A nonlinear analysis of Arabic syllabic phonology, with special reference to Makkan

Material Information

A nonlinear analysis of Arabic syllabic phonology, with special reference to Makkan
Abu-Mansour, Mahasen Hasan
Place of Publication:
Gainesville, Fla.
University of Florida
Publication Date:
Physical Description:
xiii, 309 leaves ; 28 cm.


Subjects / Keywords:
Consonants ( jstor )
Lexical stress ( jstor )
Linguistic morphology ( jstor )
Linguistics ( jstor )
Nouns ( jstor )
Phonology ( jstor )
Rhyme ( jstor )
Syllables ( jstor )
Verbs ( jstor )
Vowels ( jstor )
Arabic language -- Dialects -- Phonology
Arabic language -- Phonology -- Mecca -- Saudi Arabia
Dissertations, Academic -- Linguistics -- UF
Linguistics thesis, Ph.D.
University of Florida. ( LCSH )
Spatial Coverage:
North America -- United States of America -- Florida


General Note:
Typescript. Dissertation
General Note:
General Note:
Includes bibliographical references (leaves 297-308).

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
000940992 ( ALEPH )
AEQ2526 ( NOTIS )
16663993 ( OCLC )

Full Text







Copyright 1987


Mahasen Hasan Abu-Mansour

With love and gratitude

I dedicate this study to my parents.



I would like to express my gratitude to Professor D.

Gary Miller, chair of my doctoral committee, for his helpful

comments, constructive criticism, and many hours of

discussion, especially during the final stages of this

dissertation. His constant, challenging comments made me

think twice before making claims or drawing conclusions;

I am thankful to Professor Jean Casagrande, member of

my committee, for his advice, helpful comments, and constant


I am indebted to Professor Alice Faber, member of my

committee, for her helpful comments and for enriching my

general understanding of the Semitic background of some of

the issues discussed in this study.

I am grateful to Professor Timothy Vance, member of my

committee, for his helpful comments and kind assistance. I

would also like to acknowledge Professor Vance's prompt

response and continued support even after his move to the

University of Hawaii. I deeply appreciate his participation

in the oral defense of my dissertation.

I am grateful to Professor Marie Nelson, member of my

committee, for her reading and commenting on this study.

I would like to thank Professor Abderrafi Benhallam

from the Faculte des Lettres, Rabat, Morocco, for reading

and discussing portions of this work with me during his 1986

visit to the University of Florida.

Special thanks go to Professor Haig Der-Houssikian for

his sincere advice and continual encouragement during my

stay at the University of Florida.

I am thankful to my colleagues at the University of

Florida, especially Shirley Cole, Maggie MacDonald, and

Nellie Sieller, and to my friend Fawn Painter, for making my

stay at the University of Florida pleasant.

I would also like to thank Mrs. Barbara Smerage for the

excellent job she did in typing this dissertation, despite

the difficulty of such a job.

My greatest debt is to my parents who have always

encouraged me to pursue my goals in spite of whatever

difficulties might arise. I am grateful to them for their

unfailing love, moral support, and above all their

willingness to put up with years of being away from home and

family in order to accompany me during my study in the

States. I cannot thank them enough.

My older brother Ali and his family offered constant

encouragement and moral support, and for that I am grateful.

Special thanks go to my nephew Waheed for his love and

thoughtfulness. His letters and telephone calls have always

been my day-brightners.

Finally, and most importantly, my debt is immeasurable

to the person who shared with me the ups and downs of

graduate life--my brother and closest friend Anas. I am

grateful to Anas for his unconditional love, constant

encouragement, moral support, stimulating discussions, and

his readiness to help me at the expense of his own time and

rest. To Anas I would like to say, 'You are a super





ACKNOWLEDGMENTS..................................... v

ABSTRACT............................................. xi


ONE INTRODUCTION..... ........................... 1


2.1 Introduction........................ 4
2.2 Tiers of Representation............. 8
2.2.1 The Segmental Tier................... 9
2.2.2 The CV-Tier...... ..... ............ .. 29
2.3 Constraints and Other Principles.... 35
2.3.1 Association Lines................... 35
2.3.2 The Independence of Autosegmental
Tiers................................ 40
2.3.3 The Obligatory Contour Principle.... 42
2.4 Notes ............................... 49

SYLLABIFICATION................. ............ 51

3.1 Introduction........................ 51
3.2 Syllable Constituency............... 54
3.3 Major Approaches to Syllable
Structure............................ 67
3.3.1 Kahn (1976)........................ 67
3.3.2 Lowenstamm (1981)................... 69
3.3.3 Cairns and Feinstein (1982).......... 70
3.3.4 Steriade (1982)..................... 71
3.4 Testing the Syllable Parsing
Theories...................... ...... 72
3.5 Rules of Syllabification in Arabic.. 79
3.5.1 Syllable Parsing Rules............... 79
3.5.2 'Syllable Structure Assignment and
Resyllabification................... 87
3.5.3 Ambisyllabicity..................... 94
3.5.4 The Sonority Hierarchy.............. 97
3.6 Notes............................... 99



4.1 Preliminary Remarks................. 101
4.2 Major Approaches to CVVC and CVCC
Syllables........................... 103
4.2.1 The Superheavy Syllable Analysis.... 105
4.2.2 The Degenerate Syllable Analysis.... 109
4.3 Superheavy Syllables Reconsidered... 116
4.3.1 Distributional Constraints........... 116
4.3.2 CVVC and CVCC Syllables in CA
and MA................................ 119
4.4 Deviations and Explanations.......... 129
4.4.1 High Vowel Deletion: CVVC Syllables
in Non-final Position............... 129
4.4.2 CVVC.C., Syllable Structure and the
OCP.................................... 144
4.5 Concluding Remarks.................. 153
4.6 Notes.. .............................. 155


5.1 Introduction........................ 159
5.2 Epenthesis.......................... 160
5.2.1 General Epenthesis.................. 163 A Linear Analysis of Epenthesis..... 170 An Alternative Analysis............. 174
5.2.2 Postpausal Epenthesis............... 180 The Relevance of Syllable Structure. 185 An Alternative Analysis............. 192
5.2.3 Prepausal Epenthesis................ 194 Context and Conditioning............ 197 Analysis and Exceptions............. 203 An Alternative Analysis............. 209
5.3 Vowel Shortening.................... 217
5.4 Epenthesis and Dative Gemination.... 237
5.5 Conclusion.......................... 251
5.6 Notes............................... 253

AND SYLLABLE STRUCTURE....................... 256

6.1 Introduction........................ 256
6.2 Compensatory Lengthening in
Generative Linear Phonology.......... 257
6.3 Compensatory Lengthening in
Nonlinear Phonology.................. 260
6.4 Glottal Stops and Syllable Structure. 262
6.4.1 CV? Syllables........................ 262
6.4.2 CV?C and CVC? Syllables.............. 270
6.5 Compensatory Lengthening or
Assimilation..... .................. 275



6.6 Compensatory Lengthening, Exceptions,
and the Inalterability of Geminates.. 280
6.7 Conclusion.... ....................... 286
6.8 Notes................................. 287

SEVEN CONCLUSION.................................. 289

REFERENCES......................... .................. .. 297

BIOGRAPHICAL SKETCH..................................... 309

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment
of the Requirements for the Degree of
Doctor of Philosophy



Mahasen Hasan Abu-Mansour

May 1987

Chair: D. Gary Miller
Major Department: Linguistics

This dissertation deals with syllable structure and

superheavy syllables in Arabic, particularly Makkan Arabic

(MA). Classical Arabic is discussed whenever relevant. The

analysis of the syllable is nonlinear, that is, based on

recent advancements in autosegmental and metrical phonology.

The hierarchical view of the syllable with the minimal

onset/rhyme distinction is shown to be necessary. It is

argued that syllabification in MA consists of a universal

rule pairing CV sequences into CV syllables, plus a set of

language-specific rules that create branching rhymes and

superheavy CVVC and CVCC structures in word-final position.

The two best-known approaches to CVVC and CVCC syllables

are considered. On the basis of distribution and other

related phenomena, it is concluded that the facts of MA are

compatible with the superheavy syllable analysis of these


syllables. Superheavy syllables in MA are strictly

restricted to word-final position underlyingly; therefore,

the rule responsible for their syllabification has to be

restricted in application to that environment. The con-

straint against non-final superheavy syllables, however, is

relaxed in such a manner that high vowel deletion can yield

CVVC syllables in its output.

Two theoretical issues addressed in detail are (1) the

idea of allowing unsyllabified (extrasyllabic) consonants in

phonological representations, and (2) positing empty nuclei

under the degenerate syllable analysis. The first receives

considerable support in this study. It is argued that

phonological rules affecting syllable structure (epenthesis

and vowel shortening) are triggered by the presence of

unsyllabified consonants.

Epenthesis applies in three different positions: pre-

pausal, postpausal, and medial (general). Unsyllabified

consonants triggering general epenthesis result from poten-

tial non-final superheavy syllables, while those triggering

postpausal epenthesis result from initial consonant clusters.

Prepausal epenthesis is triggered by tautosyllabic consonant

clusters violating the sonority hierarchy, and is further

subject to the constraint that morphological classes may not

be merged. Problems with the analysis of unsyllabified

consonants as parts of degenerate syllables are discussed.


Vowel shortening (restricted to a certain morphological class

in MA) is also triggered by unsyllabified consonants.

An explanation for the phenomenon of 'dative gemination'

is attempted in terms of foot structure, syllable structure,

and the rhythmic cadences of the language.

Compensatory lengthening and assimilation are examined

both for the formal problems of their proper formulation in

the nonlinear framework and for their domain of application.

The problem of whether glides in rhyme position in MA are

eliminated by compensatory lengthening or assimilation is

discussed in the context of the MA-specific problem that

glides in certain phonological and morphological environments

are not affected.



The central thesis of this study is the so-called

superheavy syllable (CVVC or CVCC) and its relation to

syllable structure in Arabic, particularly Classical Arabic

(CA) and Makkan Arabic (MA). The study defends two general

claims. First, most of the phonological rules which modify

syllable structure in Arabic are closely related to this

syllable type. Second, an autosegmental representation of

the syllable offers a clear characterization of facts of

syllabification and resyllabification in MA and will

consequently affect the form of phonological rules.

The framework assumed in this study is that of

generative phonology enriched with the autosegmental

representations introduced in recent work by McCarthy

(1979a), Halle and Vergnaud (1979), and Clements and Keyser

(1983). The bulk of the analysis will be carried out in

autosegmental terms. Several references, however, are made

to the metrical approach (Liberman and Price 1977, Halle and

Vergnaud 1979, McCarthy 1979a, Kiparsky 1979, and Hayes

1980). A metrical explanation is expected for certain

phenomena, given the relevance of higher prosodic structure

to the overall hierarchy of speech.

In addition to outlining the basic principles of auto-

segmental phonology assumed in the following chapters,

Chapter Two singles out two (out of many) cases in the

morphology and phonology of MA that provide evidence for

autosegmental representations. The difference in the

realization of the eighth binyan in MA from that in CA is

better explained morphologically; the rarity of complex

segments is borne out by loan words. Chapter Three

introduces the rules of syllabification in MA, while Chapter

Four establishes the basic syllable types in MA and confirms

those of CA (McCarthy 1979a). The superheavy syllable may

occur underlyingly but is strictly restricted to final

position; exceptions are explained. In the chapter on

phonological rules, it is demonstrated that several rules

modifying syllable structure are triggered by potential

superheavy syllables in non-final position resulting from

the addition of new morphological material. The final

chapter provides evidence for syllable integrity and the

relevance of the superheavy syllable to this matter.

Two theoretical issues will figure prominently through-

out the analysis. The first concerns the presence of

unsyllabified (extrasyllabic, stray) consonants in the

output of basic syllabification. It will be demonstrated

that rules of epenthesis and vowel shortening crucially

refer to these consonants in their environments and

incorporate them into well-formed syllables.

The significance of the CV-tier in nonlinear represen-

tations is generally acknowledged. Besides its role in the

distinction of single and geminate segments, its presence is

indispensable to syllable-related notions like the charac-

terization of timing units and syllable integrity. The

analysis in the following chapters lends further support to

the CV-tier.

A note about the scope of this study is in order.

Since not all aspects of Arabic phonology are relevant to

the issues discussed, a full account of the phonology of

either MA or CA is not attempted. Evidence from other

dialects is cited, but no claim is made concerning the

relation between CA and any of the dialects. However,

implications with potential historical significance are

stated when relevant.


2.1 Introduction

In Classical generative phonology (Chomsky and Halle

1968), a phonological representation consists of a linear

sequence of segments, where segments are defined as

unordered sets of distinctive features. A linear represen-

tation implies two restrictions, known as the 'Bijectivity

Constraint' and the 'Integrity Constraint' (Poser 1982:122),

on the relation between feature matrices and segmental

positions. The bijectivity constraint insures that to every

segment there corresponds exactly one specification for each

feature and, conversely, that every feature specification

corresponds to exactly one segment. According to this

constraint, the partial deletion of a segment or the

insertion of an incomplete feature bundle is prohibited.

Rules that would give rise to segments violating the

constraint of 'one specification for every feature' are

disallowed by the integrity constraint. These constraints

also disallow representations where single feature

representations are shared by two or more segments, or where

a single segment is associated with two feature



2.1 Introduction

In Classical generative phonology (Chomsky and Halle

1968), a phonological representation consists of a linear

sequence of segments, where segments are defined as

unordered sets of distinctive features. A linear represen-

tation implies two restrictions, known as the 'Bijectivity

Constraint' and the 'Integrity Constraint' (Poser 1982:122),

on the relation between feature matrices and segmental

positions. The bijectivity constraint insures that to every

segment there corresponds exactly one specification for each

feature and, conversely, that every feature specification

corresponds to exactly one segment. According to this

constraint, the partial deletion of a segment or the

insertion of an incomplete feature bundle is prohibited.

Rules that would give rise to segments violating the

constraint of 'one specification for every feature' are

disallowed by the integrity constraint. These constraints

also disallow representations where single feature

representations are shared by two or more segments, or where

a single segment is associated with two feature


It was soon realized that 'subsegmental' and

'suprasegmental' phenomena, both of which violate the

constraints implied by linear representations, do occur in

many languages. Complex segments display properties which

are incompatible with the bijectivity constraint. In

English, for instance, the affricates /6/ and /'/ behave as

single segments and, yet, cannot be identified with a single

specification for the feature continuantt]. The phenomenon

of tone stability characteristic of many tone languages

whereby a vowel is deleted while its tone is realized on an

adjacent vowel is a violation, since it involves the

association of a single segmental unit with more than one

tonal unit.

In addition to the linear strings of segments, a

phonological representation in the traditional sense

includes different types of boundaries whose nature and

location are dependent on the morphological and syntactic

structure. The purpose of including these boundaries in a

phonological representation was to divide the segmental

string into substrings needed for the application of certain

phonological rules. No further grouping of segments into

hierarchical units, such as syllables, is provided for by

the theory. Nevertheless, the need for the syllable in

phonological representations became evident. The need in

the Sound Pattern of English, 1968 (henceforth SPE), to

recognize a distinction between 'weak' and 'strong' clusters

in the system of English stress is one instance where a

linear representation fails to capture the generalization

expressed by these clusters. Instead, the distinction

'weak' and 'strong' is an arbitrary one based on certain

sets of substrings particular to English. In addition, this

distinction could not be extended to other configurations

which were involved in rules other than the stress rules.

For instance, the same distinction would not account for

other rules such as flap formation, glottalization, and

r-deletion (Kahn 1976).

The combined result of the restrictions implicit in

linear representations and the lack of hierarchical units

beyond the segmental level was the clear need to modify the

traditional phonological theory in order to account for

language phenomena that fall outside the linear domain. In

fact, the theoretical modifications introduced to deal with

such phenomena do not claim to replace the traditional

linear representation completely; rather, they complement

and enrich this representation by recognizing new elements

proved to be necessary to phonological representations. The

result was the emergence of two lines of theoretical

pursuit, autosegmental and metrical phonology.

The basic assumption of autosegmental phonology is that

the standard one-tiered representation is to be split up

into several tiers, each constituting a linear arrangement

of elements. The elements of each tier are linked to each


other by association lines which indicate how they are to be

coarticulated. The autosegmental theory was originally

proposed to deal with tonal phenomena, which were

problematic to the standard theory (Goldsmith 1976, Williams

1976). It was then extended to deal with different kinds of

phenomena like complex segments, and vowel and consonant

harmony (Clements 1976). It is also possible to give

satisfying analyses of syllable structure and nonconcatena-

tive systems of morphology within autosegmental theory (see


Metrical phonology constitutes the other major modifi-

cation of the standard theory. It concerns the organization

of segments into larger units such as syllables and feet.

It was originally introduced as a theory of stress (Liberman

1974, Liberman and Prince 1977, Vergnaud and Halle 1979,

McCarthy, 1979a, Selkirk 1980, and Hayes 1980) but later

extended to cover other areas including syllable structure

(Kiparsky 1979) as well as consonant and vowel harmony.

The following is an outline of the basic assumptions

underlying the idea of separate tier representations. The

goal of this outline is to introduce the relevant aspects of

the general framework which will define the theoretical

background for the analysis presented in the following

chapters. No effort has been made to include in this

outline all aspects of the currently developing nonlinear

frameworks. The selective nature of this outline is obvious

from the exclusion of certain issues such as the treatment

of vowel harmony within the theory, and from the fact that

only occasional references are made to metrical structure.

Certain issues, however, will be the subject of further

discussion as they come up in the analysis.

2.2 Tiers of Representation

The basic insight of autosegmental phonology is that a

phonological representation is composed not of a single

linear sequences of entities, but of several simultaneous

sequences of entities or tiers. Each tier constitutes an

independent string of elements. For instance, the segmental

tier consists of a sequence of segments, the CV-tier of C

and V elements and so on.

In a nonlinear model sounds are organized into several

hierarchical units. First, segments are organized into

syllables by the universal and language-specific rules of

syllabification. Syllables and segments are mediated by a

CV-tier. The syllable has a constituent structure of its

own represented by an obligatory element, the 'Rhyme' (R),

and an optional 'Onset' (0). The Rhyme itself might be

further divided into two constituents, the obligatory

'Nucleus' (N) and an optional 'Coda' (C). Second,

syllables will be organized into higher metrical units,

i.e. metrical feet, according to some universal and

language-specific conventions. Third, metrical feet as well

as unfooted syllables will be grouped on the word structure

level. The overall model of the hierarchy of prosodic units

can be schematized as follows (McCarthy 1984:301):

word tier word

foot tier E

syllable tier a a a a a
Onset/Rhyme tier O R OR O R RO R

(Melody tiers) ? a k i -

However, only some of these tiers will be included in the

phonological representation of the examples presented in

this study. For instance, while the syllable and segmental

tiers as well as the tiers intervening between them will

often be indicated, the foot tier will be referred to only


2.2.1 The Segmental Tier

The segmental tier consists of a sequence of the

distinctive feature matrices familiar in linear phonology.

Formally, any feature may show autosegmental behavior, i.e.,

behave independently from all others, but there are some

substantial constraints on the number of tiers recognized in

languages. Most of these constraints are, in fact,

constraints on articulation, as is indicated by recent work

on this matter. Within the framework of dependency

phonology, Ewen (1980) considers the segment as an unordered

set of gestures, each of which is in itself an unordered set

of features. This concept of the segment is schematized by

van der Hulst and Smith (1982:24) as in (2):









glottal stricture


velaric suction

articulatory gesture

categorical gesture

initiatory gesture

The representation in (2) makes it possible to assume that

features which belong to different gestures may behave

independently, thus imposing some sort of constraint on

feature autosegmentalization.

Similar proposals were made within an autosegmental

framework. For instance, Thrainsson's analysis (1978) of


aspiration in Icelandic is dependent on the representation

of laryngeal and supralaryngeal features on separate tiers.

So is Steriade's analysis of h in Attic (1982). Steriade

argues that h is not a segment in the sense that it does not

take up a skeleton position by itself. This is supported by

the fact that the h-initial forms behave like vowel-initial

words in that they undergo resyllabification, e.g., domasin
h h -
emeterois 'in our home' --d domasi nemeterois (p. 155).

They also undergo contraction or elision, as in tetra ippos

'driving four horses' -- tet hrippos (p. 154).

The other basic question concerning the segmental tier

is as follows: Is it the case that only single features may

behave independently, or can clusters of features do so? In

other words, can complete segments be represented on

different tiers, and, if so, at what point in the derivation

does tier conflation take place?

With regard to the first question, the representation

of clusters of features on different tiers has shown

promising results in the analysis of nonconcatenative

systems of morphology such as that of Arabic. In his novel

extension of the notion of the autosegmental tier, McCarthy

(1979a, 1981b) has demonstrated that by splitting up the

segmental tier into three separate tiers, the consonantal

tier, the vowel tier, and the CV syllabicc skeleton) tier,

the rather complex processes of word formation in Arabic

could be accounted for in a simple and natural way.

Until the facts of Semitic were pointed out, it was

still possible to think of metrical and autosegmental

phonology as devices proposed specifically to deal with

tonal phenomena. The relevance of metrical and autoseg-

mental phonology to the analysis of nontonal systems has

been established in McCarthy's work on Semitic. He shows

very explicitly that certain striking characteristics shared

by the Semitic languages are quite analogous to phenomena

known in tone languages. This similarity, as demonstrated

below using one of McCarthy's examples, provides the

strongest argument in favor of separate tier representation

in the analysis of Semitic word formation.

The assumption underlying the separation of segmental

tiers is that particular languages may isolate certain sets

of distinctive features and represent them on a separate

autosegmental tier. The autosegmentalization of segmental

matrices is, however, subject to the constraint that

segmental matrices belong to different tiers if and only if

they belong to different morphemes (McCarthy 1979a, 1981b).

The following is a partial presentation of one of McCarthy's

examples, given here to clarify some of the issues involves

in nonconcatenative morphology.

The most salient characteristic of Arabic morphology is

that each verb may occur in several derivational classes or

'binyanim'. These indicate meanings such as the causative,

the habitual, the reciprocal, the iterative, and so on.

Within each binyan, the verb has a number of inflections

which denote aspect, voice, finiteness, and mood. This is

illustrated in the partial derivational paradigm of the root

ktb 'to write' given in (3) (McCarthy 1981b:385):1

Perfective Active
















Perfective Passive










All the forms in (3) share the stem consonants ktb, but

differ from one another with respect to the vowels and some

additional consonants. They also differ in their syllabic

skeletons, or prosodic templates in McCarthy's terminology.















The forms in (3) exhibit the following patterns (McCarthy






The rule in (5) accounts for the canonical distribution of

consonants and vowels in the binyanim (McCarthy 1981b:387):

(5) a. [({C })CV([+ seg.])CVC]
b. V --+ 0/[CVC--CVC]

Each pattern in (4) characterizes certain binyanim in (3).

For instance, the second binyan usually indicates the

meaning of the causative and is characterized by a prosodic

template of the form [CVCCVC], while the third binyan has

the meaning of the reciprocal and the template [CVCVVCVC],

and so on. The vowels carry the grammatical information

such as tense and voice. This role is exactly analogous to

that played by the tones in some tone languages, where it is

the tone sequences that carry the grammatical distinctions

in the verbal systems of those languages. The derivational

classes, however, are characterized by a prosodic template

consisting of the basic root consonants alone (as in I and

III), or, in addition to that, by an added consonant (a

prefixed t in binyanim V and VI and a prefixed n in VII), or


the reduplication of the second or third root consonants (as

in II and V, and IX and XI, respectively).

In order to account for this phenomenon of discontinu-

ous morphemes, McCarthy suggests that consonantal and

vocalic patterns be regarded as autosegments on separate

tiers and governed by the same principles observed in the

case of tonal melodies (Section 2.3.1). The derivation is

straightforward in the case of the prosodic templates where

the number of consonantal slots on the skeleton tier equals

the number of melodies (consonants) on the consonantal tier:

association is one-to-one, from left to right. As a result,

binyanim I and III will have the representations in (6a) and

(6b), respectively (McCarthy 1981b:388):

(6) a. I b. III


k tb k t
\V \V
1 u

The affixes in IV, V, VI, VII, and X are represented on

separate tiers since each belongs to a morpheme different

from the root. For instance, ? in IV is the causative

morpheme. Association starts by linking the affix with the

first consonant of the template. However, the affix in the

eighth binyan occupies the second consonantal slot rather

than the first. McCarthy suggests a Flop Rule restricted to

this binyan and its peculiar pattern. He formulates the

rule as follows (McCarthy 1981b:390):

(7) Eighth Binyan Flop

C C --+ C

t t


The rule changes the association of the affix t to the

second consonantal slot. It will not apply to the affix of

the fifth and sixth binyanim since in both cases the two

consonants are not adjacent as the rule requires.

Association, then, proceeds as usual. Since the infix t of

the eighth binyan is represented on a different tier, it

will not interfere with the rest of the association process.

The gemination of the last consonant in IX and XI is

accounted for by the principle of 'spreading'. The result

of initial association is given in (8) (McCarthy 1981b:391):

(8) a. IX b. XI


k tb k tb
V \V

Representations like the ones in (8a) and (8b) are not well-

formed. Spreading of the last melodic element to the


unassociated consonantal slot takes place, resulting in the

gemination of the last consonant (McCarthy 1981b:391):

(9) a. IX b. XI


k tb k t

1A V

Spreading of melodic elements other than the last consonant

would result in crossing association lines. It is to be

noted that the spreading of the last melodic element to the

last empty consonantal slot is the exact analog of the

spread of the last tone to all unlinked vowels in a tone


As for binyanim II and V, association followed by

spreading results in the gemination of the last radical.

Then, an Erasure Rule (McCarthy 1981b:392) disassociates the

final root consonant from the medial C. The new empty C is

now subject to the well-formedness condition and associates

with the autosegment associated with the nearest consonantal

slot, i.e. t.

The representation of elements belonging to different

morphemes on separate tiers has a further advantage in the

derivation of binyanim XII-XV, particularly XII and XIII.

These are comparatively rare--a fact which, according to

McCarthy, is captured by their forming a natural class in

the language. First, all of them are formed on the prosodic

template [CCVCCVC], and, second, they need a separate

stipulation for their peculiar affixal material: the

infixes w, n, and the suffix y. This extra statement

specifies where the affixes are to be associated on the

prosodic template, as in (10) (McCarthy 1981b:393):

(10) a. C C V C C V C b. C C V C C V C

n) y

The stipulation of the position of the affixes is followed

by left-to-right association to give binyan XV: ktanbay.

Binyan XIV requires the additional spreading of the final

consonantal melody: ktanbab. As for binyanim XII and XIII,

they require in addition to the stipulation in (10) and

spreading the Erasure Rule which is particular to the second

and fifth binyanim. The result of the initial association

followed by gemination is ktawbab for both binyanim. The

Erasure Rule now applies to ktawbab, giving the representa-

tion in (11) (McCarthy 1981b:394):

(11) P


k t b
cc cu

By virtue of representing w on a separate tier, reassocia-

tion can now be effected from the root second consonant t,

or from the infix w without violating the constraints on

association. The first association will give binyan XII,

ktawtab; the second will result in binyan XIII, ktawwab.

The final representations are given in (12) (McCarthy


(12) a. XII b. XIII


w w

ktb k t b

To recapitulate, the representation of material

belonging to different morphemes on separate tiers provides

an account of the highly elaborate morphology of the verb in

Arabic, where only a few statements are needed. Besides the

general principle of left-to-right association of consonants

and the set of prosodic templates and affixal morphemes,

only two rules are called for: the Erasure Rule and the

Flop Rule. The Erasure Rule is involved in the derivation

of several binyanim (the second, fifth, and thirteenth).

Its role might be seen as one of maximizing the expansion of

the template. (Compare the gemination of the last consonant

in the case of binyanim IX and XI and XIV.)

The Flop Rule, on the other hand, was provided mainly

to account for the eighth binyan, where the prefixed

morpheme switches places with the first consonant. No other

binyan seems to call for this rule, which might raise a

question as to its necessity. McCarthy states that rules of

this type are common in the tonal systems described by

Goldsmith (1976). In addition to the synchronic frequency

of such rules, data from Makkan Arabic, historically related

to the Classical, provide further evidence for McCarthy's

Flop Rule. The eighth binyan in Classical is the intransi-

tive reflexive or 'ergative' of the first binyan (e.g., Eng.

the vase broke). The following are illustrative:

(13) First Eighth
Binyan Binyan

a. faraq 'divide' ftaraq 'to become

b. saal 'light, staTal 'to break out
set fire' (fire)'

c. kasab 'earn' ktasab 'to earn a
living for

d. madd 'extend' mtadd 'to become

In verbs which have no form in the seventh binyan (the

passive), the reflexive meaning passes into the passive

(Wright 1971:42). These include verbs whose first radical

is one of the following consonants: ?, w, y, r, 1, or n.

Such verbs are exemplified in (14):


a. lasiq

b. nasar

c. radaT






'prevent' rtadaT

'to be /get
stuck; hang
on to'

'to become vic-
torious; to
side with

'to be pre-
vented; to

It is this group of verbs which is relevant here, i.e. verbs

where the reflexive and passive are not quite

distinguishable. In Makkan, the seventh and 'eighth'

binyanim of certain verbs seem to occur in free variation.

Consider (15):2


a. nkatab

b. nkawa

c. nkasa

d. nRabas

e. nTadal


a tkatab



a tFabas

O tladal

'to be written, to
become registered'

'to be burned
(ironed), to get

'to be dressed, to
get clothed'

'to be detained, to
be held back'

'to become straight'

Bakalla (1979) suggests that the forms in the second column

are derived from their respective forms of the seventh



binyan in the first column via a dissimilation rule. The

rule changes a prefixed n to t before roots starting with

nasals, liquids, glides, and true consonants. Indeed, this

rule is justified in the case of roots with initial n, 1, r,

y, w:



-- tnasar

----+ tlasa9

nrafaz --- trafaz

nyassar -- tyassar

nwassaT --- twassaC

'to be helped'

'to be stung,

'to be kicked'

'to become easy'

'to become wide'

It should be added that *nnasar, *nlasaT, *nrafaz, *nyassar,

and *nwassaT are all impossible. However, this is not true

in the case of the forms in (15) where both forms are

equally acceptable. Moreover, in (15) the prefix n- is

already dissimilar to k, 6, and T. To this is added the

fact that some of the verbs with first root consonant m tend

to behave more like the examples in (15) in that the n-forms

and t-forms are equally acceptable:

(17) a. nmadd

b. nmala



. tmadd

2 tmal(l)a

C tmaza)

2 tmanaT

'to be/become

'to be filled, to
become full'

'to be/become mixed'

'to be prevented,
to refrain'

(16) a.






For these reasons, I suggest that the dissimilation rule is

restricted in application to roots with initial n, 1, r, y,

w, and occasionally m. As for the forms in (15) and (17)

where n % t, the situation is different. My claim is that

the n-forms and the t-forms belong to two different

binyanim, namely, the seventh and the eighth, respectively.

In other words, the t-forms in both (15) and (17) represent

the eighth binyan of these verbs minus McCarthy's Flop Rule.

Without claiming any direction for the development of this

rule (whether the rule has been partially lost or partially

incorporated into the verbal system of MA), I hypothesize

that the difference between the eighth binyan in classical

Arabic and that of certain verbs in Makkan Arabic can be

defined in terms of this rule.3

In addition to its role in the analysis of verb

morphology in Arabic, the notion of separate tiers of

representation is further supported by certain

morphophonemic processes that operate on elements of a

single tier without affecting elements on adjacent tiers

(McCarthy 1982, 1985). McCarthy (1982) cites a language

game in Bedouin Hijazi Arabic discussed in Al-Mozainy (1981)

which freely transposes elements on the consonantal tier

only. Elements on other tiers and the canonical pattern of

the form are not affected. Thus, a verb like ?arsal 'sent'

may have one of the permutations in (18) (p. 197) (McCarthy


(18) ?arsal Base form

?aslar Game forms





But none of the alternations involve the prefix 2-, typical

of the fourth binyan. The game, then, affects the

consonantal root only. The same is true of inflectional

morphemes on verbs and nouns. Thus, in a verb like dafaTna

'we pushed' or a noun like miftaaF 'key', all permutations

are possible except those which involve the personal suffix

-na in dafaTna, or the prefix mi in miftaaF (p. 198):

(19) fidaTna mifFaat

Tafadna mitfaaF

*naTadfa *tiRmaaf

*finada *tifaam

In Hanunoo, as in Bedouin Arabic a language game

operates on the root tier only leaving unaffected the

prosodic template of a word (McCarthy 1982:199). But since

the roots in Hanunoo, unlike Arabic, are composed of both

consonantal and vocalic melodies, the game may exchange the

first and last consonant-vowel sequences in the stem with

one another, as in (20):

(20) Base Game

a. rignuk nugrik 'tame'

b. ?usah sa?uh 'one'

c. katagbu? kabugta?

However, in (20c), the heteromorphemic element ka is not

tampered with and the game form is kabugta?.

The significance of the Hanunoo game is two-fold.

First, it provides further evidence for the separation of

root and nonroot material on distinct tiers (cf. (20c)).

Second, it demonstrates that while the dichotomy of root

consonantal and vowel melodies is crucial to Arabic

morphology, it is not exhibited as such in Hanunoo.

Up to this point we have seen that the representation

of vowels and consonants on separate tiers enjoys

considerable support from Semitic-like morphological

systems. In section 2.3.3, we will see that the separation

of segmental tiers is required for the purposes of morpheme

structure conditions (the Obligatory Contour Principle).

Separate segmental tiers are also needed for at least some

early phonological rules. As an example of phonological

rules applying before tier conflation, McCarthy (1986:233)

cites a rule of Umlaut in Harga Oasis Arabic. The rule

applies to both copies of a spread stem vowel raising a to i

or i before i or u, respectively. The rule applies to stem

vowels only:

(21) a. yinzal 3rd m. sg.

tinzili 2nd f. sg.

tinzilu 2nd c. pl.

b. yikallam 3rd m. sg.

tikillimi 2nd f. sg.

tikillimu 2nd c. pl.

The rule is triggered only by the subject agreement suffixes

underlined in the data. The relevance of this example is

that none of the vowels of the imperfect personal prefixes

(ti- of the 2nd person plural in the last example of (21a)

and (21b)) is affected by the rule. Given the assumption of

separate segmental tier representations, this suggests that

the rule applies at a point where tier conflation has not

yet occurred; thus, the prefixes are still on their separate


Nevertheless, evidence also exists for rules that

require a linearized representation of the more familiar

type. For instance, Steriade (1986:138) argues that Semitic

Spirantization is such a rule. She discusses two versions

of this rule. Spirantization in Biblical Hebrew (Leben

1980, McCarthy 1981a) applies to all postvocalic stops; in

Tigrinya (Schein 1981, Kenstowicz 1982) it applies to

postvocalic velars and uvulars only. Separate tier

segmental representation can be justified for both languages

on morphological grounds similar to those in Arabic. The

derivation of Hebrew sibbeep 'to turn' and that of Tigrinya

raqqiX 'thin' serve as an illustration. The structure of

these forms before Tier Conflation is given in (22a) and

(22b), respectively:

(22) a. s b b. r q

i e a i

If Spirantization were to apply to (22), it would produce

*sippeep and *rayyiy. Also, the surface forms sibbeep and

ra'qqi would appear to involve a violation of the

'inalterability' principle (Hayes 1986), since both involve

a rule applying only to the last member of the long-distance

geminates in (22). However, if we assume that Tier

Conflation occurs prior to the application of

Spirantization, then the rule will apply to the structure

in (23):

(23) a. s i b e b b. r a q i q
I I A A I I /\ I

In (23) only the last consonant will undergo Spirantization.

The rule does not apply to the geminate configuration, since

only the first C in that configuration is in postvocalic

position. The application of Spirantization would produce

the unacceptable *sipbeep and *raXqix. However,

Spirantization applies to a sequence of nongeminate

consonants, as in /higdil/ > hiydil 'be magnified'. The

application of Spirantization to the sequence gd does not

involve a violation of the 'inalterability' mentioned above:

Consonants other than geminates have the structure

illustrated in (24) (after Tier Conflation):

(24) h i g d i 1
I I I Ii i

Therefore, Tier Conflation creates the linearized

representation required by later phonological rules. Tier

Conflation was proposed by Younes (1983) and elaborated by

McCarthy (1986). It has the function of discarding

morphological information at some point in the derivation.

In fact, McCarthy (1986) makes the claim that Tier

Conflation is to be identified with Bracket Erasure

(Kiparsky 1982a, Mohanan 1982, Halle and Mohanan 1985).

Bracket Erasure, as suggested in these studies, removes

morphological boundaries as they become inaccessible to

subsequent phonological rules. Tier Conflation generalizes

Bracket Erasure to nonconcatenative systems. McCarthy takes

no position as to where exactly Tier Conflation takes place;

rather, he concentrates on how it functions in different

analyses. In section 4.4.2 we will consider the implica-

tions of Tier Conflation and the OCP antigemination effect

for high vowel deletion in MA.

2.2.2 The CV-Tier

The segmental tier and syllabic organization are

mediated by an intervening tier which consists of the

abstract elements C and V. The role of the CV-tier in the

description of the nonconcatenative morphology needed for

certain languages has already been introduced. The other

area where the CV-tier plays a significant role is syllabic

phonology or CV phonology. The basic assumption of CV

phonology is that the property of syllabicity is separated

from the strictly segmental material and represented on an

autosegmental tier. Elements on the CV-tier are linked to

those on the 'melodic' (segmental) tier via association

lines in standard autosegmental fashion. The other signifi-

cant role of the CV-tier is that of characterizing the

elementary units of timing at the subsyllabic level. As in

autosegmental phonology, one-to-many and many-to-one

associations are allowed by the theory. By virtue of this

provision for the two possible types of association, the

dualistic behavior of geminates and complex segments can now

be represented adequately.

Recall that the standard definition of the segment

implies a constant state in the specifications of features.

This is violated, nevertheless, by the existence of

affricates and other complex segments. For instance,

affricates start as' [-continuant], and end as continuantnt].

In order to deal with this problem, the feature [delayed

release] was introduced. Phonologically, affricates count

as single segments but also have internal structure

comparable to that of sequences of segments. For instance,

in the German dialect of Zurich described in van Riemsdijk

and Smith (1973), affricates function as units underlyingly

but require separate reference to their continuant parts

later on.

In a theory with the units of timing represented on an

independent tier, the notion of complex segments is charac-

terized in terms of one-to-many association between a single

element of the CV-tier and a sequence of two elements on the

segmental tier:

(25) a. C b. C c. C

t s m b i e

affricatee) (prenasalized (short diphthong)

The monosegmental character of these segments is captured on

the CV-tier where they are represented as a single unit of

timing, while their bisegmental property is recorded on the

segmental level as two fully specified segments.

Support for representations in (25) can be drawn from

the dualistic behavior of Latin labiovelars discussed in

Steriade (1982). Steriade argues for the structures in

(26a) and (26b) as the representation for Latin labiovelars:

(26) a. C b. C

k u g u

In Latin, the labiovelars behave like single consonants in

that, unlike all other consonant clusters, they fail to

close a preceding syllable. On the other hand, they count

as a sequence of two consonants: They can occur only in

prevocalic position, in which case the second melodic unit

in the complex segment syllabifies with the following vowel.

This restriction follows from the fact that Latin word-

internal syllables obey the Sonority Sequencing

Generalization (pp. 18, 92). Under this generalization, any

sequence stop-w-consonant cannot be exhaustively

syllabified. Further evidence for the bisegmental character

of these labiovelars follows from their fate in preconsonan-

tal position. A labiovelar occurring before consonants may

lose its labial component, as in coctus 'cooked' underlying
koku-tos. The second option is for the labial glide to

syllabify as a vowel, as in lingula 'little tongue' under-
lying lingu-la. Steriade proposes two rules to account for

these two outcomes (deletion or syllabicity). The first

rule deletes the association between the labial component

and the segmental slot; the second inserts a V slot and

associates it with the segment u. It is hard to perceive

how a theory which fails to recognize the constituent parts

of these segments would handle the data without further

complications in these rules.

The other problem which the representation of

(sub)syllabic information on a different tier seems to solve

is that of geminatess' or 'long' segments. The special

behavior of these segments with respect to phonological

rules has been a well-known problem for the standard theory:

They behave sometimes like a single consonant and sometimes

like two consonants. The SPE framework allows two ways of

representing these segments: They can either be marked as

[+long], in which case they behave like a single segment, or

they can be represented as a sequence of two identical

segments when they count as two. However, several studies

have shown that neither representation can account for the

unique behavior of geminates (cf. Kenstowicz 1970, Pyle

1970, Sampson 1973, and Barkai 1974). Proposals for

modification are numerous in the literature, but most of

them are concerned with the phonological rules which apply

to geminates. Some of these proposals are constraints on

such rules, for instance, the 'Integrity Hypothesis'

(Kenstowicz and Pyle 1973), or its revision in Benhallam

(1979), and the 'Adjacency-Identity Constraint' (Guerssel

1977). Others try to specify what kind of rules would apply

to geminates, both underlying and derived (Benhallam 1980).

There are two general properties shared by these studies and

others done about the same time. First, they all maintain

the linearity hypothesis of representations, and, second,

they concentrate on phonological rules rather than the

representational aspect of geminate segments.

The representation of geminates in CV phonology is that

of many-to-one association between the CV-tier and the

segmental tier, where two C or V-slots on the CV-tier are

linked to a single melody on the segmental tier. This can

be schematized as follows:

(27) a. Long Consonant b. Long Vowel4


m i

As mentioned recently by Hayes (1986), Kenstowicz

(1970) noticed that rules which require the sequence

representation of geminates are prosodic in nature, while

those requiring the single representation are segmental

rules. A similar observation is made in Leben (1980)

concerning long consonants in Hausa. These consonants

pattern with single segments in a plural class but behave

like clusters with respect to syllable structure

constraints. Given the representation in (27), this split

in behavior can be accounted for. Since length is factored

out and represented on the CV-tier in terms of occupying two

slots, it is natural that long segments behave as such on

the same tier where syllable weight is defined, i.e. the CV-

tier. On the other hand, the monosegmental character of

geminates, as illustrated by their immunity to certain

segmental rules, is also accounted for: On the segmental

tier, geminates are single consonants.

Further, the representation in (27) predicts the

failure of certain rules such as epenthesis to apply to

segments having these structures: Any application will

result in crossing association lines (Kenstowicz 1982,

Schein and Steriade 1986):

(28) C V V C

1 m I

Finally, it follows from the above that the CV-tier

formally expresses the long recognized distinction between

light and heavy syllables. According to this distinction, a

light syllable ends in a short vowel, while a heavy syllable

in a long vowel, a diphthong, or a short vowel followed by a

consonant. Clearly the distinction concerns the nucleus and

the material following it, together called the 'rhyme'.

Given the function of the CV-tier in characterizing the

units of timing, we can now describe the distinction between

light and heavy syllables as that of 'nonbranching' vs.

'branching' rhymes. A light syllable contains a simple,

nonbranching rhyme expressed on the CV-tier in terms of

association to a single V unit. A heavy syllable, on the

contrary, has a complex branching rhyme, represented on the

CV-tier as a bipositional unit. The CV-tier also explains

why the rhyme of a heavy syllable can be one of the


following: a sequence of two short vowels (VV), a diphthong

(VG), or a short vowel followed by a consonant (VC). From

the point of view of the CV-tier, these sequences are

equivalent, regardless of the segmental content of the

matrices to which they are linked on the segmental tier.

In fact, there have been several proposals to replace the

C and V-slots with X slots (cf. Kaye and Lowenstamm 1982,

Harris 1985, Levin 1985, Schein and Steriade 1986). Thus,

the rhyme of a light syllable is represented on the 'X-tier'

by one X-slot, that of a heavy syllable by two slots--hence

the equivalence of VV, VG, and VC. Either way of represent-

ing the slots on this tier captures the distinction of heavy

vs. light as well as the structural equivalence of heavy

syllables regardless of the segmental values of the members

constituting their rhymes.

2.3 Constraints and Other Principles

2.3.1 Association Lines

Elements on different tiers are linked by association

lines, which indicate how these will eventually be

coarticulated. The association conventions are identical to

those operating in tonal systems. Goldsmith (1976) proposes

the following overriding constraints on association:

(29) Well-formedness condition

a. Every unit on one level must be associated

with at least one unit on every other level

b. Association lines may not cross

The first clause of (29) is both too weak and too strong.

It is too weak because there are different possible

associations that can satisfy (29a). In order to deal with

this indeterminacy, Haragauchi (1977) and Clements and Ford

(1979) propose several conventions to make the well-

formedness convention more specific. One of these

conventions is that mapping links elements on different

tiers on the basis of one-to-one from left to right.5

On the other hand, (29a) is too strong to allow

unassociated or dissociated elements, but the theory can be

interpreted as predicting the occurrence of such elements.

As a result, (29a) has been weakened in recent work

(Goldsmith 1979, Clements and Ford 1979, McCarthy 1979a, and

Halle and Vergnaud 1982). The crux of these proposals is

that elements are allowed to remain unassociated after the

left-to-right rule has applied, or to become dissociated,

and that (re)association should not be a matter of

convention. If these elements remain unassociated, they

will receive no phonetic realization on the surface.

In Section 2.2.2, we have seen that in addition to one-

to-one association, the theory of multi-tiered representa-

tion allows two other types of association between elements


on the CV-tier and those on the segmental tier: one-to-many

and many-to-one. The first type of association

characterizes complex segments, and the second designates

geminates. However, it is by language-specific rules that

one or both of these association types are excluded. A one-

to-many association between the CV-tier and the segmental

tier is prohibited for almost all varieties of Arabic

(McCarthy 1979a, 1981b). It follows from this that complex

segments do not occur in Arabic (but see below).

A prohibition against the one-to-many association

between the CV-tier and segmental elements is observed by

loan words in Arabic. The few loan quinqueliteral nouns

observe this constraint. In verbs derived from these nouns,

the last consonant always stays unassociated throughout the

derivation and eventually does not appear on the surface.

For instance, the verb maynat 'to magnetize' is derived from

the corresponding noun maynatiis 'magnet'. The left-to-

right association of a root like mynts gives (30) (McCarthy


(30) C C C C
m y nts

The example in (30) represents the result of one-to-one,

left-to-right association. Because in this case there are

more melodies than melody-bearing units, s stays unattached

and may not link to any of the already filled C-slots. s

does not receive any phonetic realization on the surface and

the correct form is derived.

Further evidence for the prohibition in Arabic against

a one-to-many association between units of the CV-tier and

melodies on the segmental tier (complex segments) comes also

from the behavior of the few English loans in MA. This

group of words includes forms which have either of the

complex segments c and j: check, ketchup, jug, and march

are pronounced [seek], [kesab], [zak], and [maaris],

respectively. These examples are different from maYnat in

two respects. First, the deleted element in the latter, s,

constitutes an independent melody, which would exclusively

associate to a prosodic slot, had one been available. On

the contrary, the deleted element, t or d, in these examples

does not form an independent melody (in the source language)

from the point of view of the CV-tier. In other words, the

deleted part of the complex segment in each case (t or d)

forms only a part of a prosodic unit on the CV-tier:

(31) a. C b. C

t s d z

Second, the deletion of t or d in check, ketchup, march, or

jug cannot be shown to be a consequence of the constraint on

maximum root consonants, as is the case with maYnat. None

of the borrowed examples has more than four consonants. It

is also clear that simplification of complex segments in

these words does not follow from the constraint against

syllable-initial consonant clusters. If it did, these words

could have undergone epenthesis to give *tViek, *kitVsab,

especially since epenthesis is common in the pronunciation

of words with consonant clusters (but not complex segments),

cf. ?iskuul or sikuul 'school'. Also, simplification of

complex segments through epenthesis will not exhaust the

limit on root consonants in Arabic since the resulting forms

will have either three or four consonants. It is also

interesting to note that in cases where epenthesis applies

to a loan word with a cluster, the simplification of the

complex segment also takes place; for instance, march is
v 7
pronounced as maaris, not *maaric. The evidence, then,

points to the reality of the structures in (31) and the

constraint against them even with respect to foreign words:

Epenthesis would obliterate the monosegmental structure of

the segment in the source language; a complex segment would,

on the other hand, violate constraints on the structure of

the segment in Arabic. The result is the substitution for

the complex segment of a single, nonbranching one.

The only case which seems to violate the prohibition

against the one-to-many association between the CV-tier and

segmental elements is the occurrence of /j/ in certain

varieties of Arabic. However, a consideration of the

synchronic as well as diachronic status of this particular

consonant supports in various respects the prohibition

against complex segments. First, there is a considerable

degree of variation in the pronunciation of this particular

consonant in Arabic. In MA and CA it is often pronounced

[5]; but [g], [z], and [Y] occur in other varieties as well.

Second, it is the case that even where [v] is heard, a

considerable amount of variation exists. For instance, in

MA alternation between [5] and [z] is quite common.

Consider jamal zamal 'camel', juu? U zuu? 'hunger', rijil

rizil 'foot'. Third, /3/ is the only complex segment in

the phoneme inventory of, at least, MA and CA; its voiceless

counterpart, /c/, does not occur. Fourth, if we adopt the

viewpoint that considers the modern dialects to be the

direct descendants of CA, the change of classical >

dialectical i can also be viewed as part of the tendency to

eliminate this consonant from the sound inventory of the

language. What is involved here is a change of a complex

branching consonant into a simple, nonbranching one, thus,

providing further evidence for the prohibition against

complex segments.

2.3.2 The Independence of Autosegmental Tiers

One of the constraints implied by the linear theory of

phonological representations is that in tone languages, both

segmental and tonal features should form an integrated

whole. This would mean that deletion rules could not


operate on one without affecting the other. The phenomenon

of contour tones, where vowels delete leaving their tones to

be realized on adjacent vowels is, nevertheless, a clear

indication that the two levels of representation are quite

independent. The independence of tiers of representation is

basic to the autosegmental approach, where each tier might

undergo rules that exclusively refer to elements on that

specific tier. In Etsaka (Clements 1979:100), as in many

tone languages, a vowel deletes and its tone appears on

another vowel. The expression 'each noun', for instance,

requires the reduplication of that noun:

(32) a. owa 'house'

b. owowa 'each house'

The underlying form of (32b) is given in (33):

(33) H L H L

o0 wa wa

The surface form is derived by deleting the first occurrence

of the vowel a. The deletion rule, however, leaves the tone

of the deleted vowel to be later realized on the following

vowel as a contour tone. The example also shows the

stability of the tonal tier.

The notions of independence and stability of auto-

segmental tiers extend to the CV-tier. Languages possess

rules that refer to the segmental elements along with their

positions on the CV-tier. Rules of epenthesis exemplify

this type. Languages may also have rules that refer

exclusively to elements on the segmental tier, without

reference to their position on the CV-tier. A classic

example would be rules of compensatory lengthening whereby a

segment deletes but with concomitant lengthening of another

segment. For instance, in Latin /s/ is voiced to /z/ and

later deleted before dental stops, causing compensatory

lengthening, e.g. */sisdo:/ > si:do: 'I sit', and /dis-

du:co:/ > di:du:co: 'I separate'. Compensatory lengthening

clearly violates the 'integrity constraint' on phonological

rules. It is a rule that deletes a segment but leaves its

temporal position, i.e. its slot on the CV-tier, intact. In

a theory where the segmental and subsyllabic material are

represented on one and the same level, it is hard to

represent the changes involved in a process like

compensatory lengthening.

2.3.3 The Obligatory Contour Principle

The last constraint relevant to autosegmental represen-

tation is the Obligatory Contour Principle (henceforth OCP).

The OCP, originally proposed by Leben (1973) for tonal

phonology, prohibits sequences of identical elements if they

can be replaced by a single multiply associated element. In

addition to tonal phonology, it has been demonstrated that

the OCP is operative in other areas of phonology and

morphology. Its relevance to the analysis of other than

tonal systems was examined in McCarthy (1979a, 1981b) where

it was presented as a distributional constraint on Semitic

roots. McCarthy argues more recently (1986) that the OCP

has an equally significant role throughout the phonological

derivation, as exhibited by processes of gemination and

antigemination in many languages including non-Semitic types

of languages. In what follows I present some of the

arguments offered by McCarthy (1979a, 1981b, 1986) in

support of the active role of the OCP in these areas.

The first piece of evidence for the role of the OCP in

governing lexical representations comes from the unique

explanation it offers for the absence in Arabic of geminate

roots with identical first and second radicals, e.g. *sasam.

This prohibition cannot be justified by phonetic reasons,

since identical (but heteromorphemic) consonants are

commonplace in the language, e.g. ta-takallam 'you

converse'. Nor does it follow from a restriction on

identical consonants separated by a vowel within the stem.

The absence of such forms is explained under the two assump-

tions that Arabic roots are subject to the OCP and that all

autosegmental spreading in Arabic is rightward (McCarthy

1979a, 1981b). As concluded by McCarthy, the representa-

tions in (34a) and (34b) are excluded by the OCP and right-

ward spreading respectively; operation of the two principles

gives the actual form samam as represented in (34c)

(McCarthy 1986:209).

(34) a. b. c.

*sasam *sasam samam

a a a


s s m s m s m

The OCP also plays a significant role in the structural

distinction between tautomorphemic and heteromorphemic

identical consonants. These behave differently with respect

to a well-known rule in Arabic. This difference can be

attributed to their having distinct representations, where

the role of the OCP figures prominently. The rule in

question is that of Identical Consonant Metathesis first

discussed by Brame (1970) and reformulated by McCarthy

(1979a, 1981b, and 1986). The rule applies to tautomor-

phemic identical consonants as in /samama/, but skips

heteromorphemic ones as in /ya-ta-tabbaTu/. There is no

short way of distinguishing the two types of consonants

without further complication of the rule. However, the

enforcement of the OCP as a constraint on the lexical

representation of tautomorphemic identical consonants makes

the distinction manifest. Only tautomorphemic consonants

are characterized in terms of one-to-many associations

between the melodic and skeletal tiers (cf. (34c)). The

structural distinction can thus be incorporated into the

rule itself making it inapplicable to heteromorphemic

identical consonants.

In addition to its role in the lexicon, the OCP has an

equally significant role throughout the derivation, as

emphasized by McCarthy (1986). Drawing evidence from a

variety of languages, Semitic and non-Semitic, McCarthy

argues that the OCP shows what he calls an 'antigemination'

effect. That is, instead of fusing two identical elements

into a single unit, the OCP is enforced to prevent the

creation of such sequences.

The first rule he cites is a syncope rule from Afar, a

Lowland East Cushitic language. The rule syncopates an

unstressed vowel in a peninitial two-sided open syllable.

This is illustrated in the following examples he cites from

Bliese (1981):

(35) xamila xaml-1 'swampgrass (acc./

digib-t-6 digb-e 'She/I married'

He cites the following examples where the rule regularly

fails to apply:

(36) midadi 'fruit'

xarar-e 'he burned'

danan-e 'I/he was hurt'


In (36) the vowel which fails to undergo syncope is in each

case located between two identical consonants. McCarthy

argues that the blocking of the rule is not a langauge-

specific constraint against geminates, since both tautomor-

phemic and heteromorphemic geminates occur freely in the

language. Rather he attributes the failure of syncope to

apply to the antigemination effect of the OCP. The expected

output of syncope contains a sequence of identical

tautomorphemic consonants that violates the OCP. McCarthy

illustrates the prohibited derivation in (37) (p. 221):

(37) e e

[[ c v c v ] V ] -/-V *[ C V CC ] V]
wa al wall

The OCP also interacts with yet another principle that

has been shown to be required by nonconcatenative systems

for the application of some phonological rules. The

principle of Tier Conflation (cf. section 2.2.1) achieves

the linearized representation required by certain rules.

The interaction of these two principles shows that the OCP

is also enforced after Tier Conflation. McCarthy gives

Tiberian Hebrew schwa deletion as an example. In Tiberian

Hebrew, schwa deletes in a two-sided open syllable but fails

to apply between identical consonants as in the following

examples (McCarthy 1986:234):

(38) zaaxruu 'they but saapapuu 'they
recalled' surrounded'

malXe 'kings of' but harere 'mountains

Schwa also deletes between heteromorphemic identical

consonants as in the case of verbs with final k followed by

the pronominal suffix -akaa: tapaareXXaa 'she will bless

you' (p. 235). The Hebrew data offer further support for

the enforcement of the OCP, however, with slightly different

assumptions depending on the morphological make-up of the

language and the place of schwa syncope in the overall

derivation. McCarthy notes that the application of schwa

syncope to the Hebrew example harere 'mountains of' displays

no violation of the OCP, as is evident from the output of

the rule (p. 235):

(39) a a e a e
I I I v
h r h r = *harree

The nonviolation of the OCP in this case is due to the fact

that vowels and consonants are represented on different

tiers. However, if schwa syncope is assumed to apply after

Tier Conflation, the facts of Hebrew will also be in

agreement with the OCP principle. In this case, schwa

syncope applies to the linearized representation in (40),

therefore, the violation of the OCP in the output of the


(40) e e
C V C V C + V V ---C V C C + V V
har er h a r r

There are, however, certain lapses to the antigemination

effect of the OCP. One of the rules exhibiting geminates in

its output is the rule referred to at the beginning of this

section as the rule of Identical Consonant Metathesis. The

rule plays an important role in the phonology of classical

Arabic and that of the modern dialects. The rule is

responsible for the derivation of verbs with final identical

consonants like sadd 'he pulled' from the underlying

structure sadad. The rule was formulated as a phonological

rule in Brame (1970) and McCarthy (1986). However, McCarthy

(1986) presents arguments that the rule can also be viewed

as a rule of allomorphy or a very early phonological rule.

One of the arguments, which will prove to be significant in

section 4.4.2, is that early rules have many exceptions on

the basis of lexical or morphological class membership. The

other argument, which holds in MA, is that no phonological

rule can be shown to apply before this rule. On the basis

of the early nature of such a rule, the assumption that it

applies before Tier Conflation is well motivated.

Therefore, the geminates in the output of the rule are

compatible with the OCP antigemination effect.

2.4 Notes

This is an incomplete representation of McCarthy's
original example. For instance, it excludes the active and
passive forms of the imperfect and the participle. In
addition, the forms cited in the example involve a certain
amount of abstraction. For instance, the root ktb does not
occur in all the binyanim. Ktanbay (XV) is not an actual
form, but landay 'to be stout' is. The range of meaning
expressed by each binyan besides the original meaning of the
root is given below:

I 'stative'
II 'intensive, extensive, causative'
III 'reciprocal'
IV 'causative'
V 'reflexive'
VI 'reflexive of third binyan'
VII 'reflexive of first binyan'
VIII 'reflexive or middle voice of first binyan'
IX 'express colors'
X 'reflexive or middle voice of fourth binyan'
XI 'express defects'
XII sporadic
XIII sporadic
XIV sporadic
XI sporadic

2Verbs of this group that undergo the Flop Rule of the
eighth binyan in MA are few in number. They are mostly
learned words, e.g. ?iTtadilu 'straighten!' (in prayers).
The colloquial forms used elsewhere are ?antadlu or ?atTadlu
(with no flop).

3The first possibility, however, is more compatible
with the generally observed phenomenon of simplification in
the grammar of MA and the other dialects if compared to that
of Classical Arabic.

4Certain versions of the theory allow another
representation of long vowels, whereby the vowel is
dominated by the sequence VC instead of VV as in



The evidence given by Clements and Keyser (1983) is language
specific. It relies on the observation that in certain


languages, vowel-final words behave like consonant-final
words. We will return to this issue in Chapters Five and

This convention excludes pitch-accent systems for
which association starts by linking the starred segment (*)
with the starred tone. In this case the melody may be
introduced by a phonological rule.

School has not yet made it as a loan word in Arabic.
The pronunciation cited for the example is that of a student
learning English and, thus, applying the syllable structure
constraints of MA to the target language.

Epenthesis in this particular example serves to
simplify the complex coda of march, i.e. rc.


3.1 Introduction

Since antiquity, the syllable has been recognized as a

relevant unit in phonological descriptions. Evidence to

support the role of the syllable in phonological theory

usually comes from three areas. First, it can be argued

that the syllable provides the domain within which the

phonotactic constraints in a language can be best defined

(Pike 1947, Kurylwicz 1948, O'Connor and -Trim 1953, Fudge

1969, Sampson 1970, Kahn 1976, Benhallam 1980, Selkirk 1982,

and many others). The majority of these studies demonstrate

that 'morpheme-structure conditions' and redundancy

conditions can be more adequately stated as constraints on

syllable structure.

Second, the syllable forms the proper domain for

certain phonological rules. In other words, it is only with

reference to the syllable that these rules can be properly

characterized. Studies of this kind are numerous and cover

a wide variety of languages. Among these are Basbgll's

analysis of Danish st d (1974); Hooper's argument for a

syllable-based analysis of nasal assimilation in


3.1 Introduction

Since antiquity, the syllable has been recognized as a

relevant unit in phonological descriptions. Evidence to

support the role of the syllable in phonological theory

usually comes from three areas. First, it can be argued

that the syllable provides the domain within which the

phonotactic constraints in a language can be best defined

(Pike 1947, Kurylwicz 1948, O'Connor and -Trim 1953, Fudge

1969, Sampson 1970, Kahn 1976, Benhallam 1980, Selkirk 1982,

and many others). The majority of these studies demonstrate

that 'morpheme-structure conditions' and redundancy

conditions can be more adequately stated as constraints on

syllable structure.

Second, the syllable forms the proper domain for

certain phonological rules. In other words, it is only with

reference to the syllable that these rules can be properly

characterized. Studies of this kind are numerous and cover

a wide variety of languages. Among these are Basbgll's

analysis of Danish st d (1974); Hooper's argument for a

syllable-based analysis of nasal assimilation in


(Allegretto) Spanish, of consonant devoicing in German, and

of nasalization of high vowels in Akan (1972); Hoard's

discussion of aspiration and tenseness in English (1971);

Vennemann's analysis of vowel lengthening in Modern

Icelandic (1972); Kahn's characterization of some low-level

processes in English (1976); and Broselow's (1979) as well

as Benhallam's (1980) account of emphasis in Arabic.

Third, the syllable constitutes the domain of various

suprasegmental phenomena (Trubetzkoy 1969, Cheng 1966,

Lehiste 1970, and many others).

While the evidence for the syllable as a unit of

segment organization exists in abundance, the efforts to

accord a theoretical status to it fluctuate. References to

it in pregenerative literature are hard to miss. But the

several attempts to define it either in phonetic or

phonological terms (cf. Hjemslev 1939, Stetson 1951, Haugen

1956, Hala 1961, Rosetti 1962, O'Connor and Trim 1953,

Pulgram 1970) were inconclusive, mainly because of the

complexity and variety of the linguistic phenomena related

to the syllable.

This led phonologists to leave out the syllable as a

unit of phonological organization. The standard theory of

Chomsky and Halle (1968) gives no place to the syllable in

phonological representations. The characterization of

syllabically conditioned processes was, therefore,

particularly problematic and led to the use of some

abbreviatory devices to describe these processes.

Well-known are cases of C to indicate the syllable-edge

and C { V )C to characterize the distinction between
o VC o
'weak' and 'strong' clusters. In addition to their being

arbitrary, there are other problems with such conventions.

The first device contains the elements C and #, which do not

form a natural class. Several attempts have been made to

naturalize the word-boundary by giving it specifications for

segmental features (Lass 1971, Lightner 1972). An alterna-

tive way of dealing with these boundary markers was proposed

in Pyle (1972).

The second convention mentioned above was introduced to

characterize the distinction between 'weak' and 'strong'

clusters in English stress, which is actually a distinction

between light and heavy syllables, respectively. This is a

very common distinction utilized in the description of the

stress systems of many languages, yet it cannot be expressed

using a phonological feature. This particular convention is

found to be not only arbitrary, but also cannot be motivated

on independent grounds.

In light of the increasingly clear need for the

syllable in phonological organization, there have been

several proposals for incorporating the syllable into the

generative framework. These include the use of boundary

symbols (Hooper 1972) and bracketing (Anderson and Jones

1974). They all concern themselves, in one way or another,


with the marking of syllable boundaries in the phonological

string so that rules can refer to them.

A distinction like the one between 'weak' and 'strong'

clusters, however, does not seem to depend on reference to

boundary markers; rather, what is involved is the difference

in the internal structure of the constituents to which these

clusters belong.

3.2 Syllable Constituency

According to the few proposals which incorporate the

syllable into phonology within a linear framework, a

phonological representation includes in addition to the

sequences of phonemes a number of boundary markers that

serve to delimit syllables, morphemes, and words. It is

assumed that all information crucial to the operation of

phonological rules or to the statement of phonotactic

constraints are encoded in the segments themselves or in

these boundaries. According to this theory, then, syllables

are defined in terms of their boundaries. A syllable

consists of all segments delimited by a boundary marker,

usually '$' or '.'. This view of the syllable is

represented in Hockett (1955), Haugen (1956), Pulgram

(1970), Hoard (1971), Hooper (1972, 1976), and Vennemann


Although these proposals represent a major step towards

the incorporation of the syllable into the phonological

representation, they all have one thing in common. Apart

from the delimitation of segments belonging to the same

syllable by syllable boundaries or juncture elements, no

further structure is assigned to the syllable. However,

there are certain phonological generalizations which can be

captured only with reference to the internal structure of

the syllable. First, as mentioned briefly in Section 3.1,

such a structure figures prominently in the statement of

certain phonological rules (cf. the distinction between

'strong' and 'weak' clusters). Second, in the absence of

such constituency the phonotactic constraints within the

syllable will have to be stated in some ad hoc manner.

In fact, earlier proposals viewed the syllable as

having a constituent structure of the same type as is to be

found in morphosyntactic domains (Pike and Pike 1947, Fudge

1969). For'instance, Pike (1967) argues for the onset,

peak, and coda as being the major constituents of the

syllable since it is among these positions that phonotactic

constraints hold. Similarly, Kurylowicz (1948) argues for a

constituent that groups the peak and coda together. The

basis for this grouping is that cooccurrence restrictions

are more likely to hold between the peak and coda than

between either peak or coda and the onset. Other arguments

offered for the relevance of syllable internal structure

involve certain phonological processes. For example, Pike

and Pike (1947) argue for a nucleus unit on the basis that


suprasegmental and segmental phonological rules refer to it.

Kurylowicz (1948) and Newman (1972) observe the importance

of the rhyme to the treatment of the heavy/light syllable

distinction. However, in spite of the above efforts, it was

only recently that these constituents were given a place in

the overall hierarchical structure of speech and considered

as indispensable to the structure and operation of certain

phonological phenomena.

In fact, the whole issue of the internal structure of

syllables has received a great deal of attention recently.

The reason, again, is the growing evidence for the

importance of syllable constituency. It is a generally

observed fact that most syllabically related processes do

not refer to the syllable as a unit; rather, they are

sensitive to the structure of its constituents, particularly

the rhyme. The comment made by Halle and Vergnaud (1980:93)

is relevant here:

Our studies have uncovered many phonological
processes where the constituents of the syllable--
in particular, the onset and rime--function
independently from one another. In fact it
appears to us that the superordinate unit, the
syllable, plays a much more marginal role in
phonology than do its constituents.

Evidence for the independence of the different constituents

within the syllable, as stated by Halle and Vergnaud, in

itself constitutes an argument for syllable constituency.

The syllable structure of many languages supports this

viewpoint. See, for instance, Pike and Pike (1947), Saporta


and Contreras (1962), Fudge (1969), and Newman (1972), among

others. The syllable structure of Arabic is not an

exception in this respect (McCarthy 1979a).

It is necessary to note that any evidence for syllable

constituency will argue not only against the linear repre-

sentation of the syllable but also against any theory that

does not assume an internal structure for syllables,

including some proposals made within the nonlinear

framework. For this reason, it is instructive to note that

not all versions of the nonlinear representation of the

syllable assume syllable constituency. Two viewpoints can

be distinguished in this respect, namely, the 'flat' vs.

hierarchicc' theory of syllable structure.

The flat theory is discussed mainly in Kahn (1976),

which introduces the first nonlinear treatment of syllable

structure. Apart from the association of segments that

constitute one syllable to the syllable node located on a

separate tier, Kahn does not discuss the issue of syllable

internal constituent structure. In this respect, Kahn's

proposal is quite similar to those of McCawley (1968),

Vennemann (1972), and Hooper (1976). However, Kahn's

approach differs from the boundary approach in two major

respects. First, it recognizes the syllable as a linguistic

unit separate from the strictly segmental information.

Second, unlike the boundary approach, it allows for the

representation of 'ambisyllabic' segments (Section 3.5.3).


Recently, Clements and Keyser (1983) have argued in favor of

the flat theory of the syllable. They argue that no further

structure is required between the syllable nodes and

syllabic slots. Clements and Keyser maintain that the

category 'nucleus' plays a significant role in phonological

representation and that the 'onset' and 'coda' are

unnecessary. They propose that phonological representations

consist of the syllable tier, CV-tier, the nucleus tier, and

the segmental tier. The first three tiers are defined as

structural units, which represent the organization of speech

units at higher levels. The segmental tier, on the other

hand, consists of several phonetic tiers such as the nasal

tier, the laryngeal tier, the tonal tier, and others. It

has to be emphasized that the recognition of a nucleus

constituent does not affect their basic assumption that

syllables have flat structure since the proposal nucleus

category is considered to be a simultaneous representation

on a different tier:

(1) N nucleus-tier

C C V C C CV tier

a syllable-tier

The other nonlinear representation of the syllable

recognizes an internal hierarchical structure within

syllables. This view is adopted in most studies that deal

with syllable structure (Vergnaud and Halle 1979, 1980,

McCarthy 1979a, 1980, Kiparsky 1979, Prince 1980, Selkirk

1980, Steriade 1982, Harris 1983, and many others).

Although differences might exist as to the number of con-

stituents recognized or the internal organization of certain

constituents, there seems to be a general agreement on the

division of the syllable into two major constituents: the

'Onset' and the 'Rhyme'. For an example of variation in the

number of constituents recognized, see Halle and Vergnaud

(1980) where a third constituent, the 'Appendix', is argued

for on the basis of data from Arabic, German, and Malayalam.

Variation in the internal organization of the rhyme con-

stituent also exists. For instance, Selkirk (1984a),

Pesetsky (1979), Safir (1979), Hayes (1980), Mohanan (1985),

and Benhallam (1986) recognize the further division of the

rhyme into the nucleus and coda.

An examination of the syllable structure in MA reveals

evidence that supports syllable constituency. It will be

noticed that while the discussion is directly pertinent to

the syllable in MA, most observations also hold for the

syllable in CA. Of course, certain differences exist as

well and will have considerable implications for the data we

are discussing.

We first consider the area of phonotactic constraints.

The following general observations characterize the syllable

in MA. In other than word-final position, syllables contain

at most three segments. The first two positions are

reserved for the onset and nucleus, both of which are

obligatory. The third position is occupied by an optional

coda. There are no coocurrence restrictions between the

onset and what follows. Any consonant in the phoneme

inventory of MA may occur alone as the onset, including the

glides w, y, and ?: ?amiir 'prince', ?as.wad 'black', and

yaa.bis 'hard'. Onsets consist invariably of one consonant.

Thus, the first syllables in *#ktubu, *#lbalad, *#stalam are

all ill-formed syllables, even though the overall length of

the first syllable is not excessive. The second position is

always occupied by the nucleus or the sonority peak, always

a vowel in Arabic but not necessarily so in other languages.

In the case of CVV, the second and the third positions are

exhaustively occupied by the nucleus. In the case of CVC

syllables, on the other hand, the third position is occupied

by a single consonant. The glides, however, are banned from

this particular position. The only exception is when the

glide happens to form the first member of a geminate


More restrictions exist when we move to consider

syllables that occur in word-final position. The maximal

syllable in this position consists of four segments. These

are, in fact, the same basic CVV, or CVC, plus an extra

consonant giving the syllable types CVVC, or CVCC, respec-

tively. The same restrictions on positions two and three


discussed above also hold here. Two additional restrictions

can be stated with respect to the fourth position. The

glides are excluded from this position in both types of

syllables, i.e. CVVC and CVCC, except in certain cases which

will be discussed in Chapters Five and Six. The other

restriction is specific to CVCC and concerns the last two

positions: The more sonorous member of the biconsonantal

cluster usually occupies position three; exceptions occur

where the cluster forms a geminate or in the case of a

certain morphological class.

So far, all the constraints mentioned above concern the

last two members of the syllable, the nucleus and the coda;

none of the constraints involves the onset. The claim is

that the relevance of certain parts of the syllable but not

others should be expressed in terms of a specific syllable

internal constituent structure.

In a flat theory the domains within which these

constraints hold cannot be designated without making use of

some ad hoc mode of description. The argument put forth by

Steriade (1982:73) can be summarized as follows: The fact

that in one and the same language rules can refer only to

the domains represented in (2d) and (2e) and not the others

constitutes an argument for the syllable internal

constituents they represent (the rhyme and the onset):

(2) a

c v CCC

[ ] domain (a)

[ ] domain (b)

[ ] domain (c)

[ ] domain (d)

[ ] domain (e)

As mentioned by Steriade, the domain in (2d) can be defined

in an ad hoc manner as the part of the syllable that begins

with the vowel, and the one in (2e) as the elements

preceding the vowel. However, it is no accident that the

domains described in such terms coincide with the domains

usually referred to as 'Rhyme' and 'Onset', respectively.

It seems likely that the onset and rhyme constitute the

minimum hierarchical structure within the syllable. Under a

theory that recognizes this distinction, the phonotactic

constraints holding among the last two or three positions in

the syllable can now be given some significance.

Phonotactic constraints are more likely to hold between

positions forming one and the same constituent, in this

case, the rhyme. No such constraints exist between the

onset, on the one hand, and any of the other elements within

the rhyme, on the other.

The recognition of a rhyme constituent enables us to

formulate the constraints on the maximum number of segments

that may follow the onset as the maximum number of segments


allowed in the rhyme in MA. The irrelevance of the onset to

the statement of phonotactic constraints then follows from

the onset-rhyme division within the syllable.

It can be roughly stated that in the case of MA, the

upper limit on the rhyme is three segments in word-final

position and two elsewhere, and that glides may not occur in

either position following the nucleus, except in very few

cases. It is worth noting that in CA the glides can occur

in the positions following the nucleus. Examples from CA

include fa?run 'mouse nom.', say? 'thing pausall form)'.

The upper limit of segments allowed in the rhyme in MA

is further confirmed by the impossibility of CVVC.C.

syllables. The constraint against this syllable type is

observed by both native and loan words, as we will see


The distribution of CVVC.C. syllables in CA is relevant
1 1
here. This syllable type is the least frequent in CA. It

appears rarely as the final syllable of utterances or of

words in pause form (Al-Ani and May 1978:118). The coda of

a CVVCCi syllable is always occupied by a geminate cluster.

As mentioned by Al-Ani and May, these syllables are phone-

tically realized as CWC; however, they must be phonologi-

cally considered as CVVC.C. on the basis of their behavior

in other than final position. For instance, compare maarr-

un 'passerby (nom.)' where -r becomes the onset of the

following syllable, to the pausal form maar(r) 'passerby'.

The situation in MA and most of the dialects is,

however, different. CVVCiC. syllables do not occur in final

position. Compare, for instance, zaara 'neighbor (f.)' and

zaar 'neighbor (m.)' to zaar.ra 'dragging (f.)' and zaarir

'dragging (m.)'. The expected form in place of zaarir is

*zaarr, which never occurs in that position. The gemination

of the final consonant is justified on the basis of the

feminine form zaarra. The replacement of CVVC.C. syllables

in MA by the configuration CVVCiVCi is only one way of

dealing with words that do have a CVVC.C. in CA. (See

Chapter Four for more details.)

Loan words also conform to this constraint, but,

through different mechanisms. Compare, for instance,

English bank to Arabic bank. The reduction of the nucleus

results in changing the syllabic configuration from

nonoccuring CVVCC into CVCC which conforms to the limit on

the rhyme as mentioned above. However, nucleus reduction is

not the only way in which the syllable structure of loan

words is modified to fit that of the borrowing language.

Other words are dealt with in different ways. For example,

English march is borrowed as maaris with epenthesis

simplifying the final cluster.

The general observation is that in all such examples

either the nucleus or the coda branches, but not both. If

the nucleus branches, then CVVCC syllables are realized as

CVVCVC sequences; if the coda branches, then the result is

CVCC through nucleus reduction. The significance of the

nucleus/coda distinction in the realization of CVVCC in the

language seems to require the recognition of these

constituents within the rhyme. However, this is not

necessarily the case, since the upper limit on the number of

segments within the rhyme can be stated as a restriction on

the rhyme node. Thus, if the upper limit is three positions

or segments, then CVVC is the result of rhyme branching,

while a CVCC is the result of coda branching. The branching

of both would violate the upper limit constraint.

In the above discussion, reference to the rhyme node

was shown to be crucial to the statement of certain

constraints holding within the syllable. The other reason

for recognizing the onset/rhyme division within syllables is

the frequently observed relationship between rhyme structure

and certain phonological rules, such as stress. The first

relevant observation is that onsets and certain syllable-

final segments are not relevant to the operation of stress

rules (exceptions occur (cf. Davis 1982)) but these are rare

cases. The general observation is that stress is attracted

to heavy syllables only; light syllables are usually skipped

by stress. For instance, in Makkan Arabic, heavy

penultimates are always stressed, as in raddeetu 'I returned

it', and gaalatlu 'she said to him'. However, if the

penultimate syllable is light, stress falls on the


antepenultimate syllable regardless of whether it is heavy

or light, as in wagtahum 'their time' and sazara 'tree'.

As mentioned earlier, the distinction of 'light' vs.

'heavy' syllables is represented linearly in terms of 'weak'

vs. 'strong' clusters. But the 'strong' vs. 'weak' distinc-

tion misses the generalization that what is involved is the

difference in the internal structure of the syllables. As

recently proposed by McCarthy (1979a, b) and Vergnaud and

Halle (1979), this difference in structure can be formally

represented in terms of the internal structure of the rhyme.

Given the view that long vowels are represented as under-

lying geminates, the distinction is reduced to one of

branching vs. nonbranching rhymes:

(3) a. a b. o c. a



The structures in (3) provides a formal way of character-

izing two well-known observations. First, the geometrical

structures in (3b) and (3c) capture the structural

equivalence of C V and C VC syllables, and thus provide a

justification for their similar behavior with respect to

stress as in MA. With this representation there is no need

for the use of disjunction to characterize them. Second,

these structures characterize the distinction between light

syllables, as in (3a), on the one hand, and heavy syllables,

as in (3b) and (3c), on the other. The distinction is

characterized in terms of nonbranching vs. branching rhymes.

As a result, stress rules can now refer exclusively to the

structure of the rhyme constituent (McCarthy 1979a, b).

The enrichment introduced by the incorporation of auto-

segmental representations into generative phonology is not

limited to stress rules. As will become clear in the course

of this study, other phonological rules can be stated more

adequately in a theory where the hierarchical structure of

the syllable is represented separately from the strictly

segmental information.

There have been several proposals in the literature,

all of which deal with the process of basic syllabic parsing

in the nonlinear framework. These will be outlined in the

next section.

3.3 Major Approaches to Syllable Structure

3.3.1 Kahn (1976)

In Kahn (1976) we find the first nonlinear approach to

the syllable within a generative framework. Kahn's argument

includes discussion of syllable structure assignment as well

as discussion of low level phenomena in English, such as

ambisyllabicity, aspiration, flapped /t/, /r/-loos, and the

like. Syllabification rules, in Kahn's system, can be

viewed as a parsing device which assigns tree structure to

words, and they will be considered as such here. The

analysis of a string of segments proceeds as follows:

(4) Rule I (p. 22)

With each [+ syllabic] segment of the input
string associate one syllable:

[+ syll] --+ [+ syll]

Rule II (p. 24)

a. C CV -- C Ci C C V

s s

where C. C is a permissible initial
cluster, But C. C+ C is not.
1 i+l n

b. VC C -- VC .. C. C
I 1In L / j+l1 n
x x

where C C is a permissible final
cluster, but C C. Cj+l is not.

Rule I spots a vowel and assigns the node S (syllable) to

it. Rule II, then, associates as many preceding consonants

as can make a possible word-initial cluster, and then

associates as many following consonants as can make a

possible word-final cluster.

There are two major aspects to Kahn's theory. First,

the notion of 'onset maximization' is achieved through

ordering: Preceding consonants are associated before

following ones. Second, the set of possible syllable-

initial (-final) clusters in English is established on the
basis of possible word-initial (-final) clusters. While

the first aspect of Kahn's theory enjoys considerable

support as a universal principle, the second is corroborated

only by some languages.

3.3.2 Lowenstamm (1981)

A definition of the Universal Syllable is given in the

following statements:

(5) In a string of segments, a syllable is a maximal
substring such that

a. (i) No segment is lower on the hierarchy
than both its immediate neighbors

(ii) No two segments of equal ranking on the
hierarchy are adjacent

b. The onset is maximal within the limits of (a)

Syllables are thus defined without reference to language-

specific rules. Maximization of onset is stated as a well-

formedness condition on syllabification. While Lowenstamm's

approach characterizes the syllable in many languages, cases

from many others may be cited as exceptions to the sonority

hierarchy on which this approach relies.

3.3.3 Cairns and Feinstein (1982)

This approach is basically a theory of syllable marked-

ness. Focus is on the internal structure and markedness

relations of consonant clusters in syllabic onsets. The

pivotal aspect of this view of syllabification is that

'various nodes of hierarchical syllable structure are

assigned markedness values according to universal interpre-

tive conventions, depending on the kinds of node expansions

that are chosen from a set of universal rules of syllabic

'phrase structure" (p. 193). In addition, the role of the

sonority hierarchy is fulfilled by the formal characteri-

zation of the universal component of syllabification;

language-specific well-formed syllables are defined in terms

of an explicit algorithm, and nodes in hierarchical syllable

structure are crucially labeled. Under this approach,

syllabic parsing proceeds as follows. First, the language-

specific syllable template parses a string into a 'candidate

set' of all possible syllabification. Then each syllable is

assigned a markedness index by a universal convention of

markedness evaluation, and a composite markedness value is

computed for all possible syllables (in the candidate set).

The syllabification which has the lowest composite

markedness value is then chosen. The second mechanism is a

language-specific disambiguation clause which states that,

for certain structures, candidate parsings which select them

must be preferred, other things being equal.

3.3.4 Steriade (1982)

The main claim in Steriade's approach to syllable

structure is that not all aspects of syllabic organization

are decided at the same point in the phonological derivation

but that syllabic structures are created by syllable

building rules which are ordered among the rules of the

phonological component. The syllabification rules proposed

by Steriade for the different dialects of Greek are of two

types, universal and language-specific. First, the

segmental string is parsed by a universal first rule which

crates CV syllables. The other universal aspect of syllable

parsing is implied in the requirement that other core

syllable rules must adjoin stray Cs as immediate sisters of

syllabified skeleton slots (p. 82). The language-specific

aspects of syllabification include the rules that create

branching onsets and codas, their relative order, their

unbounded or binary character, and the existence of some

segmental well-formedness conditions on their application.

Steriade maintains that there are two important

consequences to her approach. The first is that the

universal aspects of the template (i.e., each syllable

consists of a rhyme and a preceding onset if a consonant

happens to be in that position) can be factored out and

distinguished from language-specific aspects. The second is

that this approach predicts rather than stipulates that the

nuclear vowel will always be a member of the right branch


since stray skeleton slots are adjoined as immediate sisters

of the neighboring skeleton slots and not to the syllable or

any of its constituents. Therefore, a stray C adjoined to

the right of a CV syllable creates a branching rhyme,


while a stray C adjoined to the left of a CV syllable will

create a branching onset,


In both cases the vowel occupies a terminal node of the

right branch.

3.4 Testing the Syllable Parsing Theories

In this section I will discuss some of the issues con-

tingent on the extension of each of these approaches beyond

the languages for which they were originally proposed. The

issue here is not whether these proposals can be of a

universal validity (some do not claim to be). Rather, the

main point is to consider some of the results that might be

achieved from applying each approach to some other

languages, as well as the problems that might arise in the

course of doing so.

The previous section identifies two aspects to Kahn's

system of syllabification: the rules of syllabic parsing,

which divide the string into possible syllables, and a


number of low-level rules particular to English. The rules

of parsing are candidates to be tested against the syllabi-

fication systems of other languages. Drawing on an analysis

by Dresher of data from the Vespasian Psalter dialect of Old

English, Lowenstamm (1981) demonstrated that Kahn's approach

is open to objections on two grounds. First, the definition

of syllable-marginal clusters in terms of configurations

permissible at word margins is problematic. Second, some

undesirable consequences for the statement of syllable-based

rules follow from this assumption. In particular, the rule

of epenthesis in the Vespasian Psalter dialect of Old

English as well as the rule of sonorant syllabicity in both

modern English and Yiddish will have to be stated in

paradoxical terms: Syllable structure assignment presup-

poses nuclei, but the rule that creates nuclei also pre-

supposes syllable structure. For instance, the epenthesis

rule which simplifies a final consonant cluster in a word

like wetr 'water (nom.)' must be able to refer to a rep-

resentation like (6) (Lowenstamm 1981:583), where tr can be

shown to be tautosyllabic in order to trigger epenthesis:

(6) w e t r

According to Kahn's parsing rules, the final consonant in

(6) may not be associated to the syllable node since the

resulting cluster tr is not observed word-finally in

English. Similarly, the rule of modern English which

accounts for derived syllabic sonorants, e.g., 1 in simple

involves a paradox: Syllable structure assignment, as given

by Kahn's Rule (IIb), requires nuclei, but the rule which

brings the nucleus 1 into being also presupposes syllable


The facts of syllabification in Arabic contradict part

of the maximum cluster approach: While the structure of the

initial margin of the syllable can be defined in terms of

what is admissible in word-initial position, syllable-final

clusters may not be described in terms of word-final


In this language, consonant clusters are generally

allowed at word margins only, more precisely in word-final

position under very restricted circumstances.4 Word-

internal syllable margins are, however, restricted to one

consonant only.5

In the case of initial clusters, the condition in (IIa)

is not violated. Words and syllables invariably begin with

a single consonant, as in (7).

(7) a. ma.Ta 'with'

b. 'school'

c. ? /l-kitaab/ 'the book'

All of the examples in (7) start with a single consonant,

and so do all syllables regardless of their position in the

word. Therefore, (IIa) of Kahn's rules makes the correct

prediction in the case of syllable- and word-initial

clusters in Arabic.

However, the condition in (IIb) is not borne out by

final clusters. Consider the following examples of some

possible consonant clusters in world-final position. These

forms are restricted in occurrence to prepausal position:

(8) a. kalb 'dog'

b. burl 'tower'

c. jins 'race'

According to (IIb) the final clusters in (8) are expected to

be permissible at word-internal syllable margins as well.

This is not the case, however, as illustrated by the forms

in (9):

(9) a. *kalbha 'her dog'

cf. sababha 'her reason'

b. *burjna 'our tower'

cf. sababna 'our reason'

c. *jinshum 'their race'

cf. sababhum 'their reason'

The only way to achieve well-formedness for the forms in (9)

is to break up the cluster at the syllable margin in each


(10) a. kalbaha

b. burjana

c. 3insahum

The extra consonant at word margin is syllabified with the

inserted vowel. Thus, syllable-final position is restricted

to one consonant only--a situation that does not hold for

word-final position. Thus, Kahn's approach makes the right

prediction in one case but not in the other. It is worth

noting that the.unparallel structure of onsets and codas

with respect to clusters is not unique to Arabic. Pulgram

(1970) remarks that while syllable onsets always correspond

to word onsets, the same is not always true of codas.

The second objection raised by Lowenstamm against the

maximum cluster approach concerns the structure of some

epenthesis rules. Observe that these problems follow from

the conditions stated in (IIa) and (IIb), i.e., epenthesis

or sonorant syllabicity needs to refer to the tautosyl-

labicity of tr or pl cluster, but cannot, since neither is

observed word-finally. However, we have already seen that

such constraints are not without exceptions. Therefore,

many languages may put severe restrictions on clusters at

syllable-margins but, at the same time, allow syllables at

word margins to include extra elements (cf. Arabic as in

(8)). In addition, part of the paradox is due to the

requirement that epenthesis refers crucially to the

tautosyllabicity of the cluster. An alternative would be to


incorporate the conditions defining these clusters into the

rules of syllabification. Or, for languages other than

English, the removal of the conditions in (IIa) and (IIb)

will allow the consonants of the cluster to be tautosyl-

labic. Epenthesis would then modify consonant clusters that

violate the sonority hierarchy (cf. Section 5.2.3).

Lowenshamm's approach is a rearticulation of the

sonority hierarchy principle recognized long ago (Sievers

1893, Jespersen 1904, and Saussure 1915, among others).

Many languages seem to obey this principle according to

which segments within a syllable tend to occur in an order

of increasing sonority from the outside of the syllable

toward the syllable peak, which is by definition the most

sonorous segment.

These hierarchies, however, are not assumed to be

absolute. Violations of the sonority hierarchy may occur as

a result of languages having their own (language-specific)

constraints defined on their syllable structures. However,

most of these violations can be shown to be morphologically

motivated, as illustrated below.

The situation where the sonority hierarchy would be

violated is hard to come by in most varieties of Arabic

since consonant clusters are not usually allowed at syllable

margins. Yet, it is still possible for certain morphologi-

cally determined derivations to deviate from the sonority

hierarchy. Consider the following examples from Classical


(11) a. kasb 'earning' b. kabs 'pressing'

surb 'drinking' sabr 'patience'

?ams 'yesterday' ?ism 'name'

qalb 'heart' Fabl 'rope'

The cluster involved in each example in the second column

violates the hierarchy in that the members of the cluster

are not arranged in such a way that the more sonorous among

the two occurs next to the syllable peak. However, these

examples are all nouns formed in accordance with a certain

morphological template.

In fact, the sonority hierarchy principle is observed

by many languages, but deviations may still be found. A

possible suggestion is to specify where deviations from the

hierarchy occur and to interpret this principle as an

indicator of relative syllable markedness rather than an

absolute universality.

Our objection to Cairns and Feinstein's markedness

theory of syllable structure falls along the same lines as

Steriade's objections to this model: Namely, the model is

too complicated compared to the results it was established

to achieve. For instance, the role of the markedness

evaluation can be simply stated as one that assigns a V.CV

structure to the sequence VCV. Such a simple procedure does

not justify the highly complex mechanism of the markedness

evaluation. Moreover, little attention is given in this

model to the cooccurrence restrictions that usually hold

within the rhyme. This is not a genuine criticism since

similar conventions can be devised to account for the rhyme

structure of the syllable, but it is very likely that these

conventions will involve the same kind of elaborate


The main point of Steriade's approach is the distinc-

tion she draws between the universal and language-specific

aspects of syllabification. The universal part of syllabi-

fication is that all syllables consist of a rhyme and a

preceding onset if a consonant is found in that position,

and that stray consonants must be adjoined as sisters to

already adjoined C-slots. Differences among languages arise

as a result of the presence or absence of other syllabifica-

tion rules, in particular an onset rule which creates

complex onsets, and a coda rule that creates branding

rhymes. Languages might also vary with respect to the

number of applications of the onset and coda rules as well

as the constraints imposed on their application.

3.5 Rules of Syllabification in Arabic

3.5.1 Syllable Parsing Rules

It has been demonstrated in section 3.2 that syllable

constituency is a crucial aspect of the syllable. In this

section we consider how segments are organized into

syllables displaying such hierarchical constituency. The

main claim is that syllable structure organization consists

of a set of syllable-building rules, both universal and

language-specific. In this respect, my indebtedness to

Steriade's work will be evident, but certain modifications

are called for. It will be demonstrated that this view of

the rules of syllabification has the advantage of allowing a

unified description for syllable structure organization in a

number of varieties of Arabic.

I first assume that the phonological string is parsed

by a universal CV rule that takes precedence over any

language-specific rules(s).7 This rule is given in (12):

(12) CV-Assignment Rule (Universal):

C V => C V

The application of Rule (12) to strings of segments like

hiba 'donation' and kasa 'to clothe' gives the structures in

(13a) and (13b), respectively:

(13) a. o o
0 R 0 R
I I I Rule (12) I 1 I
hiba hiba

b. o
0 R 0 R
I I I I Rule (12) I I I
kasa > k a s a

The first language-specific rule of syllabification in

Arabic is the coda rule given in (14):8

(14) Coda Rule (optional) (a preliminary form)
o G

The Onset Rule in (13) and the Coda Rule in (14) together

account for the syllable structure assigned to words like
katab 'he wrote', and rasmu 'his painting', as represented
in (15a) and (15b), respectively:

(15) a. o o o o
1 1 .1 1 I I A
SI I I Rule (12) I Rule (14)1 1
ka t ab > k a t ab > k a t ab

b. o o o o
A A /\ A
cvCCv CvCCv CvCCv
I I I I IRule (12)1 1 | I Rule (14)1 1 1 1
ra s mu ---> r a s m u > r a smu

The Coda Rule as stated in (14) accounts for CVC but not

CVV--the other heavy syllable type in the langauge (cf.

section 3.2). However, it is a well established observation

that CVC and CVV syllables behave uniformly with respect to

receiving stress in certain positions (Chapter Four) as well

as triggering certain phonological rules in certain

environments--facts which suggest that they form a natural

class. The claim is that this property should be reflected

in the rule responsible for their creation. However, this

depends to a great extent on the representation assumed for

long vowels on the CV-tier. A note on the representation of

long vowels in autosegmental phonology is in order.

The representation of length in terms of a many-to-one

association between the CV-tier and the segmental tier is

commonly maintained by autosegmentalists. Yet, the repre-

sentation of the units of the skeleton tier and, in turn,

the characterization of long vowels in terms of these units

is still debated. The skeletal tier as originally proposed

by McCarthy (1979a) consists of C and V elements represent-

ing the features [cons.] and [syll.], respectively. In

subsequent work, however, C and V slots have been

interpreted as encoding only [- syll.] and [+ syll.],

respectively. The most familiar representation of a long

vowel is in terms of association to two V-slots on the

skeletal tier. But we have already seen that this


representation is not accounted for by the Coda Rule in (14)

since that rule syllabifies a post nuclear C-slot.

The other possible representation proposed for long

vowels in this framework associates a long vowel with a VC

sequence on the CV-tier (cf. Clements and Keyser 1983):

(16) V C


The significance of the difference entailed by the two

representations of long vowels seems to be language-

specific. In other words, if there is no language-specific

rule that differentiates the two structures, vowels

dominated by VV and those dominated by VC will have

identical realizations. For instance, Clements and Keyser

use the representation of long vowels as VC sequences to

characterize a certain set of stems in Turkish that end in

long vowels but behave phonologically as though they were

consonant-final. However, Clements and Keyser indicate that

phonetically the realization of vowels dominated by VV as VC

is identical.

Following a suggestion made by Halle and reported in

Steriade (p. 81) that Arabic long vowels are VC sequences,

we might assume the same representation for long vowels in

MA. Under such an assumption, as Steriade notes, long

vowels can be derived by the Coda Rule. If this assumption

is correct, then the Coda Rule as stated in (14) can be

maintained for the syllabification of CVC as well as CVV.

Consider, for instance, the derivation of naamu 'they

slept'. According to the above assumption, the long vowel

of the first syllable is dominated on the CV-tier by a VC


(17) o 0 0 o
cvCCv CvCCv CvCCv
I V I Rule (12)1 V I Rule (14)1 V II
n a m u > n a m u > n a m u

However, in the absence of phonological rules that

would justify a VC-representation for long vowels, we opt to

reformulate our coda rule. The new form of the coda rule is

given in (18):

(18) Coda Rule (optional) (Final form)

V X > V X

In (18), X covers vowels, glides, and consonants.

So far the core syllables of the language, CV, CVC,

and CVV, are accounted for by rules (12) and (18). Section

3.2, however, shows CVVC and CVCC as possible syllable types

that occur at word margins. The application of (12) and

(18) leaves the final consonants unsyllabified. In order to

account for these syllable types we propose the

Incorporation Rule in (19):

(19) Incorporation Rule
o o

SC # A--> + C #

Rule (19) says that a syllable with a branching rhyme

incorporates a single unsyllabified consonant in word-final

position only. Rule (19) will account for the syllabifica-

tion of CVVC and CVCC as in (20a), (20b), respectively:

(20) a. b.

b a b

b a b


I X\

b a b

dar s

da r s

I / ,
dar s

Rules (12) and (18)

Rule (19)

Rule (19)
== = =

Consonants which may not be exhaustively parsed into core

syllables, as is the case with superheavy syllables in