspinion (1953).
We uniformly notate ' @' for all VTVs, but it must be kept in mi nd
that some are more vowel-like than others; some sound like full-fledged
vowels, albeit very short ones, e.g. the first @ in [!tn@g@d@] !tngd
(lt-!ngd/) 'she drowned ', whereas others sound like mere stop bursts
combined with voicing, e.g. the last two in [!tn@g@d@]. Also, VTV s are
rather elusive in many cases; now you hear them, now you don't, For
instance in liznnkk 'he made you become congested', from li-!znng=kI
(Sms-congestedozms) a VTV may be heard between n and kk in some
tokens ([ !izn: @k:h]) but not in others ([ lizn .kr'[); and in yet other tokens,
we are not sure . This is different from standard cases of free variation. In
French, for instance, pas de ski 'no ski' can be pronounced either as
[padceski] or as [patski], and the two variants are categorially distinct: in
the mapping between the segments and the tones of the sentence tune , for
 VOWELL ESS SYLLABLES 137

instance.i there are only two options, depending on whether the realiza-
tion of pas de ski /paz da ski/ contains three vowels or only two.
In our previous articles on Imdlawn Tashlhiyt and in the present book
the VTV s recorded between heterorganic consonants reflect the percep-
tions of one of us, FD, who does not speak the language. The other author,
ME, is normally unaware of the existence of the VTVs in his speech , and
when the presence of one is pointed out to hirn, he finds it quite difficult
to perceive, if he perceives it at all. 3 This is all the more striking since
ME is to some extent able to introspect about other aspects of his pro-
nunciation and consciously manipulate them, such as emphatic articulation
and voicing in consonant clusters.
A fully explicit description of the phonology of Imdlawn Tashlhiyt should
characterize the distribution of the VTVs and their phonetic properties in
each environment. We are able to do this only in part because the data
are very difficult to gather. If the difficulty simply resided in the fact that
the combinations to check occur at sentence level and that there is exten-
sive free variation in many contexts, our task would be similar to that
involved in making sense of the behavior of schwa in Standard French . But,
as we have said above , the VTVs of Imdlawn Tashlhiyt do not seem to
be all-or-none entitie s akin to French schwa." A more apt comparison would
be with the very short voiced vocoids that we seem to hear after the release
of the first stop in sequences dm, bd and gm in some token s of admirer
[admive], abdiquer [abdike] or bourgmestre [buvgmestx] when we listen to
Standard French with the same attention to detail as that required to make
out the most elusive VTVs of Imdlawn Tashlhiyt. These short vocoids
may indeed be there, but they do not count for the phonology of French,
i.e. there is no known phonological process which takes their presence
into account. We hold that the same is true of the VTVs of Imdlawn
Tashlhiyt. Their distribution is to some extent language-specific, and a
grammar of Imdlawn Tashlhiyt cannot be complete unless it describes that
distribution. But if, as we claim , the other phonological processes of the
language do not take the VTV sinto account, an incomplete description
of their distribution will be without consequences for the rest of the descrip-
tion of the phonology.

6.2. VTVS ARE RELEASES WITH VOICING

Before presenting in some detail the facts presently known about the VTVs
of Imdlawn Tashlhiyt, let us first outline how we propose to account for
them. In our view the transitional vocoids of Imdlawn Tashlhiyt are nothing

2 For examples of that mapping, v. Deli (1984).

J This assertion is valid only for those VTVs which occur between heterorganic conso-
nants. Between homorganic consonants it is another matter, see below § 6.3.3.
4 Aspinion (1953: 120) likens the 'e brefs' of Astuken Tashlhiyt to French schwa.
138 CHAPTER SIX

but consonanta1 releases. In this book 'release' means the same thing as
'removal of a radical obstruction in the midsagittal region of the oral cavity,
followed by pulmonic egressive airstream'. 5
Two conditions must be met for a voiced vocoid to occur at a given point
in the pronunciation of an utterance: the vocal cords must vibrate and the
articu1ators must assume aposture which allows the air to flow unim -
peded through the vocal tract. Here is, in a nutshell, how we view a VTV
occurring between two consonants A and B which are adjacent in the
terminal representations: it corresponds to a lag between the offset of the
closure of A and the onset of the closure of B, and its voicing is an exten-
sion of that of A or B.
Consider for instance the word /t-k1a/ 'she spent the day'. For reasons
to be explained below, the closure of t must be released before that of k
is formed, and consequently the 'hold' phases of t and k must be sepa-
rated by an interval of time during which the air can flow out unimpeded.
t and k being both voiceless, the intervening vocoid is voiceless too ; it
sounds like a short voiceless i, and one hears [thkla], or more precisely,
as Coleman (1996: 191) correctly points out for a related form, [tjkla] .
Consider now /t-gla/ ' it (f) soaks ' . t and g must again be separated by a
short vocoid, but in this case the glottal vibration required by g begins as
soon as t is released. The transition between t and g sounds like a very
short lax i, and one hears [trgla]. On the other hand in /t-bla/ ' she gave (a
vice)' the transition between t and b sounds more like IPA [g].
The distinction between released and unreleased stops is not known to
play a distinctive role in the lexicon of any language. McCawley (1967),
Anderson (1974), Selkirk (1982) , Kim-Renaud (1986) and Steriade (1993a,
b, 1994) have argued that it nonetheless plays a role in linguistically sig-
nificant generalizations about individual languages. An assumption shared
by all these authors is that release should be represented in the phonolog-
ical component of a grammar. However Kim (1995) argues that in Korean,
a language in which syllable-final stops must be pronounced unreleased
in certain environments, release is not a phonological entity since its dis-
tribution can be accounted for by resorting to mies of phonetic
implementation. Following a suggestion made to us by Nick Clements,
we shall adopt the same position for Imdlawn Tashlhiyt. Whereas phono-
logical mies map categorial representations onto other categorial
representations, the mies of phonetic implementation generate repre senta-
tions in which feature values are translated into targets along continuous
phonetic dimensions, see Pierrehumbert (1980, 1991) , Liberman and
Pierrehumbert (1984), Keating (1990), Cohn (1993), Huffman (1993),
Clements and Hertz (1996) and references therein.

5 Mentioning the midsagittal region allows us to speak of the release of I, which we take
to be a noncontinuant. On the relevance of pulmonic egressive airstream, v. Kim (1994).
 VOWELLESS SYLLABLES 139

The distribution of VTVs in Imdlawn Tashlhiyt results from the inter-

play between two kinds of phonetic implementation processes, those which
determine the relative timing of the constrictions of adjacent segments
and those which govem the timing of phonation. The processes of the first
type are responsible for the duration of the vocoids which occur between
the two first consonants in t-bla, t-gla and t-kla, and also for their vowel
quality ([~] in the first word and [I] in the others); those of the second
type are responsible for the fact that the vocoid is voiced in t-bla and
t-gla, and voiceless in t-kla. In the next three sections we rely heavily on
DE (1996a).

6.3. THE DISTRIBUTION OF VTVS

In this section we describe the possibilities of occurrence of the VTVs in

various contexts. Let us first give a very rough outline of the results of
the discussion to come.
The licitness of releasing the first consonant in a ce sequence depends
primarily on whether the two consonants are homorganic and on whether
they agree in sonorancy and in continuancy. If syllable structure plays any
role at all, it is a minor one and is limited to heterorganic sequences. The
diagram (1) below summarizes the situation when both consonants are
noncontinuants. This is the case we have studied in the most systematic
fashion, because in that case VTVs are easiest to discem by purely auditory
means. In (1) sister branches represent answers 'Yes ' (+) and 'No' (-) to
a que stion which is represented by the mother node. The consonant
sequences between parentheses are examples of the various cases , and the
numbers between square brackets refer to the seetions where those cases are
discussed.
(1) [a Place]
I
-I
(tb, kt, mn, 1m)
obligatory [a sonorant]
release I
[6.3.2] -I
(nd, tl, bm)
no release [Fusion]
[6.3.1] I
-I 1+
(tHt, nn-n) (Ht, n+n)
facultative gemination
release (no release)
[6.3.3.1,6.3.3.3] [6.3.3.2]
140 CHAPT ER SIX

As indicated by (1) , release is obligatory before a heterorganic noncontin-

uant (e.g. in tb) and it is excluded between two homorganic noncontinuants
with different values for the feature [sonorant] (e.g. tl). When two hornor-
ganic noncontinuants agree in sonorancy, release is optional except in
those cases in which the rule Fusion blends them into a geminate.
The preceding summary concerns only release. VTVs are releases accom-
panied by voicing. When voicing is absent, a transitional vocoid occurs
but it is not a VTV. Voicing is excluded between voiceless consonants,
e.g. in kt. In a sequence of two obstruents which do not agree in voicing,
voicing in the second consonant is more prone to give rise to a VTV than
voicing in the first, e.g. keeping other contextual factors constant, a VTV
is more liable to occur in kd than in gt.

6.3.1 . Two generalizations

The following two generalizations are exceptionless.

(2) If an underlying sequence does not contain any voiced segment,
its phonetic realization does not comprise any laryngeal vibra-
tions.
(3) The oral closure of a noncontinuant may not be released before
a homorganic noncontinuant which differs in sonorancy.
Let us dwell on each generalization in turn. The first implies that in
the pronunciation of an expression, voicing can only occur adjacent to a
sonorant or a voiced obstruent. This generalization is true of all styles of
elocution, even the most deliberate ones . VTV s cannot occur between a
voiceless obstruent and a pause or another voiceless obstruent. 6 Take for
instance the phrase /t-s-qssf-testt/ 'you shrank it' (2-cau-narrow-2s=d03fs),
which can stand on its own as a complete utterance. It contains only voice-
less consonants in its underlying representation, and it must be pranounced
voiceless from one end to the other: [tsqs .ftst :'']. Voiced vocoids cannot
be inserted in it at any point, even for the purpose of shouting or as carriers
of intonation. It is altogether unsuitable for singing or for slogan chanting.
Let us now turn to the second generalization. When two adjacent homor-
ganic noncontinuants differ in sonorancy, e.g. in t+n or in n+t, releasing
the first is prohibited in all morpho-syntactic environments. In is=t n-dl
(intedoßrns lp-cover) 'did we cover hirn?' t must not have an oral release
(Le. it must be pronounced with nasal plosion) because it is followed by
n, which is homorganic with t but is [+son]. Similarly, d must be pronounced

6 This is in sharp contrast with the @ vowel of Rifian and Moroccan Arabic, which can
occur between two voiceless consonants (v. § 6.5, § 8.2.2), or with fast speech pronuncia -
tions such as tkila for tequila in English, where a vowel can occur between t and k in slower
speech (Hammond 1997: 34).
 VOWELLESS SYLLABLES 141

with lateral plosion. In generalization (3) and elsewhere in this book the
word 'homorganic' is meant to refer only to primary articulations; it does
not require agreement with respect to secondary articulations such as
rounding or emphasis. For instance, the sequence g+gW is homorganic in the
intended sense. Similarly the sequence [dn], where [d] is emphatic while
[n] is not, must be pronounced with a continuous coronal closure, as is
for instance the case for the cluster straddling the word boundary in It-
!bbukd nannaes/ 'his eIder sister became blind' (3fs-blind elder:sister=3s).
Like (2), prohibition (3) is enforced without taking anything into account
except the phonetic properties of the segments and their adjacency. It can
hold sway over sequences of any length, no matter what their morpho-
syntactic make-up. For instance in the sentence Idl-n=t ntl-nl 'they covered
hirn and they hid themselves' (cover-Jmpedoßms screen-3mp) the eight
consonants must be pronounced with a single uninterrupted closure in the
midsagittal region.
Generalization (3) only concerns sequences of noncontinuants which
differ in sonorancy. When the two homorganic stops agree in sonorancy,
e.g . in tH, t+d or g+kw, releasing the first stop is prohibited in some cases
and optional in others : it is prohibited when an assimilation rule has applied
or, in most morpho-syntactic contexts, when both stops are short; it is
optional otherwise. Whereas the cases covered by (3) are purely a matter
of phonetic implementation, those in which the stops agree in sonorancy
result from the interplay between the phonological component and the
phonetic component. They will be taken up later (v. § 6.3.3).
(3) suggests some kind of principle of phonetic inertia. Assurne that in
the phonetic implementation of a stop the action of the articulator effecting
the closure in the oral cavity is characterized by a single target which cor-
responds to the 'hold' phase. In a sequence such as t+n, then, the tip of
the tongue would be assigned two identical targets in a row, and the pro-
hibition against releasing t is best seen as resulting from the requirement
that an articulator must follow the shortest possible path when moving from
one target to the next. We dub this requirement MINIMAL-PATH(place).
The facts presented earlier about voicing suggest that the phonetic imple-
mentation of laryngeal features is subject to a general requirement which
is an analogue of MINIMAL-PATH(place), call this requirement MINIMAL-
PATH(voice). According to MINIMAL-PATH(voice), a transitional vocoid
must be voiceless between voiceless consonants, v. for instance the
transitional vocoid between q and k in i-snnqwk 'he wrung your neck'
(Jms-wring.neckedo'Zms), which must be pronounced [iSn:@qhkh], and it
must be voiced between voiced consonants, v. for instance the transition
between g and bb in !gbbs [!g@b:s] 'plaster!' .
Only when adjacent consonants agree in voicing does MINIMAL-
PATH(voice) make predictions about voicing in an intervening transitional
vocoid. Preliminary observations on the clusters in which one consonant
142 CHAPTER SIX

is voiced and the other is voiceless suggest the following generalizations.

If the second consonant is voiced, the vocoid is almost always voiced.
For instance one he ars an unmistakable [u] between kk" and d in
nna-neakk" !di=tt (say-3mp=all take.outedoßfs) 'they all said "take her
out!"'. If the first consonant is voiced and the second is not, there are
two cases to consider, depending on whether the first is a sonorant or an
obstruent. If it is a sonorant the vocoid is always voiced, witness the voiced
vocoid between land k in /i-xtlekm/ 'he conned you' (Jms-conedozfs),
which is pronounced [ixtl@km].7 If on the other hand the first consonant
is a voiced obstruent, the general tendency is for the transitional vocoid
to be voiceless, e.g. in ss-agg" !titt (cau-peep eye) ' let one eye show" the
glottal vibrations of ggWdo not seem to extend beyond the release of the
dorsal closure. However there are particular environments in which the
glottal vibrations of the first obstruent last until the onset of the constric-
tion of the following obstruent. This often happens before a pause, e.g.
between g and t in /t-xng-t/ 'you strangled' (2-strangle-2s), which is pro-
nounced [txn(@)g@t].

6.3.2. Release in heterorganic clusters

One striking fact about the pronunciation of Imdlawn Tashlhiyt is that an
obstruent stop is always released in an audible manner before another stop
which is not homorganic with it. It is instructive to compare the pronun-
ciation of kt in i-kti 'he remembered' and in the French word acteur (aktrey).
The velar stop must be pronounced with an audible release in ikti, which
must sound like [ilt'ti], while the release is only optional in the French word,
which can be pronounced either a[kht]eur or a[k"'t]eur. It is commonly
said that k is released in [kht] and not in [k"'t], and we will follow this
way of speaking, wh ich is harmless as long as one keeps in mind that
what is involved is a difference in the phasing between the closure of k with
that of the following consonant. In [kht] the velar closure of k is removed
before the coronal closure of t is formed. In [k"'t], on the other hand, the
coronal closure is formed with the velar closure still in place, and the sub-
sequent removal of the velar closure does not have any audible consequence.
The facts we know at present are compatible with the generalization that
in Imdlawn Tashlhiyt the oral closure of a consonant never overlaps in
time with that of an adjacent heterorganic consonant." In other words the
transition between two heterorganic consonants must always be an open

7 Homophonous with fix t-Ikml 'if she reaches' (if 3fs-reach).

8 Said, for instance, to a woman whose hair is hanging in front of her eyes.
9 Kinesthetic observations by ME suggest that sequences 1fI, bl, mll may be counterexam-
pies. We disregard these pending further research.
 VOW ELL ESS SYLLABLES 143

on e.' ? Let us assume that in Imdlawn Tashlhiyt phonetic implementation

is subj ect to the following requirement:
(4) NO-OVERLAP:
Radical constrictions effected by different articulators in the
oro-pharyngeal cavity may not overlap in time."
Requirement (4) concerns only 'radical' constrictions, i.e. it is not meant
to go vern the timing of the con strictions involved in secondary articulations
such as labialisation and dorsopharyngealization. In t-grna 'she sewed' ,
for instance, it could be the case that t is produced with a raised tongue
dorsum and rounded lips, as suggested by the fact that the transition between
t and gWsounds like a short [u] .
Consider for instance t-ss-ftq [ts.ft''q''] ' it (f) bodes ill' , where t mu st
be released before the following q. Our proposal implies that the release
of t does not have an associated phonetic target of its own. Release is simply
an unavoidable consequence of NO-OVERLAP. Because of NO-OVERLAP,
the 'hold' phase of t and that of q mu st be separated by an interval of
time during which the air can flow out unimpeded. At lea st before a
heterorganic segment, there is no need for stop releases to ha ve a coun-
terpart in the representations at any stage of the phonological deri vations.
NO-OVERLAP (4) concerns all the heterorganic sequences, but all such
sequences do not lend themsel ves equally well to the auditory detection
of intervening vocoids. These are often difficult to make out between a frica -
tive and another con son ant. For instance ak"; ' recog nize, aor ' sounds like
a[kuz] , but we cannot be sure that k" and z are actually separated by a vocoid
in our sense. It could be that coronal friction begin s as soon as the velar
clo sure is released , and that what we hear as a vocoid is simply the begin-
ning of the voiced fricati ve. Analogous remarks can be made about the VTV
which is at time s heard bet ween f and dd in It- !fddal ' it is over' (3fs-end)
and between x and z in /t-xzneak/ 's he hoarded for you' (3fs-hoard=dat2ms) .
The observations whi ch lead to NO-OVERLAP were made primarily on the
release of noncontinuants be fore noncontinuants. If the vocal cords are
vibrating at the time of rele ase, the release of the oral cl 0 sure of a non-
continuant sounds lik e a short vowel with a clearly delineated onset; it
sounds like a short bur st of aspiration if they are not. The vocal cords are
for instance vibrating when k and d are released in /i-Ibbukd/ (3ms-blind)
'he became blind' , and one hears !ibbu[k@d@]; they are not when t and
kare released in y-ut= k (Sms-strikee dozms) 'he struck you ' , and one hears
yu[thk h] .
Before a pause, ob struent stops are as a rul e released in an audible
manner, as illu strated by the final stops in the two last exampl es.

10 On 'o pen' and 'close' transitions, see Bloomfield (1933) and Catford (1977).
11 On articulatory overlap, see for instance Browman and Goldstein (1989, 1990).
144 CHAPTER SIX

Among the heterorganic clusters of noncontinuants, those in which we

find it easiest to determine by ear whether the stop is released, are those
of two obstruents (v. examples above) and also those beginning with m, n
or I, at least those in which the vocal cords are vibrating at the time the
oral closure of the sonorant is released." This happens for instance in
Isstl-x=tt/ (weigh-Isedoßfs) 'I weighed it in my hand' , which is pronounced
[s:tl@xe] and in /t-nfsl (3fs-deflate) ' it went flat', which is pronounced
[tn@fs] .
Of all the heterorganic clusters, those in which we find it most diffi-
cult to ascertain the presence of an intervening vocoid are those in which
the second consonant is a syllabic nasal or liquid. We are often unable to
distinguish by ear between [CR] and [C@R] (R a nasal or a liquid), e.g .
between [thknt h] and [thk@nt h] for tkntt (from /t -knd-t/) ' you bam-
boozled', or between ti[gm]mi and ti[g@m]mi for t-i-gmmi 'house' .
Our discussion until now implies that the only contextual factors which
are relevant for the occurrence of a VTV in a ce sequence are the phonetic
properties of the two consonants. This is true for voiceless clusters and
for homorganic ones, but other features of the environment come into play
in the other clusters. It is often the case that a given CC sequence contains
a VTV in some contexts but not in others." The preceding sentence makes
this look like an all-or-nothing phenomenon, but in some cases at least it
is rat her a gradient one: the VTV is easier to discern in some contexts
than in others. To carry out a systematic study of this phenomenon, sharper
means of observation will be required than the unaided ear. Examples are
given in (5) .
(5) a. t~.b:y [ts@b)'] t-sbv 'she painted'
a' . is.b:y [isbv ] i-sbv 'he painted'
b. ra.dl.bix [radl@bix] rad lbi-x 'I will bite '
b'. m.ral.bix [mralbix] mra lbi-x 'if 1 had bitten'
c. i.Zlx [izl@x] i-zlx 'he is dirty '
c' . su.lx [sulx] sul-x 'I survived'
d. mk.knx [m@k:nx] mkkn-x 'I tightened'
d' . i.mk.kn [imk:n] i-mkkn 'he tightened'
We have not been able to determine whether syllable structure is involved
in variations such as those in (5) . The only pattern which emerges clearly

12 Sonorant consonants are as a rule fully voiced in Imdlawn Tashlhiyt. They may devoice
when they are subject to prepausal annexation after a voiceless consonant (v. DE 1985),
but even in that context the devoicing is only a partial one.
Il This was already pointed out in DE (1985: 117).
 VOWELLESS SYLLABLES 145

from our data is the following general trend: a ce sequence is more likely
to contain a VTV when neither C is adjacent to a vowel. 14
In (5)a and (5)b the ce sequence with an intervening VTV is hetero-
syllabic. More examples of this situation are given below in (6). From
now on the raised 'h's indicating voice1ess vocoids are omitted from our
narrow phonetic transcriptions whenever convenient. Examples b.e.f.n.o.p
are included for the sake of comparison."
(6) a. dl.xas [dl@xas] Idl-x=asl cover-ls=dat3s
b. .dlx. [dl@x] Idl-xl cover-1s
c. 1.bl.tnt [!t@btnt] It - lbttnet/ 3fs-line=d03ms
d. sn:yu.ba [sn@-yuba] /s-nvuba/ cau-vanish
e. in.vu.ba [invuba] /i-nvuba/ 3ms-vanish
f. sn.nq [sn:@q] Isnnql wring:neck
g. snn.qas [sn :@qas] Isnnq=asl wring:neck=dat3s
h. snn.qxt [sn:@qxt] Isnnq-x=t/ wring :neck-1 sedoßms
i. tibll [ts@bh] It-sbhl 3fs-whip
j. t.ks.btt [tksbts't.] /t-ksbett/ 3fs-own=d03fs
k. ti.blt [tz(@)bt] /t-zbd-t/ 2-pull-2s
1. ti.bl.tas [tz@btas] It-zbd-t=asl 2-pull-2s=dat3s
m. ts.sn.gjt [!ts:n@g@t] /t-ss- !ngd-t/ 2-cau-drown-2s
n. i.xng [ix@n@g] li-xngl 3ms-strangle
o. j.xnkk [tx@n@k:] /t-xngek/ 3fs-strang1e=d02ms
p. 1.kntt [tknt:] It-knd=t/ 3fs-bamboozle=d03ms
q. tl.mtt [t!@m@t] /t-lmd-t/ 2s-learn-2s
r. sn.f] [sn@fl] Isnfll exchange
s. l.bi [!l@bz] I!1bzl knead
Let us summarize OUf presentation up to this point. VTVs never occur
adjacent to a vowel. They always occur after a consonant. They always occur
next to a voiced segment. When one examines the manner of transition
between two noncontinuants there are three cases to consider. In the first
two cases there is only one way of making the transition, regardless of
the morpho-syntactic context: a noncontinuant is released before a het-
erorganic consonant, and it is unreleased before a homorganic consonant
which differs in sonorancy. We have suggested that these two generaliza-
tions are reflections of two requirements on phonetic implementation in

14 Contextual factors other than the properties of the two consonants are also involved in
deterrnining the vowel quality of the intervening VTV. Colernan (2001) shows that the color
of VTVs is influenced by neighboring vowels.
15 The expressions in (6) have the following rneanings: (a) I covered for him; (b) I covered;
(c) she lined it (clothing) ; (d) cause to vanish! ; (e) he vanished ; (f) wring the neck! ; (g)
wring (sorneone's neck) for hirn!; (h) I wrung his neck; (i) she whipped ; (j) she owns it:
(k) you pulled; (I) you pulied for hirn; (rn) you drowned; (n) he strangled; (0) she strangled
you; (p) she barnboozled hirn; (q) you learned; (r) exchange!; (s) knead!
146 CHAPTER SIX

Imd1awn Tashlhiyt: NO-OVERLAP, see (4), and MINIMAL-PATH(place),

which dietates that an articulator should follow the shortest possible path
when moving from one target to the next. We now turn to the third case,
that of a sequenee of homorganie noneontinuants which agree in sono-
raney.

6.3.3. Release before a sibling consonant

6.3.3.1. S1BLlNG-RELEASE
Let us say that two segments are siblings when they have the same va1ues
for the features [sonorant] and [continuant], and the same primary articu-
lation . For instanee, in the underlying inventory of Imdlawn Tashlhiyt (see
§ 2.1), the siblings of It/ are It, !t, d, !d/; the siblings of Izl are ti; !z, s,
!s/; the siblings of lkJ are /k, k", g, gW/.16 To take a last example, Ifl has
only one sibling in Imdlawn Tashlhiyt, Ifl itself.
When two sibling stops stand next to one another, the transition between
them may take two forms. Their oral closures may blend into one long, unin-
terrupted, closure phase, or the oral closure of the first stop may be released
before that of the following stop is formed. These options are in free vari-
ation in some eontexts but not in others . Beeause Imdlawn Tashlhiyt allows
these two options, the loeation of releases in sequences of sibling stops may
be the only distinguishing feature between two contrasting expressions.
Consider the following sentenees.
(7) a. lis t-ttu-t/ [isr'ttut] 'did you forget?'
int 2s-forget-2s
b. lis=tt t-ut/ [istr'tut] 'did she hit her?'
int=do3fs 3fs-hit
From now on the symbol ,2, represents any stop release, voiceless (h) or
voiced (@). In the above examples ru and trt respectively represent IPA
[tht:] and [et]. (7)a may not be pronouneed as [istr'tut], and (7)b may not
be pronouneed as [isr'ttut].
Both sentenees have an alternative pronunciation in free variation: an
acceptable pronunciation for both is [istttut], with an uninterrupted coronal
closure which is unambiguously heard as a sequence of three consonants."
The eontrast in (7) involves a word sequence, but similar eontrasts are
found word-internally. Itt-tabaa=t/ 'follow hirn!' (impf-follow=do3ms) can
be pronouneed [tr'tabaat] but not [t'ttabaat] , whereas /t-ttul 'she forgot'
(3fs-forget) ean be pronouneed [rttu] but not [tt'tu]. Here is another example:

16 Note that /q/ is not included (see below).

17 The coronal closure in [istttut] is longer than that in the contrasting form [isttut]
/is=t#t-ut/ 'did she hit hirn?' (int=do3ms 3fs-hit).
 VOWELLESS SYLLABLES 147

(8) a. /i-vli d=ddirl [id'ddi] 'he started c1imbing'

3ms-c1imb prepefoot
b. /t-bidd di-s/ [idd 2di] 'she stopped with hirn'
3fs-stop prep-3s
d2dd and dd2d respectively represent [d@d:] and [d:@d]. The examples in
(8)a and (8)b each have an alternative pronunciation in free variation with
no release ([ddd]).
That the homorganie sequences appearing in most of our examples are
coronal sequences is only due to the greater availability of coronal conso -
nants : in Imdlawn Tashlhiyt it is the coronals whieh have the most complete
series of stops and they appear in numerous grammatical morphemes. The
phenomena under discussion are not peculiar to the coronals. Here is for
instance an example with velars which is parallel to (7).
(9) a. /radeak kkis-nl [radak'kkisn] 'they will dispos sess you'
fut=dat2ms dispo ssess-3mp
b. Irad=k krf-nI [rakk/krf-n] 'they will tie you up'
fut=do2ms bind-3mp
In (9)b Idl and Ik/ obligatorily blend into Ikk/ through a process of total
assimilation on which see § 3.2.1.3. [kk/k] is not an acceptable pronunci-
ation in (9)a, neither is [k 2kk] in (9)b . Both sentences in (9) have an
alternative realization in free variation where Ikkk/ is realized as [kkk] ,
i.e. with an uninterrupted velar closure whieh is unambiguously heard as
a sequence of three consonants.
The phonetic contrasts in the preceding examples are all of the form C2C:
vs. C:2C, i.e. they all involve sequences of three X slots. Analogous con-
trasts are possible for longer sequences. Sequences of five X slots , for
instance, allow a three-way contrast. The sequences t:2t:2t , rt:2t: and t:2 ft :
occur respectively in It-smun-t=t d=t -tbir-tl (2s-put:together-2s=do3ms
withebf-dove-fs) 'you put it (m) together with the dove', in lar=t tt-ttu-x/
(irnpfedoßrns impf-forget-ls) 'I forget hirn' and in It-ut=t t-ttu=tl
(3fs-strike=do3ms 3fs-forget:aor=do3ms) ' she struck hirn and forgot hirn'.
As can be seen in our examples, arelease can never intervene between
the two 'halves' of a geminate. Let us call this geminate inseparability.
Inseparability obtains für geminates resulting from assimilatory processes
(e.g. in (9)b) as weil as for those already present in lexieal representa-
tions. The inseparability of geminates follows from the conception of
geminates which we have adopted (see § 3.3) : a geminate is a single
feature bundle (a single Root node) associated with two prosodie positions. 18

18 As already stated in § 3.1, we follow the proposals of Clements and Hume (1995) con-
cerning the internal structure of segments .
148 CHAPTER SIX

1t is natural to interpret the Root node of a stop as representing an unin-

terrupted oral c1osure. Since, like simple stops, geminate stops are comprised
of a single Root node, the idea of a consonant with an intervening release
between its two 'halves' does not make any more sense for geminate stops
than it does for simple stops.
Once the source of geminate inseparability is recognized, the data in
(7)-(9) is easily understood. Consider again (7), for instance. In both (7)a
and (7)b, in the underlying representation, sequence Ittt/ is indeed comprised
of three prosodie positions, but only of two Root nodes, as shown in (10),
where the letter 't' represents the feature tree which defines the segment
t:

(10) (7)a: IHtt/ (7)b: Itt+t/

X X X X X X
I
Root
<:>
Root
<:>
Root
I
Root
I I I I
t t t t
If arelease is to occur inside the structures represented in (10), it can only
occur 'after' the first Root node . The difference between [isr'ttut] and
[istrtut] simply follows from the fact that the first t is short in lis#t-ttu-t/
«7)a) whereas it is lang in lis=tt#t-ut/ «7)b).
How is a grammar of Imdlawn Tashlhiyt to account for the fact that
the underlying contrast between It+tt/ and Itt+t/ gives rise to a phonetic
contrast between [t2tt] and [tt2t]? Given MINIMAL-PATH(place), one would
expect release always to be prohibited between homorganic stops, regard-
less of whether they agree in sonorancy. As a matter of fact, the prohibition
only holds when the two stops differ in sonorancy, i.e. in those transitions
whieh involve what Catford (1977: 219) terms an airpath change; other-
wise its enforcement is merely optional, as we have just seen . This
optionality is presumably the consequence of some other general require-
ment which is in conflict with MINIMAL-PATH(place) . Finding out what
this requirement may be is beyond the scope of this book. For the sake of
expliciteness, let us write the following rule, whieh encapsulates whatever
combination of deviees - rules, principles, and the like - is responsible
for the fact that a stop is optionally released before a sibling stop.

(11) SIBLING-RELEASE: [-cont] ~ released I - - sibling

(optional)

We have already noted the existence of alternative pronunciations with

no releases in sequences Itttl, Idddl and Ikkk/ sequences in (7), (8) and
(9). In general, releases between sibling stops are not mandatory, that is,
if an expression has one acceptable pronunciation which contains two sibling
 VOWELLESS SYLLABLES 149

stops separated by arelease, then it has another where the two closures blend
into a single uninterrupted one. Here is an example where the sibling
consonants differ in voicedness. In Ihra#t-dll ' she just covered herse1f' (just
3fs-cover) It-dl may be realized as t2d or as td, with a continuous oral
closure." The variants of the same form with and without release are not
feit to belong to different speech styles or tempos, but the Ionger a sequence
of sibling consonants the stronger the preference for pronouncing the
possible releases. For instance lar#tt-ttu-xl 'I forget' (impf impf-forget-ls)
can be realized either as [artrttux] or as [arttttux], with a preference for
the first realization. Similarly, in lar=stt tt-ttu-n/: 'they forget her'
(impf=d03fs impf-forget-3mp) the realization trtrtt
is strongly preferred.
We have come across a number of instances where release between
sibling stops is mandatory, and we have yet to disco ver the phonological
and morpho-syntactic properties which set these instances apart from the
run-of-the-mill cases, where it is only optional. Here are two examples.
Ikk+gl must be realized as kk!g in /i-nnaeyyi kkegrat-sn/ 'he told me to pass
between them' (3ms-say=datls pass:aor=between-3mp), and /t-ttl must be
realized as rttin It-ttu/ 'she forgot' when this word follows a pause ."
There is no reason to ascribe SIBLING-RELEASE to the phonological
component in the grammar of Imdlawn Tashlhiyt. We do not know of any
phonological rule which should be ordered after SIBLING-RELEASE.
The distribution of stop releases varies from one language to the next, but
the range of possible variation is probably rather limited. SIBLING-
RELEASE (11) is presumably one of a small set of options allowed by
Universal Grammar. Our policy will be to keep the formulation of the
rules of Imdlawn Tashlhiyt which directly involve release as free as possible
from contextual restrictions. Specifically, we shall assurne that SIBLING-
RELEASE (11) uniformly operates across all morpho-syntactic
environments. In those situations where release before a sibling stop is
actually forbidden, we shall try to find independent mechanisms which
exc1ude the first term of the alternative - release or no release - implied
by the optional nature of SIBLING-RELEASE.

6.3.3.2. The Fusion rule

In Imdlawn Tashlhiyt release is in most cases forbidden between two short
sibling consonants. Such is for instance the case in sequences which straddle
a word boundary. In a sequence IA#BI where A and Bare short sibling
consonants, releasing A is prohibited most of the time, unless IA#BI is a
subsequence in a Ionger sequence of siblings." Here are examples. Release

\9 There exists yet another variant, viz. dd.

20 But release is only optional in /mra t-ttu/ 'if she had forgotten ' .
2\ Instances of sequences of more than two siblings were given in the text under (9).
150 CHAPTER SIX

is forbidden between the two identical consonants in t-trm ma=s 'his mother
went down' (3fs-go:down mother=3s), where Im#mI must be realized as
mm (*m 2m), and similarly 11#11 must be realized as 11 (*[2[) in /i-fl luzinl
'he gave up the factory' (3fs-give:up factory), It#tl must be realized as tt
(*rt) in y-ut t-ili 'he struck the ewe', and Ig#gl must be realized as gg (*g2g)
in i-frg gar i-frig 'he put up a bad fence' (3ms-enclose bad u-enclosure) .
Release is also forbidden between two nonidentical siblings, as in the
fo11owing sentences:
(12) a. /t-frd t-funas-t/ 'the cow grazed'
3fs-graze bf-cow-fs
b. lis=ak kWra-n t-i-gmmi/ 'did they rent the house to you?'
int=dat2ms rent-3mp f-u-house
In (l2)a Id#tl can be pronounced dt but not d 2t; in (12)b Ik#kwI can be
pronounced as kk", i.e. as a continuous dorsal closure with no concomitant
rounding during its first half, but not as ~kw.22 Given that SIBLING-
RELEASE (11) optionally releases a stop before a sibling stop, how are
we to prevent it from generating d 2t and ~kw as optional variants of dt
and kk"! Let us assurne that the mechanism responsible for the prohibi-
tion of releases is the following phonological rule, which must operate
before SIBLING-RELEASE (11):

(13) FUSION: R~ R~ Rtj Rtj

I
Cpl
I
Cpl ~
I
Cpl
I
Cpl
I
o.
I
a
~
a
Fusion is an obligatory rule. It operates in certain morpho-syntactic envi-
ronments and not in others, as we shall see later. The rule merges two
identical primary articulations if the Root nodes which dominate them are
adjacent and if they are furthermore comprised of the same feature speci-
fications." We shall assurne that release is universally prohibited between
two Root nodes in representations containing the linked structure displayed

22 In (l2)a /d#t1 can also be pronounced as tt (but not as rt), and in (l2)b /k#k w / can also
be pronounced as a geminate kWk w (but not as kw2k W) . These free variants are due to regres-
sive assimilations in phonation type and in rounding to which we will return later. These
assimilat ions are optional in sequences of short consonants straddling a word boundary . When
regressive rounding assimilation occurs in /k#k w/ in (12)b the preceding a is articulated further
back . Only then does sentence (12)b become homophonous with the following : /iseakrk"
ra-n tigmmi/ 'the fact is that they even want the house ' (indeedeeven want-3mp house) .
23 That is, if they agree for the features [sonorant], [vocoid] and [approximant] . Clements
and Hume's [+vocoid] is the equivalent of [-consonantal] in Chomsky and Halle (1968),
and [-vocoid] is the equivalent of [+consonantal] . Vocoids and liquids are [+approximant]
whereas the other sounds are [-approximant].
 VOWELLESS SYLLABLES 151

to the right of the arrow in (13), and, more generally, between two Root
nodes which share a single primary articulation. Here is what we mean
by a primary articulation.
(14) Primary articulation:
The node in the feature tree which specifies which articulator
implements the feature [continuant], together with any nodes
which that node may dominate.
This definition is in the spirit of Sagey (1986) and Halle (1992, 1995). In
[k"], for instance, the articulator which implements [-cont] is the dorsum,
and consequently the primary articulation is the node [dorsal]. In [!z] the
articulator which implements [+cont] is the front of the tongue, and the
primary articulation is the node [coronal] together with the specification
[+anterior] which it dominates. The diagrams in (15) show how the relevant
aspects of these segments are represented in the version of feature geometry
advocated in Clements and Hume (1995):24

(15) Root
I
Oral Cavity

~nuant]
Cpl
»<.
Vpl [dors]
I
[lab]
[!z] Root

~Oi]
Oral Cavity
»<>;
Cpl [+contin]
.>'>;
Vpl [cor]

[phar]
-<:
[dors]
I
[+ant]

The representations in (16) result from the operation of Fusion in /d+tJ

and in /k+kw/ in the examples in (12).

24 The diagrams in this section represent only those aspects of the structure of segments
which are relevant.
152 CHAPTER SIX

(16) a. dt (from Id+t/) b. kk" (from lk+kwl)

Root Root Root Root

[+v~Cpl I
Cpl
I
Cpl
I
Cpl
<:>
[eor]
~
[dors] Vpl
I
[lab]

Universal Grammar bars the rules of phonetic implementation - more specif-

ieally: SIBLING-RELEASE (11) - from releasing the first stop in linked
struetures sueh as these. In (16)a, for instanee, there is a single primary
eoronal artieulation. Consequently, in the phonetie eomponent the tongue
blade is assigned a single target , i.e. dt in (16)a must have a single 'hold'
phase.
Sequenees of identieal eonsonants subjeet to Fusion are homophonous
with the eorresponding geminates, e.g. amud ds-u-dis 'the seed and the
stomaeh' (seed andeb-stomach) is homophonous with a-muddu di-s 'the trip
with hirn' (u-trip with-3s). The relevant aspeets of the strueture of a lexical
geminate Iddl are represented below in (17)a ; (17)b represents the linked
strueture whieh results from the operation of Fusion in Id+d/.

(17) a. Iddl b. Id+d1 after Fusion

X X X X
<> Root
I
Root
I
Root

[+v~Cpl [+v~
Cpl
~Oi]
Cpl
I
[eor]
<:>[eor]

The reason why release is impossible in either strueture is that both eontain
a single primary artieulation. Let us use 'Iong primary artieulation' to
refer to a primary artieulation whieh is shared by two skeletal slots, as is
the ease with [eor] in (17)a and (17)b .
Given our analysis of release, an uninterrupted long closure is ambiguous:
it may be the realization either of a long primary articulation, as in (17),
or of identical stops in which the optional rule SIBLING-RELEASE (11)
has eleeted not to apply. Consider for instanee what we transeribe as dd,
i.e, d with a closure spanning two timing slots, in eontrast with d 2d. The
terminal phonologieal representation whieh eorresponds to d2d ean only
be (18) below, for the closure before the release and that after it must be
 VOWELLESS SYLLABLES 153

manifestations of different occurrences of the same phonetic target, either

occurrence corresponding to a [cor] node of its own in the terminal phono-
logical representation.

(18) x x
I I
Root Root
I I
d d

dd too can be the phonetic implementation of representation (18), as when

/d#dd/ is realized as ddd in (8)a. In that instance the first and the second
occurrences of the letter 'd' in ddd represent time slices which correspond
to successive Root nodes. In other instances, dd is the phonetic imple-
mentation of a long primary articulation as in (17), e.g . in the previous
example /amud#d=u-dis/, where the sequence /d#d/ is subject to Fusion
and surfaces as (17)b. What makes us believe that in that case /d#dI yields
(17)b rather than (18) as its terminal phonological representation? It is
the fact that in amud deu-dis the only acceptable realization of /d#d/ is
dd. If in that example /d#dI could surface as (18), SIBLING-RELEASE (11)
would give rise to a free variant d2d, contrary to fact. More generally,
what allows us to tell apart those long closures [C:] which materialize a
sequence of identical stops from those which materialize a long primary
articulation, is the fact that the former, but not the latter, have a free variant
[C2C] .
The plausibility of the Fusion rule is enhanced by the facts of some
other Berber dialects. In the dialect of Figuig as spoken in the Zenaga
village, for instance, the geminate counterpart of /d/ is phonetically voice-
less, i.e . /ddl is always realized as [t:]. Heteromorphemic /d+dI sequences
are also always realized as [t], regardless of morpho-syntactic structure."
The latter fact shows that that dialect ha s a rule similar to Fusion (13) .
Ath Sidhar Rifian Berber has analogous facts which involve continuancy
differences between short obstruents and their geminate counterparts."
Fusion (13) does not treat the uvular stops /q, qW/ as homorganic with
the velar stops /k, k", g, gW/, e .g. the stern-final /q/ must be released in
/i-s-qllq=kI 'he made you angry' (ßms-cau-angryedo'Zms), whereas the
stern-final /k/ must not be in /i-sllk=kI 'he saved you' (Jms-saveedozms).
Uvulars must consequently differ from velars in their primary articula-
tions. In Imdlawn Tashlhiyt a velar stop optionally assimilates to a following
uvular stop, but not vice-versa. /k+q/ can be pronounced lCq or qq

25 See Saa (1995).

26 See Deli and Tangi (1992: 158-159).
154 CHAPTER SIX

in /is=ak !qrra-n/ 'did they make a confession to you?' (int=dat2ms

confess-3mp), but /q=k/ cannot be pronounced kk in /i-s-qllqek/ 'he made
you angry '. On the structure of velars and uvulars, see Trigo (1991) and
Elmedlaoui (1995a: 32ff).
Before we move on to examine various restrictions on the operation of
Fusion, let us briefly explain why the rule is formulated to merge primary
articulations rather than other nodes high er up in the feature tree.
Imagine that Fusion were to merge C-place nodes whenever they
dominate identical primary articulations and when the other conditions in
(13) are met. Assume that when two nodes are subject to merger the
resulting node is their Boolean union. If one of the C-place nodes involved
were to contain a secondary articulation, then, the result of the merger should
contain it too. This prediction is not borne out by the data. Consider the
sentence /i-!bbukd#dadda-s/ 'his eIder brother became blind' (3ms-blind
elder:brother-3s), where two stops on either side of the word boundary must
be pronounced as a single long stop . The fact that the following a is not
pronounced emphatic shows that in the long stop in question, the second
half is not emphatic, since both segments in a CV sequence must agree with
respect to emphasis at the phonetic level (see § 3.6.2.2).
The evidence is unclear in the case of secondary rounding. In /assarg'reka
ad -yi=nn/ 'only the stone anvil is there' (stone:anvil=only AD dernedir),
for instance, it is not clear whether /gw+k/ can only be realized as kWk or
whether it also can be realized as k'k". In this example and others like it
the relevant acceptability judgements are fuzzy or not consistent over time.
Judgements are clear, however, on the following point. The absence of
secondary rounding cannot spread, e.g. in the example above /g w+k/ cannot
be realized as kk, an outcome which would be predicted if Fusion were a
rule spreading to the left the C-place node of the second consonant in
(13), or any node dominating C-place.

6.3.3.3. Restrietions on fusion

If Fusion were to apply in any string meeting the conditions in (13),
SIBLING-RELEASE (11) would never have an opportunity to apply and
releases would always be prohibited between sibling stops. Let us now
turn to the restrictions on the operation of Fusion which make release
between sibling stops possible in Imdlawn Tashlhiyt. These restrictions
are of two types. Some involve length, while others have to do with morpho-
syntactic structure. Let us review these in turn.
In sequences of sibling stops the long oral closures created by Fusion
never span more than two timing units, as we shall now see.
First, Fusion ne ver affects adjacent sibling stops one of which is a
geminate. Our reason for believing this to be the case is that release is
always" possible in such sequences, i.e. the realization with an uninter-
 VOWELLESS SYLLABLES 155

rupted closure is in free variation with another in which the first conso-
nant is released. This fact has already been illustrated in our examples in
(7)-(9). In these examples the two consonants are identical except for length;
here are some in which the two sibling stops differ in phonation type. In
lis=ak ggr-nl 'did they touch you?' (int=dat2ms touch-3mp) the medial
cluster can be pronounced k!gg or kgg; in lar tt-dus-n/ 'they are becoming
sturdy' (impf impf-sturdy-3mp) the medial cluster can be pronounced tfd
or Ud. 28
Second, in sequences of three short sibling stops, Fusion never merges
three successive primary articulations into one. Like those discussed in
the preceding paragraph, these sequences may be realized with unbroken
closures, but they always have alternative realizations with releases. In
Iyat#t-tri-t/ 'one star' (one :f bf-star-fs), It#t-t/ may be pronounced either
as trt or as tu. In la-gllid#d=d#i-munl 'the king with whom he came' ,29
Id#d=dl may be pronounced as d 2dd, dd'd or ddd .
As was already pointed out in Chapter 3, Imdlawn Tashlhiyt has two
assimilation rules which are blocked if their operation would create a triply-
linked Root node . In that chapter we introduced a constraint NO-TREBLE
which forbade Root nodes linked to three adjacent X slots. The nodes
merged by Fusion are primary articulations rather than Root nodes, and
the formulation of NO-TREBLE should be made more general:
(9) NO-TREBLE:
A primary articulation may not be associated with three prosodie
positions which are adjacent."
(9) is valid over any string bounded by pauses. Note that what this
restrietion prohibits is a superlong primary articulation, not a superlong
closure. Consider again lis=tt#t-utl (see (7)b) , where Itt#t/ may be pro-
nounced tfr or tu. For the sake of convenience we reproduce here as (20)
the structure of the underlying sequence already displayed in OO)b.

27 ' " unless there has been an assimilation in phonation type or in secondary labiality.
Release is incompatible with assimilation between sibling stops, see DE (1996a : 385-388).
28 Contrary to the generalization stated at the beginning of this paragraph, there are a few
contexts in which Fusion merges a geminate with a simplex sibling. For instance, in 2nd
person imperfective forms, the prefix sequence /t-tt-/ must be realized simply as tt. Similarly,
/dd-t/ must be realized as tt in /t-bidd-t/ 'you stood up' (tbitt) . In all such cases, however,
Fusion involves the loss of a skeletal slot, and its outcome abides by NO-TREBLE (see below).
29 u-king withedir 3ms-come .
30 As formulated in (19) , NO-TREBLE is too restrictive, for it excludes languages in
which a nasal borrows its primary articulation from a geminate , for example /n+bb/ > mbb.
A more adequate formulation is given in DE (l996a: 383).
156 CHAPTER SIX

(20) x x x
<.>
Root
I
Root
I I
t t

In view of (19) Fusion cannot operate in (20), which surfaces unchanged

at the end of the phonological component. Since SIBLING-RELEASE
(11) is optional, it may apply, hence tft, or it may not, hence tu, i.e. a
continuous closure spanning three timing slots. When Itt#tI is realized as
ttt, then, what one observes from a phonetic point of view is one contin-
uous closure, but that single closure is the manifestation of two successive
primary articulations, as that notion was defined in (14). NO-TREBLE (19)
is not a constraint on the output of phonetic implementation, but on that
of the phonological component."
When three underlying short sibling stops stand in a sequence which is
not a subsequence of a longer string of siblings, the first stop and the second
cannot be released at the same time, e.g. we have not been able to find
any instance in which IHt+tI could be pronounced as frr.
If Fusion applied in all environments and if NO-TREBLE (19) was the
only restriction on its operation, release between sibling stops would only
be found adjacent to a closure spanning two timing slots, as when It#t-tI
is realized as tft in lyat#t-tri-tI 'one star' (see supra). This prediction comes
close enough to the mark. At the phonetic level, release between short
sibling closures is disallowed in most environments, and we have yet to
determine the exact range of environments which allow it. One can nonethe-
less already state one important generalization: releases between sibling
stops which are both underlyingly short are disallowed anywhere within
a 'stretch '. What we call a stretch is astring made up of a lexical morpheme
and its dependent suffixes and enclitics, in other words a stretch is a Pword
minus its prefixes. The generalization just stated can be broken down into
three subcases, which we now quickly review.
Releases between sibling stops which are both underlyingly short are
disallowed, (i) between a suffix and a preceding morpheme , and (ii) between

31 We account for the optionality of release in sequences such as /tt#tJ by assuming that
Fusion (13), which is an obligatory rule, is blocked by NO-TREBLE (19), and that SIBLING-
RELEASE (1 I) is option al. A reviewer has pointed out an alternative: Fusion would be
optional in sequences such as /tt#t/ , and SIBLING-RELEASE would be obligatory in all
contexts . We do not retain this alternative because it would force us to give up restriction
(19), which enables us to link up the distribution of releases with the blockage of rules of
complete assimilation (see Chapter 3) and with certain facts about the lexical distribution
of adjacent identical consonants, on which v. below in § 6.4.1. Yet other evidence in support
of (19) is provided by the behavior of the causative prefix before sibilant-initial kernels, v.
DE (1996a: 381-385).
 VOWELLESS SYLLABLES 157

an enclitic and a preceding morpheme, and (iii) inside morphemes. Case (i):
the clusters at the end of the words below must be pronounced as gemi-
nates no matter in what environment: li-kzin-nI (nn I *n2n) 'puppies' , /t-fllt-t/
(tt I *ft) ' you escaped', lar t-slum-m/" (mm I *m2m) 'you (p) are eating
(something powdery)', /t- !srd-t/ (tt I *d2t) 'you sued' . Case (ii): the same
holds for the sibling clusters in li-srk=k/ (kk I *e k) 'he shares you'
(3ms-share=d02ms), li-!'Yrd=dl (dd I *d 2d) 'he lay down ' (3ms-lie=dir),
li-!krd=tnl (tt I *d2 t) 'he scratched them' (3ms-scratch=d03mp),/t-!umz-t=d/
(dd I *r d) 'you seized' (2s-seize-2s=dir), /i-balakektn/ (kWkWI *ekW) 'he
evacuated you' (Jms-evacuareedozrnp).
Notice that in all the examples above in which the abutting siblings are
nonidentical, they are subject to regressive assimilations of phonation type
and rounding. These assimilations are mandatory between short sibling
consonants which belong to the same stretch.
Case (iii) of the generalization under scrutiny, which concerns tauto-
morphemic sequences, only has one exception, the plural noun It-i-mtd-inl
'loin (cut of meat)', which can be pronounced timt'din or timtdin. In all
the other morphemes with adjacent sibling consonants we have been able
to find, one of the sibling consonants is long, e.g. lttd 'coagulate'."
Fusion always applies within stretches, but not between aprefix and
the following morpheme. Release between stops which are both underly-
ingly short is acceptable in certain cases in that context, e.g. in /t-uskaed
t-tbir-t/ 'the dove came' (trt) (3fs-come=dir bf-dove-fs), in It-ut=t t-tbir-t/
(tr ft) 34 ' the dove struck hirn' (ßfs-strikeedoßms bf-dove-fs), and in
In-nkrl (n2n) ' we got up' and It-!dalbl (rd) 'she prayed' when these words
occur immediately after a pause.
Since there are instances where Fusion does not apply between aprefix
and the following morpheme, it is somewhat surprising to discover that it
applies most of the time at the boundaries between words, e.g. in (12).

6.4. THE ONLY SURFACE VOWELS ARE a, i AND u,

TWO PHONOLOGICAL ARGUMENTS

Some languages of Morocco have a fourth vowel (@) in their phonetic

representations, in addition to the full vowels a, i and u. This is for instance
the case in Rifian Berber and in Moroccan Arabic. We submit that Imdlawn
Tashlhiyt is different in that its only vowels are the full vowels . In the
surface representations of Imdlawn Tashlhiyt, consonants not separated
by a consonant or a full vowel are adj acent. In this section we discuss
two phenomena which are consistent with our analysis of Imdlawn Tashlhiyt

32 Impf 2-eat:impf-2mp.
33 On the others, see § 6.4.1.
34 Other possib1e realizations are ttr! and u'u.
158 CHAPTER SIX

and would be problematie for an analysis in which eonsonants adjaeent in

the underlying representations would in some eases be separated by an
epenthetic vowel in the surfaee forms.

6.4.1. Morphemes with adjacent identical consonants

Positing a rule of fusion which is obligatory inside 'stretches' enables us

to explain certain facts about the distribution of adjacent identieal eonso-
nants inside morphemes. Imdlawn Tashlhiyt has a few morphemes whose
lexical representations contain adjacent consonants which are identical.
Below we list all those that we have been able to find.
(21) a. zmm'm 'write down', xmm'm 'think',jnn 2n 'discern', nqlq
'stare at (while trying to identify)', slf1 'rince', un. 'treat
with caution', !nr?r 'Iiberate':"
b. brd'ddus 'marjoram', lbbnrttr 'struggle', bbrt'tts 'sprawl',
lfr t' ttu 'bat'.
In zmm'm 'write down!', and in zmm'meat 'write down! mp' (=at imper-
ative 2mp), the verb can be pronounced zmm'm or zmmm in free variation ,"
but not zm'mm nor zm2m2m nor zmm. If the lexical representation of this
verb eontains a geminate m followed by a simple m (see (22)a) the location
of the release is aeeounted for by SIBLING-RELEASE (11). SimiIarly,
brd'ddus may be pronouneed brd'ddus or brdddus, not brdd'dus nor
brd2d2dus nor brddus, and we posit the underlying representation in (22)b.
(22) a. X X X X

z
I ""J
m m
I
b. X X X X X X X

t I
r d
I ""J
d
I
u
I
S
All the morphemes listed in (21) contain two identical consonants in a
row, one of which is a geminate ." On the other hand the lexicon of Imdlawn

35 The items in (21)a are all borrowings from Arabic. They all have a free variant with i
after the second consonant, e.g. zmmim, xmmim, fnnin, etc. The free variation between i
and zero is also found in native verbs, viz. in the biconsonantal verbs where both conso-
nants are obstruents, e.g. b(i)dd ' stand up', bbii): 'pound' , kk(i)s ' remove' . In all such verbs
one of the consonants is a geminate.
36 The uninterrupted tripIe m in zmmm sounds longer than the uninterrupted double m in
tllmm 'you (mp) spun' (from It-lIm-ml).
37 In our 1985 article we stated that a simple consonant cannot immediately precede its
geminate counterpart in a lexical entry, see (48)b p. 124. The existence of the items in
(21)b shows that that assertion is false.
 VOWELLESS SYLLABLES 159

Tashlhiyt does not allow a contrast between geminates and sequences of

two identical short consonants, e.g. a contrast between /dd/ and /d+d/, where
the symbol '+' is a shorthand to indicate the presence of two distinct Root
nodes. In languages where it occurs such a gap is usually seen as a con-
sequence of a general prohibition against adjacent identical segments." Such
a prohibition cannot be invoked in the present case, however, in view of
the fact that it is violated in (21). But our Fusion rule provides an expla-
nation. Imagine that at some point in time the grammar of some speakers
contained a lexical entry with adjacent identical short consonants, e.g. /d+d/.
Since the application of Fusion is obligatory inside ' stretches' , Fusion
(13) would make all the occurrences of /d+d/ homophonous with /dd/
(geminate d), and the original /d+dI would be irretrievably lost, i.e, language
learners in the next generation would have no reason to posit anything
else but a geminate in the lexical entry of the morpheme in question. Ath
Sidhar Rifian Berber affords us an interesting comparison on this point.
Unlike the lexicon of Imdlawn Tashlhiyt, that of Ath-Sidhar Rifian Berber
allows a contrast between geminates and sequences of two identical short
consonants, see Dell and Tangi (1992 , 158-160). Consider for instance
the following forms in Ath-Sidhar : (i) li-zemm 'he wrung', from /y- !zmm/;
(ii) ye-smem 'it (dough) rose', from /y-sm+mI; (iii) semm-en 'they rose',
from /sm-rn-n/. Like Imdlawn Tashlhiyt, the Ath-Sidhar dialect has a Fusion
rule, as a result of which a sequence of two identical short consonants is
homophonous with the corresponding geminate, see /m+mI in (iii). Fusion
cannot operate in (ii) because in that form the /m+mI sequence is broken
up by the epenthesis of e, a process which occurs in the syllabification of
consonant clusters in Ath-Sidhar Rifian." In (i) that same process inserts
e before the geminate /mml.
The fact that sequences of identical short consonants are allowed in the
lexicon of one dialect but not in that of the other has nothing to do with
constraints on the underlying forms of morphemes, then. Rather, the dif-
ference follows from the different syllable structures of the two dialects .
Unlike Ath-Sidhar Rifian, Imdlawn Tashlhiyt has no phonological process
of vowel insertion which can break up underlying sequences of conso-
nants and consequently, if the lexicon of Imdlawn Tashlhiyt contained
tautomorphemic sequences of identical short consonants , these would all be
wiped out by Fusion.
In all the morphemes in (21) the identical consonants are noncontinu-
ants. We do not have any explanation for this fact. What about the fact
that there does not exist any morpheme containing a sequence of iden-
tical geminates ? This is probably a consequence of a quasi-gap which has

38 On the Obligatory Contour Principle see McCarth y (1986) , Odden (1988) and refer-
ences therein.
39 On syllabification and epenthesis in Ath Sidhar Rifian Berber, v. below in § 6.5.
160 CHAPTER SIX

nothing to do with constraints on sameness in adjacent consonants: mor-

phemes which contain adjacent geminates are very rare in Imdlawn
Tashlhiyt.

6.4.2. Regressive devoicing

Like some other Tashlhiyt dialects in the High Atlas, that spoken in Imdlawn
has a process of regressive devoicing in obstruent clusters. Devoicing is
mandatory in some cases and optional in others. We give examples below
in (23). The consonants subject to the assimilation are enclosed between
brackets for the sake of conspicuousness. Symbols separated by a slash
indicate that devoicing is only optional. The forms in parentheses are mor-
phologicaIly related words in which the voiced obstruents surface intact. 40
(23) a. /s-UZf/4 1 zzu[s]f 'discover!' (zzuzuf)
b. /rgs/ r[k]s 'hide!' (tirggas)
c. /a-!zdda/ lalsltta? '100m' (!zd)
d. /a-sds/ as[d/t]s 'trough' (isdas)
e. /l-rzq/ rr[z/s]q 'divine gift' (larzaq)
f. /l-fndqq/ lfn[dlt]qq 'caravansary' (lafnadiqq)
Regressive devoicing operates across word boundaries as weIl as inside
words. The present discussion is limited to the sequences in which both con-
sonants belong to the same kernel, the only ones for which our data is
sufficient to aIlow us to make generalizations with confidence. We will
use the expression ' vcd-vls sequence' to refer to such sequences. Devoicing
is mandatory in some kernels and optional in others, but for the purpose
of OUf argument the important fact is that it operates in aIl vcd-vls
sequences." Assuming that regressive devoicing cannot operate across a
vowel, this fact shows that the obstruents in a vcd-vls sequence are always
adjacent at the surface level, no matter the context in which the sequence
occurs. This would be difficult to explain if the surface representations of
Imdlawn Tashlhiyt contained a fourth vowel e in addition to the fuIl vowels
a, i and u.

40 Here are the glosses for the parenthesized forms: (a) id impf; (b) 'h iding ' (deverbal
noun, pluralia tantum); (c) 'weave! ' : (d-f) id p.
41 On the realization of causative Is-I, v. § 5.4.
42 In Tashlhiyt as in other Berber dialects, Iddl is generally realized as tt in emphatic
morphemes .
43 An exception must be made for vcd-vls sequences beginning with /b/, a consonant which
is in some instances immune to devoicing in Imdlawn Tashlhiyt. In Imdlawn the extent to
which regressive devoicing operates varies with the speakers' age. It is more pervasive in
the language of older speakers such as ME's father. Even in ME's less conservative dialect,
the vcd-vls sequences in which regressive devoicing is only optional all belong to kerneis
which have transparent cognates in the local variety of Moroccan Arabic.
 VOWELLESS SYLLABLES 161

Consider the perfective stern zzusf and the corresponding imperfective

zzuzuj in (23)a. In Is-uzfl the adjacency of Izl and Ifl makes regressive
devoicing possible, hence zzusf The imperfective zzuzuj is derived from
Is-uzfl in a regular fashion, i.e . a copy of the vowel is inserted in the final
cluster," thus making Izl and Ifl nonadjacent, and regressive devoicing
impossible. zzusj is trisyllabic (z.zu.sf) and our analysis implies that the
nucleus of its final syllable is f Suppose for a moment that in zzusj the
last two consonants are actually separated by a vowel e which is somehow
obscured by phonetic implementation, i.e. the perfective stern is actually
zzusef. One would then have to explain why full vowels block regressive
devoicing while e does not. This problem does not arise in our analysis,
in which e does not exist in Imdlawn Tashlhiyt. The comparison with Ath
Sidhar Rifian is instructive: as we shall see below, that dialect has a schwa
vowel, and schwa blocks regressive devoicing.
On the left in (24) below, we list words of Imdlawn Tashlhiyt in which
the voiceless consonant resulting from regressive devoicing has been lex-
icalized. The original voiced consonant can only be recovered by comparing
Imdlawn Tashlhiyt with dialects in which it has been preserved. The forms
on the right-hand side of (24) were culled from Dallet's (1982) dictionary
of At Mangellat Kabyle.
(24) Imdlawn At Mangellat
a. i-[x]ss 'bone' i-vess 'id'
b. t-a-lklsrar-t 'kneecap' t-a-g resrir-t 'id'
c. t-a-Iklssul-t 'churn' a-gessul 'bellows'
d. i-[t]qqi 'clay' i-deqqi 'id'
e. t-a-Itlkmi-t 'palmfull' t-a-dakrem-t 'id'
f. bi[k]s 'gird!' ebges 'id'
g. lt-a-Ix'Tsmar-t 'chin' !t-a-)'wesmar-t ' id'
h. a-[s]kka 'tomorrow' a-zekka 'id'
i. a-ltx"]» 'colostrum' a-dves 'id'
j. a-[t]fl 'snow ' a-dfel 'id'
k. !t-a[t]sa 'laughter' !t-a-dsa 'id ' 45
The lexicalization of devoiced consonants has all but wiped out the
vcd-vls sequences from the Berber stratum in the lexicon of Imdlawn
Tashlhiyt. Items like (23)a-c, which belong to that stratum, only amount
to a handful.46

44 On vowel insertion in the imperfective, see § 5.2.

45 While the initial vowel belongs to the stern in Imdlawn, it is an augment in At Mangellat.
46 The Berber stratum also contains vestigial pairs which do not fit anymore into produc-
tive alternation patterns, e.g. a-bggas 'bell' / biks ' gird !' (see (24)f) . The imperfective stern
of biks is tt-bikis, not tt-bigis, which shows that in biks the velar consonant is lex ically
voiceless in present-day Imdlawn Tashlhiyt.
162 CHAPTER SIX

Although regressive devoicing operates in all vcd-vls sequences, in the

examples in (23) and (24) we have chosen mainly forms in which the
consonant subject to devoicing is followed by a word-final consonant or
by CCV. Here is our reason for doing so. In those works on Tashlhiyt
which have schwas in their transcriptions, consonants which occur in these
environments are as a rule followed by a schwa, as they are in the litera-
ture which deals with Tamazight, Rifian or Kabylie Berber." For someone
claiming that Imdlawn Tashlhiyt actually has a fourth vowel e, the conso-
nants subject to devoicing in our examples in (23) and (24) are the most
likely to be separated from the following consonant by a schwa, and yet
the fact that they devoice shows that they are not.
In (23) and (24), the consonant which triggers devoicing is a nucleus
except in (24)j,k, where it is an onset. We give the syllabic parses of these
forms below in (25). The forms on the left are those in (23) and those on
the right are those in (24) . The vcd-vls sequences are in boldface for the
sake of conspicuousness.
(25) Imdlawn (23) Imdlawn (24)
a. z.zu.sf i.xss
b. r.ks ta.ks.ra.rt
c. a.st.ta ta.ks.su.lt
d. as.ts i.tq.qi
e. rr.sq ta.tk.mit
f. l.fn.tqq bi.ks
g. ta.x'ts.ma.rt
h. a.sk.ka
i. at.x'ts
j. at.fl
k. tat.sa
Regressive devoicing is similarly heedless of syllable structure in Haha
(haha) Tashlhiyt, a dialect in which simple noncoronal stops spirantize."
When preceding a voiceless consonant, Ib/ is realized as fand /g/ as x (a
velar) in all environments, for instance the noun meaning 'vagina', which
is i-bssi in Imdlawn, is i-jSsi in the Haha area. It is not reckless to assurne
that syllable structure is identical in the two dialects, in view of their great
similarity. Since the syllabic parse of /i-bSsi/ is i.bs.si, we see that in the
second syllable of Haha i.js. si the labialonset has devoiced under the
influence of a following nucleus. The first column in (26) contains other
similar Haha words provided to us by R. Ridouane, who is a native speaker

47 See e.g. Aspinion (1953).

48 On Haha Tashlhiyt, v. Ouakrim (1993) and Ridouane (1999). The lauer discusses Haha
spirantization in some detail.
 VOWELLESS SYLLABLES 163

of Haha Tashlhiyt. The words in the second column are their counterparts
in Imdlawn.
(26) Haha Imdlawn
t-a-jxlssul-t t-a-[k]ssul-t 'chum'
a-[f]hri a-[b]hriy ' sailor'
t-a-[f]qqal-t t-a-Ib]qqal-t 'grocery'
i-[f]qqa i-[b]qqa 'he stayed'
i-[f]ssi i-[b]ssi 'vagina'

6.5. EPENTHETIC VOWELS IN RIFIAN BERBER

In accord with our main goal in this chapter, which is to show that the
short voiced vocoids which one hears in Imdlawn Tashlhiyt are not vowels
(i.e. syllable nuclei), we now contrast Imdlawn Tashlhiyt with Ath Sidhar
Rifian, a dialect which does have epenthetic vowels. We shall see that in
Ath Sidhar Rifian some occurrences of @ are genuine vowels, while others
are transitions, like the VTVs of Imdlawn Tashlhiyt.
As far as can be ascertained from the literature, the syllable structure
of Rifian is rather similar to that of Tamazight and to that of Kabylie Berber,
which are also reported to have bona fide epenthetic vowels. The fol-
lowing discussion will allow us to highlight certain features by which
Tashlhiyt differs from these other dialects. It will provide evidence that
Berber dialects with very similar morphologies may differ significantly in
their syllable structures when the domain of syllabification is the stern or
some larger unit. This evidence should dispel any suspicion that syllable
structure is basically the same in all the dialects of Berber, an impression
that could be suggested by a casual reading of the literature, where the
distribution of the unstable vowel is more or less the same in the tran-
scriptions of Tashlhiyt and in those of other dialects of Berber."
The variety of Rifian Berber described here is that spoken in the village
of Bag'tg''ar, in the Ath Sidhar area, about 20 kilometers to the north-west
of the city of Nador, in north-eastern Morocco." All our data on Ath
Sidhar Rifian were gathered during joint work by one of us (PD) and
Oufae Tangi, a native speaker. Oufae Tangi has written her Doctoral dis-
sertation on the phonology of her native tongue (Tangi 1991) and she has

49 Compare for instance Basset and Picard (1948) for Kabyle, Penchoen (1973) for
Tamazight, and Aspinion (1953) and Destaing (1920) for Tashlhiyt.
50 For discussions of nearby dialects with rather similar phonological systems, see Chami
(1979) , Cadi (1981) and Chtatou (1982).
164 CHAPTER SIX

subsequently co-authored two articles on its phonology, Dell and Tangi

(1992, 1993), henceforth cited as DT (1992) and DT (1993). 51
Note weil: all our claims about Ath Sidhar Rifian, in particular those
about underlying representations, are based on analyses in which Ath Sidhar
Rifian is considered on its own terms, independently of its similarities
with other Berber dialects.

6.5.1. The basic pattern [or vowel epenthesis

Like Imdlawn Tashlhiyt, Ath Sidhar Rifian has two underlying glides
Iy, wl and three underlying vowels la, i, ul. The latter will be referred to
as 'full vowels' . Besides glides and full vowels, the surface forms of Ath
Sidhar Rifian also contain short voiced vocoids whose distribution is pre-
dictable and which sound very much like the VTVs of Imdlawn Tashlhiyt.
But whereas the VTVs which are heard in Imdlawn Tashlhiyt are not
segments, some of the short voiced vocoids of Ath Sidhar Rifian are
epenthetic vowels, as we shall see below. Like the VTV s of Imdlawn
Tashlhiyt, the short voiced vocoids of Ath Sidhar Rifian vary in vowel
quality depending on the nature of the neighboring sounds. We will abstract
away from these contextual variations and uniformly note the short voiced
vocoids as '@'. From now on let 'SVV' stand for 'short voiced vocoid'.
The SVVs of Ath Sidhar Rifian fall into two categories. The bulk ofthem
are vowels inserted in order to syllabify sequences of consonants; the
remainder are nonsegments like the VTV s of Imdlawn Tashlhiyt. Let us first
give examples of the VTVs which are vowels, i.e. syllable nuclei.
Unlike in Imdlawn Tashlhiyt, in Ath Sidhar Rifian the location of a
VTV may be the sole feature which distinguishes two expressions (i.e. words
or sequences of words). Ath Sidhar Rifian has for instance a contrast
between [C@C] and [CC] before a pause. This contrast is illustrated in (27) .
(27) a. zh@ö 'be strong' z-z@hö 'strength'
b. bh@e 'investigate' r-b@M 'investigation'
c. xr@q 'be born' s-s@rk 'wire'
d. nq@s 'diminish' n-n@fs 'breath'
e. nh@s 'bite greedily' r-m@sk 'musk'
f. xn@s 'dodge' ss-@ns 'spend the night'
The words on the left-hand side of (27) aB have underlying representa-
tions of the form ICCC/.52 It is argued in DT (1992, 1993) that in Ath Sidhar

51 Oufae Tangi' s father and mother are from the Ath Sidhar area and Berber was her first
language. She uses Berber with her parents and with other members of her family of their
generation , some of whom are monolingual. She uses Arabic with her sisters and the people
outside her family.
52 They are bare aorist sterns. As in Tashlhiyt, such sterns are used as 2s imperative forms.
 VOWELLESS SYLLABLES 165

Rifian schwas are inserted into consonant clusters to supply syllable nuclei;
if the underlying string does not contain enough vowels, some are supplied
by epenthesis. Let us use 'e' to represent the epenthetic vowel without
committing ourselves as to its featural make-up. It will be seen below that
depending on context e is realized as @, as a voiceless counterpart of one
of the voiced vocoids for which '@ ' is a cover symbol, or as the syllab-
icity of an adjacent sonorant.
Note well the difference between '@' and 'e'. As elsewhere in this book,
'@ ' stands for a voiced vocoid, i.e. an object whose presence in an expres-
sion can be ascertained by inspecting tokens of that expression. On the other
hand, ' e' represents a vowel (a nonconsonantal syllable nucleus) which is
posited to explain the distribution of [@] and other phonological regular-
ities to be discussed below. Our transcriptions of Ath Sidhar Rifian which
employ the letter 'e' are broad phonetic transcriptions akin to the phonemic
transcriptions of structuralist phonology .
Provided certain word-final consonants are marked as extrametrical (see
below), the following procedure makes predictions which are basically
correct."
(28) RIGHT-TO-LEFT SCAN:
Scanning the Pword from right to left, rewrite as CeC any CC
string which is not immediately followed by a vowel. Each step
in the scan must take as its input the output of the previous
step.
For the stem meaning 'be strong' in (27)a, for instance, RIGHT-TO-LEFT
SCAN rewrite s Izhöl as iheo in a single iteration . In /ü-xns-m/ 'you (mp)
dodged' the procedure has to apply twice . It first changes /Bxnsm/ into
lexnsern/; it then takes lexnsern/ as its input and changes it into /Bxensem/.
The role of epenthesis is to allow strings of consonants to be fitted into
syllables which are maximally CVC:,54 and consequently the epenthesis
of Ath Sidhar Rifian is reminiscent of that in Yawelmani (Archangeli
1991) or in Palestinian Arabic (Abu-Salim 1980, 1982). The distribution
of e in our transcriptions of Ath Sidhar Rifian resembles that of 'unstable
vowels' in various works on other dialects of Berber, and also in some
dealing with Moroccan Arabic."
Schwa is never inserted between the two es of a geminate. When the
formulation in (28) would lead to such an insertion, RIGHT-TO-LEFT
SCAN disregards the last C in the geminate and moves leftward by one

53 RIGHT-TO-LEFT SCAN is almost identical with the epenthesis rule proposed in Saib
(1976: 127) for the Ayt Ndhir variety of Tamazight. On that rule, see Hyman (1985: 68).
54 On how initial clusters fit in with this general scherne, see DT (1992).
55 See the references in note 49 for the works on Berber. For those on Moroccan Arabic,
the references will be found in Chapter 8.
166 CHAPTER SIX

segment; Issubbs-nl ' they had a fight' is pronounced ssubbSen, not

*ssubebSen. A form like ssnen 'they know' (lssn-nI know-3mp) illustrates
the different behaviours of geminates and sequences of identical consonants:
epenthesis breaks up the sequence In-nI but not the geminate Issl (* sesnen).
In our discussion of transitional vocoids in Imdlawn Tashlhiyt we stated
that in that dialect a short voiced vocoid can never be heard (i) between
voiceless consonants or (ii) between homorganic noncontinuants which
differ in sonorancy (v. (1)-(2) in § 6.3.1). Because of epenthesis, Ath Sidhar
Rifian has SVVs which occur in these contexts. SVVs occurring between
voiceless consonants are found in b1i@fJ, nq@s and n1i@s in (27)b,d,e.
As for case (ii), consider for instance the following pair:
(29) a. Imdlawn ssnett [s.nt:"] Issn=tt/
knowedoßfs 'know her!'
b. Ath Sidhar ssn-etet [s:n@t h] Issn-8=<t>1
know-imper.Zpedoßfs 'know (p) her! '
In Ath Sidhar Rifian the imperative 2nd person plural suffix 1-81 and the
3rd person singular object pronoun 1=<t>1 merge into a geminate Itt/. 56 When
articulation moves from n to tt in the Imdlawn Tashlhiyt form in (29)a
the coronal occlusion must be kept in place; pronunciations with a voiced
vocoid between n and tt are rejected outright. An intervening SVV is clearly
heard in the Rifian form (29)b, on the other hand.
Ath Sidhar Rifian has words which do not contain any vowel, or even
any voiced sound. The imperative 2s of ' eat' , for instance, is 155/. In a
normal pronunciation this word can anly be realized as 55 (i.e. [5:]) and
it can stand on its own as a complete utterance. Ath Sidhar Rifian has
other similar words : Ikk/ kk 'pass!', Izzl ii 'let!', Iggl gg 'da!' , IgWgWI
gWgW 'knead!' . The absence of any vowel in the le :1 words is predicted
by (28).

6.5.2. e devoicing and e absorption

Our broad transcriptions of Ath Sidhar Rifian do not portray the terminal
representations of the phonological component, although they are not far
removed. These tran scriptions adequately reflect the distribution of syllable
nuclei in the dialect, but they do not mirror faithfully what one actually
hears. If they did, one would expect every instance of the letter 'e' in
them to represent a voiced vocoid in the pronunciation, but this is not so.
e is devoiced or deleted in certain cantexts and it is absarbed into a neigh-
boring sonorant in others. We now turn to these two phenomena.

56 The angled brackets around the do3ms c1itic indicate that it is extrametr ical; in view of
this it is disregarded by RIGHT-TO-LEFT SCAN, see below.
 VOWELLESS SYLLABLES 167

e devoices or deletes between voiceless consonants. This devoicing or

deletion is optional in the last syllable of a word and obligatory elsewhere.
For instance in isolation /8-kasf/ 'she guessed' may be pronounced either
[8kas@f] or [Bkasf] . In wh at FD hears as [8kasf] we do not know whether
[s] and [f] are adjacent or whether they are separated by a vocoid which
is the voiceless counterpart of @. Let us use capital 'E' to represent the
occurrences of e which are subject to devoicing/deletion between two voice-
less consonants. Further research will be necessary to determine wh ether
'E' is actually a voiceless vocoid or the absence of any segment. From
now on we use the expression 'e devoicing' without committing ourselves
on that point. Two other examples of e devoicing are given below in (30).
The facts about each example are laid out on three lines. The output of
RIGHT-TO-LEFT SCAN and the underlying representation are given on the
first line. The other two lines show alternative pronunciations of the form
under consideration. No other pronunciations are acceptable."

(30) ttetteß /tt -tt-ü/ impf-eat:impf-imper2p

h
a. [t t:I8], i.e. ttEtt@8
a'. [t ht8], i.e. ttEttE8
fekkett /fkk-8=<8>/ release-imper.Zpedoßrns
b. [fk .rt:"], i.e. fEkk@tt
b' . [fk:ht:h], i.e . fEkkEtt

Each form contains two occurrences of e. The occurrence in the last syllable
is devoiced in one variant but not in the other: devoicing is optional in word-
final syllables. Devoicing is obligatory elsewhere, and consequently the
occurrence of ein the first syllable is devoiced in both variants. The second
variant in each pair in (30), which is voiceless throughout, can stand on
its own as an acceptable utterance. Such variants are commonplace among
the spontaneous pronunciations which we have recorded. In word-final
syllables, where the devoicing of e is optional, the choice between the
two variants is not random . In prepausal words it seems that the voiced
variant only occurs when the intonation requires a high pitch on the syllable
in question, or when the speaker wants to place special emphasis on the
word."
The devoicing of unstable vowels between voiceless consonants has been

57 The meanings of the two examples are 'eat! p' and 'release hirn! p' .
58 In Moroccan Arabic , according to Heath (1987 : 184), schwa deletion in word-final
syllables is blocked by 'Iist intonation ', and in Japanese high vowel devoicing is blocked
'when a final syllable in the devoicing environment must carry a rising intonation ' (Vance
1987: 51).
168 CHAPTER SIX

ascribed to other Berber dialects" and to Moroccan Arabic." There is an

obvious resemblance between the facts discussed here under the heading
'e devoicing' and those presented in Chapters 8 and 9, which will lead us
to formulate for Moroccan Arabic a rule of schwa epenthesis which on1y
operates at the end of Intonational Phrases. More empirical work is needed
before one can characterize precisely the similarities between the two lan-
guages in that area of their phonology.
Should subsequent research on Ath Sidhar Rifian show that 'e devoicing'
is actually adeletion process, we would not abandon our syllabification-
cum-epenthesis analysis or restriet epenthesis to word-final CC clusters,
for there is independent evidence that e occurs at some intermediate level
of representation to allow the syllabification of voiceless CC sequences
which are not word-final. This evidence is provided by fusion, a process
which is sensitive to the presence of e.
A sequence of two identical simplex consonants is homophonous with
the corresponding geminate. Whereas in ne-nbes (In-nb sI) 'we teased' one
must pronounce [n@n], i.e . two short n sounds separated by a short vocoid,
in nnebsiii 'we teased hirn' (ln-nbs=i81 lp-teaseedoßms) one must pro-
nounce a long n homophonous with the realization of a geminate n. When
the adjacent consonants both belong to the set /8 0 t d/, their fusion yields
tt or dd depending on the voicing of the second consonant in the cluster,?'
e.g, /8-01 yields dd in ddarrast (l8-orraz-<8>/), which is the bound form
of {}-a-darras-t 'weaver, f ' .62 Fusion is blocked by an intervening e, e.g.
it cannot occur in 18-obib-<8>1 {}edbiJt, the bound form of the feminine form
of a-dbib 'physician'. Similarly, fusion occurs in ttelred (l8-tl~-0/) 'you
disappeared from horne' whereas it is blocked in 18 -tl~1 {}etle r 'she
disappeared from home' . Since it is voiceless and is furthermore preceded
by a voiceless fricative, the vocoid which occurs between () and tin Hetle r
cannot be perceived directly, at least by FD, to whose ears the word sounds
like [8tlc~] , but its blocking effect on fusion is evidence that it is present
at the level of representation which is relevant for the fusion process.
Devoicing is not the only phenomenon which obscures the surface
distribution of epenthetic e. Another such phenomenon is the absorption

59 See Basset and Picard (I 948: 9), Mitchell (1957: 197-198), Penchoen (I973: 10,94).
Extensive data elicited from Fouad Saa show that a situation similar to that just described
also prevails in the Figuig dialect, on which see Kossmann (I994) and Saa (I995) .
60 E.g. HarreIl (1962a), Mitchell (I993 : 62, 64), Shoul (1995: 208).
61 8 and () do not have a geminate counterpart. Historically they derive from simplex t
and d and still alternate with them, but they must be considered independent phonemes .
One of the reasons for this state of affairs is a massive influx of Arabic loans with unspi-
rantized t and d.
62 The corresponding masculine noun is a-oarraz la-orrazl. As a rule Irrl surfaces as arr.
Irl is realized as r, ar or a depending on the context. See DT (1993) for a detailed discus-
sion of how these alternations link up with syllabification.
 VOWELLESS SYLLABLES 169

of e by a neighboring sonorant. The previously cited ttelfeo (l8-tl'l-o/) is

actually pronounced [t:l'lco]; going from the articulation of tt to that of I
is done through lateral plosion and the phonetic reflex of el is a syllabic
I. The absorbing sonorant follows e in some cases, as in the previous
example, and it precedes e in others, as in the highlighted sequence in the
following example:
(31) mayemmi=t nedh-en /maymmi=t ndh-nI whyedoßfs lead-3mp
'why did they lead her?'
In (31) the sequence t#ned is realized as [tnd], that is, with an uninter-
rupted coronal closure which begins with t and ends with the release of
d. The two examples of absorption just given might lead one to conjec-
ture that e is absorbed by a sonorant (R) whenever the sequence eR or Re
is adjacent on both sides to noncontinuants wh ich are homorganic with
the sonorant. But absorption does not always occur in such circumstances,
witness the fact that in sbeddent 'they made her stand up' (ls-bdd-n=<t>/
cau-stand-Smpedoßfs) the sequence ddent can only be pronounced [d:@nt];
it is incorrect to pronounce [d:nt], with nasal plosion in dd and a syllabic
n. Absorption requires further study/"
The devoicing of e between voiceless segments and its absorption by a
neighboring sonorant result in pronunciations which lack a SVV in places
where RIGHT-TO-LEFT SCAN predicts the occurrence of e. Ath Sidhar
Rifian also has words in which a SVV can be heard inside clusters in
which RIGHT-TO-LEFT SCAN does not insert e. These supernumerary
SVVs always occur next to a voiced consonant and we assurne that they
are like the VTVs of Imdlawn Tashlhiyt: they are not separate segments,
but mere transitions between consonants. Two examples are given below
in (32) . Each line shows (I) the underlying representation, (11) our broad
transcription, and (III) what is actually heard.
(32) I 11 III
a. /y-fqö -n/ ifeqöen [ifq@o@n] 'be concerned, prt '
b. /y-hrs/ yehres [y@h@r@s] 'he fell siek'
In (32)a one hears a SVV between q and 0, which are adjacent in the
output of RIGHT-TO-LEFT SCAN; similarly, in (32)b a SVV is heard
between hand r . Note that the realizations of the prefix /y-/ are predictable
in terms of the output of RIGHT-TO-LEFf SCAN, rather than of the vocoids
which one actually hears. As a rule, that prefix is realized as i- before CV
and as y@- before Cc. 64 In [ifq@o@n] ((32)a), because of e devoicing,

63 Absorption of an unstable vowel by a neighbouring sonorant has been ascribed to other

Berber dialects, see Mitchell (1957: 194) and Penchoen (1973: 10, 94), and 10 Moroccan
Arabic, see Milchell (1993: 63, 72) and Heath (1987: 249-253).
64 The glides of Ath Sidhar Rifian are discussed in detail in DT (1992).
170 CHAPTER SIX

no vocoid is heard between fand q. And yet the realization i- of the prefix
indicates that f is indeed an onset. In the same line of thought the realiza-
tion [y@-] of /y-/ in [y@h@r@s] (32)b shows that although the following
kernel sounds as though it began with CV, it actually begins with CC .
FD cannot hear any difference between those SVVs which are realiza-
tions of e and those which are transitional vocoids. But it is noteworthy
that Oufae Tangi's transcriptions of her language consistenly record the
former and ignore the latter, and that she has great difficulty in perceiving
the latter. The distinction between the two kinds of SVVs is the same as
that made by HarreIl (1962a) between 'major transitions' and 'minor tran-
sitions' in his seminal paper on Moroccan Arabic/" More data on transitional
vocoids in Ath Sidhar Rifian can be found in DT (1992).

6.5.3. Final CC clusters

Returning to the examples in (27), let us now consider the words in the
right-hand side column. Except for ss@ns in the last line, they are nouns
borrowed from Arabic." The source nouns in Moroccan Arabic also have
a C@CC shape. In § 4.1 we pointed out that when Imdlawn Tashlhiyt
borrows from MA it neutralizes the MA distinction between CC@C and
C@CC words. There is for instance no contrast in Imdlawn between the
reflex of the MA verb ih@d 'be strong' and that of the MA noun ih@d
'strength', witness the homophony of the two Imdlawn Tashlhiyt forms in
(33):
(33) izzd /i-s-zhd/ 'he strengthened' (3ms-cau-be:strong)67
iHd /iel-zhd/ 'to the strength' (toel-strength)
The pairs in (27) show that unlike Tashlhiyt, Ath Sidhar Rifian has the
phonological means for doing justice to the distinction between CC@C
and C@CC in MA.
If RIGHT-TO-LEFT SCAN (28) is responsible for the vowel in the
C@CC kernels in the right column of (27) as weIl as in the CC@C kernels
in the left column, the two kinds of kernels must be distinguished in some
way in the lexicon. Let us assurne that the lexical representation of each one
of the morphemes on the right-hand side of (27) contains a special mark
which forces RIGHT-TO-LEFT SCAN to 'skip' its final consonant when

65 V. also Levin's (1987) distinction between epenthesis and excrescence. The facts of
Ath Sidhar Rifian suggest that when epenthetic vowels and excrescent vocoids coexist in
the same language, they need not have different vowel qualities.
66 As in Tashlhiyt, such nouns begin with aprefix lI-I, see § 2.5.3.1. The prefix assirni-
lates to a following coronal; otherwise it surfaces as r, as do most occurrences of simplex
111 in Ath Sidhar Rifian. On Irl and nJ in Ath Sidhar Rifian see DT (1993).
67 The causative prefix assimilates to a following sibilant, v. § 5.4.
 VOWELLESS SYLLABLES 171

that consonant is word-final. We shall use ang1ed brackets to indicate those

segments which are lexically marked as 'extrametrical' .68 In (27)a, for
instance, the underlying representations of zh@o is simply Izhöl whereas
that of the kerne1 in z-z@ho is Izh<ö>l. Only obstruents can be extra-
metrical in Ath Sidhar Rifian." Verbal kernels marked for extrametricality
are only a handful, whereas nouns so marked are numerous. They are all
recent loans from Moroccan Arabic .
In Chapter 8, when we discuss the contrast between CC@C and C@CC
kernels and other similar contrasts in MA, we will not resort to extramet-
rical consonants. Our aim in the present section is not to present an analysis
of Ath Sidhar Rifian which could be compared with our analyses of Imdlawn
Tashlhiyt and MA, but to present facts which suggest that the syllable
structure of Tashlhiyt differs in important ways from that of other Berber
dialect s such as Rifian. RIGHT-TO-LEFT SCAN and extrametricality are
but convenient expository devices which allow us to give a compact pre-
sentation of the facts of Ath Sidhar Rifian. A point-by-point comparison
of syllabification in Berber and in MA within a unified framework will have
to await further research. Our goal here is to lay some of the empirical
groundwork for such a comparison.
Ath Sidhar Rifian has other contrasts between [C@C] and [CC] besides
the type illustrated in (27). We now present two of these.
In Ath Sidhar Rifian, feminine singular nouns fall into two categories,
le-z-<e>1 and II-Z-e/, where Z stands for the kernel. ?" Examples of the
first type are O-azzu-O ' search' and O-a-teJfan-t (le-a -tffah-<e>1) 'apple' :
an example of the second type is arrilteii (l1-rih-e/) 'odour' . 71 No SVV
may occur before the suffix in nouns of the first type. The suffix shows
up as t when the kernel ends in a consonant, and furthermore the conso-
nant interacts in various ways with the suffix, e.g. Im! becomes a coronal
and obstruents devoice, v. the diminutive of asrem (la-s1m!) 'fish ', which
is Oasrent (/e-a-slm-<e>/), and Biggest (/eiggz-<e>/) 'tattoo' (plural Oiggaz) .
In arribeii (Zl-rih-B') , our example of the II-Z-el nouns, the epenthetic e
can be heard as a SVV when the e devoicing, which is optional in word-
final syllables, does not occur. The presence of e prevents any interaction
between the suffix and the final consonant of the kernel, e.g. Iml does not
become a coronal in nnifmeO (ll-ni'lm-el) 'food' , and lei does not become

68 On extrametricality, see e.g. Hayes (1995) and references therein.

69 Basing themselves on the limited data in Tangi (1991), DT (1992 : 134) stated incor-
rectly that only coronal obstruents can be extrametrical. Subsequent work with Oufae Tangi
has turned up nativized loans like s-s@rk ((27)c) and r-rn@sk ((27)e), which end with non-
coronals.
70 Almost all /l-Z-61 nouns are Arabic loans, but Arabic loans are also found among the
16-Z-<6>1 nouns. 8-a-teJfah-t in the text below is a case in point.
71 The initial 11-1 assimilates to the following coronal , hence Irrl, which is realized as arr.
172 CHAPT ER SIX

t.72 Imdlawn Tashlhiyt also has two classes of feminine singular nouns,
It-Z-tl and II-Z-t/, most of the latter borrowed from Arabic. But the fs
suffix I-tl of Imdlawn Tashlhiyt nouns has the same phonological behav-
iour in both types of nouns. For instance voiced obstruents optionaIly
devoice when they immediately precede a voiceless coronal obstruent, and
suffixal I-tl triggers devoicing in both classes of nouns, e.g. /t-a-mzdav-t/
'inhabitant (f)' can be realized as tamzdavt or tamzdaxt - we shaIl write
tamzda[ y/x}t for short. Similarly /t-a-sbbav-t/, the feminine form of a-sbbav
'dyer' , is pronounced tasbba[ y/x}t, and /l-sbav-t/ 'paint'73 is pronounced
ssba[ y/x}t.
Another source of minimal pairs distinguished by the presence of @ in
Ath Sidhar Rifian is the contrast between suffixes and clitics after the
verb. Of the four suffixal PNGs which are obstruents , three are not extra-
metrical: 1-)'1 'I s', 1-01 '2s' and I-ei 'imper2p' .74 On the other hand, the
four clitics which are obstruents are aIl extrametrical: lei do3ms, Itl do3fs,
Id/ dir and Isl do2ms. The difference between the imperative 2p PNG I-ei
and the direct object 3ms pronoun 1=<e>1 gives rise to contrasts such as
the foIlowing.
(34) a. sivef Isi)'-el light-imper2p 'light! p'
b. cI>sixe Iwsi-)'=<e>1 give-l s=do3ms 'I gave it'
As indicated by our transcriptions, it is acceptable to pronounce a SVV
before the final consonant in (34)a but not in (34)b. Moreover the l s PNG
1-)'1 can devoice in (34)b , as is the case whenever an obstruent immedi -
ately precedes a voiceless coronal obstruent, whereas it must be voiced in
(34)a.
Imdlawn Tashlhiyt lacks a similar contrast. There is no evidence that it
has extrasyIlabic consonants. For instance Itl is both the suffixal part of
the 2s PNG and the direct object 3ms pronoun, and the two foIlowing
words are homophonous : It-ut-tl (2-strike-2s) ' you struck' and /t-ut=tl
(3fs-strike=do3ms) 'she struck him' . Both words are pronounced tutt. Or
again consider the foIlowing pair in Ath Sidhar Rifian.
(35) a. ezebOeo le-zbo-ol 2-puIl-2s 'you puIled'
b. eezbedd le-zbo=<d>1 3fs-puIl=dir 'she puIled hither'
In Imdlawn Tashlhiyt the corresponding forms are /t-zbd-t/, whence ti.btt
through fusion and voicing assimilation, and It-zbd=d/, whence ti.bdd
through fusion.
More generaIly, the examples in the present section illustrate a basic

72 Kossmann (\995 : 80) presents similar facts concern ing Ait Said Rifian.
73 Plural ssbayv. The MA source noun is lsbav-a (p !sbay y).
74 The fourth is extrametricaI. It is the final I-ei which mark s the feminine in
le- .. . -m-<e>1 '2fp ' and in l-n«e>1 '3fp' .
 VOWELLESS SYLLABLES 173

difference between Ath Sidhar Rifian and Imdlawn Tashlhiyt in the way
consonants which are adjacent in the underlying representations interact.
How such consonants interact in Ath Sidhar Rifian depends on whether
syllabification has inserted e between them. Because of the left-right asym-
metry of syllabification in Ath Sidhar Rifian, how two consonants which
are contiguous at the underlying level interact depends on the phonolog-
ical make-up of the string to their right within the Pword. In Iy-zö+öl 'he
grew thinner', where lö+öl represents two adjacent occurrences of lö/,
syllabification inserts e between them, which prevents them from merging
into dd , and they remain continuants in the surface form yezded. On
the other hand , in Izö+ö-nl 'they grew thinner' syllabification yields
zed+ d-en, with the two occurrences of löl still adjacent, hence the pro-
nunciation zedden, with a geminated stop resulting from fusion.
In Imdlawn Tashlhiyt, on the other hand, syllabification does not result
in vowel epenthesis. Setting aside a few special cases which are irrele-
vant to the present discussion," in lmdlawn Tashlhiyt, consonants which are
adjacent in the underlying representations are also adjacent at the surface
level . As a first approximation 76 one can say that in Imdlawn Tashlhiyt,
consonants which are adjacent at the underlying level behave in the same
manner with respect to voice assimilation and fusion, regardless of morpho-
syntactic context.

6.5.4. An outstanding issue: syllabification in kernels

One general implication of the preceding discussion is not new, but is worth
emphasizing: two dialects of the same language may have rather similar
morphologies, as is the case for Tashlhiyt and Rifian, and at the same time
they may differ significantly in their surface syllable structures.
The differences between the nonconcatenative processes which shape
kernels in Tashlhiyt and in Rifian are only minor ones . Given that the
syllable structure of a language is always a central ingredient of its non-
concatenative morphology, there is no escape from the conclusion that in
both dialects there are severallevels of representations where syllable struc-

75 V. the i-epenthesis mentioned in note 35. as weil as that presented in DE (1989 : 191-193).
Both are optional and morphologically-governed.
76 The facts which make this statement only a first approximation fall into three cate -
gories : (I) the special behaviour of prefix boundaries with respect to Fusion (v. § 6.3.3.3),
(2) the erasure of one X slot in certain Pword-internal XXX sequences in which all three X
slots are linked to sibling segments (for instance the prefix sequence /t-tt-/ must be realized
as tt when /t-/ is one of the PNG prefixes and /tt-/ is the imperfective prefix, v. DE (1989 :
193), and (3) the operation of Fusion in sequences of three or more simplex consonants which
are siblings, v. the examples illustrating the three-way contrast between t:2t :2t . rt:2t : and t:2rt:
in the text below (9) in § 6.3.3.1.
174 CHAPTER SIX

ture is defined." The most abstract level is that of kernels, where the syllable
structures of Tashlhiyt and Rifian are presumably identical. The main
sources of evidence about syllable structure at the level of kerneIs would
be (i) phonotactic regularities in kemels, and (ii) alternations in the inflec-
tion of nouns for number, in that of verbs for tense/aspect, and in the
derivational morphology. Let us briefly summarize what is presently known
about evidence of either kind.
Regarding the first, Dell and Jebbour (1991) tabulated the canonical forms
of kerneIs in over 1300 singular nouns in Tiznit Tashlhiyt. They found
that a kernel may contain at most three syllables and at most two vowels,
and that in three-syllable kernels which contain two vowels, one of these
must be in the last syllable." Like the one presented in this book for
Imdlawn Tashlhiyt, the syllabification scheme adopted in the work on
kerneIs in Tiznit Tashlhiyt allowed any consonant to be a syllable nucleus.
However, in that scheme, sonority relationships between adjacent conso-
nants did not play any role : a right-to-Ieft scan was used instead. At the
end of the article, the authors speculated that the distributional restric-
tions in noun kerneis in Tashlhiyt may reflect an earlier historical stage
still in existence in Tarifit and in Kabyle, in which the syllabification of
consonant clusters is not sonority-driven even at the sentence level.
As for morphological alternations in kerneIs, a great wealth of data
have already been collected." Since in Tashlhiyt all consonants can be
syllabic, one could expect to find morphological processes which would
treat syllabic consonants on a par with vowels, but this expectation is not
borne out. As far as morphology-governed alternations in kerneis are
concerned, the relevant distinction is not that between syllabic segments and
non-syllabic ones, but that between vowels and consonants. Consonants
all behave alike, regardless of their syllabicity in the terminal representa-
tions. We illustrate this point with an example from derivational morphology.
The TIRRUGZA derivatives are templatic nouns which denote astate
or property, e.g. tirrugza 'manhood', from argaz 'man', tirruksa 'clandes-
tinity', from rks 'hide'. 80 TIRRUGZA nouns all have a /u .. . a/ vocalic
melody and end in a light syllable. Their penultimate syllable is heavy
and its onset belongs to a geminate . Although in our examples the bases

77 See Clements (1997) for some discussion of multistratal syllabification in Imdlawn

Tashlhiyt.
78 The authors of this book have since carried out a systematic survey of nominal and
verbal kerneIs in Imdlawn Tashlhiyt and found that verbal kerneis are subject to maximal
size requirements akin to those which constrain nominal kerneIs.
79 See for instance Jebbour (1988) for number inflection in nouns (Tiznit Tashlhiyt), DE
(1991) for verbal stern formation and DE (1992) for derivational morphology (Imdlawn
Tashlhiyt) .
80 The TIRRUGZA derivative s are pluralia tantum nouns in the feminine. See DE (1992)
for a list of forms and a discussion of the templatic mapping.
 VOWELLESS SYLLABLES 175

and their derivatives are given as complete words, the inflectional material
is not involved in the correspondence between bases and derivatives. For
example, in the derivation of t-i-rrugza from a-rgaz: the templatic corre-
spondence only involves -rgaz and -rrugza. In (36) below, the words in
column I have the TIRRUGZA derivatives given in column 11. The items
in column III are the strings which are inputs to the templatic mapping.
In these strings, the segments which are not preserved in the derived nouns
are marked by a slash ."
(36) I 11 III
a. a-rgaz t-i -rrugza rgaz
b. a-maziv t-i-mmuzva mazl)'
c. i-nbgi t-i-nnubga nbgl
d. !a-nttayfu lt-i-nttuyfa lnttayfu
e. a-n-lmad t-i-nllumda nlmad
f. a-n-flus t-i-nffulsa nflus
The TIRRUGZA derivatives respect 'consonantal invariance', a general
property of nonconcatenative morphology in Tashlhiyt which was already
mentioned in § 3.3.1. Very roughly: vowels are the main targets of mor-
phological processes; except for length, these processes leave consonants
intact. The syllabic status of contoids does not impinge on their behav-
iour in morphology-governed alternations." Consider for instance the
consonant /1/ in examples (36)e,f. It is a nuc1eus in the base -nlmad (n1.mad),
whereas it is an onset in the base -nflus (nf.1us), but this difference does
not have any bearing on its fate in the derivation: it is retained in the
derivative in both cases.
We leave kernel-level syllabification as an outstanding issue . This issue
is inextricably interwoven with the analysis of the underlying glides, which
are the subject of the next chapter.

6.6. SHORT VOCOIDS IN OTHER WORKS ON TASHLHIYT

The data involved in our presentation of syllabification in Imdlawn Tashlhiyt

are mainly drawn from two areas, versification and morphologically-
governed length alternations on the one hand, and the distribution of short
vocoids on the other hand. The two kinds of data play different roles in
our discussion. Let us first pause briefly to see where the difference lies .
It is versification and the morphologically-governed length alternations
which provide the crucial evidence as to how strings of segments are parsed
into syllables; in particular, these data show that short vocoids are not

81 The base nouns in (36)-1 have the following meanings : (a) man, (b) free person , (e)
guest, (d) dummy in a eard game, (e) apprentice, (f) weaIthy person.
82 Imperfeetive gemination is an exeeption to this generalization .
176 CHAPTER SIX

taken into account in syllabification. Call this the syllabic irrelevance of

short vocoids in Tashlhiyt. It has been stated that Berber @ is a syllable
nucleus." This is not the case in Tashlhiyt, but it is in other dialects of
Berber, e.g . Rifian.
The role of our considerations about short vocoids in Imdlawn Tashlhiyt
(this chapter) is to propose an explanation for their syllabic irrelevance:
the reason why syllabification does not take them into account is that they
are not segments. Our knowledge of their distribution and of their phonetic
properties is rather incomplete, as we have seen, but what little we know
fits with the hypothesis that they are mere transitions between segments.
The syllabic irrelevance of short vocoids is a property of Tashlhiyt at
large, because the poetic conventions discussed in Chapter 4 are not specific
to people living in the Imdlawn valley. They are common to a much larger
group of Berber-speaking people living in Western Morocco. On the other
hand, our generalizations on the distribution of short voiced vocoids were
arrived at solelyon the basis of data from the Imdlawn dialect.
How much variation involving short vocoids is there among Tashlhiyt
speakers? We do not know. The extant literature on Tashlhiyt cannot be
relied upon to provide information on this point. Half a century ago, Basset
and Galand were complaining that the transcriptions used by Berberists only
gave minimal information about pronunciation," and the situation has hardly
improved since. Short vocoids are present in the transcriptions of various
dialects of Tashlhiyt, see e.g. Destaing (1920) and Aspinion (1953), but it
is unclear to what extent they mirror the actual pronunciation." The
distribution of short vowels in these works differs markedly from that which
we reported above for the VTVs in Imdlawn Tashlhiyt. It is reminiscent
of that of short vowels in works on Berber dialects outside of the Tashlhiyt
group, e.g. in Ait Iraten Kabyle (Basset and Picard 1948) or in Ayt Ndhir
Tamazight (Penchoen 1973) . But the latter authors did not intend every
occurrence of e in their transcriptions to represent a voiced vocoid actually
occurring in the pronunciation," and the same is probably true of the tran-
scriptions of Tashlhiyt by Aspinion and the other authors cited above .
As far as schwa and syllabic consonants are concerned, the literature

83 Basset (1952: 8).

84 Basset (1952: 5) and Galand (1953).
85 Some of the inconsistencies in these transcriptions probably come from variations in tran-
scriptional practice, rather than from variation in the data. See Galand (1953: 230) for examp1es
of inconsistencies in the transcriptions in Destaing (1920).
86 According to Basset and Picard (1948: 9) a is 'a vocalic element [. . .) which ranges,
depending in particular on tempo, from a well-marked vowel to nothing at all. All vowels
are voiced; it may be the case , however, that d could devoice , leaving a mere suspension
as its only residue ' . Penchoen (1973 : 10) writes that what he transcribes uniformly as d
'may be - phonetically - an [;lI, the syllabicity of a consonant such as a nasal, lateral or
Ir/, or even a simple consonant release voiced or not' .
 VOWELLESS SYLLABLES 177

on Tashlhiyt is no less confusing than that on the other Berber dialect groups:
when two descriptions disagree, it is often very difficult to sort out from
the rest those divergences which reflect actual differences in the data.
Mohamed Guerssel's work on his native Tamazight dialect is an example
of this general state of affairs.
In Guerssel's transcriptions of surface forms in his earlier work, the
distribution of schwa followed a pattern rather similar to that found in
other works on Berber, with underlying ICCCI clusters regularly surfacing
as [email protected] Then, in 1985, the author published an article in which most
of the @ C sequences occurring in the surface forms in his earlier work were
in effect reinterpreted as syllabic consonants. He gave surface forms such
as s[.dX.t.!! 'I wiped them' (from Isfd-x=tn/), which he would have tran-
scribed as s@jd@xt@n in his earlier work. According to the 1985 analysis
of Ait Seghrouchen Tamazight, all the consonants but the stops can be
syllable nuclei, and nuclear status is preferentially assigned to consonants
with greater sonority. The author does not make it clear whether the switch
from s@jd@xt@n to s[.dX.t.!! is on1y a matter of phonological interpreta-
tion, or whether he has also changed his views about what the phonetic facts
are. In Guerssel (1992), the author gives up all syllabic consonants and
reverts to Ait Seghrouchen surface forms in which ICCCI is generally
realized as C@CC. Apart from such surface forms, Guerssel's articles
provide little information about the pronunciation of his dialect.
Durand (1995/96) notes that neutral vowels become more and more
elusive as one moves westward in North Africa . He makes this observa-
tion both for Arabic and for Berber. Durand shows that in many instances,
common1y accepted transcriptions of Moroccan Arabic write a short vowel
where none is heard in the pronunciation . He suggests that as far as neutral
vowels are concerned, the transcriptions of Moroccan Arabic have been
unduly influenced by those devised for recording the varieties of Arabic
spoken in Algeria, and he seems to suggest that a similar phenomenon
may have occurred for Berber, where prior knowledge of dialects like
Kabyle, which have more clearly audible neutral vowels, may have influ-
enced the notation of dialects more to the West.88 Some transcriptions of
Tashlhiyt could contain occurrences of e which are audible as vowels only
in other dialects." We must bear in mind that the transcriptions in question
are in many ways more like conventional spelling systems than like phonetic
transcriptions, in that they are primarily for the use of people who already

87 See for instance the rule of schwa epenthesis in Guerssel (1977: 271). As we have seen
in § 6.5.1, Rifian Berber follows essentially the same pattern.
88 Galand (1953: 230) even hints that some transcriptions of Berber may have been influ-
enced by assumptions about syllabification in Arabic.
89 See also Galand (1988: 213): '. . . most often, the numerous occurrences of [;l] found
in works by Berberists reflect habits which are alien to Tashlhiyt. '
178 CHAPTER SIX

have a first hand experience with the dia1ect which they record, or with
one resemb1ing it. The peop1e who devised them seem to have tried not
to depart too far from the spelling conventions of French, since these
transcriptions were used in teaching: some native speakers of French and
Arabic were taught Berber before being sent as administrators to Berber-
speaking areas." Take for instance /tt-kks/ 'remove! impf', which can only
be pronounced as ttkks (i.e. [thk:s]) in Imd1awn Tashlhiyt. Destaing (1920:
11 0, 206) and Aspinion (1953: 271) both write ttekkes. No matter how
the word was actually pronounced in the dialects these authors were
describing, 'ttekkes' is easier than 'ttkks' to identify and to memorize for
a speaker of French (or Spanish, or German).
Louali and Puech have recently presented evidence which suggests that
some varieties of Tashlhiyt have voiced vocoids which cannot be interpreted
as mere transitions between segments. Louali and Puech (2000) report that
they have made a short list of words which contain only voiceless conso-
nants, e.g. /kf=t/ 'give it!', and they have examined audio recordings of nine
repetitions of each word by four speakers from various locations in the
Tashlhiyt-speaking region. In the case of one speaker they report the same
finding as Coleman (1999) observed in his audio recordings of ME's speech:
no voicing ever occurs, as MINIMAL-PATH(voice) would lead one to
expect (v. § 6.3.1); but they found voiced vocoids in some of the repeti-
tions produced by the other speakers .
Louali and Puech do not give any indication about the Iocation of the
epenthesis sites in the various tokens they examined. In the absence of
such information one cannot determine the relationship between the voiced
vocoids which they have observed and the syllab1e structure which we
have posited on the basis of the evidence presented in Chapters 4 and 5.
When we know more about Louali and Puech's data, it may turn out that
the distribution of the short vocoids is closely related to syllable struc-
ture, but this cannot be taken for granted apriori.
To conclude this section we discuss an article in which John Co1eman
challenges our views on the status of short vocoids in the syllable struc-
ture of Tashlhiyt.
Coleman (1996, 1999, 2001) disputes our phonological interpretation
of the syllabic consonants of TashIhiyt. Let us concentrate on Coleman
(2001), which repeats or summarizes the main facts contained in the earlier
papers and presents the author's current position. To anticipate our con-
clusions: most of the data marshalled by the author is irrelevant for choosing
between his position and ours, and the little that is relevant is not com-
pelling.
Let us first sort out what the points of contention are and what kinds
of data wou1d be decisive for choosing between our position and Co1eman's.

90 See Galand (1957) on the school founded in Rabat by General Lyautey in 1912.
 VOWELLESS SYLLABLES 179

Once this is done, we will review the various kinds of evidence presented
by Coleman.
Coleman agrees that Imdlawn Tashlhiyt has syllables which do not
contain any discemible vocoid, but he disputes our interpretation of this
fact. While we hold that the nucleus of such syllables is a consonant, his
view is that it is a vowel which is phonetically overlapped by adjacent
consonants. Consider for instance the first syllable of t-sti 'she selected',
which is pronounced [ts]. While we hold that that syllable only contains two
segments t and s, Coleman's view is that it is actually t@s, in which the
vowel is subject to a process of reduction akin to those which give rise to
English casual-speech pronunciations such as [th]night (tonight), [s]ppose
(suppose) .
Let us use the acronym 'RIPI' to refer to the Representations which
are Inputs to Phonetic Implementation." Coleman proposes what he calls
'the coproduction analysis of syllabic consonants', which implies that only
vocoids may occur as nuclei. The coproduction analysis of syllabic con-
sonants is in contradiction with the Licit Consonantal Nuclei thesis
introduced in § 4.1. According to the Licit Consonantal Nuclei thesis, any
consonant can be a nucleus in the RIPIs. The two contending positions
are illustrated below in (37) with t-sti ' she selected' .
(37) underlying pronunciation Coleman's our RIPI
representation RIPI
/t-sti/ [tsti] [email protected] It~.til

We enclose the RIPIs between vertical lines to distinguish them from the
narrow phonetic transcriptions, which are indicated by square brackets.
The latter transcriptions are only meant as means of conveying what can
be observed by inspecting the phonetic record, without any commitment
as to the content of the corresponding RIPIs. To represent the vowel subject
to reduction in the coproduction analysi s, Coleman uses the notation 'I z!' ,
To remain in agreement with our conventions elsewhere in this book, we
replace '/g/' by 'I@I'.
According to Coleman (2001) , all the occurrences of \@I are epenthetic;
they are introduced by syllabification (recall that under the coproduction
analy sis, only vocoids may occur as syllable nuclei in the RIPIs). I@I is
an empty nucleus and it is phonetic implementation that specifies its vowel
quality, which depends on the surrounding segments, and its duration. Thi s
duration may be too short for the vocoid to be discemed as a distinct
segmentation of time in the phonetic record. In [tsti], for instance, the syl-
labicity of [s] is simply due to the fact that the realization of the nucleus
I@I is completely overlapped by that of the following [s].

91 In our view the RIPIs are terminal representations of the phonological component.
180 CHAPT ER SIX

There are no dis agreements between Coleman (2001) and us on the

representations which are inputs to syllabification. Whereas in our account
the burden of accounting for the distribution of [@] is placed entirely on
phonetic implementation, in Coleman's account the work is done jointly
by the phonological component, which inserts I@I, and by phonetic imple-
mentation, which eclipses it in certain instances.
When one tries to match the 1@ls which occur in the RIPIs posited by
Coleman with the [@]s observed in pronunciation, one encounters dispar-
ities of two kinds. Compare for instance the RIPI and the pronunciation
of t-sbv-t 'you painted'.

(38) underlying pronunciation Coleman's our RIPI

representation RIPI
It-sb"(-t/ [ts@b"(t] It@sb."(@tl

First , in It@sb."(@tl, which is the RIPI under Coleman's analysis , the two
occurrences of I@I do not have reflexes in the pronunciation. A propo-
nent of the coproduction analysis may simply claim that I@I is eclipsed
by the surrounding consonants in both syllables of (38).
Second, a short vocoid is pronounced between [s] and [b], but these
segments are not separated by I@I in the RIPI. In our view the short vocoid
which is heard between [s] and [b] in the pronunciation of t-sbv-t is not
the realization of a distinct segment in the RIPI, but a transition between
[s] and [b]. Coleman too recognizes that certain vocoids are not the real-
ization of a vowel, but he acknowledges their existence only in a very
limited range of contexts. The only examples that he mentions occur next
to Irl or between the individual taps of its trilled realizations. Although
he is not explicit on this point, throughout his discussion he seems to assurne
the following :

(39) All Schwas Are Nuclei:

Except in the immediate neighborhood of Ir/, any [@] found
in the phonetic record is the reflex of a syllable nucleus in the
RIPIs.

'All Schwas Are Nuclei' is but a convenient mnemonic for (39), and we
beg the reader to bear in mind that henceforth any statement implying the
absence of transitional vocoids from Coleman's account must be understood
with the proviso 'except in the immediate vicinity of Ir/'.
Coleman's account and ours are both incomplete, in that there are certain
environments in which each fails to make definite predictions about the
occurrence of [@] in a CC sequence. Furthermore, the two accounts are
incomplete in similar ways: each account circumscribes a certain range of
environments in which it predicts that [@] may not occur in a CC sequence,
call these the 'no-schwa environments': outside of these environments
 VOWELLESS SYLLABLES 181

[@] is allowed to occur in a CC sequence, but here neither Coleman' s

account nor ours makes definite claims as to whether the occurrence of
[@] is obligatory or merely optional. When the two accounts are compared
for empirical adequacy, then , the crucial data come from environments
which are no-schwa environments.
In our account, the no-schwa environments are those singled out by
generalizations (2) and (3) at the beginning of § 6.3.1, viz voiceless
sequences (e.g. kt) and sequences of homorganic noncontinuants which
differ in sonorancy (e.g. dl), and also those in which the Fusion rule creates
geminates (v. § 6.3.3.2). Outside of these environments, a short vocoid is
heard in some cases but not in others, but the regularities have yet to be
worked out, e.g. our discussion in § 6.3.2 has nothing to say about the
fact that the pronunciation of [ts.bvtl contains [@] between [s] and [b], but
not between [b] and I"VI.
In Coleman 's account , the no-schwa environments are all those in which
syllabification does not insert I@I, e.g . [sb] in It@sb."V@tl (see (38)), but
that account does not say anything about the extent to which I@I is
phonetically overlapped in various environments, e.g. it does not capture
the fact that I@I must be totally eclipsed in [email protected] (37), but not in /i-xng/
[ixn@g] 'he strangled', a form whose RIPI is lix.n@gl in Coleman's account.
To sum up, either account is incomplete in its characterization of phonetic
implementation. Our account only makes definite factual claims about the
CC sequences involved in generalizations (2) and (3) and sequences of
sibling consonants. Coleman's account only does so for the CC sequences
which are not broken up by I@! in the RIPIs.
In Coleman's account, the syllables of Tashlhiyt all have the form
(C)V(C(C)), where V stands for a, i, U or the epenthetic vowel @. Unlike
OUfS , that account allows VCC rimes in which CC is not a geminate , e.g.
in Coleman's account /t-a-frux-t/ 'girl' is parsed as two syllables , [taf.ruxt],
with a final rime uxt, but on our analysis it must be parsed as three sylla-
bies: [taf.ru .xt]. In CC codas the second C may not be more sonorous than
the first, but otherwise the sonority of consonants does not play any role
in syllabification in Coleman's account.
We will not go any further into the details of syllabification in Coleman's
analysis, but the reader may get some sense of how it differs from syllab-
ification in our account by examining the examples in table (40) below."
In the last column the nuclear consonants are all underlined for the sake
of conspicuity. In the second column, letters in boldface indicate schwas
which are problematic for Coleman's account because they are in contra-

92 The meanings of the examples in (40) are the following: (a) it went numb ; (b) she even
hoarded ; (c) she even behaved as a miser; (d) what will happen of you? ; (e) for the cock-
roaches; (f) you drowned; (g) you painted; (h) I locked; (i) he strangled; (j) he strangled
him; (k) broken branch ; (I) he wrung (someone ' s neck) for him ; (m) take care of his motherl .
182 CHAPTER SIX

diction with the All Schwas Are Nuclei assumption (39) : they do not cor-
respond to epenthetic 1@ls in Co1eman's RIPIs. In (40)b, for instance, [x]
and [z] are not separated by I@I in the RIPI posited by Co1eman, and yet
a short vocoid can sometimes be heard between the two consonants.
(40) underlying pronunciation Co1eman's our RIPI
representation RIPI
a. li-slm/ [isl@m] is.l@m i.slm
b. /t-xzneakk'? [tx(@)znak w: ] t@x .z@ .nakk w j.xg.nakk"
c. It-bxl=akkwi [t@bxlak w: ] [email protected]@.lakk w j.bx.lakk"
d. Ima rad t-g-t/ [marat@gt] ma .rat.g@t ma.rat.gt"
e. li=t- !bdr-inl [!it@b(@)drin] [email protected] it.bd.rin
f. It-!ngd-t/ [!tn(@)g@t:] [email protected]@t tn.g!e
4

g. It-sb-y-t/ [ts(@)b-yt] [email protected]@t tß..b:yt

h. Irg1-xl [rg1@x] [email protected]@x r.glx
1. li-xngl [ixn@g] ix.n@g i.xng
j. /i-xnget/ [ixn@g@t] ix.n@gt i.xn.gj
k. l!iskdl [!isk@d] iS.k@d is .kg
1. /i-snnqeas/ [isn:@qas] [email protected] i.snn.qas
m. Inawl maes/ [nawlmas] nawl.mas na.wl.mas
Table (40) gives us the occasion to pause briefly to comment on the nature
of the difference between the syllabic parses discussed in our works of 1985
and 1988 and those discussed in this book. This difference complicates
the comparison between Coleman's position and that expounded in this
book, but it is irrelevant for assessing the va1ue of all but one of the argu-
ments presented in Co1eman (2001).
In (40) the pronunciations in the second co1umn are reproduced from
our 1985 article, but the syllabic parses in the last co1umn are not. As
explained in § 5.1, our 1985 article dealt with syllabic parses inferred
from direct questioning (IFDQ syllabifications for short). On the other hand
this book only deals with orthometric syllabification, and consequent1y
the forms at the end of each 1ine in (40) are orthometric syllabifications."
Unless the readers constant1y keep this point in mind, they are apt to find
the comparison between Co1eman's work and ours very confusing. Table
(41) below contains all the examples from (40) whose orthometric syllab-
ification differs from the IFDQ syllabification recorded in our 1985 article
(the 1ine numbering of (40) has been retained to make the comparison
easier).

93 On the deletion of Id! at the end of the future preverb Irad! , see DE (1989 : 188).
94 As a result of voice assimilation and Fusion, Idt! surfaces as geminate Itt!, see § 6.3.3.2.
95 On the reasons why we now prefer to work with orthometric syllabification, v. § 4.1.
 VOWELLESS SYLLABLES 183

(41) orthometric IFDQ

(40) (DE 1985)
b. t.xg.nak'tk" txz.nak''k"
c. 1.b~.lakwkw tbx.laktk"
d. ma.rat.gj ma.ra.tgt
f. tn.gjt tn(.)glt96
j. i.xg.g] i.xngt
k. is.kQ .iskd.
As we already noted in 1988, one important difference between ortho-
metric syllabification and IFDQ syllabification is that the latter does not
allow obstruent nuclei adjacent to a pause; it allows complex onsets and
complex codas instead, see the forms in (4l)b,c, which have a complex onset
in their IFDQ parse, and those in (41)j,k, in which the complex coda is
not a geminate." In our works of 1985 and 1988 the focus was on IFDQ
syllabification. However the analysis in these works also generated syllabic
parses which were in essence identical to the orthometric parses. The parses
in question were a significant step in the sequential derivations which led
from the underlying forms to the IFDQ syllabifications, for they were the
output of our 'Core Syllabification' procedure. In the derivation of the IFDQ
parse of (41)b, for instance, Core Syllabification would generate the parse
on the left-hand side in (41)b, which would subsequently be readjusted to
yield the IFDQ parse on the right-hand side .
The constraint-based analysis of orthometric syllabification expounded
in the present book was first presented at length in DE (l997a). Coleman
(2001) only makes passing references to that article. His arguments are
all aimed at our earlier papers .
Returning now to table (40), the highlighted schwas in the pronuncia-
tions in the second column point to what is in our opinion the main weakness
in Coleman's position, viz the All Schwas Are Nuclei assumption. Consider
for instance i- snnqe as [isn:@qas] «40)1). The sequence /nq/, which is
broken up by [@] in [isn :@qas], is heterosyllabic in Coleman's RIPI as well
as in ours . This form exemplifies one of the environments in which tran-
sitional vocoids are most systematically heard in Imdlawn Tashlhiyt:
between a sonorant and a following uvular. Note that the sonorants of
Tashlhiyt are all articulated in the front part of the mouth. In Coleman's
account, differences between the points of articulation of adjacent conso-
nants do not play any role, and consequently there is no explanation for why

96 ME was unsure whether the form contained one syllable or two.

97 The IFDQ syllabification of form (41)d violates constraint NoPICOR, which disfavors
rimes in which an obstruent nucleus is followed by a coda with the same degree of sonority,
see § 4.9.1.
184 C HA PTE R SIX

a short vocoid can occur in nq even when the sequence is heterosyllabic,

while none may occur in nt regard1ess of syllab1e structure.
Co1eman's re1uctance to accept that at least some occurrences of [@]
are not epenthetic vowels is understandable: if some occurrences of [@]
were epenthetic vowe1s while others were mere transitions between con-
sonants, the coproduction analysis would have to be supp1emented with
hypotheses about transitions between consonants, and one wou1d have to
examine if these hypotheses do not render the overlapping mechanisms
superfluous, at least as parts of an account for schwas and syllabic conso-
nants in Tashlhiyt."
We are now ready to examine Coleman's arguments. Co1eman (2001)
is in essence a 1ater version of Coleman (1996). Co1eman (2001) takes
into account some of our reactions to Co1eman (1996) in our rejoinder to
that article (DE 1996b); it also incorporates the acoustic data presented in
Coleman (1999) . These data were gathered by the author in the hope of
substantiating his 1996 position, which he has now abandoned in part: not
only did Coleman (1996) claim that in Tash1hiyt all syllabic consonants
arose from the phonetic overlap of a vowel; he claimed furthermore that
in some cases at least the overlapped vowel was [a], [i] or [u], Coleman (2001)
implicitly acknow1edges that his acoustic data do not support this latter claim
and he retreats to the weaker position that the vowels subject to phonetic
overlap are all epenthetic schwas . This retreat renders irrelevant much of
the evidence repeated by Coleman (2001) from his 1996 and 1999 articles,
for that evidence was gathered in support of a position which he no Ionger
ho1ds. Co1eman (2001) adduces four kinds of data in support of the analysis
exemplified in (37) , (38) and (40): (i) acoustic measurements, (ii) IFDQ syl-
labifications, (iii) free variation phenomena, and (iv) data from Berber
dia1ects other than Tashlhiyt. We will discuss each kind of evidence in
turn.
Coleman's acoustic measurements are based on audio recordings of ME's
speech made in the phonetic s 1aboratory in Oxford in the summer of 1995.
These recordings included all the words which occur in DE (1985). For
the purpose of this discussion, Coleman's data from acoustic measure-
ments fall into two categories: (a) data about the occurrence of [@] and
(b) data about the duration and formant frequencies of [@] in various
contexts . The latter data were gathered by the author in search of evidence
for his earlier position that the vowe1s subject to phonetic overlap were
in some cases underlying vowels . However the author has now retreated
to the claim that the eclipsed vowels are all epenthetic schwas . The data
about the acoustic properties of schwa may be of interest for the study of
those aspects of phonetic imp1ementation responsible for the duration and
the vowel quality of short voiced vocoids in various contexts, but they

98 See DE (l996b: 232) for the discussion of a case in point.

 VOWELLESS SYLLABLES 185

are irrelevant for chosing between our account and Coleman's current
account.
As data about the distribution of schwa, the author only presents
statistics which he uses to assess the overall goodness of fit of our account
and of his. " The author 's finding is that his account's overall performance
is 'marginally better' than ours in predicting the occurrences of [@]. This
can hardly be taken as compelling evidence of the superiority of the copro-
duction analysis. We think that global statistics of the kind performed by
the author are a poor substitute for data about specific forms instantiating
crucial cases in which the two accounts make different predictions. As
explained above, the crucial cases involve contexts which are no-schwa
environments in one account but not the other.
We turn next to the second kind of evidence presented by Coleman, which
is native speakers' judgements about syllable structure. The author used
his (C)V(C(C)) grammar to parse close to six hundred Tashlhiyt words. The
resulting parses were presented to a Tashlhiyt speaker from the city of
Agadir. The informant checked the parses against his own intuitive judge-
ments and dec1ared them correct in 98 percent of the cases. One would
like to know how the same informant would have rated the parses assigned
to the same words by DE's 1985 analysis . The question is of particular
interest, since Coleman (1996: 187, 206) writes that the informant, a
graduate student with extensive education in phonological theory who has
written a Master's thesi s on the phonology of Tashlhiyt, was an advocate
of our approach to syllabification. Presumably he would not have advocated
it if there had been important differences between his own judgements
and the IFDQ syllabifications recorded in our works of 1985 and 1988 .
Again, an overall success rate cannot substitute for examination of the
individual IFDQ parses which are incompatible with one or the other of
the competing analyses.
The third type of evidence presented by Coleman is cases of free vari-
ation between C" and Cu. The forms in (42) are in free variation in Imdlawn
Tashlhiyt:
(42) a. managu manag" 'when'
b. la-xurbis la-x'trbis ' small mosque'
c. a-ggurdi a_gWgWrdi 'flea'
d. !t-i-kkurda !t-i-kwkwrda 'theft '
In Coleman's paper examples such as those in (42) are given without
comment, but the context makes the author's intent c1ear. The examples
in question are given just after a discussion of the variations in the pro-
nunciation of forms such as ! kdu 'smell! '. [@] occurs between k and d in

99 Here 'our account' refers to our characterization of voiced transitional vocoids in DE

(I996a), which is the same as that presented in the earlier sections of this chapter.
186 CHAPTER SIX

the recordings of some tokens of the word but not in others, and Coleman
interprets this variation as evidence that the RIPI of !kdu is j!k@dul. It is
clear that he means to suggest that in the free variation in (42), CW is just
a realization of Cu when the vowel is eclipsed by phonetic implementa-
tion. Strictly speaking, viewing manag" as resulting from the phonetic
eclipse of the final vowel in [managu] is in contradiction with Coleman
(200l)'s position that eclipse only affects epenthetic schwas . We nonethe-
less briefly take up free variations like those in (42) because we suspect that
they will turn up again in future discussions of syllable structure in Tashlhiyt.
Coleman's presentation suggests that [manag"] is a 'reduced' variant
of [managu] in the same manner as [!kdu] is, in his view, a reduced variant
of I!k@dul. This parallel is untenable. The difference between [!kdu] and
[!k@du] is a subphonemic one, and after a pause the 'free variation' between
[kd] and [k@d] occurs in all the words beginning with 'kd' in our tran-
scriptions. The status of that variation in the grammar is comparable to
the variation in the degree of aspiration of voiceless stops after a pause in
English . On the other hand, the distinction between Cu and CW is a phonemic
one, witness contrasting pairs like those in (43) below, which can be
multiplied at will:
(43) a. aggu 'smoke' agt'g" 'peek!'
b. t-a-guri 'word' agrri 'cause'
c. t-guni 'lying down' t-g'tni 3fs-sew:neg IOO
d. t-a-skkur-t 'partridge' a-sgWgWrd 'mortar'
e. a-mrgul 'tureen' a-srg"l 'cover'
f. akuz 'weevil' akrz 'recognize!'
g. i-gulla-n 'cheeks' t-a-gWlla 'porridge'
Pronouncing CW instead of Cu in the forms on the left-hand side yields
iII-formed pronunciations, and the same is true if Cu is substituted for CW
in the pronunciation of the forms on the right-hand side. The status of the
free variation in (42) is comparable to the vacillation between [i] and [c]
in the first syllable of the English word economics. Last but not least,
whereas 'free variations' such as that between [!kdu] and [!k@du] do not
affect syllable structure, those between Cu and CW do, e.g. in singing, the
alternation between managu (ma.na.gu) and manag" (ma.nag") provides a
choice between trisyllabic LLL and disyllabic LH.
The last kind of evidence presented by Coleman is sets of cognate
words from Tashlhiyt, Tamazight and Kabyle Berber. In most sets the
Tamazight and Kabyle forms have a vowel where Tashlhiyt has [@] or a
syllabic consonant, and in the others the situation is reversed: Tamazight
and Kabyle have [@] where Tashlhiyt has a or i. Coleman does not explain

100 t-guni is the bound state form of t-a-guni. The affirmative form corresponding to
t-gWni is t-grna.
 VOWELLESS SYLLABLES 187

how these cognate sets are relevant to his argument. He probably means
to suggest that his account implies a more uniform syllable structure across
Berber dialects than ours does. This is merely begging the point, however.
As we have seen in our discussion of Rifian Berber, there are indeed dif-
ferences between the syllable structure of Tashlhiyt and that of some other
Berber dialects, and the exact extent of theses differences is an empirical
question . Cognate forms drawn from different dialects can contribute little
to that question if they are not embedded in a compari son, however sketchy,
of the grammars of the dialects under consideration, i.e. of their phonolo-
gies.
To sum up, Coleman (2001) fails to provide compelling evidence for
his claim that the syllabic consonants of Tashlhiyt result from the phonetic
overlap of an epenthetic vowel by the neighboring consonants. However,
in his search for evidence supporting an earlier version of his claim, he
has gathered acoustic data which may prove valuable for the future study
of phonetic implementation in Tashlhiyt.
This review of the treatment of short vocoids in the literature on Tashlhiyt
concludes our argumentation in favor of the Licit Consonantal Nuclei thesis,
which is one main component in our account of syllable structure in
Imdlawn Tashlhiyt. In our presentation of that account, we have devoted
special attention to the syllabification of contoids when they are adjacent
to other contoids, because this aspect of Tashlhiyt phonology is of special
interest for linguistic theory, but the constraints put forth in Chapter 4
also make definite predictions about the syllabification of contoids adjacent
to vocoids . As we shall see, these latter predictions are not always borne
out, due to certain peculiarities in the behaviour of underlying glides, a topic
which will be the center of attention in the next chapter.
 CHAPTER SEVEN

THE SYLLABIFICATION OF VOCOIDS

Like the other Berber dialects, Tashlhiyt has an underlying distinction

between the high vowels li, ul and the semivowels Iy, w/. In § 7.1 we
examine in detail how hiatus is avoided in vocoid sequences which do
not contain underlying semivowels. The rest of the chapter deals with the
phonology of the underlying semivowels. In § 7.2 we show the need for
an underlying distinction between glides and vowels. In § 7.3 we discuss
the behaviour of glides when they occur as sonority peaks in the und er-
lying representations. In some cases (§ 7.3.1, § 7.3.3) they surface as glides,
in violation of some of the constraints discussed in Chapter 4, whereas in
others (§ 7.3.2) they undergo lengthening and surface as vowel-plus-glide
sequences. Finally, § 7.4 is devoted to a discussion of geminate glides and
we end with abrief conclusion in § 7.5.

7.1. VOCOID SEQUENCES NOT CONTAINING UNDERLYING GLIDES

This section deals with the prohibition of vowel sequences and how this
prohibition is enforced in Imdlawn Tashlhiyt.
Like other dialects of Berber, Imdlawn Tashlhiyt does not tolerate
adjacent vowels in any morphosyntactic environment.' As we will see,
Imdlawn Tashlhiyt can do the following things to prevent an underlying
sequence of vocoids from being realized as a sequence of vowels. The
vocoid sequence may be realized as a long vowel, it may be broken up
by an epenthetic yod, or one of the vocoids may show up as a glide. In
all these cases, either underlying vocoid shows up as aseparate skeletal
position on the surface. Setting aside the idiosyncratic behavior of certain
grammatical morphemes, elision, i.e. the complete deletion of one vocoid,
is only used in fast speech and in poetry .
In what follows we survey the various cases in which a sequence of
vocoids could be expected to give rise to a vowel sequence, were it not
for the abovementioned prohibition. Our survey will deal with sequences

I We have come across a few cases in which hiatus may not altogether be exeluded. In all
of these, ME's judgements are not elearcut, or when they are, they are not consistent over
time. These instances all involve adjacent vowels belonging to words located on either side
of a major syntactic break . Such cases warrant further research. An example is found in
the following sentence , an interrogative eleft : i-kru a mmi ra lt -vrs-t (kid AD dat RAD
2-cut:throataor-2s) ' is it the kid whose throat you will slit?' It may be possible to put the
sentence's pitch maximum on u in ikru, the elefted noun, in which case the sentence contains
a sequence of vowels (ua) .

189
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
190 CHAPTER SEVEN

of vocoids which are brought about by concatenative mechanisms in mor-

phology or in syntax. Nonconcatenative morphology also avoids vowel
sequences.
The lexical representations of Imdlawn Tashlhiyt must distinguish
between two kinds of high vocoids depending on whether they are allowed
to surface as vowels. /y/ and /w/ are underlying segments which must always
surface as glides. ! whereas /i/ and /u/ surface as vowels in some contexts
and as glides in others. The difference between the two types of high vocoids
will be dealt with in detail in the following sections. Let us use the term
'underlying gIide' to refer to /y/ or /w/, and the term 'potential high vowel'
('potential hv ' for short) to refer to /i/ or /u/.
The underlying glides do not belong to the subject matter of the present
section. As for the potential hvs, only those located at the ends of rnor-
phemes will be considered, for Imdlawn Tashlhiyt does not possess
processes which would have the capacity of bringing together a potential
hv which is morpheme-internal and another segment located at one end
of a morpheme.' Our examples will often be given in sets to show con-
textual variants of the morphemes involved.
In our presentation of the facts we will gloss over various idiosyncra-
eies involving individual morphemes or morpheme sequences, e.g. /imma
ur/ 'while not', which is normally pronounced immawr, can also be pro-
nounced immuwr or immur in fast speech. As far as we know, only in this
particular morpheme sequence can /a+u/ be realized as uw or u, e.g. /imma
ussn/ 'while the jackal .. .' can only be pronounced as imma wssn.
The terminology and the abbreviations used below were introduced in
§ 2.2. As indicated there, 'H' represents a high vocoid, ' G' represents a
glide and '+ ' represents any morpho-syntactic boundary (-, = or #) .

7.1.1. Sequences a+H

When a is followed by a high vocoid, the high vocoid is realized as a
glide no matter in what environment. In iy i-ssa y-su (if 3ms-eat
3ms-drink:aor) 'if he has eaten, let hirn drink' the 3ms PNG must be syllabic
in the first verb but may not be in the second. Compare also ut-n 'they
struck' and t-a=da wt-n (f-sedet strike-3mp) 'the one (f) which they struck' ."
The initial segment of the encIitic is syllabic in las=iyi (shear:aor=dols)

2 Except in quite exceptional circumstances, on which v. the end of § 7.2.

3 This statement could seem to be falsified by the insertion of the chameleon vowel in the
formation of imperfective stems, as in the imperfective of gnugi 'tumble down', which is
tt-gnuguy. Chameleon insertion is better viewed as a nonconcatenative process, however.
On the chameleon vowel, v. § 5.2.
4 Homophonous, up to the second t, with t-adaw-renn-s (f-hump-fs=gen-3s) ' its hump
(camel)' .
 THE SYLLABIFICATION OF VOCOIDS 191

'crop my hair!' and i-Ias=iyt (3ms-shear:aor=IYT) 'let hirn shear' but not
in zara=yyi (seekedo l s) 'look for me!' nor in i-iara=yyt (3ms -
seek:aor=IYT) 'let hirn search' .'
For vocoid sequences other than those of the form a- H, the surface
outcome varies depending on whether + represents a morpheme boundary,
a clitic boundary or a Pword boundary. Let us examine each case in turn,
starting with Hs-a sequences.

7.1.2. Sequences tt-«

Sequences Hs a can be realized as G+a in all environments. G+a is the only
acceptable realization in some environments whereas others allow free
variation with another pronounciation. Inside words only G+a is an accept-
able pronunciation. When the sequence straddles a clitic boundary, u+a (but
not i+a) can in some cases be broken up by the insertion of yod. When it
straddles a Pword boundary, G+a is in some contexts in free variation
with a long monophthong. For word-internal sequences, here are examples
involving the 3ms PNG and the bound state prefix: i-skr ' he did' vs.
y-attuy ' it (m) is high '; deu-rga: 'and the man' vs. dvw-aman 'and the
water'. For sequences straddling a clitic boundary, here are examples
involving imperative plural at, dat3s as and demonstrative ad 'this': ldi
'pull! ' vs. ldy=at 'pull! p'; gnu 'sew! ' vs. gnw=at or gnu=yat 'sew! p';
mdi 'set a trap!' vs. mdy=as 'set a trap for hirn!'; hdu ' give away!' vs.
hdw=as or hdu=yas 'give away to hirn'; i-ldi 'sling' vs. ildy=ad 'this sling ' ;
a-gru 'frog' vs. agrw=ad or agru=yad 'this frog'. When the potential hv
and the following a belong to different words, G+a, which is always a
possible realization, is often in free variation with a long vowel: aa for
ya and DD for wa, e.g . ldi 'pull!' vs. ldy#a-skWti or ldasa-sk'ti ' pull the
mill crank!'.
In the preceding transcription the Pword boundary should not be taken
to imply the presence of a syllable boundary. The two 'halves'of the long
vowel are tautosyllabic, as in all long vowels in Imdlawn Tashlhiyt. In
this book, two identical vowel symbols in a row always represent a
(tautosyllabic) long vowel, regardless of intervening boundary symbols.
Here are other examples in which li+al has a free variant aa.
Iw-a=lli=akwkw=gi-sn i- !nrs-nl (m-s=det=all=loc-3mp prt-clever-prt) 'the
most intelligent among them', la-ydi=as fki-xl (u-dogedatßrns give-ls)
'the dog I gave him'. Here is now an example in which wa is in free vari-
ation with DD: /ss-hmu atayl (cau-warm tea) 'warm up the tea!' .
The realization of i+a and u+a as aa and DD is only possible across
Pword boundaries, but all i+a and u+a sequences with an intervening

5 The clitics iyilyyi and iytlyyt may be analyzed as beginning with geminate glides in their
lexical representations (lyyil and Iyytl) . On geminate glides, v . § 7.4.
192 CHAPTER S EVEN

word boundary cannot be realized as aa and DD. At present we do not

know what contexts allow the free variants aa and DD.

7.1.3. Sequences of potential hvs

Let us now turn to sequences of two potential hvs. Inside words they are
realized as GV;6 elsewhere they are realized as VG.7 Here are word-internal
sequences involving the 3ms PNG and the bound state prefix : i-nna 'he said'
vs. y-ut 'he struck', y-itti 'he moved away'; dssu-varas 'and the path' vs.
dcw-uday 'and the Jew', ds-y-izm 'and the lion'. The following examples
involve sequences which straddle a clitic boundary: sufu ' ill uminate !',
las=iyi 'crop my hair!', sufu=yyi 'make light for me !'; t-las=iyt 'let her
shear', t-sufu=yyt 'let her illuminate'. Here are now instances involving a
preposition: i=ysti (dat=daughter:p:my) 'to my daughters' , ddu=yiia 'under
Izza (a woman's first name) ":" d=u-frux (andeb-boy) 'and the boy' vs.
iws-x i=wjrux (help-ls dat=b-boy) 'I helped the boy '.
Finally, in the following examples the adjacent potential hvs do not
belong to the same Pword . uhu 'no', ini whu ' say no! '; izm 'lion' vs. hdu
yzm 'give the lion away!' ; ut-x=tt (strike-l sedoßfs) 'I struck her', asku wtxtt
'because I struck her' ; immi 'Mom' vs. ur lt-zri ymmi (neg 3fs-see:neg
Mom) 'she did not see Mom' .
In most contexts the glides of Imdlawn Tashlhiyt do not give the impres-
sion of being produced with a narrower constriction than the high vowels.
There are instances where our only basis for chosing between 'y' and 'i'
or between 'w ' and 'u' in our phonetic transcriptions are ME's judge-
ments about syllable count and the location of syllabic peaks. In so doing
we do not operate differently from phonologists who write [blio] and
[zolyo] when they are asked for phonetic transcriptions recording the
Parisian pronunciations of joli haut 'pretty top' and Joliot, a family name:
their main cue, or at least the overriding one, is the native speakers' judge-

6 One must bear in mind that this statement only pertains to sequences of potential hvs which
are brought about by concatenative morphology. The VG sequences created in some imper-
fective sterns by the 'insertion' of a chameleon vowel (v. note 3) are due to nonconcatenative
morphology , as is also sequence uw in plural nouns such as i-suwas 'pitch forks' , which is
related to the singular a-saw s as i-mudal 'rnountainsides' is related to the corresponding
singular a-madl .
7 In fast speech the final i of certain grammatical morphem es contracts with a following
u, yield ing uw, e.g. in Iw-a=1Ii ufi-xl (m-s=det find-I s) 'the one which I found ', lIIi#u/ ,
which is normally pronounced lliw, can also be pronounced lluw in fast speech. We only know
of one case in which lu+il may be realized as iy, that of the phrases headed by the empty
nouns bu and mmu, e.g. Ibu i-zrga-nl (ms:with bp-mill-p) 'the one with the rnills' can be
pronounced either as bu yzrgan or as bi yzrgan, v. § 2.5.3.2.
8 Although they begin with vowels, isti and iiia can occur without a prefix in the bound

state, v. § 2.5.1, note 22.

 THE SYLLABIFICATION OF VOCOIDS 193

ment that there are three syllables in joli haut and only two in Joliot? In
other instances, syllabicity differences correlate with others which even a
nonnative speaker can hear easily. Here is one such difference. Like the
sequences at the beginning of the English words yeast and woo, the
sequences yi and wu of Imdlawn Tashlhiyt are produced with a narrower
constriction at the beginning of their time course, which is not the case in
iy and uw, compare for instance tyi and tiy in y-ut y-ils (3ms-hit b-tongue)
'the tongue struck' vs. y-uti yls (3ms-Ietout tongue) 'he has his tongue
hanging out', and compare dwu with duw in y-anf=d w-udad (3ms -
take :cover:aor=dir b-ibex) ' . .. and the ibex took cover' vs. ra n-fdu wdad
(RAD l p-redeem.aor ibex) 'we shall redeem the ibex ' .

7.1.4. Sequences a+a

Let us now examine the behaviour of Ia+a/. This sequence can always
surface as aa (tautosyllabic), which is in free variation with aya in partic-
ular contexts. We have already mentioned the fact that Imdlawn Tashlhiyt
morphemes displaya contrast between a and aa, e.g. t-zba 'she took hastily'
vs. t-sbaa 'she ate her full ', bqqa-n 'they stayed' vs. fqqaa-n 'they became
angry', t-iram 'meals' (s t-irm-ti vs. la -raam 'camel', zayd 'add!' vs.
laayd 'festival'. The surface contrast between a and aa does not reflect
an underlying contrast between simple and geminate laa!, because all occur-
rences of tautomorphemic aa can be derived from 1)/, I)a! or la)/, as we
have argued in § 3.7.
The occurrences of aa which arise from la+a! are homophonous with
morpheme-internal ones, witness the homophony of !rqqa=ax (/!rqqa=axl
warm:impf=datlp) 'let us two warm up' with !rqqaa-x (/!rqq)-xl mend-
ls) 'I mended'. Another example is provided by the following pair, in which
emphasis is the only distinguishing feature at the phonetic level : i-ra aman
(3ms-want water) 'he wants water', li-raama-n /i-lr'irna-n/ 'camels'.
In Imdlawn Tashlhiyt the concatenation of morphemes inside words never
creates la+a! sequences,'? so we turn directly to those la+a! sequences which
arise through ward concatenation. As we have already said, all la+a!
sequences can be realized as aa. Except those involving the preverbs lad!
and lar/, on which v. below, all la=al sequences have a free variant aya.
Here are some examples. a-swik=ann 'that walnut (tree)', t-a-rga 'channel',

9 We do not mean to imply that there are no phonetic differences which correlate with the
syllable count difference. We simply mean that it is not necessary to know what these
differences are to be able to chose appropriately between the symbols 'i' and 'y' when tran-
scribing French sentences.
10 There are no verbal affixes beginning or ending with laJ, and in nouns the only uncon-

troversial instance of such an affix is the free state augment la-I, which can only occur
before a consonant-initial morpheme . As for the putative masculine plural suffix -an which
appears in certain nouns, it is never suffixed to nouns whose singular forrns end in a.
194 CHAPTER SEVEN

targaann or targayann 'that channel'; the notation targa(y)ann conveniently

sums up the two variants ; krz=at 'plough! p' , stta 'eat!' , 5tta=(y)at 'eat!
p'; a-ydi=da=(y)ax i-bbi-n (u-dog=det=dolp prt-bite-prt) 'the dog that bit
us'; nra=(y)akWkWmmut-n (just=all die-3mp) 'they alljust died'; mra=(y)as
nni-x (if=dat3s say-Is) 'if I had told him':" bla=(y)aga (without' jebucket)
'without a bucket' .
A special mention must be made of a=a sequences in which the first a
belongs to the preverbs larl or (R)AO. These sequences come about when
the preverb drops its final consonant and an a-initial clitic folIows. Such
sequences cannot be broken up by an epenthetic yod; they can only be
realized as aa. The complementizer lad! and the future preverb Irad/, which
we lump together under the label (R)AO, display the same set of idiosyn-
cratic alternations, on which v. OE (1989 : 188-190). In particular they
optionally lose their consonant before a vocoid-initial clitic. This happens
for instance in man a-fullus a(d)=ak y-ukr (WH u-chicken AO=dat2ms
3ms-steal) ' which chicken did he steal from you?' . As implied by our
transcription, /adeak/ may be pronounced as adak or aak, not as ayak.
How are we to interpret this fact? It looks as though what is preventing
the appearance of a hiatus-breaking yod is the final consonant in (R)AO.
Working with ordered rules, one would posit two optional rules, one
inserting yod to break up an a=a sequence , and the other deleting I(r)adl's
final consonant before a clitic beginning with a vocoid, and one would order
the yod-insertion rule before the truncation rule (counterfeeding order).
The imperfective preverb AR (lar/) obligatorily loses its final conso -
nant when it is preceded by another preverb which belongs to the same
clause (v. OE (1989 : 180ss) for details). This happens regardless of the
phonological make-up of the next word, and consequently the vowel of
larl may be brought together with that of a following clitic, as is the case
in is a=( *y)as akka-n at" t-55 (int AR=dat3s give:impf-3mp AO 3fs-eataor)
'do they give her something to eat?', where the interrogative is is itself a
preverb." Since (R)AO and larl are the only preverbs with a final conso-
nant subject to deletion , it is tempting to try for larl an analysis along the
same lines as that we have proposed for (R)AO. We could posit an oblig-
atory r-deletion rule specific to the imperfective preverb and order it after
the yod-insertion rule. But note that the contextual features to which the

11 When the second /a! in an /a=a! sequence belongs to a clitic pronoun, the sequence can
also be pronounced ay, e.g. in waxxa=as bbi-x a-ga yyu (even:if=dat3s cut-l s u-head) 'even
if I cut his head off", the first Pword can be pronounced waxxaas, wax.xayas or waxxa ys.
12 Preposition bla requires the following noun to be in the free state.
13 From AD (/ad/), whose final consonant assimilates to the following consonant.
14 Another idiosyncracy of larl is that y is obligatorily inserted between it and a preceding
word ending in a, provided the two words belong to the same c1ause (v. DE 1989: 187-8),
e.g. mra {yja n-aqqra (if AR Ip-read :impf) 'if we read'.
 THE SYLLABIFICATION OF VOCOIDS 195

r-deletion rule is sensitive are of a purely syntactic nature, whereas the

yod-insertion rule is concerned with syllabic structure.
In general sequences la#a/ can only be realized as aa, e.g. li-ssa adanl
(3ms-eat gut) 'he ate the gut' can only be pronounced issaadan . However
we have come across a number of cases in which aya is also an accept-
able pronunciation. Here are two examples: ha=tt (here :be=do3fs) 'here
she is'. ha (y)a-grtil (here:be u-mat) 'here is the mat'; manras-t
(where:be=do3ms) ' where is he?', manza (y)a-rgaz (where:be u-man) 'where
is the man?' . More work is needed to discover what factors set these cases
apart from the general case . Our present hunch is that in those instances
in which aya is acceptable, the first a belongs to a closed class of gram-
matical words which must always bear a tight syntactic relationship to the
word which immediately follows them in the sentence.
In fast speech, aa sequences which are surface reflexes of la+a/ can in
some instances be shortened to a, e.g. nttayran is an acceptable fast speech
pronunciation alongside nttaayran for Intta ad i-ra-nl (3ms AD prt-want-
prt) 'it is he who wants'. Some of the aa sequences derived from li+a/
can also be reduced to a in fast speech, e.g. Ildi askrti/ (see above in § 7.1.2),
which can be realized as ldaask'ti, can also be pronounced ldask'ti in fast
speech. The co ntex tual conditions under which these shortenings are
acceptable are yet to be worked out.

7.1.5. Sequences involving aa

Let us end this survey of vocoid sequences by looking at sequences in which

one of the abutting vocoids is aa. As mentioned earlier, tautomorphemic
aa can derive from lCi./, lCi.a/ or laCi./, but in a vocoid sequence the behav-
iour of aa is the same regardless of its underlying source. When discussing
cases such as the present one, where the source of aa at the most abstract
level of representation is irrelevant, we will include occurrences of laa/ in
the underlying representations of some of our examples, even though
according to our analysis geminate a does not occur in the lexical repre-
sentations of Imdlawn Tashlhiyt.
Like those adjacent to a, potential hvs adjacent to aa can always be
realized as glides, e.g. zri 'overtake!' vs. zry aali 'overtake Ali!' , azu 'skin! '
vs. azw aallus ' skin the calf! '. Gliding after aa gives rise to sequences of
three or even four vocoidal positions, as in the following examples : i-mmnaa
w-zal (3ms-difficult b-daylight) 'it is difficult to do it during the day' , 15

15 Literally 'the daylight is difficult'.

196 CHAPTER SEVEN

i-mmnaa w-wzal (3ms-difficult b-name) 'Awzal is a tough fellow' . 16 Note

that such sequences do not always straddle a Pword boundary, witness forms
such as laayb 'defect' , !i-saayyd 'he shouted' , i-sbaa=yyt 'let hirn eat his
full (I don 't mind)', i-sbaa=yyt=t (/=t/, do3ms) 'let hirn eat his full of it
(I don't mind)'.
The behaviour of sequences /aa+a/, /a+aa/ and /aa+aa/ paralleIs that of
the /a+a/ sequences which occur in analogous morpho-syntactic environ-
ments. Yod is inserted optionally to break up /aa=a/ sequences, e.g .
/buzmaaead/ (nameedem) 'this Boujmaa' can be pronounced buimaayad
or buimaad. Except when /aa=a/ is broken up by yod insertion, sequences
/aa+a/, /a+aa/ and /aa+aa/ are pronounced as a long a which is homopho-
nous with tautomorphemic aa. In /i-s-sbaaeak a- !kuray/ (3ms-cau-
satiate=dat2ms u-stick) 'he gave you a beating with a stick' the verb can
be pronounced issbaayak or issbaak. The latter form is homophonous with
/i-s-sbaaek/ (3ms-cau-satiate=do2ms) 'he made you eat your full '. /i-sbaa
aali/ 'Ali ate his full' can only be pronounced isbaali .

7.2. THE NEED FOR UNDERLYING GLIDES

Let us first give again two constraints which played a central role in our
discussion of sentence syllabification in Chapter 4.
(1) NoHiatus : A syllable which is not at the beginning of a syl-
labification domain has an onset.
(2) SonPeak: A sequence which is a sonority peak within the syl-
labification domain contains a syllable nucleus.
We now show that Imdlawn Tashlhiyt has an underlying distinction
between high voweis and glides, i.e. between /u/ and /w/, and between /i/
and /y/ . The reason such a distinction must be made is that the syllabifi-
cation of high vocoids is not always predictable.
There are two contexts where the syllabification of high vocoids is always
predictable on the surface: high vocoids always show up as vowels when
they occur after a contoid and before a pause, and they always show up
as glides when they are adjacent to an occurrence of a. An example of
the first context is given in (3)a ; examples of the second context are given
in (3)b,c,e.

16 u-wzal, bound state form of a-wzal , a proper name . The contrast between a and aa is
not neutrali zed before a geminate glide, V., e.g. the following quasi-minimal pair : i -sqqa
w-wzal 'Awzal is a tough fellow ' (same meanin g as the last example in the main text) vs.
i-fqqaa w-wzal 'Aw zal got upset' .
 THE SYLLABIFICATION OF VOCOIDS 197

(3) a. n-kti 'we remembered'

b. n-tt-rnm-ktay" 'we remember one another'
c. n-kty a-rgaz ' we remembered the man'
d. urs-n 'they swept' (v. next line)
e. ur ta wrs-n 'they have not swept yet'
The high vocoid in (3)a is a sonority peak occurring at the end of a syl-
labification domain . In Imdlawn Tashlhiyt glides do not occur in the context
C-II ('11' stands for a pause).
As a rule they are also excluded from the context II-c. The only known
exceptions to the preceding sentence are postpausal occurrences of the verbs
in (15) below." The high vocoids in (3)b,c,e are nonsyllabic so as to avoid
being in hiatus with a. Let us now review those cases where at the surface
level the syllabification of high vocoids is not predictable from the envi-
ronment.
First, sequences of a high vowel and a glide contrast with sequences
of a glide and a high vowel:
(4) vowel-glide glide-vowel
a. t-suy 'she let pass' t-zwi 'she beat down'
b. i-liws 'sheep hide (carpet)' a-vyul 'mule'
c. tt-gnuguy 'tumble impf' zuzwi 'be cool neg'19
As already stated in § 4.6, we adopt the commonly-held position that
glides have the same feature content as the corresponding high vowels,
and that the difference between them only has to do with their different posi-
tions within syllable structure . Consider for instance [u] and [w]. We are
assuming that they are occurrences of the same feature bundle {[-eons],
[dorsal], [labial] . . .}, call it U. [u] is an occurrence of U which occurs
as a syllable nucleus, and [w] is an occurrence of U which occurs as a
syllable margin. Similarly, the feature bundle I is notated as [i] when it is
syllabic, and it is notated as [y] when it is not. Given the assumptions we
have just made, here are the surface representations of t-suy and t-zwi in
(4)a:20

17 In tt-mm-ktay, the imperfective stern of mm-kti, a is the chameleon vowel, on which

see § 5.2. The conjugation of reciprocal verbs is presented in DE (1991: 100-102) .
18 In Rifian, on the other hand, postpausal glides preceding a consonant are commonplace,
see Dell and Tangi (1992: 144).
19 From gnugi, zuzwa .
20 The syllabic parses in (5) are those which obtain when the forms in question occur at
the beginning of a line of verse and are immediately followed by a CV sequence. The same
is true of all the syllabic parses given in this chapter.
198 CHAPTER SEVEN

I,
(5) a. o o b. o o
I
N
I
t
A\
s V I z
;1
V I
Setting aside the difference in voicedness in the initial consonant, which
is irrelevant, the lexical representations of suy and zwi must differ in some
way so as to account for the fact that it is lVI which is syllabified as a
nucleus in suy, whereas in zwi it is IV. But before we can determine what
the difference in question is, let us survey the other contexts in which glides
and high vowels contrast.
The examples in (4) involve adjacent high vocoids which differ in
backness. Analogous cases exist in which the abutting high vocoids agree
in backness:"
(6) vowel-glide glide-vowel
a. i-Ikkuwsa 'pruning hook p' i-mzwura 'first p' 22
b. !suwr 'paint' zwur ' precede'
c. t-huwt 'wander 3fs' t-hwu-t 'go down aor 2s '
d. ruwn 'fit' !rwu-n 'be good aor 3mp'
e. xuwx 'hollow out' xwu-x 'be empty aor l s'
f. t-hiyd 'rnove away 3fs' t-hyi-t 'keep alive 2s'
g. sniriy 'mimic impf'23 n-hyi 'keep alive neg Ip'
h. deu -wtil ' and the hare ' dew-uday 'and the Jew '
As implied by our transcriptions, the sequences wu and yi in the right-
hand side column sound like those at the beginning of the English words
woo and yeast. The constriction of the vocal tract is presumably narrower
in the glide than in the following vowel.
The forms in (4) and (6) abide by SonPeak (2) . In t-suy (4)a for instance,
the sonority peak uy contains u, which is a syllable nucleus. Imdlawn
Tashlhiyt also has words which violate SonPeak. In all such words a high
vocoid occurs between two contoids, or word -initially before a contoid, and
yet it is not a syllable nucleus. Compare for instance lur ' give back',
where the high vocoid is a nucleus, as is to be expected on account of the
fact that it is a sonority peak, and lwr 'flee' ," where w is an onset and
the following r is syllabic. lwr sounds like [lwor]; the short vocoid which

21 In the right-hand column in (6), the verbs in lines c-g have the following perfective sterns:
(c) hwa, (d) !rwa, (e) xwa and (f, g) Itya.
22 The corresponding singular forms are a-lkkawsu and a-mzwaru .
23 Perfective sniri.
24 lwr (impf IgWgWr) 'flee' is a rather formal word; it has a variant rwl (impf rgWgwl) which
is used in more colloquial styles of speech.
 THE SYLLABIFICATlON OF VOCOIDS 199

is heard between wand r is, in our view, a mere transition between the
two segments. It is not an independent segment at any level of represen-
tation, hence not a sonority peak. The syllabification of words such as lwr
in verse accords with their pronunciation, witness the scansion of line (7)
below, which is the 33rd line in the poem the beginning of which was parsed
in (19) in § 4.5. (7) is parsed in (8).
(7) ar i'Y=as t-!rzm-t i-bbi s-skal-at rwl-neak"
(8) 123 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a ri 'Yas tr zm tib- biSs ka 1a tr wl nak
SonPeak (2) is violated in the last word, rwi-neak, for w, which is a sonority
peak in (7), is not a syllable nucleus in (8). If it were a nucleus the parse
would be lat .rul.nak (HHH) instead of la.tr.wl.nak (LLLH). 26 The only
circumstances in which SonPeak ever incurs a violation in a well-formed
line of verse are similar to those in the example above: a high vocoid is a
sonority peak and yet it is parsed as a margin.
It is unclear how underlying glides are to be represented in the lexical
representations." For the time being let us assurne that in the representa-
tions which are inputs to syllabification the underlying glides are
distinguished in some way from the other high vocoids, and that SonPeak
is dominated by the following constraint:
(9) GlideFaith : underlying glides do not become nuclei.
GlideFaith overrides SonPeak, but it is itself overridden by NoHiatus
(1) and the other constraints in (70) in § 4.10. For instance the kerneI zwi,
which has already appeared in (4)a, is realized as zuy when the next
morpheme begins with a vowel:
(l0) a. zuyeas /zwl=asl 'beat down for hirn!'
b. zuy a baba Izwl a baba! 'beat down, 0 father!'
Consider für instance /zwl=as/. Ia! must be syllabified as a nucleus. If the
preceding /11 were also a nucleus, as it is in t-zwi in (4)a, NoHiatus would
be violated . As a matter of fact, NoHiatus can never be violated in Imdlawn
Tashlhiyt, and 1I1 must be an onset to Ia! .One might suggest that one way
to syllabify 11/ as an onset while at the same time having the preceding
Iwl abide by (9), would be for Izwl to be syllabified as an onsetless syllable,

25 • As soon as you release him, he breaks his fetters and escape s you' .
26 For other such instances, see e.g. Jouad (1995), line 6 on p. 102 and lines 1 and 12 on
p.106.
27 Guerssel (1986) has proposed an answer to this question for Ait Seghrouchen Tarnazight ,
but his proposal does not carry over to Tashlhiyt, due to differenc es between the two dialects.
200 CHAPTER SEVEN

with /z/ a nucleus and /w/ a coda (.z.w.yas.). This is impossible because
the first syllable violates the following constraint, which was one of the cor-
nerstones in our discussion of syllabification in Chapter 4:
(11) NoRR: the coda does not have a higher sonority than the nucleus.
Except in quite special circumstances on which see § 7.3.3 below, NoRR
is never violated in Imdlawn Tashlhiyt. GlideFaith must yield to NoRR,
hence the outcome zuy=as (.z!l.Ifl,s.).
One must distinguish between two kinds of high vocoids in the under-
Iying representations of Imdlawn Tashlhiyt. There are on the one hand those
high vocoids which we called 'potential hvs' in § 7.1 and which we will
simply call underlying vowels. These vocoids surface as vowels unless they
are adjacent to a vowel, in which case they surface as glides to avoid hiatus .
It is underlying vowels which are involved in the alternations in (3).
Underlying vowels are never involved in violations of SonPeak or of NoRR.
The second kind of high vocoids are those which we call underlying glides .
Underlying glides only surface as vowels in the highly restricted set of
environments exemplified in the preceding paragraph. They may occur in
strings which violate SonPeak or even NoRR, as we will see below.
Here is how the contrast between underlying glides and underlying
vowels is indicated in the letter sequences enclosed by slashes which sym-
bolize underlying representations in our running text. The underlying glides
are always noted 'w' and 'y'. Although we should use the symbols which
stand for bare feature bundles, viz. 'U' and 'I' , to represent the under-
lying vowels, we usually note these simply as 'u' and 'i' to avoid
proliferating symbols.
There is no audible difference between the surface glides which are
realizations of underlying glides and those which are realizations of under-
lying vowels. Consider for instance /wrz/, a nominal kernel meaning 'hinge',
and /urs/, the perfective stern of a verb meaning 'sweep' . The contrast
between /w/ and /u/, which is exemplified below in (12), is neutralized after
a vowel: setting aside dorsopharyngealization and differences in the con-
sonants, utterances a' and b' in (12) below sound alike:"
(12)
a. t-wrs-t (t.wr.st) 'hinge dirn b' (v. (18)b below)
b. t-urs-t (tur.st) 'you swept' (v. (3)d)
a'. ur d a-wrz !n=tamu (wr.zn) 'it is not Tamu's hinge'
b', ur ta wrs-nt anu (wr.sn) 'they haven't yet swept the weIl'

28 For reasons of convenience, the syllabic parse that we give between parentheses in an
example may represent only a portion of the complete parse of the expression under con-
sideration. In such cases the enclosing parentheses always correspond to syllable edges in
the complete parse.
 THE SYLLABIFICATION OF VOCOIDS 201

On the other hand, the surface vowels which are realizations of under-
lying vowels are in some cases distinct from those which are realizations
of underlying glides. This happens in those contexts in which /u/ is realized
as [öl as a result of u fronting ," e.g. between two coronal consonants in
a nonemphatic environment. Vocalized w is immune to u fronting. Consider
for instance the contrast between /suy/ and /swi/, which is exemplified in
(13)a and (13)b. When a vowel immediately follows /swi/, /i/ surfaces as
a glide, and /w/ as a vowel, but the realization of /swi/ is nonetheless distinct
from that of /suy/, as /u/ undergoes u fronting, whereas vocalized w does
not, witness the minimal pair (13)a',b':
(13) a. ur=tt i-suy ([söy]) 'he did not wipe with it (f)'
b. ur=tt i-swi 'it (m) is not as good as it (f),30
a' . ur i-s[ö]y a-z:ya"( 'he did not wipe with the rag (aug)'
b'. ur i-s[u]y a-z:ya-y 'it is not as good as the rag (aug)'
Outside of the contexts in which u is fronted, there is no audible differ -
ence between the vowel u and the vocalized w. Consider for instance
emphatic environments, in which u fronting does not occur. Clauses (14)a
and (14)b below are homophonous. The underlying representations of the
kemels involved are /!udn/ and /liwd/, as shown by (14)a',b', which give
the postpausal pronunciations of the forms in question, followed by the cor-
responding imperfective forms .
(14) a. mra !y-udn 'if he were in pain '
b. mra !yud-n 'if they had folded'
a' . ly-udn 'he is in pain ' (!i-tt-adn id. impf)
b' . !iwd-n 'they folded' (!tt-awd-n id. impf)
Similarly, ME does not hear any difference between the vowel in k[u]yan
'each one, m' and that which is a realization of /w/ in /ur i-xwi anu/ (neg
3ms-empty:neg 31 well) 'he did not empty the well' ur ix[u]y anu.
Our broad phonetic transcription does not record the effects of u fronting .
We shall keep it that way, with the added convention that we shall indicate
with a circumflex all the occurrences of u (i.e. vocalized /w/) which are
not subject to fronting although they occur in a fronting context. Following
this convention, we write suy in (13)a,a', sü» in (13)b', xuy in the example
at the end of the preceding paragraph, and zuy in (10)a,b.

29 On u fronting, see § 3.8.

30 swi is the negative stern of swa. The verbs in the sentences in (13) have two arguments,
a subject and a direct objecl. The direct object is the do3fs pronoun tt in (13)a,b; it is the
noun azyay in (13)a',b' .
31 Perfective xwa.
202 CHAPTER SEVEN

7.3. GLIDES WHICH ARE SONORITY PEAKS IN THE

UNDERL YING REPRESENTATIONS

In the present section we review all the words in which, as in Irwll, a

glide occurs as a sonority peak in the underlying representation. Given
the sonority-driven account of syllabification developed earlier in this book,
we expect the glide in question to be realized as a syllable nucleus. This
often does not happen , however. In some cases constraints NoRR and/or
SonPeak are violated and the glide is also a sonority peak at the surface
level, while in other cases the glide avoids such violations by becoming a
geminate. Let us first deal with some of the former cases.

7.3.1. Surface glides (onsets) which are sonority peaks

The total number of morphemes in which an underlying glide gives rise
to violations of the constraints NoRR and/or SonPeak is rather small .
When such violations occur, the glide is in some cases an onset, as in
tr.wlt (lt-rwl-t/) 'you ran away', while in other cases it is a coda, as in
~.srw.sn. (ls-rws-n/) 'they gave the impression' .32 In some morpheme s the
glide is parsed as an onset regardless of the environment, while in others
it is parsed as a coda at least in some environments.
We begin with the morphemes in which the offending glide is alway s
parsed as an onset, i.e . it violates the constraint SonPeak. As we shall
now see, the violations in question form a narrowly-circumscribed class
of cases ; in all these cases the onset glide is wand the following nucleus
is a coronal sonorant or fricative . The glide is the leftmost segment in all
of the morphemes in question. Words containing these morphemes fall
into two sets. The first set is comprised of all the inflected forms of the
following four verbal kernels:
(15) a. lwrrv (wr.ry) 'be yellow'
b. !wrrk (wr.rk) 'show off'
c. wzzl (w~.zD 'postpone'
d. wzzb (w~.zh) 'show up'
w is in free variation with u in (15)c,d; uiil and uiib are also acceptable
pronunciations. (15)a,b have no free variants. In these verbs the initial
vocoid must always surface as a glide, e.g. !i.wr.r:y 'he is yellow ', !W{f.)'ll
'they are yellow', !l.w{f.kl 'you showed off'.
The second type of onset glides occur in bound state forms of feminine
nouns in which w is kernel-initial and is followed by ace sequence. Before
we discuss these forms, let us remind the readers of their morphological
relation with the rest of the inflectional paradigm of nouns with conso-

32 On the gemination of causative is-l, v. § 5.4.

 THE SYLLABIFICATlON OF VOCOIDS 203

nant-initial sterns, and show how they contrast with the analogous forms
in nouns in which the stern beg ins with a vowel ." The forms in (16) below
are all singular forms. They are the free state and bound state forms of a
masculine noun and of the feminine diminutive derived from it. In the forms
on the left-hand side the kernel begins with /wCC/, whereas in those on
the right-hand side it begins with /uCC/.
(16) /wrz/ 'hinge' /urti/ 'garden'
mu a-wrz (a.wrz) urti (ur.ti)
mb u-wrz (u.wrz) w-urti (wur.ti)
fu t-a-wrs-t (ta.wr.st) t-urti-t (tur.tit)
fb t-wrs-t (t.wr.st) t-urti-t (tur.tit)
In (16) the forms on the left are the analogues of those in lines a and c of
the paradigm displayed in (36) in § 2.5 .1; the forms on the right corre-
spond to those in lines a and c of (35) in the same seetion. The noun on
the left-hand-side displays the morphology which is normal for nouns with
consonant-initial sterns. The augment a- which occurs in the free state forms
drops in the bound state ." Like all other nouns whose stern begins with a
vowel, the noun on the right-hand side does not have an augment. Both
forms in the second line of the table begin with the bound state prefix.
The forms which are of interest in the present discussion are the feminine
bound state forms like twrst in (16)fb. Other examples are given below in
column I in (17). Column 11 contains words where a syllabic U occurs in
similar contexts, in conformity with SonPeak.
(17) I 11
a. t-wrta-t 'ko feline' t-urti-t 'garden dirn'
b. t-wsk-in 'pendants' F lt-usrih-t 'slit olives'
c. t-wznawas-r" 'glow-worm' lt-uzlim-t 'husk" dirn'
The word-initial t-wCC sequences occur as a result of dropping the initial
vowel to form the bound state form . For instance the bound state form twrtat
((l7)I-a) is formed by dropping the prefix /-a-/ in the free state form

33 On the morphology of vowel-initial nouns, see § 2.5.1 and § 2.5.2.

34 Sinee a noun in the bound state is always preeeded by the word which govems it, it is
impossible to elieit the pronuneiation of bound state nouns after a pause . When we eite a
bound state form out of eontext , the pronuneiation we give is that oeeurring after a contoid ,
e.g. after the preposition d 'and '.
35 (Jewelry). Besides twznawast and twskin the only other similar bound state noun with
a syllabic frieative is !twidatt, a plaee name (u !tawidatt) .
36 From /t-wznawaz-t/, cf. tiwznawazin, 'id, pu' .
37 Of argan nut.
204 CHAPT ER SEVEN

/t-a-wrta-t/. Here are further examples. In the list below the first form is
the free state, and the second the bound state."
(18) a. t-i-wrtat-in twrtatin (t.wr)
b. t-a-wrs-t twrst (t.wr.st)
c. !t-a-wrda !twrda (t.wr)
d. t-a-wryu-t twryut (1. wr.yut)
e. t-a-wlk-t twlkt (1. wl.kt)
f. t-a-wlzi-t twlzit (t.wl)
g. t-a-wnza twnza (t.wn)
h. t-a-wntllis-t twntllist (1. wn .tl)
Example (18)a, a plural form (v. (17)a for the corresponding singular), is
included to illustrate the fact that in nouns with a kemel-initial glide, that
glide has the same realization in both numbers. In the forms in (15), (17)
and (18) one hears a short vocoid between wand the following conso-
nant. The timbre of this vocoid ranges between [u] and [g].
The syllables wr and ws in the bound state forms t. wr.tat, t. ws. kin
((17)a,b) are not remarkable in themselves. As we have seen in Chapter
4, syllables of the form GC (G a glide) are commonplace in Tashlhiyt as
long as they immediately follow a vowel, as in the case of the corresponding
free state forms ta.wr.tat, ti. ws.kin. Wh at is noteworthy about the bound
state forms is that the glide violates SonPeak. Whereas any combination
of a glide and a consonant may constitute a well-formed syllable as
long as SonPeak is not violated (see e.g. wt in i-wtlan 'hare, p' ,39 yJ
in mra y-fsi 'if it had melted '), the syllables in which an on set glide
violates SonPeak are tightly constrained. They all fall under the following
generalization.
(19) WC-SYLL: In a syllable in which the onset violates SonPeak,
a. the onset is w ;
b. the nucleus is a coronal sonorant or fricative.

7.3.2. Glide gemination

In t-wrta-t (t.wr.tat) and other similar forms presented in the previous
seetion, SonPeak is violated and the violation gives rise to a syllable which
falls under the parameters of WC-SYLL (19). We now turn to the other
cases where a morpheme-initial glide is a sonority peak in the underlying

38 The nouns in (18) have the following rneanings: (a) ko feline, p; (b) hinge, dirn; (c) hollow
in which to put sauce ; (d) bridle (horse) ; (e) ko leather bag, dirn; (f) ankle ; (g) fringe of
hair; (h) scroll.
39 Singular a-wtil.
 THE SYLLABIFICATION OF VOCOIDS 205

representations. We shall see that in these cases the glide is realized as a

geminate, and consequently SonPeak can be met. The cases in question
fall into two morphological categories, feminine bound state forms and
verbs.

7.3.2.1. Feminine bound state forms

Consider the feminine noun t-a-wtil-t 'hare, f' and the corresponding bound
form, which is derived thereof by dropping the prefix /-a-/. In the same
way as /zwi=as/ is realized as zuyas (see (l O)a), one might expect
/t-wtil-t/ to yield tutilt. However the correct form is tuwtilt, where the homor-
ganic sequence uw is the realization of a geminate glide, as we suggest later.
As we did in the previous section, let us first set the scene by showing
how the forms under consideration contrast with analogous forms in which
the stern begins with a vowel. The table (20) below is exactly parallel
with table (16).40
(20) /wtil/ 'hare' /uday/ 'Jew'
mu a-wtil (aw.til) uday (u.day)
mb u-wtil (uw.til) w-uday (wu .day)
fu t-a-wtil-t (taw.ti.lt) t-uday-t (tu.da.yt)
fb t-uwtil-t (tuw.ti.lt) t-uday-t (tu.da.yt)
The bound state forms which are presently of interest are those in the
feminine, see (20)fb. The masculine bound state forms «20)mb) will be
taken up later.
Here is a list of all the feminine bound state forms known to us in
which an underlying glide is realized as the homorganic vowel-glide
sequence. The first form in each line is the free state, and the second the
bound state."
(21) a. t-a-wtil-t tuwtilt (tuw)
b. t-a-wrir-t tuwrirt (tuw)
c. lt-a-wrav-t ltuwravt (tuW)42
d. t-a-wsrvin-t tuwsrvint (tuw.sr)
e. t-a-ysvl-t tiysvlt (ti.ys.vlt)
f. t-a-ynnri-t tiynnrit (ti.ynn)"

40 The meanings of the feminine forms are 'female hare' and 'Jewish woman ' .
41 The nouns in (21) have the following meanings: (a) hare, f; (b) hill; (c) yellow, f; (d)
ko medicinal plant; (e) foul-brood ; (f) ko medicinal plant, indiv; (g) goatskin (container), dirn;
(h) shepherd (the occupation); (i) dog, f; (j) ear of cereal; (k) amulet; (I) worm ; (m) old
one, f.
42 Cf. (15)a.
43 Cf. (31)a.
206 CHAPTER SEVEN

g. t-a-yddit-t tiydditt (ti. yd.ditt)44

h. t-a-yssa tiyssa (ti.ys.sa)
1. t-a-ydi-t tiydit (tiy.dit)
j. t-a-ydr-t tiydrt (tiy.drt)
k. t-a-wmmis-t tuwmmist (tu ,wm.mi.st)
l. t-a-wkka tuwkka (tu.wk.ka)
m. t-a-wssar-t tuwssart (tu.ws.sa.rt)
Consider first the examples in (21)a-d. We give in (22) the surface repre-
sentation of tuwtilt.

(22)

~ ;1/1
(J (J (J

X X [ X X X X] X

t
I <>
U
I
t
I
I
I
I
I
t

We have indicated the kernel by enclosing it between square brackets. It

will be shown later that sequences uw and iy are in some cases realiza-
tions of the geminates /ww/ and /yy/. We assurne that in the realization of
/t-wtil-t/ as tuwtilt (tuw.ti.lt) the initial glide in the kernel is geminated.
The representation in (22) conforms to GlideFaith (9) : in (22) the initial
X slot in the kernel is a margin. Conformity with GlideFaith is achieved
at the cost of violating FAITH(SHORT):
(23) FAITH(SHORT): Every segment which is short in the input
should be short in the output.
As a result of its gemination, the glide can comply with GlideFaith without
violating SonPeak (2): in tuwtilt in (22) the sonority peak uw does contain
a nucleus."
The gemination whereby /w/ surfaces as uw in tuwtilt involves creating
an additional skeleton slot. The name 'FAITH(SHORT)' was adopted for
the sake of convenience, but it should not conceal the fact that the constraint
is actually a member of the DEP family, a family of constraints prohibiting
epentheses of various kinds." Note that what needs to be inserted is merely
a skeleton slot and its association with /w/. That the epenthetic slot is parsed
as a syllable nucleus follows from the constraints invoked in Chapter Four,
notably NoOns-.

44 From m a-yddid.
45 Tautosyllabic uw and iy are realized as [u:] and [i:].
46 On DEP, see McCarthy and Prince (1995) .
 THE SYLLABIFICATION OF VOCOIDS 207

In (2l)a-d the gemination of the initial glide is motivated by the need

to avoid a violation of GlideFaith. But no such violation would occur if
the glide were not lengthened in the other forms in (21). Form (21)e, for
instance, would surface as tysvlt (t.ys. ylt), which is parallel to twskin
(t. ws.kin) in (17)b. The kernel-initial glide occurs as a sonority peak in
both forms, but only the second is well-formed. The difference between
the two forms is that twskin falls under branch a of generalization WC-
SYLL (19) whereas tysvlt does not. As a matter of fact, y can never violate
SonPeak. Similarly, if gemination did not take place, It-wmmis-t/ «21)k)
would surface as twmmist (t. wm.mi.st), in which the nucleus of syllable
wm fails to fall under case b of WC-SYLL (19), which requires a coronal.
Let us pause briefly to recapitulate what we have seen so far.
An underlying glide surfaces as a vowel only in environments in which
GlideFaith is in conflict with undominated constraints such as NoHiatus,
e.g. in Izwi=asl zuyas «(10)a). A glide which is a sonority peak in the
input representation is syllabified as an onset, in violation of SonPeak,
only if the resulting syllable meets condition WC-SYLL (19), see e.g. t.wr.tat
in (17)a. Otherwise the glide is geminated , which allows it to surface as a
margin without violating SonPeak, see e.g . tuw.ti.lt in (2l)a, where it
surfaces as a coda, and tu.wm.mi.st in (2l)k, where it surfaces as an onset.
Note the special role played by condition WC-SYLL (19) in the preceding
account : WC-SYLL defines what the permissible violations of SonPeak are.
WC-SYLL is admittedly nothing more than arestatement of the facts . It
combines the effects of two prohibitions which involve independent phonetic
dimensions: one prohibition favors nuclei with a place of articulation dif-
ferent from that of the preceding onset (w is both dorsal and labial, while
coronal consonants are neither) ; the other prohibition disfavors syllables
in which the onset is a vocoid and the nucleus a stop. The sonority curve
of such syllables deviates maximally from that of those syllables which
are presumably the most favored cross-linguistically, those with a stop as
an onset and with a vocoid as a nucleus. We shall not try to formulate
these prohibitions. Let us simply take WC-SYLL as an unanalyzed whole
and add it to OUf account as an undominated constraint. Once generaliza-
tion WC-SYLL is used as a constraint, one aspect of it which one may
find particularly worrysome is the fact that it refers to another constraint,
viz. SonPeak. One can reformulate WC-SYLL to avoid reference to
SonPeak. One can state it as a condition on syllables with a consonantal
nucleus and with a nonconsonantal onset which does not immediately follow
a vocoid. However this does not change the fact that WC-SYLL is the
description of a regularity rather than an account for that regularity.
Since the beginning of this chapter we have encountered evidence for the
constraint rankings which are represented below. In (24) a line between
two constraints indicates that they are crucially ranked, the one on top being
ranked higher.
208 CHAPTER SEVE N

(24) NoHiatus (I) NoRR (11) WC-SYLL (19)

I I
GlideFaith (9) - - - - - - - - .
I
FAITH (SHORT) (23)
I
I
SonPeak (2)

(25) gives a form which provides erucial evidenee for eaeh of the rankings
in (24) :

(25)
NoHiatus » GlideFaith /zwi=asl zü.yas *z.wi.as
. (10)
NoRR » GlideFaith /zwi=asl zü.yas *kw.yas (10)
GlideFaith » SonPeak It-wrta-t/ t.wr *.tur. (17)a
GlideFaith » FAITH(SHORT) It-wtil-t/ .tuw. *.tu. (21)a
WC-SYLL » FAITH(SHORT) It-wmmis-t/ tu.wm *t.wm (21)k
FAITH(SHORT) » SonPeak It-wrta-t/ t.wr *tu.wr (17)a

The rankings in (24) imply that an underlying glide should violate SonPeak
whenever the resulting syllable meets WC-SYLL (19) . They make an ineor-
reet prediction for one form: they prediet that the bound state form of
t-a-wssar-t ((2l)m) should be twssart (t.ws) rather than tuwssart (tu.ws).
We leave this form unaeeounted for.

7.3.2.2. Verbs and masculine bound state forms

Imdlawn Tashlhiyt has verbs whose kernel begins in uw. Here is a sample
of those verb s. For eaeh we give the perfeetive stern and the imperfeetive
stern:
(26) a. uwzn (uw.zn) tt-uwzan 'weigh'
b. uwda (uw.da) tt-uwdu 'be sufficient'
e. uwzad (uw.zad) tt-uwzad 'prepare'
d. uwkkl (u.wk.kl) tt-uwkkal 'delegate'
e. uwdda (u.wd.da) tt-uwdda 'repay'
We hold that in all these verbs the underlying form of the kerne I begins with
a nongeminate Iw/, i.e. the underlying forms in (26) are Iwzn/, Iwda/, Iwzad/,
ete ., and the initial glide is subjeet to gemination. Our justification for
this claim is the following. The items in (26) represent all the different
eanonical forms which are found in the uw-initial verbs. Under the proposed
analysis these canonical forms are the following : (a) CCC, (b) CCV, (c)
CCVC, (d) CC:C , (e) CC:V. All these eanonical forms are attested out side
of the uw-initial verbs, with the same eonjugational paradigms. For instanee
 THE SYLLABIFICATlON OF VOCOIDS 209

uwzn conjugates like bxs 'cause to become awful', whose imperfective is

tt-bxas, uwda conjugates like sda 'lean against', whose imperfective is
tt-sdu, and so on." The alternatives to this analysis are unappealing. In (26),
if uw were derived from luwl or Iww/, most of the verb types in (26)
would not have paralleis outside the uw-initial dass.
The kerneI of a uw-initial verb begins with uw no matter what the other
morphemes in the same word are, e.g. y-uwzn 'he weighed', n-uwzn=as 'we
weighted for hirn' .
Gur analysis of gemination in t-uwtil-t and the like (v. (21)) carries
over to the verbs in (26)a-c,48 but more must be said of cases (26)d,e. We
list in (27) and (28) all the verbs of types (26)d ,e which we have been
able to find.
(27) a. uwmml (u.wm.ml) 'intend'
b. uwmmn (u.wm.mn) 'promise to spare'
c. !uwxxr (u.wx.xr) 'move back'
d. uwdda (u.wd.da) 'repay'
e. uwddn (u.wd.dn) 'call to prayer'
f. !uwddr (u.wd.dr) 'squander'
g. uwttq (u. wt.tq) 'register'
h. uwkkd (u.wk.kd) 'insist'
i. uwkkl (u.wk.kl) 'delegate'
j. !uwggd (u.wg.gd) 'alert'
k. !uwqqr (u.wq.qr) 'spare'
(28) a. uwlla (u.wl.la) 'appoint to a post'
b. uwrri (u.wr.ri) 'come back'
c. !uwssa (u.ws.sa) 'leave instructions '
d. uwssaa (u.ws.saa) 'let pass'
e. uwssr (u.ws.sr) 'grow old (animalj""
In the verbs in (27) the consonant which follows the glide is not a coronal
sonorant or fricative. If their initial segment were realized as a simplex glide,
e.g. if Iwmmll (27)a were realized as wm.ml in violation of SonPeak, the
initial syllable would not meet condition WC-SYLL (19). The glide in the
verbs of (27) must geminate, then, for the same reason as that in the bound
state form tuwmmist (lt-wmmis-tl, (2l)k).
The verbs in (28) are another matter. They would meet condition WC-
SYLL (19) if their initial glide were not geminated, and the parallel with
the verbs in (15) incorrectly leads one to expect wlla instead of uwlla,

47 An exception must be made for uwrri (impf tt-uwrruy) 'come back' . The only other /CC:i1
verb we know of is nqqi 'be clean' , whose impf stern is tt-nqqi, not "u-nqquy.
48 Besides the three verbs listed in (26)a-c, the only other verb of this type we have
encountered is uwfa 'be in excess' .
49 Cf. (2I)m.
210 CHAPTER SEVEN

wrri instead of uwrri and so on. The items in (15) form quasi-minimal
pairs with those in (28) , compare for instance uwrri Iwrril (28)b, where
the initial glide geminates, with lwrrv I!wrf'yl (l5)a, where it does not. twrtat
and the other bound state forms in (17) and (18) illustrated the fact that
in nouns, initial glides which are sonority peaks in the underlying repre-
sentations must violate SonPeak whenever WC-SYLL (19) allows it. In
verb s with a similar make-up, on the other hand, SonPeak is violated in
some items, e.g. in lwrrv, but not in others, e.g . in uwlla.
An assumption in our discussion of the Iw/-initial verbs in (15) and in
the present subsection is that their initial glide is a sonority peak in the
underlying representation, which is indeed true in the forms listed in (15) ,
(27) and (28), which are sterns, and remains true in strings in which the se
sterns immediately follow a pause or a contoid. But what of the cases in
wh ich they follow a morpheme ending in a vocoid? The ans wer is that
the uw-initial verbs begin with a geminate no matter what. After a word-
final vowel, the initial geminate is realized as ww to avoid hiatus, compare
for instance uwzn-y (uw.znv) 'I weighed' (cf. (26)a), and ur ta wwzny
(taww.znv) 'I have not weighed yet', from lur ta wzn --y/. After the 3ms
prefix li-I, on the other hand, the initial geminate is realized as uw, as it
is after all the other prefixes, e.g . y-uwzn (yuw.zn) 'he weighed' . Let us
consider the two cases in turn.
The fact that the initial glide of the uw-initial verbs geminates even
after a word-final vowel is one of our reasons for assuming that syllabifi-
cation operates in successive stages. Consider again the string lur ta wzn--y/.
If syllabification operated from the start at sentence level there would be
no reason for the initial segment of Iwznl to geminate. Iwl is no more a
sonority peak in the string under consideration than it is in la-wtill 'hare' ,
and consequently it should surface as a simplex glide, as it does in awtil.
Let us recall our earlier assumption that in a sentence each stern is parsed
as an independent syllabification domain before the concatenation of the
resulting parses is subjected to sentence-level syllabification. This assump-
tion was crucial in our discussion of imperfective gemination and in that
of length alternations in the causative prefix, see Chapter 5. In the word
Iwzn--yl 'I weighed ', stern-level syllabification parses the stern as uW.zn.
In the input to sentence-level syllabification the representation of sentence
lur ta wzn--yl '1 have not weighed yet' is <urtao.uw.m.«:v>, where unsyl-
labified sequences are enclosed between angled brackets. Sentence-level
syllabification turns that representation into ur. taww. zn y.
Let us now turn to the forms with the li-I 3ms prefix, e.g. y-uwzn li-wzn/.
Why not instead i-wzn, which would meet GlideFaith and SonPeak as weIl
as yuwzn, and spare a violation of FAITH(SHORT)? Before we deal with
this case we must present the basic facts about the syllabification of the 3ms
prefix and explain how we propose to account for them.
Setting aside the cases under discussion , the syllabification of the 3ms
 T H E S YL LA B l F IC AT l ON O F VOCOID S 211

prefix in words pronounced in isolation is prett y straightforward. It surfaces

as a glide when the stern begins with a vowel, and as a vowel otherwise,
witness the following examples: y-ufa ' he found ' (cf. 3mp ufa-n) , y-itti
'he moved away' (cf. 3mp itti -n), i-waib 'he answered', i-rgl 'he locked' .50
Mo st of thi s distribution is accounted for by our constraints, notably
NoH iatus and SonPeak. What the se con straints leave unaccounted for,
however, is the fact that when the initial vowel of the kernel is high, it is
the prefix, rather than the following vowel, which surfaces as a glide to
avoid a hi atu s: /i-ufa/ yields yuf a , not iwfa? ' Let us assurne th at at the
stage of the derivations when the prefi x and the verb are first syllabified
together, the high vocoid at the beginning of ufa is already a nuc1eus,
while the pre fix is yet a skeletal slot unaffiliated with any syll abic con-
stituent. The input repre sentation /i-ufa/ and the two competing outputs yuf a
and iwfa are given below in (29).

(29) a. /i-ufa/ b. yufa

o c o o

X
I
I
N
I
X
I
X
11
I
X
I
;1
X
I
X
I
X
11
I
X
I
I U f a I U f a
c. iwfa
o o

f\ 11
X X X X
I I I I
I U f a

We assurne that when syllabification operates, its output mu st satisfy the

constraint s in such a way as to differ as little as pos sible from the input
repre sentation. yuf a and iwfa are both optimal with respect to the con straints
presented earlier, but yuf a departs less from the input representation.
At what stratum of the grammar does syllabification operate when it
changes (29)a into (29)b ? One could suppo se that the domain of syllabifi-

50 After a vowel the prefix in the last two forms is reali zed as a glid e: asku y-wai b
'be cause he answered', w-a=lli y-waib-n 'the one (ms) who answered', aiku y-rgl 'because
he locked ', t-i=lli y-rgl-n 'those (fp) who locked ' .
5\ Words beginn ing with iw are otherwise attested, e.g. iws-n 'they helped' .
212 CHAPTER SEVEN

cation is the word, but let us try not to multiply strata beyond neces sity.
Let us assurne only two levels of syllabification; at one level it is sterns
which are taken as domains of syllabification and at the other it is whole
sentences. It is sentence syllabification which changes (29)a into (29)b; there
is no word-level syllabifi cation . Recall that it is syllabification in the stern
stratum which is responsible for the fact that IwznJ and the other verbs in
(26) begin with a geminate vocoid in all their occurrences. Iwznl becomes
luw.znl in the stern stratum and the 3ms prefix surfaces as a glide in
y-uwzn for the same reason as it does in y-ufa.
As is implicit in the paragraph immediately under (29), all the constraints
in diagram (24) take precedence over the requirement that the output of
syllabification differ as little as possible from its input. This requirement
only makes a difference in situations in which more than one candidate parse
is optimal with respect to the constraints in (24). This is the case with
li-ufa/, but it would not be with /i-rgl/ 'he locked', for instance, which is
pronounced irgl. The stern Irgll is syllabified as r.gl at the output of the stern
stratum, which is parallel to u.fa . However, whereas the outputs yu.fa and
iw.fa fared equally weil with the constraints in (24), this is not so for yr. gl
and ir.gl: the former violates SonPeak while the latter does not. Con-
sequently ir.gl is selected as the grammatical output. Resyllabifying the
nucleus r as a coda is the price to pay to avoid unnecessary violations of
the constraints in (24).

7.3.2.3 . Other stem- initial glides

Our assumption that syllabification operates on sterns before it operates

on whole sentences enables us to account for the discrepancy between the
pronunciation of the uw-initial verbs and their morphology. At the same
time thi s assumption creates problems for the analysis of nouns with a
stern-initial glide. The only forms of such nouns we have discussed so far
are feminine bound state forms. Let us now deal with masculine nouns
and free state forms in either gender. Although the initial glide of the stern
in these words is not a sonority peak, here is the convenient place to deal
with them , since presumably all the relevant facts are still fresh in the
readers ' minds. We will give only one example for each case, but the reg-
ularities these examples illustrate are exceptionless.
Consider again la-wtill awtil 'hare' . If the stern I-wtill were first syl-
labified in isolation, it would yield I-uwtill, whence "awwtil after the initial
prefix is taken into consideration by sentence level syllabification." The
plural form in both states is i-wtlan , not *iwwtlan nor *yuwtlan, and sim-
ilarly the plural of t-a-w rir-t 'hill' is t-i-wrar, not *tiwwrar nor *tyuwrar.

52 Homosyllabi c aww is attested in Imdlawn Tashlh iyt, witness a-wwt if ' heap of threshed
cereaI' .
 THE SYLLABIFICATlON OF VOCOIDS 213

Masculine bound state forms raise a similar problem. The bound state
of awtil is u-wtil, not *w-uwtil. Let us recall the facts about the syllabifi-
cation of the bound state prefix lu-I and that of its fronted variant li-I
before i-initial nouns ." These facts are similar in all respects with those
presented earlier about the 3ms prefix of verbs. The bound state prefix is
realized as a glide before vowels and as vowel before consonants. Here
are free state nouns with their corresponding bound state forms: udad,
w-udad 'ibex', ilf, y-ilf 'wild boar', a-rgaz, u-rgaz 'man', i-kzin, i-kzin"
'puppy' .55 The pair a-wtil 1 u-wtil patterns just like any noun with a con-
sonant at the beginning of its stern (a-rgaz, u-rgaz), provided we find a
way to prevent its initial glide from geminating.
Before we deal with nouns like a-wtil 1 u-wtil, in which the stern begins
with an underlying glide, let us review how our analysis handles nouns
like udad 1 w-udad and ilf 1 y-ilf, in which the stern begins with a vowel.
The gliding of the bound state prefix in w-udad and y-ilf is accounted for
in the same manner as that of the 3ms prefix in y-ufa in the preceding
subsection. Consider for instance w-udad. /U-udad/, the input to syllabifi-
cation at sentence level, is a representation analogous to (29)a, which results
like (29)a from stern-level syllabification. In that representation the prefix
tul is unsyllabified while the vocoid at the beginning of the stern is already
a nucleus . By making the prefix into an onset, sentence syllabification avoids
hiatus and at the same time it leaves unchanged the syllable structure present
in the input.
We now return to a-wtil 1 u-wtil and the like. If stern-level gemination
operates on the string Iwtill taken in isolation, how can we prevent it from
geminating the stern-initial glide? Nouns like a-wtil have the same mor-
phological properties as the vowel-initial nouns in which the stern begins
with a consonant, e.g. a-rga: 1 u-rgaz: The augment (a-) and the bound state
prefix (u-) are inflectional affixes, and consequently they do not belong
to the stern . Let us assurne, however, that in nouns whose stern begins
with a glide (a-wtil) or with a consonant (a-rgaz), the augment and the bound
state prefix are included in the string which is subjected to stern-level
syllabification. Under this assumption, the initial glide of Iwtil/ is not
anymore a sonority peak in the strings which are inputs to stern-level syl-
labification and there will be no need for it to geminate.
We have just divided verbs and nouns into two categories, as far stem-
level syllabification is concerned. In the items of one category, the strings

53 On this variant, see § 2.5.2.

54 In this pair the initial vowel in the first form is the nominal augment, whereas that in
the second form is the fronted variant of the bound state prefix, see § 2.5.2.
55 After a vowel the prefix is pronounced as a glide (v. § 7.1.3): i-kti w-rga: 'the man remem-
bered' , dduew-rga: 'under the man', ikti y-kzin ' the puppy remembered', dduey-kt in 'under
the puppy' .
214 CHAPTER SEVE N

which are syllabified in the stern stratum correspond exactly to the notion
'stern' employed throughout this book : astern is what remains after a
word has been stripped of all its inflectional affixes. All the verbs belong
to this first category (e.g. gn 'sleep' , uwzn /wzn/), and also all those nouns
in which the stern begins with a vowel (e.g. aylal 'bird' , udad 'ibex'). In
the items of the other category, the strings subjected to stern-level syllab-
ification must include, in addition to the stern proper, the affix which
immediately precedes it when that affix is the augment or the bound state
prefix . This second category is comprised of all the nouns whose stern (in
the strict sense) does not begin with a vowel, e.g. a-rgaz 'man' , a-wtil 'hare' .
The above difference between the two categories, which has to do with
the delimitation of syllabification domain s, may be related to a difference
in their morphological structure. The stern may be word-initial in the first
category but not in the second category, e.g. the stern occurs at the begin-
ning of the word in gn-n 'they sIept' and in the singular free-state form aylal
'bird', but the paradigm of a-rga: 'man' does not contain any word begin-
ning with rgaz: Another fact which may be relevant here is the following :

(30) In nouns with an augment, if the initial segment of the kerne I

is a vocoid, it must be an underlying glide.

If (30) were not true, the following situation would obtain. Alongside
t-a-wtil-t and t-a-wrta-t, in which the first segment of the kernel complies
with GlideFaith when the augment drops in the bound state (cf. t-uwtil-t
(21)a and t-wrta-t (l7)a), there would exist nouns in the bound state form
of which that first segment simply surfaces as a vowel, in compliance
with SonPeak, e.g. nouns t-a-wsil-t and t-a-wrsa-t whose bound state forms
would be t-usil-t and t-ursa-t. That such a contrast does not exist is a remark-
able fact which is crying out for an explanation.
Let us finally discuss the masculine bound state forms in which the bound
state prefix immediately precedes a kernel which begins with an under-
lying /y/. The only forms we have been able to find are those listed below
(the corresponding free state forms are given in parentheses).

(31) a. u-ynnri (a-ynnri) 'ko medicinal plant col'

b. w-iydi (a-ydi) 'dog'
c. w-iyda (a-yda) 'patrimony'
d. w-iyddid (a-yddid) 'waterskin'

The first noun is the only one to behave in accordance with our analysi s.
u-ynnri behaves like u-wtil. On the other hand, the bound forms in the
other nouns look as though they have undergone syllabification in the
stern stratum , e.g. /u-ydi/ w-iydi (31)b is analogous to /i-wzn/ y-uwzn 'he
weighed' . But if these nouns were marked in the lexicon as exceptions
undergoing syllabification at the level of sterns, we would also expect
 THE SYLLABIFICATION OF VOCOIDS 215

their glides to geminate in the free state, contrary to fact (*a-yydi). We

do not have a solution to the problem posed by (31 jb.c.d. 56

7.3.3 . Surface glides (codas) which are sonority peaks

Constraint NoRR (11) is not violated in any of the data presented thus
far. However Imdlawn Tashlhiyt has rimes which violate NoRR. All such
rimes have a liquid as their nucleus and w as their coda. NoRR is for instance
violated in /t-s-rwseak/ (3fs-cau-resemble=dat2ms) 'she gave you the
impression (that ...)' .57 Whereas the analysis recapitulated in (24) predicts
tssrüsak ttss.rü.sak), the correct form is ts.srw.sak. Here is for instance a
well-formed line of verse in which the sequence /srw/ is parsed as a heavy
syllable."
(32) 1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a ys- srw sa kl na lis dil- la sl ma kan
We compare below the pronunciation of /t-s-rws=ak/ with those of forms
in which the rimes u and uw occur after an onset which is a sonorant.
(33) a. /t-s-rwseak/ ts.srw.sak [ts:( g )ru(w)srek]
b. /srus=as/ 59 s.ru.sas [srösres]
60
C . /s-nwws=as/ s.nuw.sas [snu:sres]
r gives the impression of being longer or more prominent in (33)a than in
(33)b. The parentheses around [~] in (33)a do not indicate an alternation
between two free variants, but our inability to distinguish consistently by
ear between [CR] and [C@R] (R a nasal or a liquid) when R is syllabic.?'
The whole nonconsonantal seetion between [r] and the kernel-final [s]

56 In w-iydi the bound state prefix does not assimilate to the following i, as it does e.g. in
y-ilf lu-ilfl 'wild boar, b' . There are various ways of accounting for this fact. One way
would be to order the fronting rule (rule WI/YI in § 2.5.2) before the pass of syllabifica-
tion which is responsible for the gemination of yod in Iw-ydi/. Another is to formulate rule
WIlYI so that it can only apply if the front vowel triggering fronting belongs to the kerne\.
As iIIustrated in (22), the extra slot added to the glide by gemination does not belong to
the kerne\.
57 In Imdlawn Tashlhiyt the verb Irwsl exists only in the causative. The closest item built
on a monomorphemic stern is rwas (impf tt-rwa s) 'resemble, seern', but the morphological
relationship between ss-rws and rwas in a synchronie description of Imdlawn Tashlhiyt is
not a regular one. On the gemination of the causative prefix, v. § 5.4.
58 a i-ss-rwseak I-hai ised i-lias l-makan 'Ah! you have the impression (lit, 'the situation
gives you the impression') that the world is darkening'. The metrical pattern of this line is
the same as that of the piece in Appendix II.
59 'Lay down for hirn! impf' .
60 ' Entertain for hirn!' (cau-amuseedatßs),
6\ V. § 6.3.2.
216 CHAPTER S EVEN

may sound somewhat longer in (33)a than u in (33)b, but definitely not
as long as uw in (33)c. Since it occurs between two coronals, lul is fronted
in (33)b,62 while the vocoid in (33)a shows no more signs of fronting than
that in (33)c. Finally, whereas the vocoids in (33)b,c are steady-state, that
in (33)a gives at times the impression of a closing diphthong.
Consider now the verb rwl (r.wl) 'flee' , which is the colloquial variant
of the more formal lwr cited in the text under (6). This verb allows us to
examine syllabifications of Irwl in arieher array of morphologically related
forms than Is-rwsl does. li-rwll 'he fled ' is parsed as ir.wl and sounds like
[irwal] whereas /i-rwleak/ (3ms-flee=dat2ms) 'he escaped from you' is
parsed as i.rü.lak and pronounced accordingly (with an unfronted u), and
these pronunciations are as predicted by the analysis summarized in (24).
That analysis leads one to expect t.rü.lak for /t-rwleak/ 'she escaped from
you', but the correct form is trw.lak, with r as the nucleus of the first
syllable." The examples below recapitulate all the relevant data.
(34) a. tr.wl It-rwll 3fs
tr.wlt /t-rwl-t/ 2s
a' . r.wl Irwll imper
[.wh' Irwl-'YI ls
b. trw.lak /t-rwlsak/ 3fs + dat2ms
b'. jw.lat Irwl=at/ imper=2p
c. ir.w] li-rwll 3ms
d. i.rü.lak /i-rwleak/ 3ms + dat2ms
in.na.rü.Iat /i-nna rwl=at/ 'he said "flee!"
In all these examples the onset-nucleus sequence wl sounds like [w~l] and
the rime rw sounds like that in (33)a (see text immediately under (33».
For instance tr.wl in (34)a is pronounced [t(~)rw~l] and trw.lak in (34)b is
pronounced [t(~)ru(w)lrek]. The cases whieh are problematie for the analysis
summarized in (24) occur in the forms in lines b-b', where Iwl is not
preceded by a vocoid and is not followed by a segment which can be parsed
as a nucleus, as is also the case in /t-s-rwseak/ in (33)a. These forms show
that GlideFaith dominates NoRR, contrary to what has been assumed until
now. All the cases which show the effects of this ranking fall under the
following generalization.
(35) RW-RIME: In a rime in whieh the coda violates NoRR,
a. the nucleus is a sonorant;
b. the coda is w.

62 On u fronting, see § 3.8.

63 ME does not consider trülak as absolutely unacceptable, but he finds it very sloppy .
Acceptability judgements are made more difficult by the existence of Tashlhiyt dialects in
which the normal pronunciation of this form is trulak .
 THE SYLLABIFICATION OF VOCOIDS 217

In all the data given so far to illustrate RW-RIME the consonant which
precedes w is r . Here are examples of Cw rimes in which the nucleus is
another sonorant: lur t-lwi aqqay-n/" ur.tlw.yaq.qa .yn 'she did not pick
dates'; lur t-s-nwi aman/" ur.ts.snw.ya.man 'she did not boi1 water'. The
morphemes which can be used to provide the relevant strings are very
few and we have not found any which would enable us to check the facts
for mw. The following pair illustrates the fact that sonorants can be nuclei
in Cw rimes, but obstruents cannot. /i-s -nwiead/'" 'this cooking' can only
be pronounced i.snw.yad (not *is.nu.yad), whereas li-!gzwi=ad/ 'this brook'
can only be pronounced lig .zu.yad (not *!i.g~w.yad) .67
Violations of NoRR are not very common in the languages of the world.
Two such violations are found in the varieties of English and of German
where in codas Irl is realized as a nonsyllabic nonhigh vocoid, e.g . Eng.
here [hia], Germ. Uhr [u:1-]. For a case in Ath Sidhar Rifian, which involves
rimes consisting of a syllabic Irl followed by a glide , see DT (1993: 43-46).
Let us modify our analysis by re-ranking NoRR below GlideFaith and
by adding (35) as an undominated constraint. (24) is replaced by (36):

(36) NoHiatus (1) RW-RIME (35) WC-SYLL (19)

I I ' - - 1_ _ --'I
GlideFaith (9)
I < - - I- - - - - - -

I
,

NoRR (11) FAITH (SHORT) (23)

I
I
SonPeak (2)

Forms such as Izwi=asl zuyas in the second line in table (25) now become
crucial evidence that RW-RIME (35) is ranked above GlideFaith: if /zwi=asl
were parsed as ZW.yas, with z the nucleus in the first syllable, RW-RIME
would be violated. It is GlideFaith which is violated instead, hence the
correct form zü.ya s. If, as in (36), RW(RIME and GlideFaith are the only
constraints which dominate NoRR, the implication is that in Imdlawn
Tashlhiyt all the well-formed representations which violate NoRR fall under
generalization (35), which is in fact true.
Diagram (36) indicates that WC-SYLL must be ranked higher than
GlideFaith, which diagram (24) did not, because the data examined earlier
did not provide evidence on this point. The new evidence is the fact that
/i-rwleak/ in (34)d yields i.rü.lak; with a violation of GlideFaith, rather than

64 lwi, negative of lwa 'pick unripe fruit' .

65 ss-nwi, negative stern of ss-nwa ' (cause to) cook', from nwa 'cook , ripen' .
66 isnwi, an action noun derived from nwa 'cook ' .
67 The same state of affairs obtains in Lmnabha Moroccan Arabic, v. § 9.3.3.2.
218 CHAPTER SEVEN

yrw.lak, which complies with RW-RIME and GlideFaith but violates

WC-SYLL.
Given the constraints in (36), /i-rwleak/ should be realized as i.ruw.lak,
which achieves compliance with both GlideFaith and SonPeak at the cost
of a violation of FAITH(SHORT) (23), as is the cases in t-uwtil-t /t-wtil -t/
(21)a . (36) must be supplemented with an undominated Contiguity constraint
which prohibits the 'insertion' of skeletal slots inside morphemes."

7.4. GEMINATE GLIDES

As already noted by Basset (1946), vowels do not have an underlying length

contrast. On the other hand glides do, like the other consonants. The
examples below illustrate the contrast between simple and geminate glides
after a vowel.
(37) a. t-aqqay-in 'Adam's apple p' t-aqqayy-in ' walnut p'69
b. i-rnuyag 'adze p' i-guyya 'head p' 70
c. IWI-n 'take away 3mp' iwwi 'my son '
d. a-wtil 'hare' a-wwtif 'threshed wheat'
e. t-a-ydr-t 'ear (cereal)' t-a-yydar-t 'woman' s name
f. zuyd-n 'be born 3mp' t-uyyl-t 'fly 2s'
g. ssiwd-n 'be afraid 3mp' !swiwwd-n 'burn (pain) 3mp'
h. t-aqqay-t 'walnut' i-zaraeyyt ' let hirn search'
i. t-duy -t ' you woke up' i-zlueyyt 'let hirn lose'
Apart from the fact that they have greater duration, postvocalic gem inate
glides occupy more extreme locations in the vowel space than their simplex
counterparts. When compared in such pairs as t-ayyu-t 'camel's hump' vs.
ur ta y-ut (neg yet 3ms-strike) 'he has not struck yet ' , and !w-ayyad 'other
m' vs. la-bayad ' wasteland' , yy is closer to cardinal vowel [i] than y,
which one is tempted to note [cx]. Similarly the rounding and/or protru-
sion of the lips is greater in the glide in i-zawwa-n ' strong wind p' 71 than
in a-zawar 'nagging' .
After a consonant, the first half of a geminate glide is realized as the
corresponding high vowel , as we shall see below . The examples in (38) illus-
trate the contrast between simple and geminate glides in that context. In
(38)a, for instance, the underlying representations are la-fyasl and la-fyyas/.

68 On that constraint, see Kenstowicz (I 994b) and McCarthy and Prince (1995).
69 Both nouns have identicaI. singular forrns: taqqayt.
70 From a-mayg and a-gayyu .
71 Singular a-iawwu.
 THE SYLLABIFICATION OF VOCOIDS 219

(38) a. a-fyas 'unpeeled argan nut' a-fiyas 'braggard'

b. nwa 'cook' nuwa 'intend'
c. hya 'resurrect' hiya 'be magnificent'
d. t-a-ryal-t 'ko basket' lt-a-riyas-t 'direction,n
e. t-a-rway-t ' barley porridge' t-a-ruway-t 'disorder' 73
As implied by our transcriptions in the right-hand side column in (38),
after a consonant a geminate glide is homophonous with a high vowel
followed by the homorganic glide, for instance Iwwl is homophonous with
lu+w/. Since the existence of this homophony or its systematic nature can
easily be overlooked, let us illustrate it with minimal pairs involving three
different morpho-syntactic contexts.
The kernel in Itt-swwasl 'make lots of noise impf'" is phonetically undis-
tinguishable from that in li-suwasl 'two-pronged pitchfork p' ;75 both are
pronounced suwas. Simi1arly, the sequences la-nwwasl and lanu#w-assl
at the beginning of the sentences in the following pair are homophonous
except for the final consonant.
(39) /a-nwwas" ai-ga I'=u- !dwwarl anuwasaygalvuduwar
u-informer AD 3ms-be iueb-village
'a rat (an informer), that's what he is in this village'
(40) lanu w-ass a i - g a f=u-!dwwarl anuwassaygalfuduwar
weIl m-bad.omen AD 3ms-be oneb-village
'the weIl, it is bad luck that itbrought to the village'
Gur next pair illustrates an analogous homophony before CCV :
(41) Ifwwt 77 -x=asl fuwtxas
spend-1 s=dat3s
'I spent for hirn'
(42) lasku ut-xeas/ askuwtxas
because strike-I s=dat3s
'because I struck for hirn '
In our last example the geminate high vocoid in the first sentence resu1ts
from the total assimilation of the genitive preposition Inl to the initial

72 lriys 'lead (performers), be in charge' ; lr-rays 'director, skipper, travelling musician' .

73 Adeverbal noun of shape /a-CC:aC/ derived from rwi 'stir' .
74 The corresponding perfective is /swws/ suws ([su:s]).
75 The corresponding singular is a-saw s. a-saw s forms its plurallike a-madl 'mountainside',
p i-mudal.
76 Cf. (48)g below.
77 This verb is conjugated below in (45)a.
220 CHAPT ER S EV EN

sonorant of the following noun." When the optional assimilation rule applies
the two sentences are homophonous.
(43) /i- !rza hmad ixf n=u--yyul/ lirzalunadixfuwvyul
3ms-break hmad head ofeb-donkey
'Ahmed broke the donkey's skull'
(44) /i- !rza hmad i-xfu u-vyul/
3ms-break hmad 3ms-vanish:aor b-donkey
'Ahmed broke a limb and the donkey vanished'
It is only for reasons of convenience that the three pairs above all involve
back vocoids. Lack of space prevents us from citing similar pairs to illus-
trate the homophony of /Cyy/ and /Ciy/. The homophony under discussion
follows from two facts: (i) when a geminate immediately follows a segment
of lower sonority, the first half of the geminate is syllabified as a syllable
nucleus, and (ii) when an underlying glide is syllabified as a nucleus, the
resulting surface vowel is homophonous with an analogous underlying
vowel." By (i), postconsonantal /ww/ is syllabified as /Y:{w/, and by (ii) ,
/u/ and nuclear /w/ have identical surface manifestations. Fact (i) is exem-
plified by the syllable structures of such forms as /i-hlls-n/ 'saddle prt'
and /i-tt-hllas-n/ 'saddle, impf prt ', which are i.lill.sn and itt.hl.la.sn. This
fact is accounted for by the analysis in Chapter 4, notably by constraint
NoOns-, which forbids the first half of a geminate to be an onset. Fact
(ii) was presented earlier in this chapter (§ 7.2).
Given the homophony of /ww/ and /uw/ and that of /yy/ and /iy/ after
a consonant, it is considerations about distributional restrictions or about
the morphological make-up of words which enable us to tell apart those
surface sequences which derive from a geminate glide and those which
derive from a vowel followed by a glide. Let YjGj stand for a surface
sequence uw or iy. Let us review a few cases in which the available evidence
clearly points in one direction or the other.
If a YjGi sequence is heteromorphemic and if furthermore there is no
reason to suppose that it has undergone assimilation, then that sequence
cannot be the realization of a geminate (on geminates v. chapter 3). A few
such cases have appeared earlier in our discussion. In d#u-wtil 'with the
hare' (v. (6)h), u is the bound state prefix while the following w belongs
to the kernel. In sniriy 'mimic impf' (v. (6)g) , y is the last segment in the
root," while the preceding i is a vowel which is inserted to form the imper-
fective stern (cf. the imperfective tt-gnuguy, in (4)c, and its perfective gnugi).
In i-lkkuwsa 'pruning hook p' (v. (6)a) w, which also appears in the singular

78 See § 3.2.1.2 on this assimilation.

79 Except in the case of back high vocoids in u-fronting environments, see (13).
80 That same segment is realized as a vowel in the perfective stern sniri.
 THE SYLLABIFICATION OF VOCOIDS 221

a-lkkawsu, belongs to the root, while u belongs to the /ua/ melody which
serves as a plural marker in certain nouns (cf. a-gayyu 'head' , p i-guyya).
In lt-i-mssuwra ' state of being a counsellor', a plural noun derived from
la-msawr-iy 'counsellor' (v. DE 1992), w belongs to the root while u belongs
to the plural melody /ua/.
What of the homomorphemic sequences? Let us first discuss two
instances where there are reasons to believe that they are the reflexes of
geminate glides.
First, consider the verbs whose perfective stern has the surface shape
CV jGjC.81 Apriori, their underlying forms could be either /CG:C/ or
/CVGC/. We give a few CVjGjC verbs in table (45) below, and for the
sake of comparison we add examples of CVCC and CC :C verbs. Three
sterns are given for each verb : perfective, negative and imperfective.
(45) pf neg impf
a. fuwt fuw(i)t tt-fuwat ' spend '
b. huwl huw(i)l tt-huwal 'worry'
c. qiyd qiy(i)d tt-qiyad 'record'
d. !ziyr !ziy(i)r ltt-ziyar 'tighten'
e. !rufz !ruf(i)z ltt-rufuz 'be reluctant'
f. biks bik(i)s tt-bikis 'gird'
g. hawt haw(i)t tt-hawat 'negociate'
h. sxxn sxx(i)n tt-sxxan 'dip (in sauce)'
1. bddl bdd(i)l tt-bddal 'exchange'
The CVCC verbs and the CC:C verbs insert a vowel before their last
consonant to form their imperfective stern; in the CVCC verbs the inserted
vowel is a copy of V (v. (45)e-g) while in CC:C verbs it is a (v. (45)h,i) .82
As illustrated in (45)a-d the CVPjC verbs pattern with the CC :C verbs:
they insert a in the imperfective. This fact is our first reason for assuming
that in the CV jGjC verbs the V.G, sequence is a geminate glide, e.g. the
underlying form of [uwt (v. (45)a) is /fwwt/ and not /fuwtl. One cannot
attribute the ill-formedness of *tt-fuwut and *tt-qiyid (from qiyd, v. (45)c)
to a constraint against sequences uwu and iyi , for such sequences occur
elsewhere in Imdlawn Tashlhiyt. uwu occurs in nuwu and quwu, which
are the aorist sterns of nuwa 'intend' and quwa 'be strong', and iyi occurs
in the negative sterns of the CVjGjC verbs, e.g. in qiyid and lziyir «45)c,d);
iyi also occurs in hiyi, the negative stern of hiya 'be magnificent', and in
its ls and 2s forms in the perfective, hiyi-x, t-hiyi-t.
Our second reason for assuming that in CVjGjC verbs the VjGj sequence
derives from a geminate glide, is the existence of verbs of the form aaG:C,

8\ Such verbs were already given in the left-hand side colurnn in (6)b,e.
82 On irnperfective sterns, see § 5.2.
222 CHAPTER SEVEN

e.g. aayyb 'criticize' . As has been shown in § 3.7, in Imdlawn Tashlhiyt

all occurrences of tautomorphemic aa can be derived from 111 by the
following obligatory rule, which has very few exceptions:
(46) 1-TO-aa: 1 ~ aa I except when adjacent to a high vowel
For instance 6ssisl ' mount guard' is realized as aassis, /i-d'ia' 'pray for
someone, 3ms' is realized as idaa." while the corresponding aor form
/i-d'iu/ is realized as idfu. As implied by rule (46) , fis adjacent to a high
vowel in all the surface forms in which it occurs, e.g . fis ' survive', Yum
'swim' , lbif 'sale', labyuf 'sale p' .
Consider the three aaG :C verbs which are given in (47)a-c below.
These verbs conjugate like ICC:CI verbs (v. (45)h-i); the 11C:CI verbs in
(47)d-f are given for the sake of comparison:
(47) pf neg impf
a. 11yybl aayyb aayy(i)b tt-aayyab 'criticize'
b. 11yydl aayyd aayy(i)d tt-aayyad 'spend time"
c. 11wwll aawwl aaww(i)l tt-aawwal 'count on'
d. 11rrsl aarrs aarr(i)s tt-aarras 'stand in full view '
e. Imml aallm aall(i)m tt-aallam 'put a mark on'
f. l !lmmrl !aammr !aamm(i)r ltt-aammar 'fill'
The underlying forms of the verbs in (47)a-c can only be 6G:CI, e.g . aayyb
can only derive from 11yyb/. Deriving aayyb from Ifjiybl would require
allowing J)I to be tumed into aa in front of a high vowel, something which
otherwise never happens. Wh at the existence of the aaG:C verbs shows,
then, is that the lexicon of Imdlawn Tashlhiyt must contain some ICG:CI
verbs. A distribution which would only allow 111 as an initial consonant
in the underlying form of the ICG:CI verbs would be oddly skewed.
Analyzing all the CVP;C verbs as ICG:CI would avoid this problem.
Our third reason for assuming that in the CVjGjC verbs the VjGj sequence
is a geminate glide has to do with the memberships of the three classes
of verbs exemplified in table (45) . The CC:C verbs are numerous, and so
are the CVjG jC verbs, while there are not so many CVCC verbs. If we
take the CVjGjC verbs to be ICC :CI verbs of a particular kind, then the
fact that there are many CV jGjC verbs is simply a special case of the fact
that there are many ICC:CI verbs. On the other hand, if we took the CVjG jC
verbs to be ICVCCI verbs, then we would have a rather peculiar distribu-
tion among the ICVCCI verbs, most of them containing a high vowel
followed by a homorganic glide.

83 Unless fis adjacent to a high vowel, sequences /'ia/ and /a'i/ surface as aa.
84 During a religious festival.
 THE SYLLABIFICATION OF VOCOIDS 223

The second instance where we have evidence that a YjGj sequence is

the surface reflex of a geminate glide is that of the a-CV;GiaC nouns,
which in many cases are clearly instances of the templatic a-Cc.·aC nouns.
These are illustrated below in (48):
(48) a. a-sbbav 'dyer' sbv 'dye'
b. !a-nddam 'poet' !ndm 'cornpose poetry'
c. a-zmmal 'wholesaler' zmml 'buy wholesale'
d. t-a-qiyat-t 'lqayd status' l-qayd 'ko judge'
e. la-siyad 'fisherrnan' !siyd 'go fishing '
f. a-fiyas 'braggard' fiys 'show off'
g. a-nuwas 'informer' nuws 'malign'
h. a-quwad 'pimp' quwd 'mediate'
i. !a-suwag 'cattle hand' !sug ' drive (herd)'
The a-Cc.·aC nouns are occupational nouns. Since Imdlawn Tashlhiyt does
not have analogous nouns built on a template a-CVCaC,85 the YjGj sequence
in the aV;GiaC nouns in (48)d-h must be the reflex of a geminate glide.
a-fiyas (48)f represents an underlying /a-fyyas/.
A class of kernels with a YjGj sequence which cannot be traced back
to a geminate glide in the underlying representations is that of the
nouns ending in iy whose plurals end in ay, e.g. la-frdiy / !i-frday 'tree',
lt-a-zrbiy-t / lt-i-zrbay 'ko carpet' . In most of these the kernel has the
shape CCCiy. If iy were a reflex of /yy/ the canonical form of these kernels
would be /CCCC :/, and these would be the only existing nouns with an
initial vowel and a /CCCC:/ kernel. On the other hand, if we assume these
nouns to have a /CCCiy/ kernels, they fall together we other nouns of the
well-attested /CCCYC/ type, e.g. la -grnrid / ti-g''nrad, 'quail' a-lmbub /
i-bnbab 'pipe' .
We devote the end of this section to a review of the cases in which a
geminate glide is not realized as a vocoid, as has always been the case
thus far. These cases all share the following two properties: (i) the glide
involved is /ww/, which surfaces as gWgW (gg in delabializing contexts);
(ii) /w/ is the medial glide in a three-segment root and its gemination is
due to nonconcatenative morphology. The words involved fall into four
classes A to D. Below we give for each class a complete list of the words
we have encountered. Class A is comprised of the three-segment verbs

85 This is a simplification. Imdlawn Tashlhiyt actually has a few nouns built on a template
a-CVCaC, but they all have to do with physical defects. Here is a list of all those which
we have come across, with their source verb between parentheses when there is one.
!a-bukad 'blind' (!b(b)ukd 'be blind'), a-kusam 'paralyzed' (kusm 'be paralysed'), !a-bidar
'lame' (!b(b)idr ' limp' ), a-siban 'old' (sib 'grow old'), a-titaw 'stutterer', a-ziwal 'cross-
eyed' .
224 CHAPTER SEVEN

with a medial w. w becomes gWgW in all those which form their imperfec-
tive stern by geminating their second consonant, e.g. the impf stern of rwl
'flee' is rgwg"'l.86 Class B contains two a-CC:aC action nouns, a_rgWgWay
(rwi 'stir') and a_zgWgWay (zwi 'beat down')." Class C contains the
perfective sterns of two 'quality verbs', zgWgWay (aor izwiy) 'be red' and
IgWgWa y (aor ilwiy) 'be soft', and the corresponding nouns a_zgWgWay 'red'
and a_lgWgWay 'soft' .88 Finally, class D contains two singular nouns, !a-
dgWgWal 'member of the family of one's spouse' (p !i-dula-n) and a-zgWgWar
(p i-züra-ni 'jujube tree'. If we assurne that in the singular the underlying
kerneIs of these nouns are I!dwwal/ and /zwwarl, the alternations between
singular and plural are exactly parallel to those in other nouns such as
a-sllab / i-slba-n 'whip', a-skkif / i-skfa-n 'soup' .89 In the plural kerneIs
/!dwla/ and /zwra/ the glide can only be syllabified as a nucleus, see the
preceding sections. /w/ is also syllabic in the deverbal nouns t-a-rüla 'flight'
(rwl 'flee', class A above), t-a-zuvi 'redness' (zgWgWay 'be red' , class C)
and t-a-luvt 'softness' (lgWgWa y 'be soft', class C).

7.5. CONCLUSION

Imdlawn Tashlhiyt has an underlying contrast between high vowels and

semivowels. The contrast is neutralized next to a vowel, where underlying
vowels surface as glides to avoid hiatus . In other environments, the contrast
is in most cases maintained in surface forms, and the underlying glides have
three manners of realization, depending on context. In some contexts they
surface as high vowels, in compliance with the requirements of sonority-
driven syllabification. In other contexts they surface as glides, in violation
of those requirements. In yet a third class of cases, the underlying glides
geminate and surface as homorganic vowel-plus-glide sequences .
As it involves epenthesizing a skeletal slot, glide gemination is analo-
gous to the schwa epenthesis found in other dialects of Berber such as
Ath Sidhar Rifian , but whereas schwa epenthesis can occur next to any
consonant, in Imdlawn Tashlhiyt an empty skeletal slot can be inserted
only before an underlying glide. If, as is conceivable, Tashlhiyt had in earlier
times an across-the-board schwa epenthesis process analogous to that of Ath
Sidhar Rifian, glide gemination is its last remaining vestige within the

86 The other verbs in dass Aare lwr (lgWgWr), a formal variant of rwl, nwa (ngWgWa) 'be
cooked ', rwi (rgWgwi) 'stir (a liquid , e.g.)' , zwa (zggu) ' become dry ' , zwi (zgWg wi) ' beat
down', !zwi (!zgWgWi) 'ladle' , and zwur or zwar (zggur) 'be first' . The latter verb is excep-
tional in that it is the only CCVC verb which geminates a consonant in the imperfective.
87 Compare with a-bzzag 'swelling' ibrg 'swell ') , a-fttak 'sprain' (ftk 'sprain one' s X).
88 Compare with !mqqur (aor !imyur) 'grow' . The nouns a_zgWgWay 'red' and a_lgWgWay
'soft ' have regular plural forms: i-zgWgWa y_n and i_lgWgWay_n .
89 But /ww/ does not surface as a stop in !a-duwar ' vilIage' (/!dwwar/; p !i-dura-n).
 THE SYLLABIFICATION OF VOCOIDS 225

phonological component, but this is irrelevant from the point of view of

language typology: whatever the situation in other dialects or in earlier
stages of Tashlhiyt, the facts of present-day Imdlawn Tashlhiyt show that
there are languages in which any segment can be a syllable nucleus.
The discussion of the high vocoids ends our analysis of the syllable
structure of Tashlhiyt. We have argued that in Tashlhiyt any expression (any
word or sequence of words) may be exhaustively parsed into syllables
meeting the following conditions, among others:
(A) A syllable is of the form (O)N(D). The onset (0) is comprised of a
single skeletal slot. The coda (D) may be comprised of two skeletal
slots if these belong to the same geminate.
(B) The nucleus (N) is comprised of a single skeletal slot. It may be asso-
ciated with any segment (feature bundle) .
Let us briefly recapitulate the evidence for (A) and (B) . Part of the
evidence in support of (A) comes from poetry (text-to-tune alignment,
with its distinction between light and heavy syllables); the remaining
evidence is provided by two morphologically-governed alternations (imper-
fective gemination , causative prefix lengthening).
Proposition (B) goes hand in hand with our contention that none of the
short vocoids which one hears in Imdlawn Tashlhiyt is a vowel (a syllabic
vocoid). In fact, we claim that these vocoids are not even segments, i.e. they
do not have a skeletal slot of their own.
Our evidence in favor of that claim is twofold. On the one hand, there
are phonological processes which operate only on adjacent segments (fusion,
regressive devoicing). In Imdlawn Tashlhiyt these processes are only
blocked when the two consonants are separated by a full vowel, whereas
in Ath Sidhar Rifian, a Berber dialect with genuine epenthetic vowels,
consonant sequences which are broken up by vowel epenthesis are rendered
immune to fusion and to regressive devoicing .
On the other hand, although the regularities that govern the distribu-
tion of short vocoids have not been fully worked out, those which are known
only make reference to the feature content of the abutting consonants.
Moreover, they do so in a manner which is more suggestive of the effects
of phonetic implementation than of those of phonological epenthesis. Finally,
even in Berber dialects with genuine epenthetic vowels, vowel epenthesis
cannot account for all the short vocoids which are heard : in addition to bona
fide epenthetic vowels, Ath Sidhar Rifian has transitional vocoids like those
of Imdlawn Tashlhiyt.
Tashlhiyt is in close contact with Moroccan Arabic. The Moghrebian
dialects of Arabic have undergone massive vowel deletions, a phenom-
enon attributed by many to the influence of Berber. The phonology of
Moroccan Arabic is usually examined against the backdrop of the vari-
eties of Arabic spoken further East, which have better retained the vowels
226 CHAPTER SEVEN

of Classical Arabic. However, our discussion of Imdlawn Tashlhiyt and

Ath Sidhar Rifian has shown that two languages with very similar mor-
phologies may nonetheless differ markedly in their syllable structures. In
the next chapters we examine Moroccan Arabic in the light of what we
know about its neighbour Tashlhiyt, a language which allows vowelless
syllabIes .
 CHAPTER EIGHT

SYLLABLE STRUCTURE IN
MOROCCAN ARABIC

8.1. INTRODUCTION

The varieties of Berber and of Arabic spoken in Morocco have rather similar
surface phonologies, e.g. their segment inventories do not differ much.
Reading the literature on Moroccan Arabic (henceforth MA) might give
the impression that Tashlhiyt and MA have almost identical syllable
structures, at least at those levels of representation which are near the
terminal representations of the phonological component. One of our aims
in the forthcoming discussion is to show that such an impression would
be mistaken. Our outline of the syllable structure of MA in surface repre-
sentations will be detailed enough to allow us to point out the basic
differences with Tashlhiyt. It is also meant as an original contribution,
both factual and analytic, to the study of MA phonology.
The facts which could lead one to suspect virtually identical syllable
structures in MA and in Tashlhiyt come from two areas, transcriptional
practice and versification.
As already noted in Chapter 6, the transcriptions commonly used for MA
and those used in works on various Moroccan dialects of Berber show
very similar restrictions on consonant clusters and on the occurrences of the
unstable vowel. In some of these works, furthermore, these transcriptions
are accompanied by phonetic facts analogous with those we have adduced
as evidence in favor of our analysis of syllabification in Imdlawn Tashlhiyt.
For instance HarreIl (1962a) notes that MA allows relatively long voice-
less stretches of speech . He illustrates his point with 'an utterance of seven
syllables of which the first six are voiceless in ordinary conversational
delivery', which he transcribes /xossok tf;:)tt;:)s fassotta/ .' On the basis of
similar facts, Durand (1995/96) claims that short voiced vocoids, when they
are present at all, are but phonetic manifestations of the syllabicity of an
adjacent consonant. In his view, the word meaning 'he wrote', which is pro-
nounced [kt@b], is a monosyllable Iktbl in which the nucleus is /b/ .
The second area in which the facts of Tashlhiyt and those of MA may
at first sight look very similar, if not identical, is versification. In DE (1988:
10) the authors pointed out the close resemblance of syllabification in MA
songs to that in Tashlhiyt. The evidence adduced was the scansion of a
four line stanza from a MA song. At the end of a book devoted to versi-

1 !xess=ek t-fettes fe=s= sett-a, in the transcription adopted here (see infra). The sentence
means 'you have 10 inspecl at six o'c\ock' .

227
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
228 CHAPTER EIGHT

fication in some Moroccan dialects of Berber, Jouad (1995 : 304-314)

presents syllabic scansions of material drawn from several MA songs.
Jouad's transcription and his discussion give no inkling of the differences
between orthometric syllabification in MA and in Tashlhiyt.
Unlike in Berber, root-and-pattern morphology of the most unmistak-
able sort is alI-pervasive in MA , which makes it easier to discern the
respective contributions of word-formation and of phonology proper in
the sound pattern of the language. In particular, it provides compelling
evidence for the difference between those short voiced vocoids which are
associated with syllable nuclei and those which are not. Also, MA is rather
uniform, in contrast with the dialectal fragmentation of Berber, which means
that lingui stic variation is less of an obstacle to studies based on evidence
gathered from different consultants.'
Consider the form in (1) below, which means 'you (p) were manhandled' .
In (1) we give (a) a standard transcription for that form (v. below about
standard transcriptions), (b) its underlying representation, (c) its actual
pronunciation at a normal speech rate in the variety of MA spoken in Oujda,
and (d) the syllabic parse of (c), with dots to indicate syllable edges (this
syllabification will be justified later) :
(1) a b c d
ttekseftu /tt-ksf-tu/ ttks@ftu .t.tk [email protected].
We must provide an analysis which relates the underlying representation
(1)b and the pronunciation (1)c . Our analysis must in particular account
for the fact that the unstable vowel appears between sand f rather than
between k and s or betweenfand t. This will be done by reference to syllable
structure.
Rather than immediately present our analysis in its final form , we will
go through the three stages depicted below in (2). This manner of exposi-
tion makes it easier to show exactIy how each new item of data we present
along the way contributes to shaping our final analysis.
In the first stage, evidence from versification will lead us to analyse
the form as containing four syllables (see (1)d). We will first develop an
analysis in which the third syllable, . s@f, which has a voiced vocoid,
represents the normal case. We will assurne an intermediate representa-
tion et.tek.sef.tu in which every syllable contains a vowel and we
will attribute the lack of any vocoid in syllables .t. and .tk. to superficial
phenomena such as deletion in word-initial position or devoicing between
voiceless consonants. At this point of our discussion, in syllable sefthe rime
ef will contain a nucleus e and a coda J, as represented in (2)a below .

2 For an overview of the history and dialects of Arabic in Morocco , see Colin (1985) .
 SYLLABL E STRUCT URE IN MOROCCAN ARABIC 229

(2) a. (j b. o c. o
->.R
0
»<.R
0
-<.R
0

N
r-.D I
N N
I

x x
I I
x x x
-<.x x
I
x
I
s
I I I
S
I I I I
e f e f S f

However, joint consideration of phonotactics and syllable weight in versi-

fication will lead us to the conclusion that J actually belong s to the nucleus
of se! At the end of the second stage of our analysis, schwa forms a complex
nucleu s with the consonant which follows it, as shown in (2)b .3 Relyin g
on this assumption, we will develop a set of constraints which go some
way towards assigning syllable structure to any expre ssion in the langu age.
The constraints will also allow us to pinpoint some of the differences
between the syllable structure of MA and that of Tashlhiyt.
In the third stage of our analysis we will reconsider the internal struc-
ture of syllable s such as seJ and suggest that their nucleu s is not a complex
one, after all. We will argue that the analysis with a vowel is needed only
in special contexts and that in the other contexts seJ only contains two
segments, the onset s and the nucleus J, see (2)c. Except under special
circumstances, the short voiced vocoids which occur immediately after onset
consonants are not segments; they are mere transitions between the onset
and the nucleu s. We will show that this final analys is ties up in a simple
way variou s loose ends of our second analysis.
Thi s chapter is devoted to the stages of our discussion which are rep-
resented in (2)a and (2)b; (2)c will be the subject of Chapter 9.
Thi s chapter is organized as follow s. § 8.2 is devoted to prel imin ary
remark s on the present state of our knowledge about the pronunc iation of
MA and on the nature of the data. In § 8.3 we set up an inventory of the
syllable types of MA, drawing our evidence from syllabification in chanting
and singing as weil as from the phonotactics of words. § 8.4 shows that MA
allow s violations of the con stra int SonPeak which cau se ill-formedness
in Tashlhiyt. § 8.5 provides the groundwork for a constraint-based account
of word-level syllabification in MA which is consi stent with the syllable
inventory drawn up in § 8.3.

3 On the syllabification of postpausal consonant clusters, see § 8.3.I.

230 CHAPT ER E IG HT

8.2. STANDARD TRANSCRIPTIONS

For MA as for Berber, in the present state of research on the phonology

of the language, one important difficulty resides in the problematic nature
of part of the data. In this section we attempt to draw the border between
those areas of the data for which the transcriptions of MA found in various
works (including the present one) are a rather direct reflection of what
one can actually hear in MA speech, and those areas in which implicit
assumptions about the phonology and the morphology of the language
play an importantrole in shaping the transcriptions. Unle ss stated otherwise
our data is drawn from the dialect spoken by ME,which is in essence that
of Lmnabha, in Western Morocco, where he spent his childhood (see
§ 1.6).

8.2.1. The distribution 0/ 'e ' in standard transcriptions

In our discusssion of MA, ' word' is as a rule an abbreviation for 'Pword'. 4
When we want to refer to syntactic words we will use the expression 'syn-
tactic word' in full. For the rest, unless stated otherwise, in our discussion
of MA we use the symbols for boundaries and speech sounds and the various
abbreviatory conventions in the same manner as in our discussion of Berber.
Besides using '= ' to mark the left edge of enclitics (direct object pronouns
and possessive determiners) we will also use this symbol to mark the right
edge of the definite article 11=1, that of prepositions whose underlying form
is comprised of a single consonant, e.g. /b=1 ' with' , and that of ulw 'and'.
Like that of Ath Sidhar Rifian, the vowel inventory of MA consists of
four voweIs, the three full vowels la, i, ul and an unstable vowel e, whose
realization depends to some extent on the nature of the surrounding
segmenrs.' In the following discussion we use the letter 'e' with two values ,
depending on the context. It may be part of a 's tandard transcription' (see
below) or it may represent a voweI in a representation which we posit in
the phonological component. It is commonplace for an occurrence of 'e ' not
to have a corresponding vocoid in the pronunciation. On the other hand ,
in this chapter as everywhere else in this book, the symbol '@' always
stands for an audible object, namely a short voiced vocoid.
MA has a systematic distinction between simplex consonants and their
geminate counterparts, but vowels do not have an analogous length

4 As defined in § 2.2, a Pword is a sequence which comprises a word together with all
the c1itics attached to it.
5 Some authors posit two contrasting unstable vowels e and o. We interpret the rounded
unstable vowel as a variant of e due to a neighboring labialized consonant, e.g. we posit
kWeli and x'eb: where HarreIl (I 962b) writes koll ('all') and xob z ('bread'). On this analysis,
see Heath (1987: 254-263).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 231

contrast." The reflexes of e in most contexts sound shorter than the full
vowels. Although the full vowels are systematically recorded as long vowels
by some authors,' they do not give the impression of having an especially
long duration. Furthermore, it will be seen later (v. § 8.3.2) that in versi-
fication, syllables ending with a full vowel always count as light, while they
should count as heavy if full vowels were geminates.
As is the case in Imdlawn Tashlhiyt and in Ath Sidhar Rifian, a stop is
as a rule pronounced with an audible release when it precedes a pause or
a non-homorganic stop.
We will often need to set apart the unstable vowel from all the other
segments. Following the terminology of Heath (1987), by a full segment we
mean a full vowel or a consonant. Whereas the lexicon and morphology
play a central part in determining the nature and sequential ordering of
full segments in words, they play only a marginal role in determining the
distribution of the unstable vowel, which is to a large extent govemed by
phonotactic regularities valid for all words , regardless of their morpho-
logical structure. Consider for instance the ms active participle kateb
'writing' and its mp counterpart katb-in. The lexical entry of the verb kteb
'write' and the morphology jointly produce the sequence !katbl as the stern
of the active participle. In the mp form l#katb-in#1 the vowel of the mp
suffix I-inl allows Itbl to surface as a licit intervocalic cluster, hence the
output katb in. On the other hand, in the ms form, which is the naked stern
l#katb#l, e appears in order to avoid a word-final cluster preceded by a
full vowel, an ill-formed sequence in MA, hence the surface form kateb.
Let us use the expression 'standard transcriptions' as a cover term for
the transcription in Harrell (1962b) and similar transcriptions used in Harris
(1942), Harrell and Sobelman (1966), Abdel-Massih (1973), Keegan (1986)
and Youssi (1992), which do not differ much from one another as far as
the distribution of the unstable vowel is concemed. According to the
standard transcriptions the distribution of the unstable vowel and the
consonant clusters of MA are not very different from those we have found
in Rifian Berber.

6 By 'an analogous contrast' we mean an underlying contrast between vowels linked to

one and two skeletal position s. MA actually has a surface contrast between short vowels
and long ones. Sequence iy is realized as a long i when it is tautosyllabic, and so is the
sequence customari ly transcribed as eyy. Such sequences occur in imiyl=u 'his little camel'
and in meyyl=u 'he put hirn aside' . The two words are homophonous from m onwards; they
both end in [mi:lu]. Similarly the following two expressions, the first a sentence and the second
a noun phrase, are homophonous from s onwards: ma sewwl-u=ni 'they did not interrogate
me' (v. sewwel 'he interrogated') and lun kaseu w=lun=i 'the color of his glass and my color'
(color glass=3ms andecolore l s). Both expressions end in [su:luni].
7 For claims that the full vowels of MA are underlyingly long or geminate, see e.g. Cantineau
(1950), Lowenstamm (1991) , and also Kouloughli (1978) for an Arabic dialect of Eastern
Aigeria with facts similar to a certain extent.
232 CHAPTER EIGHT

As far as the full segments are concerned, standard transcriptions pose

no special problem: they can be taken as 'broad' phonetic transcriptions,
and adequate ones on the whole. Matters are different as far as the letter
'e' is concerned. It would be greatly mistaken to interpret the occurrences
of 'e' in the standard transcriptions as we do those of 'a' or 'i", i.e. as
consistently representing a voiced vocoid in the pronounced forms . Our
position about 'e' is stated in (3) for later reference:
(3) The letter 'e ' in standard transcriptions merely indicates that
the preceding consonant is a syllable onset.
In many instances the following nucleus is not (must not be) realized with
a voiced vocoid; ample evidence for (3) will be provided in the next section
when we examine versification. Consider again the syllabic parse t.tk.s@ftu
in (2)d and the corresponding standard transcription ttekseftu in (2)a. The
form in question contains three syllables without full vowels . The standard
transcription contains no 'e' for the initial syllable .t. because it lacks an
onset (t is a nucleus). In the two other syllables the onsets are t and S, as
indicated by the location of the occurrences of 'e'; the first 'e' has no
corresponding voiced vocoid in the pronunciation. Rather than a stand-in
for a class of speech sounds, we propose to view 'e' in standard tran-
scriptions as a device for encoding constituent structure. 'e' is comparable
to the dots used to indicate syllable edges. The analogy is not immedi-
ately apparent because as a rule the symbols used by linguists to mark
constituent boundaries have the following two properties: (i) they are not
letters of the alphabet, and (ii) they are located outside the strings which
they delimit, e.g. the dot marking the beginning of a syllable is located
before the onset, whereas 'e' occurs after the onset. As far as the unstable
vowel is concerned, then, the standard transcription of an expression reflects
more directly the syllable structure of that expression than its pronuncia-
tion. Once this is understood, discussing the distribution of e in the standard
transcriptions is a convenient means of getting acquainted with the basic
facts of syllabification in MA.
The following generalizations can be stated about the distribution of e
in words in the standard transcriptions: (i) e immediately follows a con-
sonant, and (ii) e is immediately followed by a CC sequence or by a
word-final C; in short , e only occurs in closed syllables. Generalization
(i) is exceptionless; generalization (ii) has a few exceptions, on which see
next chapter (§ 9.5). Here are forms which illustrate the distribution of e.

(4) I 11
a. qerd 'rnonkey' qerd-a 'id, f'
b . suret 'he locked' SUft-U 'they locked '
c. qeddem 'he presented' qeddm-u 'they presented'
d. dheb 'gold' dehb-i 'golden'
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 233

e. lzevret 'he ululated' lzvert-u 'they ululated'"

f. tte-xle? 'he got scared' tt-xel'i-et ' she got scared'
g. s=sder 'the jujube, col' s=sedr-a 'the jujube, indiv'
h. sekk 'he doubted' sekk-u 'they doubted'
In each line in (4) the second word is formed by adding a vowel-initial suffix
to the first word, in which the rightmost vowel is e. Except in the first
and last line s, adding a vowel-initial suffix changes the distribution of e
in the preceding stern. Some occurrences of e disappear, and the 'dele-
tions' may affect other occurrences besides the rightmost, see for instance
lines e and f. Furthermore the suffixed forms contain occurrences of e in
locations where the corresponding unsuffixed form has a consonant cluster,
see lines d to g.
The altemations in (4) are driven by the need to comply with certain lim-
itations on well-formed consonant clusters. As a provisional characterization
of what these limitations are, admittedly a crude one, let us accept for the
moment that all occurrences of e present in the standard transcriptions result
from the operation of the following procedure:"
(5) RIGHT-TO-LEFT SCAN:
Scanning the Pword from right to left, rewrite as CeC any CC
string which is not immediately followed by a full vowel or by
e. Each step in the scan must take as its input the output of the
previous step.
The procedure rewrites for instance I!Zl'rt/ as lzevret and I!zl'rt-u/ as fzyertu
((4)e). We assurne that e has a skeletal position of its own. The universal
inseparability of geminates prevents e from occurring between the two
halves of a geminate, and when (5) predicts that this should happen, the
site of insertion is shifted before the geminate. If RIGHT- TO-LEFT SCAN
operated without taking the universal inseparability of geminates into
account, in (4)c-II, for instance, one would expect /qddm-u/ to yield qdedmu,
but this is an impossible outcome because in the input /dd/ is a geminate,
and we get qeddmu instead. As in Berber, not all the geminates are under-
lying. In (4)g-I, for instance, geminate inseparability causes /s=sdr/ to
become ssder, not sesder. /s=sdr/ comes from a more abstract /l=sdr/, where
/1=/ is the definite article, and the initial geminate results from the total

8 An exclamation point prefixed to a form indicates that the form contain s dorsopharyn-
geaIized segments. Like Berber, MA has dorsopharyngealized coronals in the lexical
representations and dorsopharyngealization spreads to the neighboring segments at the phonetic
level.
9 This procedure was invoked in § 6.5 for Rifian Berber.
234 CHAPTER EIGHT

assimilation of /1=/ to the following coronaI. 10 The geminate ss is hetero-

morphemic, i.e. it is a single bundle of distinctive features Iinked to two
heteromorphemic skeletal slots."
RIGHT-TO-LEFT SCAN excIudes all word-final CC sequences except
geminates (e.g. sekk in (4)h-I) and it excIudes all medial and initial CCC
sequences except those beginning with a geminate (e.g. qeddmu in (4)c-II
and ttxel'iet in (4)f-II). The statements (6) and (7) below are logically equiv-
alent reformulations of these predictions.
(6) At the end of words
a. consonant sequences contain at most two skeletal positions ;
b. two-position sequences are geminates.
(7) EIsewhere than at the end of words
a. consonant sequences contain at most three skeletal posi -
tions;
b. in three-position consonant sequences, the first two posi-
tions make up a geminate.
Statements (6)a, (7)a and (7)b are factua:Ily accurate, disregarding a few
cases which are either isolated or unclear." On the other hand, (6)b is
only a first approximation to the facts, e.g. it does not square with qerd
'monkey', cited in (4)a-I. This statement will be replaced by a more accurate
one later on, but let us stay with (6) and (7) for the time being.

10 As a rule the definite article /1=/ completely assimilates to a follow ing coronal contoid.
11 In our transcription 'sesder' in (4)g-I a boundary symbol occurs between the two halves
of a geminate. As in all other transcriptions in this book, the boundary symbols only provide
information about morphemic affiliation at the skeletal tier ; they do not impl y anything
about the organisation of the distinctive features associated with the skeletal positions. The
symbol sequence ' s=s' implie s that we are dealing with two heteromorphemic skeletal posi-
tions but it doe s not indicate whether each skeletal position has its own associated feature
bundle or whether the two positions share a single feature bundle, in other words it does
not say whether we are dealing with two occurrences of simplex /s/ or with a geminate Ist.
A consequence of what precedes is that when a symbol '=' or '-' is sandwiched between
two occurrences of the same simplex consonant, it does not imply anything about the release
of the first consonant. In fact , in all the example s below, two occurrences of the same letter
with an intervening ' =' or '-' represent a geminate or a sequence homophonous with a
geminate. Qualifications will be added when we discuss consonant releases in the next chapter
(§ 9.4) .
12 (6)a does not take into account the negative enclitic /=8/, which has no phonological effect
on what precedes it (Heath 1987: 243) . (7)a disregards certain prefixes which give rise to
initi al CCCC sequen ces in HarreIl (l962b), as in n-tt-xel'i-u 'for us to be scared' (p. 49) ,
see also pp. 34, 46 and 57 for other instances. Empirical work has yet to be done on the
pronunci ation of such clusters as weil as on their behaviour in versification, to determine their
status with respect to syllable structure. The cases mentioned in this footnote will be ignored
in the following discuss ion . Some of the counter-examples to (7)b are initial clu sters
analogous with those ju st mentioned, e.g. n-dfen ' he was buried ' (HarrelI 1962b: 34, see
also pp. 46 and 57).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 235

Let us assurne that in MA a C is always tautosyllabic with a following

V, and that intervocalic CC sequences are all heterosyllabic. Under these
assumptions, RIGHT- TO-LEFT SCAN predicts that all occurrences of e
should occur in rimes whose codas are either single skeletal slots or
geminates, see, e.g., the initial syllables in deh.bi «4)d-1I) and in qedd.mu
«4)c-lI).

8.2.2. Uncontroversial schwas vs. putative ones

We will use a standard transcription when we discuss syllabification in

MA versification later in this chapter, but before we turn to versification
let us explain why, as far as the unstable vowel is concerned, the standard
transcriptions must not be construed as representations, however 'broad', of
the phonetic facts . Two points must be made : (i) 'e' does not always have
a corresponding voiced vocoid in the pronunciation, and (ii) one hears short
voiced vocoids in places where no ' e' occurs in the standard transcrip-
tions.
In what follows we will be in constant need to distinguish between two
kinds of sequences inside words, which we will call W-final and W-internal.
Here is how we define them :
(8) W-final, W-internal:
A sequence is W-final if it is contained in the last syllable of
a word; it is W-internal otherwise.
Consider n-ketb=u 'I write it', in which the last syllable is bu. Following
OUf definition, sequences band bu are W-final while sequences tb, ket
and nk are W-internal. Note that since a single segment is a sequence of
length 1, we can say that the e in zaweb 'he answered' is W-final, in the
special sense of 'W-final' we have j ust defined. This cannot lead to any
confusion with the normal sense of 'word-final' because e cannot be the
last segment of a word in MA.
We can now turn to point (i), i.e. 'e' does not always have a corre-
sponding voiced vocoid in the pronunciation. Let us first dweIl on W-final
sequences, which are the only environment in which the presence or absence
of 'e' in the standard transcriptions is always a faithful reflection of the
phonetic facts." MA has a contrast between W-final [C@C] and [CC], as
in the following pairs:

13 Cf. Heath (1987: 243).

236 CHAPTER EIGHT

(9) a. !shet 'he slashed' lseht 'slash (n)'

b. !nqes 'he diminished' !neqs 'shortage'
c. !zlet 'he ruined' !zelt 'ruin (n)'
d. !qnet 'he feIt bored' !qent 'boredom'
e. kasf-et 'she guessed' kasef-t 'I guessed' 14
f. nbet 'he pushed' neb-t 'I acted as a proxy'
g. lbesleek 'your onion' 15 beselk 'with a wire'
A voiced vocoid must be pronounced immediately before the final con-
sonant in the forms on the left, whereas no voiced vocoid may separate
the last two consonants in the forms on the right. Except in the last two
pairs, we have chosen examples in which the last two consonants are homor-
ganic or both voiceless. There are two reasons for our choice. First, voiced
vocoids, even very short ones, are particularly easy to hear when they
break up such clusters. Second, preliminary observations on the pronunci-
ation of the varieties of MA spoken in Oujda and in Lmnabha suggest
that these clusters never give rise to voiced transitional vocoids, i.e. to
voiced vocoids which are not segments in their own right, but rather mere
transitions between segments (see below).
The phonetic difference between CCeC and CeCC is often not as straight-
forward as in the examples in (9). Consider for instance dien 'he wandered'
and seit:' Ashlhiy' . The release of I is accompanied by glottal vibrations
in both words. In dien the release is the beginning of a vocoid which is
auditorily more salient than I. In seih, on the other hand, the voiced vocoid
n
between land gives the impression of having a shorter duration than
the preceding sonorant, which carries the main part of the pitch contour
on the word. The short vocoid which one hears between land in seih n
is merely a transition from one consonant to the next.
Examples (9)a-d illustrate the well-known contrast between CCeC forms
and CeCC forms. The CCeC forms comprise the citation forms of all the
triconsonantal 'strong' verbs and some nouns." On the other hand the CeCC
forms are mostly nouns, some of them derived from CCeC verbs, as in
(9)a-d. 17 (9)e exemplifies the contrast between the 3fs perfective suffix,
which must be pronounced with a schwa in most environments, and its ls
counterpart, which may not. The last pair in (9) illustrates the fact that
the 2s clitic must be pronounced with a schwa in most environments. The

14 Cf. kasef 'he guessed' .

15 Cf. !bsei 'onion ' .
16 The citation forms of verbs are their perfective 3ms forms, which are naked sterns.
17 Contrary to what the examples in (9) might suggest, CeCC nouns do not all contain an
emphatic consonant, nor do they all end in a voiceless coronal obstruent, e.g. qeib ' heart' ,
seili 'Ashlhiy', kehf 'cave' , senq 'hanging ', mesk 'rnusk ' . On the contrast between CCeC
and CeCC in MA nouns, see Amimi and Bohas (1996) and Zeroual (2000).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 237

contrast between [CC] and [C@C] at the end of a word exists in all vari-
eties of Arabic spoken in Morocco.
We turn next to sequences which are not W-final. It may be that the
standard transcriptions adequately represent syllable structure at an abstract
level, but elsewhere than at the end of words they are not to be taken as
accurate depictions of the sequencing of vocoids and contoids at the phonetic
level. This has already been pointed out by some authors.
Mitchell (1993: 60ff) begins his discussion of consonant clusters in
MA with remarks on the problems one runs into when one tries to identify
consonant clusters in the first place . According to Mitchell one difference
between MA and varieties of Arabic spoken further east is that in MA the
phonetic differences between CC clusters and C@C sequences are rather
elusive in many contexts, a difference that he ascribes to the influence of
Berber. Take for instance the MA expressions meaning 'the shoulder' and
'his shoulder', pronounced in isolation at normal tempo. Most authors
writing about MA would respectively transcribe them as leckte] and ketf=u.
Mitchell notes that the different occurrences of e in these transcriptions have
quite different phonetic correlates. While a voiced vocoid is clearly audible
between t and j in lektef, none occurs between k and t in ketfu. In ketfu
the unstable vowel is only manifested by a 'voiceless, very rapid, barely
audible transition' (p. 62), which is nonetheless different, according to
Mitchell, from 'the closer, inaudible passage from lkJ to /t/ in lekte!, (p. 62).
Mitchell does not voice any objection against postulating a phonological
vowel after the fir st consonant in ketfu and lektef, as long as it is clear
that 'from a phonetic point of view, these [forms] are typically pronounced
with [initial] three-consonant clusters' (p. 62).
In a similar vein, Heath (1987: 266) writes that in the form meaning 'they
wrote ', which other authors would represent as ketb-u, with a vowel between
k and t, his judgement is that in the variety of MA spoken in the FeslMeknes
region 'the transition [between k and t] does not constitute a true surface
segment; rather, it is simply the minimally necessary articulatory transi-
tion between segments' . He goes on to remark that no transition at all can
be heard in analogous forms involving fricatives, e.g. he cannot hear any
difference between the beginning of the form meaning 'they got drunk',
which other authors would transcribe as sekr-u, and that of Eng. screw. 18
To give an example of our own, we cannot hear any difference between
the sequence noted ket in ma ketr-u=s 'they have not been numerous' and
kt in those Parisian pronunciations of the French word actrice [aktxis]
'actress' in which the closure of k is released before that of t begins. Some
authors have ascribed instances like these to the devoicing of the unstable
vowel between voiceless consonants, see for example Harrell (1962a) .

18 Durand (1994, 1995/96) also insists that many schwas in the transcriptions do not
correspond to any vocoid .
238 CHAPTER EIGHT

Other instances of surface clusters have been seen as resulting from

the absorption of the unstable vowel by a neighboring sonorant which
becomes a syllable nucleus, see e.g. Harris (1942), Harrell (1962a), Heath
(1987) . !tenz 'haste' may for instance be pronounced [!t@nz] or [!tnz]. In
the latter realization the oral closure is maintained throughout the articu-
lation of [tn] and [tl is pronounced with nasal plosion . We chose an example
which involves a sequence of homorganic stops because auditory and/or
kinesthetic detection of an intervening vocoid is easiest in such sequences.
As already stated in § 6.3 .2, we often find it difficult to ascertain the
presence of an intervening vocoid in other CR sequences (R represents a
nasal or a liquid), e.g. between [q] and [n] in !qent 'boredom' or between
[z] and [m] in zemt 'choking'. It would be cumbersome constantly to remind
the readers of this uncertainty in our narrow phonetic transcriptions. We will
adopt a special convention about the narrow transcription of CR sequences
in which the two consonants are not homorganic noncontinuants. When such
a sequence occurs befare a consonant or a pause we uniformly represent
its pronunciation as [CR] without thereby implying the absence of a short
voiced vocoid between the two consonants. We will for instance repre-
sent the pronunciations of !qent and zemt as [!qnt] and [zmt]. The
transcription '[!qnt]' makes no commitment as to the presence of a voiced
vocoid between [q] and [n], because it falls in the purview of our convention.
In '[!tnz]' , on the other hand, the adjacency of ' t ' and 'n ' implies that
there is no intervening vocoid, for that transcription is not concerned by
our convention.
The problem with the standard transcriptions is not that they do not
faithfully record consonant clustering at the phonetic level. It is that their
relation to the phonetic facts is not known, because at present the phonetic
facts themselves are to a great extent a terra incognita, at least those
pertaining to consonant clustering in W-internal sequences. How little we
actually know at present about the surface distribution of short vocoids in
MA is concealed by the fact that most authors writing on MA take the
standard transcriptions for granted. 19
Dialeetal variation is one important factor in the confused state of our
present knowledge." In Morocco as in other parts of the Arab world, the
language used in teaching and in most formal aspects of public life is
Classical Arabic, a language which has no native speakers. In the colloquial
language dialectal variation is found primarily in the lexicon and in minor
aspects of the phonology. Unlike in Berber, linguistic variation in MA is
never an obstacle to intercomprehension. As far as phonology is con-
cerned, there is as yet nothing which could be called Common Moroccan

19 In dubious cases, the facts are sometimes inferred from other cases, rather than ascer-
tained by direct observation, see note 4 in Kaye (1987: 158) for an example.
20 On the language situation in Morocco, see Boukous (1995).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 239

Arabic, although such a dialect may be gradually emerging now in

Casablanca and Rabat. The discrepancy between W-internal occurrences
of ' e' in the standard transcriptions and short vocoids in the actual pro-
nunciation varies across dialects. lt is greater in the cities of Casablanca and
Rabat and in Southern MA (the area comprising Marrakesh and Taroudant),
and lesser in North-Eastern MA and in the so-called Aroubi dialects spoken
in the countryside around Casablanca and Rabat.
Dialeetal differences involving the unstable vowels are not merely a
matter of phonetic implementation, as shown by certain systematic differ-
ences we have observed between the dialect of Oujda, in North-Eastern
Morocco, and the Beduin dialect of Lmnabha, to the East of Taroudant.
There is a wider gamut of surface contrasts between CCeC and CeCC in
Oujda than in Lmnabha." Let '0' stand for an obstruent and ' R' stand
for a liquid or a nasal. Let us compare the possibilities of contrast between
/CCeO/ and /CeCO/ sequences in the two dialects, depending on whether
the medial C is a sonorant or an obstruent. In the two minimal pairs below,
which exist in both dialects, the first item is a verb and the second is an
action noun derived from that verb:
(l0) a. !dreb 'he struck' !derb 'striking'
b. !sxet 'he cursed' !sext 'cursing'
Consider now the related forms with the suffix -u (3p) and the possessive
clitic =u (3ms). In both dialects the contrast is neutralized regardless of
the nature of the medial consonant. Here are the forms in the Oujda dialect.
The necessity of avoiding occurrences of schwa in open syllables is what
seems to motivate the apparent metathesis in the verbal forms.
(11) a. lderb-u 'they struck' lderbeu 'his striking'
b. !sext-u 'they cursed' !sext=u 'his cursing'
Whereas the Lmnabha pronunciations of the forms in (ll)a do not differ
audibly from their pronunciations in Oujda, the forms in (ll)b are not
pronounced alike in the two dialects. Whereas glottal vibrations may occur
between the first two consonants in the Oujda pronunciation, none may
occur in Lmnabha and one hears [!sxtu].
Consider next analogous forms with the suffix -na (l p) and the posses-
sive clitic =na (lp). In Oujda the contrast is maintained in both pairs:
(12) a. !dreb-na 'we struck' lderbena 'our striking'
b. !sxet-na 'we cursed' !sext=na 'our cursing'
The Lmnabha dialect, on the other hand, has contrast (l2)a but not contrast
(l2)b. In Lmnabha the two forms in (l2)b are homophonous. They are

21 We thank Chakir ZerouaI for answering our questions about the Oujda diale ct,
240 CHAPTER E IG H T

pronounced as [!sxtna) , with no glottal vibrations before the onset of the

nasal consonant.
To sum up: Lmnabha allows a narrower range of surface contrasts than
Oujda in W-internal sequences . Immediately before CV, sequences CReO
and CeRO contrast in both dialects, whereas a contrast between COeO
and CeOO exists in Oujda but not in Lmnabha .
We now dweil brieflyon the second type of discrepancy between standard
transcriptions of MA and the distribution of voiced vocoids at the phonetic
level: certain short voiced vocoids are systematically glossed over in the
standard transcriptions, even though they do not seem to be phonetically
different from others which are instances of the unstable vowel. Two
examples will suffice, which are valid both in Oujda and in Lmnabha.
The expression /f=l=dbiz/ 'in futility' would standardly be transcribed as
feddbiz, and yet dd and bare separated by a short voiced vocoid which is
no less audible than that betweenfand dd. As our second example, consider
the forms meaning 'he diminished' and 'we diminished', respectively !nqes
and Inqes-na in the standard transcriptions. It is plain that a voiced vocoid
separates the oral release of n from the closure of q. One hears [!n@q@s)
and [!n@q@sna) .
The reasons why the standard transcriptions are selective in the short
vocoids they note are usually not made explicit, but they are not hard to
find. The standard transcriptions are to a certain extent embodiments of
analyses which relate the phonetic facts to certain morphological regular-
ities. The notation ' !nqes' rather than '!neqes' for 'he diminished' is based
on the recognition that the form in question is parallel to the 3ms perfec-
tive forms of other triliteral strong verbs, e.g. kteb 'he wrote' or !shet 'he
slashed' , and on the assumption that the occurrence of a short vocoid
between the first two consonants in !nqes is a superficial phenomenon
tied to the particular consonants involved." A similar case can presum-
ably be made for the short vocoid which occurs between dd and b ui feddbiz;
our other example. ddbiz (l1=dbiz/) is the definite form of a CCiC action
noun derived from dbe z 'fool around'. Whether a short voiced vocoid occurs
between the first two consonants in analogous forms derived from other
verbs depends on the nature of these consonants. No such vocoid can for
instance be heard between ss
and t in !sstih (l1=!stihl), from ! steh 'dance' ,
or between nn and tin nntif (/l=ntif/), from ntef 'pluck (feathers)'.
Our preliminary observations on the varieties of MA spoken in Lmnabha
and in Oujda suggest that the short voiced vocoids not recorded by the
standard transcriptions occur in the same types of clusters as give rise to

22 V. Heath (1987: 251, 263-265) for passages where such assumptions are made explicit.
V. also MitcheII, who writes (1993: 68): ' -;}- may be less of a phonetic segment in a given
case than a phonological fiction recogni zed [. . .) with a view of facilitating the formula-
tion of general structural patterns '.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 241

transitional vocoids in Imdlawn Tashlhiyt, e.g. they regularly occur in

sequences of two heterorganic stops the second of which is voiced." Pending
systematic research, one may speculate that MA could be like Ath Sidhar
Rifian in that it would possess short voiced vocoids of two kinds: some
would be vowels, i.e. syllable nuclei, while others would not be segments
in their own right; they would be mere transitions between segments.
Let us assurne that at some level of representation not too far removed
from the surface all syllable nuclei in MA are vowels. Depending on how
much is known at present about them, one can distinguish between two types
of vowels in MA, which we shall henceforth refer to as the 'uncontrover-
sial' vowels and the 'putative' ones. The uncontroversial vowels comprise
all the full vowels, all the W-final schwas as well as some W-internal ones
such as those in (12)a. We call these vowels uncontroversial because the
evidence as to their location in strings is clearcut. The uncontroversial
schwas give rise to surface contrasts like those in (9), (10) and (12). The
other syllable nuclei are what we call the 'putative' vowels . Given what
is presently known about the pronunciation of MA and about syllable
structure in other languages, the claim that W-internal occurrences of 'e'
in standard transcriptions mirror exactly the distribution of the unstable
vowel in (near-) surface representations is only a plausible surmise, and
empirical evidence is needed before this surmise can be turned into an estab-
lished fact. The required evidence will be presented in the next section.

8.3. THE STRUCTURE OF SYLLABLES IN MA

In this section we propose an analysis of the internal structure of sylla-

bles in MA. Our evidence is drawn primarily from chanting and singing.
The data of orthometric syllabification allow us to substantiate the claim
we made in (3) about the proper interpretation of 'e' in standard tran-
scriptions, a claim which may be formulated more precisely as follows:
the letter 'e' occurs if and only if the preceding consonant is an onset in
a syllable which does not contain a full vowel. Orthometric syllabifica-
tion shows that in word-initial clusters all the consonants but the last one
belong to a rime. Syllable weight distinctions in versification lead us to
the conclusion that some consonants belong to a syllable nucleus . Finally,
orthometric syllabification agrees by and large with the syllable structures
which can independently be inferred from the distribution of vowels and
consonants in the standard transcriptions.

23 See Chapter 6.
242 CHAPTER EIGHT

8.3.1. Hinge syllabies; syllable-final schwas

Versification provides us with a rieh source of evidence about the syllable

strueture of MA at a surface level. The MA material examined below will
be presented using the standard transcriptions, but only the consonants
and the uncontroversial vowels will playa crucial role in our aligning the
linguistic material with the metrical patterns imposed by versifieation. In
this subsection we first show that putative 'e's in the transcriptions only
occur in locations where verse structure requires syllable nuclei. This fact
is strong evidence that the surmise in the last paragraph of the preceding
seetion is basically correct."
Although the unstable vowel always occurs in closed syllables when
words are considered in isolation, it sometimes oecurs in an open syllable
as a result of resyllabification across a word boundary. In the latter part
of this subsection we make use of this phenomenon to highlight the dif-
ference between MA, which has an unstable vowel, and Tashlhiyt, which
does not.
We give below the text of a nusery rhyme which is well-known in the
Marrakesh-Taroudant area and in Oujda. We first give the text and its
translation.
(13) 1. lbid-a !bid-a Iillah
2. b=as n-zewweq luh-tei
3. u=lun-t=i llend ltaleb"
4. !we26=t=taleb fe=z=zenn-a 27
5. we=z=zenn-a me-hlul-a
6. helleleha mulaena
7. rnulaena mulaena
8. la lte-qte? lrZa=na
9. fi sabil m'Tiammed"
10. muhammed lweeshabeu
11. fe=z=zenn-a lye-nsab-u

24 In our special use of 'uncontroversial' and 'putative' , these adjectives are predicated of
vowels , i.e. of phonological objects, but for the sake of convenience we will allow our-
selves to predicate them also of letters in transcriptions , e.g. we shall say that, of the two
occurrences of 'e' in lbesleek ' your onion' ((9)g), the first is putative and the second is
uncontroversial.
25 lta leb, caretaker of a mosque; often acts as a schoolteacher.
26 u 'and' has a free variant w before a consonant cluster.
27 The initial geminates in !t=taleb and z=zenna result from the assimilation of the definite
article 11=1 to the following coronal: /l=!taleb/, II=zenna/. Henceforth in all surface sequences
of the form #K=K or =K=K in which the two 'K's represent the same coronal , the first
' K' represents the assimilated variant of the definite article. ' K=K' represents a heteromor-
phemic geminate, i.e. two heteromorphemic skeletal positions linked with the same bundle
of distinctive features.
28 Variant of muliammed, see following line.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 243

(14) 1. Egg, egg, for God's sake

2. With what could I decorate my writing tab let?
3. And my writing tablet is with a teacher
4. And the teacher is in Paradise
5. And Paradise is open
6. Our Lord made it accessible
7. Our Lord, our Lord
8. Do not shatter our hope
9. In Muhammad's way
10. Muhammad and his companions
11. In Paradise they are dwelling
When the nusery rhyme is performed, it is chanted in such a way as to make
the syllables coincide with beats in a rhythmic pattern. AllIines are chanted
to the same pattern, which is represented in (15), with the text of the first
line aligned undemeath."
(15) * * * *
* * * * * * * *
bi da bi da lil lah"
In (15) the stars in the bottom line represent points in time separated by
equal durations; items which are vertically aligned are simultaneous.
Columns with two stars represent strong beats. Every line in the nusery
rhyme is six syllable long. The first five syllables are evenly spaced in time,
and so are the first, third, fifth and sixth syllables, which occur on strong
beats." The last syllable of one line and the first syllable in the following
line are aligned with successive strang beats.
The alignment of each line of the text in (13) with the rhythmic pattern
in (15) is given below.
(16) * * * *
* * * * * * * *
1. bi da bi da lil lah
2. ba s#n zew weq luh ti
3. u luh ti llend ta leb
4. wet ta leb fez zen na

29 On the use of 'metrical grids ' such as (15) to represent musical rhythm, v. Lerdahl and
Jackendoff (1983).
30 Here and below we omit from the syllabic parses thc exclamation points representing
emphasis .
31 The rhythmic pattern of the chant is the same as that of the first line of the well-known
French song ' Au clair de la lune' (v. § 4.3). Here is that line, with the strong beats indi-
cated by capitals : AU clair DE la LU-NE.
244 CHAPTER EIGHT

5. wez zen na meh lu la

6. hel lei ha mu la na
7. mu la na mu la na
8. la teq te c;#r za na
9. fi sa bi I#m w ham med
10. mu ham med wes ha bu
11. fez zen na yen sa bu
For the sake of conspicuity we have highlighted every putative vowel
together with the consonant which immediately precedes it. The informa-
tion displayed in (16) is a mixture of observed facts and of inferences
based on the examination of a larger corpus of MA verse. The empirical
observations concern the rhythmic alignment of onset consonants and that
of uncontroversial vowels, while the inferences concern the coda conso-
nants. Let us first dweIl on the empirical observations embodied in (16).
What the rhythmic delivery of the nursery rhyme allows us to observe
is that every line is divided up into six successive chunks and that every
chunk contains an uncontroversial vowel together with the consonant which
immediately precedes it, or it contains a consonant which immediately
precedes a putative vowel. (16) contains two kinds of putative vowels, some
represented by a highlighted 'e', others by a '#' sign . Let us look at these
in turn .
For every occurrence of a highlighted 'e' in the transcription in (16) there
is a corresponding syllable nucleus in the phonological string. If, as proposed
at the end of the preceding section, we assurne that all syllable nuclei in
MA are vowels, (16) is one piece of evidence suggesting that like the W-
final occurrences of 'e', the W-internal occurrences all correspond to vowels
in the phonological string.
Let us now turn to the nuclei represented by '#'. When a word ending
in a consonant is immediately followed by another beginning with a con-
sonant cluster, a syllable straddling the two words is formed. In line 2,
for instance, b=as#n-zewweq is parsed as ba.s#n.zew.weq, with the second
syllable straddling the two words . We call 'hinge syllabIes' the syllabIes
which straddle two words and we say that a syllable is 'hollow' when it
does not contain a full vowel. s#n is a hollow hinge syllabie. Let us assurne
that like those of Tashlhiyt, the syllables of MA do not allow complex
onsets." Forming hollow hinge syllables such as s#n is a way to avoid
complex onsets while leaving no segment unsyllabified. Under our assump-

32 Cantineau (1960: 118-119) suggests that there are no comp1ex onsets inside words in
the modern diaIects of Arabic. In his detailed and insightful discussion of the phonology of
a Bedouin dialect of Eastern Algeria, KouIoughli (1978: 104ff., 256ff.) gives several argu-
ments leading to the same conclusion . Some of his evidence is drawn from versification in
songs.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 245

tion that all syllable nuclei in MA are vowels, every # in (16) actually stands
for a vowel. The phonetic manifestations of # nuclei are at present as murky
as those of other putative vowels. It would be consistent to use the letter
'e ' to represent the nuclei of hinge syllables in our parses of MA poetry,
but for the sake of conspicuity we will continue to note these nuclei as #.
Hinge syllables occur as well in nonpoetic speech styles, witness the
following pairs of homophonous expressions:
(17) a. qlib argan 'overturning of the argan trees"
a'. qli b=argan 'fry with argan oil!'
b. mat lhmar 'a donkey is dead '
b' . ma lte-h mar 'you will not blush"
c. zab dZaz-a 'he brought a hen'
c'. za beedzaz-a 'he came with a hen'
d. zab dzaz-a 'he brought a hen' (v. item c above)
d' . zabed Zaz-a 'pulling a window pane'
e. !serm !1=gansa 'a rip in the gizzard'V
e' . !sermel !gansa 'he put dressing on a gizzard'
In each pair we are comparing two pronunciations of a certain sequence
of segments, e.g. Iba/ in pair (17)a-a'. The sequence in question straddles
a word boundary in the first member of each pair but not in the second.
In the first pair the nucleus of the hinge syllable is a full vowel. The hinge
syllables involved in the other pairs are hollow ones. Pairs b-b' and c-c'
illustrate the fact that C#CCV is homophonous with #CeCCV. Pairs d-d'
and e-e' illustrate the fact that in certain contexts, at a normal speech rate,
C#CCV is homophonous with CeC#CV. 36
These remarks on hinge syllables conclude what we have to say about
those aspects of table (16) which are observed facts . We now turn to the
information in (16) which is not derived from direct observation. This
information concerns the consonants which do not immediately precede a
vowel , e.g. h in lului in line 3. We have not tried to observe how chanting
apportions these consonants between successive chunks. Our decision to
group them with the preceding vowel rather than with the following con-
sonant is in accordance with our assumption that complex onsets are not
allowed.
We have stated at the outset that e does not occur in open syllables, It
is not surprising, then , that in (16) every e but one is followed by a

33 qlib is a noun derived from qleb 'overtum'.

34 As, e.g., in ma !te-limar ma te-sfar 'you will neither blush nor turn pale' .
35 !serm is a noun derived from the verb lsrem 'tear, rip' .
36 In very deliberate speech, or if the speaker intends to put special emphasis on the first
word, the homophony disappears in example d', see next chapter (§ 9.2).
246 CHAPTER EIGHT

tautosyllabic consonant. The one occurrence of e which is in an open

syllable is the W-final e in !teqte f, in the third syllable in line 8.
The sequence tef#ria gives rise to a hinge syllable ; W-final f becomes
the onset of a syllable which has word-initial r as its coda. Let us assume
that in MA, syllabification operates in successive stages defined by more
inclusive domains, two such domains being the word and the line. /1)./ is
syllabified as a coda at the word level (.teq.te f.) and it is later resyllabi-
fied as an onset when the successive words in a line form a single
syllabification domain: teq.te. f#r.ia.na. This example is but one instance
of a general fact of MA poetry: at the line level, W-final schwas can occur
in an open syllable as a result of resyllabification, and in this they behave
just like the full vowels, see e.g. a in ba. Mn, in line 2. In the present case
(tef#ria) the hinge syllable resulting from resyllabification happens to be
a hollow syllable, but the consonant in a word-final eC sequence can become
an onset to a full vowel as weIl, e.g. when !ria=na is replaced with amal>i
'my hope' in line 8, the resulting line ends in te. Ya.ma.ll, which is still
perfectly suited to the rhythmic pattern (15).
When a word-final consonant is parsed as an onset, we assume that this
is a result of resyllabification rather than ambisyllabicity. To see what we
mean by this, let us use as an example the sequence which results when
in line 2 of (13) the ls imperfective verb n-zewweq is replaced by its 3ms
counterpart i-zewweq. The resulting sequence, which is a well-formed line,
is shown in (18)a below and its scansion is shown in (18)b:
(18) a. b=as i-zewweq lun-t=i
b. ba.si.zew.weq .Iuh.ti
Consider s in ba. si in (18)b. The skeletal position of s is first made a
coda to the preceding vowel by word-level syllabification; line-level syl-
labification subsequently deletes the grouping of s with the preceding a
and turns sinto an onset to the following i. Alternatively one could assume
that s is ambisyllabic, i.e. that there exists a level of representation at
which s is both a coda in one syllable and an onset in the following syllable.
The skeletal position of s would be contained in two overlapping sylla-
bles, as represented in (19):

o o
.r-:R
(19)

0 0
»<.R

0~
I
N
I
X X X # X
I I I I
b a s
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 247

This view is inconsistent with the framework adopted until now, in wich
syllables do not overlap, and it would lead to an incorrect prediction about
syllable weight in the syllabification of CVC#C sequences in versifica-
tion. We shall see later that in poetry, when V is a full vowel, CV syllables
count as light whereas CVC syllables count as heavy . If a word-final C
retained its status as a coda when it is resyllabified as an onset, as dia-
grammed in (19), word-final CVC strings in which V is a full vowel would
always count as heavy syllables no matter what folIows. In fact, in the parse
of /CVC#V/ as CV.CV in verse, the first syllable always counts as light."
Let us revert to sequences like te. f#r in line 8 of (16). e is indeed in a
cIosed syllable when we consider syllabification at the word level (.ter),
but we know of no evidence which would suggest that in such sequences
e is still in a cIosed syllable when we consider syllabification at the line-
level. Indeed, such syllables always count as light in poetic scansion."
Parse te. f#r and the like are dramatic illustrations of the difference
between MA, which has an unstable vowel, and Imdlawn Tashlhiyt, which
does not have any. The difference can be brought horne in an even more
striking manner by devising similar sequences of segments in MA and in
Tashlhiyt and looking at how they align to the same rhythmic pattern.
Tashlhiyt happens to have a 'trick or treat' nursery rhyme which is chanted
to the same rhythmic pattern as the MA nursery rhyme in (13). Let us
first give the text of the nursery rhyme and its translation.
(20) a. a-k''ndrris a-swih
b. yan=ax=t=id ur i-fki-n
c. ar !i-tt-zzig t-a-ydi-t
d. ar i-ss-ndu )'=u-hlas
(21) a. Meatballs." rag
b. He who does not give us any
c. Let hirn milk a bitch
d. And churn the milk in a packsaddle?"
Here is how the nursery rhyme must be chanted:
(22) * * * *
* * * * * * * *
a. a kWn dr ri sas wih
b. ya nax ti du rif kin
c. a rit tz zig tay dit
d. a ris sn du )'un las

37 See for instance the second syllable in lines 2a and 3a in (30) below.
38 See for instance the second syllable in line 11b in (31) below.
39 More precisely, balls made with dried tripe.
40 With the hollow facing up.
248 CHAPTER EIGHT

We want to compare line 8 of the MA nursery rhyme in (13) with a similar

sequence in Imdlawn Tashlhiyt. The crucial word in that line is !teqte r,
where the final consonant happens to be r, a consonant with a very limited
distribution in Imdlawn Tashlhiyt. To compare strings which are as similar
as possible, let us replace qte r in line 8 by another verb, e.g. by nsef 'under-
mine' . The resulting sequence is also a well-formed line:
(23) la te-nsef !ria=na41
We will compare the scansion of (23) in MA with that of the following
Tashlhiyt sequence:
(24) ra y-nsf rbu-neteid"
Let us align the syllabic parse of (23) in MA with that of (24) in Tashlhiyt:
(25) a. (23) MA la ten se f#r Za na
b. (24) Tash ra yns fr bun tid
c. *(24) ra yn s- fr bun tid
As can be seen in (25)b, the Tashlhiyt sequence in (24) only has five
syllables and cannot be chanted to the rhythm of (15) and (22). (25)c
aligns the segments in (24) with those in the MA line in (23) so as to
highlight the crucial difference between the two sequences. The under-
score in (25)c indicates the segment whose absence makes (24) shorter
by one syllable than (23): the last two consonants in the MA verb te-nsef
in (23) are separated by a vowel, whereas the last two consonants in the
Tashlhiyt verb y-nsf in (24) are not. In (23) a voiced vocoid must be pro-
nounced between sand f in te-nsef even if the sentence is uttered in a
style of diction appropriate in everyday conversation.
Finally, consider the following Tashlhiyt sequence:
(26) ra y=nsr fl-n=as=t=id 43
(26) has one more syllable than the Tashlhiyt sequence (24) because the
presence of r between sand f in (26) yields one more sonority peak. This
is made clear in (27), where (26) is aligned with (23) and (24).
(27) a. (23) MA la ten se f#r za na
b. (24) Tash ra yns fr bun tid
c. (26) Tash ra yn sr fl nas tid

41 neg 3fs:irnpf-undermine hope=lp.

42 fut 3rns-graze carry .aor-Smpedoßmsedir, 'he will graze (his skin) and they will carry
hirn on their backs' . In the underlying representation the line begins with /rad i-nSfl. The
altemations involving the final consonant of the future marker /radi are described in DE
(1989 : 188-190).
43 fut 3rns-graze leave .aor-Smpedatßsedoßrnsedir, 'he will graze (his skin) and they will
leave it (rn) with hirn' . nsr is synonyrnous with nij in (24) .
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 249

Let us sum up our discussion of the two nursery rhymes. Our compar-
ison illustrates the following difference between the two languages: In
MA, but not in Tashlhiyt, a hollow syllable occurring inside a word sequence
may contain a single consonant, see e.g . .s@ . in (27)a . Such syllables are
made possible in MA by the fact that MA has a fourth vowel in addition
to the full vowels /a, i, u/ . In Tashlhiyt, on the other hand, there is no
vowel schwa and consequently hollow syllables which are not postpausal
must contain at least two consonants.

8.3.2 . Inventory of syllable types

In this subsection we draw up an inventory of the syllable types occur-

ring in orthometric syllabification. This inventory coincides with that which
can independently be inferred from the distribution of vowels and conso-
nants in standard transcriptions.
Like Tashlhiyt, MA distinguishes between light and heavy syllables.
We now turn to a song which makes systematic use of this distinction, unlike
the nursery rhyme in (16) . The song belongs to a genre of singing called
melliun, which is widely known throughout Morocco and in Algeria." The
melhun tunes have their own distinctive musical style." The language of
the lyrics comprises many lexical items used only in singing." Apart from
occasional occurrences of the glottal stop, its phonology does not differ from
that of everyday MA. The meters used in melhun are different from those
of poetry composed in Classical Arabic (henceforth CA). Even today many
composers and performers - the same person can be both - are illiterate,
not to speak of the experienced listeners. Some practitioners of melhun
are professionals, others are amateurs from all walks of life. The length
of a lqsid-a, a melhun song , ranges between twenty lines and several
hundred. The transmi ssion of melhun songs is mostly oral, but occasion-
ally some are recorded using the Arabic script. The melhun is only one
among several MA poetic genres in widespread use in Morocco today.
The melhun is a pan-MA genre . As far as we know, speakers from dif-
ferent parts of Morocco only need to make minor alterations in the lyrics
when they sing a given !qsida in their own dialect of MA. This fact suggests
that syllable structure changes little when one moves from one dialect of
MA to another.
Tahar (1975) was the first to state the basic principles which govern

44 For general presentations of melhun, see AI-Jirari (1970), Tahar (1975), Pellat (1987),
Al-Malhun i (l990b) and Jouad (1995). Al-Fassi (1997) contains a representative sampie
of songs.
45 On the music of melhun, see Aydoun (1994) for a quick overview and for references
to more detailed work.
46 See Oe Premare and Alaoui (1989).
250 CHAPTER EIGHT

versification in the melhun. The following discussion is based on OUf

examination of the songs collected in pp. 253-364 in Al-Malhuni (1990a)
and in AI-Fassi (1997).47 Some melhun pieces are organized in stanzas
with several different meters combined in a stanza . Other pieces have all
their lines built on the same meter. Most of our examples are drawn from
pieces with two-line stanzas.
We give below the text of the first twelve couplets in 'the !qsida of
Fatma', a song by Muhammed Lehmer Lermyaq (AI-Fassi 1997: 15-18).
An English translation folIows.
(28) 1a. mir !le=-ywrem xil=u !fe=l=nerb !zatm-a
1b. dakk l=hedd-a fuq selw-i me-lzum
2a. !we=l=-ywram i-seyyeb we=d=dat saqm-a
2b. nett l=-yiwan ma !y-weqqer me-yrum
3a. hakdaek !zra l=i !qess-a lweterzm-a
3b. weanaya ma dri-t !fe=l=nerb qyum
4a. ha sbab hwa=ya !xWennar xatm-a
4b. vab-et 'lenn=i w='lad qelb=i me-hmum
5a. fi knan=i !-yess-et !-ywess-at samm-a
Sb. kif n=nisan dar xebl-a fe=l=qum
6a. vab-et 'lli=ya weelemrehz-at haym-a
6b. qWel-t a 'lezb=i !d=dedd feebn-at l=yum
7a. !tal-et l=-yib-a ya le=-ywzal !fatm-a
7b. 'lib !I:::heir-an ya sbiv-et Ieenyum
8a. kan-et mlafei fe=I='lwessaq hakm-a
8b. we=l=yum sqa-t=ni mhayen weehmum
9a. hWebb=ha xella l=i dat=i m-faqm-a
9b. we=l=hwebb !s'lib qal-et dha-t l=qum
lOa. sal qis I=me-znun 'lla l=hakm-a
lOb. lil-a hiya sbab qelb=u me-Ydum
lla. sal sif l=yazan 'lla z=za'i'm-a
llb. msex ula -ywrab lwadeh me-hkum
12a. sewwel 'i'la !'lenter we=l=qum zairn-a
12b. le=-ywram i-luh le='i'Siq l=le=hmum

47 Jouad (1995 : 304-314) presents ninety odd lines of melhun with their syllabic parses
and French translations. DE (1988 : 10) is an earlier instance of syllabic scansion of MA verse.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 251

(29) la. Love, the great prince, launches his cava1ry into battle
lb. Pounding my side under his impressive bearing
2a. Love has made my hair white and my body weak
2b. Amorous yearning does not spare any lover
3a. Here is what happened to me, here is the story, here is
the chronicle
3b. I, who am inept in the arts of warfare
4a. The object of my passion is a hard-headed beauty
4b . She has left me, and my heart is troubled
Sa. At the bottom of my heart, deep and mortal regrets
Sb. Like arrows which lay waste to the fighters' ranks
6a. She 1eft me and my soul is distraught
6b. How obstinate, I realize, girls are nowadays
7a. Absence has lasted too long, 0 Fatma, my gazelle
7b. Your being away is no good, 0 you with khol-painted eyes
8a. She was my tame one, she used to bewitch lovers
8b. Now she showers me with adversity and worries
9a. My love for her has unsettled my body
9b. Love is hard, as people with experience say
10a. Ask Qays the madman about the woman who bewitched
him"
lOb. Layla, who bruised his heart
lla. Ask AI-Yazan about his daring one
llb. Better a transfixing curse than a pure and bewitching love
12a. Ask Antar (on how he feels) when armies rush forward
12b. Love plunges the lover into worry
The text in (28) is parsed below. The two lines in each couplet have
different meters. The couplets' first lines are parsed in (30); the second lines
are parsed in (31). Hand L respectively stand for 'heavy' and 'light'.

48 Qays, Al- Yazan (li ne lla) and Antar (line l2a) are well-known characters from the
literature .
252 CH APTER EIGHT

(30) 1 2 3 4 5 6 7 8 9
H L L L L L H H L
la. mir le)' w rem XI lu fel
herb zat ma
2a. welv " ra mi sey- yeb wed- dat saq ma
3a. hak da k#z ra li qe s- saw terz ma
4a. has ba b#h wa ya xWen- nar xat ma
)' w
5a. fik na ni )'es- set es_ sat sam- ma
6a . )'ab t#'l li ya wel m reh z.at hay ma
7a . tal t#l )'1 ba ya le)'w zal fat ma
8a. kan t#m la fi fel 'lwes sa q hak ma
9a. h Webb ha xel- la li da tim faq ma
lOa. sal qi s#l mez nu n#'l laI hak ma
lla. sal si f#l ya za n#'l laz- za'l ma
12a. seww 1#) la 'len ter wel qum za'l ma

(31) 1 2 3 4 5 6 7 8 9
L L L H H L L L H
lb. dak- k#l hed- *da fuq seI wi mel zum
2b. het- t#1 )'i wan may weq- qer mev rum
3b . wa na ya mad rit fel her b#q yum
4b. )'a bet 'len- niw 'lad qel bi meh mum
5b. ki f#n- ni sa n dar xeb la fel qum
6b. qWel ta 'lez bid- dedd feb na t#1 yum
7b. 'li b#l hel. ran yas bi )'et len yum
Sb. wel yu m# s qat nim ha yen weh mum
9b. wel h Web- b#s 'lib qal t#d ha t#1 qum
lOb. li la hi yas bab qel bu me? dum
llb. #m se xu la)' w rab wa deh meh kum
12b. le)'w ra mi luh le'l si q#l- leh mum

There are no complex onsets. Onsetless syllables are allowed onl y at

the beginning of a line (see line llb for an example); in particular, hiatu s
is prohibited. The syllable type s which occur in melhun are listed below
in the fir st column in table (32) , where we follow the same conventions
as in table (28) in § 4.6.49

49 N and 0 respec tively stand for Nucleus and Coda. Z- indicates that Z is linked with
the first skeletal position in a geminate.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 253

(32) nme N = full V N =e

a. N xi (la:4) se (llb:2)
b. N D- dak- (lb:l) het- (2b:l)
c. ND mir (la:l) yeb (2a:5)
d. N D-D hubb dedd (6b:5)
e. ND D *h arb herb (la:7)

Each syllable type is exemplified with two instantiations drawn from the
couplets cited above." The parenthesized numbers which follow the
examples indicate the examples' coordinates in (30) and (31); a colon sep-
arates the line number from the position of the syllable in the line. The table
does not include onsetless syllabIes, which will be dealt with separately.
Since all the syllables within the purview of the table have an onset, the
relevant differences between them all have to do with their rimes.
In table (32) codas comprise at most two skeletal positions; after a full
vowel the only complex codas allowed are geminates, as indicated by the
starred form in the bottom line of (32). The evidence summarized in (32)
pertains to syllabification in poetry. Not surprisingly, an examination of
the phonotactics of MA words leads to the same inventory of syllable types.
We have made a preliminary survey of the canonical forms of MA words
noted in the standard transcriptions. It turns out that once word-initial syl-
lables are set aside, all and only the syllable types listed in (32) are necessary
to parse the isolation forms of MA words. This observation is illustrated
below in (33) . The ten lines in (33) correspond to the ten syllable types
in (32). The syllables which are instantiations of the various syllable types
are highlighted. Each word is followed by information about its morpho-
logical make up. The syllabification of initial clusters in the examples in
(33) will be justified later.
(33) a. sa.fu 'they saw'; saf-u
a' . s.ke.ku 'his plowshares'; skek=u; skek 'plowshares"
b. tt.haz.zu 'they lifted one another'; u-hazz-u
b'. sek.ku 'they doubted' ; sekk-u
c. kas.fu 'they guessed'; kasf-u; kasef 'he guessed'
c' . der.bu 'they struck'; !derb -u; !dreb 'he struck'

50 The syllable type intantiated as hubb in the table happens not to occur in the song cited
above, but is attested in others, e.g. hubb occurs in the first line of the first stanza in a
ballad recorded on tape by Toulali which will be cited in § 8.5.6. See exact reference there.
5\ The singular form is sekk-a . On the special status of e in skek and other similar plural
nouns, see § 9.5.
254 CHAPTER EIGHT

d. m.qidd.sa 'rnischieveous girl'; m-qidds-a"

d'. t.ness.ri 'spread impf 2fs' ; t-nessr-i 53
e terz.ma 'Iife story ' ; terzm-a
e'. *ta rZ.ma (ill-formed)
Whenever possible we have used syntactic words rather than Pwords in
the examples in (33). The only syllable type which does not occur in isolated
syntactic words is that with a hollow open rime, see (33)a', the only case
in which we were forced to use an example containing a clitic. Apart from
case (33)a', the set of syllable types necessary to parse Pwords is the same
as that necessary to parse syntactic words. Since complex onsets are dis-
allowed, all word-internal clusters comprised of three skeletal positions
begin with a complex coda.
Anticipating the conclusions of our discussion of line-initial syllables
(see below), we are faced with the fact that two different routes lead us
to the same set of syllable types. One line of investigation is concerned with
the orthometric syllables in songs, while the other deals with the phono-
tactics of words. We take this convergence as evidence that there is no
difference between the syllable structure of MA in its everyday usage and
that in singing. From now on we consider the two structures as one and
the same thing and we assurne that as far as syllabification is concerned,
lines of verse are treated as instances of a higher-Ievel phonological con-
stituent which we call the Phonological Utterance. We define a Phonological
Utterance as a minimal stretch bounded by pauses. A Phonological Utterance
can contain several sentences in a row, or it can be comprised of no more
than a single word, as when a word is pronounced in isolation .
Let us now turn to line-initial orthometric syllabies. Since line-initial syl-
lables may lack an onset, the syllable types one would expect to find at
the beginning of lines are all those in table (32) plus their onsetIess coun -
terparts. In fact the hollow rimes in the last column in (32) only exist in
syllables with an onset, i.e. schwa is never line-initial. On the other hand,
at the beginning of lines, and only there, there occur syllables which are
made of a single consonant, e.g. the initial syllable of line llb in (31)
consists of a mere m. Let us suppose that such syllables are actually onset-
less syllables in which the nucleus is e, and that syllable-initial occurrences
of e are never realized as vocoids. This analysis enables us to account for
the peculiarities of line-initial syllables while preserving our assumption
that in MA all syllables contain a vowel. It allows us to reconcile the
existence of word-initial consonant clusters with our assumption that MA
disallows complex onsets : word-initial clusters are not syllable-initial
clusters ; their first part is actually the coda of a hollow onsetless syllable,

52 The masculine form is mqiddes.

53 nesser is the intensive stern derived from nser 'spread' .
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 255

e.g. the surface syllable structure of skeku 'his plowshares' (v. (33)a') is
/.es.ke.ku./. 54 In MA a short voiced vocoid is never heard at the begin-
ning of a word pronounced in isolation.
Viewed in terms of syllable structure, the observations above can be sum-
marized in the statement below, which we number for later reference.
(34) Only after an onset can e be realized as a vocoid.
Clearly, (34) is not a principle of grammar, but a fact which must be
accounted for. We will propose an explanation for it in Chapter 9. Till
then, we will treat (34) as a primitive in our analysis. In a word such as
(33)a', which is pronounced [sk@ku] but is syllabified as .es.ke.ku., we
will not attempt to explain exactly how after a pause the syllabicity of e
is transferred to the following s, for the question will not arise anymore
in our final analysis.
Judging from the songs we have examined, lines beginning with an onset-
less syllable seem to be far less common in MA than in Tashlhiyt. A detailed
inventory of the various possibilities allowed in line-initial syllables would
require a larger corpus. Let us simply give two more examples. They are
the first lines in couplets #6 and #22 in 'the !qsida of Muni', a song recorded
in Al-Fassi (1997 : 5-6), see below (36)c,d. Before these lines we give the
first line of the song as an instance of a line beginning with an onset «36)a),
and the first line of couplet #26 to exemplify the case of vowel-initial
lines «36)b). We first give the text of these lines, then their syllabic
parses:"
(35) a. qelbei be=l=nwebb !sar me-fni
b. u=!n-u sef qeddeha sesani
c. !llah hsib men vwaeha
d. sqaeni kas men !mrar-a
(36) L L L H H L L
a. qel bi bel n Webb sar mef ni
b. u nu sef qedd has- sa ni
c. 1- la h#n sib menv wa ha
d. s qa ni kas menm ra ra
Given a sequence of morphemes or words in MA, what is the proce-

54 For an analysis of word-initial consonant clusters in Ath Sidhar Rifian along the same
lines, see DT (1992). Hyman (1985: 68) already proposes that in Ayt Ndhir Tamazight Berber,
in which the distribution of schwa is similar to that discussed here, words with initial con-
sonant clusters begin with syllabic consonants .
55 Here are the translations of the four lines: (a) my heart is wrecked by love; (b) and I
describe her radiant silhouette ; (c) God will judge those who diverted her; (d) he has handed
me a cup of bitterness.
256 CHAPTER E IGHT

dure which parses it into a sequence of syllables? We cannot provide a

complete answer to this question, but we can at least answer it inasmuch
as is relevant to our main aim in this discussion, which is to highlight the
main differences between the surface syllable structure of MA and that of
Tashlhiyt.
One can view syllabification in MA poetry as operating in succe ssive
stages, the last two of which are word-level syllabification and line-level
syllabification. After syllabification has operated at the level of words,
i.e. once each word in a line has been syllabified as aseparate unit, syl-
labification operates again , taking the whole line as its domain . Constraints
on well-formed syllables are the same at the two levels, and what line -
level syllabification does is to reshuffle syllable structure at junctures
between words. What motivates all the reshufflings is the need to avoid con-
tiguous rimes . Here are the most common ways this result is achieved . First,
certain word-final codas are turned into onsets. In line 2a of (28), for
instance, the final consonant of le=yWram, a coda in its word, becomes
an onset to the initial vowel of i-seyyeb, and in line Ib the second skeletal
position of kk becomes an onset in a hollow hinge syllable k#l whose coda
is the initial consonant in l=hedda . Second, certain vowel sequences are
broken up by the insertion of a glide, as when Isra#atayl 'he bought tea'
is pronounced srayatay. Third , a nucleus loses its syllabicity next to another.
If that nucleus is a high vowel it becomes a glide, as in the realization of
itim 'orphan' in 19a ytim 'he found an orphan'. If the nucleus in question
is e, it disappears altogether, as happens at the beginning of line 4a in
(28) : in the representations which are the output of word-level syllabifi-
cation the line begins as I.ha.es.ba . . .!, where the second syllable lacks
an onset. Line-level syllabification changes I.ha.es.! into I.has.!.
Our reason for viewing line-syllabification as operating on already syl-
labified words is that schwas in a given word occur much in the same places
regardless of whether the word is uttered in isolation or strung with other
words in a line. Let us call this 'the stability of word-level nuclei':
(37) Stability of word-level nuclei:
Line-level syllabification preserves the syllable nuclei resulting
from word-level syllabification to the extent that their preser-
vation is compatible with the avoidance of contiguous rimes (i.e.
of line-internal onsetless syllables).
We view (37) as a fact which a grammar of MA must account for and not
as one component in the machinery comprised in the grammar of MA. Going
back to the three cases of resyllabification reviewed in the text above (37),
we see that in the first two cases, line-level syllable structure preserves
all the word-level nuclei. As for I.ha.es.! in the third case, apart from
inserting a consonant between the two nuclei , an option which is not avail-
able in MA when one of them is e, the deletion of e is the most economical
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 257

way of getting rid of the adjacent rimes." The view that e is deleted after
a vowel is only a provisional one adopted here for the sake of explicit-
ness. In our final analysis we will suggest a more parsimonious account
in which there is no difference between the phenomenon which we inter-
pret here as schwa deletion, and turning a high vowel into the corresponding
glide, see the next chapter.

8.3.3. Complex nuclei; evidence from syllable weight

Like that of Berber, the versification of MA makes a distinction between
heavy (H) and light (L) syllables, In a nutshell, syllables ending with a vowel
and hollow syllables ending with a single C are light, while all the others
are heavy. According to Pell at (1987) the distinction was first proposed
by Tahar (1975), whose book deals with Algerian melhun."
The table below indicates the weight associated with each of the syllable
types listed in table (32). As in table (32), the strings in the middle columns
are instantiations of the syllable types listed in the first column.

(38) I II
rime N = full V N =e
a. N Xl se L

b. N 0- dak- het-
L or H
c. ND mir yeb
I
d. N 0-0 hubb dedd
H
e. ND 0 *harb herb

As in Tashlhiyt, certain syllable types are alternately H or L depending

on what the meter requires. The rimes which are ambiguous in this manner
are (i) those which end in the first half of a geminate (line b) and (ii) the
hollow rimes which only have one skeletal position after e (box c-II). We
have already encountered weight ambiguity of type (i) in Tashlhiyt versi-
fication, see § 4.6, where rimes of type (i) were dubbed rimes with hinged
codas . As an example of the two-valuedness of rimes with hinged codas,
compare dak-; the first syllable of line lb in (31), which counts as L, and
sam-, the eighth syllable in line 5a in (30), which counts as H. Hollow rimes
with only one skeletal position after e count as light in the overwhelming
majority of cases, but the lines in which such syllables occupy H posi-

56 e is also deleted elsewhere than after a vowel, e.g. in the first word in Iines 6a, 7a and
8a. These cases will receive a different interpretation in § 9. I.
57 In support of his distinction, Tahar (pp. 73-74) cites Cantineau (1960: 119), who states
that in Maghrebian dialects H syllables are 'a littIe longer' than other syllabIes.
258 CHAPTER EIGHT

tions are not judged awkward by the experienced listener. The only exarnple
of such a line in the twelve couplets in (28) is line 12b in (31), with zer
counting as a H syllable." Let us use the expression 'eC rirne' to refer to
a rirne which has only one skeletal position after e.
In what foIIows we will have nothing to say about weight arnbivalence
in rirnes with hinged codas . As for that in eC rirnes, it is ternpting to see
a connection between their rather infrequent occurrence in H positions
and the rather infrequent occurrence of e in open syllables at the word level.
Syllables like .se., whose rirnes do not contain consonants, and H syIIa-
bles with eC rirnes, are special in that they are the only ones in which schwa
is equivalent to a fuII vowel, as far as syIIable weight is concerned. In
the other cases a fuII vowel contributes as rnuch to syllable weight as e
together with the following C. Let us first discuss the weight equivalence
between fuII vowels and tautosyIIabic eC sequences.
Let us tabulate again the rirne types in (38) , leaving aside the rirnes
with hinged codas (line b) and those ending with tautosyIIabic gerninates
(line d). In the table below the sequences used to instantiate the rirne types
have been selected for expository convenience. Onsets, which are not
relevant, have been left out. To rnake cornparison easier, the lowercase letters
at the beginning of line s match with those in table (38) . Line c of table
(38) has been split to separate the cases in which eC counts as L frorn
those in which eC counts as H.

(39) I II
nme N = fuII V N =e
a. N a (e)
L
c. ND *arn ern
Cf.ND arn (ern)
H
e. ND D *arns erns
Three kinds of rirnes appear in table (39), sorne rnarked with an asterisk,
others with parentheses and others yet without any rnarking. An asterisk
indicates ill-formedness, absolute as in line e (a fuII vowel cannot be
foIIowed by two consonants), or in cornbination with the weight indicated
to the right of the table, e.g. am, which is well-forrned, cannot count as L
(line c). Parentheses indicate rirne types which we caII 'secondary': rirnes
lacking any fuII segment and eC rirnes counting as H. The rernaining four
rirne types are those we caII prirnary. Let us first focus on these .
In prirnary rirne types , syllable weight seerns to depend only on the fuII

58 Contrast that syllable with the eighth syllable (md) in line lOb in (31), which counts
as L.
 SYLLABLE STRUCTURE IN MOROCCA N ARABIC 259

segments, as can be seen by inspecting the table below, in which we have

re-tabulated the primary rimes of (39) .

(40) I 11
N = full V N =e
a, c a em L

c', e am ems H

In (40) the rimes with one full segment are Land those with more are H.
Furthermore, a full vowel can onl y be followed by one tautosyllabic
segment, whereas e can be followed by two. This suggests that full vowels
and eC sequences play analogous roles in syllable stru cture . Until now
we have assumed implicitly that in a rime all con sonants belong to the coda.
Let us abandon this assumption and suppose that in the hollow rimes in (40)
e shares its nucleus node with the segment which immediately follow s it.
We propose the structures below for the rime s of table (40):

(41) a am em ems
R R R R

N
I ~
N D
I
N N
r----------D
X
I I
X X
I »<.X
X X
»<.X X
I
I I I I I I I I
a a m e m e m s

In the analysis wh ich we are proposing here, only e can be associated

with the first position of a complex nucleus, and only a con sonant can be
associated with the second position. Under that analysis, syllable weight
does not depend on the number of skeletal positions in the rime, but on
the existen ce of a coda. Rimes am and em both contain two skeletal posi-
tion s; am, which contains a cod a, is H, while em, which does not, is L.
Let us now go back to the secondary rimes, i.e. those parenthesized in
table (39) . Their structure is that represented below.

(42) e em

r-.
R R
I
N N D
I I I
X X X
I I I
e e m
260 CHAPTER EIGHT

The distinguishing property of the secondary rimes is that e does not share
its nucleus node with another segment. Compare the rime em in (41) with
the rime em in (42). The former is L (it is coda-less) whereas the latter is
H (it has a coda). The relative rarity of heavy eC rimes in versification
suggests that in eC rimes the final consonant normally belongs to the nucleus
and that it is construed as a coda only to satisfy the needs of versifica-
tion.
The introduction of complex nuclei forces us to rearrange the contents
of table (32), in which the syllable types of MA were all listed and arranged
according to the structure of their rimes . Table (32) was devised under
the assumption that any consonant belonging to a rime is a coda . We give
below the old version of (32) and the rearranged one. Table (43) is (32) with
two modifications which do not change its import. First , since the differ-
ence between lines band c in (32) will not be relevant later, they have
been collapsed, but the lowercase letters at the beginning of lines in (43)
match with those in (32). Second, in (32) the strings used as instantia-
tions of the syllable types all actually occur in lines of melhun examined
by us. In (43) these have been replaced by invented ones chosen to facil-
itate the discussion of later versions of the table .

(43) rime N = full V N=e

a. N xi se
b, c ND mib wet
d. ND-D huzz hezz
e. NDD *h ast hezb
Table (44) below is the result of rearrangeing (43) in a manner consi s-
tent with our new conception of the internal structure of rimes; while the
nucleus is ein the last column in (43) it is eC in (44). Parentheses indicate
syllables with secondary rimes .
(44) rime N=V N = eC

a. N xi (se) wet
b, c ND mib (wet) hezb, hezz
d. N D-D huzz ?benss
e. *N D D *n ast *benst

The rime types in lines a-e are the same in (44) as in (43) , but some
strings have migrated from one box to another, e.g. liezb has a complex coda
in (43)e whereas it has a simple one in (44)b,c. The only complex codas
allowed under the new analysis are those in which the two skeletal posi-
tions in the coda form a geminate (see line d), and line e, which now lacks
any well-formed rimes , has been kept only to draw attention to this fact.
In line d the interrogation mark indicates a problematic case not men -
 SYLLABL E STR UCTURE IN MOROCCA N ARABIC 261

tioned before, that of hollow syllables in which the coda would be a

geminate. Certain facts about versification suggest that such syllables are
well-formed, but the evidence would take us too far afield and we willleave
the issue pending.
Finally, the displa y below bring s together the prim ary rimes of MA,
which are reproduced in the top row, with their analogues in Tashlhiyt
(bottom row).

(45) MA rimes (top) and their analogue s in Tashlhiyt (bottom)

a am em ems
R R R R
I
N N
r-,
D
I
N N
r---------D
I
X
I
X
I
X X
»<.X X
~
X
I
X
I I I I I I I I
a a m e m e m s

a am m ms
R R R R

N
I r-.D
N
I
N N
~
D
I I I I I I
X X X X X X
I I I I I I
a a m m m s

8.4. VIOLATIONS OF SONPEAK IN MA

In this section we propose an extended version of constraint SonPeak which

allows us to compare the role of sonority in the syllable structures of MA
and Tashlhiyt. To do this, we must devise means of abstracting away from
the fact that Tashlhiyt allows syllabic consonants while MA does not, and
that MA allows complex nucle i while Tashlhi yt doe s not. Extending
SonPeak is not merel y a matter of devis ing a yardstick with which to
compare the role played by sonority in the syllable structure of MA and
in that of Tashlh iyt. The extended vers ion of SonPeak will also play a
role in our account of word syllabification in MA when the language is
considered on its own terms in the following section s.
Let us state two differences between MA and Tashlhiyt:"

59 Similar differences were found between Tashlhiyt and Ath Sidhar Rifian, see § 6.5.
262 CHAPTER EIGHT

(46) Nature of nuclei :

a. MA : Every nucleus contains a vocoid:" there is an
unstable vowel (e).
b. Tashlhiyt: Any contoid can be a nucleus all by itself; there
are no vowels besides the full vowels .
(47) Location of nuclei other than full vowels:
a. MA: Determined by RIGHT-TO-LEFT SCAN (5);
sonority is irrelevant to locating epenthesis sites .?'
b. Tashlhiyt: Determined by sonority relationships, v. SonPeak,
NORR and other constraints in Chapter 4.
As already noted in § 4.1, (46) and (47) are logically independent. To see
this , note that there is no contradiction involved in the idea of a language
which would combine MA's strictures on the class of nuclei with Tashlhiyt's
sonority-driven distribution of syllabicity. Imagine for instance a language
which resorts to vowel epenthesis to break up consonant clusters, with the
site of insertion always located immediately to the left of a sonority peak
of the input string. In such a language /smtl/ would surface as sem.tel and
/smlt/ would surface as smelt. Nor is there anything contradictory in the idea
of a language which would be like Tashlhiyt in allowing any consonant
to be a nucleus, but in which sonority would not have any influence in
the syllabification of consonant sequences. Imagine for instance a language
with a procedure which is like RIGHT- TO-LEFT-SCAN (5) , except that
it turns consonants into nuclei instead of inserting e. The procedure in
question scans words starting from the end and syllabifies as .Ce.. any
CC string which is not immediately followed by a vowel or by a C which
has been made a nucleus in a previous iteration .f In such a language
sequences /smtl/ and /smlt/ would be syllabified respectively as sm.tl and
sm.!!.
MA and Tashlhiyt both have hollow syllables. In MA the nuclei of hollow
syllables are occurrences of e or eC sequences while in Tashlhiyt they are
consonants. In spite of this difference the hollow syllables occupy similar
positions in the overall syllable structure of either language. When one
compares the syllable types of MA with those of Tashlhiyt, the two inven-
tories show obvious analogies. Among the hollow syllabies, in particular,
the two languages show parallel weight contrasts. These are repre sented,
details set aside, in the table below .

60 We are abstracting away from those cases in which the vocoid is absorbed by an adjacent
sonorant (see above in § 8.2.2).
61 As mentioned earlier, RIGHT-TO-LEFr SCAN is only a temporary device adopted for
expository convenience .
62 Here and below, underlinings indicate nuclei.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 263

(48) Tashlhiyt MA

Light Ce!:.
Heavy Ce!:.C
What we intend to do is compare the distribution of the nuclear Cs in the
two languages. We will see that those of MA are much less under the
influence of restrictions on sonority than those in Tashlhiyt.
The reader may recall that a key factor in the distribution of sonority
in the syllables ofTashlhiyt is the constraint SonPeak. Let us reproduce here
our formulation of the constraint in § 4.7.
(49) SonPeak: A sequence which is a sonority peak within the
syllabification domain contains a syllable nucleus."
SonPeak was formulated under the implicit assumption that nuclei always
consist of a single skeletal position, an assumption which we abandoned
when we posited complex nuclei in MA. We modify slightly our formula-
tion of SonPeak to adapt it to representations which contain complex nuclei.
(50) SonPeak: a sequence which is a sonority peak within the
syllabification domain contains a skeletal position
which belongs to a syllable nucleus .
For the reader's convenience we have italicized the words which were added
to (49) to yield (50). Even with this new formulation, SonPeak cannot
help us much in bringing out the differences between syllabification in
MA and in Tashlhiyt, and it is not difficult to see why. We have assumed
(i) that in MA every nucleus contains a vocoid, and (ii) that vocoids have
a higher degree of sonority than contoids. Taken together, these assump-
tions do not leave much room for violations of SonPeak in MA. As far as
SonPeak is concerned, MA is as different from Tashlhiyt as, say, French
iso
And yet it makes sense to want to be able to use SonPeak in order to
compare the distribution of sonority in hollow syllables in the two languages.
Take for instance the sequence /zrb/. Let us compare forms of either
language which can be minimal embodiments of this sequence, i.e. surface
forms which contain as few distinctive feature specifications as possible
in addition to those already present in /zrb/. In MA the forms in question
are zerb and zreb, which are both well-formed: zreb 'he was in a hurry'
vs. zerb 'hurriedness'. In Tashlhiyt /zrb/ can only give rise to zrb, a verb

63 A sonority peak is any sequenee which is a loeal maximum of sonority, see § 4.7. We
are assuming that MA has the same sonority seale as that proposed for Tashlhiyt in § 4.7.
This assumption will be modified later, see § 8.5.3.
264 CHAPTER EIGHT

with the same meaning as MA zreb/" The respective syllabic parses of

MA zerb and zreb are .zerb. and ez.reb: that of Tashlhiyt zrb is .zrb.. In
Tashlhiyt, dissyllabic z.rb, in which z and bare nuclei, is an impossible
surface form in the language. Its ill-formedness is due to a violation of
SonPeak: r is a sonority peak and yet it is not a nucleus. We would like
to be able to consider paralleIs like the following :
(51) Tashlhiyt MA
a. .zjb. .zerb.
b. *.l,.r12. .ez.reb.
In a given line in (51) the apportionment of the consonants between onsets
and rimes is the same in both forms, e.g. r belongs to a rime in both forms
in (5l)a whereas it is an onset in both forms in (51 )b. (51) is but one
particular illustration of the fact that MA is more liberal than Tashlhiyt in
the range of such apportionments it allows. This fact is an empirical one;
it remains unaltered if one claims, contrary to the analysis we are advo-
cating, that in Tashlhiyt all hollow syllables have complex nuclei of the form
eC.
Strictly speaking, neither of the MA forms in (51) violates SonPeak,
for the sonority peak is e in both . In order to compare MA with Tashlhiyt,
we want to introduce an extended version of SonPeak which would enable
us to say that .z.reb. is a violation in that extended sense, while .zerb. is
not. Since MA has a vowel e and Tashlhiyt does not, we want to leave
out the occurrences of e in our computations of sonority contours in MA.
For the purpose of comparing MA and Tashlhiyt we want sequences zreb
and zerb to have the same sonority contour; we furthermore want this
sonority contour to be the same as in the schwa-less sequence zrb , This
can be achieved in the following way.
Let us suppose that in the phonological derivations of MA there is an
intermediate level of representation in which the unstable vowel is an empty
skeletal position, i.e. a skeletal position which does not have any feature
bundle attached to it. 65 We give below the representations at the level in
question for the verb zreb 'be in a hurry' and its deverbal noun zerb
'hurriedness', and we add for the sake of comparison the surface repre-
sentation of the Tashlhiyt verb zrb, which also means 'be in a hurry'.

64 In Tashlhiyt the deverbal noun of zrb is zzrb (/I-zrb!).

65 HarreIl (l962a) already proposed that MA schwa does not have any supra-Iaryngeal
features of its own.
 SYLLABLE ST RUC TU RE I N MORO C CA N A RA B I C 265

(52) a. zreb b. zerb

o o o
I
R
-<:R
0
»<.
0 R
I I ~
N N N D

x
»<.x x x
-<.x x
->.
x x
I
x
I I I I I I
z r b z r b
c. zrb
o

0
-<.R
r-.
N D
I I
x x x
I I I
z r b
Since the three phonological objects in (52) contain the same sequence of
melodie unit s, they have the same sonority contour. In (52) , MA zerb
complies with SonPeak, as Tashlhiyt zrb does."
There is admittedly something paradoxieal about disregarding the feature
content of schwas when evaluating MA forms for violations of SonPeak.
The raison d' etre of schwa epenthesis is presumabl y to enable the nuclei
of all the syllables of MA to be above a certain sonority thre shold, and
we have j ust propo sed that the feature specifications introduced as a result
of schwa epenthe sis be ignored in the computation of sonority contours. We
will see in § 9.1 that the paradox is only apparent.
To give the reade r a feeling for the empirical import of SonPe ak (50),
let us review a few instance s in which SonPeak is met or violated. Consider
line la in (28), which we reproduce again here with its syllabie parse:

(53) a. mir !le='Ywrem xil=u !fe=l=herb !zatm-a

b. mir le'Y w rem xi lu fel herb zat ma
H L L L L L H H L
1 2 3 4 5 6 7 8 9

66 MA zerb and Tashlhiyt zrb are homophonous. On such homophon ies, v. § 9.3.
266 CHAPTER EIGHT

The occurrences of r in !"'/rem and llierb are sonority peaks. The latter
occurrence complies with SonPeak, since it belongs to a nucleus (see
syllable #7), but the former does not, since it is an onset (see syllable #3).
As another example, consider line l2b in (28), which we give again here:

(54) a. le=)'wram i-luh le=lJSiq l=le=hmum

b. lev" ra mi luh lell si q#l- leh mum
L L L H H L L L H
1 2 3 4 5 6 7 8 9
The first position in the sonority peak II straddling syllables #7 and #8
belongs to a nucleus, while the occurrences of l in syllables #1 and #5, which
are also sonority peaks, are onsets and consequently violate SonPeak . Here
are a few other examples which the readers may check for themselves in
(30) and (31) : l meets SonPeak in 7a:2 and IOa:3. Sonority peaks vio-
lating SonPeak are f in 6b:6 and m in 11b:8.
As illustrated by the above sampie, MA allows violations of SonPeak
which cause -ill-formedness in Tashlhiyt. We will show later that these
violations have two sources. One is that syllabification is influenced by
morpho-syntactic structure to a much greater extent in MA than in Tashlhiyt.
The other resides in the sonority scale of MA, which is less differentiated
than that of Tashlhiyt.
Glides are a very common source of SonPeak violations in MA, as they
are in Tashlhiyt, e.g. at the beginning of line 2a in (28) the sequence
/w=l=)'wra/ surfaces as .wel-v'.ra. (HL), with a violation of SonPeak in
the first syllable: /w/ is a sonority peak (it is more sonorous than /11) and
yet it does not belong to a nucleus. We will deal later with certain aspects
of the syllabification of high vocoids in MA. We simply note at this point
that the high vocoids of MA raise basically the same problem as those in
Tashlhiyt." The representations which are inputs to word-level syllabifi-
cation must distinguish high vocoids of a special kind, which we call
underlying glides . As far as morphology is concerned the underlying glides
behave like consonants. Whereas a run-of-the-mill high vocoid may be
syllabified as a vowel in some environments and as a glide in others,
underlying glides as a rule surface as glides in all environments. As in
Tashlhiyt, in MA some of the underlying glides have their source in tem-
platic morphology, others in lexical representations. Reverting now to our
example, an idiosyncracy of the word meaning 'and' is that its lexical
entry contains two allomorphs, /u=/ and /w=/ which are in free variation
in some contexts." It is the latter which occurs at the beginning of line

67 See Chapter 7.
68 See Heath (1987: 288).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 267

2a in (28). Using the former allomorph instead would have resulted in a line
beginning with u.lev'.ra (LLL), which does not accord with the meter
required by the song.

8.5. THE SYLLABLE STRUCTURE OF WORDS

The widespread violation of SonPeak in MA should come as no surprise,

given the way word-level syllabifieation operates in that language.
All violations of SonPeak which oecur at the line level are due to word-
level syllabification. Here is why. SonPeak violations can only occur in
plaees where there oeeurs a sonority peak. A moment's thought shows
that concatenating two strings cannot add new sonority peaks to those
already present in each string eonsidered separately." Consequently stringing
words together to make lines does not ereate new opportunities for SonPeak
violations to oecur.
In the following discussion we will take preliminary steps towards an
overall characterization of syllabification at the word level in MA, foeussing
on those aspeets which are most revealing for our comparison of the syllable
struetures of MA and Tashlhiyt.
In the kind of eharacterization of syllabifieation we have in mind for MA,
the eentral eomponent is a procedure whieh takes any string of MA and
parses it into suceessive syllables all meeting eertain eonditions on syllable
well-formedness. Sinee the distribution of eis to a great extent predietable,
we would like the predictable oecurrences of e to be inserted as a result
of applying syllabification." The only schwas whose location would be
indieated in the input strings would be the unpredietable ones. To see what
we have in mind, eonsider the following pair.
(55) (a) (b) (e)
ltenzra .tenz.ra. ftnh-aI 'billy-can'
!tnezna .et.nez.na. ftnz-nal 'we made fun of'
Each line gives (a) the pronunciation of a word recorded in a standard
transeription, (b) the eorresponding syllabic parse and (c) the string which
we are assuming is the input to syllabifieation. A eomplete aecount of
syllabification in MA must indicate how one gets from (c) to (b). In MA
syllabifieation, aside from glide epenthesis, which was mentioned briefly
near the end of § 8.3.2, the only ehanges employed to make strings syl-
labifiable are schwa epenthesis and schwa deletion. Syllabification does not

69 More precisely, for any two strings A, Band their concatenation AB, every segment
which belongs to a sonority peak in AB also belongs to one in A or B. Let A and B be
the strings qfxbsn and mtiw . The sonority peaks in Aare fx and n, those in Bare m and
iw, and those in AB are fx, nm and iw.
70 This was already the general spirit of Keegan' s (1986) pioneering article.
268 CHAPTER EIGHT

insert, metathesize, (de)geminate or delete full segments, nor does it modify

their feature content." Let us assurne that the grammar of MA contains
various constraints which jointly define the set of strings which are well-
formed syllabies. Let us temporarily lump all these constraints together
under the label SWF (Syllable Well-Formedness). SWF has the same
function as the syllable templates used by various authors." A complete
characterization of the syllable structure of MA requires more than a char-
acterization of SWF ; SWF must be supplemented with what one might
call a syntax of syllabIes. The syntax of syllables is a set of conditions
limiting the privileges of occurrence of the various syllable types within
words. Here are two reasons why such conditions are needed . First, con-
catenating two well-formed syllables does not always give rise to a
well-formed structure. To take a trivial example, it and ak are well-formed
syllables of MA, but it.ak is not well-formed, because in a MA word only
the first syllable may lack an onset. A second reason is that an input string
may have several syllabic parses which are compatible with SWF, and SWF
has nothing to say about the fact that some are well-formed and others
are not. According to SWF, for instance, syllables with rimes eC and eCC
are both well-formed . Consequently, in the first line of (55), where the gram-
matical form teni.ra contains a syllable with a eCC rime, one must explain
why the alternative parse et.nei.ra is iII-formed although it is also com-
patible with SWE
Our distinction between SWF and syllable syntax does not imply the exis-
tence of two distinct components in that part of the grammar of MA which
deals with syllable structure; it is intended merely as a typology of con-
straints which we find useful in explaining how the following discussion
fits into a complete account of syllabification in MA. We will see that
restrictions on sonority are involved both in SWF and in syllable syntax.
In the following discussion we sketch the basics of word-Ievel syllabi-
fication in MA, devoting special attention to those aspects which are relevant
in a comparison with Tashlhiyt.

71 The following fact may look like a counter-example to some readers. In perfective verbs
with sterns ending in aC the vowel becomes e when the suffix begins with a consonant:
saf ' he saw' , sef-na 'we saw ' (-na is the Ip desinence). But note that *saf-na would be
well-formed, as far as syllable structure is concerned, cf. saf=na ' he saw us' (=na is the
dolp c1itic). One of our basic assumptions is that the phonotactics of any language is due
in part to constraints on the form of words which are blind to morphological structure, and
it is constraints of that nature which are at the center of our attention in this chapter. The
alternation under consideration here is indeed sensitive to morphological structure: First,
suffixes trigger it whereas c1itics do not, and yet the difference between suffixes and c1itics
is otherwise irrelevant to syllabification in MA. Second, the alternation is limited to verbs,
e.g. /!far-t=i/ 'my female rat' , the Is possessivized form of lfar-a 'fernale rat' , is pronounced
lfarti, not *Iferti .
72 See e.g. Selkirk (1982). For MA, see Keegan (1986).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 269

8.5.1. A constraint-based analogue of right-to-left scan

Until now we have used RIGHT-TO-LEFT SCAN (5) as a convenient

expository device encapsulating a first approximation of the basic facts about
the predictable occurrences of e. We reproduce RIGHT-TO-LEFT SCAN
below:
(56) RIGHT-TO-LEFT SCAN:
Scanning the Pword from right to left, rewrite as CeC any CC
string which is not immediately followed by a vowel. Each step
in the scan must take as its input the output of the previous
step.
In (55) RIGHT-TO-LEFT SCAN correctly predicts the occurrence of e in
Itnezna. not that in !tenira ; that form contains a CCC cluster which RIGHT-
TO-LEFT SCAN should have broken up, as it has the /nznl sequence in
/!tnz-na/. Basing our description of MA on an iterative procedure like
RIGHT- TO-LEFT SCAN would not make it easy to compare MA and
Tashlhiyt, for our analysis of Tashlhiyt syllabification was couched in terms
of constraints. So as to compare MA and Tashlhiyt on equal terms , let us
discard RIGHT-TO-LEFT SCAN and replace it with constraints which do
the same work. We will first posit various constraints which will be designed
in such a way that their simultaneous enforcement will require the occur-
rence of e precisely in the locations in which RIGHT-TO-LEFT SCAN
inserts e. These constraints fall into several categories.
First, certain conditions jointly circumscribe the set of syllable types rep-
resented in (44). We reproduce here (44) as (57) to make comparison easier,"

(57) I II
rime N=V N = eC
a. N xi (se) wet
b, c ND mib (wet) hezb, hezz
d. N D-D huzz ?benss

Let us recall the most important conditions which contribute to confine

licit syllables to the types listed in (57) and assign them structures like those
displayed in (41) and (42). There are no complex onsets . In complex nuclei
the first segment must be e and the second must be a consonant. The only
complex codas are geminates following a full vowel. We will not state these
conditions formally. We simply assurne that this can be done and lump them
together under the cover term SYLL. For the sake of explicitness we also
include in SYLL two conditions of a very general nature.

73 Une e in (44), which is now irrelevant, has been left out.

270 CHAPTER EIGHT

(58) SYLL: In a structure which satisfies SYLL the following

conditions are all met:
a. every syllable belongs to one of the syllable types listed in
(57);
b. every skeletal position belongs to a syllable and there are
no overlapping syllables;"
c. geminate inseparability is respected.
The reader may recall that each of the syllable types in (57) can be instan-
tiated with an onset or without, but that onsetless syllables can only occur
at the beginning of words in isolation or at the beginning of lines . Let us
supplement SYLL with the following requirement, which is already familiar
from our discussion of Tashlhiyt.
(59) NoHiatus: A syllable which is not located at the beginning of
a syllabification domain has an onset.
As already stated in § 8.3.2/ 5 we are assuming that the units which are
domains of syllabification in MA are the word and the Phonological
Utterance. As far as syllabification is concerned, the line in poetry plays the
same role as the Phonological Utterance in other speech styles.
From this point on, unless stated otherwise the only structures which
we will take into consideration as candidates, i.e. as possible grammatical
outputs, are those which meet SYLL (58) and NoHiatus (59) . Let us call
these structures viable candidates. Given any input string there exist many
viable candidates which are compatible with that string, and our task is to
devise means for selecting one of these (or several, in cases of free varia-
tion) as the grammatical output(s). Here are for instance the viable
candidates which are compatible with the input string /ktb/ for kteb 'he
wrote': .ke.te.be., .ek.te.be., .ekt.be., .ket.be., .ke.teb., .ketb., .ek.teb.. Note
that RIGHT-TO-LEFT SCAN correctly rules them all out except the last.
For the purpose of mimicking RIGHT-TO-LEFT SCAN, SYLL is too
liberal in a number of ways. First, it allows occurrences of e not followed
by a tautosyllabic consonant (see box a-I in (57»; such occurrences are only
found in special circumstances . Second, SYLL allows occurrences of e
followed by two tautosyllabic consonants (b, e-Il) . The following condi-
tions are designed to outlaw such occurrences.
(60) NoLoneSchwa: e occurs in a complex nucleus.
(61) NoCoda: avoid codas.
NoLoneSchwa rules out syllables like se (a-I in (57» and NoCoda rules
out syllables like hezb (b, e-Il) ; on how these syllables nonetheless end

74 On overiapping syllabies, see (19) in § 8.3.1.

75 See the text below (33).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 271

up in well-formed outputs, see below. Let us now see how the various
conditions work together. Consider the following paradigm:
(62) a. qleb 'he overturned'
b. qleb-na 'we overturned'
c. qelb-u 'they overturned'
Here is how q/eb 'he overturned' derives from the input Iqlb/. All the viable
candidates are listed below:
(63) .qe.le .be. .qe.leb. .eq.le .be. .qel.be. .eql.be.
.qelb. .eq.leb.
All the items in the first line violate NoLoneSchwa. In the second line
the first item violates NoCoda while the second violates neither constraint
(on how it is so, see below). Consequently .eq.leb. is the winning contes-
tant.
Strictly speaking, the linear transcription .eq.leb. can represent two
phonological objects which differ only in the syllabic affiliation of b. In
the first object, b is a sibling of e under the nucleus node; in the second,
b is a coda. The two structures are displayed below.

(64) a. R b. R
I
N N
r-. D
-<.X
X X
I I
X
I I I I
e b e b

Since the second structure violates both NoLoneSchwa and NoCoda it will
always be disfavored and from now on it will simply not be taken into
consideration when we list the various candidates.
Going back to paradigm (62) , it is easy to see that the derivation of
the items in line b proceeds in the same manner as that of qleb . Let us
turn to line c. We list below all the viable candidates:
(65) .qe.le.bu. .eq.le .bu. .eql.bu. .qel.bu.
The first three candidates lose against the fourth, the only one which does
not violate NoLoneSchwa nor NoCoda.
In the same way as our analysis associates outputs of the form .eC.Cec.
to inputs of the form ICCCI (see above our discussion of .eq .leb. 'he over-
turned'), it correctly accounts for the fact that all naked sterns of ICCCCI
verbs are pronounced as CeCCeC, e.g. !beiyet (/!bz"(tl) 'he babbled'.
Con sider the following paradigm:
272 CHAPTER EIGHT

(66) a. lbezvet 'he babbled'

b. lbezvet-na 'we babbled'
e. lbzevt-u 'they babbled'
All the viable eandidates for I!bz)'t/ «66)a) are listed below:
(67) a. .be .ze .ve.te. a' . .eb.ze.ve.te. alt..bez.ve.te.
b. .be.zev.te. b' . .eb.zev.te. b". .be.zevt.
e. .ebz.)'e. te. c' . .bezv.te. e". .eb.zevt.
d. .ebz .vet . d' . .bez.ver,
Violations of NoLoneSehwa oeeur in all items in the first three lines exeept
in e"; violations of NoCoda occur in b", in all the items in the third line
and in d. Item d' is the only one not to eontain a violation of either eon-
straint and is therefore the winner. Here are now all the viable eandidates
for I!bz)'t-u/ «66-e):
(68) a. .be.ze.ve.tu. a' . .eb.ze .ve .tu. alt. .bez.ve.tu.
b. .be .zev.tu. b'. .ebz, ve. tu. b" . .bez)'.tu.
e. .eb.zev.tu.
Violations of NoLoneSehwa oeeur everywhere in the first line , as weIl as
in the first two eandidates in the seeond line; items b' and b" violate NoCoda.
Candidate e is preferred for it is the only one whieh violates neither eon-
straint.
Violations of SYLL or NoHiatus are never found to oeeur in lieit forms .
When NoLoneSehwa and NoCoda conflict, NoCoda gives way, as shown
for instance by the fact that /kaSf-u/ 'they guessed' yields kas.fu, not
ka.se.fu. Geminates are another souree of violations of NoCoda. Consider
for instanee the derivation of sekk 'he doubted' from /skk/. There are only
three viable eandidates for /skk/. We list them below besides the repre-
sentation of the input /skk/.

(69) x X X a. .esk.ke. b. .sek.ke. e. .sekk .

I
s
<>
k
.es.kek. and other eandidates whieh violate geminate inseparability are not
listed in (69) beeause they are not viable eandidates : viable eandidates
must by definition eomply with SYLL, which implies that they must respeet
geminate inseparability, see eondition e in the definition of SYLL in (58).
The three eandidates in (69) are represented in (70):
 SYLLABLE STRUCTURE IN MORO CCA N ARABIC 273

(70) a. .esk.ke. b. .sek.ke.

o o o o
I
R
/~
0 R
.r-:R
0 0
->.R
~ I I I
N D N N N

x
->.x I
x x x
1
x
->.x x
x
I
x
I
e
I
s
<::
k
I
e
I
s
I
e
<:
k
I
e

c. .sekk.
o
»<.R
o
r-----------D
N
~ I
x x x x
11
s e
<:
k

The first candidate violates NoLoneSchwa and NoCoda at the same time ;
the second only violate s NoLoneSchwa, while the last only violate s NoCoda.
That the last candidate is one which corresponds to the correct output shows
again that NoLoneSchwa dominates NoCoda.
Verbs like fe rtet 'pick off' and llienze: ' stare' do not violate geminate
in sep arability. We assume that MA is like Ath Sidhar Rifian in that the
representations which are inputs to syllabification allow a contrast between
a geminate, as in !1iett ' lay down ' and hezz 'take ', and adj acent occur-
rence s of the same segment, as in fe rtet and llienzez . Here are for instance
the represent ations of !1iett and fertet in the input to syllabific ation:

(71) x X X X X X X
I
1i
Vt 1
f
I
r
I
t
I
t

8.5.2. Kerneis ending in eCC; FinH

Geminate inseparability accounts for the eCCJ sequence in sekk, but not
all eCCJ sequences can be explained in that manner. We already gave in
(9)a-d minimal pairs CCeC vs. CeCC in which the only distinguishing factor
274 CHAPTER EIGHT

is the location of e. These were verb-noun pairs. Here are pairs in which
both members are nouns."
(72) a. !tres 'deaf person' !ted 'slap (in the face)'
b. !)req 'sweat' )erq 'vein'
c. !fre) 'defloration' !fer) 'branch'
d. )neb 'grapes' denb 'peach'
These examples show that the position of schwa is not always predictable
in three-consonant kernels, even if one only considers nouns. In the pairs
above the middle consonant is a liquid or a nasal, but schwa is also unpre-
dictable with other middle consonants, as suggested by the following pairs:
(73) a. qfez 'cage' !gebs 'gypsum'
b. )des 'lentils' zehS 'ass's foal'
c. dheb 'gold' sehb 'tributary'
d. sqef 'ceiling' kehf 'cave'
Before going any further, let us remind the readers of what we mean
by 'kerne!', a notion which will playan important role below. We use the
term 'kerne!' with the same meaning as in our discussion of Berber. The
kernel of a word is what remains when the word has been stripped of its
clitics and of all its prefixes and suffixes, derivational as weIl as inflectiona!.
In me-hlul-in, the mp passive participle of bell 'open', the kernel is hlul.
In the affixless word sekkek 'he caused to suspect' the kernel is sekkek.
These two examples were chosen to make it clear that kernels may be
morphologically complex, e.g. sekkek is a causative verb derived from
sekk 'to suspect', but this fact is irrelevant, as far as the notion 'kerne!'
is concerned.
An important observation about MA is that all the unpredictable schwas
occur inside kernels." Furthermore, unpredictable schwas are only found
in kernels which do not contain any full vowel. The location of full vowels
in kernels is in many cases dictated by templatic morphology. We will
ass urne that it is also templatic morphology which is responsible for the
contrast between CCeC and CeCC in nouns. We assurne that the distin-
guishing feature of the CeCC nouns is FinH, a morphological template
which requires a H syllable at the right edge of the kerne!.
(74) FinH: The right edge of the kerne1 must coincide with the
right edge of a heavy syllable.
FinH is an alignment constraint in the spirit of McCarthy and Prince (1993),
but we will sometimes call it a template, as areminder of its role in our
analysis. Recall that a H syllable is a syllable with a coda. In the input to

76 Some of these pairs were pointed out by Amimi and Bohas (1996).
77 The only exceptions are the 3fs suffix -et and the 2s clitic =ek (see infra).
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 275

syllabification, the representation of qelb 'heart' is the pair (lqlb/, FinH)

whereas that of qleb 'he overturned' is simply /qlb/. " The set of all the
viable candidates for qelb is the same as that in (63), which gave rise to
qleb 'he knocked over ' :
(75) .qe.le.be . .qe.leb . .eq.le.be. .qel.be . .eql.be.
.qelb . .eq.leb.
The need to satisfy the template FinH takes precedence over NoCoda . The
winning candidate is in consequence .qelb..
The paradigms below illustrate the fact that the contrast between CCeC
and CeCC is preserved before the affixes and clitics which begin with a
consonant, while it is neutralized before those which begin with a vowel:
(76) a. qleb 'he knocked over' qelb 'heart'
b. qleb=na 'he knocked us over ' qelbena 'our heart'
c. qelbeu 'he knocked hirn over' qelb=u 'his heart'
The preservation of the contrast follows from our analysis . Let us review
the derivations of all the forms in (76). Those of the paradigm on the left
have already been discussed, see the text around (63) and (65) . In the
paradigm on the right, in which the kernel is lexically associated with
FinH (74), the derivation of qelb=na is the same as that of qelb, modulo
the clitic. Finally, consider qelb eu 'his heart'. The set of viable candi-
dates is the same as that given in (65) for qelb-u 'they knocked over' . We
give it again in (77):
(77) .qe.le.bu. .eq.le.bu. .eql.bu. .qel.bu.
qel.bu is again the preferred candidate, as it was in (65), for it is the only
one not to violate NoLoneSchwa nor NoCoda. An important point to note
is that all four candidates in (77) violate FinH. Given the input /qlb=u/, with
/qlb/ lexically marked as subject to FinH, an output cannot simultaneously
satisfy NoHiatus and the template FinH. Forms like qelbeu 'his heart' show
how the enforeement of eonstraints on syllable strueture ean result in vio-
lations of a morphologie al template. "
The template FinH is also needed to distinguish two classes of /CCCC/
nouns. All /CCCC/ verbs have naked kernels of the form CeCCeC, as
already stated in our diseussion of lbeivet 'babble' (see (66)). Among
monomorphemic nouns, on the other hand , some have naked kernels of
the form CeCCeC, while in others the naked kernels are of the form CCeCC;
for instanee :

78 Our assumption about the underlying representations of the CCeC words will be modified
later, see § 8.5.5.
79 The interaction between FinH on the one hand, and SYLL and NoHiatus on the other
hand, is doing the same work as the cyclic derivation s proposed by Keegan (1986).
276 CHAPTER EIGHT

(78) !<i'egreb 'scorpion ' frenk 'franc ' Sfenz 'doughnut'

While the lexical entry of the first noun in (78) is a mere sequence of
four consonants, those of the two others must furthermore indicate that
the nouns in question are subject to template FinH, e.g . the lexical entry
of frenk is (lfrnkl, FinH) .80
We are now ready to turn to the role of sonority. Let us first recapitu-
late the constraints posited so far.

(79) SYLL (58) [undominated]

NoHiatus (59) [undominated]
~ NoLoneSchwa (60)
~ FinH (74) - - - - - ,
NoCoda(61)---------~

The machinery in (79) does not impose any constraint on sonority, aside
from the requirement that every nucleus contain a vocoid, a requirement
which is part of SYLL. As a matter of fact SYLL should be made more
restrictive, to account for certain limitations on complex rimes, as we shall
now see.

8.5.3. Sonority in rimes; NoRR

Whereas the freedom of occurrence of consonants in CCeC kern els is limited
only by cooccurrence restrictions on consonants in roots, a review of the
CeCC kernels in which ce is not a geminate leads to the following gen-
eralization.
(80) In word-final consonant clusters occurring in surface forms, if
the final C is a sonorant contoid, it must have a lesser degree
of sonority than the preceding C.81
(80) is stated as a generalization about clusters at the end of words, rather
than at the end of CeCC kernels, because, as it turns out, it is valid for
MA words in general, no matter what their morphological structure. We
are assuming here the following sonority scale:
(81) vocoids
liquids (L)
nasals (N)
other contoids (0)

80 frenk came into the language as a borrowing but its origin is not anymore traceable in
present-day MA. franc has lost its final velar in modern French . FaithAdapt is irrelevant
here. On FaithAdapt see below in § 8.5.3.
8\ (80) extends to all words an observation of Heath (1987: 265) about the shape of CeCC
words.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 277

Sinee the seale in (81) lumps together all the eontoids whieh are not
liquids and nasals, it is less differentiated than the one we employed in
our diseussion of Tashlhiyt (see § 4 .7).82 Consequently the eonstraints
SonPeak (50) and NoRR (see (83) below) impose weaker limitations on
syllabic parses in MA than in Tashlhiyt. Consider for instanee the under-
lying sequenee Iddfbl in line 6b in (28). Beeause frieatives are more
sonorous than stops in the Tashlhiyt sonority seale, Ifl is a sonority peak
in Iddfbl aeeording to that seale, and syllabifying it as an onset would
result in a violation of SonPeak. Aeeording to seale (81), on the other
hand, Ifl is not a sonority peak , and no violation of SonPeak oeeurs in
line 6b in (31), where it is syllabified as an onset.
As a particular eonsequenee of the generalization in (80), MA forbids
CeCC words with the three types of final clusters listed below in (82) .83
Each line in (82) eorresponds to a class of forbidden clusters. It first gives
examples of excluded sequenees ; these are followed by CCeC words whieh
show that the final sequenees in question are not excluded when they are
broken up by e. '0', 'N' and 'L' respeetively stand for ' obstruent' , 'nasal'
and 'liquid' .
(82) a. *ON bn, fn, dm, firn tben, kfen, gdem, shem"
b. *OL bl, kl, zr, hr zbel, skel, !szer, !sher 85
e. *NL mr, ml !lmer, zmel"
Consider qebl, an ill-formed word. Sinee in that word the final eonso-
nant is a sonority peak, it would seem that the ill-formedness of qebl has
its souree in the violation of eonstraint SonPeak (50). Note however that
the liquid also violates SonPeak in qleb 'he overturned', whieh is well-
formed nonetheless. We submit that the eonstraint whieh is violated in a
erucial way here is NoRR, which played an important part in our discus-
sion of Tashlhiyt syllabification in Chapter 4. NoRR is repeated below :
(83) NoRR (No Rising Rimes):
the coda does not have a higher sonority than the nucleus .
The monosyllable qebl is ill-formed beeause it violates NoRR. Note that
in monosyllabic words like x'eb: 'bread' the final cluster does not violate
NoRR, because stops and fricatives belong to the same rung of scale (81).
Being a constraint on codas, NoRR predicts that generalization (80),

82 Zeroual (2000) advocates a more differentiated scale . with h between the liquids and
the nasals, and fbetween the nasals and the obstruents. His conclusions are based on a detailed
survey of the CC clusters in CeCC nouns. Unfortunately, this work became available too
late for us to take it into account.
83 Counterexamples will be taken up later.
84 tben ' straw', kfen ' shroud', gdem 'he el', shern 'grease' .
85 zbel ' rnountain' , skel 'shape', !szer 'trees , col' , !sher 'month' .
86 ! ~ mer 'length of life' , zmel 'camel' .
278 CHAPTER EIGHT

which was stated for word-final clusters, is also valid for preconsonantal
CC clusters. The prediction is borne out by the facts . We have not found
any counter-example in HarreIl and Sobelman's dictionary. Here is a sample
of the words with internal eCC rimes.
(84) fermli 'nurse' berd'i -a 'pack saddle'
!xenfr-a 'nose' genbri 'banjo'
<ienkbut ' spider web' zerdb-a 'ko feline, f'
festq-a 'pistachio' !<ieskr-i 'soldier'
!qWestl-i 'chestnut (color)' zeb-tehum 'I brought them'
In the list above, the CCC sequences are all tautomorphemic except that
in the last item, zeb-tehum, where the eCC rime coincides with the end
of a syntactic word . Since the consonant of the 1s suffix I-tl is an obstruent,
it cannot give rise to eCC rimes violating NoRR, no matter what conso-
nant occurs at the end of the preceding stem."
We have not found any words in which an internal CCC sequence has
a steadily rising sonority profile, i.e. words such as fetmli or seblyun ."
This is additional evidence that the constraint responsible for generalization
(80) is NoRR, rather than SonPeak. If there existed a wordfetmli, its syllabic
parse would have to be fetm.li. This parse does not violate SonPeak, for
m is not a sonority peak within the domain of syllabification, that is, within
the word (see the formulation of SonPeak in (50». fetm.li violates NoRR,
on the other hand ."
We now turn to forms which fly in the face of generalization (80). All
of them are CeCC nouns. Examples are given below.
(85) I II
a. lsetr'" satr ' line'
b. !feqr faqr 'poverty'
c. <iedl <iadl 'justice'
d. resm rasm 'drawing'
e. fehm fahm 'understanding'
f. wezn wazn 'weight'
Column I contains counterexamples to generalization (80). Column 11
contains their counterparts in CA ; the relevance of the latter will be
explained below. The last four items in (85) are the only exceptions to
generalization (80) which we have found in the lists in Amimi and Bohas

87 On the ls suffix I-tl see § 8.5.5.

88 Cf. the noun meaning 'the Spanish, col'. which is !sbelyun, not ! seblyun .
89 For other languages with a prohibition against syllable-final clusters with a rising sonority
contour, v. e.g. Cantineau (1960: 114-115, 118-119), Kenstowicz (1986), Kouloughli (1978) ,
Hayward (1988) and Bohas (1999) .
90 This word is distinct from its everyday synonym !ster, see below .
 SYLLABLE STRUCTURE I N MOROCCAN ARABIC 279

(1996). In the remainder of this subsection we show that the existence of

the words in column I does not invalidate our claim that NoRR is cru-
cially involved in shaping well-formed codas in MA.
Before we go any further, let us make sure that the items in the first
column of (85) are indeed monosyllables with complex rimes rather than
dissyllables, i.e. that the word !setr in (85)a, for instance, is .setr. and not
se.ter.
That the words under consideration are monosyllabic is shown by their
prosodie silhouette and by the fact that their schwa must be pronounced
with glottal vibrations no matter what the surrounding consonants. Let us
take up these points in turn.
In !setr, for instance, eis auditorily more salient than r. The word sounds
rather like the English word sucker, mutatis mutandi s. When it is uttered
under a sentence-final rising intonation the main pitch event is located on
e. We now turn to the voicedness of schwa. If the correct syllabic parse
of the words in (85)1 were Ce.CeC, one would expect the first schwa to
devo ice when the adjacent consonants are both voiceless, see (85)a,b . In
fact schwa must be pronounced voiced in such cases. Compare for instance
!setr, in which the voicing of schwa is mandatory, with !setter 'he drew a
line', in which the first schwa must devoice (one hears [!str]). As elsewhere
in this book , the indications just given pertain to isolation forms, i.e.
prepausal ones. When a suffix or clitic follows !feqr, schwa devoice s in
the expected fashion , e.g. it must be voiced in a prepausal occurrence of
!feqr 'poverty' but it is voiceless in !feqr=u 'his poverty'.
There is no denying that the items in the first column of (85) belong
to MA. Indeed, the list of such items can be extended ad libitum, as will
become clear shortly. However it must be noted that they all belong to a
special speech register, a crucial feature which is not indicated in HarreIl
and Sobelman's dictionary or in the lists in Amimi and Bohas (1996), where
they are listed indiscriminately alongside other MA words.
All the forms under consideration belong to classicized MA. Classicized
MA is a formal speech style intermediate between everyday MA and CA.
It is commonly used by educated speakers in certain kinds of face-to-face
interaction involving some degree of formality, e.g . by professors and
students in university lecture halls or by participants in political caucuses."
Very roughly, to use classicized MA is to speak MA with borrowings from
the CA lexicon . In the process the borrowed CA words are moroccanized,
that is, they are adapted to the phonology of MA. Except for a few dif-
ferences, the syntax and inflectional morphology of classicized MA are those
of everyday MA, and so is the phonology. One phonological difference is
the addition of the glottal stop to the set of underlying consonants. Another
is the use of CeCC words such as !feqr, which violate NoRR. While they

91 For some discussion of classici zed MA, see Heath (1989) and Youssi (1992).
280 CHAPTER EIGHT

are speaking in the classicized register, the speakers are directly tapping
their knowledge of CA. Virtually any lexical item of CA that they happen
to know is eligible for use, once they have adapted it to the phonology of
MA. It is quite common for speakers to borrow from CA a word which they
have never heard used as a classicism before. In such cases the CA word
is moroccanized off the cuff.
In classicized MA, CA words of the form CVCC are uniformly adapted
as CeCC, as examplified in (86):
(86) MA CA
a. beht bahe 'research, monograph'
b. neht naht 'sculpture'
c. nehz nahz 'rnethod'
This pattern of adaptation is mimicking the regular correspondences
which relate the everyday words of MA with their cognates in CA. In par-
ticular, when a CA word is of the form CVCC its reflex in (vernacular)
MA is almost always a CeCC word, as examplified below :
(87) MA CA
a. xWebz xubz 'bread'
b. kelb kalb 'dog'
c. sebt sabt 'Saturday'
d. teht taht 'under '
e. be'ld ba'ld 'after'
The regularity illustrated in (87) breaks down, however, when we consider
the CVCC words of CA in which the final cluster has a rising sonority
contour. The shape of their reflexes in colloquial MA is not CeCC, which
would violate NoRR; it is CCeC instead:
(88) MA CA
a. !'ldem 'laQm 'bone'
b. hbel habl 'rope'
c. !bher banr 'sea'
d. !qWten qutn 'cotton'
e. !sher sahr 'month'
The upshot is that NoRR is always unviolated in MA except in one
class of words: classicisms of the form CeCC, in which, presumably, NoRR
is overriden by a constraint which requires adapted forms to mirror as
faithfully as possible the canonical shapes of their CA sources. To the
ears of the MA speakers, the final rising sonority contour of !feqr 'poverty'
is a telltale sign that it is a classicism. Classicisms which become estab-
lished words in the common language have a tendency to become
regularized to meet NoRR. The classicism !setr 'line' (see (85)a) has
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 281

spawned the regularized form !ster, which is nowadays in everyday use

along its older synonym l xett.
In conclusion, NoRR does indeed playa role in restricting the privi-
leges of occurrences of consonants in present-day MA, and the existence
of classicisms which violate NoRR does not disprove this claim."
In the lexical entries of nouns such as kelb 'dog', telu 'lower part' , the
template FinH is an idiosyncratic property on a par with the consonant
sequences /klb/ and /tht/. FinH is also used to derive nouns from tricon-
sonantal verbs , e.g. bret ' he plowed' liert 'plowing' (see (9) for other
examples) . What would happen if template FinH were used to derive a noun
frorn a CCeC verb like ster 'cover' , in which the final consonant is a
sonorant contoid with a higher degree of sonority than the medial conso-
nant? If we assume that NoRR is ordered above FinH , the result would
be a CCeC noun. To see this, consider the set of all viable candidates for
the input (/str/, FinH):
(89) .se.te.re. .se.ter. .es.te.re. .set.re. .est.re.
.setr. .es.ter.
The only candidates not violating NoLoneSchwa are those of the second
line, viz .setr. and .es.ter.. The former conforms to the template but violates
NoRR , while the latter abides by NoRR, but does not conform to the
template. If the grarnrnar of MA ranks NoRR more highly than FinH, the
result of rnapping /str/ to template FinH is ster. Our grammar accounts
for the inexistence of deverbal nouns of the form CeCC which violate
NoRR. The ternplate FinH, which normally gives rise to CeCC nouns, is
overridden by the prohibition of rising rirnes . Let us recapitulate the
constraints proposed so far:

92 In his account of forms like those in (85), Kaye (1987) relies on two debatable assump-
tions. One assumption implies that all forms which are counterexamples to generalization (80)
are nominalized verbs ('masdars'). This assumption is incorrect, as implied by our presen-
tation of c1assicized MA. Here are words which contradict (80) and are not masdars: !setr
(85)a, !f eqr (85)b, neim 'star, famous performer' (cf. CA nazm), ! fesr 'era' (cf. CA fa:l'r).
The author also assumes implicitly that in MA all nominalized verbs are built on (what
amounts to) a CeCC template . Actually , the nominali zations of many CCeC verbs are of
the form CCuC or CCiC (e.g. dxul 'entering' , from dxel 'enter' , srit 'plowing', from sret
'plow' ). As for nouns lacking a full vowel which are related to CCeC verbs, many of them
are of the shape CCeC, e.g. fqe s 'disembowelment ' (cf. fqe s 'disembowel ' . fteS ' thirst' (cf.
fteS 'be thirsty') .
282 CHAPTER EIGHT

(90) SYLL (58) [undominated]

NoHiatus (59) [undominated]
FaithAdapt I

CE
NoRR(83)
NoLoneSchwa (60)
FinH (74) ~
NoCoda (61)

In the hierarchy in (90), FaithAdapt is a cover term for whatever constraints

are at work to guarantee that when words from other languages are adapted
into MA, the moroccanized forms bear some ressemblance to the original
words." Consider the following nouns, both of the form CeCC: zens 'ass's
foal', an everyday word, and ! Yerd 'formal talk, presentation', a classi-
cism derived from CA Yard In our analysis the CeCC shape has a different
origin in the two words. Whereas the CeCC form of zens 'ass's foal' is
due to the template FinH, that of ! Yerd 'presentation' has its source in
FaithAdapt. The fact that the everyday word zens is the historical reflex
of CA ians 'ass's foal' is irrelevant for the characterization of that form
in a synchronie account of MA. Neither has this fact anything to do with
FaithAdapt, a component in the mental machinery which enables speakers
to moroccanize words from other languages, CA among them.
NoRR is higher than FinH in the constraint hierarchy, while it is lower
than FaithAdapt. In other words, syllable well-formedness overrides a
template of the language, but it yields to some of the likeness constraints
which shape moroccanized words. The latter situation is not unusual. Many
languages loosen their phonotactics to accomodate loanwords. In Parisian
French, for instance, a glide cannot be followed by a tautosyllabic conso-
nant, but this prohibition is not enforced in recent loans, e.g. design [dizayn],
Nike (a commercial brand) [nayk].

8.5.4. Favoring sonority peaks as nuclei; SonPeak

We have used SonPeak (50) in § 8.4 as a means of pointing out system-

atic differences between the syllable structures of Tashlhiyt and MA, but
SonPeak does not play any role in the analysis of MA syllabification which
we have started to develop in the present section. This analysis implies
that the basie pattern for syllabifying /CCCCV/ is eC.CeC.CV rather than

93 On phonological adaptations, see e.g. Vip (1993), Paradis and LaCharite (1997) and
references therein. In the case of MA, see Heath (1989). Some of the constraints which are
needed specifically to account for moroccanizations do not belong to FaithAdapt. An example
of a phenomenon which falls outside the purview of FaithAdapt is the fact that the short
vowel s of CA are represented by e or zero in MA c1assicisms, e.g. CA fadl 'justice' is
moroccanized as Yedl, not Yadel; despite the fact that fadel is a well-formed MA kerneI,
cf. fbatel 'calumny' and sabel 'shad'.
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 283

CeCC.CV, i.e. light syllables are preferred, except when certain overriding
factors comme into play. One such factor, as we have just seen, is the
template FinH. The ranking of FinH above NoCoda is wh at makes CeCC
nouns possible. We now wish to suggest that SonPeak is another factor
responsible for the occurrence of heavy hollow syllables. We will present
two facts in favor of this claim. One, which concerns syllabification in
certain nouns, will be discussed presently. The other, a free variation which
occurs in versification and in colloquial speech, will be presented in § 8.5.6.
The only hollow H syllables for which the analysis recapitulated in
(90) makes allowances are those whose right edge coincides with the right
edge of a kernel, e.g. those in qelb 'heart' and qelb ena 'our heart'. Not
all violations of NoCoda can be attributed to FinH, however. One area in
which they crop up is the nouns of the form ICCCCVI in which ICCCCI
is monomorphemic. Consider the following pair:
(91) a. mzebna /mzbn-a/ .em.zeb.na. 'large intestine'
b. !xenfra I!xnfr-a/ .xenf.ra. 'nose'
The syllabification in (9l)a is that predicted by our analysis in its present
state. /mzbn-a/ yields miebna for the same reasons as I!bz"{t-ul 'they
babbled' yields !bieytu (on the derivation of that form, see (68) and the
surrounding text) . We sub mit that in (91)b the realization of I!xnfr-a/ as
!xenfra rather than !xnefra is due to SonPeak: parsing Ixnfra/ as xenf.ra
has the effect of including In/, which is a sonority peak, into a nucleus.
We have culled Harrell and Sobelman's dictionary for all the nouns in which
a full vowel is preceded by four consonants; they comply with SonPeak
in their overwhelming majority. In these nouns, ICCCCVI sequences in
which the second consonant is a sonority peak surface as CeCCCV. Nouns
of this type have already been given in the first three lines in (84) . Here
are other examples.
(92) !merstan 'asylum' sertl-a 'set of gold bracelets'
sensl-a 'chain' sendgur-a 'germander'
!xenfra and forms like those in (92) suggest that SonPeak is ranked higher
than NoCoda: in the correct parse xenf.ra the sonority peak Inl is con-
tained in a nucleus, as required by SonPeak (50), and the price to pay for
compliance with SonPeak is a violation of NoCoda. SonPeak must also
be ranked below the template FinH, as shown by the fact that there exist
CeCC nouns in which the initial consonant is a sonority peak, e.g. left
'turnip', mesk 'musk', nelis 'bad luck', weld 'boy'. According to OUf
analysis the lexical entry of left 'turnip' is (lIft!, FinH). That FinH is
ranked higher than SonPeak is shown by the fact that left, which violates
SonPeak but complies with FinH, is preferred to *lfet (.el.fet.), which
complies with SonPeak but violates FinH. There also exist longer nouns
lexically associated with FinH which compel us to adopt the same ranking,
284 CHAPTER EIGHT

e.g. frenk 'franc' (v. (78» and !setreni ' chess' . If SonPeak took prece-
dence over FinH these words would be pronounced *f ernek and *!sternei .
With these additional rankings, our analysis is now that shown in (93).

(93) SYLL (58) [undominated]

NoHiatus (59) [undominated]
FaithAdapt - - -
NoRR (83) - - -
.-----+--NoLoneSchwa (60)
~inH(74)

SonPeak (50)
I
l-----NoCoda (61)------'

The analysis predicts that ICCCCVI sequences in which the first conso-
nant is a sonority peak surface as CCeCCV (i.e. .ee. Ceü. CV), and the
prediction is borne out by the nouns in HarreIl and Sobelman 's dictionary.
We have already encountered such a noun, i.e. mzebna in (9l)a. Here are
others .
(94) mxezn-i 'ko herald' !rbe)tas 'fourteen'
mheqb-a 'flower pot' !msetr-a 'ruler'
We have examined the facts about ICCCCVI sequences in which the first
or the second C is a sonority peak. Those in which the third C is a sonority
peak all surface as CCeCCV, e.g . bdenial 'eggplant', sberdil-a 'pair of
sneakers', but this fact is not additional evidence in favor of the role of
SonPeak. If the sequences in question surfaced as CeCCCV, e.g. if the
noun meaning 'eggplant' were *bedni al, not only would these sequences
violate SonPeak, they would also run afoul of NoRR, which is dominated
only by FaithAdapt in the constraint hierarchy of MA.

8.5.5. Kerneis ending in eC; FinL

The analysis summarized in (93) now runs into the following problem. Since
SonPeak is ranked higher than NoCoda, how come the influence of SonPeak
does not override that of NoCoda in verbs? If this were the case, Iqlbl 'he
overturned', which is pronounced qleb (see the paradigm in (62», would
be pronounced qelb instead, and I!hlt/ 'he mixed', which is pronounced
! iexlet, would be pronounced ! ixelt, and similarly for all the verbs with
kerneis of the form ICCCI or ICCCCI in which the penultimate C is a
sonority peak. Recall that we assumed earlier that ICCCI and ICCCCI
verbs have no associated morphological template and that the consonant
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 285

sequence contained in their lexical entry is all that is needed to predict

their syllable structure . This assumption now turns out to be mistaken.
Suppose that ICCCI and ICCCCI verbs are all lexically associated with
a template, call it FinL, which requires their kerneis to end with light
syllabies; suppose furthermore that FinL is ranked above SonPeak. The
lexical entries of kteb 'write' and lbezvet 'babble', for instance, are now
(lktb/, FinL) and (I!bz)'t/, FinL) . Let us assume furthermore that FinL and
SonPeak are both ranked below Nol.oneSchwa." In this modified analysis
Iqlbl ' he overturned' and Iqlb-ul 'they overturned' still surface as qleb
and qelbu, and similarly I!ixlt/ 'he mixed ' and I !ixlt-ul 'they mixed' still
yield the observed forms ! iexlet and ! ixeliu. 95 The template FinL is also
needed to account for the CCeC and CeCCeC nouns in which the penul-
timate C is a sonority peak, e.g. ltres 'deaf person' (v. (72)a) and !feqreb
'scorpion', for which we would otherwise expect the correct pronunciations
to be *lters and *! Yqerb, if SonPeak had its way.
Since the only syllables concerned by FinL are those at the end of kernels,
our analysis still predicts that codas should arise whenever required by
SonPeak, in syllables whose right edge does not coincide with the right edge
of a kernel. As an example of this prediction , consider !sentrei 'chess' .96
Under our assumptions the lexical representation of this noun is (I!sntrz/,
FinL) . For the sake of brevity, the set of candidate parses considered below
in (95) does not comprise all the viable candidates, but only those viable
candidates which meet NoLoneSchwa (60) and NoRR (83), two constraints
which are ranked high in the constraint hierarchy displayed in (93).

(95) (I!sntrZ/, FinL) FinL SonPeak

a. .sen.terz. *! ok
b. ~ .sent.rez ok *
c. .es.net.rez. ok * *!
The ranking of FinL above SonPeak prevents the occurrence of a coda in
the kernel-final syllable (line a), but not in the first syllable; minimizing
SonPeak violations is what makes (95)b a better candidate than (95)c.
We have just seen how FinL guarantees that ICCCI verbs are all of the

94 The ranking of NoLoneSchwa above SonPeak is need ed in order to account for the
CCeC words in which the medial C is a sonority peak, e.g. !sreb 'he drank' . Candidates
.ser.be. and .es.reb. both meet FinL. .es.reb., which violates SonPeak, is preferred to .ser.be.,
which satisfies SonPeak but violates NoLoneSchwa, because NoLoneSchwa is ranked higher
than SonPeak.
95 The reader is referred to our prev ious discussion of qleb and qelb-u , whose viable
candidate sets were given in (63) and (65), and to that of !beiyet and !bieyt-u, for which
see (67) and (68).
96 This is the word given for 'chess' in HarreIl and Sobelman (1966: 156). It is unknown
to ME, who only knows the variant l setrerd;
286 CHAPTER EIGHT

form CCeC, SonPeak notwithstanding. The introduction of FinL also brings

us closer to the solution of another problem. MA has two verbal desinences
which are distinguished only by schwa: I-tl and I-et/, which respectively
mark the 1st person singular and the 3rd person feminine singular." Here
are examples.
(96) 3ms Is 3fs
bze-y bze-y-t bezv-et 'crush'
lbezvet lbezvet-t lbzevt-et 'babble'
kellern kellem-t kellm-et 'speak to'
bukes bukes-t buks-et 'brim'
Let us assurne that a schwa is already present in the underlying form of
the 3fs desinence l_et/. 98 The problematic forms are those in the 1st person
singular, in which schwa must be prevented from appearing before the It/
of the desinence. The forms to compare here are bzey-t 'I crushed' and
!beiyet 'hebabbled'. The same reasons which make our present analysis
derive the correct output !beiyet for (/!bz-yt/, FinL) 'he babbled', make it
predict the incorrect form bezv-et for 'I crushed': NoCoda favors a final
L syllable in both cases. Let us examine the viable candidates for the input
/bz-Yk tl ('I crushed'), where the bracket labelIed L indicates the right
edge of a kerne I which has an associated template FinL. We consider only
those viable candidates which satisfy NoLoneSchwa.
(97) .e b.z e -Y]L 1.

The correct pronunciation is actually that which corresponds to the third

candidate, but our analysis elects the first candidate, for the two others
violate NoCoda. FinL is of no help here because all three candidates violate
it: in none of them does the right edge of the kerneI coincide with the
right edge of a syllable. Note however that whereas the last segment of
the kernel is an onset in the first two candidates, in the third it OCCUfS at
the end of a nucleus. The third candidate would be the preferred one if
the prosodic constituent involved in FinL were a nucleus rather than a
light syllable:
(98) FinL: The right edge of the kernel must coincide with the right
edge of a nucleus.
This slight reformulation of FinL does not impinge on the other predic-
tions made by our analysis. For FinL (98) to favor bzeyt over beiyet in
(97) , FinL must be ranked above NoCoda, a ranking which is consistent
with the fact that FinL dominates SonPeak (see (95)) and SonPeak domi-

97 On these two desinences see e.g. Heath (1987: 233).

98 The other morpheme of MA which must possess a schwa in its underlying form is the
2s clitic l=ekJ, see e.g. beZy=ek 'he crushed you' (*biey=k) .
 SYLLABLE STRUCTURE IN MOROCCAN ARABIC 287

nates NoCoda (see (93)). We end up with the following hierarchy of con-
straints:

(99) SYLL (58) [undominated]

NoHiatus (59) [undominated]
FaithAdapt =:J
NoRR (83)
...---+--- NoLoneSchwa (60) - - - - - ,
FinH (74) ---f-------,

~nl (98)---'

SonPeak (50)----+------'
I
' - - - - - NoCoda (61) -------'

Further development of the analysis summarized in (99) will require addi-

tional empirical work . We have not examined the pronunciation of the
suffixed forms of four-consonant verbs, for instance , and it remains to be
seen whether these forms are impervious to the demands of SonPeak, as
the standard transcriptions imply. When we had to cite an instance of a four-
consonant verb, in (66), we selected lbe ivet, which has no sonority peak
in terms of the sonority scale which we have adopted for MA (see (81)).
Before we leave (99), let us point out an empirical inadequacy which
we will leave as a problem for further research. The problem has to do
with those nouns containing ICCCCVI sequences in which the first two con-
sonants both belong to a sonority peak . Our analysi s incorrectly predicts
that the grammatical output should be CCeCCV. Consider for instance nouns
like menzl-a ' flu' and mendb-a 'a lot' , in which the first two consonants
make up a sonority peak . Our analysis incorrectly predicts *mnezla and
*mnedba. Here is for instance the set of viable candidates for Imnzl-a/ :
(100) .me.ne.ze.la. .me.nez.la. .men.ze.la. .emn.ze.la.
.em.ne.ze.la. .em.nez.la. .menz.la .
All the candidates but the last two are fatally marred by violations of
NoLoneSchwa. The penultimate candidate is preferred by our analysis,
because it does not violate any of the constraints in (99), while the last
candidate, which is the actually occurring form, violates NoCoda. The same
problem arises with words such as festq-a 'pistachio' and similar ones in
(84), in which the consonant cluster does not contain a sonority peak.
The problem which motivated our discussion of FinL was that of
accounting for the fact that verbs such as qleb withstand the pressure of
SonPeak. We now come to data which show that template FinL some-
times yields to that pressure.
288 CHAPTER EIGHT

8.5.6. Free variants in which SonPeak overrides Finl:

In the paragraph below we use 'CReC' to refer to any CCeC sequence in

which the middle C is a sonorant and has a high er degree of sonority than
the surrounding consonants. In MA some CReC sequences have an alter-
native realization which complies with SonPeak, viz CeRC. 99 This
phenomenon occurs both in everyday speech and in singing. In Lmnabha
MA W-internal CReC sequences regularly have a free variant CeRC, e.g.
gleb-na 'we overturned' can also be pronounced gelb-nal'" The free vari-
ation can also occur in W-final sequences, but in that environment it is
subject to restrictions which have yet to be worked out. The option of
pronouncing CeRC instead of CReC also exists in some other dialects, but
with limitations different from those which obtain in Lmnabha.'?' Although
the author does not explicitly say so, the data presented in EI Mejjad
(1984: 103-104; 141-144) suggest that in Marrakesh any W-final CReC
sequence may also be pronounced as CeRC. In the variety of MA described
in Heath (1987), whose main consultants were from 'the Fes/Meknes region'
(p. 2), CReC and CeRC sequences remain distinct unless the medial con-
sonant is r, in which case they are homophonous in all contexts (pp.
249-253).
Melhun lines in which CReC must be pronounced CeRC are quite
common, and we now present a few examples to illustrate this phenomenon.
To begin, let us consider the first line in the 22nd couplet of the 'ballad
of Fatma', the beginning of which was cited earlier in (28). We give the text
of the line in (101) and its parse in (102) (see (30) for other lines with
the same meter).
(101) !tleq-t=ek be=flal=ek te-msi lm-qasm-a'?'
(102) 1 2 3 4 5 6 7 8 9
H L L L L L H H L
tleq tek bef 1a lek tem sim qas ma
This line is well-formed. Its first word is !tleq-t=ek 'I released you' (release-
ls=d02s), a word which begins with CCeC, as all analogous verbal forms

99 A simiIar phenomenon is attested in Berber. In the Tamazight diaIect of Ait Seghrouchen,

according to GuersseI (1977: 274-275), CCeC is obligatorily changed into CeCC whenever
the middle C is a sonorant.
100 CeRC sequences in which R is a sonority peak may not be pronounced CReC, e.g.
kelbena 'our dog' does not have a free variant klebna .
101 Certain nouns have undergone lexical change, e.g. the words meaning ' muslim' and
'earthen dish' are mselm and mterd in Lmnabha, while they are recorded as meslem and metred
in HarreIl and Sobelman (1966), who state (p. ix) that their dictionary is based on the speech
of educated speakers from Fes, Rabat and Casablanca .
102 'I let you go, crushed under the weight of your deeds' .
 SYLLABLE STRUCTURE TN MORO CCAN ARABTC 289

(cf. kteb-t ' I wrote'). The expected pronunciation is ltleq-t, but here it
must be pronounced Itelq-t instead .
Let us give two more examples of the same phenomenon. One is taken
from ' the seizure of Oujda' , a song by Hasem Sshdani to be found in AI-
Malhuni (I 990a: 253-255). The relevant line is the second in the 16th
couplet. In that song, both lines in each couplet have the same meter. We
first give the first line of the couplet for the sake of comparison.
(103) a. zar lweet-zebber weeblev mentha fzureu
b. la bqa l=u illa kesra i-xlef kesra 103
(104) 1 2 3 4 5 6 7 8 9 10
H L L L L L H H L L
a. zar wet zeb ber web lev ment haf zu ru
b. lab qa lu yel la kes ray xelf kes ra
The relevant form in this example is i-xlef 'he replaced' , near the end
of line b, in which the kernel must be pronounced xelf if line (l04)b is to
be well -formed,
Our last example is taken from our own trans cription of a song in a
tape by L'Haj L'Houcine Toulali (1=1iaii 1=1iusin t=tulal-i).l04 It is the sixth
line in the fourth stanza. Thi s song has a rather complex stanza structure.
For the sake of comparison we fir st give the sixth line of the pre ced ing
stanza, which has the same meter.
(105) a. was ntiya l-yum sem'ieek !sami
b. fi west l-qelb zned !nar !dram=i 105
(106) L L L H H L L L L
1 2 3 4 5 6 7 8 9
a. wa s#n ti yal yum sem c;ek sa mi
b. fi wes t#1 qelb zend na r#d ra mi
For line b to be well-formed, zned 'he lit up' must be pronounced zend,
and this is indeed the singer 's pronunciation in the recording.

103 (a) 'He became unfair and tyrannical, and reached the depths of sin': (b) 'It only remained
for hirn to name hirnself Kisraa so as to replace Kisraa' (CA kis raa = Caesar).
104 Side One of tape TCK684 (no date or place indicated on the casse tte's packag ing).
The title of the song is 'Fatma', like that of the song to which line (102) belongs; it was
composed by Dris Ben-Ali Elmalki.
105 (a) 'Have you now become deaf?' ; (b) 'At the bottom of the heart he lit up the fire of
passion' .
290 CHAPTER EIGHT

8.6 . SUMMARY

In this chapter we first presented the basic properties of syllable structure

in MA, drawing our evidence from versification and from phonotactics.
In addition to the full vowels a, i , u, the surface representations of MA
contain a vowel schwa. Abstracting away from differences in the nuclei
of hollow syllabies , the inventory of syllable types of MA is similar with
that of Tashlhiyt. There are no complex onsets, and only geminate conso-
nants can give rise to complex codas. Hiatus is prohibited.
In the second half of the chapter we laid the groundwork for a constraint-
based account of word-level syllabification. In MA, surface syllabification
is sensitive to morphological structure to a much greater extent than in
Tashlhiyt. The syllabification of words results from the interplay of two
kinds of con straints, the alignment constraints and the phonotactic con-
straints. The alignment constraints in our analysis play the same role as
templates in classic analyses of CA. Some of these are imposed by mor-
phology while others are lexically specified. The two alignment constraints
which we propose restriet syllable structure at the end of kemels. The phono-
tactic constraints operate regardless of morphological structure. Many of
them , most notably NoHiatus, override the alignment constraints.
Sonority plays a role in the syllable structure of MA. This role is less
conspicuous than in Tashlhiyt because the influence of constraint SonPeak
is sometimes masked by that of the alignment constraints. The sonority scale
needed for MA has fewer sonority classes; all the consonants less sonorous
than the nasals are lumped together into a single sonority class.
The evidence from versification and from phonotactics led us to the
conclusion that in most instances hollow syllables have a consonant in
their nucleus. By positing complex nuclei of the form ee, we were able
to accept that conclusion while maintaining the assumption that in MA every
syllable contains a vowel. This assumption will be reexamined in the next
chapter.
 CHAPTER NINE

VOWELLESS SYLLABLES IN
MOROCCAN ARABIC

In this chapter we explore a modified version of our analysis of syllable

structure in MA. In this new version, syllable s with a schwa would all be
reanalyzed as vowelless. In the end we settle on a mixed account , in which
hollow syllables are vowelless, except in special circumstances, in which
they contain schwas.
According to the analys is of MA developed in the preceding chapter,
all hollow syllable s have complex nuclei consisting of two segments, schwa
and a consonant. In this chapter we ask to what extent e can be dispensed
with, i.e. whether MA is not after all a language which, like Tashlhiyt, allows
syllable nucle i which have consonants as their sole content. The idea is
not new. It was suggested in DE (1988) on the basis of versification . Durand
(1994, 1995/96) advocates phonemic transcriptions with no schwas: in his
transcriptions, some occurrences of ' e' in the standard transcriptions are
replaced by a syllabicity mark under the following consonant while others
are omitted altogether.
Con sider the verb !snet 'he slashed' and the noun !sent ' slash'. The
surface representations of these words in our present analysis are displayed
below in (1); the modified representations which we will be exploring are
displayed in (2):

(1) a. !snet [!sn@t] b. !sen t [!s@nt]

o o o
I »:-:R »<.
r-------
R 0 0 R
I I
N N N D

X
»<.X X X
»<.X X
-<.X
X
I
X

e
I I
s n
I I
e
I
t
1
S
I
e
I
n t
I

291
F. Dell et al., Syllables in Tashlhiyt Berber and in Moroccan Arabic
© Kluwer Academic Publishers 2002
292 C H A PTE R N INE

(2) a. !snet [!sn @t] b. !sen t [!s @nt]

o o o
I
R
»<.R
0 0
-<.R
I
N
I
N
r-.
N D
I I I I
x x x x x x
I
s
I
'h
I 1
s
I
n
I
t
t
The affiliation of the consonants to syllabic constituents is the same in
both analyses, which only differ in how they account for schwa. As indi-
cated by the transcriptions [!sn@t] and [!s@nt], the most salient auditory
difference between the two form s is the location of @ . In (I) this differ-
ence is represented directly in the surface representations. The letter ' e'
represents a certain bundle of feature specifications which is realized as [@].
Thi s feature bundle is linked to a skeletal po sition wh ich immediately
follow s the syllable onset and belongs to the nucleus, like the following
consonant. In (2), on the other hand, the skeletal position of the onset and
that of the nuclear consonant are adjacent, and so are the feature bundles
linked to these positions. We will explore the claim that the voiced vocoid
which is heard bet ween the onset and the nucl ear co nso nant is not the
phonet ic manife station of a segment (of a bundle of distinctive features),
but mere ly a transition f rom the onset consonant into the nuclear conso-
nant.
As we shall see, this claim cannot be maint ained to the end in this radic al
formulation . Our strategy will be to drive this extrem e version as hard as
we can and mitigate it onl y as a last resort. Our argumentation in favor
of the new analy sis implied by (2) will be twofold. First, the new anal ysis
is simpler. Second, it allows a more insightful characterization of the dif -
ference s between the syllable structures of MA and Tashlhi yt; it also permits
a more straightforward account for their phonetic ressembl ance s.
In what follows , we will often need to refer to examples and to constraints
given in the preceding chapter. To avoid con stant repetition, let us agree
that while numbers between parentheses refer to item s occurring in the
present chapter, numbers between angled brackets refer to items contained
in Chapter 8.

9. 1. THE NEW ANALYSIS IS SIMPLER

The function of some comp onents in our previous analysis was to account
for the fac t th at e is alw ays preceded by an onset, while that of others
was to account for the fact that e is in most cases followed by a nuclear
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 293

segment. These parts of the machinery can altogether be dispensed with

in the new analysis, in which e stands for a transition between an onset
and the following nucleus.
In (1) !snet, which is pronounced [!sn@t], is syllabified as es.net, with
an initial e to which no voiced vocoid corresponds in the pronunciation.
This is but a particular instance of our general observation that only after
an onset can e be realized as a vocoid, see (34) in § 8.3.2, where the
observation was left unaccounted for. Under our new proposal, the problem
disappears: what we took to be the segment e is actually the transition
from an onset to the following segment, and since in the dissyllabic word
! snet the first syllable lacks an onset, there is no reason for us to expect
it to be pronounced with an initial vocoid.
If there is no such thing as a postpausal occurrence of e, there is also
no reason to worry about postvocalic schwa deletion, a que stion men-
tioned near the end of § 8.3.2. Consider the sequence iwa !snet 'then he
slashed', whose syllable structure is i. was.liet. If the first syllable of ! snet
is .s. rather than .es., no deletion is involved in the derivation of the hinge
syllable . was. by utterance-level syllabification. All that is involved is the
resyllabification of a nucleus as a coda, in a manner exactly parallel to
the gliding of the initial vowel of itim 'orphan' , which becomes a glide
in lga ytim 'he found an orphan' .
Another phenomenon which ceases to be seen as a case of 'schwa
deletion' is the alternation between @ and zero in words ending in VCeC.
'Schwa deletion' occurs in particular when the following word begin s with
an onsetle ss syllable, as in the sequence vabet Yliya in line 6a in (28) . Under
our new analysis, word-level syllabification parses vabet as . va.bt. and fliya
as . f.li.ya. To avoid hiatus between the nuclei t and f , the sequence is
reparsed as fab .tf.li.ya by line-level syllabification.' On the alternation
between VC@C# and VCC# in other contexts, see below in § 9.2.
In our discussion of SonPeak in § 8.4 we agreed that the sonority contour
of the surface forms of MA was to be computed without taking into con-
sideration the occurrences of schwa and we commented on the paradoxical
nature of that convention. We now see that there was nothing paradox-
ical, after all . We are proposing now that the surface representations of
MA do not contain any vowel schwa, i.e. that the only feature bundles
characterizing syllabic vocoids are those which characterize a, i and u.
Feature specifications not present in a representation cannot contribute to
the sonority contour of that representation.
A pause is in order to clarify our use of the symbol '@' in what folIows.
Let us first recall the meaning of 'e'. We use the symbol 'e' with two values,
depending on the context. In some cases 'e' represents a phonological object ,
namely a certain feature bundle belonging to a syllable nucleus. In other

J For other similar instances in (28), see lines 7a, 8a, 9b and 12a.
294 CHAPTER NINE

cases 'e' is a letter in the standard transcriptions of MA; that letter indicates
that the preceding consonant is a syllable onset, see (3). Turning now to
' @' , the reader may recall that it has been used until now to represent
any voiced vocoid which is not a 'full segment', i.e. which is neither a
full vowel nor a consonant. When the symbol '@' was first introduced in
this book (v. § 2.2), it was indicated that the timber of @ varies depending
on the surrounding segments. In § 8.2 we explained that two kinds of
short voiced vocoids occur in MA. Some are manifestations of the presence
of a syllable nucleus whereas others are not. To cite an earlier example,
the pronunciation of !nqes 'he diminished' is [lnxqas], which normally
we note simply as [!n@q@s]. Until the beginning of the present section,
our analysis was that the @ between n and q was a transitional vocoid,
whereas the @ between q and s was the realization of the segment e, and
our analysis was only meant to account for the second occurrence of @.
We are now entertaining the hypothesis that both kinds of @ are tran-
sitional vocoids . Let us use the expression 'post-onset @' and others like
it to refer to a short voiced vocoid which immediately follows an onset. Our
new analysis will be like the analysis which it replaces, in that it will only
account for post-onset occurrences of @, e.g. it will have nothing to say
about the first @ in the pronunciation [!n@q@s].
We are now ready to examine how our new proposal impinges on the
formulation of the constraints. In a nutshell: the only constraints in (99)
which are affected are SYLL, which must be modified slightly, and
NoLoneSchwa, which vanishes .
Consider constraint SYLL, which requires, among other things, that every
syllable belong to the syllable types listed in table (57), see the formula-
tion in (58)a. We give in (3) the modified version of table (57) which results
from replacing the eC nuclei by the corresponding simple C nuclei in hollow
rimes:
(3) rime N = [-cons] N = [+cons]
a. N xi (se) wl
b, c ND mib (wet) nzb, nzz
d. N D-D huzz ?bnss

The underscores indicate nuclear consonants. Like Tashlhiyt, MA now

allows any segment to be a nucleus. For the time being we set aside the
syllables with secondary rimes, which are indicated by parentheses.' The
discussion will first proceed as though such syllables did not occur in
MA. They will be dwelt upon in § 9.5.
A central fixture in our previous analysis was the constraint
NoLoneSchwa (60), which required e to occur in a complex nucleus. Our

2 On secondary rirnes, see (42) and the surrounding text.

 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 295

new proposal does not contain any constraint which would be the coun-
terpart of NoLoneSchwa. Recall the facts which NoLoneSchwa was meant
to account for: eC rimes count in most instances as light in versification,
and post-onset @ does not occur word-finally or before cv. These facts
now simply follow from our new conception of post-onset @. Post-onset
@ is a vocoid which is produced when the articulation moves from the
implementation of a feature bundle in an onset to that of a feature bundle
in a nucleus. Post-onset @ cannot occur anywhere else than 'before' a
nucleus, then. If constraint NoLoneSchwa disappears from the analysis,
so must its rankings with the other constraints. We have given evidence
for three such rankings: with FinL, SonPeak and NoCoda, and NoLoneSchwa
was the higher-ranking constraint in all three cases, see (99). Our previous
grammar contained one independent stipulation for each of these rankings.
In our new account the separate facts which previously justified the three
rankings are but manifestations of the same fact: the distribution of tran-
sitional vocoids is governed by the machinery of phonetic implementation,
and phonetic implementation operates later than constraints FinL, SonPeak
and NoCoda, which belong to the phonological component.
Note that discarding the segment e greatly reduces the size of the set
of viable candidates for any input. Take for instance the derivation of qleb
'he overturned', which was first discussed in § 8.5.1 and was discussed
again at the beginning of section § 8.5.5. The underlying representation
of this form is (lqlb/, FinL). We reproduce again in (4) the set of viable
candidates given earlier in (63).
(4) .qe.le.be. .qe.leb. .eq.le.be. .qel.be. .eql.be.
.qelb. .eq.leb.
As pointed out in our discussion of (63), all the candidates in the first
line violated NoLoneSchwa. None of these candidates has a counterpart
in the new analy sis. To see this , take for instance the first candidate,
.qe.le.be.. Since in the new analysis there is no segment e, the counter-
part of .qe.le.be. in this analysis would be .q.l.b ., a sequence of three
syllables each comprised only of an onset. Such a structure clearly violates
SYLL, i.e. it is not a viable candidate . The only members of (4) to have
counterparts in the new analysis are the two items in the second line. The
candidate set in (4) becomes that in (5):
(5) .qlb. .q.lb.
.q.lb. is preferred over .qlb. for the same reason that .eq.leb. was preferred
over .qelb. in the old analysis. The highest-ranked constraint relevant here
is FinL, and .q.lb. complies with FinL whereas .qlb. violates it. To conclude :
under the new analysis examplified in (2), in the terminal representations
of the phonological component the grammatical output for (lqlb/ , Finl.)
would be Q.lh, to be phonetically implemented as [ql@b].
296 CHAPTER NI NE

The fact that sets of viable candidates are much smaller does not con-
stitute an argument in favor of the new analysis. The point of discussing
the derivation of qleb 'he overturned' was rather to illustrate that the one-
to-one correspondence between the derivations under the old analysis and
those under the new analysis is straightforward.
(6) below is the constraint hierarchy which results from removing
NoLoneSchwa from (99) and reformulating SYLL (58) so that the well-
formed syllable types become those listed in (3) .3

(6) SYLL [undominated]

NoHiatus (59) [undominated]
PaithAdapt ~

CS
---NoRR (83)
---PinH (74)----.

inL (98)

SonPeak (50)

~ocoda (61)---'

9.2. EXPANDED HOLLOW SYLLABLES

Consider again (2)a, the representation which our new analysis assigns to
! sliet 'he slashed' . When uttered in isolation, this word must be pronounced
[!sn@t], with a voiced vocoid between li and t. How can one maintain
that this voiced vocoid is only a transition between li and t, while both
consonants are voiceless? According to our preliminary observations on
MA, both in the Oujda dialect and in the Lmnabha dialect, a short voice-
less vocoid cannot occur between two voiceless consonants unless the first
consonant is a syllable onset. If the post-onset schwas occurring between
voiceless consonants were all manifestations of a vowel, one could claim
that phonetic implementation in MA must meet the same requirement
MINIMAL-PATH(voice) as in Imdlawn Tashlhiyt."
Similarly, how can one maintain that the voiced vocoid which must occur
in the isolation form of lmet 'perjury' is only a transition between n and
t? Again, preliminary observations on the Lmnabha dialect suggest that a
short voiced vocoid cannot occur between two homorganic stops differing
in sonorancy unless the first stop is an onset. If post-onset schwas occur-
ring in such sequences were all realizations of a vowel, one could claim that

3 On geminate inseparability in the new analysis, see below in § 9.4.3.

4 On MINIMAL-PATH(voice), see § 6.3.1.
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 297

phonetic implementation in Lmnabha MA must meet the same require-

ment MINIMAL-PATH(place) as Imdlawn Tashlhiyt.'
In the face of these objections we will weaken our claim rather than
abandon it altogether. We now claim that in the general case the nucleus
of a hollow syllable is a single skeletal position linked with a consonant,
but that there are particular contexts in which that nucleus takes on an
expanded form , with a vocoid preceding the nuclear consonant. This alter-
nation is akin to phrase-final lengthening in other languages."
Con sider the following sentence.
(7) ma fateh bab=u 'he is not opening his door '
In (7) the participlefateh ' opening' can be pronounced with a voiced vocoid
between t and h or without. Whenfateh is pronounced [fat@n] the sentence
can only have the meaning indicated in (7). Whenfateh is pronounced [fath],
on the other hand , sentence (7) is homophonous with sentence (8) below;
it is ambiguous between meanings (7) and (8) :
(8) ma fat hbabeu 'he has not abandoned his close relations'
Sentence (8) has only one pronunciation; a voiced vocoid may not occur
between the final t of fat and the initial h of hbab=u.
The alternative pronunciations of [ateli in (7) are an example of what
Heath (1987 : 248-249) calls Forward Syncope, a free variation which affects
word-finallight syllables when the preceding syllable contains a full vowel. '
The comparison between (7) and (8) shows that Forward Syncope does
not impinge on the syllable count. Both variants of sentence (7) have five
syllables: ma.fa.t(eth.ba.bu. In one variant the syllable .teb. is pronounced
with an 'uncontroversial' schwa, while in the other it is voiceless throughout
and homophonous with the hinge syllable .t# h. in (8) . We propose that
the alternation between the two pronunciations of fateli is one between a
simple nucleus and a complex one in the last syllable, as represented below
in (9).

5 On MINIMAL-PATH(place), see § 6.3.1.

6 On final lengthening , see e.g. Beckman and Edwards (1990).
7 See also Harrell (1962b: 17).
298 CHAPTER NINE

(9) a. [fath] b. [fat@n]

o o o
.r-:
o

0
»<.R 0
->.R 0
»<.R 0 R
I I I I
N N N N

x x
I
x x
I
x x
I
x x
-<.x
I I I I I I I I I
f a t n f a t e n

In (9)b the letter 'e' stands for a tree-geometric Root node dominating the
specifications [-cons] and [+voice] . These specifications guarantee that
the segment occurring after t will be a voiced vocoid.
Let us say that a hollow syllable is basic when it has a simple nucleus,
and that it is expanded when it has a complex nucleus. (9)a ends in a basic
hollow syllable while (9)b ends in an expanded one . We take the basic
hollow syllables to be the normal case, with the expanded hollow sylla-
bles occurring only under special circumstances. At present we are unable
to spell out the exact distribution of the basic/expanded distinction; we
can only make a few suggestions.
ME feels that the difference between the two pronunciations of (7) is
primarily a matter of tempo: only in relatively slow pronunciations does
it feel natural to pronounce a voiced vocoid in the final syllable in fateh
in (7) . Let us speculate that the difference between the two variants of (7)
is one of intonational phrasing: whereas the sentence comprises only one
Intonational Phrase (henceforth: IP) at a normal rate of delivery, it is broken
down into two IPs when uttered at a slower rate :"
(10) a. [ma fateh babeul., (normal rate, v. (9)a)
b. [ma faten],p [bab=u]IP (slower rate, v. (9)b)
We are assuming here that only when it is at the end of an IP can the last
syllable of fateli be pronounced with a complex nucleus. Here is another
example, which suggests in a more direct manner that intonational phrasing
is indeed involved in the alternation between basic and expanded hollow
syllables, Consider the pair of sentences in (11), where the brackets indicate
the edges of a subordinate clause:

8 In (10) below the labelIed brackets indicate the edges ofIntonational Phrases. On the phono-
logical constituent 'Intonational Phrase ', see Selkirk (1978) and work in its wake , e.g.
Nespor and Vogel (1982), Rice (1987), Selkirk (1984).
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 299

(11) a. [ila ma sket] fi sa'1-a te-sxef

'if he does not shut up, she will faint right away'
b. [ila ma sket fi sa'1-al te-sxef
'if he does not shut up right away, she will faint'
The two sentences differ only in the syntactic affiliation of the phrase f;
sa fa 'immediately'. The phrase belongs to the main clause in (ll)a whereas
it belongs to the conditional in (1l)b. In (ll)a sket 'he fell silent' must
be pronounced with a voiced vocoid between k and t, a pronunciation which
is only optional in (11)b. This difference is to be expected, if in (ll)a the
occurrence of the right edge of the subordinate clause calls for that of an
IP.
Why should hollow syllables take complex nuclei when they occur at the
end of an IP? Recall that complex nuclei contain voiced vocoids. The
possession of a voiced vocoid is presumably a useful feature for a syllable
to have, if that syllable is to be the locus of suprasegmental phenomena
associated with the end of an IP. That this conjecture is on the right track
is suggested by two facts .
The first fact has to do with pitch before a pause . For the last syllable
of ZVCeC words like fateli to take on their expanded form, occurrence at
the end of an IP is a necessary condition but not a sufficient one, witness
the fact that before a pause these words can appear either as [ZVC@C] or
as [ZVCC]. However, as Heath (1987: 184) has pointed out, only [ZVC@C]
is allowed when the intonation requires a pitch rise on the last syllable.
The second fact has to do with the nature of the nuclear consonant. Not
all ZVCeC words show an alternation between [ZVC@C] and [ZVCC].
Those ending in an obstruent do, while those ending in a sonorant do not,
e.g. !damen 'guarantor', saken 'inhabitant', qawel 'he promised', bayen
'apparent' only have one acceptable pronunciation before a pause, with
prominence on the last syllable no matter what the intonation." Such words
do not show variations dependent on tempo analogous to those exempli-
fied in (7). Consider for instance the following pair of sentences:
(12) a. kan '1amel farina m'1a s=s'1ir
aux do:prt wheat with defebarley
'he grew wheat with barley'
b. kan '1am l=farina m'ia s=s'1ir
aux year defewheat with defebarley
'(that year) was a year with wheat and barley'

9 This generalization simplifies slightly. Words of the form ZVL@N (L a liquid, N a nasal)
allow the alternation between [ZVL@Nl and [ZVLN1,e.g. salem 'unharmed' , !dalem 'unjust'.
The alternation in these forms is left unaccounted for by the analysis proposed below .
300 CHAPTER NINE

The structural difference between (l2)a and (l2)b is parallel with that
between (7) and (8), and yet the facts about pronunciation are different.
Whereas (7) has two contrasting pronunciations, one of which cannot be
a pronunciation of (8), (l2)a can only be pronounced in one way ; it is
homophonous with (l2)b unless the ambiguity is avoided by artificial means
such as the insertion of pauses, i.e. the momentary cessation of articula-
tion.
Whereas the final syllable in fateli has the two variants depicted in (9),
the final syllable in Yamel has only one. Which variant does it lack? Since
.met. is homophonous with the hinge syllable .m#t. in (l2)b and since
syllables of the form .C#c. have simple nuclei, we are lead to the conclu-
sion that the final syllable of Yamel does not have an expanded variant.
We conjecture that it does not have one because it does not need one: the
sonorant consonant in its nucleus provides enough elbow room for the
deployment of suprasegmental features. We have reached the following
generalization: hollow syllables have simple (vowelless) nuclei, except when
they are IP-final and their nuclear consonant is an obstruent, in which
case their nucleus is complex, with schwa occupying the first skeletal
position.
It should be possible to modify the constraints recapitulated in (6) so
as to make them select well-formed representations meeting the above
generalization, but we will not attempt such a reformulation, because we
feel that little genuine insight would be gained thereby. Instead, as a
makeshift device for the sake of explicitness, we posit the following rule,
which operates on the syllable structures generated by the constraints in
(6):
(l3) IP-Final Epenthesis:
In the last syllable of an Intonational Phrase, if the nucleus
does not contain a sonorant, make it complex by inserting e
before the nuclear consonant.
Take for instance bent 'perjury' and te-bnet 'you perjured yourself' .
The underlying forms of these words are respectively (lhnt/, FinH) and
(lt-hnt/, FinL). The syllable structures assigned to these inputs by the con-
straints are displayed below in (l4)a and (l4)b.
 VOW ELL E S S S Y L LA B LE S I N MORO C CA N ARABI C 301

(14) a. bent (all env) b. te-lmet (not IP-final)

o o o

0
->.R 0
-<.R 0
->.R
N
r-. D N
I
N
I
I I I I
X X X X X X X
I I I I I I I
ti n t t ti n t
c. te-bnet (IP-final)
o o

0
~
R 0
.>:R
I I
N N
I ~
X X X X X
I I I I I
t ti n e t

Since in (14)a the nucleus is a sonorant, the epenthe sis rule does not apply
and (14)a is the repre sentation of bent in all contexts. In (l4)b, on the
other hand, there are two syllables neither of which has a sonorant in its
nucleus. When (14)b occurs at the end of an IP the rule turns (14)b into
(14)c , but elsewhere (14)b remain s unchanged . Since the rule operates
only on IP-fin al syllables, it leaves the first syllable of (14)b unaffected
no matter in what context. In our analy sis there is no proce ss of schwa
devoicing.
IP-Final Epenthe sis guarantees that at least one sonorant must OCCUf in
an Intonational Phrase. Thi s generalisation seems more accurate than
Mitchell 's (1993 : 61), according to whom words spoken in isolation and
utte rances must all contain at least one voiced vocoid. Mitchell 's state-
ment is not true of the varieties of MA spoken in Lmnabha and Oujda, where
even in isolation a word like Jtel-t 'I rolled ' may very weIl be pronoun ced
[ftlt], with /tlt/ realized with an uninterrupted period of coronal closure.
A question which comes up naturally at this point is that of stress in MA.
Author s who are otherwi se mindful of phonetic detail deny the existence
of any clear pattern of word accentuation in MA. 1O We greatly doubt that

10 See Heath (1987: 266), Mitchell (1993: 144) and Durand (1995/96). EI Mejjad (1985: 158
ff.) reports a predictable difference between final stress and penultimate stress in Marrakesh
MA. Benhallam (1990) and Durand (1995/96) contain partial surveys of the meagre litera-
ture on stress in MA.
302 CHAPTER NINE

Lmnabha MA possesses anything that one could call word stress. The
little evidence that we have managed to gather suggests that the stress of
Lmnabha MA is rather similar to that of Parisian French , where stress is
a property of phonological constituents larger than the word, presumably
IPs, and where the main stress of an IP occurs on the last syllable or on
the penultimate." Prominent syllabIes in Lmnabha MA seem to have even
less auditory salience than in Pari sian French, witness the fact that in
Lmnabha MA it is even more difficult than in Parisian French to devise
pairs of sentences with patterns of prominence which are in clear contrast.
We give one such pair in (5).

(5) a. za hemmu, la tub hrir mli'a=h

come :3ms Hemmu (proper name), not cloth silk with=3ms
'Hemmu came (and) he had no silk cloth with hirn'
b. zahem mula tub hrir
jostle:3ms the:one:with cloth silk
'he jostled the one with silk cloth'

If the third syllable in (5)b (mu) is pronounced with the degree of pro-
eminence required on that in (5)a, the resulting pronunciation of (5)b is
ill-formed. Note that in order to secure a clearcut contrast in (5), we had
to make use of a break between clauses , as we did in (11).
In the alternative pronunciations of laten in (7) and in those of sket in
(l l jb an uncontroversial schwa alternates with a putative one. When we
made the distinction between uncontroversial schwas and putative ones at
the end of § 8.2, OUf discussion implied that the phonological constituent
relevant for describing their distribution was the word. We have just seen
that the constituent involved is larger than the word. The reason the dif-
ference between W-final and W-internal contexts seemed relevant then
(see (8» is simply that at the time we only considered words pronounced
in isolation, i.e. words which are prepausal and postpausal at the same time.
Facts about the edges of a word pronounced in isolation are actually facts
about the edges of a Phonological Utterance coextensive with that word.
IP-final syllables with nuclear obstruents are not the only hollow sylla-
bles in which positing a vowel seems unavoidable in Lmnabha MA. A vowel
schwa must also be posited in [@w] diphthongs and in one class of tem-
platic plural nouns. Discussion of these is deferred to § 9.3.3. and § 9.5
for reasons of expository convenience.
As noted in section § 8.2.2, MA shows dialectal variation as to which
hollow syllables may/mu st be realized with a voiced vocoid between the
onset and the nuclear consonant. Consider example (I) /tt-ksf-tu/ 'you (p)

11 See Deli (1984).

 VOWELLESS SYLLABLES IN MOROCCA N ARABIC 303

were manhandled', whose syllabic parse is t.tk.sf.tu. That form may be

pronounced [tks@ftu] in Oujda, but not in Lmnabha, where only [t.ksftu]
is acceptable. In terms of the analysis set forth in this chapter, the
penultimate syllable in t.tk.sf.tu. may have a complex nucleus in Oujda
but not in Lmnabha. Impressionistic observations by ME suggest that
Southern MA , which includes the dialect spoken in Lmnabha, stands at
one end of the spectrum of variation, with complex nuclei occurring only
in a very limited range of contexts. The dialects which stand at the other
end of the spectrum are certain varieties of MA spoken in the 'Oriental'
(Morocco's eastern province) and also the Aroubi (i.e . bedouin) dialects
in use in the countryside around Casablanca and Rabat.
Let us go back to [tks@ftu] and [t.ksftu]. Both are acceptable pronun-
ciations of /tt-ksf-tu/ in Oujda whereas only the latter is in Lmnabha.
According to our analysis, what is involved in the free variation in Oujda
as well as in the dialectal variation between Oujda and Lmnabha is the
difference between a complex nucleus (ei) and a simple one (j). It is
important to note that the representations involved in this claim are those
extant at the most superficial level of the phonological component. Our
analysis of the alternations between post-onset schwa and zero has no
obvious bearing on the question whether post-onset schwas are epenthetic,
as we claim they are, or are instead derived from underlying vowels. "

9.3. COMPARING TASHLHIYT AND MA

In our new analysis all the hollow syllables except certain IP-final ones
are vowelless. As a consequence of this, only minor differences now separate
the internal structure of syllables in MA from that in Tashlhiyt. In this
section we discuss two consequences of this state of affairs for character-
izing the differences between MA and Tashlhiyt. First, the main difference
between syllabification in the two languages now resides in the fact that
constraint SonPeak is more strictly enforced in Tashlhiyt than in MA .
Second, the reason why certain sequences in MA are homophonous with
certain sequences in Tashlhiyt is straightforward: these sequences have
identical representations at the output of the phonological component in both
languages. We now take these points in turn .

9.3.1. Well-Jormed sequences 01 syllables in MA and in Tashlhiyt

When one compares the surface phonologies of Imdlawn Tashlhiyt (lTB)

and Lmnabha MA, one conspicuous difference is that the occurrence of short
voiced vocoids is required by certain contexts of MA, while it is forbidden

12 A good example of the latter view is Kouloughli ' s (1978) analys is of an Aigerian dialect ,
304 CHAPT ER NINE

in those same contexts in Imdlawn Tashlhiyt. Consider the following

pair.
(16) a. (lTB) s')'=t [sxt] /s')'=t/ 'buy it (m)' 13
b. (MA) !sext [!s@xt] /sxt/, FinH 'cursing' 14
Both forms are heavy monosyllables with [x] in the nucleus (.sxt.) but
neither is an acceptable isolation form in the other language. MINIMAL-
PATH(voice) forbids the occurrence of a voiced transitional vocoid between
sand x in the Imdlawn Tashlhiyt form in (l6)a, while IP-Final Epenthesis
requires the occurrence of a vowel between sand x in the MA form in (16)b.
This difference between the two languages is a striking one, especially
when one concentrates on words pronounced in isolation, a context in which
the effects of IP-Fin al Epenthesis are most readily observed. However the
difference should not obscure a more central one, which has to do with
syllabification itself, rather than with the distribution of short voiced
vocoids. Very roughly: the syllable types available in both languages are the
same, but one language imposes stricter limitations than the other on syllable
sequen ces. If the analysis proposed above is adopted, a crude characteri-
zation of the difference between the syllable structures of Tashlhiyt and MA
would run like this: setting aside special cases such as IP-final hollow
syllables with nuclear obstruents (see § 9.2) and CCeC plural forms (see
§ 9.5), there is little difference between the internal structure of syllables
in Tashlhiyt and in MA;15 the differences between the two languages reside
rather in the range of possible syllabic parses which each language allows
for a given string. In Tashlhiyt the only information present in the input
strings which is relevant for syllabification is the feature content of the
segments and their linear ordering. Except in special case s," a given string
of segments can only be syllabified in one way. In MA, on the other hand,
in addition to distinctive features, syllabification must also take into account
morpho-syntactic structure. MA often allows the same string of segments
to be parsed in several different ways depending on the location of word
boundaries and on the internaI structure of words.
Consider the expressions in (17), the first two from MA and the last from
Tashlhiyt.

13 buy:aor=do3 s. Underlying /'Y/ devoices before the following /t/.

14 Cf. !sxet 'he cursed '.
15 Compare table (3) above with the inventory of syll able type s given for Ta shlh iyt in
table (28) in § 4.6.
16 See § 4.9.
 VOWELL ESS SYLLABLES IN MOROCCAN ARABIC 305

(17) a. (MA) l=<iadel hnet 'the notary committed perjury'

b. (MA) had l=nent 'this perjury' 17
c. (lTB) dlh-net 'they smashed into hirn' 18
ME cannot detect any difference between his pronunciation of the string
/dlhnt/ in the MA expression in (17)b and his pronunciation of the Tashlhiyt
expression in (17) c. He furthermore reports that a characteristic feature of
the pronunciation of MA by native speakers of Tashlhiyt is the faulty pro-
nunciation of CCeC forms like the verb hnet at the end of (17)a.
The expressions in (17) are reproduced below side by side with their
syllabic parse s:
(18) a. (MA) l=<iadel hnet .1. <ia.dlh.nt,
b. (MA) had l=nent .ha.dl.hnt.
c. (ITB) dlh-net .dl.hnt.
The parses of the MA expressions in lines a and bare those consistent
with our new analysi s, before IP-F inal Epenthesis changes t, the last nucleus
in (18)a, into et. Th at of the Tashlhiyt expression in line c is consistent
with the an aly sis of Tashlhiyt syllabification propounded in Ch apter 4 .
The representation s of the last two syllables in the expression s in (18) are
given below in (19).

(19) i. .dlh. nt. (MA) 11. .dl.h nt. (MA and ITB)
o o o o
.>.R .>.R ->.R 0.>.R
r-.
0 0 0
~
D
I I N D
N N N
I I I I I I
X X X X X X X X X X
I I I I I I I I I I
d 1 h n t d 1 'h n t
In the terminal representations of the phonological component, the string
/dlhnt/ in the MA expression in (18)b and that in the Tashlhiyt expression
in (18) c have the same representation, viz (l9)ii, a fact which is consis-
tent with their bein g pronounced alike. As for dlh.nt (see (19)i) , th at
representation is ill-formed in Tashlhiyt because n violates SonPeak.
According to our anal ysis, the reason the CCeC words of MA which violate
SonPeak are difficult to pronounce for Tashlhiyt speakers, is that the phono-
logical component of Tashlhiyt cannot generate the appropriate syllable

17 lient is a noun derived from hnet '10 perjure oneseIf' .

18 percuss-Jmpedoßms.
306 CH APT ER NINE

structure for these form s. Tashlhiyt speakers learning MA are already

endowed by their mother tongue with all the mach inery neces sary for a
correct pronunciation of MA dl.lmt . To pronounce correctly dlli.nt, on the
other hand, the se speakers have to learn to compute syllable structures
with SonPeak violations disallowed by their native grammars. We have
explained earlier how SonPeak violations which would be lethaI in Tashlhiyt
come about in well-formed MA utterances: word-level syllabification creates
SonPeak violations (FinL is ranked higher than SonPeak), and these vio-
lations are not mended by utterance-level syllabification. Another source
of such violations is the fact that the sonority scale of MA is less differ-
entiated than that of Tashlhiyt.

9.3.2. Strings pronounced alike in MA and in Tashlhiyt

With the old analysis , in which every hollow syllable of MA contains a

vowel , we face an awkward situation when we compare sequences which
sound alike in MA and in Tashlhiyt: the same voiced vocoid is the real-
ization of a vowel in MA, while in Tashlhiyt it is only a transition between
con sonants. Consider for instan ce the noun meaning ' marj oram', which
has identical pronunciations in MA and in Tashlhi yt: [brd@d :us] .19 The
surface representation of the MA form, according to our present analysis,
is that displayed below in (20)a; (20)b will become relevant later.
(20) a. o o o
»<.R
0
.>-;
0 R
-<:R
0
I
N N
I
N
-<.D
I I I I
X X X X X X X
I
b
I
r
I
d
<>
d
I
u
I
s
b. o o o
-<:R
0
-<:R
0
-<.R
0

N
I I
N
.>.D
N

X
-<.X
X X X
»<.X X
I
X
I
X
I
b
I
e
I
r
I
d
I
e
<>
d
I
u
I
s

19 The Tashlhiyt form was discussed briefly in § 6.4. I . The MA noun has a variant
merdeddus, the only one given in Harrel and Sobelman (1966).
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 307

(20)a is also the terminal representation of the Tashlhiyt form according

to the analysis expounded earlier in this book. Under the present analysis
of MA, then, the reason why the MA word and the Tashlhiyt word sound
exactly alike is simply that they have identical terminal representations.
Under our previous analysis of MA, on the other hand, the terminal
representation of the MA word is (20)b, and we are left with the problem
of explaining how the phonetic implementation components in the grammars
of Tashlhiyt and MA associate homophonous pronunciations with (20)a and
(20)b.
For the same terminal representation (20)a to give rise to homopho-
nous pronunciations in Tashlhiyt and in MA, phonetic implementation
must operate on (20)a in the same way in both languages. The little relevant
data which we have gathered suggests that there is indeed a close simi-
larity between the phonetic implementation components of the two
grammars.
To give just one example, let us take the verb Izbdl 'pull', which exists
both in Lmnabha MA and in Imdlawn Tashlhiyt, and let us compare the
pronunciations of similar forms of the MA and the Tashlhiyt verbs. We
will compare them first in isolation, a context in which they are pronounced
differently, and then embedded in a sentence in which they sound alike .
(21) a. (MA) It-zbdl, FinL .tz.bed, [tzb@d]
b. (ITB) /t-zbd/ .tz.bd. [tzb@d]
(21)a is the 3fs imperfective form, which the standard transcriptions of
MA note teibed; (21)b is the 3fs perfective form in Tashlhiyt, tibd in the
transcription used elsewhere in this book. The syllabic parse given above
for the MA form is that which obtains in isolation, a context in which
IP-Final Epenthesis operates. We give [tzb@d] as the narrow phonetic
transcription of the isolation forms of both words, but that transcription
leaves unexpressed clear differences between the two pronunciations.
In the isolation form of MA te-ibed, [@] sounds like a full-blown
vowel and the second syllable is more prominent than the first, e.g. it carries
the main pitch event in the word. In the isolation form of Tashlhiyt t-ibd,
on the other hand, [@] sounds much shorter and the main pitch movement
is on [z] . According to our analysis, these differences are due to the oper-
ation ofIP-Final Epenthesis (13). @ is the realization of a vowel in the MA
form whereas it is a mere transition from b to d in the Tashlhiyt form .
While they must be pronounced differently in isolation, the two forms
can be homophonous in nonprepausal environments. Consider the following
pair of (incomplete) sentences:
308 CHAPTER NINE

(22) a. (MA) bas te-zbed bezzaff

'for her to pull hard'
b. (ITB) a wr t-zbd bzzaff'"
'that she did not pull hard'
When (22)a is pronounced at a normal speed, ME finds it homophonous,
from t onwards, with (22)b. At a normal speed (22)a presumably constitutes
a single Intonational Phrase, and consequently IP-Final Epenthesis does
not operate in the last syllable of tZ.bd. According to our analysis, in the
output of the phonological component (22)a and (22)b have representa-
tions which are identical from t onwards : tz.bd.bz.zajf. If the relevant aspects
of phonetic implementation are identical in the two languages, one expects
the pronunciations of (22)a and (22)b to be identical from t onwards, as
indeed they are."

9.3.3. Glides which are sonority peaks

In this book we are assuming that glides have the same feature content as
the corresponding high vowels ." Under this assumption, syllabifying a glide
as a syllable nucleus turns it into a high vowel. According to the analysis
of MA proposed at the beginning of this chapter, a consonant may be the
sole content of a nucleus in certain environments; when glides occur in those
environments, then, the analysis predicts that they should surface as high
vowels. As we shall see below, there exists one class of cases in Lmnabha
MA where the prediction is incorrect. Before we turn to these cases, let
us present our background assumptions about high vocoids in the under-
Iying representations of MA and iIIustrate them by examining cases which
are not problematic for our analysis.
We assurne that in the representations which are inputs to syllabifica-
tion the difference between high vowels and glides is that the former, but
not the latter, are already associated with a nucleus node ." Let 'U' stand
for the feature bundle [-cons, +high, labial . . .]. In the input to syllabifi-
cation, 'u' represents U associated with a N node, whereas 'w' represents
U without any associated syllable structure:

20 a wr, AD neg (lad ur/).

21 (22)a allows an alternative pronunciation for which (22)b has no counterpart. That
pronunciation , which is slower, differs from the first in that the vocoid between band d at
the end of the verb is auditorily more saIient. We assurne that that longer vocoid is the
reaIization of the vowel inserted by IP=Final Epenthesis when (22)a contains two IPs: [bas
te-zbed] [bezzaff).
22 See Chapter 7. For some interesting discussion of the syllabification of high vocoids in
MA, see Keegan (1986) and Heath (1987).
23 In his discussion of the high vocoids in the Ait Seghrouchen variety of Tamazight Berber,
Guerssel (1986) proposed that high vowels have a preassociated rime node.
 VOWELLESS SYLLABLES IN MOROC CAN A R A B IC 309

(23) N
I
/u/ : u /w/ : u
Similarly, if T stands for the feature bundle [-cons, +high, coronal . . .],
'i ' in the input to syllabification represents I with an attached N node and
'y ' stands for a bare I.

9.3.3.1. Yocalized glides

In the input to syll abification, then, glides do not differ from the other
consonants , which are also without asso ciated syllable structure, and one
expects glides and contoids to behave alike in syllabification. Th is
expectation is fulfilled to a point.
Consider the CCeC adjective !byed 'white' and its fs form lbid-a. The
underlying form of !byed is (/ !byd/, FinL), which is syllabified as .b.yd.,
whence .b.y@d. by IP-Final Epenthesis (13). In (24) below are displayed
(a) the representation of !byed prior to syllabification, (b) the well-formed
output !byed, (c) for the sake of comparison, the surface form lbid , which
is not a po ssible pronunciation of !byed, and (d) the surface rep resenta-
tion of lbid-a.

(24) a. / !byd/ b. !byed

o o
I ~
R 0 R
I I
N N
I I
X X X
I I I
b I d

c. *!bid d. !bida
o o o

o
-<:R 0
~
R 0
->.R
-<. N D
I I
N N
I I I I
X X X X X X X
I II
b
I
d
I I I I
b I d a
310 CHAPTER NINE

The bracket labelled L in (24)a represents the template FinL. Note that
the representation displayed in (24)b is not the final output .b.y@d., but
rather .b.yd., which obtains before epenthesis applies. Since IP-Final
Epenthesis is irrelevant to the present discussion, we abstract away from
it whenever convenient.
.b.yd. is a better candidate than .bid. for the same reason as (lqlb/, FinL)
'he overturned' yields .q.lb. rather than .qlb. (qleb rather than qelb, in
standard transcriptions): unlike .qlb. , .q.lb. violates SonPeak, but this
violation is the cost to pay in order to avoid violating template FinL, which
is ranked higher than SonPeak (see § 8.5.5) .
Consider next !bid-a (I!byd-a/), the fs form of !byed, which is represented
above in (24)d. The medial 11/ in the kernel is syllabified as a nucleus in
/!byd-a/ for the same reason as /1/ is syllabified as a nucleus in /qlb-u/
'they overturned', which is realized as ql.bu."
Another example of underlying glides which surface as high vowels as
predicted by our analysis is found at the beginning of CCeC verbs whose
first C is a glide, e.g. (lwld/, FinL) 'give birth' . The pronunciation of /wld/
'he gave birth' is [ul@d] or [?ul@d] and that of /wld-u/ 'they gave birth'
is [w( @)ldu], where the parentheses around '@' are areminder of our
general inability to distinguish auditorily between [CR] and [C@R] when
R, a consonantal sonorant, is syllabic ." These surface forms, uled and
weld-u in standard transcriptions, are represented below in (25). The syllabic
parses of qleb 'he overturned' and qelb-u 'they overturned' are added
undemeath to illustrate the parallelism with run-of-the-rnill CCeC verbs.

(25) a. uled, qleb b. weld-u, qelb-u

o o o o

R
I
0
-<.R ->.R 0-<;R
0
I I I I
N N N N
I I I I
X X X X X X X
I I I I I I I
[ul@d] U I d [wldu] U I d U
[ql@b] q I b [qlbu] q b U

MA is much more liberal than Tashlhiyt in the range of violations of

SonPeak which are allowed to syllable onsets. The reader may recall that

24 On the derivation of this form, v. § 8.5.1.

25 On this inability, see § 6.3.2. and § 8.2.2. Parenthesized '@' will be omitted from such
sequences in the forms cited below.
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 311

in Imdlawn Tashlhiyt the only onsets which are found to violate SonPeak
are occurrences of w preceding a nucleus which is a coronal consonant or
fricative," e.g. in the medial syllable in t. wn.za, the bound form of t-a-wnza
'fringe of hair' . In MA, on the other hand, both y and w can be onsets in
violation of SonPeak, and there are no restrictions on the consonants which
may be nuclei after such onsets, see e.g. the nuclear d which follows onset
y in !byed ((24)b). MA's greater tolerance of onset glides violating SonPeak
is merely a particular instance of its greater tolerance of SonPeak violations,
which is due, as has been explained earlier, to the ranking of SonPeak below
FinH and FinL.
As in Tashlhiyt, hollow syllables with a glide in onset position give
the impression of beginning with an opening diphthong. Consider for
instance Iwdn-inl 'ears', for which our analysis predicts the terminal rep-
resentation wd.nin, with d as the nucleus in the first syllable. This word
sounds like [wudnin). Its first syllable does not meet the conditions of
IP-Final Epenthesis (13) . According to our analysis, the vocoid which is
heard in . wd. when the articulators move from w to d is not the manifes-
tation of a segment, but a mere tran sition. Later we will encounter short
vocoids adjacent to glides which cannot be construed in this fashion, but
before we turn to such cases, let us review a few more in which the glides
behave as our analysis leads one to expect.
When a geminate glide follows a consonant its first half is realized as
the corresponding high vowel." On each line in the examples below, the
form on the right contains a medial geminate:
(26) a. herz ' amulet' herrez 'rnake amulets'
b. zeld 'skin' zelled 'cover with a skin'
c. xuf 'fear' xuwef 'frighten'
d gid 'fetter' giy@d ' put fetters on'
We give below in (27) the representations of Ixwwfl x uwef 'he frightened'
and Ixwwf-ul xuwf-u 'they frightened', prior to the operation of IP-Final
Epenthesis. To illustrate the parallelism with the other CeC:eC verbs , we
add underneath the surface forms of Izlldl zelled 'he covered with a skin'
and Izlld-ul zelld-u 'they covered with a skin'.

26 See§7.3.1.
27 For similar facts in Imdlawn Tashlhiyt, see § 7.4. As in Tashlhiyt, a crucial factor in
the syllabification of geminates is the constraint NoOns-, which forbids the first half of a
geminate to be an onset. On the role of No Öns- in Tashlhiyt, v. § 4.8. The role of this
constraint in MA will be taken up later, see the text below (46) in § 9.4.3.
312 CHAPTER N I N E

(27) xuweJ, ielled

o o
»<.R
0
->.R
0
I I
N
N
I I
x x x x
[xuw@f]
I
x
<>
u
I
f
[zl:@d] z d

xuwf-u, ielld-u
o o

0
-<:R 0
»<.R
N
»<.D I
N
I I I
x x x x x
[xu:fu] x
I <>
U
I
f
I
U
[zl:du] Z d U

As a last case in which our analysis doe s not require any additional
machinery to accord with the data, consider the CeCC forms with medial
glides . We saw earlier that ICCCI kernel s surface as CCeC when they are
assoc iated with template FinL , and as CeCC when they are associ ated
with template FinH. When the medial consonant is a glide, (lCCCI, FinL)
still surfaces as CCeC, as illustrated by !byed in (24)b . As for (lCCCI, FinH)
form s, our analy sis predicts that they should be phonetically indistin-
guishable from forms derived from ICVCI with a medial high vowel, e.g.
(lxwf/, FinH) 'fear ' surfaces as xuf, «26)c) for the same reason as (/zld/ ,
FinH) surfaces as i eld «26)b). The surface forms of xuf and ield are dis-
played below in (28).
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 313

(28) xuJ, leid

o
-<.R
0

N
-<:D
I I
x x x
I I I
[xuf] x U f
[zld] Z d

Like Tashlhiyt, MA has a process of u fronting which affects u but not

wand operates in similar contexts." This process regularly applies to the
occurrences of u which are surface reflexes of Iw/, e.g. those in /ksw-t=ul
ksutu 'his suit' and Ishw-t=ul shutu 'his appetite' . 29 The actual pronunci-
ations of these forms are [ksötu] and [shötu].

9.3.3.2. @w diphthongs; NoRR violations

Our analysis predicts that glides which are syllabified as nuc1ei should be
realized as high vowels. In Lmnabha MA the prediction is borne out in most
cases, as illustrated in the above discussion, but not in all. Nuc1ear Iwl is
in some contexts realized as the diphthong @w. Consider for instance
Iwsws-ul 'they mislead'r" This form can be pronounced either us@ws-u
or usus-u, the latter variant characteristic of a faster tempo. usus-u is
homophonous with u#sus-u 'and let them shake down!'." In either expres-
sion the second syllable contains an occurrence of the full vowel u.
In us@ws-u, on the other hand, the second syllable contains a closing
diphthong. That diphthong cannot be construed as an instance of uw:
us@ws-u is not homophonous with lu#swws-ul u#suwsu 'and they are
worm-eaten', an expression in which the second syllable contains a steady-
state long u (Iusu.sul)." We give other examples below. The forms on the
right are derived from those on the left by adding the fs suffix I-al or the
3ms possessive c1itic I=ul ('cross-eyed, f", 'his time-table', etc.):

28 On u fronting in TashIhiyt , v. § 3.8. On that process in MA, see Elmedlaoui (1995a :

222-227) . u fronting also applies in the Moroccan pronunciation of C1assical Arabic.
29 The uncliticized forms are kesw-a and sehw-a. The suffix I-al becomes I-tl before a
possessive c1itic.
30 Cf. Iwswsl [usw@s] 'he mislead' .
31 Cf. sas ' he shook down', sus ' shake down!'.
32 Cf. Iswwsl [suw@s] 'it is worm-eaten' .
314 CHAPTER NI N E

(29)
a. /hwl/ hwel 'cross-eyed' /hwl-a/ h @w.la / hula
b. /f{wz/ f{wez 'twisted' /f{wz-a/ f{@w.za / f{uZa
c. I!f{wr/ !f{wer 'one-eyed' /!f{wr-a/ !f{ @w.ra / !f{ura
d. /zdwl/ zedwel 'time-table' /zdwl=u/ [email protected]
e. /mzwd/ mezwed 'ko bag ' /mzwdeu/ [email protected]
In some forms, e.g . in (29)a-c, nuclear /w/ can be realized either as
@W or as u depending on speech rate , but there are others in which it can
only be realized as @w, v. examples (29)d ,e. We cannot go into the details
about the contexts which allow a nucle ar glide to surface as a diphthong.
Suffice it to say that in Lmnabha MA this can only happen with /w/, and
that the diphthong @w cannot occur at the beginning or at the end of a word .
Once it is assumed that [u] and [w] are but two manifestations of the
same bundle of dist inctive features, one is forced to construe the schwa
in @w diphthongs as a segment in its own right, and not as a mere tran-
sition between segments.
The nuclear diphthong s of Lmnabha MA are the same thing as the
' syllabic semivowels' which Heath (1987) has reported in Fes/Meknes MA .
In that dialect, accord ing to Heath, underlying glide s which occur as syllable
nucle i have two types of realization s depending on context. Some undergo
vocalization and become phonetically indistinguishable from full vowel s,
whereas others surface as segments which he calls 'syllabic semivowels'
(v. pp . 238 , 269 , 288) . According to the author the syllabic semivowels
do not show the same allophonic vari ations as the full vowels (e .g. in
' emphatic' contexts), and the transiti on between them and the preceding
con son ant 'may resemble a faint schwa ' (p. 269) . As we have ju st seen,
Heath 's tran sition is actually a segment.
The representation s of the medial syllables in the free varia nts
us@ws-u and usus-u are displayed below in (30)a,a'.

(30)
a. [s@w] a'. [su] b. [t@h] b'. [th]
o o o o
-<:R
0
~
0 R
-<:R
0
-<:R
0
I I I I
N N N N

X X
-<:X X
I
X X X
~
X X X
I
I I I I I I I I I I
s e U s U t e h t h
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 315

The displays on the right-hand side are given for the sake of comparison.
They represent the variant pronunciations of the last syllable in taten in
(10) ; they are reproduced from (9).
The pairs in (30) represent the only two c1asses of cases so far in which
Lmnabha MA has a surface contrast which requires positing a segment
schwa." The two categories of contrast differ in their context of occur-
rence and in the feature content of the nuc1ear segment. The displays on
the right-hand side illustrate a contrast which only occurs in word-final
syllables whose nuc1ear consonant is an obstruent. (30)b and (30)b' are
contextual variants; the choice between them depends on phrasing, as
explained in § 9.2. The displays on the left -hand side in (30) illustrate a
contrast which only occurs in nonfinal syllables with U in their nuc1eus. Not
all words in which @w occurs have an alternative variant with u.
The distinction between simple and complex nuc1ei illustrated on the
right-hand side of (30) is needed for heavy syllables as well as for light
ones. On the other hand the distinction between @w and u seems to be
limited to light syllables; [CuC] syllables are easy to come by, but we
have not encountered any syllable of the shape [C@wC]. In view of this
gap, one might wonder whether [C@w] syllables are indeed light, as implied
by the representation in (30)a. That they are is shown by their behaviour
in singing. We give below one line of poetry coined by ME on the meter
of (30). In that line the second syllable of the word [email protected] (v. (29)e)
occupies a L position and the line is well-formed. For the sake of com-
parison we give another line in which the same L position is occupied by
the first syllable in /mzwd/ mezwed (v. (29)e), whose status as a light syllable
is not in doubt:
(31) H L L L L L H H L
a. gim t#m z@w du ya leh bab bay na
b. gim t#l mez wed ya leh bab bay na
The text of these lines is given in (32) and their meaning in (33) :
(32) a. gim-t mzewdeu, ya le=nbab, bayn-a
b. gim-t lernezwed, ya leehbab, bayn -a
(33) a. the value of his bag, 0 (my) friends, is obvious
b. the value of the bag, 0 (my) friends, is obvious
Before leaving the glides we must dwell brieflyon violations of con-
straint NoRR, which disallows rimes in which the coda has a higher sonority
than the nuc1eus, v. § 8.5.3. We saw in § 7.3.3 that Imdlawn Tashlhiyt allows
hollow rimes with a [+son, +cons] segment in the nuc1eus and w in the coda.

33 A third case will be discussed in section § 9.5.

316 CHAPTER NINE

MA has similar violations of constraint NoRR, in which the offending

coda glide can be y as weil as w.
Consider for instance the participle /m-drws/ mderwes 'feeling dizzy' and
its fs form /m -drws-a/. In view of what precedes one might expect the
pronunciation of /m-drws-a/ to be med.ru.sa or [email protected]. This form must
actually be realized as m.drw.sa, with ras the nucleus of its second syllable.
The facts about the pronunciation of syllables such as .dj w. are the same
as in Tashlhiyt, see § 7.3.3 . The vocoid which follows r differs from
realizations of lul and Iwwl in the same environment. The form m.drw.sa
contrasts with Im-drus-a/ med.ru.sa, the fs participle of dres 'thresh', whose
kernet has the underlying shape ICCuCI, as in the case of any other ICCCI
verb. The vocoid which follows r in m.drw.sa is furthermore different
from that in /t-rwwn-u/ t.ruw.nu 'that you (p) mess up' . We give below other
exam ples with .CCw. syllables in Lmnabha MA. In these examples the forms
on the right are derived from those on the left by adding the 3p desinence
I-ul or the 3ms possessive clitic I=ul ('they dried suddenly', 'his make-up
pencil', etc .)."
(34)
a. Islwnl seiweh 'he dried suddenly' Islwn-ul .slw.nu.
b. Imrwd/ merwed 'make-up pencil' Imrwd=ul .rnrw.du .
c. I!frytl !feryet 'he scrapped' I!fryt-ul .fry.tu .
d. Iknw-a/ kenwa 'nickname' Iknw-t=ul .knw.tu.
e. Iqlw-a/ qelwa 'testicle' /qlw-teu/ .qlw.tu.
Standard transcriptions of the forms on the right-hand side would be selwnu,
merwdu and so on.
In Lmnabha MA, then, glides are licit codas in hollow syllables, provided
the preceding consonant is a sonorant. When Iwl occurs in forms analo-
gous to those in the right-hand side of (34), but with an obstruent preceding
it, it must belong to a nucleus. This nucleus is @w in some cases, u in others,
and @w in free variation with U yet in others. The complementary distri-
bution between coda wand @w is for instance found in most ICCwCI
kernels, compare (34)a,b, in which Iwl follows a sonorant, with (29)d,e,
in which Iwl follows an obstruent. As for that between coda wand u, it
occurs for instance in the possessed forms of the ICCw-a/ nouns: w is a coda
in knw.tu 'his nickname' and other similar forms in which it is preceded
by a sonorant (v. (34)d,e), whereas it surfaces as a full vowel when the
preceding consonant is an obstruent, as in Iskw-t=ul skutu 'his goatskin
bottle' or /nsw-teu' nsutu 'his addiction' .35 Finally, for a case in which coda
w is in complementary distribution with u and @w in free variation, compare

34 The fs suffix I-al becomes I-tl before a possessive c1itic.

35 The uncliticized forms are sekw-a, nesw-a.
 VOWELLESS S YLLABLES IN MOROCCAN ARABIC 317

(34)a,b with the already cited Iwsws-ul ' they mislead', which can be
pronounced either usus-u or us@ws-u.
MA shows much dialectal variation in the pronunciation of syllable nuclei
which contain an underlying glide. The facts presented above are those of
the variety of Southern MA spoken in Lmnabha. As explained earlier,
Southern MA requires complex nuclei to occur in a narrower range of
contexts than other dialects. The dialect of Oujda lies near the other end
of the spectrum of variation, as far as complex nuclei are concerned. In
that dialect all the forms on the right-hand side of (29) must be pronounced
with a diphthong in their penultimate syllable, e.g . in (29)a /hwl-a/ must
be realized as n@w!a, and the same is true of many forms such as
Iksw-t=ul 'his suit', in which Iwl is realized as u in Lmnabha. The bedouin
dialects spoken around Oujda allow an even wider distribution of complex
nuclei. Whereas the diphthong @y does not exist in the dialect of most
people living in the city proper, it does in its periphery, where I!byd-al
'white, fs', can be pronounced Ib@yda (cf. (24)d) . Speakers living in the
outskirts of Oujda and the surrounding countryside also pronounce a
diphthong for the first half of a geminate glide which occurs after a con-
sonant, e.g. they pronounce Iswwll 'he interrogated' and Izyyrl 'he
whitewashed' as s@ww@! and z@yy@r, whereas in the city itself these
words are pronounced suw@! and ziy@r, as they are in Lmnabha.

9.4. RELEASES IN SEQUENCES OF SlBLING CONSONANTS

In § 6.3.3.1 we agreed to say that two segments are siblings when they
have the same primary articulation and the same values for the features
[sonorant] and [continuant] . Examples of sibling sequences are t+d and t+t.
In MA some underlying sequences of siblings give rise to surface geminates
while others do not, hence surface contrasts in which the distinguishing
feature is the release of an oral closure. A consonant release is apt to be
mis-analyzed as a vestigial vowel, especially when it occurs before a sibling
consonant. Consider for instance the sequence t+t in a word in which the
first t may be pronounced with an oral release. At the end of the 'hold'
period of the first t the tongue is already in pIace for the closure of the
following consonant. Why should the tongue not simply maintain a fixed
posture throughout the articulation of the whole sequence? Could the release
not be due to the presence of an intervening phonetic target specifying a
vocoid? This line of thought is particularly tempting when the released
consonant is an onset: the following vocoid would then be a nucleus .
We shall see, however, that syllable structure only impinges in a rather
indirect way on the distribution of releases in sibling sequences.
318 CHAPTER NINE

9.4.1. Fusion and NO-TREBLE

In § 6.3.3 we showed that Tash1hiyt has a process of fusion whereby two

adjacent sibling consonants merge their primary articu1ations. A prelimi-
nary survey of consonant sequences suggests that MA also possesses a
fusion process. This surmise is based on the foIIowing observation: except
in special cases to be described later, when two short sibling consonants
occur in a sequence the first must not be re1eased and the sequence is
homophonous with the corresponding geminate. Here are examples with
such sequences (the forms in the second column are given for the sake of
comparison).
(35) a. fertt-u 'they picked off' fertet 'he picked off'
b. mqet-ti 'you (f) loathed' mqet 'he loathed'
c. smenena 'our butter' smen 'butter'
d. selk=kum 'your (p) wire' se1k 'wire'
e. ka t-teb'i-u 'you (p) foIIow' tbe? 'he foIIowed'
f. ka !d-dekr-u 'you (p) mention' ldk''er 'he mentioned'
g. berret-teu 'I cooled it (m)' berred 'he cooled'
The highlighted sequences in the forms in the first column must aII be
pronounced as though they were geminates. The sequences It-d/ and Id-t/
in lines fand g, which must respectively be pronounced as [d:] and [t],
exemplify regressive voicing assimilation between sibling stops. We do
not know whether the domain of fusion in MA is the word or some larger
unit, for we have not investigated analogous sequences straddling a word
boundary.
Let us assume that MA has a fusion process identical with that which
we have posited in Tashlhiyt. We reproduce below the rule which we gave
in § 6.3.3 .2.

(36) FUSION R~ R~ Rtj R~

I
Cpl
I
Cpl
I
Cpl
I
Cpl
I
IX
I
IX
<> IX

(36) does two things at once. It states that the configuration to the left of
the arrow must be avoided, and it specifies how that configuration is to
be modified to give rise to a well-formed output. If (36) is to be inte-
grated into our account of MA, which is constraint-based, it must be replaced
by a whole complex of constraints designed to prohibit the configuration
to the left of the arrow in (36) and to guarantee that infringements are
avoided by merging the offending adjacent primary articulations. Let us
simply assume that this can be done and use the term Fusion to refer to
the contraint complex in question.
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 319

As in Tashlhiyt, in MA three skeletal positions in a row cannot share

the same primary articulation. When sibling consonants are linked with three
successive skeletal slots, their 'hold' periods need not merge, i.e. one sibling
consonant can be released before the constriction of the next is effected.
Here are examples in which this phenomenon is observed. As in Chapter
6, 'z, indicates that the oral constriction of the preceding consonant is
released, i.e. it notes a voiceless burst ([h]) between voiceless stops and a
short voiced vocoid ([@]) between voiced ones." The parenthesized form
at the end of each line is the bare kernel of the preceding word, i.e. a singular
noun or a 3ms perfective verb .
(37) a. sekkz=kum 'your (p) doubt' (sekk)
a'. sekk'k-u 'they caused to doubt' (sekkek)
b. ldenn/ena 'our idea' (!denn)
b' . zennzn-u 'they annoyed' (zennen)
c. nelf=l=u 'he opened for hirn' (nell)
c'. neUzl=u 'he made it (m) licit' (hellel)
d. gettZ-t=u 'his stack' (gett-a)
d' . getrteu 'he stacked it (m)' (gettet)
As in analogous cases in Tashlhiyt, each example has an alternative
pronunciation in which the two closures blend into a single uninterrupted
one. In (37)a,a', for instance, the sequence noted kICk may be pronounced
either as [k:hk] or as [k::], where the doubling of the length mark indi-
cates an extralong 'hold' period distinct from the long 'hold' period of a
geminate. Unless stated otherwise, release is also optional in all the
examples to come which contain an occurrence of the symbol ,z'.
As we did for analogous sequences in Tashlhiyt, we assurne that both
pronunciations are realizations of a single terminal representation. That of
sekk=kum «37)a) is given in (38) below.
(38) o c

0
»<.R 0
-<:R
-<. ~
N D N D
I I I I
X X X X X X
I
s
<.>
k k
I I
u
I
m

36 Except for the replacement of some occurences of e by the release symbol 2, the tran-
scriptions in (37) and others below are standard transcriptions.
320 CHAPTER NINE

In (38) the realization of /kkek/ as kICk ([k:hk]) is but a particular instance

of the fact that in MA as in Tashlhiyt, the closure of a stop is released before
that of a following stop is formed . The alternative realisation kkk ([k::])
is presumably due to a principle akin to MINIMAL-PATH(place), whereby
an articulator must follow the shortest possible path when moving from
one phonetic target to the next. 37
As in Tashlhiyt, Fusion is overriden by NO-TREBLE, an undominated
constraint which prevents a primary articulation from being shared by
more than two adjacent skeletal positions ." Fusion and NO-TREBLE jointly
make the prediction that a consonant can be released before a sibling only
if one of the consonants is a geminate. This prediction allows for instance
the surface sequences t'u, trt and tru, but not ri. in which both stops are
short. The prediction is borne out by the data, with important qualifica-
tions to be discussed later.

9.4.2. Earlier views on releases in sibling sequences

Releases between sibling consonants have not been satisfactorily dealt

with in the previous literature on MA. It is worthy of notice that our analysis
makes it possible to clarify the confusion surrounding them. In a nutshell ,
there is no difference between releases in sibling sequences and other
releases. The seemingly special status of releases in sibling sequences is
only an artifact of alphabetic transcriptions in which gemination is indicated
by letter doubling.
Unlike releases between nonsiblings, releases between siblings cannot be
omitted from an adequate transcription of MA, however 'broad' , because
they give rise to surface contrasts of a particular kind. Consider the fol-
lowing pair.
(39) a. fertet -t=u [frt't.u] 'I picked it (m) off'
b. sett 2t=u [sthtu] 'he scattered it (m)'
In (39)a an underlying sequence of three occurrences of It/ is realized as
a short t followed by a long one. In (39)b an underlying sequence of a
geminate Ittl and a simple Itl is realized as a long t followed by a short
one. The feature which distinguishes [tht] from [tht] is the timing of the
medial release. Clearly, it would not do simply to transcribe both sequences
as 'ttt'. Whereas some previous analyses interpret the release in (39)a as
a regular occurrence of schwa, they must introduce additional stipulations
to characterize the release in (39)b.
Our own account of the contrast in (39) is implicit in the discussion in

37 On MINIMAL-PATH(place), § 6.3.1.
38 See § 6.3.3.
 VOW ELLESS SYLLABLES IN MOROCCAN AR ABIC 321

the preceding sections. The lexical representations of the two verb s in (39)
are given in (40) below .

(40) a. /frt+t/ b. /stt+t/

X X X X X X X X
I I I I I V I
f r t t s t t

In the line at the top of (40) the sequences between slashes are no more than
typographically convenient stand-ins for the plurilinear objects underneath,
which are the lexical repre sentations in question. In trans criptions noting
underlying repre sentations we follow the convention introduced in § 6.4.1
when we discussed analogous structures in Tashlhiyt: Inside amorpheme,
adjac ent occurrences of the same symbol indicate a gemin ate; an intervening
'+' between two occurrences of the same symbol ind icates that each
occurrence represent s a distinct feature bundle.
The verb fertet in (40) a was already mentioned at the end of sec tion
§ 8.5.1 , see (71) . Its lexical representation end s with two occurrences of
the feature bundle characterizing t. settet in (40)b is a four-consonant verb
with a medial gem inate in which the last two consonants are identical. 39 The
surface representations of the two forms in (39) are given below in
(41) -1. The material below line I in (41) will become relevant later. In
(41) the skeletal positions linked with t have been repre sented as digits to
make reference easier. In (41) and below, the right edge of the kernel is
indicated by a right bracket when it is rele vant in the discus sion.

(41) a. fertet-teu «39)a) and sket-tu «42)a)

o o c
-<.R
0
.>.R
0
-<.R
0
I I I
N N N
I I I
X X 2 ] 3 X
I
I
f
I
r
<>
t
I
u
11 s k u

Figure (41) continued overleaf

39 Other such verbs are sekkek, iennen, heltel and gettet in (37). See Harrell (l962b: 31)
für others.
322 CHAPTER NINE

(41) b. sett 2t=u ((39)b) and sekkt=u ((42)b)

o
.r-:
o

o R o
->.R
N
-<.D I
N
I I I
x I 2 3 X

I
I
s <> t
I
t
I
u
11 s k u

Release can only occur at the end of a 'hold' period. Each occurrence of
t in (41) has an associated 'hold' period. Limiting our attention to the
transition between positions land 2 and to that between positions 2 and
3, we see that in (41)a, release can only occur between land 2, while in
(41)b it can only occur between 2 and 3.40
The pair in (39) is in all relevant respects the exact parallel of that in
(42) below, which does not contain sibling consonants:
(42) a. sket-tu [sk't.u] 'you (p) fell silent'
b. sekkt=u [skr'tu] 'he shut hirn Up '41
The surface representations of the two forms in (42) are those resulting from
the association of the feature bundles in line 11 in (41) with the tree struc-
tures displayed there above line I. Whatever devices in the phonetic
implementation component of the grammar of MA are sufficient to account
for the occurrence of releases in the consonant sequences in (42), are also
sufficient to account for those in (39). What makes pair (39) seem special
is that unlike (42) it poses a problem for alphabetic transcriptions, which
must represent phonological structure as a unilinear sequence. Let us briefly
go over the views of HarreIl (1962b) and Heath (1987) about releases
between siblings.
HarreIl (1962b) kept to a strictly unilinear conception of phonological
structure which was commonly held at the time of his writing, and he
managed to do so without resorting to a special symbol for consonant
releases, but this was at the price of misconstruing syllable structure in some
instances.

40 In (41)a Fusion has rnerged the stern-final Itl with the following I-tl suffix. Merger of
the stern-final Itl with the preceding Itl would have created a violation of constraint NoOns-,
see below (47) and (48).
41 sekket 'cause to fall silent' is a causative derived frorn sket 'fall silent' .
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 323

The consonant which is released in (4l)a is an onset, as indieated by

the following 'e' in the standard transcriptions, and previous analyses in the
literature, e.g. Keegan (1986), would interpret the release as a realization
of schwa. In (41 )b, on the other hand, the released consonant is a coda,
and yet HarreIl interprets the releases in such cases as occurrences of e.
On p. 18, where he discusses alternations between e and zero, e.g. ! sifet
'he sent' vs. !sift=u 'he sent him' , he presents these alternations as
motivated by the need to prevent schwas from occurring in open sylla-
bles . He then notes the existence of forms such as (39)b and states that
they represent the only cases in which the unstable vowel may occur in
an open syllable. Harrell's example is /smm-rn-u/ ' they poisoned', cf.
semmem [sm:@m] ' he poisoned'. According to our analysis /smm-rn-u/ has
a surface representation .smm.mu. analogous to (41)b. .smm.mu . may be
realized either as smm'mu [sm:@mu] or as smmmu [sm ::u], which Harrell
transcribes respectively as ' semmemu' and 'semmmu' .
In the sequence 'mmern' in Harrell's first transcription, 'e' notes the
oral release of a consonant. Harrell mistakenly interpreted that release as
the manifestation of a vowel. Indeed, noncontinuants must be released before
a vowel, but they are also released in other contexts, e.g . before a het-
erorganie consonant, as mm is before n in hemm=na [hm:@na] 'our worry'.
/hmm=na/ and /smm-rn-u/ have identical syllable structures: .hmm .na.,
.smm.mu. The reason why Harrell's transcription records mm's release in
.smm.mu. but not in .hmm.na. is that in .hmm.na. this release is predictable
on the sole basis of the graphie sequence 'mmn', whereas in .smm.mu.
the graphie sequence ' mmm' is ambiguous; it can correspond to either of
the surface structures displayed in (43) .
(43) a. X X X b. X X X

m
I Vm Vm m
I
In (43)a a release can occur after the first skeletal position but not after
the second, whereas in (43)b it can occur after the second position but
not after the first.
Heath (1987) devotes special attention to releases in sequences of sibling
consonants. In the surface representations of MA, according to Heath, some
ce sequences are tightly-knit while others are not (p. 219) . What Heath
calls secondary gemination, i.e. Fusion in our terms, is a special case of
tight clustering. The distinction between tight ce clusters and loose ones
is neutralized when the two consonants are not siblings (p. 281). In loose
sequences of siblings, the two consonants 'retain separate articulations
and releases' (p. 231) . Heath calls this phenomenon 'hiatus' and uses a
special symbol (") to record it in his transcriptions of surface forms. Heath's
hiatus marker corresponds to our symbol ,2,. In Heath's view hiatus is
A

sometimes the trace of a syncopated schwa (pp. 219, 232). Although Heath
324 CHAPTER NINE

plainly sees that there is a close similarity between hiatus and schwa, he
never explicates the exact connection between them. The reason for this,
we believe, is that in Heath's analysis schwa is a segment while hiatus is
not; hiatus is merely a phonetic cue for the edge of a tight ce cluster.
To repeat our own position, hiatus is just a special name given to con-
sonant release when it occurs between siblings. The releases in sibling
sequences are subject to the same restrictions as other releases, with the
additional restrictions imposed by Fusion and NO-TREBLE: whereas a short
stop must be released before a short heterorganic stop, the joint enforce-
ment of Fusion and NO-TREBLE guarantees that in most cases a sequence
of two short siblings surfaces as a geminate, and consequently the first
sibling cannot be released.
In our view, release before a sibling consonant is no more represented
in terminal representations than release before a nonsibling, compare for
instance the terminal representations of setrt=u 'he scattered it (m)' and
sekkt=u 'he shut hirn up', which are both displayed in (41)b . The main
reason why release before a sibling has received more attention than other
releases, is that it poses a special problem to alphabetic transcriptions in
which geminates are represented with doubled letters."

9.4.3. Releases between short sibling stops

Until now syllable structure has not played any role in our discussion of
releases in sibling sequences. We have given full syllabic representations
for some of our examples, see (38) and (41), but in these representations
all the information relevant for locating possible releases is contained in the
two lines at the bottom, which depict the mapping between the feature
bundles and the skeletal positions. The rule of thumb for 'reading off'
consonant releases from the lower part of these diagrams is a very simple
one: arelease can occur 'after' a skeletal position whenever the feature
bundle associated with that skeletal position is not also associated with a
following skeletal position.
In some of our examples in section § 9.4.2 the consonant released before

42 The machinery which Heath sets up to deal with hiatus is also put to use in his account
of the length alternation in the passive prefix (p. 280 ff.). Lack of space prevents us from
going into the details of Heath's discussion . We simply suggest an alternative account based
on syllable structure .
In the dialect described by Heath the passive prefix is a simplex t in some forms and a
geminate tt in others . It is for instance long in tte-qtel 'he was killed' and short in t-qetl-u
'they were killed '. In terms of our own account of the syllable structure of MA, it seems
that the surface form of the passive prefix must meet the following requirement: a skeletal
position belonging to the prefix must be included in a rime . The prefix takes on its long
form only if its short form cannot meet this requirement. The forms cited above meet the
requirement in question . The syllabic parse of t-qetl-u is .t.qt.lu. and that of tte-qtel is .t.tq.tl.
 VOW ELL ESS SYLLABL ES I N MOROC CA N ARABIC 325

a sibling is an onset, see (41)a, but in others it is a coda , see (38) and
(41)b. The latter fact must be faced by anyone toying with the idea that
in MA the relea se of a consonant before a sibling is the realization of a
vowel.
As already stated before, Fusion and NO-TREBLE jointly make the
prediction that in a sibling sequence where release occurs, one of the
consonants must be a geminate. There are however sequence s in which both
consonants are short, witness the following example .
(44) fertet-na 'we picked off'
The lexical repre sentation of Jert et 'pluck', which was given in (40)a, is
reproduced below in (45)a. The terminal representation of [ertet-na is
displayed in (45)b.

(45) a. Ifrt+tJ b. fertetna

o o o

0
~
R
->.R
0
~
0 R
I I I
N N N

X X X X
I I I
X X X X X X
I I I I I I I I I I
f r t t f r t t n a

The medial syllable in (45)b contains two occurrences of the feature bundle
characteristic of t. The closure corresponding to the first t can be released
before the closure corresponding to the second t is effected. Why has Fusion
not merged the two occurrences of ItJ in the lexical representation (45)a into
a geminate IttJ?
Before we can answer this question we must pause briefly to examine
the role of geminate inseparability in an analysis with vowelles s syllables.
In our earlier account , the special behaviour of geminate s with respect
to syllabification was due to geminate inseparability, which forbids inserting
material between the two halve s of a geminate, see the text around (69)
in section § 8.5.1. Once we adopt an analysis in which syllables can be
vowelless, geminate inseparability leave s cert ain properties of the MA
geminates une xplained. Con sider !denn ' he thought' , whose underlying
representation is I !dnn/. This form must be pronounced [!d(@)n:]. Geminate
inseparability may be the reason why [!dn@n] is not a well-formed real-
ization of I!dnn/ , but it does not expl ain why [!d(@)n:] must count as a
heavy syllable (.dnn.) and not as a sequence of two light ones instead (d.nn).
The two syllabic parses are represented in (46)a,b below. For the sake of
comp arison we have added the terminal representation of !dmen 'he guar-
326 C HAPTE R NI NE

anteed', a three-consonant verb whose last two skeletal position are not
occupied by a geminate .

(46) a. .dnn. b. *.d.nn. c. .d.mn.

.r:
0" 0" 0" 0" 0"

0 R
I
R
.>.R
0
I
R 0
»<R
N
->.D I I
N
I I
N
N N
I I I I I I
X X X X X X X X X

d
I <>
n
I
d
<>
n
I
d m
I I
n

The device which exclude s (46)b is the constraint NoOns-, which forbid s
the fir st half of a gem inat e to be an onset. This constraint was already
invoked for Tashlhiyt, see § 4.8. It is given again below.

(47) NoOns- : prohibit o (0 = onset

I Rt = Root node)
X X
V
Rt

The role played by NoOns- in our new account is analogou s to that played
by gemin ate inseparability in our earlier analysis. We assume that NoOn s-
is undominated and that it is encapsulated in the analogue of SYLL in our
new analy sis, see condition c in SYLL (58).
We can now go back to the derivation of Jertet-na ((44» , which is well-
formed , in spite of the fact that it violates Fus ion (36) , see the surface
representation of fe rtet-na in (45)b.
We submit that the viol ation of Fusion in (45)b is a lesser evil: the
other viable candidates violate constraints which are more highl y ranked
than Fusion . We propose the rankings displayed in (48):

C
(48) Noo ns- (47)
FinL (98) - - - ,
Fusion (36) ------'

Recall that four-con sonant verbs are associated with template FinL (see
§ 8.5.5). Except for (45)b , the viable candidates for the input (lfrt+t-na/,
FinL) all violate NoOns- or FinL, see (49) below.
 VOWELLESS SYLLABL ES IN MOROC CAN ARABIC 327

(49) a. *ferttna
o o o
-<.R
0
»<.R
0
~
0 R
I I I
N N N
I I I
x x x x x x
I
f
I
r
<:: I
n
I
a

b. *frettna
o o o
I
R 0
-<:R ~
0 R
I
N N
-<.D I
N
I I I I
x x x x x x
f
I
r
I <>
t
I
n
I
a

(49)a violates NoOns- and (49)b violates FinL, which are more highly
ranked than Fusion, while (45)b does not violate either constraint and is
in consequence the grammatical output.
Fusion is violated only if the violation allow s the output to comply
with FinL, witne ss the pronun ciation of /frt-t-u/ 'they plucked', in which
the first Itl may not be released (see (35) a). All viable candidates for
/frt+t-u/ must violate finL, because the stern-final /tl rnust be an onset before
the follow ing vowel.
We summarize this discussion with the two tableaux in (50).
(50) /frt+tlLna/ NoOn s- FinL Fusion
.fr.ttj .na. (49)a *!
.f.rtt1.na. (49)b *!
~ .fr,t2t 1.na. (45)b *

Ifrt+tl Lul
~ .frt.tju, (35)a *
.frt't]u. * *!
We have encountered two classes of words in which a short consonant
may be rele ased before a short sibling. All the word s in one class are
verbal forms likeJertet andJertet-na, i.e. the kernel is word-final or followed
328 CHAPTER NI NE

by a suffix or clitic which begins with a consonant, and it ends with two
identical short consonants ." In the other dass of words, release occurs
between aprefix and the stern . Here are examples, all borrowed from
Heath (1987: 283 ff.):
(51) a. e-dwaz 'letting pass' (dewwez)
b. !e-twal 'lengthening' (!tewwel)
c. sr-tna 'he excluded"
d. ma lt2-dreb=s 'do not hit! ' ( ldreb)
We do not have an account for the cases in this second dass. The data is
complicated by the fact that some of the cases allow free variation, e.g.
(5l)d has another variant in which Fusion has occurred (ma lddrebs) ,
whereas Fusion is excluded in the other examples in (51). The following
fact suggests that syllable structure is also at play in this second dass of
words with release between short siblings: in all of these words the second
sibling immediately precedes an onset, and consequently the first sibling
is in the appropriate environment for being syllabified as an onset. As for
the fact that Fusion is enforced less severely at the juncture between prefix
and stern than at other locations in the word, the reader is referred to
§ 6.3.3 .3, where we noted that a similar situation prevailed in Tashlhiyt.

9.5. STABLE SCHWAS

In § 9.2 the pronunciation of certain IP-final syllables forced us to retain

e as a segment, albeit with a limited distribution. As we shall now see ,
certain plural nouns of the form CCeC pose an even more serious chal-
lenge to a schwa-Iess analysis of MA, for they contain syllables in which
schwa is the sole content of the nucleus of an open syllable. Ironically
enough, these plural nouns also turn out to raise a serious problem for an
analysis in which all hollow syllables would contain schwas.
Certain feminine nouns of the form CCC-a have plural forms with the
shape CCeC. Here are examples.
(52) singular plural
a. gerb-a greb 'goatskin for water '
b. sekk-a skek 'plowshare'
c. sebk-a sbek 'net'
d. sell-a sIel 'basket'
e. kff-a kfef 'pan (of scales)'

43 Other such verbs are !gertet 'cut off', f eynen 'hum', hernen 'grumble ' .
44 A classicism derived from CA sta-Bnaa.
 VOWELLESS SYLLABLES IN MOROCCA N ARABIC 329

When a clitic is appended to a CCeC form which is the plural of a

CCC-a noun , the schwa in that form does not behave as that of other
CCeC forms . Recall that when a clitic beginning with a vowel is appended
to a CCeC word, the word is realized as CeCC, i.e. its second consonant
becomes nuclear, e.g . l Yreg 'sweat' yields !ferg=u ' his sweat', with a
nuclear r. This does not happen with the CCeC plurals of CCC-a nouns .
greb 'goatskins' ((52)a) yields greb=u 'his goatskins' (*gerbu), skek 'plow-
shares' yields skek=u 'his plowshares' (*sekku) .
In the Lmnabha dialect, most of the CCC-a feminine nouns with CCeC
plurals which we have been able to find are like forms (52)b,d,e: the last
two consonantal positions form a geminate which is split by e in the plural
form. Plural s like greb ((52)a) and sbek ((52)c) are important in that they
show that the special behaviour of schwa in skek and the like has nothing
to do with the fact that schwa is sandwiched between identical consonants."
To present the other property of schwa which is special to these plural
forms, let us temporarily revert to the parlance of the old analysis, in
which every hollow syllable contains an occurrence of e. When it occurs
in plural forms like those in (52), schwa cannot devoice between voice-
less consonants, nor can it be absorbed by an adjacent sonorant." For
devoicing, let us compare cliticized forms of skek 'plowshares' and anal-
ogous forms of sekkek 'cause to doubt' , which represent the normal case.
In sekkek=na 'he caused us to doubt', schwa must be voiceless, and one
hears [sk :hkna]. When =u is cliticized to sekkek the resulting form is
sekk!k=u, where ,2, represents the release of the preceding consonant, and
one hears [sk: hku].4 7 In skeke na 'our plowshares' and skek=u 'his plow-
shares', on the other hand, the two occurrences of k must be separated by
a voiced vocoid: [sk@kna], [sk@ku] . Pronunciations such as [sk'kna],
[sk'ku] or [sk:na], [sk:u] are unacceptable.
Turning now to absorption by an adjacent sonorant , let us compare cliti-
cized forms of slel 'baskets' ((52)d) and of seilel 'rince ' , which represents
the normal case. Consider seilel=ha 'he rinced her'. (53) gives (a) the
syllabic parse in the new analysis, (b) the same in the old analysis , which
some readers may find easier to relate to the standard transcriptions, and
(c)-(d) the two possible pronunciations of seilel=ha.
(53) a b c d
.sl.ll.ha. .sel.lel.ha. [sl:@lha] [sl::ha]
In the notation used in § 9.4.1 (see (37)), (53)c and (53)d would respec-

45 See McCarthy (1986) on antigemination effects.

46 On devoicing and absorption, see OUT discussion of W-internal sequences in § 8.2.2.
47 The terminal representation of sekk!k=u is identical with that displayed in (38), except
for the last consonant.
330 CHAPTER NINE

tively be transcribed as slz2lha and slllha. (54) displays the phonological

object represented as .sl.ll.ha. in (53)a.

(54) o o o

o
-<.R ~
o R o
-<.R
I I I
N N N
I I I
X X X X X X
I
s <>
I
I
I
I
h a
I
Pronunciations (53)c and (53)d are both realizations of the structure depicted
in (54). The two pronunciations differ in the manner of transition between
the onset and the nucleus in the second syllabie. We take this difference
to be a matter of phonetic implementation, as explained when we dis-
cussed analogous examples in § 9.4.1. In (53)c the con striction of the
geminate II is relaxed before a similar constriction is effected for the
articulation of the following simple I. In (53)d, on the other hand, the 'hold'
period of II blends with that of the following land the result is an unin-
terrupted, extralong, 1.48 Under the older analysis, one would say that in
(53)d the schwa in the second syllable has been absorbed by the following
/1/, with which it shared its nucleus node.
'Absorption' is not possible in sleleha 'her baskets' , which can only
be pronounced [sl@lha]. [sl:ha] is weII-formed, but only as a realization
of sellssha 'he extracted her' (v. sell 'he extracted').
Why do the CCeC words which are plural forms of CCC-a nouns not
behave like the other CCeC words? Recall that the other CCeC words result
from the syllabification of inputs of the form (lCCC/, FinL). We propose
that CCeC plurals like those in (52) are different in that their vowel comes
from a template akin to those which give rise to kerneis with a full vowel.
In the same way as the morphology of MA has a template CCaC respon-
sible for the derivation of klab 'dogs' and Ymam 'paternal uncle s' from
the singular forms kelb and Yemm, or a template CCuC underlying ilud
'skins' and xdud 'cheeks' , which are derived from the singular forms ield
and x edd , we propose that it has a template CCeC for the plurals in (52).
One reason to think that in the CCeC plurals of CCC-a nouns the input
to syllabification is not simply (lCCC/, FinL), is that only in these plurals
does MA allow kerneis like skek, slel, in which a geminate is split by an
occurrence of schwa. In particular MA does not have any CCeC verb or

48 These extra-long consonant s contrast with the geminates, as already pointed out in Harrell
(1962a) .
 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 331

singular noun with the shape CCjeC; ('Ci' . . Ci' stand for identical con-
sonants)." As is well-known, phonological epenthesis cannot split
geminates, whereas templatic morphology can."
In the CCeC template which we posit for the CCeC plurals of CCC-a
nouns, the vowel e is the same feature bundle as that which appears in
expanded hollow syllables such as the final syllables in (9)b and in (l4)c,
which is why plural forms like those in (52) cannot be distinguished on a
purely auditory basis from other CCeC forms, e.g . setting aside the dif-
ference in the first consonant, greb 'goatskins' does not sound different from
zreb 'he hurried', and sbek 'nets' is homophonous with sbek 'he tied (e.g .
strings) into a net' .
(55)a below is the representation which results from mapping the singular
noun kelb 'dog' onto template CCaC to form the plural klab; similarly, (55)b
is the representation which results from mapping the singular noun
gerb-a 'goatskin' onto template CCeC to form the plural greb.
(55) a. Iklabl b. Igrebl
X X X X X X X X
I I I I I I I I
k lab g r e b

Consider the words grebeha 'her goatskins' and greb=u 'his goatskins'.
In the input to syllabification the representations for these words both
contain (55)b. In grebeha, lei and Ibl will be syllabified as a complex
nucleus, in compliance with whatever mechanisms are responsible for the
shape of expanded hollow syllables. In greb=u, on the other hand, Ibl will
be syllabified as an on set to lu/, and Irl as one to lei, and the resulting
parse will be g.re.bu, in which the sole content of the nucleus of the medial
syllable is e.
greb and other plurals like it give us a further reason to prefer our new
analysis to that summarized in (99), in which all hollow syllables contain
an occurrence of e. Under the new analysis, a plural form like greb only
has one special property: r remains an on set regardless whether the
morpheme following greb begins with a vowel or a con sonant, a property
which we claim originates in the CCeC template. Under the analysis
summarized in (99) , the plurals in question have yet another peculiarity:
their schwa cannot devoice or be absorbed by a sonorant. It is not clear how
that analysis would explain the concomitance of the two peculiarities. Under
the present analysis there are no such processes as the devoicing of schwa
or its absorption by a sonorant. Consequently there is no concomitance to
explain.

49 Ath Sidhar Rifian has CCieCi verbs, on the other hand, as we have seen in § 6.5.3.
50 See e.g. Benhallam (1980: 14Iff), Hyman (1985: 126, note 22).
332 CHAPTER NINE

The rime in the medial syllable of greb=u 'his goatskins' (g.re.bu) is

what we called earlier a secondary rime, i.e. one in which schwa is the
sole content of the nucleus." The secondary rimes in c1iticized plurals
like greb=u are W-internal. Secondary rimes also occur in W-final position,
as a result of resyllabification across word boundaries. An example is the
second syllable in tq.te. fr.ia, which is the syllabic parse of /t-qt'l#!rta/ in
line 8 in (16). Accounting for W-final secondary rimes can only be done
within a reasonably detailed account of utterance-level syllabification. We
leave this problem for further research.

9.6. SUMMARY OF CHAPTER 9 AND ISSUES FOR FURTHER RESEARCH

Abandoning our earlier assumption that in MA every syllable contains a

vowel, we have argued that some post-onset schwas are transitional vocoids
analogous to those of Tashlhiyt. The hollow syllables of MA divide into
three types depending on the content of the rime, call these A, Band B'.
In type A the nucleus is schwa and there is no coda, as in the second syllable
of ~.k~.k!!, the terminal representation of skek=u 'his plowshares'. Syllables
of type A only occur in certain templatic kerneis and as a result of
resyllabification across a word boundary. In types Band B' the nucleus
contains a consonant. The run-of-the-mill hollow syllables are those of
type B, in which the nucleus is a bare consonant, as e.g. in the first two
syllables of b~.l'l.ng, the terminal representation of lbeivet-na, 'we
babbled' . As in Tashlhiyt, any consonant may be a nucleus. When a type
B syllable in which the nucleus is an obstruent occurs at the end of an
Intonational Phrase, it takes on a special form with a complex nucleus in
which the obstruent is preceded by schwa, as e.g. in the last syllable in
b~.l'et, the terminal representation of lbezvet 'he babbled' . Such sylla-
bles constitute class B'. In addition to these, class B' also contains syllables
whose nucleus is the diphthong ew.
Our modified account of syllable structure in MA involves surface
representations which are closer to the actual pronunciations than those in
the preceding chapter. This account offers a better starting point for com-
paring the phonologies of MA and Tashlhiyt: in certain cases where a
bilingual speaker feels that the pronunciation of a MA word is indistin-
guishable from that of a Tashlhiyt word, our new analysis assigns these
words identical terminal representations.
In conclusion, let us take a brief look at the ground we have covered
and raise some still-unresolved questions.
When we started writing this book, our central goal was to present the
evidence relevant to Tashlhiyt's peculiar syllable structure and embed it
in an overall picture of Tashlhiyt phonology and morphology. We wanted

51 See the text under (39) at the end of § 8.3.3.

 VOWELLESS SYLLABLES IN MOROCCAN ARABIC 333

to do this in a way that would help readers to see how our work fit into
the larger landscape of the variegated linguistic literature on Berber dialects .
Finding one 's bearings in that literature is not an easy task for non-
Berberists. It is often difficult to sort out genuine linguistic differences
among dialects from apparent differences which result from the divergent
theoretical outlooks and expository styles of different authors. In partic-
ular, we needed to puncture the illusion that as far as syllabification and
vowel epenthesis are concerned, the differences between the various Berber
dialects are only of a marginal nature. Our foray into the phonology of
Ath Sidhar Rifian has enabled us to point out some important differences
between Rifian, which has genuine epenthetic vowels, and Tashlhiyt, which
only has transitional vocoids.
Moving on further to MA was a natural thing to do. After many centuries
of contact, the pronunciations of Tashlhiyt and MA have much in common.
In cases where Tashlhiyt has a certain word containing only vowelless
syllables, and MA has a word which is homophonous with it, we might
be tempted to conlcude that MA also has vowelless syllables. But clearly,
the mere homophony of two forms could not take us very far unless both
are considered against the backdrop of the overall sound patterns which give
rise to them . We were thus led to study MA on its own terms .
Our work leaves many unanswered questions, having to do with
phonology as weIl as with phonetics. Several important phonological issues
were pointed out along the way. We now mention two phonetic questions
of immediate concern.
If we are correct in our contention that the short vocoids of Tashlhiyt
are all transitions between segments, a detailed study of these vocoids would
be of considerable general interest for understanding how phonetic imple-
mentation works and for determining to what extent it can vary across
languages. Our knowledge about the distribution of short vocoids in
Tashlhiyt is still very patchy, and making that knowledge more system-
atic will require tools of observation more accurate than the unaided ear.
Instrumental studies must of necessity concentrate on a few points of special
interest which have been previously identified as a result of broader surveys
of the terrain carried out with more primitive means. We hope that our work
in this book can serve as such a preliminary survey.
Whereas in Tashlhiyt all short vocoids are transitional, in Ath Sidhar
Rifian and in MA some are transitional while others are genuine vowels .
An intriguing question is what the phonetic differences are between the
two kinds of short vocoids, apart from differences in voicing in voiceless
environrnents." Our discussion in this book leaves the ans wer to this

52 The occurrence of glottal vibrations between two voiceless consonants is a sure symptom
of the presence of an intervening vowel, see e.g. § 9.2.
334 CHAPTER NINE

question completely open. Since the only feature specifications we have

assumed for the epenthetic vowel in MA are [-cons] and [+voice],53 it
may very weIl be the case that the mechanisms of phonetic implementa-
tion which account for the quality of transitional vocoids are also responsible
for the quality of epenthetic vowels. Between voiced consonants, then,
the only difference between transitional vocoids and epenthetic vowels
would be that the lauer, but not the former, have associated skeletal
positions. Whether this difference translates into systematic differences in
duration is a question worth investigating.

53 See the text immediately under (9).

 APPENDIX ON E

PRELIMINARIES TO APPENDICES 11 AND 111

In this appendix we give background information about the data to be pre-

sented in the next two sections. Our aim is not simply to help the readers
understand the content of these appendices, but also to give them some sense
of the proce ss of elaboration which leads from the raw data in our sources
to the sequences of Tashlhiyt word s which are our starting point in our
discussions of syllabification in singing in various places in this book.
The Tashlhiyt poems presented in Appendices 11 and III were composed
by !rrways, viz professional Ashlhiy musicians. The music and poetry of
the !rrway s borrow widely from those performed by village aficionados
during evening parties called ahwas. Being themselves actively involved
in altwas singing and dancing, villagers are a knowledgeable audience for
the productions of the !rrways (Schuyler 1979: 49-52, 237 ff.). As Schuyler
has already observed (p. 271), traditional singing is in sharp decline among
the urbanized Ashlhiys. ahwas evenings are difficult to organize in an urban
environment and the musical tastes of the young city-bred Ashlhiys tend
to be the same as those of their Arab peers. As the older generations dis-
appear and as the ties with the mountain villages slacken, experienced
listeners become less numerous among city-dwelling Ashlhiys.
Our main reason for drawing our data from songs composed by !rrways
is that ME is knowledgeable about the !rrways' production and is profi-
cient in their singing style. It would be a worthy task to examine text-to-tune
alignment in recent Ashlhiy songs whose style departs from the traditional
canons. We leave it to younger Ashlhiy researchers with an ear attuned to
these canons.
Appendices 11 and III present two songs drawn respectively from Amarir
(1975 : 147 ff.) and from Asid and Lachgar (1996 : 23 ff.), two collections
of Tashlhiyt songs written down using the Arabic script. The authors of
the two volumes are Ashlhiys. They worked from recordings on tape . Going
from their Arabic transcription to the one we give below cannot be done
in a mechanical fashion ; some amount of interpretation is unavoidable. I
Let us explain briefly how the Arabic script is put to use to notate
Tashlhiyt. The reader may recall that in the Arab world the only variety
of Arabic normally used in writing is Classical Arabic (CA). As they learn
to read and write, children also leam CA . While CA is quite different
from the local Arabic dialect the children speak, the correspondences

I Recordings of Tashlhiyt songs are easily available, but for practical reasons, using written
sources was the only option open to us at the time we did our work on versification .

335
336 APPENDIX ONE

between them are systematic enough to be readily apparent to the literate

speakers, who generally view the variety of Arabic they speak as a
corrupted version of CA.
In what follows, forms written in the Arabic script are represented by
sequences of roman letters enclosed between angled brackets. Thus (s)
represents the Arabic letter siin , which notates the sound s, and (ktb) stands
for a sequence of three Arabic letters, those representing the sounds k, t
and b. As is well known, conventional Arabic spelling only represents
consonants. Short vowels are not represented at all. The word taliduti is
written (thdü), also the spelling of the words tuhdatt, tuliditi, tahaddaü
and tuliadditi (consonant gemination is not represented either). Long vowels
are represented by consonant letters: uu by (w), ii by (y) and aa by the letter
aleph , which we represent here by (A). While katab is written (ktb), kaatab
is written (kAtb) and katabaa is written (ktbA) ; while wuzin is written (wzn) ,
wuzinuu is written (wznw) and wuuzinuu is written (wwznw).
Since the 8th century the Arabs have devised diacritics to represent
short vowels and consonant gemination. Aside from its systematic use in
the Koran, the spelling which comprises these diacritics, 'vocalized
tmaskuul) spelling', as it is caIled, is mainly used for pedagogical purposes.
Vocalized spelling is also resorted to in order to notate MA or Tashlhiyt
in those rare instances where these languages are committed to writing.
In the Arabic script the diacritics are written above or below the letters,
but we will represent them as superscripts immediately after the letters to
which they are attached. In a vocalized text, geminates are indicated by a
special diacritic which we represent by (2), and each occurrence of a vowel
is represented by a diacritic sign attached to the letter corresponding to
the consonant which precedes that vowel. /a!, /i/ and /u/, the three vowels
of CA, each have their own diacritic, which we represent here by (a), (i)
and (U). The vocalized spelling of tuhaddib, a word normally written simply
as (thdü), is (t" h a d 2, i 8°), with two diacritics attached to (d), one to indicate
gemination and the other to represent the fact that dd is immediately
followed by i. The raised zero at the end of the above spelling stands for
the sukuun diacritic, a sign which indicates that the consonant to which it
is attached is not followed by any vowel. Other illustrations of the use of
the sukuun are found in such vocalized spellings as (t" h 0 da 8°) for tulidat)
(standard spelling (thd8»), (k" A t" b0) for kaatab (standard spelling (kAtb»)
and (m" 5° k" w 1°) for maskuul (standard spelling (mskwlr). The last two
examples illustrates the notation of long vowels in the vocalized spelling.
Before we see how the vocalized spelling is put to use to notate Tashlhiyt
and MA , a caveat is in order. Although the Arabic script has sporadically
been used for many centuries to record poems in Tashlhiyt.' this practice

2 See, e.g., Boogert (1997, 1998) and Stroomer (1992).

 PRELIMINARIES TO APPENDICES II AND III 337

has not evolved its own system of conventions.' At present there is no

way of writing down Tashlhiyt which is deemed the only correct one. The
problems faced by the authors of the Arabic transcriptions are of the same
nature as those faced by the authors of phrase books for travellers, in
which sentences in one language are notated using the spelling of another.
Different authors may use different letter combinations to represent the same
sound, and alternative renditions or even inconsistencies can be found within
the same phrase book.
The Arabic transcriptions of Tashlhiyt poems published during this
century present variations in the delimitation of words and in the notation
of full vowels, but they are consistent on the following point: As far as
vocoids are concerned, these transcriptions represent all the occurrences
of the glides and full vowels present in the Tashlhiyt material, and only
those. If a consonant is not immediately followed by a full vowel in the
Tashlhiyt material, its written counterpart in the Arabic transcription has
a sukuun diacritic or no diacritic. We illustrate this generalization with three
lines from a Tashlhiyt poem by A. Hafidi which the author hirnself has
notated with Arabic letters."
(1) a. bayn-n i-tra-n xet-agut-t masesul' n-ss-ntal
b. l-!yamar-t nm a t-i-ddukkla n-ssnetnteakrk"
c. ss-ifif-veit kraed i-s-lulli w-a-fus" nx
We give below the author's rendition of these lines in the Arabic script.
(2) a. b" A y n2, 0 i t0 ra A n x ta A g'' W t2, 0 m" A
S2, U w 1 n° S2,0 n° t A 10
b. 10 ya A m'' A r t0 n m r
ti y d2, U w k0 1 A
n" S2.0 n° t n° t ?a k , 0
0 2

C. S2, i Y r y t0 )'i Y t k0 ra A d0
i S0 l'' W 12, i Y W
a A f" W S0 n° x0
The metrical pattern of the poem requires each line to have 12 syllabies,
with a H rime in the 3rd, 7th and 12th syllable:

3 Such conventions have recently been proposed, see Chafik (1990, 1991) and Elmedlaoui
(1999) and references therein.
4 The Iines below are Iines 7, 10 and 14 in the poem in Hafidi (1996: 33-34). Here are

the meanings of these Iines: (a) The stars appeared through the ciouds; there is not anymore
anything to hide. (b) The signs of friendship, this I know very weIl. (c) Let me then sift all
that my arm would have grinded.
5 From /mad=sul/.

6 Bound state form of a-fus 'arm', in which the initial vowel fails todrop in order to meet
the needs of the meter. In Imdlawn, only u-fus is acceptable in nonpoetic speech.
338 APPENDIX ONE

(3)
1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L H L L L L H
a. ba yn- nit ra nx ta gutt mas- su Ins- sn tal
b. 1 ya mar tn ma tid- dukk 1a ns- sn tn takWk W
c. s- si fif )'i tk ra dis lu1- li wa fu snx
Besides illustrating our observation about the notation of full vowels of
Tashlhiyt and that of their absence, the lines in (2) are also typica1 of various
prob1ematic aspects of Arabic transcriptions. The 1abia1ization of /kwkw/
at the end of 1ine (3)b is simply glossed over in the Arabic transcription.
Unlike Tashlhiyt and MA, CA does not possess any labialized consonants
in its phonemic inventory and the Arabic script does not have 1etters
representing 1abialized consonants. Note also the variation in the repre-
sentation of the full vowe1s. CA has an underlying contrast in vowe1length,
which Tash1hiyt does not. The full vowels of Tash1hiyt are spelled as long
in some instances and as short in others, e.g. the rightmost a in line (a)
and the two occurrences of a in Il-yamar-t in line (b) are spelled as long,
whereas the rightmost occurrence ofc in line (b) is spelled as short. These
variations do not reflect any phonological distinction in Tashlhiyt. Finally
there are typographical uncertainties. There is actually a sukuun diacritic
in the second word of 1ine (3)b, but the typographicallayout in our source
makes it unclear whether it belongs to (n) or to (m). Ditto for another
sukuun, near the end of the first word in the next line, where it is unclear
whether the intended spelling is (y0t) or (yt0).
While the authors' Arabic transcriptions enable a speaker of Tash1hiyt
to retrieve without ambiguity the words in the original recordings, they
do not a1ways enable one to make the appropriate choice between alter-
native pronunciations of the same word. The authors have a tendency to
transcribe the words as they are pronounced in isolation. ME has sung
each 1ine before re-transcribing it. When a sequence of words transcribed
in Arabic 1etters could be pronounced in severa1 ways, he chose the pro-
nunciation which he feIt sang most naturally to the tune. Here is an example
to give an idea of the kind of decisions ME had to make. The words in
(4) below are those of line 31 in the song whose first lines were parsed
in (19) in § 4.5:
(4) is aeka i-siggil ag=giwn afi-n l-mvafl-t'
The sequence of words in (4) can be pronounced in two ways depending
on whether the final segment in afi-n is assimilated to the following lateral.

7 ' He is merely seeking distractions from you' . The phonological representation aeka is
lar=ka/ and that of ag=giwn is lad=giwnl .
 PRELIMINARIES TO APPENDICES II AND III 339

One can pronounce /n! without assimilation to the following /1/, hence the
sequence of segments represented in (5)a below, which can only be parsed
as (5)b in view of our discussion of syllabification in Chapter 4:8
(5) a. is aka isiggil aggiwn afin lmvaflt
b. i sa kay sig- gi lag- giw na fi nlm -ya fit
1 2 3 4 5 6 7 8 9 10 11 12
On the other hand, if the optional assimilation takes effect, (4) is realized
as the segment sequence in (6)a, whose orthometric parse is given in (6)b:
(6) a. is aka isiggil aggiwn afil lmvaflt
b. i sa kay sig- gi lag- giw na fil- 1m -ya fit
12345678 9 10 11 12
While syllable #10 is H in (5), it is L in (6). (6) sings without a hiteh,
which is not the case for (5); this accords with the fact that the meter of
the song requires a L syllable in the tenth syllable. In this particular line,
then, the optional assimilation should apply, and ME chooses pronuncia-
tion (6)a.
In re-transcribing the songs ME has merely chosen between variant
pronunciations compatible with the Arabic transcriptions; he has not
straightened the lines out. Some lines in the songs are ill-formed. Such lines
are marked with an asterisk whose location indicates the point where the
meter is violated. Some of the violations may be due to transcription errors
in our sourees .
One kind of violation is worth mentioning here although it does not occur
in the two songs presented below. In Ashlhiy singing it is not unfrequent
for a line to have one syllable less than the number required by the meter.
In singing, the gap created by the missing syllable is patched by stretching
a neighboring syllable . Here is for instance line 32 in the song by Hmad
Biyzmawn already cited in (33) in § 4.6:
(7) a yan u-tbir i-bbi flla !laxbar nns"
We give the scansion of this line below in (8)b, together with that of line
(33)a in § 4.6, which is reproduced as (8)a for the sake of comparison:
(8) 1 2 3 4 5 6 7 8 9 10 11 12
L L H L L L L L L H L H
a. a ma nad ra ru mk- ki ns ta ray za -yarn
b. a ya nut bi rib- bi fI- la lax ba rnns

8 In (5)a the spaces between words are given only for the readers 'convenience.
9 Ah! the loved one (lit 'a dove'), he stops sending news.
340 APPENDIX ONE

In the singing of line (8)b the syllable bi is used as a carrier of the portion
of the tune which is carried by ra and ru in (8)a. In some instances of
this kind the stretching of a syllable over two positions sounds so natural
to ME that he may not notice the metrical violation immediately. This,
however, should not make us lose sight of the following point: although
lines like (8)b lack one syllable , they also give us data about syllabifica-
tion when their alignment with a tune is examined.
In Tashlhiyt as in French, all the pronunciations acceptable in non-
poetic speech are also acceptable in singing, but the converse is not true.
In our transcriptions of the songs the pronunciation is always that in use
in Imdlawn, except for the realizations of lfi/. In Imdlawn this phoneme
is realized as a long a in some contexts (v. § 3.7) but in some other Tashlhiyt
dialects it is always realized as a consonant. The people of Imdlawn are
used to hearing that pronunciation from other Ashlhiys and they can use
it themselves in singing . Setting lfil aside, those pronunciations notated in
our transcriptions which are not acceptable in everyday language in Imdlawn
are all pointed out in footnotes . 10
Our transcription is the same as that used elsewhere in this book, with
the following modifications. The exclamation point indicating emphasis
(dorsopharyngealization) is prefixed to the morpheme which contains
emphasis in the underlying representations. Parentheses around a vowel
indicate an underlying vowel which is elided (the contraction of two occur-
rences of the same vowel into a single short vowel does not occur outside
of the poetic language). We have notated with a capital 'A' the vocative
particle a and the vowel a often used as a stopgap syllable at the begin-
ning of lines, to distinguish them from other words pronounced a, which
are realizations of ladl with its consonant deleted. 'y' between square
brackets represents the hiatus-breaking glide .
In the texts of Appendices 11 and III, two successive occurrences of the
same letter not separated by aspace always represent a geminate, regard -
less whether they belong to the same morpheme. Let us review three kinds
of heteromorphemic geminates which are a common occurrence in the songs
cited below.
First, Berber nouns which are loanwords from Arabic begin with the
prefix lI-I, which assimilates to a following coronal (v. § 2.5.3.1). In the
texts below, all the words which begin with two identical letters separated
by a hyphen are nouns with an underlying shape II-Z/, see e.g. 11:3 (third
line of the song in Appendix 11).

10 Some of these are in use in everyday language in other Tashlhiyt dialects, but this is
irrelevant for our purposes in this book. We will not dweil on syntactic irregularities. In
II:51, for instance, the pronounfllas should immediately follow the verb i-Idr, and in III:62
the noun t-i-Irzi, which is govemed by apreposition, should be in the bound state. On the
distinctive characteristics of the syntax of poetic language, see the works of Galand-Pernet,
Jouad and Bounfour.
 PRELIMINARIES TO APPENDICES II AND III 341

Second, the consonant of the grammatical morpheme lad! often assim-

ilates to the next consonant. In the texts below, all the occurrences of aC
in which C is identical to the consonant at the beginning of the next word,
are realizations of lad/, see e.g. 11:12, 50, 51.
Third, in certain contexts word-final Inl optionally assimilates to a
following sonorant. In the texts below, all the occurrences of a word with
the shape R=, where R represents a sonorant identical to that which follows
the = boundary, are realizations of the genitive preposition n, see e.g.
11:25,39 and III:l,49.
 APPENDIX TWO

SONG

This poem is a song by Rqiya Tandmsirt, a professional singer (!tarrayst).

ME knew the original tune of this song, and he sang every line to that
tune in order to transcribe it.' Every line ends with the vowel i, which is
omitted below.

1. A [y] a-marg ureak nzi-v ula sllm-veak

2. A kiyin del-hubbia) ad=an)'2 i-kkis-n t-iram
3. ula s-lsah-t inu t-mda flla lahh l-lun
4. nkki n-niy-t inu a=yy(i) i-hlk-n llieka !~mmr-)'
5. n-gaenn r-Irza )'=bnadm fki-neav i=t-illas
6. i-sggras-n a [y] ad ludr-v I-iyam ur n-ssin
7. is i-lla )'k-a=d lli gi-v )'assa )'=I-hayat
8. nkka kullu t-i-rnizar yar' t-!trf neyat
9. kullu maned n-kka s=u-!dar n-kkaetn s=I-!frh
10. A zud I-hna neu-gadir ur i-li t-taman
11. walayinni" t-a-mazir-t mra y-ufi yan
12. ag=gis i-skr t-a-mazir-t n-s" aru-n arraw
13. i-ga sus zud l-hizz i)' y-ufa yan I-mal
14. asesrs i-sv I-mlk i-genn kullu t-i-Imitas
15. ukan i-vz l-Yin i-genn fllas l-mutur
16. i-genn !rrZa nn-s )'=!rbb(i) a=ys6 i-kmml s=l-!xir
17. !ullah amradd ur I-hmm nn-un a wi-nnav
18. A tt-ini lzdar-av ad ut-v ukan a- lvaras
19. a mmied uski-)' a mmied ut-v ukan a-lvaras
20. a n-issan ma mmi=k=id n-fl a winu )'=u-fus
21. A winu wahqq !rbbi ini sisn t-amn-t
22. A t-asa nu tt-ini gis yan amr kiyin
23. ar ggan-v ar=i 7=k=id i-s-mala yan I-mlk

I V. Tune I in Appendix IV.

2 Variant of the Ip object pronoun ax. This variant is only used in poetry.
3 Even in singing the release of this consonant is only optional, v. § 6.3.3.
4 This pronunciation is acceptable only in poetry. The normal pronunciation is walaynni.
5 From Inn-sl 'of hirn/her' . In poetry, the initial morpheme in certain possessive determiners,
which is normally pronounced nn, can be pronounced with a simplex consonant.
6 Underlyingly lad=as/.
7 The Is pronoun liyi/ , which is realized as yyi or as iyi depending on the context , has a
variant liy/, which is realized as yy or iy depending on the context. In poetry the lauer
realization can be shortened to i.

343
344 APPENDIX TWO

24. ar=i 8=d i-s-mala t-iddi nn-k ul(a) awal nn-un

25. A lsbhan wallillah a kra u=w-awal nn-un
26. i-ga zun d r-ribab i-y=tn y-ut u- lsnna'i
27. i-ga zun d l-luz i-y i-sli y-ili t-ammn-t
28. A zun i-tt-yara -y=iggi n=t-asa nu sel-qlm
29. ur a=yyi=kwn i-ssirid u-!nzar ul(a) a-Ismmid
30. lrwah a n-mun nkki dik yan i-ra-n i-mun
31. lrwah a n-mun i=t-udr-t n-mun i=s-siwal
32. n-mun i=s- !sirat inn -y ur t-uki-t n-ss-akriek
33. lrwah a gWma set-mazir-t nn-av akek awi-v
34. adeak n-bnu t-a-gadir-t ammas new-aman
35. A !brra nn-s adeas n-g r-ryal i-la-n a-fus
36. a-g'tns nn-s adeas n-g l-hrir net-ukay-in
37. awi=yyis=darun i-y ur t-lrdi-t akek awi-v
38. ad=ak n-g t-a-wayya -y=u-nwal i=ma t-uru-t
39. adeak n-tt-asi madedark i-lla-n uew-arraw
40. adeukan t-nni-t mbarka s nni-v n\'am
41. A t-anna ur n-lsuwib n-ss=fllas a-Ikuray
42. !imrbba s=yan i-tt-azzal-n l-firaq nn-av
43 . at=tn gi-n d u-safu gi-n w-aman a-safar
44 . amar i-bna-n a-gadir s=i-!msd net-ammn-t
45 . i-g=as=nn t-i-gzda deu-lsvar n-k" a r-lrihan
46 . i-g=as a-k'faf net-i-lbinsr-t t-luh aman
47 . t-i-flw-in ti n=z-zaz l-qfl wi nen-Jnqqr-t
48. t-a-saru-t n=d-dahab i-fili win l-mlf
49. ukan i-fkeak a bab n=d-!draf-t t-a-saru-t
50. !imrbba set-asa y-ugi-n t-!ayyad at t-livar"
51. akekullu t-g l-lbarud i-Idr u-safu fllas

H H H
2 4
3 5 6 7 8 9 10 11 12
1. a ya mar gu ra kn *zi -yu la sl- Im -yak
2. a ki yin dl bub- ba dan -yik- ki sn ti ram
3. u las- sab ti nu tm *da fl- la lah- bl- lun
4. nk- kin- niy *tin way- yi hlk nl- li ka 'lm- mr-y
5. n gan- nrr za -yb na dmf ki na -yi til- las
6. sgW_ gW as na ya du dr-y li ya mu rns- sin
7. i sil- la-y ka dl- li gi-y -yas- sa -yl ba yat
8. nk- ka kul- lu ti mi zar ya tt- tr fn yat

8
See note 7.
9
From /nn-k/: v. note 5.
10
From lad t-Iirav/,
 SONG 345

9. kul- lu man dnk- ka su *da rnk- ka tn si frh

10. a zu dlh na nu ga *di TU ri lit- ta man
11. wa la yin- ni ta ma zir tm ra yu fi yan
12. ag- gi sis kr ta ma zir tn sa TU nar- raw
13. i ga sus zu dl hiz- Zi-y yu fa ya nl mal
14. as- sr sis 'Y I ml ki gnn kul- lu ti mi tas
15. u ka ni'Y zl 'li ni gnn fl- la si mu tur
16. i gn- nrr zan- ns 'Y rb- bay si km- · ml si xir
17. ul- la harn rad- du rl *hm- mn- nu na win- na'Y
18. at- ti niz da ra 'Ya dut 'Y u ka na 'Y a ras
19. am- ml dus ki varn- mi dut 'Y u ka na 'Y a ras
20. a nis- san mam- mi ki dnf la wi nu 'Y u fus
21. a wi *nu wa hq- qrb- biy ni si sn ta mnt
22. a ta *sa nut- ti ni gis ya na mr ki yin
23. a rg- gan 'Y a ri ki dis ma la ya nl mlk
24. a ri dis ma la tid- dinn ku la wa In- nun
25. a sb han wal- lil- la hak raw- wa wa In- nun
26. ga zun dr- ri ba bi'Y tn yu tu sn- na'l
27. ga zun dl- lu zi 'Y is li yi li tam- mnt
28. a zu nitt ya ra vig- gin ta sa nu si qlm
29. u ray- yikW nis- si n dun za TU la sm- mid
30. r wa han mun- nk- ki dik ya ni ra ni mun
31. r wa han mu m tu *dr tn mu nis-si wal
32. n mu nis- si ra *tinn 'Y ur tu ki tns-sa kWik
33. r wa hagW ma st ma zir tn- na 'Y ak-
ka wiv
34. a da knb nu ta ga dir tam- ma sn wa man
35. a br- rann sa da sn grr ya li la na fus
36. a gW n snn sa da sn glh ri m tu ka yin
37. a wiy- yis da ru ni 'Y ur tr di tak- ka wi'Y
38. a da kng ta way- ya 'Y un wa li ma tu rut
39. a da knt- ta si mad- dar kil- la nu- war- raw
40. a du kan tn- ni tm bar ka sn- ni 'Y n 'lam
41. a tan- naw rn su- wi bnss fl- la sa ku ray
42. mrb- bas ya nit- taz- zal nl fi ra qn- na'Y
43. at- tn gin du sa fu gin wa ma na sa far
44. a ma rib na na ga dir si ms dn tarn- mnt
45. i ga snn ti gz da dus 'Ya rn kar- ri han
46. i ga sakW fa fn ti bin sr tt lu ha man
47. ti fl win ti nz- Za zlq fl wi nn- nq- qrt
48. ta sa ru tnd- da ha *bi fi li wi nl mlf
49. u ka nif ka ka ba bndd ra ft- ta sa rut
50. i mrb- bas ta sa yu gin tay- ya dat- ti 'Y ar
51. ak- kul- lut gl ba ru did ru sa fu fl- las
346 APPENDIX TWO

1. Passion for music, I will never forgive you;

2. Because of you and of love I have lost taste for food ;
3. My health is ruined, my energy is gone.
4. Me, it is the good will I display which has doomed me.
5. I had invested my hope in human beings, and I am plunged into
darkness.
6. I have been stirring the days for years without ever suspecting
7. That life may lead one to a situation such as mine.
8. I have visited country after country;
9. Everywhere I set foot I did so in joy,
10. But the tranquility of Agadir is something money can't buy.
11. Ah what a country! If only one could
12. Have it as one's own and have children there!
13. The Sous is like the Holy Places if one can afford
14. To buy a plot of land and grow tomatoes,
15. To dig a weIl and set up a pump,
16. And if one can trust that God will grant one's wishes.
17. Were I not worried for you, 0 mine, I swear
18. That I could never travel all the way to here.
19. What made me go all the way
20. Was to know to whose hands I am trusting you, 0 mine!
21. 0 mine! I swear by God -glorify Hirn and believe in Him-
22. That the thought of no one but you obsesses me.
23. In my sleep a ghost shines your image to me;
24. He conjures up your figure as weIl as your voice .
25. God be praised! What a voice, your voice!
26. It is like the one-stringed fiddle under the virtuoso's fingers,
27. Like almonds roasted and soaked in honey ;
28. It is as though a stylus had engraved it on my heart;
29. Wind nor rain can erase it.
30. Come! Let us be together, you and I, no matter what the others do.
31. Come! Let us share life and purgatory.
32. Together, let us cross the Sirat isthmus, and where you falter, I will
support you.
33. Come along, brother, I will take you to my land.
34. There I will build for you a horne in the midst of the water.
35. On the outside it will be decorated with silver reals;
36. The inside will be lined with tukayin" silk.
37. Take me to your place, if you would be humiliated by my taking you
to our place.
38. And there I will be a servant to your children.

11 Word unknown 10 ME.

 SONG 347

39. 1 will rock the babies which you will beget.

40. It will be enough for you to call out 'Mbarka!', and 1 will answer 'Yes' .
41. Let me be punished for whatever 1 fail in.
42. Whoever works at separating us,
43. Let them become a firebrand and let water be their remedy!
44. Let them build Agadir with honeycombs;
45. Let them fit it with myrtle joists,
46. With slanted roofs to let water flow down,
47. With glass doors with silver locks,
48. With golden keys on silk key-holders,
49. And in the end offer you the keys, 0 loved one!
50. Let the heart which hates become dry,
51. Let it be turned into gunpowder and be touched with a firebrand .
 APPENDIX THREE

ORATORICAL ENCOUNTER

The lines presented below were improvised during an oratorical contest.

Later on the authors themselves (Asid and Lachgar) wrote them down using
the Arabic script. Since he has not heard a recording of the encounter,
ME does not know to which tune(s) the line s were sung, and consequently
he does not know whether they were sung with an additional vowel at the
end , like those in Appendix 11. However, ME knows several tunes which
fit the meter of these lines. He chose one such tune and sung each line to
it before transcribing it. I
Even if it is not obvious from our translation, to an Ashlhiy audience
the stanzas below form a coherent succession of retorts. On covert meaning
in Ashlhiy tales and poetry, v. Galand-Pemet (1972) et Jouad (1989).
Lachgar:
1. A t-a-bra-t i-ss - lrsa u-fus I=I-lalim
2. i-gaeysnt l-haq a-mhas i=u-g Wmmay
3. at=tnt y-azn yan iel-xlq i-s-hmma-n
4. wa-lli i-ra -n akw=kWnt !aqqra-n i-fhmekrnt
Asid:
5. l-Yaql a i-ss-!rsaw-n arra bla s-smx
6. kullu kra i-bna s-lsah-t a=f a=t bnnu-n
7. ar aqqra-rrr' mddn xel-msayl zri-nin
8. awi-ned l-lxabar i=xti-ll(i) ur !zrra-n
Lachgar:
9. l-aman-t ur i-zz-nfam yan a-yd usi-x
10. i-qqaned a n-mmav a ur=i 3 t-!rz )'=ufus
11. imm(a) ur sul i-lli yanegis i-ltthlla-n
Asid:
12. d-din i-Idfur-n bab nn-s i-qqaned ukan
13. y-ufeasn at=tn y-addu i-srs i=l-hmm n-s"
14. at=t ur i-tt-n'iat !udad is )'ussa-n

I V. Tune 2 in Appendix IV.

2 From laqqra-nJ through assimilation .
3 V. note 7 in Appendix 11. The unreduced variant is found below in !ine 64.
4 V. note 5 in Appendix 11.

349
350 APPENDIX THREE

Lachgar :
15. A t-a-nsa nei-bnkal-n i-ffav l-aman
16. t-lli-t xeu-Ivaras uregis i-z~m yan
17. i-ra u-!dar at=t n-ss-anf iet-awdiw-in
Asid:
18. i-ga l-Yaqql t-ifaw-t ur i-tt-Ivrra-n
19. kada neyan aded i-!di x=gr=t-imday-in
20. kada neyan i-flet ar=t i-tt-!amz l-xuf
Lachgar:
21. i-ga u-Ivaras a-sawn i-vlb kuyan
22. labudd a i-ss-!rmuy wa-nnaed y-iwn-n
23. i')' i-!zr(a) a-mdlu dl-n kullu i-gnw-an
24. ur a=d y-akk(a) aman is=tn=k(a) i-gli r-rih
Asid:
25. wa-nna i-la-n a-fud iet-wada w=!')'aras5
26. ur ar6=t i-ss-iwid u-drar i-!d~a=[y]as
27. ur aregis i-tt- !rza l-himma bla yat
28. i')' i-llas l-hal urtaenn i-zri man
29. ula i- lruh xet-ifaw-t ar i-tt- lbrram
Lachgar :
30. i-ra u-zrg i-mndi i-ra y-igr aman
31. i-ra bnadm 1- !usiy-t at=tn ldfur-nt
32. i-ra l-muhndiz l-lsas ad g-n yan
33. ad i-bnu f=l-naqq a ur i-!ttar dlhin
Asid:
34. hati z-zman-ad aex uresul i-!dhr yat
35. yak l-lsas ur i-dus-n a=f i-bna s-!sur
36. imikk sul at=t i_qWqWay_n !imma i-Idr nit
37. wa-nna i-ra-n a-malu nn-s i-lla ')'=l-unl-t
Lachgar:
38. i-ga l-bnya wH 7 l-uqqt ak=k(a) i-!sbr yan
39. wa-lli x hawl-n i-fass-n ar=ax=d i-Ittar
40. i-ggut ma=mm(i) i-tt-af l-Ivrur a i-bn s-sas
41. vinn a=x !zrra-n mddl" l-himma drus-nt

5 From Iwada neu -lvaras/. In! assimilates to the following sonorant, whence a geminate high
vocoid. In Imdlawn the degemination of the high vocoid is acceptable only in poetry.
6 In this line and in the next, the preverb larl retain s its final consonant despite the fact
that it is preceded by another preverb. Outside of poetry, the final Irl obligatorily drops in
such an environment, see DE (1989: 180).
7 From Iwin!.
8 From Imddn! .
 ORATORICAL ENCOUNTER 351

Asid:
42 . l-haqq a=s a y-!atta l-Yin i=I-)'11-at
43 . walayni? t-!rza=[y]ax t-rga t- Ivrreax
44 . i-lli i-rufa-n is i- !qqur ur=d i-xlf
45 . i-lli i-ra-n imikk uregis i- !dhr yat
Lachgar:
46. t-addar-t a-lztta nn-un wr'? rad !«i'mmr-n
47. usi-ned i-fass-n t-i-zlaf-in del-muss
48. gabl-n t-a-wala s-sll-t ag=gis gWmmr-n
49. ur=akwkw=sul skr-n uss-an l=l-!«i'nsr-t
Asid:
50. A t-ladfi wa-lli=tt i=myar-n t-huwlet
51. i-qqaned abebdda y-asi aggu d=w-aman
52 . awi-ned a-faruz nn-s akwi-n !«i'mmr-n
53. !mqqar gis l-haqq nn-k ur=ak=d i-ffuv
54. i-lli x=t=inn t-fl-t a=x=sul ur )'ama-n
Lachgar:
55. hann a gWma t-a-mazir-t-lli x=d n-lul
56 . !Hah abla bzziz a=s=tnd ll=d i-ffv l-xlq
57 . n-Ihrmesul a-Ivaras x=is i-zri yan
58. yan a i-ldi !rubas i-xlu=t=id w-asif
Asid:
59 . s-si'ir ix i-lla xeixf ur a s-hnna-n
60 . maxx is t-ufi-t a l-lasl ak=k i-zlu yan
61. lmqqaregis ur i-dawm i-bidd fllas
62. !mqqar=nit ur i-ssugr i=t-i- !rzi yat
Lachgar:
63. n-!zuzd abebddaegis n-ili tthnna-v
64. walayn(i) ureiyiegis maef n-tt-I'immar
65. mani=x n-ufa 1-!mdars i=mayd uru-x
66. ul(a) a-!dbib Ur=iI2=gis ma=yy(i) i-tt-dawa-n

Another poetic variant of walaynni , v. note 4 in Appendix II.

10 Even outside of poetry the negation ur has free variant Iwrl when it immediately precedes
Irad! (future).
11 From Itnt/.
12 V. note 7 in Appendix II.
352 APPENDIX THREE

Asid:
67. arraw-da net-mazir-t a=f i-bna s-lslh
68. ntt(a) a i-ra-n at=tnd=d i-bnu nix hlknt
69. i-nna xwa-i" id-bab nn-s i-ga i-Ivrm-an
70. ur raennegis i-ttrs, a l-lsas, u-!zru nn-un
Lachgar:
71. hann a-drar ur ugr-n l-!uda dew-asif
72. mllix t-Ili-t a d-Iduw xeimi y14=!dura-n
73. blhaqq bagnziz i-!ktr 'Y=u-glif n-x"
74. i-ssa kra i-!zda sul t-nkr t-vuyyi-t
Asid:
75. A t-izzw(a) aggu neilamm-n i-!zza=d=giwnt
76. i-Izzued w-!adu wala f=ti-da i-grmmr-n
77. kuyat=ka dei-frg-an x=as i-tt -'Ywi r-ris
78. t-lkm t-vuyyi-t n-snt" bab i17=i-gnw-an
Lachgar:
79 . t-affa=nn y-aggug-n f=u-!nzar n-hmlekrnt
80. walayni l-mnazileka x n-ldfr s-sur
81. ar ix n-ufa wadd i-Izlay-n d-dmn-t n-x"
Asid:
82. issn-t maexeann i-ffal bu -s-lsab-t !mnass
83. ass-an x i-suq i-zznzed ur lumz -n yat
84 . gar alwa t-i-!mudan=k(a) as=sul=d i-ffal
85. ix ur xlf-n l- luzur ti-lli zri-nin

13 From Ixwa-nJ. The optional assimilation of InJ to the following sonorant gives rise 10 a
geminate vocoid, which constraint NoOns- (§ 4.8) prevents from syllabifying as yi. xwayyd
is also a possible pronunciation outside of poetry.
14 From li mi nei- ldura-n/ , Preposition Inl optionally assimilates to lil and the resulting
rime iyy is shortened to iy. This pronunciation is also acceptable outside of poetry.
15 From Inn-x/; v. note 5 in Appendix 11.
16 From Inn-snt/; v. note 5 in Appendix 11.
17 From In=i-gnwa-nJ through the assimilation of the preposition In/.
18 From Inn-x/; v. note 5 in Appendix 11.
 ORATO R ICAL ENCOUNTE R 353

H H H H
2 3 4 5 6 7 8 9 10

1. a tab ra tis- sr saw fu sll ~a Iim

2. gay sn tl ha qam ha siw gW m_ may
3. at- tnt ya zn ya nil xl qis hm- man
4. wal- liy ra nak'r- kWn taqq ra nif hm kWnt
5. I ~ aq la ys- sr saw nar- rab las-smx
6. kul- luk ra yb nas- sah ta fat bn- nun
7. a raqq ram- md- dn xlm sa ylz ri nin
8. a win dl xa ba rix til- lur zr- ran
9. la man tu riz- zn zam ya nay du six
10. iq- qan da nm- ma "{aw ri trz "{U fus
11. im- mur su lil- li yan gi sitt hl- lan
12. *d- di nid fu m ba bnn siq- qan du kan
13. yu fas nat- tn yad- duy sr sil hm- mns
14. at- tu rit- tn ~a tu da dis "{us- san
15. a tan sa ni bn kaI nif- fav la man
16. u- lit xu "{a ra sur gl siz ~m yan
17. raw da rat- tns- san fi taw di win
18. i gal ~aq- ql ti faw tu ritt "{r- ran
19. ka dan ya nad- di dix gr tim da yin
20. ka dan ya ru fl tar tit- tarn zi xuf
21. i gaw "{a ra sa saw ni "{Ib ku yan
22. la bud- da ys- sr muy wan- nad yi wnn
23. "{lZ ra md lu dln kul- luy gn wan
24. u rad yak- ka ma ms tn kig Iir- rih
25. wan- nay Ia na fu dit wa daw "{a ras
26. u rar tis- si Wl dud ra rid ~a yas
27. u rar gi sit- tr zal him- mab Ia yat
28. "{il- Ia sI na lur tan- niz ri man
29. u Iay ru hx ti faw ta ritt br- ram
30. raw zr gi mn diy ra yig ra man
31. rab na dm Iu siy tat- tnd fu mt
32. ral mu hn di zll sa sad gn yan
33. a dib nu fl ~ aq- qaw rit- tar dl hin
34. ha tizz ma na da xur su lid hr yat
35. ya kll sa su ri dus na fib nas- sur
36. i mikk su Iat- tiqw_ qW ay nim- may dr nit
37. wan- nay ra na ma Iunn sil- lav Iu olt
354 APPEND IX THREE

38. i gal bn ya wil- luqq tak- kis br yan

39. wal- lix na wl ni fass na rax dit- tar
40. ig- gut mam- mit- ta fly ru ray bns- sas
41. -yin- nax zr- ra nmd- dll him- mad ru snt
42. naq- qa sa yat- tal lli nil -yl- lat
43. wa lay ni tr za yax tr gat -yr- rax
44. il- liy ru fa m siq- qu rur di xlf
45. u- liy ra ni mik - kur gi sid hr yat
46. tad- dar ta zt- tan- nun wr rad llm- mrn
47. u sin di fas- sn tiz la fin dl muss
48. ga bIn ta wa las- sll tag- gis gW m_ mrn
49. u rakwkwsu ls kr nus- sa nll lln srt
50. a tad fi wal- lit- tim ya rnt hu- wlt
51. iq- qan dab- bd- da yas yag- gud wa man
52. a win da fa ru znn sa k'Tn llm- mrn
53. rnq- qar gi sI liaq- qnn ku rak dif- fuv
54. n- lix tin- nt fl tax su lur -ya man
55. han- nag W ma ta ma zir tl- lix dn lul
56. *1- la hab la bz- Zl zas tnd- diff -yl xlq
57. n nrm su la -ya ras xi siz ri yan
58. ya nay di ru ba six lu tid wa sif
59. s- sill ri xil- la xix fu ras hn- nan
60. max- xis tu fi ta las lak- kif lu yan
61. mq- qar gi su n daw mi bidd fl- las
62. mq- qar ni tu ris- sug ri tir Zl yat
63. n zuz dab- bd- da gis ni litt hn- na-y
64. wa lay nu n- yi gis ma fntt llm- mar
65. ma nix nu fa 1m dar si may du rux
66. u lad bi bu n gis may- yitt da wan
67. ar- raw da nt ma Zlr ta fib nas- slh
68. nt- tay ra nat- tnd- dib nu nix hl knt
69. in- nax way- yd ba bnn si gay -yr man
70. ur- rann gi sit- tr sall sa suz run- nun
71. han- nad ra ru ru grn lu dad wa sif
72. ml- lix tl- li tad- duw xi miy du ran
73. bl naqq ba gn zi zik tr -yug li fnx
74. u- sak ra yz da sul tn krt -yuy- yit
75. a tizz wag- gu ni lamm niz- zad gl wnt
76. lZ- zud wa du wa laf ti day grrn- mrn
77. ku yat ka di fr gan xa sitt -ywir_ ris
78. tl kmt -yuy- yi tn snt ba biy gn wan
 ORATORICAL ENCOUNTER 355

79. taf- fann yag- gu gn fun za rnh ml kWnt

80. wa lay ni 1m na zil ka xnd frs- sur
81. a fiX nu fa wad- diz la yndd mn tnx
82. is- snt ma xan- nif- fal bus- sab tm nass
83. as- san xi su qiz- znz du rum zn yat
84. ga ral wa ti mu dan kas- sul dif- fal
85. i xur xl fn lu zur ur- liz ri mn

1. The message which the scholar's hand has written in a perfect

manner,
2. God the Just supplements it with vowel signs, for its reading aloud,
3. So that it can be sent to the conscious man,
4. To hirn who is able to read it and understand it.
5. Reason makes writings perfect without using ink.
6. Everything that Reason builds, it builds it on solid bases.
7. Men meditate on events of the past
8. In order to know something about events which they have never
witnessed.
9. I have embarked on a delicate mission;
10. It is up to me to do my utmost to prevent its shattering in my hands,
11. For there remains nobody else to take care of it.
12. A debt befalling one cannot be avoided.
13. The debtor would do weIl to repay and relieve hirnself
14. To avoid continually being pointed at as unreliable.
15. 0 hole of perfidious snakes,
16. You are at the side of the path; no one dares tread there;
17. My foot beseeches me to keep it away from nasty surprises .
18. Reason is a light that never misleads .
19. Many are those it has saved in the swell.
20. Many others it has abandoned, and they were left the prey of fear.
21. The path is steep; it challenges everyone.
22. It cannot but tire whoever goes upward
23. When he sees the clouds that cover the sky,
24. Rainless, smothered by the winds.
25. He who has strength to set forth,
26. Does not let hirnself be awed by the heights; they obey hirn.
27. Only one thing stops hirn:
28. It is when night falls before he reaches the stopping place,
29. And, not having arrived by daylight, he goes in circles.
30. The mill wants grains, and the field wants water;
356 APPENDIX THREE

31. People need edifying words that they can carry along.
32. The architect needs foundations which form a coherent whole,
33. So as to build correctly and avoid premature destruction.
34. In the present times the prospects are not encouraging.
35. Was the wall built on flimsy foundations?
36. It hardly stays up; it is ready to fall down.
37. He is taking chances, who hopes to take cover in its shade.
38. Those are works of nowadays; one has to make do.
39. All those in which too many hands are involved collapse .
40. Many people prefer precariousness to consolidation work;
41. That is where poor morality shows.
42. Springs obey Equity in pouring out water for crops,
43. But our crack-ridden aqueducts bias the apportionment.
44. Places lacking water have become dry and barren.
45. Half-dry places do not breed hope.
46. 0 beehive, your honeycombs haven't got a chance to fill up.
47. So many hands are carrying containers and knives
48. And are in line to harvest in the basket (to empty it).
49. They no longer even respect the lansart (?)19 season.
50. He who has grown fond of delights lives in anxiety.
51. He is condemned continually to carry around a water pipe and water
52. To fill its china vessel.
53. Even if you are entitled to your share, it will not be spared ;
54. It will not remain where you left it.
55. About the land where we were born, I swear, 0 brother,
56. That we left it against our will.
57. There is still no path to get there;
58. The path which Robas had built, a flood ruined it.
59. There is no serenity in a mind inhabited by passion .
60. Can one ever forget one's place of origin,
61. Even if one does not live there,
62. Even if there is nothing to be done against ill luck?
63. I had a deep desire to settle there and live in peace,
64. But I don't find there a base for my undertakings.
65. Where is a school to be found for my children ?
66. A doctor to look after me, I don't find one there either.

19 This word is unknown to ME, and so are those followed by an interrogation mark in
lines 73 and 84.
 ORATORICAL ENCOUNTER 357

67. It is on the child of the land that prosperity rests.

68. He will rehabilitate it, or we are done for.
69. A place abandoned by its owners is all ruin and desolation.
70. There, foundation stones do not stand fast.
71. Mountains are no different from plains and valleys,
72. Provided electric power reaches the village entrance.
73. But the beehive is full of bagnzi: (?)
74. Which have eaten up what the bees made ; furthermore crys can be
heard.
75. Bees, the smoke of burned bran pursues you.
76. Those of you who go out on their quest are caught out by the winds;
77. Their wings get caught in the hedges .
78. Their cries have reached the Master of the sky.
79. 1 avoid slopes too remote for rain to reach them .
80. 1 scrutinize the seasons, in wait for the opportunity
81. When 1 find someone to redeem my patrimony.
82. Do you know on what occasion the farmer loses half of his harvest?
83. This happens on the day he goes to the market and the takings of his
sale are small .
84. A bad alwa (?) only brings ills
85. If divine reward does not compensate for earlier losses.
 APPENDIX FOUR

FIVE ASHLHIY TUNES

The Ashlhiy tunes given below are appropriate for singing the lines used
as examples in Chapter 4 and the whole pieces presented in Appendices
II and III.
Tune I is that of the song by !Rqiya Tandmsirt transcribed in Appendix
II. It can also be used to sing the lines in examples (19) and (20) in § 4.5.
Tune 2 was composed by Lhazz Bl'iid. It can be used for singing the
lines in Appendix III.
Tune 3 can be used for singing the lines in examples (30) and (33) in
§ 4.6 .
Tune 4 can be used to sing the lines in example (41) in § 4.7.
Tune 5 is the tune to which the Imdlawn sing the winnowing song (46)
in § 4.8.

Except for the last, these tunes are appropriate carriers of two-line stanzas .
They are made of two halves , one for each line in a stanza. The two halves
are rhytmically parallel.
The numbers under the notes in the scores indicate the text-to-tune align-
ment, e.g. '4' under a note indicates that that note must carry the fourth
syllable in a line sung to the tune under consideration; in singing the song
in Appendix II, for instance, the fourth note in Tune 1 must be associated
with gu in line I, with dl in line 2, and so on.
When there is no number associated with a note, that note must carry
the same syllable as the preceding note, e.g. in singing the song in Appendix
II, the third syllable in line I, (mar) must bear the last two notes in the
first bar of Tune I, and similarly the seventh syllable (zr) must bear the
first two notes in the third bar.
The bracketted 'i' represents a line-final stopgap [i] vowel which must
be sung to the corresponding note. That vowel is omitted from our tran-
scription. Some tunes require a line-final stopgap vowel while others do not.

359
360 APPENDIX FOU R

Tune 1

1 2 3 4 5 6 7 8 9 10 11 12 m

1 2 3 4 5 6 7 8 9 10 11 12 [il

Tune 2

2 3 4 5 6 7 8 9 10 [i] 2 3

~ ~ ~ ~
4 6 7 9 10 [i]

Tune 3

'ä#l# U
1 2 4 5 6 7 8 9 10 11

, ä#ltt@
12 [i) 1 2 3 4 5 6 7 8 9 10 11

-f äl'lä r12
p p
[i)
, 11
 FIVE ASHLHIY TUN ES 361

Tune 4

~ä~l~ij~~
I 234 5 6 7 8 9 10

~ äää1ä 'f ~
[i] 2 3 4 6 7 8 9 10 [i]

Tune 5

2 3 456 7 8 9 10 11 12
 APPENDIX FIVE

LIST OF VERBS WITH

IMPERFECTIVE GEMINATION

We list below all the triconsonantal verbs we have been able to find in
Imdlawn Tashlhiyt which resort to gemination in the imperfective (see
§ 5.2). The forms listed are perfective sterns. As a rule imperfective sterns
only differ from their perfective counterparts by the gemination of one
consonant. When there are additional differences, the imperfective is given
between parentheses. Such additional differences are only found in two
situations: (i) the initial consonant becomes a in the imperfective, and (ii)
the geminated consonant is one of those which regularly undergo strength-
ening when they are subject to morphologically-governed gemination:
w > gWgW, 'Y> qq, f> qWqW, !d> !tt.
The verbs are classified according to the sonority contours of their per-
fective sterns. Each sonority contour is characterized by a two-letter
sequence in which 'R', 'E' and 'F' respectively mean 'rising ' , 'even' and
'falling'. A letter characterizes the sonority slope between two adjacent
consonants. For instance the sonority contour of the sequence xtl is FR,
as the sequence xt has a falling sonority contour (x is more sonorous than
t) and the sequence tl has a rising contour (t is less sonorous than I).

1. Verbs which geminate the second segment in the imperfective

(RR)
!xmr 'ferment'
gW rnr 'hunt'
bzr (azzr) ' pluck (feathers)'
!dhr 'appear'
gzr 'slaughter (animal)'
'Y ml 'mould'
hml 'enjoy (someone's company)'
bxl 'be stingy'
ksm 'enter'
gzm 'slice off'

(ER)
!'Y fr 'forgive'
fsr (assr) 'spread'
!xsr 'be damaged'
zhr 'blaze up'
lhsr 'stop'

363
364 A P P E N D I X FI V E

bdr (addr ) ' rnention'

xzn 'hoard'

(FF)
!msd 'comb'
lmhd 'poison'
n-yd (nqqd) 'refine'
!ns d ' be happ y'
rsq ' be happ y'

(FE)
!rxs ' be cheap'
msx ' metamorphose'
mh s 'vocalize (in writing Arabic)'
lnsh 'advise'
nzn 's ucceed'
nsf ' scrape (skin)'
!ngd ' drown'
ftk 'sprain'
zbd 'pull '

(FR)
19r 'lock'
lwr (lgWgW r) ' flee '
mgr 'harvest'
ndr 'rnoan'
!ndr (l nttr) 'jump'
nfr ' blow one's nose'
nkr 'get up'
nsr 'graze (skin)'
!str 'protect'
!zbr 'prune'
zgr 'go acro ss '
lh gr 'underestimate'
zdr 'burn '
rwl (rgWgwl) 'flee'
!rdl (!rttl) 'borrow'
rgl 'lock'
!rn l 'move (house)'
!mdl (Iattl) 'bury'
msl (ass l) 'plug'
nzl 'spur'
ntl 'take shelter'
 LIST OF VERBS WITH IMPERFECTIVE GEMINATION 365

xtl 'feint'
!rzm 'open'
lrsm 'mark'
lrzm 'stone'
lrhm 'be merciful'
rkm 'rot'
rgm ' insult'
lkm 'reach'
Ihm 'solder'
ndm 'regret'
!ndm (!nttm) 'compose (poem)'
nzm 'remain unharmed'
zdm 'gather firewood '
hkm 'have within one ' s reach '
!stn 'bother '
l' wbn 'lash'
rks 'hide'
stl' ' split'
zdl' 'inhabit'
!lbz ' squash'

2. Verbs which geminate the first consonant in the imperfective

(RF)
lhrm 'be forbidden by religion'
krm 'be dried out'
trm 'shimmy down'
frn 'sift'
krf 'tie up'
!"'{rf (!qqrf) 'flatten (dough)'
frs 'be sharp'
!l'rs (!qqrs) 'slit the throat'
mrz ' wound in the head'
krz 'plough'
!qrs 'reopen (wound)'
frs 'deceive'
hrs 'feel slightly ill'
!hrs 'be smart'
!lid 'be angry'
!srb 'have diarrhoea'
!krd 'scrape'
!srd 'sue'
frd 'graze (animal)'
!frd 'clear a field'
366 APPENDIX FIVE

!'Yrd (!qqrd) 'lie down'

frk 'gue ss'
srk 'own with others'
frg 'fence in'
srg 'have a miscarriage'
!hrg 'burn '
!zlm 'peeI'
zlf 'singe'
zlx 'be dirty'
kls 's lash (meat)'
qlb 'knock out'
hlb 'eat (liquid food)'
!xld 'mix'
kWms ' tie into a neat bundle'
krrnz 's cratch'
smt 'dupe'
smd 'add'
hmd 'thank (God)'
krrnd 'wither prematurely'
!'Y ns (!qqns) 'lose a bad habit'
knd 'dupe '
!qnd 'be bored'
xng 'choke'

(EF)
fsd 'be spoiled'
 REFERENCES

Abdel-Massih, Ernest T., 1968: Tamazight Verb Structure; A Generative Approach.

Bloomington and The Hague: Indiana University and Mouton .
Abdel-Massih, Ernest T., 1973: An Introdu ction to Moroccan Arabic. Ann Arbor: University
of Miehigan .
Abu-Salim, Issam, 1980: 'Epenthesis and geminate consonants in Palestinian Arabie', Studies
in the Lingu istic Seiences 10(2), I-lI.
Abu-Salim,Issam, 1982: 'Syllable structure and syllabifieation in Palestinian Arabie', Studies
in the Lingui stic Seiences 12(1), 1-28 .
AI-Fassi, Mohamed , 1986: rubaa 'iiyaatu nisaa 7i faas (al- 'iuruwbiyaat) [Women's quatrains
from Fez]. Casablanca: '1uyuunu I-maqaalaat.
AI-Fassi, Mohamed , 1997: Ma'ilamatu l-malhuwn: mi7atu qasiidatin wa qasiidahfiy mi7ati
vaaniyatin wa vaan iyah. [The encyclopedia of malhun : 101 poems about 101 beautiful
women]. Rabat: Publications de I' Academie du Royaume du Maroc.
AI-Ghadi, Abdellatif, 1990: Moro ccan Arabic Plurals and the Organization 0/ the Lexicon ,
Dipl öme d'Etudes Superieures , Faculte des Lettres, Universire Mohamed V, Rabat.
AI-Jirari, Abbas, 1970: Al-qasiidah [The poem] . Rabat : edit ions Maktabat At-taal ib;
Imprimerie AI-?umniyah .
Al-Malhuni, Abderrahman, 1990a: 7adabu lmuqaawamati min xilaali si'iri l-mal1iuuni wa
l-muraddadaati sSifaahiyah. daaru I-manaahili li ttibaahati wa nnasri , [The literature of
resistance as instantiated in rnelhun poetry and oral tradition. Moroccan Ministry of
Cultural Affairs.]
Al-Malhuni , Abderrahman , 1990b: kitaabatu ssi 'iri ssa'ibiyi l-malhuwni: 7iskaaliyatun min
7iskaaliyaati l-Yinsaadi wa ttadwiyn [Writing down melhun poetry: a problem in singing
and transcribing] . diywaanu l-rnalhuwn : silsilatu ?abhaatin wa diraasaatin fi I-qasidati
zzazaliyati. Depot legal la Bibliotheque Generale de Rabat 1990-411.
ä

Amarir, Omar , 1975: assi'iru l-maghribii l-Yamaariivii [Berber poetry in Morocco] .

Casablanca: Imprimeries Daar I-kitaab.
Amimi, Abdeljebbar and Georges Bohas, 1996: 'Les formes nominales [CICC] et [CCfC]
en arabe marocain ou la persistance des schemes ' , Canadian Journal 0/ Linguistic s
41(1), 1-28 .
Anderson , Stephen R., 1974: The Organi zation 0/ Phonology. New York: Academic Press.
Anderson, Stephen R., 1978: 'Syllables, segments and the Northwest Caucasian languages ' ,
in Alan Bell and Joan B. Hooper (eds.), Syllables and Segments, pp. 47-58. Amsterdam :
North-Holland .
Applegate, Joseph R., 1958: An Outline 0/ the Structure 0/ Shilha . New York: American
Council of Learned Societies .
Applegate, Joseph R., 1970: 'The Berber languages ' , in Thomas A. Sebeok (ed.), Current
Trends in Linguistics, vol. 6: Linguistics in South West Asia and North Africa , pp. 586-661 .
The Hague: Mouton .
Archangel i, Diana, 1991: 'Syllabification and prosodie templates in Yawelmani' , Natural
Language and Linguistic Theory 9, 231-283.
Asid, Ahmad and Brahim Lachgar, 1996: 7an'iibar, 7amarg n usays: diywaan l-muhaawaraat
s-si'iriyat [Jousting with verse, poetry on the commons: a collection of poetieal exchanges].
Rabat: Depot legal la Bibliotheque Generale 846/96.
ä

Aspinion, R., 1953: Apprenons le berbere; initiation aux diale ctes chleuhs. Rabat: Editions
Felix Moncho.

367
368 REFERENCES

Aydoun, Ahrned, 1994: Musiques du Maroe. Casablanca : Editions Eddif.

Bader , Y. and M. Kenstowicz, 1987: 'Syllables and case in Kabylie Berber ', Lingua 73,
279-299.
Bagemihl, Bruce, 1991: 'Syllable structure in Bella Coola', Linguistie lnquiry 22, 589-646.
Basset, Andre, 1929: La langue herbere. Morphologie. Le verbe. - Etude de themes. Paris:
Leroux.
Basset, Andre, 1932: 'Note sur l'etat d'annexion en berbere' , Bulletin de la Societe de
Linguistique de Paris 33, 173-174.
Basset, Andre, 1945: 'Sur la voyelle initiale en berbere ', Revue Afrieaine , 82-88; reprinted
in Basset 1959, Articles de dialeetologie herbere, pp. 83-89, Paris: Klincksieck.
Basset, Andre , 1946: 'Le systeme phonologique du berbere', Comptes Rendus du GLECS
(Groupe Linguistique d 'Etudes Chamito-Semitiques) IV, pp. 33-36. Paris: Ecole Pratique
des Hautes Etudes.
Basset, Andre, 1949: 'Sur le participe berbere' , Comptes Rendus du Graupe Linguistique
d'Etudes Chamito-Semitiques V, pp. 34-36. Paris: Ecole Pratique des Hautes Etudes.
Basset, Andre, 1952. La langue berbere . Oxford: Oxford University Press.
Basset, Andre , 1952/1987: 'Sur la metrique berbere' , Etude s et documents berberes 2,
85-90.
Basset, Andre and Andre Picard, 1948: Elements de grammaire berbere (Kabylie/lrjen) . Alger:
'La Typo-Litho' et Jules Carbone\.
Beckrnan, Mary E. and Jan Edwards, 1990: 'Lengthenings and shortenings and the nature
of prosodie constituency ' , in Kingston and Beckman, eds., pp. 152-178 .
Bendjaballah, Sabrina, 1995: Aspeets du systeme verbal du berbere (kabyle), Dipl öme d'Etudes
Approfondies, Universite Paris 7.
Benhallam, Abderrafi, 1980: Syllable strueture and rule types in Arabie, Doctoral Dissertation,
University of Florida.
Benhallam , Abderrafi, 1990: 'Native speaker intuition s about Moroccan Arabic stress ', in
Jochen Pleines, ed., La linguistique au Maghreb, pp. 91-109. Rabat: Editions Okad.
Bentolila, F., 1969: 'Les modalites d'orientation du proces en berbere ' , La Linguistique 1,
85-96; 2, 91-11 I.
Berque, Jacques, 1955/1978: Struetures sociales du haut-atlas. Paris: Presses Universitaires
de France.
Billerey , Roger, 1999: 'Optimality Theory : Can Onset be dispensed with? Evidence from
I1okano and Imdlawn Tashlh iyt Berber , and new insights on Berber syllabification' .
Unpublished, University of California at Los Angeles .
Bloomfield, Leonard, 1933: Language. London: George Allen and Unwin.
Bohas, Georges, 1999: ' La logique du signe "vocalisateur" (mhagyänä) et du signe "accelera-
teur" (rnarhtänä) dans la phonologie du syriaque: une approche de la structure syllabique',
Bulletin d'Etudes Orientales LI, 127-149.
Boogert , van den, Nico, 1997: The Berber literary tradition of the Sous. With an edition
and translation of 'The Oeean of Tears' by Mohammad Awzal. Publications of the 'De
Goeje Fund' XXVII. Leyden: Nederlands Instituut voor het Nabije Oosten.
Boogert , van den, Nico, 1998: "La revelation des emigmes ", lexiques arabo-berberes des
XVlle et XVllle siecles (trans\. by Claude Brenier-Estrine from an unpub . original in
English) . Aix-en-Provence: Travaux et Documents de I'IREMAM n° 19.
Booij, Geert, 1995: The Phonology of Duteh. Oxford: Clarendon Press.
Bougchiche, Lamara, 1997: Langues et litteratures berberes des origines a nos jours;
bibliographie internationale . Paris: Ibis Press.
Boukous, Ahmed, 1987a: Phonotaetique et domaines prosodiques en berbere (parler taehelhit
d'Agadir, Maroe), These de Doctorat, Universire Paris 8.
Boukous, Ahmed, 1987b: Syllabe et syllabation en berbere, Awal 3, 67-81.
Boukous, Ahmed, 1995: Societe, langues et eultures au Maroe; enjeux symboliques. Rabat:
Publications de la Faculte des Lettres et des Seiences Humaines .
 REFERENCES 369

Boukous , Ahmed, 2000: 'L' amazighe: perte irreversible ou changement linguistique?' , in

Salem Chaker and Andrzej Zaborski (eds .), Etudes berberes et chamito-semitiques;
melanges offerts Cl Karl-G . Prasse , pp. 43-59. Paris and Louvain : Peeters.
Bounfour, Abdellah, 1984: Linguistique et litterature: etudes sur la litterature orale maro-
caine, These de Doctorat, Universire Paris 3.
Brenier-Estrine, Claude , 1994: Bibliographie berbere annotee, 1992-1993, Travaux et
Documents de I'IREMAM. Aix-en-Provence: CNRS, Universires d' Aix-Marseille-Il-Ill.
Brenier-Estrine, Claude, 1995: Bibliographie berb ere annotee, 1993-1994 , Travaux et
Documents de I'IREMAM. Aix-en-Provence : CNRS, Universires d' Aix-Marseille-II-III.
Brenier-Estrine, Claude, 1995-1997: 'Chronique bibliographique berbere", Annuaire de
I'Afrique du Nord XXXIV, 1257-1296; XXXV, 1037-1082; XXXVI, 669-726. Paris:
CNRS Editions .
Broselow, Ellen, 1995: 'Skeletal positions and moras', in Goldsmith (ed.), pp. 175-205 .
Browman , Cather ine P. and Louis Goldstein, 1989: ' Articulatory gestures as phonological
units', Phonology 6, 201-25 I.
Browman, Catherine P. and Louis Goldstein, 1990: 'Tiers in articulatory phonology, with
some implications for casual speech', in Kingston and Beckman (eds.), pp. 341-376.
Bynon , J., 1978: "The internal reconstruction of Berber vowels and semivowels' , in P.
Fronzaroli (ed.), Atti del secondo congresso interna zionale di linguistica camito -semitica,
Firen re, 16-19 aprile 1974, Universitä di Firenze : Quaderni di semitistica 5.
Cadi, Kaddour, 1981: Le verbe en tarifit (Maroc-nord) : formes , structures et valences,
Doctorat de Troisieme cycle, Universire Paris 3.
Camps , Gabriel , 1984: 'Avertissement' , in Gabriel Camps (ed.), Encyclopedie berb ere I.
Aix-en Provence: Edisud.
Cantineau, Jean, 1950: 'Reflexions sur la phonologie de I'arabe marocain ' , Hesperis 37,
193-207; reprinted in Cantineau 1960, pp. 241-255 .
Cantineau, Jean, 1960: Etudes de linguistique arabe. Paris: Klincksieck.
Catford, John C; 1977: Fundamental Problems in Phoneti cs. Edinburgh: Edinburgh University
Press.
Chafik, Mohamed, 1990: al-mu'iiamu l-Yarabiyu l-?amaa ziyyiy [Arabic-Berber dictionary] ,
vol 1. Rabat: Publication s de I' Acadernie du Royaume du Maroc.
Chafik , Mohamed, 1991: ?arba Yatun wa ?arba 'iuuna darsan fiy lluvati 1- Zamaaziy-viyati
(naliwun wa sarfun wa stiqaq) [Forty four Berber lessons (syntax , morphology and
derivations)]. Rabat: an-nasru l-Yarabiyu I-?ifriyqiyu.
Chaker, Salem, 1975: 'Les parametres acoustiques de la tension consonantique en berbere
(kabyle)' , Travaux de l 'lnstitut de Phonetique d 'Aix 2,151-168.
Chaker, Salem, 1984: Textes en linguistique berbere. Paris: Editions du CNRS.
Chaker, Salem, 1992: Une decennie d'etudes berberes (1980-1990). Bibliographie crit ique.
Langue, litterature, identite. Algiers: Editions Bouchene.
Chaker, Salem, 1994: 'Chleuh (linguistique/litterature)' , in Gabriel Camps (ed.), Encyclopedie
berbere XIII, 1926-1933. Aix-en Provence: Edisud.
Chaker , Salem and Abdellah Bounfour, 1996: Langues et litteratures herberes; chroniques
des etud es XIII (1994-1995). Paris: L'Harmattan.
Chami, Mohamed, 1979: Un parler amazigh du Rif maro cain, Doctorat de Troisierne Cycle,
Universire Paris 5.
Chomsk y, Noam, 1986: Knowledge of Language: Its Nature , Origin and Use. New York:
Praeger.
Chomsky , Noam and Morris Halle, 1968: The Sound Pattern of English . New York: Harper
and Row.
Chtatou, Mohamed, 1982: Aspects of the Phonology of a Berber Dialeet of the Rif, Doctoral
Dissertation , SOAS, University of London.
Clements, George N., 1990: "The role of the sonority cycle in core syllabification', in Kingston
and Beckman (eds.), pp. 283-333 .
370 REFERENCES

Clements, George N., 1997: 'Berber syllabification: Derivations or Constra ints?' , in Iggy Roca
(ed.), Derivations and Constraints in Phonology, pp. 289-330. Oxford , Clarendon Press.
Clements, George N. and Elizabeth V. Hume, 1995: 'The internal organization of segments',
in John Golsdmith (ed.), pp. 245-317.
Clements, George N. and Susan R. Hertz, 1996: 'An integrated approach to phonology and
phonetics' , in Durand and Laks (eds.), pp. 145-175.
Cohn , Abigail C., 1993 : 'Nasalization in English: phonology and phonetics' , Phonology
10,43-81 .
Coleman, John , 1996: ' Declarative syllab ification in Tashlhit Berber' , in Durand and Laks
(eds.), pp. 177-218.
Coleman, John, 1999: 'The nature of vocoids associated with syllabic consonants in Tashlhiyt
Berber' , in John J. Ohala et al. (eds .), Proceedings of the XIVth International Congress
of Phonetic Sciences , vol. I, pp. 735-738. Berkeley : Universi ty of Califomia.
Coleman, John, 200 I: "The phonetics and phonology of Tashlhiyt Berber syllabic consonants',
Transaetions of the Philological Socie ty 99, 29-64.
Colin , G.S., 1985: 'Apercu linguistique' : section VII in entry 'AI-maghrib', Encyclopedie
de l'Islam (Nouvelle edition) . Leiden : Brill.
Cornul ier, Benoit de, 1995: Art Poetique . Lyon : Presses Universitaires de Lyon.
Dallet, J.-M., 1982: Dictionnaire kabyle -francais. Paris : SELAF.
Davenson, Henri, 1955: Le livre des chansons. Neuch ätel: La Baconniere.
Oe Premare, Alfred-Louis and Assia Alaoui , 1989: La litterature maghrebine: le malhun
ll; lexique, index. Ministere de I'Education Nationale, Centre National d'Education ä

Distance; Centre de Vanves.

Deli , F., 1984: 'L' accentuation dans les phrases en francais ' , in Deli , Hirst and Vergnaud
(eds .), pp. 65-122.
Deli, F., 1989: 'Concordances rythmiques entre la musique et les paroies dans le chant; l'accent
et l'e muet dans la chanson francai se' , in Mare Dominicy (ed.), Le souci des apparences ,
pp. 121-136. Bruxelles: Editions de l'Universite de Bruxelles.
Deli , F. and M. Elmedlaoui, 1985: ' Syllabic consonants and syllabification in Imdl awn
Tashlhiyt Berber' , Journal of Afri can Languages and Linguistics 7, 105-130.
Deli, F. and M. Elmedlaoui, 1988: ' Syllabic consonants in Berber: some new evidence',
Journal of Afri can Languages and Linguistics 10, 1-17 .
Deli, F. and M. Elmedl aoui, 1989: 'Clitic ordering , morphology and phonology in the verbal
complex of Imdlawn Tashlhiyt Berber ' , part I, Langues Orientales Anciennes Philologie
et Lingui stique 2, 165-194.
Deli , F. and M. Elmedl aoui , 1991: ' Clitic ordering , morphology and phonology in the verbal
complex of Imdlawn Tashlhiyt Berber, part 11' , Langue s Orientale s Ancienne s Philologie
et Lingui stique 3, 77-104.
Deli, F. and M. Elmedlaoui, 1992: 'Quantitative transfer in the nonconcatenative
morphology of Imdlawn Tashlhiyt Berber', Journal of Afroasiatic Languages 3, 89-125.
Deli, F. and M. Elmedlaoui, 1996a: ' On consonant release s in Imdlawn Tashlhiyt Berber',
Linguistics 34, 357-395.
Deli, F. and M. Elmedlaoui, 1996b: 'Nonsyllabic transitional vocoids in Imdlawn Tashlhiyt
Berber' , in Durand and Laks (eds.), pp. 219-246.
Deli, F. and M. Elmedlaoui, 1997a: 'La syllabation et les geminees dans la poesie berbere
du Maroc (dialecte chleuh)' , Cahiers de Grammaire 22, 1-95 .
Deli, F. and M. Elmedlaoui, 1997b: 'Les geminees en herbere", Linguistique Africaine 19,
5-55.
Deli, F., Daniel Hirst and Jean-Roger Vergnaud (eds.), 1984: Forme sonore du langage. Paris :
Hermann.
Deli, F. and A. Jebbour, 1991: 'Phonotactique des noms voyelle initiale en berbere (chleuh
ä

de Tiznit, Maroc)' , Linguistic Analysis 21, 119-147.

Deli , F. and Abdelkrim Jebbour, 1995: ' Sur la morphologie des noms en berbere (chleuh
 REFERENCES 371

de Tiznit, Maroc)', Langues Orientales Anciennes Philologie et Linguistique 5-6,

211-232.
Deli, F. and Oufae Tangi, 1992: 'Syllabification and empty nuclei in Ath-Sidhar Rifian
Berber' , Journal of African Languages and Linguistics 13, 125-162.
Deli, F. and Oufae Tangi, 1993: 'On the vocalization of Irl in Ath-Sidhar Rifian Berber' ,
Linguistica Communicatio V (1-2), 5-53.
Destaing, Edmond, 1920: Etude sur la tachelhit du Sous I; vocabulaire francais-berbere.
Paris: Leroux.
Destaing , Edmond, 1937: Textes arabes en parler des Chleuhs du Sous (Maroc). Paris:
Geuthner.
Durand, Jacques and Bemard Laks (eds.), 1996: Current Trends in Phonology: Models and
Methods . University of Salford, European Studies Research Institute.
Durand, Olivier , 1994: Profilo di Arabo Marocchino. Rome: Universitä degli studi 'La
Sapienza' .
Durand, Olivier, 1995/96: 'Le vocalisme bref et la question de I'accent tonique en arabe
marocain et berbere' , Rivista degli Studi Orientali LXIX, 1-2 (1995) [sie], pp. 11-31.
Rome: Bardi Editore 1996 [sie].
EI Mejjad, Khadija, 1985: Le parler de Marrakech; quelques aspects prosodiques, Doctorat
de Troisieme Cycle, Universite Paris 7.
Elmedlaoui, M., 1985: Le parler berbere chleuh d'Imdlawn (Maroc) ; segments et syllaba-
tion , Doctorat de Troisieme Cycle, Universire Paris 8.
Elmedlaoui, M., 1988: 'De Ja gemination' , Langues Orientales Anciennes Philologie et
Linguistique 1, 117-156 .
Elmedlaoui, M., 1993: ' Gemination and Spirantization in Hebrew, Berber and Tigrinya: a
"fortis-lenis module" analysis', Linguistica Communicatio V (1-2), 121-176 .
Elmedlaoui, M., 1995a: Aspects des representations phonologiques dans certaines langues
chamito -semitiques . Rabat: Universire Mohammed V, Publications de la Faculte des Lettres
et des Seiences Humaines.
Elmedlaoui, M., 1995b: ' Geometrie des restrictions de cooccurrence de traits en semitique
et en berbere: synchronie et diachronie', Canadian Journal of Linguistics 40(1), 39-76.
Elmedlaoui, M., 1999: Principes d 'orthographe berbere en graphie arabe ou latine. Oujda:
Publications de la Faculte des Lettres et des Seiences Humaines d'Oujda n° 25.
Frampton, John, 1999: 'SPE extensions, conditions on representations and defect driven mies',
to appear in Morris Halle and Bert Vaux (eds.), Phonological Perspectives: Rules and
Constraints in Contemporary Phonological Theory , Cambridge University Press.
Galand, Lionel, 1953: 'La phonetique en dialectologie berbere', Orbis 11/1 , 225-233 .
Galand, Lionel, 1957: 'The Rabat Institute of Advanced Moroccan Studies' , Report on current
research - 1957 . Washington: The Middle East Institute.
Galand, Lionel, 1975: section V, 'Langues' in the entry ' Berberes' , Encyclopedie de l'Islam
I, 1215-1220 . Leiden: Brill.
Galand, Lionel, 1979: Langues et litterature berberes; vingt cinq ans d 'etudes . Paris: Editions
du Centre National de la Recherche Scientifique.
Galand, Lionel, 1985: 'La langue berbere existe-t-elle?', in Melanges linguistiques offerts a
Maxime Rodinson, pp. 175-184, Paris: Geuthner.
Galand, Lionel, 1988: 'Le berbere', in Daniel Cohen and Jean Perrot (eds.), Les langues
dans le monde ancien et moderne. Troisieme partie: les langues chamito-semitiques,
207-242. Paris: Editions du CNRS.
Galand, Lionel, 1989: 'Les langues berberes', in Istvän Fodor and Claude Hagege (eds.),
Language Reform: History and Future, vol. IV, pp. 335-353 . Hamburg: Helmut Buske.
Galand, Lionel, 1997: 'Les consonnes tendues du berbere et leur notation', in Miloud Taffi
(ed.), Voisinage . Melanges en hommage a la memoire de Kaddour Cadi, pp. 99-120. Dhar
EI Mahraz - Fes: Publications de la Faculte des Lettres et des Seiences Humaines .
Reprinted in Linguistique Africaine 19 (1997), 57-77.
372 REFERENCES

Galand-Pernet, Paulette, 1967: 'A propos d'une langue litteraire berbere du Maroc: la koine
des chleuhs ' , in L. E. Schmitt (ed .), Verhandlungen des zweiten internationalen
Dialektologenkongresses; Zeitschrift für Mundartforschung, Beihefte, Neue Folge Nr. 3,
pp. 260-267.
Galand-Pernet, Paulette, 1969: 'Apostrophe, delirnitation prosodique de l'enonce et faits de
style en berbere', Cahiers Ferdinand de Saussure 25, 101-113.
Galand-Pernet, Paulette, 1972: Recueil de poemes chleuhs. Paris : Klincksieck.
Galand-Pernet, Paulette and Haim Zafrani, 1974: 'Sur la transcription en caracteres hebraiques
d'une version herbere de la Hagaddah de Pesah', in Daniei Cohen (ed .), Actes du Premier
Congres International de linguistique semitique et chamito-semitique, pp. 113-146. The
Hague: Mouton.
Goldsmith, lohn (ed.), 1995: The Handbook of Phonological Theory. Oxford: Blackwell.
Goldsmith, lohn and Gary Larson, 1992: 'Local modeling and syllabification'. in Michael
Ziolkowski et al. (eds.), Papers from the 26th Regional Meeting of the Chicago Linguistic
Society, Volume 2, The Parasession on the Syllable in Phonetics and Phonology, pp .
129-141, Chicago: Chicago Linguistic Society.
Guerssel, Mohamed, 1976: Issues in Berber Phonology , MA thesis, University ofWashington.
Guerssel, Mohamed, 1977 : 'Constraints on phonological rules', Linguistic Analysis 3,
267-305.
Guerssel, Mohamed, 1983: 'A phonological analysis of the construct state in Berber',
Linguistic Analysis 11, 309-330.
Guerssel, Mohamed, 1985. 'The role of sonority in Berber syllabification', Awall, 81-110.
Guerssel, Mohamed, 1986: 'Glides in Berber and syllabicity' , Lingui stic Inquiry 17, 1-12.
Guerssel Mohamed, 1992 : 'The phonology of Berber derivational morphology by affixa-
tion' , Linguistic Analysis 22, 3-60.
Hafidi, Abdellah, 1996 : tayri d unkkid - Amour et amertume; poemes berberes. Rabat:
depöt legal 849/96.
Halle, Morris, 1992 : ' Phonological features' , in William Bright (ed .), International
Encyclopedia of Linguistics, vol. 3, pp. 207-212. Oxford: Oxford University Press.
Halle, Morris, 1995: 'Feature geometry and feature spreading', Linguistic Inquiry 26, 1-46.
Hammond, Michael, 1997: 'Optimality Theory and prosody', in Diana Archangeli and Terence
Langendoen (eds .), Optimality Theory. An Overview, pp. 33-58. Oxford: Blackwell.
Harreli, Richard S., 1962a: 'Consonant vowel and syllable in Moroccan Arabic ', in
Proceedings of the IVth International Congress of Phonetic Sciences, pp . 643-647,
Mouton: The Hague.
HarrelI , Richard S., 1962b : A Short Reference Grammar of Moroccan Arabic. Washington
D.C .: Georgetown University Press.
HarrelI, Richard S. and Harvey Sobelman (eds .), 1966: A Dictionary of Moroccan Arabic.
Washington, D.C.: Georgetown University Press.
Harris, James W., 1983 : Syllable structure and Stress in Spanish . Cambridge, Mass .: MIT
Press .
Harris, James W. and Ellen M. Kaisse, 1999 : 'Palatal vowels, glides and obstruents in
Argentinian Spanish' , Phonology 16, 117-190.
Harris, Zellig, 1942: 'The phonemes of Moroccan Arabic ', Journal of the American Oriental
Society 62, 309-318.
Hayes, Bruce, 1989: 'Compensatory lengthening in moraic phonology', Linguistic Inquiry 20,
253-306.
Hayes, Bruce, 1990: 'Diphthongisation and coindexing' , Phonology 7,31-71.
Haye s, Bruce, 1995: Metrical Stress Theory . Chicago: University of Chicago Press .
Hayes, Bruce and Abigail Kaun , 1996: "The role of phonological phrasing in sung and chan ted
verse", The Linguistic Review 13, 243-303.
Hayward, R.J ., 1988 : 'In defence of the skeletal tier', Studies in African Linguistics 19,
131-172.
 REFERENCES 373

Heath, Jeffrey, 1987 : Ablaut and Ambiguity; Phonology of a Moroccan Arabic Dialect.
Albany: State University of New York Press .
Heath, Jeffrey, 1989: From Code-switching to Borrowing: Foreign and Diglossie Mixing in
Moroc can Arabic. London and New York : Kegan Paul International.
Huffman, Marie K., 1993 : 'Phonetic Patterns of Nasalization and Implications for Feature
Specification' , in Marie K. Huffman and Rena A. Krakow (eds.) , Nasal s, Nasal ization and
the Velum, pp . 303-327. New York : Academic Press .
Hyman , Larry, M., 1985: A Theory of Phonological Weight . Dordrecht: Foris .
Inkelas , Sharon and Young-mee Y. Cho, 1993: 'Inalterability as prespecification' , Language
69, 529-574.
Jackendoff, Ray, 1997: The Architecture ofthe Language Faculty . Cambridge, Mass.: MIT
Press .
Jebbour, Abdelkrim, 1988: Processus de formation du pluriel nominal en tamaz ight (tachelhit
de Tiznit) , Memoire de D.E.S ., Departement de Langue et de Litterature Francaises,
Universire Mohamed V, Rabat.
Jebbour, Abdelkrim, 1995 : ' Mores et poids prosodique en berbere ', Langues Orientales
Anciennes Philologie et Linguistique 5-6, 167-192.
Jebbour, Abdelkrim, 1996: Morph ologie et contraintes pro sodiques en berbere (tachlh it de
Tiznit). Analyse linguistique et traitement automatique. Doctorat d'Etat, Universire
Mohamed V, Rabat.
Jebbour, Abdelkrim, 1999: 'Syllable weight and syllable nuclei in Tachelhit Berber of Tiznit' ,
Cahiers de Grammaire 24, 95-116.
Jouad, Hassan, 1983 : Les elements de la versification en berbere marocain tamazighi et
tachelhit. Doctorat de Troisierne Cycle , Universire Paris 3.
Jouad, Hassan , 1986: ' Metres et rythmes de la poesie orale en berbere marocain' , Cahier s
de poetique comparee 12, 105-127.
Jouad, Hassan, 1987 : ' Les tribulations d'un lettre en pays chleuh' , Etudes et documents
berberes 2, 27-41.
Jouad, Hassan, 1989: 'Le langage Im' fna: l'esthetique de I'implicite', Etudes et documents
berberes 6, 158-168.
Jouad, Hassan , 1990: 'La metrique matricielle: nombre, perception et sens ' , Bulletin de la
societe de linguistique de Paris LXXXV(l), 267-310.
Jouad , Hassan , 1995: Le calcul inconscient de l'improvisation . Paris-Louvain: Peeters.
Jouad, Hassan and Bernard Lortat-Jacob, 1982: 'Les modeles metriques dans la poesie de
tradition orale et leur traitement musical', Revu e de musicologie 68, 174-197.
Kaye, Jonathan, 1987: 'The Case of Moroccan Arabic ', The Lingui stic Review 6, 131-159.
Kaye, Jonathan D. and Jean Lowenstamm, 1984 : 'De la syllabierte', in Deli , Hirst and
Vergnaud (eds.), pp. 123-159.
Keating, Patr icia A., 1990: 'Phonetic representations in a generative grammar' , Journal 0/
Phonetics 18, 321-334.
Keegan, John M., 1986: "The role of syllabic structure in the phonology of Moroccan Arabic ',
in Gerrit J. Dimmendaal (ed .), Current Approaches to African Linguistics, pp. 209-226.
Dordrecht: Foris .
Kenstowicz, Michael, 1986 : ' Notes on syllable structure in three Arabic dialects', Revue
Quebecoise de Linguistique 16, 101-128.
Kenstowicz, Michael, 1994a: Phonology in Generative Grammar. Oxford: Blackwell.
Kenstowicz, Michael, 1994b: ' Syllabification in Chuckchee: a constraint-based analysis' , in
Alice Davidson et al. (eds .), Proceedings ofthe Formal Linguistics Society of Mid-America
4, 160-181. Iowa City : Department of Linguistics.
Kenstowicz, Michael, 1996: 'Base-Identity and Uniform Exponence: alternatives to cyclicity',
in Durand and Laks (eds .), pp . 365-395 .
Kirn, Hyunsoon, 1994: 'A phonetic characterization of release/nonrelease: the case of Korean
and English', Paper presented at the Lingu istic Society of America.
374 REFERENCES

Kim, Hyunsoon, 1995: 'The status of release and nonrel ease in Korean: phonologieal or
phonetie ?', Paris: Institut de Phonetique, unpublished.
Kim-Renaud, Young-Key, 1986: Studies in Korean Linguistics. Seoul: Hanshin .
Kingston, John and Mary E. Beckman (eds .), 1990: Papers in Laboratory Phonology I,
Between Grammar and the Physics of Speech. Cambridge: Cambridge University Press.
Kiparsky , Paul, 2000: ' Opacity and cyclieity ', The Linguistic Review 17, 351-365.
Kossmann, Maarten G., 1994: Grammaire du berbere de Figuig , Doctoral Dissertation,
University of Leiden.
Kossmann, Maarten G., 1995: ' Schwa en berbere' , Journal of African Languag es and
Linguistics 16, 71-82.
Kouloughli , Djamel-Eddine, 1978: Contribution Cl la phonologie generative de l'arabe: le
systeme verbal du parler arabe du Sra (Nord Constantinois, Algerie). Doctorat de
Troisiem e Cycle, Universire Paris 7.
Ladefoged, Peter and lan Maddieson, 1996: The Sounds of the World' s Languages. Oxford :
Blackwell .
Leben, William R., 1980: 'A metrical analysis of length', Linguistic Inquiry 11, 497-509.
Leguil , A., 1981: 'Le schema d'incidence en berbere ' , Bulletin des Etudes Africa ines 1(1),
35-42.
Lerdahl , Fred and Ray Jackendoff, 1983: A Generative Theory of Tonal Music. Cambridge,
Mass.: MIT Press.
Levin, Juliette , 1985: A Metrical Theory of Syllabicity, Doctoral Dissertation, Massachusetts
Institute of Technology .
Levin, Juliette, 1987: ' Between epenthetic and excrescent vowels (or What happens after
redundancy rules)', in Megan Crowhurst (ed.), Proceedings of the West Coast Conference
on Formal Linguistics, vol. 6, pp. 187-201. Stanford: The Stanford Linguisties Association .
Liberman, Mark and Janet Pierrehumbert, 1984: ' Intonational invarian ce under changes in
pitch range and length ' , in Mark Aronoff and Riehard T. Oehrle (eds.), Language Sound
Structure ; Studies in Phonology Presented to Morris Halle by his Teacher and Students,
pp. 156-244. Cambridge : MIT Press .
Lortat-Jacob, Bernard, 1980: Musique et fetes au Haut-Atlas . Paris: Mouton and Ecole des
Hautes Etudes en Seiences Sociales .
Louali, Naima and Gilbert Puech, 1994: 'Les consonnes "tendues" du berbere: indices per-
ceptuels et correlats phonetiques ' , Etudes et documents berberes 11,217-223.
Louali, Naima and Gilbert Puech, 1999: 'La syllabe en berbere tachelhit ', in Sophie Wauquier-
Gravelines and Sandrine Ferre (eds.), Syllabes. Ilemes Journee s d'Etudes Linguistiques,
pp. 35-40. Universire de Nantes.
Louali, Nauna and Gilbert Puech, 2000: 'Etude sur l' Implementation du schwa pour quatre
locuteurs berbe res de tachelhit' , XXI/femes Journees d 'Etude sur la Parole, Aussois,
19-23 juin .
Lowenstamm, Jean, 1991: 'Vocalic length and centralization in two branches of Semitic
(Ethiopic and Arabie)', in Alan S. Kaye (ed.), Semitic Studies in Honor of Wolf Leslau
on the Occasion of His Eighty-fifth Birthday, vol. 11, pp. 949-965. Wiesbaden:
Harrassowitz .
Malone , Joseph L., 1996: 'Orthometrie patterns and phonetic interfacing: a reply to Manaster
Ramer ', Phonology 13, 119-125.
Marantz, Alec, 1982: 'Re reduplication' , Linguistic Inquiry 13, 435-482.
McCarthy, John J., 1979: Formal Problems in Semitic Phonology and Morphology , Doctoral
Dissertation, Massachusetts Institute of Technology.
McCarthy , John J., 1981: 'A prosodie theory of nonconcatenative morphology ', Linguistic
lnqu iry 12, 373-418.
McCarth y, John J., 1986: 'OCP effects: geminat ion and antigemination' , Linguistic Inquiry
17, 207-263.
 REFERENCES 375

McCarthy , John J. and Alan Prince, 1993: 'Generalized alignment', in Geert Booij and Jaap
van Marle (eds.), Yearbook of Morphology 1993, pp. 79-153 . Dordrecht: Kluwer.
McCarthy, John J. and Alan S. Prince, 1995: 'Faithfulness and reduplicative identity' , in
Jill N. Beckman et al. (eds.), Papers in Optimality Theory, University of Massaehusetts
Oeeasional Papers i8, pp. 249-384. Amherst: GLSA, UMass.
McCawley, James D., 1967: 'Le röle d'un systerne de traits phonologiques dans une theorie
du langage' , Langages 8,112-131.
Mestaoui , Mohamed , 1996: rrays lliaii BIYid; liayaatuhu wa qasaa?idu muxtaaratun min
si'lrih [The Rays EI-Haj BI'Iid; life and colle cted poems] . Casablanca: Imprimeries
Matbahat AI-naiaah al-iadiydah.
MitchelI, Terence F., 1957: ' Long consonants in phonology and phonetics' , in Studies in
Linguistie Analysis, pp. 182-205. Oxford , Blackwell.
MitchelI, Terence F., 1993: Pronouncing Arabie, vol 2. Oxford : Oxford University Press.
Nespor, Marina and Irene Vogel, 1982: ' Prosodie domains of external sandhi rules' , in van
der Hulst and Smith (eds.), vol I, pp. 225-255.
Odden , David, 1988: 'Ant i antigemination', Linguistie lnquiry 19,451-475.
Ouakrim,Omar, 1993: Fonetica y Fonologia del Bereber, Doctoral Dissertation, Universitat
Autönoma de Barcelona.
Ouakrim, Omar, 1995: Fonetica y Fonologia del Bereber . Cerdanyola des Valles : Universitat
Autönorna de Barcelona , Servei de Publicacion s.
Paradis, Carole and Darlene Laf'harite, 1997: ' Preservation and minimality in loanword
adaptation ', Journal of Linguisties 33, 379-430.
Pellat, CharIes, 1987: ' Malhün', Encyclopedie de l'islam (Nouvelle edition). Leiden: Brill.
Penchoen, Thomas G., 1973: Tamazight of the Ayt Ndhir . Los Angeles : Undena Publications.
Perlmutter, David, 1995: ' Phonological quantity and multiple association' , in Goldsmith (ed.),
pp. 307-317.
Pierrehumbert, Janet, 1980: The Phonology and Phoneties of English Intonation, Doctoral
Dissertation, Massachusetts Institute of Technology.
Pierrehumbert, Janet, 1991: "The whole theory of sound structure' , Phonetiea 48, 223-232.
Prince, A., and P. Smolensky , 1993: Optimality Theory: Constraint Interaction in Generative
Grammar, Technical Report #2 of the Rutgers Center for Cognitive Science, Rutgers
University .
Puech, Gilbert and Naima Louali , 1999: 'Syllabification in Berber : the case of Tashlhiyt',
in John J. Ohala et al. (eds.), Proeeedings ofthe XIVth International Congress of Phonetie
Seiences. vol. I, pp. 747-750. Berkeley : University of California .
Rice, Keren D., 1987: 'On defining the Intonational Phrase: evidence from Slave' , Phonology
Yearbook 4, 37-59.
Ridouane , Rachid , 1999: La spirantisation dans un parler berbere du Maroe (parler ehleuh
de Haha), Diplöme d'Etudes Approfondies, Universire Paris 3.
Roux, Arsene , 1955: La vie berbere par les textes; parler du sud-ouest marocain (Tashlhit},
Ure partie : la vie materielle . Paris: Larose.
Rubach, Jerzy , 1993: The Lexical Phonology of Slovak . Oxford : Clarendon Press.
Saa, Fouad, 1995: Aspeets de la morphologie et de la phonologie du berbere parle dans le
ksar Zenaga Cl Figuig (Maroe). These de Doctorat, Universire Paris 3.
Sage y, Elizabeth c., 1986: The Representation of Features and Relations in Non-lin ear
Phonology , Doctoral Dissertation, Massachusetts Institute of Technology.
Saib, Jilali: 1974: 'Gemination and spirantization in Berber: diachrony and synchrony ', Studies
in Afriean Linguisties 5, 1-26.
Saib , Jilali, 1976: A Phonologieal Study of Tamazight Berber: Dialeet of the Ayt Ndh ir,
Doctoral Dissertation, University of California Los Angeles .
Saib, Jilali, 1977: "The treatment of geminates: eviden ce from Berber ' , Studies in Afri ean
Linguisties 8, 299-316.
376 REFERENCES

Saib, Jilali , 1978: 'Segment organization and the syllable in Tamazight Berber ', in Alan
Bell and Joan B. Hooper (eds.), Syllables and Segments, pp 93-103. Amsterdam : North-
Holland.
Sehuyler , Philip D., 1979: ARepertory of Ideas : The Music of the rwais, "nv ais ", Berber
Professional Musicians from Southwestern Morocco , Doctoral Dissertation, University
of Washington .
Seobbie, James M., 1993: 'Constraint violation and eonflict from the perspeetive of Declarative
Phonology', Canadian Journal of Linguistics 38, 155-167 .
Selkirk , Elisabeth 0 ., 1978: 'On prosodie structure and its relation to syntactie strueture',
in T. Fretheim (ed.), Nordic Prosody /I . Trondheim : TAPIR.
Selkirk, Elisabeth 0., 1981: ' Epenthesis and degenerate syllables in Cairene Arabie', in
Hagit Borer and Youssef Aoun (eds.), MIT Working Papers in Linguistics 3, 209-232.
Selkirk, Elisabeth 0., 1982: 'The Syllable', in van der Hulst and Smith (eds.), vo!. 11, pp.
337-383.
Selkirk, Elisabeth 0 ., 1984: Phonology and Syntax. The Relation Between Sound and
Structure. Cambridge, Mass.: MIT Press.
Selkirk, Elisabeth 0., 1990: 'A two-root theory of length', in Elaine Dunlap and Jaye Padgett
(eds .), UMass Occasional Papers 14, 123-171. Amherst , MA.: University of
Massachusetts .
Shaw, Patricia A., 1993: "The prosodie constitueney of minor syllabIes', in Proceedings of
the Annual Meeting of the West Coast Conference on Formal Linguistics (WCCFL 12).
University of California Santa Cruz.
Shaw, Patrieia A, 1996: "The non-nuclear status of syllabie obstruents in Berber'. Handout
of a talk given at the Linguistic Soeiety of America, January 5.
Shoul, Mostafa, 1995: Etude phonologique et acoustique de l'arabe du Maroc Oriental, These
de Doetorat, Universire Lava!.
Steriade, Donca, 1993a: 'Closure, release and nasal eontours', in Marie K. Huffman and Rena
A. Krakow (eds.), Nasals, Nasalization and the Velum, pp. 401-470. New York: Aeademic
Press.
Steriade, Donea, 1993b: 'Segments, eontours and clusters', in Andre Crochetiere et al.
(eds.), Proceedings of the 15th International Congress of Linguists, pp. 71-82. Sainte Foy:
Presses de l'Universite de Lava!.
Steriade, Donea, 1994: 'Complex onsets as single segments: the Mazateeo pattern' , in Jennifer
Cole and CharIes Kisseberth (eds.), Perspectives in Phonology, pp. 203-291. Stanford :
CSLI Publieations .
Stroomer, Harry, 1992: 'On religious poetry in Tashlhiyt', in La recherche scientifique au
service du developpement. Actes de la troisieme rencontre universitaire maroco-neer-
landaise, pp. 185-193. Rabat : Publications de la Faculte des Lettres et des Seienees
Humaines, Serie Colloques et Seminaires n° 22.
Stroomer, Harry, 1998: 'Dialect differentiation in Taehelhiyt Berber (Moroeeo)', in Mohamed
EI Medlaoui, Said Gafaiti and Fouad Saa (eds.), Actes du 1er congres Chamito-Semitique
de Fes, pp. 37-49. Fes: Faculte des Lettres et Scienees Humaines Sais-Fes.
Tahar, Ahmed, 1975: La poesie populaire algerienne (melhün) ; rythme, metre et formes .
Aiger : Societe Nationale d'Edition et de Diffusion.
Tangi, Oufae, 1991: Aspects de la phonologie d 'un parler berbere du Maroc : Ath-Sidhar (Rif),
These de Doetorat, Universire Paris 8.
Trigo, Loren, 1991: ' On pharynx-larynx interaetions' , Phonology 8,113-136.
van der Hulst, Harry and Norval Smith (eds.), 1982: The Structure of Phonological
Representations. Dordrecht: Foris.
Vanee, Timothy, J., 1987: An Introduction to Japanese Phonology. Albany: SUNY Press.
Vaux, Bert, 1998: The Phonology of Armenian. Oxford: Clarendon Press.
Vergnaud, Jean-Roger, Morris Halle et al., 1979: 'Metrical structures in phonology ' .
Unpublished , Massaehusetts Institute of Technology.
 REFERENCES 377

Willms, Alfred, 1962: 'Zur Phonologie des langen Konsonanten im Kabylischen' , Zeitschrift
fü r Phonetik, Sprachwissenschaft und Kommunikationsf orschung 15, 103-109.
Willms, Alfred, 1991: 'Beraber (linguistique)', in Gabriel Camps (ed.), Encyclopedie herbere
X, 1473-1475. Aix-en Provence: Edisud.
Yip, Moira, 1993: 'Cantonese loanword phonology and Optimality Theory' , Journal of East
Asian Linguistics 2, 26 1-291.
Youssi, Abderrahim , 1992: Gramm aire et lexiqu e de l 'arab e marocain moderne. Casab1anca:
Wallada.
Zec, 0 ., 1995: 'Sonority constraints on syllable structure', Phonology 12, 85- 129.
Zeroual , Chakir , 2000: Propos controverses sur la phonetique et la phonologie de l 'arab e
marocain . These de Doctorat, Universite Paris 8.
 INDEX

Square brackets indicate pages where terms and symbols are defined , rules
and constraints are stated, etc.

2 ( stoprelease) 146, 319 acceptabil ity judgement 89

: (portmanteau) 15 AD (lad!) complementizer 48, 194,341
! (dorsopharyngealization) 13, 340 adaptation 279, 282
+ (boundar y) 15 adjacent identical consonants (tautomor-
- (morpheme boundary) 15 phemic) 158-160,273,325
<Z> (Ath Sidhar) extrametricality 166, 171 adjective inflection 18
= (clitic boundary) 15 affix 72
# (word boundary) 15 affrication 60
# (nucleus in a hinge syllab1e) 244 Agadir 185
+ [159], 321 agreement 17, 18
@ [16],75, [136], [164-165] , [176], [230], aHwash 79, 335
[294] Ait Ayache Tamazight 44
- (gemination) 86 Ait Iraten Kabyle 39, 57, 176
Ait Said Rifian 172
la-I (augment) 30, 36, 37 Ait Seghrouchen Tamazight 25, 177, 199,
I-al MA (n, fs) 35 288, 308
lbu=1 32, 37 alternation
I!dd! realized as [!tt] 35, 55, 58, 67, 160 - length 124, 324
li-I (augment) 31, 37 - V - zero 25
li-I (v, 3ms) 23, 190, 191,210 - vocalic 25
li-I (v, prt) 23 Amazigh 5
l i=1 (prep) 100 Amimi, A. and G. Bohas 274, 279
lid-I (n, mp) 37 aorist 23, 24, 120
I-inl (n, fp) 29, 36 AR (lar/) impf 194
listt-I (n, fp) 37 Arabic 35
11-1 (n, prefix) 35, 76, 340 Arabic script 8, 80, 84, 335-338
Immu=1 32, 37 Armenian 77
In=1 (prep) 34, 46, 341 ashlhiy 7
I-nI (n, mp) 29, 36 Ashtuken Tashlhiyt 137
I-ni (v, prt) 23 assimilation 32, 34, 35,46,48, 125, 150, 153,
Irad! (fut) 194 154,156,157,160,215,234,339,347
Is-I (v, cau) 124 At Mangellat Kabyle 161
It-I (n, f) 29, 34 Ath Sidhar Rifian 42,121 ,153,159.163-173,
It-I (v, 2s, 2p) 24, 173 197,217,230,231,241,255
I-tl (n, fs) 29, 35, 172 AugDel31
Itt-I (v, impf) 24-26, 117, 173 augment [28], 29, 31, 32, 193,203,214,337
lu -I (bound state) 29, 191,213 augmentative 26, 36
lw -I (pronoun , m) 33, 37 Ayt Ndhir Tamazight 57, 85, 165, 176, 255

aa 66, 195,222 B(e)raber 6

ablaut 118 base [25], 26
absorption 169,238,329,331 - in causatives 124, 130, 133

379
380 INDEX

- secondary 26 consonants which differ only with respect to

basic hollow syllable [298] voicing 67
Bella Coola 71 constraint 98, I 14
Berber languages 5-6 constraints vs. sequentially-ordered rules 4
Billerey, R. 113 construct state 35
bound state 17,27,29-31,46,192,202,205, Contiguity constraint 218
214 contoid [14], 175
Bounfour, A. 85 cooccurrence restriction 37, 67, 276
coproduction analysis [179], 184
C (con sonant) 14 Core Syllabification I 16, 183
C-place 154 coronal 34, 35, 58, 61, 62, 64, 68, 125, 127,
CA 238, 279, 313, 335 147, 171, 202
categorial 62, 63, 136, 138 countable 26
causative 25, 124-134, 156
CausLength [128], 132 definite article 35, 233, 242
chameleon vowel 42,117, [118], 161, 190, degemination 93
192, 221 deletion 155, 173, 189, 194
chleuh 7 demonstrative 15
classicism 279, 280 derivational affix 27
clause 18, 194 derivational morphology 23, 26, 27, 35
Clements, N. 138 DETACH [105], 107, 109
clitic 15, 18, 22, 268 devoicing 24, 58, 66, 75,144,160-163,172,
- adverb 19,20 237, 279, 304
- dative 19,20 dialect I, [4,6], 28, 39, 41, 50, 58, 71, 76,
- imperative 19 121, 129, 159, 163, 165, 168, 169,
- list of pronouns 19 171, 173, 176, 177, 187,216,238,
- object 20 239,249,288, 302, 317, 340
- prepositional phrase 19, 20 diminutive 26, 36
- pronoun 19 direct ional 19,20
- verbal 19 dorsopharyngealization 13, 58-65, 233
clitic boundary [15], 191 duration 57, 59, 139, 218, 334
cliticizable preposition [19], 20 Dutch 56
closure 138, 141
closure vs. primary articulation 153, 156 e [165, 167,230], 232, [293]
coda 74, 90, 325 e-devoicing 167, 170, 171, 301, 329, 331
Coleman , J. 145, 178-187 eC rime 258
collective 26 EF-G sequence [104]
complementizer 18,48 emphas is span 59, 61-62, [63],68, 154
complex [30] emphatic phoneme 62
complex coda 93, 183 emphatic word 61
complex nucleu s 229, 257-261 English 14, 140,179,186,198,217,237,279
complex obstruent rime [109], 110 epenthetic consonant 28, 63, 64, 194
complex onset 98, 101, 183,244,254 epenthetic vowel 74, 77, 163, 173, 179,300,
compliant [108] 315
compound 37 epenthetic yod 189, 191, 194, 196, 256, 340
concatenati ve 27, 190 exception 29, 33, 35, 66, 67, 125, 127, 132,
configurational analysis of length 41 157,209,214
conjoined construction 23 excrescent 170
conjunction 18 expanded hollow syllable [298]
consonant [14], 51, 174 explo sion burst 16, 59, 136, 319
consonant cluster 72 expression 59
con sonant-initial noun 34-37 extralong 146, 147, 319, 330
consonantal invariance [53], 54, 175 extrametrical 171, 172
 INDEX 381

FAITH(SHORT) [206] Heath, J. 314, 323, 328

FaithAdapt [282] heavy 50, 85, 92, 93, 257
featural analysis of length 41 hiatus 16,30,97, 113, 189,211,293
feature bundle 40, 43, 53-55, 90, 98, 107, high voeoid 67, 91, 112, 190, 200
197,234 high vowel 17,55 ,56
feature geometry 41, 127 hinge syllable [244], 293
feminine 26, 29, 30, 34, 35, 171,202,205 hinged coda 92, 93, 257
Figuig 57, 58, 121, 129, 153, 168 hollow syllable [122], [244], 262
FinH [274], 281 homorganic [141]
FinL 285, [286], 327 homorganic noneontinuant 140, 166, 296
fortis 41
free state 17, 27, 28 I,U 56, 90, 91, 197,200,308
free variation 136, 146, 185, 191, 193, 202, IFDQ syllabifieation [115], 182, 183, 185
288, 328 Ifni Tashlhiyt 57, 71
Freneh 56,83, 116, 136, 137, 142, 192, 193, Igliwa Tashlhiyt 85
237, 282, 302 Imdlawn 10, 79, 176
frieative 57, 59, 111, 125, 127 Imdlawn Tashlhiyt 4,10,41 ,42,66,79,129,
friction noise 60 170, 172, 178
full segment [231] imperat ive 21, 24, 26, 120
full vowel [16], 75, 157, 160, 231, 337 imperfeetive 24, 25, 42, 221, 224
fusion 42, 43, 48, [150],159,168, 172, 173, imperfeetive gemination 25, 117-124, 134,
318,327 363
future 18, 48 individuative 36
infleetional morphology 23
G (glide) 14 initial vowel [28], 29
geminable verb [118] input to syllabifieation 91, 199,211 , 308
geminate 14,33-35, [41],41-58,92,93, 104, intereomprehension 6, 8, 238
111, 124, 152, 154, 159, 165, 230, intonation 14, 17, 167, 189,279
234, 320, 321, 336, 340 Intonational Phrase 298, 300, 301
geminate fission 42 InV 28
geminate inseparability [147], 233, 270, 272, IP-Final Epenthesis [300], 304, 307
273,325, 327, 331 Italian 14
geminating verb [118], 133 iy,uw 205, 206, 219, 220, 231
gender 17, 20, 27 IYT 19,20
generative phonology 3
genitive 33 Japanese 167
German 217 Jebbour , A. 122-130
glide [14], 16, 30, 34, 68 Jouad, H. 85, 88
- Freneh 83
- geminate 10,46, 55, 68, 191, 196, 218, Kabyle 5, 42, 121, 162, 163, 174
311 kernel [25],35,61 ,64,67, 134, 173, 174,
- underlying 17, 164, 196 [274]
- voealization 199,201 ,309,313 Korean 138
- vs. high vowel 56, 90, 192, 196, 308
glide gemination 204-215 I-assimilation 35
GlideFaith [199] labial dissimilation 55
gliding 16, 190-192,256,293 labialized eonsonant 13, 51, 154, 185, 230,
glottal 68 338
Guerssel, M. 177 lateral plosion 141 , 169
lexical representation 158, 170, 198, 199,321
H (high voeoid) 14 Lieit Consonantal Nuclei thesis [73], 78, 134,
Haha Tashlhiyt 9, 50, 57, 58, 162 179
HarrelI, R. 170, 322, 323 lieit parse 103, 105, 107, 109
382 INDEX

light 50, 85, 92, 93, 257 Obligatory Contour Principle 159
line 85, 92, 96, 254, 270 obstruent 112, 171, 172
line-level syllabification 246, 256 obstruent nucleu s 95, 111, 183
liquid 215 occlusivization 55
Lmnabha 4, 11, 230, 239, 288 onset 74, 90, 92, 325
loanword 28, 34, 35, 65, 75, 76, 158, 160, onsetless syllable 96, 254, 270
168, 170-172, 282 optimal parse 105, 107
local maximum of sonority 77, 100 ordinal 33
long 14,39 orthometric syllabification 85, 182,241 ,253,
long closure 152-154 254
long primary articulation [152], 153 OT 98
long vowel 14, 59, 86, 191 Oujda II
Louali , N. and G. Puech 57, 178 Oujda MA 228, 239, 317
lowering 55
palatalization 63
MA 36, 59, 69, 75,165,168-171 ,217 Palestinian Arabic 165
MA 's influence on Tashlhiyt 8 paradigm 19, 214
margin [90], 197 - noun 27, 28
masculine 26, 29 - PNGs and clitic s 19
melHun 3, 85, 249, 288 - verb (total: 42 forms) 26
melodi c unit 45, 49, 53, 54, 98 participie 22-24, 26
metrical pattern [85], 88 passive 25
midsagittal 138 pause 143, 144, 149, 155, 157, 164, 167, 183,
MINIMAL-PATH(place) [141], 146, 148, 186, 197,203,254,279,299
297, 320 perfective 23, 24
MINIMAL-PATH(voice) [141], 296, 304 pharyngeal 68
Mon-Khmer 71 phoneme 13, 62
mora 124, 125 phonetic implementation 72, 143, 152, 179,
morpheme boundary [15] 181, 187,295,307,308, 333
music 79 - vs. phonological component 138, 141,
156,180
nasal plosion 140, 169, 238 phonetic inertia 141
negation 21 phonetic representation 15
negative 24, 25 phonetic target 138,141 ,143,152, 153,317
NO-OVERLAP [143], 146 phonetic transcription
no-schwa environment [180], 185 - broad 15, 16,60,72, 165, 166,201,232
NO-TREBLE [45], 48, [155], 156,318,320 - intermediate 16
NoCoda [270] - narrow 16, 60, 73, 145, 179, 307
NoHiatus [92], 100, [196], 199, [270] - of isolation forms 16
NoLoneSchwa [270], 294 - u-fronting 69
nominal inflection 2, 17, 27 phonological component 149
nominal morphology 26-37 Phonological Utteran ce 100, 270, 302
nonconcatenative 25, 27, 36, 190,223 - defined 254
nonemphatic word 61 phonotactics 174, 254
NoOns- [105], 107, 116, 206, 220, 327 phrase-final lengthening 297
NoPICOR [111], 183 PIAug 31
NoRR [102], 113, [200], 215-217, 276-284, plural 27, 31, 36, 37, 61, 67,192
315 PNG [17], 19-20,24,26
Northwest Caucasian 71 poetry 79
Northwe st Pacific Coast 71 possessive 33
Ntifa 121 post-onset @ [294], 303
nucleus 71, 74, 83, 90, 197 potential hv [190], 200
number in nouns 17, 26 prefix 20, 24, 157,328
 INDEX 383

prepausal backing 59 sonorant46,59, 144, 169,216,300,301,316

preposition 19, 22, 27 sonority 181
preverb [18], 20, 194 sonority peak [100]
prirnary artieulation [151] sonority plateau 100, 109, 112
prirnary base 25 sonority scale [76, 98], 112, [276]
prirnary gender 27 Sonority-Driven Syllabification thesis [73],
prirnary rirne [258], 261 74, 77, 119, 134
pronoun 18 SonPeak 99, [100],102,113, [196],198,199,
prosodie position 40, 43, 45, 49, 51, 54, 89 204, [263],277,306
putative vowel [241], 302 - asyrnrnetry in 77, 99
Pword [15], 156, 165, 191,230,254 Sous 7, 11
Spanish 56
(R)AD 48, 194 speech tempo 190, 192, 195,308,314
reciprocal 25 spirantization 162, 168
relative c1ause 21-23 stable schwa 328
release 43, 61, [138], [142], 143, 146, 231, standard transcription [231], 232
317,324 state 17,27
- rnandatory 149 stern 25, 28, 29, 36, 115, 124, 130, 132, 163,
- optional 146-148, 156,319,343 214
- prohibited 149, 156 - basie 119, 120
- surface-contrastive 146, 147 - naked 24, 26, 27, 120, 164,236
release in heterorganie clusters 140, 142- 146, - nominal [27]
322 - verbal [24]
release in hornorganie clusters 140, 320 stern-level syllabification 210, 212-214
release in voieeless clusters 140, 319 stop 57, 111
resyllabifieation 211, 242, 246, 256, 293, 332 Stray Erasure 47
Rif 6 stress 14, 301
RIGHT-TO-LEFT SCAN [165], 169, 170, stretch [156], 159
[233,269] suffix 20, 24, 268
rirne 90, 92 sukuun 336
RirneSize 92 SVV [164],169,319
RIPI 179 - nonsegment 164, 241
root 26, 54, 61, [62],276 - syllabie 164, 241
Root node 40, 41, 45, 98, 148 SYLL [270]
root-and-pattern 54, 228 syllabic [14], 32, 56, [90]
!rrays 63, 80, 219, 335 syIlabie consonant 71-73
RW-RIME [216] syllabie parse [88, 93], 197, 200
syllabification dornain [100], 102
schwa 42, 76, 137, 162, 165, 180 syllable 90
secondary base 25, 26 syllable boundary 83
secondary gender 27, 36, 37 syllable count 78, 128, 192
secondary rirne [258], 259, 294, 332 syllable weight 92, 93, 122, 124,241,247,
sentence-level syllabification 210, 212, 213, 257
256
Shaw, P. 94, 96,110,119 Tahaggart 5
shlHa 7 Tahar, A. 85
shortening 195, 196, 340, 343, 346, 347 Tarnasheq 5
sibilant harrnony 55, 125, 127 Tarnazight 5-8, 42, 84, 121, 129, 162, 163
sibling [146],154,317,324 Tangi, O. 163, 170
SIBLING-RELEASE 148, 156, 158 Taqbaylit 5, 129
Skeletal slot 49, 89, 189,206, 234, 259, 334 Tarifit 6, 8, 162, 174
Slovak 56 Tashlhiyt 5-11 , 84, 176, 178
sonorancy 140, 141 ternplate 274, 285, 330, 331
384 INDEX

templatic 2, 51, 66, 130, 174, 223, 274 voiced 60

tense 39-41 , 56, 57 voiced pharyngeal 65-68, 222, 340
terminal representation 49, 63, 73, 77, 138, voiced vocoid 16, 138, 165, 178, 232, 296,
152,179,305,307,319,324 299
text-to-beat alignment 78, 81-84, 244, 248 - auditory detection 236, 238, 310
2fronttilde 94 voiceless cluster 10, 140, 141, 144, 166, 167,
TIRRUGZA word 174 178, 181,236,296
Tiznit Tashlhiyt 36, 57, 129, 174 voiceless fricative 60
transcription 177, 320 voiceless pharyngeal 55, 60
- Berber I, 163, 176 voiceless sentence 60, 72, 140, 166, 167,227
- four kinds used in this book 16 voiceless syllable 87, 96
- MA 177,227,323 voiceles s transitional vocoid 74, 138, 141,
- Tashlhiyt 162, 176 145
transitional vocoid 14, 15, 136, 137, 140,294 voicing 57, 60, 136, 138, 140, 141
- and syllable structure 144 voicing in VTVs 141, 142
- auditory detection 136, 143, 144,215,236 vowel [14], 16,29,34,51 ,59, 174
treble segment 45 vowel quality of VTVs 136, 139, 145, 204
Tuareg 5 vowel-initial noun [28]-34
VTV 16, 72, [135], 137, 141,236,240
U 201 - not a segment 72, 176, 180, 199
u-fronting 68-69, 201, 216, 313 - not represented in broad phonetic
UKRIS word 51, 53 transcriptions 16, 72
uncontroversial segment [135], 136
uncontroversial vowel [241], 242, 302 W-final [235], 288, 302
underly ing glide [91], 190,200,214,308, W-intemal [235], 238, 239, 241, 288, 302
317 WC-SYLL [204], 207
underlying vowel [200] WIIYI 31, 215
undominated 98, 102, 114, 217 word [15], 35, 58, 133, [230], 276, 302
uvular 183 word-level syllabification 246, 256, 267, 270
writing 5
V (syllabic vocoid) 14
vcd-vls sequence [160] y,w 200
verbal inflection 24 Yawelmani 165
verbal morphology 2, 23-26 Yuman 71
viable candidate [270], 295
vocoid [15], 83, [135] Zeroual, C. 277
vocoid SEGMENT [14] Zuara 57,71
 KluwerInternational Handbooks of Linguistics

1. L. Haegeman (00.): Elements ofGrammar. Handbook in Generative Syntax. 1997

ISBN 0-7923-4297-6; Pb 0-7923-4298-4

KluwerAcademic Publishers - Dordrecht I BostonI London