c o "'0 oC --l ~ ..:::L ... o >­ ~ cu (6)l~ c :c '" ~"~ Nl.WYt'l~~~!~~" 11·••••••••• _

DEDICATION Copyright © 1988 Stuart Davis All Rights Reserved To my parents and brothers.

Library of Congress Cataloging-in.publication Data

Davis, Stuart. Topics in syllable geometry JStuart Davis. p. em. - (Outstanding dissertations in linguistics) Originally presented as the author's thesis (ph. D. -University of Arizona, 1985) 8ibliography: p. ISBN 0-8140-5181-5 1. Grammar, Comparative and general-Syllable. I. Rhyme. 3. Accents and accentuation. I. Title. II. Series. PI36.D38 1988 414-dc19 88--16509

Printed on acid-free, ISO-year.life paper Manufactured in the United States of America ACKNOWLEDGEMENTS

I would like to thank several people for making the completion of this dissertation possible. I thank the three members of my cOmMittee, Dick Demers, Dick

Oehrle, and Deirdre Wheeler, as well as Richard Janda and Natsuko Tsujimura for their valuable comments and encouragement. Much of the dissertation was shaped through discussion with them. I would also like to thank Susan Steele and Ann Farmer for their input on earlier versions of Chapter two. Furthermore,

I offer thanks to the Graduate College of the Univer­ sity of Arizona and to Sigma xi for financial support. r~ NllNY;;k~~!!~D" Ill••••••••••••••••

TABLE OF CONTENTS

Page

CHAPTER

1. OVERVIEW 1

1.0. Introduction 1 1.1. Brief Background of the Syllable in Phonological Theory . • . . . . 1 1.2. Background on Syllable-Internal Constituency ...... 8

2. ON THE ARGUMENTS FOR THE RHYME AS A SUB SYLLABIC CONSTITUENT 17

2.0. Introduction 17 2.1. Phonotactics 19 2.1.0. Introduction 19 2.1.1. Constraints between Onset and Other Syllable Parts for English ...... • . 23 2.1.2. Constraints between Onset and Nucleus for Other Languages . 27 2.1.3. Constraints between Onset and Coda for Other Languages 29 2.1.4. Conclusion . 30 2.2. Rule Types Referring to the Rhyme 31 2.2.1. Stress Rules .....• 32 2.2.2. Phonetic-Spreading Rules 46 2.2.3. Compensatory Lengthening Rules ...... 48 2.2.4. Conclusion . 55 2.3. The Rhyme as a Durational Unit 56 2.3.1. Background on Durational Experiments of VC-Sequences 58 2.3.2. English •.••.• 64 2.3.3. Korean •..... 68 2.3.4. French and Russian 70 2.3.5. Other Languages 72 2.3.6. Surrnnary 74 2.4. Conclusion ...... 75 ...... - N!W 'i'n.... ••':;-:'::::.""" ,,------­

TABLE OF CONTENTS--Continued CHAPTER 1

Page OVERVIEW

S~LABLE STRUCTURE AND STRESS RULES FROM DIFFERENT LANGUAGES 79 1.0. Introduction

The goal of this cha~ter is to furnish back­ 3 • 0 • Introduction . 79 3.1. Standard Stress Systems 82 3.1.0. Introduction •. 82 ground information on the role of the syllable in recent 3.1.1. Same Syllab:e Stress S3 3.1.2. Long Vowel Stress B5 phonologlcal theory, and, then. to put my thesis in proper 3.1.3. Closed Syllable stresS 87 3.1.4. Heavy syllable Stress B9 perspective by presentin9 vdxious views on the nature of Oncet-Sensi~ive Stress Languages 97 3. 2. ~yllable-internal con~tituency. 3.2.0. IntrOductio~. 97 In this dissertation, 3.2.1. Western Aranda 97 3.2.2. Alyawarra 107 : arque ~hat none of these various Views provides an 3.2.3. Madimadi :16 3.2.4. Piraha' 127 adequate representation of the syllable. 1 prOpose that 3.3. Conclusion •• 132 the syllable universally consists ot three (sister) "MOVEMENT" PHENOMENA AS EVIDENCE FOR SYLLABLE-INTBRNAL CONSTITUENCY 13' components: an onset (o~tional), a nucleus, and a coda

4.0 • Introduction • • • • 138 (optional) • 4 • 1 • Speech Errors .,. 139 4.1.0. Introduction 139 4.1.1. Some Problems With the Use of 1.1. Brief Background of the Syllable speech Errors as E~idence 143 in Phonolosical Theory 4.1.2. Speech Error Evidence and the Rhyme • • • • • • • • • • 146 Much recent work in phonology has focused on the 4.:.3. Speech Error E~idence and sUbsy11abic Constituency 151 ntudy Of the syllable. In the early years of generative 4.1.4. Conclusion. 161 4.2. Lan9uage Games • • • • • • • • • 162 phonology, though, the role of the syllable in phonolo­ 4.2.0. Introduction •.•••• 162 4.2.1. Typology of Language Games 163 qiC~l description was largely ignored. one reason for 4.2.1.1. Expunsion 163 4.2.1.2. Contraction 167 this was that the structuralists never l>ucceeded il1 4.2_1.3. substitution 168 4.2.1.4. Rearrangement 169 ~aoqudtely defil1ing the syllable as a physical ~eality. 4.2.1.5. Combinations •• 173 4.2.2. Language Games and subsy11abic '!/(ttnon 11951:21. for e~ample, attempted to define the Constituency 176 4 • 3 • Concl\:sion 184 !lVl J.'lblo Physically by associating it with a chest pulse

1ST OF REFERENCES . . . 185 1 2 3 induced by muscular activity. "careful experimentation consonant clusters do not break down into a combination with all types of syllables and consonants makes it cer­ of a ~ossible word-final cluster followed by a word­ l tain that every syllable has its chest pulse delimited initial cluster. by chest muscles (intercostals) or by the constriction Even though the syllable cannot be defined

(complete or partial) of the consonant or by both_" physically, it can still function as a relevant phonolo­

Contrary to Stetson's experiments, Ladefoged (1967) has gical unit. And, in fact, significant phonological shown that chest pulses are not always an accurate generalizations are missed when phonological rules are diagnostic for the syllable: some syllables are produced not allowed to refer to syllables. Por example, it has without a chest pulse, and some chest pulses do not been noted by Kahn (1976) and others that many rules correspond to syllables. Jespersen, as cited in Pulgram that are written with the environment {i} miss the (1970) defined the syllable in terms of vocalic sonority; generalization that the environment for the rules is in other words, te believed that sounds grouped themselves actually a syllable boundary. Another case of a missed into syllables by their sonority. However, this view has qeneralization, discussed both by Vennemann (1972) and also been criticized (e.g., in Ladefoged 1975) since Wheeler (1981) among others, relates to Chomsky and sonority can either increase or decrease before reaching llJlle's (1968) formulation of the main stress rule in the syllabic peak - e.g., in the word ££l there is an ~nglish. Main stress for nouns and certain adjectives in increase in sonority between the Ikl and the /r/, but J<;nglish is usually assigned to a penultimate vowel if it in the word ~ there is a decrease in sonority between tn tensed or followed by at least two consonants; the lsi and the /p/. Yet both words constitute a single syllable. Haugen (1956) and others defined the syllable 1. Not all structuralists, however, accepted Ilw oyllable. For example, Malmberg (1955) cites a number in terms of distribution of phonemes. In other words, fI(1 ,1.inguists who view the syllable as not being signifi­ I~nt since no physical correlate of it could be found. they viewed the syllable as a unit over which phonotactic !ilmJ,larly, Kohler (1966) conclUdes that the syllable is "1mpQusible" and "harmful. II He argues this on the basis constraints hold. This view, though, has been criticized 1/1' the claim that syllable division is often not discer­ lIn)J10 and is arbitrary. For further discussion of the by Bell (1976) for several reasons. One reason is that (I I ff'oring views on the phonetic reality of the syllable rH1f1 1)u19ram (1970) and Vogel (1977). such a view is unable to handle languages in which medial ~ '':~

5 4 Another example of a generalization that is missed utherwise, the antepenultimate vowel is stressed. The wlthout reference to the syllable is found in Hooper's oxnmples in (1) illustrate this:

/ (1972) discussion of Harris' (1969) analysis of Spanish (1) Ariz6na ag~nda vigilant nasalization. Harris observes that nasals assimilate llowever, note the following exceptions: before adjacent obstruents in the same word or across (2) discipline IGdicrous ~loquent word boundaries as is shown below (examples from Hooper Exceptions, like those in (2), led Chomsky and Halle to formulate the concept of "weak cluster" which is essen­ 1972:525) : [umbeso] 'a kiss' tially defined as in (3): (4) un beso [uncarko] 'a pool' (3) Weak cluster un charco [u'jgato] 'a cat' v Cl un gato o ['" voc ] [-tense] rA cons Harris also observes that, before glides, nasal assimi­

They then refer to the notion of weak cluster in their lation takes place only across word boundaries and not main stress rule (as well as in four other rules ­ word internally, as the two examples in (5) illustrate: Auxiliary Reduction, Pre-Cluster Laxing, u-Tensing, and (5) a. miel [myel] 'honey' Tensing before CIV). But by not recognizing the syllable b. un hielo [unyelo] 'an ice' Chomsky and Halle miss a generalization. Weak clusters Hooper notes that in Harris' system the rules for the are just those clusters that are possible syllable-initial two cases of nasal assimilation (in (4) and (5b) above) clusters. That is, both their members are syllabified cannot be collapsed, although they describe the same with the following (ultimate) vowel, so that the penulti­ process. Hooper argues, however, that once the syllable mate vowel is in an open syllable. Thus, in the data is recognized (and it is formally recognized by syllable above, stress is on the antepenultimate vowel when the boundary insertion rules), an obvious generalization penultimate vowel is both lax and in an open syllable. follows: Nasal assimilation occurs before a consonant

This generalization was missed by Chomsky and Halle or a glide if a syllable boundary intervenes. because they did not give any status to the syllable. 7 6

Similarly, Hoard (1971) recognizes the syllable support for Kahn's (1976) proposal to treat the syllable in his analysis of the different realizations of the autosegrnentally. In the autosegrnental treatment, the intervocalic ItI in the two words veto and motto. In the syllable constitutes a distinct tier onto which the former the ItI is aspirated (or tensed) but is not elements from the segmental tier are mapped; it is flapped; in the latter, it is flapped but not aspirated. then possible for a consonant to become arnbisyllabic by

Hoard (p.137) accounts for the different realizations of having the consonant linking to more than one syllable

ItI by first having a syllable boundary insertion rule as long as the anti-crossing lines constraint (Goldsmith (as in (6) which has the affect of putting the boundary 1976) is not violated. Specifically, Kahn (1976) has before the ItI in veto but after the ItI in motto, and phonemes first linking to the syllable tier (by his rules afterwards, having a rule (as in (7)) which basically I and II, p.22 and 24, respectively), and then, only tenses a consonant before a stressed vowel (M=maximal under certain conditions (by rules III and IV, p.28), can 2 initial cluster and [ • ]=syllable boundary). some consonants (optional10 become arnbisyllabic. Thus, [+SYll] the advantage of Kahn's theory is that it allows for (6) rIJ ---~ [ . ] I [+syll] Co __

(i.e., it can be ambisyllabic). Ambisyllabicity provides

....AL 9 8

1.2. Background on Syllable-Internal Constituency postnuclear margin (containing syllable-final 4 While the importance of the syllable for phono­ consonants). logical theory has become widely recognized, the issue Hockett has been credited (by Haugen 1956) with ~ of the type of constituent structure below the syllable inventing the terms (syllable-initial consonantism),

(or, rather, the nature of syllable-internal consti­ peak (vowel or syllabic consonantl, and coda (syllable­ tuents) has become controversial. Interestingly, not final consonantism) to describe the subconstituents of only has it become recently controversial, but it was the syllable. Hockett argued that the syllable could also controversial in the structuralist period, as well. contain another element in addition to these three. This

Pike and Pike (1947) made one of the earliest other element was called the interlude, which he defined

(if not the earliest) proposals for dividing the syllable as follows (1955:52): "An interlude is coda-like and into various subsy11abic constituents. They divided the onset-like at the same time, and structurally it belongs

Mazatec syllable into a margin (syllable-initial conso­ to the syllable that contains the preceding peak and to nant or consonants) and a nucleus (from one to three that which contains the following peak." Further, when there is an interlude, there is no syllable division vowels in Mazatec). Pike and Pike proposed this division .~ {z for two reasons: First, the pitch contours on vowels between coda and onset. This is because Hockett proposed

(i.e., in the nucleus) form contrasts in such a way as the interlude in order to handle arnbisyllabicity. He to differentiate meaning, while such is not the case cited such examples as 'nitrate', where the first /t/ with consonants; and second, phonotactic constraints can go with either syllable, and, thus, claimed (1955:64) hold within the elements of the nucleus and within elements of the margin. In later work, Pike (1967) 4. pike and Pike (1947) are sometimes mentioned as proposing a syllable in which the nucleus and the further divided the margin into a pre-nuclear margin post-nuclear margin form one constituent and the pre­ * nuclear margin the other. However, this is not proposed (containing syllable-initial consonants) and a in their article. What they do propose is that within the nucleus or within the margin there is a hierarchical structure. Thus, if a margin has three consonants, two of the consonants would have a tighter bond between them as against the third one.

.~ 11 10 that "syllable division in an interlude is structurally postulated the interlude as a constituent to which

irrelevant." Essentially, the interlude would be the Ambisyllabic consonants belonged.

name of the constituent where coda and onset are simul­ In most of the structuralist work on sub-syllabic

taneous - since, because of ambisyllabicity, the division constituents, there are no claims about possible hierar­

between coda and onset cannot be pinpointed. chical relationships among these constituents. It is

Haugen (1956:218) criticized Hockett's proposal not usually discussed, for example, whether any two of

regarding the interlude for the following reason: "If the sub-syllabic constituents form a unit together, or we are to make use of the syllable as an Ie [immediate whether all the subsyllabic constituents are independent

constituent] of the macrosegrnent it does not make good of one another. Pike (1967:386), though, did suggest

sense to leave the syllables all attached to each other the possibility of the nucleus and the post-nuclear

by indivisible segments." Haugen argued (p.220) that margin forming a constituent (e.g., in a tone language

non-overlapping syllble division was important, for he if the contrastive tone extends over these) :

saw, as mentioned above, "the syllable as a unit of ••• there may be postulated a dichotomous break between the nucleus on the one hand and the phonotactic structure." Thus, the coda and the onset had two margins on the other (if the nucleus is somehow independent of the two margins), or a to be kept separate. He also argued that one could dichotomous break between the pre-nuclear margin versus the nucleus and the post-nu always separate coda and onset, because the syllable clear margin in a language in which the domain of significant pitch extends over the post­ division usually does not permit the formulation of a nuclear margin as well as the nucleus.

new coda or a new onset (Haugen 1956:218-219): "In Pike's work on syllable structure had little

'nitrate' a division /naytr. eyt/ would introduce a non­ influence at first on generative phonologists because

existent final cluster -tr; but the division /nay. treyt/ of their belief that the syllable played no role in

or /nayt. reyt/ would both fit with the existing positions phonology. However, because of the evidence for the

and their members." Essentially, then, Haugen maintained syllable in Kahn's (1976) dissertation, the importance

that the syllable constituents consisted of just onset, of the syllable for phonological theory has largely

peak (or nucleus) and coda, while Hockett further t been recognized. Kahn, though, viewed the syllable as having no internal structure; the syllable branched into

~ 13 12 (9) Rhyme-Structure Analysis a otring of segments, as in (8a). Kahn's view of syl­ a. syllable b. syllable luble-internal structure is also held by Clements and ~ Keysor (1983) with some modification. Clements and e onset rhyme

Kayser propose that a CV-tier intervene between the onset/7:: nuc. coda syllable tier and the segmental tier (as in (8b». Both Selkirk (1978) and Halle and vergnaud (1980) among

(8) The Segmental Hypothesis many others essentially claim that the rhyme is a

a. syllable b. syllable universal in the strong sense; that is, all syllables ~ ~ in all languages contain them. Halle & Vergnaud's p r I n t C C V C C I I I I I characterization of the rhyme, tnough, differs from that p r I n t of Selkirk's in that they do not view the rhyme as Clements and Keyser recognize the nucleus as a consti­ possessing labelled constituents; for them, the syllable tuent since elements in the nucleus can be displayed consists of just an onset and a rhyme, as in (9b). or projected. (Throughout the remainder of this disser­ Another view of subsyllabic constituency is that tation the syllable structures in (8) will be referred the syllable contains a body (a constituent that consists to as the single-segment hypothesis or single-segment of an onset and a nucleus - after Vennemann 1984). A analysis.) syllable containing body structure is illustrated in (10): Another view of subsyllabic structure analyzes (10) Body-Structure Analysis the syllable as consisting of an onset (the syllable­ syllable initial consonant or consonants), a nucleus (the peak of sonority within the syllable), and a coda (the syllable­ A~ final consonant or consonants). The nucleus and the onset nuc. coda coda, in turn, are obligatorily grouped together to form /\ I /\ p r I n t a subsyllabic constituent, the "rhyme" (or rime), as A number of linguists have suggested that the syllable in (9a) (nuc.=nucleus). could have the structure in (10). McCarthy (1976) viewed

the syllable as either possessing the structure of (9) 14 15 or of (10). Specifically, he argued that both (9) and onset, the nucleus, and the coda), as in (l1): 5

(10) are possible syllable types in Estonian. There, (11) Level Syllable Structure

if the coda is an obstruent, syllable structure is as syllable in (9) because the nucleus and the coda form a "tight ons~oda phonetic unit." But, if the coda is a sonorant, syllable In chapter two of this dissertation I argue in

structure is as in (10) since there would be no tight detail against the proposal of Selkirk (1978) and Halle &

phonetic relationship between nucleus and coda. Bach Vergnaud (1980) among others that the rhyme is an obliga­

and Wheeler (1981) have argued that (10) reflects tory (universal) syllable constituent. In chapter three

syllable structure in Korean because of the phonotactic I discuss several stress rules in the metrical framework

constraints that exist in that language between the incorporating the different syllable structures in (9-11),

onset and the nucleus. More recently, Vennemann (1984) and I show that it is only an analysis incorporating a

has argued that (10) is a possible syllable structure. level syllable structure that is compatible with all the

He argues that syllables do not have any structure per different types of stress rules to be considered.

~, but structure is imposed upon them by language Finally, in chapter four, I look at phonological "move­

specific phonological rules and processes. Thus, in ment" phenomena and consider what type of syllable inter­ his view, if a rule refers to elements within the onset nal constituent structure it supports. Specifically, I and the nucleus then syllables involved in the rule (or argue that evidence from speech errors and language games process) would have the structure in (10).

In this dissertation, though, I will argue that 5. I do not consider other different proposals for syllable-internal structure. Among those not none of the structures in (8-10) provide an adequate considered are the elaborated rhyme structure and onset structure proposals of Cairns and Feinstein (1982), the representation of the syllable. Instead, I argue that N-bar syllable structure proposal of Levin (1984) in which syllable structure parallels X-bar syntax, and there is only a single syllable structure, and that is the proposal of Hyman (1984) in which an essentially moraic analysis of the syllable is presented. There a (rhymeless) structure in which the syllable internal is also the view that at least for some languages the syllable is not relevant whatsoever. Higurashi (1983) constituents are sisters (i.e., consisting of just the has proposed this for Japanese and Hyman (1982) has argued thus for Gokana.

---- 16 are most compatible with the syllable structure in (11) CHAPTER 2 and thus provides additional evidence for syllable structure consisting of just an onset, a nucleus, and ON THE ARGUMENTS FOR THE RHYME AS A SUBSYLLABIC CONSTITUENT a coda.

2.0. Introduction

In this chapter, I challenge the widespread

\jtJt'ion that the rhyme is an obligatory constituent in a

I IH'Qry of syllable structure. According to the most

""norally held view the nucleus and the coda are analyzed

.\1\ forming a single constituent, the rhyme; nucleus and

•-,lila thus together have a privileged status. But here

I I'l'view three major arguments that are often cited as

Illtlluating the universality of this privileged status

1,,'1 woon nucleus and coda, and, after summarizing each

·~I 'Imnont I show that it is invalid when a wider range of

"" 1dc'nce is considered. The first argument to be discussed is found in

qulklrk (1978) and Vergnaud & Halle (1978). They argue

Foil Lila rhyme constituent by appealing to the existence

"r pllonotactic constraints holding between nucleus and

".,,1.1. However, this argument for the rhyme, which is

I All "II to establish it as a universal, makes an implicit

1'1 u, II /' tion that can be shown to be f alse (namely, that

tl" ,dl,motactic constraints occur between onset and nucleus

17 16 are most compatible with the syllable structure in (11)

and thus provides additional evidence for syllable CHAPTER 2

structure consisting of just an onset, a nucleus, and ON THE ARGUMENTS FOR THE RHYME a coda. AS A SUBSYLLABIC CONSTITUENT

2.0. Introduction

In this chapter, I challenge the widespread

notion that the rhyme is an obligatory constituent in a

theory of syllable structure. According to the most

generally held view the nucleus and the coda are analyzed

as forming a single constituent, the rhyme; nucleus and

coda thus together have a privileged status. But here

I review three major arguments that are often cited as

indicating the universality of this privileged status

between nucleus and coda, and, after summarizing each

argument I show that it is invalid when a wider range of

evidence is considered.

The first argument to be discussed is found in

Selkirk (1978) and Vergnaud & Halle (1978). They argue

for the rhyme constituent by appealing to the existence

of phonotactic constraints holding between nucleus and

coda. However, this argument for the rhyme, which is

taken to establish it as a universal, makes an implicit

prediction that can be shown to be false (namely, that

no phonotactic constraints occur between onset and nucleus

17 :if) 18 none of the major arguments adduced for the constituonoy or onset and coda).· The second argument for the rhyme 1 of the rhyme actually supports it. derives from the observation that certain rule types only make reference to elements in the nucleus and coda. An 2.1. Phonotactics example of a rule type to be discussed here are stress

assignment rules which are most commonly mentioned as 2.1.0. Introduction supporting the constituency of the rhyme. Halle & Selkirk argues for the universality of the rhymo

Vergnaud (1980:93) argue for its constituency by claiming on the basis of the existence of phonotactic restric­

that " ••• in all languages known to us, stress assignment tions between nucleus and coda. She argues (1978:5) that:

rules are sensitive to the structure of the syllable The grouping of peak and coda into a constituent is advocated as a universal of syllable compo­ rime, but disregard completely the character of the sition•••• The claim made is that cooccurrence restrictions between peak and coda are always onset." However, it can be shown that there are in fact more likely to exist (and indeed are quite common) than restrictions between either peak languages where the structure of the onset is important or coda and the onset. The explanation offered is that the former two comprise a constituent. for assigning stress. Thus, stress assignment rules do Vergnaud & Halle (1978:41) employ a similar argument not provide evidence for the universality of the rhyme. for justifying the constituency of the rhyme in English: A third argument which Selkirk (1978) gives to support "Whereas practically any onset can combine with any rime the constituency of the rhyme is the claim that nucleus to form a proper English syllable, there are severe and coda form durational units. However, a closer reading limitations on what peak can precede what coda." The of the phonetics literature, as well as my own investi­

gations using spectrographic evidence, reveals that the 1. Another often-mentioned argument for the durational relationship here is between a vowel and a subsyllabic constituent rhyme relates to the poetic notion "rhyme". It has been claimed that, in rhyming following consonant (regardless of tautosyllabicity) and words, only the onset changes, but the rhyme remains the same. Clements & Keyser (1983:24) point out, how­ not between nucleus and coda. Thus the relative duration ever, that the poetic rhyme is actually sensitive to tho of segments offers no evidence for the rhyme. Hence, stressed syllable plus everything to the right of it (e.g., slnister:mfnister). It is not equivalent to the subsyllabic constituent rhyme, and consequently, it does not constitute an argument for the rhyme constitutimL. 21 20 reason for this, so their argument goes, is that the that many other languages also possess phonotactic con­ nucleus and the coda form a constituent. otraints holding between constituents other than nucleus 3 It would be predicted from the above arguments and coda. Consequently, the existence of such con­ that any phonotactic restrictions existing between the otraints in many languages militates against the use of onset, on the one hand, and nucleus or coda, on the phonotactics to argue for the rhyme. other hand, must be highly irregular and infrequent; such However, before presenting the phonotactic con­ constraints would not be considered to be systematic, but straints holding between the onset and the other parts rather, only accidental gaps in the phonotactics. of the syllable, I must first make clear the view that

However, there is strong evidence that this prediction I am adopting concerning the location of syllabifica­ is incorrect, since there are systematic phonotactic tion in the phonology. I am accepting the view of Selkirk constraints between other position slots than nucleus (1981), McCarthy (1982a) and Clements and Keyser (1983) and coda. Recognition of such constraints eluded among others that syllabification is present in underlying structuralist phonologists because they concentrated on representation. Probably the strongest case for the con­ constraints holding between adjacent slots, while trary view has been put forth by Broselow (1979). She neglecting to look for systematic constraints holding proposes that syllabification occurs late in the phonolo­ between nonadjacent slots. 2 But I have shown previously gical derivation. She argues for this view based on

(e.g., Davis 1984), that a number of such restrictions hold in English. In the following subsections I first 3. The manner in which phonotactic constraints are expressed is somewhat controversial. The view adopted present these phonotactic restrictions, and then show here is that of Stanley (1967) who essentially proposed that such restrictions are expressed either by positive conditions, negative conditions, or if-then statements. As for the question at what level such restrictions apply, 2. Twaddell (1939, 1940), though, drew con­ the view of Shibatani (1973) is basically adopted here; clusions concerning the cooccurrence frequencies of that is, constraints apply at both the underlying level pre- and post-vocalic consonants of stressed German and the phonetic one. However, constraints at the two syllables. He also cited a study by Jespersen of mono­ levels may differ in view of the fact that there are syllabic words for English in which Jespersen basically some well-motivated cases of positing abstract underlying stated that any constraints holding between syllable­ representations, and there are cases of fast speech rules initial and syllable-final consonants in English are that produce surface outputs that do not occur accidental. underlyingly. 23 22 evidence from Cairene Arabic where syllabification occurs Selkirk (1981) has argued that syllabification does across word boundaries (i.e., the domain of syllabifi­ indeed occur at the underlying level in Cairene Arabic, cation is the phrase, not the word). For example, the contra Broselow. Selkirk shows how the epenthesis and

Cairene phrase mi~ ('not') ~ ('I') is syllabified deletion facts are elegantly handled by allowing an mi.~a.na (where the dot represents a syllable boundary), empty node to be incorporated in the underlying syllabi­ and the phrase bint ('girl') kibiira ('big') is realized fication of Cairene words. On the phonetic level, then, as bin.tik. bii.ra, after an epenthesis and a deletion either an epenthetic vowel fills the empty node, or (if rule have applied. Consequently, syllabification takes the node is word-final) an initial vowel from an imme­ place late in the derivation after rules of epenthesis diately following vowel-initial word fills the node. and deletion. However, her proposal cannot handle data ThUS, syllabification always occurs on the underlying from the dialect of Palestinian Arabic described by representation, and, in cases like Cairene Arabic empty

Abu Salim (1980). In this dialect syllabification must nodes are incorporated. In the remainder of this disser­ precede epenthesis. This is because the stress rule tation, then, I, too, will accept the view that

(which must apply after syllabification since stress syllabification occurs on underlying strings. is sensitive to the makeup of a syllable) does not 2.1.1. Constraints between Onset and Other Syllable (normally) put stress on an epenthetic penultimate vowel Parts for English that is part of a heavy syllable (other heavy penultimate English has a number of constraints that hold syllables receive stress--unless the final syllable is between onset and coda, as well as some that hold between superheavy). For example, the word [?fbin-ha] ('her onset and nucleus. The only mention of these in the son'), from the underlying form I?ibn-ha/, has stress on literature (besides that in my own work) is by Fudge the initial syllable in spite of the fact that the (1969) and Clements & Keyser (1983). Fudge (1969:279­ epenthetic [i] has made the penultimate syllable heavy. 181) observes that no syllable has sn in the onset and

Thus, stress here must apply prior to epenthesis, and, an 1 initial cluster in the coda. He also observes consequently, syllabification occurs at the underlying that the second slot of an onset cannot be the same as representation prior to the rule of epenthesis. Moreover, the first slot of the coda. (Actually, there are a 25 24 (1) Constraints Holding Within the Syllable number of exceptions to this: state, stat, and stet. I a. *s [-cont ] [-cons] -cont ] state the relevant constraint later.) Moreover, he says -cor +syl -cor 0<. ant oC.ant that syllables do not both begin and end with an s + stop [ ~ voice .~ voice cluster (though, this is part of a broader generaliza­ b. * [+cons] [+cons] -cons] [+cons] -son +lat [ +syl +lat tion to be expressed by my constraint (2a)), and, finally, c. * [+cons] [+nas] -consl ( [+cons]) +cons ] he mentions that no word has a nasal in both the second [+syl [ +nasal slot of the onset and the second slot in the coda. By (la), if a [-coronal] stop follows an lsI in a sylla­

Further, Clements & Keyser (1983:20) have noticed the ble's onset, then that same consonant cannot be initial in following two constraints for English that hold between the syllable's coda. Thus forms like *~ or *skak are onset and nucleus: "Voiced fricatives and IC11 clusters not possible. By (lb), if III is the second consonant are excluded before lUI ••• Stop plus Iwl clusters are of an onset cluster, then III cannot be the initial con­ excluded before lu:, U, " , awl: *kwu:t, etc." Interes­ sonant of a coda. Thus, a form like *slil is not tingly, this last constraint partly follows from the fact possible. And, by (lc), if a nasal occurs in the second that Iwl plus high back vowel sequences are rare in slot of an onset no nasal occurs in the coda. Thus English. As I will mention again later, Kawasaki & forms like *~ or *smin are impossible. The second

Ohala (1980) point out that such a constraint is common constraint has also been noted independently by Clements in the world's languages, and can be understood accousti­ & Keyser (1983:20). Besides these general conditions on cally, since the transition between a Iwl and a high syllable shapes, English also possesses at least two back vowel is not distinct. onset-coda constraints that only hold within a mono­

Probably the most extensive collection of onset morphemic syllable. These are given below: coda constraints holding in English is given in my own (2) Constraints Holding Within the Monomorphemic Syllable previous work (Davis 1982, 1984). For example, I have a. *C.C.VC.C noted the following exceptionless negative constraints ~ J ~ holding within the syllable: 26 27 b. [+cons] * +cons] -cons] [+cons] pattern in final position (with the parenthesis meaning +son [+son +son [ -nas -nas nonoccurring) : al..lat ol-lat -lp- -It- -1<:- -lk­ Constraint (2a) disallows monomorphemic syllables in -lb- -ld- -13'- (-lg-) which the initial consonant of an onset cluster is the Moreover, in final position /r/ can occur with any stop same as the initial consonant of the coda cluster. Thus consonant after it, including /g/ (e.g., in morgue). monomorphemic forms like *fraft are impossible, but These facts suggest the lack of final ~ is truly acci­ spaced (/speyst/) is possible since it is not mono- dental; /1/ does not pattern like /r/ in this regard, and morphemic. (Two exceptions to constraint (2a) are the /g/ does not pattern like any of the other stop conso­ words ~ and klux - unless the sound represented by nants. On the other hand, the constraints proposed in orthographic ~ actually represents a single affricate (1-5) do make reference to natural classes; hence, they [ks ] and not the sequence [ks].) Constraint (2b) rules are not arbitrary. Furthermore, systematic constraints out coda clusters beginning with a liquid (/r/ or /1/) holding between onset and nucleus and between onset and when that same liquid occupies the only slot of an onset. coda are also found in other languages. In the followinq Thus, monomorphemic forms like *rorp cannot occur, but a sections, I list some of these. nonmonomorphemic form like roared (/rowrd/) is permitted.

The only exception to constraint (2b) is the word lilt. 2.1.2. Constraints Between Onset and Nucleus for Other Languages Contrary to what Selkirk (1978) and Vergnaud & Constraints holding between onset and nucleus Halle (1978) argue, constraints holding between positions are not uncommon. In fact, Kawasaki & Ohala (1980) have other than nucleus and coda are not infrequent in English, suggested that there is a universal tendency against nor are they irregular or arbitrary; they are systematic initial *~ and *~, due to the lack here of an acousti­ because they involve natural classes. Algeo (1978) cites cally clear transition between the glide and the corres­ an example involving the case of the nonoccurrence of ponding high vowel. Besides this universal tendency, final ~ in English in order to illustrate an accidental many languages display specific onset-nucleus constraints. gap in the phonotactics. Algeo notes the following I mention just a few here. 28 29 For Mazatec, Pike and Pike (1947:87) cite the In Korean, according to Cho (1967), a number of following phonotactic restriction involving onset and constraints hold between onset and nucleus. Two of nucleus: "The nasalized vowels may not be preceded by these are that fronted vowels do not occur after labial v, y, 1, r, or their clusters, nor by m, n, ~." More consonants and that the vowel Iii does not occur after simply put: the class of sonorant consonants cannot occur any of the alveolar stops. before nasalized vowels. That it is a natural class Finally, according to the description of the which cannot occur before nasalized vowels suggests that Nigerian language Efik given in Dunstan (1969), the the restriction is systematic (it is possible that Ivl high vowels (which are Iii and lui in this language) do is an underlying sonorant in Mazateci if not, its non­ not occur after syllable-initial clusters. occurrence before nasalized vowels is most likely an accidental gap). 2.1.3. Constraints Between Onset and Coda for Other Languages For Piro, ~n Arawakan language of eastern Peru, Constraints holding between onset and coda are Matteson (1965:29) describes the following constraint: also found in languages other than English. For German, "The consonants It~/, lxi, Itsl, lsi, Itxl and Iyl do not "" Twaddell (1939, 1940) notes a tendency for the same con­ precede lui." These consonants are all the alveopalatal, sonant not to occur both pre- and post-vocalically in tho palatal, and velar consonants that occur in this language. same syllable (except for Ipl and lsi) i the more sonorous Thus, to state the constraint more simply, high conso­ the consonant, the more likely the restriction is to nants do not precede lui. hold. Moreover, in Chinese, according to Fudge (1969: In Kagate, a Tibeto-Burman language of Nepal 226-7), if a glide is the second member of an onset, then described by Hoehlig & Hari (1976), an onset cluster con­ no glide can appear in the coda. And for Yindjibarndi, sisting of a consonant followed by Iyl cannot precede a a language of Western Australia, Wordick (1982:14) notes front vowel. This constraint perhaps reflects an that Irl does not occur in both the onset and the coda extension to all front vowels of the universal tendency of the same syllable. Similarly, in Njangumad, a lan­ noted by Kawasaki & Ohala against Yi-sequences. guage of Northern Australia, O'Grady (1957) observes

that identical liquids do not occur in a CVC syllable, 30 31 and also that a nasal the coda (except n) does not in onset-nucleus constraints, and nucleus-coda constraints) ­ Occur with a homorganic stop in the onset. 4 would yield the following syllable structure with "double

2.1.4. Conclusion motherhood" • 0­ The above data on phonotactic constraints indi­ cate--contra Selkirk (1978) and Vergnaud & Halle (1978)-­ ~ onset nucleus coda that there do exist systematic dependencies between onsets The implausibility of such a structure (e.g., it violates and codas and between onsets and nuclei. Thus, the pro­ the anti-crossing lines constraint) leads one to conclude posed limitation of phonotactic constraints to members that, in fact, phonotactic constraints are not a test of the syllable rhyme cannot be maintained, since this for constituency. Moreover, this conclusion is not sur­ would make it impossible to state the restrictions dis­ prising in light of the similar situation in syntax where cussed in this section. A theory of syllable structure cooccurrence restrictions are not a necessary indica­ which incorporates the rhyme as an obligatory universal tor of constituency. For example, subject verb agree­ constituent therefore cannot be based on an argument from ment in English (or subject-NP predicate-adjective phonotactic constraints, since such an argument - applied agreement for number and gender in Spanish) fails to to English (which has onset-coda constraints, establish the constituency of subject-NP and verb (or

4. It is worth mentioning that there are lan­ subject-NP and predicate adjective). The existence of guages in which restrictions hold between consonants in separate syllables. For example, according to Schutz such restrictions, then, cannot be used to establish the (1976:78), Fiji syllables that begin with a velar do not appear before another syllable that begins with a constituency of the rhyme, or, for that matter, velar. Geers (1945) reports that in Late Akkadian, two or more emphatic (i.e., pharyngealized) consonants could subsyllabic constituency in general. not COoccur in the same word, though no such restric­ tion held in Old Akkadian. Some investigators might suggest that the Occurrence of these heterosyllabic 2.2. Rule Types Referring to the Rhyme constraints argues for higher level constituents (e.g., the foot and the word). However, since I am only focusing The next argument for the rhyme to be considered on the relationship between phonotactic constraints and sUbsyllabic constituency, I do not further pursue the here derives from the observation that some phonological issue of heterosyllabic constraints. rule types supposedly make reference to elements only 32 33 in the nucleus and coda. Rules of stress placement, as ~hcy argue for the universality of the rhyme on the well as phonetic-spreading rules and compensatory leng­ basis of the following claim (1980:93): "••• in all thening rules, are often claimed to make reference to languages known to us, stress assignment rules are sensi­ only these elements, and thus provide support for the tive to the structure of the syllable rime, but disregard rhyme as a single constituent. In the following sub­ completely the character of the onset." A possible con­ sections, I discuss these and show that the argument sequence of this argument for the rhyme has been noted from rule types fails to establish the constituency of by McCarthy (1978:8): "••• if the rhyme is a structural the rhyme because the onset (in addition to the nucleus unit then no language can assign stress by reference to and coda) can sometimes be mentioned in these rule types. the weight or any other property of CoV sequence." Moreover, elements in neighboring syllables can also In the remainder of this section, I will show sometimes be mentioned in phonetic spreading rules and that Halle & Vergnaud's claim is false on two accounts, compensatory lengthening rules. Since these rule types and that it thus fails to establish the universality of can refer to elements other than nucleus and coda, they the rhyme. First, stress assignment rules are not always do not provide convincing evidence for nucleus and coda sensitive to the structure of the syllable rhyme. And, together forming a single constituent. second, there ,are languages where the character of the

2.2.1. Stress Rules onset is not disregarded in stress placement. Conse­

Halle & Vergnaud argue that the rhyme is obliga­ quently, the argument from stress assignment is not valiu, tory; every syllable in every language has one. They and so obviously does not establish that the rhyme is an s tate (19 80 : 93) : obligatory (universal) syllable constituent, as Halle &

Our proposal is that skeleta in all languages Vergnaud claim. are subdivided into sUbsequences to which the term syllable has traditionally been attached. Halle & Vergnaud's statement that "••• in all Furthermore, we wish to argue that the syllables themselves possess internal constituent struc­ languages known to us stress assignment is sensitive to ture••• The obligatory constituent, which we shall call rime, dominates at least one V slot the structure of the syllable rime••• " could imply that in the skeleton. there are no languages where stress placement fails to

be sensitive to the particular make-up of the rhyme. .J3" 34 .\re languages in which stress is sensitive to the pre- However, the various surveys of stress rules in the nonce of an onset, such as in the Arandic and Paman literature do not support this claim. Only in a languages of Australia. And there are languages in minority (albeit a sizable one) of languages are stress which stress is sensitive to a particular type of onset, rules sensitive to the structure of the rhyme; in most ~s in Madimadi and Piraha. In the following section, I languages stress rules operate irrespective of its make­ discuss these languages, as well as others that have up. Ohsiek (1978:35), who surveyed stress rules in 140 onset-sensitive stress rules. different languages, found that stress rules were sensi­ tive to the rhyme in only thirty of these languages. 2.2.1.1. Onset-Sensitive Stress Rules Hyman (1977:59-66) lists only twenty-four languages (out In this section, I discuss several languages The existence of more than 300) in which the rhyme plays a role in that have onset-sensitive stress rules. stress placement, although Hayes (1981:55) notes that of these rules constitutes clear counterexamples to there are some mistakes in Hyman's survey. Many of these Halle & Vergnaud's claim of the onset-ignoring nature of

300 languages have consistent initial, penultimate, or stress rules. I delay until chapter three, though, the final stress. In these languages, where stress is always (metrical) analysis of some of the stress rules presented on the same syllable in a word, the particular struc­ here. ture of the rhyme is irrelevant for stress placement. In a number of Australian languages, stresS ruJ.OO

Thus, if Halle & Vergnaud are interpreted as arguing are found that are sensitive to the presence of an onset for the obligatoriness of the rhyme on the grounds that Such stress rules are found in the Arandic consonant. stress rules are sensitive to its structure in all lan­ languages of Central Australia, in many of the Paman guages, then their claim is simply false, and hence languages and the Lamalamic languages of the Cape York constitutes no argument at all. Peninsula, and in the isolate language Mbabaram. Moreover, Halle & Vergnaud's claim that "stress In Western Aranda the onset is crucial for assignment rules ••• disregard completely the character streSS placement in words of more than two determining of the onset" is likewise false, and so also does not (Bisyllabic words always have initial stress.) syllables. support the constituency of the rhyme, either. There 37 36 case for two syllable words (unlike in Western Aranda) Strehlow (1942:299-301) gives the following stress rule us the following examples show: for Western Aranda: (4) Alyawarra Main Stress Exemplified If a trisyllabic word begins with a consonant, the stress falls on the first syllable... If ingwa 'night' kw~tja 'water'

the trisyllabic word begins with a vowel, the /, stress falls on the second syllable... If a iylp~ 'ear' p~ynta 'spring' word of four syllables begins with a consonant, the main stress falls on the first syllable••• Moreover, note the following words that begin with a If a word of four syllables begins with a vowel, the stress falls on the second syllable... If glide (/y/ or /w/): a word of five syllables begins with a conso­ nant, the main stress falls on the first syl­ (5) Stress on Glide-Initial Words in Alyawarra lable, and a weak secondary stress is usually ( placed on the third syllable or on the fourth ••• wal~yrnparra 'pelican' If a word of five syllables begins with a vowel, the main stress normally falls on the second yukuntya 'ash' syllable, and a weak secondary stress is placed on the fourth syllable. In glide-initial words, stress falls on the second syl­

Thus, in Western Aranda, main stress is on the first lable. Thus, we see that Alyawarra stress is sensitive syllable containing an onset,. The following examples, not only to the presence of an onset, but also to the taken from Strehlow (1942), illustrate the rule in nature of that onset - i.e., to whether it is [+consonan­ question: tal] or [-consonantal]. Thus, the existence of such rules

(3) Western Aranda Main Stress Exemplified falsifies Halle & Vergnaud's claim about the onset­

tararna 'to .laugh' ankata 'lizard (sp.)' ignoring nature of stress assignment rules.

kbtu')ula 'ceremonial ar~lkarna 'to yawn' Onset-sensitive stress rules are also found in assistant' the Paman Languages of the Cape York Peninsula in the to:turatura 'marsupial ul~ularnba 'fowl (sp.)' mole' northeastern part of Australia. Many of these languages

In another Arandic language, Alyawarra, a possess an onset-sensitive stress rule that resembles similar stress rule is found. According to Yallop (1977: the Arandic one. For example, in Uradhi, according to

43), main stress in a word falls on the first vowel Hale (1976:44), "••• stress is assigned by means of a following the first consonant (i.e., stress is on the rule that is apparently shared by all the Northern Paman

languages, namely: V--~ V/#(V)C • That is, stress first syllable containing an onset). This is even the 1 38 3tl the first post consonantal vowel in the word." Or, in In sum, we have noted that four separate Austra­ terms of syllable structure, stress the first syllable lian language families (Arandic, Paman, Lamalamic, and with an onset. Some examples showing this are /ut~ga/ Mbabaram) have virtually the same onset-sensitive stress 'dog' and /mfnhitji/ 'bird'. Hale (1966) cites another rule (i.e., stress the first syllable containing an Northern Paman language, Linngithig, as possessing the onset), though the details differ, depending on the lan­ same stress rule. Some examples are given in (6): guage. In Western Aranda, for example, the rule applies (6) Linngithig Stress Exemplified only in words that have more than two syllables, while, k~pan / 'short' ayo') 'sky' in Alyawarra, the onset must contain a [+consonantalJ g 9 6"tro 'black' a1~ 'crow' segment for the syllable to receive stress. Thus, it is And Rigsby (1976) cites two other Paman languages, Kuku- evident that, in these languages, it is not the proper­ Thaypan and Urnbindhamu, as having this same onset­ ties of the rhyme constituent that is being referred to, sensitive stress rule. in the stress-assignment rules, but, rather, the proper­ Besides the Faman languages, other languages of ties of the onset. And, in this regard, they constitute the York Peninsula share the same stress rule. Laycock evidence against Halle & Vergnaud's argument for the (1971) establishes as separate from Paman the Lamalamic universality of the rhyme based on the claim that rules family, which O'Grady, Voegelin, and Voegelin (1966) had assigning stress are sensitive only to the properties classified as a subfamily of Paman. Laycock (1971:78) of nucleus and coda, but not to those of the onset. observes that, "... by far the most cornmon stress place­ Before discussing other languages that have onset­ ment is on the first syllable beginning with a conso­ sensitive stress rules, I want to consider the develop­ nant." In other words, stress falls on the first syllable ment of the almost identical stress rule found in Arandic, containing an onset. Also, Dixon (1970) notes that Faman, Lamalamic, and Mbabaram. I do this because Mbabaram, a language isolate spoken in the southeastern supporters of the rhyme might argue that, since the stress part of the Cape York Peninsula, has the identical stress rule common to these languages has a historical explana­ rule. tion, it cannot be considered a legitimate counterexample 41 40 to Halle & Vergnaud's claim of the onset-ignoring nature have some historical explanation (e.g., in early Latin, of stress rules. according to Allen (1965:82), stress was assigned from Like all stress rules, the synchronic stress the beginning of the word, and it was only later that rule shared by Arandic, Paman, Lamalamic, and Mbabaram a rule assigning stress from the end of the word sensi­ (stress the first syllable containing an onset) has a tive to syllable quantity developed), but this should not diachronic account. Alpher (1976) relates the stress be considered in a synchronic analysis of those stress rule to a process termed "initial-dropping" which has rules. This is because children learning a language (and occurred independently in these branches, according to the stress rules of that language) do not know the history Dixon (1970) and Capell (1979). Alpher (1976:85) notes of that language (nor of its stress rules); they construct that: "Most, but not all, of the ID [initial-dropping] stress rules solely based on the data that they are languages assign the most prominent stress in the word exposed to. Thus, in spite of its historical explana­ according to the following principle: stress the first tion, the Australian onset-sensitive stress rule still vowel that follows a consonant." Alpher then posits a constitutes a valid counterexample to Halle & vergnaud'6 5 development whereby an original initial-syllable stress claim. in these languages shifted onto the second syllable, Not only are there languages in which stress is thereby causing the loss of an initial consonant and the sensitive to the presence of an onset, as we have just following vowel if that vowel was short, and, if that illustrated, but also, there are other languages (be­ vowel was long, then it was shortened. The result of sides Alyawarra) in which stress is sensitive to the this development, as Alpher (1976:85) observes, "•.. would particular nature of the onset. For example, in njapu, have been the synchronic rule that the first vowel that a language of North Central Australia described by Morphy follows a consonant is stressed." However, the fact (1983) , main stress normally falls on the initial that this synchronic stress rule has a diachronic account syllable. Morphy (p.25) notes the following exception: does not mean that it cannot be considered a valid counterexample to Halle & Vergnaud's claim about the 5. See Janda (1984:102-103) for some relevant remarks on the invalidity of certain types of diachrnoic onset-ignoring nature of stress rules. Most stress rules explanation for synchronic phenomenon. 42 43 " exceptions to the general rUle are predominately no coronal consonant appears in the onset of the second words in which the second syllable begins with /d/, for oyllable. Madimadi, thus, provides a good example of example budapthu-N 'go down and cross' and band~ny 'dry, n language that has a stress rule that is crucially clean'." So, if the onset of the second syllable consists dependent on the particular features of an onset of /d/, then main stress is attracted to that syllable. consonant.

A second example in which stress is sensitive to A somewhat different example of a language having the particular nature of an onset is provided by Madimadi, an onset-sensitive stress rule is Piraha, an Amazonian an Australian language Spoken in New South Wales. Accor­ language discussed by Everett and Everett (1984). The ding to Hercus's (1969) description, primary stress nor­ location of main stress in Piraha sometimes depends on mally falls on the first syllable of a word. However, the particular nature of the onset and the particular if the second syllable begins with an interdental, nature of the nucleus. Briefly, main stress normally alveolar, retroflex, or palatal consonant (i.e., roughly, falls on one of the last three syllables of the word, a coronal consonant), then primary stress is attracted whichever one has a long vowel or a diphthong. However, onto that syllable, and a weak secondary stress is rea­ if more than one of the last three syllables has a long lized on the first syllable. The following data illus­ vowel or a diphthong, then stress falls on the rightmost trate the Madimadi stress pattern: syllable that has a voiceless consonant in the onset; (7) Madimadi Stress Exemplified examples illustrating this are given in (8): / wlrandU 'his sinews' buryae~ 'to pullout' (8) Pirana Stress Exemplified (stressed syllables butu:ryaeA 'to smash' gupae~ 'to drink' are underlined; tones are not indicated) , / / mana~ai 'my hand' dE:maeA 'to hear' bii sai 'red'

~ / / bijali 'red gum' walwae~ 'to burn' kai bai 'monkey'

In the words of the first column, in which the second ?i bao sai 'her cloth' syllable begins with a coronal consonant, primary stress Further, if there are no long vowels or diphthongs in falls on the second syllable. In the words of the the last three syllables of the word, then stress falls second column, stress is on the initial syllable, since on the rightmost syllable (among the last three) that has 44 45 a voiceless consonant in its onset; otherwise, it falls placement is Gadsup, a language of New Guinea. Frantz on the final syllable. Thus, the Piraha stress rule & Frantz (1966:9) report that among the many factors

provides another counterex~~ple to Halle a..d Vergnaud's helping to determine stress placement in Gadsup is the

claim. make-up of the onset. They observe that ~I ••• syllables Before concluding this subsection, it is worth ... with a phonetic stop onset have more stress than

pointing out that there are other cases of onset­ those with nonstop onset. II Finally, for English, Oehrle sensitive stress rules mentioned in the literature, but (1971:26) observes that bisyllabic nouns with the prefix these either do not seem to be as productive as the pre­ dis have stress on the first syllable if the (onset in

viously discussed onset-sensitive stress rUles. For the) root part of the word begins with Ikl or lei. Some example, Breen (1981:25) mentions that, in Ngawun (an examples include, discard, discount, and discharge. Australian language spoken in Queensland), there is a In conclusion, the existence of such stress- tendency for stress in words having three or more syl­ assignment rules, which are sensitive to the presence lables to shift from the first syllable to the second of and sometimes also the character of the syllable's

if the first syllable starts with a peripheral stop or onset, constitutes evidence against Halle & Vergnaud's nasal and if the second syllable begins with a lateral argument for the (universality of the) rhyme. For - to or rhotic. Dixon (1977:102) notes the opposite tendency repeat - as McCarthy (1978:8) has observed, " .•• if the Y in Yidin • There, if the second syllable is assigned rhyme is a structural unit, then no language can assign stress, and if that syllable begins with a lateral or stress by reference to the weight or any other properties rhotic, and if the third syllable begins with a stop or of CoV sequences." Since there are such languages, then, Iwl, then stress shifts to the third syllable. Thus, it appears that the rhyme cannot be an obligatory (uni­ Dixon notes that, in YidinY : "The main preference seems versal) structural unit within the syllable. (Further, to be for the stressed syllable to begin with a stop or in chapter three, I will consider - and reject ­ wand for it not to commence with a lateral or rhotic." (metrical) analyses of onset-sensitive stress rules that Another example of a language in which the par­ incorporate syllable structure possessing the rhyme.) ticular nature of the onset plays a minor role in stress 47 46 lnnture [+nasal] and [+round] is the onset plus the 2.2.2. Phonetic-Spreading Rules nucleus, and thus would constitute evidence for their Another rule type said to have the rhyme as its l'omprising a subsyl1abic constituent. domain is phonetic-spreading rules. For example, McCarthy In Nupe, it is well known that front vowels (1979:454) argues for the rhyme on the basis of the condition palatalization of the preceding onset con­ following: "French vowels assimilate in nasality to the uonant, while back vowels condition rounding of the pre­ following tautosyllabic consonants. Therefore, the ceding onset consonant (Hyman 1970). Thus, the Nupe rhyme is the domain of the feature [+nasal]." Against domain of the features [+round] and [+high] is also this claim, one can mention that phonetic-spreading processes, such as nasalization, do not necessarily just the onset plus the nucleus. Finally, in English, the domain of the feature have nucleus and coda as their domain - as McCarthy's [+high] (when triggered by a high back vowel) is often French nasalization example would seem to imply. Rather just the onset and the nucleus, as is apparent from nasalization, as well as labialization and palatalization, such words as ~ [kYut], ~ [vYu], and ~ [mYut]. often apply over the onset and the nucleus. Examples of By McCarthy's (1979) logic, then, that the domain of the such languages include Lugisu, a Bantu language, and feature [+high] is the onset and the nucleus constitutes Nupe, a Kwa language. evidence for them forming a single subsyllabic consti­ In Lugisu, a nasal consonant nasalizes the tuent. On the other hand, the English domain of nasali­ following vowel, as the examples below from Brown (1970: zation is often the nucleus and the coda, as is 13) illustrate: exemplified by such words as can [kren], ten [tIn], and (9) /ka + Jl u + Pi/ --7 [kapuJli] 'bird' sing [sr~]; McCarthy would interpret this as being /u + mu + kana/ --~ [umukana] 'girl' evidence for the rhyme. Interestingly, sometimes, in Also, in Lugisu, a rounded vowel causes the preceding the same word both nasalization and palatalization can onset consonant to become round, as is seen in (10): occur; this is exemplified by such words as ~ [fYum] (10) /i + N + goko/ --~ [i,gWOkwo ] 'hen' and ~ [hYun]. By McCarthy'S logic, such words would /Bu + Bu + eni/ --~ [BwuBwwe:ni] 'forehead' have a syllable structure in which there is "double By McCarthy's reasoning, then, the Lugisu domain of the 4<)

48 Immediately preceding syllabic nucleus. All other empty motherhood" for the nucleus: the onset and the nucleus nodes are to be pruned." The convention is illustrated form one constituent, and the nucleus and the coda com­ by the Latin example Isisdo:1 'sit (present)' which, prise another. Consequently, phonetic-spreading rules

segment associated with the nucleus then fills. He

refers to this as the "Empty Node Convention" (p.47l):

"Empty w nodes which are part of a syllabic coda are to

be associated with the terminal element dominated by the 'H 50 (13) /hinbit/ [hibbit] 'he looked' (12) $ $ $ $ $ $ $ $ $ $ A /\ A /\R~ A !\ A f\. A f".- a R a R a R a R a R a R a R \ a R a R a l---j I (\ ~ \ /\ \ 1\ I 1\ I . I 1\ I I I /\ \ f\-::::=::t 1/\ C v C C v c o e s m l. o e m i C v C C v C C v C C v C III I I ",\ I I \\\\I I \ I h i bit h i n bit h i bit [haaRo~] 'the head' $ $ (14) /haRRos( A /'­ $ a R a R $ $ $ ~ $ $ I 1'---.\ I A 1'­ A j\ /'\ 1\ o e m i ~ a R a R a R a R II\\I\~ \{\ 1/\ l Al 1\-( CVCCVC c v C c v C Such a derivation suggests that compensatory lengthening C v C C v C I l··' \ I I I I I I I I I I IlL 0 h a R 0 s h a R S really does not involve the rhyme constituent as much as h a R R 0 5 a language (or dialect) specific preference for direction (15) (bi9ger( [biger] 'he burned' $ of spreading. In the Lesbian Greek example, the empty $ $ $ /\ /\ A a R node is filled by leftward spreading, while, in the Latin a R I'R a R 1/\ 1/\ \ f\ \/\ example, the empty node is filled by rightward spreading. C v C C v C C v C C v c I I I I I I I 1 I \ I 9 e r Furthermore, in Tiberian Hebrew, compensatory b i 9 9 e r b i lengthening that involves a xC VC C Y sequence may be (the symbol 9 represents a voiced pharyngeal 1 2 3 fricative) resolved by leftward spreading, rightward spreading, or Hebrew seems to have a preference for leftward spreading no spreading, depending on the nature of C • Basically 2 (as in (13». Rightward spreading occurs only when left­ (omitting certain details not relevant here), if c is 2 ward spreading is not possible (or, rather, when leftward a nasal that deletes, leftward spreading follows, as in spreading would produce the original input). Lowenstamm (13) below. And, if C is a deleted guttural (or z that leftward spreading is and Kaye (1984) suggest pharyngeal) then, depending on the particular guttural, preferred in languages that have geminates (as in either there is rightward spreading, as in (14), or there is no spreading, as in (15): S3 52

Hebrew).6 Thus, also in Hebrew, compensatory lengthening 11110 gives the example latJ-l:lk-al which has the derivn- does not involve mention of the rhyme constituent, but IJon in (17): rather a preference for a specific direction of spreading. (17) lat:>-Dk-al [atJIUaa] $ $ $ Furthermore, there are cases of compensatory $ $ $ $ $ I A A (\ R1 o/).· n ~ OnI). ~ 01\·n ~ lengthening that are triggered by the deletion of an onset R On Ri On Ri On Ri I I I I I I I I I \ I I I I C V C V consonant, which Ingria (p.471) predicts should not occur: V C V C V C V V C V I I I I I I I I " ,I I I I I U k a 1 ;) k a a t ;) 1 "All other empty nodes [Le., those that are not part of a t J ~ a syllabic coda] are pruned." That such cases do occur clearly illustrates that the rhyme is not the domain of The deletion of an onset consonant in Maasai causes the compensatory lengthening. One example of this kind of following vowel to fill its slot. Another language that compensatory lengthening involves k-deletion in Maasai, has a similar compensatory lengthening process is the a Nilo-Saharan language. In Maasai, when intervocalic Modern Greek dialect of Samothraki, discussed in a foot­

Ikl (which is an onset consonant) deletes, the following note by de Chene and Anderson (1979:508). In this dia­ vowel is lengthened. Levergood (1984:48) cites Wallace lect, the loss of prevocalic Irl (an onset consonant)

(1981) who gives the following rule of k-deletion: triggers lengthening of the following vowel, as is shown

(16 ) [-hi] K [+10] in (18) : 1033 1 2 3 :::=::::} 'ask! ' (18) Irotal [o:taJ "delete k between a non-hi vowel and a, lengthen a." Iprasinosl [pa:snus] 'green' lantrasl [ada: s] 'man' 6. Lowenstamm & Kaye (1984) also provide an The last example in (18) is most illuminating because analysis of the Hebrew compensatory lengthening in which, essentially, there is leftward spreading if the deleted it shows that, in samothraki, a deleted coda consonant consonant is part of the coda, and rightward spreading if it is part of the nucleus (and not the coda). They (/nl in the last example) fails to trigger compensatory argue that syllable-final gutturals are not in codas, but are more vowel-like since they are resonants. They lengthening, but a deleted onset consonant does trigger do not consider the interesting cases where there is no compensatory lengthening (which is not uncommon). They say that such examples are not counterexamples to their it. analysis, but only to the data!