AN OPTIMALITY-THEORETIC ANALYSIS OF STRESS IN THE

ENGLISH OF NATIVE ARABIC SPEAKERS

A DISSERTATION

SUBMITTED TO THE ENGLISH DEPARTMENT

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE

DOCTOR OF PHILOSOPHY

BY

RASHEED S. AL-JARRAH

DISSERTATION ADVISOR: DR. FRANK TRECHSEL

BALL STATE UNIVERSITY

MUNCIE, INDIANA

APRIL 2002
 AN OPTIMALITY-THEORETIC ANALYSIS OF STRESS IN THE
ENGLISH OF NATIVE ARABIC SPEAKERS

A DISSERTATION
SUBMITTED TO THE ENGLISH DEPARTMENT
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE
DOCTOR OF PHILOSOPHY

BY

RASHEED S. AL-JARRAH

APPROVED BY:

Committee Chairman Date

Committee Member Date

Committee Member Date

Committee Member Date

BALL STATE UNIVERSITY

MUNCIE, INDIANA

APRIL 2002

ii
Acknowledgements
I am indebted to my committee members who contributed substantially to this

dissertation and provided constant encouragement. I am especially indebted to my

supervisor Dr. Frank Trechsel, without whose guidance and support, this dissertation

would not have been possible. His pertinent and detailed comments on earlier drafts of

this dissertation have been significant in shaping the final product. To him, I say “thank

you Frank”.

I am grateful to Dr. Herbert Stahlke for his invaluable comments and helpful

suggestions. I am also indebted to Dr. Carolyn Mackay for her insightful comments on an

earlier version of this text. Sincere thanks also go to Dr. Gail Bader for her constant

encouragement and support. I am thankful to all of them for devoting much of their time

and effort to make my task easier. Responsibility of shortcomings in this dissertation is

mine and mine alone.

I am grateful indeed for the department of English at Ball State University for

providing me with a wonderful environment to complete my dissertation. I should also

register my acknowledgements to Yarmouk University-Jordan for the financial support.

It goes without saying that thanks in million go to my family members for their

love and confidence over the years. I should admit that this dissertation would not have

been possible without the love and understanding of my wife Khitam and my little

daughter Zayneb.

iii
 DEDICATION

TO MY MOTHER

AND

TO THE MEMEORY OF MY FATHER

iv
 TABLE OF CONTENTS

AN OPTIMALITY-THEORETIC ANALYSIS OF STRESS IN THE ENGLISH OF

NATIVE ARABIC SPEAKERS ......................................................................................... i
Acknowledgements........................................................................................................ iii
ABSTRACT...................................................................................................................... vii
Chapter One: Introduction .................................................................................................. 1
1.1 Rationale of the study ......................................................................................... 1
1.2 The Phonetics of Stress....................................................................................... 6
1.3 Stress and word recognition................................................................................ 7
1.4 Universals ......................................................................................................... 13
1.5.1 Optimality Theory: an overview................................................................... 16
1.5.1.a Constraints ............................................................................................... 18
1.5.1.b Constraint interaction............................................................................... 23
1.5.2 Optimality Theory: exemplification ............................................................. 24
1.5.3 Optimality Theory: concluding remarks....................................................... 27
1.6.1 Statement of the Problem.............................................................................. 29
1.6.2 Methodology ................................................................................................. 37
Chapter Two: Word-Stress ............................................................................................... 39
2.1 The phonology of stress .................................................................................... 39
2.1.1 Generative Theoretical Approaches to word-stress ...................................... 40
2.1.1.a Linear SPE Approach .............................................................................. 41
2.1.1.b Nonlinear Approach................................................................................. 45
2.1.1.c Optimality-Theoretic Analysis................................................................. 53
Chapter Three: Constraint Interaction in Arabic and English .......................................... 71
3.1 Overview........................................................................................................... 71
3.2 The constraints .................................................................................................. 73
3.3 An Optimality Analysis of Stress in Arabic and English ................................. 89
3.3.1 Stress Assignment in Arabic ............................................................. 90
3.3.2 Stress assignment in English ............................................................. 94
3.3.3 English Stress vs. Arabic Stress: Additional Remarks ............... 98
3.4 Constraint Interaction in Arabic...................................................................... 100
3.4.1 Conclusion ........................................................................................... 150
3.5 Constraint interaction in English .................................................................... 152
3.5.1 Stress in monomorphemic words ................................................... 153
3.5.1.a Mora count ............................................................................................. 154
3.5.1.b The Analysis .......................................................................................... 156
3.5.2 Stress in polymorphemic words ..................................................... 177
3.5.3 Secondary stresses ............................................................................. 182
3.5.4 Conclusion ........................................................................................... 184
3.6 Endnotes.......................................................................................................... 186
Chapter Four: Stress Acquisition .................................................................................... 189
4.1 Acquisition of English Stress.......................................................................... 189

v
 4.2 Constraint Demotion....................................................................................... 190
4.3 Constraint ranking in Arabic and English....................................................... 202
4.3.1 Basic facts and generalizations ....................................................... 202
4.3.2 Stress errors in the English of native speakers of Arabic ........ 209
4.3.2.a Lx=Pr, MAIN-RIGHT, TROCHAIC..................................................... 210
4.3.2.b FAITH and NONFINAL........................................................................ 215
4.3.2.c REDUCTION ........................................................................................ 217
4.3.2.d BIMORAIC............................................................................................ 221
4.3.2.e WSP ....................................................................................................... 223
4.3.2.f ALL-FEET-RIGHT and ALL-FEET-LEFT .......................................... 226
4.3.2.g FOOT-BINARITY and PARSEσ .......................................................... 232
4.3.2.h PARSEσ and NONFINAL .................................................................... 235
4.3.2.i FOOT-BINARITY................................................................................. 238
4.3.2.j FOOT-BINARITYσ and WSP .............................................................. 241
4.3.2.k MAIN-RIGHT and WSP(≥µµ).............................................................. 248
4.4 Conclusion ...................................................................................................... 255
References....................................................................................................................... 260
Appendixes ..................................................................................................................... 271

vi
ABSTRACT

The overall purpose of this study is to analyze the acquisition of English word-stress by
Arabic speakers in light of advancements in Optimality Theory. It has been reported that
Arab second language learners of English have difficulty in acquiring the various patterns
of English word stress. According to OT, the reason for this difficulty is that although
these speakers, like native speakers, have full command of the universal and violable
constraints that are operative in determining where stress falls in the word, they fail to
capture or induce the exact ordering of these constraints. The basic premise of OT is that
each grammar is a unique way of ordering the set of universal and violable constraints
that determine the actual output form of a certain linguistic feature, say word-stress in
this case. In other words, whereas Arabic word-stress and English word-stress are both
subject to the same set of universal and violable constraints, they differ in one respect: the
ordering of these constraints. The sole task of the learner then is to capture the correct
ordering that determines which syllable in each word carries main stress.
This study consists of four chapters. In chapter one, we introduce the problem of
the study and the basic background information for an OT analysis, the task we undertake
for word stress in subsequent chapters. Chapter two reviews word-stress placement in
three competing models: linear approach (Chomsky and Halle 1968), nonlinear approach
(Liberman and Prince 1977; McCarthy 1979; Hayes 1980, 1982, 1991), and finally
Optimality Theory (Prince and Smolensky 1993; McCarthy and Prince 1993a, b). In
chapter three, we introduce the set of constraints that are relevant for predicting the place
of stress, not just in English and Arabic, but in all languages. Hence, these constraints are
literally present in all languages, though their ranking is language-specific. Then, we
develop a ranking of the set of constraints particular to Arabic and another one particular
to English. In chapter four, we set out to compare the two constraint rankings in order to
(1) predict stress errors in the interlanguage of native speakers of Arabic when learning
English, and (2) demonstrate how, by making use of the notion of constraint demotion,
those learners can make their English more native-like with respect to stress placement.
This study has diverted from a standard OT analysis in at least two ways. First,
we allow for some alignment constraint (namely MAIN-RIGHT) to be interpreted as a
nongradient constraint. Second, we allow for constraint parameterization. NONFINAL is
parameterized to account for Arabic word stress; and WSP is parameterized to account
for English word stress.
This study has shown that there are significant differences between Arabic and
English as far as the ranking of the universal and violable constraints is concerned.
Among the major differences are the following. (1) WSP is irrelevant for stress
placement in Arabic. (2) Arabic requires that FOOT-BINARITY be interpreted under a
moraic analysis, but English requires it to be interpreted under a syllabic analysis. (3)
Arabic requires constructing meterical feet from left to right (i.e. ALL-FEET-LEFT >>
ALL-FEET RIGHT), English require that it be the other way around (i.e. ALL-FEET-
RIGHT >> ALL-FEET-LEFT). (4) in plysyllabic words, whereas a final syllable that
weighs two or more moras is parsed in English, only a final syllable that weighs three
moras is parsed in Arabic. (5) Arabic requires that PARSEσ dominates FOOT-
BINARITY, but English requires the opposite ranking.

vii
Chapter One: Introduction

1.1 Rationale of the study

This study was first sparked off by the observations of the researcher regarding the

mastery of English stress by Arabic-speaking learners. As a teacher of English, it

sometimes occurred to me that some of the effort that second language teachers put into

teaching some aspects of the target language is fruitless. For example, when it comes to

teaching aspects of the suprasegmentals of English (e.g. word-stress; sentence stress,

intonation, etc.) to students at the university level, many teachers would agree that

classroom instruction becomes no more than a remedial task, that is, just to correct

students’ errors. Not only this, but even after spending one whole class period on the

difference between words such as ‘record’ and ‘record’, ‘rebel’ and ‘rebel’, ‘produce’

and ‘produce’1, etc., it comes as no surprise to the teacher of English to have many, if not

most, students confuse the two as far as stress is concerned in less controlled speaking

drills. Many of the teachers I know have repeatedly made statements that would cast

doubt on the effectiveness of explicit instruction in second language teaching settings.

1
Bold here means stressed
That is, they may only reluctantly accept the idea that explicit knowledge can be

converted into implicit knowledge through practice.2

Another observation that contributed to drawing the attention of the researcher to

this particular acquisition problem, i.e., acquisition of English word-stress by Arabic

speaking adults, is related to native English speakers’ “interpretations” of some erroneous

utterances produced by nonnative speakers. It is not uncommon that in cases of

communication breakdowns, the native speaker reproduces the whole or part of what the

nonnative was trying to say, taking linguistic and situational factors into account, of

course. The phonetic execution of the utterance (including the acoustic properties of the

phonemes and the accompanying suprasegmental features such as intonation, stress,

juncture, etc.) by the nonnative is among the things that the native speaker should pay

attention to when ‘interpreting’ the utterance. It turns out sometimes that what the native

speaker thinks he has heard does not correspond with what the nonnative speaker has

actually said as far as the segmental information of the word is concerned. Examples of

‘mis-interpretation’ such as “come out” for “about”, “must fear” for “atmosphere”,

“retain” for “written” etc., are often cited in the literature (Bansal 1966; Lea 1973; Garnes

and Bond 1975; Browman 1978; Cutler 1980; Cutler and Clifton 1983; Cutler 1984: 79-

80, Benrabah 1997; among others). I myself have noticed this. When a friend of mine

2
As regards the impact of explicit instruction on second language acquisition, two viewpoints have
emerged: the interface view and the noninterface view. Whereas the proponents of the former claim that
explicit knowledge can facilitate implicit knowledge, those of the latter claim that explicit knowledge
cannot result in implicit learning. For the effect of explicit instruction on second language acquisition, see
de Graaff (1997); Hulstijn (1997); Robinson (1997, 1996); Alanen (1995); VanPatten & Sanz (1995);
Hulstijin & de Graaff (1994); VanPatten (1994); VanPatten & Cadierno (1993); DeKeyster (1993); Fotos
(1993); Doughty (1991); Yan-Ping (1991); Harley (1989); Scott (1989); among others). For a complete
overview, see de Graaff (1997) and Hulstijn (1997).

2
who is an Arabic-speaking learner of English said [firist] for [fiгst] in a certain context,

the native speaker’s response was: “excuse me, you said “the gift”?

The crucial factor here is that the segmental information does not help the native

speaker to figure out precisely what the nonnative speaker is trying to say. Stress

placement has played a more decisive role. That is, the native speaker singles out a native

word that has the same stress pattern as the word that the nonnative speaker has actually

mispronounced. When the native speaker perceives “retain” for “written”, for example,

he then relies on the stress pattern “weak-strong” used by the nonnative speaker when

uttering “written”. Stress placement can then determine how well native speakers will

understand nonnative talk (Fudge, 1984: 4; Kenworthy 1987: 28). This is tantamount to

saying that stress misplacement is very detrimental to intelligibility, probably more than

any distortion to segmental pronunciation (Garnes and Bond, 1975; Hubicka 1980, 1981;

Aitchison, 1994; Dalton and Seidlhoffer 1994: 39; Benrabah 1997).

Besides, when stress is misplaced, other aspects of pronunciation are affected as

well. Vowel quality, accent and intonation are some examples that are often referred to in

the literature (Allen 1971; Roach 1983 [1991]; among others). As for vowel quality, there

is a tendency in English that unstressed vowels are reduced to a schwa. When the syllable

containing a reduced form is wrongly stressed by the nonnative speaker, the strong form

of the vowel is also wrongly retained. This is what actually happens with Arabic-

speaking learners of English when they communicate in English: They hardly, if ever,

make use of reduced forms. The English vowel reduction system is foreign to them as

their native language preserves all vowels in their full forms. This may then lead to

3
having those unreduced, but unstressed, forms “compete with the stressed element for the

(English) listener’s attention” (Allen 1971: 81).

The issue is not very much better when it comes to accent (or sentence stress), which

is, grossly speaking, assigned to the most important word in the utterance in that “[t]he

more important a word is, the stronger is its stress’’ (Jones 1972: 262). Coupled with the

fact that incorrect placement of sentence stress may affect the overall intonational pattern

of the whole utterance, a staccato-like rhythm will be characteristic of learners’ speech,

and thus can adversely affect the comprehensibility of the message they intend to convey.

To illustrate, intonation is said to have an “accentual function” in that it signals the word

that is most prominent in the tone unit (Roach 1983: 143). This is done by dividing up the

discourse into information units, and assigning an accent to one content word in each

unit-- the one that carries new information (Mathesius 1975; Clark & Clark 1977; Prince

1981, 1992, Bardovi-Harlig 1983; Avery and Enrlich 1992: 76; among others). So in

“George has plans to leave”, if sentence stress is placed on ‘plans’, then the sentence will

be interpreted in a way completely different from that where stress shifts to “leave”, for

example3. In a nutshell, stress misplacement by the nonnative speaker does not often help

the listener (the native speaker in this case) identify points of focus in the speech stream,

and thus, fails to draw a line of demarcation between what is given and what is new as it

3
Following Chafe (1970: 213), we would like to assume that, as a general tendency, “those surface
structures items which reflect new information are (with some exceptions) spoken with a higher pitch (and
greater amplitude) than those which reflect old information.” The discussion here excludes altogether the
assignment of stress for emphasis or contrast, simply because every lexical item in a certain context,
including prepositions and form words, which are marked for “their high word frequency and their low
sentence stress” (Philips 1983: 487), can be stressed contrastively. Bolinger (1985: 85) argues that “the
focusing of a preposition is like the focusing of any other word”. So, I think we need to draw a distinction
between those stressed contrastively, and those stressed noncontrastively. Otherwise, the speaker may
choose to make a “declared contribution” (Gussenhoven 1983: 383) to the context, and thus, highlight the
meaning of any lexical item, for “whenever emphatic stress occurs in a sentence, it overrules normal stress”
(Hogg and McCully 1987: 4).

4
would in normal native-to-native interaction. This may not only lead to communication

breakdowns, but it may also result in serious misunderstanding. Consider the following

example where the nonnative speaker wrongly stresses ‘can’ (i.e., uses the strong form

where the weak form is more appropriate):

‘I can (kæn) do it’ instead of ‘I can (kn) do it’

As Kenworthy (1987: 124) puts it, “It will sound as if the speaker is protesting or denying

a previous statement (‘I can do it though you say I can’t’) when this meaning is not

intended.”

A third observation that attracted the attention of the researcher to this particular

pronunciation problem on the part of Arabic-speaking learners of English is related to

those learners’ ability to perceive stress as an auditory feature characteristic of the speech

of native speakers. In short, they are most of the time unable to detect the syllable which

is most prominent even if produced by a native speaker of the language. After further

probing into the problem, it turned out that native speakers of Arabic have serious

problems detecting the stressed syllables in even their native language (i.e., Arabic), and

they often confuse it with another related pronunciation aspect, namely, gemination. In

order to experimentally prove that, the researcher conducted a simple perception test on a

small sample of Arabic-speaking subjects. The description and findings of the test are

discussed in Appendix (2) below.

5
1.2 The Phonetics of Stress

It is time, I believe, to back up a little to briefly comment on what it means to say that a

certain syllable of a polysyllabic word is stressed. Phoneticians (and phonologists as

well) generally agree that when a syllable in a word is relatively more prominent than all

adjacent syllables, it is said to be stressed. They further agree that there are different

degrees of stress: primary, secondary, tertiary and even quaternary (or zero) stress

(Chomsky and Halle 1968: 16; for details see Cruttenden 1986:21-23). However, the

most debatable issue in this regard is related to how a certain syllable can be made more

prominent than another.

Scholars assume that stress is a composite phenomenon that can be understood by

reference to a number of independent parameters. In addition, they believe that stress has

some phonetic correlates (cf. articulatory, auditory and acoustic correlates4). Flege and

Bohn (1989: 38), for example, point out that “tension in vocal folds, change in timing of

articulatory movements, and change in subglottal pressures” are the articulatory

correlates of stress. The corresponding auditory correlates, they mention, include “change

in pitch, change in duration and change in loudness”5. It is unclear, nevertheless, that

these parameters are of equal importance either in production or perception. Some

researchers (Cf: Fry 1958; Juffs 1990) claim that pitch height is the most important

auditory parameter. Bolinger (1958)6 too thinks that loudness and duration serve as

4
For a detailed survey of the classical works on this point, see Adams (1979), especially chapters 4, 5, and
6.
5
The acoustic correlates include frequency, length, and amplitude/intensity, but they are seen as falling
outside this research proper.
6
“A Theory of Pitch Accent in English” reprinted from Word 14, 1958, (pp.109-49), in Forms of English,
p. 17

6
auxiliary clues as far as the phonetic correlates of stress are concerned, with pitch

prominence being the main cue. The correlation between these parameters is quite patent

in the sense that they are dependent on each other, and stress cannot be attributed to one

single parameter (Roach 1983: 74; Flege and Bohn 1989: 38). Besides, these parameters

may not be equally important cross-linguistically, either. Whereas loudness and duration

serve subsidiary function relative to pitch prominence in English, for example, they may

not be less important in Arabic.

1.3 Stress and word recognition

A preliminary question that should be addressed here is: How important is stress

in word recognition? First of all, it should be noted that the discussion here assumes a

holistic approach to sound discrimination and word recognition. That is, there is no one

single factor that is crucial to word recognition during language comprehension at all

times. Rather, it is the interplay of segmental and non-segmental properties that usually

influence speech-sound discrimination. Within the framework of this paper, stress is seen

as one of the major factors that can determine what listeners “think” they hear. And any

disruption of this feature may “significantly influence the discrimination of target

sounds” (Goldman et al. 1980: 153). Echoing Cutler (1984:80), “stress pattern identity

can precipitate false recognition, often in defiance of segmental evidence.” Gilbert (1984:

19) goes even further to claim that lexical entries are stored in the mental lexicon

according to their stress patterns. This is tantamount to saying that when a word is

7
pronounced with the wrong stress pattern, "the listener may spend some time searching

stored words in the wrong category”.

The claim that word stress is crucial in word recognition can best be seen if we

hypothesize that speech is not perceived as a linear sequence of discrete sound units

(Goldman et al 1980; Grosjean and Gee 1987; Dalton and Seidlhoffer 1994: 39). One

piece of evidence in favor of this hypothesis may in part be based on the impossibility on

the part of listeners to hear everything when listening to someone speaking; rather, they

“use a variety of clues to actively reconstruct the words they recognize” (Aitchison 1994:

84). In addition, there are various properties of spoken words that affect their recognition,

such as frequency of usage (see Howes 1957; Rubenstein & Pollack 1963), their length

(see Grosjean, 1980), their phonotactic configuration (see Jakimik 1979), and their

uniqueness point ( see Marslen-Wilson 1984).

Within the framework of this paper, stress is seen as one of the most crucial

auditory clues that listeners usually pay attention to. Following the proposal made by

Cutler (1976), Bradley (1980), McClelland and Elman (1986), and Grosjean and Gee

(1987), the belief that words are recognized in a linear fashion, one after another, left-to-

right (see Cole and Jakimik 1979: 133-134) is abandoned. A more convenient proposal is

to hypothesize that word recognition is based on the prosodic structure of the speech

stream (Grosjean and Gee 1987)7.

In this model, the domain of application of word recognition is the phonological

unit (e.g., phonological word, phonological phrase or even intonational phrase) which

7
The proposal here assumes that speech recognition in human beings is different from that in machines

8
may or may not correspond to the traditional syntactic unit (e.g., word, phrase, clause).

The phonological unit that is of primary importance as far as stress and speech

recognition are concerned is the phonological word defined along the following lines:

“The phonological word is a tightly bound phonological unit that is made up of one

stressed syllable and a number of weak (unstressed) syllables that are phonologically

linked to it” (Grosjean and Gee 1987: 142)

In this model, which relies on the prosodic structure of the speech stream, word

recognition involves two types of analysis: lexical search analysis and weak syllable

analysis. Whereas the former is initiated by the stressed syllable(s), the latter is

responsible for identifying weak syllables that fall on either sides of the stressed syllable.

These include (all examples are taken from Grosjean and Gee 1987):

1. unstressed syllables of content words (e.g., “poral” in “temporal”)

2. affixes attached to a stem (e.g., “un” and “ful” in “unhelpful”)

3. clitics attached to content words (e.g., French “j’” in “j’peux”)

4. reduced functors lexically attached to content words (e.g., have to = “hafta”, out of =

“outta”)

5. function words phonologically linked to content words (e.g., “a” in “a dog”, “him” in

“hit him”).

It is worth mentioning here that lexical access for stressed and unstressed syllables is

processed differently: One is triggered by the stressed syllable; the other is “identified by

means of a pattern-recognition-like analysis and with the help of the listener’s knowledge

of phonotactic and morphophonemic rules” of the language (Grosjean and Gee 1987:

9
145). However, although the two are processed differently, they constantly interact with

one another and with other sources of information such as the immediate situation,

speaker’s knowledge of the world, etc. So as not to sacrifice the complexity of the issue,

it is more convenient to view spoken word recognition as “a rather complex process that

involves various narrowing-in and monitoring stages, correcting stages, post-access

decision stages, and even look-ahead and look-back operations” (ibid. p. 136)—for

further information on this subject, see Grosjean (1980) and Swinney (1982)

What really matters here is the proposal that lexical access is a process that makes use

of the prosodic structure of the speech stream, in that it puts special emphasis on stressed

syllables. Let’s see how this prosodic organization helps give special emphasis to the

stressed syllable, which will then greatly influence speech recognition on the part of the

listener.

The suprasegmental, hierarchical organization of the word (or even utterance) may be

defined in terms of binary branching metrical trees (see McCarthy 1979a, b) where all the

sister nodes in the tree bear strong/weak (s/w) prominence relations to each other as

illustrated by Figure 1 below:

s
w

s w s w

s w

10
The syllable that is dominated exclusively by S’s throughout the tree structure is the one

that is most prominent8. Words are not then recognized in a linear fashion, one-to-one,

left-to-right (see Cole and Jakimik 1979: 133-134; McClelland and Elman 1986); rather,

the whole speech stream is parsed in terms of weak and strong syllables, where the most

prominent syllable (that which is dominated by S) is picked out to initiate a lexical search

that will invoke all word-like entries in the lexicon, that is, those with the same number of

syllables and, more importantly, with the same stress pattern. The best candidate is

selected based upon the stressed syllable, and is then judged by how well this syllable fits

with the other unstressed syllables to its right and left (Grosjean and Gee 1987; Dalton

and Seidlhoffer 1994: 39). So when the native speaker perceives the nonnative’s

production of “suitable” as “the level” or “normally” as “no money”, “upset” as “absent”,

etc., he relies on the stress pattern produced by the nonnative (i.e., suitable, normally and

upset) to interpret what the nonnative speaker is trying to say.

The problem of the second language learner in mastering the stress pattern of

English is not restricted to his mastery of primary stress (i.e., which syllable of the word

carries primary stress). In addition, the learner must also master of zero, as well as

secondary and even tertiary, stress due to the alternating stress pattern of English

(Dickerson 1978: 128). Research on primary stress, however, has not reached a point

where the information gained from the theory about primary stress can be successfully

translated into pedagogically viable generalizations—let alone research on other degrees

of stress (i.e., secondary stress in a three-level system or tertiary stress in a four-level

system). Even with the fine treatments of word stress that have emerged lately, the

8
We will discuss how the tree is built in subsequent sections

11
traditional debate concerning the linguistic foundation of English word stress (i.e,

whether stress is derived by rule or is part of the mental lexicon) remains unresolved.

Many of the traditional approaches to this particular pronunciation problem are still the

sole inspiration to classroom practice as far as stress is concerned. The bottom line to

almost all those pedagogical treatments is: We cannot help it; it is unpredictable

(O’Connor 1976:115 [1980]; Gimson 1980; Roach 1983 [1991]; Kreidler 1989; inter

alia)

The belief that much of English stress can be predicted (Chomsky and Halle

1968; Halle and Keyser 1971; Liberman and Prince 1977; McCarthy 1979; Hayes 19809,

1982, 199110; Halle and Vergnaud 1987; Prince and Smolensky 1993, McCarthy and

Prince 1993a, 1993b) has not yet been fully translated into viable pedagogical treatments

that would make the job of the second language learner less tedious. Several reasons have

led to this state of affairs. First of all, classroom practice (at least in those situations I am

familiar with) has been working under the tenet that second language learning is more

habit-governed than rule-governed. So, the formulation of cognitive generalizations has

at best been devoted to the segmentals, leaving altogether the suprasegmentals to the

learner’s ability of induction. In other words, what most second language teaching texts

do is teach the learners the sounds of the word; but they do not teach those learners how

to articulate those sounds. According to Dickerson (1978: 132), this is due to the neglect

of word-level phonology in second language teaching settings.

9
[Distributed 1981: IULC] [Published 1985: New York, Graland]
10
[Published 1995: Chicago, University of Chicago Press]

12
1.4 Universals

Although learner variables such as age, exposure to the target language, aptitude,

motivation, attitude towards the target language, etc., are always relevant in shaping the

acquisition of target language phonology, they are seen as falling outside the research

undertaken here. The assumption of the present research is that there are language

universals underlying the acquisition of LANGUAGE—be it first or second, etc11. A

good understanding of the role these universals play in shaping the learner’s

interlanguage helps us understand the way s/he performs in that language. Conversely, a

careful examination of the learner’s interlanguage system may also help us uncover those

underlying language universals.

A preliminary question that we need to tackle at this point is: what do we mean by

language universals? As early as the outset of the sixties, Greenberg (1962, 1966) started

the search for those aspects of language that are characteristic of many, if not all,

languages. The search has yielded what are now known as implicational language

universals of the form (if…, then…). The idea that there are language universals has been

pushed to the limit within the framework of Chomsky’s model of ‘universal grammar’.

According to Chomsky (1981b: 38), “UG consists of a highly structured and restrictive

system of principles with certain open parameters”. The argument underlying this line of

research is that the linguistic properties that all languages have in common are greater

than what surface representations actually reveal. Languages vary in two main ways: (1)

each language has fixed a parameter in a different way; or (2) a parameter is operative in

11
For an alternative view, see Wong, R. (1987b)

13
one language but not operative in the other (White 1990: 48). These differences between

languages, it is assumed, are rather insignificant, as the parameters along which

principles of universal grammar vary are limited (White 1990). Otherwise, the task of the

language learner would be impossible if these parameters are infinite. This argumentation

bears on the assumption that human beings are all equipped with the same innate

“language acquisition device” (LAD) (Chomsky 1981b). Children, irrespective of their

native language(s), go through almost identical developmental stages when acquiring

their native language(s). Even when the input is scanty or imperfect, children are still

capable of acquiring language (Chomsky 1981b; Hornstein and Lightfoot 1981; Lightfoot

1982). One way to explain this is to assume that human beings are genetically

predisposed to acquire LANGUAGE, and there is then every reason to believe that they

have a deductive ability which enables them to formulate hypotheses, test them, modify

them, and ultimately adopt or reject them altogether (see Chomsky 1972). A substantial

body of research has also shown that second language acquisition goes through similar

steps as first language acquisition (see Dulay and Burt 1974, White 1990). Be that as it

may, the relevant idea here is that there are ‘principles’ (which are given in all languages)

and ‘parameters’ (which allow for variations between languages). One crucial point

inherent in this line of thought is the belief that for a linguistic property or generalization

to count as a language universal (i.e. principle), it must not be violated by any language

whatsoever. In other words, absolute universals are considered inviolate.

Some linguists have argued that this view is too strong. A more realistic approach,

they believe, is to draw on linguistic oppositions. That is, a linguistic structure, be it

phonological, semantic, etc., has an ‘unmarked’, as well as a ‘marked’, value.

14
‘Unmarked’ is used in the sense of ‘more universal’. In Greenberg’s model (1962, 1966)

‘unmarked’ is equivalent to ‘more frequently attested’ cross-linguistically. To Chomsky

(1972, 1981b), it means ‘more basic’ and ‘first acquired’. For example, when we

hypothesize that voiceless stops are less marked than their voiced counterparts, we are

committing ourselves to the belief that if language X, for example, has a voiced stop, it

also has its voiceless counterpart, but not necessarily the other way around; or that the

voiceless stop is acquired before (or perhaps simultaneous with) the voiced counterpart,

hence ‘more basic’. It is clear then that ‘markedness’ is a relative term (i.e., form X is

only unmarked in comparison with form Y). Such being the case, the function of

‘markedness’ remains at best evaluative (for details see Kager 1999: 2-3).

The need, then, is to maintain the basic idea underlying the theory of

‘markedness’ and that underlying the theory of ‘principles-and-parameters’, as neither of

them communicates the whole truth by itself. An approach that makes use of both of

them would have more explanatory power. Such an attempt is made within the

framework of Optimality Theory (McCarthy and Prince 1993a, b; Prince and Smolensky

1993). As this research project is couched within this framework, we will first introduce

the basic concepts of Optimality Theory (1.5.1 below), and then we will go into details

with some exemplification of the ideas (1.5.2). In (1.6) we will state the problem that this

research study sets out to investigate.

15
1.5.1 Optimality Theory: an overview

The belief that each language uniquely reflects the structure of LANGUAGE has been

carried over into Optimality Theory (Prince and Smolensky 1993, McCarthy and Prince

1993a, b). This has come about as a result of the realization that there are linguistic

properties that are common to all languages, and that one major goal of linguistic theory

is to uncover those underlying properties (along the same lines of Greenbergian

implicational universals and Chomskyian ‘principles-and-parameters’). Yet, there is also

the realization that different languages show surface differences that are rarely trivial.

Within Optimality Theory this twofold goal (that languages share the same set of

universal principles, and yet allow for variation) is achieved by proposing the concept of

“conflicting universal constraints”. That is, there are constraints which are ‘universal’ in

the sense that they are equally active in all languages; meanwhile, those constraints may

be in conflict with each other as the realization of one constraint in an actual surface

representation may hinder the realization of one or more other constraints in that same

surface form. In this theory where “grammaticality equals optimality” (Prince and

Smolensky 1997: 1604) what the grammar of each language does is resolve this conflict

between constraints. As each grammar ends up with a different surface form, each

grammar then resolves the conflict uniquely: By invoking the notion of ‘constraint

ranking’. In other words, although the Grammar of Language A and the Grammar of

Language B exhibit the same set of universal constraints, each grammar ranks these

constraint in a unique way, and ultimately the two languages end up with two different

surface forms, each of which is optimal.

16
All LANGUAGES SHARE THE SAME SET OF UNIVERSAL CONSTRAINTS AND
THEY DIFFER IN THE RANKING OF THESE CONSTRAINTS.

Optimality Theory sets itself apart from earlier approaches in at least two other

ways. First, it has abandoned the notion of rewrite rules. Second, it has recognized the

notion of violable constraints. In a rule-based grammar, rules may be linearly ordered in

that they may apply one after the other A, B, C, etc., and more importantly, the output of

the application of Rule A is the input of Rule B, etc. (Chomsky and Halle 1968). Within

Optimality Theory, however, constraints are structural requirements on surface

representations; so, unlike the traditional model, those constraints (or the rule in the

traditional sense) need not wait until their structural description is met; rather, they all

apply on the output form in parallel fashion. If it turns out that two or more constraints

are in conflict with regard to a certain output form, they both apply to that same surface

form. Hence the structural description of the constraint is the output form itself.

In autosegmental (nonlinear) approaches (e.g., Liberman and Prince 1977;

McCarthy 1979a, b; Hayes 1980, 1982, 1991; among others) there have been two

advancements: (1) the notion of rule-ordering is abandoned; and (2) the idea that

languages share a core set of fixed universal properties in the form of ‘principles’ on

which some specifications are set in the form of parameters is incorporated (White 1990).

Yet, there remains one basic idea that is shared with the traditional approaches of

generative grammar, namely, the belief that those principles are inviolate. It is only

within the framework of optimality theory (McCarthy and Prince 1993a, b; Prince and

Smolensky 1993) that universal constraints are seen as VIOLABLE, hence the existence

of marked structures in almost all languages of the world. The term ‘violable’ is used

17
here to mean that (1) an output form may violate a constraint and yet still be grammatical,

and (2) it is not a necessary condition that all constraints be satisfied for the output form

to be grammatical.

CONSTRAINTS APPLY TO SURFACE FORMS AND THEY ARE VIOLABLE

Having introduced the basic concepts of optimality, let’s examine these notions in

some detail. In section (1.5.1.a), we will examine the notion of constraints. In (1.5.1b),

we will examine constraint interaction. In (1.5.2), we will provide an exemplification of

these notions.

1.5.1.a Constraints

In this section we will try to provide an answer to one important question that should be

tackled when talking about Optimality theory: What is a constraint?

In Optimality Theory, a constraint is a structural requirement on surface

representation. Consider the following constraint based on some English data (used only

for exposition purposes)

Constraint A
English basic word order requires that grammatical subjects
come before verbs and grammatical objects

If this is truly a constraint, then we are making the assumption that a sentence like (1)

below is grammatical (which is true):

(1) The principal has left the meeting.

But is this constraint always satisfied? Consider for example (2) below:

(2) Has the principal left the meeting yet?

18
In (2), the constraint is not satisfied as the auxiliary verb ‘has’ comes before the

grammatical subject ‘the principal’. To remedy the damage, we may have two options:

(a) to assume that the sentence is ungrammatical because it does not satisfy the constraint

(which is not true, of course); or (b) modify the constraint a little so as to accommodate

example (2), and keep modifying it until it has the power to account for all English data

with regard to word order. If this seems to be a difficult and time-consuming task for one

single language, the reader could imagine then how much tedious the task is if we have to

account for all cross-linguistic variations for each constraint that we may claim to be

universal. Fortunately, within optimality theory we do neither. All we need do is accept

the idea that an output form may violate some constraint(s) like (A) above and still

grammatical as in (2) above. The question that promptly arises is: How can (2) be

grammatical if it violates the constraint in (A) above? The answer to this question lies in

assuming that in English there is another constraint at work and it is stronger (or ranked

higher) than the one stated in (A) above, namely:

Constraint B
In yes/no questions, auxiliary verbs precede grammatical subjects

Here the two constraints (A & B) are then in conflict; that is, the output may satisfy one

constraint but only at the expense of the other. Needless to say, when two (or more)

constraints are in conflict, they can never be satisfied perfectly, i.e., the output form

cannot meet the structural requirements of both of them simultaneously, but they can be

satisfied optimally, i.e., the stronger constraint is satisfied at the expense of the weaker

one. In OP terms, one constraint is ranked higher than the other, and the violation of a

higher ranked constraint is a more serious business than the violation of a lower-ranked

19
constraint. The violation of X (as a higher-ranked constraint) may be fatal, i.e., yielding

ungrammatical output, but the violation of Y (as a lower-ranked constraint) may be less

serious and yields a grammatical output. So when constraint (A) is violated in example

(2) above, the output form is still grammatical, but if constraint (b) is violated in the same

example (like having the auxiliary verb in some place other than before the grammatical

subject, for example), then the output will undoubtedly be ungrammatical. This is a

sufficient reason to claim that B is ranked higher than A in English as far as that matter is

concerned.

VIOLATION IS NEVER A BINARY CRITERION. THERE ARE DEGREES OF

VIOLATION

In OT literature, two main classes of constraints are usually distinguished:

Markedness Constraints and Faithfulness Constraints. Markedness constraints are

structural requirements stated on the well-formedness of output forms (surface structure)

irrespective of the input forms, e.g.,

syllables must have onsets

syllables must not have codas
etc.

Faithfulness constraints, on the other hand, are needed to match the input forms with their

corresponding output forms, requiring that they be as similar as possible, e.g.,

output and input must be voiced

output syllable and input syllable must be of the shape CVC.
etc.

20
 Grossly speaking, whereas markedness constraints work on the output form,

faithfulness constraints pay attention to both input and output forms12. In so doing, they

keep the symmetry between a linguistic form and its different realizations or variants,

enhancing the basic shape of a given form in all its variants. For example, in order to

ensure that a hypothetical form like ‘wxyz’ is not realized as any combination of these

four characters irrespective of their order, or even worse, as some form that changes,

deletes or adds some other characters, we need a Faithfulness constraint to limit the

correct realization to that which best corresponds to the input form ‘wxyz’, ideally

‘wxyz’.

Both types of constraints are needed simply because if we have markedness

constraints only, then we expect output forms to be simple structures all the time. Let’s

imagine that the only types of constraints that are operative in syllable structure, for

example, are markedness constraints, then we may expect constraints like the following:

all syllables have one-consonant onsets

all syllables end with zero consonants (i.e., open syllables)

Etc.

And if we add to this the markedness constraints on vowels and consonants, we may end

up with no more than a couple of syllable types that are allowed by the grammar of all

languages. Conversely, if we have faithfulness constraints only, then we assume that

there are no linguistic generalizations whatsoever; hence those types of constraints “favor

replication of the input, even at the expense of structural complexity” (Prince and

Smolensky 1997: 1605).

12
Within this Model, there are no constraints that refer exclusively to the input

21
 One more thing that we need to mention here about these two types of constraints

is that they both are universal and violable. Constraints are universal in the sense that

they are all part of UG, and thus, part of the grammar of all languages. As Prince and

Smolensky (1997: 1605) succinctly put it, those constraints are “not just universally

available to be chosen from, but literally present in every language.” On the other hand,

these same universal constraints are violable in the sense that output forms never satisfy

both types of constraints perfectly, but only optimally.

But why are the constraints violable? The straightforward answer to this question

is that it is legitimate to violate a constraint only to avoid violating another. The one that

we tolerate a violation of is a lower-ranked constraint, but the one that we cannot tolerate

a violation of is a higher ranked constraint. This notion of violability is embodied in the

concept of ‘DOMINANCE’ in Optimality Theory: Constraints dominate each other.

Constraint B above dominates constraint A, simply because satisfying B is far more

important than satisfying A.

The last question that I think I need to tackle here before I move to the next

section is this: How is a constraint put forward? Or, what are the criteria that should be

taken into account in order to claim that a linguistic property could count as a universal

constraint? At least two conditions must be met. First, it must be attested in a large

number of natural languages. In other words, the cross-linguistic evidence should be

overwhelming for an intuitive linguistic observation to count as a constraint. For example

when we posit a constraint against syllable codas like C below, we then assume that the

majority (if not all) languages of the world tend to have syllables that do not have codas.

22
 Constraint C
NO CODA

But is this the only reason why an intuitive linguistic observation may be posited

as a constraint? Not at all. The other half of the story is that a constraint must be

phonetically motivated (from a production or perception point of view). A Constraint like

(C) above, in addition to being cross-linguistically motivated, is phonetically plausible.

For it is unlikely to assume that CVC, for example, is less marked than CV even if,

hypothetically speaking, it is found in the vast majority of the world languages. The

reason is simple: it is phonetically more marked than CV.

FOR A LINGUISTIC OBSERVATION TO COUNT AS A CONSTRAINT IT MUST

BE PHONETICALLY AND CROSS-LINGUISTICALLY MOTIVATED

1.5.1.b Constraint interaction

Languages share two things. First they share the set of universal constraints, and second

they share the way these constraints interact with each other. On the other hand,

languages are seen to differ in only one thing: the ranking of those constraints; each

language ranks the constraints in a unique way.

How do those universal and violable constraints interact?

Given the set of potential and actual output forms, the grammar of each language

evaluates these forms against the set of universal constraints that are literally available in

that grammar. The conflict among constraints continues to be resolved until precisely one

output form for each input form wins the competition. This implies that (1) we have to

have in advance all the possible output structures available, and (2) each language

23
evaluates those candidates against the set of universal constraints uniquely, hence the

realization of different surface forms in each language.

In OT, the first task is accomplished by the function Gen (generator) while the

second task is accomplished by Eval (evaluation). In order to retrieve all the possible

candidate forms for some input, Gen acts on a LEXICON that contains all of those

lexical forms. This is tantamount to saying “every language considers exactly the same

set of options for realizing an input” (Prince and Smolensky 1997: 1605):

Gen (input) ⇒ {cand1, cand2…candn}

This reads as: Being a function of UG, Gen applies to an input form and produces all the

possible candidate output structures irrespective of the structural context in which the

form occurs (i.e., it is context-blind). The function Eval (evaluation), on the other hand,

is language-specific. That is, each language ranks the set of universal constraints in a

unique way, so that only one form will ultimately win the competition

Eval {cand1, cand2…cand3} ⇒ {candk} output

This reads as: Eval evaluates all the candidate forms supplied by Gen, and singles out

precisely one output for each input with respect to a given ranking of constraints.

1.5.2 Optimality Theory: exemplification

In a rule-based theory, the input is mapped onto output through linearly ordered rewrite

rules of the shape

(1) A → B

24
For rule (1) above to apply, its structural description must be met in the input or in the

output of an immediately preceding rule. Let’s rewrite the above rule as a context

sensitive rule as in (2) below:

(2) A → B ⁄ C

Then the input configuration CA is realized as CB, but never as CA, or AC, for example.

Given the input CA, the configuration CB can be obtained in an optimality-

theoretic approach in the following way. First, we need a markedness constraint that

defines the configuration CA as ungrammatical:

(3) *CA

Second, we need a faithfulness constraint that guarantees that the output is identical with

the input, so as to avoid deriving a configuration like CD, for example. One way is to

have a constraint that defines any change on the input as ungrammatical as in (4) below:

(4) *A → B

If you notice, what we propose is in itself contradictory; that is, we have two conflicting

constraints: one defines the configuration CA as ungrammatical, and the other defines

anything other than CA as ungrammatical. Put more succinctly, one constraint acts

against the realization of A after C, and the other acts against the realization of anything

other than A after C. In OT literature, this kind of problem is represented in a tableau like

(1.1) below. Let’s use the Tableau to expose the violations of both constraints where the

constraints are arranged horizontally, and the outputs vertically (where * means violation

of a constraint):

25
Tableau (1.1)
*CA *A → B
CA *
CB *

In one instance the constraint *CA is violated; the other time *A → B is violated. In no

way, then, can the two constraints be met simultaneously. To resolve this conflict, OT

researchers propose that constraints can be satisfied optimally but never perfectly. What

this basically means is that EVAL, being a function of UG, evaluates the candidate

outputs (CA, CB, and possibly AC, BC, BA, AB, etc.) against the set of conflicting

constraints. But how does EVAL do this?

EVAL makes use of the notion of dominance:

Constraint X dominates Y if X is a higher ranked constraint and Y is a lower

ranked constraint.

For any conflict of this kind to be resolved, then, one of the constraints has to dominate

the other(s). The one that dominates is called a higher-ranking constraint, and the one that

is dominated is a lower-ranking constraint --usually listed in the Tableau in descending

order from left to right.

For the sake of exposition, let’s assume that, for example, (3) is a higher-ranking

constraint and (4) is a lower ranking constraint. Then the violation of (3) is more serious

than the violation of (4). To use OT terms, any violation of a higher-ranking constraint is

fatal, indicated by the exclamation point ‘!’. In this case, we expect the conflict to be

resolved by having a configuration that satisfies (3), i.e., CB as in (5a) below. Note that

the upper left-hand corner is always occupied by the input--CA in this example. All the

possible output configurations--CA and CB in this example-- are listed underneath the

26
input. The constraints-- *CA and *A → B occupy the uppermost cells across the tableau,

and the pointing hand " indicates the winning configuration:

Tableau (1.2)
CA *CA *A → B
CA *!
"CB *

By the same token, the output configuration CA will only win if we assume that (4) is a

higher-ranking constraint and (3) is a lower-ranking constraint. In this way, we may

sacrifice (3) only to avoid violation of (4). Hence, the violation of a higher ranked

constraint is ‘fatal’.

Tableau (1.3)
CA *A → B *CA
"CA *
CB *!

1.5.3 Optimality Theory: concluding remarks

Unlike the application of rules in a rule-based theory, the application of both markedness

and faithfulness constraints here depends crucially upon their ‘eventual consequences’ on

the surface. That is, the output (the ultimate form) itself determines whether or not a

certain constraint is violated; hence, OT is a surface-oriented theory. In a rule-based

theory, this is not the case since the output is the ultimate realization of the application of

all rules, so it is irrelevant whether the output violates any of the rules that are applied at

earlier stages of the derivation.

27
 The Grammar of each language then consists of both GEN and EVAL. Whereas

GEN, being a function of UG, provides for each and every language the same set of

possible surface forms for a given input form, the function EVAL, being a function of

UG too, evaluates that set of possible surface forms against the set of conflicting

universal and violable constraints by invoking the notion of DOMINANCE. Ultimately,

only one form wins, optimally but never perfectly.

This is the kind of grammar that we will be using in order to shed further light

upon the phenomenon of language learnability. In particular, we will be investigating the

performance of Arabic-speaking subjects in acquiring English stress patterns. The

proposal here is that the Optimality Theory model can better account for how second

language learners perform in acquiring the target language patterns. The basic assumption

is that adult second language learners, like children acquiring their first language, have

available to them the same set of universal and violable constraints. In particular, both

Markedness constraints and Faithfulness constraints are operative in the phonology of

second language learners. The only difference between first and second language

acquisition is related to how these constraints are ranked with respect to each other. To

account for why nonnative speakers’ phonology may not be 100% identical to that of

native speakers, it is claimed, following the basic premises of Optimality Theory, that

second language learners rank the constraints differently from native speakers. Let’s

assume for a moment that some second language learner of English, for example, fails to

produce a consonant cluster at the beginning of an English word like /skuul/ ‘school’, and

instead produces it as / kuul/, /iskuul/, /sikuul/, etc. This can be easily accounted for

within the framework of Optimality Theory by assuming that the phonology of this

28
second language learner still ranks certain Markedness constraints (those which favor

simple structures) over Faithfulness constraints (those which favor replication of the

input). As this learner moves on and starts producing the English /sk/ cluster, s/he realizes

the importance of having Faithfulness constraints outrank Markedness constraints, at

least in this example. We conclude by claiming that the phonology of the second

language learner gets improved as the learner realizes the importance of re-ranking the

set of universal constraints, which her/his phonology consists of in the first place.

Detailed examples will be given in subsequent chapters where we introduce the whole set

of constraints that are required to account for second language learners’ stress patterns.

1.6.1 Statement of the Problem

Natives of Arabic speaking countries are used to having some formal Arabic

Language teaching in schools. That is, because of the diglossic situation in the Arab

world, some variety of Arabic, usually called the high variety (H), is taught at school the

same way a foreign language is taught: Students are given grammar lessons,

pronunciation drills, vocabulary exercises, etc., in this “superposed variety”. All in all,

“the actual learning of H is chiefly accomplished by the means of formal education”

(Ferguson 1972: 293)13. Be that as it may, the point that I would like to make here is that

although I was educated in that way, I never remember that we had had any exposure to

13
Ferguson’s article first appeared in Word (1959), 325-340. However, we are using here its reprinting in
Giglioli (1972), 232-251.

29
the notion of stress in the Arabic Language classes. And when I tried to find out whether

native speakers of Arabic from other Arab countries have had exposure to the notion of

stress in their formal Arabic Language education in schools, it turned out that most of

them confuse the term ‘stress’ with another seemingly related phenomenon, that is,

‘gemination’14. What might have contributed to this confusion is the fact that while

Arabic orthography, for example, makes use of a diacritic called šaddah “ّ” for the

geminated sound, no similar diacritic is used for stress. In addition, as is mentioned above

no attention is given to the subject of accentuation in Arabic Language studies. Mitchell

(1960: 368) is “understandably puzzled by the failure of the Arab grammarians to

mention the subject of accentuation”.

The reason for this negligence of teaching stress in grammar classes, as well as

for not marking the stressed syllable with a diacritic in the orthography, may be due to

the fact that word stress in Arabic is “predictable and therefore non-phonemic” (Mitchell

1960; Larudee 1973; among others). The assumption is that when a native speaker is

confronted with a novel word, s/he makes no mistakes regarding which syllable of the

word should be stressed. This renders stress markings in Arabic textbooks altogether

unnecessary.

14
To verify this, I included in the list of words which I used to test the perception of stress by native
speakers of Arabic five words where stress falls on some syllable other than the syllables which contain
either leg of the geminated sound. For example, in a word like “?assamaa?u” (the sky), although the left
leg of the geminated sound /ss/ is the coda of the first syllable and its right leg is the onset of the second
syllable (counting from the left), neither of which is stressed; stress goes to the penult. The test, however,
shows that most of the subjects thought that stress falls on either the first or the second syllable, i.e. that
which contains one or the other leg of the geminated sound (for details see the discussion of the test results
in Appendix 2 below). Notice that although native speakers of Arabic can produce the word with the
correct stress pattern, they may not be able to figure out which syllable is stressed when listening to
someone’s production of the word)

30
 Larudee (1973) verifies the proposal that native speakers of Arabic never fail to

stress novel words in a predictable pattern by asking eight native speakers of Cairene

Arabic to read a list of ten nonsense words (i.e., words that are not actual but possible).

He shows that a native speaker may vocalize the unfamiliar word in one or more ways.

“But no matter how he reads it he follows a predictable pattern in choosing the stressed

syllable” (Larudee 1973: 32). Table (1) below shows the ten nonsense words and their

reading(s) by native speakers of Cairene Arabic:

Table 1
Nonsense Arabic words Readings

1. fdrt 1.1. fadarat

1.2. faddarat
1.3. fadarti
1.4. faddarti

2. fdrtk 2.1. fadartik

2.2. faddartak
2.3. fadaritik

3. mfsntkm 3.1. mafasnitkum

3.2. mafsantukum
3.3. mufassantikum

4. ksrtnlhm 4.1. kassartanilhum

4.2. kasartanilhum

5. šftlsnthm 5.1. šuftilsanithum

5.2. šuftalsintuhum

6. tbsntlkm 6.1. tabasnitilkum

6.2. tabsintilkum
6.3. tabasnitlukum

7. mkstšt 7.1. makastašat

7.2. maksatšat

8. fsntltkm 8.1. fasantiltukum

31
 8.2. fasantilitkum

9. hlmtrnsk 9.1. hilimtirinsik

9.2. hilimtirnisik
9.3. hilmitirinsik

10. drstnlkm 10.1. darastinilkum

10.2. darastinlukum
10.3. darsitinilkum

All of the readings of the nonsense words in the right hand side column are stressed

according to the predictable pattern of stressing words in Arabic. Larudee’s (1973: 32)

conclusion is that the native speaker of Arabic “has internalized a stress system for his

language which he employs in novel situations”.

This proposal has been extended to the area of second language acquisition to

explain why Arab learners of English fail to both produce and perceive the English stress

pattern correctly. Errors in stress assignment by second language learners of English,

which we will be using throughout, are taken from three studies: Aziz (1980) for Iraqis,

Anani (1989) for Jordanians, and Ghaith (1993) for Saudis.

The problems that native speakers of Arabic face when learning English as far as

word stress is concerned can be categorized as follows:

First, words ending in heavy unstressed syllables15 pose considerable difficulties

to Arab second language learners of English. Table (2) below shows examples of these

words and the way they are produced as far as primary word stress is concerned by

Arabic-speaking learners (Aziz 1980: 105-106):

15
Anani (1989: 16), for example, uses the term ‘long syllable’ to refer to any syllable which contains a long
vowel (cvv) or ends with more than one consonant (vcc). According to Larudee (1973: 33) cvc is
considered long in nonfinal position, but short in final position.

32
Table 2
English words Learners’ pronunciation in
terms of stress
- abbreviate, complicate, imitate - abbreviate, complicate, imitate
- obsolete, concrete - obsolete, concrete
- amplify, classify - amplify, classify
- fertile, hostile - fertile, hostile
- carbine, incline - carbine, incline
- empire, satire - empire, satire
- communism, colonialism - communism, colonialism
- novelist, scientist - novelist, scientist
- advertise, apologize, realize - advertise, apologize, realize
- expert - expert
- occupy - occupy
- multitude, solitude - multitude, solitude
- capsule, ridicule - capsule, ridicule
- contribute, substitute - contribute, constitute

Anani (1989: 18) reports that disyllabic and tri-syllabic words whose ultima is heavy (e.g.

dedicate, concrete, inside, Chinese, submarine, lemonade, photograph, etc.) are stressed

on that syllable by all of his subjects (natives of Jordan). Gaith (1993: 385), too, shows

that his subjects “fared the worst on three-or-more-syllables ones ending in heavy

syllables, such as exercise, monopolize, humidify, complement and manifest”. Gaith

argues that such words are stressed by the subjects on the basis of their final syllable

structure.

Second, words containing a long (heavy) unstressed penult pose a problem for

Arabic-speaking learners of English. Consider the following table:

Table 3
English words Learners’ pronunciation in
terms of stress
- ancestor, calendar - ancestor, calendar

33
Gaith (1993: 385) reports that words such as presidency, dependency, etc. are very much

problematic to his subjects. He believes that the difficulty may be due to the fact that the

suffixes here “do not affect the assignment of the primary stress on the bases”:

presidency = president, dependency = depend. However, we argue that this problem is

rather caused by native language transfer. Since suffixed and suffixed words are treated

alike as far as stress placement is concerned in Arabic, stress assignment cannot then be

not morphologically conditioned in that language. Words such as presidency,

dependency, etc., contain a heavy unstressed penult, and thus pose considerable

difficulties to Arab learners of English.

Third, English words which are stressed on a pre-antepenult syllable, especially if

the word ends in short (light) syllables, also pose a problem for Arab second language

learners of English. Consider the following table:

Table 4
English words Learners’ pronunciation in
terms of stress
- necessary - necessary16
- occasionally, vocabulary - occasionally, vocabulary17

Notice here that this problem underlies (1) shifting stress to one of the last three syllables,

and (2) choosing which of these three syllables should be stressed on the basis of their

phonological nature and their distribution relative to the other syllables in the word.

16
May be this is one of the differences between Iraqi Arabic (and some other dialects including Bedouin
Jordanian Arabic) and most other Arabic dialects (including Rural and Urban Jordanian Arabic). Whereas
words having the forms cv-cvc-cv-cv and cv-cv-cv-cv are stressed on the antepenult in Iraqi Arabic, they
are stressed on the penult in Jordanian Arabic. So necessary would be pronounced by Jordanian learners of
English as necessary.
17
According to Aziz, then, in Iraqi Arabic long words ending in two short (light) syllables are stressed on
the antepenult irrespective of its internal structure and the number of syllables that separate it from a
preceding heavy syllable or word boundary. In Jordanian Arabic, on the other hand, stress falls on a light
antepenult provided that it is separated from a preceding heavy syllable or word boundary by an even
number of syllables. So whereas occasionally, vocabulary are stressed in the same way by Iraqis and
Jordanians, necessary is produced differently by the two groups (see footnote 3 above).

34
 Fourth, English words whose stress pattern is dependent upon their parts of

speech (i.e. verb, noun, adjective) pose an additional problem to Arab second language

learners of English. This is so because in Arabic the syntactic information of the word is

irrelevant for stress placement. Stress in Arabic is a function of (1) the internal

phonological structure of the syllable (i.e. its weight), and (2) the position of that syllable

relative to the other syllables in the word. So, it is expected that native speakers of Arabic

stress English words based on their syllable structure, irrespective of their parts of speech

(i.e., verb, noun, adjective). Consider the following table:

Table 5
English words Learners’ pronunciation in
terms of stress
- collect (n), combine (n), compound (n) collect (n), combine (n), compound (n)
- rebel (v) rebel (v)

Fifth, English words with complex morphological composition pose problems to

Arab second language learners in at least two ways: (1) single-stressed compounds (e.g.

blackboard, breakfast, policeman) are viewed as consisting of two independent words,

each with primary stress. Consider the following table:

Table 6
English words Learners’ pronunciation in
terms of stress
- blackboard, breakfast, policeman blackboard, breakfast, policeman

And (2) polymorphemic words with secondary stresses are produced with only single

primary stress. Consider Table (7) below (primarily stressed in bold, secondarily stressed

in italics):

Table 7
English words Learners’ pronunciation in
terms of stress
- defrost, prejudge, rewrite - defrost, prejudge, rewrite

35
Ghaith (1993: 385) argues that the stressing of such words by his subjects can be

explained on morphological grounds, i.e. “they usually receive stress on the base”. We,

however, argue that these errors, too, are phonologically governed.

Sixth, words containing reduced vowels pose problems to second language

learners. Because of the influence of their native language, Arab learners of English tend

to retain the full value of the vowel irrespective of its stress status. Consider Table (8)

below (reduced syllables underlined):

Table 8
English words Learners’ pronunciation in
terms of stress
-several, preparatory, legendary, basically - several, preparatory

Finally, I should mention that native speakers of Arabic sometimes use other

simplification processes to modify on the English syllable structure, such as breaking

consonant clusters, replacing long vowels with their long ‘equivalents’ or vice versa, etc.

(Aziz 106-107). These processes are obviously motivated by the native language

grammar. Table (9) below gives some illustrative examples:

Table 9
English words Learners’ pronunciation in
terms of stress
- basically /beisikli/, severl /sevrl/ - basically /beisikæli/, severl /severel/
- trustee, shampoo - trustee /tr٨sti/, shampoo /šæmpu/

Notice that given the new phonological structure of the word, stress is assigned in

conformity with the stress assignment rules of Arabic.

36
 Aziz (1980), Anani (1989) and Ghaith (1993), among others, then, agree that

native speakers of Arabic carry over their native language fixed word stress system to

English test words.

1.6.2 Methodology

In the previous section, it is established that native speakers of Arabic face considerable

difficulties in acquiring the English stress system. Traditionally, it is thought that one

possible reason for this difficulty is that English stress is not highly regular. There is, for

example, a wider range of stress patterns in English than in Arabic. Different kinds of

information (i.e., syntactic, morphological, phonological) should all be taken into account

in order to capture the generalizations regarding the place of English stress. Arabic, on

the other hand, makes use of only limited phonological information stress placement

purposes. To illustrate, stress can be easily predicted given (1) the phonological nature of

the syllable (namely its weight: light, heavy, etc.), and (2) its distribution relative to the

other syllables in the world.

The overall purpose of this study is to account for stress errors by Arab learners of

English within the framework of Optimality Theory (OT). Since this theory views each

grammar as a set of universal, violable constraints, and cross-linguistic variations result

from different ranking of these constraints, the researcher needs to find out:

1. the set of constraints that are relevant for predicting the place of stress, not just in

English and Arabic, but in all languages. Hence, these constraints are literally

present in all languages, though their ranking is language-specific;

37
2. the ranking of the set of constraints in the learners’ native language (Arabic in this

case);

3. the ranking of the set of constraints in the target language (English in this case);

4. the exact problems that learners face in learning the target language;

5. how to account for these problems by comparing the two languages’ different

rankings of the constraints; and

6. how learners can improve his/her performance. Here we will see that this can be

done by making use of the notion of constraint demotion.

Each of these issues is dealt with in subsequent chapters.

38
Chapter Two: Word-Stress

2.1 The phonology of stress

Much research has been done to uncover the general linguistic principles that govern

stress placement in one language and cross-linguistically. Among the most influential

analyses of stress placement are Chomsky and Halle (1968), Liberman and Prince (1977),

McCarthy (1979), Hayes (1980, 1982, 1991), Halle and Vergnaud (1987), Prince and

Smolensky (1993), McCarthy and Prince (1993a, 1993b). The basic question that they all

set out to answer is: How can we predict where stress will fall in Language X (where X

can be any language)? Of special interest to all those approaches are languages whose

word-stress patterns are not fixed such as Arabic, English, etc. Needless to say, there are

some languages that have fixed stress patterns. What we mean by ‘fixed stress pattern’ is

that stress falls on one fixed syllable (maybe the initial as in Finnish, the ultimate as in

Turkish, the penult as in Spanish, etc.) within every word. Languages such as these do

not pose a challenge to cross-linguistic phonological investigations about word-stress

behavior.
 The various treatments of word-stress are based on different types of information.

They have achieved some success when dealing with some (but not all) languages. The

SPE model (Chomsky and Halle 1968), for example, is relatively successful in

accounting for English stress patterns. However, many researchers (Abu-Salim 1982;

Alghazo 1987; inter alia) have questioned the capability of that model to handle Arabic

word-stress patterns18. They claim that the nonlinear approach, for example, can better

handle Arabic word-stress. Despite the dispute over which approach can best handle the

data, the simple fact remains that almost all these approaches have admitted some

residual data that cannot be easily accounted for, and all of them encounter theoretical

problems that require extensive future research.

Moreover, as far as stress is concerned, the main focus of some of these

approaches has been to account for primary-word stress. It turns out then that these

approaches have to survive another test: to account not only for primary word-stress, but

for other degrees of stress (secondary, tertiary, etc.) as well. Unfortunately, very little has

been done in this direction.

2.1.1 Generative Theoretical Approaches to word-stress

In this section, I will try to give an introduction to the basic tools used to account

for word-stress placement in three competing models: linear approach (Chomsky and

Halle 1968), nonlinear approach (Liberman and Prince 1977; McCarthy 1979; Hayes

18
For an analysis of Arabic word-stress based on the SPE model, see Abdo (1969) and Brame (1971, 1973,
1974).

40
1980, 1982, 1991), and finally Optimality Theory (Prince and Smolensky 1993;

McCarthy and Prince 1993a, b).

2.1.1.a Linear SPE Approach

According to Fudge (1984: 11), the SPE model relies on two types of information

to account for word-stress: “the segmental make-up of the word” and “the internal

structure of the word”. Chomsky and Halle (1968: 18) assume that stress, like any other

phonological feature, can be accounted for by simple rules, so they make use of two rules

(Cf. Compound Stress Rule and Nuclear Stress Rule19) to account for the distribution of

stressed and unstressed syllables. Besides, their analysis relies on the claim that lexical

items are hierarchically organized. As far as stress is concerned, for example, they

incorporate the idea of cyclic application of the primary stress rule in that [1 stress] is

assigned to the most deeply embedded constituent, and so on. Meanwhile, when a

constituent receives [1 stress] all other stresses in the same domain of application are

reduced by one level.

One of the finest treatments of stress that is modeled on the SPE approach is

Brame’s analysis of Arabic stress assignment (see Brame 1971, 1973, 1974). According

to Brame (1971: 560; 1973: 15; 1974: 40), the place of main word-stress in Arabic can be

captured by the following rewrite rule:

(1) V → [+stress]/ — C0 ((VC) VC

19
Whereas CSR assigns primary stress to the leftmost stressed vowel of a lexical category (e.g., black
bird), NSR assigns primary stress the rightmost stressed vowel of a phrasal category (e.g., black bird)

41
As stated, (1) imposes “an interpretation of disjunctiveness”. That is, the rule in (1) is an

abbreviation of three rewrite rules (Brame 1971: 560):

(2)

(a) V → [+stress]/ — C0VCVC #

(b) V → [+stress]/ — C0 VC #

(c) V → [+stress]/ — C0

The rules in (3) should read as follows: (a) assigns antepenult stress to long words whose

final syllable is not superheavy and whose penult is light; (b) assigns penult stress to

disyllabic words whose last syllable is not superheavy; and (c) assigns stress to

monosyllabic words. When applied, the largest expansion of the rule (2a) takes

precedence over the next largest expansion (2b) which, in turn, takes precedence over

(2c). In other words, (2a) applies before (2b) which, in turn, applies before (2c)

However, in order to account for the distribution of stressed and unstressed

syllables in Palestinian Arabic [fhim-na] ‘we understood’ and [fihim-na] ‘he understood

us’20 in the SPE model, Brame (1971, 1973, 1974) claims that

a. Stress rule applies cyclically

b. Stress applies to subject suffixes on the first cycle, but to object suffixes on the

second cycle.

This means that [fhim-na] ‘we understood’ and [fihim-na] ‘he understood us’ are

different in terms of their constituent structure: Whereas the subject suffix /na/ plus the

base /fihim/ form one internal layer of structure in [fhim-na] ‘we understood’, the object

suffix /na/ adds an additional layer of structure to the base form [[fihim]-na] ‘he

20
Notice that [-na] is both a subject suffix and an object suffix.

42
understood us’. So the underlying structure of [fhim-na] ‘we understood’ and [fihim-na]

‘he understood us’ can be represented as in (3a) and (3b) respectively:

(3a) [fihim-na] “we understood”

(3b) [[fihim]–na] “he understood us

The question that promptly arises here is how stress is assigned in both cases. It

suffices here to say that in Arabic, stress goes to the penult if it is heavy (i.e. cvv, cvc),

but if the penult is not heavy, stress goes to “the antepenult of words with three or more

syllables”, and to the “ first syllable of disyllabic and monosyllabic words” (Brame 1971:

558)21.

Accordingly, stress assignment to [fhimna] and [fihimna] proceeds as follows. In

(3a), the domain of application of stress is the whole constituent (the base form plus the

suffix /na/), and so stress goes to the penult [fihim-na]. In the case of (3b), however,

stress applies twice: once to the most deeply embedded constituent [fihim], and a second

time to the whole constituent [fihim-na]. On the first circle, main stress goes to the penult

of [fihim] by virtue of the fact that it is a disyllabic word whose final syllable is not

superheavy. On the second cycle, main stress goes to the penult of [fihim-na] due to the

fact that the penult is heavy. Notice that stressing the penult of the whole constituent on

the second cycle requires the demotion of stress on the antepenult by one level. This

results in main stress on the penult and secondary stress on the antepenult of [fihim-na].

The foregoing discussion is summarized in (4) below:

(4a) [fihim-na] “we understood” (4b) [[fihim]–na] “he understood us

1st cycle stress fihimna fihim
2nd cycle stress n.a f`ihim-na

21
Stress assignment rules in Arabic are discussed in details in subsequent sections.

43
What this means is that stress applies once in the case of (4a) but twice in the case of

(4b) simply because the former consists of one cyclic domain (i.e., [fihim-na]22), but the

latter is made up of two cyclic domains, namely the verb root [fihim] and the root plus

object suffix [[fihim]–na]23.

However, there remains one more problem: How can we account for the fact that

[fihim-na] “we understood” surfaces as [fhimna], but [[fihim]-na] “he understood us”

surfaces as [fihimna]? In other words, the vowel of the first syllable of [fihim-na] gets

syncopated, but it doesn’t in [[fihim]-na]. According to Brame (1974: 20), this is due to

syncope formulated as follows24:

(5) V
+ high → ∅ / — CV
-stress

The rule in (5) requires a high vowel in an open unstressed syllable to syncopate.

Accordingly, the correct surface forms of [fihim-na] ‘we understood’ and [fihim]-na] ‘he

understood us’ can be obtained if syncope is ordered after stress so as to have stress bleed

syncope. In so doing, the unstressed high vowel of the first syllable in (4a) syncopates,

but the stressed syllable of the first syllable in (24b) does not. The foregoing discussion is

summarized in (6) below:

22
Notice that this is only a theoretical claim put forward to distinguish the subjective form [fhim-na] from
the accusative form [fihim-na].
23
Although the present analysis can account for a large number of cases, it fails to account for cases such
as [allimt-u] “she taught him”, and [baarikt-u] where the epenthetic vowel is stressed. To derive the
correct surface form in Chomsky and Halle’s model, Kenstowicz and Abdul-Karim (1980: 67) suggest a
new rule that assigns stress to the epenthetic vowel when preceded by a heavy syllable (for a formal
representation of the rule and for further details, see Kenstowicz and Abdul-Karim (1980)).
24
The formulation of the rule in Brame (1974: 40) is:
ĭ → ∅ /  CV

44
 (6a) (6b)
[fihim-na] “we understood” [[fihim]–na] “he understood us”
1st cycle stress fihimna fihim
nd
2 cycle stress ……… f`ihim-na
syncope fhim-na …….

Surface form fhim-na fihim-na25

2.1.1.b Nonlinear Approach

The nonlinear approach (Liberman and Prince 197726, McCarthy 1979a,b and

Hayes 1980, 1982, 1991), on the other hand, adopts the view that stress is no longer a

phonological feature analogous to [nasal], for example; rather, it is an autonomous

phonological entity, reflecting a hierarchically organized rhythmic structure or grid

(Liberman and Prince 1977). Under this proposal, stress is viewed as a function of the

prominence relation between syllables. Word stress reflects the grouping of syllables into

larger metrical units (e.g., foot, prosodic word, etc.), and so syllables are mapped onto a

rhythmic alternation of peaks and troughs as in (7) below, where the bold-faced syllable

represents the peak, and non-bold-faced syllable represents the trough (see Kenstowicz

1994: 548):

(7a) CVCVCVCVCV, or

(7b) CVCVCVCVCV

A series of alternating stressed and unstressed syllables is grouped into binary units

called metrical feet formed by either left-to-right or right-to left grouping. The decision to

have a left-to-right or right-to-left grouping is language specific. Arabic, for example,

25
The fact that [fihim]-na] “he understood us” dos not surface with secondary stress is due to a post cyclic
rule of destressing.
26
The idea of the grid originated in Liberman (1975).

45
happens to adopt the left-to-right option. So with a left-to-right parsing (7a) will become

(8a) below, but with a right-to left parsing, it will become (8b):

(8a) (CVCV) (CVCV) (CV)

(8b) (CV)(CVCV)(CVCV)27

This parsing may be graphically represented in terms of peaks and troughs as in (9a) and

(9b), respectively:

(9a) (x .)(x .)(x)

(9b) (x)(. x)(. x)

Another related notion in grid theory is syllable weight. In some versions of this

theory, the weight of the syllable is determined by the number of moras (i.e. unit of

weight) it contains: A light syllable consists of just one unit of weight (i.e., one mora),

but a heavy syllable consists of two moras, and a superheavy syllable may contain three

or more moras. Each segment in the rhyme of the syllable counts as one mora. If we use

an asterisk to represent the mora, then light, heavy and superheavy syllables can be

sketched as (10a), (10b), and (10c), respectively (where the underlined are the rhymes of

the syllable):

*
(10a) c v

** **
(10b) c v c or cvv

*** ***
(10c) c v c c or cvvc

27
Notice that a left-over CV is parsed into a degenerate foot.

46
 In order to account for the observation that segments at the edge (usually right) of

the prosodic word do not add to the weight of the syllable, and ultimately to the overall

stress pattern, the notion of extrametricality is invoked (Hayes 1979; Prince 1983; Harris

1983). In a sequence like (11) below, for example, when the final mora is considered

extrametrical (cf. enclosed between angled brackets), the whole word then consists of just

light and heavy syllables:

** * * * **<*>
(11) cv c cv cv cv cvv c

When applying this model to an Arabic word like /adwiyatuhu/ ‘his medicine’,

for example, the stress pattern of the word can be obtained through the following steps

(12a-f):

(12)

(a) Syllable-weighting: the mora is the stress-bearing unit

** * * * *
 ad wi ya tu hu

(b) Extrametricality: final syllables which are not superheavy are extrametrical

** * * * < *>
ad wi ya tu hu

(c) Binary Foot-construction: form a foot by grouping every two moras together
from left to right (or right to left—this is language specific)

(**) (* *) (*) < *>28

ad wi ya tu hu

28
Notice here that the last foot consists of only one mora. This is so because it is the only mora left after
parsing the entire word from left to right

47
 (d) Head-specification: make left-headed (or right-headed-- this is also language
specific) feet where the first mora of each foot is the peak

x . x . x
(**) (* *) (*) < *>
ad wi ya tu hu

(e) Unbounded foot-construction: all the peaks and troughs make up an

unbounded constituent (i.e., foot)

(x . x . x)
(**) (* *) (*) < *>
ad wi ya tu hu

(f) Main-stress assignment: the syllable represented as the rightmost peak of the
unbounded constituent carries primary stress (i.e., the unbounded foot is right-
headed in this case, but it can also be the other way around— again language
specific issue)

x
(x . x . x )
(**) (* *) (*) < *>
ad wi ya tu hu

So the model predicts correctly that stress falls on /tu/.

Another version of the nonlinear grid theory is found in Hayes (1995). He

assumes that the Arabic basic foot template is what he calls a moraic Trochee. The foot

consists of two moras, the first of which is stronger; and any single heavy syllable forms

a metrical unit or foot by itself. This is illustrated in 13 below (where υ represents a light

syllable and — a heavy syllable):

(13) (x .) (x)
υ υ or —

48
Applying this model to our previous example, the derivation goes as in (14) below:

(14) ( x ) word level

(x) (x .) (x .) foot level
— υ υ υ υ mora level
ad wi ya tu hu

Notice here that the first syllable (i.e., /ad/) is a heavy syllable (represented with —),

and thus forms a metrical unit by itself. All the other light syllables (represented with υ)

pair together to form metrical units (i.e., feet) that consist of two moras each. The model

once more correctly predicts where in the word main stress occurs.

A point worthy of mention here is that within this framework only the final

consonant, not the final syllable, is extrametrical29. Although the tools are modified

slightly, the correct stress pattern is derived.

Another approach that works along the same lines, but with somewhat different

tools, is the metrical theory of McCarthy (1979). Within this theory, stress rules are rules

for the construction and labeling of metrical trees. The syllable σ consists of two parts:

the onset (which is weak) and the rhyme (which is strong):

(15) σ

w s

The rhyme in its turn consists of the syllable nucleus (which is strong) plus any following

vowel or consonant (which is weak) as in (16) below:

29
If the final syllable is considered extrametrical, words ending with heavy and superheavy syllables will
be stressed on the wrong syllable. For example, in a word like /mudarrisaat/, stress will not go to the final
superheavy syllable (which it should) if that syllable is considered extrametrical.

49
(16)

(a) Light Syllables σ

w s

(b) Heavy Syllables σ

w s

s w

The construction of metrical tree structure proceeds as follows:

(17)

(a) Rhyme-construction: project all the rhymes of the word (i.e. “the rhyme
consists of the syllable nucleus and any following vowel of consonant”)

a d wi ya tu hu

(b) Foot-construction: form binary feet over pairs of light syllables30, from left to
right

a d wi ya tu hu31

30
A heavy syllable makes up a foot by itself as it projects two rhymal units.
31
Notice here that the geometry of a heavy syllable is identical to the geometry of a foot.

50
 (c) Tree-branching: gather the structure into a right branching word-level tree

a d wi ya tu hu

(d) a right node is strong iff (if and only if) it branches

w s

s w w s

s w s w
 a d wi ya tu hu

The model is therefore capable of predicting which syllable carries stress.

Al-Mozainy et al. (1985) provides one piece of evidence favoring this metrical

account over all other approaches. The argument is based on the phenomenon of stress

shift. When the vowel that carries stress gets deleted for one reason or another, stress

shift, according to Al-Moziany et al. (1985), can be accounted for under the tree-structure

approach. Consider an example like (18) below where the stressed vowel of the

antepenult is deleted, and stress shifts to the penult:

(18) inkasarat > inksarat

Under the tree-structure interpretation, stress shifts to the right position in the tree:

51
 (19)

(a) before vowel deletion

w
s w

s w

i n ka sa ra<t>

(b) after vowel deletion

w
s w

i n ksa ra<t>

The argument here is that ka and sa are sister nodes in (19a) as they pair together to make

up one foot. So the deletion of the nucleus of one of them causes stress to shift to the

nucleus of the other. Therefore, the behavior of stress under syncope becomes predictable

in that the remaining mora (or syllable) of the foot which loses one of its two syllables

becomes the most likely candidate to receive primary word stress (Cf. 19b).

Although the above analysis gives the correct result, I believe that a grid-based

analysis could also yield the same result. Consider example (18) repeated here as (20a-b):

52
 (20)

(a) before deletion

x
(x . x .)
(* *) (* *)
 i n ka sa <rat>

(b) after deletion

x
(x. x)
(* *) (* )
 i n ksa <rat>

The second foot has undergone a process of re-syllabification where the remaining

consonant of the syncopated syllable becomes part of the onset of the next syllable, and

so does not influence the weight of the syllable as it does not count as a mora. All in all,

both approaches, I believe, can account for stress shift under the influence of vowel

deletion.

2.1.1.c Optimality-Theoretic Analysis

Finally, we move to see how word-stress is dealt with from an optimality-

theoretic approach. The proposal adopted here is that the optimality-theoretic approach is

more comprehensive than all its predecessors as it incorporates most of the pieces of

information that the previous models consider for stress assignment purposes. To

illustrate, the above discussion has shown that the different models rely on different types

of information to account for word-stress patterns cross-linguistically: some of them

incorporate information regarding “the segmental make-up” and “the internal structure”

53
of the word (Chomsky and Halle 1968); others employ the notion of syllable weight

(Prince and Libermann 1977; McCarthy 1979b), extrametricality (Hayes 1979, 1980),

etc. In optimality theory, however, all these pieces of information are seen as relevant to

stress placement. Information regarding the morphological and syntactic structure of the

constituents, their rhythmic organization, weight, etc. are all regarded as conflicting

forces (i.e., can be translated into violable constraints) that determine the distribution of

stressed and unstressed syllables in polysyllabic words, not just within one language but

cross-linguistically as well. This is based on the premise of optimality theory that stress

placement, like all other phonological features, is subject to constraint satisfaction. For

example, it has long been assumed that ‘stress is a rhythmic phenomenon” (Liberman and

Prince 1977). This means that there is a strong-weak relation between all the syllables of

the prosodic word. Some languages show that every odd-numbered syllable in a

polysyllabic word is stressed; other languages stress every even-numbered syllable. Still

others place primary stress on a fixed syllable (be it first, antepenult, penult, ultimate,

etc.) and secondary stress on every other syllable on both sides of the primary stress. In

order to see how this works, just consider the stress patterns of the two English examples

below where, under the influence of rhythm, the main stress on the word ‘fourteen’ shifts

one step leftward in the second example, so as to avoid a stress clash (Prince 1983: 21)

(21a) Fourteen
(21b) Fourteen women

It is also noted that there are other forces that determine the distribution of

stressed and unstressed syllables in polysyllabic words cross-linguistically. One such

strong force is the relative intrinsic prominence (i.e., or weight) of the syllable. Syllables

54
are said to be light, heavy, or superheavy depending upon the number of moras (units of

weight) that occupy nuclear and post-nuclear positions: the more moras, the heavier the

syllable. A light syllable has only one mora; a heavy syllable two moras; and a

superheavy syllable three or more moras. Many investigations have targeted the internal

structure of the syllable to determine its weight relative to the other syllables in the same

word. The findings have shown that in order to account for the distribution of stressed

and unstressed syllables, there is a need to postulate other prosodic units such as rime,

metrical foot, prosodic word, etc. Extensive research has provided ample evidence that

the syllable, for example, has tripartite organization: an onset, nucleus and coda, of which

only the nucleus is universally obligatory. For stress assignment purposes, only elements

that occupy nuclear and post-nuclear positions (namely nucleus and coda) count32. The

crucial point here is that stress can be assigned in a polysyllabic word based on the

relative weight of the syllables. Superheavy syllables are more likely to receive main

stress than heavy syllables which, in turn, are more likely to receive stress than light

syllables33. Consider the following examples form Arabic:

(22) šaabaat ‘young ladies’

katabna ‘we wrote’

In these two examples the stressed syllable (in bold) gets stressed simply because it is

heavier than all the other syllables in the same word.

It is also noted that the position of the syllable (be it initial, final, etc) is relevant

to stress placement. In this regard, stress tends not to occur word-finally cross-

linguistically. Consider the following examples from Arabic:

32
The only reported exception to this generalization comes from Aranda- a language spoken in central
Australia (For details the reader is referred to Davis (1988).
33
For argument against this traditional dichotomy, see de Lacy (1997).

55
 (23) muallim ‘teacher’
šaabaatun ‘young ladies’
mustašfaa ‘hospital’

In these examples, although the ultimate and penultimate syllable are of equal weight

(both heavy syllables), stress goes to the penultimate. What is more revealing is that

stress may go to a light non-final syllable even if the final syllable is heavy as in the

following examples from Arabic:

(24) maktabah ‘library’

raaa ‘saw’
šajarah ‘tree’

The same may also apply to English where for purposes of stress assignment, we

sometimes need to (1) count the number of syllables backward from the end of the word,

(2) consider the type of syllables, and (3) consider the morphological make-up of the

word, etc. All of this points towards one conclusion: the rhythmic principle, syllable

weight, the position of the syllable in the word, etc., can all simultaneously play a role in

stress assignment. Within optimality theory, these are seen as conflicting forces that

struggle for optimality.

Observations like these have contributed to the development of a model of

analysis that is based on constraint interaction. To illustrate, it is recognized that there is a

need to observe syllable weight, as it may sometimes be very decisive to stress

placement. In optimality terms, a constraint like the weight-to-stress principle (henceforth

WSP) acts to this effect. What this principle does is make sure that a heavy syllable takes

priority over light syllables to carry main stress. In the meantime, it is realized that there

is also a need to respect the rhythmic principle (i.e. the alternations of weak and stress

syllables is pretty regular) even if it may turn out that respecting this principle may only

56
be at the expense of some other principle, e.g. the WSP. To this effect, the Parse-syllable

constraint is put forward. So when this constraint is ranked high enough, it will, in

addition to predicting the position of main stress, show that secondary stresses occur

where they should.

Still, there has always been the need to take morphological information into

consideration when accounting for stress placement. And to this effect, many constraints

(usually alignment constraints) are advanced. Examples include Main right/left; All-feet

right/left, word-foot right/left, Iambic/trochaic, and non-final/non-initial

constraints34. Having all this in mind, the inevitable conclusion has been this: stress

placement is a realization of the interplay between all these constraints simultaneously.

Notice that it is difficult, if not impossible, to satisfy all these needs (constraints)

perfectly, but they can be satisfied optimally. In other words, it is assumed that these

constraints are all conflicting forces that struggle for satisfaction, but the satisfaction of

one (or more) of these forces may only be possible at the expense of the other(s), and

thus, here rises the notion of constraint ranking. This notion lies at the heart of Optimality

theory. Each grammar is a means to solve the conflict between these constraints by

ranking the constraints in a unique way.

Let us take some examples to illustrate this state of affairs35. In one of the

Jordanian varieties, namely, Bedouin, the second syllable of disyllabic words, rather than

the first as in other varieties of Jordanian Arabic, receives main stress as shown in (3)

below:

34
Definition and examples for each will be provided later.
35
The examples are taken from Hung (1995).

57
 (25) jimal
bina
aduw36

In addition, in polysyllabic words whose final syllable is not superheavy and whose

penult is not heavy, stress goes to either the penult or the antepenult whichever is

separated from a preceding heavy syllable or word-boundary by an odd number of

syllables37 as in (26):

(26) gahawah
ankitalaw
zalamatak

Since in this variety, as well as in all the other varieties of Arabic, the weight of the

syllable is a determining factor in stress assignment, then there should be a constraint that

acts to this effect. Following the insight of Myers (1987), the constraint is formulated

along the following lines:

Weight-to-Stress Principle (WSP)

Heavy syllables must be stressed38

To determine where stress falls in the word, the syllables of the word are grouped into a

higher-level metrical constituent, namely the foot. Each foot consists of two moras (unit

of weight). A pair of light syllables (i.e. /cvcv/) make up one foot, and a heavy syllable

(i.e. /cvv/ or /cvc/) make up one metrical foot as below:

(27) (* *)
ælә

36
Notice here that the first syllable should be light.
37
This also applies to the other Jordanian dialects except that the condition specifies that the stressed
syllable is whichever is separated by an even number of syllables.
38
The wording is due to Tesar (1996: 3).

58
To make sure that all the syllables of the word are grouped into metrical feet, a constraint

like Parse-syllable is suggested:

Parse-Syllable
A syllable must be footed

For each foot, one leg is strong (called head) and the other is weak (called non-head or

tail). If the head of the foot is initial, then the foot is called trochee; if it is the other way

around, then it is iamb. Yet when a word consists of more than one foot, then the head of

one of those feet has to dominate; that is, become more prominent than all the other heads

as below:

(28)
*
(* .) (* .)
æ lә bæ mә “Alabama”

Bedouin Jordanian Arabic chooses to have the head on the right. Other languages,

however, may choose left-headed foot construction. There is then a need to postulate

some alignment constraints like the one below that takes care of this issue:

Iambic
Align the head-syllable with its foot, on the right edge

Still, in Bedouin Jordanian Arabic, as well as in all the other dialects of Arabic, stress

tends not to occur word finally unless the ultimate syllable is superheavy (i.e., tri-

moraic). In pre-Optimality literature this has been referred to as the theory of

extrametricality (Hayes 1979; 1982a). The effect of extrametricality is translated into a

constraint in Optimality theory along the following lines:

Non-Final
The final syllable should not be footed

59
Finally, we need a Faithfulness constraint that militates against deletion or insertion of

segments. In other words, we need a constraint that acts to the effect of having the output

identical with the input. For ease of exposition, let’s call this constraint IO-OI.

IO-OI
Output segments must have input correspondents (no epenthesis), and input
segments must have output correspondents (no deletion)

In order to account for stress placement in (20) and (21) above, the conflict between the

following five constraints must be resolved by making use of the notion of constraint

ranking. For ease of exposition, the constraints are reproduced as follows:

Weight-to-Stress Principle (WSP)

Heavy syllables must be stressed

Parse-Syllable
A syllable must be footed

Iambic
Align the head-syllable with its foot on the right edge

Non-Final
The final syllable should not be footed

IO-OI
Output segments must have input correspondents (no epenthesis), and input
segments must have output correspondents (no deletion)

The question arises, then, as to how the conflict between these constraints can be resolved

to yield the correct stress pattern in a word like ‘bi'na’ (25 above), for example.

First, the function Gen provides both actual outputs and potential outputs which

are all evaluated by Eval against the set of universal constraints as shown in Tableau (1)

below39. The form /bina/ in the left-most upper corner is the input; all the candidate forms

39
The dotted lines in the Tableau mean that the example at hand does not provide evidence as to how the
constraints are ranked relative to each other.

60
(actual and potential) which are provided by GEN are listed vertically; and the constraints

are arranged horizontally (dashed lines mean that the constraints are not ranked relative

to each other).

Tableau (1)
/bina/ IO-OI Parse-syllable Iambic Nonfinal
" (bi'na) *
('bina) *! *
('bi)na *! *
('bi:)na *! * *
etc.

Let’s take a moment to explain how the ranking of constraints in Tableau (1) above can

yield the desired form, namely ‘bi'na’.

Notice that all candidates (actual or potential) violate one or more constraints. The

actual form (bi'na) violates Nonfinal. So, (bi'na) can surface as the optimal candidate

provided that the constraint which militates against it is ranked very low in the tableau, so

that the violation of this constraint is not fatal. In other words, all other constraints which

are violated by the competitors of (bi'na) must dominate Nonfinal. ('bina), for example,

violates IAMBIC, so IAMBIC should outrank NONFINAL. Ranking IAMBIC above

NONFINAL means that any output form which violates Iambic loses the competition

altogether since it violates a higher-ranking constraint. The violation of Iambic is fatal

relative to the violation of Nonfinal. Tableau (2) below shows the ranking of Iambic and

Nonfinal (solid line means that a ranking argument is established):

61
Tableau (2)
/bina/ Iambic Nonfinal
" (bi'na) *
('bina) *! *

Notice that although ('bina) violates Nonfinal, this is irrelevant (shaded area). Hence, the

exclusion of ('bina) is due to its violation of a higher-ranking constraint, namely

IAMBIC.

Another competitor to (bi'na) is ('bi)na. This competitor violates another

constraint, namely Parse-syllable, which requires that every syllable of the prosodic word

be footed. Again, in OT terms, the only way to get this candidate out of the way is to

have Parse-syllable ranked higher in the tableau than Nonfinal. What this basically means

is that the violation of Nonfinal is less serious than the violation of Parse-syllable.

Accordingly, when two candidate forms compete for optimality, then the one that violates

Parse-syllable loses the competition to the one that violates the lower-ranked constraint

Nonfinal. Tableau (3) below displays this ranking argument:

Tableau (3)
/bina/ Parse-syllable Nonfinal
" (bi'na) *
('bi)na *!

Another potential candidate that may also want to compete for optimality is

('bi:)na. Notice that this potential candidate differs from the actual output form (bi'na) in

at least two respects. First, it violates Parse-syllable as one of its syllables is left unfooted.

Second, it violates another constraint, namely IO-OI—the constraint which militates

against the addition or deletion of some segments: the output must be identical with the

62
input and vice versa. This candidate ('bi:)na loses the competition for optimality on

similar grounds: it violates higher-ranked constraints, namely Parse-syllable and IO-OI.

Tableau 4
/bina/ IO-OI Parse-syllable Nonfinal
" (bi'na) *
('bi:)na *! *

Notice that IO-OI and Parse-syllable are ranked relative to Nonfinal (separated by a solid

line), but they are not ranked relative to each other (separated by dotted line).

To make the story short, it should be borne in mind that in order to get (bi'na) as

the optimal form, then any constraint which is violated by this form must be ranked lower

in the tableau than any other constraint(s) that other potential candidates violate. Since

(bi'na) violates only Nonfinal, then this constraint should be ranked lower than all the

other constraints in the Tableau; and thus, violation of that constraint becomes

permissible as it lets the optimal form avoid the violation of higher-ranked constraints

such as IO-OI, Parse-syllable and Iambic. Incorporating this ranking argument into

Tableau (1) above, Nonfinal should be separated from the other constraints by a solid line

as in Tableau (5) below:

Tableau (5)
/bina/ IO-OI Parse-syllable Iambic Nonfinal
" (bi'na) *
('bina) *! *
('bi)na *! *
('bi:)na *! * *
etc.

Tableau (5) basically shows that IO-OI, Parse-syllable and Iambic all dominate Nonfinal,

but they are not ranked relative to each other (separated by dotted lines).

63
 One final remark here is that not all the constraints that we suggested earlier may

get involved in the competition for optimality for every particular output form. For

example, Weight-to-Stress (i.e, which requires that heavy syllables be stressed) does not

take part in the case of (bi'na). The reason is simple: None of the syllables of (bi'na) is

heavy, so the constraint which acts on heavy syllables becomes irrelevant. For it to take

part, one of the syllables of the input form must be heavy as in (ga'hawah), for example,

where the final syllable is a heavy one. Consider the following tableau:

Tableau (6)
/gahawah/ Nonfinal WSP
" (ga'ha)wah *
(ga'ha)( 'wah) *!
etc.

Notice here that the violation of Nonfinal becomes a serious matter. In order to yield the

desired result, we need to have WSP ranked lower than Nonfinal, so that the candidate

form that violates WSP (e.g. ‘ga'hawah’) may still have the chance to surface as the

optimal form. Incorporating this ranking argument into Tableau (6) above yields Tableau

(7) below (where Nonfinal and WSP are separated by solid line):

Tableau(7)
/gahawah/ Nonfinal WSP
" (ga'ha)wah *
(ga'ha)('wah) *!
etc.

Let’s consider more complicated cases to see if this constraint-based model is

capable of correctly predicting the stress pattern. Consider the example of syncopated

accusatives in Arabic which is discussed in (2.1.1.1) in light of the transformational cycle

(Chomsky and Halle 1968). The example is reproduced here for ease of exposition:

64
 (1a) (1b)
[fihim-na] “we understood” [[fihim]–na] “he understood us
1st cycle stress fihimna fihim
2nd cycle stress ……… f`ihim-na
i-syncope fhim-na …….

Notice that the stress pattern in (1a) is different from (1b) due to the claim that the stress

rule applies once in (1a) but twice in (1b) (see Brame 1974). This is entailed by the claim

that the former consists of one cyclic domain (i.e. [fihim-na]), but the latter consists of

two, namely the verb root [fihim] and the root plus object suffix [[fihim]–na]. In other

words, there is an additional layer of morphological structure to which the rule reapplies

in the case of the accusative form (1b).

Before we show how stress patterns can be accounted for in these examples under

a constraint-based approach, let’s recall that syncope does not apply in the case of (1b)

because the high short vowel [i] of the first syllable of “fihim” is stressed. In other words,

the ordering of the stress rule before syncope results in blocking the application of the

latter, so stress application here bleeds syncope. We move now to see how these

observations are accounted for in a constraint-based framework.

Kager (1999: 281- 287) assumes that one major difference between subject forms

(e.g. [fhimna]) and accusative forms (e.g. [fihimna]) is that the former do not have

corresponding base forms, but the latter do. He assumes that the base of every accusative

form is the corresponding subject form. Consider the following examples:

65
 Accusative form Base

fihimna “he understood us” fihim “he understood”

aTaani “he gave me” aTaa “he gave”
abaraka “he informed you” ?abar “he informed”
jamaahum “he gathered them” jama “he gathered”
darasaha “he studied it” daras “he studied”

According to Kager, /i/ does not syncopate in the accusative, though it occurs in a

light open syllable, because it has a correspondent that is stressed in the base. Let us also

point out that in the case of (1b) the base form [fihim] and the derived form [fihimna] are

both “free-standing output forms of the language” (Kager 1999: 280), i.e., both are

independent words in the language; [fihim] means “he understood”, and [fihimna] means

“he understood us”. But the basic insight is that the base form [fihim] and the derived

form [fihimna] are “compositionally related” in that the former contains a proper subset

of the morpho-semantic features of the latter (Kager 1999: 281). One such feature is that

the /i/ of the first syllable does not syncopate in either form despite the fact that it occurs

in a light syllable. In OT terms, this requirement can be stated in the form of a

correspondence constraint which matches an output form with another output form (in

this case accusative form with the base). Kager (1999: 283) formulates a constraint that

requires the stressed vowel in the base [fihim] to have a correspondent in the derived

accusative form [fihimna]:

66
 HeadMax-BA40
Every segment in the base’s prosodic head has a correspondent in the affixed
form (italics in original)

HeadMax-BA basically requires the high front vowel of [fihimna] not to syncopate

because its correspondent in the base [fihim] is the prosodic head of the word, i.e. it is

stressed.

HeadMax-BA will inevitably be in conflict with another constraint which triggers

i-deletion in open syllables:

No-[i]
/i/ is not allowed in light syllables

How do these constraints interact?41 In the case of accusatives where /i/ is not

deleted, even though it is in an open syllable, No[i] should be dominated by the higher-

ranking constraint HeadMax-BA. Consider the following tableau where the two

candidate forms [fi'himna] and ['fhimna] compete for optimality:

Tableau (8)
Input: /fihim-na/ HeadMax-BA *i
Base: ['fihim]
" [fi'himna] *
['fhimna] *!
etc.

Notice that the stressed vowel in the base (upper left-hand corner) does not have a

correspondent in ['fhimna]. This violation of HeadMax-BA turns out to be fatal.

40
It should be stated that the Base-Output constraint is different from another constraint that requires that
every segment in the input have a correspondent in the output. This latter requirement can be translated into
a maximality constraint like Max (I/O) (McCarthy & Prince 1995)
Max (Input/Output)
Every segment in the input has a correspondent in the Output
The sole difference between Headmax Base-Output and Max Input/output is that whereas the former only
affects the prosodic head of the input (i.e., stressed segment), the latter affects every segment in the string,
whether stressed or not.
41
The basic insights are found in Kager (1997, 1999: 281-287)

67
[fi'himna], on the other hand satisfies this constraint, and thus surfaces as the optimal

form even though it violates the lower-ranking constraint No[i]. The ranking of the

constraints introduced so far is as follows:

Constraint Ranking (I):

HeadMax-BA > No[i]

Having this in mind, let’s turn to the derivation of [fhimna]? The basic insight

here is that [fhimna] does not have a corresponding base form. Consequently, HeadMAx-

BA becomes even irrelevant. [fhimna] then surfaces as optimal simply because its

competitor [fihimna] violates No[i]. Consider Tableau (9) below:

Tableau (9)
Input: /fihim-na/ HeadMax-BA *i
Base: None
" [fi'himna] *!
['fhimna]
etc.

One may also wonder why main stress in [fihimna] surfaces on the second

syllable, but not the first syllable whose correspondent in the base is stressed [fihim]. In

other words, why does the accusative form [fihimna] not surface as [fihimna] where the

first syllable receives main stress as in the base form? The straightforward answer to this

question is that this potential form [fihimna] violates another constraint, namely Weight-

to–Stress, whereby heavy syllables are stressed. Consider Tableau (4) below:

Tableau (10)
/fihim-na/ WSP HeadMax-BA No[i]
" [fi'himna] *
['fhimna] *!
['fihimna] *! *

68
Clearly, since [fihimna] and [fihimna] fair equally on HeadMax-BA and No[i] (that is,

both satisfy HeadMax-BA, and both violate No[i]), the conflict between them is resolved

by the satisfaction (or lack thereof) of WSP. The second syllable of the actual output

[fihimna] receives stress and this satisfies WSP, but that same heavy syllable is not

stressed in its competitor [fihimna], and so the latter fails altogether to survive as the

optimal form. Incorporating WSP into (I) above, we get:

Constraint Ranking (II):

WSP, HeadMax-BA > No[i]

But what about the base form itself ['fihim]? We have said earlier that the base

form is a word of the language, and thus, optimal. But notice that it fails to comply with

the requirement of WSP in that the second syllable, which is heavy, is not stressed. The

answer once more lies in the assumption that this output form, though it violates WSP,

still satisfies another higher-ranking constraint, namely Nonfinal (which requires that a

final syllable be not footed, and so does not receive stress):

Tableau (11)
/fihim/ Nonfinal WSP HeadMax-BA No[i]
" ['fihim] *! *
[fi'him] *! *

Tableau (11) succinctly shows that [fi'him], as a potential candidate form, does not

surface as optimal simply because it violates Nonfinal.

Constraint Ranking (III)

Nonfinal > WSP, HeadMax-BA > No[i]

Having shown how the suffixed form [fihimna] “ he understood us” and the base

form [fihim]) are derived, there remains to show how [fhimna] “we understood” surfaces

as an optimal form, too. The argument here is that although [fhimna] has the same input

69
as [fihimna], it does not have a corresponding base form, and so it no longer violates

HeadMax-BA. In fact, HeadMax-BA is irrelevant in the absence of a ‘base’. Consider

Tableau (12) below:

Tableau (12)
Input: /fihim-na/ WSP HeadMax-BA No[i]
Base: None
" ['fhimna]42
[fi'himna] *!
['fihimna] *! *

The examples above then show how the ranking of constraints can determine

which candidate form surfaces as the optimal form. To reiterate, none of the candidate

forms satisfies all the constraints perfectly; rather, each one of them violates one or more

constraints. The crucial point is that the phonology consists of a set of ranked constraints

where the violation of constraint X, for example, may be more serious than the violation

of constraint Y. Consequently, the form that satisfies the higher-ranking constraint (X in

this case) will ultimately win the competition by being the optimal form.

42
['fhimna] violates Input-Output which requires that every segment in the input have a correspondent in
the output (see footnote 41). Input-Output should then be ranked lower than */i/.

70
Chapter Three: Constraint Interaction in Arabic and English

3.1 Overview

The overall purpose of this study is to analyze the acquisition of English word-stress by

Arabic speakers in light of advancements in Optimality Theory. It has been shown that

Arab second language learners of English have difficulty in acquiring the pattern of

English word stress. According to OT, the reason for this difficulty is that although these

speakers, like native speakers, have full command of the universal and violable

constraints that are operative in determining where stress falls in the word, they fail to

capture or induce the exact ordering of these constraints. The basic premise of OT is that

each grammar is a unique way of ordering the set of universal and violable constraints

that determine the actual output form of a certain linguistic feature, say word-stress in

this case. In other words, whereas Arabic word-stress and English word-stress are both

subject to the same set of universal and violable constraints, they differ in one respect: the

ordering of these constraints. The sole task of the learner then is to capture the correct

ordering of the constraints that interact to determine which syllable of each word in the

language carries stress. In reality, however, there are three possibilities as far as the

ranking of constraints is concerned. In the case of Arabic-speaking learners of English,

these three possibilities are:

 I. learners successfully acquire the target language (i.e. English) ranking. This is

tantamount to saying that Arabic-speaking learners of English produce

English word-stress with native proficiency--a possibility that is often ruled

out based upon the existing literature that examines the performance of

Arabic-speaking learners of English in terms of word-stress;

II. learners transfer the native language ranking (i.e. Arabic) to English. Implicit

in this is the assumption that erroneous tokens produced by second language

learners can be attributed to the influence of L1—a possibility that is

supported by a large body of research; or

III. learners do neither. They have their own ranking which is neither that of their

native language nor that of the target language—a possibility that we need

also to pursue. As their interlanguage stress system is a naturally occurring

stress system, then it should be described on its own by special ranking of the

relevant constraints.

In OT terms, for Arabic-speaking learners of English to acquire the correct stress patterns

of English, all they need to do is learn the English-specific ranking of the set of universal

and violable constraints. Before showing how this goal can be reached, we need to find

out the followings:

I. the set of constraints that are relevant for predicting the place of stress, not

just in English and Arabic, but in all languages. Hence, these constraints are

literally present in all languages, though their ranking is language-specific

(see section 3.2 below);

72
 II. the ranking of the set of constraints in the learners’ native language, namely

Arabic (see 3.3.4 below);

III. the ranking of the set of constraints in the target language, namely English

(see 3.3.5 below);

IV. the exact problems that learners face in learning the target language;

Once these issues are accounted for, we can embark on addressing how nonnative

speakers can modify their ranking of the set of universal constraints. This issue of

constraint re-ranking will be addressed in the next chapter.

3.2 The constraints

In the OT literature, it is assumed that the set of constraints that determine the behavior of

linguistic features (e.g. syllable structure, stress, etc.) is fixed in all languages. As for

stress, Tesar (1996) lists 12 universal constraints that interact to yield the correct stress

pattern in any language. These are:

WSP heavy syllables must be stressed

PARSE-σ a syllable must be footed
MAIN-RIGHT align the head-foot with the word, on the right edge
MAIN-LEFT align the head-foot with the word, on the left edge
ALL-FEET-RIGHT align each foot with the word, on the right edge
ALL-FEET-LEFT align each foot with the word, on the left edge
WORD-FOOT-RIGHT align the right edge of the word with a foot
WORD-FOOT-LEFT align the left edge of the word with a foot
IAMBIC align the head-syllable with its foot on the right
edge
TROCHAIC align the head-syllable with its foot, on the left edge
NONFINAL the final syllable should not be footed
NONINITIAL the initial syllable should not be footed

Other treatments may add to this list FOOT-BINARITY, and NOCLASH, NOLAPSE
and RHYTHM:

73
 FOOT BINARITY feet are binary
NOCLASH adjacent Stressed syllables are prohibited
NOLAPSE one of two adjacent syllables must be stressed
RHYTHM H must be followed by L at the end of the prosodic
word

In this section, I will try to shed light on the function of each constraint. That will

include its structural description, when it is satisfied, when it is violated, the

consequences of its satisfaction or violation, and most importantly its interaction with

other constraints.

The Weight-to-Stress Principle (henceforth WSP43) is formulated in Prince and

Smolensky (1993) as follows:

WSP
Heavy syllables are stressed

WSP is basically an OT mechanism to capture the tendency that some syllables, due to

their internal constituent structure, are more likely to receive stress than others. What this

constraint does is relate the weight of the syllable, compared with the other syllables in

the word (e.g. light, heavy, superheavy), to its prominence. WSP can then be violated iff

(if and only if) it turns out that a heavy syllable receives no stress. It is possible then to

have multiple violations of this constraint in the same word if that word contains more

than one unstressed heavy syllable. Consider Tableau (1) below:

Tableau (1)
Input: /LHLHL/ WSP
a- LHLHL **

43
For pre-optimality literature on this principle, see Prince (1983, 1990).

74
 b- " LHLHL

Output (b) is optimal (indicated by the pointing hand ") in tableau (1) above because it

satisfies the requirement of WSP44.

PARSE-σ 45(Prince and Smolensky 1993) requires that all syllables be parsed into

feet:

Parse-σ
Syllables must be parsed into feet

Any syllable that is not part of a foot constitutes a violation of PARSE-SYLLABLE.

Consider Tableau (2) below:

Tableau (2)
Input: /LLLLL/ Parse-Syllable
a- (LL) LLL ***
b- " (LL) (LL) L *

As PARSE-SYLLABLE mitigates against unfooted σ, (a) incurs three violations, but (b)

incurs only one violation. All things being equal, (b) wins the competition.

It is worth mentioning here that PARSE-SYLLABLE enforces the requirement

that the syllable, being a universal prosodic category, is part of a still higher-level

category, namely the foot46. When stressed, the syllable becomes the head of the foot.

This relationship can be graphically represented as in (1) below:

(1)
F

44
Notice that the mirror image of WSP is not true; that is, if it turns out that a light syllable receives stress,
this should not constitute a violation to WSP.
45
For pre-optimality literature on this constraint, see Hayes (1980); Halle and Vergnaud (1987)
46
Check the following for pre-optimality Literature on this constraint:
Strict Layer Hypothesis: A category of level i in the hierarchy immediately dominates a (sequence of)
categories of the level i-1 (Selkirk 1984a: 26).
Prosodic licensing: All phonological units must be prosodically licensed, i.e. belong to higher prosodic
structures (Ito 1986: 2).

75
 σ

Another constraint that also helps to further define the foot is FOOT-BINARITY

(Prince and Smolensky 1993)47:

Foot Binarity
Feet are binary under syllabic or moraic analysis

This constraint requires that each foot consist of two syllables (σσ), or two moras (µµ).

Under a syllabic analysis, we get the following foot forms (LL), (LH), (HL) and (HH).

Any foot that consists of just one syllable, (L) or (H), or consists of more than two

syllables (LLH), (HLH), etc., constitutes a violation of FOOT-BINARITY. Consider (2)

below (an asterisk means a violation):

(2)
a. *(L)
b. *(H)
c. *(LHL)
d. (LL)
e. (LH)
g. (HL)
h. (HH)

Under moraic analysis, however, any foot that consists of less than one mora (µ), or more

than two moras (µµµ) constitutes a violation of FOOT-BINARITY. Two light syllables

(LL) and one heavy syllable (H) can make up one foot, and thus satisfy FOOT-

BINARITY. Consider (3) below:

(3)
a. *(L)
b. *(LLL)

47
For pre-optimality literature on this constraint, see Prince (1980), Kager, R. (1989)

76
 c. *(LH)
d. (H)
e. (LL)

It is stated in (1) above that the foot as a universal prosodic category must have a lower

level category, namely the syllable (or the mora under a moraic analysis), as its head.

However, due to FOOT-BINARITY, which requires that each foot consist of two legs,

i.e. two syllables (σσ) or two moras (µµ), we need to single out which syllable (or mora)

is the head of the foot. McCarthy and Price (1993a: 32) suggest a number of alignment

constraints that conspire to yield the correct distribution of head syllables, feet, etc. in any

language. All alignment constraints are of the form in (4) below:

(4) ALIGN (GCat, GEdge, PCat, PEdge)

What this means is that the grammatical edge (right or left) of any grammatical category

(e.g. root, stem, suffix, morphological word, etc.) must coincide with the prosodic edge of

some prosodic category (e.g. syllable, foot, prosodic word, etc.)

As for the head syllable, there are two alternative constraints designated for this

purpose, namely IAMBIC and TROCHAIC. Whereas the former requires that the head

syllable be aligned with the right edge of the foot, the latter requires that the head syllable

be aligned with the left edge of the foot:

Iambic
Align the head-syllable with its foot, on the right edge.

Trochaic
Align the head-syllable with its foot, on the left edge.

Under a syllabic analysis, (LL), (HL), (LH), (HH) satisfy IAMBIC, but they all violate

TROCHAIC, which requires that the left syllable, rather than the right one, be the head of

the foot. Consider Tableau (3a) below:

77
Tableau (3a)
Input: /LLLHH/ IAMBIC TROCHAIC
a- (LL) (LH) (H) **
b- (LL )(LH) (H) **

Under a moraic analysis, however, (H) and (LL) satisfy IAMBIC; and (H) and (LL)

satisfy TROCHAIC. Consider tableau (3b) below:

Tableau (3b)
Input: /σσσσσ/ IAMBIC TROCHAIC
a- (LL) (LL) (H) **
c- (LL) (LL) (H) **

Notice that when all syllables are parsed into binary feet (the requirements of both

PARSE-SYLLABLE and FOOT-BINARITY), then degenerate feet, i.e., feet which

consist of only one light syllable (L), are disallowed. A parsed light syllable (L) will

constitute a violation of FOOT-BINARITY; and an unparsed light syllable L will

constitute a violation of PARSE-SYLLABLE.

But what happens when the number of syllables or moras is odd? A degenerate

foot (L) or an unparsed L is then inevitable. Consider, for example the sequence in (5)

below:

(5) LLLLL

Due to FOOT-BINARITY, the parsing of (5) into feet results in two binary feet and one

degenerate foot (L), or two binary feet and one unparsed L as in (6a) and (6b)

respectively:

(6)
a. (LL) (LL) (L)
b. (LL) (LL) L

78
The important point here is that whether we adopt (6a) or (6b), there remains a problem

relating to the distribution of feet. (7) below shows three possible distributions for (6b)

above:

(7)
a. (LL) (LL) L
b. (LL) L (LL)
c. L (LL) (LL)

It seems that the problem then lies in the location of the unparsed syllable (or degenerate

foot). This can be resolved by making use of some other alignment constraint- this time

to decide upfront the distribution of syllables into meterical feet. As far as the foot is

concerned, the general alignment requirement in (4) above can be broken into two

specific constraints: ALL-FEET-RIGHT and ALL-FEET-LEFT. Whereas the former

aligns each foot with the word on the right edge, the latter aligns each foot with the word

on the left edge.

All-Feet-Right
Align each foot with the word on the right edge.

All-Feet-Left
Align each foot with the word on the left edge.

Notice that any parsing of a word into feet that results in more than one foot for each

prosodic word constitutes a violation of both ALL-FEET-RIGHT and ALL-FEET-LEFT.

Of all the parses in (8) below, only (a) satisfies ALL-FEET-RIGHT, and (b) satisfies

ALL-FEET-LEFT; (c-h) each incur one or more violations of both constraints.

(8)
a. L L L (LL)
b. (LL) L L L
c. (LL) (LL) L

79
 d. (LL) L (LL)
f. L (LL) (LL)
g. L (LL) LL
h. L L (LL) L

In each parse in (8c-h), there is at least one foot that this not at the right or left edge of the

prosodic word.

The question arises then as to how violations of ALL-FEET-RIGHT and ALL-

FEET-LEFT are counted, especially when the parsing of a word results in more than one

foot as in (8c-f) above. The answer lies in the fundamental claim of OT that all

constraints are violable, but violation must be minimal. The degree of violation, rather

than absolute violation, is what matters here. Let’s take some of the examples in (8)

above to illustrate this fundamental principle. (8c), for example, reproduced as (9) below,

violates both ALL-FEET-LEFT and ALL-FEET-RIGHT.

(9) (LL) (LL) L

However, its violation of ALL-FEET-LEFT is not as severe as its violation of ALL-

FEET-RIGHT because only one foot is not at the left edge of the prosodic word.

Conversely (8f) above, reproduced as (10) below, violates ALL-FEET-LEFT more

severely than ALL-FEET-RIGHT because only one foot is not on the right edge:

(10) L (LL) (LL)

What about the number of violations? Consider (8d) reproduced as (11) below:

(11) (LL) L (LL)

To determine the degree of violations (not just the presence or absence of violation),

violations of ALL-FEET-LEFT and ALL-FEET-RIGHT are counted in syllables rather

than in feet. That is, we count the number of syllables that separate the foot from a

80
specified edge of the word. In (11) above, for example, there are three syllables that

separate the leftmost foot from the right edge of the word, so (11) incurs three violations

of ALL-FEET-RIGHT. Similarly, (11) incurs three violations of ALL-FEET-LEFT as the

rightmost foot is separated from the left edge of the word by three syllables. Consider

Tableau (4) below (where ALL-FEET-RIGHT dominates ALL-FEET-LEFT):

Tableau (4)
Input: /LLLLL/ ALL-FEET-RIGHT ALL-FEET-LEFT
a- (LL) L (LL) *** ***
b- (LL) (LL) L * , *** **
c- " L(LL) (LL) ** * , ***

Tableau (5) below, however, shows that (b) is optimal under the opposite ranking:

Tableau (5)
Input: /LLLLL/ ALL-FEET-LEFT ALL-FEET-RIGHT
a- (LL) L (LL) *** ***
b-" (LL) (LL) L ** *,***
c- L(LL) (LL) *,*** **

Notice that ALL-FEET-LEFT and ALL-FEET-RIGHT together conspire to prevent the

occurrence of unparsed syllables in medial positions. In other words, what Tableau (5)

and Tableau (6) above show is that, all things being equal, under no circumstances can

(a) be the optimal form. One of the other alternatives, (b) or (c), fares better regardless of

which ranking of constraint is adopted. If ALL-FEET-RIGHT dominates ALL-FEET-

LEFT, then (b) is optimal; if it is the other way round, then (c) surfaces as optimal.

Whereas outputs (b) and (c), then, each incur two violations under one ranking, (a) incurs

three violations under either ranking.

Two other alignment constraints are usually appealed to to get the correct stress

patterns cross-linguistically: WORD-FOOT-RIGHT and WORD-FOOT-LEFT. Whereas

81
the former requires that the right edge of the word be aligned with a foot, the latter

requires that the left edge of the word be aligned with a foot. In other words, WORD-

FOOT-RIGHT makes it obligatory for each prosodic word to end with a foot, and

WORD-FOOT-LEFT makes it obligatory for each prosodic word to begin with a foot.

Word-Foot-Right align the right edge of the word with a foot

Word-Foot-Left align the left edge of the word with a foot

Accordingly, whereas (12a) violates WORD-FOOT-RIGHT, (12b) violates WORD-

FOOT-LEFT:

(12)
a- (LL) (LL) L
b- L (LL) (LL)

Here, too, the degree of violation is important. Violations of both constraints are counted

in syllables. Violation of WORD-FOOT-RIGHT is counted by the number of syllables

that separate the rightmost foot from the right edge of the word. Violation of WORD-

FOOT-LEFT, on the other hand, is counted by the number of syllables that separate the

leftmost foot from the left edge of the word. In tableau (7) below, (a) is then less

harmonious than (b) as far as WORD-FOOT-RIGHT is concerned, for example:

Tableau (7)
Input: /LLLLL/ WORD-FOOT-RIGHT
a- (LL)(LL)L *!
b- " L(LL)(LL)

Output (a) loses the competition because it violates WORD-FOOT-RIGHT as the

rightmost foot is separated from the right edge of the word by one syllable; meanwhile,

(b) incurs no violation of WORD-FOOT-RIGHT.

82
Tableau (8) below, however, shows that output (a) is optimal as far as WORD-FOOT-

LEFT is concerned.

Tableau (8)
Input: /LLLLL/ WORD-FOOT-LEFT
a- " (LL)(LL)L
b- L (LL)(LL) *!

The question arises here as to how WORD-FOOT-RIGHT and WORD-FOOT-

LEFT differ from ALL-FEET-RIGHT and ALL-FEET-LEFT. First of all, whereas

violations of WORD-FOOT-RIGHT and WORD-FOOT-LEFT are counted by the

number of syllables that separate the rightmost and leftmost foot from the right and left

edge of the word respectively, violations of ALL-FEET-RIGHT and ALL-FEET-LEFT

are counted by the number of syllables that separate all feet from the right and left edge

of the word. Consider the following Tableau:

Tableau (9)
Input: /LLLLL/ WORD-FOOT-LEFT ALL-FOOT-LEFT
a- (LL)(LL)L **
b- L (LL)(LL) * *, ***

Notice that output (a) does not violate WORD-FOOT-LEFT as there are no syllables

separating the leftmost foot from the left edge of the word. However, (a) incurs two

violations of ALL-FOOT-LEFT because although there are no syllables separating the

leftmost foot from the left edge of the syllable, there are two syllables separating the

other foot from the designated edge of the syllables. Output (b), on the other hand, incurs

one violation of WORD-FOOT-LEFT, but four violations of ALL-FOOT-LEFT.

83
 Now, why both constraints? Suppose that in Language X, for example, feet are

aligned with the word on the right edge (i.e. ALL-FEET-RIGHT) to get main word-stress

on the penult:

(13) L(LL)(LL)
L(LL)(LL)(LL)

Suppose, too, that in this language the first syllable of each word receives secondary

stress. If no foot is erected on the leftmost syllable, then in no way can that syllable be

secondarily stressed. So, the parsing of (13) needs some modification, so that the first

syllable can be part of a foot as follows:

(14) (LL)L(LL)
(LL)L(LL)(LL)

Up to this point, we have introduced the constraints that interact to yield the

correct parsing of syllables into feet crosslinguistically, but we have said almost nothing

about main stress assignment. That is, the constraints introduced so far do not specify

which foot serves as the head of the higher-level prosodic category, namely the prosodic

word. As each foot has one syllable functioning as its head, each prosodic word has to

have one of its feet as its head. So (1) above can be modified as (15) below:

(15)
PW = Prosodic Word

F = Foot

σ = Syllable

The goal now is not just to show the alternation of stressed and unstressed syllables (i.e.,

stress pattern), but to figure out where the main word-stress falls. This can be achieved

84
through the interaction of two more alignment constraints, namely MAIN-RIGHT and

MAIN-LEFT:

Main-Right Align the head-foot with the word on the right edge
Main-Left Align the head-foot with the word on the left edge

The former requires that the head foot be aligned with the word on the right edge as in

(16a) below; the latter requires that the head foot be aligned with the word on the left

edge as in (16b):

(16)
( ) Head Foot
a- (LL) (LL) Head Syllable

( ) Head Foot
b- (LL) (LL) Head Syllable

MAIN-RIGHT is violated if the head foot does not occur on the rightmost edge of the

prosodic word. MAIN-LEFT, on the other hand, is violated if the head foot does not

occur on the leftmost edge of the prosodic word. Consider Tableau (10) below:

Tableau (10)
Input: /LLLLL/ MAIN-RIGHT MAIN-LEFT
a- (LL) (LL)48 L *
b- L(LL) (LL) *

If MAIN-RIGHT is undominated, then each word ends with a main stress foot. If MAIN-

LEFT, however, is undominated, each word begins with a main stress foot.

All in all, these alignment constraints conspire not only to yield the correct stress

patterns in the world’s languages, but they also help to single out which syllable carries

primary stress.

48
Notice that the head syllable is determined by whether the system is iambic (LL) or trochaic (LL).

85
 However, there are other factors involved in determining which syllable surfaces

as most prominent relative to all the other syllables in the word. For example, cross-

linguistically there is a tendency to avoid stress clash, i.e., stressed syllables should not be

adjacent. In OT, this requirement is established by the following constraint:

NOCLASH49
Adjacent stressed syllables are prohibited

Notice that this constraint is responsible for creating alternating stress patterns as in (13)

below, for example:

(17) thirteen women

Consider Tableau (11) below which shows how alternating stresses can become optimal:

Tableau (11)
Input: /σσσσσ/ NOCLASH
a- " (LL) (LL) L
b- (LL) (LL) L *
c- (H)(LL) (LL) *

Another fact concerning word prominence crosslinguistically is that many

languages tend not to stress the last syllable of the word. In pre-Optimality literature, this

requirement was addressed under the notion of extrametricality, where some prosodic

head (e.g., mora, syllable, foot) is considered “invisible for the purposes of creating

metrical structure” (Hayes 1991: 47)50. In optimality literature, this has been translated

into a constraint:

Nonfinal
The final prosodic unit (e.g. mora, syllable) should not be footed

49
For pre-optimality literature on this point, see Liberman and Prince (1977), Prince (1983), Hammond
(1984).
50
For pre-optimality literature, see Hayes (1979, 1980, 1982a, 1991), and McCarthy (1979).

86
The problem now arises as to which prosodic category (e.g. mora, syllable, foot) should

be considered extrametrical, i.e., invisible to stress rules. Consider (18) below:

(18) PW = Prosodic Word

F = Foot

σ = Syllable

µ = Mora

What (18) basically says is that the mora µ is the prosodic head of the syllable σ, which

in turn is the prosodic head of a higher-level prosodic category, namely the foot F, which

is also the head of a still higher-level prosodic category, the prosodic word PW. The

debate is then over which of these prosodic categories is invisible to parsing for stress

placement purposes.

Extensive literature on this subject (Cf. McCarthy 1979 for Arabic; Hayes 1980,

1982 for English and Latin; Prince 1980 for Estonian; inter alia) has shown that this is a

language-particular phenomenon. Whereas some languages (e.g. Arabic) require that the

last mora be extrametrical, others (e.g. Latin) require that the last syllable be

extrametrical; and still others consider the last foot to be extrametrical51. Consider

Tableau (12) below (where the syllable is considered extrametrical):

Tableau (12)
Input: /LLLLL/ NONFINAL (σ)
a- " (LL)(LL)L
b- (LL)(LL)(L) *

51
Hayes (1982) argues for consonant, rhyme and suffix extrametricality when accounting for English stress
behavior.

87
 Apart from the controversy over which prosodic head (i.e. mora, syllable, or foot)

should be extrametrical, all languages share the fundamental fact that extrametricality is

blocked if it renders the whole word extrametrical. This is usually referred to in the

literature under the notion of Nonexhaustivity (Hayes 1991). The other cross-linguistic

restriction on extrametricality is that the right-edge of the word is less marked than the

left edge of the word. This is usually called Edge Markedness (Hayes 1991). Yet, some

languages impose a similar requirement on the initial prosodic head so as to avoid

stressing the initial syllable:

Noninitial
The initial prosodic unit (e.g. mora, syllable) should not be footed.

Consider tableau (13) below:

Tableau (13)
Input: /LLLLL/ NONINITIAL (σ)
a- " (L)(LL)(LL)
b- L(LL)(LL) *

In section 3.2.1 above, we surveyed the constraints which are responsible for stress

placement in all languages. The important point here is that although these constraints are

assumed to be operative in every language, the output forms in those languages are

distinct. OT offers the following straightforward explanation for this fact: constraints do

not have equal weight in the world’s languages. On the contrary, constraints may

dominate other constraints and this hierarchical domination relationship between the

constraints results in a language-particular ranking of constraints. When two constraints

like X and Y, for example, state conflicting demands on some input form, the output form

that results when X dominates Y is often different from that when Y dominates X. Not

88
only should we then be interested in the satisfaction (or lack thereof) of these constraints,

but we should also be interested in how these constraints interact to yield surface

structures.

3.3 An Optimality Analysis of Stress in Arabic and English

In this section, we will show how the ranking of the constraints introduced in 3.2.1 can

account for the place of stress in Arabic and English. However, before we undertake this

task, we will need to lay down the basic facts and generalizations regarding stress in both

languages. In 3.3.1, the basic facts and generalizations regarding word-stress in Arabic

will be sketched out. In 3.3.2, we will present the basic facts and generalizations

89
regarding word-stress in English. Some additional remarks regarding the two systems

will be sketched out in 3.3.3. In 3.3.4, we will develop an optimality analysis of word

stress in Arabic. In 3.3.5, we will develop an optimality analysis of word-stress in

English.

3.3.1 Stress Assignment in Arabic52

For the distribution of stressed and unstressed syllables in Arabic, two main factors are

usually taken into consideration: (1) the internal structure of the syllable (specifically

“weight”), and (2) the position of the stressed syllable relative to all other syllables in the

same word. As for rhythmicity (the alternation of stressed and unstressed syllables),

Arabic is an unbounded system in that it allows long strings of unstressed syllables. In

Arabic, each word has a single main word-stress. None of the studies that I have come

across makes the claim that classical Arabic has secondary stress; and only very few

studies (Cf. Mitchell 1960; Welden 1980) claim that some varieties of Arabic (e.g.

Egyptian Arabic) have secondary stress. Needless to say, this kind of language provides

a challenge to any theory which aims to uncover the rules underlying the distribution of

stress53.

As for the internal structure of the syllable, Arabic has mainly three syllable types:

light /cv/, heavy /cvc, cvv/ and superheavy /cvcc/, /cvvc/. Consider the following

examples:

52
For some background information about Arabic, see the Endnotes section for this chapter
53
In some languages, stress is an arbitrary, unpredictable property of the lexical item in which the syllable
occurs (cf. Russian and Japanese). In some languages, syllables are stressed according to their position in
the word (cf. penultimate syllable in Polish)

90
 (19)

Light: /CV/ muallim, kaatabaa, yataqaataluuna, arabt, kataba

Heavy: /CVV/ mustašfaa, kaatabaa, kaatabataa, yataqaataluuna,
šaabaatun
/CVC/ muallim, šaabaatun, mustašfaa
54
Superheavy /CVVC/ muštaaqaat, šaabaat, amaal, yasaluun
/CVCC/ arabt, akalt, samit

As for the position of the stressed syllable relative to all other syllables in the word,

stress in Arabic falls in one of three categories: oxytonic (stress on the ultimate),

paroxytonic (stress on the penult), or proparoxytonic (stress on the antepenult) (Mitchell

1960). Stress in Arabic, then, never falls on a pre-antepenult syllable. Having all of this in

mind, primary word-stress in most Arabic dialects (including Jordanian Arabic) is

distributed along the following lines:

(20)

i. Stress falls on the last syllable of the word iff (if and only if) it is superheavy, i.e.,

/cvcc/ or /cvvc/

e.g., darast, dukkaan, babeen, etc.

ii. If the final syllable is not superheavy, stress goes to the penult if it is heavy, i.e.,

/cvc/ or /cvv/

e.g., baarak, katabna, maktabna, kitaabah, maktabitna, etc.

54
It should be mentioned that although light and heavy syllables can occupy initial, medial and final
positions, superheavy syllables only occur word-finally:
CV (occur anywhere in the word)
CVC (occur anywhere in the word)
CVV (occur only in ultimate and antepenult positions)
CVVC (occur only in ultimate position)
CVCC (occur only in ultimate position)

91
 iii. Otherwise, stress falls on either the penult or the antepenult whichever is

separated from a preceding heavy syllable (or word boundary) by an even number of

light syllables (Langendoen 1968: 102).

e.g., kataba, maktabah, katabataa, šajarah, šajaratun, šajaratuhu, murtabiatun,

adwiyatuhum adwiyatuhumaa, etc.

Although these are the general principles governing the distribution of stressed and

unstressed syllables in Arabic, there are a few remarks that I think I should mention here.

First of all, the differences between the different Arabic dialects as far as word-stress is

concerned are only minor55 (Kenstowicz 1982, 1994; Abu-Salim1987; McCarthy1980;

Irshied & Kenstowicz 1984; Welden1980). In Jordanian Arabic, for example, there are

three main varieties: Urban, Rural and Bedouin. The first two varieties are identical with

the standard variety as far as stress is concerned. Bedouin Jordanian Arabic differs from

the other two varieties (and the rest of other Arabic varieties) in at least two ways:

(21)

(a) In disyllabic words whose second syllable is not superheavy and first
syllable is light, stress goes to that second syllable, whereas in the other
dialects, it goes to the first syllable following rule (iii) above. Consider the
following examples:

Bedouin Urban Rural Gloss

bina bana bana “he built”

jimal jamal jamal “camel”

55
For example, whereas Classical Arabic cvv in final position is not stressed (raaa), Egyptian cvv in
final position is stressed, e.g., šuftii “you saw him?” (see Welden 1980: 102)

92
 bişal başal başal “onion”
gidam56 gadam adam “foot”

(b) In polysyllabic words whose final syllable is not superheavy and whose
penultimate syllable is not heavy, stress goes to either the penult or the
antepenult whichever is separated from a preceding heavy syllable or word
boundary by an odd number of syllables:

Bedouin Urban Rural Gloss

ankitalaw inatalu inkatalu “they were killed”

zalamatak zalmatak zalmatak “your man”

Notice that although the segmental string may be changed from one variety to another,

stress assignment is subject to the same guiding principles as in (20) above for Urban and

Rural dialects, and as in (20) and (21) above for Bedouin.

Second, in almost all varieties of Arabic, superheavy syllables (i.e., /cvcc/ and

/cvvc/) are only found in final position. In order to maintain the contrast between light

and heavy syllables throughout the entire word, the notion of extrametricality is often

invoked in pre-Optimality literature. That is, when the last mora of each prosodic word is

considered extrametrical, the whole prosodic word is then viewed as consisting of light

(CV) and heavy (CVC, CVV) syllables. In Optimality theory literature, this has been

taken care of by putting forward a violable constraint, namely NONFINAL.

Third, with very minor exceptions, suffixed and unsuffixed forms behave alike as

far as word-stress is concerned57; so the “inclusion or exclusion of suffixes is irrelevant to

the facts of prominence” (Mitchell 1960: 371). Consider:

(22)

56
According to Hung (1995: 30), these examples can be accounted for if the iambic, rather than trochaic,
analysis of foot form is adopted.
57
Interestingly enough, Arabic orthography successfully handles this issue in that suffixes, prefixes
(including even the particle prefix ‘al’) are fused with the word.

93
 ki.taab (rule 1) “book”
al-ki.taab (rule 1) “the book”
ki.taa.b-un (rule 2) “a book”
ki.taa.b-ak58 (rule 2) “your book”
kitabu-ha (rule 3) “her book”
kitabu-huma (rule 3) “their (dual) book”

Fourth, in Classical Arabic, each word has two forms: a pause- and non-pause

form. For example, the pause form of the word “teacher’ is “muallim” but the non-

pause form is muallimun. Notice here that although the stressed syllable in the two

cases is different, both forms are stressed according the rules stated above. All of this

shows that “prominence is a function of the total syllable pattern” (Mitchell 1960).

3.3.2 Stress assignment in English59

Treatments of stress placement in English have made reference to three kinds of

information. These include syntactic information (whether the word is noun, verb, etc.,

e.g. subject vs. subject), morphological structure (whether the word is mono- or

polymorphemic, e.g. limit vs. limitation), and phonological information (a composite

variable that provides information about syllable structure, i.e., light vs. heavy syllable,

number of syllables, and syllable position in the word). It will turn out that all these kinds

of information are relevant in determining the place of stress in English words.

As for word class, most textbooks (Cf. Cruttenden 1986, Kreidler 1989, among

others) distinguish between stress assignment rules for verbs, adjectives, and nouns.

58
Notice here that when ki.taab is suffixed with –ak, the whole word is re-syllabified: the coda of the last
syllable of ki.taab has become the onset of the suffix.
59
For stress variation in British and American English, see Berg (1999)

94
Cruttenden (1986: 19), for example, proposes a set of stress assignment rules in English

for verbs and adjectives on the one hand and nouns on the other:

(23)

(i) Verbs and adjectives

(a) stress is on the penultimate syllable when final syllable has a short vowel in an
open syllable or is followed by no more than one consonant, e.g.

surrender, polish, astonish, rigid, explicit

(b) otherwise, stress is on the final syllable (subject to rule (ii) below), e.g.

relate, maintain, sublime, severe, defend, abrupt

(ii) Nouns

(a) if the final syllable has a short vowel, disregard it and apply rules under (i)

above, e.g.

elephant, moment, complexion, surrender

(b) if the final syllable has a long vowel, it is stressed (subject to (iii) below, e.g.

police, machine, dispute, campaign

(iii) Words of more than two syllables with a long final vowel: stress on the
antepenult syllable, e.g.

anecdote, Fahrenheit, pedigree, organize, escalate, moribund, erudite

Kreidler (1989: 204) proposes a similar set of rules, but assumes that adjectives are

divided into two groups as far as stress assignment is concerned: (1) adjectives that

follow the stress rules of verbs, and (2) adjectives that follow the stress rules of nouns.

According to Kreidler (19989: 197-218), there are two basic types of stress rules: stress

95
rules for verbs, and stress rules for nouns. He incorporates these rules into decision trees

like (a) and (b) below:

(24)

(a) Decision tree for verbs

Is the ult stressable (V: or VCC)

No Yes

Penult is stressed How many syllables in the verb?

(e.g. solicit)

Two syllables More than two syllables

Ult is stressed Antepenult is stressed

(e.g. contain, erupt) (e.g. exaggerate, supplement)

(b) Decision tree for nouns

How many syllables are there in the noun?

Two syllables More than two syllables

Penult is stressed Does the ult have a free vowel?60

(e.g. promise, tribune)

60
According to Kreidler (1989: 199), “a free vowel is one which can occur at the end of a one-syllable
word” illustrated in the following words: see, bay, by, sue, go, cow, spa, law, toy.

96
 Yes No
Antepenult is stressed Is the penult stressable
(e.g. anecdote)

No Yes

Antepenult Penult is
is stressed stressed
(e.g. cinema) (e.g. aroma)

Notice that all these rules make reference to the internal structure of the syllable (light vs.

heavy in the case of Cruttenden, having (or lack thereof) free vowels in the case of

Kreidler, etc.). In addition, they involve counting the number of syllables (sometimes

from the end of the word, sometimes the total number of syllables in the word). When

accounting for these generalizations from an optimality-theoretic approach in section

3.3.4 below, we will see that there are constraint requirements, such as Peak-prominence,

Edgemost, Nonfinality, etc. (Prince and Smolensky 1993: Chapter Four) that all demand

satisfaction for stressing English words.

Not only this, but the morphology of the word in English turns out sometimes to

be very decisive for stress placement purposes. In this regard, Kreidler (1989: 200) lists a

whole set of neutral suffixes that should be recognized before the above stress rules are

set to work. Cruttenden (1986: 20), on the other hand, divides English suffixes into three

main categories as far as stress placement is concerned:

(25)

97
(a) suffixes which leave the stress on the stem unaffected, e.g. fulfil/fulfilment;

usual/usually

(b) suffixes which themselves take the stress, e.g. limit/limitation;

picture/picturesque; China/Chinese

(c) suffixes which shift the stress on the stem, e.g. economy/economic;

curious/curiosity; apply/applicant; maintain/maintenance

Despite these fine treatments, everyone seems to admit that the English stress rules are

limited in their domains, and often “murky and exception-ridden” (Hammond 1999: 192).

However, the crucial point here is that the treatment of stress assignment in English in

Optimality theory (which is after all a theory of linguistic variation) will show that stress

assignment is subject to demands of conflicting forces, and the realization of stress is a

matter of constraint satisfaction. As each grammar is a unique way of solving the conflict

between the violable constraints by making use of the notion of constraint ranking, it is

expected that English has its own ranking of the conflicting constraints that sets it apart

from all other languages.

3.3.3 English Stress vs. Arabic Stress: Additional Remarks

Although English and Arabic word-stress assignment are subject to demands imposed by

the morphological structure of the constituents, their rhythmic organization, intrinsic

prominence of the syllable (i.e., its weight), number and position of the syllables in the

word, there are some differences between the two systems, including function of stress,

position of stress, degrees of stress, and vowel reduction.

98
 As for the function of stress, whereas word-stress in English is phonemic, it is

allophonic (i.e. predictable) in Arabic. This has surely contributed a lot to the fact that

Arabic word-stress is more fixed. Most treatments of Arabic word-stress leave no (or

very little) residual data unaccounted for. On the other hand, when surveying the

principles that govern the distribution of English word-stress, long sections are usually

devoted to discussing the details of exceptions of many kinds.

As for degrees of stress, most studies of Arabic word-stress admit that every word

in Arabic has only one main stress; secondary stress is rarely reported (Cf. Mitchell 1960,

for Egyptian Arabic); tertiary stress never reported. On the other hand, English

distinguishes up to four degrees of stress: primary stress, secondary stress, tertiary stress

and zero stress (unstressed). Textbooks usually devote the bulk of discussion about word

stress to the primary stress. Only a few textbooks deal with secondary stress. The other

degrees of stress are rarely dealt with. However, most treatments of English word stress

admit that the alternation of stressed and unstressed syllables is basically rhythmic.

Consider, for example, the distribution of secondary stress in English (Fudge 1984:31):

(26)

(i) If there is only one syllable before the one with main stress, no secondary

stress is assigned, e.g. lapel, veranda, America

(ii) If there are two syllables before the one with main stress, secondary stress

is always assigned to the first of these (i.e. two syllables back from main

stress), e.g. aluminium, panorama61

(iii) If there are three or more syllables before the one with main stress, then:

61
primary stress in bold, secondary stress in italics.

99
 (a) If there is a strong syllable two syllables back from main stress, it

takes secondary stress, e.g. en.cyc.lo.pe.di.a

(b) If there is a weak syllable two syllables back from main stress, the

third syllable back from main stress takes secondary stress, e.g.

far.ma.co.pe.a, sep.tu.a.ge.si.ma

The last point that I would like to raise concerning the differences between Arabic

and English as far as word stress is concerned is related to vowel reduction. In English,

vowels that occupy the nucleus of unstressed syllables are subject to vowel reduction: the

more the syllable is unstressed, the more likely its nucleus gets reduced to a schwa.

Grammatical and functional words are good examples; they always appear in their weak

forms if there is no good reason to do otherwise. Unstressed vowels in Arabic, on the

other hand, are always realized in their full forms.

Having surveyed the constraints that are thought to be responsible for the

realization of different stress patterns in the world languages in 3.2.1 above, our task

shifts now to apply the theory to genuine metrical systems. In 3.3.4 below, we will

analyize Arabic stress patterns in a constraint-based framework. In 3.3.4, we will analyize

English stress patterns in a constraint-based framework. Our main task will be to find out

the constraint ranking particular to Arabic and English.

3.4 Constraint Interaction in Arabic

100
 The general principles governing the distribution of stressed and unstressed

syllables in almost all Arabic dialects, including Jordanian Arabic, are discussed in 3.3.1

above. In this section, I undertake the task of accounting for the place of main word-

stress in Arabic in a constraint-based framework. In order to do just that, our task should

be twofold: (1) we need to find out the constraints that are relevant when accounting for

the stress patterns in Arabic, i.e. the constraints that are not ranked low enough to the

extent that their impact may not be even noticeable; and (2) we need to find out how

these constraints are ranked relative to each other.

First of all, although there are monosyllabic words in Arabic, there is still a

minimality requirement on what constitutes a possible lexical word. A monosyllabic

word in Arabic must at least consist of a heavy or superheavy syllable (i.e. σµµ or σµµµ),

but never a light syllable (i.e. σµ). Whereas monosyllabic words such as those in (27a-c)

below are possible (and in fact actual) Arabic words, those in (27d) are not:

(27)
a. min, ab, etc.
b. fii, maa
c. umm
d. *mi, fi

In OT terms, this minimality requirement is entailed by the interaction of two

constraints, namely, FOOT-BINARITY and Lx=Pr (where Lx stands for lexical word

and Pr for prosodic word):

Foot-Binarity
Feet are binary

Lx=Pr
Every lexical word must consist of a prosodic word

101
The requirement that every lexical word should consist of a prosodic word can be further

paraphrased as follows: Every prosodic word (PW), as a high-level prosodic category,

must have the next available lower-level prosodic category as its head. The hierarchy in

(18) above is reproduced as (28) below:

(28) PW = Prosodic Word

F = Foot

σ = Syllable

(28) basically states that the next available prosodic category to function as the head of

the prosodic word (PW) is the foot (F). The minimality requirement in Arabic is now

clear: Every lexical word must contain at least one foot functioning as its head. And since

Arabic is a quantity-sensitive language as far as stress is concerned, the constraint that

defines the form of the foot (i.e. FOOT-BINARITY) is interpreted under a moraic

analysis. FOOT-BINARITY in Arabic then is reformulated as follows:

Foot-Binarityµ
Feet are binary under moraic analysis (i.e. consist of exactly two moras)

FOOT-BINARITYµ explains why the words in (27a-c) above are possible Arabic words,

but those in (27d) are not. Under this analysis, any foot that consists of one mora (µ)62

constitutes a violation of FOOT-BINARITYµ. Two light syllables (LL) or one heavy

62
By the same token, any foot that consists of more than two moras (e.g. (µµµ)) constitutes a violation of
FOOT-BINARITYµ

102
syllable (H) can make up a possible foot in Arabic, and ultimately a possible prosodic

word:

Tableau (14)
Input: /fii/ FOOT-BINARITY-µ
a- " (fii)63
b- (fi)i *!

Tableau (15)
Input: /LL/ FOOT-BINARITY-µ
a- (rama)64
b- (ra)ma *!
c- ra(ma) *!

Accordingly, the optimal parsing of a word like /adwiyatuhu/ into moraic feet is

determined:

Tableau (16)
Input: /?adwiyatuhu/ FOOT-BINARITY-µ
a- (ad) (wiya) (tuhu)
b- (ad wi) (yatu) (hu) **
c- (ad) (wiyatu) (hu) **
d- (ad) (wiya)(tu) hu *
e- (ad)(wi)(yatu)hu *

63
One may wonder why this syllable is footed. In pre-Optimality literature, it is often assumed that the last
syllable in Arabic is extrametrical. Observing this requirement in an OT analysis means that the last
syllable should be unfooted (i.e. NONFINAL). We will see later that NONFINAL is outranked by the
constraint which requires that at least one foot be constructed over the prosodic word, that is, Lx=Pr , so
that (fii) is a better parse than fii.
64
The reason why a pointing hand "is not used to mark the output form that incurs fewer violations of
FOOT-BINARITY-µ is that output (a) in Tableau (16) is not the one that ultimately wins the competition;
rather, it is (b) that wins the competition. This is so because FOOT-BINARITY-µ is dominated by
NONFINALσ which, in turn, is dominated by Lx=Pr.

103
According to Tableau (16) above, if FOOT-BINARITYµ were undominated, output (a)

would win the competition because it incurs no violation of FOOT-BINARITYµ. Output

(a) in Tableau (16) above shows that /adwiyatuhu/ is optimally parsed into three feet,

each of which consists of two moras. Output (b), on the other hand, loses the competition

because it incurs two violations of FOOT-BINARITYµ: its first foot is constructed over

three moras, and its third foot over one mora. Similarly, output (c) incurs two violations

of FOOT-BINARITY-µ: its second foot is erected over three moras, and its third foot

over one mora. Outputs (d-e) each incur one violation of FOOT-BINARITYµ because

there is one foot that is constructed over one mora in each case.

However, in order to get the correct stress pattern in a word like /adwiyatuhu/,

FOOT-BINARITYµ is not the only constraint that needs to be satisfied. Other universal

constraints demand satisfaction, too. However, simultaneous satisfaction of these

constraints is impossible. It turns out that violation of some constraints (e.g. FOOT-

BINARITYµ) is sometimes inevitable. Lower-ranking constraints are violated so as to

avoid violation of higher-ranking constraints. In order to demonstrate this, we need to

bring the constraints that interact with FOOT-BINARITY-µ into play. These include

NONFINAL, TROCHAIC, MAIN-RIGHT, PARSE-σ, ALL-FEET-LEFT, and ALL-

FEET-RIGHT.

Nonfinal
The final syllable should not be footed65

Parse-σ

65
As for final superheavy syllables which surface with main word-stress in Arabic, it will turn out later that
NONFINAL is outranked by a constraint that requires final superheavy syllables to be stressed.

104
 A syllable must be footed

Main-Right66
Align the head-foot with the word, on the right edge

All-Feet-Right
Align each foot with the word, on the right edge

All-Feet-Left
Align each foot with the word, on the left edge

In a word like /šajaratuhu/, for example, the last syllable, due to NONFINAL, should not

be footed. (29a) then is a better parse than (29b):

(29)
a. šajaratu<hu>
b. šajaratuhu

According to PARSE-σ, (30a) then is a better parse than (30b):

(30)
a. (šaja)(ratu)<hu>
b. (šaja)ratu<hu>

Once the requirements of NONFINAL and FOOT-BINARITY are observed, /šajaratuhu/

is, due to ALL-FEET-RIGHT or ALL-FEET-LEFT, parsed into two binary feet.

Adopting a left-to-right or right-to-left parsing creates the same output form:

(31)
a. (šaja)(ratu)<hu>
b. (šaja)(ratu)<hu>67

66
A point worthy of mention here is that the rightmost foot is assumed to contain the syllable with main
word-stress. Throughout this work, we assume that MAIN-RIGHT is not a gradient constraint whose
violation is counted by the number of intervening constituents; rather we assume that MAIN-RIGHT is
satisfied if the rightmost foot is the head of the prosodic word, and it is violated if the head foot is not the
rightmost one. MAIN-RIGHT is not violated by (an) unparsed syllable(s) to the right of the rightmost foot,
but it is violated if main word stress terminates on a syllable that is not part of the rightmost foot. In this
way, we avoid any interaction between MAIN-RIGHT and NONFINAL. The reason why we want MAIN-
RIGHT to work independently of NONFINAL is that we want to avoid an ordering paradox. Let us see
what happens if MAIN-RIGHT and NONFINAL interact. If the final syllable is light or heavy, we want
NONFINAL to dominate MAIN-RIGHT so that stress terminates on a nonfinal syllable ((baT)Tah); if the
final syllable is superheavy, we want MAIN-RIGHT to dominate NONFINAL so that stress terminates on
that final superheavy syllable (e.g. (šan)(taat)).

105
According to TROCHAIC, (32a) below is a better parse than (32b):

(32)
a. (šaja)(ratu)<hu>
b. (šaja)(ratu)<hu>

According to MAIN-RIGHT, (33a) below is a better parse than (33b):

(33)
a. (šaja)(ratu)<hu>
b. (šaja)(ratu)<hu>

The model predicts correctly that in a word like /šajaratuhu/, stress falls on the

antepenult. The question arises as to how (šaja)(ratu)<hu> defeats all other competitors.

Notice that meeting the requirements of all constraints simultaneously is

impossible. This suggests that there is constraint conflict. Whereas PARSE-σ, for

example, favors a footed form over an unfooted form, NONFINAL requires that the final

syllable be unfooted. Similarily, whereas ALL-FEET-RIGHT requires the foot be aligned

with the right edge of the word, NONFINAL requires the last syllable to be unfooted, etc.

In a constraint-based framework, the conflict between the constraints is resolved by

constraint ranking. That is, some constraints are higher-ranking, while others are lower-

ranking. Violation of a lower-ranking constraint is allowed if it results in satisfaction of a

higher-ranking constraint.

Another point worthy of mention is that the whole model is based upon the idea

of constraint demotion. That is, a constraint is assumed to be top-ranked unless there

arises evidence for its demotion to a lower position. At the initial stage, all constraints are

assumed to be undominated. Since our task here is to find out the relative ranking of

67
We will see later that in case the word exhibits an even number of moras (e.g. adwiyatuhu), adopting a
left-to-right or a right-to-left creates different output forms.

106
FOOT-BINRITYµ, NONFINAL, TROCHAIC, MAIN-RIGHT, PARSE-σ, ALL-FEET-

LEFT, and ALL-FEET-RIGHT in Arabic, we start out with the assumption that all of

them are undominated

Initial Stage
FOOT-BINRITYµ, NONFINAL, TROCHAIC, MAIN-RIGHT, PARSE-σ, ALL-
FEET-LEFT, and ALL-FEET-RIGHT

The demotion of any one of these constraints has be motivated by the fact that its

demotion results in satisfaction of higher-ranking constraints which serve to rule out

suboptimal forms. Hence, the constraints that are violated by the optimal form are lower-

ranking, but the constraints that are violated by the suboptimal forms are higher-ranking.

Let us start with the directionality of footing in Arabic; that is, whether feet are

constructed on a left-to-right (i.e. ALL-FEET-LEFT) or a right-to-left (i.e. ALL-FEET-

RIGHT) basis. Here, we need to do two things. One, we need to see the ranking of ALL-

FEET-LEFT and ALL-FEET-RIGHT relative to the remaining set of constraints, so as to

find out whether ALL-FEET-LEFT and ALL-FEET-RIGHT are higher-ranking or lower-

ranking constraints. If they are higher-ranking, their violation always results in

suboptimal forms; if they are lower-ranking, the form which violates them still has the

chance to surface as the optimal form. Second, we need to establish the ranking of ALL-

FEET-LEFT and ALL-FEET-RIGHT relative to each other.

As for the ranking of ALL-FEET-LEFT and ALL-FEET-RIGHT relative to the

other constraints, crucial here is the ranking of ALL-FEET-RIGHT and ALL-FEET-

LEFT relative to FOOT-BINARITYµ, NONFINALσ, MAIN-RIGHT and TROCHAIC.

107
 We need NONFINAL to dominate ALL-FEET-RIGHT and ALL-FEET-LEFT, so

that the final syllable does not get footed. This is so because in Arabic a final syllable is

often not stressed.68 In /šajaratuhu/, if the last syllable were footed (e.g. outputs d in

Tableau 17 below), stress would wrongly terminate either on the penult (if it pairs with

the penult), or on the final syllable (if it makes up a foot by itself). We also need FOOT-

BINARITYµ to be undominated, so that degenerate feet are not constructed unless

absolutely necessary69. In other words, we want degenerate feet to be sanctioned only by

higher-ranking constraints. We also need the feet to be left-headed (i.e. TROCHAIC). It

is true that TROCHAIC, and ALL-FEET-RIGHT and ALL-FEET-LEFT do not interact,

but we need to make sure that TROCHAIC is top-ranked because its violation always

results in suboptimal form (Cf. output c in Tableau 17 below). Finally, we need MAIN-

RIGHT to be undominated, so that the head foot is always rightmost. Consider how the

the correct output form for /šajaratuhu/ is obtained when invoking higher-ranking

constraints, which outrank ALL-FEET-LEFT and ALL-FEET-RIGHT:

Tableau (17)
Input: /šajaratuhu/ FBµ NF TR MR A-F-L A-F-R
a. " (šaja)(ratu)hu ** *,***
b. (šaja)(ratu)hu *! ** *,***
c. (šaja)(ratu)hu *! ** *,***
d. (šaja)(ratu)(hu) *! **,**** *,***
e. (šaja)(ra)(tu)hu *!* **,*** *,**,***

68
The final syllable receives main word-stress iff it is superheavy. We will see later that this can be
achieved in two ways: (1) factoring out NONFINAL, or (2) parameterizing WSP. These two alternative
analyses will be discussed in detail later.
69
Notice that FOOT-BINARITYµ, as a constraint on the shape of the foot, does not interact directly with
ALL-FEET-LEFT and ALL-FEET-RIGHT. Whether feet are constructed on a left-to-right or right-to-left
basis, the requirement of FOOT-BINARITYµ still holds in the same way. However, the domination
relation between FOOT-BINARITYµ on the one hand, and ALL-FEET-LEFT and ALL-FEET-RIGHT on
the other, is obtained by transitivity.

108
 According to Tableau (17) above, output (e) is ruled out because it violates

FOOT-BINARITYµ. Output (d) is ruled out because it violates NONFINAL; output (c)

is ruled out because it violates TROCHAIC; and output (b) is ruled out because it violates

MAIN-RIGHT. Output (a), on the other hand, satisfies FOOT-BINARITYµ,

NONFINAL, TROCHAIC and MAIN-RIGHT simultaneously. The fact that (a) incurs

two violations of ALL-FEET-LEFT and four violations of ALL-FEET-RIGHT does not

prevent it from winning the competition. This constitutes evidence that ALL-FEET-

LEFT and ALL-FEET-RIGHT are not higher-ranking constraints. The output form which

violates these two constraints still surfaces as the optimal form. In OT terms, the

constraints that are violated by the actual form (output (a) in this case) are always lower-

ranking constraints. ALL-FEET-LEFT and ALL-FEET-RIGHT, which are the only

constraints violated by (a), must be outranked by all the constraints that are violated by its

competitors. The constraints that outrank ALL-FEET-LEFT and ALL-FEET-RIGHT are

FOOT-BINARITYµ, NONFINAL, TROCHAIC, and MAIN-RIGHT70.

The previous example shows that ALL-FEET-LEFT and ALL-FEET-RIGHT are

outranked by FOOT-BINARITYµ, NONFINAL, TROCHAIC, and MAIN RIGHT

(separated by solid lines). FOOT-BINARITYµ, NONFINAL, TROCHAIC, and MAIN-

RIGHT are higher-ranking constraints in Arabic because their violation serves to rule out

suboptimal structures. In terms of constraint ranking, all the constraints that are violated

70
Notice that ALL-FEET-LEFT and ALL-FEET-RIGHT are separated from the constraints that outrank
them by a solid line. The constraints that are separated from each other by a dashed line are not ranked
relative to each other.

109
by the suboptimal forms dominate the constraints that are violated by the optimal form.

The constraint ranking obtained so far looks like (34) below:

(34) Constraint Ranking

FOOT-BINARITY, NONFINAL, TROCHAIC, MAIN-RIGHT >> ALL-
FEET-LEFT, ALL-FEET-RIGHT

Having shown what constraints outrank ALL-FEET-LEFT and ALL-FEET-

RIGHT, let us see how ALL-FEET-LEFT and ALL-FEET-RIGHT are ranked relative to

each other. To establish a ranking between these two alignment constraints, we need an

example that, after observing the requirement of NONFINAL, still contains a degenerate

foot (i.e. a foot that consists of only one mora).71 The location of this degenerate foot is

very decisive for stress placement. The example introduced in Tableau (16) above, i.e.

/adwiyatuhu/, is basically what we need:

Due to NONFINAL, the last syllable is not footed (i.e. extrametrical):

(35) ad wiyatu<hu>

If ALL-FEET-RIGHT applies, we get:

71
Notice that in case where the parse does not contain a degenerate foot, a left-to-right, or a right-to-left
parsing results in exact number of violations. Consider the following Tableau:
Tableau
Input: /šajaratuhu/71 A-F-R A-F-L
a- (šaja)(ratu)hu *,*** **
b- ša(jara)(tuhu) ** *,***
According to this Tableau, (a) incurs four violations of ALL-FEET-RIGHT, and two violations of ALL-
FEET-LEFT. Conversely, (b) incurs two violations of ALL-FEET-RIGHT, and four violations of ALL-
FEET-LEFT. All things being equal, output (a) wins if the word is parsed on a left-to-right basis, i.e. ALL-
FEET-LEFT >> ALL-FEET-RIGHT, and output (b) wins if the word is parsed on a right-to-left basis, i.e.
ALL-FEET-RIGHT >> ALL-FEET-LEFT. As each parse incurs the same number of violations under
either ranking, the correct output form is obtained by invoking other constraints (see Tableau 17 above).

110
 (36a) (ad)(wi)(yatu)<hu>

And due to MAIN-RIGHT and TROCHAIC, stress terminates on the antepenult:

(36b) *(ad)(wi)(yatu)<hu>

If, however, ALL-FEET-LEFT applies, we get:

(37a) (ad)(wiya)(tu)<hu>

And due to MAIN-RIGHT, main-word stress terminates correctly on the the penult:

(37b) (ad)(wiya)(tu)<hu>72

It turns out that the decision concerning the relative ranking of ALL-FEET-LEFT

and ALL-FEET-RIGHT in this example is crucial. If ALL-FEET-RIGHT is ranked

higher than ALL-FEET-LEFT, the wrong output form is obtained (Cf. 36b). If ALL-

FEET-LEFT is ranked higher than ALL-FEET-RIGHT, the correct output form is

obtained (Cf. 37b). In other words, whereas the left-to-right parsing of syllables into

metrical feet yields the correct surface form, the right-to-left parsing yields the wrong

output form. Examples like this one provide an argument for ranking ALL-FEET-RIGHT

below ALL-FEET-LEFT.

Another piece of evidence that supports the claim that ALL-FEET-LEFT

dominates ALL-FEET-RIGHT in Arabic comes from the fact that stress does not fall on a

pre-antepenultimate syllable (see 3.3.1 above). In Arabic, a light penult or antepenult can

be stressed provided that it is separated from a preceding heavy syllable or the initial

word boundary by an even number of light syllables (including zero). This counting

72
Notice that (36b) and (37b) violate FOOT-BINARITYµ equally. As the optimal form violates FOOT-
BINARITYµ, this suggests that FOOT-BINARITYµ is a lower ranking constraint. We will see later what
constraint(s) dominate(s) it.

111
phenomenon in Arabic is interesting because it provides evidence as to the direction of

footing. Stressing a light penult or antepenult in long words in Arabic requires that at

least one independent foot be constructed over two light syllables which separate the

stressed syllable (whether the penult or the antepenult) from a preceding heavy syllable73

or word boundary. In words with an odd number of syllables, this is achieved through

either a left-to-right or right-to-left parsing.74 In words with even number of syllables,

however, the correct output form can be obtained only with left-to-right parsing. Consider

the following examples, where the syllables are parsed into metrical feet on a left-to-right

basis:

(38)
a. (CVCV)<CV>
b. (CVCV)(CV)<CV>
c. (CVCV)(CVCV)<CV>
d. (CVCV)(CVCV)(CV)<CV>
e. (CVCV)(CVCV)(CVCV)<CV>
f. ETC

(38a), (38c), and (38e) provide no clue as to the direction of footing. Because these

consist of an odd number of light syllables, a left-to-right or a right-to-left parsing will

result in the same footing. The last syllable, due to NONFINAL, remains unfooted; and

stress, due to MAIN-RIGHT and TROCHAIC, goes to the antepenult.

(38b) and (38d), however, exhibit an even number of light syllables. Here, the

direction of footing is critical. Under a left-to-right parsing, the penult is stressed

correctly in (38b) and (38d). After observing the requirement of NONFINAL, an

exhaustive left-to-right parsing of syllables into metrical feet causes the penult to make

73
Due to FOOT-BINARITYµ, a heavy syllable makes up a foot by itself.
74
Here, we talk about light syllables where each one of them weighs only one mora.

112
up a degenerate foot75; and stress, due to MAIN-RIGHT and TROCHAIC, falls on the

penult.

It should be pointed out that penult stress (Cf. 38b and 38d) cannot be obtained

with a right-to-left parsing. Consider (39) below, with right-to-left parsing:

(39)
a. (CVCV)<CV>
b. *(CV)(CVCV)<CV>
c. (CVCV)(CVCV)<CV>
d. *(CV)(CVCV)(CVCV)<CV>
e. (CVCV)(CVCV)(CVCV)<CV>

As can be seen, a right-to-left parsing of syllables into feet in (39) yields the correct stress

pattern for (39a), (39c) and (39e), but fails to yield the correct stress pattern for (39b) and

(39d). In fact, it fails to yield the correct output form for all words where stress falls on a

degenerate foot (Cf. 38b and 38d above). Put more succinctly, it yields main word-stress

on a light antepenult in (39a), (39c), and (39e), but it fails to yield main word stress on a

light penult in (39b) and (39d). The reason is that a right-to-left parsing always results in

having the degenerate foot as far to the left as possible (e.g. 39b and 39d)- a state of

affairs that we want to avoid. What we actually need is to have the degenerate foot fall as

far to the right as possible (e.g. 38b and 38d above). This can never be achieved when

parsing takes place on a right-to-left basis, because the light penult pairs with the light

antepenult in one foot76, and stress, due to MAIN-RIGHT and TROCHAIC, always falls

on the antepenult (e.g. 38b and 38d). Such being the case, a right-to-left parsing can

75
Notice that forming a degenerate foot requires the violation of FOOT-BINARITYµ. We will see later
that in Arabic, FOOT-BINARITYµ is dominated by PARSEσ, which in turn is dominated by NONFINAL.
This amounts to saying that once the final syllable is considered invisible to parsing, the remaining
syllables of the prosodic word are exhaustively parsed.
76
Notice that this is always the case because once the final syllable is left unfooted, the light penult and
antepenult pair together to make up a moraic trochee (LL).

113
account for the stressing of a light antepenult, but it can never account for the stressing of

a light penult.

In the case of a preceding heavy syllable (σµµ), that heavy syllable (which

constitutes a foot by itself) acts like a word boundary and footing always starts anew after

that syllable:

(40)
a. (CVC)(CVCV)<CV>
b. (CVC)(CVCV)(CV)<CV>

With a right-to-left footing, (41) will look like (42) below:

(41)
a. (CVC)(CVCV)<CV>
b. *(CVC)(CV)(CVCV)<CV>

Here, too, although a right-to-left parsing correctly predicts that stress falls on the

antepenult in (41a), it wrongly predicts that it falls on the antepenult in (41b). The reason

is that a right-to-left parsing requires the light penult and light antepenult together to

make up one foot which, due to TROCHAIC, will be always left-headed. In other words,

right-to-left can never account for the stressing of a light penult (cf. 38b, 38d and 41b).

A left-to-right parsing, on the other hand, can account for stressing a light

antepenult as well as a light penult. Recall that for a light penult or antepenult to be

stressed in long words, there must always be an even number of light syllables (including

zero) between the stressed syllable and a preceding heavy syllable or word boundary as in

(38 and 40) above. If stress falls on the antepenult, that syllable must pair with the light

penult to make up a metrical foot which is left-headed (e.g. 38a, 38c, 38e, and 40a). If,

however, stress falls on the penult, that syllable must make up a degenerate foot (e.g. 38b,

38d, and 40b). The degenerate foot cannot be constructed over the light penult unless

114
parsing takes place on a left-to-right basis, i.e. ALL-FEET-LEFT >> ALL-FEET-

RIGHT.

(42) Constraint Ranking

FOOT-BINARITY, NONFINAL, TROCHAIC, MAIN-RIGHT >> ALL-
FEET-LEFT >> ALL-FEET-RIGHT

Having established a ranking of ALL-FEET-LEFT and ALL-FEET-RIGHT

relative, first, to the other constraints, and, second, to each other, let’s back up a little to

see what evidence there is to support postulation of the other constraints (i.e.

TROCHAIC, FOOT-BINARITY-µ, NONFINALσ, and MAIN-RIGHT).

The easiest to motivate is TROCHAIC (i.e. feet are left-headed):

Trochaic
Align the head-syllable with its foot on the left edge77

Consider Tableau (18) below where the two candidate forms (a) and (b) differ only with

respect to the place of stress in the final foot:

Tableau (18)
Input: /šajaratuhu/ FBµ NF MR TR A-F-L A-F-R
a- " (šaja)(ratu)hu ** *,***
b. (šaja)(ratu)hu *! ** *,***

Recall from (28) above that the foot, as a higher-level prosodic category, must

have the next available lower-level prosodic category (i.e., the syllable) as its head. In

case the head foot is constructed over two syllables, one of them must function as the

head of the foot, and therefore must be stressed. If the left-hand syllable surfaces stressed,

the foot is said to be left-headed (i.e. TROCHAIC); if the right-hand syllable happens to

surface with main word-stress, the foot is then right-headed (i.e. IAMBIC). The fact that
77
Notice that according to this formulation, a foot that is constructed over a stressed heavy syllable (H),
and a degenerate foot that is constructed over a stressed light syllable (L) satisfy TROCHAIC and IAMBIC
simultaneously as the head syllable is aligned with both the left and right edges of the foot.

115
/šajaratuhu/ in Tableau (18) above surfaces with stress on the antepenult (which is the left

leg of the rightmost foot) demonstrates that the foot in Arabic is left-headed.78 As for the

ranking of TROCHAIC relative to the other constraints, it suffices here to say that

TROCHAIC rarely interacts with other constraints.79 In Arabic, TROCHAIC is never

violated.

Having resolved this issue, let us turn to see whether FOOT-BINARITY in

Arabic should be interpreted under a moraic or under a syllabic analysis. If FOOT-

BINARITY is interpreted under a syllabic analysis, then it is satisfied if the foot is

constructed over two syllables irrespective of their intrinsic weight (e.g. LL, LH, HL, or

HH). If it is interpreted under a moraic analysis, then it is satisfied if the foot is

constructed over two light syllables (LL) or over one heavy syllable (H).

We argue that the interpretation of FOOT-BINARITY under a syllabic analysis

does not work to account for some stress patterns in Arabic. Precisely, it does not work to

account for the placement of stress on a light penult in Arabic (e.g. HLL, etc). In a word

like [maktabah], for example, stress goes to the light penult because it is the syllable that

is separated from the preceding heavy syllable by an even number of light syllables (zero

in this case). If we assume that FOOT-BINARITY is interpreted under syllabic analysis,

then after observing the requirement of NONFINAL80, the light penult has to pair with

78
Further evidence comes from that the fact that Arabic words that consist of two light syllables, e.g. rama,
maša, etc., are stressed on their first syllables.
79
TROCHAIC may, for example, interact with WSP in stress systems where FOOT-BINARITY is
interpreted under a syllabic analysis. Suppose in language X, a foot is constructed over two syllables where
the first is light and the second is heavy (LH). All things being equal, TROCHAIC dominates WSP if stress
goes to the light syllable. If stress goes to the heavy syllable, then WSP is said to dominates TROCHAIC.
80
This entails that NONFINAL outranks FOOT-BINARITYµ. We will see later that this is true.

116
the heavy antepenult to make up one foot, e.g., (HL)L; and stress, due to TROCHAIC,

falls on the heavy penult, e.g., (HL)L, which is not correct.81

However, if FOOT-BINARITY is interpreted under a moraic analysis, stressing

the light penult in the sequence /HLL/ can be accounted. But to do just that, we first need

to acknowledge that FOOT-BINARITY is not undominated; rather, it is dominated by a

higher-ranking constraint, namely, PARSE-σ.

Parse-σ
A syllable must be footed

Let us see how under a moraic interpretaion of FOOT-BINARITY, the sequence /HLL/

can surface with main word-stress on the penult.

Due to NONFINAL, the final syllable is unfooted as in (43a) below:

(43a) HL<L>

Due to FOOT-BINARITYµ, a foot is constructed over the heavy syllable as in (43b)

below:

(43b) (H)L<L>

Due to PARSE-σ, another foot is constructed over the light penult as in (43c) below:

(43c) (H)(L)<L>82

81
Although the interpretation of FOOT-BINARITY under a syllabic analysis may not work for Arabic, we
will see later that it works just fine for English. The difference between English and Arabic in this respect
is that in English a heavy antepenult receives stress when it pairs with a light penult to make up one foot,
e.g., HLL. In a word like [asterisk], for example, stress goes to the heavy antepenult. The fact that stress in
English can also go to a heavy penult, e.g., LHL, can still be accounted for even when FOOT-BINARITY
is interpreted under syllabic analysis. One way to get stress on a heavy penult in English is to posit that
WSP outranks FOOT-BINARITYσ, so that L(H)L is a better parse than (LH)L. We will see later that this
is the case in English. FOOT-BINARITY should then be interpreted under moraic analysis to account for
the Arabic data, but under syllabic analysis to account for the English data. And this is one of the major
differences between the two languages as far as the place of main word-stress is concerned.
82
Notice that PARSEσ does not force the final syllable to be footed. This entails that the need to satisfy the
requirement of NONFINALσ is more important than the need to have each and every syllable be part of a

117
And stress, due to MAIN-RIGHT and TROCHAIC, falls correctly on the penult as in

(43d) below:

(43d) (H)(L)<L>

The interpretation of FOOT-BINARITY under a moraic analysis can then yield the

desired output form.

Foot-Binarityµ
Feet are binary under moraic analysis (i.e. consist of exactly two moras)83

Notice that to obtain the correct stress pattern in the sequence /HLL/, FOOT-

BINARITYµ should be ranked below PARSE-σ. The constraint ranking in (42) above is

modifies as in (44) below:

(44) Constraint Ranking

NONFINAL, TROCHAIC, MAIN-RIGHT, PARSE-σ84 >> FOOT-
BINARITY >> ALL-FEET-LEFT >> ALL-FEET-RIGHT

Another piece of evidence to support the demotion of FOOT-BINARITYµ below

comes from disyllabic words. In Arabic, words consisting of two syllables are divided

into two groups as far as stress is concerned:

(a) Disyllabic words which are stressed on the first syllable. This category
includes all disyllabic words that consist of light (L) and/or (H) syllables
only:

metrical foot. In terms of constraint ranking NONFINALσ >> PARSEσ. We will see in a minute that this is
exactly the case.
83
This basically means that a heavy syllable, which consists of two moras (σµµ), makes up a foot by itself;
and two light syllables where each one of them consists of one mora (σµσµ) make up one foot. In the case
of the foot that is constructed over a heavy syllable (σµµ), it does not matter which mora is the head of the
syllable because the two moras are part of one syllable which ultimately surfaces as the one that carries
stress (i.e. (σµµ)). In the case of the foot that is constructed over two light syllables (σµσµ), it does matter
which one of them is the head of the foot because each mora belongs to an independent syllable and only
one of the two light syllables can surface as the head of the foot (i.e. (σµσµ) or (σµσµ). In this latter case, we
have seen that there is evidence (Cf. Tableau 18 above) that the foot is left-headed in Arabic. That is, in
each foot that consists of a pair of light syllables, it is the leftmost syllable that must be the head of the foot,
and ultimately the one that carries stress (i.e. (σµσµ)).
84
Once a constraint is introduced for the first time, we assume that it is undominated unless evidence to the
contrary arises.

118
 LL rama, maša
LH ibir, qalam
HL igra, imši, kursi
HH ijma, baTTah

(b) Disyllabic words which are stressed on the second syllable. This category
includes all disyllabic words whose second syllable is superheavy:85

CVVC šantaat, kasaat

CVCC darast

Notice that the stressing of the final syllable is possible iff (if and only if) it is

superheavy. In OT terms, we will see later that this is sanctioned by higher-ranking

constraints. For the time being, let us just focus on the examples on (a) where stress

surfaces on a nonofinal syllable.

If FOOT-BINARITYµ were undominated, the examples in (a) would be parsed as

follows:

(45)
Footing Actual words Gloss
a. (LL) ra.ma ‘throw’ (mas. pas.)
b. L(H) i.bir ‘ink’
c. (H)L kur.si ‘chair’
d. (H)(H) baT.Tah ‘goose’

Notice that there is a discrepancy between foot formation and stress assignment in some

of the examples in (45) above.86 To illustrate, due to FOOT-BINARITYµ, TROCHAIC,

and MAIN-RIGHT87, (45) should look like (46) below:

85
The internal structure of the syllable in Arabic is discussed in 3.3.1 above.
86
There is another way to look at the facts. Hayes (1995) suggests that feet are moraic trochees, i.e. (H) or
(LL), in Arabic. In OT terms, what this means is that FOOT-BINARITYµ and TROCHAIC may be
collapsed into just one constraint. The reason why we keep FOOT-BINARITYµ and TROCHAIC as two
independent constraints is that our analysis has already suggested that in some cases the need to have each
foot be part of a metrical foot (Cf. degenerate feet in 43c above) outweighs the need to comply with the
requirement of FOOT-BINARITYµ, i.e. PARSE-σ >> FOOT-BINARITYµ. The important point is that

119
 (46) Parsing into (H) and (LL) Actual words
a. (LL) ka.na
*b. L(H) i.bir
c. (H)L kur.si
*d. (H)(H) baT.Tah

Outputs (46a, and c) show no discrepancy between the parsing of syllables into

feet and the placement of stress in the actual words. In (46a), the whole word consists of

two light syllables (σµσµ), so parsing these two light syllables into a moraic trochee

yields the desired stress pattern (* .).88 In (46c), the word consists of a heavy syllable

(σµµ) that makes up a foot by itself plus one unparsed light syllable. Being the only foot

that is constructed over the word, the heavy syllable over which the foot is constructed

ultimately surfaces as the one that carries main word-stress. All in all, if FOOT-

BINARITYµ were never violated in Arabic, the parsing of the words into binary feet

yields the correct stress patterns for only (46a) and (46c):

(47) Parsing into (H) or (LL) Actual word

(LL) ka.na
(H)L kur.si

The other examples, however, show two types of problems. First, in (46b), only

one foot is erected over the prosodic word, but stress happens to fall on the light syllable

that is not even part of that metrical foot. Consider (46b) above reproduced as (48) below:

although PARSE-σ dominates FOOT-BINARITYµ, it does not dominate TROCHAIC. PARSE-σ does not
even interact with TROCHAIC. If we choose to collapse FOOT-BINARITYµ and TROCHAIC into just
one constraint that is dominated by PARSE-σ, we commit ourselves to the claim that PARSE-σ dominates
TROCHAIC as well—something we surely do not want to do, for TROCHAIC is a top-ranked constraint
that is never violated in Arabic.
87
Notice that these three constraints do not interact, so their satisfaction is independent of one another.
88
Again, this is true only if FOOT-BINARITYµ were undominated.

120
 (48) Parsing into (H) and (LL) Actual word

L(H) i.bir

The second problem in having FOOT-BINARITYµ undominated is manifest in

(46d) above. Here, two feet are constructed over two heavy syllables; and stress, due to

MAIN-RIGHT terminates on the head syllable of the rightmost foot.

(49) Parsing into (H) and (LL) Actual word

(H)(H) baT.Tah

Fortunately, in a constraint-based framework, problems like these can be solved

by assuming that constraints are violable. That is, although some forms violate some

constraint(s), they can still surface as the actual output forms. The violation of some

constraint is sometimes inevitable. Actual output forms violate lower-ranking constraints

in order to satisfy higher-ranking ones. And this is exactly what is happening here.

In order to account for the place of stress in /hibir/ and in /baT.Tah/, we need to

do two things. One, we need to know why the final syllable does not surface with main

word-stress; and second, we need to see how stress terminates on the first syllable. As for

the stresslessness of the final syllable, recall that in Arabic final light and heavy syllables

are never stressed. This suggests that the constraint which militates against the footing of

final syllables should be invoked to account for the stress patterns in Arabic.

Nonfinal
The final syllable should not be footed.

NONFINAL is incorporated into the constraint ranking in (44) above as follows:

(50) Constraint Ranking

121
 NONFINAL89, MAIN-RIGHT, TROCHAIC, PARSEσ >> FOOT-BINARITYµ
>> ALL-FEET-LEFT >> ALL-FEET-RIGHT

Consider how [baT.Tah] can now be optimally derived:

Tableau (19)
Input: / baT.Tah/ TR MR NF Pσ FBµ A-F-L A-F-R
a- (baT)(Tah) *! * * *
b- (baT)(Tah) *! * *
c- " (baT)Tah * *

Notice that with the introduction of NONFINAL, neither (a) nor (b) wins the

competition. Rather, it is output (c) that wins because it incurs fewer violations than (a)

and (b) on the whole set of constraints: output (a) violates MAIN-RIGHT and

NONFINAL; output (b) violates NONFINAL; output (c) violates neither MAIN-RIGHT

nor NONFINAL.90 The important point here is that outputs (a) and (b) lose the

competition because they violate NONFINAL as a higher-ranking constraint.

The introduction of NONFINAL also helps to explain why the final syllable in

/ibir/ surfaces stressless.

Tableau (20)
Input: /Hibir/ TR MR NF Pσ FBµ A-F-L A-F-R
a- i(bir) *! * *
b- i.bir91 *!*

89
Once a constraint is incorporated into the domination hierarchy for the first time, we assume that it is
undominated until evidence to the contrary arises.
90
Notice that this solution creates a problem. It cannot explain why [šantaat] is stressed on the final
syllable. We will attend to this problem in detail later. It will turn out that NONFINAL is dominated by a
constraint which requires superheavy syllables to surface with main word-stress.
91
Notice that neither (a) nor (b) ultimately wins the competition. (a) loses the competition because it
violates NONFINALσ; and (b) loses the competition because no foot is constructed over the word.

122
However, unlike /baTTah/, the problem in /ibir/ is that stress surfaces on the first

syllable, which is not even part of a metrical foot. (46b) is reproduced in (51) below:

(51) Parsing into (H) or (LL) Actual word

L<H> i.bir

Here, we need not only to establish why the final syllable surfaces stressless, but we also

need to establish why the first syllable is stressed. The discrepancy between parsing

syllables into (H) or (LL) and the stress pattern in LH words (i.e. words that consist of

two syllables, where the first is light and the second is heavy)92 can be avoided if (1)

NONFINAL is undominated, and (2) if FOOT-BINARITYµ is violated.

If NONFINAL is undominated, the final syllable does not receive stress simply

because it is not footed at all as in (52) below:

(52 ) Moraic trochees, i.e. (H) and (LL) Actual word

LH i.bir

The final syllable is no longer a candidate to receive main word-stress because it is not

part of a metrical foot that can function as the head of the prosodic word. But notice that

the first syllable is not footed, either. The unfooting of the first syllable is due to FOOT-

BINARITYµ which requires each and every foot to consist of a heavy syllable (H) or two

consecutive light syllables (LL). This entails that the first syllable cannot be stressed

because it is not part of a metrical foot, either. But the fact of the matter is that the first

syllable does receive main word-stress. How can we overcome this problem?

92
Notice that in all words of this shape, stress falls on the first light syllable, e.g. ibir, katib, etc.

123
 In order for the first syllable to receive stress, it has to be footed. This constitutes

evidence that parsing the first syllable into a metrical foot is more important than

satisfying the requirement of FOOT-BINARITYµ. The constraint which requires the first

syllable of /Hibir/ to make up a degenerate foot by itself is PARSEσ:

Parse σ
A syllable must be footed

PARSE-σ, which forces the first light syllable of /Hibir/ to be part of a metrical foot,

must dominate FOOT-BINARITYµ. Otherwise, there would be no way to get penult

stress in /Hibir/. Consider how the first syllable of [ibir] can receive main word-stress if

FOOT-BINARITYµ is dominated by PARSE-σ:

Tableau (21)
Input: /Hibir/ TR MR NF Pσ FBµ A-F-L A-F-R
a- " (i)bir * * *
b- i.bir *!*

Although output (a) incurs a violation of FOOT-BINARITYµ, it surfaces as the

optimal form because its competitor (b) incurs two violations of PARSE-σ, one time

because, like (a), the last syllable is not footed, and a second time because the first

syllable is not footed.

Let us focus on the parsing of the first syllable of /hibir/ into a degenerate foot. In

order to have it form a metrical foot by itself, FOOT-BINARITYµ has to be violated.

PARSE-σ must dominate FOOT-BINARITYµ, so that (i)bir becomes a better parse

than Hibir. The constraint ranking in (50) above then yields the correct surface form.

124
 A point worthy of note here is that (i)bir cannot win the competion without

defeating another competitor, namely (i)(bir). Parse-σ requires that every syllable be

part of a metrical foot. If Parse-σ were undominated, (i)(bir) would then be a better

parse than (i)bir; and due to MAIN-RIGHT, (i)(bir) would surface as (i)(bir). To

avoid this unhappy conclusion, NONFINAL, which requires the last syllable to be

unfooted, must outrank PARSE-σ. That is, the need to have the last syllable unfooted is

more important than the need to have each and every foot be part of a metrical foot.

Consider Tableau (22) below:

Tableau (22)
Input: /ibir/ TR MR NF Pσ FBµ A-F-L A-F-R
a- " (i)bir * * *
b- i.bir *!*
c- (i)(bir) *! * * *

According to Tableau (22) above, output (c) fails to surface as the optimal form because

it violates NONFINAL as a higher-ranking constraint (ranked-higher than PARSEσ).

Given this relative ranking of NONFINAL and PARSE-σ, we make sure that the last

syllable is always unfooted, and thus is not a potential stress carrier.

(53) Constraint Ranking

TROCHAIC, MAIN-RIGHT, NONFINAL >> PARSEσ >> FOOT-
BINARITYµ >> ALL-FEET-LEFT >> ALL-FEET-RIGHT

Notice how this new ranking of constraints provides a uniform analysis of stress

for all the examples in (45) above. Let us demonstrate this.

First, because NONFINAL is undominated, the final syllable is left unfooted

125
 (54a) Parsing Actual words
a. L<L> ka.na
b. L<H> i.bir
c. H<L> kur.si
d. H<H> baT.Tah

Second, due to FOOT-BINARITYµ, feet are constructed over two light syllables (LL)

and over one heavy syllable (H) form left to right

(54b) Parsing Actual words

a. L<L> ka.na
b. L<H> i.bir
c. (H)<L> kur.si
d. (H)<H> baT.Tah

Third, due to PARSEσ, an unparsed syllable makes up a foot by itself

(54c) Parsing Actual words

a. (L)<L> ka.na
b. (L)<H> i.bir
c. (H)<L> kur.si
d. (H)<H> baT.Tah

Finally, due to TROCHAIC and MAIN-RIGHT, the correct stress pattern is derived

(54b) Parsing Actual words

a. (L)<L> ka.na
b. (L)<H> i.bir
c. (H)<L> kur.si
d. (H)<H> baT.Tah

The constraint ranking in (53) above, however, raises two interesting questions:

(1) the place of stress on monosyllabic words, and (2) the place of stress on final

superheavy syllables.

If NONFINAL is undominated as in (53) above, the final syllable can never

receive stress because it will never be footed. But we know that monosyllabic words

126
surface with main word-stress in Arabic, and we also know that all Arabic words that end

in a superheavy syllable receive stress on that syllable.

To show how this can be accounted for, we maintain the relative ranking of

NONFINAL and PARSEσ as in (56) above, so that we can account for why polysyllabic

words that end with light or heavy syllables do not receive stress on that final syllable,

but we also assume that there is(are) some constraint(s) that outrank(s) NONFINAL.

Again, the proposal that NONFINAL should be dominated by some higher-ranking

constraint is needed to account for two things:

(a) why monosyllabic words surface with main word-stress, and

(b) why words that end with a superheavy syllable receive main word-stress

on that syllable.

We can account for why monosyllabic words surface with main word-stress by

positing that Lx=Pr outranks NONFINAL.

Lx = Pr
Every lexical word must consist of a prosodic word.93

This basically means that every lexical word must contain at least one foot. This

constraint is violated if no foot is constructed over the word. Consider how an Arabic

monosyllabic word like /min/ surfaces with main word-stress:

Tableau (23)
Input: /min/ TR MR Lx=Pr NF Pσ FBµ A-F-L A-F-R

93
In pre-Optimality literature, this requirement is formulated as a restriction on extrametricality; that is,
extrametricality is blocked “if it would render the entire domain of the stress rules extrametrical” (see Hung
1995: 7).

127
 a- " (min) *
b- min *! *

According to Tableau (23), output (a) violates NONFINAL, and output (b) violates

Lx=Pr. Neither of them violates both NONFINAL and Lx=Pr simultaneously. If

NONFINAL dominates Lx=Pr, output (b) wins the competition; if Lx=Pr dominates

NONFINAL, output (a) wins the competition. Because output (a) is the actual output

form, Lx=Pr must dominate NONFINAL. Lx=Pr is, then, a higher-ranking constraint:

(55) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr94 >> NONFINAL >> PARSEσ >>
FOOT-BINARITYµ >> ALL-FEET-LEFT >> ALL-FEET-RIGHT
94
One may suggest here that the introduction of Lx=Pr as a higher-ranking constraint can account for why
stress terminates on the light penult on disyllabic words such as [ibir] without the need to have PARSE-σ
outrank FOOT-BINARITYµ as suggested earlier. Hence, Lx=Pr requires that at least one foot be
constructed over the prosodic word, and does not make any specifications on the form of the syllable.
Consider the following Tableau, where Lx=Pr is a high-ranking constraint:

Input: /ibir/ TR MR Lx=Pr NF PS FBµ A-F-L A-F-R

a- " (i)bir * * *
b- i.bir *! **

What this Tableau suggests is that (a) surfaces as optimal not because it incurs fewer violations on
PARSEσ than (b) as is posited in Tableau (22) above, but because it does not violate Lx=Pr. If this were
true, then we would no longer need PARSEσ to dominate FOOT-BINARITYµ. Although this may seem to
be an attractive solution, it turns out that this proposal may not work to predict the place of stress on a
penult in longer words. Consider the following example where stress falls on a light penult:
(a) [katabaha] ‘he wrote it’
Due to NONFINAL, the last syllable is not footed:
(b) kataba<ha>
The remaining part of the word is, due to FOOT-BINARITYµ, parsed as follows:
(c) (kata)ba<ha>
Notice that the penult is not footed because it is a light syllable that does not pair with another light syllable
to make up a foot. Were FOOT-BINARITYµ not outranked by PARSEσ, there would be no way to have
the penult footed; and stress, due to MAIN-RIGHT and TROCHAIC, would fall on a pre-antepenult
syllable, which of course never occurs in Arabic:
(d) *(kata)ba<ha>
The important point here is that Lx=Pr cannot force the penult of /katabaha/ to form a degenerate foot so
that it later attracts main word-stress as is the case in /Hibir/ simply because in the case of /katabha/, Lx=Pr
is already satisfied without the need to erect a foot over the penult (see d above). All in all, the proposal that
Lx=Pr is undominated cannot solve the problem of stressing a light penult in long words. For /katabaha/ to
be stressed on the penult, PARSEσ has to dominate FOOT-BINARITYµ as in the following Tableau:

128
 Having solved the problem of the place of stress in monosyllabic words, let’s turn

to the other problem that the existing ranking of constraints in (55) above still cannot

handle: The place of stress on final superheavy syllables.

In Arabic, polysyllabic words that end in superheavy syllables (σµµµ) are stressed

on that syllable:

(56)
a šan.taat
b. ka.saat, da.rast

If NONFINAL were undominated, the last syllable of each word in (56) above would be

invisible to parsing, and thus, would never receive stress. But the fact of the matter is that

all Arabic words that end in superheavy syllables must receive stress on that final

superheavy syllable.95

For polysyllabic words which end in superheavy syllables to get stress on the last

syllable, that syllable must be footed. But parsing the last syllable into a metrical foot

violates NONFINAL. Because the need to have the final superheavy syllable footed is

greater than the need to satisfy the requirement of NONFINAL, we then must allow the

violation of NONFINAL. One way to have the last syllable footed is to have PARSEσ

Input: /Hibir/ TR MR Lx=Pr NF Pσ FBµ A-F-L A-F-R

a- " (kata)(ba)ha * * ** *,**
b- (kata)baha *!* **

95
Notice that the ranking of Lx=Pr above NONFINAL does not solve the problem of final stress on words
that end in superheavy syllables. This is so because Lx=Pr requires that one foot be constructed over the
prosodic word. This condition might be satisfied without the need to construct an independent foot on the
last syllable of words ending in superheavy syllables, e.g (šan)taat. This parsing cannot result in stress on
the final superheavy syllable because that syllable is not footed. What we need is a constraint that forces
final superheavy syllables to be footed irrespective whether or not other feet are constructed over the same
prosodic word, e.g. (šan)(taat), so that main word-stress, due to MAIN-RIGHT, surfaces on that final
superheavy syllable.

129
outrank NONFINAL. However, it has already been demonstrated that NONFINAL must

outrank PARSEσ; otherwise, we cannot account for the fact that light (σµ) and heavy

syllables (σµµ) never receive stress in final position. The alternatives then are as follows:

NONFINAL outranks PARSEσ so as to account for the fact that light and heavy syllables

are stressless in final position (see 57 below); or PARSEσ outranks NONFINAL so as to

account for the fact that final superheavy syllables (σµµµ) are always stressed ( see 58):

(57) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr >> NONFINAL >> PARSEσ >>
FOOT-BINARITYµ >> ALL-FEET-LEFT >> ALL-FEET-RIGHT

(58) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, PARSEσ96 >> NONFINAL >>
FOOT-BINARITYµ >> ALL-FEET-LEFT >> ALL-FEET-RIGHT

It is suggested earlier that nonfinality is motivated on independent grounds, and it

is only blocked if it renders the whole word unfooted. That is, only Lx=Pr dominates

NONFINAL. The suggestion that PARSEσ dominate NONFINAL will create more

problems than it can solve. It seems then that we need to capitalize on the former option

where NONFINAL outranks PARSEσ. With this relative ranking, we account for why

light and heavy syllables surface stressless in final position. Meanwhile, to account for

how superheavy syllables are stressed in final position, we posit a new constraint that

forces a superheavy syllable to be footed. This new constraint has to be ranked above

NONFINAL. Satisfying the demands of this new constraint is far more compelling than

satisfying the demands of NONFINAL.

96
Notice that no ranking has been established between PARSEσ on the one hand, and MAIN-RIGHT and
LX=Pr on the other.

130
 In fact, there are at least two ways to account for how superheavy syllables (σµµµ),

in contrast to light (σµ) and heavy syllables (σµµ), receive stress in final position. One

way is to factor out NONFINAL, that is, to assume that NONFINAL, while applicable to

light and heavy syllables, is not applicable to superheavy syllables. The other alternative

is to parameterize the Weight-to-Stress Principle (WSP). That is, to assume that

NONFINAL is outranked by another constraint that requires superheavy syllables to

receive main word-stress. The remaining discussion will be devoted to demonstrating

how these two alternatives can account for the fact that superheavy syllables, in contrast

to light and heavy syllables, receive main word-stress in final position. The two

alternatives (i.e. parameterizing NONFINAL and parameterizing WSP) are potential

ways of solving the issue of how final superheavy syllables attract main word-stress, but

final light and heavy syllables do not. The decision to choose between these two

alternatives is not a trivial matter. Hence, whichever approach is adopted, there is the

issue of constraint parameterization. We need to see which approach can best account for

Arabic word stress patterns.

As for parameterizing NONFINAL, we need to make sure that final light and

heavy syllables are more invisible to parsing than superheavy syllables. There are at least

two ranked constraints:

(59) 97
NONFINAL (σµ ,σµµ)>>… >> NONFINAL (σµµµ)

This basically means that NONFINAL is more likely to apply to light and heavy syllables

than to superheavy syllables. Whereas NONFINAL(σµ ,σµµ) requires final light and heavy

97
The basic insight of (59) is found in Hammond (1999: 264) when accounting for English word-stress.

131
syllables to be unfooted, NONFINAL(σµµµ) requires final superheavy syllables to be

unfooted:

Nonfinal(σµ ,σµµ)
Final light and heavy syllable should not be footed

Nonfinal(σµµµ)
Final superheavy syllables should not be footed

As for the ranking of NONFINAL(σµ ,σµµ) and NONFINAL(σµµµ) relative to the

other set of universal constraints, crucial here is the ranking of these two parameterized

constraints relative to PARSEσ. To account for why final heavy syllables are unstressed

in final position, PARSE-σ must be dominated by NONFINAL(σµ ,σµµ); but it must

dominate NONFINAL(σµµµ) to account for why superheavy syllables are stressed in that

same position. The constraint ranking in (59) above is modified so that PARSE-σ is

interleaved:

(60)
NONFINAL (σµ ,σµµ) >> PARSE-σ >> NONFINAL (σµµµ)

Given this relative ranking of NONFINAL(σµ ,σµµ) and PARSE-σ, consider how

the final heavy syllable of /baTTah/ surfaces stressless:

Tableau (24)
Input: /baT.Tah/ Lx=Pr TR MR NF Pσ NF FBµ A-F-L A-F-R
(σµ ,σµµ) (σµµµ)
a-"(baT)Tah * *
b- (baT)(Tah) *! * *

According to Tableau (24) above, the optimal form (indicated by the pointing hand)

violates PARSE-σ, and its competitor (b) violates NONFINAL(σµ ,σµµ). It turns out that

the violation of PARSEσ is not as serious as the violation of NONFINAL(σµ ,σµµ). The

132
violation of PARSE-σ does not prevent output (a) from surfacing as the actual output

form. On the other hand, the violation of NONFINAL(σµ ,σµµ) results in a suboptimal

structure. As the constraint violated by the suboptimal form always dominates the

constraint violated by the optimal form, NONFINAL(σµ ,σµµ) must dominate PARSE-σ.

Ranking NONFINAL(σµ ,σµµ) above PARSE-σ guarantees that a final heavy syllable is

always invisible to parsing, and therefore not a possible carrier of main word-stress.

For final superheavy syllables to receive main word-stress, however, PARSE-σ

must dominate NONFINAL(σµµµ). Consider the following Tableau for /šan.taat/:

Tableau (25)
Input: /šan.taat/ Lx=Pr TR MR NF Pσ NF FBµ A-F-L A-F-R
(σµ ,σµµ) (σµµµ)
a-" (šan)(taat) * * * *
b- (šan)taat *! *

Output (b) in Tableau (25) above loses the competition because it violates PARSEσ. Its

competitor (a) wins the competition because it satisfies PARSEσ. The fact that output (a)

in Tableau (25a) violates NONFINAL(σµµµ) does not prevent it from surfacing as the

optimal form because NONFINAL(σµµµ) is a lower-ranking constraint, i.e. lower-ranked

than PARSEσ:

(61) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr >> NONFINAL(σµµ) >> PARSEσ
>> NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >>
ALL-FEET-RIGHT98

98
One may need to posit that NONFINAL(σµµµ) should dominate FOOT-BINARITY so as to allow
(šan)(taat) to defeat another competitor, namely (šan)(taa)t, where FOOT-BINARITY is not violated. This
reasoning is invalid because (šan)(taa)t already violates a higher-ranking constraint, namely
NONFINAL(σµµ):

133
 All in all, whereas NONFINAL(σµ,σµµ) outranks PARSEσ, PARSEσ outranks

NONFINAL(σµµµ). This accounts for the fact that that light and heavy syllables (σµ , σµµ)

are more likely to be invisible to footing than superheavy syllables (σµµµ).

If this analysis is to be adopted, we still need to address one further question:

What prosodic unit (e.g. foot, syllable, mora) should be unfooted (i.e., extrametrical) at

the end of the prosodic word?

In general, NONFINAL requires that some prosodic unit (e.g., mora, syllable, etc)

be unfooted:

NONFINAL
The final prosodic unit (i.e., mora, syllable, etc.) should not be footed

In the foregoing discussion, it has been assumed that it is the last syllable of the word that

should not be footed. NONFINAL is, then, formulated as follows:

Nonfinal(σ)
The final syllable should not be footed

When adopting the parameterized formulation of NONFINAL, final light and

heavy syllables surface stressless in final position because NONFINAL(σµ,σµµ) outranks

PARSE-σ (Cf. Tableau 24 above). Final superheavy syllables, on the other hand, receive

Input: /šan.taat/ Lx=Pr TR MR NF Pσ NF FBµ A-F-L A-F-R

(σµ ,σµµ) (σµµµ)
a-" (šan)(taat) * * * *
b- (šan)taat *! *
c- (šan)(taa)t *!

In spite of the fact that the constraint ranking in (61) rules out (c), this sub-optimal form still raises the issue
regarding which prosodic unit (e.g. mora, syllable, etc.), should be left unfooted. We return to this issue in
detail below.

134
main word-stress because PARSE-σ dominates NONFINAL(σµµµ) (Cf. Tableau 25

above).

However, in pre-Optimality literature, it is often assumed that the final mora is

extrametrical in Arabic (e.g. McCarthy 1979a, 1979b).99 In OT terms, what this means is

that NONFINAL should be interpreted under a moraic analysis. In what follows, we try

to show that a moraic analysis of NONFINAL cannot do the job. It creates an ordering

paradox. Precisely, a moraic analysis of NONFINAL would require that PARSE-σ

outrank FOOT-BINARITYµ to account for the stress pattern in /katabaha/, but it requires

the opposite ordering to account for the stress pattern in /baTTah/. Let us demonstrate

this ordering paradox.

Nonfinal (µ)
The final mora (µ) should not be stressed

When NONFINALµ applies to (45a-f) above, the strings are re-footed as follows:

(62)

Word Mora Count Footing Gloss

a- ra.ma a.a ra.<ma> ‘throw’ (mas.pas.)
b- i.bir i.ir (i.bi)<r> ‘ink’
c- kur.si ur.i (kur).s<i> ‘chair’
d- baT.Tah aT.ah (baT).Ta<h> ‘goose’

What NONFINAL(µ) does is render a final light syllable altogether extrametrical (e.g.

62a,c) 100 and a heavy syllable light (e.g. 62b, d).101

99
Recall from 2.1.1.2 above that, according to McCarthy (1979), a right node is stressed iff (if and only if)
it branches. What this means is that if the last mora is extrametrical, a heavy syllable would no longer
branch, and thus would not be stressed. A superheavy syllable, on the other hand, still branches even if the
last mora is considered extrametrical, and thus receives main word-stress.
100
We have seen that in order to account for the stress pattern in words like (62a), FOOT-BINARITYµ has
to be dominated by NONFINAL which in turn has to be dominated by Lx=Pr. The domination relation
between NONFINAL and FOOT-BINARITYµ has been obtained by transitivity. That is, we have argued

135
 The problem of interpreting NONFINAL under a moraic analysis arises in cases

like (62d), where a final heavy syllable is not stressed. Consider how the stress pattern of

/baTTah/ is obtained under the new analysis of NONFINALµ:

Due to NONFINAL(µ), the last mora is left unfooted:

(63a) baTTa<h>

Due to FOOT-BINARITYµ and ALL-FEET-LEFT, the remaining part of the prosodic

word is footed as follows:

(63b) (baT)Ta<h>

(63b) shows that a light syllable (i.e. Ta) to the right of the bimoraic foot is left unparsed,

resulting in a violation of PARSEσ:

Parse-σ
A syllable must be footed

Should we then allow for the parsing of this syllable into a degenerate foot as below?

(64) (baT)(Ta)<h>

that PARSE-σ must dominate FOOT-BINARITYµ, so that we can account for the stressing of a light
penult, and that PARSE-σ must be dominated by NONFINAL, so that we can account for how a final light
and heavy syllable fails to surface with main word-stress. We then conclude that NONFINAL dominates
FOOT-BINARITYµ as in the following partially ranked domination hierarchy:

Constraint Ranking
Lx=Pr >> NONFINAL >> PARSEσ >> FOOT-BINARITYµ

Input: /rama/ Lx=Pr NONFINAL PARSEσ FOOT-BINARITYµ

a- (ra) <ma> * *
b- (rama) *!
c- ra<ma> *! **

101
It also renders a final superheavy syllable (e.g. kasaat) heavy.

136
In so doing, we commit ourselves to the violation of FOOT-BINARITYµ. Consider

Tableau (26) below:

Tableau (26)
Input: /baTTah/ NONFINALµ FOOT-BINARITYµ Parse-σ
a- (baT)Ta<h> *
b- (baT)(Ta)<h> *

The two candidates, (a) and (b), each violate either PARSE-σ or FOOT-BINARITYµ. If

PARSE-σ dominates FOOT-BINARITYµ, (b) becomes the optimal output form. If the

ordering is the other way around, (a) wins the competition. To get the correct surface

form (i.e. baTTah), FOOT-BINARITYµ must dominate PARSE-σ:

Tableau (27)
Input: /baTTah/ FOOT-BINARITYµ Parse-σ
a-"(baT)Ta<h> *
b- (baT)(Ta)<h> *!

(65) Constraint ranking

FOOT-BINARITYµ >> PARSE-σ

Notice that this ranking contradicts our previous conclusion that PARSE-σ must

dominate FOOT-BINARITYµ which is needed to insure that a light penult receives main

word stress (e.g. katabaha). If we posit (65) where FOOT-BINARITYµ dominates

PARSE-σ, we wrongly predict that stress falls on the first syllable of /katabaha/:

Tableau (28)
Input: /katabaha/ FOOT-BINARITYµ PARSE-σ
a- !" (kata)ba<ha>102 *
b- (kata)(ba)<ha> *!

102
Here the final mora, not the final syllable, is meant to be unfooted. But the angled brackets enclose the
whole final syllable so as not to confuse the parse in (a) with another parse (kata)ba.h<a> where it may be
wrongly perceived that the syllable which falls to the right of the foot is a heavy syllable.

137
 To derive the correct stress pattern in /katabaha/, we need to posit that PARSE-σ

dominates FOOT-BINARITYµ:

Tableau (29)
Input: /katabaha/ PARSE-σ FOOT-BINARITYµ
a- (kata)ba<ha> *!
b- " (kata)(ba)<ha> *

(66) Constraint ranking

PARSE-σ >> FOOT-BINARITYµ

The problem then is as follows: The constraint ranking in (65) can account for the

stress pattern in words like /baTTah/, but cannot account for the stress pattern in words

like /katabaha/. The constraint ranking in (66) can do exactly the opposite: it can account

for the stress pattern in /katabaha/, but cannot account for the stress pattern in /baTTah/.

The question is then: Which ranking should we adopt?

Given the fact that ALL-FEET-LEFT dominates ALL-FEET-RIGHT in Arabic, it

is almost impossible to have a light penult surface with main word-stress without having

it make up a degenerate foot by itself (see 38 and 39 above). In other words, for stress to

terminate on a light penult in Arabic, PARSE-σ must dominate FOOT-BINARITYµ. So

the constraint ranking in (65) is difficult to motivate.

Given the fact that PARSEσ must dominate FOOT-BINARITYµ (Cf. 66 above),

one may wonder if there is still some way to get the correct surface form for /baTTah/.

Notice that when the requirement of NONFINALµ is met, the remaining part of /baTTah/

is parsed as follows:

(67) (baT)(Ta)<h>

138
If this parse were correct, we could propose that stress terminates on the syllable over

which the leftmost foot is erected because that syllable is heavy. One possible solution to

have /baTTah/ surface with penult stress is then to posit a constraint that requires the

stressing of a heavy syllable. This constraint is WSP:

WSP
Heavy syllables are stressed.

However, in order to do get penult stress on /baTTah/, we need to have WSP outrank

MAIN-RIGHT:

Tableau (30)
Input: /baTTah/ WSP MAIN-RIGHT
b-" (baT)(Ta)<h> *
c- (baT)(Ta)<h> *!

The problem with this analysis is that it predicts that a heavy syllable will be more

likely to receive main word-stress, even if it falls outside the three-syllable window, than

the head syllable of the rightmost foot of the prosodic word. Recall that in Arabic stress

falls on one of the last three syllables of the word; a light penult or antepenult is more

likely to receive stress than a heavy syllable in pre-antepenult position. The ordering of

WSP above MAIN-RIGHT does not guarantee this; rather, it yields incorrect stress

patterns whenever a heavy syllable falls outside the three-syllable window. Consider the

following example:

(68) muštarikah “a female participant”

Tableau (31)
Input: /muštarika/ Lx=Pr WSP MR Pσ FBµ
a-!"(muš)(tari)<ka> * *
b- (muš)(tari)<ka> *! *

139
 To have main word-stress fall on the light antepenult, not on the heavy pre-

antepenult, in /muštarikah/, we need to keep MAIN-RIGHT undominated:

Tableau (32)
Input: /muštarika/ Lx=Pr MR WSP Pσ FBµ
a- (muš)(tari)<ka> *! *
b- " (muš)(tari)<ka> * *

(69) Constraint ranking

Lx=Pr, MAIN-RIGHT >> WSP, PARSEσ>> FOOT-BINARITYµ,

But again, the ranking of MAIN-RIGHT above WSP creates the problem why a word

like /baTTah/ is stressed on the heavy penult:

Tableau (33)
Input: /baTTah/ Lx=Pr MR WSP Pσ FF
a-!" (baT)(Ta)<h> *! *
b- (baT)(Ta)<h> *

We again run into an ordering paradox: Ranking WSP above MAIN-RIGHT can account

for penult stress in /baTTah/, but cannot account for antepenult stress in /muštarikah/.On

the other hand, ranking WSP above MAIN-RIGHT can do exactly the opposite.

We conclude that ordering paradoxes such as these emerge only when

NONFINAL is interpreted under a moraic analysis. Given the fact that PARSEσ has to

dominate FOOT-BINARITYµ and MAIN-RIGHT is undominated, we can obtain a

uniform analysis for /baTTah/, /katabaha/ and /muštarikah/ by resorting to our original

suggestion that NONFINAL be interpreted under a syllabic analysis, i.e., the final

syllable, not just the final mora, is unfooted:

Nonfinal(σ)
The final syllable should not be footed

140
Consider how the correct surface forms for /katabaha/, /baTTah/, and /muštarika/ are

derived:

Tableau (34)
Input: /katabaha/ TR Lx=Pr MR NFσ Pσ FBµ
a- " (kata)(ba)<ha> *
b- (kata)ba<ha> *!

Tableau (35)
Input: /baTTah/ TR Lx=Pr MR NFσ Pσ FBµ
a- " (baT)Tah *
b- (baT)(Tah) *!

Tableau (36)
Input: /muštarika/ TR Lx=Pr MR NFσ Pσ FBµ
a- "(muš)(tari)<ka> *
b- (muš)(tari)<ka> *! *

This constitutes evidence that NONFINAL must be interpreted under the syllabic

analysis103.

Having shown that parameterizing NONFINALσ can account for how superheavy

syllables (in contrast to light and heavy syllables) attract stress in final position, let us

consider how parameterizing WSP can do the same job. If it turns out that it does, then

we need to choose between the two alternatives.

It has been shown in 3.3.1 above that stress in Arabic is subject to syllable weight

restrictions, where superheavy syllables are more likely to receive stress than heavy

syllables which, in turn, are more likely to receive stress than light syllables. In OT

analyses, the WEIGHT-TO-STRESS PRINCIPLE (WSP) accounts for the prominence of

heavy syllables:
103
Notice that under this analysis, we need not invoke WSP at all. The remaining constraints conspire to
yield stress on a heavy syllable when needed.

141
 WSP
Heavy syllables must be stressed.

Notice that the WSP does not differentiate between heavy and superheavy syllables as far

as word prominence is concerned. WSP only makes a two-way contrast: light syllables

vs. heavy and superheavy syllables. In parameterizing WSP, we acknowledge that

superheavy syllables in Arabic do really contrast with heavy syllables as far as

prominence is concerned. In other words, WSP involves a three-way contrast in Arabic:

superheavy syllables on the one hand, versus heavy and light syllables on the other. This

contrast is only manifest in final position in Arabic. Due to the fact that a superheavy

syllable (σµµµ) is more likely to attract main word-stress than a heavy syllable (σµµ),

WSP(σµµµ) should then outrank WSP(σµµ), which in turn outranks WSP (σµ)

(70)
WSP(σµµµ) >> WSP(σµµ) >> WSP(σµ)104

What we need to do next is to find out how this parameterization of WSP can

account for the fact that stress appears on the final syllable iff (if and only if) it is

superheavy. This requires that we find out not only the relative ranking of WSP(σµµµ)

and WSP(σµµ) with respect to each other, but also the relative ranking of WSP(σµµµ) and

WSP(σµµ) with respect to other constraints. Crucial here is the relative ranking of

WSP(σµµµ), WSP(σµµ), and NONFINALσ.

In order for superheavy syllables to receive stress, WSP(σµµµ) must outrank

NONFINALσ. Otherwise, a final superheavy syllable cannot be footed, and thus cannot

104
As final superheavy syllables are contrasted with both light and heavy syllables, only WSP(σµµ) is used
to contrast with WSP(σµµµ).

142
receive main word-stress. On the other hand, NONFINALσ must outrank WSP(σµµ), so

that a final heavy syllable is not footed, and so surfaces stressless in final position.

(71)
WSP(σµµµ) >> NONFINAL >> WSP(σµµ)

The basic insight of (72) is that while a heavy syllable can be skipped over, due to

NONFINAL, a superheavy syllable cannot be. In terms of constraint ranking, a

superheavy syllable receives stress in final position because WSP(σµµµ) outranks

NONFINAL, but a heavy syllable fails to receive stress because WSP(σµµ) is dominated

by NONFINAL. Consider the two following Tableaus:

Tableau (37)
Input: /baT.Tah/ Lx=Pr TR MR WSP NF WSP Pσ FBµ A-F- A-F-R
(σµµµ) (σµµ) L
a-"(baT)Tah * * *
b- (baT)(Tah) *! * * *

Tableau (38)
Input: /šan.taat/ Lx=Pr TR MR WSP NF WSP Pσ FBµ A-F-L A-F-R
(σµµµ) (σµµ)

a-" (šan)(taat) * * * * *
b- (šan)taat *! * *

In both cases, the correct surface form is derived. In Tableau (37), a final heavy syllable

(σµµ) fails to surface with main word-stress simply because it would otherwise violate

NONFINAL which dominates WSP(σµµ). In tableau (38), a final superheavy syllable

(σµµµ) attracts stress to satisfy WSP(σµµµ) which dominates NONFINAL.

(72) Constraint Ranking

TROCHAIC, Lx=Pr, MAIN-RIGHT, WSP(σµµµ) >> NONFINAL >>
WSP(σµµ), PARSEσ >> FOOT-BINARITYµ >> ALL-FEET-LEFT >>
ALL-FEET-RIGHT

143
 We conclude that Parametrizing WSP can also account for the fact that

superheavy syllables, unlike light and heavy syllables, attract stress in final position. The

question that should be addressed now is: Which alternative should be adopted?

Before resolving this issue, a further point is in order. When the function GEN

provides an output form that contains more than one stressed syllable, MAIN-RIGHT

guarantees that one of them is main, and the remaining ones are secondary, for implicit in

MAIN-RIGHT is the assumption that of all the feet constructed over the word, the

rightmost contains the syllable with main word-stress. The question that promptly arises

is: How does one rule out multiple stresses? For, in Arabic, a heavy syllable that does not

constitute the rightmost foot is never stressed. To clarify this point, let us consider the

following Tableau:

Tableau (39)
Input: /jaadalatu/ Lx=Pr MR WSP Pσ A-F-L A-F-R
a- (jaa)(dala)<tu> * * * *,***
b- !" (jaa)(dala)<tu> * * *,***
c- (jaa)(dala)<tu> *! * * *,***

According to Tableau (39), the unattested [jaadalatu] fares better than the actual form

[jaadalatu]. For [jaadalatu] satisfies WSP and MAIN-RIGHT simultaneously, but

[jaadalatu] violates WSP. The constraint ranking in (72) above helps to settle the conflict

between (a) and (c) in favor of (a) in Tableau (39) above, but it does not help to settle the

competition between (a) and (b). On the contrary, it wrongly predicts that the unattested

[jaadalatu] fares better than the actual surface form [jaadalatu] on the whole set of

constraints. For further clarification, consider how the unattested [maktabna] defeats the

actual form [maktabna] in Tableau (40) below:

144
 Tableau (45)
Input: /maktabna/ Lx=Pr MR WSP Pσ FBµ A-F-L A-F-R
a- (mak)(tab)<na> * * * *,**
b-!(mak)(tab)<na> * * *,**
c-(mak)(tab)<na> *! * * *,**

According to Tableau (40) above, output (c) is excluded altogether because it violates

MAIN-RIGHT, a high-ranking constraint. The competition between (a) and (b) cannot be

resolved on this basis because the unattested output (b), in addition to satisfying MAIN-

RIGHT, incurs fewer violations than the actual output form on the entire set of

constraints. Specifically, whereas the actual form violates WSP, the unattested form

satisfies it.

To resolve the conflict between the forms in (a) and (b) in Tableaus (39) and (40)

above, we once more need to posit a new constraint which is violated by the unattested

forms [maktabna] and [jaadalatu], but satisfied by the actual forms [maktabna] and

[jaadalatu]. This new constraint has to outrank WSP which is violated by the actual

output forms.

Recall from 3.3.3 above that most studies on Arabic word-stress confirm that

Arabic does not distinguish between degrees of stress. In Arabic, every word has only

one main stress, with no secondary stresses whatsoever (see Mitchell 1960 among

others). This fact is not only particular to Arabic, many languages do exactly the same

thing. To account for this tendency in an OT framework, Hammond (1999: 308) proposes

the following constraint:

*Secondary (*2)
Secondary stresses are dispreferred

145
Consider Tableaus (39) and (40) reproduced as (41) and (42) respectively, where

*SECONDARY is undominated (main stress in bold, secondary stress in italics):

Tableau (41)
Input: /jaadalatu/ Lx=Pr MR *2 WSP NF Pσ FF A-F-L A-F-R
a-" (jaa)(dala)<tu> * * * *,***
b- (jaa)(dala)<tu> *! * * *,***

Tableau (42)
Input: /maktabna/ Lx=Pr MR *2 WSP NF Pσ FF A-F-L A-F-R
a-" (mak)(tab)<na> * * * *,**
b- (mak)(tab)<na> *! * * *,**

According to Tableaus (41) and (42), although the actual forms violate WSP, they win

the competition because they do not violate *SECONDARY. Their competitors (output b

in each Tableau) lose the competition because they each violate this higher-ranking

constraint (i.e. higher-ranked than WSP).

(73) Constraint Ranking

TROCHAIC, Lx=Pr, MAIN-RIGHT, *SECONDARY >> WSP >> NONFINAL
>> PARSEσ >> FOOT-FORM >> ALL-FEET-LEFT >> ALL-FEET-RIGHT

Given the constraint ranking in (73) above, consider how the actual form can defeat its

competitors.

Tableau (43)
Input: /maktabna/ TR Lx=PR MR *2 NF Pσ FBµ A-F-L A-F-R WSP105
a- maktabna *!
b. (mak)(tab)<na> *! * * *,**
b- (mak)(tab)<na> *! * * *,**
c- maktab(na) *! * ** **
d- mak(tab)<na> *!* * * *
e- (ma)k(tab)<na> * *! * **
f-"(mak)(tab)<na> * * ** *

105
WSP is ranked very low because we will see later that we can make do without it.

146
 Tableau (43) above shows that the optimal form can defeat all other competitors

without even invoking WSP. This suggests that the former alternative (i.e. parameterizing

NONFINAL) should be adopted because it allows us to dispense with WSP altogether.

One reason why we choose to dispense with WSP (or at least rank it far down the

hierarchy) is that a heavy syllable may be stressless in Arabic. A heavy syllable that is

outside the three-syllable window is not stressed at all; a heavy syllable in

antepenultimate position is not stressed; and a heavy syllable in final position is not

stressed, either. It is only when the heavy syllable occurs in penult position that it is

stressed. Consider the following examples (heavy syllables underlined):

(74)
a. ?adwiyatuhu
b. jaadalatu 106
c. maktabah
d. mak tabna

In terms of constraint ranking, this amounts to saying that WSP is ranked low enough in

the domination hierarchy that its impact is not even noticeable. The fact that a heavy

syllable is stressed in penultimate position, and a superheavy syllable is stressed in final

position, falls out as a result of the interaction of the remaining set of constraints. The

interaction of NONFINAL(σµµ), PARSEσ, FOOT-BINARIYµ, TROCHAIC, and MAIN-

RIGHT guarantees that a heavy syllable in penult position gets main word-stress (e.g.

(baT)Tah). The interaction of PARSE-σ, NONFINAL(σµµµ), and TROCHAIC, MAIN-

RIGHT guarantees that a superheavy syllable gets main word-stress (e.g. (ka)(saat)).

106
In some dialects of Arabic, e.g. Urban Hijazi Arabic (Abaalkhail 1998: 220-221), stress goes to the
heavy antepenult.

147
 Unlike possibly all previous treatments of Arabic word-stress, we assume that

stressing a syllable in penultimate or in final position in Arabic is not related to its

weight, but to whether or not it gets parsed. As a heavy syllable is invisible to parsing in

final position, it escapes stress; but as a superheavy syllable is more visible to parsing in

that position, it ultimately surfaces stressed. By the same token, a heavy syllable in

penultimate position surfaces stressed because it gets parsed. The more visible to parsing

the syllable is, the more likely it surfaces stressed. Since MAIN-RIGHT is undominated

in Arabic, our analysis entails that the rightmost foot is the head of the prosodic word in

Arabic, irrespective of the intrinsic weight of the syllable(s) over which that foot is

erected. This top-most ranking of MAIN-RIGHT is supported by the fact that a light

penult and a light antepenult surface with main word-stress even if they are preceded by a

heavy syllable (e.g. jaadalatu, maktabah).

A uniform analysis of the place of main stress in Arabic can be accounted for

without invoking WSP at all. The other constraints conspire to yield the correct stress

patterns whenever a heavy syllable surfaces with main word-stress. This is so because

FOOT-BINARITY, interpreted under a moraic analysis in Arabic, makes it impossible to

have syllables of varying weight pair together to make up one single foot. WSP is

defintely needed when FOOT-BINARITY is interpreted under a syllabic analysis. In this

latter case, when the foot is constructed over two syllables of different weight, i.e. (LH)

or (HL), WSP must conspire with some alignment constraint (e.g. IAMBIC,

TROCHAIC) to yield the correct stress patterns107. In a trochaic system, the problem

107
Coversely, in an iambic system the problem arises when the left leg of the foot is a heavy syllable that
surfaces stressed (HL). In this case, IAMBIC must be ranked above FOOT-BINARITYσ.

148
arises when the right leg of the foot is a heavy syllable that surfaces stressed, i.e. (LH). In

this case, we need to invoke a constraint such as WSP to account for the prominence of

the heavy syllable. To do just that, we need to allow the violation of FOOT-

BINARITYσ. Specifically, we need WSP to dominate FOOT-BINARITYσ, so that a

string like LH is footed as L(H) instead of (LH). This problem does not arise under the

moraic interpretation of FOOT-BINARITY in Arabic because a heavy syllable, due to

FOOT-BINARITYµ, makes up a foot by itself. All in all, the need for WSP diminishes

because there are no feet of the form (LH) or (HL).108

(75) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, *2nd >> NONFINAL(σµµ) >>
PARSEσ >> NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-
LEFT >> ALL-FEET-RIGHT >>…>> WSP

With the constraint ranking in (75) above, we conclude the discussion of how the

set of universal and violable constraints which interact to yield main word-stress are

ranked in Arabic. However, before we close up this section, a further point is in order. In

3.2.1 above, the whole set of universal and violable constraints is surveyed. One may

wonder why some of the constraints that are introduced in 3.2.1 above (namely, WORD-

FOOT-RIGHT, WORD-FOOT-LEFT, NOLAPSE and RHYTHM) are not invoked to

account for the place of main word-stress in Arabic, though the assumption that is

adopted throughout is that all these constraints are literally present in each and every

108
It will be shown later that WSP is needed in English for this reason. In English, where FOOT-
BINARITY is interpreted under a syllabic analysis, feet are trochaic. However, in a word like ‘agenda’, it
turns out that stress goes to the heavy penult. The problem is clearer when we know that in English the final
syllable is, due to NONFINAL, unfooted, so /agenda/ is, due to FOOT-BINARITYσ and NONFINALσ,
parsed as (agen)<da>. If this parse were optimal, stress would, due to TROCHAIC, go to the antepenult.
To avoid this unhappy conclusion, a constraint like WSP is badly needed. When WSP is ranked higher than
FOOT-BINARITYσ, a(gen)<da> becomes a better parse than (agen)da, and stress, due to MAIN-RIGHT
terminates correctly on the heavy penult, i.e. a(gen)da.

149
language. It suffices here to say that those constraints are ranked low enough to the extent

that their impact is not even noticeable in Arabic. They are dominated by the whole set of

constraint; even if some output forms violate these constraints, they still have the

opportunity to surface as the actual output form as long as they satisfy higher-ranking

constraints.

(76) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, *2nd >> NONFINAL(σµµ) >> PARSEσ >>
NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >> ALL-FEET-
RIGHT >>…>> WSP, WORD-FOOT-RIGHT, WORD-FOOT-LEFT, NOLAPSE
and RHYTHM

3.4.1 Conclusion

In order to account for main word-stress in Arabic, the following constraint ranking is put

forward:

(77) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, *2nd >> NONFINAL(σµµ) >> PARSEσ >>
NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >> ALL-FEET-
RIGHT >>109

This ranking has been obtained by advancing the following following generalizations:

I. TROCHAIC, MAIN-RIGHT, Lx=Pr, and *SECONDARY are undominated

in Arabic110:

109
The other constraints are left out because they are ranked low enough that their impact is not even
noticeable in Arabic.
110
Under the analysis that WSP is parameterized, the following generalizations hold:
a. WSP(σµµµ) must dominate NONFINALσ. (šan)(taat) is a better parse than (šan)taat.
b. WSP(σµµµ) must dominate FOOT-BINARITYµ. (šan)(taat) is a better parse than
(šan)(taa)t
c. MAIN-RIGHT must dominate WSP. (jaa)(dala)<tu> is a better parse than
(jaa)(dala)<tu>
d. *SECONDARY must dominate WSP. (jaa)(dala)<tu> is a better parse than
(jaa)(dala)<tu>

150
 a. TROCHAIC must be undominated. (jaa)(dala)<tu> is a better parse
than(jaa)(dala)<tu>.
b. MAIN-RIGHT must be undominated. (jaa)(dala)<tu> is a better parse
than (jaa)(dala)<tu>
c. Lx=Pr must dominate NONFINALσ. (min) is a better parse than min.
d. *SECONDARY must be undominated. (jaa)(dala)<tu> is a better parse
than (jaa)(dala)<tu>

II. The other constraints, although they are all dominated by higher ranking
constraints, they are ranked relative to each other:

a. NONFINAL(σµµ) must dominate WSP. (baT)Tah is a better parse than

(baT)(Tah).
b. NONFINAL(σµµ) must dominate FOOT-BINARITY-µ. (i)bir is a better
parse than i(bir), and (ra)ma is a better parse than (rama).
c. NONFINAL(σµµ) must dominate PARSE-σ. (i)bir is a better parse than
(i)(bir), and (ka.ta)ba is a better parse than (ka.ta)(ba)
d. PARSEσ must dominate NONFINAL(σµµµ). (šan)(taat) is a better parse
than (šan)taat.
e. PARSE-σ must dominate FOOT-BINARITY. (i)<bir> is a better parse
than Hi<bir>, and (kata)(ba)<ha> is a better parse than (kata)ba<ha>.
f. ALL-FEET-LEFT must dominate ALL-FEET-RIGHT.
(ad)(wiya)(tu)<hu> is a better parse than (ad)(wi)(yatu)<hu>

151
3.5 Constraint interaction in English111

In the previous section, the facts and generalizations regarding stress assignment in

Arabic from an optimality-theoretic standpoint are laid out. The goal has been to find out

how the universal and violable constraints interact to yield the correct stress pattern in

that language. In this section, I try to do the same for English. However, I should make

clear that a comprehensive treatment of the English stress system from this standpoint is

beyond the scope of this research project for at least two reasons.

First, since the primary goal of this research project is to see how the grammar of

stress in L1 influences the acquisition of the stress system in L2, our focus should be on

those aspects of L1 that transfer to L2. Given the constraint ranking in L1, we can make

predictions concerning the performance of the learners in L2. In OT terms, we need to

find out how the ranking of constraints in L1 transfers to L2. As for the English stress

system, what is important to us is to see how the basic facts and generalizations regarding

English stress can be utilized in the context of second language acquisition. Specifically,

we need to develop a constraint ranking particular to English to show what native

speakers of Arabic need to do to acquire the English stress system.

Second, the stress system in English exhibits irregularities. As the bulk of the

discussion assumes a uniform analysis of stress, we mainly attend to cases that exhibit

distributional regularities. This comprises the distribution of main word-stress in

monomorphemic words. Non-uniform cases such as stress in polymorphemic words and

111
The basic facts and generalizations laid out in this section are entirely based upon Hammond (1999) and
Pater (1995).

152
the place of secondary stresses are “extremely murky and exception-ridden” in English

(Hammond 1999: 192), and so will only be broadly sketched out.

Our main goal here is to figure out the basic facts and generalizations regarding

the distribution of main word-stress in monomorphemic words in English. In 3.5.1 below,

we try to develop a uniform analysis of stress in monomorphemic words from an

optimality-theoretic standpoint. In 3.5.2 and 3.5.3, the place of stress in polymorphemic

words and the basic facts regarding the distribution of secondary stresses will be broadly

sketched out.

3.5.1 Stress in monomorphemic words

In monomorphemic words, not only is primary stress almost always rightmost, but it also

falls on one of the last three syllables (i.e. ultima, penult and antepenult). In other words,

stress in monomorphemic words does not fall on a pre-antepenult syllable. According to

Hammond (1999: 192-330), the rules governing the distribution of stress among the last

three syllables in monomorphemic content words are as follows:

(78)

I. The ultima is stressed under the following condition(s)

a. iff (if and only if) it is at least bimoraic

II. The penult is stressed under the following condition(s)

a. the ultima does not contain a long vowel
b. the ultima does not end in a consonant cluster112

112
It may end with a consonant cluster in the case of nouns but only if the second consonant of the cluster
is a coronal (a coronal is nonmoraic)

153
 III. The antepenult is stressed under the following condition(s)
a. the ultima must be a syllabic [r] or [n] in the case of verbs and
adjectives, and cannot be closed by a cluster in the case of nouns.
b. the penult cannot be bimoraic (i.e., a heavy syllable cannot be skipped
over in this position) 113

It is quite patent that these generalizations depend crucially on the notion of the mora.

The most relevant question that has to be addressed here is: What constitutes a mora in

English? In the following section, we try to investigate this issue in order to see how the

notion of the mora can be utilized for stress placement purposes.

3.5.1.a Mora count

Syllable weight in English turns out to be crucial for stress assignment, so the question

arises as to what makes a syllable light, heavy, or even superheavy. Hammond (1999:

103-148) suggests that vowels, diphthongs and coda consonants be divided into subsets

depending on how many moras each contributes to the weight of the syllable. Some

subsets of phonemes contribute one mora; some of them optionally contribute one mora;

some of them contribute two moras; some of them contribute three moras; and some of

them contribute none. The following chart is taken from Hammond (1999:137):

(79)
Intrinsic mora count
Lax vowels µ
Tense vowels µµ
[ay,yu] µµ
[aw, y]114 µµµ

113
The penult can be bimoraic and still skipped over in the following two cases (i) it contains [i, u, e, o]
prevocalically, e.g., Bedouin, or (ii) the ultima contains a syllabic [r], e.g. cylinder, or [i] burgundy.

154
 coronals115 (µ)
Noncoronals µ
[ž, ŋ] µµ
[, r] Ø
[] Ø

In addition to proposing the mora counts in (79) above, Hammond puts forward the

following mora assignment principles that refer to neighboring segments (the designated

sound is underlined):

(80)

1µ/liquid/-C a liquid contributes one mora pre-consonantally, e.g. bulk

(Hammond 1999: 147)

PRE-MORA a consonant must contribute precisely one mora if it precedes a

moraic consonant, e.g. ask (Hammond 1999: 141)

POST-MORA a coronal is moraic if it follows a noncoronal consonant, e.g. ax

[æks] (Hammond 1999: 141)

PRE stressless preconsonantal vowels are nonmoraic, e.g. gallop

(Hammond 1999: 207)

According to this proposal, coronals contribute a mora under specific circumstances.

Hammond (1999: 141) argues that coronals fail to contribute a mora in the following

cases:

114
Hammond (1999: 135-6) imposes a trimoraic limit in that no single phoneme contributes more than
three moras to the weight of the syllable in which it occurs, and that no syllable contains more than three
moras.
115
Coronals are sounds produced with the front of the tongue (the corona) as the main articulator. This
class includes dental and alveolar sounds [θ, , t, d, s, z, n, l, r] (Hammond 1999: 4).

155
 (81) 116

a. if preceded by [aw, y], e.g. pout [pawt], simply because these diphthongs are
trimoraic(µµµ)
b. if preceded by a tense vowel (that is bimoraic) or diphthong (that is bimoraic) plus
a consonant, e.g. pounce [pawns].
c. the second member in the sequence lax vowel + C+ COR + COR, e.g. text [tєkst]

The difference between [pawns] in (81b) and [tekst] in (81c) above as far as coronals are

concerned is that in [pawns] neither [n] nor [s] contributes moras, but in [tekst] the first

coronal contributes a mora, but the second does not.117

3.5.1.b The Analysis

The overall purpose of dividing vowels, diphthongs and coda consonants into subsets

according to mora count is to distinguish stressed and unstressed syllables. Stressed

syllables are at least bimoraic; syllables that fall to the right of the syllable which carries

main stress are at most monomoraic (Hammond 1999: 202). However, following the

usual tack of Optimality Theory, we assume that all syllables (not just stressed syllables)

are bimoraic in English:

116
According to Hammond (1999: 135-6), the overall purpose of this distribution is to maintain a minimum
and maximum size restriction. For example, lax vowels do not occur word-finally because a word must be
at least bimoraic. Words like *[b] are ill-formed. Likewise, only a coronal consonant can close a syllable
whose nucleus is [aw] or [y] because syllables are maximally trimoraic.
117
The /n/ in [pawns] does not contribute a mora because of (81a) above
The /s/ in [pawns] does not contribute a mora because of POST-MORA in (80) above
The /s/ in [tekst] contributes a mora because of POST-MORA in (80) above
The /t/ in [tekst] does not contribute a mora because of (81c) above.

156
 Bimoraic118
*σ

The fact that some syllables are not bimoraic on the surface can be accounted for by

assuming that BIMORAIC is dominated by some higher-ranking constraint(s). For

example, Hammond (1999: 249) argues that only a schwa is allowed as the nucleus of

final closed syllables (e.g. hammock, dollop, gamut, etc.). With the proposal that the

schwa is moraless (see 79 above), only monomoraic and zero-moraic syllables can occur

to the right of any stressed syllable. The key constraint in Hammond’s analysis is

REDUCTION, that forces the nucleus of a closed syllable to reduce (and thereby be

unstressed):

Reduction119
* vc]σ

For a final closed syllable to surface stressless, REDUCTION must outrank

BIMORAICITY. REDUCTION forces the vowel occupying the nucleus of the syllable to

reduce to schwa. This basically means that the final closed syllable in [hammk] counts

as one mora: the /k/ contributes a mora, but the // does not (see 79 above), and so we

obtain the right result as in Tableau (44) below:

118
Hammond’s formulation of this constraint is:
All syllables are at least bimoraic (Hammond 1999: 207)
119
Hammond’s formulation of this constraint is:
Stressless preconsonantal vowels are nonmoraic (Hammond 1999: 207).

157
 Tableau (44)
Input: /hammock/ RED BIM
a- hammock120 *!
b-" hammck *

As expected, (a) does not surface as optimal because it violates the higher-ranking

constraint REDUCTION. Output (b), on the other hand, wins the competition because it

satisfies REDUCTION. The fact that (b) violates BIMORAIC does not prevent it from

winning the competition. This should establish the domination relation between

REDUCTION and BIMORAIC:

(82) Constraint Ranking

REDUCTION >> BIMORAIC

In terms of mora count, Hammond (1999: 201) proposes that “penultimate stress in nouns

is possible only if the word-final coda contributes no more than one mora”. This is

consistent with the proposal stated above that only monomoraic and zero-moraic

syllables can occur to the right of a stressed syllable, the penult in this case.

If we adopt this analysis, we commit ourselves to the claim that, due to

REDUCTION, every final closed syllable surfaces stressless. This is not a happy

conclusion because we have words like [bassinet] where the final closed syllable is

stressed. One way to solve this problem is to propose that REDUCTION is dominated by

some higher-ranking constraint. Hammond (1999:269) adopts the solution (first outlined

in Halle and Vergnaud 1987; Hammond 1989a, 1989b) that words like [bassinet] are

exceptional cases that are accented underlyingly, and so escape REDUCTION. In other

120
For the time being, we ignore how the word is footed.

158
words, a syllable that would otherwise surface as monomoraic or zero-moraic is accented

underlyingly, and so surfaces as bimoraic. Hammond calls this constraint FAITH121:

Faith
Accented elements are stressed

Consider Tableau (45) below (the accented syllable is in bold):

Tableau (45)
Input: /bassinet/ FA RED BIMO
a-" (bassi)(net) *
b- (bassi)(nt) *! *

According to Tableau (45) above, output (a) violates REDUCTION, and output (b)

violates FAITH. The fact that /bassinet/ surfaces with ultimate stress suggests that

FAITH should dominate REDUCTION. Output (a) surfaces as optimal because, even

though it violates REDUCTION, it satisfies FAITH. This constitutes evidence for the

relative ranking of FAITH and REDUCTION:

(83) Constraint Ranking

FAITH >> REDUCTION

The constraints introduced so far are: BIMORAIC, REDUCTION and FAITH. A

ranking of these constraints relative to each other is established between REDUCTION

and BIMORAICITY on the one hand and FAITH and REDUCTION on the other. It is

shown that REDUCTION dominates BIMORAIC, but in turn is dominated by FAITH.

By making use of transitivity (which is crucial in establishing ordering relations among

conflicting constraints in OT analysis), we conclude that FAITH dominates

121
This constraint is a Faithfulness constraint; it is undominated in English. What this means is that in
English, Faithfulness constraints dominate Markedness constraints. In Arabic, however, it has been shown
that markedness constraints are undominated. It will turn out that this is one of the basic differences
between English and Arabic as far as main word-stress patterns are concerned. We will talk about this point
in detail later.

159
BIMORACITY, too. The partially stratified hierarchy reached so far looks like (84)

below:

(84) Constraint Ranking

FAITH >> REDUCTION >> BIMORAIC

Let us introduce some other constraints that are needed to derive the correct stress

patterns in English monomorphemic words. First of all, every content word must receive

at least one stress. This requirement can be captured by assuming that Lx=Pr is

undominated in English (where Lx stands for lexical word, and Pr for prosodic word):

Lx=Pr
Every lexical word must be a prosodic word

According to the prosodic hierarchy in (85) below, there is a minimum size

restriction on words. Every higher-level prosodic category must have the next available

lower-level prosodic category as its head:

(85)
PW = Prosodic Word

F = Foot

σ = Syllable

µ = Mora

Accordingly, the next available prosodic category to function as the head of the prosodic

word (PW) is the foot (F). Without a foot constructed over it, the prosodic word surfaces

headless, i.e., with no stress whatsoever. Therefore, each content word must minimally

contain a foot. The constraint that defines the form of the foot is FOOT-BINARITY

formulated below:

160
 Foot-Binarity
Feet are binary under syllabic or moraic analysis (i.e. consist of exactly two
syllables or two moras)

Since English, like Arabic, is a quantity-sensitive language as far as word prominence is

concerned, we propose that the constraint that defines the form of the foot (i.e. Foot-

Binarity) should be interpreted under a moraic analysis.

Foot-Binarity-µ
Feet are binary under moraic analysis (i.e. consist of exactly two moras)

Under this analysis, any foot that consists of one mora (µ), or more than two moras (e.g.

(µµµ)) constitutes a violation of FOOT-BINARITYµ. Only two light syllables (σµσµ) or

one heavy syllable (σµµ) could make up an optimal foot in English.

There is, however, one problem in interpreting FOOT-BINARITY under a moraic

analysis in English. We have asserted earlier that all English syllables must satisfy

BIMORAIC, i.e. must be at least bimoraic. If this is true, then each and every syllable

makes up a foot by itself, vacuously satisfying FOOT-BINARITY-µ. What this means is

that no new insight would be gained by proposing the notion of the foot because each

syllable would make up a foot by itself. To avoid this conclusion, we need to adopt the

proposal that FOOT-BINARITY should be interpreted under a syllabic analysis:

Foot-Binarity-σ
Feet are binary under syllabic analysis (i.e. each consists of exactly two
syllables)122

Under this analysis, the foot is independently motivated because no other prosodic unit

can do its job. Consider Tableau (46) below:

122
For an alternative view, see Lee (1999: 39).

161
Tableau (46)
Input: /America/ FOOT-BINARITY-σ
a- A(me)rica *!
b- "A(meri)ca

According to Tableau (46), (b) is a better parse than (a) because it satisfies FOOT-

BINARITY-σ, which requires that every foot consist of two syllables (σσ), irrespective

of the number of moras each syllable contributes.

As for the ranking of FOOT-BINARITY-σ relative to the other constraints, it

should be pointed out that if FOOT-BINARITY-σ were never violated, output (c) in

Tableau (47) below would wrongly be the optimal parse of /America/:

Tableau (47)
Input: /America/ FOOT-BINARITY-σ
a- Ameri(ca) *!
b- A(me)(rica) *!
c- (Ame)(rica)
d- (A)(meri)ca *!
ETC

In fact, none of the candidates in Tableau (47) above wins the competition. The optimal

parse for /America/ is A(meri)ca. Let us see how this can happen.

A possible constraint that outranks FOOT-BINARITY is NONFINALσ, a

constraint that requires the last syllable of the lexical word to remain unfooted. In order to

parse /America/ into metrical feet, the first step is to render the last syllable unfooted as

follows:

(86) Ameri<ca>

162
Once the last syllable is unfooted, then the remaining part of the word can be parsed into

binary feet as below:

(87)
a. A(meri)ca
b. (Ame)rica

The only difference between (87a) and (87b) is related to the directionality of footing.

(87a) is the optimal parse if syllables are grouped into binary feet from left to right; (87b)

is optimal if it is the other way around, i.e., from right to left. We will attend to this issue

later. For the time being, what is important is the fact that NONFINALσ must dominate

FOOT-BINARITYσ:

(88) NONFINALσ >> FOOT-BINARITYσ

Incorporating (88) into (84) above yields (89) below:

(89) Constraint Ranking

Lx=Pr, FAITH, NONFINALσ123 >> FOOT-BINARITY-σ124,
REDUCTION >> BIMORAIC

However, it should be made clear that NONFINALσ itself is not top-ranked in English,

either. Otherwise, a word like /hat/ would be entirely unfooted125. To avoid this unhappy

conclusion, we need to have Lx=Pr dominate NONFINALσ:

(90) Constraint Ranking

Lx=Pr, FAITH >> NONFINALσ, REDUCTION >> FOOT-BINARITY-σ,
BIMORAIC

123
Until now, NONFINAL is undominated, so it is top-ranked. If it turns out that it is dominated by some
other constraint, it will be demoted one step below the constraint which dominates it.
124
Notice that FOOT-BINARITY-σ is demoted one step below NONFINAL. There is no need to demote it
to a lower position unless it turns out that it is dominated by some constraint(s) that occupy a lower position
in the hierarchy. In other words, demotion must be minimal.
125
Rendering the entire word extrametrical (i.e. unfooted) is a violation of the nonexhaustivity restriction
on extrametricality (see Hung 1995: 7).

163
Consider Tableau (48) below:126

Tableau (48)
Input: /hæt/ Lx=Pr NONFINALσ FOOT-BINARITYσ
a- hæt *!
b- (hæt) * *

The reason why output (a) fails to win the competition is that it fails to comply

with the requirement that at least one foot be constructed over every word. If this

violation were permissible, one would expect some lexical words to surface with no

stress whatsoever. As each and every English content word must be stressed, a violation

of Lx=Pr is then fatal. According to Tableau (48) above, the violation of NONFINALσ

and FOOT-BINARITYσ is inevitable so as to avoid violation of Lx=Pr. This is done on

the principle that the constraint(s) violated by the suboptimal structure (output a in this

case) must dominate all constraints violated by the optimal structure (output b). Lx=Pr

must dominate both NONFINALσ and FOOT-BINARITYσ. The requirement to have at

least one foot constructed over the prosodic word is more compelling than the

requirement to have the last syllable unfooted or to have a foot of a certain type.

Another constraint that must dominate NONFINAL is FAITH. The fact that a

final closed syllable can be stressed (i.e. bassinet) due to FAITH tells us that FAITH

must outrank NONFINAL; otherwise, that syllable can never be part of a metrical foot,

and thus can never be stressed. Consider Tableau (49) where FAITH and NONFINAL are

still unranked relative to each other:

126
Notice also that output (b) in Tableau (48) above, which wins the competition, violates FOOT-
BINARITY-σ. After all, /hat/ is a monosyllabic word.

164
Tableau (49)
Input: /bassinet/ FAITH NONFINALσ
a- (bassi)net *
b- (bassi)(net) *

Output (a) violates FAITH, and output (b) violates NONFINALσ. To get [bassinet], we

need to insure that FAITH dominates NONFINALσ:

Tableau (50)
Input: /bassinet/ FAITH127 NONFINALσ
a- (bassi)net *!
b-" (bassi)(net) *

The domination hierarchy in (90) correctly shows that FAITH dominates REDUCTION

which in turn dominates BIMORAIC.

In a word like /America/, unfooting the last syllable does not render the entire

word extrametrical because at least one foot can be constructed over the remaining

syllables. Consider Tableau (51) below:

Tableau (51)
Input: /America/ Lx=Pr NONFINALσ
a- A(meri)(ca) *!
b- "A(meri)ca

Output (b) is a better parse than (a) simply because (b), in addition to satisfying Lx=Pr,

satisfies NONFINALσ. The question arises as to how main word-stress terminates on the

antepenult in /America/.

127
It should be pointed out that any output form that satisfies FAITH trivially satisfies Lx=Pr by virtue of
the fact that at least one foot is constructed over the prosodic word. FAITH and Lx=Pr are undominated in
English.

165
 First of all, it should be pointed out that output (b) in Tableau (51) above is not

the only parse of /America/ that satisfies NONFINALσ. There are other parses that can

be equally competitive. /America/ can be parsed as (Ame)rica, A(meri)ca, (A)(meri)ca,

(Ame)(ri)ca, (Ame)rica, etc. Given the set of constraints introduced so far, it is unclear

which of these is the optimal parse. Other constraints need to come into play.

We need, for example, a constraint that forces syllables to be footed. To this end,

PARSE-σ is put forward:

Parse-σ
Syllables must be footed

This constraint favors footed output to unfooted output. As for the ranking of PARSEσ

relative to other constraints, recall that we have already suggested that the final syllable

in polysyllabic words, due to NONFINALσ, should not be footed, and that feet

constructed over polysyllabic words, due to FOOT-BINARITYσ, should be binary. What

this means is that both NONFINAL and FOOT-BINARITYσ must dominate PARSE-σ.

NONFINAL must dominate PARSE-σ, so as to allow a final syllable in polysyllabic

words to surface stressless. A(meri)ca is then a better parse than A(meri)(ca) or any other

parse where the final syllable is part of a metrical foot. In other words, it is not a

necessary condition that syllables be exhaustively parsed into feet. We assume that some

syllables may surface unfooted.

FOOT-BINARITYσ must also dominate PARSEσ so as to allow for some

syllables, other than the final syllable, to surface unfooted. (A)(meri)ca and (Ame)(ri)ca

are then ruled out because they violate FOOT-BINARITYσ. The reason why we may

want our domination hierarchy to rule out these two parses is that stress in (Ame)(ri)ca,

166
for example, would surface on the penult128- a state of affairs that we want to avoid

anyway.

(91) Constraint Ranking

Lx=Pr, FAITH >> NONFINALσ, REDUCTION >> FOOT-BINARITYσ,
BIMORAIC >> PARSEσ129

As for /America/, once NONFINALσ and FOOT-BINARITYσ are satisfied, the

competition for optimality is limited to A(meri)ca and (Ame)rica. Consider the following

Tableau:

Tableau (52)
Input: /America/ Lx=Pr Nonfinalσ Foot-Binarityσ Parseσ
a- A(meri)ca **
b- (Ame)rica **
c- (A)(meri)ca *! *
d- (Ame)(ri)ca *! *
e- Ameri(ca) *! * ***
f- America *! ****

According to Tableau (66) above, outputs (e) and (f) are excluded from the competition

because they violate the higher ranking constraints Lx=Pr and NONFINALσ,

respectively. Outputs (c) and (d) are, too, ruled out because they violate FOOT-

BINARITYσ. Outputs (a) and (b) fare equally on all the constraints introduced so far:

They do not incur any violations of Lx=Pr, NONFINALσ or FOOT-BINARITYσ; and

they each incur two violations of PARSEσ.

128
If FOOT-BINARITY-σ were dominated by PARSE-σ, stress would surface on the penult in (Ame)(ri)ca
because MAIN-RIGHT (which requires the rightmost foot to be the head of the prosodic word) is
undominated in English. This idea will be discussed in detail later.
129
If the relative ranking of FOOT-BINARITY-σ and PARSE-σ were correct, one may wonder if
degenerate feet (i.e. feet that consist of one syllable) are sanctioned at all. We argue that this ranking is
correct and that degenerate feet are sanctioned by higher-ranking constraints such as Lx=Pr (e.g. hat),
FAITH (e.g. bassinet) and WSP (e.g. agenda).

167
 To decide between (a) and (b), we still need one more constraint. Recall that in

English stress in monomorphemic words does not fall on a pre-antepenult syllable. This

suggests that feet should be aligned with the right edge of the word in English.130

All-Feet-Right
Align each foot with the word on the right edge.131

Let us see how this works for /America/:

Tableau (53)
Input: /America/ Lx=Pr Nonfinalσ Foot-Binarityσ A-F-R Parseσ
a- A(meri)ca * **
b- (Ame)rica *!* **
c- (A)(meri)ca *! *,*** *
d- (Ame)(ri)ca *! *,** *
e- Ameri(ca) *! * ***
f- America *! ****

130
Notice that in Arabic, too, stress does not fall on a pre-antepenult syllable; yet A-F-L is said to dominate
A-F-R. This is so because in English no counting between the stressed syllable and a preceding heavy
syllable or word boundary in terms of light syllables is involved. In Arabic, stressing the penult or the
antepenult in long words requires that an independent foot separate the stressed syllable from the preceding
heavy syllable (which makes up a foot by itself due to the fact that Arabic feet are moraic trochees) or the
word boundary. In case of words consisting of three light syllables (CVCV)CV, the last syllable remains
unfooted due to NONFINAL; and stress, due to TROCHAIC, goes to the antepenult which is the head of
the foot. In the case of four light syllables (CVCV)(CV)CV, the last syllable is unfooted due to
NONFINAL; stress, due to MAIN-RIGHT, goes to the penult (notice that it is the one that is separated
from the word boundary by an even number of light syllables). This conclusion cannot be reached without
having a left-to-right parsing where the first and second syllable make up one foot and the third syllable
makes up a degenerate foot by itself (which will, due to MAIN-RIGHT, be the one that carries primary
stress). In words consisting of four light syllables, if syllables are parsed into metrical feet on a right-to-left
basis, the second syllable and third syllable will make up one foot and the first syllable will make up a
degenerate foot by itself (CV)(CVCV)CV, and stress will, due to TROCHAIC, go to the antepenult
(CV)(CVCV)CV-- which is not a happy conclusion, of course. In the case of a preceding heavy syllable,
that syllable will act like a word boundary and footing always starts anew after that syllable. In Arabic,
then, there must always be an independent foot consisting of two light syllables that separates the stressed
syllable (be it the penult or the antepenult) from the left edge of the word or a preceding heavy syllable.
131
Any foot that is not aligned with the rightmost edge of the word constitutes a violation to ALL-FEET-
RIGHT. The number of violations is important; violations of ALL-FEET-RIGHT are counted by the total
number of syllables separating that foot from the right edge of the word.

168
Due to ALL-FEET-RIGHT, the competition between (a) and (b) is now resolved in favor

of (a). Whereas (a) incurs one violation of ALL-FEET-RIGHT, its competitor (b) incurs

two violations.

The issue regarding the best parsing of /America/ is, thus, resolved by exploiting

ALL-FEET-RIGHT. Notice that this analysis requires, too, the ranking of NONFINAL

(which is already dominated by Lx=Pr) above ALL-FEET-RIGHT. Otherwise,

(Ame)(rica) would be a better parse than A(meri)ca. The ranking of constraints

introduced so far is as in (92) below:

(92) Constraint Ranking

Lx=Pr, FAITH >> NONFINAL, REDUCTION >> FOOT-BINARITY-σ,
BIMORAIC, ALL-FEET-RIGHT >> PARSE-σ

With this domination hierarchy, let’s see now why A(meri)ca is a better parse than

(Ame)(rica).

To get the correct output (with antepenultimate stress), we need to decide on two

things upfront: (1) whether the head foot (if there is more than one) is aligned with right

or left edge of the word, and (2) whether the head syllable of each foot is aligned with the

right or left edge of the foot. As for the head foot, two constraints are posited:

Main-Right
Align the head-foot with the word on the right edge
Main-Left
Align the head-foot with the word on the left edge

As for the head syllable, two other constraints are posited:

Trochaic
Align the head-syllable with its foot on the right edge
Iambic
Align the head-syllable with its foot on the left-edge

169
 The fact that main word-stress in English tends to occur as close to the right as

possible tells us that feet are aligned with the right edge of the word. If this were true,

then the optimal parse for /America/ should never be (Ame)(rica); otherwise stress would,

due to MAIN-RIGHT, always terminate on the ultimate or the penult, the two syllables

over which the rightmost foot is constructed. However, if NONFINALσ dominates ALL-

FEET-RIGHT (Cf. 92 above), the requirement to have the last syllable unfooted becomes

stronger than the requirement to align all feet, especially the rightmost one, with the right

edge of the word. In this case, the place of stress in /America/ can be accounted for

because the rightmost foot which, due to MAIN-RIGHT, is the head foot is now

constructed over the penult and the antepenult. In other words, once the final syllable is

considered invisible to parsing, there remain three candidate forms, namely (Ame)(ri)ca,

(A)(meri)ca, and A(meri)ca.

As for (Ame)(ri)ca and (A)(meri)ca, we have already suggested that these two

parses are ruled out by the proposal that FOOT-BINARITYσ dominates PARSEσ. When

FOOT-BINARITYσ dominates PARSEσ, A(meri)ca is then more harmonious than the

other two competitors, (Ame)(ri)ca and (A)(meri)ca. Not only this, but also (Ame)(ri)ca

and (A)(meri)ca each incur more violations of ALL-FEET-RIGHT than A(meri)ca (Cf.

Tableau 53 above). Once A(meri)ca wins as the optimal parse, MAIN-RIGHT comes into

play to predict that the rightmost foot is the head of the prosodic word. (93) below shows

that MAIN-RIGHT is undominated in English:

(93) Constraint Ranking

Lx=Pr, FAITH, MAIN-RIGHT >> NONFINALσ, REDUCTION >> FOOT-
BINARITYσ, BIMORAIC, ALL-FEET-RIGHT >> PARSEσ

170
 Having established that due to NONFINALσ, FOOT-BINARITYσ, ALL-FEET-

RIGHT, and MAIN-RIGHT, A(meri)ca is a better parse than all its competitors, there

remains to see which syllable in the rightmost foot is the head of the foot, i.e., the one

that receives main word-stress. Consider the following Tableau:

Tableau (54)
Input:/America/ Lx=Pr MR TR IM NFσ FBσ A-F-R Pσ
a- A(meri)ca * * **
b- A(meri)ca * * **

According to Tableau (54) above, the only difference between outputs (a) and (b) above

is whether the main foot is TROCHAIC (i.e. left-headed) or IAMBIC (i.e. right-headed).

If we assume that English feet are trochaic, output (a) wins the competition, but if we

assume that they are iambic, output (b) wins. The fact that /America/ surfaces with

antepenult stress gives us straightforward evidence that English feet are left-headed.

TROCHAIC is, like Lx=Pr. FAITH and MAIN-RIGHT, undominated in English:

(94) Constraint Ranking

Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> NONFINALσ, REDUCTION
>> FOOT-BINARITYσ, BIMORAIC, ALL-FEET-RIGHT >> PARSEσ

It is hypothesized earlier that stress is assigned to monomorphemic words by

appealing to mora count (see 78 above). Syllable weight then plays some role in stress

placement in monomorphemic words. For example, if the ultima contains a long vowel

(e.g. agree, maroon, etc.), it receives main stress. This is true for all words irrespective of

their syntactic categories. The ultima may be skipped over if it is monomoraic in verbs

and adjectives, but it may be stressed in nouns, provided that it is underlyingly accented

(i.e. due to FAITH as in bassinet). Similar restrictions apply on penult and antepenult

stressed syllables. The penult, for example, receives main word-stress iff (if and only if)

171
it is heavy (e.g. agenda, etc.). All of this points to the fact that syllable weight restrictions

are crucial for purposes of stress placement in English.

The fact that a final syllable with long vowel (e.g. agree, maroon, etc.) receives

main word-stress entails that NONFINAL is dominated by another constraint that forces

syllables of certain weight to receive stress. This constraint that refers to the weight of the

syllable is the Weight-to-Stress principle (WSP):

WSP
Heavy syllables must be stressed.

In 3.5.1.a above, it is pointed out that the number of moras that the nucleus and the coda

of the syllable can contribute determines syllable weight. Vowels, diphthongs and coda

consonants are divided into subsets depending on how many moras each subset

contributes to the weight of the syllable. Some subsets, it is hypothesized, contribute one

mora; some of them optionally contribute one mora; some of them contribute two moras;

some of them contribute three moras; and some of them contribute none (see 79 above).

It is also hypothesized that English syllables are maximally trimoraic. Trimoraic syllables

are heavier than bimoraic syllables, and so on. This proposal helps to explain why some

syllables are never skipped over as far as the place of stress is concerned while others are

skipped over in some positions but never in other positions. For example, whereas a CVV

syllable type is never skipped over in final position (e.g. agree, maroon, etc), a CVC

syllable type may be skipped over in that same position (e.g. animal, etc.), but never in

penult position (e.g. agenda, etc.). To capture this insight, Hammond (1999: 264)

proposes that WSP be “instantiated as a universally ranked series of parameterized

constraints”:

172
 (95) …>> WSP(VV) >>…>> WSP(VC) >>

(95) basically says that stressing a /CVV/ syllable type is more compelling than stressing

a /CVC/ syllable type. In terms of mora count, we propose that WSP(VV) counts as three

moras (i.e. WSPµµµ), and WSP(VC) as one mora (i.e. WSPµ).132

The fact that a /CVV/ syllable type is always stressed in final position proves that

WSPµµµ dominates NONFINALσ. Conversely, the fact that a CVC syllable type may be

skipped over in final position proves that WSPµ is dominated by NONFINALσ133. So

(95) above can be reproduced as (96) below:

(96)…>> WSPµµµ >> NONFINALσ >> WSPµ >>…

(96) basically says that the weight constraint WSPµµµ may dominate some non-weight

constraint (e.g. NONFINAL) which in turn may dominate another weight constraint (e.g.

WSPµ). For example, in a word like /agree/, the ultima gets stressed not because of

FAITH, but because of WSPµµµ which dominates NONFINAL as in the following

Tableau:

Tableau (55)
Input: /agree/ WSPµµµ NF
a- (ag)ree *!
b-"ag(ree) *

To avoid having stress on the final syllable of /animal/, we need to have NONFINAL

outrank WSPµ. Consider:

132
We also propose that when a closed syllable is stressed (e.g. agenda), it counts more than one mora (i.e.
WSPµµ).
133
The fact that bassinet is stressed on the final syllable, even though it is a CVC syllable type, is
accounted for by the assumption that stress here is underlying and that FAITH >> NONFINALσ.

173
Tableau (56)
Input: /animal/ WSPµµ NF WSPµ
a- (ani)(ml) *!
b-" (ani)ml *

Incorporating the domination relation in (96) into (94) above yields (97) below:

(97) Constraint Ranking

Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC, WSPµµµ >> NONFINAL,
REDUCTION >> WSPµ , FOOT-BINARITY(σ), BIMORAICITY, ALL-FEET-
RIGHT >> PARSE(σ)

In a word like [agenda] where a closed syllable (CVC) gets stressed, the penult

weighs more at least two moras134. In this instance, we need to postulate three

parameterized WSPs: one requires the stressing of a syllable that contributes three moras

(WSPµµµ); another requires the stressing of a syllable that contributes two moras

(WSPµµ); and still a third constraint that requires the stressing of a syllable that

contributes less than two moras (WSPµ). The parameterization of WSP then is as below:

(98) WSµµµ … WSPµµ … WSPµ135

Consider Tableau (57) below:

Tableau (57)
input /agenda/ TR WSPµµµ WSPµµ FBσ
a- "a(gen)da *
b- (agen)da *!
c- (agen)da *!

Notice that this accounts for the generalization that if the penult is heavy, it is then

stressed. A point worthy of mention here is that WSPµµ has to dominate FOOT-

134
The nucleus of the penult is a tense vowel which weighs two moras (see 79 above).
135
The domination relation between these three parameterized constraints is established by universal
principles in that WSµµµ dominates WSPµµ which in turn dominates WSPµ.

174
BINARITYσ. Otherwise, the penult in /agenda/ would pair up with the antepenult, and

due to TROCHAIC, stress would terminate on the antepenult, which is not correct.

As for syllables in final position that are closed by more than one consonant (e.g.

alarm, etc.), we also propose that they are stressed iff (if and only if) they weigh more

than one mora. Consider the following Tableau for /alarm/:

Tableau (56)
Input: /alarm/ WSPµµµ WSPµµ NF FBσ
a- "a(larm) *! *
b- (a)larm *! *

The question promptly arises as to what happens in a word like /harvest/. Recall

that coronals do not necessarily add to the weight of the syllable. By appealing to the

mora count in (79) above, we see that the difference between /alarm/ on the one hand and

/harvest/ on the other is that the final syllable in /alarm/ contributes more than one mora,

but the final syllable of /harvest/ contributes less than two moras.

As for the ranking of WSPµµ and WSPµ relative to the other constraints, crucial

here is their ordering relative to NONFINAL. In order to get ultimate stress in /alarm/,

WSPµµ has to outrank NONFINAL, but in order to avoid having /harvest/ surface with

ultimate stress, WSPµ has to be outranked by NONFINAL (see Tableau 59 below):

Tableau (59)
Input: /harvest/ WSPµµµ WSPµµ NF WSPµ FB
a- " (har)vst * *
b- (har)(vst) *!

What this analysis actually does is put final syllables that contribute more than one mora

in one category as opposed to final syllables that contribute less than two moras. In other

words, it puts word like [alarm] and [agree)in one category, and words like [novelist]

175
and [animal] in another category. In terms of constraint ranking, WSPµµ has to outrank

WSPµ and NONFINAL has to be interleaved.

We have then argued that WSPµµµ and WSPµµ dominate both FOOT-BINARITYσ

and NONFINALσ, but WSPµ is dominated by NONFINAL. The current domination

hierarchy is as in (99) below:

(99) Constraint Ranking

Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC, WSPµµµ, WSPµµ >> NONFINAL,
REDUCTION >> WSPµ , FOOT-BINARITY(σ), BIMORAICITY, ALL-FEET-
RIGHT >> PARSE(σ)

Finally, the fact that stress in monomorphemic words in English does not fall on a

pre-antepenult syllable even if that syllable is of a /CVV/ type proves that WSPµµµ and

WSPµµ are dominated by some higher ranking constraint(s). The most likely constraint to

dominate WSP is MAIN-RIGHT, so that when more than one foot is constructed over the

prosodic word, main word-stress goes to one of the syllables that make up the rightmost

foot even if the feet to the left of it contain a syllable that is bimoraic or even trimoraic.

The ranking of constraints in (100) below shows that MAIN-RIGHT dominates both

WSPµµµ and WSPµµ:

(100) Constraint Ranking

Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSPµµµ, WSPµµ >>
NONFINAL, REDUCTION >> WSPµ , FOOT-BINARITY(σ), BIMORAICITY,
ALL-FEET-RIGHT >> PARSE(σ)

Notice that the current domination hierarchy houses WSPµµµ, and WSPµµ, in same rank

relative to the other constraints. They are both dominated by MAIN-RIGHT and they

both dominate NONFINAL. We can collapse them into just one constraint. We suggest

that WSP be parameterized as below:

176
 (101) WSP(≥µµ) >> WSP(<µµ)136

What this means is that we put syllables that weigh two or more moras in one category as

opposed to those that weigh less than two moras. The final domination hierarchy is as in

(102) below:

(102) Constraint Ranking

Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ) >>
NONFINAL137, REDUCTION >> WSP(<µµ), FOOT-BINARITY(σ),
BIMORAICITY, ALL-FEET-RIGHT >> PARSE(σ)

3.5.2 Stress in polymorphemic words

Recall from 3.3.3 above that one of the basic differences between Arabic and English as

far the place of main word-stress is concerned is that in Arabic no distinction is made

between stress assignment in monomorphemic words and that in polymorphemic words.

In Arabic, suffixes, prefixes and infixes are fused within the phonological word, and

stress assignment is “a function of the total syllable pattern” (Mitchell 1960). Suffixed

and unsuffixed forms are treated alike as far as prominence is concerned in Arabic. This

is, however, not the case in English. Stress patterns characteristic of monomorphemic

136
In Arabic, we have seen that syllables that contribute one mora (i.e. CV) and syllables that contribute
two moras ( i.e. CVV or CVC) are put in one category as opposed to syllables that contribute three moras
(i.e. CVVC or CVCC).
WSP>µµ << WSP≤µµ
In English, however, syllables that contribute two or more moras are put in one category as opposed to
syllables that contribute less than two moras.
137
In Arabic, we have seen that NONFINAL is parameterized and WSP is considered irrelevant. We cannot
do the same for English because we cannot account for why a heavy syllable in penult position is stressed.
This is so because in English, unlike in Arabic, FOOT-BINARITY is interpreted under a syllabic analysis.
It seems impossible to get stress on a heavy penult without having it make up a syllable by itself and this
requires that a constraint such as WSP be invoked.

177
words in English are distinct from those characteristic of polymorphemic words. For

example, whereas stress falls on one of the last three syllables in monomorphemic words

(this is called the three-syllable window), it may fall further to the left in polymorphemic

words. Consider the following examples (stressed syllable in bold; suffix underlined):

(103) appropriateness, similarly, necessary, interested, jettisoning, alienist,

audiences, etc.

Suffixed forms138 also differ from unsuffixed forms in that they allow a heavy

syllable to be skipped over (Hammond 1999: 323). Consider the following examples

(stressed syllable in bold, skipped over heavy syllable in italics, suffix underlined):

(104) blandishment, instantly, columnist, messenger, motorman, Washington,

rancidness, allergy, similarly, comfortable, etc.

The words in (103) and (104) are exceptionally stressed on syllables that would

otherwise surface stressless. This is due to the influence of the suffix. In monomorphemic

words, the rightmost heavy syllable receives main word-stress (e.g. balloon, agenda,

anarchy, etc.). If the words in (104) were monomorphemic words, they would all receive

stress on the penult because it is heavy.139

Let us see how an optimality-theoretic analysis can account for these

generalizations. According to Hammond (1999: 327), stressing a syllable outside the

three-syllable window (Cf. 103 above), and skipping over a heavy syllable (Cf. 104) in

polymorphemic words can be obtained by positing a constraint that penalizes words with

138
In 3.3.2 above, it is shown that English suffixes fall into three categories as far as stress placement is
concerned (Cf. Cruttenden 1986: 20):
a. suffixes which leave the stress on the stem unaffected, e.g. fulfil/fulfilment; usual/usually
b. suffixes which themselves take the stress, e.g. limit/limitation; picture/picturesque; China/Chinese
c. suffixes which shift the stress on the stem, e.g. economy/economic; curious/curiosity; apply/applicant;
maintain/maintenance
139
As for the suffixes themselves, we will see shortly that they are extraprosodic, i.e. fall outside the stress
domain.

178
suffixes. Suffixes are considered extraprosodic, i.e., not part of the prosodic word. In OT

terms, extraprodicity can be achieved by positing NONFINAL as a high-ranking

constraint. For a suffixed word to surface with stress outside the three-syllable window

(e.g. necessary), or with stress on some syllable other than the rightmost heavy syllable

(motorman), there should be another constraint that dominates all the constraints that

yield main word-stress in monomorphemic words. Let us consider how stress terminates

on some syllable other than the rightmost heavy one in polymorphemic words.

The constraint that is responsible for the stressing of a heavy syllable is WSP.

WSP
A heavy syllable must be stressed

In /agenda/, for example, the heavy penult is stressed because of WSP. Consider the

following Tableau:

Tableau (60)
Input: /agenda/ NFσ WSP
a- " a(gen)da
b- (a)genda *!
c- agen(da) *! *

Likewise, if WSP were undominated, stress would wrongly terminate on the heavy penult

in /motorman/:

Tableau (61)
Input: /motorman/ NFσ WSP
a- !" mo(tor)man *
b- (mo)torman *!*
c- motor(man) *! *

179
To avoid this conclusion, we need to posit a constraint that outranks WSP, so that stress

surfaces on the antepenult in /motorman/. Specifically, we need a constraint that calls for

the skipping over of the heavy penult in /motorman/, but not the heavy penult in /agenda/.

Hammond (1999: 327) proposes a constraint that penalizes all words with

suffixes:

Neutral140
NoSuffix141

Neutral is posited to explain (1) how stress may fall on some syllable outside the

three-syllable window, and (2) how a heavy syllable may be skipped over in a

polymorphemic word. When some suffix in the word is considered extraprosodic, the set

of constraints that act on the remaining part of the word interact to yield the correct stress

pattern. For example, a word that receives stress on some syllable other than the

rightmost heavy one (e.g. blandishment) can be optimal provided that NEUTRAL is

undominated. Consider Tableau (62) below:

Tableau (62)
Input: /blandishment/142 NEUTRAL NFσ WSP
a- " (bland)dish≠ment143 **
b- (bland)(dish)≠ment *! **
c- (bland)(dish)ment *! **

140
Whereas NEUTRAL is high-ranked in English, it is low-ranked in Arabic. We will see later that for
Arabic-speaking learners of English to acquire the stress system of English, they need to keep NEUTRAL
undominated. This can be done by demoting the constraints that dominate NEUTRAL to a lower position.
141
Hammond’s formulation of this constraint is as below:
Certain affixes cannot be in the prosodic word (Hammond 1999: 327).
142
Notice that the suffix (i.e. ment) is not a possible carrier of stress because of high-ranking
NONFINALσ.
143
≠ is used to indicate the boundary of the prosodic word. Notice that once the suffix is not part of the
prosodic word, the penult escapes stress because of NONFINAL.

180
According to Tableau (62) above, the competition between (a) and (b) is resolved in

favor of (a) because (b) violates NONFINALσ144. However, the competition between (a)

and (c) must be resolved by appealing to some other constraint(s) because both (a) and

(c) satisfy NONFINAL(σ). What we need then is to find out why output (a) surfaces as

the actual form. In [blandishment], the heavy penult (i.e. /dish/) is skipped over.

Although this is a violation of WSP, it is licensed by NEUTRAL which penalizes

polymorphemic words (Cf. output c in Tableau 62 above). For /blandishment/ to surface

with antepenultimate stress, WSP has to be dominated by NEUTRAL, so that the output

form that violates NEUTRAL loses the competition for optimality irrespective of how

well it fares on WSP.145 The same should apply to cases where stress falls on a pre-

antepenult syllable (Cf. 82 above). All in all, for stress to surface on some syllable

outside the three-syllable window and for a rightmost heavy syllable in the prosodic word

to be skipped over in suffixed forms, NEUTRAL should dominate all constraints that

militate against this state of affairs:

(105) Constraint Ranking

NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ)
>> NONFINALσ, REDUCTION >> WSP(<µµ), FOOT-BINARITY(σ),
BIMORAICITY, ALL-FEET-RIGHT >> PARSE(σ)

144
Notice that NONFINALσ only affects stemsyllables.
145
We have seen in 3.3.4 above that a heavy penult is never skipped over in Arabic. We predict that words
such as those in (87) above pose considerable difficulty for native speakers of Arabic as far as prominence
is concerned.

181
3.5.3 Secondary stresses

Another area that will be broadly sketched out is the place of secondary stresses in

English. Recall from section 3.4 above that most studies of word-stress in Arabic show

that every word has only one main stress and no secondary stress(es) (Cf. Mitchell 1960,

for Egyptian Arabic). Secondary stresses are disallowed in Arabic. English, on the other

hand, distinguishes up to four degrees of stress: primary stress, secondary stress, tertiary

stress and zero stress (unstressed). This basically means that the constraint that militates

against secondary stress is undominated in Arabic, but low-ranked in English146:

*Non-Primary stresses147
Nonprimary stresses are are disallowed

Apart from the issue whether or not languages recognize secondary stresses, there

is still a more complicated issue (which we will not pursue in detail here) related to the

location of secondary stresses relative to primary stress. In English, for example,

secondary stresses usually fall to the left of the primary stress in monomorphemic words.

Hence, the rightmost stress is, generally speaking, the main stress of the word. Consider

the following examples (syllable with primary stress in bold, syllable with secondary

stress in italics):

(106) gasoline, minaret, Alexandria, souvenir, Oklahoma, pantaloon

However, this generalization does not always hold. Consider the following examples

(syllable with main stress in bold, syllable with secondary stress in italics):

146
Again we will see later that for native speakers of Arabic to acquire English secondary stress, they need
to demote this constraint (which is top-ranked in their native language) to a lower level.
147
Hammond (1999) calls this constraint *2nd.

182
 (107) hygiene [hayjin], asset, blaspheme, ribald, apartheid, kerosene [kєrәsin],
Amazon, Afghanistan, cucumber, pinochle, hierarchy, etc.

According to Hammond (1999: 322), the generalizations regarding nonfinal primary

stress are as follows: (1) “main stress is never separated from the right edge of the word

by more than one secondary stress”, and (2) “the secondary stress that falls to the right is

always either the final syllable or the penult (only if the final syllable is a syllabic

sonorant)”. To capture this generalization, Hammond (1999: 318), postulates that in all

cases where a secondary stress falls to the right of primary stress (Cf. 107 above), the foot

which contains the syllable with the rightmost secondary stress is degenerate.148

Implicit in this is the assumption that for secondary stress to be realized, more

than one foot must be constructed over the word. However, constructing more than one

foot entails more violations of ALL-FEET-RIGHT and ALL-FEET-LEFT. Hence, a

candidate with one foot incurs fewer violations of ALL-FEET-RIGHT and ALL-FEET-

LEFT than a candidate with two feet, and so on. On the other hand, constructing more

feet over the word entails fewer violations of PARSEσ. Since in English it is more

important to construct more than one foot over the word than to have each and every foot

coincides with the right edge of the word, we must tolerate more violations of ALL-

FEEt-RIGHT. In terms of constraint ranking, we need PARSEσ to dominate ALL-FEET-

RIGHT:

148
For how English secondary stress is accounted for in an optimality theoretic approach, see Pater (1995).
It suffices here to say that in a word like / Amazon/ where the final syllable is secondarily stressed, that
final syllable must make up a foot by itself. In terms of constraint ranking, the constraint that calls for the
footing of the final syllable of /Amazon/ must outrank NONFINAL. For ease of exposition, lets assume
that the constraint which calls for the realization of a secondary stress on the final syllable of /Amazon/ is
FAITH2. FAITH2 has then to dominate NONFINALσ.

183
 (108) Constraint ranking
NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ)
>> NONFINALσ, REDUCTION >> WSP(<µµ), FOOT-BINARITY(σ),
BIMORAICITY >> PARSEσ >> ALL-FEET-RIGHT >>

As for *NON-PRIMARY, it is violated in English fairly frequently. In terms of

constraint ranking, it is ranked low enough in the domination hierarchy that its impact is

inconsequential in English.

(109) Constraint ranking

NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ)
>> NONFINALσ, REDUCTION >> WSP(<µµ), FOOT-BINARITY(σ),
BIMORAICITY >> PARSE(σ) >> ALL-FEET-RIGHT >> …>> *NON-
PRIMARY

3.5.4 Conclusion

The ranking of constraints in (109) above, while it does not account for every single

token in English, captures the basic generalizations concerning the English stress system,

especially in monomorphemic words. These generalizations include:

I. Lx=Pr, FAITH, MAIN-RIGHT, and TROCHAIC are all undominated in

English.

a. Lx=Pr is undominated because each and every content word has to have at
least one foot constructed over it; otherwise, that word will surface
stressless (Hence it won’t have a head). A(meri)ca is then a better parse
than America.
b. Lx=Pr must dominate NONFINAL so as to avoid rendering the whole
word extrametrical (i.e. unfooted). (cat) is then a better parse than cat.
c. Lx=Pr must dominate FOOT-BINARITY. The need to have at least one
foot constructed over monosyllabic words is greater than the need to have
the foot fit a certain shape. (hat) is a always optimal.
d. FAITH must dominate NONFINAL so that a final closed syllable can be
exceptionally stressed. (bassi)(net) is a better parse than (bassi)net.

184
 e. FAITH must dominate REDUCTION, so that a final closed syllable that is
stressed does not undergo reduction. (bassi)(net) is a better parse than
(bassi)(nt).
f. MAIN-RIGHT must dominate WSP so as to allow stress to fall on one of
the last three syllables. (bassi)(net) is a better parse than (bassi)(net).
g. TROCHAIC must dominate IAMBIC, so as to allow feet to be left-
headed. A(meri)ca is a better parse than A(meri)ca.

II. All the other constraints are dominated. Yet some of them are ranked relative
to each other:

a. WSP(≥µµ) must dominate NONFINAL, so as to allow words ending with

long vowels to receive stress on that syllable. ag(ree) is a better parse than
(ag)ree; and to allow stress to surface on ultimate syllables that are at least
bimoraic. a(larm) is a better parse than (a)larm.
b. NONFINAL must dominate WSP(<µµ) so as to allow a closed syllable in
final position to be skipped over: (ani)mal is a better parse than (ani)(mal);
and to allow a syllable that is closed by consonant cluster that contributes
less than two moras to be skipped over and thereby surface stressless:
(nove)list is a better parse than (nove)(list)
c. NONFINAL must dominate ALL-FEET-RIGHT, so as to allow stress to
fall on an antepenult syllable. A(meri)ca is a better parse than Ame(rica).
d. NONFINAL must dominate FOOT-BINARITY. (ham)mock is a better
parse than (hammock); and A(meri)ca is a better parse than (Ame)(rica).
e. REDUCTION must dominate BIMORACITY, so as to allow the nucleus
of unstressed syllables to reduce to schwa. hammck is a better parse than
hammock.
f. WSP(≥µµ) must dominate FOOT-BINARITY, so as to allow a closed
syllable that contributes more than one mora to be stressed in penult
position. A(gen)da is a better parse than (agen)da
g. FOOT-BINARITY must dominate PARSE-σ. A(meri)ca is a better parse
than (A)(meri)ca or (Ame)(ri)ca.
h. PARESσ must dominate ALL-FEET-RIGHT. (mina)(ret) is a better parse
than mina(ret).

For language acquisition purposes, learners of English need to contrast this

hierarchy with that of their native language(s). In cases where the two systems differ, the

learner needs to move towards the target system. In the next chapter, we will see that this

is achieved through constraint demotion.

185
3.6 Endnotes

Endnote 1
Jordanian Arabic
Whenever a research project sets out to study how Arabic-speaking subjects acquire another linguistic
system, the debate concerning the subjects’ actual native language becomes inevitable— Is the first
language of the subjects Standard Arabic (H) or some local variety (L)?149 Without going into the
intricacies of the subject matter, the position adopted in this research project is that L rather than H is the
subjects’ first language. In addition to the overwhelming evidence that L should be considered the first
language, this has been adopted here for at least two other reasons. First, this will save us time and effort to
delving into a no-less debatable issue— the nature of stress in Classical Arabic. For many have questioned
the essence of stress in Classical Arabic. Ferguson (1956: 384), for example, claims that “Classical Arabic
has no word stress at all”. His reasoning is that the stress pattern that was dominant in Eastern non-Arabian
Arabic such as Egypt, Syria, etc., was applied to Classical Arabic when this variety came at one point in
time to be the dominant language150. Second, and most importantly, even if we assume, contrary to
Ferguson’s assumption, that stress in classical Arabic is original, the evidence is available that the
differences between classical Arabic and the regional varieties as far as word-stress is concerned are minor
(see section 2.4 below for a straightforward comparison between Classical Arabic and Jordanian Arabic).
Since one of the objectives of this research is to come to a better understanding of when and how
the native language patterns will transfer to the second language, it would be appropriate to briefly
comment on those features of the native language that are most relevant to the research questions.
Meanwhile, a general idea about the basic features of Jordanian Arabic will help the reader who is
unfamiliar with the language.
First of all, I should mention that the term Jordanian Arabic is not a conclusive term referring to
one single variety spoken throughout the whole country; rather, there are many local varieties. Although all
of them are mutually intelligible, they all exhibit certain characteristics that make their native speakers
easily identifiable. Interestingly enough, some of these varieties may not be as close to each other as they
are to other varieties of Arabic spoken outside the borders of Jordan (see Abdel-Jawwad 1981, 1986; Al-
Khatib 1989; Irshied 1984; Irshied and Kenstowicz 1984; and Palva 1976, 1980, 1986). For example, the
local varieties spoken in the Eastern and Southern parts of Jordan may share more features with the dialects
spoken in Saudi Arabia and the Bedouin Arabic of Negev than they share with the varieties spoken in the
rural and urban centers in the country (Irshied 1984; Bani Yasin and Owens 1984). Likewise, the varieties
spoken in the central and Northern parts of Jordan may have much in common with those spoken in Syria,
Palestine and Lebanon (Abdel Jawad 1981, 1986). Overall, there are mainly three undisputed local varieties
in Jordan: Urban, Rural and Bedouin151. The present paper, which aims at studying the acquisition of

149
This is related to the diglossic situation in the Arab World where there is a one-to-one relationship
between language usage and social context, that is, each variety (usually referred to as L and H) serves
certain purposes, e.g., ordinary conversation vs. formal speech, in certain settings (e.g., home, school,
church, etc.). Crucially, speakers in diglossia situations use one variety at a time, simply because each
variety can be seen as having a distinct place or function within the local speech repertoire. In diglossia
situations then code alternation is a conscious process, i.e., participants immersed in the interaction are
usually aware which code is used at any one time (For an informative discussion of diglossia in the Arab
World, see Ferugson 19-)
150
For an alternative view, see Abdo 1969.
151
Sakarna (1999), for example, objects to the generalization that the members of the many Bedouin tribes
in Jordan speak on dialect called the Bedouin dialect; rather, he believes that there are as many Bedouin
varieties as there are tribes. He recognizes at least five of them: Bani Hassan Bedouin variety, Bani Saxr
Bedouin variety, Bduul Bedouin variety, HweTaat Bedouin variety, and Al-9ajarma Bedouin variety. The
same may also apply to the urban and rural varieties. I myself who is a native speaker of one of the rural
varieties have noticed that there are some features characteristic of the variety spoken in my area that make

186
English stress patterns by Jordanian adult second language learners of English, sets out with the premise
that these dialects are highly congruous as far as stress placement is concerned. Most of the studies that are
partially or totally devoted to investigating stress patterns in Jordanian Arabic varieties make the claim that
stress placement is subject to the same rules in almost all these varieties (Abu-Salim 1982; AlGhazo 1981;
Al-Sughayer 1990; Sakarna 1999, among others). For the purpose of this research, we will incorporate data
from many of these varieties in our attempt to uncover the stress patterns in Jordanian Arabic. If it turns out
that there are some minor differences, they will be highlighted and discussed in as much they are relevant
to the purpose of analysis.

Jordanian Arabic vs. Standard Arabic: general idea

The purpose of including this section is twofold. First, we intend to give the reader who is unfamiliar with
the diglossic situation in the Arab world a brief idea about the two varieties. Second it serves presentational
purposes. When some lexical entry happens to be represented by somewhat different forms in two varieties
(Here we mainly refer to the colloquial varieties usually referred to as L and the standard as H), the reader
may need to know how the two forms are related to each other. For example, the corresponding Urban and
Rural colloquial forms for the Classical Arabic word /kayfa/ are /ki:f/ and /čeef/, respectively. Since our
goal is to show how H and L forms are related as far as stress is concerned, not all of the features that
characterize each variety have been surveyed. Only major correspondences and differences between the
two varieties are presented. We will focus on the systematic differences between the two main varieties
with respect to the inflectional and case ending systems on the one hand, and the vowel and consonant
systems, on the other.
Inflectional and case endings
In a nutshell, case and inflectional endings of classical Arabic have been lost altogether in almost all
varieties of colloquial Arabic. For example, a word like /kitaab/ ‘book’ which has only one form in all
colloquial Arabic varieties has six case endings in Classical Arabic: ?al.kitaabu, kitaabun, ?al.kitaaba,
kitaaban, ?al.kitaabi, kitaabin152. So in order to see the correspondence between /kayfa/ and /ki:f/ /čeef/, for
example, the first step is to see that the inflectional ending <a> in /kayfa/ is lost altogether in /ki:f/ and
/čeef/.

Consonant and vowel systems

The consonant and vowel inventories of both Standard Arabic and Jordanian Arabic are presented in Tables
(1) and (2) below
Table (1): consonants
bilabial Labio- (inter) alveolar Post- palatal velar uvula Pharyn- glottal
dental dental alveolar r geal
Plosive ‫ب‬ ‫د ت‬ ‫)( ك‬ ‫ق‬ ‫ء‬
Nasal ‫م‬
Trill ‫ن‬
Flap ‫ر‬
Fricat. ‫ف‬ ‫ذ ث‬ ‫ز س‬ ‫ش‬ ‫ع ح غ خ‬ ‫ﻩ‬
Affri- () ‫ج‬
Approx ‫و‬ ‫ي‬
lateral ‫ل‬
Emph.S ‫ط‬ ‫ض‬
Emph.F ‫)( ظ‬ ‫ص‬

it non-identical with the other varieties spoken in other areas. Be it as it may, for the purpose of this
research it is assumed that those local varieties (or dialects) are more or less identical as far as prominence
is concerned. Sakarna (1999: 27) himself notes that as far as stress is concerned, there is very little, if any,
differences between the dialect he investigates and the other Jordanian dialects.
152
u, a, and i are the case endings for the nominative, accusative and genitive, respectively. /n/ is added if
the noun is indefinite

187
As far as the consonantal system is concerned, there some differences between classical Arabic and
Jordanian Arabic. First, there are some sounds that are part of the consonantal inventory of CA but are
nonexistent in Jordanian Arabic, namely the boldfaced voiceless uvular plosive ‫( ق‬which has been replaced
by the voiced velar stop /g/ in the Rural and Bedouin varieties and by the voiceless glottal stop /?/ in the
Urban variety in almost all cases153) and the voiced alveolar emphatic stop ‫( ض‬which has been replaced by
the voiced dental emphatic alveolar ‫ ظ‬in the Rural and Bedouin varieties but continues to be used in the
Urban variety). The interdental emphatic fricative ‫ ظ‬is only changed in some urban varieties into (ż). There
are other changes such as those which affect ‫ث‬, ‫ذ‬, etc., but those changes may affect one dialect or the
other154.
Conversely JA has two consonants which are lacking in Classical Arabic: the voiced velar stop /g/ and the
voiced post-alveolar (palato-alveolar) affricate č. Those two sounds are not represented by separate
symbols in the Arabic consonantal system of writing: ‫ ق‬is used in place of /g/ and ‫ ك‬is used in place of /č/.
Table (2): Vowels
i ii u uu a aa ay aw
high + + -
back - + -

As for the vowel system, Classical Arabic is characterized by having a very simple vowel system: three
short vowels with three corresponding long counterparts. Besides, it has two diphthongs (see Table (2)
above). JA, on the other hand, happens to have developed a more complex vowel inventory: continues to
have the six vowels of CA, but at the same time makes use of four more vowels, namely /e/, /ee/ , /o/ and
/oo/ (‘beet’ ‘home’, and yoom ‘day’). However, the diphthongs are used very sporadically in this dialect,
and probably in almost all the dialects of Arabic:

i ii u uu a aa e ee o oo
high + + - - -
back - + - - +
low - - + - -

It is noteworthy that in most cases the diphthong /ay/ is replaced in JA by /ee/ as in /bayt/ > /beet/ ‘home’,
/sayf/ > /seef/ ‘sword’; and /aw/ is replaced with /oo/ as in /dawr/ > /door/ (turn), /θawr/ > /θoor/ ‘bull’,
etc155.
As far as the orthography of Arabic is concerned, the long vowels are represented by separate symbols,
namely ‫ي‬, ‫و‬, and ‫ ا‬for ii, uu and aa, respectively. ‫ ي‬and ‫ و‬are also used to represent the diphthongs /ay/ and
/aw/, respectively. The short vowels /i/, /u/ and /a/ are only represented as diacritics, namelyِ ,َ, andُ , called
kasra, fatha and Damma by Arabic grammarians (for more details, see Mitchell 1960; Abdo 1969; Abdl-
Al-Jawad 1981; Abu-Salim 1982; Bani Yasin and Owens 1984; Al-Ghazo 1987; Al-Sughayer 1990; and
Zakarna 1999). We will not go into further differences between Classical and Colloquial Arabic varieties as
this is not one of the objective of the study. For a list of the differences between the two varieties, see Abdo
(1969: 1-3).

153
The voiceless uvular plosive ? is still in both dialects in the same way in a very limited like in
/?al.qahira/ “Cairo”
154
For the differences between Classical Arabic and Urban dialects in the consonant and vowel inventories,
see Abdel-Jawad and Awwad (1989), for Classical Arabic and Bedouin dialects see Sakarna (1999)
155
See Abdo (1969: 8) for cases where this rule does not apply

188
Chapter Four: Stress Acquisition

4.1 Acquisition of English Stress

In the previous chapter, our goal was to develop an optimality-theoretic analysis of the

Arabic and English stress systems. The constraint ranking in (77) in Chapter Three above

captures the basic generalizations regarding the place of stress in Arabic. The constraint

ranking in (109) captures the basic generalizations regarding the place of stress in

monomorphemic words in English. In this chapter, our primary goal is to compare the

constraint rankings obtained for the two systems so that we can find out (1) what stress

errors occur in the English of native Arabic speakers, and (2) what native speakers of

Arabic who learn English need to do to improve their English with respect to stress

placement.

As Optimality Theory views each grammar as a set of universal, violable

constraints, and cross-linguistic variation results from different rankings of these

constraints, the sole task of the language learner is to re-rank the constraints of his/her

native language so that they match that of the target language. In the case of native

speakers of Arabic learning English, they need to modify the constraint ranking in (77) so

that it becomes like (109) above. The question that promptly arises is: How can the

learner accomplish this task?

 Within the framework of Optimality Theory, when the language learner finds that

the constraints that yield the correct output form for a particular linguistic aspect (say

stress) are ranked differently in his/her native language and in the target language, s/he

needs to employ the notion of constraint demotion to reach the target grammar. In the rest

of this chapter, we set out to do two things. In 4.2 below, we explain how the notion of

constraint demotion works. In 4.3, we predict the errors in the English of native speakers

of Arabic (i.e. compare the Arabic and the English rankings of constraints with regard to

specific erroneous tokens) and show how the language learner, when making use of the

notion of constraint demotion, could improve his/her English in terms of stress.

4.2 Constraint Demotion 156

In this section, we elaborate on how the language learner infers the ranking of constraints

for some language in order to acquire it. According to OT, the language learner has

available to him/her: (1) the set of universal constraints, (2) the set of universal inputs157,

and (3) a set of language-particular output forms. The task of the language learner is to
156
This section based on the following cited works:
Tesar, Bruce (1995). Computational Optimality Theory. [Ph.D. Diss, University of Colorado, Boulder].
[ROA-90, http://ruccs.rutgers.edu/roa.html]
Tesar, Bruce (1996). An iterative strategy for learning metrical stress in Optimality Theory. [ROA-177,
http://ruccs.rutgers.edu/roa.html]
Tesar, Bruce (1997). Multi-recursive constraint demotion, [ROA-197, http://ruccs.rutgers.edu/roa.html]
Tesar, Bruce (1998). Robust interpretive parsing in metrical stress theory. [ROA-262,
http://ruccs.rutgers.edu/roa.html]
Tesar, Bruce and Paul Smolensky (1996). Learnability in Optimality Theory (long version) [ROA-156,
http://ruccs.rutgers.edu/roa.html]
Tesar, Bruce and Paul Smolensky (1996). Learnability in Optimality Theory (short version) [ROA-155,
http://ruccs.rutgers.edu/roa.html]
Tesar, Bruce and Paul Smolensky (1993). The learnability of Optimality Theory: An algorithm and some
basic complexity results. [ROA-2, http://ruccs.rutgers.edu/roa.html]
157
The belief that the set of inputs is universal is usually referred to in the literature as Richness of the Base
(Prince and Smolensky 1993: 191). The basic idea here is that cross-linguistic variation is due to one single
reason: constraint variation, so the input to the language-particular grammars is the same.

190
deduce a language-particular grammar, that is, a ranked set of universal constraints.

Hence, particular grammars are language-particular rankings of the set of universal

constraints. As Tesar (1996: 1) puts it “to learn the core grammar of a language is to learn

the language-specific ranking of the universal constraints”. Given language-particular

output forms, the language learner deduces the correct ranking of constraints by adopting

a simple strategy: demoting all constraints that are violated in some output form (which is

the optimal form). This is based on a prior assumption that if an output form violates

some constraint, that constraint must be dominated by some higher-ranked constraint(s),

which the output form does not violate, but which can be violated by some potential

candidate form. The output form, like all the other candidate forms that are supplied by

GEN, can, and should, violate some constraints, but those constraints are never higher

ranking than the constraints that are not violated. The language learner thus has to figure

out what is dominating what, that is, which constraints are higher-ranking, and which are

lower-ranking.

Tesar and Smolensky’s (1993) algorithm is a formal mechanism to show how

domination relationships between conflicting constraints are determined by the language

learner. Already noted is the assumption that when an output form violates some

constraint, that constraint is believed to be dominated by some higher-ranking one(s),

which the output form never violates, but which is violated by an alternative candidate

form. What the language learner does is compare the output form (which will ultimately

win the competition by being optimal but never perfect) with every other candidate form

supplied by GEN. When evaluating the actual and potential forms against the set of

universal constraints, any constraint that is violated by the output form is demoted. This

191
implies that at the initial stage the learner has all the constraints unranked relative to each

other as below:

(1) Constraints unranked {C4, C1, C5, C6...}

The final stage is reached after demoting all the constraints which are violated by the

output form to some position below the constraints which are violated by all its

competitors. Suppose that C5 and C10, for example, are violated by the output form F1,

and that C3, C6 and C7 are violated by its competitors F2, F3, and F4, then C5 and C10

(those violated by the output form F1) must be demoted one step below C3, C6, and C7

(those violated by the suboptimal forms F2, F3, F4), and thus, C3, C6, and C7 dominate

C5 and C10.

(2) Initial ranking {C3, C5, C6, C7, C10)

Domination relation {C3, C6, C7}

>
{C5, C10}

In this way, the language learner can guarantee that the consequences of violating C5 and

C10 (hence lower-ranking) are never as serious as the consequences of violating C3, C6,

and C7 as far as the input form in question is concerned.

According to the algorithm developed by Tesar and Smolensky (1993), the

process of evaluating the candidate forms in pairs (consisting each time of the output

form and another potential form) against the set of constraints usually results in

intermediate stages where the target hierarchy has not been finalized. These intermediate

stages are subject to change as they only represent the knowledge the learner has obtained

from some, but not all, candidate forms. Only when the learner tests all candidate forms

against the whole set of constraints does his knowledge becomes permanent.

192
 Let’s take a hypothetical example to demonstrate how this works. Suppose that

we have five constraints that interact to yield the correct output form in Language L, and

suppose that GEN has provided the learner with four potential forms that incur a number

of violations against the set of constraints as illustrated in Tableau (1) below:

Tableau (1)
input (F) C1 C2 C3 C4 C5
"F1 * ** ***
F2 *! * * *
F3 *! * ** * **
F4 *!* ** **

Tableau (1) above shows that only one output form is optimal, namely F1 marked with

" (informally called the winner). F2, F3, and F4 (informally called the losers) do not

surface as optimal forms as they fatally violate higher-ranked constraints, namely C1 and

C2158. F1 surfaces as optimal because it does not violate C1 and C2, though it does

violate C3, C4 and C5. The fact that the optimal form F1 incurs more violations of C4

and C5 than any other candidate form does not prevent it from surfacing as optimal

because these are lower-ranked constraints. The challenge is then to deduce that C1 and

C2 (which are violated by the suboptimal forms) dominate C3, C4 and C5 (which are

violated by the optimal form). In terms of language acquisition, the question arises as to

how the language learner can infer this ranking.

According to the Tesar and Smolensky’s (1993) algorithm, the language learner

starts off by comparing the optimal form (F1 in this case) with every other candidate

158
Notice that the difference between F3 and F4 is that the latter incurs double violation of the higher-
ranking constraint C1.

193
form (F2, F3, and F4) one at a time (i.e. F1 vs. F2, F1 vs. F3, and F1 vs. F4). The

comparison of F1 with F2 results in tableau (2) below (usually called a mark-data pair):

Tableau (2)
input C1 C2 C3 C4 C5
"F1 * ** ***
F2 *! * * *

The comparison of F1 with F3 results in tableau (3) below:

Tableau (3)
input C1 C2 C3 C4 C5
"F1 * ** ***
F3 *! * ** * **

The comparison of F1 with F4 results in tableau (4) below:

Tableau (4)
input C1 C2 C3 C4 C5
"F1 * ** ***
F4 *!* ** **

The next step is that the language learner tries to figure out which piece(s) of

information is (are) decisive for ranking purposes in these reduced tableaux. For example,

since both F1 and F2 violate C3 in the same way in Tableau (2) above, this piece of

information may not have any consequences on the ranking of constraints because the

optimal form fares neither better nor worse than the suboptimal form on that particular

constraint. The language learner may then need to ignore this piece of information

altogether, as it is not decisive for constraint ranking purposes. The algorithm requires

that such unnecessary information be eliminated altogether. One way to do that is to

delete each occurrence of * which is shared by F1 and any other single candidate form.

194
Consider Tableaux 2, 3, and 4, which are reproduced as 5, 6, and 7 respectively, after

being purged of unnecessary information:

Tableau (4)
input C1 C2 C3 C4 C5
"F1 * **
F2 *!

Tableau (6)
input C1 C2 C3 C4 C5
"F1 * *
F3 *! * *

Tableau (7)
input C1 C2 C3 C4 C5
"F1 * ** *
F4 *!* **

It is clear then that we do not need to mark violations (*) in the same column.

Tableau (7), for example, shows that each constraint (C1, C2, C3, C4 and C5) is violated

by either F1 or F4, but never by both of them: C1 is violated by F4; C2 by F4; C3 by F1;

C4 by F1; C5 by F1. It no longer shows any constraint violations common to both F1 and

F4159.

After this cancellation, if it turns out that some form (whether optimal or

suboptimal) still incurs more than one violation of each constraint, then that piece of

information should also be discarded simply because what matters now is not the number

of violations of each constraint, but the fact that the constraint is being violated. In other

159
This whole process is referred to in the literature as Mark Cancellation, stated formally in Tesar (1995:
78) as:
for each pair (loser marks, winner-marks) in mark data:
a. For each occurrence of a mark *C in both loser-markers and winner-marks in the
same pair, remove that occurrence of *C from both
b. If, as a result, no winner-marks remain, remove the pair from mark data

195
words, the presence, not the magnitude, of the violation is what counts here160. So, we

need another round of “purification” of the reduced tableaux, this time to reduce multiple

violations of the same constraint by any one form to just one. Consider tableaux 5, 6, and

7, which are reproduced as 8, 9, and 10 respectively, after multiple violations are reduced

to one:

Tableau (8)
input C1 C2 C3 C4 C5
"F1 * *
F2 *!

Tableau (9)
input C1 C2 C3 C4 C5
"F1 * *
F3 *! * *

Tableau (10)
input C1 C2 C3 C4 C5
"F1 * * *
F4 *! *

Now the Tableaux only contain the necessary information that the language

learner needs to reach the correct ranking of the constraints. Before an illustrative

example is provided, a number of points need to be reiterated.

First, as already noted, the language learner starts with an a priori assumption that

in the initial stage the constraints are unranked relative to each other. So, constraints can

be put in any order as below:

160
Notice that we do not reduce multiple asterisks to one prior to canceling the violations assessed by the
optimal and suboptimal forms.

196
 (3) Random ordering161 {C4, C1, C3, C5, C2}
{C2, C5, C3, C4, C2}
Etc.

Second, the language learner is not expected to compare all potential candidate

forms (optimal and suboptimal forms) in a fixed way (F1 before F2 and so on). He may,

compare F1 (the optimal form) with F4, for example, before comparing it, let’s say, with

F2 or F3. This is so because the candidate forms that are supplied by GEN are not

structured. F5 may be supplied by GEN before F2, and so on. The algorithm is equipped

to work taking any pair comparison as its starting point: we can start with Tableau 8 and

end with 10, or vice versa.

Third, the algorithm is based on the idea of constraint demotion. That is, the

constraints that are violated by the suboptimal forms F2, F3, and F4 dominate those

violated by the optimal form F1. Hence the former are higher-ranking constraints, but the

latter are lower-ranking constraints. In other words, having two constraints- one violated

by the optimal structure and the other violated by the suboptimal competitor- the one

violated by the optimal structure must be dominated by the one violated by its

competitor, denoted as suboptimal >> optimal. This implies that if at one stage two

constraints, let’s say X and Y, are housed in the same stratum as in (4a) below, the one

which is violated by the optimal form should be demoted to the immediately following

stratum as in (4b) (where X is violated by the optimal structure and Y by its competitor):

(4)
(a) Stratum 1 {W, X, Y, Z}
>>
Stratum 2 {M, N}
161
To claim that the constraints in the initial stratum are randomly ordered or relatively unranked at all may
not have any practical consequences. Hence the decisive factor here is the position of the stratum with
regards to the other strata in the hierarchy (Tesar and Smolensky 1996: 32)

197
 (b) Stratum 1 {W, Y, Z}
>>
Stratum 2 {M, N, X}

It should be noted that although the constraints in each stratum are unranked

relative to each other, each constraint in stratum 1 dominates all the constraints in stratum

2 which in turn dominate all constraints in subsequent strata, and so on. Whereas X and Y

in (4a) above do not dominate each other, Y dominates X in (4b) because X is demoted

one step downward. The process of constraint demotion is repeated over and again until

no positive evidence is available for further demotion to take place.

Having all of this in mind, let’s see how the algorithm is equipped to yield the

correct ranking for the constraints in Tableau (1) above. Recall that the important point as

far as the ranking of constraints is concerned in Tableau (1) is that F2, F3, and F4 do not

surface as optimal forms because they fatally violate the higher-ranked constraints C1

and C2. F1, on the other hand, wins the competition because it does not violate C1 and

C2, though it violates C3, C4 and C5. We have, therefore, to prove that C1 and C2

dominate C3, C4 and C5.

First, as already noted, the language learner starts with an a priori assumption that

the constraints are unranked relative to each other, so the algorithm considers the

constraints in (5) below to be unranked relative to each other:

(5) Initial stratum = {C4, C1, C3, C5, C2}

The next step is to find out what dominates what in each of the reduced “purified”

tableaux 8, 9, and 10 above. This is done on the principle that the constraints violated by

the suboptimal structure dominate those violated by the optimal structure. From tableau

8, we learn that C2 (which is violated by the suboptimal structure) dominates C4 and C5

198
(which are violated by the optimal structure). The initial stratum in (5) above does not

show that C2 dominates C4 and C5 as all constraints there are housed in one stratum, and

thus unranked relative to each other. To show the domination relation exemplified in

Tableau 8 (i.e., C2 dominates C4 and C5), C4 and C5 should be demoted to the next

available stratum. In case there isn’t a next stratum, a new one is created as below:

(6) Initial stratum = {C1, C3, C2}

>>
Stratum 1 = {C4, C5}

Notice that this demotion results not only in having C2 dominate C4 and C5, but also in

having all the other constraints housed in the same stratum with C2 (namely C1 and C3)

dominate C4 and C5. The partially stratified hierarchy in (6) above reads as: while C1,

C2, and C3 are all unranked relative to each other, they each dominate C4 and C5, which

in turn are unranked with regard to each other.

Then we move to tableau 9 to look for positive evidence for further demotion

(i.e., if there are some constraint(s) that dominate(s) others). From there, we learn that

C1, C2, and C3 dominate C4 and C5. Again, this is because C1, C2, and C3 are violated

by the suboptimal form, and thus must outrank C4 and C5 that are violated by the optimal

form. To the attentive reader, it is clear that no changes need be made to (6) above as C1,

C2, and C3 already dominate C4 and C5. So (6) is reproduced as (7) below:

(7) Initial stratum = {C1, C3, C2}

>>
Stratum 1 = {C4, C5}

Moving to Tableau (10), we learn that C1 and C2 dominate C3, C4 and C5. Here

the situation is more complex. Notice that (7) shows that C1 and C2 dominate C4 and C5,

so no further demotion takes place as far as C4 and C5 are concerned. But what (7) does

199
not show is that C1 and C2 dominate C3; they all are housed in the initial stratum. We

then need to demote C3 to the immediately following stratum, which is Stratum 1, of

course. The resulting domination relation after this round will be something like (8)

below:

(8) Initial stratum = {C1, C2}

>>
Stratum 1 = {C4, C5, C3}

But (8) is still not the final hierarchy. Having surveyed the three tableaux in the

first round, we may need to start all over again to see if some evidence has emerged to

improve the existing hierarchy, for it is possible that some evidence may emerge after

each round to improve the existing domination hierarchy. And this is what actually

happens in our example. Whereas Tableau 8 adds no additional information, as it only

requires that C2 dominate C4 and C5, a condition already satisfied in (8), Tableau 9

provides evidence that the existing hierarchy in (8) above can be further modified. Notice

that Tableau 9 requires that C1, C2, and C3 dominate C4 and C5. Whereas (8) above

shows that C1 and C2 dominate C4 and C5, it does not show that C3 dominates C4 and

C5. Therefore, C4 and C5 which are housed in the same stratum with C3 in (8) must be

demoted one step down the hierarchy, resulting in (9) below:

(9) Initial stratum = {C1, C2}

>>
Stratum 1 = {C3}
>>
Stratum 2 = {C4, C5}

(9) reads as: C1 and C2, although they are unranked relative to each other, dominate C3

which, in turn, dominates C4 and C5.

200
 Finally, the whole process is repeated over and over again until no positive

evidence is available to support further demotion of some constraint(s). Once no such

evidence emerges, the result is a fully stratified hierarchy (Tesar and Smolensky 1996:

22), believed to represent the permanent knowledge of the language learner as regards the

constraints that determine the optimal form of some input. If this algorithm yields a

unique ranking, it would then be safe to conclude that “constraint ranking is learnable”

(Tesar 1996: 1).

The hierarchy in (9) can be converted into a tableau to show the domination

relation between all the constraints involved:

Tableau (11)
input C1 C2 C3 C4 C5
"F1
F2
F3
F4
F5

Tableau (11) shows that C1 and C2 are unranked with respect to each other (separated by

dashed lines), but they each dominate all other constraints to the right of them (separated

by solid lines). Likewise, C3, while dominated by the constraints to its left (C1 and C2),

dominates all constraints to its right (C4 and C5). C4 and C5, on the other hand, are

unranked with regard to each other, and do not dominate any constraints. Tableau (11),

then, differs from Tableau (1) which we started with in that the domination relation

between the five constraints is now established. Tableau (1) is reproduced as (12) below

with domination relations being incorporated.

Tableau (12)

201
 input (F) C1 C2 C3 C4 C5
"F1 * ** ***
F2 *! * * *
F3 *! * ** * **
F4 *!* ** **

Tableau (12) clearly reflects the stratified hierarchy in (9) above, namely:

Initial Stratum {C1, C2) >> Stratum 1 {C3} >> Stratum 2 {C4, C5}.

The important point here is that C1 and C2 which are violated by the suboptimal forms

F2, F3 and F4 dominate C3, C4 and C5 which are violated by the optimal form F1,

indicating that the violation of C3, C4 and C5 is never as serious as the violation of C1

and C2. In other words, C1 and C2 are higher-ranking constraints, but C3, C4 and C5 are

lower-ranking constraints.

4.3 Constraint ranking in Arabic and English

4.3.1 Basic facts and generalizations

The constraint ranking in (77) in Chapter Three above communicates some basic facts

and generalizations regarding the domination relations among the constraints that interact

to yield the correct stress patterns in Arabic. The constraint ranking in (77) is reproduced

as (10) below:

202
 (10) Constraint Ranking
TROCHAIC, MAIN-RIGHT, Lx=Pr, *NON-PRIMARY >> NONFINAL(σµµ) >>
PARSEσ >> NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >>
ALL-FEET-RIGHT

This ranking is motivated by the following generalizations

I. TROCHAIC, MAIN-RIGHT, Lx=Pr, and *NON-PRIMARY are undominated in

Arabic:

a. TROCHAIC must be undominated. (jaa)(dala)<tu> is a better parse

than(jaa)(dala)<tu>.
b. MAIN-RIGHT must be undominated. (jaa)(dala)<tu> is a better parse
than (jaa)(dala)<tu>
c. Lx=Pr must dominate NONFINALσ. (min) is a better parse than min.
d. *NONPRIMARY must be undominated. (jaa)(dala)<tu> is a better
parse than (jaa)(dala)<tu>162
III. The other constraints, although they are all dominated by higher ranking
constraints, are ranked relative to each other:

f. NONFINAL(σµµ) must dominate WSP.(baT)Tah is a better parse than

(baT)(Tah).
g. NONFINAL(σµµ) must dominate FOOT-BINARITY-µ. (i)bir is a
better parse than i(bir), and (ra)ma is a better parse than (rama).
h. NONFINAL(σµµ) must dominate PARSE-σ. (i)bir is a better parse
than (i)(bir), and (ka.ta)ba is a better parse than (ka.ta)(ba)
i PARSEσ must dominate NONFINAL(σµµµ). (šan)(taat) is a better parse
than (šan)taat.
j PARSE-σ must dominate FOOT-BINARITY. (i)<bir> is a better
parse than Hi<bir>, and (kata)(ba)<ha> is a better parse than
(kata)ba<ha>.
k ALL-FEET-LEFT must dominate ALL-FEET-RIGHT.
(ad)(wiya)(tu)<hu> is a better parse than (ad)(wi)(yatu)<hu>

As for English, the following constraint ranking is obtained:

(11) Constraint ranking in English

162
Notice that Lx=Pr and WSP(σµµµ), for example, do not only dominate NONFINAL, but they also
dominate all the constraints that are dominated by NONFINAL. This is called transitivity of ranking in
which we assume that some ranking exists between all constraints even those which do not interact (see
Kager 1999: 21).

203
 NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ)
>> NONFINAL, REDUCTION >> WSP(<µµ), FOOT-BINARITY(σ),
BIMORAIC, ALL-FEET-RIGHT >> PARSE(σ) >> *NON-PRIMARY

The ranking of constraints in (11) above, while it does not account for every single token

in English, captures the basic generalizations concerning the English stress system,

especially in monomorphemic words. These generalizations include:

I. Lx=Pr, FAITH, MAIN-RIGHT, and TROCHAIC are all undominated in

English.

a. Lx=Pr is undominated because each and every content word has to

have at least one foot constructed over it; otherwise, that word will
surface stressless (Hence it won’t have a head). A(meri)ca is then a
better parse than America.
b. Lx=Pr must dominate NONFINAL so as to avoid rendering the whole
word unfooted (i.e. extrametrical). (cat) is then a better parse than cat.
c. Lx=Pr must dominate FOOT-BINARITY. The need to have at least
one foot constructed over monosyllabic words is greater than the need
to have the foot fit a certain shape. (hat) is a always optimal.
d. FAITH must dominate NONFINAL so that a final closed syllable can
be stressed. (bassi)(net) is a better parse than (bassi)net.
e. FAITH must dominate REDUCTION, so that a final closed syllable
that is stressed does not undergo reduction. (bassi)(net) is a better
parse than (bassi)(nt).
f. MAIN-RIGHT must dominate WSP so as to allow stress to fall on one
of the last three syllables. (bassi)(net) is a better parse than
(bassi)(net).
g. TROCHAIC must dominate IAMBIC, so as to allow feet to be left-
headed. A(meri)ca is a better parse than A(meri)ca.

II All the other constraints are dominated. Yet some of them are ranked
relative to each other:

h. WSP(≥µµ) must dominate NONFINAL, so as to allow words ending

with long vowels to receive stress on that syllable. ag(ree) is a better
parse than (ag)ree; and to allow stress to surface on ultimate syllables
that are closed by a consonant cluster. a(larm) is a better parse than
(a)larm.
i. NONFINAL must dominate WSP(<µµ) so as to allow a closed
syllable in final position to be skipped over: (ani)mal is a better parse

204
 than (ani)(mal); and to allow a syllable that is closed by a consonant
cluster that contributes less than two moras to surface stressless:
(nove)list is a better parse than (nove)(list)
j. NONFINAL must dominate ALL-FEET-RIGHT, so as to allow stress
to fall on an antepenult syllable. A(meri)ca is a better parse than
Ame(rica).
k. NONFINAL must dominate FOOT-BINARITY. (ham)mock is a better
parse than (hammock); and A(meri)ca is a better parse than
(Ame)(rica).
l. REDUCTION must dominate BIMORACITY, so as to allow the
nucleus of unstressed syllables to reduce to schwa. hammck is a
better parse than hammock.
m. WSP(≥µµ) must dominate FOOT-BINARITY, so as to allow a closed
syllable that contributes more than one mora to be stressed in penult
position. A(gen)da is a better parse than (agen)da.

n. FOOT-BINARITY must dominate PARSEσ. A(meri)ca is a better

parse than (A)(meri)ca or (Ame)(ri)ca.
o. PARSEσ must dominate ALL-FEET-RIGHT. (mina)(ret) is a better
parse than mina(ret).

Before we embark on comparing the constraint ranking in Arabic with the

constraint ranking in English, three points are in order. First, some of the constraints that

are required to account for the place of stress in one language are not invoked to account

for the place of stress in the other language. For example, whereas NEUTRAL,

REDUCTION and BIMORAIC are needed to account for the place of main word-stress

in English, they are not invoked to account for the place of main word-stress in Arabic.

This raises the question whether the constraint set is really universal. The proposal

adopted throughout is that the constraint set is literally present in each and every

language, and that language variation results from different rankings of the set of

constraints. In terms of constraint ranking, what this basically means is that the

constraints which are not invoked to account for the place of stress in one language are

205
ranked so low that their impact is not even noticeable. If we want to incorporate

NEUTRAL, REDUCTION and BIMORAIC, for example, into the constraint ranking of

Arabic, (10) above becomes (12) below:

(12) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, *NON-PRIMARY >> NONFINAL(σµµ) >>
PARSEσ >> NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >>
ALL-FEET-RIGHT >>…>> NEUTRAL, REDUCTION, BIMORAIC

The second point concerns the interpretation of constraints. The most relevant

example is FOOT-BINARITY, the constraint that defines the shape of the foot in the two

languages. In 3.3.4 above, it is argued that FOOT-BINARITY in Arabic is interpreted

under a moraic analysis (i.e. FOOT-BINARITYµ). In Arabic, foot formation is sensitive

to the weight of the syllable, whereby each foot consists of two moras. On the other hand,

English requires that each foot consist of two syllables irrespective of their weight.

FOOT-BINARITY is, then, interpreted under a syllabic analysis (i.e. FOOT-

BINARITYσ) in English. Degenerate feet are only sanctioned by higher ranking

constraints such as WSP and FAITH.

For puposes of comparison, we may want to assume that there is a family of

constraints as far as foot formation is concerned (e.g. FOOT-BINARITYµ, FOOT-

BINARITYσ, etc.). Because FOOT-BINARITYσ is perhaps irrelevant to account for the

place of main word-stress in Arabic, it must be ranked low enough that it does not

dominate any other constraint. The domination hierarchy in (12) above now looks like

(13) below:

(13) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, *NON-PRIMARY >> NONFINAL(σµµ) >>
PARSEσ >> NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >>

206
 ALL-FEET-RIGHT >>…>> NEUTRAL, REDUCTION, BIMORACITY,
FOOT-BINARITYσ

Likewise, because FOOT-BINARITYµ is irrelevant when accounting for the

English stress system, (11) above becomes like (14) below:

(14) Constraint ranking In English

NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ)
>> NONFINAL, REDUCTION >> WSP(<µµ), FOOT-BINARITY(σ),
BIMORAIC >> PARSEσ >> ALL-FEET-RIGHT >> *NON-PRIMARY
FOOT-BINARITYµ

A third point that is worthy of mention here is that some constraints are

parameterized in one language, but not in the other language. The most important

examples are: (1) the parameterization of WSP in English, and (2) the parameterization of

NONFINAL in Arabic.

In order to account for stress placement in English, WSP is parameterized as

follows: WSP(≥µµ) and WSP(<µµ).163 For purpose of comparison with Arabic, we treat

WSP(≥µµ), and WSP(<µµ) as independent constraints. Again, because we can make do

without these constraints in Arabic, they are housed in the lowest rank in the domination

hierarchy. The constraint ranking in (13) can thus be further modified as in (15) below:

(15) Constraint Ranking

TROCHAIC, MAIN-RIGHT, Lx=Pr, *NON-PRIMARY >> NONFINAL(σµµ) >>
PARSEσ >> NONFINAL(σµµµ), FOOT-BINARITYµ >> ALL-FEET-LEFT >>

163
A cross-linguistic comparison may not be well-motivated in this regard for the simple reason that weight
is not uniform in the two languages. It is true that the notion of the mora is invoked to make it unitary, but
still what counts as a mora is language-specific. For example, whereas all nuclear and post-nuclear
segments each contribute one mora to the weight of the syllable in Arabic, nuclear and post nuclear
segments are divided into subsets depending on how many moras each subset contributes to the weight of
the syllable in English. Whereas a syllable that ends with a long vowel contributes precisely two moras in
Arabic, which is of course not enough to make it stressed in ultimate position (e.g. kitabuhumaa), a syllable
that ends with long vowel contributes enough moras to make it stressed in final position in English (e.g.
kangaroo). Examples like this one suggest that these parameterized constraints be viewed as independent
constraints.

207
 ALL-FEET-RIGHT >>…>> NEUTRAL, REDUCTION, BIMORACITY,
FOOT-BINARITYσ, WSP(≥µµ), and WSP(<µµ)

As for the parameterization of NONFINAL in Arabic, the constraint ranking in

(15) above shows that final light and heavy syllables are invisible to parsing in Arabic,

but final superheavy syllables are not. A distinction is drawn between final light and

heavy syllables on the one hand, and final superheavy syllables on the other. In terms of

constraint ranking, whereas NONFINAL(σµµ) dominates PARSEσ, NONFINAL(σµµµ) is

dominated by PARSEσ. All in all, the stressing of a final syllable in Arabic is contingent

upon whether or not that syllable is parsed. In English, however, all final syllables are

invisible to parsing164, and the stressing of a final syllable is only sanctioned by higher

ranking constraint. This suggests that the parameterized NONFINAL(σµµ) and

NONFINAL(σµµµ) are low ranked in English. The constraint ranking in (14) above is

now as in (16) below:

(16) Constraint ranking

NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC >> WSP(≥µµ) >>
NONFINALσ, REDUCTION >> WSP(<µµ), FOOT-BINARITYσ, BIMORAIC
>> PARSE(σ) >> ALL-FEET-RIGHT >> *NON-PRIMARY, FOOT-
BINARITYµ, NONFINAL(σµµ), NONFINAL(σµµµ)

To sum up, whereas REDUCTION, BIMORAIC, FOOT-BINARITYσ,

WSP(≥µµ), and WSP(<µµ) are all ranked so low in Arabic that they do not dominate

other constraints, *NON-PRIMARY, FOOT-BINARITYµ, NONFINAL(σµµ),

NONFINAL(σµµµ) are dominated by the entire set of constraints in English. The reason

164
The fact that some final syllables receive stress in English is accounted for by the proposal that
NONFINAL is dominated by some higher ranking constraints such as WSP(≥µµ) (e.g. kangaroo, alarm),
FAITH (e.g. bassinet), etc.

208
we incorporate all of the constraints into the domination hierarchy is to maintain the basic

premise of Optimality Theory, that the set of constraints is universal and that language

variation stems from the fact that these universal constraints are ranked differently in

different languages. In other words, the suggestion that REDUCTION, BIMORACITY,

FOOT-BINARITY(σ) WSP(≥µµ), and WSP(<µµ) need not be considered when

accounting for the place of stress in Arabic does not mean that they are not part of the

grammar of Arabic; it simply means that they are ranked so low that their impact is not

noticeable. Likewise, the fact that NONFINAL is parameterized in Arabic but not in

English should not make us lose the basic insight of OT that language variation stems

from different rankings of the set of universal and violable constraints.

4.3.2 Stress errors in the English of native speakers of Arabic

(15) and (16) above show that there are basic differences between Arabic and English as

far as the ranking of constraints is concerned. These differences, it is hypothesized, are

major sources of stress misplacement in the English of native speakers of Arabic. In this

section, we set out to compare the constraint ranking in Arabic with the constraint

ranking in English in an attempt to categorize stress errors made by native speakers of

Arabic when learning English. Our goal is two-fold. One, we predict the kinds of errors

that are observed in the English of native speakers of Arabic; and two, we envisage what

those learners need to do to make their English more native-like as far as prominence is

concerned.

209
4.3.2.a Lx=Pr, MAIN-RIGHT, TROCHAIC

Whereas (15) shows that TROCHAIC, MAIN-RIGHT, Lx=Pr, *NON-PRIMARY are

undominated in Arabic, (16) shows that NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC are all undominated in English. This basically means that English and

Arabic have three things in common:

(1) Lx=Pr is undominated in both Arabic and English

(2) TROCHAIC is undominated in both Arabic and English

(3) MAIN-RIGHT is undominated in both Arabic and English

The fact that Lx=Pr is undominated in both languages is easy to explain. There is

a minimum size restriction on content words in both languages. According to (17) below,

every prosodic word (PW), as a high-level prosodic category, must have the foot, the next

available lower-level prosodic category, as its head. The foot, in turn, has to have the

syllable as its head, and so on:

(17) PW = Prosodic Word

F = Foot

σ = Syllable

µ = mora

210
If (17) were violated, words would surface headless, i.e., with no stress whatsoever. As

no lexical word in the two languages surfaces stressless, Lx=Pr (i.e., the constraint that

requires each content word to minimally contain a foot that functions as the head of the

prosodic word) must be undominated in the two languages. Consider Tableau (13) for the

Arabic word /min/ and Tableau (14) for the English word /hæt/, and see how the

candidate form which violates Lx=Pr fails to surface as the optimal form:

Tableau (13)
Input: /min/ Lx=Pr
a- min *!
b- " (min)

Tableau (14)
Input: /hæt/ Lx=Pr
a- hæt *!
b- " (hæt)

The pedagogical implication of this is straightforward. Native speakers of Arabic

who learn English do not experience problems regarding the fact that every English

content word must receive main stress. In fact, none of the studies that I have come

across report that subjects produce any English word unstressed165.

165
Native speakers of Arabic encounter two problems in this regard. The first problem is related to the
acquistion of English weak forms (e.g., frm, kn, etc). For, in Arabic no distinction is made between
content words and grammatical words as far as word prominence is concerned. Each word, irrespective of
its content, must receive main stress. In English, however, form words such as prepositions, determiners,
etc. may, due to REDUCTION, surface stressless. In OT terms, what this means is that Lx=Pr is applicable
to all Arabic words irrespective of their lexical status, but Lx=Pr is only applicable to lexical words in
English. However, this is seen as falling outside the scope of this research project.
The second problem is that most studies on the acquistion of English word-stress by native speakers of
Arabic report that single-stressed compounds (e.g. blackboard, breakfast) are produced with two main
word-stresses (e.g. blackboard, breakfast). This may be so because those learners view these words as
consisting of two independent words, each with primary stress. Hence, each word has to have at least one
foot constructed over it, which, due to undominated Lx=Pr, becomes the head foot.

211
 The problem that native speakers of Arabic encounter when learning the English

stress system is not whether or not each content word must surface with main stress, but

where in the word stress falls. Monosyllabic content words pose no problems with respect

to stress placement, as the place of stress in these words is predictable.166 In OT terms,

the constraint that requires every monosyllabic and polysyllabic word to surface with

main word-stress is undominated in the two languages. Such being the case, the language

learner is never exposed to data that cause re-ranking of this constraint. On the contrary,

the fact that this constraint is undominated in both English and Arabic is a major source

of positive transfer.

The second constraint that is undominated in both languages is MAIN-RIGHT.

Recall that in both Arabic and English, main stress tends to occur as close to the right

edge of the word as possible. Stress falls on either the ultimate, penultimate, or

antepenultimate syllable (called the three-syllable window), but never on a pre-

antepenultimate syllable in either language. In addition, the penult and antepenult are

stressed iff (if and only if) the final syllable is considered extrametrical (i.e. unfooted).

Once the final syllable gets footed, it surfaces stressed in both languages. This suggests

that the foot closest to the right edge of the word should be the one that contains the head

syllable in both languages. This, too, should be a major source of positive transfer

Not only do the two languages choose to have the foot closest to the right edge of

the word as the head of the prosodic word, but they also choose to have the head syllable

in each foot coincide with its left-edge. In OT terms, this means that TROCHAIC is

166
In addition, since in both languages Lx=Pr dominates NONFINAL, no problem arises with regard to the
fact that each and every content word (including monosyllabic ones) must surface with main stress.

212
undominated in both Arabic and English. This, too, should foster the acquistion of

English word-stress by native speakers of Arabic.

All in all, the fact that Lx=Pr, MAIN-RIGHT and TROCHAIC are undominated

in both languages helps explain why words English words like hat, alarm, agenda, rigid

and pedigree are never reported as problematic to native speakers of Arabic with respect

to stress placement. In all of these words, only one foot is constructed over the word

which, due to MAIN-RIGHT, functions as the head of the word; and that foot contains

the head syllable which, due to undominated TROCHAIC, coincides with the left edge of

the foot:

(18)
(hat)
a(larm)
a(gen)da
(ri)gid
(pedi)gree167

In [hat], for example, only one foot is constructed over the word. In [alarm], the

the rightmost foot is constructed over the final syllable. In [agenda], the rightmost foot is

constructed over the heavy penult. In [rigid], the rightmost foot is constructed over the

light penult. In [pedigree], the rightmost foot is constructed over the the penultimate and

the antepenultimate syllable which, due to TROCHIAC, surfaces left-headed. All of this

is true for Arabic. If these words were in Arabic, they would surface with the same stress

patterns.168

167
We have seen earlier why these words are parsed this way.
168
It is true that in Arabic a foot is constructed over the heavy penult because of FOOT-BINARITYµ, and
in English because of WSP(≥µµ), but the important point here is that in both cases the end result is the
same. That is, the rightmost foot is constructed over the heavy penult. Once this foot is constructed, the
undominated MAIN-RIGHT and TROCHAIC come into play to make the word surface with penult stress
in the two languages.

213
 In terms of stress acquisition, the question that arises here is: What should the

language learner do in this case?

As far as Lx=Pr, MAIN-RIGHT, and TROCHAIC are concerned, no evidence has

emerged to motivate their demotion to a lower position. The learner is not exposed to

data that require the demotion of any one of these constraints to a lower position. Hence,

any candidate form that violates any of these constraints fails to surface as optimal in

both Arabic and English. All in all, no English data should contain evidence as to

disconfirm the ranking of Lx=Pr, MAIN-RIGHT and TROCHAIC that those learners

have already acquired as part of the grammar of their native language. Such being the

case, native speakers of Arabic get the correct stress patterns in hat, alarm, agenda, rigid

and pedigree for free. Let us see how this should work out.

When learning English, native speakers of Arabic start off with the assumption

that all of the universal and violable constraints are unranked relative to each other169. Let

us call this the Initial Stratum:

(19) Initial Stratum

{NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT, TROCHAIC, WSP(≥µµ),
NONFINALσ, REDUCTION, WSP(<µµ), FOOT-BINARITYσ,
BIMORAICITY, ALL-FEET-RIGHT, PARSE(σ), *NON-PRIMARY, FOOT-
BINARITYµ, NONFINAL(σµµ), NONFINAL(σµµµ)}

As no crucial evidence concerning the re-ranking of Lx=Pr, MAIN-RIGHT and

TROCHAIC emerges, the ranking of these constraints in (19) above does not get altered.

Apart from this, things get fuzzier because the remaining constraints are ranked

differently in both English and Arabic. Production errors of all sorts are therefore much

169
Notice that whether native speakers of Arabic start off with the proposal that the constraints are
unranked relative to each other or that they are ranked as in their native language, the end result will be the
same, as what matters here is the idea of constraint demotion.

214
more likely to occur. In what follows, we set out to categorize some of these errors that

depend on the constraint rankings obtained in (15) and (16) above; and then suggest what

those Arabic learners of English need to do in order to acquire the target form.

4.3.2.b FAITH and NONFINAL

One type of stress error that is often reported in the English of native speakers of

Arabic is the place of stress in words like [bassinet]. Almost all the studies that I have

come across claim that native speakers of Arabic experience great difficulty in acquiring

the stress pattern in English words where stress falls on a final syllable that is not

superheavy (i.e. /cvvc/ or /cvcc/).

Recall that [bassinet] is stressed in English because of FAITH. Native speakers of

Arabic are expected to experience some difficulty in acquiring its stress pattern because

FAITH is ranked so low in Arabic. When confronted with words like bassinet, souvenir,

toucanet, etc. native speakers of Arabic find out that FAITH is a high ranking constraint

in English; it must stay undominated. Examples like these show that the ranking of

FAITH in (19) above does not get altered, either.

What else should native speakers of Arabic acquire when confronted by words

like [bassinet]? One important thing is that in English, unlike in their native language,

FAITH dominates NONFINAL. One possible reason why native speakers of Arabic fail

to produce /bassinet/ with the correct stress pattern is that FAITH is dominated by

NONFINAL in their native language.

215
 Previous studies have all assumed that native speakers of Arabic fail to produce

the correct stress pattern in /bassinet/ because stress here falls on an ultimate syllable that

is not superheavy. The analysis above suggests that this may not be the case at all, for in

English there is more to it than just that. In English, the need to satisfy FAITH is more

important than the need to satisfy NONFINAL. In Arabic, however, it is the other way

around: NONFINAL dominates FAITH. Applying the constraint ranking in Arabic to

/bassinet/ results in penultimate stress:

Tableau (15)
input /bassinet/ NF Pσ NF FBµ AFL AFR FA
(σµµ) (σµµµ)
a- (bassi)(net) *! ** *
b- (bassi)net *! * *
c-!" (bas)(si)net * * ** *

Native speakers of Arabic who want to learn the correct stress pattern of bassinet,

souvenir, toucanet, etc. need to learn to demote NONFINAL170 at least one step below

FAITH. The constraint ranking in (19) above now looks like (20) below, where

NONFINALσ is demoted one step down the hierarchy:

(20) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), REDUCTION, WSP(<µµ),
FOOT-BINARITYσ, BIMORAICITY, ALL-FEET-
RIGHT, PARSE(σ), *NON-PRIMARY, FOOT-
BINARITYµ}

>>

Stratum 1 {NONFINALσ}

170
Notice that since in English NONFINAL refers to each and every final syllable irrespective of its
internal structure, i.e. both NONFINAL(σµµ) and NONFINAL(σµµµ). We use NONFINALσ to refer to
both of them.

216
4.3.2.c REDUCTION

Another obvious observation concerning the English of native speakers of Arabic is that

unstressed syllables are not reduced. In 1.6.1 above, it is shown that most previous

studies report that native speakers of Arabic who learn English almost always fail to

reduce the nuclei of unstressed syllables. They retain the full value of the vowel

irrespective of its stress status (e.g., /several/, /preparatori/, /basikali/, /legendari/ for

/sevrl/, /prepartori/ basikli/, /legendri/, etc.).171

In English, the final closed syllable of /hammock/, for example, surfaces

stressless because of REDUCTION which requires the nucleus of that syllable to reduce

to a schwa (e.g. [animl], [hammk], etc.):

Tableau (16)
Input: /hammock / Lx=Pr MR TR FA RED NF FBσ AFR AFL
a- (ham)mock *! * * **
b- (hammck) *!
c-" (ham)mck * * **

The source of confusion for native speakers of Arabic comes from the fact that in

their native language, REDUCTION is dominated by the entire set of constraints. That is,

it is ranked so low in the domination hierarchy that it becomes irrelevant to the

171
When applying the constraint ranking particular to Arabic, all of these words, with the exception of
several, surface with incorrect stress patterns: preparatory, basically, and legendary.

217
determination of the candidate form that ultimately wins the competition. Hence, all

vowels in Arabic remain unreduced.

On the other hand, in English, REDUCTION is dominated by FAITH, so that

words like [bassinet] escape REDUCTION:

Tableau (17)
input /bassinet/ Lx=Pr MR TR FA RED NF FB AFR AFL
a-" (bassi)(net) * * * * **
b- (bassi)(nt) *! * * * **

Otherwise, /bassinet/, like /hammock/, surfaces as in (c) below:

Tableau (18)
Input: /bassinet/ Lx=Pr MR TR RED NF FB AFR AFL
a- (bassi)(net) *! * * * **
b- (bassi)(nt) *! * * **
c-!" (bassi) nt *

To native speakers of Arabic who make use of their native language’s ranking of

the constraints, the form that satisfies NONFINAL surfaces as optimal even if it violates

REDUCTION:

Tableau (19)
Input: /hammock/ NF Pσ NF FBµ AFL AFR FA RED
(σµµ) (σµµµ)
a- (ham)(mock) *! ** * * *
b-" (ham)mock * * * ** *

According to Tableau (19) above, native speakers of Arabic produce /hammock/ with the

correct stress pattern, but they fail to reduce the nucleus of the final syllable.

So far, two features are obvious in the English of native speakers of Arabic: (1)

there are no exceptional cases that are accented underlyingly; and (2) unstressed vowels

do not reduce. This is so because FAITH and REDUCTION are ranked lowest in the

218
hierarchy in Arabic, and thus, their satisfaction or violation is irrelevant in determing the

actual output form.

For acquisition purposes, native speakers of Arabic need to learn two things to

acquire the correct stress pattern in /hammock/ and /bassinet/. First, they need to learn

that FAITH is undominated in English, and so should not be demoted to a lower position.

Second, they need to learn that REDUCTION is dominated by FAITH. In other words,

they need to demote REDUCTION one step below FAITH. The domination hierarchy in

(20) above now looks like (21) below:

(21) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), WSP(<µµ), FOOT-
BINARITYσ, BIMORAICITY, ALL-FEET-RIGHT,
PARSE(σ), *NON-PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}172

In terms of the stages of language acquistion, consider what native speakers of

Arabic need to go through to acquire words like hammock and bassinet:

Tableau (20)
Input: /bassinet/ Lx=Pr MR TR FA RED NF FB AFR AFL
a- (bassi)net *! *! * * **
b- (bassi)nt *! * * * **
c-" (bassi)(net)173 * * * * **

172
Notice that (22) shows not only that FAITH dominates REDUCTION and NONFINAL, but also that the
constraints that are housed in the initial stratum with FAITH dominate all the constraints that are housed in
stratum 1. In other words, (22) shows that REDUCTION is still dominated by all the constraints in the
Initial Stratum. Unless all other constraints are housed in their appropriate strata so that REDUCTION
happens to outrank some of them, the learner will fail to reduce the nuclei of unstressed syllable to a schwa.
Hence, there are still examples in the language that require the existing domination hierarchy in (21) to be
modified.

219
 When applying the constraint ranking of their native language, native speakers of

Arabic produce (a), where the final syllable is unfooted, and not reduced. In order to

produce the correct target form, they need to through at least two stages. First, they need

to learn that a final closed syllable, due to REDUCTION, surfaces with a schwa (Cf.

output b in Tableau 20 above). Second, they need to learn that REDUCTION is

outranked by FAITH, so that a bimoraic final syllable with underlying stress surfaces

stressed and unreduced (Cf. output c in Tableau 20 above).

Language learners who produce (b) in Tableau 20 above are at least one step

ahead of those who produce (a), and they are only one step away from the target form (c).

This conclusion can be supported by the fact that the learners who produce (b) should

produce the final syllable of [hammock] with a schwa:

Tableau (21)
Input: /hammock / Lx=Pr MR TR FA RED NF FB AFR AFL
a- (ham)mock *! * * **
b- " (ham)mck * * **

Learners who have not yet reached the point where they can produce unstressed

syllables with a schwa still need to make this move before they learn that REDUCTION

is sometimes violated so as to satisfy higher ranking constraints such as FAITH. That is,

they have to deduce that in English FAITH dominates REDUCTION and NONFINAL.

173
As shown in (3.3.5) above, syllables that are stressed due to FAITH surface as bimoraic, so that they
satisfy BIMORAIC.

220
4.3.2.d BIMORAIC

Another source of difficulty for native speakers of Arabic when learning English comes

from the fact that each syllable in English must at least be bimoraic, and that monomoraic

or zero-moraic syllables (e.g. animl, hammck, etc.] are only sanctioned by higher-

ranking constraints such as REDUCTION.

Implicit in this is the suggestion that native speakers of Arabic need to learn the

English way of counting moras. According to Hammond (1999: 103-148), English

vowels, diphthongs and coda consonants are divided into subsets depending on the

number of moras each subset contributes to the weight of the syllable: Some of them

must contribute one mora, some of them optionally contribute one mora, some of them

contribute two moras, some of them contribute three moras, and some of them contribute

none. The following chart is taken from Hammond (1999: 137):

(22) Intrinsic mora count

Lax vowels µ
Tense vowels µµ
[ay,yu] µµ
[aw, y] µµµ
coronals (µ)
Noncoronals µ
[ž, ŋ] µµ
[, r] Ø

221
 [] Ø

This is surely one of the most basic differences between English and Arabic. In

Arabic, each nuclear and post-nuclear segment contributes precisely one mora to the

weight of the syllable. This is probably why native speakers of Arabic produce words like

/novelist/, /expert/, /collect/, /defrost/, etc. with ultimate stress (e.g. novelist, expert,

collect, defrost, etc.). To those learners, the final syllable in each of these words weighs

three moras, and so surfaces with main stress. In addition, a /cv/ syllable type weighs

only one mora in Arabic, irrespective of the nature of that vowel. In English, however, all

syllables are at least bimoraic, and the realization of monomoraic or even zero moraic

syllables can only be sanctioned by higher ranking constraints.

Unlike English, BIMORACITY is ranked so low in Arabic that it is dominated by

the entire set of constraints. This entails that native speakers of Arabic when learning

English need to make sure that BIMORAIC is high ranking; it is demoted only one step

below REDUCTION.

(23) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), WSP(<µµ), FOOT-
BINARITYσ, ALL-FEET-RIGHT, PARSE(σ), *NON-
PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

222
 Stratum 2 {BIMORAIC}174

4.3.2.e WSP

In English, a bimoraic syllable gets footed in final position, and thus surfaces

stressed (e.g. bassinet). This is achieved by having WSP(≥µµ) dominate NONFINALσ.

In Arabic, on the other hand, a final syllable that weighs two moras or less is not footed at

all.175 This is achieved by having NONFINAL(µµ) dominate PARSEσ in Arabic.176

For acquistion purposes, native speakers of Arabic who learn English need to

learn two things in this respect. First, they need to learn the English intrinsic mora

count.177 Second, they need to demote NONFINALσ at least one step below WSP(≥µµ)

When those learners acquire the English way of counting moras, they will no

longer produce words like /novelist/, /expert/, /collect/, etc. with ultimate stress. Rather,

they will learn that the final syllable in each of these words weighs less than two moras.

Hammond (1999: 137) argues that final coronals do not necessarily contribute to the

174
Notice that as long as the other constraints are not demoted to their proper position in the hierarchy,
learners still fail to get the idea that BIMORAIC is relevant when accounting for English stress. For the
present hierarchy still shows that BIMORAIC is dominated by the entire set of constraint.
175
Recall that final light (CV) and heavy syllables (CVC, CVV) are never stressed in final position in
Arabic.
176
What is interesting here is for a final syllable to get stressed in polysyllabic words in English, it must
weigh two moras. In Arabic, however, a final syllable is never stressed unless it weighs three moras. This
basically entails that light and heavy syllables pair together as opposed to superheavy syllables in Arabic,
but heavy and superheavy syllables pair together in English. In other words, in Arabic we have light and
heavy syllables in one category and superheavy syllables in another category; but in English we have light
syllables in one category and heavy and superheavy in another category.
177
This is seen as falling outside the domain of this reaseach as it requires a high-ranking constraint that
works at the segmental level. The task of this constraint would be to assign each segment its weight in
terms of moras.

223
weight of the syllable. The final syllable of /novelist/ does not count as two moras, and so

fails to surface with main word-stress.

In terms of constraint ranking, NONFINAL dominates WSP(<µµ), and thus stress

falls on either the penult or the antepenult in these words. Consider the following Tableau

for /novelist/:

Tableau (22)
Input: /novelist/ Lx=Pr MR TR FA WSP NF WSP AFR AFL
(≥µµ) (<µµ)
a- " (nove)list * *
b- (nove)(list) *! * **

On the other hand, the final syllable that weighs at least two moras surfaces

stressed because WSP(≥µµ) dominates NONFINAL. Consider the following Tableau for

/alarm/:

Tableau (23)
Input: /alarm/ Lx=Pr MR TR FA WSP NF WSP FBσ AF AFL
(≥µµ) (<µµ) R
a- (a)larm *! * *
b- "(larm) * * *

According to Tableau (23) above, Output (b) wins the competition despite the fact that it

violates NONFINAL, because its competitor violates a higher-ranking constraint, namely

WSP(≥µµ).

The domination hierarchy in (23) above does not show the ranking of WSP(≥µµ)

and WSP(<µµ) relative to NONFINALσ, for both are still housed in the same stratum.

Recall that because a final syllable that weighs at least two moras surfaces stressed in

English and a final syllable that weighs less than two moras surfaces stressless in English,

WSP(≥µµ) must dominate NONFINAL, but WSP(<µµ) must be dominated by

224
NONFINAL. When learning English, native speakers of Arabic need to learn to demote

WSP(<µµ) at least one step below NONFINAL:

(23) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ, ALL-
FEET-RIGHT, PARSE(σ), *NON-PRIMARY, FOOT-
BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

Stratum 2 { WSP(<µµ), BIMORAIC}

Although native speakers of Arabic do not encounter a problem with respect to

final syllables that weigh less than two moras, they face great difficulty in stressing a

final syllable that weighs two moras (e.g. bassinet, souvenir, toucanet, etc.). This is so

because a final heavy syllable is not stressed in Arabic (i.e. (mu)(dar)ris is a better parse

than (mu)(dar)(ris) ). Native speakers of Arabic then need to be exposed to data (e.g.

bassinet, souvenir, toucanet, kangaroo, agree, etc.) that make them acquire that

WSP(≥µµ) dominates NONFINAL.

225
4.3.2.f ALL-FEET-RIGHT and ALL-FEET-LEFT178

As for the directionality of footing, our analysis has shown that Arabic and

English are similar with respect to the ordering of ALL-FEET-RIGHT and ALL-FEET-

LEFT relative to NONFINAL, but they are very different with respect to the ordering of

ALL-FEET-RIGHT and ALL-FEET-LEFT relative to each other. In both Arabic and

English, NONFINAL dominates ALL-FEET-RIGHT and ALL-FEET-LEFT, something

that should facilitate learning. However, one source of stress errors in the English of

native speakers of Arabic comes from the claim that feet are parsed from left to right in

Arabic (i.e. ALL-FEET-LEFT >> ALL-FEET-RIGHT), but from right to left in English

(i.e. ALL-FEET-RIGHT >> ALL-FEET-LEFT).

Notice that for almost all the examples introduced so far, it makes no difference

whether the word is parsed from right to left or from left to right, because only one foot is

constructed over the word. All that the previous examples show as far as the parsing of

syllables into feet is concerned is that NONFINAL must dominate ALL-FEET-RIGHT

and ALL-FEET-LEFT for both Arabic and English. Let us recall the evidence why
178
The ordering of ALL-FEET-LEFT and ALL-FEET-RIGHT relative to each other is partially determined
by UG in that one of them must dominate the other. This is so because parsing has to take place either from
right to left or from left to right, but not in both directions.

226
NONFINAL must dominate ALL-FEET-RIGHT and ALL-FEET-LEFT in the two

languages before we embark on demonstrating how the ranking of ALL-FEET-RIGHT

and ALL-FEET-LEFT relative to each other causes transfer errors in the English of

native Arabic speakers.

In Arabic, NONFINAL must dominate ALL-FEET-LEFT and ALL-FEET-

RIGHT; otherwise, it would be impossible to account for antepenultimate stress.179 If

NONFINAL were outranked by ALL-FEET-LEFT, the final syllable would pair with the

penult or would make up an independent foot by itself as in (25) below:

(25)
(a) (jada)(latu)
(b) *(kaa)(taba)(ha)
(c) *(?ad)(wiya)(tuhu)(maa)

If the need to have the leftmost foot coincide with the left edge of the word in

Arabic outweighs the need to have the last syllable unfooted, then stress, due to

TROCHAIC and MAIN-RIGHT, always terminates on the penult as in (25a) above, or

the final syllable as in (25b-c), but never on the antepenult. But if NONFINAL is said to

dominate ALL-FEET-LEFT, it becomes straightforward to account for penultimate as

well as antepenultimate stress. Consider how the correct stress pattern can be derived if

NONFINAL is a higher-ranking constraint (at least higher ranked than ALL-FEET-

LEFT) in Arabic:

(26)
(a) (jada)(la)<tu>
(b) (kaa)(taba)<ha>
(c) (?ad)(wiya)(tuhu)<maa>

179
Recall that ALL-FEET-LEFT >> ALL-FEET-RIGHT in Arabic

227
 In English, on the other hand, NONFINAL must dominate ALL-FEET-RIGHT180,

so as to account for antepenult stress:

(27) (ani)mal

Otherwise, stress, due to MAIN-RIGHT and TROCHAIC, would always terminate on

either the ultimate or the penultimate syllable:181

(28)
(a) *(ani)(mal)
(b) *(a)(nimal)

Let us see what the learners need to do to acquire the ranking of NONFINAL

relative to ALL-FEET-RIGHT and ALL-FEET-LEFT. In a nutshell, learners need to

demote both ALL-FEET-RIGHT and ALL-FEET-LEFT one step below NONFINAL.

So, (22) above can be modified as (29) below:

(29) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ,
PARSE(σ), *NON-PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

180
Recall that ALL-FEET-RIGHT >> ALL-FEET-LEFT in English
181
The fact that the final syllable sometimes surfaces stressed in both languages can be accounted for by
higher-ranking constraints. As for Arabic, it is shown that a final syllable is stressed in two cases (1) if the
word is a monosyllabic word (e.g. min > min); and (2) if the final syllable is superheavy (e.g. kasaat >
kasaat). Otherwise, the final syllable always remains unstressed. In terms of constraint ranking, stressing a
monosyllabic word is accounted for by having Lx=Pr dominate NONFINAL; and stressing a final
superheavy syllable is accounted for by having PARSEσ dominate NONFINAL(σµµµ).
As for English, the final syllable is stressed in three cases: (1) if the word is monosyllabic (e.g. cat > cat);
(2) if the word ends in a long vowel (e.g. agree > agree); and (3) if the word is underlyingly accented on
the final syllable (e.g. bassinet > bassinet). Otherwise, the final syllable is always unstressed. In terms of
constraint ranking, /cat/ surfaces stressed because Lx=Pr dominates NONFINAL; the final syllable in
/agree/ surfaces stressed because WSP(≥µµ) dominates NONFINAL; and the final syllable in /bassinet/
surfaces stressed because FAITH dominates NONFINAL.

228
 Stratum 2 {ALL-FEET-RIGHT, ALL-FEET-LEFT, WSP(<µµ),
BIMORAIC}

As shown above, native speakers of Arabic who learn English get this step for

free because the acquistion of stress in their native language requires the demotion of

ALL-FEET-RIGHT and ALL-FEET-LEFT below NONFINAL.

However, what the two languages seem not to have in common as far as the

directionality of parsing is concerned is the ordering of ALL-FEET-RIGHT and ALL-

FEET-LEFT relative to each other. In 3.4 above, it is shown that syllables are footed on a

left-to-right basis in Arabic; and in 3.5 above, it is shown that syllables are footed on a

right-to-left basis in English. In other words, whereas ALL-FEET-LEFT dominates ALL-

FEET-RIGHT in Arabic, it is the other way around in English. Words like /America/ are

parsed as (Ame)rica not A(meri)ca by native speakers of Arabic. And this is one of the

major sources of stress errors in the English of native Arabic speakers

In 3.4 above, it is shown that it is almost impossible to account for stressing a

light penult in Arabic (e.g. ?adwiyatuhu) without parsing the syllables from left to right.

Recall that a light penult is stressed in Arabic iff (if and only if) it is separated from a

preceding heavy syllable (or word boundary) by an even number of light syllables. What

this basically means is that stressing a light penult in Arabic requires that a degenerate

foot be constructed over it. Only a left-to-right parsing results in having the degenerate

foot constructed over the light penult in all cases:

(30)

a (CV)<CV>
b. (CVCV)(CV)<CV>
c. (CVCV)(CVCV)(CV)<CV>

229
 e. ETC

A right-to-left parsing results in having the degenrate foot as close to the left edge of the

word as possible:

(31)

a. (CV)<CV>
*b. (CV)(CVCV)<CV>
*c. (CV)(CVCV)(CVCV)<CV>
d. ETC

Consider how this applies to an actual Arabic word like /?adwiyatuhu/:

First, due to NONFINAL, the last syllable should not be footed (i.e. extrametrical):

(32a) ?ad wiyatu<hu>

Second, if ALL-FEET-LEFT applies, we get:

(32b) (?ad)(wiya)(tu)<hu>

If, however, ALL-FEET-RIGHT applies, we get:

*(32c) (?ad)(wi)(yatu)<hu>

The decision whether to adopt a left-to-right or right-to-left parsing is crucial. If

ALL-FEET-RIGHT is adopted, then stress, due to TROCHAIC and MAIN-RIGHT, falls

on the antepenult--which is not correct. If ALL-FEET-LEFT is adopted, then stress, due

to TROCHAIC and MAIN-RIGHT, falls on the penult--which is correct. All in all, the

actual output form can only be accounted for with a left-to-right parsing.

Let us now turn to English for the ranking of ALL-FEET-RIGHT and ALL-

FEET-LEFT relative to each other. Consider how the correct stress pattern for /America/

is obtained:

230
Tableau (24)
Input: /America/ Lx=Pr NF FB TR Pσ A-F-R A-F-L
a- (Ame)rica ** *!* *
b-" A(meri)ca ** * **

Notice that, due to ALL-FEET-RIGHT, the competition between (a) and (b) is resolved

in favor of (b). Whereas (b) incurs one violation of ALL-FEET-RIGHT, its competitor

(a) incurs two violations. The proposal that ALL-FEET-RIGHT should dominate ALL-

FEET-LEFT in English is supported by the fact that stress surfaces on the antepenult;

otherwise, stress will surface on an pre-antepenultimate syllable (cf. output (a) in Tableau

22 above).182 After observing the requirement of NONFINAL, the correct output form is

obtained by having the penult and the antepenult pair together to make up one foot

(Hence FOOT-BINARITYσ) and having stress, due to TROCHAIC, fall on the

antepenult.

In terms of stress acquisition, native speakers of Arabic are expected to

experience problems to get the correct stress pattern in /America/ because they parse its

syllables into metrical feet from left to right. However, before we see what those learners

need to do to acquire the target language way of parsing, there is one more wrinkle that

should be tackled first, namely the relative ranking of PARSEσ and FOOT-BINARITY

in the two languages.

182
Notice that this is true only if FOOT-BINARITYσ dominates PARSEσ.

231
4.3.2.g FOOT-BINARITY and PARSEσ

Because of the influence of their native language, native speakers of Arabic produce

/America/ as [America], with main stress on the penult. This is so because in /America/,

the penult is the syllable that is separated from the word boundary by an even number of

light syllables. Compare an erroneous token like [America] with an actual Arabic word

like [katabaha].

Because ALL-FEET-LEFT dominates ALL-FEET-RIGHT in Arabic, native

speakers of Arabic are expected to parse /America/ into feet as in (33) below:

(33) (Ame)ri<ca>183

The question that arises is: How then do these learners produce a word like /America/

with penultimate stress?

Recall that in Arabic PARSEσ outranks FOOT-BINARITY. That is, the

requirement that each and every syllable be part of a metrical foot is more important than

the requirement to have every foot be of a certain shape. Given the constraint ranking in

Arabic, consider how /America/ is stressed:

Tableau (25)
Input: /America/ Lx=Pr MR NF Pσ FB A-F-L
a- (Ame)rica *!* *
b-!" (Ame)(ri)ca * * **

According to tableau (25) above, when employing the Arabic way of parsing syllables

into metrical feet, output (b) wins the competition because it incurs fewer violations of
183
Compare English (Ame)(ri)ca with Arabic (kata)(ba)ha. As PARSEσ dominates FOOT-BINARITY
(kata)(ba)ha is a better parse than (kata)baha.

232
PARSEσ than its competitor (a). The fact that (a) does not incur any violations of FOOT-

BINARITY is of no consequence here because PARSEσ dominates FOOT-BINARITY.

Thus, the candidate form that incurs more violations of PARSEσ is the one that loses the

competition, irrespective of whether or not it violates the lower-ranking constraint

FOOT-BINARITY.184

Let us demonstrate why native speakers of Arabic who transfer the constraint

ranking in their native language to English produce /America/ with penult stress.

First, the final syllable, due to NONFINAL, is unfooted:

(34a) Ameri<ca>

Second, syllables are grouped into moraic trochees, i.e. (LL) and (H) from left-to-right

(hence ALL-FEET-LEFT > ALL-FEET-RIGHT):

(34b) (Ame)ri<ca>

Third, due to PARSEσ, a degenerate foot is constructed over the penult:

(34c) (Ame)(ri)ca

Finally, due to TROCHAIC and MAIN-RIGHT, stress surfaces on the penult:

(34d) (Ame)(ri)ca

For acquisition purposes, native speakers of Arabic who learn English need to do

two things as far as the parsing of syllables into metrical feet is concerned. One, they

need to demote ALL-FEET-LEFT one step down the hierarchy, so that ALL-FEET-

RIGHT outranks ALL-FEET-LEFT. The constraint ranking in (29) above should now

look like (35) below:

184
In English, however, the need to satisfy PARSEσ is not as compelling as the need to fulfill the
requirement of FOOT-BINARITYσ. Hence, A(meri)ca is a better parse than (A)(meri)ca.

233
 (35) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ,
PARSE(σ), *NON-PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

Stratum 2 {ALL-FEET-RIGHT, WSP(<µµ), BIMORAIC}

Stratum 3 = { ALL-FEET-LEFT}

Second, they need to demote PARSEσ one step below FOOT-BINARITY. The

domination hierarchy in (35) above can be further modified as in (36) below:

(36) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ, *NON-
PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {PARSEσ, REDUCTION, NONFINALσ}

>>

Stratum 2 {ALL-FEET-RIGHT, WSP(<µµ), BIMORAIC}

Stratum 3 = { ALL-FEET-LEFT}

234
The current domination hierarchy in (36) above raises two interesting points: (1) the

relative ranking of PARSEσ and NONFINAL; and (2) the shape of the optimal foot in

the two languages.

4.3.2.h PARSEσ and NONFINAL

Disregarding the issue of final superheavy syllables in Arabic, final light and heavy

syllables in Arabic fail to surface stressed because NONFINAL dominates PARSEσ. For

a word like /katabaha/, (kata)ba is a better parse than (kata)(ba). Similarly,

NONFINALσ dominates PARSEσ in English. A(meri)ca is then a better parse than

A(meri)(ca).

Because NONFINAL dominates PARSEσ in the two languages, native speakers

of Arabic never produce words like /America/ with stress on the final syllables. In other

words, no transfer errors are expected to occur here. Such being the case, the constraint

domination hierarchy in (36) above automatically corrects itself, yielding (37) below,

where PARSEσ is demoted one step down the hierarchy:

(37) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ, *NON-
PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

Stratum 2 {PARSEσ, ALL-FEET-RIGHT, WSP(<µµ), BIMORAIC}

>>

235
 Stratum 3 = { ALL-FEET-LEFT}

However, this demotion has resulted in having PARSEσ housed in the same stratum with

ALL-FEET-RIGHT, which entails that these two constraints are unranked relative to

each other. Recall that for the learner to be able to construct more than one foot over the

word, PARSEσ has to dominate ALL-FEET-RIGHT and ALL-FEET-LEFT. Since

PARSEσ dominates ALL-FEET-RIGHT and ALL-FEET-LEFT in both Arabic and

English, the learner gets the demotion of ALL-FEET-RIGHT one step down the

hierarchy for free:

(38) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ, *NON-
PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

Stratum 2 {PARSEσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 = { ALL-FEET-RIGHT, ALL-FEET-LEFT}

The current domination hierarchy results in a violation of an already established

ranking; that is, the relative ranking of ALL-FEET-RIGHT and ALL-FEET-LEFT. We

have shown that words like /America/ should force the learner to demote ALL-FEET-

236
LEFT one step down the hierarchy so that the hierarchy continues to show the relative

ranking of these two alignment constraints:

(39) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ, *NON-
PRIMARY, FOOT-BINARITYµ}

>>

Stratum 1 {REDUCTION, NONFINALσ}

>>

Stratum 2 {PARSEσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 = { ALL-FEET-RIGHT}

>>

Stratum 4 = {ALL-FEET-LEFT}

237
4.3.2.i FOOT-BINARITY

One major source of errors in the English of native speakers of Arabic with respect to

stress placement comes from the fact that the shape of the foot is different in the two

languages. Whereas Arabic requires that each foot consist of two moras, English requires

that each foot consist of two syllables. In other words, FOOT-BINARITY is interpreted

under a moraic analysis in Arabic, but under syllabic analysis in English. This is expected

to be a significant source of stress errors in the English of native speakers of Arabic.

In Arabic, a heavy syllable (σµµ) makes up a foot by itself; and two light syllables

(σµσµ) make up one foot.185 We have shown in 3.4 above that a syllabic analysis of

FOOT-BINARITY does not work for Arabic because it cannot account for the place of

stress on a light penult (e.g. HLL). In a word like [maktabah], for example, stress goes to

the light penult. Given the fact that Arabic is a trochaic system, the stress pattern of

[maktabah] cannot be accounted for if we assumed that FOOT-BINARITY is interpreted

under syllabic analysis because the light penult would then pair with the heavy antepenult

185
In case the foot is constructed over a heavy syllable (σµµ), it does not matter which mora is the head of
the syllable because both moras are part of one syllable which ultimately surfaces as the one with main
stress (i.e. (σµµ)). In case the foot is constructed over two light syllables (σµσµ), it really does matter which
one is the head of the foot simply because each mora belongs to an independent syllable and only one of
the two light syllables must surface as the head of the foot, and ultimately carriy main word-stress. In this
latter case, we have shown that there is ample evidence that the foot is left-headed in Arabic. That is, in
each foot that consists of a pair of light syllables, it is the leftmost syllable that must be the head of the foot,
and ultimately the one that carries main word-stress (i.e. (σµσµ)).

238
to make up one foot (e.g., (HL)L186), and stress, due to TROCHAIC, would fall on the

heavy penult, e.g., (HL)L, which is not correct.

In English, however, FOOT-BINARITY must be interpreted under syllabic

analysis, i.e., that each foot must consist of two syllables irrespective of their weight, e.g.,

(LL), (LH), (HL), or (HH). The reason is that in English, unlike Arabic, a heavy

antepenult receives stress when it pairs with a light penult to make up one foot, e.g.,

(HL)L187. In a word like [asterisk], stress goes to the heavy antepenult. All in all, the

interpretation of FOOT-BINARITY under a syllabic analysis can account for the English

data, but cannot account for the Arabic data.

This is probably why native speakers of Arabic produce words like /asterisk/ with

penult stress. In order to acquire the correct stress pattern, those learners need to learn to

parse /asterisk/ as in (40a) below, not as in (40b):

(40)

a. (aste)risk
b. (as)(te)risk

In terms of constraint ranking, FOOT-BINARITYµ is very-low ranked in English

because its violation is irrelevant in determing the actual output form. This entails that

native speakers of Arabic need to demote this constraint to the lowest stratum possible, so

that it is outranked by the entire set of constaints:

(41) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), FOOT-BINARITYσ, *NON-
PRIMARY}

186
Notice that this is true irrespective of the directionality of footing.
187
This is also true in some Arabic dialects (e.g. Hijazi Arabic, see Abaalkhail 1998)

239
 >>

Stratum 1 {REDUCTION, NONFINALσ}

>>

Stratum 2 {PARSEσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 { ALL-FEET-RIGHT }

>>

Stratum 4 { ALL-FEET-LEFT}

>>

Stratum 5 { FOOT-BINARITYµ}

To sum up, because FOOT-BINARITY is interpreted under a moraic analysis to

account for the Arabic data, but under a syllabic analysis to account for the English data,

negative transfer is likely to occur. In particular, it makes native speakers of Arabic fail to

stress the heavy antepenult in words like /asterisk/.

240
4.3.2.j FOOT-BINARITYσ and WSP

If FOOT-BINARITYσ were undominated in English, then stress, due to

TROCHAIC, would always terminate on the antepenult. To illustrate, we have already

shown that the final syllable, due to NONFINAL, is unfooted in English, and that the

remaining syllables are parsed into binary feet from right to left. If FOOT-BINARITYσ

were undominated, the penult and the antepenult would always pair together to make up a

left-headed foot (because English is a trochaic system):

(42)
(a) a(meri)ca
(b) (ani)mal
*(c) (agen)da

Although this yields the desired output forms in (42a) and (42b), it yields the

wrong form in (42c). The question arises as to how we can account for the stress pattern

in /agenda/. Given the fact that English feet consist of two syllables, we have asserted

earlier that the answer to this problem is to assume that FOOT-BINARITYσ is outranked

by another constraint that requires the penult in words like (42c) above to surface with

main word-stress. We have suggested that the penult in (42c) is stressed because it is a

heavy syllable. This basically means that FOOT-BINARITYσ is dominated by the

constraint which calls for the stressing of a heavy syllable, namely WSP(≥µµ).

241
 Interestingly enough, native speakers of Arabic are never reported to have

difficulty in acquiring the stress pattern of words like /agenda/. The reason is that a heavy

penult always surfaces with main stress in both Arabic and English. In Arabic, the place

of stress on a heavy penult is accounted for by FOOT-BINARITYµ. Hence, a heavy

penult always make up a foot by itself, and stress, due to MAIN-RIGHT, terminates on

that foot. Consider how stress surfaces on the heavy penult of /fihimna/:

The final syllable is, due to NONFINAL, unfooted:

(43a) fihim<na> ‘he understood us’

The penult, due to FOOT-BINARITYµ, makes up a foot by itself:

(43b) fi(him)<na>

And stress, due to MAIN-RIGHT and TROCHAIC, surfaces on the penult

(43c) fi(him)<na>

In English, however, stressing a heavy penult (e.g., LHL) is accounted for by

positing that WSP(≥µµ) outranks FOOT-BINARITYσ, so that L(H)L is a better parse

than (LH)L. Consider Tableau (26) below with a partial ranking of the constraints:

Tableau (26)
input /agenda/ WSP(VV) NF WSP(≥µµ) FBσ
a- (agen)(da) *! *
b- (agen)da *!
c-" a(gen)da *

For acquistion purposes, despite the fact that native speakers do not encounter

difficulty in acquiring the stress pattern in /agenda/, they still need to learn to demote

FOOT-BINARITYσ one step below WSP(≥µµ):

(44) Domination hierarchy

242
 Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,
TROCHAIC, WSP(≥µµ), *NON-PRIMARY}

>>

Stratum 1 { FOOT-BINARITYσ, REDUCTION, NONFINALσ}

>>

Stratum 2 {PARSEσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 { ALL-FEET-RIGHT }

>>

Stratum 4 {ALL-FEET-LEFT}

>>

Stratum 5 {FOOT-BINARITYµ}

This demotion results in having FOOT-BINARITY(σ) housed in Stratum 1 with

NONFINAL. Recall that in English the need to satisfy NONFINAL is more important

that the need to satisfy FOOT-BINARITYσ. Hence, A(meri)ca is a better parse than

(Ame)(rica). In Arabic, too, NONFINAL dominates FOOT-BINARITY. Hence,

(jada)(la)tu is a better parse than (jada)(latu). The domination hierarchy in (53)

automatically corrects itself by demoting FOOT-BINARITY(σ) one step below

NONFINAL:

(45) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), *NON-PRIMARY}

>>

243
 Stratum 1 { REDUCTION, NONFINALσ}

>>

Stratum 2 { FOOT-BINARITYσ, PARSEσ, WSP(<µµ),

BIMORAIC}

>>

Stratum 3 { ALL-FEET-RIGHT}

>>

Stratum 4 { ALL-FEET-LEFT}

>>

Stratum 5 {FOOT-BINARITYµ}

At this stage, negative transfer is very likely to occur again. This is so because the

demotion of FOOT-BINARITYσ one step down the hierarchy has resulted in having it

housed with PARSEσ in the same stratum, wrongly suggesting that these two constaints

are unranked relative to each other. In Arabic, it is shown that the need to make each and

every syllable part of a metrical foot outweighs the need to satisfy the requirement

imposed on the shape of the foot, so that (jada)(la)tu is a better parse than (jada)latu. In

English, however, it is the other way around. FOOT-BINARITYσ dominates PARSEσ,

so that A(meri)ca is a better parse than (A)(meri)ca or (Ame)(ri)ca. Accordingly, learners

should be exposed to data that force the demotion of PARSEσ one step further down the

hierarchy:

(46) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), *NON-PRIMARY}

244
 >>

Stratum 1 { REDUCTION, NONFINALσ}

>>

Stratum 2 { FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 { PARSEσ, ALL-FEET-RIGHT}

>>

Stratumb 4 { ALL-FEET-LEFT}

>>

Stratum 5 {FOOT-BINARITYµ}

This demotion results in the violation of another already established ranking,

namely the ranking of PARSEσ with respect to ALL-FEET-RIGHT. In order to keep the

relative ranking of PARSEσ and ALL-FEET-RIGHT, the latter must be demoted one

step:

(47) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), *NON-PRIMARY}

>>

Stratum 1 { REDUCTION, NONFINALσ}

>>

Stratum 2 { FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

245
 Stratum 3 { PARSEσ}

>>

Stratum 5 {ALL-FEET-RIGHT, ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

Again, this demotion results in the violation of the relative ranking of ALL-

FEET-RIGHT and ALL-FEET-LEFT. Learners thus need to be exposed to data that

again force the demotion of ALL-FEET-LEFT one step down:

(48) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), *NON-PRIMARY}

>>

Stratum 1 { REDUCTION, NONFINALσ}

>>

Stratum 2 { FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 { PARSEσ}

>>

Stratum 4 {ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT, FOOT-BINARITYµ}

246
 The learner has also to be aware that FOOT-BINARITYµ is violated all over the

place in English, so that its violation is not crucial in determining the actual output form.

FOOT-BINARITYµ should then be housed in the lowest stratum, so that it is dominated

by the entire set of constraints, including ALL-FEET-LEFT:

(49) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, WSP(≥µµ), *NON-PRIMARY}

>>

Stratum 1 { REDUCTION, NONFINALσ}

>>

Stratum 2 { FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 { PARSEσ}

>>

Stratum 4 {ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

247
4.3.2.k MAIN-RIGHT and WSP(≥µµ)

Recall that in English, syllables that end with a long vowel may occur in nonfinal

position (e.g., aroma, boomerang, bailiwick, etc.). In addition, they may not be the ones

that carry main word-stress (e.g. souvenir, toucanet, etc.). What this basically means is

that WSP(≥µµ) is not top-ranked in English. Words like souvenir and toucanet suggest

that an independent foot is constructed over the final syllable which carries main word-

stress:

(50)
a. (souve)(nir)
b. (touca)(net)

And, due to MAIN-RIGHT, stress terminates on the final syllable:

(51)
a. (souve)(nir)
b. (touca)(net)

This basically means that MAIN-RIGHT outranks WSP(≥µµ), which in turn outranks

NONFINAL in English. Accordingly, the domination hierarchy in (49) should be

modified as in (52) below, where WSP(≥µµ) is demoted one step down the hierarchy:

(52) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

248
 >>

Stratum 1 {WSP(≥µµ), REDUCTION, NONFINALσ}

>>

Stratum 2 {FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

Stratum 3 { PARSEσ}

>>

Stratum 4 {ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

Notice that this move requires the domination hierarchy to correct itself several times.

First, WSP(≥µµ) is now housed in the same stratum with NONFINAL, which is

not correct because a final syllable that weighs two moras (or more) surfaces with main

stress (e.g. agree, kangaroo, etc.). This requires the demotion of NONFINAL to the next

available stratum:

(53) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ), REDUCTION}

249
 >>

Stratum 2 { NONFINALσ, FOOT-BINARITYσ, WSP(<µµ),

BIMORAIC}
>>

Stratum 3 { PARSEσ}

>>

Stratum 4 {ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

The domination hierarchy has also to show that WSP(≥µµ) dominates

REDUCTION, so that, unlike the final syllable in hammck, animl, etc., the final

syllable in bassinet, souvenir, toucanet, etc. does not get reduced:

(54) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ) }

>>

Stratum 2 {REDUCTION, NONFINALσ, FOOT-BINARITYσ,

WSP(<µµ), BIMORAIC}

>>

Stratum 3 { PARSEσ}

250
 >>

Stratum 4 {ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

The domination hierarchy still needs to correct itself; it has to show that NONFINAL

dominates WSP(<µµ) and FOOT-BINARITYσ, and that REDUCTION dominates

BIMORAIC:

(55) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ) }

>>

Stratum 2 {REDUCTION, NONFINALσ}

>>

Stratum 3 {FOOT-BINARITYσ, WSP(<µµ), BIMORAIC, PARSEσ}

>>

Stratum 4 {ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT}

>>

251
 Stratum 6 {FOOT-BINARITYµ}

The domination hierarchy in (55) violates the relative ranking of FOOT-BINARITY and

PARSEσ, and thus requires the demotion of PARSEσ one step down the hierarchy:

(56) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ) }

>>

Stratum 2 {REDUCTION, NONFINALσ}

>>

Stratum 3 {FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

Stratum 4 { PARSEσ, ALL-FEET-RIGHT}

>>

Stratum 5 {ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

This move results in a violation of the relative ranking of PARSEσ and ALL-FEET-

RIGHT, requiring that the latter be demoted to a lower position:

(57) Domination hierarchy

252
 Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,
TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ) }

>>

Stratum 2 {REDUCTION, NONFINALσ}

>>

Stratum 3 {WSP(<µµ), FOOT-BINARITYσ, BIMORAIC}

>>

Stratum 4 { PARSEσ}

>>

Stratum 5 { ALL-FEET-RIGHT, ALL-FEET-LEFT}

>>

Stratum 6 {FOOT-BINARITYµ}

The resulting domination hierarchy violates the relative ranking of ALL-FEET-RIGHT

dominates ALL-FEET-LEFT. To show just that, ALL-FEET-LEFT must be demoted to

the next available stratum:

(58) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ) }

>>

253
 Stratum 2 {REDUCTION, NONFINALσ}

>>

Stratum 3 {FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

Stratum 4 { PARSEσ}

>>

Stratum 5 { ALL-FEET-RIGHT}

>>

Stratum 6 { ALL-FEET-LEFT, FOOT-BINARITYµ}

But still the domination hierarchy has to correct itself again to show that FOOT-

BINARITYµ is dominated by the entire set of constraint, and thus must be demoted one

step down the hierarchy , creating a new stratum:

(59) Domination hierarchy

Initial Stratum {NEUTRAL, Lx=Pr, FAITH, MAIN-RIGHT,

TROCHAIC, *NON-PRIMARY}

>>

Stratum 1 {WSP(≥µµ) }

>>

Stratum 2 {REDUCTION, NONFINALσ}

>>

Stratum 3 {FOOT-BINARITYσ, WSP(<µµ), BIMORAIC}

>>

254
 Stratum 4 { PARSEσ}

>>

Stratum 5 { ALL-FEET-RIGHT}

>>

Stratum 6 { ALL-FEET-LEFT}

>>

Stratum 7 {FOOT-BINARITYµ}

4.4 Conclusion

(59) is a stratified hierarchy showing the domination relations that exist among the

various constraints that are needed to account for main word-stress in English. This

hierarchy is motivated by the following generalizations:

I. Lx=Pr, FAITH, MAIN-RIGHT, and TROCHAIC are all undominated in

English.

a. Lx=Pr is undominated because each and every content word has to

have at least one foot constructed over it; otherwise, that word will
surface stressless (Hence it won’t have a head). A(meri)ca is then a
better parse than America.
b. Lx=Pr must dominate NONFINAL so as to avoid rendering the whole
word unfooted (i.e. extrametrical). (cat) is then a better parse than cat.
c. Lx=Pr must dominate FOOT-BINARITY. The need to have at least
one foot constructed over monosyllabic words is greater than the need
to have the foot fit a certain shape. (hat) is a always optimal.
d. FAITH must dominate NONFINAL so that a final closed syllable can
be stressed. (bassi)(net) is a better parse than (bassi)net.
e. FAITH must dominate REDUCTION, so that a final closed syllable
that is stressed does not undergo reduction. (bassi)(net) is a better
parse than (bassi)(nt).

255
 f. MAIN-RIGHT must dominate WSP so as to allow stress to fall on one
of the last three syllables. (bassi)(net) is a better parse than
(bassi)(net).
g. TROCHAIC must dominate IAMBIC, so as to allow feet to be left-
headed. A(meri)ca is a better parse than A(meri)ca.

II All the other constraints are dominated. Yet some of them are ranked
relative to each other:

a. WSP(≥µµ) must dominate NONFINAL, so as to allow words ending

with long vowels to receive stress on that syllable. ag(ree) is a better
parse than (ag)ree; and to allow stress to surface on ultimate syllables
that are closed by a consonant cluster. a(larm) is a better parse than
(a)larm.
b. NONFINAL must dominate WSP(<µµ) so as to allow a closed
syllable in final position to be skipped over: (ani)mal is a better parse
than (ani)(mal); and to allow a syllable that is closed by a consonant
cluster that contributes less than two moras to surface stressless:
(nove)list is a better parse than (nove)(list)
c. NONFINAL must dominate ALL-FEET-RIGHT, so as to allow stress
to fall on an antepenult syllable. A(meri)ca is a better parse than
Ame(rica).
d. NONFINAL must dominate FOOT-BINARITY. (ham)mock is a better
parse than (hammock); and A(meri)ca is a better parse than
(Ame)(rica).
e. REDUCTION must dominate BIMORACITY, so as to allow the
nucleus of unstressed syllables to reduce to schwa. hammck is a
better parse than hammock.
f. WSP(≥µµ) must dominate FOOT-BINARITY, so as to allow a closed
syllable that contributes more than one mora to be stressed in penult
position. A(gen)da is a better parse than (agen)da
g. FOOT-BINARITY must dominate PARSE-SYLLABLE. A(meri)ca is
a better parse than (A)(meri)ca or (Ame)(ri)ca.

The domination hierarchy in (59) should be compared with the domination hierarchy

needed to account for Arabic main word-stres, which is motivated by the following

generalizations:

I. TROCHAIC, MAIN-RIGHT, Lx=Pr, and *NON-PRIMARY are undominated

in Arabic:

256
 a. TROCHAIC must be undominated. (jaa)(dala)<tu> is a better parse
than(jaa)(dala)<tu>.
b. MAIN-RIGHT must be undominated. (jaa)(dala)<tu> is a better parse
than (jaa)(dala)<tu>
c. Lx=Pr must dominate NONFINALσ. (min) is a better parse than min.
d. *NONPRIMARY must be undominated. (jaa)(dala)<tu> is a better parse
than (jaa)(dala)<tu>188

II. The other constraints, although they are all dominated by higher ranking
constraints, are ranked relative to each other:

a. NONFINAL(σµµ) must dominate WSP.(baT)Tah is a better parse than

(baT)(Tah).
b. NONFINAL(σµµ) must dominate FOOT-BINARITY-µ. (i)bir is a better
parse than i(bir), and (ra)ma is a better parse than (rama).
c. NONFINAL(σµµ) must dominate PARSE-σ. (i)bir is a better parse than
(i)(bir), and (ka.ta)ba is a better parse than (ka.ta)(ba)
d. PARSEσ must dominate NONFINAL(σµµµ). (šan)(taat) is a better parse
than (šan)taat.
e. PARSE-σ must dominate FOOT-BINARITY. (i)<bir> is a better parse
than Hi<bir>, and (kata)(ba)<ha> is a better parse than (kata)ba<ha>.
f. ALL-FEET-LEFT must dominate ALL-FEET-RIGHT.
(ad)(wiya)(tu)<hu> is a better parse than (ad)(wi)(yatu)<hu>

The comparision has shown that there are some features in common as far as the

ranking of constraints is concerned, but they differ in other features. When two or more

constraints are ranked in the same way relative to each other in the two languages, no

stress errors occur; but when they are ranked differently, stress errors are very likely to

occur. With respect to constraint ranking, among the major differences between the two

systems are the following:

188
Notice that Lx=Pr and WSP(σµµµ), for example, do not only dominate NONFINAL, but they also
dominate all the constraints that are dominated by NONFINAL. This is called transitivity of ranking in
which we assume that some ranking exists between all constraints even those which do not interact (see
Kager 1999: 21).

257
 a. The ranking of *NONPRIMARY relative to the entire set of constraints. Whereas

*NONPRIMARY is never violated in Arabic (i.e. undominated), it is violated all

over the place in English. Secondary stress is thus a major problem in the English

of native Arabic speakers.189

b. The ranking of NEUTRAL relative to the entire set of constraints. NEUTRAL is

undominated in English, but it is dominated by the entire set of constaints in

Arabic. Suffixed English words are thus very much problematic to native speakers

of Arabic with respect to stress placement.190

c. FOOT-BINARITY is interpreted under a moraic analysis in Arabic, but under a

syllabic analysis in English. In terms of constraint ranking, FOOT-BINARITYµ

is even irrelevant in English, and so it is dominated by the whole set of

constraints. By transferring the ranking of FOOT-BINARITY in their L1 to

English, native speakers of Arabic fail to produce /asterisk/ with antepenult stress.

d. Syllables are parsed into metrical feet from left to right in Arabic, but form right

to left in English. In other words, whereas ALL-FEET-LEFT dominates ALL-

FEET-RIGHT in Arabic, it is the other way around in English. So, native speakers

of Arabic parse /America/ as (Ame)ri<ca> not as A(meri)ca.

e. PARSEσ dominates FOOT-BINARITY in Arabic, allowing for degenerate feet to

be constructed over the word. In English, however, FOOT-BINARITY dominates

PARSEσ. Degenerate feet are only sanctioned by higher-ranking constraints in

English. So, native speakers of Arabic parse /America/ as (Ame)(ri)<ca>.

189
Because our focus has been on main stress, we devoted little time to talk about this point.
190
Because our focus has been on monomorphemic words, we devoted little time to talk about this point

258
f. FAITH dominates REDUCTION in English, but both FAITH and REDUCTION

are so low ranked in Arabic that they are irrelevant in determining the ctual output

form. So, native speakers of Arabic produce /bassinet/ as bassi<net>.

g. A final syllable that weighs two moras is stressed in English, but a afinal syllable

that weighs two moras is always stressless in Arabic. In other words, whereas a

syllable that weighs two moras fall in the same category as the syllable that

weighs three moras in English, a syllable that weighs two moras falls in the

category with the syllable that weighs one mora in Arabic. In terms of constraint

ranking, that WSP(≥µµ) outranks NONFINAL in English. Native speakers of

Arabic produce /bassinet/ as (bas)(si)<net>.

h. REDUCTION dominates BIMORAIC in English, but both REDUCTION and

BIMORAIC are so low ranked in Arabic that they are irrelevant in determining

the actual output form. So native speakers of Arabic produce /hammock/ as

ham(mock).

259
References

Abaalkhail, F. (1998). Syllabification and metrification in Urban Hijazi Arabic:

between rules and constraints. Ph.D. Diss. University of Essex.

Abdel-Al-Jawad, H. (1981). Lexical and phonological variation in spoken Arabic

in Amman. Ph.D. Diss. University of Pennsylvania.

Abdel-Al-Jawad, H. (1986). The emergence of an urban dialect in the Jordanian

urban centers. International Journal of the Sociology of Language 61, 53-63.

Abdel-Al-Jawad, H. & Awwad, M. (1989). Reflexes of classical Arabic

interdentals: A study in historical sociolinguistics. Linguistische Berichte 122, 259-282.

Abdo, D. (1969). Stress and Arabic phonology. Ph.D. Diss. Urbana: University
of Illinois.

Abu-Salim, I. (1982). A reanalysis of some aspects of Arabic phonology: A

metrical approach. Ph.D. Diss. Urbana: University of Illinois.

Abu-Salim, I. (1987). Vowel Harmony in Palestinian Arabic: a metrical

perspective. Journal of Linguistics 23, 1-24.

Adams, C. (1979). English speech rhythm and the foreign learner. The Hague:
Mouton Publishers.

Aitchison, J. (1994). Understanding words. In G. Brown, K. Malmkjor, A. Pollitt

and J. Williams (eds.) Language and understanding. Oxford: Oxford University Press,
83-95.

Alanen, R. (1995). Input enhancement and rule presentation in second language

acquisition. In R. Schmidt (ed.). Attention & awareness in foreign language learning.
Manoa, HI: Second Language Teaching & Curriculum Center, 259-302.

Alghazo, M. (1987). Syncope and epenthesis in Levantine Arabic: A nonlinear

approach. Ph.D Diss. University of Illinois at Urbana-Champaign.

Al-Khatib, M. (1988). Sociolinguistic change in an expanding urban center: A

case study of Irbid city, Jordan. Ph.D. Diss. University of Durham.

Allen, V. F. (1971). Teaching intonation: From theory to practice. TESOL

Quarterly 4 (1), 73-81.

260
 Al-Mozainy H., Bley-Vroman, R. & McCarthy J. (1985). Stress shift and metrical
structure. Linguistic Inquiry 16, 135-44.

Al-Sughayer, K. (1990). Aspects of comparative Jordanian and modern standard

Arabic phonology. Ph.D. Diss. Michigan State University.

Anani, M. (1989). Incorrect stress placement in the case of Arab learners of

English. IRAL 27 (1), l5-21.

Avery, P. & Ehrlich, S. (1992). Teaching American English pronunciation.

Oxford: Oxford University Press.

Aziz, W. (1980). Some problems of English word stress for Iraqi learners. English
Language Teaching Journal 2, 104-109.

Bani Yasin, R. & Owens, J. (1984). The Bduul dialect of Jordan. Anthropological
Linguistics 26 (2), 202-235.

Bansal, R.K. (1966). The intelligibility of Indian English. Ph.D. Diss. London:
London University.

Bardovi-Harlig, Kathleen. 1983. Pronouns: When ‘Given’ and ‘New’ coincide. In

A. Chukerman, M. Marks, & J. F. Richardson (eds.). Papers from the Ninteenth Regional
Meeting. Chicago: Chicago Linguistic Society, 15-26.

Benrabah, M. (1997). Word stress – a source of unintelligibility in English. IRAL

36 (3), 157-165.

Berg, T. (1999). Stress variation in British and American English. World

Englishes 18 (2), 123-143.

Bolinger, D. (1958). A theory of pitch accent in English. Word 17, 309-317.

Bolinger, D. (1985). Two views of accent. Journal of Linguistics 21, 79-125.

Bradley, D. (1980). Lexical representation of derivational relation. In M. Aronoff

& M.L. Kean (eds.). Juncture. Saratoga, CA: Anma Libri.

Brame, M. (1971). Stress in Arabic and generative phonology. Foundations of

Language 7, 556-591.

Brame, M. (1973). On stress assignment in two Arabic dialects. In S. Anderson &

P. Kiparsky (eds.). A festschrift for Morris Halle. New York: Holt, Rinehart and Winston.

261
 Brame, M. (1974). The cycle in phonology: Stress in Palestinian, Maltese and
Spanish. Linguistic Inquiry 5, 39-60.

Browman, C. P. (1978). Tip of the tongue and slips of the ear: Implications for
language processing. UCLA Working Papers in Phonetics 42.

Chafe, W. (1970). Meaning and the structure of language. Chicago: University of

Chicago Press.

Chomsky, N. (1972). Language and mind. New York: Harcourt Brace

Jovanovich.

Chomsky, N. (1981). Principles and parameters in syntactic theory. In N.

Hornstein & D. Lightfoot (eds). Explanation in Linguistics: The logical problem of
language acquisition. London: Longman.

Chomsky, N. & Halle, M. (1968). The sound pattern of English. New York:
Harper & Row.

Clark, H. & Clark, E. (1977). Psychology and language. New York: Harcourt.

Cole, R. A. & Jakimik, J. (1979). A model of speech perception. In R. Cole (ed.)

Perception and production of fluent speech. Hillsdale, NJ: Erlbaum.

Cruttenden, A. (1986). Intonation. Cambridge: Cambridge University Press.

Cutler, A. (1976). Phoneme monitoring reaction time as a function of preceding

intonation contour. Perception and Psychophysics 20, 55-60.

Cutler, A. (1980). Errors of stress and intonation. In V.A. Fromkin (ed.). Errors in
linguistic performance: Slips of the tongue, ear, pen and hand. New York: Academic
Press, 67-80.

Cutler, A. (1984). Stress and accent in language production and understanding. In

D. Gibbon & H. Richter (eds.). Intonation, accent and rhythm: Studies in discourse
phonology. Berlin: Walter de Gruyter.

Cutler, A. & Clifton, C. E. (1983). Lexical stress effects on phonetic

categorization in auditory word perception. Paper Presented at the Tenth International
Congress of Phonetic Sciences, Utrecht, 77-90.

Dalton, C. & Seidlhoffer, B. (1994). Pronunciation. Oxford: Oxford University

Press.

262
 Davis, S. (1988). Syllable onsets as a factor in stress rules. Phonology 26, 183-
187.

de Graaff, R. (1997). The experanto experiment: Effects of explicit instruction on

second language acquisition. SSLA 19, 249-275.

DeKeyser, R. M. (1993). The effect of error correction on L2 grammar knowledge

and oral proficiency. The Modern Language Journal 77, 501-514.

de Lacy, P. (1997). Prosodic Categorization. [M.A. Thesis. The University of

Auckland. [ROA-336, http://ruccs.rutgers.edu/roa.html].

Dickerson, W. (1978). English orthography: A guide to word stress and vowel

quality. IRAL 16 (2), 127-147

Doughty, C. (1991). Second language instruction does make a difference:

Evidence from an empirical study of SL relativization. Studies in Second Language
Acquisition 13, 431-469.

Dulay, H. & Burt, M. (1974). Natural sequences in child second language

acquisition. Language Learning 24, 37-53.

Ferguson, C. (1956). The emphatic /l/ in Arabic. Language 32 (3), 446-452.

Ferguson, C. (1959). Diglossia. Word. Reprinted in P. Giglioli (ed.). Language

and social context: selected readings (1972). Baltimore: Penguin Books, 232-251.

Flege, J. & Bohn, O. (1989). An instrumental study of vowel reduction and stress
placement in Spanish accented English. SSLA 11, 35-62.

Fotos, S. (1993). Consciousness raising and noticing through focus on form:

Grammar task performance vs. formal instruction. Applied Linguistics 14, 385-407.

Fry, D. (1958). Experiments on the perception of stress. Language and Speech 1,

126-152.

Fudge, E. (1984). English word-stress. London: George Allen & Unwin.

Garnes, S. and Bond, Z. (1975). Slips of the ear: Errors in perception of casual
speech. Papers from the Eleventh Regional Meeting, Chicago Linguistic Society 214-225.

Ghaith, S. (1993). The assignment of primary stress to words by some Arab

speakers. System 21 (3), 381-390.

263
 Gilbert, J. B. (1984). Clear Speech: Pronunciation and listening comprehension
in American English. Cambridge: Cambridge University Press.

Gimson, A. (1980). An introduction to the pronunciation of English. 3rd edition.

London: Edward Arnold.

Goldman, R., Armour, L. & Schwartz, A. (1980). The effects of phonetic context
on speech-sound discrimination. Journal of Communication Disorders 13, 153-157.

Greenberg, J. (1966). Language universals, with special reference to feature

hierarchies. The Hague: Mouton.

Grosjean, F. (1980). Spoken word recognition processes and the gating paradigm.
Perception and Psychophysics 28 (4), 267-283.

Grosjean, F. & Gee, J. (1987). Prosodic structure and spoken word recognition. In
U. H. Frauenfelder & L. K. Tyler (eds.). Spoken word recognition. Cambridge, Mass.:
MIT Press.

Gussenhoven, C. (1983). Focus, mode and the nucleus. Linguistics, 19, 377-417.

Halle, M. & Keyser, S. J. (1971). English stress: Its form, its growth and its role
in verse. New York: Harper and Row.

Halle, M. & Vergnaud, R. (1987). An essay on stress. Cambridge, Mass.: MIT

Press.

Hammond, M. (1984). Constraining metrical theory: A modular theory of rhythm

and destressing. PhD. Diss. Unversity of California. [Published 1988, New York:
Garland.]

Hammond, M. (1989a). Cyclic Secondary stresses in English. West Coast

Conference on Formal Linguistics 8.

Hammond, M. (1989b). Lexical stresses in Macedonian and Polish. Phonology 6,

19-38.

Hammond, M. (1999). The phonology of English: A prosodic optimality-theoretic

approach. Oxford: Oxford University Press.

Harley, B. (1989). Functional grammar in French immersion: A classroom

experiment. Applied Linguistics 19, 331-359.

264
 Harris, J. (1983). Syallable structure and stress in Spanish. Cambridge, Mass.:
MIT Press.

Hayes, B. (1979). Extrametricality. MIT Working Papers in Linguistics 1, 77-87.

Hayes. B. (1980). A metrical theory of stress rules. Ph.D. Dissertation, MIT.

[Distributed 1981: IULC] [Published 1985, New York: Garland Press].

Hayes, B. (1982). Extrametricality and English stress. Linguistic Inquiry, 13 (2),

227-276.

Hayes, B. (1995). Metrical stress theory: principles and case studies. Chicago:
University of Chicago Press.

Hogg, R. and McCully, C. 1987. Metrical phonology: A coursebook. London:

Cambridge University Press.

Hornstein, N. & Lightfoot, D. (eds.) (1981). Explanation in Linguistics: The

logical problem of language acquistion. London: Longman.

Howes, D. (1957). On the relation between the intelligibility and frequency of

occurrence of English words. Journal of the Acoustical Society of America 29, 296-305.

Hubicka, O. (1980). Why bother about phonology? Practical English Teaching 1

(1), 22-24.

Hubicka, O. (1981). Phonology: Stress. Practical English Teaching 1 (3), 20-23.

Hulstijn, J. H. (1997). Introduction. Second language acquisition research in the

laboratory: Possibilities and limitations. SSLA 19, 131-144.

Hulstijn, J. H. & de Graaff, R. (1994). Under what conditions does explicit

knowledge of a second language facilitate the acquisition of implicit knowledge? A
research proposal. AILA Review 11, 97-113.

Hung, H. J. (1995). The rhythmic and prosodic organization of edge constituents:

An optimality-theoretic account. Bloomington, IN: IULC.

Irshid, O. (1984). The phonology of Arabic; Bani-Hassan, A Bedouin Jordanian

dialect. Ph.D. Diss. Urbana: University of Illinois at Urbana-Champaign.

Irshid, O. & Kenstowicz, M. (1984). Some phonological rules of Bani-Hassan

Arabic: A Bedouin dialect. Studies in the Linguistic Sciences 14 (1), 109-147.

265
 Ito, J. (1986). Syllable theory in prosodic phonology. PhD Diss. Umass at
Amherest. [Published 1988, New York: Garland.]

Jakimik, J. (1979). Word recognition and the lexicon. In J. Wolf & D. Klatt (eds.)
Speech Communication Papers Presented at the 97th Meeting of the Acoustical Society of
America. New York: Acoustical Society of America..

Jones, D. (1972). An outline of English phonetics. 9th edition. Cambridge:

Cambridge University Press.

Juffs, A. (1990). Tone, syllable structure and interlanguage phonology: Chinese

learners’ stress errors. IRAL 28 (2), 99-117.

Kager, R. (1997). Surface opacity of metrical structure in Optimality Theory.

[ROA 207, http://ruccs.rutgers.edu/roa.html].

Kager, R. (1989). A metrical theory of stress and destressing in English and

Dutch. Dordrecht: Foris.

Kager, R. (1999). Optimality Theory. Cambridge: Cambridge University Press.

Kenstowicz, M. (1980). Notes on Cairene syncope. (1980). Studies in the

Linguistic Sciences 10 (2), 39-53.

Kenstowicz M. (1994). Phonology in generative grammar. Cambridge, MA. &

Oxford, UK: Blackwell.

Kenstowicz, M. & Abdul-Karim, K. (1980). Cyclic stress in Levantine Arabic.

Studies in the Linguistic Sciences 10 (2), 55-76.

Kenworthy, J. (1987). Teaching English pronunciation. London: Longman.

Kreidler, C. (l989). The pronunciation of English: A course book in phonology.

New York: Basil Blackwell.

Langendoen, D. (1968). The London school of linguistics: A study of the linguistic

theories of B. Malinowski and J. R. Firth. Cambridge, Mass.: MIT press.

Larudee, F. (1973). Word stress in the spoken Arabic of Cairo. Language

Sciences 26, 31-34.

Lea, W. A. (1973). Prosodic aids to speech recognition II: Syntactic segmentation

and stressed syllable location. Sperry Univac Technical Report No. PX 10232.

266
 Lee, Y. (1999). Extrametricality in English. In K. Baertsch and D. A. Dinnsen
(eds.). Optimal green ideas in linguistics. Indiana University: IULC.

Liberman, M. (1975). The intonational system of English. Ph.D. Diss. MIT.

Liberman, M. & Prince. A. (1977). On stress and linguistic rhythm. Linguistic

Inquiry 8, 249-336.

Lightfoot, D. (1982). The language lottery: Towards a biology of grammars.

Cambridge, Mass.: MIT Press.

Marslen-Wilson, W. (1984). Function and process in spoken word recognition. In

H. Bouma & D. Bouwhuis (eds.). Attention and performance: Control of language
processes. Hillsdale, NJ: Erlbaum.

Mathesius, Vilem. (1975). Excerpts from A Functional Analysis of Present Day

English. In J. Vacheck (ed.); L. Duskova (trans.). Prague: Academia. Pp. 81-85, 100-103,
153-160, 192-202.

McCarthy, J. (1979a). Formal problems in Semitic phonology and morphology.

Ph.D. MIT [Published 1985: New York: Garland Press]

McCarthy, J. (1979b). On stress and syllabification. Linguistic Inquiry 10, 443-

466.

McCarthy, J. (1980). A note on the accentuation of Damascene Arabic. Studies in

the Linguistic Sciences 10 (2), 77- 98.

McCarthy, J. & Prince, A (1993a). Prosodic Morphology I: Constraint interaction

and satisfaction. Ms. University of Massachusetts, Amherst, and Rutgers University,
New Brunswick, NJ.

McCarthy, J. & Prince, A (1993b). Generalized alignment. Yearbook of

Morphology, 79-153.

McClelland, J. & Elman, J. (1986). The trace model of speech perception.

Cognitive Psychology 18, 1-86.

Mitchell, T. F. (1960). Prominence and syllabification in Arabic. Bulletin of the

School of Oriental and African Studies, 2, 369-389.

Myers, S. (1987). Vowel shortening in English. Natural Language and Linguistic

Theory 5, 485-518.

267
 O’Connor, J. (1980). Better English Pronunciation. 2nd edition. Cambridge: Cam-
bridge University Press.

Pater, J. (1995) On the nonuniformity of weight-to-stress and stress preservation

effects in English. [ROA-107, http://ruccs.rutgers.edu/roa.html].

Phillips, B. (1983). Lexical diffusion and function words. Linguistics 21, 487-499.

Prince, A. (1980). A metrical theory for Estonian quantity. Linguistic Inquiry 11,
511-562.

Prince, A. (1983). Relating to the grid. Linguistic Inquiry 14, 19-100.

Prince, A. (1990). Quantitative consequences of rhythmic organization. In M.

Ziolkowski, M. Noske, & K. Deaton (eds.). Papers from the Chicago Linguistic Society,
26 (2), 355-398.

Prince, A. & Smolensky P. (1993). Optimality theory: Constraint interaction in

generative grammar. Rutgers University, New Brunswick, NJ. Report RUCCS TR-2.

Prince, A. & Smolensky, P. (1997). Optimality: From neural networks to

universal grammar. Science 275, 1604-1610.

Prince, E. (1981). Toward a taxonomy of given–new information. In P. Cole

(ed.). Radical pragmatics. New York: Academic Press, 223-255.

Prince, E. (1992). The ZPG letter: Subjects, definiteness, and information-status.

In W. C. Mann & S. A. Thompson (eds.). Discourse description. Amsterdam: John
Benjamins Publishing Co, 295-325.

Roach, P. (1983). English phonetics and phonology. 2nd edition. Cambridge:

Cambridge University Press.

Robinson. P. (1996). Learning simple and complex second language rules under
implicit, incidental, rule-search, and instructed conditions. SSLA, 18, 27-68.

Robinson, P. (1997). Generalizability and automaticity of second language

learning under implicit, incidental, enhanced, and instructed conditions. SSLA, 19, 223-
248.

Ruhenstein, H. & Pollack. I. (1963). Word predictability and intelligibility.

Journal of Verbal Learning and Verbal Behavior, 2, 147-158.

Sakarna, A. (1999). Phonological aspects of abadi Arabic: A Bedouin Jordanian

dialect. Ph.D. Diss. University of Wisconsin, Madison.

268
 Scott, V. M. (1989). An empirical study of explicit and implicit teaching
strategies in French. The Modern Language Learning Journal 73, 14-22.

Selkirk, E. (1984). Phonology and syntax: The relation between sound and
structure. Cambridge, Mass.: MIT Press.

Swinney, D. (1982). The structure and time-course of information interaction

during speech comprehension: Lexical segmentation, access and interpretation. In J.
Mehler, E. Walker & M. Garrett (eds.). Perspectives on mental representation. Hillsdale,
NJ: Erlbaum.

Tesar, B. (1995). Computational optimality theory. Ph.D. Diss. University of

Colorado, Boulder. [ROA-90, http://ruccs.rutgers.edu/roa.html]

Tesar, B. (1996). An iterative strategy for learning metrical stress in optimality

theory. [ROA-177, http://ruccs.rutgers.edu/roa.html]

Tesar, B. (1997). Multi-recursive constraint demotion, [ROA-2,

http://ruccs.rutgers.edu/roa.html]

Tesar, B. (1998). Robust interpretive parsing in metrical stress theory. [ROA-262,

http://ruccs.rutgers.edu/roa.html]

Tesar, B. & Smolensky, P. (1993). The learnability of optimality Theory: An

algorithm and some basic complexity results. [ROA-2, http://ruccs.rutgers.edu/roa.html]

Tesar, B. & Smolensky, P. (1996). Learnability in optimality theory (long

version) [ROA-156, http://ruccs.rutgers.edu/roa.html]

Tesar, B. & Smolensky, P. (1996). Learnability in optimality theory (short

version) [ROA-155, http://ruccs.rutgers.edu/roa.html]

VanPatten, B. (1994). Evaluating the role of consciousness in second language

acquisition: Terms, linguistic features and research methodology. AILA Review 11, 27-36.

VanPatten, B. & Cadierno, T. (1993). Explicit instruction and input processing.

SSLA 15, 225-243.

VanPatten, B. & Sanz, C. (1995). From input to output: Processing instruction

and communicative tasks. In F. Eckman, D. Highland, P. Lee, J. Mileham & R.
Rutkowski Weber (eds.). Second language acquisition theory and pedagogy. Mahwah,
NJ: Erlbaum.

269
 Welden, A. (1980). Stress in Cairo Arabic. Studies in the Linguistic Sciences 10
(2), 99-120.

White, L. (1990). The “pro-drop” parameter in adult second language acquisition.

Language Learning 35 (1), 47-61.

Wong, R. (1987b). Learner variables and prepronunciation considerations in

teaching pronunciation. In J. Morley (ed.). Current perspectives on pronunciation:
Practices anchored in theory. Washington, D.C.: Teachers of English to speakers of other
languages, 13-28.

Yan-Ping, Z. (1991). The effect of explicit instruction on the acquisition of

English grammatical structures by Chinese learners. In C. J. James & P. Garret (eds.).
Language awareness in the classroom. New York: Longman, 255-277.

270
Appendixes

Appendix 1: List of symbols

 ‫ ء‬voiceless glottal stop

q ‫ ق‬voiceless uvular stop
 ‫ غ‬voiceless uvular fricative
x ‫ خ‬voiceless uvular fricative
 ‫ ح‬voiceless pharyngeal fricative
 ‫ ع‬voiced pharyngeal fricative
š ‫ ش‬voiceless palato-alveolar fricative
j ‫ ج‬voiced palato-alveolar affricate
 ‫ ض‬emphatic /d/
T ‫ ط‬emphatc /t/
S ‫ ص‬emphatic /s/
 ‫ ظ‬emphatic //

Appendix 2: Preception of Arabic word-stress by native speakers of Arabic

Native speakers of Arabic are often unable single out the syllable which is most

prominent in a word even if produced by a native speaker of the language. This is

probably because word stress in Arabic is nonphonemic; “not only … rules of

prominence are stable for a given colloquial but also … correspondence between

colloqials or between a colloquial and a given ‘Classical’ pronunciation is equally

regular’ (Mitchel 1960: 369).

In order to see that this is really the case, a simple perception test is carried out.

Ten native speakers of Arabic were asked to single out the syllable that is most prominent

18 words produced by native speaker of Arabic. Each word is produced as many times as

the listeners demanded. The following Table shows the responses of the listeners on the

perception test :

271
 L1 L2 L3 L4 L5 L6 L7 L8 L8 L9

adwiyatuhumaa ad hu wi wi/hu ad ad tu ad/tu tu ad

marifatuhu ma ma tu ma ma ma tu ma/tu fa ma

jalalatuhu la la la la/la la la/la tu tu tu la

inkasara in ka ka in/ka ka in ka in in in

inbasaTu in ba in ba in in ba in/Tu in in

mustamaliin mus ta mus mus mus mus ta muus/ta mus/ta mus

mastalmathaaš tal tal mas mas mas tal haaš mat tal ….

maqaaid id id qaa qaa id id qaa id qaa qaa

almustašfaa mus mus taš taš taš mus taš al/mus/taš mus taš

muqaddimaat qad di di qad di qad qad di qad di

mutaTallabaat Tal ta ta ta/Tal Tal Tal mu ta/Tal/la Tal la

maraaji ji ji raa raa raa ji raa ji raa ji

almauub ma ma ma uub uub ma uub ma mu ma

falayubattikunna bat ti ti bat bat bat yu bat/ti/na. bat ti

baddada bad da bad bad bad bad bad bad/da bad da

jajaah ja ja ja ja/ja ja ja ja ja/ja/ah ja ja

kataba ta ka ta ka ka ta ka ba ba …

mutaTallabatuhunna Tal hun na mu/ta Tal Tal tu ta/Tal/la tu na

272
The table above shows two important observations:

(1) None of the subjects perceived all of the the test words with the correct stress

patterns. In fact, on the 18 test token, the highest score is 10 and the lowest score

is 0.

(2) None of the test tokens is perceived with the correct stress pattern by all the

subjects. All of this shows that native speakers of Arabic are not aware of this

linguistic feature. To all of them, stress is nonphonemic.

273