PROSODIC DOMAINS IN OPTIMALITY THEORY

Dominique Roder

Department of Linguistics

McGill University, Montréal

August 1998

A thesis submitted to the

Facul ty of Graduatc Studies and Reseacch

in partial fulfillment of

the requiremen ts for the degree of

Doctor of Philosophy

O Dominique Rodier August 1998

National Library 1*1 of Canada Bibliothèque nationale du Canada

Acquisitions and Acquisitions et Bibliographie Services services bibliographiques

395 Wellington Street 395, rue Wellington OttawaON K l A W Ottawa ON K I A ON4 Canada Canada

The author has granted a non- exclusive licence allowing the National Library of Canada to reproduce, loan, distribute or seil copies of this thesis in microform, paper or electronic formats.

L'auteur a accordé une licence non exclusive permettant à la Bibliothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la foxme de microficbe/lfilm, de reproduction sur papier ou sur format électronique.

The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantid extracts fkom it Ni la thèse ni des extraits substantiels may be printed or otherwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation.

PROSODIC DOMAINS IN OPTIMALITY THEORY

Cross-linguistically, the notion 'minimal word' bas proveâ fruidul grounds for esplanatory

accoun ts of requi rements i m posed on morpholo@cal and phonological constituents. Word

minimality requires that a le-ical word includes the main-stressed f m t of the language- As a

result, suh in imal words arc augmented to a bimoraic foot through diverse strategies like

vowel lengthening, syllaMe addition, etc. Even languages with numerous monornoraic

!esical words ma>* impose a minimality requirement on derived words that would othenvise

be smaller than a well-forrned foot. In addition, the minimal word has been argued to play a

central role in characterizing a p d i c base within some morpho-prdic constituent for

the application of processes such as redupt ication and idisation.

The goal of this thesis is to offer an explmation as to why and in which contests

grammars may preîer a p r d i c constituent w hich may not be reduci Me to a bimoraic foot.

1 proiide esplanatory accounts for a number of cases urhere the prosodic structure of

morphological o r phonological constituents cannot be defined as coestensive with the main

stressed foot of the lanpage. To this end, 1 propose to add to the theory of Prosodic

Structure (Chen 1987; Selkirk 1984, 1986, 1989, 1995; Selkirk and Shen 19%)) within a n

optimalic-theoretic frarnework by providing evidence for a near level within the Prosodic

Hierarchy, that of the P r d i c Stem (PrStem).

An important aspect of the mode1 of p r d i c structure proposed hem is a notion of

headship which follows directly from the Mit Hierarchy itself and h m the metrical

grouping of prosodic constituents. A theocy oî prosodic heads is developed which assumes

that structurai consûaints can impose wel1-fmedness requirernents on the prosodic shape

and the distri bu tion of heads wi thin morphological and phonolqical consti tuents.

LES DOMAINES PROSODIQUES DANS LA THEORIE DE L'OPTIMALITE

A travers les systèmes linguistiques, la notion d e 'mot minimal' s'est trouvée être très riche

en solutions explicatives en ce qui concerne les demandes prosodiques imposées à certains

constituants morphologiques et phonologiques. La notion & rninimalité demande que le

mot lesical wntienne au moins un pied bien formé, celui recevant l'accent fort du mot. Le

mot sous-minimal est donc augmenté de diverses façons afin & devenir bimotaïque. Même

les langues aux nombreux mots lexicaux ne consistant que d'un pied monomoraique

peuvent demander que les mots dérivés morphologiquement soient au moins bimorai'ques.

De plus, le mot minimal a été proposé comme le facteur principal dans la délimitation d'une

base prosalique à l'intérieur d'un constituant prosodique ou morphologique pour certains

procédés comme la réduplication et I'infkation.

Le but d e cette th* est d'offrir une esplication, ii savoir pourquoi e t dans quel

conteste une grammaire peut choisir un constituant prosodique qui ne soit pas coextensif

avec le pied bimoraïque. Dans ce but, la présente thèse propose une extension à la théorie

de la Structure Prosodique (Chen 1987; Selkirk 1984, 19û6, 1989, 1995; Selkirk and Shen

1990) à l'intérieur de la théorie de I'Optirnalité avec introduction d'un nouveau niveau

métrique à l'intérieur de la Hiérarchie Prosodique, celui du radical prosodique (RStem).

Un aspect important du modèle de la structure prosodique propose dans cette thèse

est une notion d e la tête prosodique qui découle directement d e la Hiérarchie Prosodiqw et

des groupements métriques des constituants prosodiques. La théorie développée dans cette

t h h e concernant les têtes prosodiques présuppose des contraintes structuraies imposant

certaines conditions à la f o m e prosodique et à la distribution des têtes à l'intérieur des

constituants morphologiques e t phondogiques.

Acknowledgments

1 t is entirel y due to Glyne Pi ggott that this dissertation was ever completed. 1 am especially

indebted to him, not just for his comments and crïticisms, but also for his patience and

good humour. He tau@ me phonology, but more important& he taught me to be open-

minded and even enthusiastic when f a d with novel approaches and new ideas.

To the whole department of Iinguistics at McGill, 1 also owe a great deal of gratitude for

their support al1 through these years.

Je tiens aussi a remercier mes parents pour leur amour et compréhension. Je tiens à les

remercier pour la confiance qu'ils m'ont accordée à üavers toutes la etapes qui ont mené à

la complétion de cette th&. Je leur suis redevable d'une Y de précieuse.

Table o f Contents

.................................................................................. 1 . Introduction 1 ................................................. 1.1 The Theory of Prosodic Structure - 2

................................................................... 1.2 Optirnality Theory - 4 ............................................................ 1.3 Some Roblematic Cases 5

........................................... 1.3.1 Minimally Disyllabic Domains 5 ...................... 13.2 Bimoraic Syllables as the Main-Stressed Foo t.. -7

..................................................... 1.4 A Mode1 of Prosodic Structure 9 ........................................................ 1.4.1 The Prosodic Stem 9

....................................................... 1.4.1 Prosodic Structure 1 1 ............................... 1.4.2 Headedness and Metrical Constituents -13

.................................................................... 1 -5 Thesis Overview -14

..................................................................... 2 . Japanese Truncation -16 ............................................. 2.1 The Bimoraic Foot as Minimal PrWd 17

........................................................ 3.3 The Zoose' Minimal Word 19 .................................................................. 3.3 Previous Analyses 2 1

............................ 2.4 Japanese Truncation and Correspondence T heory -27 .................................... 2.4.1 The Emergence of the Unmarked -27

.......................................... 2.4.2 The Emergence of the RStem 31 ............................................................ 2.4.3 Compoundi ng 36 ........................................................... 2.5 Bound Truncated Stems -37

................................................ 2.5. L Hlpocoristic Formation -38

............................................ 2.5.2 R d i c Subcategorization -41

............................................ 2.5.3 Lesically-Specific Ranljng -44

3 . -4rabic Stem Patterns ....................................................................Sû .................................................................. 3.1 Previous Anal y ses -51

............................................ 3 -2 Root-and-Pattern Morphology in OT -55 ...................................... 3.2.1 The Prosodic Structure of Arabic 55

.......................................... 3.2.2 Consniant 'Extrarnetricali ty' -57

..................................................... 32.3 The Disyllabic Foot -62 ....................................................... 3.2.4 Mora Fai thfulness -66

.......................................................... 33 The Verbal Stem System -70 .................................... 3 3 . 1 Output-teoutput Correspondence -72

....................................................... 3.3.2 Double Refixation 7û

........................................................... 4 . Heads and Crid Prominence 8 1 ................................................. 4.1 Stress as Rominence on the Grid 81

.................................................... 4.2 The Weight-to-Stress Principle 83 .......................................................... 4.3.1 Classicd Arabic 83

............................................................ 3.2.2 Cairene Arabic 90

........................................................ 3.2.3 Palestinian Arabic 95

............................................... 5 . Binaq and Ternary Stress Systems 103

............................... 5.1 The Binary Stress System of Guugu Y imidhirr 103 ......................................... 5.1.1 The Initial Disyllabic Domain 104

................................... 5.1.1 Disfllabicity in Optimality Theory 105

................................. 5.1.3 The M t e m and Initial Disyllabicity 109

.................................... 5.1.4 lambic Feet in a Trochaic System 113 ................................... 5.1.5 Headedness and Grid Prominence 116

........................................................... 5.2 Temary Stress Systems 119 ............................................................... 5-21 Ca>uvava 120

.......................................... 5.2.1.1 Previous Analyses 131 .......................... S.?. 1.2 Temary systems and the PrStem 125

................................................................. 4.3.2 Estonian 131 ............................ 5.2.'. 1 The Stress System of Estonian 133

....................... 5.2.2.2 B i n q and Ternaq Stress Patterns 136

......................................... 5.2.2.3 Overlong Syllables 142

......................................... 6 . The Prosodic Stem and Syllable Weight 1 4 ............................................................................ 6.1 Chugach 148

...................................................... 6.1.1 Revious Analysis 150 ....................... 6.1.2 The Chugach Stress System and the M t e m 153

............................................................................ 6.2 Mohawk 159 ..................... 6.3.1 Epenthesis and the Stress System of Mohawk 160

.............................. 6.2.2 The Rinciple and Parameter Approach 162

......... 6.23 Tonic Lengthening in a 'Quantity-Insensitive' Languap 165 ................................................ 6.2.4 Epenthesis in Mohawk 169

..................................................................................... Conclusion 176

..................................................................................... References 17û

Chapter 1

Introduction

Cross-linguistically, the notion 'minimal word' has proven to bc fruitful grounds for

c s p l a n a t o ~ accounts of prosodic requiremenîs i mposed on morphological wnsti tuents

(McCarthy and Prince 1986, 199ûab; Spring 1990; Poser 1990; Mester 1990). Word

minirnality requires that a lexical word include at least the main stressed foot of the

language. As a result, subminimal words are augmented to a bimoraic foot through diverse

strategies like vowel lengthening, syllaMe addition, etc. Even languages that have

nurnerous monomoraic lexical words may impose some mi nimalit). requirement on

morphological i y d e r i d words that u-ould othenvise be smaller than a well-formed foot. 1 n

addition, the minimal word has been argued to play a central r d e in delimi ting a prosodic

base within some morpho-prosodic constituent for the application of processes such as

reduplication and infixation. Intcrestingl y, word minimali ty requirement on 'dcrived' words

often brings ~5th i t the additional requirement that these words be no larger than one foot.

The goal of this thesis is to offer an esplanation as to why and in which coritem

grammars may prefer a prosodie constituent which is not reduciMe to a bimorak foot In

this thesis, a number of cases where the prosodic stnicture of some morpholo@cal or

phonological constituents cannot be defined as cœxtensivc with the main stressed foot of

the language are aven a unified account within an optimality-theoretic frarnework To this

end I propose to add to the theory of Prosodic Structure (Chen 1987; Selkirk 1984, 19'86,

1989; Selkirk and Tateishi 1Bû; Selkirk and Shen 1990) by providing evidence for a new

metrical level within the Prosodic Hierarchy, that of the P r d i c Stem (PrStem). 1 will

show that the addition of the PrStem to the Prosodic Hierarchy elirninates the need to resort

to espiiclt statements such as "P-Stems can be no smaller than two syllables" (Downing

19981, or "The Reduplifant is left and nght-aligned with the corraponding e d p of

different syllables" (McCarthy and Prince 1993b) in order to account for disyllabic

requirements irnposed on prosodic constituents.

My proposal hdds that this sort of disyllabic requirements c m be @en a unified

account by means of structural wnstraints that are active throughout the phonology of the

language. It will be shown that the choice between the different prosodie structures follows

from the specific rankngs of these structural mnstraints in some language. 1 will further

argue that these consuaints play an important part not only in the prosodic structure of

lexical and derived words but aiso in the stress system of languages.

An important aspect of the mode1 of p r d c structure proposod here is a notion of

headship which follows directly from the Prosodic Hierarchy itself and from the metrical

groupings of prosodic constituents. A th- of prosodic heads is developed which

assumes that structural constraints can impose well-formecines requirements on the

prosodic shape and the distribution of heads within morphoIogicai and phonologkai

consti tuents.

This chaprer is laid out as follows: In 8 1.1. a brief outline of the basic tenets of the

thcoq of Prosodic Strucnire is presented. 1.2 descri bes the central claims of Optimality

Theory. In 5 1.3, a number of cases where some morphological or phonological dornain

cannot be easily defined as cœxtensive with the main-stressed f a t of the language are

csamined. In 5 1.4, the mode1 of prosodic structure developed in thk thesis is introduced.

FinaIly. 5 1.5 presents an overview of the thesis.

1.1 The Theory of Pmsodic Structure

The theory of Prosodic Structure hot& that the prosodic mgankation of sentence

phonology is defined on morpho-prmdic constituents that correspond to. but are distinct

frorn, the morpho-syntactic constituents of the sentence. According to this theory of the

s yn taxlprosod y interface, sentences are orgnized into pnisodic structures defined in terms

of the categones specified by the Pmsodic Hierarchy.

1. The Prosodie Hiemrchy (Selkirk 1995)

Utt Utterance I

IP in tonational phrase I

PPh phonological phrase I

PWd prosodic word I

Ft foot I

The Strict Layer Hypothesis (Selkirk 1984; Nespor and Vogel 1986) impcxses structural

constraints on the p r d i c categories specified by the Prosodic Hierarchy. Conceived as a

singie constra.int, the Stnct m e r Hypothesis requires that a prosodie categq of level i

immediately dominates only categories of the nest level i-1. Inlcelas (1989) and Itô and

Mester ( 1992) have argued, hoivever, that the Strict Layer Hlpothesis should instead be

subdiwded into a number of independent well-fonnedness conditions. Selkirk ( 1 9 5 ) , for

esample, proposes the following set of constraints on prosodic domination.

3. Constraints on Prosodic Domination (where Cn = some pr&c categov)

( i ) Lqeredness No Ci dominates a O, j z i e.g. "No a dominates a Ft."

( i i ) Headehess Any Ci must dominate a Ci-l (errcept if Ci = O)

e-g- "A PrWd must dominate a Ft."

(ii i ) Exhamivity No Ci immediately dominatcs a constituent Ci, j < i- 1 e.g. 'No PrWd irnmediately dominates a a. "

( i v) Nonrecursiviîy No Ci dominates a constituent CJ, j = i e.g. "No Ft dominates a Ft,"

Taken together, Liyeredness and Headedness define the hierarchical organization of the

Prosodic Hierarchy. Both appear to hold universally. Euhaustivity, on the other hand, is

reguiarl y violated in numerous languages where syllables are found immediatel y dominateci

by a PrWd (cf. Inkelas 19û9; Itô and Mester 1992; Rinœ and Smolensky 1993; McCarthy

and Pnnce 1993ab; Kager 1996ab). No~ecursivity has also been argued to be a violable

constraint, at least at the level of the PrWd and higher (Inkelas 1989; McCarthy and Prince

1993ab; Selkirk 1995; Kager 1996a).

Another class of coristraints crucial to the well-fonnedness of prosodic structure is

defined in ternis of the alignment of edges between morpho-syniactic structure and prosodic

structure. These constraints require that the right/left eûge of some grammatical constituent

coincides with the corresponding edge of some phonoiogical constituent

The t h e o ~ of Generalized Alignment propowd by McCarthy and Rince (195r3ab)

within Opmality Theory extends the notion of edge a l i m e n t to include a correspondence

of edges between word-interna1 morphological constituents and word-internai metrical

constituents. These constraints are defined following the general scherna given below:

where

Cat 1, Cat2 E PCat U GCat ( R d i c and Grammatical categories) Edgei, Edge2 E {Right, Left)

GCat consists of morphological categories (Word, Stem, Affix, Root, etc.), uhile PCat

consists of prosodic categories (Mora, Syllable, Foot and Rosodic Word). Informaily,

Align(Catl, ..., Cat2, ...) is r a d as follows: "the designated edge of every Cati coincides

with the designated egde of some Caiz".

1.2 Optimality Theory

Optimali ty Theory hdds that grammars consist of a set of u n i v e d constraints. Surfacing

outputs result from the lanpge-specific d n g s of these constraints. The main insight of

this theory is that the selected outputs need not be "perfect" with respect to the various and

sometimes conflicting constrain&. The chosen or surfacing outputs need only to be optimal,

i-e. the best of al1 possible outputs within a partndar grarnmar, given the underlying

segmenta1 structure of the input and the hierarchical dominance of certain constraints over

others in a particular grammar.

In the optimaiity-theoretic frarnework, constraints can be roughl y divided into two

classes. Those that demand perfect identity between input and output forms (faithfulness

constraints) and those that impose well-formedness requirements on p r d i c constituents

(structural constraints). One important aspect of Optimalit?; Theory is the daim that the

complesity of surfacing outputs results f r m the domination by faithfulness wnstxaints of

rnost of the high-ranking structural constraints of the language. Lanpuage diversi ty, on the

other hand, follows from the way in which each language's grarnrnar chooses to rank those

tnro sets of constraints among themselves and ~ l t h respect to each other.

1.3 Some Problematic Cases

According to the Strict h y e r Hypothesis of Selkirk ( 1 W), the ideal situation is a prowâic

~vord that only consists of one or more feet, either bimoraic or disyllabic, i.e. no syilable is

parsed directly by the PrWd (i.e. no Exhaustivity \-iolations). This is not aiways possible,

how.ever, particularly in odd-syllable words or in words in which the position of heavy

s y1 lables in tederes w i th the metricd grou pi ng of syllables into b i q cons ti tuents.

At issue here is how to a m u n t for morphdogical phenomena and stress patterns in

which the distribution of prosodic constituents l aves syllables unfooted in contests where

a more satisfactory parse in regad to the Strict Layer Hypothcsis could be obiaincd. Also at

issue is the treatment of minimally disyllabic constituents in languages where the bimoraic

syilablc is the main-stressed foot.

1.3.1 Minimally Disyllabic Domains

One wcll-known example of a minimal1 y disyllabic requirement on prosodic consti tuents

cornes from the formation d- loanword abbreviations in Japanese (1 tô 1990; 1 tô and Mester

1992). As extensively demonstrated in Poser ( 1984, 1990), narne shortenin? in Japanese

confoms to a bimoraic template, either (9%) or (c+& Loanword clippings. however,

must be minimally disyllabic. The proMem is compwncied by the fact that trisyIIabic

constituenis are perfeztly acceptaMe word clippings, as shown in (4c-d).

4. Word Clippings (1 tô 1990; Itô and Mester 1 =)

a ( a 4 su to (raiki) 's tri ke' hi su (terii) 'Hysterie' (Gcrman) ra bo (ratori) 'laborator).'

b. ( P r) (P P) to ri Lw ro (metirehi) 'trichlorethylene' koN bi ni (ensu) 'convenient store' baa teN (daa) 'bartender'

C. (PCC) P d - (au) r day ya (moNdo) 'diamond' te re bi (2on) 'television' koN bi (neehn) 'combination' b a s u ke (tto) 'basket'

Word clippings have a minimal p d i c shape of two syllables and a maximal p d c

shape of four. The word clippings in (442-d) represent what Itô and Mester (19!Z) refer to

words that consist of a bimoraic f m t followed by a syllable as loose' minimal words.

An even more sîxiking pattern is found in Guugu Yimidhirr. an Australian language

s p k e n in Queensland (Haviland 1979). Both i t s morphology and phonology regulariy

refer to an initial prosodic domain that is precisely two syllables, irrespective of syllable

wcight. For esample, main stress and vowel length contrasts are restricted to the first two

syllables of the word. As shown by the data below, main stress is initial (Sa-d), unless a

long vowel is found in the second syllable of the word, in which case the long vowel

receives main stress (Se-g). When both the first and second syllables contain a long vowel,

both syllables receive main stress (31).

5. The Distribution of vowel Iengrh (Kager 19%a) a. miil 'eye' e. dawdac

b. IV&@ 'crow ' f. bu1 bducmbul

c . w q i @ 'moon' g. Qam&cbinà

d. guu[umùgu 'meat hawk' h. daarslali)àn

'star'

'phcasan t'

'magpie goose'

'kangaroo'

The secondary stress ptem is detennined by the position of long vowels within the word.

If there is only one heavy syllable in second position, then stress is iambic, othenvise the

secondary stress pattern is trochaic. Obviously, the initial disyllabic domain of Guugu

Y irnidhirr cannot be defined as a fwt, since it mas consist of two bimoraic syllables (5h).

1 At Bimoraic Syllables as the Main-Stressed Foot

In the Rght-headed stress system of Chugach, one of two dialects of Pacific Y upik, long

\.owels and diphthongs are obligatorily parseâ as a unique foot (Leer 198Sab). Closed

syilables, on the other hand, are usually parsed as part of a disyllabic foot. In fact they are

often found in the w e . position of a foot.

6. Chugach Ternury Stress Sysrem (Hayes 199s)

a. a. tu.qu.ni.ki 'if he (refl.) uses them' b. an.& .qu. kut 'well go out'

c. nuy.jrai 'her hai r'

'apparentlu reading i t' 'she lied to me'

'1 will backpack'

Word-initially, however, a closed syilable is always parsed as a monosyllabic foot (6b, e-

0. In addition, the onset of a long vowel or diphthone becomes geminated when preceded

by a word-initial short open syllabie in order for the initial syilable to be parsed as a foot

(6c). The challenge presented by Chugach, apart f r m the probiem of weight variability in

closed syllabies, is how to define the trisyllabic p u p i n g L U and the disyllabic p u p i n g

HL, urhile acmunting for the fact that a bimoraic syltabie is never parsed as the head of a

disyllabic foot (LH), usually regarded as the mmt harmonic iambic foot (McCarthy and

Prince 1 99Ob; Prince 1990).

The Icft-headed stress system of Mohawk presents a similar problern. Although the

ianguage does not have ünderlying long vowels, whenever the stressai syllable is open, the

vowel is lengthened, resulting in a HL sequence at the nght edge. Mohawk has main stress

o n the rightmost nonfinai syllaMe. Interestingly, as s h m n by the &ta belou., the presence

of an epenthetic (underlined) vowel in the final or penultimate syllaMe results in the

retraction of main stress to the antepenultimate syllable (7d-f).

7. Mohaw Stress Paneni (Piggott 1995) a. k-atirut-ha? @catinitha? 1 '1 pull it' b . uak-ashet-u [wakas hé: tu] '1 have counted i t'

c. k-hyatu-s [khy&tus] '1 write'

d. te-k-nk-s [&griks] 'I put them togethet' e. t-A-k-ahsutr-A-? [th k a h d t g r ~ 31 'I will splice it' f . A -k-arat-3 ka-rate? 1 'I lay myself down' g. rdwt-ot-7 [rokii:to@ J 'he has a bump on his nose'

W hat is most stri king about the data above is that a stressed vowel never lengthens when

immediately fdlowed by an epenthetic vowel(7d-e). Another aspect of the dismptive effect

of epenthetic vowels c m be found in their invisibility to the word-minimality condition.

Subminimal stems (Le. stems with only one undedying vowel) are a u p e n t e d through the

insertion of a prothetic [il.

8. a. k - y ~ - s [iiqhs] b. k-ek-s [i:kelrsJ

c. s-riht [isgriht]

d . t-n-ehr-3 [itgnehrg? ]

As show in (&-d), augmentation resul ts in a metrical grouping that is minimdl y trisyllabic

and ma~imall y quadrisyllabic, whenever an epenthetic vowel is present. Thus, although the

main stress f m t is bimoraic, the minimal word in Mohawk is not reducible to either a

bimoraic or disyliabic foot.

An even more interesting pattern is found in the stress system of Estonian (Prince

1980). Although the language seems at first glance to have heavy closed syllables (Le. final

CvCC syllables are stressed), closed syllables are found in both the strong and the weak

positions of a disyllabic foot.

9. Binary Stress in Estonian Tematy Stress in Estonian

a. té.rav&md.t tt5.rava.màl.t 'more skillful, gensg.'

b. pAri.màt.te1.t *#ri.mat.t&l.t 'the best, abl.pl.'

c. pi.mes.tà.va1e pi.mes. t a x i l e 'blinding, ill.sg.'

d. pLmes.tà.t.as.se p i mes. ta.vàs.se 'blinding, 111-sg.'

i. pi.mes.tàt.tu. te *pi.mes.tat.tù. te 'the dazelled, gen.pLf

j. u.1is.îà.vamài.t u-lis. ta.viî.mai. t I f . . .

The Estonian stress system differentiates between closed and open syllabies in two ways.

First, a final superheavy syllabie (CwC and CVCC) that cannot be parsed by a disyllabic

foot aluTays fonns a foot on its own. This is shown in the t e m q pattern (i.e. second

column) of (9a-b) and in the binary pattern (i.e. the first column) of (9j) Second, in the

optional te- stress pattern, o d y an open syllable can be left unfooted. A closed syllabie

must be parsed as the head of a following foot (9b, i).

Thus, the ?stem of Estmian must ei ther allow for a ('LH) trochaic f o d or one must

assume that weight is variable in the language. The problem wi th the latter assumption, is

how to define the contest in which a closed syllable is treated as light (or heay). Any

analysis of the stress systern of Estonian must also account for the variably disyllabic and

tnsyllabic groupinp.

1.4 A Model of Prosodic Structure

The fact that, in languages as diverse as Japanese. Guugu Y imidhirr and Mohawk where

the main stress foot is bimoraic, some prosodic or morphological constituent may require a

minimally disyllabic shape shows that the assumption thai word minimality is reducible to

the notion of f m t minimality cannot account for al1 the minirnality requirements that may be

imposed on prosodic constitwnts. Any mnstihmt that is minimally two qdlables and

ma. mal 1 y mro feet m o t be treated as cœstensive with the main stressed foot.

1.4.1 The Prosodic Stem

Following Inkelas ( LW), Downing ( 1998) proposes that the Rosodic Hierarchy should

be entended to include the sub-lexical m o r p h o - p d c constituents P-Stem and P-Rwt.1

- -- -

I Affixes. like fuactioa wads, w d d not be part of the Prosodic Hierarchy.

9

T hese morpho-prosodic mnstituents smaller than the phonologid word are distinct from

the metrical consti tuents which are ususaily considerd to be the sub-lexical mernbers of the

Prosodic Hierarchy.

1 O. The P-Hierarchy VS. The Metrical Hierarchy Utterance

1 i ntonational phrase Posi-lexical

I phonolo@cal phrase

1 ----- phonological word ----

I Prosodic Word

1

Downing's proposal allows for mismatches between some morphological constituent (M-

Constituent) and i ts corresponding morph~prosodic constituent ( P-Consti tuent). In other

words, the two consti tuents may be misaligneci to improve the structural well- formedness

of the P-Cons ti tuen t (al ways assuming that phonological processes are &fi ned on prosodic

domai ns rather than morph*syntactic domains).

Estendine Downing's P-Hierarchy and Metrical Hierarchy further, 1 propose t h t

the disyllabic requirement imposed on morph~prosodic and prosodic metrical constituents

reflecr the propcrtics of a metrical constituent, the Prosodic Stem (PrStem) which is found

bctwwm the RWd and the foot within the Pmsadic Hierarchy.

1 1. Metrical Hiemrchy Prosodic Word

I Prosodic Stem

I Foo t

I S y llable

Such a constituent bas also been proposed by Iiô (1990) for Japawse worâ dippings. but her assumptions about the prosodic stem are closer to &ose of Downing than to mine. She assumes the prosodic Stem to be rnorpholopically defined and thai t a m r y domains resuit f' tbe addition of a light syllable Affix to an obligatoril y bimoraic stem. Hcr proposal is discusscd in man: detail in 92.1.

According to this malel, the f m t is immediately dominated by some M t e m which by

transitivity is also defined as the head of the PrWd.

Downing assumes that there i s direct correspondence between M-constituents and P-

constituents. In my proposai, no such correlation is assurned berneen a phonological stem

(P-Stern) and a PrStem. In fact, given the possibility of recursive PrWd structures, not

a - c q R W d needs be in direct correspondence with some phonologicai word (P-Word).

Similarly, a phonologd stem may include more than one M t e m \rithin its morpho-

prosodic domain.

1.4.1 Prosodic Structure

Centrai to our conœrns in this thesis is the prosodic organization of the PrStem and its

distri bu lion within a PrWd. I will start by assuming that al1 sublcaical metrical constituents

are maximallu bina?, including the RStem.

12. BINARYBRANCHING COND Heads of metrical constituents are ma~imally b i n q .

This uni\.ersally undominated constraïnt demands that the sublesical metrical constituents

PrStem. foot and syllable maximally consist of a head and a non-head constituent. The

candidates in ( 13) il lustrate the prosodic organïzation of a M t e m uniquel y dominated by a

PrWd. Note that the bimoraic foot in the structures beiow also stands for a disyllabic foot.

13 a. PrWd b. PrWd c. PrWd I I I

PrS k m PrStem PrS tem I b

Ft Ft Ft I I I

;Uthou,oh there have been a few analyses that assumed othemise. rhe siandard assurnption in p r d c phonology is thai feet are maumally disyllabic and syllables mauxnaily bimoraic.

Thc srnailest prosodic structure in (13a) satisfies the Strict h y e r Hypothesis, as d œ s the

candidate structure in ( 13c). The candidate in ( 13b), on the other hand, minimal1 y violates

Eshausitivity. Two additional candidates which also violate Exhaustivity are given belou..

14 a. PrWd b. PrWd I

PrStem A

In ( 1 4 ) , the rightrnost syllabte is directly parsed by the PrWd, violating E~haustivity,

while both feet are immediately dominated by the PrStem, satisfying the constraint In

( 14b), the rightrnost f w t and the medial monomoraic syllaMe vidaie Eshaustivity

1 propose that of the three RStem structures which violate Exhausitivity, those in

( 13 b) and ( 14b) should be evaluated as more optimal than the candidate in ( 14a) with

respect to the parsing of pmsodic constituents.

15. PARSE-PRCAT Everv prosodic category is parsed in the head or non-head position of some metrical consiituent.

Hcre 1 assume that a relation behveen a head and a non-head is restricted to the same level

or one level down within the P r o d i c Hierarchy. That is, a syllable directly parsed by the

PrWd cannot be defined as k i n g in a non- hcad da t ion with either the head PrStcm or the

head foot of a PrStern. This notion d I d i t y can be expressed in ternis of 1 tô and Mester's

( 1992) Hierarchical Lncality condition w hich li mi ts the access to structural information to

on1 y one level down.

1 6. H i e r d i c a i Lmaiiiy A condition operatine at pr&c level Ci has access only to structurai information at Ci and at the subjacent levcl Ci-1.

According to the I d i t y condition, the head bot of a PrStem can only be in a relation with

another foot (level Ci) or a syllable (levei Ci-1) immediately dominated by the RStem. The

head PrStern of a RWd, on the other hand, cannot be in a head-non-head relation with a

syllable (level Ci-2 ) immediatly dominated by the RWd.

f 4.2 Headedness and Metrical Constituents

In the mode1 of prosodic structure developed hem, the Rosoâîc Hierarchy is assumed to

encode both constituent domination and headshi p. A problem inherent to the standard Men.

of most proponents of the intrinsic nature of heads lies in the assumption that abçence of

prominence on che g i d fdlows fmm the absence of foot structure in the output For those

rhat view headedness as imposed on metrical consti tuents, lack of prominence is assurned

CO result from the parsing of a headless constituent (Hune 1994; Crowhurst 1996; Hewitt

and Crowhurst 1996).

Satisfaction of Headedness, however, is usually assumed to be universal at least

with respect to the prosodic word and higher prosodic consthents (cf. Selkirk 1995). I f

Headedness is a requirement imposed on d l (non-terminal) suble?tical memcal constituents

of the Prososic Hierarchy, then headless feet (and headles Mtems) could never be well-

formed prosodic consti tuents?

Here 1 propose that the notion of headship resulting frorn the Rosodic Hierarchy is

independent of the notion of prominence of the grid My p m p d is laid out as follows: the

metrical grid only reflects relative p m i n e n c e between grid marks projected by the heads of

consti tucnts. Well-formedness constraints dealing with syllable structure and how syllables

are to be par& into metrical a~s t i t uen t s detemine the position of heads within the RWd.

The locafion of grid marks to the metrical grid, on the other hand, is deterrnined by

constraints defining what type of head may project ont0 the grid. Thus, lack of prominence

on the grid only reflects the non-projection of a grid mark by a ceriain type of head rather

than the absence of consti tuency or the existence of headless consti tuents.

The prosodic stnicrure of syllables, on the other W. is dependcnt on the segrnenid and moraic amtent of thc lexical inpic Although moras are regarded as the head of a syiiable, rhe maraic amtent of a syliaMe is often unpredictable. allowing for headless syUables. Nso. in coatrast with the metrical amstituents in thc Prosodic Hicrarchy, nioras are COIlfincd to the syUabie itself.

1.5 Thesis Ovewiew

In the rest of the thesis, theoretical and empirical evidence for the PrStem will be presented.

The analysis of strict1 y or minimally disyllabic constinients in languages where the bimonic

syllable (41i) is a possible foot uill be shown to d f e r strong eviâence for the PrStem as a

new metrical constituent within the Rosodic Hierarchy. This conception of an e ~ c h e d

Prosodic Hierarchy uill be shown to offer a restrictive and unified account of the possible

prosodic shapes that define the minimal RWd accross languages. Analyses of b i w and

t e m q stress systems will also provide support for the introduction of the M t e m by

demonstnting the r d e played by this new metncal constituent in longer words, where the

PrStem is not coextensive with the whole word.

In chapters 3 and 3, the diverse prosodic shapes of truncated words in Japanese and

the nominal and v e M systems of Arabic are a n a l y d and the r d e piayed by the M t e m is

shown to be crucial in the minimal shape of truncated nouns in Japanese and of nominal

and verbal stems in Arabic. In chapter 4, the stress systems of a number of Arabic dialects

also offer evidence for the position taken in this dissertation that headedness and grid

prominence are relateci but independent notions.

The first section of chapter 5 presents an anaiysis of the strictly disyllabic system of

Guugu Y imidhin: A number of properties of the main stress domain are accounted for by

assuming that the PrStem immediately dominates the main-stresscd foot of the langage.

The second section deais with the temary stress system of Cayuvava. It is arped that with

the addition of only one constraint and by p tu l a t i ng some rerankings for the constraints

afready introduced in the preceding chapters, most of the dirferences behveen the te-

pattern of Cayuvava and the binaq pattern of Guugu Yimidhirr can be accounted for in a

suaightfonvard manner. In the last section, the opona l ly binay and ternaq stress patterns

of Estonian are shown to follow frorn the rerankings of the structural constraints introduced

eariier in the dissertation.

Enally, chapter 6 presents an analysis of the iambic and trochaic stress systems of

Chugach and Mohawk respectively. It is argued that the variable weight of closed syllables

in some mntexts and the requirement that a heavy syllable be uniquely parsed as a foot

fol Ionv from requirernents on the parsing of metrical consti twnts and of their heads.

Chapter 2

Japanese Truncation

I n Optimality Theory (Rince and Smolensky 1993; McCarthy and Prince 1993ab, 1994),

language variation fdlo\vs from the way in which grammars choose to order the set of

univcrsally available constraints with respect to one another. One important aspect of this

framework is that the complexi ty of surfacing outputs rcsults from the domination of some

raithfulness constraints over most of the hiph-rankinp structurai constraints of the language.

1 n phonolo@cally-conditimed morphological processes such as redu pl ication and

tnincation, sets of correspondence constraints distinct from t h e regulating input-output

faithfulness are assumeci. One of the most interesting predictions of this frarnework is that

rcduplicated or t m c a t e d forms should allow for the surfacine of 'unmarked' structures --

'unmarked' in terrns of the high-ranking structural çonstraints of a patticular language

( McCarthy and Prince's ( 1994) "the emergence of the unmarked" or TETU).

McCarthy and Prince ( 1993 b, 1994) argue for e.xampIe that, when not constrained

by higher-ranked faithfulness constraints, the optimal PrWd consists of a t least one foot to

satisfy the Prosodic Hierarchy (HXEDSZS), every syllable is immediately dominated by a

foot (PARSE-~), which is minimaily bimoraic (FrBrs) and p e r k t l y a l i p e d with the edge d-

a PrWd (ALL-FT-L~/RJGHT).

As pointed out by Itô and Mester ( 1W-), the existence of a large class of syllable

groupings larger than a s inge bimoraic foot has dways been proMematic for any theory

that predicts word minimality to be reducible to foot minimality. Also problematic is the

esistence of di fferent phonologicall y-empty morphdogicai consti tuents wi thin a single

Ianguage that pduc t ive ly differentiate b e ~ e e n a bimoraic bot and a disyllabic f a . In

this chapter, a unified account of truncation phenomena in Japanese is provided in terms of

'the emergence of the unmarkci'. 1 will argue that the various prosodic shapes of the

truncation morphemes in Japanese resul t from the satisfaction of a number of h i g h - M n g

stnictural constraints which are minimailp violated in the general phonology of bpanese.

2.1 The Bimoraic Foot as Minimal PrWd

Itô ( 1990) makes the observation that Tobo Japanese has numemus monomoraic words

(e .g . su 'vinegar', n 'rice field', ko 'child'). These words are lexical mots that are not

augmented to bimoraici ty. in other words, Japanese allows for a subminimal foot as the

minimal PrWd. Bimoraicity imposed on 'derived' words, on the other hand, is a pervasive

process in the language. Poser's (1984, 1990) extensive studp of personal name shortening

in Japanese has demonstrated that morphological truncation conforrns to a bimoraic

constituent, i.e. a bimoraic trochaic foot. Three different types of truncation are discussed

in this section.

Hypocoristic formation in lapanese shortens personal names to a single bimoraic

b o t (Poser 1990; Mester 1990; 1 6 1990). (Only a placeless nasal (N). the first part of a

geminate or a nasal homorganic to a following stop may be found at the ri@ edge of a

syllabie in Japanese.)

1. Hypocoristic Fomution (Poser 1984, 1990) Midori MidcktN, Mii-&N Yooko Yok&aN, Yoo-ÇaN Akin A k i - k N Hiromi Hir&aN, Romi-CaN Mari ko Mari-&IN, Mako-CaN, Riko-CaN, Maa-ZaN JuNko JuN-i.nN Hanako Hana-&N, Haa-CaN, Hat-* Kazu hi ko Kazu-CaN

For a discussion of Iapanese hypocotistics in Optimali ty Theory . see Benua ( 1995). Suzuki ( 1995).

17

The tmncated f o m s are always followed by the Hypocoristic suffix 4aN. A bimoraic

shape, however, is imposed on d l the truncated forms.

In the second pattern, the Rustic Girls' narne formation, fernale names are also

shortencd to a bimoraic foot, Rustic Girls' names fake the homrific prefix O-.

2. Rustic Girls' Names (Poser 1990; Mester 1990)

Y ukiko Fumie Y ooko RaNko I c W

*Y uki *Furni *Y00 O-RaN *Ie (input is a last name)?

KirErna O-Kiri (input is a last name)

Thesc shortened names differ from Hqpconstics in that the truncatcd fonns in the Rustic

Girls' formation are an identical wpy of the initiai bimoraic foot of the base. No variation is

allowed in the mapping of the base to the bimotaic fmt.

The third type of truncation discussed in Poser (1990) involves the names of regular

clients of Geisha Houes and Japanese-style bars and restaurants. These clients are referred

to by a shortened form of their family names. Formation of the Geisha House Client narnes

differs from the two previous patterns in that these truncated forrns mnsist of only one

bimoraic syllable.

3. Geisha House Client Names (Poser 1990; Mester 1940)

Tanaka O-Taa-saN * o-TaN-saN * o-Tana-saN HoNda O-Hm-saN O-HoN-saN Sait0 O-Saa-saN O-Sai-saN *O- Sato-saN Ha ttori O-Haa-saN 'O-Has-saN * O-Hato-saN Y asuda O-Y aa-saN **Y as-saN *O-Y~su-saN

The truncated forms are obligatorily followed by the suffis -SUN and are preceded by the

honorific prefis O-.

Al1 three truncative morphemes surface wi th a prosodic shape that can be defined as

a minimal PrWd: i.e. an output rom whose prosodie shape is no smaller but also no larger

than one bimoraic foot. In the next section, two additional morphdogical processes in

The last two exampies are from Suzukï ( 1995).

18

Japanese are discussed. Both processes are characterized by a canonical3 prrisodic shape

that is larger that a foot.

2.2 The 'Loose' Minimal Word

The first pattern involving tnincated forms with a canonid prosodic shape larger than a

foot is that of word clippings discussed in Itô (1990) and Itô and Mester ( 1 9 E ) . This

system of loanword abbreviations is quite productive in Modem Japanese. One of the most

interesting characteristics of these truncaîed rorms is the ban on bimoraic monosyllabic

words. Tnincated forms such as *& (cf. daiyamondo 'diamond'), *koN (cf. koNbineeSon

'combination') o r *p (cf. paamaneNto 'permanent') are impossible word clippings in

Japanese. A nother i nteresting characteristic or these loanword abbreviations is that they can

only begin with a foot, and not a syllable, i.e. ('app) or ('apop) (cf. derno vs. *demoN

from demoNsutoreeSoN 'demonstntion'). The full nnge of word clippings is illustrated

4. Word Clippings (Itô 1990; Itô and Mester lm-) a. (a 0)

su to (raiki) hi su (terii) ra bo (ratori)

'stri ke' 'Hysterie' (German) 'laborator)rt

O pe (reesion) 'operation'

b- (r r) (c' CI) to ri ku ro (roetireN) a su pa ra (gasu) koN bi ni (eNsu) baa tcN (daa)

koN bi (neebn) ap pe (Ndisitisus)

'trichlorethylene' 'asparagus' 'convenient store' 'bartender'

'permanent (perm hair style)' 'diamond' 'combination' 'Appendizi ti s' (German)

1 use the term cm~onical pmsodic shape to rder to the maximal expansicm of a partida posodic category.

d. (0 4 P

te re bi (ion) ba su ke (tto) a ru mi (puurnu)

'television' 'basket' 'aluminium'

'animation'

Word clippings have a minimal prosodic shape of two syllables and a masimal prosodic

shapc of four. The word clippings in (4c-d) represent what 1 6 and Mester ( 1 W-) refer to

as 'loose' minimal words: i.e. words that consist of a bimoraic foot fdlowed by a

monomoraic syllaMe. Those in (4b) are described as supraminimal words.

Another set of data involving a prosodic shape larger than a single foot is provided

by zuuja-go, a secret language among Japanese jazz musicians. in their analysis of the

reversi ng argot muja-go, 1 tô and Mester ( 1992) show that the output patterns of this word

game can be roughly divided into hvo catepries depending on the size of the input. I f the

input has three moras o r less, the output is usually trimoraic. If, on the other hand, the

input is four moras or more, the output must consist of a sequenœ of syllables that can be

parsed into two feet.

5. Input

$ 3 ~

2 4 ~ 1 4p koohii 4p ikebana

Sp m a n e e k 6p toroNbooN

m u si i SUU

Van0

Ft + hii bana

iaa boN

le y es'

'f ire' 'jazz' 'food ' 'base' 'piano'

'coffee' 'flower arrangement'

'manager'

'recorder'

Cnicially, the zuuja-go pattern is characterized by an initial bimoraic fmt fdlotved by

another prosodic categq, either a syllable o r a foot. In the next section, an overview oC

previous analyses of Japanese truncation within a serial framework is presented.

2.3 Previous Analyses

The problem presenied by tnincated forms in Japanese is twdold. First, we need to explain

the prosodic requirements thal impose restrictions on the parsing of a bimoraic f a : Le. the

ban on disyllabic feet (opp) in the formation of Geisha House Client narnes and the ban

on monosyllabic feet (avci) in the formation of rvord clippinp. Second, we need to account

for p r o d i c shapes that are larger than the main-stressed f w t the trimonic (pp)op and

quadrimoraic (pp)(pp) f o m s in word clippings. An additional challenge is how to restrict

the maximal prosodic shape of the truncated forms to two feet

Most of the previous analyses of Japanese truncation have dealt wiih only a subset

of the data i ntroduced in the preceding section. Mester ( 1990), for example, focuses on the

three productive processes or personal name shortening presented above. In keeping with

the theory of P r d i c Morphology (McCarthy and Rince 19-b; Lombard and

McCarthy 199 1 ), Mester proposes to differentiate between the di fferent shortening

processes by appealing to the theory of prosodic circumscription (McCanhy and Prince

199ûa).

In the mode1 of Prosodic Morphdogy (McCarthy and Prince 1986, 1988, 1990ab),

tmncaîion is described as invdvinp the mapping of the base melody onto a prosodically

defined template. Mester (1990) argues that truncation templates can operate in two

independently established ways: (a) as a prosodic target to whch the base melody is

rnapped; (b) as a pTOGOdic delimiter fhat effectively reduœs base f o m s to a certain prosodic

size. Mester proposes that the choice between these two strategies is what underlies the

different prosodie shapes of the Hypocoristics and of the Rustic Girls' names. Mester

analyzes the Hypocoristic morpheme as a bimoraic target template onto which the base f o m

is mapped. As illustrated in (@, the mapping to the bimoraic l w t template bp the base

rnelody is not constrained by the prosodic structure of the base.

The Rustic Girls' morpheme, on the othcr hand, requires a bimoraic foot ternplate which is

defined as a prosodic delimiter that isdates the first foot of the base. The honorific prefix o-

is then afrised to the p i tively circumscribed base, with the residue left unparsed.

Finally, Mester argues that the derivation o f Geisha House Client names involves two

di ffcrent templates. First, a delimiting template circumscri bes the initial sy llable of the base.

This truncated base is then mapped onto a birnoraic target template.

8a. O

1 Uno ---> u no ---> u O-Uu-saN

I \

O [ P CL] O-HoN-saN I I \ / \ I

HoNda ---> h o N d a ---> h o N I / \

The problem with delirniting templates is that, as pointed out by McCarthy (1993, prosodic

circumscnption views as purely accidental the fact that the Rustic Girls' name formation

both circumscnbes a faot and maps to a foot template. That is, the prosodic shape of some

Rustic Girls' names is that of a specific bimoraic foot (either monosyllabic or disyllabic),

because p r d i c circumscription restricts the base melody to the initial bimoraic f a t in the

input However, the fact that both the delimiting template and the mapping template involve

a foot (or in the case of Geisha House Client formation, a syllable) is regarded as purely

coi nci den tal.

Furthemore, as Mester himself notes, the possibility of targeting a monomoraic

syllable in the Geisha House Client name formation seems proMematic with regard to the

proposal advanced in McCarthy and Rince ( 1-b) that the delimiting use of' prosodic

templates should be restricted to the minimai word. As supporting evidence for the proposal

that delimiting templaks may target a single syllable, Mester briefly discusses a different

tnincation morpheme, one which obligatorily attaches to the Hypocoristic suffis -ko. The

main charactcristic of this truncation morpheme is that it is restncted to a mmosyllabic

shape. 1 n contrast with the Geisha House Client narnes, this syllabie can be either bimoraic

or monomoraic. Mester gives as an example Hirorni which can surface as Hi.ko, Ro.ko,

Mi. ko, Hik. ko, Hii-ko, RoN. &o. etc. but not as *Hiro.ko, *Roini. ko or *Hi.ro.nri. koS4

9. H'pocoristic Formation (based on Mester 1990)

Hirorni Hi-ko, Ro-ko, Hii-ko, Hik-ko *Hiro-ko, *Romi-ko

Tanaka Ta-ko, Na-ko, Taa-ko, TaN-ko *Tana- ko, * Naka-ko

HoNda Ho-ko, Hoo-ko, HoN-ko

Sait0 Sa-ko, Saa-ko, Sai- ko

Hatton Ha-ko, Haa- ko, Hak- ko *Hato-ko

Y a u & Y a-ko, Su-ko, Yaa-ko, Y ak-ko *Yasu-ko, 'Suda-ko

The new set of data, however, only adds to the problem of how to uniformally account for

the tmcation phenornena in Japanese in that the tnincated forms in (9) are shown to

optionall y surface as a monomoraic syllabie (the truncation morpheme cannot be viewed as

minimally consisting of a well-formed foot).

The suffis is distinct f- the usual ending -&O fa women's aames found in many of the e.xampies in (2) and (3) above. These Hypoanistic forms are introduced in Tateishi (to appear).

23

1 tô ( 1990) deals u i th a di fferent type of problem posed by the Jûpanese data. She

discusses almost exclusively the differences and similarities between Hypocoristin ( 1) and

word clippings (4). Her analysis of the two p r e s s e s relies on the observation that

truncated names are never found in ida t ion (Le. they are always follorved by the affin

-&IN), while word clippings involve free-standing words. 1 tô proposes a morphdo@cally

based prosodic distinction between (bound) stem and (free) word in Japanese. She argues

that the two morpho-prosodic categories Smc and W m must satisfy separate minimalie

requirements.

10. Minimal Stem Requirement Min(STEM) = F = (p p)

Minimal Word Requirement: Min(W0RD) > a

In other words, the minimal prosodic stem is a bimoraic foot, while the minimal p r d i c

urord is at Ieast disyllabic. Both the prosodic stem and the prosodie word are defined as

morphological constituents with a specific prosodic shape.

Itô assumes that, in contrast with underived words which corne with their own

underl ying templates, Hypocoristics and word clipping are derived from their base f m s ,

and therefore must satisfy the minimahg requirements of the l a n p g e . The idea is that

since stems are part of words and every word must contain a stem, then, the disyllabic

prosodic words must also properiy contain the bimoraic prosodic stem. And since Japanese

is a suffising language, Itô argues that the unmarked position of the bimoraic stem is ai the

lcft edge of the prosodic word. Under this view, the wntrast between the hypocoristic soi

(-cm 'li ttle Sai ko' and the word clippi ng saik from saikederikku 'psyc hedel ic' fol lows

dirccdy from the minimal size requirements in ( 10).

11 a. Word b. Word c. Word / \ / \ I

Stem Suffix Stem Suffix Stem / \ 1 \ / \

(sai) &N (sail ke s u to

In both (1 la) and (1 1 b), the disyllabic word contains a bimoraic stem followed by a

syllable, which is defined by Itô as a quasi-suffis in the iwrd clipping form. This suffixed

syliablc can be either monomoraic (1 1 b) or bimoraic (as in (baa).(teN) from baateNctaa

'bartender'). In the word clipping suto from sutoraiki 'strike' ( 1 lc), on the other hand, the

stem and the word completely overlap.

As shown by the latter form, the quasi-suffix is not an obligatory part of the

prosodic word. Itô notes that the minimalit?; requirements cannot prevent the indusion of

additional materiai to the tmncated forms. This is exemplified by the free variation found in

fonns iike herikopuuia 'heliwpter' (disyllabic [[hc.riJsJlvd us. trisyllabic [[he.rilst. k ~ ] ~ ~ ~ ) .

As a result, there is no esplanation as to nhy word clippings should be restricted to two

metrical constituents, i.e. two syllables, two feet, or a foot followed by a syllable. Itô's

proposal is also problematic from an empirical potnt of view. since it cannot esplain either

the disyllabic ban of the Geisha House narnes or the optimal monomoraic shape of the

Hjpocoristic forms atîached to the suffis -ka

A different proposal which anticipates many aspects of Optimalit). Theov is offered

by 1 tô and Mester ( 1%). In their anal ysis of Japanese word clippings, I t ô and Mester

accouni for the loose' minimal word template in loanword abbreviations by proposing that

the rcquirements goveming prosodic constituents should be viewed as resulting from the

intcnction bchveen distinct but related conditions on p r d i c licensinp.

As a minimal well-formedness condition governine the Prosodic Hierarchy, Itô and

Mestcr propose Proper Headedness.

1 2 . Proper Headedness Every (non-terminal) prosodic category of level i must have a head, that is, it must immediately dominate a category of levcl i - 1.

To account for unîooted syllables within a PrWd, they propose that the strong version of

maximal parsing be replaced by a weaker version (the Weak Layering Hypothesis).

13. Maxi& Parsing Prosodic structure is ma~irnally parsed, within the lirnits i m p d by other (universal and language-particular) constraints on p r d i c form.

1 tô and Mester then propose that the canonical prosodie shapc of urord clippings is achieved

through thc imposition of a wcll-formedness wnstraint restncted to words that are reiated to

more basic forms by means of prosodie-morphological processes.

14. Word Binarity P-dcrived words rnust be prosodidl y b i n q .

To account for the minimal prosodic shape of tvord clippings, Itô and Mester argue that the

rnoraic content of syllables is outside the domain of word-level conditions. The?; speculate

that the opacity of syllable intemal structure with respect to the word constraint may follow

from a general locality condition limiting the accessibiliq of intemd p s a i ~ c structure.

That is, any constraint that speçifically defines the prosodic shape of the PrWd domain

should only sec syllables and feet.5

1 5. Hierarchicai k a l i & A condition operating at prosodjc level Ci has access on1 y to structural information at Ci and at the subjaccnt level Ci-1.

Under this view, the moraic content of syllables is invisible to well-forrnedness constraints

imposed on PrWds, since only the foot-internal structure would be accessible at the word-

la-el. Requirements regarding foot well-formedness, on the other hand, urould have access

to syllable-intemal structure.

CrucialIy, Itô and Mcster's anaîysis pro\-ides an approach in which the minimal

prosodic shapc of wwd clippings can be defineci solely in terms of constraints on the shapc

of specific prosdic categories. That is, the prosodic shape d word clipping is defincd by

Foot Binarity and Word Binarity, while the left edge requiremcnt is &fined as a constraint

that favors the coincidence of the left edge of a PrWd with that of a foot.

16. Lefi Edge Matching Left word edges preferentiall y coincide with fmt edges.

5 1 tô and Mester ( 1992) assume ihat a mora is the terminal elcment in the Prosodic Hierarchy. Following McCarthy and Rince (1986,1990ab. 1993ab). I assume that syllables. uot m m . are tbe terminal elements in the Prosodic Hierarçby, with moras bang M I as attributes of syilables and segments. See also Hung ( 1994). for artouments tbat m m are outside of the domain of the Rosodic Hierarchy wiihin the framework of Optimality Thcoq.

S till problematic for their analysis, however, is the fact that Word Binruity, a rninimality

requirement, must also be vicwed as a maumality requirement. Moreover, Word Binarity

cannot be viewed as a condition imposed on al1 derived words in Japanese, since it is

clearly not satisfied in the Rustic Girls' name formation (cf. O-Yoo and O-RaN in (2)).6

2.4 Japanese Truneation and Correspondence Theory

Optimality Theon seems p;iruculariy well-suited to deal with the non-unifbrmity of the

trunwtion phenomena in Japanese. Central to the framework is the notion that hi@-ranking

stnicturai constraints may be minimaiiy violwd in some contents in order to satisfy higher-

ranked faithfulness constraints. The prediction then is that satisfaction of the h i g h - W n g

stmctural constraints should occur in contests where faithfulness between the lexical input

and the surfacing output does not play a rde.

Correspondence theory ( McCarthy and Prince 1 995) pos tulates that phono1 o g i d 1 y-

cmpty morphologid constituents corne with their own set of faithfulness constraints which

rcgulate identity relations between a morphologicaily or prosodically-specified base and an

output. Gi\-cn the r a g e of possible rankings thai can be posited between the faithfulness

constraints thcmselves and between these constraints and the structural constraints of the

language, non-uniformity is predicted to be the norm rather than the exception.

2.4.1 The Emergence of the Unmadced

To my knotvledge, there has been no previous attempt to prot-ide a unified account of the

two types of Japanese truncation morphemes, bound and free-sading, wi thin Optimaiiiy

Thcoq. 1 begin by briefly discussing two analyses that offer an optimality-theoretic account

for some of the truncation morphemes presented in 82.1 and 42.3.

In their study of over- and underapplication in reduplication, McCarthy and Prince

( 1995) develop a theory of correspondence where reduplicaiion involves a relation between

- - - -

Note that to assume the homnific @x O- to be part of ihe tnmcaied form would lx problematic, s i m the prefix does not obey the Left E â s Requirement. Here 1 assume that tbe prefis O- is an affixai clitic parsed by a nxwsive W d (cf. Selkirk 1995)

the base and the reduplicant (BR Identity), a relation between the lesical input and the base

(IO Faith), and in some cases a relation between the input and the reduplicant (IR Faith).

17. Input: /AffixM -Stem/

I R - F . m Jf' If IO-F.m Output Red s Base

Estendinp McCarthy and Prince's mode1 to tmncation morphology, Benua (1995) proposes

that tmncation involves two correspondence relations, an input-to-base relation (IO-Faith)

and an output- to-output relation behveen two distinct (P-)Stems (BT-Identity). In wntns t

ni th reduplication rnorphology where base and reduplicant affk are part of the samc input

and simu1tancously generated, in inincation morphology, the base and the truncated forrn

surface as two separate words. The correspondence relations between the lesical input and

the base, and between the base and the tnincated forrn are encded in the schematic model

below (Benua 1995; McCarthy and Prince 1995).

18. Input: Root

IO-F.m tJ Output Base StemTninc

BT-I mm

The input is mapped to the base by IO correspondence, and BT correspondence relates the

base to the truncated form. Benua also assumes that in languages with multiple tnincation

morphcmes, each truncation morpheme cornes with its own set of BT-Identity constraints

(cf. McCarthy and Prince 1995; Urbanczyk 1995).

Following McCarthy and Prince (lW), Benua argues that tmncation to a bimot-aic

foot in Japanese is a case of the emergence of the unmarked PrWd A PrWd is unmarked

when i t uniquely dominates a birnoraic foot that is perfèctly alignai at the edge of the

PrWd. Such an unmarked prosodic structure results from the dominaiion of Mixm by

what Benua calls the 'PrWd restrictor' constraints, Ti'-Bw, Au-FT-Lm and P.WE-a.

19. FT- BIN (McCarthy and Prince 1993a) Feet are b i n a s at some level of analysis (p, a)

ALL- FT-LEFT Align (Ft, Left, PrWd, Left)

"The left edgc of eveF foot coincides wi th the left edge of some RWd."

PARSE-0 E\.cry syllable is parsed by a foot.

This is il1 ustrated in the tableau below with the Hypcoristic fonn ~ i d o - & N frorn MidoB.

Note that the input structure of the Hqpocoristic rom is incornpiete in that it abtracts the

truncation morpheme away from its obligatory suffis - C ~ N .

Fr-BIS is assumed here to rcquire strict binarity so that the candidates in (2ûb-c) violate the

consuaint in k i n g either too large (20b) or too srnall (2043. The candidate in (20c) also

violates loiver-ranked AL-FT-Lm. As for the last candidate, i t fatal1 y violates PARSE-a

Intcrcstingiy, Benua dœs not discuss the optional bimoraic form Mi-&, so there is no

account as to why the latter forni is regarded as optimal as Mide&, even though Mii-&hl

crcatcs two additional violations of MA..^. (It does rcsult, howcver, in onc less violation

of AL-FT-LR. )

Benua ( 1995) also provides an analysis of the Rustic Girls' name formation which

reformulates Mester's ( 1940) prosodic circurnscriptional anaiysis in optimality-theoretid

Candidates CiPCi

Base: mi.do.ri u CI

a. d (mi-do)

Ci C i C i

b. (mi .do. ri)

Pli Ci

C. (mi.do).(n)

U P Ci

d. (mi.do).ri

FT-B 1s

*!

*!

*!

P . ~ s E - o AL- FT-

k ~ r

X *

M.-I.BT

I

*S

tenn S. Adopti ng a proposal by McCarth y ( 1993, s he proposes that the perfect rai Mu1 ness

of the tnincated form to the initial bimoraic foot of the base follows h m two constraints

that requi re the correspondence of peripheral segments wi th some prosodic constituent.

2 1. ASCHOR-R(R) "Every correspondent of a foot-final segment is foot-final. "

22. ASCHOR-L(FT) "Evexy correspondent of a foot-initial segment is foot-initial."

Such 'circumscri ption- typei mnstraints, however, fail short of proïiding an explanation for

the suictiy monosyllabic sbape of Geisha House names, e.g, O-Hoo-SUN from (HoN)da

and O-Tm-SUN from (Tana)ka. This type of prosoctic identity constraint becornes even more

problematic for the op t iod ly bimoraic and monomoraic f m s found with the Hypoconstic

suffis 40, e.g. Mi-ko, RoN-ko and Hik-ko from (Hiro)rni.

A différent challenge presented by the Japanese &ta with respect to 'the emergenœ

of the unmarked' notion of word minimality is how to define the various prosodic shapes

of the ivord cli ppings. In the case of 'loose' minimal words. McCarthy and Prince ( 1994)

assume that for some morpheme, AIL-FT-Lm is satisfied. while P.RSE-u is minimally

violated. This account does not so readily estend to loanword abbreviations in Japanese,

however, s i n e a perfectly well-forrned word clipping consisting of two feet satisfies

Another solution, adopted by Suzuki (1995). is to assume that a PrWd larger than a

foot follows from satisfying N ~ m s u r r y . ~

31. NONFINALHW No PrWd is ri@-aligned with its head foot.

The idea behind this proposal is that any word that begins with a heavy syllable ai11 have to

include a following syllable to satisfy the nonîïnality consuaint (cf. disyllabic niai.ku from

' These word clippines that consisi of two fcet are usuall y dismissed as nrpraminirnai. thus outside of the domain of any atialysis of word minimaiity in Japanese.

Sumici's (1995) optimadity-theoreticai analysis of the Japanese data. bowever. is based on the assumption that XfinWd is a prosoctic requirinwat iinposed on RWâs. oaitra .McCarrhy and Rince (lm).

nurikurohoN 'microphone'). Under this 1-iew, tri syllabic forms li ke te-re. bi from terebizOn

'television' reflect the need to fu11y satisfy N~F~sALHFT.

Although Suzuki's proposal provides a straightfonvard esplanation for the lac); of

monosyllabic word clippinp. it fails to account for the very productive wwd clipping

pattern that consists of a single disyllatic foot. The espectation is that in a form with three

initial Iight syllabia. the third syllable of the base should be parc of the output in order to

satisfy N O ~ ~ A L H F T . Furthemore, the free variation noted by Itô (1990) for forrns like

hetikoputua 'hel icopter' (disyllabic heeri vs. trisyllabic he.ri. ko). strongl y argues against

such an analysis. Suzuki's andysis dœs no< offer any expianation for qurdrisyllabic forms

l i k (a.su).(pa.ra) from usqmragusu 'asparagus' which are dismissed as supraminimal and

outside of any anal ysis of word minimality in Japanese.

2.4.2 The Emergence of the PrStem

1 propose that the various prosodic shapes of the truncation morphemes in Japanese can be

gi\.en a unified account by assuming a Pr&c Hierarchy of metrical constituents in which

the R W d immediately dominates a PrStem. 1 will demonstrate how this mode1 of prosodic

stmcture accounts for the tmncation morphemes in Japanese in terms of 'the emergence of

the unrnarked' by showing that the truncated forms fully satisfy hi@-ranking structural

constraints that are minimally violated in the general phonoloey of the language as a result

of hig he r-rankcd F m r o cons trai nts.

As il l ustrated by the different candidate stnictures in (24), ai1 of the prosodic shapcs

of the truncated forms can be organized as a PrWd that uniquel y dominates a PrStem.

34 a. PTWd b. PrWd c. PrWd d. PrWd I I I I

PrStem PrStem PrStem PrStem I I

F\

h Ft Ft Ft Ft I I \ I I

Ocici 9 Oc, "cici % %c, %lr

The first question to be addressed here is what makes the bimoraic PrWd in (Ma) pcrfectl y

n-c I l - formed as a suffised shortened name, yet unacceptable as a free-stand ng abbreviated

Imword. The prosodic shapes which charactenze word clipping formation (24bd) have in

common a minimally disyllabic constituent at some level of prosdic orpizat ion. T o

account for this rninimality requirement, 1 propose a binarity constraint imposed on the

PrStem, that is defined in terms of headship.

25. BRANCHHEAD The head of a PrStern is in a rhythmic relation with a non-head constituent at some Icwl of analysis. (a, Ft).

In a PrStem that uniquely dominates its head foot as in (24b), the head syllable must be in a

rhythmic relation with a non-head syllable within the foot. In a PrStem where the head foot

dominates a bimoraic syllable, BRLYCHHE~U) can be satisfied in one of two ways: by adding

a syllable (24c) or a foot (246) in the non-head position of the PrStem.

By contrast, none of the PrWd structures in (26) are well-forrned tmcated stems.

The nest question then is in what makes these candidates unacœpcable word clippings.

26 a. PrWd b. PrWd c . PrWd

The candidate in (36b) is easil y dismissed since i t clearl y adds an unnecessary violation of

U - F f - L m . The candidate in (26c), on the other hand, d œ s not add to the number of

AL-Fr-Lm vioIaiions that can be round in the well-formed u'ord clipping in (24d). I

propose that this candidate is el iminated as a resul t of creating a fatal violation of the high-

ranking structural constraint ALJGN-P~Srn-Lrn. In contrast with AU-FT-Lm, this latter

27. ALIGN- PRS TEM- LE^ Align (PrStem, Left, PrWd, k f t )

"The left edge of e v e v PrStem coincides with the left edge of some PrWd."

Consider now the candidate in (26a). This prosodic structure is characterized by a final

syllable directly parsed by the PrWd, with the leftmost urirooted syllable k i n g parsed by

the PrS tem. Each parsing resul ts in a violation of Eul.\csî~m (see 0 1.1 ). T o account for

the unacceptability of (26a) as a well-fonned word clipping, 1 propose that PARSE-ET should

be defined in terms of headnon-head relation. Based on this definition of syllable parsing,

only the rightmost syllable wodd vidate the constraint.

23. PARSE-a A syllable is parsed in the head or non-head position of some m e t r i d constituent.

Here 1 assume that some localiiy condition, similar to Itô and Mester's (1992) Hierarcbid

k a l i ty, 1 imits the access to stnictural inrormation, so that a syllable direcrl y parsed by the

PrWd cannot be defined as k i n g in a non-head relation with the head PrStem.

29. Hierarchical LocaIity A condition operating ai p r o d c level Ci has access o d y to structural information at Ci and at the subjacent level Ci,l.

To sum up, a wfell-fomed word-clipping M 1 y satisfies BRL\.KCHHG\D, A ~ ~ r n - P R S r n f - L r n

and PARSE-a, while rninimizi ng \.iolations of A u - FT-Lm. Two- fmted structures will be

discussed in §1.5.3. This proposal is illustrated in the tableau belonr. (The parentheses

represent foot boundaries, while the square brackets indicate RStem boundaries.)

P.USE-O ALIGX-

PRST-LEFI-

Candidates IiP P Plr PP

Base: kon. bi.nee.ioN I

w P a. I/ [(koN).bi]

w v Pl' b. [(koN) . bi] .(me)

w P P c. [(koN).(bi.ne)]

Pl'

d. I(koN)I

Ar--Fr-

rn

* !*

*!

-

BRLXCH-

HEU)

*!

In (30a), the optimal output satisfies the four hi&-mking stmctural constraints at the çost

of multiple Mi.m violations. Smaller and larger word clippings, on the other hand, vidate

high-ranking BR~~CHHUD in (306) and Am-Fr-Ln=r in (30b-c), respective1 y.

The choice of the prmodic shape of a particular word clipping is usually restrïcted

by the segrnenial structure of the base (BT-Identity). Here 1 assume that of the four hi@-

ranhng stuctural constraints in (30), AIL-FT-LEFI- is the only on that is cnicially dominated

by BT-Identity constraints, wïth the exception of M Q X ~ . Some forms, however, have a

base that allours for more possibilities, as illustrated in the tableau below.

3 1. BRXX;CHHE~ >> AUGS-PRST-LEFT >> P-ARSE-O >> AU-FT-LEFI. » M.-KBT

In (3 la-b), the two outputs satisfy the four hi@-ranking structural constraints at the cost of

multiple M - I s ~ ~ violations. A PrStem that consists of two feet, howvcr, violates lower-

ranked Au-FT-Lm, as shown in (31c). In (3 Id), the bimoraic foot violates high-ranking

B ~kh-a~i-km.

The question to be addressed at this point is how to proleide for the optional choices

available to some of the word clipping forms, as shown in (3 la-b). An important aspect of

Optimalis. Theory is that whenever high-ranking constraints are satisfied, the choice of the

optimal output is left to lower-ranked consfraints. The prediction then is that there should be

on1 y one optimal output for every wwd clipping.

Candidates u P P P P ! '

Base: he.ri.ko.pu.taa C u

a. d [(he-ri)]

C u b. d [(he.n).ko]

C u u It c. [(he.n).(ko. pu)]

w d- I(hee)l

-

BR.LVCH-

, H w

*! -

AL-FT-

Lm- J

* !+ ,

AUGS-

PRST-L~T

P ~ E - a

Building on an idea of f iger ( 19%b), 1 propose that variability in the choice of an

output cornes from the variaMe ran)iings of two codict ing constraints within the crnstxa.int

hierarchy. The choice beween disyllabic [(he-ri) ] and trisyllabic ((fie. ri ).ko] 1 propose,

follows directly from the variable rankngs between the tu'o constraints No?.mlu-HFT and

33. NONFINALHF~ No PrWd is ri@-aliped with its head foot.

33. ALIGNHFT A PrWd is right-aligned with its head foot.

The constraint AUGSHFT requires that the head foot of a PrWd be final, while NOWISAL-

HFT rcquires the opposite, that the head f m t never be final in a RWd. No output will ever

satisfy both constraints simultaneously. Whch comtra.int urill be satisfied in the output then

follows from the choice of a specific ranking behveen the constraints. This is illlustrated by

the double tableau below.

Candidates 1 1 BR.LVCH- C L P U C L P P

Base: he.ri. k0.pu.m H L ~ P P

a. d [(he-ri)] P I L IL

b. [(he-ri). ko] t ' c c I ' P

c. [(he.ri)-(k0.p~) 1

Candidates 11 P P P P I r C i

Base: he.ri .ko.pu. taa

P P a. [(he-ri)]

l ' c c Cc 6. d [(he.ri).ko]

P P P I ' c. [(he.n).(ko.pu)] Ir*

As shown in the tableau above, AUGXHFT is assumed to be gradiently violable. Note that

cvcn if we were to assume that AUGNHFT is categorically violated, lower-ranked &-Fr-

LH=T wouid still choose [(he.ri).ko] as the output of the second tableau.

1 tô ( 1990) notes that once an abbreviated fonn is established, other possiMe forms

are highly disfavored. The assumption then is that each newly created word clipping is

stored with some specific ranking behveen Nmm..v.HFr and AuG~HFT as part of its

lesical entry. According to Itô, however, there are some loanwords that have more than one

abbreviated forms. The different prosodic shapes of thc word clippings then follow directly

from their underlying prasodic shape and from the choice of a speci fic ranking benveen two

lotver-ranked conflicting constraints.

2.4.3 Compounding

Word clippings that consist of two feet are usuall y abbreviated compounds. Each member

of a compound is an independent word which c m form a word ciipping on its own. As

noted by Itô ( I W O ) , a Ioanword such as koNpanii 'pars, company' can be shortened to

disyllabic koNpa. As part of the compound kuraîu-koNpa 'class par^ ', the sarne form c m

further shorten to monosyllabic &ON as in kura-koN. However, no member of a shortened

compound may be parsed as a monomoraic syllable: *Lu-koN, * kura-ko, cven though the

latter tntncated form is an acceptable word clipping.

Here 1 propose that both the requirernent demanding that a word clipping be foot

initiai and the requirernent that cach member of a compound be minimal1 y parsed as a foot

are the surface effects of a single constraint on the alignmcnt of edges behveen a sub-lesical

morpho- prmodic constituent and a suù-lesical metrical constituent (Downing 1998).

35. ALIGN- P-STEM- LEFT Align (P-Stem, Left, Ft, Left)

"The left e d g of every P-Stem coincides with the left e d p of some fmt."

This is i llustrated in the tableau below. (The straight lines indicak P-Stern boundaries.)

Both forrns in (36c) and (366) are eliminated through a fatal violation of AI~GS-P--1-

m. The choice of the optimal output is thus left to lower-ranked Au-FT-LEFT. Double-

fmted word clippings that cannot be defined as compounds will be discussed in $2.5.3.

Candidates C i Ci rr r

Base: ku-rasu - koN.pa r r CIP

a. v' [ I (Lu-ra). I(koN)]

F r r P r

b. [ 1 (ku.ra).su]. 1 (koN)

Ci w c. [ 1 ku. 1 (koN)]

Ir

d. [ 1 (ku. 1 ko)]

2.5 Bound Truncated Stems

The question to be addressed now is why the prosodic shapes of shortened names should

be restricted to a foot, either bimoraic or disyllabic. 1 propose, following Itô (1990), that

the difference between shortencd narnes and word clippings lies in large part in a distinction

between bound stems and free words. The crucial assumption here is that with respect to

bound rnorphemes, i.c. truncated stems that are obligatorily followed by an affis, structural

constraints are imposed on both the truncated stem and the affix simultaneously. In other

~\.ords, the solution as to why shortened names are restricted to a bimoraic foot shape lies in

part in the optimality-theoretic notion b t the prosodic shapes of phonologically-empty

morphemes result [rom the satisfaction of high-ranking structural mnstraints, mther than

from lcsicall y-specified prosoâic requiremen ts im posed on those morphemes.

The correspondence relations involved in generating the suffixecl name shorteninp

are encoded in the schematic mode1 belou..

AUGS-P- BRXXCH-

SEM-LRT j H m

*!

*!

AUGX-

PRST-Lm

AU-Fr-

LEET

**

+**!

*

37. Input: Roo t A fris IO-F.m ct k t IO-F-m

Output Base a StemT,, -Affi?r

The input root is mapped to the base through R m - I O - F . m constraints, while the affis is

rnapped to the output through AFFJX-IO-F.U~H. In mm, the B T - I m m constraints relate

the base to the truncated forms. As shown in (33, the tnincated morpheme is lesically

rnarked as a morphological stem (M-Stem). In the next sections. it tvill be shown that

~xriability in the choice of an output in sme of the mmcated name formations foilows in

part from the satisfaction of a constraint on the alignment of P-Stem cdges.

2.5.1 Hypocorfstic Formation

Consider first the Hypo~oristic forms. The challenge here is how to restrict the prosodic

shape of these truncated forms to a bimoraic fwt, while allowing for both a di-llabic and a

monos~.llabic output, irrespecti~e of the prosodic structure of the base. As shown in (38)

beIow , both Midon' and Yooko have a disyilabic and monayllabic form.

38. ~ p o c h o ~ s t i c Fannation (Poser 1984, 1990) ,Mi dori M i d o - a , Mii-CaN Y m k o YokcwkN, Yoo-bN Hi rom i Hircka.N, R o m i - a , Hi-* Mariko Mari -W, Mako-ikN, R i k o - W , Ma-* JuNko JuN-&IN Hanako Hana-W, Haa-kN, Ha?-CaN

The analysis proposed here assumes that the prosodic shapes of tnmcated names are in part

the result of the domination of Afk-IO- FA^ over the hi@-ranking structural constraints

of the language. Fdlowing Itô (1990). 1 assume that the H~poconstic suffis and the

truncated stem are evaluated simultaneously by the constraint hierarchy. This is shown in

the tableau below.

in tableau (39) above, the choice of a monosyllabic foot for the truncated stem (39b) results

from satismng I O - F m and the high-ranking structural consüaints, while at the same time

minirnizing violations of AwFT-Lm. In (39d), the truncated stem is left-aligned with an

unfmtcd syllable \iolaiing Ams-P-Smr-LEFI-. In (3%) and (39e), the PrStem dominates a

disyllabic foot, creating a fatal violation of AU-FT-Lm. T o assume that the bimomic suffin

is directiy parscd by the PrStem [(mi.&).bNl to satisfy AU-FT-Lm fully, on the other

hand, \vould result in a fatal violation of WEIGHT-IDSTRESS-

Variabiliîy in the prosodic shape of :he Hlpocoristic forms, hoivever, suggests that

somcthing more is at work. As shown in the tableau above, with regard to AU-FT-Lm,

(3%) is l e s optimal than (3%). The prediction then is that [(mi.do)].(€aN) should not be a

possible output. Part of the answer as to why bath foot shapes are possible Hypocoristic

f o m s lies in the lexical marking of the H~pocoristic morpheme as a morphologid stem.

Following McCarthy and Rince (l993ab), 1 assume that the minimally bimoraic shape of a

phonological l yempty morpheme marked as a morphological stem fol lows from satisfying a

Candidates P P P Plr

Base: mi.do.ri -&IN

(r Cr Ct Cr

a. [(mi.do)].(hN)

w Crlr

b. d [(mii).(ëaN)]

( r u !'

C. [(mi.do).n]

!' Crlr

d. [mi.(taN)]

C P Cr UCr

e. [(mi.do).ri].(bN)

AFFIS-

F.ml"

* !**

-

ALL-Fî-

LEFI- 1

**! 1

* A

X

1

**!*

ALJGS-P-

Smf-LEIT

*!

BR-LXH-

H m

cons traint requiring the edges of a corresponding morph~prosodic constituent to be d i @

n-ith those o f a PrWd.

40. P-STEM=PR WD The left and ngh t edge of a P-Stem coincide wi th the left and ri@ t edge of a PrWd.

Satisfaction of this a l i g n e n t constraint when imposed on a bound stem, however, results

in a rccursive PrWd structure violating NO~-RECLR (Selkirk 1995).

41. NONREC No PrWd dominatcs a PrWd.

1 assume here that BR.LVCHHE\D dominates P-STEM=PRWD and N ~ R E c , with AIL-FT-Lm

king Iowcr-tanked. The variable ranhng of the hvo potentiaily codicting constrain& P-

STEM=PRWD and NON- in turn results in an optional choice between a bimoraic and a

disyllabic fmt. (The square bnckets in bold indicate PrWd boundanes.)

Since AFFK-F.UTHIO is always satisfied in the output, it is left out of this and subsequent

tableaux. Note that the variable rankings of STEM=PRWD and NmREC do not play a d e in

the formation of word clippings, due to the latter's status as free-standing word. Similady,

the variable rankings of NOS'FIS.'U.HFT and ALIGSHFT do not affect the choice of the output

in the obligatorily affised îruncaGed names: i.e. N ~ ' F c I ; . ~ H F T is alaays satisfied, while

2.5.2 Prosodic Subcategorization

Consider now the Geisha House Client narnes. in contnst with Hypocoristic forms, this

tnincation morpheme presents very little variation with respect to a choice of outputs. What

is strihng about this particular shortening formation is that the truncateâ stem only suifaces

as a monosyllabic fmt

43. Geisha House Client Nanies

Tanaka HoNda Saï to Hattori

The solution 1 propose in order to acmunt for the distinct prosodic shape chat di fferentiates

the Geisha House Client names from the Hpocoristics lies in the assumption that an affix

can be su bcategorized for a specific prosodic base (McCarthy and Prince 1993ab). In other

n,ords, the answer as to why the truncated stems of the Geisha-House Client names are

rcstricted to a monosyllabic shape lies in their status as a 'base' for the obligatory affïsation

of -saN. 1 propose a lesically-specific alignment constraint that expresses the p r d i c

su bcategorization requirement of the suffis-saN.

44. AFFIX-TO-Ha The left edge of the tnincation affis -saN is right-aliped with the head syllable of some PrWd.

This is illustrated in the tableau below.

Note that the domination of Ar=-mHa and BR-\S&EZD over N ~ R E c and P - S ~ I =

P~Wo(cf . tableau (42)) ensures that the only possible output is the one in (4%). The little

ixriability akailable to this truncation morpheme is further reslricted by the domination of

the BT-Identity constraint Sm-ROLE~ (iaa.saN) over M d T *(taN.saN).

Supporting evidence for the proposal that the Geisha House stem is prosodicall y

determioed by the subcategorizaiion constraint A m - m H o , mther than from the trucarion

morpheme being lexically-specifiied for sane pmsodic requimment can be found in the

prosodic shapes of the H~vpocoristic morpheme when associated with the sufis -ko. As

nord by Mester (l990), this pariicular Hgxxoristic suffis dso takes a syllable as its base.

1 n conwt tvith the Geisha House names, however, this syllable GUI be either bimoraic or

rnonomoraic.

46. Hypocons~ic Fannation (bûsed on Mester 1440)

Hiromi Hi-ko, Ro-ko, Hii-ko, Hik-ko *Hire-ko, *Romi-ko

Tanalia Ta-ko, Na-ko, Taa-ko, TaN-ko *Tana-ko, * Naka-ko

HoNda Ho-ko, Hoo-ko, HoN-ko

Sait0 Sa-ko, Sa -ko , Sai-ko Hattori Ha-ko, Haa-ko, Hak-ko *Hato-ko

Y asuda Y a-ko, Su-ko, Yaa-ko, Y&-ko *Y asu-ko, * Suda-ko

B y assuming that Hypocoristic-ko subcategorizes for the sarne A m - m t f a constraint as

the suffix -s& the monosyllabic shape of the Hypocoristic stem can be accounted for in a

straightfoward manner, using the same structurai wnstraint rankings that were proposed

for Hpocoristic -&Np as shown by the tableau below.

Candidates 1 P V V P

Base: hi.ro.mi -ko w CI

a. [ [(hii). ko] 1

What is striting about the possible prosodic shapes of these Hypocoristic forms is that they

follow directly from the variable d n g s of A ~ N H F T and NOSFKALHF~, irrespective of

the foilowing suffix. That is, the variable weight of the momsyllabic H~'p0coristic stem

when attached to the suffrx -ko, and the variable choice in the type of foot that characteri=

the Hypocoristic stem when attached to the suffïx -CL& are given the sarne e?<planatwq.

account. Di fferences between the two Hypocaistic f m s mme from the lexical 1 y-speci fic

subcategorization constraint of the suffix -ko (Am- HU) and f m the segmentai

content of the two hypocoristic suffixes (Amx-Fm[O). Similarly. the bimoraic shape of

the suftis -saN in addition to the subcategorization constra.int A m - m - H a ensures a

monosyllabic foot shape for the Geisha House truncated stem.

T o surnmarize, the different pmsodic shapes of both word clippings and suffixed

tnincated names have been s h o w to satis@ a number of hi@-ranking structural constnints

that are minirnally violated in the g e n e d phonology of the language: AUGS-P-Sm[-Lm,

AUGS-PRST-Lm, P.\RsE-a and BRXXHHE~. In addition, the more restricted prosodic

shapes of the suffised truncated names have been argued to follow in part from their status

as bound stems, while the variability found uithin truncated morphemes has been argued to

resuit fmm the variabie rankings of potentiaily conflicting lower-ran);ed constraints,

2.1.3 Lexically-Specific Ranking

In this study of truncation phenornena in Japanese, the constraint BRLYCHHEAD has k e n

psi ted as the force behind a num ber of restrictions imposed on the prosodic shape of both

word cl i ppings and bound truncated stems, with in some cases l o u r e r - d e d AU-Fr-Lm

restricting even further the prosodic shapes of the truncated stems. There are, however, a

nurn ber of truncated forrns that appear to be esceptional in that they are marked structures

[ri th regard to ei ther BRIXRHEAD or ALL-R-LEFI-. The one parûcularl y glaring esample of

a marked structure with regard to BR~~YCHHW is that of the Rustic Girls' namc formation.

48. Rustic Girls' Names

Yukiko O-Yuki %-Y uu, *O-Y uko Fumie o-Fumi *O-FUU, *o-Fue Y ooko O-Yoo RaNko O-RaN

A different problem cornes from non-compound f o m s like those in (49) that surface with a

dou ble-footed stnicture, resul ti ng in seemingl y unnecessar).. violations of AU-R-LEFI-.

49. a m . pa. ra asuparagasu 'asparagus'

koN. bi .ni koNbi nieNsu 'convenient store' fu.ra.su.to f uras u t o r e & o ~ 'frustration'

iN.to.ro i N t ~ r o d a l , ~ o N 'introduction'

koN. sa. ba koNsaabati bu 'conservative (in attire)'

Pater ( 19%) proposes that exceptionality with rcspect to the general phonoloe)i of the

Ianguage should be treated as the result of a le?ucdly-specific constraint r a n g that only

appIies to a subset of the lexicon.

Consider first the Rustic Girls' formation. In marked contrast with the Hypocoristic

fonns and Geisha House Client names, this tmncation morpheme presents a perfect copy of

the initial foot of the base.

5 1. Rustic Girls' Nantes Y uhko Fumie Ymko RaNko ietaiia KI ri s'i ma

(input is a last n a m ~ ) ~ (input is a iast name)

Bcnua (1995) provides an andysis of Rustic Giris' narnes that reformulates the prosodic

circumscriptiond analysis of Mester (1990) in ternis of prowdic identity between

corresponding outputs: ASCHOR-R(R) and Asc~m-L(Fr) (see 52-41 ).

The problem posed by an analysis of perfect copy effects in tenns of prosodic

identity is that i t only works for words that contain one f m t As shown by the taMeau

belo\v, a two-footed base allows for more variability in the choice of an output. Note that

the structural constraints in (52) are espressed according to the mode1 of prosodie structure

1 assume, Le. a syllable that is directly parsed by the PrStem does not result in a violation

-~ ~ - - -

'ïhe last two e.samples are from Suzuki (1995).

45

In order to get perfect correspondence between a longer base and a wncated form, Cmrm

must be added to the set of hi@-ranking BT-constraints of the Rustic Giris' formation.

53. CONTIGUCTYB* The portion of Si standing in correspondence forms a contiguous string.

(no skipping) The portion of S2 standing in correspondence f o m s a contiguous string.

(no intmsion)

Consider nou. the candidate output in (52b). This f o m violaies ASCHOR-R(Fr) since the

initial segment of the unfooted syllable corresponds to a segment ivhich is fa-init ial . In a

trisyilabic base, hotirever, an output stem like for esample [(Yo.);i).ko] would correspond

pcrfectly to the prosodic shape of the base without violating either of the prmodic identity

constnints. Recall that such trisyllabic forms are perfedy acceptable word clippings, so

thcy cannot be dismisscd out of hand as impossible truncated forms.

The question then is how on the maximail y bimoraic prosodic shape of the Rustic

Girls' names be differentiated frorn the minimally disyllabic shape of the word clippings. 1

begin by proposing a reformulation of Benua's analysis in ternis of the standard constraints

WT-1 D A T ~ , C a ~ m and ASCHOR-Lm, thus eliminating the prosodie identity constraints

fmrn the sets of BT-Identity constraints.

54. ANCHOR-LEFT Any element at the left edge of Si has a correspondent at the left edp of S2.

55. WT- IDENT (McCarthy 1995) If a and fl are in a correspondence relation,

and a is monomoraic, then f3 is monomotaic (no lengthening). and a is bimoraic, then f3 is bimoraic (no shortening).

Full satisfaction of these consmaints ensures that the initial foot of the base is mapped in the

output, in contrast wi th the Hypocoristic forms Rico-citN and Mako-caN from M m e h and

Yoko-caN from Yooko.

In (56b), the initial segment of the truncated fonn does not correspond to the initial segment

of the base, resulting in a violation of ASCHOR-Lm. In (56~) . the initial segments are the

same, but the output string that corresponds to the base string s k p the second syllable of

the base, rcsulting in a rata1 violation of Cmm. Finally. the lengthening of the initial

vourcl in (56d) creates a \violation of undominated WT-IDEXT~T

The problem to be addressed at this point is how to account for the exceptionality of

monosyllabic outputs with respect to BR~xCHHEU) and how to prevent tnsyllabic outputs.

Candidates P P P

Base: yo.ki.ko w

a. v' [ (o .k i ) ]

C !'

b. [(ki-ko)]

C Cr

C. [(yo.ko)] or P

d. [(yoo).ki]

1 propose that what looks like 'circumscriptional' effects follows from a lesically

\.ersion of AuûsHFi for the Rustic Girls' truncation morpheme.10

A similar proposal is first introduced in Prince and Smolensky (1993: 101) who assume constraiat FREE-V: 'A word f i vowel must not be parsed' is Limited to the nominative in Lardil.

47

WT- CO~TBT i ASCHOR-

I D E \ T ~ i j L J ~ 1

1

i 1 ! I *! 1 $

i *!* i

i

*! ; i i i

specifïc

chat ihe

BRV~CH-

H w

PVI;L.BO

*t

*.

**

IcX

This is illustrated in the tableau below.

The truncated stem k i n g lesically-specified for the h ie r - r anked version of AUGSHFT-y,

thc monosyllabic foot in (5&) is the optimal output, even though it violates BRASCHHEU).

When AucsHR-y d œ s not apply, as with the word clippings, the disyllabic form in (5%)

The question left 10 be ansarered with respect to truncation phenornena is how to

accoun t for the non-compound forms in (60) which surface with a double-footed structure

(1 tô 1990). What is striking about these forms is that each can generate a different output

that would not violate AL-Fr-Lm.

60. a m . para * a m , *a.su.pa amparagasu 'asparagusI

koN. bi-ni * koN. bi koNbinieNsu 'convenien t storc' fu.ra.su.to *fu.ra, *fu.ra.su furasutorceioN 'frustration'

iN. to.ro *iN.to i ~ t o r o d a k u S o ~ 'introductionf

koN.sa.ba *koN.sa koNsaabati bu 'consexvative (in attire)'

Here 1 propose that there is a subset of the truncated stems which are lesically-speçifid for

a \.ersion of AIJGS-P-SEM-RIGW which is ranked between N~FCIALHFT and AUGXHFT.

6 1. ALIGN-P-STEM-RIGHT-y Align (P-Stem, Right, Ft, Right)

"The right edge of eveq P-Stem coincides with the right edge of sorne footw

Note that this constraint is always satisfied in the abbreviated compounds as a result of fully

satisfying A LIGS-P-Sm !-Lm. A L I G S - P R S ~ + L ~ and P-ARSE-a, while minimizing ALL-

FT-Lm violations.

To surnmarize, the open system of truncation in Japanese offers the same possibilities as

does the p n e d phonology of the Ianguage: i.e. prosodie subcategorization of affises,

lesical specification of a subset of the l e i c o n for the hi&-ranking of some constraint, and

the satisfaction of high-ranljng structural constraints when not dominated by faithfulness

constraints.

Candidates 1 I r C i P C r P P

Base: a-su.pamgasu 1

P Cr P a. 4 [(a.su).(pa.ra)]

lr P P

b . [(a su). pal I

Cr Ii

c. [(a-su)]

N~'FI .v~-

HFT

*!

AUGS-P-

STEM-R-y

*!

ALIGSHFT

**

t

ALL-FT-

LEF~

** 1

* 1

L

Chapter 3

Arabic Stem Patterns

Anothcr challenge to standard assumptions about word minimality is found in the root-and-

pattern system of Arabic. In the Arahc nominal sqstem, the minimal prosodic shape is that

of a monosyllabic f w t (apr), followed by an 'extrameuid' consonant, while the maximal

prosodic shapc of nominal stems is disyllabic, irrespective of syllable weight. In fact, every

p s i ble sequcncc impol\*ing heavy and light syllables is actually used in the nominal

system. The prosodic shapes of nominal stems are presented in the table below, abstracting

away from prefixes and suffises (McCarthy and Prince I99ûa).

1. Basic N0fni~l Paiîerns

Also problematic is the fact that the minimal pfosodic shape of verbai stems is a disyllabic

Biliîerals

Tn'lierais

Quadn'liierals

foot ( o p p ) . Although Arabic is viewed as a quantity-sensitive language. the rnonosyllabic

bimonic foot is not a possible verbal stem. The m a i m a l prosodic shape of verbal stems is

al so di sy 1 labic, abstracti ne away from the obl igatory final non-moraic consonant.

2 Morus

barr

bahr

4 Moras

jaaruur

jab baar

ras maal

sabab

rn

3 Moras

jadiid

%taan

&r

kaatib

unjar

2. Busic Verbal Systenz finite fom)

1 n chapter 2, 1 proposed that disyllabic domains in languages where a hcar-y syllable is

usually pancd as a fmt (i-e. Weight-to-Stress) can be accounted for by assuming that the

PrStem is the head of the PrWd. 1 argued that the ban on bimoraic syllaMes as the minimal

prosodic shape of PrWds in the Japanese tmncation s'stem folloM.ed from the high-ranking

of BRLLXHHE~ within the constraint hierarchy of the ianguage.

In this study of the prosodic morphology of Arabic, 1 propose that a wiified account

of the nominal and verbal stem systems can be pmvided by assurning that the minimal word

is a PrWd that contains oniy one PrStern. I intend to show that in this particular m d e l of

prosodic structure, the different shapes of nominal stems and of relate- verbal stems within

a paradigrn on be acmunted for by the different sets of correspondence relations involved.

Font1 1

, h alal

faCd

1

3.1 Previous Analyses

In their anaiysis of verbai and nominal systems of Arabic, McCarthy and Prince's (19%)

strict adherence to the f m t i n r e n t q (i.e. moraic trochec, syllabic trochm and iamb) l a d s

IO the proposal that the canonical prosodic shape of verbal stems cannot be anaiyzed as an

underived template, i.e. as a template uniquely speciried as a syllable, a f m t or a PrWd.

Consequently, they argue that the disyllabic prosodic shape of the verbal stems should be

deconstnicted into two independent templates: a base and a suffix (cf. also 1 tô 1990).

McCarthy and Rince propose that in a disyllabic verbai stem, an initial monomoraic

syllable represents the p r d i c template d Form 1 (i.e. [faral] ) and of i ts closel y related

Fonn 2

hallai

fana i

Fom 3

haaid

faakl

Ql

dahra j

F o m 4

?ahlai

?afW

Fonn 7

nhalal

nfaCal

Form 8

htalal

ftaW

Fonn 9

hlalal

fTdd

dhar ja J

Fomt 10

stahlal

stafTai

Fonn 11

hlaalai

f W a l

forms (Forms 7, 8, and 9), while an initiai bimoraic s~*llable is the default ternplate, since it

appears uith d l the other, unrelated forms. The monosyllabic suffix is argued to have two

different prosodic shapes, a light syllable which defines the finite form and a heavy syllable

that characterizes d l the other, nonfinite forms. Finite forms would thus have either of the

tn70 prosodic shapes below.

3. a. Base + Affis 1 1

Form 1 Finite

b. Base + Affis 1 I

Default Finite

According to McCarthy and Pnnce's proposal, then, the minimal disyllabic shape of verbd

stems follows from their morphological structure. They further propose that the masimai

disyllabic shape rcsults from a condition stipulating that a ternplate contains at most two

syllablcs. 1

4. Mainml Stem Constraint Templates are maximally disyllabic.

The final 'estrametrical' consonant is derived from a d e requiring that al1 nominal and

verbal stems end with an incomplete syllable (Le. the final consonant is parsed as the onset

of a non-monic syllable).

5. Final Incompleteness

0 - - d l s t w i

In addition. they suggest that in forms such as [f.ta.Tal], the initial consonant should be

analyzed as k i n g parsed by an estnsyllabic mom, which they argue is syllabified as the

coda of an incomplete syllable. Fdlowing Moore ( 1990), McCarthy and Pnnce view this

estrasyllabic mora as a separate morpheme, i.e. a detransititlsing verbal prefix. Thus, the

1 SfcCarthy and Prince argue &ai such a descriptive constraint is not as a r b i m as it seems. They assume that tbere are gencrai coosicieratioas of locality tbat require that no d e counts to greater than two. S ine the des specifying the Arabic templates are subjed to l d i t y . the IMaximai Stem Constraint can be obtained from a general principle of' UG.

morphologïcal structure of a vertal stem in the finite fonn has the ma~imal prosodic shape

shown belour:

6 . a. Aff + Base + Aff b. Aff + Base + Aff I I I I f I

Detr. Fom 1 Finite Detr. Default Finite

Finally, the general conditions of prosodic well-fmedness in Arabic, such as Melody

Consenvation ( McCarthy and Prince 1986:46,1 OS), requi ring that al1 mot consonants be

Iinkcd to the verbal stem template. and the Ma..imal Stem Constm.int, requiring that thc

prosodic shape of a stem be maimally disyllabic, ensure that the surface realization of QI

for quadriliteral r o m is p r o d c a l l y parsed as an initiai birnonic syliable f01lou:ed by a

rnonomoraic syllable: [dah. ra. j J .

The difference between Forms 2 and 3 for biliterai and tnlitenl roots, hourever,

m u t be accounteà for by a morpholo@cally-trigpred grnination d e , since it should not

be possible to distinguish betu7een the tu.0 types of heavy syllables.

Thus, McCarthy and Aince are able to @\.e an account of the verbal stem system in Arabic

which respects the principles of Rosodic Morphology by subdividing the verbal stems into

different rnorpholo@cal constituents that have each their own template.

More problematic is the analysis pro~ided by McCarthy and Rince to account for

the canonical prosodic shapes of nouns. Their analysis assumes that the nominal stems can

draw on a srnail set of bûsic noun templates. Mapping to these templates must obey the

well- formedness conditions of Arabic: the Mafimal Stem Constraint, Final Incom pleteness,

Melody Conservation, etc. Each nominal template therefore d s to be specified for very

Ii ttlc information to denve the correct surface form. So for esample in (3a-b), the hvo types

of bimonic foot need only be specifed for either a rnonocsyllabic trochee FQT(U) or a

disyllabic trochee F@Yu). The different nominal templates are illusiraied in (8).

CJ

c. CvCvvC (jadiid)

Since the iambic foot is assurned to dway surface as its maimal expansion (upupp), there

is no necd to spca fy anything more than the type of foot (FI). A nominal template c m also

be specificd as containine more than one f w t (84. The quadriliteral roots in (&-O, on the

other hand, need only be specified as to whether the final syllable is rnonomoraic or

d i~yI lab ic .~

Basically, templates that are larger than a f m t are handled by two distinct strategies.

Wi thin the nominal system, templaies are defined either as a mhaic foot (qp) or ( U ~ C Y ~ ) .

an iambic foot (uvopp) or as a sequenœ of NO trochaic feet (qp)(up,). The prosodic

shape of stems. on the other hand, are defined through the rnorphological

concatenation of hvo different templates. In addition, the maumal prosodic shape of stems

is simply stipulated as k i n g disy llabic. Consequentl y although the individual templates

satisîy the Prosodic Morphology H>pthesis (i.e. uniquel y specified as a syllable. a foot or

a PrWd), the c a n o n i d prosodic shapes of b t h the nominal and verbal stems do no t

Furthemore, the surface realization of templates may be affected by lengthening rules (i.e.

m o n affisation) and gemination niles that can be triggered either by morphological or

lesical speci fication.

In the following sections, I present an analysis of the nominal and verbal stem

systems in Arabic wi thin the framework of Optimality Theor).. 1 wiU show that, ~vithin the

maiel of prosdic structure proposed here, the theor). of Correspondence / McCarth y 1995;

McCarthy and Prince 1995; Benua 1995) provides for a unified account of the mot-and-

pattern morphology of Anbic.

3.2 Root-and-Pattern Morphology in OT

Following McCarthy and Prince ( 1995) and Benua ( 1995), 1 assume that root-and-pattern

morphology, like tnuication, is governed by two distinct sets of correspondence relations: a

relation between an input (a consonantal mot and a vocalic melody) and an output (IO-

Fa-m), and a relation betwen a base and d l its related output f m s within a paradigm

(BO-IDEX-~-). 1 propose that IO-F.m constraints play a crucial mie in differentiating

betivcen the various prosodic shapes of the nominal stems, uthile distinct sets of BO-

ID^^^ constraints are involved in charackrizing a large part of the prosodic shapes of

verbal stems uithin a paradim.

3.2.1 The Prosodic Structure of Arabic

Considcr first the problem of hour to characterize ai! of the possible prosodic shapes of

nominals in Arabic. As noted before, there is only one type of nominal stem that does not

begin a e i t h the head syllable of a trochaic foot ( U ~ U ~ ~ ) . In this rnodel, the right-headed

prosodic constituent is assumed to be organized as a nght-headed M m , i.e. a light

syllable followû by a bimoraic f a : [ap(app)],. This proposal is consistent with the

standard assumption that Arabic is a trochaic language. The canonical prosodic shapes of

the nominal stems are giwn again below.3

There are a few monmoraic and non-moraic nouns (i.e. consisting of onl y one consonauta1 mot) but they arc quite rare. McCarthy a d Rince regard them as exceptionai in d y every resped.

9. Basic Norw%al Patterns

1 2 Morm 1 3 Moras

Each of the prosodic shapes of nominal stems can be organized as a

4 Moras

jabbaat

PrStem. This is illustrated with the f i b V e candidates below. Note that the verbal stems are

restricted to the prosodic structures in (lob) and (1Od-e).

10 a. PrWd b. PrWd c. PrWd d. PrWd e. PrWd I l I I I

PrS tern PrStem PrS tem PrStem PrSîem 1 1 ,h F\

h Ft Ft Ft Ft I I \ I I

% OP OP "P a~~ "w OP %P %c,

Every candidate satisfies AUGX-PRSTEM-Lm and PARSE-O fully. The canstraint Au-FT-

L m , on the other hand, is minimal1 y \-iolated by the candidates in ( 10c) and ( 10c).

1 1 . ALIGN-PRSTEM-LEFT Align (PrStem, Left, PrWd, Laft)

"The left e d p of e v e q PrStem coincides with the left edge of some RWd."

13. ALL-FT-LEFT Align (Ft, Lcft, PrWd, Left)

"The left e d p of every foot wincides tri th the left edge of some RWd."

What is interesting about the pmsodic organization of the candidates in ( 10) is that they al1

end in a foot with the exception of (Lod) where the PrStem ends in a light syllable. This last

candidate has a very restricted distribution in the nominal system, yet is widespread in the

verbal system. As noted by McCarthy and Rince (199ûa), if we exclu& qdriliterals,

nearl y dl the (ap&, nominals resdt from a sinde morphological pmcess which derives

the Form 1 active particifle W b f m the corresponding finite verb katab. Here 1 assume

that the deverbal stems result from an output-tcmutput relation behveen a verbai base and

an output nominal stem (BO-Imm), rather than k i n g derived from a nominal lexkal

mot (IO-F-m). 1 therefore regard them as k i n g outside of the dornain of study of the

nominal system.

3.2.2 Consonant 'Extrametricality'

An important characteristic of nominal and verbal stems in Arabic is the presence of a final

'estrametriai' consonant. McCarthy and Prince treat this final consonant as the onsel of an

empty syllable. Within the optimality-theoretic framework, 'extrametrical' constitwnts are

often seen as the result of satisfgng some nonfinality constra.int For esample, the parsing

of a final 'estrametrical' syllable by the PrWd rather than by a foot results from satisfyang

No>rn-.~u,HFr, while the tack of weight found in a final consonant follows from satisfying

N o s ~ ~ s . x H a .

13. NONFINALHCY No PrWd is right-aliped wi th its head qllable.

To see why the behavior of final consonants in the Arabic stem system seems to challenge

this t-ictv of estrametncality, consider again the prosodic structures in ( 10). Each candidate

ends in a root consonant which does not conm bute to ei ther the disÿllabic requirernent of

the PrWd o r the weight of the preceding syllable. If we assume, following McCarthy and

Prince ( 19%). that the final consonant is parsed as the onset of a headless syllable, the

prosodic shapes of the nominal stems would be as shown belou.

14a. ('opp)C, b. ('qa,)C, c. 4( 'q&o d- ('qLp)q& e- ( ' q L p ) ( ~ ~ ~ ) C a

In (Ma-c), the final consonant ensures that the head foot is not final in a PrWd. In ( 14d-e),

howcver, satisfaction of Nohm.u.HF~ does not hinge on the parsing of a consonant as the

onset of a final headless syllable, i.e. the head foot is already nonfinal. The question then is

what forces the final consonant in (14-e) to be parsed as the onset of a headless syllable.

short of stipulating that Arabic stems must end in an incomplete syllable, as McCanhy and

Pn ncc ( 1 B o a ) in façt do.4

1 propose that the solution as to why a wrbd or nominal stem must end in a consonant

lies in Broselow, Chen and Huffman's ( 1997) thwry of coda weight. Their mode1 assumes

that the weight of consonants in closed qllabies is an e f f e t of a universally hi@-ranking

constra.int which requires every coda consonant to be dependent on a mora

15. MORAICCODA (Broselow et ai. 1997:64) AI1 coda consonants must be dominated by a m o n

Broselow et al. propose that the way in which languages differ as to whether a consonant

dependent on a mom adds weight to a syllable or not follows from the language-specific

ranknp of two violable constraints: NOCMORA which demands thai the head of a m m be a

vowel (resulting in weightless coda consonanis sharing a mora with a preceding vowel) and

NOSH.UIEDMORA which requires that every mora be linked to a single segment (resulting in

bimoraic closed syllables).

16. NOCMORA The head of a mora must be a vowel.

17. NOSHAREDMORA Moras should be linked to s inde segments.

In their study of syllable weight, Broselow et ai. propose the ran)cing MouucCo~ »

N C S H - U I E D M ~ » N a m 4 for b t h Levantine Arabic and E ~ p t i a n Arabic. They also

note that the lack of weight in final CVC syllables in the Arabic stress system may possi M y

rcsult from a constra.int forbidding a final consonant to head a m o n Adopting Broselow et

al's anal ysis of Arabic coda weight, 1 propose that the answer as to why final consonants

are iwightless in Arabic lies in a ranking between a consuaint forbidding a final consonant

Or more d i d y that a RWd must end in a consonant as in McCarthy and Prince (1993b:23) FISAL-C: "Every PrlVd is coasoaaat-final". In my analysis. rhe parsing of the stem r d consonant as the omet of an empty-hearbrl syUabie fdlow~s from the interaction of a number of constraints. rather than from a specific consuaint rcquiring stems to be consonant fid.

to be the head of a mora ( N o F M L C M ~ ) and a constra.int requiring a syllaûie to dominate

a moxaic head (NOHLUXBSSIU),

18. NoFr NALCMORA No PrWd final consonant is the head of a m o n

19. NOHEADLESSSYLL A syllable must dominate a m o n

The parsing of a final consonant as the onset of a headless syllable in a nominal or verbai

stem would rhus be enforced by the following ranking:

These proposed rankings are illustraied belou. wi th a quadriliteral mot as the lexical input.

Nok thar in tableau ( I l ) , A m x - P R S m f - b is satisfied fully in d l the candidates. while

P ~ R S E - a is \.iolated in ( 2 1 a) to satisfy NoFrsa CMow without violating any of the higher-

In (21b), the final consonant is the head of a mora, vidating NOFN.V.CMORL In (Zlc), on

Noun Candidates P

x n j r a L

w !'

a. I / [(san).jal.r w !'ci

b. [@an).( jar)] w IL

c. [(xan). jar]

the other hand, the final mora is shared by the consonant and a preceding vowel, rcsul ting

in a fatal violation of N~SH-IREDMORA. The optimal candidate in (?-la) violates ody loarer-

Mo~uc 1 M d 0 i

C ~ A 1 I f f

i

1 %

rankcd NOHEIDLESSSYU. and PARSE-a.

N c Û ~ u i a >

MORA

*!

NOFRX

CM-

*!

*HEWUSS

SIU

t

Now consider another possible candidate for the quadriliteral rom: * [(san) .( ja ra) ] .

As shoirn by the l e s i d input of tableau (21). the vocalic melody is assurned to consist of

one monomoraic vowel which is copied whenever necessq . Following McCarthy ( 1995).

1 assume that the centrai difference betwveen concatenative and nonconcatenative languages

lies in the 10s.-ranking of O - C ~ T in nonconcatenative languaees.

32. O-CONTIGUITY The portion of S2 standing in correspondence forms a contiguous string.

(no intrusion)

Another important aspect of the mot-and-pattern system of Arabic is the domination of

I S ~ E Z R ~ ~ by D&O which forces copying of input segments over epenthesis ivhenever there

is not enough lesicai material to satisfy higher-ranked structural constraints.

33. ~ N T E C R I T Y ~ ~ No element of Si has multiple correspondents in S2.

This is illustrated in the tableau below.

As shown in (24), the domination of I S T E G R ~ ~ and lowcr-ranked N o H ~ . u x ~ s s S i u by

high-ranlung D d o and by the structural constrain& which determine the weight of coda

consonants edorccs a disyllabic shape on the PrStem and the parsinp of a final headless

syllable by the PrWd. A triliteral form is illustrated belou..

Noun Candidates P

sn j r a

w P a. 4 [(san). ja1.r

w P U b - [( san) . ( ja. n) ]

w P c. [(san). jI1.r .

w P!' d. [(san).(jar)]

DE$O

*!

NCSHW

MOFL~

I~TE~GRIO

*

**!

%

NOFISUC

MM\

"!

*HE~DLESS

SIU

*

rl:

Note that to have the final consonant parsed directly by the PrWd in either candidate \vouid

simply add an unnecessary violation of PME-a.

Consider now the minimal prosodic shape of a biliteral mot. In contrast with the

trilitcd and quaciditeral mots, the tanking NOFLXALCM~A » I s m d G is insufficient to

ensure the parsing of a final root consonant as the onset of a headless syllable. Both the

bimoraic forrn [(bar).r .] and disyllabic [(bara) j result in esactly one violation of l s n d o .

In other words, N o ~ w H F T , n7hich is h i g h - r d n g in the stress system of Anbic (see

Noun Candidates Ci

b h r a

w a. 4 [(bah).r .)

CI CL b. [(baha).r.]

chapter 4), also plays a role in the nominal stem system as shown below in (27).

36. N O N F I N A L H ~ No PrWd is right-aligned with its head foot.

N ~ H A R ~ N o F w C

MORA

I?.TEGR~*

*!

* m m

SIX

*

1

PARSE-O

In (Yb). the disyllabic foot requires a copy of the input vowel to fil1 the head of the second

syllablc, resulùng in a second and fatal vidation of I s m d O . The candidate in (37d), on the

other hand, fully satisfies I s m d o but vidaies both NOF~~ALCMORA and N~~TcI;MHFT.

The tableaux in (24) and (25) above demonstrate that the minimal p r d c shape of

trilitcd and quaciditeral nominal stems oui be accounted for by languap-specific r;uiliings

of IO-F .m constraints and their interaction uith the structural constraints determining the

weight of coda consonants. As for biliteral nominal stems. N ~ ~ A L H F T has been shown

to play a central role in determining their minimal proaodic shape. Crucially, nominal stems

in Arabic have a minimal prosodic shape thai is largely predictable from their consonantal

rooi (IO-F-uni). What those minimal prosodic shapes have in cornmon is that they satisfy

Nohm.aHFr. A u ~ s - P R S m i - L m and AL-&-Lm full y, n i th P.*€--a violations found

on1 y in quadriliteral stems. Another interesting aspect of the minimal prosodic shapes of the

nominal stems is that they al1 satisfy BR-LKHHUD. In the next sections, this structural

constraint will be shown to play an important role in the prosodic shape of nominal stems.

3.2.3 The Disyllabic Foot

The problem to be addressed at this point is how to account for biliteral and triliteral stems

that surface as a disyllabic f o o ~ Consider the prosodic shapes of the nominal stems. shown

again in (28).

38. Basic Nominal Pattern

Having dismissed the deverbal stems from the domain of analpis and having accounted for

the minimal prosodic shapes of the biliteral, triliteral and quadnl i ted stems, ihis leaves us

ui th the nominal stems in the second, third and fifth columns. The main difference between

large subsct of the nominal -stem is lesically-specified for a hi&-h-ranking version of the

29. ALICN-PRST-RIGHT-y Align (PrStem, Right, Foot, Right)

"The right edge of every PrStem coincides with the right e d g of some foot."

AUGS-PRST-RI-, howcver, c a ~ o t by itsel C ensure a di y llabic structure in triliterais or

biliterals (Le. *[(bw)].r YS. [(bu.ra)].r ). Some other force must be at work to force the

parsi ng of a disyllabic f w t in the nominal stems. 1 propose that this disyllabic requirement

c m be straïghtfonvardly accounted for by the constraint BRXXCHHWD. which requires the

head of a RStem to be in a relation with a prosodic consinient in a non-head position.

3 1. BRANCHHEAD The head of a PrStern is in a rhythmic relation with a non-head constituent at sorne level of anaiysis. (a, Ft).

The tableau in (32) shows the interaction betrveen BRL\SHEXD, which is assumed to be

undominated in the pneral phonology of Arabic, and AUGS-PRST-RIGHI. urhich is lesicaily

specified as high-rrurking for a subset of the nominal lexicon.

AL-FT-

LEF~

ISTM;RIO

**

&

&

AUGS-

PRST-R-y

* f

Noun Candidates P

s b a

Ci !'

a. [(saba)].b . w

b. [(sab).b.]

w c. [(sab)j.b.

PARSE-O

X

*

BRXKH-

HEW

*!

0bsen.e that the smallest nominal stem [(sab).b .] in (32b) satisfies BRLXHHEAD, allowing

for thc optimality of the disyllabic candidate whenever a bilited or trititeral stem is not

lcsicall y speci fied for Am-PRST-RIGHI--y (i .e. in the general phonologq. of the language).

The candidate in (32c), on thc other hmd, violates B~rlvcHHw. The central role played by

AUGS-PRST-RIGW in the parsing of a quadriliteral root is illustrated in the tableau below.

Although violaton of A UGS-PRST-RIGHT-y eliminates the left-headed candidate in (33b).

saisfaction of the constnint does not force a two-footed structure. In a quadnlilteral, the

parsing of the RStem-final syllable as a Tm< through lengthening of the vowel is not the

only way that AIJGS-PRST-RGKT-y oould be satisfied without violating undominated

U.\sro. Another psi ble candidate is [ra.(sam)].l., which ties wi th [ ( ras ) . ( r~a) ] . l . for the

1 ower-ran ked wnstmints I ~ T ~ c R ~ O , PARSE-o and ALL- FT-Lm.

What is striking aboui [ra.(sam)].l . is that it is not a well-fomed stem in Arabic.

Nominal and verbal stems with the sinichire [('apr)(opr) JC, or [ap( 'opd]C, always end

in a foot with a long voiel, never a closed syllable (McCarthy and Prince 1-43)? Oniy

the bimoraic stem allows (in fact requires) a moraic coda in this position, Le. adjacent to a

fi na1 headless syllable.

So far 1 have argued that the requirement of a final headless syllable for Arabic

stems should be analyzed as resulting h m the language-specilic ranking of the weight

Noun Candidates C

r s m l a 1

w !'CL

a. 4 [(ras).(maa)].l . w !'

b. [(ras).rna].l.

There is al ways a few lesical exceptions. like [disamSq] Damasas'.

64

BRLLVCH-

H m

AuGL\;-

PRST-R-y

*!

IATEGRIO

t

It

PARSE-O

*

t

AL-FT-

La-T

*

constraints of Broselow et al. with NOFKALCMORA and N o H E ~ s S ~ U The rankings

alone, hoivever, cannot esplain the absence of coda consonants in a foot followed by a

hcadless syllable which is directiy parsed by the PrWd. Note that to assume a constraint

forbidding a moraic consonant to be adjacent to a final headless syllable would incorrectiy

cvaiuate [(sab).b .] as non-optimal. Such a constraint could not distinguish between a

headiess syllable parsed by a RStem, which ailows a preceding moraic coda, and one

parsed by a M d , which does not. Crucial1 y, the moraic coda of the riphtmost foot is

optimal on1 y when the following syllable satisfies PARSE-a.

Downing (1998) proposes that this son of dependency between two consuaints is

best espressed by conjoining the two consuaints to defïne a new constraint. Following.

work by Hewitt and Crowhurst (19%). she assumes that this type of conjunction is

satisfied if and only if there is some domain in which both consrraints are satisfied. To

account for why moraic codas are not acceptable when fdlowed by a .s$laMe directly

parsed by a PrWd, 1 propose a distinct consinint conjoini ng NOFNALCMORA and PARSE-a.

One advantage of choosing N o F r s x ~ C M r n ~ as one of the wnjoined constraints is that the

domain of the new constraint is limited to c a b s that are at m a t one syllable away from the

nght cdge of the PrWd- As s h o w by the rankings a t the top of taMeau (34) belon., the

conjoined constraint is regarded as distinct from the constraint PARSE-a.

1 Noun Candidates

P r s m l a

I

w PCr a. V' [(ras).(rnaa)].I .

1

P w b. [ra(sam)].l .

I

w Ci

ç. [(ras).ma].l.

NOFISALCMOR~

n PARSE-a

(4 *! (*)

-

M d 0 ALIGS-

PRST-R-y

*!

Au-FT-

LEE=T

8

*

As shown in (Mb), the conjoineâ constraint is violated if either of the two constraints is

violated and satisfied only if both are satisfied. Coda consonants that are more than one

syllable away from the ri@t edge of a PrWd are not subject to NOFINALCMORA and are

thercfore outside of the domain of the conjoined wnstraint-

3.2.4 Mora Faithfulness

Up to this point, the minimal prosodic shapes of biliteral. triliteral, and quadriliteral nominal

stems have k e n shown to rcsult from satisfying hi&-rdcing faithfulness and structurai

constraints, while the maximal p r o d i c shapcs of these stems have been iirgued to follow

fiom a les icd y-specific version of Am-PRST-RIGHT assigned to a subset of the nominal

lesicon.

Consider now biliteral and triliteral stems which are in fact charactenzed by an initial

unfooted syllabe (e-g. triliteral [?a..(taa)].n .). What is striking about this unespected shape

---'unespectedf with respect to the structural constraints already discussed--- is that the

most straightfoward solution which is to assume faithfulness to an input bimomic vowel

nVill not work. Faithfulness to underlqing quantitative distinctions is usually expressed by

the constrai nt WT-i DESTP

35. WT- [DENT (McCarthy 1995) If a and fl are in a correspondence relation,

and a is monomoraic, then is rnonomoraic (no Iengthening). and a is bimoraic, then f3 is bimoraic (no shortening).

R d 1 that in the case of the quadriliteral [(ras).(ntaa].l ., weight mrrespondence behveen

the monomoraic input vowel and the bimoraic vowel in the second foot of the output

L-ioiates W T - I ~ T V In [ & . ( t ~ ) ] . n ., on the other hand, it would be the first vowel of the

output that vidates W T - I ~ T I O , assuming tfiat the input vo\trel is bimoraic.

To account for the violation of the faithfulness constraint in both the quadrilitenl

and the triliteral, I p q m e that two-fwted structures in Arabic violate a consbaint requiring

that every foot dominated by a M t e m be parsed as i ts head.

Satisfaction of M~SOHEADED~ESS which requires every prosodic constituent to be uniquely

headcd rcsults in a vidation of F T - D P R S ~ I for each two-footed but one-headed PrStem,

as shown in (37d).6

Chi>- (37c) violaces high-ranlring AUGX-PRST-RIGHI--y, whi le a violation of F T - D P R S ~ I

climinates (37d). The choice of the optima! candidate is thus left to W T - I ~ T I ~ .

The twdooted quadriliterals. on the other hand, resul t from the domination of FT-

DPRSTEN by M . d o and AUGS-PRST-WGHT-y, as shown by the tableau below .

Noun Candidates w

3111 a 1

w a. 4 [?a.(taa)].n .

Ci u b. [(%ta)j.n .

w Ci

c. [(?aa).ta].n. Cilr !w

d. [(?aa).(taa)].n .

Soie that in Japanese. FT-TO-PRSTE!U would have to be very low-ranking.

67

AUGN-

PRST-R-y

*!

FT-m-

PRSTEM

*!

WT-

I m - r I O

*

**!

*

I o

t

*

*

*

ALL-FT-

Lm-

*

I

t

A

Observe thai the main di fference between the quadrili teral form in (38) and the trilitcd form

in (37) follows directly from their number of root consonants and the ranking ML.-IO 22

Fr-m- miSrni » WT-IDESI-IO.

Consider now the hm- footcd structure in tri1 i terah li ke [ jabbaar]. These norni na1

Noun Candidates P

r s m l a w PP

a. d [(ras).(maa)].l .

w v b. [(ras).rna)-1.

stems have a consonantal mot that could be defined with hvo identical medial consonants,

making h e m somewhat esceptional. McCarthy and Prince ( l m ) , however, note that

media1 gernination applies productively in the noun of profession or habitual action and

semi-productively in one type of broken plural.

The analysis proposed here wkes the position thai forms like [jabbaarj result from a

floating mora which is a lexical fealure of some consonantal roots7 The obligatory pmsing

of the mora by the consonantal r w t is assumai to follow from a faithfulness constraint

w-hich requires for every mora in the input to be associated with an output segment: p-

~ m ' o . In contrast with WT-IDE~TIO, which forbids both lengthening and shortening. p-

M d 0 only blocks the l o s or an input mora and can be satisfied through cornpensatory

lengthening.

P w 1 c. [ra.(sarn)l.l. I I I I

Mmro

In the morphological deridon of nouns of profession or habitual action. however. the gemination of the media1 consonant would be a n a l y d as rhe idkation of a tloating mora to the head sylfable of the nominai stem. (Sce the analysis of the verbal systcm in the wxt section.)

NOFISALC~

n PARSE-a

A ~ G s -

PRST-R-y

*!

FT-m-

PRSTEM

t

WT-

I ~ T I O

*

39. ~ M A x I O Eve- m o n of Sr has a correspondent linked to a segment in S2.

This is illustratcd in the tableau below.

Anothcr p s i ble candidate [. j. ( ja ba)).r . creates a fatal violation of OSSEL 8

Noun Candidates CI

U j b r a 1

w !'P

a. 4 [(jab).(baa)].r . w !'

b. [(jab).ba].r .

Cr CI

c. [(ja.ba)J.r.

C'w d. [ja-(bab)].r .

Finally, we are left with the nominal fonns in the last column, which contain two

long \-owels. McCarthy and Rince (1990:44) point out that triliteral stems like [jaamuus]

rue rather rare and not used in any systematic way by the morphology. As noted by

McCarthy and Pnnce, Le\-y ( 197 1) even refers to these forms as only 'semi-canonical'. 1

ti-ill thereforc assume that the 'semi-canonical' status of this latter type of nominal stems

rcsul ts from a vowel melody containine one or two long vowels and a l e s i d l y specific

version of WT-IDBT which domi nates Fr-TO-PRSTEM in the constrai nt hierarchy, ensuri ng

that cach of the two long vowels wiII be parsed as the head of a foot.

Sote that in the case of / &.(tao)].n . wiih an input long vowel. r - ~ ~ l o is satisfied in the output form sincc both input moras are par& by an output segment.

p-M.d0

*!

NoFrm~Cp

n P-ARSE-0

(4 *! (9

AUGS

PRST-R-y

*!

FT-TO-

P R S T ~

*

WT-

I ~ T I O

t

To sum up, within the mode1 of prosodic structure developed hem, 1 have shown that the

minimal p r o d i c shapes of nominal stems can be accounted for in a straighdonvard way

by the interaction between the hifi-ranking faithfulness and structural constraints of the

language. Less predictable prosodic shapes, on the other hand. have been shown to follow

from the ie'rically specific version of AUGS-PRST-KGHT assigned to a subset of the nominal

lesicon. 1 will tum now to the verbal system of Arabic.

3.3 The Verbal Stem System

The phonolog- of the verbal system shows many similarities with that of the nominal

system. The minimal p r d i c shape of verbal stems is disyllabic [(opp)]u, while its

maximal prosodic shape in the non-finite forms contains two monosyllabic feet

[(O~~)(O~~)]O. In the finite forrn, however, the verbal stem may contain only one foot

rvhich 1s alway left-aiiped with the PrStem.9

42. Basic Verbal Systent Cffnite fornt)

1 propose that m a t of the mot-and-pattern morpholopy of the verbal system (Le. the

conj ugation of the Arabic verb) is best represented as a relation between the basic verb forni

(Form 1) and al1 the other related forms within a patadigrn. l0

Fonrr I

hala1

faCd

-

Sote that foms 5 and 6, tv-CvCiCjvC and tv-CvvCvC respectivcly, aloag uith form Q2 tv-CvCCvC are not included in the rableau. Tbese forms can be ùandied by assurning that the affix must lx prefisd to a rnorphological stem: -4ligu (MLx. Right. Stem. Left). The an?s must be furtber spccified so as to attach ody to forms 2, 3 and Q I . Crucialiy. prefixes and suffixes that are rquired to be alignai with a morphoIogica1 stem are outsi& of the domain of the verbai stem system.

The numerals in Form 1. Form 2. etc. are the desi@- for the Western system of classification.

Fonrt 2

hallal

faTfal

Fornt 3

haalal

faam

QI

dahra J

Fornr 4

W ~ l a l

3afTal 1

Fonn 7

nhaiai

nfaTal

Fonn 8

htalai

ftaTal

Fonn 9

hialai

fCdd

e4 dharja j

Fonrt IO

stahlal

stafTal

Fonrt I f

hlaalal

fTaaial

2

L

43. Input: Cons-Root 1 VocMeldy

IO-F.m /f OUtplt: Base~orml c=; Skm~onn 2. Form 3. etc.

Unl ih the nominal -stem, the minimal prasodic shape of the verbal base is a disyllabic

foot. Anothcr difierence berneen the hvo systems is the absence of hvo-fmted structures in

thc finite forrn of verbal stems. These hvo obsen-atioris can be @\.en a unified account by

assuming that in the finite rom, therc is a lesically specific version of ALIGS-PRST-RIGHT

assigneci to verbal stems, one which is dominated by FT-TD-PRSTEU.

The d e played by A L I G S - P R S T - R I ~ - ~ ~ in the verbal system is illustrated in the tableau

1 n a uiliteral or biliteral mot, the lexically specific alignment constraint forces a disyllabic

Coot sbpe on the verbal stem, as shown in (4%). In a quadriliteral root, however, the

domination of Aucs -PRS~-Ricc r r -~ by F j r - m - P ~ S r n ~ results in a violation of the former in

the optimal verbal stem in (4%) below.

IS~EGRIO

t*

x*

*

WGS-

PRST-R-6

*!

"!

FT-TO-

PRSTR.I

Verb Candidates IL

h l a I

C P a. 4 [(ha.la)].l .

C Ii

b. [(ha.la).l .]

Ir(r

c. [(hal).l .]

BR.IXCH- i N o F w C p

HEAD j n P~SE-a

; (4 4 (4

Art-Fr-Lm

As shown by the two tableaus above, the minimal p r d i c shapes of verbal stems can be

straigh tfonvard t y accounted for by some reranking between Fr-mPRSmr and a Iesically

speci fic version of AUGS-PRST-RIGH~.

3.3.1 Output-to-Output Correspondence

The problem that has to be addressed now is how to account for the prosodic shapes of the

output stems related to the base forms. Abstracting away from initid and final consonants

parsed by headless syllables, al1 of the output stems can bc characterized as a bimoraic f m t

followed by a light qllable. Differences between these output stems aithin the finite f o m

paradip corne from çome weight specification usually associated with a segment in the

head syllable of the base. Form 3, for esample, lcngthens the vowel in the head syllable of

the base. Following different patterns, îorms 4, 8, 9 and Q4 assign a mora to the initiai

consonant of the base. In Fonn 2, on the other hand, the second consonant of the base is

gcminated resulting in a bimoraic head syllabie, whilc in Form 7, an initial mora is linkcd to

a consonantal prefix.

W hat these mo~hologicai fonns have in common is ( i ) the ad& tion of a mora to an

input element, either a consonani;ll prefin or a segment from the base, and (ii) that this

addi tiond weight targets in m e way the initiai syllable of the base. T o account for these

Verb Candidates !'

d h r j a 1

lrlr Cr

a. 4 [(&h).ra].j .

w PP b . [(dah). (raa) 1. j .

w UP

r=. [(dafO.(mj)l-j -

ALIGX-

PRST-R-p

t

I ~ ~ f o

t

t

tt

BRASCH- N O F W C ~

HEW i n P.WE-O

; (4*! (*)

FT-ID

PRSTEI~~

*!

X

obsen.ations, 1 propose that the output stems are 'derived' from the base by the affisation

of a floating mora to the phonologicaliy-empty stems. The differences between the output

stems themselves are assumed to follow from the distinct sets of BO-Identity constraints

spccific to each of the forms within the fini te paradi gm-

47. Input: ConsRoot / Voc. Melody

In addition 1 propose that the moraic affix is lexicaliy subcategorized for a spcci fic prosodic

base. The subcategorization constraint associated with the moraic affix is s h o w below.

48. AFFI X-TOI H u The right edge of the pA~ffin is left-aliped with the head syllable of some PrWd.

1dent.i~. between the base and the output stems is detennined by the interaction of the BO

constraints with ~ - W Y I O . which requires an affixed mora to surface in the output. and with

the subcategorization constraint A m - D H O .

Consider first the relation between the base and the Form 2 stem. The formulation

of A m - r o - H a in (48) &man& perfect d ig rnen t between the right edge of the prefised

mora and the left edge of sorne head syllable. In Form 2, however, the right edge of the

affised mora is associaied with the right edge of the head syllaMe instead of the left edge-

This typc of infixation can be acmunted for if we assume that ASCHOR-LR=~~ dominates

A m - m - H a in Form 2 of the verbal stem.

49. ANCHOR-LEF~ Any elernent at the lcft edge of S 1 has a correspondent at the left edge of S2.

The gemination of the second consonant of the base, on the other hand, fdlows from the

domination of I h r n ~ ~ by W T - I o ~ T ~ . The proposais are illustrated in the tableau below.

(Note that 'lFomrl should be rcad as P ~ f f i ~ - StemlFo,,,,il. )

In (50c) the affised mora is perfectl y aligned with the left edge of the head syllable, but i ts

association with an initial vowel resd ts in a fatal kiolation of ASCHOR-LmB0. In (SOd), the

non-parsîng of the mora, while vacuously satisfying Am-TO-Ho, violates undominateci p-

M.L.~*. The choice of the optimal output is thus left to lower-ranked WT-i DESTBO. Note that

c ~ c q - candidate in (50) satisfies Fr-TO-PRSTEM, while only those in (50c-ci) satisfy lower-

rankcd A UGS-PRST-RIGw-~~.

1 n contrast with Form 3, the output of Form 3 is characterized by the lengthening of

thc vo~.el in the head syllable. To account for this differençe in output forms, we need to

assume that some of the BO-Identity constraints specific to Form 3 are ranked differently

from the constraints involved in the relation b e ~ e e n Fom 2 and the base. Cnicially, the

relation between Form 3 and the base is characterizad by the dominance of WT-IDE~T~O by

1 h - r ~ ~ ~ ~ ~ . This is illustrated in the tableau beiow.

F

Verb Candidates I i ! '

Base: hala-1 ~ [ F O I Y I I ~ ~ l

w P

a. J [(hal).la].l .

w IL b. [(haa).Ia].l .

Ii P P

c. [a(hala)].l . P !'

d. [(ha.la)].I .

p-Mdo

*!

I S E R ~

*

* I

6

WT-

IDRT~O

*!

AXCHOR-

IJFrm

*!

AFFIS-~o-

HU

*

*

Note again chat the domination of A m - m - H o by A S C H O R - L E F ~ ~ forces the infixation of

the mora to the right edge of the head syllablc rather than its prefixation at the left edge of

the head syliabk.

There is, however, a third possible output not shown in the tableau above which

associates a prefixed mora to the initial consonant of the base: F.(lala)].l . Inkrestingl y,

WT-

ID EST^

* I

this prosdic shape is in Tact used to define Form 9 in the verbal paradigm. 1 propose that

the main difference bçtween Form 9 and the two previous Foms is the ranking of the

fai thfulness constrainl LP.€~R~O.

I ~ T E G R ~

*!

52. LINEARITY~O Si is consistent with the preceding structure of S2, and vice versa

Verb Candidates P P

Base: hala1 ~ [ F O ~ I r

w P

a. d [(haa).laJ.l . w P

b. [(hal).la].l .

Lnu~m requires that the linear order of the base segments be presewed in the linear order

of the output stem. Here 1 propose that what is specific to the relation between the Fonn 9

output stem and the base is the low-ranhng of LI~EwBO. This would contrast with the

high-ranking of L I S ~ R B O in Forms 2 and 3. The iableau illustrating Form 3 is @ven again

in (53) below to shown the r d e played by LB- in selecting the optimal candidate. The

d n g of L I ~ E I R ~ O in Form 9 is illustrateci in the nest tableau (54).

A Y ~ R -

L m B o

p-Mdo A F ~ - T O -

Ha

t

*

-

( Verb Candidates

R d 1 that the same ranking for L R E - \ R ~ is assumed in the relation between F o m 1 and

the base. Form 9 is shown below.

54. ; Mx@ » BRAKHHUD » ASCHOR-Lmm » A F ~ - m H o » WT-

Verb Candidates P P

Base: ha.lal p ~ ~ ~ 9 ]

As show in (54a), full satisfaction of ASCH~R-LETT~ and A m - D H U results in the first

consonant of the base k i n g parsed as the coda of an empty-headed syllable. Observe aiso

that satisfaction of the IWO high-ranking constraints results in the satisfaction of lower-

ranked A LIGS- PRST-RIGW-p.

Consider now the Form 4 verbal stem. This output f o m differs from the previous

stems, in that the affisation of a mora dœs not result in either lengthening or gemination. In

fact, Form 4 can be characterized by the absence of violation for the tu'o identie constraints

1 Y ~ R B O and W T - I ~ T B O . Imtead, metathesis between the initial root consonant and the

follori~ing \-owel &es place, vioiating both A S C H O R - L ~ ~ and LISE~BO.

The distinct rankings of the BO-Identity constraints specific to the relation behveen

Form 4 and the base are il f ustrated in the following tableau.

55. p-M-dO ; Ma* » BRIKCHHWD » WT-ID EST^ ; I S I E G R ~ >> Amy-TO-HU

» AS(HOR-LET~ ; LISEARBO » F r - n k S m ~ » A=-PRST-RIW-~ » PARSE-O » ALL-FT-LW » LL\E-\R~O

Verb Candidates !'

Base:ha.lal pi^^^^] w u

a. 4 [(ah).la]-1 . w

b. [(haala)].l . !' I r P

c. [h.(la.la)].l

PC P d. [(hal.la)j.l .

Note that in (55a), in spite of the metathesized first t ~ - o segments of the base, the aîatfisal

m o n is stitl assumed to be a i i p e d rtith the r@t edge of the kad syllable, resulting in the

\-iolation of Ams--Ha.

Considcr finally the prosodic shape of Q4 verbal stems. 1 propose that the parsing

of the initial root consonant as the monic coda of a headless syllable car^ be accoun ted for

in a straightfonvard manner by assuming that A m x - m - H a dominates AUGS-PRST-Lm in

the \.erbal system. This is illustrateci in the tableau bclow.

3.3.2 Double Prefixation

Additionai evidence for the analysis proposed above cornes from the bchavior of the

consonantai prefises in Forms 7 and 8. The initial p s i tion in the output of the prefix n- in

Form 7, compared with the second position of the prefix f - in Form 8. strongly suggests

that both prefises are subcategorized for AFFK-TDH~ and are subject to the sets of BO-

Identity constraints specified for the output stems. Form 7 can be defincd as resulting [rom

the prcfisation of msonantal n- and of a moraic affis to the FOrm 4 stem, while Form 8

rcsults from the prefiwtion of cmsonantal t- and of a moraic affix to the Form 9 stem.

This is shown in the two tableaux below.

IXTEGR~O

1:

AUGS-

PRST-Lm

*

AFFIS-D

Ha

*

*!

Verb Candidates PU P

Base: dah .ra. j p1~41

!' Pt' Ir a d d.[(har).ja].j .

lrlr P

b. [(dah).ra]. j .

w Plr c. [(dah).(raa)].j.

p-U\s"

*!

AXCHOR-

Imr Bo

Verb Candidates I .Ci

1 Bsse: ha.la.1 n-pr~om4]

I d . [(ah).lal.l.

In ( 5 3 , the constraint is specified for both affixes simultaneously. Each affis which is not

left-aliped nvith the head syllable ad& a violation mark to the colurnn. Note that (5%) is

iess optimal than ( 5 7 ~ ) . since it has one additional violation. The optimal candidate in (57a)

homever, does not violate either of the two instances of the subcategorization constraint.

1 Yerb Candidates

Ci l ' C i 1 b. 4 [ h.(ta.la)].l .

Ir Ci

Base: ha-la1 t - p I ~ ~ m ~ ~ v

hZL. Io Ln=rBo

In the nest chapter, additional evïdence for the PrStem is provided by an analysis of the

stress system of a number of Arabic dialects. In conjunction with this anaiysis of the Arabic

stress systems, a number of theoretical issues regarding heads of prosodic wnstituents and

their projection on the grid will be discussed.

Chapter 4

Heads and Grid Prominence

In chaptcrs 2 and 3, an acwunt of word minimality has been offcred that builds on the

proposai that an intermediate level, that of the PrStem, is found benveen the PrWd and the

foot in the Prosodic Hierarchy. To provide further support for the PrStem, 1 intend to show

that this ncw metrical constituent plays an important d e in longer fonns, where the M t e m

is not coestensive with the phonological word.

4.1 Stress as Prominence on the Grid

In thcir influentid study of stress systems, Halle and Vergnaud ( 1987) propose that stress

is a relation betueen strong and weak elements that is computed on an autosegmental plane

or grid projected from the segments and associated skeletal slots. On this metricd gnd, al1

stressable eIements (Le. syllabic segments that can bear stress) project a grid mark at the

loivcst levei (line O). These stress-bearing elements are organized into headed çonstituents

ei ther b i n q or unbounded, wi th the metrical groupings k i n g directional and exhaustive,

forcing the construction of degenerate u n q constituents. Constituent heads then project a

grid mark on the nest Içvel (level 1). In quanti@-sensitive languages, a bimoraic syllabie (in

Hal te and Vergnaud's terms, a syllable with a branching rime) also projects a grid mark on

Icipel 1. A single head at an edge then receives a grid mark on level 2 in accordance with the

languagc's particular End Rule.

Halle and Vergnaud's m d e l of stress as prominence on the grid is elaborated on

further by Seikirk ( 1984), Hayes ( 1985, 1987, 1995), McCarthy and Prince ( 1986, 1988,

199Oab). They independentl y propose that relative prominenœ should be derived from the

basic properties of syllables and foot structure. These relations are then directly mapped

onto the grid, with heavy syllables and foot heads obligatonly projecting a grid mark at

level 1 under compulsion from the Weight-to-Stress Principle (Prince 1990). A single foot

head is then chosen to project a grid mark at level 2 in accordance with the language's End

Rule.

At issue here is how to acwunt for stress systems where iterative foot parsing and

the location of heavy qllables play a central rote in determinin? the position of main stress,

>et where the predicted secondary stresses are absent In Classical Arabic, for enample,

presence o r absence of heavy syllabies and their positions within the word are crucial in

determining the location of main stress. S e o o n d q stresses are absent, however, even in

n.ords nrith more than one heavy syllable. A sirnilar pattern is found in Cairene Arabic. In

this dialect, both the location of hemy syllables and iterative f m t distribution determine the

Iacation of main stress. Again no m n d a r y stresses are found in the final output. The

question then is houf to resolve the apparent confiict behveen the absence of prominence on

the grid for sorne prosodic constituents with their role in positioning the mai n-stressed foot

ni th respect to the edge of a PrWd.

In this chapter, 1 reanalyse the stress V t e m s of a number of Arabic dialects in

terms of the mode1 developed here. 1 propose thai these stress systems can be accounted for

in a straightfonvard manner by assurning that headship within metrical constituents is often

cocstensivc with but actually distinct rrom prominence on the grid.

4.2 The Weight-to-Stress Principle

Before reandqzing the Arabic stress systems, 1 will begin by discussing the status of the

Weight-testress Rinciple in the mode1 of prosodic structure proposed here. The standard

interpretation of Weight-to-Stress within Optirnality Th- is as a requirement imposed on

foot structure and on grid prominence.

1. WEIGHT-TO- STRESS (Prince and Smolensky 199353) Heavy syllables are prominent in foot structure and on the grid.

Hcre 1 propose that Weight-to-Strxss should be subdivided into two distinct requirements

on heads: WHG~-TO-HE;V) and STRESS-DHEW. WEIGW-TO-HWD requires weight to be

assessed in ternis of the headship of metrical mnstituents, while STRESS--Hm restricts

projection ont0 the grid to some specific type of head.

3. WEIGHT-TO-H(PRCAT) A bimoraic syllable is parsed as the head of some prosodic categot)..

3. STRESS-TOIH(PRCAT) Prominence is only assigned to the head of some prosodic category.

For example, W E I G W - ~ - H ( R ) wuires hem? syllables to be parsed as heads of feet,

n.hilc STRESS-~H(kSm.1) only allows the head of a foot parseci by a M t e m to project

onto the grid. Wac~r-m-H(PRSmi), on the other hand, requires a h v y -Ilable to be

uniquely parsed as the head f w t of a PrStern. while STRESS-ID-H(FT) requires eeey foot

head to project ont0 the gnd, irrespective of weight 1 t is important to note that there is no

potential c o ~ i c t between the hvo types of cons train^ In the nest sections. analyses of a

number of Arabic diafects will demonstrate the role piayed by these constraints in the stress

system of languages.

4.2.1 Classical Arabic

Consider first the stress system of Classical Arabic. The data below are from McCarthp and

Pri nce ( 1 WOa).

4. Stress Placernenr

Final

t3ir jama

The conditions on stress placement in Classicd Arabic can be describeci as follows:

5 . Stressasuperhea~ysyllableattherightedge; Othewise stress the rightmost nonfinal heavy syllable;

Othenvise stress the initial syllable.

These conditions are even better illustrated, when we restrict our attention to words that are

three syllables and longer, as shown belocv. The additional data is from McCarthy (1979).

6. Initial Mahatun mdmlakatun

In Classical Arabic, the location of main stress depends cmcially on the position of heavy

svllables uithin the PrWd, even though only the rightmost non-final one is stressed. When

there is no heay syllable in non-final position, an initial light syllable receives main stress.

The first problem to be addressed is how to ensure that a rightmost non-final h e a q

syllable is chosen as the main stressed foot over a word-initial disyllabic foot If as is often

assumed there is no secondary stress in Classical Arabic, then an additional problem is how

to account for the absence of secon* stress in words with more than one non-final heavy

syiiable. As a solution to the first problem, 1 propose a constraint which requires ha \ -y

syllables to be parsed, not as the head of a foot (the standard definition of the WSP), but

rather as the head or a PrStem.

7. WEIGHT-TO-H(PRSTEM) A bimoraic syllable is parsed as the head fmt of a PrStem.

Satisfaction of this constraint entails that a h a v y syllable must be uniquel y parsed as a fmt

that is itself parsed by a PrStem.

The initial main stress found in words uithout non-final heavy syllables, on the

other hand, is accounied for by assuming ihai A U G S - P R S ~ I - L m , although dominated by

WEIGHI--1-0-H(PRS~I), is still hi@-ranking in the language. The absence of main stress on

final long vowels is assumed to follow from the domination of W E I G K ~ - ~ - H ( P R S ~ ) by

8. ALIGN- PRSTEM-LEFT Align (PrStem. Left, RWd, Left)

"The left edge of every PrStem coincides with the left edge of some RWd. "

9. NONFINALHF~ No PrWd is right-aligned with its head fmt.

Thcse proposais are illustrated in the taMeau below. As shown in (lûa), A w s - P R S ~ I -

L m is catcgorically evaluated. Every PrStem which is not Ieft-aligned with a PrWd counts

as onc violation of AUGN-PRSTEX-LI~T, irrespective of its position within the word.

1 n words wi thout non-final heavy syllables, on the other hand, satisfaction of AUGN-PRST-

Candidates

m d i l u !' Plr * C L

a. J[ma.(naa)].[(dii).lu]

CL w P l t P

b. [ma(naa)].(dii).lu

L m rcsults in the initial syllable receiving main stress.1 Note that the ranla'ng PARSE-O >>

AL-FT-LEI- is assurned to result in a trisyllabic PrStem in odd-numbered words.

Nosnu,i\~-

HFT

Consider now words which contain only one final closed syllable. Here 1 assume that the

WHGW-

HEAD

*!

Candidates

Sa jaratuhu

C C L C L

a. 4 [(s'a. ja).ra].(tu. hu)

)r lr U l t CL

b. [(Sa. ja)].(ra.tu).hu

Y ! ' P Y Ir

c. (Sa. ja). [(ra-tu). hu]

final syllable is monomoraic to satisfy undominated NoFc laCMm.

12. NOFINALCMORA No PrWd final consonant is the head of a rnora.

AUGN-

PRST-LEFI-

t

- --

I The word [Ja jasatubu] 'his irec (nom.)' is listed as a Caircne Classical Amtic form in the data on Cairene Arabic in Hayes (1995).

PXRSE-a

I

a

*

NOSFIS+L-

Hm

P . ~ s E - a

*!

ALIGS-

PRST-Lm

"!

AL-Fr-

Ln=r

ttt

1

t*

**

This is illustrated in the tableau belotv.

Candidates

mamlakatun

0bsen.e that with respect to the f o m [mamlakatun], al1 of the structural consu-ainis already

introduced are satisfied by an initial main-stressed sq-llable, uith the esception of ALL-FT-

Consider now che problem of how to ensure that only the main stressed foot will

have prominence on the grid. In Halle and Vergnaud's ( 1987) frarnework, heavy syllables

are assigned a grid mark on line 1 under compulsion of the WSP, with the exception of

final heavy syllabies which are made prosodicdly invisible through the estrametricality of

the last segment in the word. Left-headed unbounded constituents are then constructed o n

line O, with the constitucnt heads projecting a grid mark on line 1. The rightmost grid mark

on Iinc 1 then projects a grid mark o n line 2. This is illustrated with the f o m s [ y u f i h l

and [bdahatun]. The last segment is treated as estrametrical.

yu saa r i ka^ ba la ha tu cn>

iine 2 line 1 line O

To account for the la& of secondary stress, Halle and Vergnaud propose a rule which

conflates lines 1 and 2 after main stress has been assigned According to their pro@,

\i.hen hvo Iines in a metrical grid are confiaîed, a constituent on the lower line is presewed

on1 y if its head is also a hcad on the higher line. Crucially, the loss of grid mark on line 1 is

assurned to entail both the loss of the head from which it was pmjected and the los of the

constituent i tsel f.

Conflation follows from the standard assurnption that mecricai constituents must have

a head. Consequently the l as of a head p i t i o n on the metrical grid results in the los of

the metrical constituent itself' This standard assumption is dso part of the melrical theorïes

i ndependen tl y proposed i n Hayes ( 1995) and McCarth y and Pn nce ( 1986). Processes such

as Conflation and constituent erasure, however, are intrinsically repair strategïes that can

onl y be part of a derivational framework In Optirnali ty Theory, such repair strategks have

n o possi bie status, since only the input and the output are available for evaluation.

The account proposed here takes the position that constituent heads and prominence

on the grid are hvo distinct requirements. The lack of secondary stress in Classical Arabic is

assumed to follow from a wnstraint requiring that only the head of a PrWd may project a

grid mark on the grid at any level.

15. STRESS-TO-H(PRWD) Prominence is only assigned to the head of a PrWd.

Whe9 a PrWd dominates more than one PrStem as shown in ( 16), only one can qualify as

ihe hcad of the PrWd in order to satisfy M O S O H E ~ ~ ' E S S . =

16. PrWd I

PrStern PrStern I 1

Ft Ft i I

17. MONOHEADEDNESS (Crowhurst 1996) Prosodic constituents are uniquely headed.

S e e the anal y sis of Guugu Y imidhirr in chapta 5 wbere the domination of XfONOHE4DW'ESS by FT- TO-PRSTEIrl ensures that both fecr within a head PrStem will receive main stress.

In any language where -MOSOHE.U>EDT\IESS is undominated, the default option for the head

PrStern would foilow from the choice of either AUGN-PRST-LEFI- or AUGS-PRST-R~GW by

the grammar. However, in a stress system where the more specific requirement (Stress the

rightmost non-final heavy syllable) is at one edge, yet the default location of main stress is

at the opposite edge (Ohenvise stress the initiai syllabe), the position of the head PrStem is

assumed to result from the satisfaction of a constraint demanding that of al1 the possible

heads of a PrWd, the rightmost one be chosen.

18. RICHTMOST The rightmost PrStem is defined as the head of a PrWd.

Satisfaction of this constraint only requires that the rightmost PrStern qualify as the head of

a PrWd. Rrcmrosr is indifferent as to whether other PrStems within the word are aiso

defined as PrWd heads. T o have more than one PrStem as head of the PrWd, however

~vould violate Moso~--Ess.These arguments are illustrated in the tableau below.

Satisfaction of STRESS-~~-H(PRWD), on the other hand, demands that on1 y the head(s) of

the PrWd be ailoured to project on the metric. grid. This means that only the head syliable

of thc PrWd can project a grid mark on line 1. As the head f o d of a PrStem, it ais0 projects

a mark on line 2. Finally, as the only head found o n line 2, it is then assigned a grid mark

on line 3.

Candidates

mamadiilu CL VIL w f'

a. J[ma.(naa) j.[(diï).lu]

P CiCi w P

b. [ma(nda)].[(dil).lu]

P w P P P

c. [ma (nja)] . [(dii). l u]

Mo.- W ~ m r o s r

HE;\DED i

*! ;

*!

WHGKF

HEAD

ALJGS-

PRST-L

*

P-ARSE-a

,.

t

*

. . * - line 3 H(PrWd) . . * . line 3 H(PrStem) . . (9 - line 1 H( Ft)

Consider nonr a form which has no heavy syllables in non-final position. As illustrated in

the earlier tableau ( 1 1 ), satisfaction of AUGN-PRST-LEFI- results in an ini tid main- stressed

foo t .

2 1. * . . . . line 3 H(PrWd) * . . . . line 2 H(PrStem)

(*) . - . line 1 H m

P P P C L C 1 [(Sa. Ja).ra].(tu. hu)

The benefits of subdividing the WSP into hvo distinct constraints are clearly seen from the

analysis just presented. On the one hand, the constm.int smEss-rn-H(PRW~) eliminates the

need for Conflation as a repair stragegy. On the other hand, the consvaint WHGKT-TO-H

( PRSTEN) eliminates the need for unbounded constituents which have no place in restrictive

thcories of foot structure?

To summarize, the analysis of the Classical Arabic stress system presented here

offers support for a mode1 of prosodic structure ivhich estends the Prosodic Hierarchy to

include a new metrical constituent, the PrStem, behveen the foot and the PrWd. An

important aspect of this mode1 is chat headship follows directly from the Prosdic

Hierarchy. Whether the head of some prosodic category is assigned prominence on the

metrical grid or not, on the other hand, results from a specific type of mnstxa.int STRESS--

HE=- Under this view, stress systems where syllable structure and foot distribution play a

role in determining the location of main stress yet where seconckq stress is absent are

predicted by the theory rather than k i n g problematic for it.

In my analysis, the position of main stress in Classical .eabic is acc~unted for in a straighdorward manner. The prcsence or absence of secondaq stress in the language is a separate issue.

In the nest two sections, an analysis of the stress systern of Cairene Arabic and that

of Palesthian Arabic will be presented based on the proposais introduced in this section. It

will be shown that the stress sytems of these two dialects share many similarities with that

of Classical Arabic, as would be eapected from different dialects of the same language.

3.2.2 Cairene Arabic

Consider f int the stress system of colloquid Ambic as spoken in Cairo. The data are taken

from Hayes ( 1995) and include f m s that are defined as 'Cairene Classical Arabic'. As in

the other Arabic dialecis, syllables in Cairene include light Cv, heavy CvC and Cvv, and

superhcavy CvCC and CwC that are restricted to final position.

The stress pattern of Cairene Arabic as descri bed in iis f in t formal metrical account

by McCarihy ( 1979) is as follows: stress a final superheavy syllable o r a final Cvv in non-

classical f ~ r m s : ~

'1 wrote' 'pilgrimmages' Cl.

'cake'

O t h e n t k stress the penult, provided it is heavy:

'your (m-sg.) house'

'you (msg.) wrote' CI.

'teacher' 'these (m.dua1)' Cl.

Othen~rise, stress the penult o r the antepenult, whichever is separateci by an even number of

syllables from the closest preceding heavy syllable, o r if there is no such syllable from the

beginning of the word.

'he u n d e r s t d ' 'tree (nom.)' Cl.

According to Hune (lm). from a personal communication by McCarthy, final stressed vowels as in (12c) are most Iikel y superheavy syiiables with a final -h.

c. katabitu d. 5ajaraiuhfimaa5

c. qattala

f. mud-sit g. 3adwiyattihu

'she wrote it (m.)' 'their (dual) tree (nom.)' CI.

'he klled' CI.

'teacher (f .constnict)' 'his dmgs (nom.)' CI.

'mothert

'his dmgs (nom.)' Cl.

'he nTrote' CI - 'tree (clause-final)'

'his tree (nom-)' Cl.

As shown above, the position of hewy syllables within a word p l a y an important role in

dctcrmining main stress. In contrast with Classical Arabic, however, main stress is never

found farther left than the antepenult. 1 propose that the first and most crucial difference

behveen the tu.0 dialects cornes from the domination of AUGX-PRST-Lm by Am-PRWP

RIGKT in Cairene Arabic.

26. AL~CN- PR WD- RIGHT Align (PrWd, Right, PrStem, Right)

"The right edp of every R W d wincides with the right edge of some PrStem."

The second difference between the two Anbic dialects wmes from the di fferent argument

chosen by the nonfinality constraini. That is, in Cairene NO SIS^ requires for the PrWd

hcad syllable to be non-final, rather that for the head f m t to be non-final.

27, NONFINALHIJ No PrWd is right-aligned with its head syllable.

Thesc proposais are illustnted in the tableau below. Nm'~c~;..u.Ho is left out of the tableau

since it is never violated in the optimal outputs. The form in the tableau below is from

Crowhurst ( 1996).

The f o m in botb (216) and (25b) from btitchell(1960) appear orith a long vowel word-fïdly- However, in Classical words, final Cvv is always oounted as lipht. In fact. as noted by Hayes (1995). many are doubtful about the s t a t u of such syllables (Harreii 1%0; Harm 198 1).

Note that undominated R I G ~ ~ O S T ensures that the final M t e m resulting from satisfying

AUGS-PRWDRK~H~ is always parsed as the head of the PrWd. Whether the head foot is

linal or not, however, is determined in part by the location of ka\- syllables within the

\fiord-

In forms ending with three light syllables,, fmt distribution becomes crucial for the

positioning of the main stress f w t at the right edg, as shown in the two tableaus below.

ALL-FT-

h~

ttttt*

As shown in (29b), the choice of the optimal output lies squarely on the position of the

second foot within the PrWd.

AUGN-

PRST-LEFI-

*

t

Candidates

mabsatuhunna v

!w Ci P C r w C1

0 . 4 [(mad). ra] .(sa. tu). [(hun). na]

Ar--Fr-

IAT

*

**!

!'!' ! ' P C C w Ci

b. [(mad)].[(ra.sa).tu].[(hun).na]

AUGX-

PRWD-RGHT

Candidates

rir~kasara

u!' I r ! ' ! '

a. 21 [(ag)].[(lra.sa).ra]

!'Ci P !'

b. [(Cir~).ka].[(sa.ra)]

-

W n m -

TO-HEAD

**!

WHGW-TD

H(PRST)

ALIGS-

PRWPRIGHT

ta***

ALJGS-

PRST-LEFT

*

t

In (30a), the optimal candidate has its second foot tnro syllables away from the left edge

\.iolating AU-Fr-Lm hvice. In (30b), there are three syllables between the second f w t and

&-Fr-

LEZT

** I

***!

***!*

the left edge of the PrWd. In (3&), the first foot is one syllable away from the left edge,

P.ms~-a

whilc the second foot is three syllablcs away from the lcft edge, resdting in four violations

Aum-

PRST-Lm

*

*

*

Candidates

Sa jaratuhu 1

C I P C P P a. 4 [(Sa. ja)]. [(d-tu). hu]

P PCI C P b. [(Sa. ja).ra]. [( tu hu)]

P P P ci!' c. [:a.(ja.ra)].[(tu.hu)]

The elimination of the non-optimal candidates in (30b-c) has been shown to follow

~ G S -

PRWM'~GHT

from thc nurnber of i.iolations that apply to Au-FT-Lm. This proposal, however, becomes

problematic when we consider forrns with an initial bimoraic foot. This is illustrated in the

tableau below.

Candidates

3 adwiyatuhu 1

Plr P P Clr a. [(?ad)].(wi.ya).[(tu.hu)]

P C C r P b. I/ [(?ad). wi]. [(yd-tu). hu]

PP P CL P l ' c. [(?ad).wi]. [ya.(tu.hu)j

~ G X -

PRWD-RIGHI-

WHGW-

-Hm

AUGS-

PRST-LM

t

*

t

AL-Fr-

b = r I

***!*

*X

#

***!

if we sirnply assume that AL-FT-Lm is a gradiently vidable constraint which &mands

perfect alipment with a d e s i p t e d edge, then according to tableau (31)- the candidate

output in (3 1 b) should be the optimal choice. The correct output, however, is (3 la). What

differentiates the candidate in (3 la) from those in (3 1b-c) is perfect adjacency behveen feet

(a b i n q pattern), compared with the temary pattern urhich characterizes (3 lb).

In metrical theory, fmt distribution reflects two different processes: directiondit?;

and iterativiy. Within Optimality Theoy, the standard assumption has been that Au-Fr-

reflects both processes through the fact that its violations are gradient rather than

categorical. Any theory of stress systems that allows for te- constinients, however,

faces a problem sirnilar to that found in tableau (3 1) in that maximally temary constituents

tend in some cases to reduœ the number of AU-Fr-Lm vidations in a word. This is

i I l ustnted below.

32 a [ ( a a ) a j [ ( a u ) a j b- [(a ab m ( 0 4 1

In (32a), foot parsing results in o d y three violations of AU-FC-Lm, while (32b) results in

sis violations. Foot parsing in the temary pattern in (33a) results in nine violations, while

the binary pattern in (33b) results in esactly twelve violations.

Adopting the position taken in Kager ( l9%b), 1 propose that Au-m-Lm should be

subdividcd into AU-ET~-LEFT >> Au-FT2-Lm » Au-FT~-LEF~ etc. The crucial aspect

of this subdi\isîon is that AL-Fr-Lm evaluates the distance of each foot to the left edge on

a foot-bu-foot basis. This means that, irrespective of the number of feet in a PrWd, a

violation of Au-Frl-LET is always more costly than any number of violations of AU-FT~-

m, while any violation of A u - m - L m is counted as worse than hvo or more violations

of subsequent feet. Similarl y, AL-Fr-Ricm evaluates the distance between cach fmt and

the right edge of a PrWd, Le. A u - F r l - R I ~ m is the nghtmost foot.

0bsen.e that b t h (34b) and (34c) have more thari one violation of A L - F T ~ - L ~ , while in

Candidates

3 adwi yatuhu

PP PCC P P

a. d[(?ad)].(wi.ya).[(tu.hu)]

P P C L C C P ! '

b. [(?ad).wij.[(yA-tu).hu]

PCC P P P ! '

C. [(?ad). wi]. ba( tu .hu) J

( 34) A L - F T ~ - L ~ is violated only once. So, even though (34b) has a lesser number of

~.ioIations for ALL-fi--, i t is still evaluated as less optimal than (344.

ALIGS-

PRWDRIGHT

3.2.3 Palestinian Arabic

Consider now the stress pattern of Palestinian Arabic which is illustrated belou., The data

W n m -

DHW

arc from Hayes ( 1995).

*

d

*

First, stress a final superheavy syliable.

J1 *? * f * 3

J1 **!?

JI **!*?

AUGS-

PRST-Lm

35. Final Superheu- Syllables

a. daraist

b. baabéen (--w babéen)

c. dukk5an

d. mooladeen

Ar-FT-

UFI-

'1 studied'

'two doors'

's hop '

'two feasts'

Othewise stress the penult, provided it is k v y :

'he blesscd'

'office' 'bribe (3 m-sg. imperf)'

'we wrote'

'our offices'

'our feast'

'OUT COW'

'she blessed us'

In words ending with two light syllables the stress pattern is as follow: in disyllabic or

trisyliabic words, stress the initial syllable.

37. Disyllabies and Trisyllab les a- ?Ana

b, katab

c. kiitabu

' 1 '

'he wrote'

'they wrote'

Othewise stress the antepenuit, pmvided it is heavy:

38. Heu- Antepenultimate Syllable

a baarako 'he blessed him' b. Tallamat 'she taught'

c. ? id la îu 'pay (pl. imper.)'

d. makhtibi (--> makaatbi) 'mu offices'

Othentisc, stress the preantepenultimate in four syllable words without heavy syllables or

stress the antepenul tirnate.

3 9 . Preantepenu ltirnate Sy 1 fable

a. b M t o (--> bakarto) 'his cow'

b. darabato (--> dart>ato) 'shc hit him' c. kijaranin 'a tree' Cl.

40. Antepenultimaîe Syllable a TdlAmato 'she taught him'

b. m a k ~ b i t o (--> mak-tabto) 'his l i b r q ' c. Sa jd tuhu 'his tree' CI.

Obsenre first that the head foot is only finai in disyllabic words in Palestinian Arabic. In

this rcgard, Palestinian Arabic differs from Cairene but is similar to Classical Arabic. The

fact that main stress is never initial in words longer than four syllables, on the other hand,

strongly suggests that AUGK-PRWDRIGHT is high-ranking in Palestinian Arabic. These

p ropods are illustrated in the following tableaux.

As noted before, Palestinian Arabic diffcrs from Cairene in that the head foot cannot be

Candidates

Sa jaratuhu

I r P P P a. I/ [(Ça. ja)]. [ (d tu ) . hu]

Cr P C ! '

b. [(Sa. ja).ra].[(tu. hu)]

I ' F P P!' c. [(SA. ja).ra].(tu.hu)

final in a PrWd. This however does not explain why main stress is initial in words which

contain four light syllables. Here 1 propose that the preantepenultirnate stress follows from

NOMTSAL-

HFT

*!

satisfying ALIGS-PRWDRGHT and AUG:-PRST-LE~. This is shown in the tableau below.

Aue';-

PRWDRIGW

* !

In (421)~ 1 assume that the PTStem parses turo feet in order to satisfy b t h AUGS-PRWD-

Candidates

Sa jaratun

CI Ci C i C i

a. [(SA. ja).(ra.tun)]

! ' P U

b. Sa.[(j&ra).tun]

P C i P

c. [(Si. ja).ra] .tun

RIGI-rr and AUGX-PRST-Lm. Such a tua-footed structure, however, violates the constra.int

ALIGK-

PRST-Lm

*

It

F T - ~ P R S ~ , which rcquires ewry foot parsed by a PrStem to be its head.

Au-FT-

Lm l

dl **2

dl ***?

dl

N o ~ ~ L -

HF?-

43. FT-TO- PRSTEM Every foot dominated by a PrStem is parsed as i i s head.

AUGX-

PRWBRIGM~-

*!

AUGS-

PRST-Lm

*!

ALL-FT-

LEZT

4 **2

$1

Y/ i

The role played by FT-TO-PRS~I is in evidence in amther dialect, Egyptian Radio Arabic

\vhich differs from the other diaiects in that al1 feet are stressed. In other words, STRESS-rn

H(FT) is undominated in Emptian Radio Arabic. Interestin&, in this particular diaiect

discussed in Hmell ( 1960), words ending in four light syllabies have the position of main

stress \.ar)i ng freel y between the pend timate and the p-eantepenultimate syIlaMe:

[katabahu] vs. [katabàhu] .

Even more interesting is the free variation found in trisyllabic words with an initial

44. Free Vmation between the Penult and Antepenultimate Syllable a. h&ai.hi h à a a i.hi 'this'

b. rnuS.ki.la mùS. ki. la 'problem '

Strikingly, there is no secondary stress on the penult when main stress is on the antepenult.

Gi\.en that seconClary stress is found on the penult in qdr i sy l lab ic forms, the question

here is what is particdar to quadrisyIlabic forms that the trisyllabic forms in (44) do not

share.

The solution 1 propose lies in assuming that F T - - D P R S ~ I is in fact high-radine in

Eg5ptian Radio Arabic (and Palestinian Arabic). The question then is what would force the

i-iolation of F ~ - r n - M m f in quadrisyllabic forms but not in tnsyllabic forms. In his

andusis of ternary systems, Kager ( 1996b) proposes the constraint PA ME-^ Lo restrict

sequences of unparsed syllables.

45. P A R S E - ~ One of ovo adjacent stress units must be parsed by a foot.

This constraint, however, wouid piuse both quadrisyllabic forms and triqllabic fonns as

ci ther two feet uniquely parsed by a PrStem: [(ka ta)]. [(M. hu)] and [(hh)]. [(a i. hi)], o r as

a t1c.o-footed PrStem: [(kàta).(M.hu)] and * [(hh).(ai.hi)]. Here 1 propose a different type

of constraint, W W . ~ E C ~ D , which 1 assume is undominated ~n iversa i ly .~

This proposai is based oa Rince 's (1980) idea that no prosodic domain ~may end in more than two weak syllables.

46. WEAKPARSECOND A mora in a non-head syllable is adjacent to at mmt one mora in a non- head syl lable.

This constraint requires that a mora not parsed by the head of a f m t should be aâjacent to at

most one mora that is not the head of a foot. The d e played by the WWARSECCSD in

eliminating the unacceptable form * [(kata).ba]. hu in contmst wîth [(h&).l il. hi îs s h o w

Candidates

katabahu P !' )r P

a.d[(kà.ta) J. [(ba hu)]

Ci C I ! ' ! '

b . [(k&ta).ba].hu

C i ! ' I L ! '

c. [(kAta).(bà.hu)]

In the analysis of variability of truncated names in Japanese, 1 proposed that variability in

the choice of an optimal output follows from the variable rankings of two conflicting

constraints wiihin the constraint hierarchy. Here 1 propose that the variability in the choice

of the optimal output in Egyptian Radio Arabic follows from the variable rankings o f two

WWPARSE

C ~ D

*!

A r r ~ s -

PRWD-R

*!

WGS-

PRST-Lm

*

Candidates

haaa ihi L

PP P IL a. ./ [(hàa)].[(a i h ) J

w I L I L b. [(hiia).a i1.h

w P P

c. [(h&).(ai. hi)]

FT-TO-

PRSTEM

*!

P ~ s E - a

r(:

WE\KP.L\RSE

C m

Aum-

PRWDR

*

FT-TO-

P R S T ~

*!

AUGS-

PRST-Lm

*

PXRSE-O

X

1

-

constraints: NQ~SALHFT and FT-TO-PRSTEM, within the hierarchy. The crucial rankings

are i 1 lustrated in the hierarchical structure belo~'.~

This is illustrated in the tsvo tableaux bclow.

Candidates 1

kata bahu C I r P

a. d [ (kata) . (bah~)] P P ! ' I r

b. [(ka.ta).ba].hu Ii Ir P P

c. [(kà.ta)].[(ba.hu)l

Candidates II

For a morc detailcd discussion of a stress system mith variable stress patterns. see the analysis of the optionall y b i n q and te- stress patterns of Estoniiui in chapter 5.

Candidates

haaB ihi

Candidates II

In conclusion, the different stress systems of a number of Arabic dialects have been

W W A R S E

C m

accounted for by assuming a modcl of prosodic constituency where al1 metrical constituents

are i ntnnsicall y associated ~ i t h a head but where heads d o not necessari1 y project a grid

mark on the metrical grid. The main advantag of this mode1 with regard to the stress

systems of Arabic is chat metrical constituency is not dependent on prominence on the grid.

This means that the distribution of unstressed metrical constituents and the position of their

heads within the PrWd c m influence the position of main stress without having to posit

defooting rules o r reparsing niles, o r in the case of optimality-theoretic analyses, without

having to assume headless constituents. Furthemore, the proposal that not al1 metrical

FT-'10-

PRSml

*!

ALJGS-

PRWSRIGH~

*!

No';n.ï.u,

Hm

t

h d s need obligatorily projet a grid mark on the metrical grid allows us to eliminate Line

Conflation as a repair strategy.

Another advantage of this mode1 comcs from the introduction of the PrStem in the

Prosodic Werarchy- Wi thin the revised P r o d i c Hierarchy, the only feet that are possible

heads for the PrWd are those parsed as the head of a PrStem. As a result, every foot

directly parsed by the PrWd will be ipored by the consuaint ~ G H ~ ~ o s ~ . ~

In the analysis presented above. the parsing of final superfieavy syilabies as illuslrated in (35) is aot discusscd. Here 1 assume that these forms d in an mpty-headcd syiiable resulting from a çonstraint which prohi bi ts branc hing of a mora in a nonitead p s i tim (cf. \frai ker ( 1941); Sprouse ( 19%)).

Chapter 5

Binary and Ternary Stress Systems

In the study of word minimalis presented in chapter 2 and chapter 3, both disyllabic and

trisyllabic constituents have been argued to result from the satisfaction of a nurnber of high-

&ng structural constraints that are minimally violated in the general phonology of the

language. In this chapter, 1 intcnd to show the central role played by these same structural

constraints in the g e n e d phonology of three v e s different languages. First, 1 analqze the

binary system of Guugu Yimidhirr. Then, the t e m q system of C;.uvava is examined.

Finally, 1 present an andysis of the variably b i n q and t e r n q stress patterns found in

Estonian.

5.1 The Binary Stress System of Guugu Yirnidhirr

Guugu Yimidhirr, an Australian language spoken in Queensland (Haviland 1979)' presents

an intcresting problem in that both morphology and phonology regularly refer to an initial

prosodic domain that is precisely t\vo syllables, irrespective of syllable weight. In contrast

ni th the morphologicall y-de fi ned word formation processes of Japanesc and Arabic, there

is no distinct rnorpholo@cal domain upon which a minimally disyllabic requirement can be

irnposed in Guugu Y imidhirr.

1 The anal ysis of G u u p Yimidhirr presented in fhis section is based eniireiy on tbe data and insights found in b g e r ( 199th). Aithougb Kager's analysis is also within the framework of Optimality Theory. many of the conciusions reached hem difia markd y from those prescated in his paper (see 55.1.2).

5.1.1 The Initial Disyllabic Domain

Guupu Y imidhirr is a language with a contrast between short and long vowels. Both types

of vowels are found in closed syllables, but short-vowelled closed syllables pattern with

s hort-vouVelled open syllables wï th regard to the stress system. Moreover, on1 y coronal

sonorants c m be parsed as a syllable coda, and on1 y @ides and l iquids can be final in a

long-~owelled syllable.' Both oblipatory onsets and optional codas are non-complex.

Main stress falls within the word-initial disyllabic domain. In addition, stress falls

on avenp heavy syllable withn this domain. In words that begn with a main stressed

syllaMe, the remaining odd-numbered syllaMes have secondary stress. In words wi th main

stress on the second syllable, the remaining even-numbercd syllables receive seconchy

stress. Interestingly, whenever stress falls on two heavy syllables within the initial domain,

the s e c o n d q stress pattern starts on the next syllable. in addition, secondary stressed

syllables can be found as the head of what look like a unaq f6ot at the right edge of the

PrWd (e.g. [mArbugànbigu] 'still in the cave').

1. Stress Panems in Guugu Yirnidhirr (Kager 1996a)

#LL naambal 'stone-ABS' mar. bu-gàn 'cave-.4t3S'

dur.bin.bi.gu '1 ndian Head (place narne)'

W. bu.gàn.bi.gù 'cave-LX-E‘rPHf ('still in the cave')

#HL guu.gu b u u - s a yày

buu-sa. yhy.gu

q & ~ ba.i)àl.galà

'Ianguage-.-S' 'water-LOCI ('in the water')

'water-LOC-WH' ('still in the waterf)

'ask-RED4 M F ('keep aski ng')

'branch-i\Bs ' 'stone-.uL' ('f rom the stone')

'branch-PL-EMPH' ('just branches')

#HH buu.r&y 'water-ABs'

buu.r&y.bi.gu 'water-LOC-WH' ('sti Il in the waterf)

Therc are a few mots wbich end in a long-vowelled syllable closed by a nasal: e.g. burutun 'one'.

104

The stress pattern of Guugu Yimidhirr reveais hiPo important characteristics of the

Ianguage. The first one is the initial disyllabic domain. The second one is the !anpage's

sensitivity to weight. That is, although the initial disyllabic domain may consist of both

hcavy and light syllables, a heavy syllable must still be parsed as the head of a foot. In fact,

according to Haviland (I979), in words that begin \vith a heavy-heavy sequence, both

syl iables receive equal prominence.

Further etidence for the initial disylfabic domain in the language m e s [rom the fact

that, al though Guugu Y imidhi rr has a contrast between shon and long vowels, the latter are

restricted to the first two syllables.

3. The Distribution of vowel length (Kager 1996a) a. miil 'eye' e. daw6ac 'star'

b. \vhQa ' crowr f. gambuugu 'head'

c. wiiacigàn 'moon' g. bulbilucmbul 'pheasant'

d. guu[umùgu 'meat ha&' h. daaraali~h 'kangaroo'

There are, however, a few monosy llabic words in Guugu Y irnidhirr.

3. 4 nouns: dii ' t e . (ban) , miil 'eye', Quul 'guts', buur 'bird's nest'

4 particles: paa 'that, there' (root /@), yii 'this, there' (mot /yi/), yuu 'ves', aa (agreement).

Although these f o m s are rare, the lengthening of the function words (urhen in isolation)

strongly suggests that word minimality in Guugu Yimidhirr can be satisfied by a birnoraic

foot.

5.1.2 Disyllabicity in Optimality Theory

In M c M h y and Prince (1993a), the disyllabic shape of the redupliçant in Asininca Campa

is argued to resul t from the satisfaction of a constraint requiring that the left and nght edges

of some morphological domain be alipned with the eâges of two different syllables, DWL.

Kagcr (1996a) in tum proposes that in order to account for the initial disyllabic constituent

of Guugu Yirnidhir the domain of application of Disiu. should be extended to allow for the

alignement of edges of t\vo prosodic consti tuents.

4. DISYLL The left and right edges of the RWd must coincide, respectivelu, with the left and right edges of diflerenr syliables.

DM-LL, however, cm only impose some minimality requirement; i t cannot restrict prosodic

constituent. to a maimally disyllabic shape. Kager must assume that a phonological word

of more than two syllables is parsed as a recursive W d . The recursive structure is q u e d

by Kager to result from the high-ranking of Au~s-a in the language.

5. ALIGN-a The right edge of e\:ery syllable coincides with the right edge of some PrWd.

The ianlung D t n u >> AUGS-a ensures that the latter constraint can ncver be fu11y satisfied

in an optimal output Violation of AUGN-a can be minimized, however, as long as RWds

are allowed to be recursive.

6. [ [a a] a.. . J

The prediction then is that these recursive structures should result in multiple violations of

PARSE-o. K a g r , however, argues that the bin- stress panem of the language reflecîs the

dominance of ALIGS-O by FT-BK and PARSE-O, resulting in a double violation of AUGX-a

for cach recursive M d . In addition, w e r assumes that the hi&-ran)ring of Alip-Wd:

Align (PrWd, Left, Stem, Left) accounts for the word-initial position of the Head-PrWd.

Following Kiparsky ( 1991) and Kager ( 19931, Kager further assumes that final unary feet

in f o m s like [mA~bugànbigùj and [buurayhy] reflect the presence of an empty syllable at

the end of the word, with which a b i w foot can be formed. Catalexis in Guugu Yimidhirr

follotvs from the ranking Fr-Brs, Parse-a » A m - R I G W (Align(LexWd, Right, PrWd,

Right)). Kager's proposal is shown in the following tableau. (The square brackets indicate

PrWd boundaries.)

Candidates

buurayay

[(buu).(ri yay)]

[[(buu). sa] .yay]

Note that in order to get the correct result, the constraint AUGX-a must be assumed to allow

for endient violations. In (&), each misaligned syllable is one syllable away from the nght

edge of some R W d In (gb), the leftmost sylllable is two syllables away from the e d p ,

resulting in a third violation for A~Gs-o. If AUGS-u violations were counted catego~cally,

(a) and (8b) would have two violations each, Le. they each have nvo syllables that are not

right-aligned with a PrWd edge.

9. Fr-BIS ; DISYLL » PARSE-a » AUGS-a

As ~~~~~~n in (9b). prosodic consti tuents larger than two syllables create an unnecessary and

fatal violation of AUGN-a.

Candidates

buurayaygu

a v' [[(buu.ra)].(yày.gu)]

b. [[(buu).(rà. yay) J .(gù.a)]

c. [[[(buu)]. ( rà. yay).(gù.a)]

d. [[(buu).ra]-(yày-gu)]

i FT-BK [ nu. 1 PARSE-a

1 I i !

i 1

*! I

*!

AUGS-a

xx

***!*

**

IC*

AIthough Kager dismisses the idea of an intermediate prosodic unit between the f w t

and the PrWd, he achowledges that there seems to be such a coherent domain of reference

in the m m a r of Guugu Yimidhirr. This domain is that of the most embedded RWd,

tvhich he refers to as the Head-PrWd.3

First, the Head-PrWd is the on1 y domain for main stress. Second, vowel length is

only found within that initial domain. Furthemore, suffis-induced vowel lengthening and

shortening only affect the second syllable of the Head-RWd.

10. Lengthening suffies

a. /gaigai/

b. /mifia/

c. /wulur~gur/

~ a l ~ l - p ' s m o ke-PL-RP' mipaa-QU 'meat-P1.W"

wului~gur-gu 'lightning, flame-PLW' (* rvul uqpuur-r)u)

1 1. Shortening su#ikes: a. Igabïisl gabi S-IF

b. /nubuun/ nubun-il

c. /gaapal/ ga@ -ay

(*gqd-ay)

Kager relates the attraction of main stress by heavy syllables to the undominated constra.int

Weight-to-Stress Principle ( WSP}. He further assumes that in a doubl y stressed sequence

[E-IH], both feet must stand in the Head-PrWd because of undominated HdWd-

12. WSP Havy syllables have ma~imal prominence.

13. HDWD The head of the PrWd is the innermost PrWd,

Note, howcver, that the latter constm.int is i n t r o c i d specifically to handle the position of

main stress in Guugu Yimidhirr. There is no mention by Kager of any other language that

would j u s ti fy the postulation of HDWD as a universal well -formedness rnnstra.int.

Wer S t a t e s in a faorwie tbat ihere does not seem to be enough evidenœ cross-linguisticaily for cxtcnding the Prosodic Hierarchy. However, Kager does wondcr, at some point in the analysis. as to whether thc tcrm 'Prosodic Stem' would not be more -te thm Head-RUrd.

Kager accounts for the secondary stress patterns by assuming two high-ranked

constraints: R~wE.=T which favors trochaic default fect, and HL\R\KN which demands

that prominence of secondarq. stressed feet (trochaic o r iam bic) hannonizes with that of the

Head-PrWd. That is, within the Head-PrWd, trochaic feet are optimal uniess it leads to a

violation of WSP. Secondary stressed feet ta);e the same R k i r i ~ as the Head-PrWd, to

satisfy hi&-ranking. HUWO?X.

Al though Kager's analysis accounts for every aspect of the binary stress system of

Guugu Y imidhirr, it fias very linle espianatoy power with respect to language variability.

Onc centrai aspect of Optimality Theay is chat variability resutts from the reranking of

universal constraints. The w&ess of Kager's anaIysis is that almost every aspect of the

b i n q pattern that characterizes the stress system of Guugu Y imidhirr is accountcd for by

some new constra.int. In the next sections, I propose an analysis of the b i q system of

Guugu Yimidhirr which is predicated on the notion that variability results from language-

s peci fic re ranki ngs of uni versal constraints.

S. 1.3 The PrStem and Initial Disyllabicity

The analysis presented here reformulates Kager's insights in terms of the mode1 of prosodic

structure developed in this dissertation. For Kager, the position of the Head-PrWd resd ts

from a specific headedness requirement imposed by the constraint HDWD on recursive

PrWds. In my analysis, the initial position of the head of the PrWd follows directly from

assuming that A u c ~ - P R S r n - L r n is undominated in the languagc, with the righnvard

di recîional J t). effects resulling from hi@-ranking AL-FT-Lm rather than from Ams-a.

14. ALIGN- PRSTEM-LEFT Align ( M t e m , Lefi, PrWd, Left)

"The left edge of every PrStem coincides with the left edge of some PrWd. "

The disyllabicity of the initial PrStern is accounted for by assuming thai BRA-KCHHEXD is

aiso undominated in the language.

15. BRANCHHEAD The head of a PrStem is in a rh>.thmic relation with a non-head constituent at some level of andysis. (a, Ft).

Ncst consider the position of hem? syllables within a word. Kager's WSP is refonnulated

in terrns of the constraint WE~GK~-~~-H(PRSTE%). Recall that in chapter 4, this constraint

\\.as introduced to explain why main stress is preferably assigned to a non-final heavy

s>.llable, irrespective of its position within the word in the Classical Arabic stress ?stem.

16. WEIGHT-TO-H(PRSTEM) A bimoraic syllaMe is parsed as the head of a PrStern.

The restriction of h w y syllables to the head of a R W d in tum follows directly from the

assurnption that A u ~ x - P ~ S r n - L r n is undominated in the pmmar. The prediction is that

in any language where WEIGHT-~H(PRSTEM) and AUGS-PEXSEM-LEF~ are undominated,

hea? syliables will be restricted t o the main-stressed foot. These proposais are illustrated

in the tableau below with the form buurdüybigu 'still in the w7aterf.

Satisfaction of WEEKI=~-H(PRSTEM) demands that e v e y heavy syllabie be parsed as the

Candidates

buuraaybigu

a v' [(b~u).(rAay)].(bi.gu)

b. [(buu)].(ràay) .(bi.gu)

c. [(buu).(rhy)].[(bi.gu)]

head foot of a PrStem. The location of main stress on both bimoraïc syllables in the initial

PsStem indicates that MOSOHE~EDESS must be dominated by WEIGH~--DH(PRS~!) .

i AUGJI- j BRLKH- W ~ i ~ r - r r - ~

PRST-Lm 1 H u o ! H(PRSTELL) I

1 I

IC

*!

Consider also rk fact that head syllables receive secondary stress outside of the

PARSE-O

,

l

initial PrS tem. T hese second. stresses show that STRESS-TO-H(R) is hi gh-ranking in the

ianguage, with satisfaction of STRESS-mH(FT) resulting in eveq foot head projecting a

gnd mark on level 1, whether they are parsed by a PrStem or not. In addition, domination

of Mosmwmmss by Wmw-m-H(PRSmf) ensures that every head foot in a PrStern

will project a gnd mark on level2, and then on level3.

18a x x line 3 H(PrWd) * x line 2 H(PrStem) (*) (*) (* -) line 1 WFt)

In a nVord beginning with two heavy syllables, the mauimally disyllabic shape of the initial

PrStern fotlows from fully satisfying both AUGS-PRST-Lm and Wncm--ruH(P~Smi ) .

In forms with on1 y one initial h e a t ~ syllable, however, the disyllabic shape of the PrStem

follo\c.s from satisfying BRA.XCHHE~. This is illustrated below mith the trisyllabic form

As shown by the output form in ( Mc), 1 adopt Kager's proposal that final unary feet are

parsed as monomoraic disyllabic feet (ap a), violating N o H ~ w m S i i ~ . This proposai is

Candidates

buurayay

a- 4 [(b@.ra].(yày.a)

b. [(buu)].(ràyay)

c. [(buu)].[(rà.yay)]

d. [(buu).ra].yay

discusscd in more detail in the nest section.

Consider now trisyllabic forrns without heavy syllables like [wulu~gur ] lightning,

flarne'. BR~KHHFAD only requires that the head syliable or the head f w t of a PrStem be in

ALJGS- i BR-IYCH W -

PRST-LEFT H m H ( P R S m )

*!

*! *!

P-UGE-G

I

*!

a rhj-thmic relation with a syllable in a non-head position. MaUmally disyllabic domains in

longer words ivould have to follow from some other constm.int to mle out candidates like

*[(wi.lu).!Igur] and *[(buu).(rByay)]. To account for the m a ~ m d l y disyllabic PrStem. 1

propose that strict diqllabicity resuits from the interaction of BRXXCHHEW with two hiph-

ranhng wnstraints: AUGS-PRWDRIGHI. which rquires ihat a RWd end in a bot, and hi

TO-PRSTEM which requires a f m t parsed by a PrStem to be its head.

20. FT-TOI PRS TEM Every foot dominated by a PrStern is parsed as i ts head.

2 1. ALIGN- PR WD- RIGHT Align (PrWd, Right, Ft, Right)

"The right edge of every RWd coincides with the ri@ edge d =me foot."

Note that in a two-footed structure, there is a conflict between MCX'OHWDEDSESS and Fr--

P~Srn l . Here 1 assume that the crucial ranking aith respect to such structures is as follow:

W n ~ w - m - H ( P ~ S r n f ) » MON-- » FT-TO-PRS~I. The foie @a@ by these

constraints and by Am-PRWIFRIGHT is illustrated in the hvo tableaus below.

Candidates

wulu9gur

a I/ [(wU.lu)].(!Igur.a)

In (22bc) , both f e r are parsed by the PrStem. Satisfaction of M O S O H E ~ ~ ' E S S in (23b)

rcsults in a violation of FT-TO-PIXSTEM, while in (2%) satisfaction of FT-TO-PRSm create~

a fatal \-iolation of MONOHEADEDWS. The candidate in (22d) violates AUGN-PRWDRIGKT.

Candidates B~1'r;u-i- WUGHT-KF Mom- FT-m-

buu r ayay H w i H(PRSm) 1 PRSTEM I

As show in (23a), the optimal candidate satisfies the high-ranking strucniral constrainfs of

the language, with the exception of A u - F T - m .

In this mode1 then, the restriction of main stress and of heavy syllables to the initial

disyllabic constituent follows from satisfying the two undominated constraints WEIGHT-TI)

H(PRSTR~) and AUGX-PRST-Lm. The strictly disyllabic shape of the initial constituent is

the result of satisf>ing BRQXCHH~D and MOS~H~DEDSESS without violating the lower-

ran ked stmct ural constraints FT-'10- PRSTEM and AUGN- PR W BRIGHT'.

5.1.4 lambic Feet in a Trochaic System

An important aspect of the stress system of Guugu Y imidhirr is that it allows for either an

iambic or a trochaic pattern depending on the prosodic shape of PrWds. More precisely. the

rhythmic pattern of any RWd is dependent on the presence of heavy syllables within the

initial disyllabic domain. Whenever this initial domain is left-headed, subsequent feet are

also left-headed. When the initial prosodic domain is ri@-headed, al1 following feet are

right-headed. Following Kager (19%a), 1 assume that feet are trochaic, unless there is a

conflict with the hcadedness of the initial p r d i c domain.

Before discussing in more defails my anaiysis of Guugu Yimidhirr's trochaic

system, consider how metrical analyses have dealt with the problem of a language with

variable fmt-headedness within its stress system. T o my howledge, there is no recent

analysis of Guugu Y imidhirr within the metrical framewwk. However, the stress system of

Yidip, a closely related langage, presents a similar problern which has been eatensiïely

studied. The Y i d i ~ l stress system is described by Halle and Vergnaud ( 1987) as f011ows.~

14. Y idij: Stress fdls on even-numbered syllables, if the word contains an even- nurnbered syllaMe with a long vowel; othenvise, stress falls on odd- nwnbered syllables. (Hayes 198 1 ; Dixon 1977)

Gi \-en the variabili ty of foot- heads in Y idip, Halle and Vergnaud propose that constituencp

on the metrical grid be implemented on two different planes simultaneously. On Plane 1.

the line O constituents are right-tieaded; on Rane 2, the' are left-headed. The inappropriate

plane is deIeted by a subsequent de: whenever there is a constituent head dominating a

long \.oarel on P l , PI is deleted; otherwise, Pl is deleted Crucially, in this model,

hadedness is dependent on consti tuency, but metrical consti tuents i nherentl y have a head.

This proposal, however, cannot be readily extended to the Guugu Y irnidhirr stress systern,

most particularly when the initial prosodic domain coniains two h e a y syllables. According

io Halle and Vergnaud's formulation. given that in an initial disyllabic domain containiop

hvo monosyllabic feet there wilI be both a constituent head dominating a Iong v o w l on P 1

and a wnsti tuent head dominating a long vo\vel on P?. One would then predict an iam bic

pattern, rather than the trochaic one found in these forms. An additional mlc therefore is

necded to ensure that whenever the= are two potential heads within the initial diqilabic

domain, it is P l that deletes. That move, howekrer, would require allowing for (HH)

consti tuents.

A different proposai by Crowhurst and Heuitt (1995) relies instead on the presence

of headless feet throughout most of the metrical derivation. Following Hammond (1989),

Halle ( 19901, and Halle and Vergnaud ( 1987). they assume that metrical groupings and the

assignment of stress-bearing heads to those metrical constituents are independent steps in a

derivation. Crowhurst and Hewitt (1995) argue that there is a laver of Ceet assigned early in

S e c Hung ( 1994) for an aoaiysis of YidiJi within the Optimaliiy framework.

114

the dcrivation which is unheaded and insensitive to weight distinctions and which p l a y a

role in the morphology. The assignment of metricai heads, on the other hand, resul t from

quantity-sensitive d e s which apply within the pst-lexical clitic domain.

35 a Quanrity-Sensitive Head Assignnlent Within a PrWd, assign a head to the syllable mntaining the rightmost long V.

b. YidinJ Initiai Head Assignmenr (defauit rule) Within a RWd, assign a head to the initial syllable.

A g i n . an initial prosodic domain containine two monosyllabic feet is problematic for the

proposeci analysis, since both the rightmost heavy syllabIe and the initial syilable receive

main stress in these fonns.

These two metrical analyses cnicially rely on the fact that in Yidi~, undedying heavy

syllables are always found in an even-numùered p i t i o n . This is obviously not the case in

Guugu Yimidhirr and that is where both proposais fail to account for its stress system.

Consider now Kagerls (1996a) analpis of the stress system of Guugu Yimidhirr in

an optimality-theoretic framework Under OT, formulations of the type: "do this, except

when" a n treated as resul ting from satisfying some structural consuaint at the expense of a

loiver-ranked one. Thus, for Kager, bot-headedness within the Head-RWd follows fmm

h c Weight-&Stress Principle dominating F T - F o R V ~ R ~ ~ ~ C ] . Within the whole RWd,

satisfaction of the constra.int HARCD~Y results in al1 additional k t sharing the same

headedness as the main-stressed f o d

26. HARMONY (Kager 1996aA5) Fect have identical headedness wi thin a PrWd.

Kager's proposal is based on îsvo crucial assurnptions: (a) that P.ms~-a is undominated in

the languagc in order to account for an initial (LH) foot and (b) thai minimal violation of

m û s - o forces the parsing of uneven trochaic feet (HL), even though bimoraic syllables

-- -

"a notes thai hIcCarlhy and Ri- (1986) propose a similar foor h o n y process for their analysis of Yidi~i stress.

can be parsed as a unique f m t within the HdWd. Thus, in his anaiysis, both iambs and

uncvcn trochees are allowed by the language.

In the nest section 1 reformulate Kapr ' s anal ysis of metrical prominence in Guugu

Yimidhirr in terms of the mode1 of prosodic structure developed in this disserîation.

S. 1.5 Headedness and Grid Prominence

In Guugu Y imidhirr, the rhythmic patîem of a PrWd is dependent on the location of weight

aithin the initial disyllabic domain. Note, howcver, that if the initiai disyllabic domain is

parsed as a RSRS as assumed here. then the iambic pattern cannot simply fdlow from the

l m t i o n of the heaw syllabie. An LH constituent can be p;used as either a branching

PrStem [ap( 'apr)] or an iambic foot [(ap1oPp)] What is striljng about the iambic pattern

is that, although a right-headed RStem does not violate either WEIGHI--~H(PRS'EW) o r

FT-FOR~ r~~ocmic J, the optimal output resui ts in the bj moraic syllable k i n g parsed as the

head of an iambic f m t violating both hi@-ranking constraints.

T o force an iam bic pattern on the initial LH sequence. 1 propose that AUGS-P-Sm-

LtFT already introduced in the analysis of J a p e s e truncation is undominated in Guugu

Y imidhirr.

37. ALICN-P-STEM-LEFT Align (P-Stem, Left, Ft, Left)

"The left edge of every P-Stem coincides with the left edp of somc fmt."

This constrai nt would thus dominate b t h Wncm-m-H(PRSrnr) and FT-FoR\~~RocH.~Ic)

in ihc language. Fdlowing Kagcr, 1 assume the domination of FT-FoR\~~REHAIC] by

28. HARMONY Fect have identical headedness within a PrWd.

As a consequence of the latter constraint's domination of R-FoR\I, any initial iarnbic fmt

rcsulting from satisfying AUGN-P-SIRFLET and W T - l o n ~ will enforce further violations

of FT-FOR~~~ROCH.LUC] t h m g h o u t the RWd domain. These ranhngs are shown in tableau

(29) beiow with the form [magilgaygù] 'branch-~~-EMPH' ('just branches').

Candidates

a v/[(ma..giil)].(i~ay.gù)

As shown by the candidate in (29a), satisfaction of AUGS-P-Sm{-Lm rules out an initial

unfooted syllable, resulting in violations of both WEIGHT-ID-H(PRSTBI) and FT-FORM

/J%ocK-\Ic]. In words with a bimoraic yl lable at the right e d p , on the other hand. therc is

n o nced to assume, as does Kager, that the disyllabic constituent parses an unel-en trochaic

foot.

1 Candidates

In fact as shown by the non-optimal output in (30b), an uneven trochaic f m t incun a fatal

Final1 y considfr the prcsence of cataiectic syllables word-final1 y in a numbcr of odd-

num bered syllable words. Here 1 assume that the presence of a caiaiectic syllable ai the right

edge fol lows direcii y from satisfying AUGN-PRWDRIGHT. Wi thout catalexis. the presence

of only one initial PrStem in the language would leme an unboted syllable to be parsed

dircctly by the PrWd, violating both AUGN-PRWIIRIGK~ and P.~RsE-a.

Supporting evidence for the catalcxis analysis comes from the iambic stress pattern.

Within this pattern, a linal moraic syllable is ncver stressed. Note that under the assumption

ihat non-rnoraic syllaMes are headless. the final moraic syllable is the only available fooi

Consider first a PrWd without a final cabîectic syllabie, as in the quadrisyllabic

27b. * d~ line 2 [. (*)] (* .) line 1

CL ClCL CL Cil [(ma.giil)].(gaygu)

Eveq hot-head projets a linc I grid mark. Satisfaction of HUWOW, hoivever, requires

thüt e v e y foot have identical headedness, as illustrateci by the non-optimal output in (Tb).

On line 2, there is a grid mark for evev PrStem-head.

As shown by the metrical grid in (32). only moraic heads may project a line 1 pnd

mark. This requirement can be regarded as an instantiation of the Conrinuous Colunm

Consrraint of Halle and Vergnaud ( 1987), under the assumption that a mora is the head of

a syllable.

28b.e x linc 2 [. ( 9 1 ( *) line 1

Cr ClCr C I - [(ga-mbuu)] .(gu.u)

I l (gu) a7ere treated as a unaq foot, there would be no principled reason for the fact that the

final syllable is unstresscd in the iambic pattern but stressed in the trochaic pattem (cf.

(wu.lu).(C)gùr.u) 'lightning, flame' and (mAr.bu).(gàn.bi).(@.u) 'still in the cave'). Notice

that forms like [buu-rday.bi.gu] 'still in the water' offer clear evidence aginst any analysis

The anal ysis of hfohawk in g5.2 discusses in more detail this partïcular assumption. See also the analysis proposed by Cohn and hfclarihy (1994) for the behavior of schwa in the stress system of Indonesian.

of *[ga?buu.gùJ in terms of stress clash. On the other hand, by assuming that the head

syllable of a disyllabic monomoraic foot cannot project a grid mark at level 1 whenever

such a projection would result in a violation of undominateâ i-Luwon, we can provide a

straightfonvard accuunt for the lack of secondary stress on the final moraic yllablc of odd-

numbcred words in the iambic stress pattern. These proposais are illustratcd in the taMeau

In (33a), the final unstresseci foot of the optimal candidate output is assumed to mtisfy Fr-

F o R \ ~ ~ R ~ ~ K ~ I c ] . This assumption follows direct1 y from the position taken here that

metrical constituent headedness and grid prominence are two interretated but independent

notions.

To sum up, in my analysis of the stress system of Guugu Yimidhirr, the l d o n of

main stress follows directly from the proposal that AUGS-PRSmt-LET is undominated in

the language. That the PrStem may contain two main-stressed feet, on the other hand, is

assumed to follow from the domination of M O S O H E ~ ' E S S by Wn~w-m-H(PRsm!) ,

while the iambic pattern is assumed to follow from the domination of both Wmw--rio-

H( PRSTM) and ~-FORV(TROCHAIC] by AIJGS-P-Sm4-Lm. .

Candidates

gambuugu l

a. I/ [(ga.mbuu)].(gu.u)

b. [(gambuu) j.(gù.a)

5.2 Ternary Stress Systerns

I n standard Optimality Thmry, directionality effects in languages wi th bi nary stress

systems are captureci through a constraint that reqüires the left or right edge of every foot io

h m - P - HAR\K)~Y

S m r - L m i

* !

STRESS-m-

H( FT)

a

FT-FORCI

[T ROCHAIC]

* tt

coincidc with the cormponding edge of some RWd. Satisfaction of AU-Fr, however, is

depcndcnt on its interaction with iwo higher-ranked constraints FT-BIS and P-ARSE-a In

most languages, Au-Fr t m l a t e s into minimal (or pidual) violations in words longer than

two syllables. T e m q stress patterns present an interesting problem for such a view of foot

distribution in that optimal outputs creatsd b5. t e w ?stems usuaily display numerous

~~iolat ions of PARSE-a whjch d o minimize violations of Am-FT violations, but never result

in full satisfaction of the latter constraint in words longer than four syllables.

In the model of prosodic structure developed here, the constraint A u ~ s - h S m i -

Lwr has been shown to provide a uni fied acoount Cor a number of phenornena in languages

as diverse as Japanese, Arabic and Guugu Yimidhirr- Directionality of foot pming, on the

other hand, has been argued to follow from a version of Au,Fr-Lrn/Rr~~r that requires

the alignment constraint to bc suwivided into AU-FT~, Am-Pr?, ArriFT3,.. etc. (cf. Kager

l996b).

Neither ALIGS-PRS-EM nor Au-- Fr requires that more than one foot be parsed by

sornc PrStem in a PrWd, however. The Prosodic Hierarchy only requires that wery RWd

have at least one PrStem as its head, and that e v e s PrStem have at l m t one head foot. Any

number of feet may be immediatel y dominated by the PrWd. So far in the study of stress

systems presented in this thesis, the only constraint that has been argued to crucially interact

with AUX-PRSTEM, rcsulting in minimal violation of the alignment constraint, is WEIGHF

I D - H ( P K S ~ I ) in Arabic. In the next two sections, 1 will show that t e m q stress systems

present strong evidence that some lmguagcs must allow for PrWds with multiple M t e m s .

5.2.1 Cayuvava

In ordcr to illustrate how ternary stress systems are optimally genemted in this model of

prosodic structure, 1 begin with a fairiy straightfonvard system, that of Cayuvava (Kager

1 W6b) .'

The Cayuvava data is onginally fmm Key (1x1). The languagc's ternary pattern has bcen anal@ in metrical rheory by Halle and Vergnaud (lm, Lcvin ( 1988). Dresha and Labiri (199 1). and Haycs (1 995).

120

3 4. Cayuvava Ternar), Stress Sysrem

a. d a p a

b. t6.mo.ho

c. ari.po.ro

d. a.ri.piro.to

e. a-ri. hi.hi. be-e

f. ma.rA.ha-ha.é.i.ki

g. i. k i - t apa re. ré. pe. ha h. ta.a.di.r6.bo.~.ni.ni.ce

'canoe'

'srna11 utater contai nef'

'he already tumed around'

'al r a d y planted'

'1 have al ready put the top on'

'their blankets'

'the water is clean' 'ninety-nine (first digit)'

Cayuvava is a language without vowel length distinction. Any vowel sequence represents

tnro syllables. Main stress is initial in disyllabic and trisyllabic words and is found on the

antepenultimate in longer words. When the word is long enouph, stress fails on every third

syllable munting leftward from the main stressed syllable.

5.2.1.1 Previous Analyses

Two recent anaiyses, that of ffiger ( I9%b) and Green and Kens to~icz ( 1993, have

rcformulated traditional accounts of temary stress systems wi thin the optimality-theoretic

Sramework. Kager adopts Hayes' (1995) view that temary rhythm is achieved at the cost of

cshaustivity with regard to syllable parsing. Green and Kenstowicz, by contrast, follow

Halle and Vergnaud ( 1987) in postulating an additional type of bot: the amphybrach foot,

in which the hcad is one syllaMe away from each foot edge.

In the serial frarnework of Hayes ( 1995), foot parsing is iterative and is assumed to

csami ne a finite 'window' (syllable sequence) for each iteration.

35. Foor P arsing ibcafity Parameter

a. Strong Local Parsing M e n a foot has k e n constructed, d i en the window for

further parsing at the next unfootcd syllable.

(unrnarked value of the parameter)

b. Weak Lacal P arsing When a foot has k e n constructed, al ign the window for

further parsing by skipping over /"/, wherc possi bie.

(marked value of the parameter)

Under this foot parsing algorith, Strong Local Parsing results in a b i n q stress pattern,

tvhile Weak k a i Parsing creates a te- stress pattern. Hayes proposes the follo\vlng

andysis for Cayvava

3 6a. Sy ilable Exrrametricality 0 --> (0) 1 -J H ' O ~

b. Foot Comtntction Fonn syllabic trochees f r m right to left:

i. Employ weak local parsing. . . LI. Degenerate feet are allowed in strong position. . . . I I I . Footing is non-persistent.

c. Word b e r Construction End Ruie Right

S tri king1 y, Cayuvava di ffers from most temary systems (cf. Chugach and Es to~an) in that

adjacent feet never surface. Whenever the grouping of syllables into k m q constituents

rcsults in two left over syllables at an edge, the two syllables are never regrouped into a

b i n q constituent (36d-g). Thus, in words consisting of 3n+2 syllables, we find a double

upbeat: i.e. the first twvo syllables are stressless. In Hayes' framework, the initial double

upbeat is derived from the assumption that foot construction is not persistent in the

37. Persistent Footing If two stray syllables are left at the end of the initial parse, they are regrouped into a syllabic trochee.

The end rcsult of the leftward iterative foot parsing is the assipment of a degcnerate u n q

foot at the left cdge in 3n+2 words. When the word l a y r is constructed over the foot Iayer,

this degenarate foot is deleted.

1 n Kager's constraint- based anal ysis, undominated Fr-Brs ensures that no unary

fect are ever created. Weal local parsing is refomulated as a constraint which forbids

adjacent feet, with non-exhaustivi ty k i n g enforced by the ranlring *ErFr >> PARSE-a.

38. *FTFT Fcet must not be adjacent.

The hi@-rankinp of the consimint *F-rFr, however, cannot explain the choice of (3%) over

the ungrammatical (39b) which does not viofate either *FTFT or Au-FT-RIGHT.

39a. (td.mo).ho b. * to.(m&ho)

Neither Hayes and Kager can relate the choice of (3%) over (39b) through satisfaction of

either the weak parsing parameter or the constraint *FTFT. Thus Hayes must apped to the

notion of ex&tricaIiv and assume t h a word-final syllable is invisible to foot parsing.

Similarly, Kager uses the OT version of estrarnetrïcality and proposes a dominancc relation

between NOWLUL and AL- Fr-RGW.

40. *FTFT ; N0hTW.a >> P.~RsE-o » AL-FT-NGHT

Kager's anal ysis adequatel y descn bes the stress system of Cayuvava. There is, however,

Little e~~idence provided that supports the introduction of *FTFT as a universal constraint,

cscept to spccificall y address the question of how to esplain the lack of foot adjacency that

chanccenzes t e m q stress patterns.*

In their analysis of t e m q systems, Grccn and Kenstowicz dso need some version

of the *FTFT constraint which the? formulate as *@(a. In contrast with Kager, however,

Candidates 1

a. J a.ri.(pLro).to

b. (a. ri).(pi.ro). to

c. (ai.ri).pi.(r6,to)

thcir analysis is based on the proposal that Fr-BIS should be decornposeci into hvo distinct

constraints: MIS-2(p,a), which set a lower bound on foot si=, and L . . E (Green 1995).

4 1. MIN- 2p, MIN- 2a: a metrical foot contains two moras or two syllables.

LAPSE- p, LAPSE-a adjacent unstressed moras or syllables must be separated by a

foot boundary.

i *FTFT j NO~;FK.AL

I

j *! i i

!

1 *!

The role played by Kager's *FTFT conshnt is furrher examineci in the account of the optionally stress pattern of Estonian in $52.2.

PARSE-a

*tlt

It

t

AU;-FT-R

*

Ir*** I

ttt

Green and Kenstowicz generate the temary stress pattern of Cayuvuva by assuming the

following constra.int ranking: LAPE-O >> Au-h-RIGHI- >> PARSE-O- Since there are no

restrictions imposed on possible leet, Green & Kenstowin propose that the way to satisfy

L-WSE-a in five-sylIabe words while minimizing ALL-F~-R.IGHT violations is to assume that

the main strcssed syllable is parsed as the head of an arnphibrach fwt: [a(ri.pLro).to]. This

givcs the w o n g result, however, for longer words as shown below

42a ( ta.a).di .(ro.bo). p.(ni.ni).ce b. * ta(adiro).bo.(@-ni.ni).ce

Green & Kenstowin argue thai the tailure of the double amphibrach in nine-syllabie words

indicatcs that some additional constraint musr dominate Au-FT. They assume a constraint

~vhich favors Mt-heaâed feet: HEAD-L The arnphibrach imposed on five-syllable words,

howeïer, becomes problematic if HEXD-L dominates ALL-FT. Thus the need to introduce

*a)@ as another h i g h - d n g constraint in the stress system of Cayuvava. Again however,

these constraints are not enough to account for the stress pattem in [rna.(d. ha). ha.(& .i).ki].

Green and Kenstoslcz must therefore add M.-UN(STRESS)-L to their analysis in order to

generate the correct stress pattern in the seven-syllable word.

This analysis presents a number of problems. The loss of pnerality resulting from

the decomposition of Fi-BR' makes Coot parjing something much more like a guessing

game than one would like on acquisition grounds. That is, the typology of possible feet

includes now u n q , binary, t e m q and unbounded constituents, rather than k i n g

restricted to binary constituents. On e m p i r i d grounds, the fact that nearly e v c y odd-

numbcred syllable word require the introduction of an additional consimint in order io be

correc tl y pnerated brings into question the explanatory power of Green and Kenstowin's

model.9

Xlthough Grem and Kenstowicz (1995) aiso disasss Lhe optionai tcnÿny system of Estouïan. nonc of the tbree constraints introdisced in ader to specif~cally explain the iemat)l patterns in Capvava =ms to play a rolc in the Estonim stress system. I-&W-L, however. is argued to play a role in the stress system of Piraha.

1 n the nest section, an analysis of the stress system of Ca)-uvava is proposed that

on1 y requires the addition of one structurai constraint and some rerankng of the constraints

already introduced for Arabic, Japanese and Guugu Y imidhirr in order to genemte the

temary patterns of Cayuvava

5.2.1.2 Ternary systems and the PrStem

At issue in this section is how to account for te- systems in terms of the structural

constraints already introduced in the preceding chapters. In the mode1 of prosodic structure

dcveioped here, the most harmonic PrWd is assurned to follow from satisfying AUGS-

P R S ~ 1, P..IRSE-O and ALL-FT-LEF~/RIGHI-, with minimal violation of the laiter rnnstra.int

enforced in some contexts by the hi&-ran)ring of BR.LVCHHE.\D. In longer words, minimal

iViolation of AUGN-P~Srni has been argued to result from satisfying fully W ~ ~ m - l r ,

H(PRSmt). Here 1 propose that the t e m q pattern of Cayuvava, in which ternary

constituents iteratively parse the whole word, c m be accounted for by assumine the

domination of ALJGS-PRSTBI by some constraint which requires the left edge of every foot

to coincide with the left edge of some PrStem.

40. ALIGN- Fr-LEFT Align (Foot, Left, PrStem, k f t )

"The left edge of every foot coincides with the left edge of some PrStem."

Satisfaction of this constraint requires IWO conditions to be met: (i) that every foot be parsed

bq' a PrStern and (ii) that each fmt be left-di p e d wi th a PrStem. The candidates in (4 1) al1

41 a. PrWd I

PrS tem

Ft Ft I I

b. PrWd I ""

OPcr

c. PrWd

I m\ Ft Ft

I i

Far each of the candidates, the left edge of the final foot is not aligned with the left edge of

the PrStem. By contrast, only (41a) violaies FT-TO-PRSmf.

43. FT-TOI PRSTEM Every foot dominated by a PrStem is parsed as i ts head.

In a ianguagc where both AUGN-Fr-LER and F r - n P ~ S m i are high-ranking, the prosaiic

structure in (4 1 a) is the least harmonic candidate. The candidate in (41 b), on the other hand,

\-iolates AUGS-Fr-Lm, but would satisfy AUGN-FT-RIGKT. Crucial for this analysis of the

t c r n q pattern of Cayuvava is that the prosodic structure in (41 c), in which a foot is parsed

directly by the PrWd, always violates AUGX-FT-L~/RGHT. The claim is that in languages

nvhere AUGS-Fr dominates AUGX-PRSTEM, the result is that each foot is dominated by a

different PrStern. I o

Consider at this point the stress systern of Cayuvava 1 t is clear from forms such as

(dapa) and (r6mo)ho that feet are trochaic. In addition, the leftward directiondi ty effcct of

foot distribution and the initial double upbeat in some forms strongl y suggest that AUGS-

P~Srni-RIGHI- dominata P.ms~-a in the ianguage.

43. ALIGN- PRSTEM- RIGHT Atign (PrStem, Right, PrWd, Right)

"The right edge of every PrStem coincides with the right edge of some PrWd."

Obscrve how main stress is always three syllaMes from the r i e t edge, even when strict

dis>'llabici ty could be obtained.

44a. a. (ri. po). ro b. *(&ri).(@.ro)

Kager argues that such f o m s present strong evidence for undominated N O S ~ A L H F T . In

contrast with Kager's proposal, 1 argue that the nonfinality d main stress is an effect of

ALIGS-Fr-LEFI' and Au~x-P~Sm~-Rrccrr rather than NOS~ALHFT. This is illustrated in

thc two tableaux below.

In the analysis of Japanese tnmcation, WGN-FIiLE4T. although low-ranlàag. oever receives more Lhan one violation. That is, the optimai output had at most me foot tbat was not left-aligned with a PrStem. eithm by k n g part of a bipodal RStem or by king parsed directly by the RWd.

126

Note that AU-Fr-RIG~ evaluates how far each fmt is from the right edge. That is,

In the optimal candidate, AwFT-RIGWT is the only violaied constrainL In a disyllabic Tom

such as [(dapa)], on the other hand, al1 the constraints in the tableau are satisfied fu11y.

As illustrated in the tableau below with the parsing of a four syllabie word, the

domination of PARSE-:-a by both AUGX-FT-Lm and AUGS-PRST-WGHT results in an initial

syllable directly parscd by the PrWd.

AL-FT-

RIGHT I

*

./i

t - .-

- -

Candidates

ari poro

a. d a.[(ripo).ro]

Candidates

tomoho

a. d [(t6.m0).h0]

b. [to.(m6.ho)]

c. [(to.mo)] .ho

b. [(A. ri) J . [(M. ro) j

c. [(&ri)] .(@. ro)

AUGS-PRST-

RIGHI-

G - -

IAT

*!

P-*SE-O

*! t

The final foot in (46c) violates AUGN-Fr-LEFI-, since i t is parsed direct1 y by the RWd. In

(a), Au~s-FT-LEFI- is violated because the rightmost f w t cano t tx ieft-aligned with the

PrS tcm. 1 n (4th) the optimal candidate violates lower-ranked PARSE-O and ALL-FI-WHT.

The nest tableau illustrates the constraint rankings involved in the optimal parsing of

a fi\,e-sytlable word. As noted befme, what is rnost striking about 3n + 2 syllablc words is

that they always surface with a double upbeat. This initial doubie upbeat, 1 have argued,

foiiows directly from the satisfaction of AUGX-PRST€~~-RIC~HT.

I n the optimai candidate in (47a) both AUGS-FT-Lm and Am?;-PRST-RIGHI- are satisfied

fully Interestin&, in al1 the fonns reviewed so f a , no optimal candidate violates AUGN-

PRST-RIGHT. This is not the case, however, in words with more than five syllables. The

Candidates

aripiroto

a. 4 an.[(pi.ro).to]

b. (in). [( pi-ro). to ]

c. [ ( ir i ) ] . [(pi.ro).to]

d. [(a.ri).pi].[(r6.to)l

problem here is how to ensure that in such words no more than two syliables will be left

unparscd.

In his analysis of ternary systerns, Kager ( l996b) propose the constraint P.ARSE-~ to

UGS-R-

L E F ~

*!

rcstrict sequences of unparsed syllablcs. This constraint is argued to completely eliminate

the necd for PARSE-o. Acmrding to Kager, in Cayuvava, PARSE-? is dominated by both

* FTFT and Nmm..v..

UGN- PRST-

RIGHT

*!

*!

48. P A R S E - ~ Onc of two adjacent stress uni& must be parsed by a foot.

PARSE-a

tic

A~--FT-

REHT

* 1

~t ~ X X

~c tt*, -

4, ***?

1 n chapter 4,1 proposed a di fferen t type of constraint, W ~ K P A R S E C ~ D , whkh is assumed

to bc undominated universah, to restnct the number of unparsed sy Ilables in PrWds.

49. WEAKPARSECOND A rnora in a non-head syllable is adjacent to at rnost one mora in a non-head syllable.

1 n a Ie f hwrd trochaic system 1 ike Cayuvava, W UKPARSECCSD on1 y alt ows double up beats

at the left edge. l As noted by Hayes ( 1995), in languages where every foot is assigned

some stress, main stress on the fourth syllable from the right edge seems to be completely

unattested ( *[(a.ri).pol.ro ). This would follow direct& from the WE.KP.A.RSEC~D.

Thesc proposais are illustrated in the tluec tableaux bclow.

As shown in (50b) above and (Sld) below, although P- SE-c7 is low-mnkïng in the system,

the undominated status of the WE.IW.ARSECCYSD ensures that an' syllable not parsed as the

hcad of a foot will not be adjacent to more than one mora in a non-head syllable.

Candidates

ari hi hi bee

a. 4 [(&ri).hi].[(hLbe).e]

b. sri-hi. [(hi.be).eJ

c. a(ri.hi).[(hL be).e]

Sote that the optimal candidate output in (47) does not violate the WWKP.4RSECOhl3. since each mora in a non-hcad syîiable is adjacent to onl y m e other mora in a non-head syllable.

r i t least in langvages without headless syllables. See the analysis of hiohawli in $6.2 for a detailed discussion of the d e played by headless syllables in stress systems.

W ~ - u c P m s ~ A~GX-FT-

COKD La-r

*! i

*!

AUGS-PRST-

RIGHI-

~t

PARSE-O

t ~ i l t

t

-

Candidates

In (51 b), the non-optimal output is eliminated through a fatai violation of A m - k - L m ,

while in (Slc), the parsing of three PrStems results in a fatai violaton of ALIGN-PRST-

RIGHI-. In the nest tableau belou., a nine qllable word enforces the parsing of three

PrStems in order t o satisfy the undominated W W A R S E C ~ D and hi@-ranking AUGS-Fr-

L m .

Candidates

taadirobo~runiçe

Thus, in thc mode1 of prosodic structure developed here, the ternary patterns of Cayuvava

arc @\.en a straightfonvard account simply by assuming an al ipment constraint requiring

the coincidence of edges betwecn a foot and a PrStem. An important advantage presented

by this analysis over previous nile-baxd and constraint-based analyses (e.g. the Weak

Local Parsing of Hayes (1995) and the *RFT constraint of Kager (1996b) respectively) 1s

that the ternary system or Cayuvava can be explained without resorting to sorne constraint

o r condition on f m t pcusing that is specific to temary patterns. In the mode1 proposed hem,

the difference behveen b i n q and ternary patterns is aaiounted for by the crucial fanking of

mro alignment constraints: ALIGN-fi-Lm » AUGS-RST-RICm in a t e m q system like

Cayu\*ava and AUGS-PRST-Lm >> AUGS-Fi in a binary v s t e m li ke Guugu Y irnidhirr. In

the nest section, the variable ternary and b i n w systems of E~tonian are esarnined.

4.2.2 Estonian

The complcx stress patterns of Estonian present an interesting challenge to any t h e q of

metrical parsing. The Estonian stress system is optionally b i n q or tema?. with a ihrce-

a a y distinction of syllabie weight, i.e. short, long, and overlong syllables. In Estonian.

main stress is found on the first syilable of the phonologid word. Secondary stress is also

largely predictable, although it can be variably ternary o r b i q , depending on word length

and syllable weight Exceptions to initial main stress are found in some loan words, and

there is a restricted set of stems and affixes that carv o r trigger i d i q n c r a t i c secon-

stress. 13

Estonian is sensitive to the weight of syllables in that a syllable mntaining a heavy

diphthong or a syllable closed by two consonants is stressed a t the right edge of a PrWd,

whiie a light syllable is not. T h e Tact that CVC is oounted as light, while CvCC is heavy,

also indicrites that final consonants are 'extmnetrical' in the language. Sensitivity to syllable

weight, however, is sometimes lacking word-medially. In the binary pattern, for example,

the secondary stress system parses ail non-final feet into strictly disyllabic domains. This

mcans that closed syllables may be parsed either in the strong or in the weak position of a

disyllabic foot, depending o n their position aithin the string. In the optional ternary pattern,

on the other hand, secondary stress faits on every third syllable f m m the initiai main stress,

but closed syllables m o t be left unfooted. (The data are from K a p r 19%b.)

I-Iint (1973) groups these exceptions together refening to them as morphologically bound stress. Hint also argues that e v q morphologically bound stress begins its own phondogicai word (Hayes 1995: 16).

13 1

53. Binas, Stress in Estonian a. pA.1at.t

b.

c. ka. v a làt. t

d. ré.te.li.le

e. té.ravàmd.t

f. pd.ri.màt.te1.t

g pi.mes.tà.vale

h. pimes.i2.vas.se

i. pi.mes.tàt.tu. te

j. u.1is.tà.vamai.t k. 'yp.pet.tA.jat.tèk.s

1. 6.sa.và.ma.lè. ki

m. hi.li.sè.mat-t&le

n. v5.ra.sèi.mat.tè.k o. ' yp.pet.tùs. te.Iè. ki

p. U-sal.tàt.ta.vàtmat. tèk.s

Ternar)? Stress in Estonian 'piece, parts&

pimes.tav 'blinding'

'cunning, part-se.'

'ladder, allsg.'

té.ra.vamàl.t 'more skillful, gen-sg.'

'the best, abl-pl.'

pi-mes. t a -v i le 'blinding, ill.sg.'

pi.mes. ta.vàs.se 'blinding, ill-sg.'

'the dazelled, gen-pl.'

u.lis.ta.v&mai. t I ... t yp. pet. ta-jàt. tek-s ' 1 . . *

6.sava.mà.le. ki ' a h more skillful, abl-sg.'

hi.li-se.màt.te.le 'later, all.pl.'

'earliest, all.pl.' 'lessons, too, al 1. pl .'

1 1 ...

The Tact that only open syllables can be skipped in the temary pattern has been offered as

cvidence that closed syllables are t r d as bimoraic by the stress system of Estonian.

Another parûcdarly interesting aspect of Estonian is the presence of a third weight

distinction, that of overlong syllablcs. Ovcrlength is characterized as a heavy syllable which

receives some estra duration: Cvv: (Bu: kèle), CvC: ($1: kèsse), CvCC: (kint lùstetèki) and

CiVvC: (tms: kùstesse). Such overlong syllables are od y found word-initially. As with the

ivords beginning wi th a syllable of ordinary length duration, the secondary stress pattem of

words with initial overiength has an optional temary or b i n q pattern in some contexts.

54. Bina? Stress Sys f ern Ternary Stress Sysrern

a. \16nk:.rit-t 'carriage, partsg. ' b. kAu:.k&.le l 4 'far away'

c. jal:.kè.test jal:. ke. tèst 'trick, ell.pl.'

14 Eek (1975) allows a ternary variant for the form in (Sb). In Hint (1973). ody the binary pattern is allowed for this type of fonn (Hayes lWS:3S 1 ). W l e my anal ysis is based mosd y on the patterns dlowed hy IIinr, the tmuaq pattem for tbe rom in (54b) can be accounted for with only a small modification of m e of the constraints involved.

jUl: .kès.se

tniu: . tù-se. lè. ki h&i:.kùs.test. t

tms:. kùs. tes-sc

3iu: .sàt, te.le

téot: .Et. tut. tèl. t

kint:.Iùs.te.lè.ki

kau:. kèt. tes-sè. ki

'bold, ill.sg.'

tniu:.tu.sè.le.ki I l ... hdi:. kus-tèst. t 'deseasc, ell. pl .' tWs: . kus. tès.se ' i n d u s q , ill.sg.'

du: .sat.iè. le 'honest, ai l .sgt

téot:. tat-tùt tel. t 'backe, abl-pl.'

kint:.lus.tè.le.ki I I ...

kau:. ket. t2s.se.ki 1 1 .. .

As noted before, one of the most puzzling aspects of Estonian is the seeming absence of

sensitivih to weight dispiayed by f m t parsing in the binary stress pattern. Monosyllabic

feet are only h u n d word-initiaily or word-finally. In fact, an important feature of ovedong

syllables is that they always begin a new phonologïcal word. Furthemore, words with

initial overlength d lows for a following closed syllable to be left unfooted in the t e r n q

pattcrn (ti%k: kustesse). B y contrast, od y an open syllable can be skpped by foot parsing

whcn following some disyllabic foot in the t e w pattern (* pimestattute vs. pimestavàte).

A non-final closed syllable that is not preceded by an overlong syllable is al ways parsed by

a disyllabic foot (pfmestàttute).

5.2.2.1 The Stress Systern of Estonian

Consider foot parsine in Estonian in words without overlong sy

arc only found word-finally and they always takc the f o m of a superheavy syllable, i.e.

Ilables. Monosy llabic feet

thcy end in two consonants or in a heavy diphthong and a consonant. On the other hand,

ordinary cioscd syllables in final position are always treated as liai. To account for the fact

that a final consonant docs not add to the weight of a prcceding syllable, 1 assume that the

conswa.int NoFcLuCMORA is undominated in Estonian (cf. the anaiysis of Arabic final

consonants in chapters 3 and 4).

55. NOFINALCMORA No PrWd final consonant is the head of a mora.

This proposal allows for a final bimoraic syllable to be parsed as the head of a disyllabic

foot [('q,li.u)] The assumption here is that ail feet are disyllabic in Estonian. The status of

overlong syllables is discussed in the final section-

Ncst consider the initial position of main stress and the rightulard directionality

effcct found in the binary stress pattern. This sugpsts that AUGS-PRST~-LE~T and A&-

FT-Lm are high-ranking in the laquage.

56. ALIGN-PRSTEM-LEFT Align (PrStem, h f t , PrWd, Left)

"The left edge of evev PrStem coincides with the left edge of some M d . "

57. ALL- FT-LEFT Align (Foot, Left, PrWd, Left)

"The left edge of every fcmt coincides with the lert edge of some PrWd."

The optional t e m q stress patteni, on the other hand, strongly suggests that AUGS-FT-LEFT

dominates AUGN-PRSTEM-Lm in Estonian.

58. ALIGN- FT-LEFT Align (Fmt, Left, PrStem, k f t )

"The left edge of every f m t coincides with the left edge of some PrStem."

Words containin? less than five syllables prcsent an intercsting problern in that they show

no variability with regard to the binay and ternary stress pattern, Le. there is no optionai

t e m q pattern. By assuming that PARSE-u alço dominates AUGW-P~Srnt-Lm such lack of

\.ariabiIie can be straightfomrardly accounted for.

AL-FT-

La=r

**

**

AUGS-

PRST-Lm

I

Candidates

reteiile Ci P Ir!'

a. 4 [(ré. te)]. [(li.le)]

P Ir ! ' C c

b. [(ré. te)].(lLle)

P P P P c. [(r&te).li~.Ie

A~GN-FT-

Lm-

*!

PARSE-O

*!

As shown in (59b), the domination of A m - P R S m i - L m by ALIGS-FT-LER forces the

parsing of the final f w t by a second PrStem. Note that the double upbeat at the right edge

i n (5%) also violatcs undominated W EAKP.ARsECO~D.

60. WEAKP ARSECOND A mora in a non-head syllable is adjacent to at most one mora in a non-head syllable.

The crucial rote played by this constraint in the stress system of Estcnian will bc esarnined

in the section on overlong syllables.

Nest consider a fonn with a final superheavy syllabte. As noted above, final CvCC

syllables are assumed to suface as a disyllabic f m t (opp. a) in Estonian. This is shown in

the tableau below, in which the fonn kavalm is treated as a four syllable word.

In (6 1 b), the final bimoraic syllable violates undominated NoFN.-u,CMOR-\, while in (61c),

the loss of the mora associated with the geminate results in a violation of p-M-do. Note

that parsing the final syIlable in (61c) as monomoraic does not result in a violation of the

la i thfulness constraint WT-IDESTIO which requires every segment linked to one mora in the

input to bc associated with one mora in the output.

As shown by the next tableau, the optimal candidate for a trisyllabic word ending in

an ordinary closed syllabie does not vidate either p-Mny10 or A U G S - P R S T ~ I - L ~ .

1

ALIGN-

PRST-Lm

*

t

Candidates

hvalatt

P u PP

a. 4 [(ki\ra)].[(iàt.t )]

P PP b. [(ka\-a)J.[(làt)J

C L P P c. [(ka va) .iat]

NOFCSALCP

*!

AUGN- FT-

LEI-

p-Mdo

*!

Candidates

pi mestav

ppp

u !' P!' b. [(pimes)].[(tàv.\-)]

Ir CI Ci!' c- [(pi'.mes)].[(tàv)]

Nok that the gernination of the final consonant in the non-optimal candidate (62b) also

results in an unnecessary violation of NOHEWLESSSIU

What is striking a b u t the candidates in tableau (62) is that the first closed syllable is

al ways parscd in the weak position of the main-sîressed ho t In the and ysis of A rabic, f m t

parsing was shown to be dependent on language-specific rankings of the weight constraints

which resulted in heavy closed syllable-s word-internally and in light closed syllables word-

finally. Here 1 propose that the weight of coda consonants is in part dependent on some

language-speci fic ranki ng beween the high-ranking structural canstraints and the weight

constraints. The ranking of the weight consiraints with respect to each other is given below.

In the foliowing sections, the interaction between the structurai wnstraints and the weight

constxaints is esamined in detail.

5.2.2.2 Binary and Ternary Stress Patterns

I n Es tonian, NOSH~REDMORA is dominated by NOFCLZLCMOR~. Word-internally, however,

some other constraint must interact wi th NOSHAREDM~A to enforce violations of the latter

constraint. Here 1 propose that N~SHAREDMORA vidations follow in part [rom the

optimi~ation of foot parsing. That is, in contrast with what has been assurned in the

prcvious tableaux, the optional b i n q pattern in Estonian is assurneci to f o l l o ~ from the

high-ranking of AU-F;T-Lm. This sarne h i g h - d n g is viewed as responsibk for the

wight ~ariability of closed syllables, word-internally.

In the optimd output, the parsing of the closed syllables as light ensures that AU-FT~-LEFI-

is minimally violated. In (64b). on the other hand, both ctosed syllables are parsed as a

bimoraic fmt, resulting in an additionai violation of AL-FT~-LEF~. In (WC), the fatal

1-iolation of A u - F - r 3 - L ~ , results from having the first closed syllable parsed directly by

the PrStcm. The high-ranking of ALL-FT-Lm thus imposes a binary pattern on five syllable

words ~ i i h a closed syllable in tbird position, irrcspectivc of whether the closed syllable is

trcatcd as heavy or light in the candidate outputs.

Following Broselow et al. (lm, 1 assume that an input moraic consonant when

sharîng a mora with a preceding vowel in the output does not result in a violation of WT-

IDBT. The faithfulness constraint o d y rquires that a segment linked to one mora in the

input be in a corresponcence relation with a sgmen t also linked to a mora t n the output --

the numbcr of segments sharing that particular mora in the output is irrelevant. The Tact that

the consonant still surfaces as a geminate, however, seems to indicate that somc sort of

prcscrvation of structure is at work hem. For this anaiysis, 1 wiIl simply assume that not Co

parse the second half of a geminak violates undominated MAX~O.

Aum-

PRST-LEF~

Ir

s ~ f

*

Candidates

parematkit Pl' Cr !'

i a. d [(@.re)].[(mAttelt)]

Cr Ir Pl' P!'

b. [(pire)]. [(màt)].[(tel. t ) ] u l ' !' !'!'

b. [(@.re).mat]. [(tel-t )]

Au~s-Fr-

LEF~

AuFr-

LEFT

4 *2

J1 **!?

dl **!?

N&HW

MORA

**

*

Observe now that even in a rive syllable word with o d y open syllables outside of

the initial disyllabic f a , only the bina^ pattern is optimal, Le. it violates Au-FT?-LEF~

minimally. This is shown in the tableau belout.

Thc question here is, givcn that in (WC) the third closed syllable is treated as light, i.e.

csactly like the third open syllable in (65b), then what would ensure that the t e r n e pattern

ALIGS-

PRST-LET

*

t

t

is optional for the latter but impossible for the former?

In the analysis of the stress system of Egyptian Radio Arabic I assumed that

Candidates

pimestavale P P P!' P

a. 4 [(pi.mes)].[(tàva)-le]

CL P ! 'C r b. phes es).^ [(vi le)]

CL w P Cr c. [pi.(més)]. [(tà.va).le J

variabilie in the choice of an optimal output follows direct1 y from the variable rankings

WGS-FT-

IR=I-

*!

1

ALL-FI--

La-T

4 *2

dl '*!?

*1 **2

bctuPeen two constraints within the constraint hienrchy. Here 1 propose that the optionally

N c S ~ m m

MORA

*

*

b i n q and temary stress patterns result from the variable rankinp of AL-Fr-Lm within

thc consirain hierarchy. The dactyl effect round in the t e m q pattern of five syllable words,

on the other hand, would follow from the high-ranl;ing of a constraint rcquirïng that a fmt

66. ALICN- PR WD- RICHT Align (PrWd, Right, foot, Right)

"The right edge of every R W d coincides with the right edge of some fmt."

Thc constraint ranbngs that are relevant to this analpis of the stress system of Estonian

nvould bc as follow.

This is illustrated in the hvo îableaus belonr.

Candidates 1

~imesîavale

Candidates II

pimestavale P Ii P P P

a. [(pimes)]. [(tà.va).le] P P P P

b. d [(pimes). ta]. [(v&le)]

As shown above, whenever ALL-Fr-Lm- y is bypassed and AUGS-PRST-LEFI- ties with

one violation for each candidate output, the choice o f the optimal output falls on AUGN-

PRWDE~GK. Note that d l the candidate outputs tie with one violation each for

N C S I - L ~ ~ E D M ~ A in order to satisfy AUGN-Fr-Lm.

Consider now the more problematic case displayed in the tableau belou.. As shown

in the lower tableau, in a five syllablc word with a ciosed syllable in third position, both the

temary pattern in (69b) and the binary pattern in (6%) results in a fatal violation of

NOS~-LVIEDMORL This resulis in the choice of *[(pLmes)].[(tàt)].[(tu.te)] as the opma l

output.

69. W E . ~ P . A R S E C O ~ ; NOFIS.U.CMOR.\ ; AUGY- F r - L m » PARSE-u >> ALL-FT-Lm- y

Candidates II 1 AUGS-Fr- 1 NOSHWD 1 AUGS- 1 Au-FT-

Candidates 1

pimestattute

What this mcans is that although optimal foot parsing has been shown to result in disyllabic

pimestattute

!' P l r P !' a.v'[(pi.mes) J.[(tàt)]. [(tù-te)]

P P P U

b . [(pi.mes).tat]. [(tù. te)]

P P P l r P c. [(pimes) 1. [(tàt. tu). te]

feet for most of the data, we need to assume that BEWKHHEAD is high-mking in the

AIAGN-FT-

L m - y

grarnmar's hierarchy to cnsure that closcd syllables will nor be parsed as a bimoraic bo t ,

N&HW

MORA

L - F T - L E F ~

!' P ! ' C r ! '

a. 4 [(pLmes)]. [(tàt. tu). te]

P U !' P P

b . [(pimes). tat]. [(tu-te)]

L m

70. BRANCHHEAD The head of a PrStem is in a rhythmic relation with a non-head constituent at some levct of analysis. (a, Ft).

ALIGX-

PRST-Lm

t f

t x

JI *2

JI **!?

*

t

MORA

X

**!

**!

PRST-LEFT

**

*

*

LW+

4 *Z

dl '5

4 **2

Thc proposai that the obligatory tnnary pattern in five syllable words with a dosed syllable

in third position fdlows directi y from the domination of NOSHMEDMORI by B R ~ ~ H E A D

and PARSE-a is illustraieci in the iableau below. Note that eveqzime there is a closed syllable

in lhird or îourth position in longer words, AUGS-PRWDRIGHT becomes irrelevant with

regard to the choice of the optimal output. It it thus not included in the tableau,.

7 1. WEZAKPARSECOSD ; NOFI~LUCMOR~ ; A UGS- FT-Lm » BR~LKCHHCU~ >> P.USE-O

» Au-Fr-Lm- >> NaSnuuz~Mo~~ >> p - M d O >> A u ~ s - P R S T - L m »

ALIGS-PRWIIRIGHI- » AL-FT-Lm-g - -

Candidates 1

pi mestattute I L

C P C I C

a. U/ [(pi. mes)]. [(et. tu). te]

U C C P!' b . [(pimes).tat]. [(th-te)]

I Candidates II I pimestattutc U C U P C

a. [(pi. mes)]. [( tàt. ru). te]

P I iC u C b. [(pi-mes)]. [(tàt). tu J. te

U C I Ir P V

C. [(pLmes).tat J. [(tù. te) 1 u Ir P!' CI CI

d. [(pLmes) ].[(tàt)].[(tù.te)]

The parsing of a six syllable word with a ciosed syilabic in third position preiients furthcr

evidcnce for the proposal ofîered above.

ALIGS-

FRST-Lm

*

t

BRLKCH-

HGID

7

AL-FT-

L m - y

J i "2

J1 ** !2

NOSHNGD

M m

*%

tt

In the nest section, the status of overlong syllables and their behavior with regard to the

stress system is discussed.

5.2.2.3 Overlong Syllables

Considcr no\v words with overiong syllablcs in initial position. Most analyses of

overlength in Estonian are built on the idea that ok7erlong syllables may be treated on a par

with disyllabic feet. For Kager (MW), the assignment of two grid marks to overiong

syllables is one way to treat them as if they were disyllabic. Hayes ( 19!35), in tum, chooses

to view oterlong syllables as trimoraic, with the added option of treating them as disyllabic

(Hayes 1995325).

73- [[y PI [PI Io

r

Candidates 1

tii lisemattele r

P P U P P !' a.Jf(hi.ii)]. [(sè.mat)]. [(tè-le)]

U V P P P P

b . [(hili).se].[(màt.te).leJ r

Candidates II

hi lisemattele 1

P IL P P Ir P

a. [(hi.li)j.[(sè.mat)].[(tè.le)]

P P P P P P b. d[(hi.li).se). [(màt. te).lel

w

P P P PP P P

c. [(hili).se].[(màt).te].le

Overlong syllables in Estonian his tor id l y derive frorn ciisyllabic sequences (Tauli 1954;

Mürk 1981). Overlong syllables also differ from ordinary heavy syllables in that thcy are

AU--FT-

La=r

4 **3

J1 **!?

N~SHUZED

M m

*

*

BRASCH-

H m

P - w s o

*!

NOSHWED

M m 4

t

t

WGS-

PRST-Lm

**!

*

*

AUGS-

PRST-Lm

rl*

I

t

I

AL-FT-

IBT

4 *2 *%

4 **2 I

dl

felt by native speakers to have 'extra heavy stress'. They also have special pitch patterns

(Hayes 1995). In addition, the fact that overlong syllables are always initial in a

phonological word is assumed in most analyses to mean that any non-initial overlong

syllable must begin a PrWd.

Accordingl y, 1 propose to characterize overlong syllables as superheavy syllables,

i.c. e v e s overlong syllable has the form CvCC or CvvC in the input, even those wïthout a

fi na1 consonant For esample, 1 assume that a word li ke [mu: tuseleki] has the lexical rom

[ruuu~tuseleki], nrhiie the word Ijul:liesse] has the lexical fonn Ijullkesse]. Note that

whcther we assume for the geminated consonant to be associateci with one m o n or uith

two skeletal dots in the le?iicon will not make any ciifference since the geminated consonant

always closes a bimoraic syllable in the output form.

The result is that every overlong syllaMe will be parsed as an initial displlabic foot

(%ri.a), as illustrated by the p i n g of the data below.

74. Bina- Stress System Tema- Stress System

a. (vAn.k ).(rit.t )

b. (k5u.w ).(kè.le)

c. (jal.1 ).(kè.tes).t . (jd.1 ).ke.(tès. t )

d. (ju1.1 ).(kes.se)

c. (tniu-w ).(tù.se).(lè.ki) ( t n i u . ~ ).tu.(sè.lc).);i

f. (h6i.y ).(kùs. test). t (h&y ). hxs.(tèst.t )

g (t66.s ).(kùs.tes).se (t66.s ),kus.(tès.se)

h. (6u.w ).(sàt.te).le (au.w ).sat.(tè.le)

i . (té0.t ).(tàt.tut).(tèl.t ) (téo. t ).tat.(tÙt.tel).t

j. (kin.t ).(lùs.te).(lè.ki) (kin. t ).Ius.(tè.le).ki

k. (kau.ur ).(kèt.tes).(sè.ki) ( k a u . ~ ).ket.(tès.sc).ki

'carnage, part-sg.'

'far away'

'trick, ell.pl.'

' M d , ilI.sg.' 9 f -.. 'desease, ell-pl.'

'industry, ill-se.'

'honest, d l -sg . '

'backe, abl-pl.' t 1 . . . ' 1

S . .

Now consider the stress pattern of words with an initial overlong syllable. Again, words

with less than five syHables c m o t take the optimal temary patterns. As shown by the two

tableaux below, satisfaction of NOFWALCMORA and of PVZLYIO results in the absence or

t e n q patterns in words with overlong syllables.

Rccall that in the previous section, 1 noted that the los of the second half of a geminate

sccms to be preserved whether the first ha1 f of the geminate heads i ts own mora or shares i t

w i th the preccding vowei. In (75d) then, it is the 105s of the consonant not of the mora that

rcsults in a fatal violation of MdO.

Notc that in the non-optimal outputs (7%-c), the W m W . w ~ C m o is satisfied since

the two m o m in a non-head syllable are adjacent to a mora in the head syllable. Crucially in

(75c), both moras of the non-head sy llable are adjacent to no more than one mora in a non-

hcad syllable.

As shown in (76b) below, in a four syllable word that ends in an ordinary open

syllable, the tcrnary pattern does incur a fatal violation of WUKPAEXSEC~D, whenever the

ciosed syllable is bimoraic- That is, the second mora of the non-bead syllable is adjacent to

two moras that are not parsed by a head syllable. In ( 7 6 ~ ) ~ on the d h e r hand, the parsing of

thc closed syllable as monomoraic satifies the WE~UCPARSECO~D, but results in a fatai

violation of PARSE-a.

Candidates

vankri tt

P* *!' a. [ ( v h - k )].[(rit.t .)]

Plr CiCr b. [(van. k ).(rit. t) J

PP y r c. [(van.k).nt]

*Ci Cr d. [(van.k ).rit]

Mc10 / NoFrsiu, UGS-FT-

i cp i

*!

: *! i

* i

ALL-FT-

L m - y

**

tt

NOSHWD

Mou4

t

Candidates

iull kesse

Consider now the stress patterns of five and six syllable words with an overlong syllable.

1 n contrast with words beginning with an o r d i n q disyllabic foot, initial overlong syllables

allow for a hewy syllable to be parsed directly by a RStem. as illustmted in the tableaux

bel ow .

77. WEXP.~SECOXD ; M d o ; NoF[';..u,CMm ; ALIGS-FT-Lm >> B R L ~ ~ H E \ D >> P - ~ s E - o » AIL-FT-Lm- y » N O S H ~ ~ M O R A » p - h k ~ I O » A U G S - ~ S T -

Lwr » AUGK-PRWM~GW >> Am-FT-Lm-f.3

Candidates 1

hai vhustesse

a. 4 [ ( h k y )].[(kùs.tes).se] liP Pli Cc v 1 b [(hai-y ).kur].[(tfs.se)]

Candidates II

haivhs tesse

Obscn-e that with regard to a PrStem that contains an initial overlong syllable, every mora

in the non-head syllable is adjacent to a head syllable. In other words, the W E M X R S E C ~ D

is satisficd in both the binay and thc temary pattern in f m s that are five syliables or

longer.

78. W E X P X R S E C ~ ; M-do ; NoF(sALCMOR-\ ; AUGV-FT-LEFT' » BRXKCHHEU) >>

P.-IRSE-O » AL.-FT-Lm- y » NOSUWMORA >> p-Mxx10 >> AUGS-PRST-

L m » AUGS-PRWDRIGHT >> Au-FT-Lm-p

Candidates 1

Candidates II

{kin) lusteleki

Pl' Ii Cr P P

As was the case for words without overlong syllables, the parsine in (7&) of a non-final

closcd syllable as a bimoraic foot in the terne pattern results in a fatal violation of PARSE-

a. This results in the choice of the t e m q pattern falling back to Auc;s-PRST-L~-

We arc left with one residual problem. As Kager (1996b) notes, therc secms to be

somc preference for thc temary pattern with word-initial overlength when the result is the

parsing of a light syllable in the weak position of a PrStem and of a heavy syllable as the

hcad of a non-final fwt. In fact, in some words with initial overlong syllables, only the

ternary pattern is found.

82. Obligatory Temary Partent

a. "[(kah. t )]. [(lè.vai).le] [();ah.t ).le].[(vài).le] 'doubtful, all.pl.' 6. *[(kyh.t )].[(lè.jat)].[(tèk.s )-] [ ( k h - t ).le].[(jàt).teks] 'hesi tant, transi-pl.'

c. * [(kul-t )]. [(sè.mat)]. [(tè-le)] [[(Lail. t ).se]. [(màt.te).le] 'more golden, d l . pl' d. *[(kau.w )].((kè.mat)].[(tès.se)] [[(kau-w ).ke].[(rnàt.tes).se] 'far away, ill.pl.'

There is at least one esample in Hayes ( 19953 19), however, that shows an optional b ina5

pattern for a similar fonn: [(we.s )].[(tù.set) j. [(tà.ki) ] and [( tyes ).tu].[(sèt.ta).ki]. These

two foms suggest that the obligatory t e m q pattern must be lexically specified somchow.

I have n o clear answer, however, as to why this lexical specification would be round only

in nords with ovedong syllables and only when the overlong syllable is followed by a

sequence of open + closed syilables.

In conclusion, the analysis of the stress system of Estonian presents a stnlang

csamplc of the esplanatory power of a theory that takes non-uniformity as a central aspect

of grammars. The notion of minimal violation as a means for ensuring smctural well-

formedness allows for a unified account of variable wcight in consonants. The claim that

ianguage variation results from the rerankng of one o r two constraints, on the other hand,

offers a straightfonvard account of the restrictions imposed on the optional binary and

t e r n q stress patterns in Estonian. 1 have shown here that varïability behveen temary and

b i n q stress patterns follows from two aspects of the laquage: (i) the undominated status

of AUGS-FT-Lm and (ii) the variable ranhng of A u - F r - L m in the pmrnar 's hierarchy.

One of the most interesting results achieved here, however, is how the interaction

betwcen NOS~WREDMORA, AU-FT-Lm and BR~XHHEW allows for an anaiysis of

Estonian that completely eliminates the need for the highly-marked ('LH) and ('HH)

trochaic feet. The proposal that BR~VCHHEAD and PARSE-O dominate both NOSH.AREDMOW

and AL-FT-Lm was s h o w to be crucial to the parsing of closed syllables in the temary

pattern.

Chapter 6

The Prosodie Stem and Syllable Weight

In this chapter, two languages, one wi th an iambic stress system (Chugach) and the other

ui th a trochaic stress systcrn (Mohawk) are under anaiysis. 1 t will be shown that these two

languages demonstrate a strong preference for a main-stressed foot that uniquely dominates

a h c a q syllable. In optimality-theoretic terrns, both languages treat a bimoraic (app) fmt as

more h m o n i c than either an uneven foot (ap~upp) in the case of the iarnbic system or a

disyllabic f w t ('opv) in the case of the trochaic system.

6.1 Chugach

Chugach is one of two major dialects of Pacific Yupik (also calleci Alutiiq). It is an iambic

system wi th ri ghtward di rectionaii @- and a t e m q stress pattern (Hayes 1995: Rice 1992).

Although the language has both long vowels and diphthongs, only word-initial closed

syllablcs are treated as bimoraic by the stress system. Words with only light syllables show

a distinct t e m q pattern in that the second 'Ilable is strcssed, and additional stresses fa11

o n cvcry third syllable thereafter. In contrast with the ternary stress system of Cayuvava,

maximal f't parsing in Chugach takes precedence over t e p rhythm, shorvn in ( 1 b, e).?

1 . Only Lighr Syllables

a. a.ta.ka b. a.h%.ta.m&

c. a. tci.qu.ni.ki'

'my father' 'kind of food (abl-sg.)'

'if hc (refl.) uses thern'

The Pacific \'upi k data, taken from Hayes (1 995) a d Ricc ( 199î). is o"@nall y from Leer ( l98%, 1989). The language's stress pattern has also becn analyzed within the rneiricai framework by Hammond (lm), HdIe (1990). Hewitt (1991) and m e r ( l m ) .

Although schwa seems to bear stress in Chugach. this only happens in closed syllables. In open syllables, schwa is either deieted or dcvoiced, with stress showing up leftward within the foot (Leer 1985a).

'if he (refl.) is goinp t o hunt' 'if he (refl.) is going to hunt porpoise'

In ~ ~ o r d s begiming with a long vowel o r closed syllable, the first syllable is stressed and

additional stresses fall on every third syllable therdter . Again, double upbeats are avoidcd.

'mine (pl.)' 'm y older brother'

'we'll g o out'

'apparently reading it' ' ou ' re goi ng to backpack'

'if he (refl.) is going to get bemes '

Both long vowels and diphthongs interfere with the ternary stress pattcrn in that they are

obligatonly stressed. Word-intemally, closed syllables d o not pattern with bimoraic vowels

i n that regard. 1 nterestingl y, long vowels and diphthongh in pst- ini tial position eriforce

initial stress by triggering gemination in a preceding open syllable, as shown in the fonns in

(3c) and (3e). (The geminated consonant is in bold.)

3 . Non -initial Heavy Sy Ilables 'she's reading i t'

'pocket'

'her hai r' '1'11 g o out'

'1'11 go out'

'she lied to me'

'1 will backpack'

'if you ta)re a long time' 'he is going to watch her'

'I'm going to go out'

Chugach prcscnts an intercsting problem for any framework which assumes that the uneven

iamb is the most harmonic f w t of an iarnbic stress system. Aithough the head of every

disyllabic foot in Chugach is subject to Iambic Lengthening, heavy syllables are obIigatorily

parscd üs a unique fwt . This was first noted in Leer (1983, who demonstnted that

consonant fortition in Yupik correlates with a foot-initial boundary.3 In Chupch, e v e q

hcavy syllable is described as beginning with a fortis consonant. Another problem that has

already bocn mentioned is that, although he language allows for long vowels, closed

syl lables are treakd as heavy only in word-ini tial position.

6.1.1 Previous Analysis

Consider Hayes' (1995) metrical analysis of the iambic system of Chugach. In order to

csplai n the variable weight of closed syllables, Haycs simply assumes that thc domain of

application of the Weight-by-Fosition rule cm be resüicted to the initial syllab1e. The

parsing of heavy syllables as a unique foot, howcvcr, presents more of a challenge for

Hayes' model, since both (LH) and (H) are perfect iambs.

Hayes proposes the following analysis for the temary stress pattern of Chugach.

4a. Foot Construction Fom iambs from left to ri@, under weak local parsing;

footing is persistent.

The assumption that footing is persistent, in conjunction with the parsing of (LH) feet as a

perfcct iamb, is problematic for any metrical anaiysis of Chugach because it cannot explain

the rcquiremcnt that every underlying heavy syllable should be parsec! as a unique foot.

Consider for example, quadrisyllabic f m s that have both initial and final heavy yllables

(cf. (at.ma.fi.quA]). Foot construction as constraincd by Weak Local Parsing derives the

incorrect form [(At).max.(ii.qui)]. In order to denve [(at).(maxii:).(qua)1, Hayes must

Fortition is described as complete lack of voicing in voiœless c~nsoaatlîs and prcclosure. Rcclosure results in extra Iength for the fortis amonants as cornparcd witb tbeir lenis counterparrs.

assume h a t the thrd syllable is heavy before foot construction applies. T o ensure that the

third syllable in such f o m s is heavy, Hayes relies on two surface characteristics of stressed

upords in Chugach: the lengthening of heads in disyllabic feet and consonant fortition in

foot-initial position. (Fortition is shown in bold character.)"

5. (zin).&(quii) (kd).(mda).nuk (in).ki.(qu.kut) (&it).(mau.èi:).(qui) (a. tk).qu.(ni.kf) (iq).Suk.(qiï).r~a

(nh).qu.(ma.lu:).ku (&n).(ku. ta)O.(t~@ (aku). tar.(tu. nir). tuq

Ha).=' analysis relates the fortition of a consonant at the Ieft of a heavy syllable and the

lcngthening of the preceding vowel to a d e that inserts a m m t o the Ieft of an underiyinp

Cvv syllable prior to f w t construction.

6. Pre-Long Strengfhening O Conditions: (a) pi links to @ where basic I \ syllabi fication pennits

0 --> pi 1 - ~1 ci (b) Othenvise, 11; Iinks to b t h

/ \ I a and p. a [-cons]

Prc-Long Strengthening is formulated so as to account for both simple gemination in forms

likc lnu.yai/ --> [nuyyaij and the lenpthening found in forms like /at.ma,.Ci.qua/ -->

[at.ma.i.EL.qua]. When the mora is inserted after an initial Cv syllable. Weight-by-Pbsition

allows the added mora to link only to the geminated consonant. In word-intemal position,

the insertcd mora must be linked to both the onset consonant and the preceding vowel. The

samc process is assumed to apply to closed syllables preceding any Cvv syllable.

- - n u y a i ... c i q u a ... t a t u a

Even &ou& no phondic contrast holds betweca fortis and lenis initidly. Leer descrÎbes al1 wad-initiai consonants as fortis.

As shown in (8), f w t construction then applies and parses every bimoraic syllaMe as a

foot. Follocving f m t construction, persistent footing reparses stray syllables in the weak

p i tion of a following monosyllabic foot.

8a. (nuy).(yai) (at). mai. (ei : ).(qua) (an).ku.(ta)O .(tua)

b. (nuy)-(yai) (at).(max.Ei:).(qua) (an).(ku. tay).(tua)

As shown in (8b), fortition also applies in word-initial position and at the left edge of a

disyllabic fmt. This is viewed by Hayes as the result of a late phonetic d e .

9. Phonetic Foot -initial Lengthening Realize a non-moraic f a - i n i tial consonant as slightly longer and more fortis.

Hayes defends the view that iambic lengthening in disyllabic Teet resulis from the

application of hvo distinct rules: Pre-Long Strengthening which applies befwe foot

construction and Iarnbic Lengthening which applies late in the derivation. In a similar way,

fortition is the result or two different processes: Pre-Long Strengthening for underlying

heaq syllables and Foot-initial Lengthening which applies at the phonetic levcl.

Even more problematic, however, is the generation of trisytlabic forms with one

l ight syllable caught behveen IWO heavy syllabies (cf. [&n.ti.qui]). The application of Pre-

Long Strengthening should malce the medial -Ilable bimoraic, resulting in three bimoraic

fcct. To climinate such forms, Hayes proposes that a sequence of three stressed syllables is

ill-forrncd because of a double clash. Whenever foot construction creates such a sequence

of stresses, the ill-fomed sequcnce is repaired through destressing and subscquent loss of

the degenerate syllable.

1 Oa. * * * t . * b . * * * (an).(Ei: .)(qua) --> (an).(Ci)(quA) --> (an). Ci .(quA) -> (an).(ki.quA)

Loss of Foot Constniction Dcstressi ng Dcgenerate Foot Persis rent Foo ti ng

Persistent footing results in a final (LH) foot. A shown above. such foot parsing inmrrecily

prcd@s that the onset consonant of the initial monomotaic syllable of the (LH) foot should

be strenghtened by foot-initial fortition. The Corn (&~).&(quii). however, on1 y shows

fortition on the final heay syllable. Hayes must therefore assume that footing is persistent

only in the initial stages of foot construction in Yupik and that persistent footing cannot

adjust metrical structures resulting from destressing. How persistent footing can distinguish

bctu-een the result of foot construction and that of destressing is not examined.

In the next section, an analysis of Chugach is presented that relies on the notion that

the PrStem is the head of the RWd. Phonological processes that are sensitive to foot

structure but which d o not influence foot parsing, such as Consonant Fortition and Iambic

Lengthening, are outside of the domain of this analysis.

6.1.2 The Chugach Stress System and the PrStem

Non- consider the t e m q stress pattern of Chugach witbin the mode1 of prosodic structure

presented hcre. As demonstnted in the analysis of the stress system of Cayuvava, a ternary

pattern, rhis timc with rightward nther than Icftward directiondis, u n be attributed in part

to the domination of AUGK-Fr-Lm over AUGS-PRSTEM-LEFT. In addition, the a v o i b c e of

a final double upbeat clearfy demonsrnies that PARSE-a crucially dominates AUGS-PRSTEM-

L m in Chugach. These proposais arc illustnted in the tableau below.

1 n l o n g r even-num bered words the domination of AUGS-Fr-LEFI- over AUGS-PRST-LDT

rcsults in a tcmary pattern.

Candidates

ahtarnak 1

P Ci F Ci

a. J [(aku)] . [(ta. ma k)]

P P C L C L

b. [(a. Lai)]. (ta. mak)

AuG.;-FT- P-WE-a

IJFI-

*!

AUGS- PRST-

Lnir 1

* I

l

P P P P [(a. ku).ta].mak C. *!

I Candidates

P P P !'l' b. [(pi.su)].(qu.ta).(qu.nf)

Pl' Ir 21 l' Ir c. [(pi.sU).[(qu.ta) J. [(qu.nK) J

!'P P U F U d. [(pi-su)]. [(qu-tA).qu].ni

Since Chugach is an iambic language, the final double upbeat would not violate the

LEFI- j

13. WEAKPARSECOND A mora in a non-head syllablc is adjacent to at most one mora in a non-head syllable.

Lm-

Odd-numbcred syllable words, on the other hand, offer evidence of the role playxi by ALL-

Fr-RIGHI- in the choice of the optimal output, as shown belou..

In (lQ), the optimal candidate output satisfies Au;FTI -RIGHI- sincc the rightmost syllable

is pesfectly aligned with the ri@ edge of the RWd. The candidate in (l4b), o n the 0 t h

hand, has a syllable between the RWd ri@-edp and the rightmost bo t , resulting in a fatal

violation of AU-FT~ -RIGw.

Candidates

atuquniki L

P P CL !' IL a. 4 [(a.tu).qu].[(ni.ki)1

P Cc C L P P b. [(a.tu)].[(qu.ni').ki]

AUGS-

PRST-Lm

*

X

&JGX--- PARSE-a

i

AIL- FT-

RIGHI- I

JI ***2

X ! XX*, 1 -

Consider now the problem posed by the obligatory bimoraici ty of an initial closed

syllable and by the fact that long vowels in second position trtgger gemination in an initiai

open syllable-

1 Sa. [(th ) ]. [(sh) .ni] b. * [(Css&i).~i]

In the analyses of Arabic and Guugu Yimihdirr, the constraint Wnccrr-~o-H(PRSmi) was

introduced in order to esplain the requirement that main stress is preferably assigned to a

heavy syllable.

16. WEIGHT-TO- H(PRSTEM) A bimoraic s yllaMe is parsed as the head of a PrStem.

This constraint requires a bimoraic syllable to be parsed as the head foot of a PrStem, rather

than as the head syllable of a fwt. The prediction then is that, even within an iambic

systcm, a bimoraic syllable cannot be optimally parsed as the head of a disyllabic foot when

W n ~ m - ~ o - H ( h S m ) is undorninated in the constraint systcm.

The fact that gemination rather than vowel lengthening applies to the initial syllable

is assumed to follow from the dominance of W T - I ~ T over I~TE%R'O, as shown in the

tableau below .

1 ~TE%RI*

*

Candidates

haaui

!'Ci PCi P a. 4 [(&k)].[(sAa).~i]

!' Ci!' P

b. [(Ca. s h ) .ni)

w Cr!' v c. [(ha)]. [(sia).isi]

!' C(P P d. ta. [(sk).ai]

Wnccrr-m- f ~ G N - F T - WT-1 BT / PARSE-a

H(PRSTEN) f LEFI. !

*!

*! t

*!

1 n ( 1 7c), thc non-optimal ondidate violates WT-1 m ~ , since the input monomoraic vowel

is bimonic in the output. In the optimal ouput (l7a), the second rnora is assigned to a copy

of the precedi ng consonant, resulting in a violation of I s n x i ~ I O .

Consider now the obligatory bimoraicity of closed spllables in initial position. E v e v

other closed y l lable is viewed as monomoraic for the purpose of foot parsing. This word-

initial cffect, however, cannot be the result of satisfying either AUGX-Fr-Lm or PARSE-a.

as show-n by a form like [h.Ei.qu. kut] 'we'll go out'. AIJG';-FT--L~ and PARSE-a are

The obiigatory parsing of an initial closed syllable as a bimoraic fmt, I propose, follows

from a rcquirement that a R W d be left-alipcd with the head of a foot.

19. ALICN-PRWD-LEIV Align (PrWd, Left, Hu, Left)

"Every PrWd is left-aligncd with some hcad syllable."

Undcr the assumption that FT-FoRU~[..~.MBIC] is undominated in Chugach, the only way to

satisîy AUGS-PRWDLER. is for the initial syllabie to be parsed as a bimoraic f m t 5

Note that ALIGN-PRWDLEFT crucially dominates Am';-PRST-Lm, as illustrateci in

thc tableau below.

In Kager (1996a). ihe oMigatory parsing of a heavy syllable at the righi edge of the RNrd is simply statcd as AUGS-H The right edge of every RWd coincides with the right edge of some heavy syiiable." In his analysis of the incomplete phase in Ronimaa, on the d e r band. McCarthy (1995). formulates the constnint ISC-PH ris: .4lign (StemInc.Ph, Right [u)Ft, Right) (or -4iign (Stdnc-Ph, Right. pp. Right). i-e. "Every incomplete-phasc stem ends in a monosyllabic h o t (or heavy syllable)".

Furthemore, by assuming the dominancc of W T - I ~ T and IXTEGRIO over AUGS-PRWD

L m , one predicts that lengthening and gemination will never apply to satisfy the alignment

constnint. Only a non-rnoraic input segment is allowed to receive a mora in the output so as

to satisfy AUGS-PRWDLE~.

2 1 a. [(a tii).qu]. [(ni. ki) ] b. *[(at).tu].[(qu.ni).ki]

Candidates WGS- ALIGS- AL- FT-RIGHT

atmaktaXptan P R W D L ~ PRST-Lm

a. 4 [(at).ma].[(ku.tallQ]. [(tu. tan)]

The parsing of forms containing both long vowels an3 an initial closed syllable are

illustrated in the two iableaus (22) and (23) belo~~.

C i ! ' Cr P ! '

b. [fat-ma). hm]. [(ta)d tu)- tan]

rr Cr P r C r c. [(At)]. [(ma. hi).ta)a. [(tu. tan)]

u!' r C1 P P P

d. [(at)] . [(ma. ku)]. [( tax tu). tan]

*!

Candidates

atmaeiqua

PCi P P PP

a d [(At)]. [(mm-tf)]. [(qui)] Ci!' P Cr !'CL

b. [(At). mas]. [(&.qui) J Pl' !' !' PP

c. [(At).mas]. [&.(qua)]

P C C P !'CL

d. l(at.m&x).Ci]. [(qui)]

*

**

**

1 Wacw-m- Aum-Fr-

I H(PRS-mr) ! L m

8

i !

*! I i I

*! 1

* %**%, I -

J1 ***!3****" - 3

*! ***,***** 1 - 3

Aucs-

PRWDLEI.

"!

AUCI;-

PRST-LEF~

*X

*

* 1

* L

In (23b), the non-optimal output parses the final diphthong as the head of a disyllabic foot

n ther than as the head f m t of a PrStem, incurring a \.iolation of Wnc~n--ro-H(P~zSm~). In

(22c), on the other hand, the Mt-edge of the final foot does not coincide with that of the

PrStem, resultinp in a fatai violation of AUGS-Fr-Lm. Findly in (22d), the non-optimal

ouput violates AUGS-PRWDLEF~, since the initial ctosed syllable is in the weak position of

a disyllabic f a . Another possible candidate [(At).mau].ii. [ (qua incurs a fatal violation of

P.=sE-a.

31. W W X R S E Com ; WT-IDBT ; WEIGK~-TO-H(PRS~I) ; AUGX-FT-Lm : PARSE-O >>

Ixmxro >> AUGS-PRWDLEFT » AUGS-PRST-Lm >> A u - F T - R I G ~

I t Candidates

In this analysis, the bimoraic status of initial closed syllablcs and the obligatory grnination

iqSukqiiqa

PP P IiP P S. v' [(iq).+uk].[(qii).ga]

UP P Vlr P

b. [(iq) ].[(Suk.qTi).']a] P PP P

c. [(iq-Mc)]. [(qii).qaJ

process iound in the ini tial input sequence CvCvv fdl together in thai they both follow from

fully satis-ing the t\vo constraints WEIGI~T-TO-H(PRSTBI) and AUGS-PRWDLET.

H(PRSTE\I) j 1

1 *! i

I i *!

Finally, with regard to the equal promincncc assigned by the stress system to cvery

PrStcm, 1 propose that this pattern follows from the domination of M O N O H E ~ S E S S by

24. PRSTEM-TOI PRWD Evcry PrStem is parsed as the head of a PrWd.

25. MONOHEADEDNESS Prosodic consti tuen ts are uniquel y headed.

The assumption hem is that satisfaction of PRSmt-TO-PRWD rcsults in the RWd king

=signcd ttvo grid marks at level 3.

26. 8 * line 3 H(PrWd) * * . line 2 H(PrS tem)

(*) - (*) - line 1 H( FI)

The consuaint system pertaining to this analysis of the stress in Chugach is shown beloa.

Observe that d l of the crucial constraints p r o p e d to account for thc ternary stress system

of Chugach are part of the srnall set of constraints that has already been proposed to explain

both the b i n q pattern of Guugu Yimdhirr and the t e m q pattern of Ca>uva\.a.

29. Guug u Yirnidhirr A~~Gs-PRSTE~~~-LW ; PARSE-O >> AU-FT-RIGHI- >> AUGS-FT-RKW.

In addition 10 the variable rankings of the consuaints above, differenccs betweecn the three

languagcs have becn shown to follow from the choice of FT-FORM. the arguments chosen

by the constraint WEIGW-IO- HL\D and S T R E S S - ~ > - H ~ , and the interaction of faithfulness

constraints with thc genenl phonology of the language, i.e. with the constnints on the

prosodic shape of mcuical constituents (includinp syllables) and on thcir distribution wilhin

larger prosodic consti tuents.

6.2 Mohawk

We tum now to the analysis of the t roch ic stress systern of Mohawk a Northem Iroquoian

language spoken mainly in New York, Ontario and Québec. The problem presented by

Mohawk is twofold. First, the proccss of Tonic Lengthening (i.e. the lengthening of a main

strcsscd open syllabie) in a trochaic language that o t h e m i s @\les the appcarance of k i n g

quantity-insensitive (Le. no long vowels underiyingly, c l d syllables that behave as light

cscept w hen stressed, and an obligatorily disyllabic minimal word). Second, the variable

visi bility of epenthetic syîlablcs with regard to the stress system. The behavior of the three

di ffcrent epenthetic voufels in Mohawk is descnbed below .

6.2.1 Epenthesis and the Stress System of Mohawk

According to Michelson (1989), therc are thrce distinct cpenthetic vowels e, a. i. in

Mohawk.6 The first turo behave in the same manner uith regard to stress: invisible to it

wvhen in an open syllable; visiMe to it in a closed syllaMe. The major di fferencc bctween the

tuu nith rcgard to stress is found in a scquenœ of two adjacent epenthesizeû syllables in a

ulord- In such cases, the first one is usually visible to the stress system. However, if a is

one of the two epenthesized vowels, it is always the one participating in the stress system.

By contrast, the prothetic vowel i is always stressed and obligatorily found in initial

p s i tion in \i.ords tri th on1 y one non-epenthetic s yllable. The data are from Piggoti ( 1945).

Linder1 i ned segments rcpresent cpenthetic vowcls.

30. The Distn'bution of prothetic [i J a. k - y ~ - s [ i k y ~ s j '1 put it'

b. k-tat-s [iktats j '1 offcr i tf

c. k-ck-s [i:keks] '1 eat'

d. s-riht [isgriht] 'Cook!' c . t-n-ehr-3 [itgnehrg? 1 ' o u and 1 want'

Epcnthetic e is found in thrce different contcxts, as noted by Michelson ( 1989). In the first

contes t, e is inscrtcd h twcen a consonant and n, r, o r W .

,\mrding to bfichelson. both epenthctic e and a have the same phonelc realizaiion as their uaddying CounierpartS.

3 1. a. k-runyu-s [kçninqws) '1 sketch' b. ~- ) i - r -~ - ? [A k g n 31 '1 will put into a container'

c. tc-k-nk-s [lékgri ks 1 1 put them together' d . t-n- k-ahsutr-A-? [tr\ kahsutgn ?] '1 wili splice it'

e. M.-akra-s [ w&gas] 'It smells'

Thc second contest involves a sequence of three consonants, the fim of which cannot be h

or < and the second of which cannot be h o r S. The epenthetic vowel is insertcd betureen

the first conLwnant and the following cluster. (The p v e accent represcnts a falling tone.)

32. a. k-the3t-ha' [kgthè: tha? ] '1 pound, am pounding'

b. wak-nyak-s [wakenyaks] '1 get mamed'

c. s-rho-s [grhos) 'you coat it with something' d . te-k-ahsutr-ha? [tekahsut&ha3 1 '1 splice it'

e . s-k-ahkt-s [skiih-ts j '1 got back'

Third. the epenthetic vowel is inserted betwreen a consonant and a word-Pinal glottal stop.

Note, howet'er. that closed syllables of the type Ce7 are alway treated as light.

33. a. A-k-ant-3 [AM rate? 1 '1 lay myself down'

b. ro-ht-01-3 [ r M : t0tg3 ] 'he has a bump on his nose'

c. wa? -t-k-abt-nak-3 fwa7 tkatatgna)cg3 ] '1 scntched mye1 T

d. t-A-k-rik-3 [ t ~ k g i l & ] '1 will put Logether side by side'

c. 0-nrahl-3 [ongrahtg7 ] 'leaf'

With regard to n 'the joincr', Michelson states that it is inserted betwecn two consonants al

a morphcmc boundary within the vcrb base: e.g. betrvan an incorporaicd noun root and a

follou~ing den\ational suffis. In c o n m t with e, the joiner may break up cluskrs that are

normal1 y allo\ved in Mohawk-

34. a. tc-k-a?shar-rik-s [teka? sM:w)cs] '1 put the );nives side by side'

b . ka-naw- ku [kana: wgku] 'in the swamp' c. wak-i? tuhkw-rho-s [wa)ri3 tuhkw&hos] '1 have a fever'

d. k-r-kw-as [Iqx&was] '1 take i t out of something'

c. te-kahniw-nyu [tekahruw&y] 'manu objects put in o u r p t h '

Epenthetic a can also be found idiosyncratidly within morphemes which cannot bc

andyzcd sqnchronidly into smaller morphemes. That is, there are some cases of non-

moraic a which cannot bC trcated as instances of the joiner.

6.2.2 The Principle and Parameter Appmach

A comprehcnsive analysis of the stress system in Mohawk within the Pnnciple and

Parameter frameworlr has been proposed in Piggott ( 1995).7 His analysis is bascd on the

assumption that 'weightless' (i-e. non-moraic) syllables are part of the invcntory of

possible syllables. Piggott's propsal assumcs that 'unlicensed' consonants can be parscd

by an cmpty-headed syllable and that epcnthetic vowels are insertcd as late as p s i ble in

thc p h o n o l ~ g y . ~

To account for the prosodic invisibility of epenthesized syllabies not closed by a

consonant, stress in Mohawk is argued to be sensitive to moras rather than syllables, i .es

non-rnoraic (epenthetic) syllables are ignored- Piggott argues that the hcad foot in Mohawk

is the rnoraic trochee and that coda consonants are assiped a mon by the Wcight-by-

Position rulc before stress assignment, irrespective of whether the syllable head is itseif

moraic or not9

35. Weight-by-Position A 'coda' consonant is assigned a mora in the course of syllabification (parameter).

To account for thc rcgular penultimate stress, Piggott proposes that a non-itentive trochaic

stress foot bc assipncd to the right-edgc. Given that the mora is the stress-bearing element

in Mohawk, non-moraic syllabks are skipped by the stress foot, resulting in tnsyllabic feet,

as shown in (36) wi th the fonns w-akra-s 'i t smells' and O-nrahr- 2'leaf':

36a (wA. kg. ras) b. (6. ngrah). tg3

Eiger (1993) also offcrs an d y s i s of Mohawk which assumes the pmxnce of 'weightiess' syilables in thc 1 anguagc. Sec Pi ggott ( 1992) for a di scussioo of Mohauï k minimal word. This latencss to mie application is rclated by Piggott (1995) to the economy prioQple of p m m i m in

Chomsky ( 1992). calicd Procastinate. Rccall. however. that a fiinai glottal stop does not add weight to a syllables.

Piggott assumes that non-moraic syllables within a foot are later prosodically liccnscd by

adjunction to the head of the fwt, a structure similar to the one proposed by Dresher and

Lahiri (1991) for the Gerrnanïc foot.

One problematic aspect of Piggott's analysis, however, is thc lack of stress on

closcd syllables in final or penultimate p s i tion when followgi by a non-moraic syllable. I f

coda consonants are moraic and stress is assigneci to a moraic tmchce at the right edge, the

cspcctation is that the rightxnost bimoraic syllable will be assignai main stress.

The moraic tmchec analysis also mns into problems with the minimal word requirernent.

Onc u-ould e'tpcct any fonn that contains a closed syllable to satisfy the minirnality

rcquircmcnt in Mohawk. Such enpectation is contradicted by the &ta in (37).

38. a. k - p s [ikyns)

b. k-ek-s [i:keksj

c. s-riht [fsgriht]

d . t-n-ehr-3 [itgnehrc? ]

To account for the consistent lack of stress on a rightmost closed syllable, Pi@gott assumes

a nonfinal ity condition rcquinng that the head qllable of a foot not be final. Hoarever, this

condition has to appiy to the metrical grid rather than to the RWd iiself in ordcr to gcneraie

the correct result, since an epenthetic syllable is invisible to rhis condition (e-g. in ongmhre

'lealT). Piggott thus assumes the avoidance condition of Idsardi ( 1 W-).

39. Avoid (s #

This is illustnted in (40) below.

B y using grid marks rathcr than mons to account for the nonfinality effect, Rggott can

account for b t h thc lack of stress on the nghtmost closed syllable and for the assignment

of a prothetic vowel. Since the mora is argued to be the stress-bcaring unit in Mohawk,

only moraïc syllables can project a line O grid mark.

Another important aspect of Rggott's analysis is the tieatment of the interaction of

cpenthesizcd syllables ~vith the stress system of Mohawk. To esplain the variaMc visibility

of the thrce cpenthetic t-owcls, Rggott adopts the derivational mode1 of hsical Phonology

(Kiparsky lm; Mohanan 1986). This mode1 h a at least twro welldefined levels at which

cpcnthetic vowels ma'; be inserted: the icxical and the postlcsical Icvcls. Piggon proposes

that epcnthctic vowels are inserted as Iate as possibIe in the derivation. The assumption is

that on1 y \.owels insemd at the lesical level are assiped a mora-

Undcr this \.iew, the visibilit). of epenthetic closed syllables follows directly from

the application of Weight-by-Position before foot assignment rather than from the insertion

of the cpcnthetic vowel. The prothetic vowel, however, is inserted lesically to satisfy the

mi nimal ity rcquircment Eady insertion is also triggered by any sequence of t u a cpenthetic

syl labtes. Assuming sornc notion of government between syllable positions (Kaye 1990;

Charette 1 99 1 ) , Piggott argues that non-moraic syllables must be licensed (i.e. govemcd)

by a moraic syllable to thcir right. 10 In a sequence of two non-moraic syllables, the first

one is not propcrl y govcrncd. A prosodic structure wi th two adjacent non-moraic sfllablcs

is thus ill-formcd and must bc repaired before stress assignment c m apply. This is done

through the earl y insertion of a vowel in the head psi tion of the non-moraic syllablc. On

the one hanci, epcnthetic [el is insertcd at the pst-lexical level. On the other hand, the joiner

[a], k i n g morphoiogically condition&, is inserted lexically but after stress assignrncnt.

Thus, the choice of [a] over [el in sequences containing both follows from the assumption

that [a] is insertod at the lexical level, while [el is inserted at the pst-lesical Icvel.

I o As aoted by Piggott. both Kaye ( l m ) and Charettç (1991) assume thal satisfaction of h o p tiovcniment is detcrmined h m right to lcft. This proposai is a m s i s t m t wilh the daim that final wcightiess syllables arc: dways propcrly govaned.

Finally, P i g o t t proposes an account of Tonic Lcngthening based on Rince's

( 1990) Trwhaic Rhythmic Harmony S d e , in \\hich a trochee coniaining one hem-y

syllable or tu.0 light syllablcs is r a i d higher than an uneven (HL) trochee. Piggott adop&

the Harrnonic Scale but rank H as more harmonic than LL, his proposal bci ng that Tonic

Lengthening is a repair strategy used to better a t i s fy the Hannonic Sdc.

4 1. Trochaic Enhancement Transfer wcight from the dependent to the head of the trochee.

This analysis also awiounts for the lack of lengthening in stressed syllables lollowed by an

epenthecic [el since the 'weightiess' syllablc has n o m m to transfer to the head -Ilable.

As n o t d bcfore, there are some problcmatic aspects to Pjggott's analysis. The main

problem to his approach is the assumption that stress in Mohawk is sensitivc only to moms

and that it ignores non-rnoraic syllables. This results in the parring of trisyllabic feet and the

assumption that nonfindi ty of the main-stressed foot applies to the metrical grid rather than

to thc PrWd itself. in addition, by treating Tonic Lengthening as a phonological process

that applies to an already constructed disyllabic foot rather than as a condition imposed o n

foot parsing. the obligatory bimomicity of the main-stresscd f m t m o t be reducible to

cither word minimality or foot parsing. In the neat sections. an optirnaiity-theoretic analysis

is proposcd that characterizcs the trochaic stress system of Mohawk in t e m s of structural

constrainis on prosodic constituents. 1 will focus o n the epenthetic schwa and prothetic [il.

6.2.3 Tonic Lengthening in a fQuantity-lnsensitivef Language

Consider first the stress system of Mohawk in words without epentheùc vowels. A word

has only one stress that rcgularly surfaces on the penultimate syllable. In an open syllable, a

main-strcssd vowel in penuitimate position is always lengthened.

42. Basic Stress Paftern a. k-atirut-ha? b. wak-ashet-u c. k - a k ~ 3rokew-as d. k-hyatu-s e. s-ho-ahk-u

@mtinitha?] '1 pull it' [wakas hé: tu] '1 have counted il' [b 3roke: w as 1 '1 am dusting' [khyk tus] '1 write' [shohktu] 'he wcnt back'

The absence of s e c o n d q stresses indicates that AuGs-PRSTR~-R~C;FTI- is high-raniiing in

the language, nhile the absence of main stress on a final c l o r d syllable follows frorn

43. NONFINALHF~ No PrWd is right-alipcd with its head fmt.

The o b l i g a t o ~ bimoraicity of the strcssed syllable, on the other hand, can bc given a vcq'

simple esplanation by assuming that lengthening follows directly from trying to minimize

violations of a structural constraint rhat requires the head syllable of a PrWd to be final.

44. AIAGNHO A PrWd is right-aligncd nith its hcad syllablc.

Thesc proposais arc illustrated in the following tableau.

1 n (4%) and (45d). the parsing of a disyl Iabic main stressed foot results in two violations

of A uc,sHu. 1 n (45b). the alignment of the disyllabic foot ~ ' i t h the right cdgc of the PrWd

also crcates a fatai violation of Noh,m-\~HF-r. In (54a), the penultimate vowel is lengthcned

in ordcr to minimizc violation of Au~sHa without violating undominatcd NO~SALHFT.

To Iengthen the main stresscd syllable in (45c), on the other hand, creates a fatal violation

of A UGS-PRST-RIGH~.

P.-IRSE-a

* 2

*

*

Candidates

k-&A rokew-as r

U C L l l C I ! '

a . 4 ka-(kil7.ro).[(ké:).\vasJ t P C t CI C1 I(

b. h ( k n ?r0). [(ké.was)].

!' u l l C I v !'

c. (ka.k~?).[(ro:).kcl.~~as

P PCL u d. (ka. kh 7). [(ro. ke) .\vas J

f 1

N m m ~ - i AUGS- ?

HFT 1 PRST-RIGHI-

Ar-rcsHa

?

t

i *! 4

j

i *!

j

1

X

*

**

**!

The problcm to be addressed at this point is how to ensure that the assigrment of a

m o n to a coda consonant is more optimal than to lengthen a vowel. Following Broselow et

al. (lm), 1 assume t k t languages in which coda consonants never contribute 10 syllable

uvei g h ~ escept under pressure from higher-ranked constninis, have the following ranking.

In Mohauvk, botfi vowel length and weighted coda consonants are restricted to the main

stressed foot. This should follow from the domination of both NOSH-\REDMORL\ and Wr-

IDESI-IO by ALIGSHO. Given the fact. howcvcr, that thcrc is no etidence for underlying long

\-ouVel s in the langage, I propose that both the lack of long vowels and the weightless coda

consonants resul t from the satisfaction of a constraint which requires that every rnoraic

segment be uniquely dominated by the head mora In other words, a segment cannot be in

any rra). ~tssociated with a mora in a non-head position.

47. S EGMENT-TOIMORA Evcq. segment dominated by a mora is uniqueiy parsed as ils head.

The candidates in (48) illustnte the prosodic structure of morriic segments. Note that of al1

the candidates ody (48d-e) satisf-y SEL~MEXT-DM~A fully by k i n g dominaied only by the

hcad mora.

1 n (4%) and (&), at least one segment is linkcd to a non-head mon, whilc in (48b), both

scgmcnts are 1in)rcd to a non-head mora Note that both (48b) and (48d) violate N o S u m a

Mc)R.\, whilc on1 y (48c) violates NOCMORL

The rolc played by SEGMES~-TO-MORA with respect to closed syllables is illustraied

b e l o ~ * .

As shown in (49c), it is more optimal to a s i p a m o n to a coda consonant, which results

Candidates

k-run YU-s r

Ir)r !'

a. 4 b. [(nin).yus]

w !' b. kc.[(ni:n).vusj

in one violation of SEGMEST-~-MOR.I~ than to lengthen a rnonomoraic vowcl in a closed

syl Iablc, which satisfies NoCMm4 but resul ts in wo violations of S~~~SEST-TO-MORLA.

Consider now the disyllabic requirement imposcd on minimal words as shown in

Aues-

PRST-RIGCCT-

thc data in (30). given again below.

50. The Distribution of prothetic [il a. k-yh-s [ikyhsj '1 put it'

b. k-ut-s [iktatsl '1 offer it'

c. k-ek-s [i: kcks] '1 mt '

d. s-riht [ isgi ht ] 'Cook!' c. t-n-ehr-7 [ i~neh-3 ] 'you and 1 want'

A u ~ s H a

*

x

In the prcccding chapters, such a disyllabic requirement has bcen argued to result from the

satisfaction of BRASCHHW which demands thc head of a prosodic constituent to bc in a

S E % S ~ T -

TO-MORA

*

r +!

rhythmic relation with some prosodic constituent in a non-heaâ position. In Mohawk,

NOCMORA

l

t

I

howevcr, undominaied N~T;T.v.ALHFT is cnough to forcc both the insertion of a prothctic

vowcl and its Icngthening at lcast in a-ords without epenthetic syliabies. This is illustratcd

in the tableau below.

Candidates NOSSISAL- Dd" AUGS- AUGNHU S E G ~ ~ T -

k-ek-s H FT PRST-RIGHI' TO-MORA 1

lrCl l' X * x

a. 4 [(i:).kcks]

!? *! t * w

ç. [(kcks)] *! *

To summariza, in words w-ithout epenthetic vo\vcls, bimoraicity of the main stressed

s>*llablc follotvs from the interaction between Lwo slmctural constraints: undominated

Nosm.-u ,HF~ and high-ranking Au~sHa.

6.2.4 Epenthesis in Mohawk

Consider now the rolc ptayed by the epenthetic vowcls in the positiontng of main stress.

Whenc\.cr an open epenthetic syllable is found in thc pcnult, the result is the retraction of

the main stress to the penultimate syllablc. Furthemore, the main stressed syllabie is ne iw

lcngrhcncd bcforc epcnthetic [el.

52. a. k-runyu-s [kgninps J '1 sketch1 b. A-k-r-A-? [A kgii ?] '1 niIl put into a container'

c. te-k-rik-s [tc'kg-iks 1 '1 put them togcthcr' d - l - ~ - k - a h s u ~ - ~ - 3 [thkahsulgr~? ] '1 will splice i t'

c. M.-ab-s [wakgas 1 'It smells'

1 propose that the answer as to why cpenthctic i.o\vcls cannot receive main strcss lics in

their non-moraic status. In my analysis of nominal stems in Anbic, it was assumed rhat the

obligatory parsing of a Iloating mora by the consonantal mot follows from a faithfulness

constrainl which requircs evcry mora in the input to be associated with an output scgment:

p-M.d0. In contrast with W T - I ~ B ? ~ ~ , which forbids both Iengthening and shortening, and

S~ihtn-r-TO-Mcm, which oniy forbids lengthening, p-Mn.10 bloch the loss of an input

mor4 although the wnstraint can be satisfied through the üansfer of a mora from a vowel

to a coda consonant.

53. p - M d o Every mora of S 1 has a correspondent l i d e d to a segment in S2.

The counierpart of p - M d o is p - D d o which requires that e v e s mora in the output be

associalcd with a segment in the input

9. ~ - D E P ~ O Ewry mora of S2 is l inked to a s e p e n t wi th a correspondent in S .

An)- mora Iinked to a cak consonant satisfies p-Dml0, and so does a lengthcned vowel.

Any m o n associated with an epenthetic vowel, on the other band, results in a \iolation oC

the faithfulncss constraint. Thesc atgumcnls arc îllustrated in the taMcau bclow.

In (Sb), the association of m o n s to thc cpenthetic v o w l in order to minimis2 violations of

A i ~ c i s H o results in a fatal violation of p-DEPIO. T o lengthen the preceding moraic vowel as

in (5%). results in a tie with (5%) with respcct to A~GXHO, while adding a fatal violation

for S m~-ro-Mm-~.

Consider now the augmentation of monosyllabic forrns that coniain cpenthetic [el.

Alderete ( 1995) proposes a constraint HUD-DEP to explain the lack of visibility of

epcnthetic vowels uith rcgard to the stress systcm.

SF~~ME~T-

TO-MORA

*

* l

r

Candidates

t -~-k-ihsutr-~-? B

P P P

a. (th.kah).[(su.t-).r~ 3 )

U V w u b. tli.(kah.su).[(t&J.rh 31

P P w * c. (th.kah).[(su:.te).r~ 3 )

Noms.u.-

HFT I

p-Ddo

*!*

A u ~ s H o

X*

*

**

56. HEAD( PCAT) - DEP E\.ery s c p e n t contained in a prosodic head in S2 has a correspondent in Si,

That is, any prosodic head in the output m u t be in a correspondence relation with a head

segment in the i npu t However, the constraint HWDDEP m o t expiain why [il is inserted

to augment a subminimal word, when anothcr epenthctic vowcl is dready available. since

both \-iolatc the prosodic constraint.

57.

AIderctefs sol utian d o p t s McCarthy's ( 1995) proposal that thcre arc prosodic fai thlulness

Candidates

s-ri ht

Ci

a. [(i. se).riht]

w b. [sc.(riht))

1

w P [(sg:).ri ht] C.

constraints which place segments in a correspondence relation mediated by their prosodic

structure. My analysis, on the othcr hand, assumes that Mohawk syllabifiation disfavors

NOWK-V.-

HFT

*!

the assa-iation of moras with segments which ha\.c n o correspondent in the input, p-D&O.

HEADDEP

*

*

Candidates

S-riht

(r Ci

a. [(i. se).nht]

w b. sg.[(riht)J

w P c. [(sg).riht]

AIJGSHCI

X X !

NOCMORA

*

Nmm-.4~-

HFT

*!

*

p-DwI0

*

w+!

*

AucsHa

X t

*

S ~ , ? r ~ s r -

TO-MORA

X

*

Considcr now the behavior of epenthetic vowels in closed syllables, as prcsented in the data

belo~v.

59. a. k-th&-haî [kgthè: tha3 ]

b. wak-ny&-s [wak&aks]

c. s-rho-s [sgrhos] d . te-k-ahsutr- ha3 [tekahsut&ha3 ]

e , s- k-ahhl-s [skahktgs ]

As notcd bcforc, a closcd epenthetic syllabie is prosodica

'1 pound, am pounding'

'1 get married'

'you coat it with something' '1 splice it'

'1 got back'

11' visible in penultimatc position.

Here 1 propose that the undorninated unistrain t MorurcCm.~ forces the insertion of a mon

in a hcadless closed syilable.

0bsen.c that in thc wse of a closed syllable, only the mora associateci with the epenthetic

\.o\ircl ~vould rcsult in a violation of p-Dd*. If WC assume that the initiai mord is also

linked to thc coda consonant, howe\.er, p D d 0 is satisfied fully in b t h (a) and (a). This is rcflcctcd in the numbcr of violations of S~cixrrr\rr-ro-Mm~ in (60a).

Now consider sorne forms ~ l t h a word-final epcnthetic syllable, and in the case of

(6 lc-f), with an additiond word-interna1 cpenthetic [cl.

6 1. Word-Firial Epenihetic Syllable a. A-k-arat-7 [~Mra@ 1 b. ro-Lit-ot-7 [du: totg? 1

'1 lay myself down'

'he has a bump on his nose'

S E G S ~ T -

TD-MORA

**

Candidates

te- k-ahsu tr- ha?

C L P U w P a. d te.(kah.su).[(t&). ha3 ]

CL CL P

b. (te. kah).[(su.tg).ha? ]

P P PCL

c. ( tc. kah). [(su. tg) . ha' ]

p-DdO MORAICCWA

*!

.

A i ~ ~ s H c 3

*

**

**!

c. \va? -1-k-atat-ncik-3 [wa? tkaiatgnakc? ] '1 scratched mysclf

d. t-A-k-rik-7 [ti\kgrikg? J '1 will put togethcr sidc by side'

e . p t - r - 7 [?O: ~ w - 7 1 '1 t's in the dishlglass'

S. w m h t - 3 [6-rahtg? ] 'leaf'

The problem presented by the word-final epenthetic syllables is threefold: (a) the absence of

main stress on the penul timate syllablc; (b) the lengthening of a main stressed vowel in the

antepenulùmaie syllable; and (c) the lack of lengthcning when stress is four syilables h m

Hcrc 1 propase to explain dl of these generalizations by assuming an undominated

constraint that requires a non-moraic syllable to be parsed in the weak position of a fwt.

62. NONHEADCOND A hcadless syllable is parsed in the non-head position of a fmt.

Crucially. any headless syllable thai is parsed directly by a RStem or a RWd violates the

N ~ H E u X ~ D . That only the penultimate cpenlhetic vowel mny rcceive a m o n on the

othcr hand. follo\vs from satisfying the constraint N o G L I J ~ ~ ~ u M o R ~ . As noted by Pigrnt

(1995). syllables closcd by a glottai stop are never trcated as heavy in Mohawk. Here 1

assume that the domination of N o G m U r r r ~ by MOR-IICCCOA forces the glottal stop to

link to the Root nodc of the preccding vowel. When the glottal stop links to a moraic

\.o\vcl, howevcr, the result is assumcd to result in a violation of N O G ~ X M O R A in order

to satisîy both p-M.-\sIo and Mu;1o.

Thesc proposais are i ilustrated in the tableau beiow. N o k a p n that in the evaluation

of' AIL-FT-RIGHT. ALL-FT~-RIGI~T evaluates the rightmost fmt .

As shown in the tableaux abovc, an- headless syllable parsed directly by the PrSkm or the

PrWd incurs a fatal violation of undominatcd N O N H F ~ O W .

Hme 1 assume that the head syllablc of a foot mus t itself have a heack i.c. a m o n

Candidates

yo- t-r- 3 l

w !'

a. I/ [(ycj:)(tg.r~9)]11

C

b. [(4'6.tg).rg1

P Cr

c. [ (y6 tg) .rg? j

Ci!' P

c. [(i. yo).( tg. rg3 ) ]

N m H m

CG~D

*!

N o G m . - w

MORA

*!

p D d 0

*

*

**!

Au~sHo

i

t*

1

** 1

x* .

%t*

Hcre, thc behavior of b t h the epenthetic vowel [el and the prothctic vowel [il in Mohawk

is accounted for with the introduction of two constnjnts on the parsing and licensing of

hcadl css syllables: NCXHEADC~D and DEPIO PI O. In addition, the wcight constraints have

bccn shown to crucially interact with some hi@-ranking structural constraints. resulting in

a bimoraic syllable as the main-stressed fwt in a laquage without underiying weight

contrasts.

Candidates

t-n-e hr-3 1

P C'

a- 4 [(i. te)-(neh.rgT)] u w

b. [(i.tg)].[(néh).rg?] Ir(i

c. tc. [ ( n é h ) . ~ ? 1

p-Dd"

*

*

*

- NCKHGID-

C ~ D

*!

* !*

AL~GX-

PRST-RIGHT

A L ~ G s H ~

1

*** 1

t

1

t

Conclusion

1 n this dissertation, 1 have shown that the introduction of the Prosodic Stem (PrStem) as a

ncn9 metrical constituent \vithin the P r d i c Merarchy allows for a more restrictive theory

of mcuical constituents. By assuming that the main-stressecl foot is irnmediately dominated

by a PrStem, a number of different prosodic shapes i m p e d on metrical constituents and a

numbcr of different stress systems could be anal>zed using a very limited set of constraints

on the prosodic shape and distribution of metrical constituents.

In chapter 2, 1 ha\.e argued that the difference khveen shortened names and word

clippings lies in large part in a distinction between bound stems and free words. That 1s. in

tnrncated stems that are obiigatoril y followed by an fis, strucnird constraints are imposed

on both the truncated stem and the affis simultaneously. In other words, the solution as to

nvhy shortened names are restricted to a bimoraic f m t shape lies in part in the optimality-

theoretic notion that the prosodic shapes of phonologically-empty morphemes result from

the satisfaction of high-ranking structural constraints, rather than from lesically-specified

prosodic requirements impmed on those morphemes. 1 have presented a unified account of

thc minimal and m a ~ m a l prosodic shapes of both tnincated forms and word clippings in

tcrms of 'the emergence of the unmarked' which relies on the proposal that a number of the

structural consûaints of the language impose some prosodic requirement on the PrStem. In

chapter 3, this mode1 of prosodic structure has been extended to include the root-and-

pattern sysrcm of the nominal and verbal stems in Arabic.

T h e study of the stress systems of a n u m k r of Arabic dialects in chapter 4 has

shown that the role played by the PrStem with respect to minimality requirements can be

estended to include longer words which are not coextensive with the whole word. An

important aspect of this mode1 is that the P r d i c Hierarchy encodes both headship and

p r o d i c domi nation. Whether the head of some prosodic category i s assigned prominence

on the metncal grid or not, on the other hand, has been argued to result from a constraint

which imposes restrictions on the type of heads that may project on the grid. In this view,

stress systems where syllable structure and f w t distribution play a role in detemining the

location of main stress yet where secondary stress is absent are predicted by the theorq.

nther than k i n g problematic for it.

The concerns of chapter 5 are in shonPing the central role played by the PrStem in

the p n e d phonologp of three very different languages. In the mode1 of prosodic structure

de\-eloped here, it is proposed that the di fference between b i n e and te- patterns on be

accounted for in a strai&tfoward manner by the crucial reranking of a v e q lirnited set of

structural constraints. The advancage offered by the introduction of the PrStcm within the

Prosodic Hierarchy is that it accounts for the fact ht, in both the binary system of Guugu

Y imidhim and the ternap system of Cayuvava, the preferred metrical grouping is larger

than the main-stressed f w t of the lanpuage. Chapter 6, on the other hand, has demonstrated

the role played bp the RStern in languages where the main-stressed foot of choice is

smaller than what appears to be the most h m o n i c foot for either the quantity-sensitive

imbic system of Chugach or the quantity-insensitive trcxhaic system of Mohawk.

Finally, this thesis has shown how Optirnality Theory seems parûculady welI-suited

to deal with the non-uniformity of various phenornena. Central to this framework is the

notion that hi@-ranking structural constraints may be minimdly violatcd in some contexts

in order to satisfy higher-ranked structural or fai thfulness constrain ts.This has k e n clearl y

demonstrated in the study of non-uniform phenomena such as truncation in Japanese and in

thc study of the \.ariability in stress patterns and in syllaMe weight that may be found within

a single lanpuage and across languages.

References

Alderete, John (1995). Faithfulness to prosodic heads. Ms., University of Massachusetts, Amherst.

Benua, Laura (1995). Identity effects in morphological tmncation. Jill Beckman, Laura Walsh D i c w and Suzanne Urbancqk, eds. Papers in Optimdity Theory. University of Massachusetts Occasional Papers in Linguistics 1 8. 77- 136.

Brosclonl, Ellen, Su-1 Chen and Marie Huffman. ( 1997). Syllable weight: convergence of p honology and phonetics. Phonology 1 4. 47-82.

Chen, Matthew ( 1987). The syntas of Xiamen tone sandhi- Phonology 4. 109-50.

Chomsky, Noam ( 1995). The Minimalist Progranr Cam bridge, M A : MIT Press.

Cohn, Abigail and John McCarthy (1994). Alignment and parallelism in Indonesian phonology. Ms., Comell University and University of Massachusetts. Amherst.

Crowhurst, M e p J. ( 19%). An optimal alternative to conîlation. Phonology 1 3. 409-24.

Crowhurst, Megan J. and Mark Hewitt (1995). Rosodic oi-erlay and Headless feet in Y idinS. Phonology 1 2. 39-84.

Dison, R. M. W. ( 1977). A Gramnar of Yidinv . Cam bridge: Cambridge Uni versi ty Press.

Doivning, Laura J. ( 1998). On the prosodic misalignment of onsetiess syllables. N a r a i iunguage and Linguistic Theory 1 6. 1 - 52.

Dreshcr, B. Elan and Aditi Lahiri ( 1991). The Gerrnanic foot metrical coherence in Old Engl is h. Linguistic Inquiry 2 2. 35 1-286.

Eck, A n o (1983,). Stress and a s s i a t e d phenomena: a sunreg with examples from Estonian. 1. Esronian Papers in Phonetics 1980- 1% 1, Academ y of Sciences of the Estonian SSR, Insti tute of Languages and Litentures. 20-58.

Gol dsmi th, John ( 1 990). Autosegnrerunl und Metrical Phomlogy. Oxford: Basil B lackwel 1 .

Green, Thomas ( 1995). The stress window in Pirahik a reanalysis of rhythm in Optimality Theory. Ms., MIT.

Green, Thomas and Michael Kcnstowicz ( 1995). The lape constraint. In Proceedings of FLSM VI, vol.1. Bloomington IN: Indiana University Linguistics Club.

Hal le, Morris ( 1490). Respecting metrical structure. Narural language and Linguistic Theory8. 149-176.

Halle, Morris and Michael Kenstowicz (1991). The free element condition and c y l i c vs. noncyclic stress. Linguistic 1nquir-y 2 2. 457-50 1 -

Halle, Morris and JeamRoger Vergnaud ( 1987). An Essay on Stress. Cambridge, MA: MIT Press.

Hammond, Michael ( l m ) . Metrical theory and learnabili ty. Ms., University of Arizona, Tucson.

Harms, Robert (1981). A backward metrical approach to Cairo Arabic stress. Linguistic Analysis 7. 429-50.

Harrell, Richard (1960). A iinguistic analysis of Egqptian Radio Arabic. In Charles A. Ferguson, cd., Contributions to Arabie Linguistics, Harvard Uni versi t y Press, Cam bridge. 3-77.

Haviland, J. (1979). Guugu Yimidhirr. In RM.W. Dixon & B..J Blake (eds). Handbook of Australiun Languages, Vol.1. Amsterdam: John Benjamins.

Hayes, Bruce ( 198 1). A nterncal t h e o v of stress rules. Ph.D. dissertation, MIT. Revised version distributed by Indiana University Linguistics Club. Published 1985, New Y or k: Garland.

Hayes, Bruce (1985). lambic and trochaic rhythm in stress d e s . In Proceedings of the Berkeley Linguistic Sociery 1 3. 429-446.

Hqes, Bruce (1987). A rwised parametric metrical theory. North-Emem Linguistics Society 1 3:429-446.

Hayes, B ruce ( 1995). Metrical Stress Theory: Principles and Case S td ie s . Chicago: Uni versi ty of Chicago Press.

Hew i tt, Mark ( 1 =). Vertical nauximization and mehical theoty . Ph. D. dissertation, Brandeis Univcrsi ty, Wal tham, Massachusetts.

Hewirt, Mark and Megan J. Crowhurst ( 19%). Conjunctive constraints and templates in Optimality T h e o ~ N a m l Language and Linguistic Theov 1 1. 38 1432.

Hint, Mati (1973). Eesti keele sonafonoloogia I. Tdi im , Estonia: Eesti NSV Teaduste A kadeemia

Hune, Henrietta ( 1994). The rhythmic and prosodic organization of edge constituents. Ph.D. dissertation, Brandeis University.

1 nkclas, Sharon ( 1989). Prosodic constituenq in the lexicon. Ph. D. dissertation, Stanford University.

Idsadi, William J. ( 1992). The contputation of prosody. Ph.D. dissertation, MIT.

1 tô, Jun ko ( 1 990). Prosodic minimali ty in Japanese. CLS 26, Papers fiont the Parasession on the SyIlable in Phonetics and Phonologu.

1 tô, J u d o and A m i n Mester ( lm-). Weak layering. and word binari ty. Ms., University of Cali fornia, San ta Cruz

Kager, René (1993). Alternatives to the iambic-troçhaic law. Narural Language and Linguistic Theory 1 1. 381-432.

Kagcr, René (19%). Stem disyllabicity in Guugu Yimidhirr. In Proceedings of the Holland lnstitute of Linguistics Phonology Conference 1994. Utrecht Uni\.ersiV.

Kager, René (l996b). Ternary rhythm in alignment theory. Ms. Utrecht University.

Kaye, Jonathan, Jean Lowenstarnm and Jean-Roger Vergnaud (1985). The intemal structure of phondogical elements: a theory of charm and govemment. Phonology Yearbook 2:305-328.

Kenstowicz, Michael J. ( 19%). On metrical constituents: unbalanceci trochees and degenerate feet. Jennifer Cole and Charles Kisseberth, eds. Frontiers in Phonology: Proceedings of the University of Illinois Conference on the Organiianon of Phonology, Features and Domaim. Center for the Study of Language and Information, Smfo rd , CA.

Key, Harold ( 1 % 1). Phonotactics of Cayuvava International Journal of American Linguistics 2 7. 143- 1 50.

Kiparsky, Paul (199-). Catalexis. Handout, University of California, Berkeley.

Lcer, Jeff. ( 198%). Prosody in Alutiiq. 1 n Krauss, ed., Yupik Eskirno prosodic systenzs: descriptive and comparative studies, rai rbanks: Alas ka Native Language Center. 77- 133.

Lccr, Jeff. ( 1985b). Toward a metrical interpretation of Y upik prosody. In Krauss, ed., Yupik Eskimo prosodic systenrs.: descriptive and comparative studies, Fairbanks: Alaska Native Language Center. 159-172.

Levin @levins], Juliette (1983). Reduplication and prosodic structure. Ms., MIT.

n [Blevins], Juliette ( 1985)- Generating temary feet. Texas Linguistic Forum 2 9. 97- 113.

n [Blevins], Juliette ( 1988). A metrical theop of syllabicity. Ph.D. dissertation, MIT.

Lm-in [Blcvins], Juliette ( 1989) The autonomy of the skeleton: evidence from Micronesian. Ms., University of Texas at Austin.

Levy, Mary ( 197 1 ). The plural of the noun in Modern Standard Arabie. Ph.D. dissertation, University of Mich gan.

Liberman, Mark and Alan Prince (1977). On stress and linguistic rhythm. Linguistic Inquiry 8. 249- 336.

Lorn bardi, Linda and John J. McCarthy ( 199 1 ). Prosodic circumscription in Choctaw morphology. Phonology 8. 37-7 1.

McCarthy , John. ( 1979). F o d problem in Semitic phonology and morphology. Ph. D. dissertation, MIT.

McCarthy, John ( 1995). Extensions of rai thfulness: Rotuman revisi ted. Ms., Univeni ty of Massachusetts, Amherst.

McCarthy. John and Alan Pnnce ( 1986). Prosodic morphology. Ms., University of Massachusetts, Amherst, and Brandcis University, Wal tharn, Mass.

McCarthy, John and Alan Rince (1988). Quantitative transfer in reduplicative and templatic morphology. In Linguistics in the Morning Calm 2, ed. Linguistic Society of Korea, 3 -35. Seoul: Hanshi n Pub1 ishing Co.

McCarthy, John and Alan Prinœ (l990a). Foot and word in R d i c Morphology: The Arabic broken plural. Natural Lunguage and Linguistic Theop 8. 209- 282.

McCarthy, John and Alan Prince (1990b). Frosodic Morpholoey and templatic morpholog y. 1 n Perspectives on Arabic Linguistics: Papers fiom the Second Symposium, ed. M. Eid and J- McCarthy, 1-54. Amsterdam: Benjamins, Amsterdam.

McCarthy, John and Alan Prince (1993a). P r d i c Morphology 1: Constra.int interaction and satisfaction. Ms., University of Massachusetts, Amherst, and Rutgers University. To appear, MIT Press.

McCarthy, John and Alan Prinœ (1993b). Generalized digrunent In Geert Bwi j & Jaap van Made (eds.), Yearbook of Morphlogy 1993. Dordrecht: Kiuwer. 79- 153.

McCarthy, John and Alan Prince (1994). The emergence of the unmarked: Optimality in Prosodic Morphology. In Mer& Gonzàlez (ed.), Proceedings of the North East Linguistic Society 24. Amherst, M A : Graduate Linguistic Student Association. 333 -379.

McCarthy, John and Alan Prince (1995). Faithfulness and reduplicative identitv. Ji11 Beckman, Laura Walsh Dickey and Suzanne U r h c z y k , eds. Papers in ~ ~ t . m a l i t ~ Theory . University of Massachusetts Occasional Papers in Linguistics 1 8. 249- 384. Also to appear in René Kager, Hamy van der Hulst, and Wim Z o ~ e v e l d , ed S., The Prosody Morphology Interjiace.

Mester, Ra1 f- Amin ( 1940). Patterns of tnincation. Linguistic Inqui- 2 1. 478- 485.

Michelson, Karïn ( 1989). invisibility: Vowels without a timing dot in Mohawk. In Donna Ge rd ts and Karen Michelson , eds. Theoretical Perspectives on Native American Languages. Albany, NY: SUNY Press.

Mitchell, T.F. ( 1963). Collquial Arabie: the living langmge of Egypt. London: O ~ o r d University Press.

Moore, John (19%). Doubled verbs and vowel association. In Perspectives on Arabic Linguistics: Papers from the Second Symposium, ed. M. Eid and J. McCarthy Amsterdam: Benjamins.

Mürk, Hamy W. ( 199 1 ). The structure and development of Estonian morplzology. Ph. D. dissertation, Indiana University. [Distributed by UMI Dissertation Senlices.]

Ncspor, Marina and Irene VogeI(1986). Prosodic Phonology. Dordrecht: Foris.

Orgun, C. Orhan ( 1994). Monotonic cyclicity and Optimality Th-. In Mercè G o d e z (ed. Proceedings of the 24th Annual Meeting of the North-& Linguistic Socieîy. Amherst, MA: Graduate Linguistic Student Association. 4 6 1 -474.

Paradis, Carole ( 1988a). On wnstraints and repair strategies. The Linguistic Review 6. 7 1 - 97.

Paradis, Carole ( 1 9t38b). Towards a theory of cons traint violations. McGill Working Papers in Linguistics 5. 1 -44.

Pater, Joe (19%). Comequences of consfraint ranking. Ph.D. dissertation, McGill University.

Piggott, G. L. ( 1992). Satisfying the minimal word. Ms.. McGiil University, Montrkal.

Pi ggott, G. L. ( 1995). Epenthesis and syllable weight. Natural Language and Linguistic Theoty 1 3. 283 - 326.

Poser, William J. ( 1984). Hypocoristic formation in Japanese. WCCFL 3. 318-29.

Poser, William J. ( 1990). Esidence for Foot Structure in Japanese. Languuge 6 6. 78- 105.

Potter, Brian ( 1994). Seriai Optimality in Mohawk prosod y. Proceedings of the Chicago Linguistic Society 3 0.

Pnnce, .4lan S. ( 1980). A metrical theon; for Estonian quantity. Linguistic Inquiry 1 1. 51 1-56?.

Piince, Alan S. ( 1983). Relating to the grid. Linguistic Inquiry 1 4. 19- 100.

Pn ncc, A lan S. ( 1 Wû). Quantitative consequences of rh>-thmic organization. Proceedings of the Chicago Linguistic Society 26- I I . 353-98.

Prince, Alan and Paul Smolenshy (1993). Optimaîity theory: consiraint interaction in generative grmmr . T o appcar, MIT Press. TR-2, Rutgers University Copni tive Science Center.

Ri ce, Curtis ( 1 =). Binarity and temriiy in niem'cal theory: parametn'c extensions. Ph. D. dissertation, University of Texas at Austin.

183

Rice, Curtis ( 1993). Metrical coherence in Chugach. Ms., University of Trondheim.

Selkirk, Elisabeth 0. (1980). The rôle of prosodic categories in English word stress. Linguistic Inquiry 1 1 . 563-605.

Selkirk, EIisabeth 0. ( 1984). Phonology and Syntax: The Relation between Sound and Structure. Cambridge, MA: MIT Press.

Selkirk, Elisabeth 0. (1986). On derived domains in sentence phonology. Phonology Yearbook 3 . 37 1-405.

Selkirk, Eiisabeth 0. (1995). The p r d i c stmcture of functional elements: affixes, clitics, and words. In Ji11 Beckman, Laura Walsh Dickey and Suzanne Urbanczyk, eds. Papers in Optimal ity Th-. University of Massachusetts Occasional Papers in Linguistics 1 8. 439470.

Selkirk, Elisabeth O. and Koichi Tateishi (1988). Minor phrase formation in Japanese. Proceedings of the Chicago Linguistic Society 2 4. 3 16-36.

Selkirk, Elisabeth O. and Tong Shen ( lm). Prosodic domains in Shanghai Chinese. In Sharon Inkelas and Draga Zec, eds. The S y n t d h o n o l o g y Connection. CLSI: Uni versi ty of Chi gaco Press, 3 13-3&470.

Sloan, Kelly D. (1991). Syllables and templates: evidenœ [rom Southern S i e n Miwok. Ph.D. dissertation, MIT.

Smeaton, B.H. ( 1973). Lexical l3pansion Due tu Technical Change. Bloomington, Ind: Indiana University Press.

Spting, Cari (1990). Implications of Axininca Campa for prosodic morphology and reduplicarion. Ph. D. dissertation, University of Arizona, Tucson.

Sprouse, R. (1996). Vowels chat borrow moras: geminates and weight in OT. North- Eastern Linguistics Society 2 6. 393 - 408.

Stcriade, Donca (1988). Reduplication and syllable transfer in Sanskrit and elsewhere. Ms., UCLA.

Suzuh , Hisami ( 1995). Minimal words in Japanese. Proceedings of the Chicago Linguistic Society 3 1: Main Session.

Tauli, Vdter ( 1954). The origin of the quantitative system in Estonian. Journal the la société finno-ougrienne 5 7 . 1 - 19.

Tauli, Vaiter (l-), On quanti- and stress in Estonian. Acta LiBguistica Hafiiiema 9 , 145- 162.

Urbanczyk, Suzanne (1994). Double reduplications in parallel. In Ji11 Beckman et al., eds. 1995. Also to appear in R e d Kager, Ham; van der Hulst, and Wim anneveld, eds. The Prosody Morphology Interface.

Urbanczyk, Suzanne (1995). The case of the rnigrating suffis: Out-of-control in Saiish. Ms,, University of Massachusetts, Amherst.

Walker, Rachel (1994). Buriat syllable weight and head prominence. f h o m l o ~ ut Santa Cruz 3. 99-1 10.

Yip, Moira 1W-. Reduplication and C-V skeleta in Chinese secret languages. Linguistic Inquiry 13, 637-661.

Yip, Moira 1988. Template morphologv and the direction of association. Naturai h g u a g e and Linguistic Theory 6 , 55 1 - 577.

Yi p, Moira ( 1992). On quanti tu-insensitive languages. Ms., Univcrsi ty of California, Inrine.