Syllable Structure and Stress in Dutch [Reprint 2021 ed.] 9783112419984, 9783112419977

Citation preview

Syllable Structure and Stress in Dutch

Linguistic Models The publications in this series tackle crucial problems, both empirical and conceptual, within the context of progressive research programs. In particular, Linguistic Models will address the development of formal methods in the study of language with special reference to the interaction of grammatical components. Series Editors: Teun Hoekstra Harry van der Hulst Michael Moortgat

Other books in this series: 1 Michael Moortgat, Harry van der Hulst and Teun Hoekstra (eds.) The scope of lexical rules 2 Harry van der Hulst and Norval Smith (eds.) The structure of phonological representation. Part I. 3 Harry van der Hulst and Norval Smith (eds.) The structure of phonological representation. Part II. 4 Gerald Gazdar, Ewan Klein and Geoffrey K. Pullum (eds.) Order, Concord and Constituency 5 W. de Geest and Y. Putseys (eds.) Sentential Complementation 6 Teun Hoekstra Transitivity. Grammatical Relations in Government-Binding Theory 7 Harry van der Hulst and Norval Smith (eds.) Advances in Nonlinear Phonology

Syllable Structure and Stress in Dutch

Harry van der Hülst INL, Leiden

1984

FORI S PUBLICATIONS Dordrecht - Holland/Cinnaminson - U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Sole distributor for the U.S.A. and Canada: Foris Publications U.S.A. P.O. Box C-50 Cinnaminson N.J. 08077 U.S.A.

C IP-DATA KONINKLIJKE

BIBLIOTHEEK,

DEN HAAG

Hülst, Harry van der Syllable Structure and Stress in Dutch/Harry van der Hülst. - Dordrecht [etc.]: Foris Publications. - (Linguistic Models; 8) Also published as thesis Leiden. ISBN 90-6765-037-4 bound ISBN 90-6765-038-2 paper SISO *837 UDC 803.931-4/5 Subject headings: Syllable Structure; Dutch Language/Stress; Dutch Language.

ISBN 90 6765 037 4 (Bound) ISBN 90 6765 038 2 (Paper) © 1984 Foris Publications - Dordrecht. No part of this publication maybe reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. Printed in the Netherlands by ICG Printing, Dordrecht.

for Ella, Femmy, Lieke and Niels

Table of Contents

Acknowledgements Samenvatting

xi xiii

PAST I Int.roduct.ion Chapter 1 The framework of nonlinear phonology

3

1.1.

Introduction

3

1.2.

A sketch of the nonlinear framework

3

1.3.

Autosegmental phonology

10

1.3.1.

The characterization of complex segments

10

1.3.2.

Supporting arguments for representing features on independent tiers

14

1.3.3.

Principles of autosegmental phonology

17

1.3.4.

Extension of autosegmental theory to vowel harmony

22

1.4.

Metrical phonology

29

1.5.

Contents of this study

33

PART II SYLLABLE STRUCTURE

Chapter 2 The representation of syllable structure

37

2.1.

37

Introduction

2.2.

Theories of syllable structure

37

2.2.1.

Introduction

37

2.2.1.1.

Linear structure: universal

38

2.2.1.2.

Linear structure: language-specific

39

2.2.1.3.

Hierarchical structure: universal

40

2.2.1.4.

Hierarchical structure: language-specific

40

2.2.1.5.

Autosegmental structure: universal

40

2.2.1.6.

Autosegmental structure: language-specific

2.2.2.

A multivalued feature [sonority]

42 42

2.2.3.

Constituent structure

51

2.2.4.

Syllabification rules

62

2.3.

Conclusions

73

Chapter 3 Dutch syllable structure

75

3.1.

Introduction

75

3.2.

Preliminary remarks

75

3.3.

Linear structure

78

3.3.1.

X

78

and X

3.3.2.

1 2 The syllabic prefix

84

3.3.3.

X , X4, X5

89

3.3.4.

The appendix

3.3.5.

Exceptions

3.4.

Hierarchical structure, syllabification and related issues

98 100

100

3.5.

Trommelen (1983) and the diminutive suffix

109

3.5.1.

Trommelen (1983)

109

3.5.2.

The diminutive suffix

117

3.5.2.1.

Introduction

117

3.5.2.2.

Trommelen1s analysis

120

3.5.2.3.

Alternative analysis I

122

3.5.2.4.

Alternative analysis II

125

3.6.

Conclusions

131

PART III STRESS

Chapter 4 The representation of stress

135

4.1.

Introduction

135

4.2.

Grids and trees

135

4.3.

A comparison of grid and tree theory

144

4.3.1.

Word stress systems

144

4.3.1.1.

Quantity insensitive systems: bounded

145

4.3.1.2.

Quantity insensitive systems: unbounded

152

4.3.1.3.

Quantity sensitive systems: bounded

154

4.3.1.4.

Quantity sensitive systems: unbounded

160

4.3.1.5.

Limited stress systems and syllable weight

163

4.3.2.

A conclusion and a proposal

170

4.3.3.

The main stress first theory

178

4.3.4.

Fixed stress and secondary stress

187

4.3.4.1.

Introduction

187

4.3.4.2.

Secondary stress in Italian

190

4.4.

Metrical feet

197

4.5.

Conclusions

200

Chapter 5 Dutch stress

201

5.1.

Introduction

201

5.2.

Stress and syllable weight in Dutch

201

5-3.

The stress pattern of Dutch monomorphemic words

216

5.3.1.

Nouns

216

5.3.1.1.

Bisyllabic nouns

216

5.3.1.2.

Trisyllabic nouns

222

5.3.1.3.

A metrical analysis

229

5.3.1.4.

Extension of the data base

232

5.3.1.5. 5.3.2.

Syllables with a schwa and an alternative analysis Adjectives and verbs

238 249

5.4.

Conclusions

251

References

253

Acknowledgements

In 1980 I was asked to give a talk about recent developments in phonology and in particular about autosegmental and metrical phonology. Until then I had been primarily interested in the diachronic background of phonological rules that govern synchronic segmental alternations. At that time the explosion of literature in which new ways were proposed of analyzing suprasegmental phenomena such as stress and tone was still to come. It took me some time to find out what was going on in phonology and what was happening to the theory of generative phonology, which had caused such fruitful discussions with respect to the proper treatment of morphophonemic alternations. When I gave the talk I was still puzzled about many aspects of the new proposals and in particular about the way in which modifications to the SPE-theory, which were motivated on the basis of a related but still diverse set of phenomena, had to be combined to lead to a new theory. In the subsequent years some of these puzzles were solved, either by a growing understanding of the new proposals on my part or by modifications that were suggested. At present it is still the case that generative phonology is in a phase in which proposals that deal with suprasegmentals are advanced and rejected with rapid speed. The present study is meant as a contribution and an introduction to this lively way of doing

linguistics.

interrelated

I have

phenomena,

limited

viz.

myself

syllable

to the discussion

structure

and

of two

stress, but

an

introductory chapter offers an overview of current approaches to vowel harmony and tone as well. During the past few years I have had the opportunity to profit from the opinions and personal guidance of excellent linguists with respect to the subject matter of this book, or closely related topics. Some of them I talked to nearly every day during the preparation of this book and they are not

even

phonologists.

Explaining

to

a government-and-binding

syn-

tactician and a possible-world semanticist why a particular phonological proposal makes sense can be frustrating, but it also forces one to take

little for granted. Both Teun Hoekstra and Michael Moortgat contributed more than they may think to this study. They are sharp linguists and good friends. I am not sure that I would have survived in the jungle of metrical trees, association degenerate

lines and

floating

tones

syllables) without Norval

(not to mention Smith,

extrametrical

a companion with

great

knowledge of challenging data and a lively imagination. I am sure that our joint enterprises will continue to flourish. I twice had the opportunity to visit the Salzburg Summer Institute (1979, 1982) and once the Summer Institute held at UCLA (1983 ). This gave me the chance to take courses with Nick Clements, Wolfgang Dressier, John Goldsmith, Bruce Hayes, Larry Hyman, Joan Mascaré and Paul Kiparsky (and many other abled linguists whose courses have no direct bearing on the topic of this study). Without these courses and the discussions before, during and after classes I would not be in phonology where I am now. I thank the Netherlands Organization for the Advancement of Pure Research (ZWO) for sponsoring my attendance at these institutes. A third visit to Salzburg was possible in 1983 because Gaberell Drachman was kind enough to offer me the opportunity of giving a course myself. The preparation of this course, the classes and the discussions I had with Angeliki Malikouti-Drachman all contributed to my own understanding of

the emerging nonlinear framework.

Various other people have contributed in their own way either to the writing of this study. I thank Simone Langeweg for helping to gather the data on which chapter 5 is based, for reading and commenting on the entire manuscript and for her assistance in typing parts of earlier versions of this text. I thank Randal Rhyverthus and Stanley H. Vurrhard (temporarily assigned to the INL computer section) for typing the final version and producing the camera-ready print-out. Finally, I thank my wife Ella and my children, Femmy, Lieke and Niels, for their indispensable mental support, too often taken for granted, and my parents, without whom I would not be were I am now, in phonology or any other place.

Samenvatting

De

Studie

van

zgn.

suprasegmentale

verschijnselen

(o.a.

klemtoon,

lettergreepstructuur, toon en intonatie, vocaalharmonie) heeft binnen de generatieve taalkunde de laatste jaren een hoge vlucht genomen. Hierbi j is gebleken dat het theoretisch kader, voorgesteld in Chomsky and Halle's The Sound Pattern of English (1968), op een aantal punten in aanzienli jke mate gewijzigd

moet worden.

In deze

Studie

Staat de vraag

centraal hoe

suprasegmentele verschijnselen verantwoord moeten worden, waarbij ik me beperkt heb tot twee nauw samenhangende verschijnselen, nl. lettergreepstructuur en klemtoon. Het boek

is opgedeeld

in drie delen. Deel I bevat een hoofdstuk

(hoofdstuk 1), waarin een algemeen overzicht wordt gegeven van de huidige stand van fonologische theorievorming binnen de generatieve taalkunde met betrekking tot de structuur van fonologische representaties. Hierbi j wordt met name aandacht geschonken aan een aantal suprasegmentele verschi jnselen (zoals toon en vocaalharmonie) die in de overige hoofdstukken niet centraal staan. Op deze wijze kan de lezer zieh enigszins een totaalbeeld vormen van het gehele veld van suprasegmentele verschi jnselen en de wijze waarop deze verschijnselen in hun samenhang worden benaderd. Deel II en III zijn respectievelijk gewijd aan de representatie van lettergreepstructuur en klemtoon. Beide delen bevatten twee hoofdstukken, waarvan het eerste telkens gericht is op een kritische beschouwing van de beschikbare benaderingen en het tweede verschillende voorstellen toetst aan het Nederlands. Aldus biedt hoofdstuk 2 een overzicht van verschillende opvattingen over

de

lettergreep.

concurrerende

Een belangrijke

benaderingen

wordt

rol bij de

toegekend

aan

evaluatie

van

de mogelijkheid

de een

verklärend verband te leggen tussen de structuur van lettergrepen en het gedrag van lettergrepen ten

opzichte van klemtoontoekenning. Tevens

wordt de Stelling verdedigd dat een meerwaardig kenmerk [sonoriteit] een

belangrijke roi moet spelen bij de analyse van lettergrepen. In hoofdstuk 3 wordt een volledige analyse gegeven van de Nederlandse lettergreep. Deze analyse wordt vergeleken met de analyse die onlangs is voorgesteld door Mieke Trommelen (Trommelen 1983). De vergelijking wordt toegespitst

op

de

mogelijkheden

die

beide

analyses

bieden

bij

de

verantwoording van de verschillende verschijningsvormen van het verkleiningssuffix (bijv. tje, etje enz.). In hoofdstuk

4 staat de representatie van

klemtoonverschijnselen

centraal. Uitgebreid wordt ingegaan op twee Varianten van de zgn. metrische klemtoontheorie.

In de

"ortodoxe" variant

spelen binair

vertakkende

boomstructuren een belangrijke roi, terwijl in de "ketterse" variant klemtoonpatronen worden gekarakteriseerd met behulp van een eenvoudiger hierarchische structuur, waarin een groepering in constituenten geen roi speelt. Ik betoog dat aan beide Varianten, in een enigszins gewijzigde vorm, een complémentaire roi kan worden toebedeeld. Voorts stel ik voor dat de relatie tussen hoofdklemtoontoekenning en toekenning van nevenaccenten anders gezien moet worden dan gebruikelijk is in beide Varianten van de metrische theorie. Nevenaccenten kunnen namelijk het best geanalyseerd worden als nauwelijks taalspecifieke bijverschijnselen van het hoofdaccent, niet als iets dat aan hoofdklemtoontoekenning ten grondslag ligt. In hoofdstuk 5 wordt het klemtoonpatroon dat we in ongelede Nederlandse woorden aantreffen uitgebreid onderzocht. Eerst wordt de relatie tussen lettergreepstructuur

en

klemtoon,

zoals

we

die

in

het

Nederlands

aantreffen, besproken. Vervolgens wordt een inventaris opgesteld van zgn. dominante klemtoonpatronen, op basis van een corpus bestaande uit ongelede substantieven en adjectieven. Tenslotte worden verschillende metrische analyses voorgesteld en vergeleken. De voorkeur gaat uiteindeli jk uit naar een analyse die in verschillende opzichten aansluit bij de voorstellen die in hoofdstuk 4 op meer algemene gronden werden gedaan.

Chapter 1

The Framework of Nonlinear Phonology

1.1. Introduction

In this chapter I will discuss current views on the structure of phonological representations within the theory of generative grammar, thus providing the context within which the present study of syllable structure and stress is placed. After a brief sketch of these views in section 1.2, sections 1.3 and 1.4 will provide a discussion of the theory of autosegmental phonology and the theory of metrical phonology. It is my intention to provide insight into the basic structure and principles of the two theories and to make clear the type of argumentation that has led to their development. Since chapters 2 and 4 will be concerned with issues that fall within the scope of metrical phonology, this introductory chapter is devoted primarily to autosegmental phonology. It is not my intention to give a complete introduction to the theory of nonlinear phonology here. I will discuss neither all phenomena for which nonlinear treatments have been suggested nor all the different competing proposals that occur in the literature. I have selected topics that involve some of the central issues and will limit myself in several places to giving the views I adhere to. Detailed introductions to both autosegmental and metrical phonology can be found in Van der Hulst and Smith (1982a) and McCarthy (1982).

1.2. A sketch of the nonlinear framework

The theories discussed in this section have both emerged from the research program that was initiated in Chomsky and Halle's The Sound Pattern of English (SPE; Chomsky and Halle 1968) . Although there are many differences

4

between the SPE approach and current approaches, a number of fundamental assumptions have been maintained without change, such as the distinction between underlying and surface representations, rule ordering and the belief that the significance of generalizations corresponds to formal simplicity. The theory proposed in SPE has a derivational aspect and a representational aspect. The former aspect involves issues such as the formulation of phonological rules, rule application and rule ordering, whereas the latter aspect involves the structure of phonological representations at each level of the derivation. One can say that the changes that I will discuss here all involve the representational aspect. The changes that we have witnessed over the past few years have been caused, not by changing the methodology in any way, but by extending the empirical domain of the theory considerably. The following quote from Chomsky (1955: 29) could apply to a large extent to SPE, where suprasegmental features are approached as if they were segmental features: In this study, suprasegmental features (pitch, stress, juncture) have not been seriously considered. Ultimately of course, these phenomena must be incorporated into any full syntactic theory, and it may be that this extension still requires a more elaborate system of representation. In SPE the phonological representation is unilinear, i.e. it consists of a single sequence of segments and boundary symbols. Segments consist of an unordered set of features, each of which has a binary value. This sequence is associated with a hierarchical structure that is non-phonological, but morphological and syntactic. Currently a phonological representation is considered to be a three-dimensional object, in which we find not just one sequence of segments, but several sequences. Hence the representation is called multilinear or nonlinear. These sequences (called tiers) are linked to a central tier that consists of abstract units to which the segments on the other tiers are associated. This pivotal tier also constitutes the interface between the

(morpho)syntactic hierarchical structure and a

prosodic hierarchical structure. From the point of view we have today the quote given above sounds almost like an understatement. I will now sketch in abstract terms the two ways in which the SPE conception of phonological

representations has been altered, before

proceeding with a discussion of the type of data that have motivated the changes. I will also pay some attention to the question how the changes that have been proposed are to be combined.

5

The view of segments as unordered sets of specified features has been abandoned for the following reason. It has been shown that the scope of a specified feature need not be a single segment, implying that not all features are synchronized by the same temporal function. The scope of a feature may both be smaller and bigger than a single segment. Before giving some examples let me emphasize that the term segment must be redefined if features are to be allowed to have different scopes. We must first determine what it is that features may have within their scope. The current view is that features have scope over abstract units, called slots (also sometimes referred to as timing points) . These slots are essentially traditional SPEsegments deprived of all (or most) of their features, leaving only two brackets. According to one view slots are completely unspecified units, represented with the symbol "X", according to another we find two types of slots, usually represented by the symbols "C" and "V". These different views will be discussed in section 2.2.1.3.: (1)

t

]

The sequence of pairs of brackets is referred to as the central tier. There is of course no objection to calling the units that constitute the central tier segments, as long as we realize that the interpretation of this term has been changed. The second point to be precise about is the status of features. The idea is that features are also segments; segments on their own, hence autosegments. If, in a particular language, there is good reason to represent the feature [F] as having scope over more than one slot then a sequence of such features is regarded as a sequence of [F]-segments. On the assumption that all other features are synchronized (i.e. have the same scope, in this example over one slot

only), the following representation

is the result: (2)

[+F] [

]

[

[-F] ]

[

[+F]

I /\ ]

[

]

[

+G

-G

-G

+G

-G

-H

-H

+H

-H

+H

+N

-N

+N

+N

-N

[F]-tier :

central tier

An understandable, but strictly speaking confusing, term to use for the

6

third tier is segmental tier. It is possible to maintain that such a cumulative tier does not even exist and that it just happens to be the case here that the [G]-tier, the [H]-tier and all other tiers (except the [F]tier) are subject to the same function that associates them to the central tier. However, since wide scope is the special case rather than the norm I will assume that the third tier has a theoretical status and that features are bundled in the traditional way, unless there is evidence to the contrary. Instead of wide scope a feature may also have narrow scope: (3)

[-F] C

C+F] ]

+G -H +N_ The need for giving features both wide and narrow scope (the reasons for which I will discuss in the next section) has led to the development of the theory of autosegmental phonology. The second change in our conception of phonological representations is logically independent of the first. As noted above, the only hierarchical structure that is imposed on the row of segments in SPE is of a morphosyntactic nature. This hierarchical structure tells us that substrings of segments constitute morphemes, words and finally phrases and sentences. Even in SPE it is pointed out that syntactic bracketing is not appropriate in all cases to characterize the domain over which intonational contours extend. To remedy this defect certain rebracketing operations are suggested. In subsequent work it has been pointed out that the mismatch between morpho-syntactic structure and some other kind of structure is more serious. The clearest examples involve rules that specify what sequences of segments are wellformed in a particular language. In SPE it was assumed that the domain of these rules was the morpheme, but many phonologists have pointed out that another unit would be more appropriate. This unit was the syllable. The logical conclusion of introducing phonologically motivated units comprising substrings of segments was that a complete hierarchical structure was assumed, distinct from the morpho-syntactic structure, although not unrelated to it. This view did not only arise from a need to

7

have an intonational phrase

or a syllable. New views on stress completed

the development by providing arguments in favor of several constituents smaller than intonational phrases and larger than syllables, viz. the foot and the (phonological) word. The new theory of stress was termed metrical phonology but soon afterwards this term was used to refer to a theory of phonological constituent structure in general. The independence of morpho-syntactic and phonological

constituent

structure is acknowledged by assuming two distinct tree structures imposed on the string of segments, that is the

interface between these two

structures. (4)

In (4) I have indicated the string of segments as pairs of brackets. Above I referred to such pairs of brackets as slots, to which features or feature bundles present on different tiers are linked. It turns out now that these slots can be interpreted as the smallest units, the terminal symbols, of the phonological constituent structure. Putting the two theories together in this way we can no longer look upon a phonological representation as a twodimensional object. In fact this would have been impossible anyway for a case in which two features were autosegmentalized. This explains why the theory that is discussed here is called three-dimensional (as well as nonlinear). The combination of the autosegmental conception and the two-sided hierarchical conception has another consequence. An important insight captured

in

autosegmental

phonology

is

that

the

relation

between

autosegments and slots need not always be stipulated. In certain cases this relation is predicted by rule. In the simplest case the relation can be brought about by associating autosegments to slots in a directional fashion, going from left to right, associating autosegments and slots in a

8

one-to-one fashion: (5)

[+F] [-F] [+F] ... 1 i i ! ! I [' ] [ ' ] [ ' ] . . .

The dotted lines indicate the structural change of the rule that introduces the association lines, which are themselves represented by a closed line (cf. 6). If a one-to-one association was the only possibility it would be less obvious to distinguish between autosegments and slots in the first place. However, autosegmental phonology was invented precisely because the association is not always one-to-one. The original claim of autosegmental phonology is that, where deviations from the one-to-one pattern arise, the number of autosegments is different from the number of slots. Two types of situations may arise. Either there are more slots or there are more autosegments s (6)

a.

[+F] [-F]

II

b.

[ ] [ ] [ ]

[+F] C-f: [+F]

II

[ :[ :

According to the theory of autosegmental phonology representations as in (6) may lead to cases in which features have either wide or narrow scope: (7)

a.

[+F] [-F] I

[ ] [ ] [ ]

b.

[+F] [—F J [+FJ I [

] C

L-'-' ]

In (7a) we find spreading and in (7b) dumping; these are technical terms that I will continue to use here. Ignoring the issue as to whether spreading and dumping are the norm or the exception, it will be clear that a mismatch between the number of slots and the number of segments always holds within a particular

domain. The most general position

is that

autosegmental

association can in principle be bound to all domains that the theory defines. This includes morpho-syntactic domains as well as phonological domains. Let us be more precise about what it means to say that association is bound to a particular domain. We must make a distinction between two types of cases. Suppose we have one autosegment and three slots. It is then either the case that the autosegment is already associated to one of the slots in

9 the lexical representation by stipulation (because it is not predictable b y rule) or that there is no such pregiven association. If the autosegment is lexically associated we might simply say that the association rule is bound to a particular domain D: (8)

(r []. [][, ] )D C+F:

The autosegment that is associated to the first slot in D will spread to the other slots in this domain. But now consider the other logical possibility where the autosegment is not associated to any particular slot, and suppose furthermore that the representation consists of several autosegments and D's in a row. To make this more concrete imagine a language in which whole syllables are characterized by the presence or absence of a particular feature, such that all segments belonging to the same syllable are either C+F] or [ - F ] . A case in point could be the phenomenon of emphasis occurring in many Arabic dialects

(9) ([

(see Van der Hulst and Smith

C-F]

[+F]

] [ ] ) ( [ ] D

c :

1982b):

C-F] [

])

D

(c

c

]

]).

This time it is clearly insufficient to say that the association rule is bound to a particular domain, because it is not clear to which syllable each autosegment should go in the first place. Since (11) is the association that we want to derive, (10) seems to be a more appropriate input representation than

(9):

(10)

[-F] [

]

c

]

D

[

]

[-F]

(ID (C

]

C

C-F]

C+F] [

]

[

:

D

L

[

:

D

(C

]

C

]

J:

D

„

C-F]

C+F] ])

:

C

])

D

(C

]

C

])

D

It turns out to be the case then that we must be able to express that a particular domain comprises not only a sequence of slots, but also one or more autosegments. The following figure illustrates this conception

of

10

phonological representations ("AS" stands for autosegment): (12) ... word level AS1 s foot level AS 1 s syllable level AS 1s segment level AS's

morpheme level AS's word level AS's phrase level AS's

To mention just one other example, in most vowel harmony systems we find morpheme level autosegments. I should emphasize that the view on the combination of autosegmental theory and the theory of domains presented here is not a commonly accepted view, to the extent that the issue is addressed at all. The study of autosegmental features in domains other than the morpheme or the word (usually meant as a morpho-syntactic domain) has so far not supplied us with very well documented examples. The present view is advanced in Van der Hulst and Smith (1982b), who were inspired by Hart (1981), where a slightly different view is offered. In forthcoming work Vago applies and elaborates this version of autosegmental phonology, where autosegments occur on different levels. The above suffices to give the reader an idea of the theory of nonlinear phonology. In the next section I will discuss the various types of data that have played a crucial role in the emergence and further development of autosegmental phonology.

1.3. Aut.osegment.al phonology

1.3.1. The characterization of complex segments The standard theory is characterized by what Goldsmith (1976) has called the "absolute slicing hypothesis". An abstract representation of speech sound is split up into slices, called segments. Each slice is specified with

11

exactly one value for each feature of the total set of features that is required to represent speech sounds. Hence segments have no linearly ordered subparts. In the majority of cases segments can be interpreted as functions from points in time to a particular

state of articulatory

organs. So if a segment is specified as [+nasal] this means that at the time of producing this segment the velum is in lowered position. In actual life it may be the case that the velum is lowered slightly earlier causing a preceding segment

to be nasalized during part of its production. The

absolute slicing hypothesis embodies the claim that such a half nasalized vowel is not represented as [-nasal, +nasal]. The definition of segments on which this hypothesis is based only allows specifications such as [@F], where @ ranges over + or -. In the SPE theory it is also possible to have an integer as the feature value, indicating a certain degree of, for example, nasalization. Since in this example nasalization is dependent on the presence of a neighbouring nasal, the integer is only required as part of the

phonetic

representation,

and

not

as

part

of

the

underlying

representation, where the vowel in question is specified as [-nasal], thus abstracting away from the nasalization that occurs at the surface. In some cases, however, we cannot abstract away from the fact that an articulatory state changes during the production of one slice, because in such cases the change is an intrinsic property of the segment, i.e. it is not dependent on the

presence

of

some

neighbouring

sound.

Examples

are

numerous:

affricates, pre- and postnasalized consonants, pre- and postaspirated consonants, (short) diphthongs (see Ewen 1982 for a discussion of many of these segment types, usually referred to as complex segments) . During the production of an affricate we have a change from a stop to a fricative. Strict obedience to the absolute slicing hypothesis requires that we characterize such segments with a feature that is specified with an integer or a separate feature that directly refers to this change. In SPE the first possibility is blocked because all features are binary at the phonological level. The new feature to characterize affricates is called [delayed release]. Similarly, we will need features like [prenasal], [postnasal], [preaspirated], [postaspirated], [diphthong] etc. With respect to tones too, there is need for what we might call here contour features (as opposed to level features). In this case we need features like [rise] and [fall]. There is no objection as such to adding these features to the inventory. The reason for questioning this strategy comes from the fact that we miss certain generalizations by using contour features. The crucial argument has been advanced by Anderson (1976) with special reference to the contour features [prenasal] and [postnasal] and by proponents of autosegmental

12

phonology (Goldsmith 1976) with reference to tonal contour features. Consider the following example. In many African tone languages a phenomenon occurs called downdrift. The term downdrift refers to a gradual lowering of the pitch height of tones that are phonologically speaking the same. More specifically we find that a high tone is lowered slightly when it follows a low tone. So in a sequence . . .HLH. . . the second H is lower in pitch than the first. Sequences of high tones stay at the same pitch height. The following rule expresses this fact: (13)

H

->

( 1 1H 1 indicates

!H 1

/

L -

lowered high' ) . In languages that have both downdrift and

falling tones it may be the case that H is lowered after both a low tone and a falling tone. Hence the rule must be complicated in that case: (14)

H

—>

1H

/

{L, F}

-

Conjunctions comprising contour tone features and level tone features are not exceptional. They show up again and again and may constitute the norm rather than the exception. When the context bar is on the right L appears together with F, when it is on the left L appears with R(ise). Recurrent conjunctions require an explanation. There must be something that the conjoined environments have in common and our formalism must be able to express this. In the case at hand an explanation is available if we decide to abandon contour features and replace them with sequences of level features. So F is replaced by a sequence consisting of a high tone feature and a low tone feature (HL) . It will be clear that given such a move we can return to the simpler downdrift rule in (13), but more importantly we no longer face the problem of recurrent conjunctions. The mere decision to abandon contour features does not necessarily entail that we must also abandon the absolute slicing hypothesis embodied in SPE. In fact the decision to eliminate tonal contour features had already been taken by Woo (1969) . The conclusion she drew from this was that it is impossible for short vowels to have contour tones. Assuming that long vowels can underlyingly be represented as a sequence of two short vowels Woo predicted that only long vowels can bear contour tones. There are indeed cases where contour tones are not permitted to occur on short vowels. A case in point is Lithuanian where complex tones (i.e. rising or falling tones) can only occur on long vowels or sequences of short vowels

13

and sonorant consonants (cf. Kenstowicz 1970). This in itself supports the view to represent these tones with two level features. Languages of this type can be said to have a constraint prohibiting more than one tonal feature linked to a single segment. Unfortunately the prediction as such is false. There are many cases where short vowels bear a contour tone, either underlyingly or at a later stage in the derivation. It seems then that our conception of segments must be altered in order to allow single segments to have two different specifications for one feature. The same conclusion follows from Anderson (1976), who gives an argument that

is

completely

parallel

to

the

"downdrift-argument"

involving

nasalization of vowels preceding either a nasal consonant or a prenasal consonant. In the cases he discusses there is no evidence for representing the complex consonants as two segments underlyingly, hence the decision to represent a prenasal consonant as [+nasal] and [-nasal] directly implies that we must revise our conception of segments. The model that has been sketched in the previous section finds part of its motivation in supplying a representation of complex segments, using sequences of level features. As I already indicated above (cf. 3), in this model a complex segment is represented as consisting of minimally three tiers (cf. 15a). This view on complex segments is not the only one that one will find in the autosegmental literature. Clements and Keyser (1983) represent complex segments by associating two fully specified segments to a single slot (cf. 15b), while Kaye and Vergnaud (1984) propose to represent complex segments without assuming linear order of the relevant components (as in 15c): (15)

a.

+N

-N

[

]

[N]-tier central tier segmental tier

b.

m

b

[

]

c.

m C

]

b The view expressed in (15a) relies on the fact that we may have segment-

14

level autosegments, i.e. autosegments that do not automatically spread to neighbouring

segments.

The alternative

would

imply that, given

the

presence of complex nasals in a certain language, all segments for which the feature [nasal] is relevant, are associated with this feature, that is represented on an autosegmental tier, which itself does not belong to a specific domain. By saying that complex segments can be characterized in terms of segment-level autosegments we do not exclude the possibility that complex segments may arise by being associated

to two higher

level

autosegments. This is typically the case with contour tones in African languages, if a short vowel comes to bear a contour tone when a neighbouring vowel

is deleted.

In most

such cases

the tones are

morpheme-level

autosegments: (16)

o w 0 This example is taken from Elimelech (1976) . Here I will not try to decide on the question as to how complex segments must be represented, because the consequences of the various alternatives for the theory as a whole are not always discussed in the literature and hence not completely clear. The essential point is that in the more widely accepted alternatives (expressed in 15a and 15b) complex segments are characterized in a way that is incompatible with the absolute slicing hypothesis. This is not the case in the proposal advanced by Kaye and Vergnaud (1984), so it remains to be seen to what extent their theory makes crucial use of different tiers in order to characterize complex segments.

1.3.2. Supporting arguments for representing features on independent tiers The study of tone has given the main impetus to the development of autosegmental phonology and its application in this area has convinced many more than its application to other areas such as vowel harmony. The first phenomenon that supports the autosegmentalization of certain features involves what has been called stability. In (16) an example of vowel deletion was given, leading to the emergence of a contour tone. However, I did not stress the fact that, while the vowel was deleted, the tone stayed behind. Given the autosegmentalization of tonal features this phenomenon does not strike us as very unexpected. This is precisely the

15

point. By representing features on different tiers we predict independent behaviour and this is what we find. Of course independent behaviour does not always involve deletion, but this is something that can be considered as the extreme case. A short reflection on the issue will reveal that a strictly segmental (i.e. SPE type of) approach is going to miss the point completely. To handle the facts in (16) we need two rules. The first rule copies the tone of the vowel to be deleted on the next vowel. The second rule deletes the vowel. In an analysis of this type we have not explained why the tone has stayed behind. A second and even stronger argument in favor of separating a tonal and a segmental tier involves defective morphemes. In tonal analyses one will encounter morphemes that consist exclusively of tone and also morphemes that (although they contain a vowel or, better, a tone-bearing unit; TBU) have

no tone.

In a strictly

segmental

framework

problematical,

especially the

"segmentless"

such

entities

tones. Within the

are

auto-

segmental model defective morphemes are not real anomalies. If morphemes consist of two independent tiers then there is no reason why one could not be lacking. A third argument involves the phenomenon that words consisting of different numbers of syllables may show a behaviour which strongly suggests that they have the same tonal melody. Consider the following example, discussed in Odden (1980). In Shona a certain class of prefixes triggers a rule that lowers sequences of high toned syllables in the stem, no matter how many syllables this stem contains. This suggests that sequences of high toned syllables are associated with a single autosegment H. The rule triggered by the prefixes changes this H to L. Another argument involving the notion tonal melody is the following. In Edmundson and Bendor-Samuel (1966) Etung is described as having the following melodies on words consisting of one, two and three syllables: (17)

1 syllable

2 syllables

3 syllables

L

L L

L L L

H

H H

H H H

LH

L H

L H H

HL

H L L HL

H L L L H L

H LH L LH

L L H

H HL

H H L

H L H

16

The

notation Hi, indicates

a contour

tone. Goldsmith

(1976:132-134)

discusses these facts: The conclusion is clear: we have in Etung a small class of tone formulas that may be spread over words of one, two or three syllables, proceeding from left to right. These melodies are L, LH, HL, H, LHL, LLH, HHL, and HLH. The occurrence of contour tones that motivate Leben's left-to-right mapping in Mende occur here, but the melodies HHL and LLH make it clear that the Obligatory Contour Principle is too strong in fact. There

are several

issues

that

are relevant here, but let us first

concentrate on the argumentation in favor of the autosegmental mode of representation. The facts from Etung provide us with three arguments. Firstly, it is clear that in a strictly segmental framework it is not possible to express the generalization that there are eight melodies, simply because the notion melody has no status in such a framework. Secondly, the gaps receive a straightforward explanation, which would not be available if we were working with contour features. We can account for the gaps in table (17) by saying that Etung has a constraint which prohibits more than two different tone features being associated to one tone-bearing unit (cf. Halle and Vergnaud 1982): (18)

T

T

T

V where l=/=2=/=3 Thirdly, consider the distribution of contour tones: they only occur on the last vowel. As stated in the preceding section and also in the quote from Goldsmith (1976), tones are associated to tone-bearing units (TBU's) in a one-to-one fashion, going from left-to-right. If the tones outnumber the tone-bearing units, left-over tones are dumped on the last tone-bearing unit. This explains why contour tones in Etung are found on the final vowel only. If tones were characterized by segmental features it would remain a mystery why a bisyllabic word cannot have a contour tone on the first vowel. At this point the most important arguments supporting the autosegmental model have been discussed. Proceeding on the assumption that the model has been

sufficiently

motivated

I will

discuss

a

number

of

essential

17

conventions and principles in the next section.

1.3.3. Principles of autosegmental phonology In the quote from Gold smith (1976) reference is made to two principles: the Obligatory Contour Principle (OCP) and left-to-right mapping. According to the OCP, proposed by Leben (1971, 1973), we exclude the possibility that adjacent autosegments have the same value. Hence we exclude ...HH ... or ... L L ... in all cases. Goldsmith points out that the facts from Etung are problematical for such a view since the total set of melodies comprises all eight possibilities that we have with two tones and at most three of them in a row. If the OCP is a genuine universal principle there should be only six melodies: (19)

NO OCP

OCP

L L L

L

L H H H H H

H iL II H H

H L L

H L

H H L

H L

H L H

H L

L H L

L H

L L H

It seems that Goldsmith is correct in saying that the OCP cannot be maintained as a universal principle (cf. Halle and Vergnaud 1982). Even Mende, the language that Leben used to argue in favor of the OCP, appears to have melodies of the forbidden type (cf. Dwyer 1978, and Conteh et al. 1983). This reduces the OCP to a principle that allows one to collapse identical autosegments if there is no reason to leave them separate. The mapping rule too dates back to Leben (1971). In his original conception this rule merged the tonal tier and the segmental tier, so that the surface representation conformed to the SPE-theory. In this form the rule is also found in Williams (1971), with the difference that Williams' version of the mapping rule did not contain the dumping clause. It will be evident that the mapping-as-merger rule could not be maintained because the whole issue of short vowels with contour tones remains problematical in that case. Hence Goldsmith (1976) proposes to let the rule introduce

18

association lines, leaving the tonal and segmental features on different tiers. Leaving tones and segments on different tiers raises the question how they are related. This question is ambiguous. There are, to be precise, two questions to be answered: (18)

a. What constitutes a wellformed relation between the tonal and the segmental tier? b. How does this relation come into being?

Goldsmith provides the answer to the first question by formulating a Wellformedness Condition (WFC): (19)

Wellformedness Condition a. Association lines do not cross (no X-ing) b. All TBU's are associated to at least one tone c. All tones are associated to at least one TBU

The answer to the second question involves first of all three Association Conventions (AC1s): (20)

Association conventions a. Mapping Insert association lines between one tone and one

TBU

-going from left-to-right/right-to-left -starting with the left/rightmost tone and TBU b. Dumping Left-over tones are associated to the nearest TBU (to their right/left) c. Spreading Left-over TBU's are associated to the nearest tone (to their left/right) The part between parentheses is relevant if there is a choice and the first option mentioned in each case is considered to be unmarked. In the study of tone I am aware of no example where tones are associated from right-to-left, so it is perhaps not the case that we need the different options for all three association conventions. I refer to Haraguchi (1977) and Clements and Ford

(1979) for further detailed discussion of the association con-

19

ventions. Goldsmith's original idea was that the WFC holds at every level of the derivation. If in the course of the derivation a violation of the WFC arises (for example because a TBU is deleted) the relevant AC applies to make the representation conform to the demands of the WFC. I will return to this view below. Subsequent

developments

have

shown

that

the

AC' s are

not

only

parametrized in the sense that the left/right options must be fixed, but also in the sense that it depends on the language in question whether they apply at all. In Clements and Ford (1979) it is argued that tones that are left over after one-to-one mapping or set afloat in the course of a derivation may either be left floating

(in which case they are not

phonetically interpreted) or be associated by a language-specific rule. This position, which was also embodied in Williams'Mapping Rule, is adopted in Halle and Vergnaud (1982) as well. A consequence is that the third clause of the WFC must be eliminated and also that dumping can no longer be regarded as a universal convention. Halle and Vergnaud (1982) go one step further and argue that automatic spreading too must be rejected. Whether or not spreading takes place is a language-particular

matter,

they

argue.

In

languages

that have

no

automatic spreading TBU's that remain unassociated will surface with the "unmarked tone value" that each TBU is supposed to have as a segmental specification. Hence clause b of the WFC can be eliminated too. Instead of assuming segmental values, others (e.g. Kiparsky 1983, Pulleyblank 1982) have proposed to leave all segments unspecified for tone. At the end of the derivation TBU's that are not associated with an autosegment are supplied with a default value. The result of the development just discussed is that both the WFC and the inventory of AC's are considerably reduced: (21)

WFC: no X-ing AC: mapping

In Pulleyblank

(1982) we find a defence of this type of impoverished

autosegmental theory. A sceptical tonologist could argue at this point that the universal principles of autosegmental phonology given in (21) embody the claim that the least marked situation is the one in which each TBU is associated to precisely one tone (and vice versa) and furthermore that a model in which tones are segmental features expresses this claim much more straight-

20

forwardly. Only the first part of this statement is correct, however. The second part, meant as a rejection of the autosegmental representation of tones, totally ignores the arguments that have been provided in favor of this model and can therefore not be taken seriously. The issue whether spreading and dumping are universal conventions or not is logically independent from the fact that a strictly segmental model is incapable of explaining both the representation and the particular distribution of contour tones. The same holds for the other arguments that were discussed in the previous section. Let us proceed with a discussion of the association conventions and turn to the last one, viz. mapping. In Goldsmith (1976) it is realized that the mapping rule as formulated in (20a) is inadequate to handle all types of tone languages. The following example reveals the crucial point to be made here. In Haraguchi (1977) the autosegmental model is applied to Japanese dialects. In Standard Japanese words have a tonal pattern starting with at most one low toned syllable, followed by a rise to high, that lasts for at least one syllable, possibly followed by a fall to low. Consider the following examples: (22)

a.

i |n o t i

b.

k o|k o |r o

c.

a11 a m a

Haraguchi's analysis is to assume a word melody HL and a rule that lowers the high tone of the first syllable if the second syllable is high as well. Hence he starts out with the following underlying representations: (23)

a.

H L

b.

H L

inoti

c.

H L

kokoro

atama_

If we let the AC's apply, the representations in (23) will be changed into those in (24i), whereas a correct characterization of the tonal pattern of these words requires (24ii) (ignoring initial lowering for the moment): "H L

b.

H L

b.

1 K\ 1 l _inoti 1K

_inoti_

H L

c. " H L

H L

c.

1K J kokoro NX

kokoro_

1 K 1 l\ atama

H L ^ atama

21 Let us keep in mind here that Haraguchi assumed that both spreading and dumping are universal conventions. Only in the first case we obtain the correct result. The problem at issue is solved by calling upon the notion of a starred syllable, which is essentially the same notion as accented syllable, as used in McCawley (1977). It is claimed that in each word a particular TBU is designated as starred and furthermore that mapping is preceded by what is called an Initial Tone Association Rule (ITAR), that can be formulated as follows: (25)

Initial Tone Association Rule Associate the H tone to the starred syllable

Once this ITAR has applied the AC's can take over, which, in Haraguchi1s version of the autosegmental model, means spreading, dumping and mapping: (26) i.

H L i

b.

c.

\

inoti

_*

H L

H L \\ \

ITAR \

kokoro

atama

H L

H L

kokoro

atama

*

b.

H L \\ \S.

*_

AC'S inoti

The presence of dumping necessitates adding a contour tone simplification rule to the analysis, to explain why the final syllable of atama does not have a falling tone. The necessity to have rules like (25) entails that it is, strictly speaking, not correct to say that the WFC holds at every level. Rather we must say that the WFC comes into play after all language-specific rules, that refer to floating autosegments (like the one in 25), have applied. I refer to Odden (1984) for a discussion of this point. Languages using stars are traditionally called pitch-accent languages, whereas languages not using stars have been referred to as (true or lexical) tone languages. An illuminating discussion of this distinction is found in McCawley (1978), and also in Clements and Ford (1979). In the examples just given the location of the star is unpredictable and hence a lexical property of the words in question. Haraguchi also analyzes cases, however, in which the location of the first syllable with a high tone

22

is predictable. This raises the question as to whether in such cases there is a rule that assigns stars, applying prior to (25), or whether we assume another type of ITAR: Assign the H tone to the TBU in position P

(27)

If (27) is adopted the next question might be whether we need the star at all. It would also be possible to say that alleged lexically starred syllables are simply lexically associated to a high tone. Hyman (1982) defends the view that we can do without stars and this position is also held by Pulleyblank (1982) . Going into this issue in more detail would go beyond the limited goals of this chapter, however. The consequences of eliminating stars for an autosegmental analysis of the data analyzed in Haraguchi (1977) are discussed in Van der Hulst (forthc.). The autosegmental model discussed in this section has not only been applied to tone, but also to other suprasegmental phenomena such as intonation and vowel harmony. In the next section I will show how the autosegmental model has been extended, limiting myself, however, to the area of vowel harmony. For the autosegmental approach toward intonation I refer

to

Liberman

(1975),

Goldsmith

(1976)

and,

in

particular

to

Pierrehumbert (1980).

1.3.4. Extension of autosegmental theory to vowel harmony The proposal to extend the empirical domain of the autosegmental theory to vowel harmony is made in Clements (1976). Clements claims that it is possible to apply autosegmental phonology to vowel harmony, and that this model explains a number of essential properties of this phenomenon. The crucial point is this. If we say that a certain feature [F] is autosegmental in a language L this is essentially all we have to say to account for the fact that L has harmony involving the feature [F] . In particular we do not have to write a language-specific harmony rule. Assuming the WFC as a universal principle, the autosegment [@F] will be associated with all "[F]-bearing units" that are in its scope. Let me clarify this with a schematic example. In many cases of harmony affixes harmonize to the stem they are attached to; this

is

called

root-control

by

Vago

(1980a) who

offers

introduction to the phenomenon and analysis of vowel harmony:

a

useful

23

+F

(28)

[

]

[ ']

[

J

L

J

L

]

C

]

Thus we explain why harmony is unbounded and bidirectional. In the current segmental treatments there are various differences. With respect to the representation of roots we have at least two choices. Either we specify one of the vowels (usually the first) as [+F], leaving the others unspecified, or we specify all vowels as [+F] in the underlying representation. In the first case we need a mirror image iterative rule to account for the values of the remaining stem vowels and the vowels of the affixes. In the second case we need two rules, a redundancy rule specifying that all segments of a stem must agree in their value for [F] and in addition the mirror image iterative rule to account for the affix vowels. Of course, various variants can and have been proposed. The two special properties of the segmental rule that is required in both variants of the segmental approach (viz. mirror image,

iterativity)

correspond exactly to the two things that could not be otherwise given the autosegmentalization of [F] (viz. unboundedness and Now

suppose

that the

theory

of

autosegmental

bidirectionality) . phonology was not

independently motivated by tonal phenomena. It would then be possibile to argue that vowel harmony is segmental, but that we do not need language specific rules, because there is a universal convention which carries out the function of the

mirror image iterative rule:

(29)

The autosegmental and the segmental approach are now equally simple: there is only one [+F] specification, one universal convention and no languagespecific rule. The only problem that the segmental approach faces is that it is to a certain extent arbitrary to specify the first vowel as [+F] rather than some other vowel. In a toneless world this would provide us with one argument in favor of the autosegmental approach. Clements

(1976, 63) argues that,

in addition,

the

autosegmental

approach offers a better approach toward exceptions: Within the present framework, since vowel harmony is viewed as a consequence of general well-formedness conventions rather than as a set of language-particular rules, exceptions to harmony cannot be

2-4 accounted for by [...] rule features. Rather, they must in all cases be built directly into the lexical representation of morphemes: vowels which invariably exhibit a given feature F are lexically bound to that feature, unless (as in the case of neutral vowels [...]) there is an independent explanation for their failure to show surface alternation. Segments

that

are bound

to

an

autosegment, before

the

association

conventions (AC's) apply, are called opaque. Segments may be opaque on an item-to-item basis or predictably. An example of predictable opaqueness is found in Akan where the low vowel a is invariably linked to the autosegment [-ATR] (cf. Clements 1981). This fact is represented by linking all the occurrences of the low vowel to [-ATR] by a language-specific rule which takes precedence over the AC's: (30)

-A C+low]

->

This treatment of opaque segments explains a number of properties of such segments. They are not only exempted from undergoing a harmonic change, they also block the spreading of a harmonic feature to their left to segments to their right (and vice versa). Instead those segments become associated to the autosegment of the blocker. In short, opaque segments are nonundergoers, blockers and spreaders (cf. Clements and Sezer 1982 for more details) . Consider the following example from Turkish, taken from the article just referred to: (31)

-R ' gEl -B

+R N

I IyEr

^ Im

I--"' +B

The suffix /lyor/ is disharmonic. Its second vowel does not harmonize to the value of the root. Observe that the autosegment of the root cannot spread further than the first vowel of the suffix. The second vowel then is opaque: it does not become associated to the autosegments of the root, it blocks propagation of both [-R] and [-B] and it finally spreads to the following suffix. A strictly segmental approach of this type of phenomenon requires

25

exception features of various types. Generally speaking it can be said that part of the autosegmental program is to eliminate exception features from the theory. The crucial point is that exceptionality in the autosegmental framework does not lead to the formulation of special

(sub)rules or

assigning special features. The irregular behaviour (being non-alternating) is encoded in the most straightforward way there is, given the autosegmental formalism. It seems then that an autosegmental approach toward vowel harmony has several advantages and this would still be true, even if the crucial principles were established to explain this phenomenon only. But tonelanguages exist and this implies that the autosegmental approach toward vowel harmony merely makes use of principles that have been independently motivated. The question should therefore not be: why should we treat vowel harmony autosegmentally? (as Ringen 1984 puts it), but rather: why should we continue to treat it segmentally? Stewart (1983) proposes a segmental approach toward vowel harmony that is different from the two mentioned above in that use is made of word level wellformedness conditions (called word structure conditions) and rules (called Automatic rules, or A-rules) that specify how violations that may arise in the process of affixation are eliminated. Stewart applied his theory to Akan for which Clements analysi s. Clements

(1981) had given an autosegmental

(1984) offers a critical assessment of Stewart's

theory. It may be the case that Stewart's approach (and the somewhat similar approach offered in Crothers and Shibatani 1980) is the most preferred linear one, not facing the problem of stating the same generalization twice (i.e. as a morpheme structure rule and as a phonological rule) or the problem of using 'blanks' in phonological representations. A decision between the two types of treatments must then be made on the basis of their capacities to deal with complex sets of data in an explanatory way. At this point it is simply too early to say that either Stewart's approach or the autosegmental approach is more successful in this respect. So far I have made use of a highly simplified (but essentially correct) picture of vowel harmony. I will now investigate in somewhat more detail the consequences

of

treating

vowel harmony

autosegmentally,

adopting a

slightly more critical perspective with respect to the claim that the principles used to explain vowel harmony and the principles used to explain tone are indeed the same. My intention here is to raise a few questions, rather than to provide answers. In sections 1.3.1.-2 a number of reasons were discussed for treating tone autosegmentally. A sceptical student of harmony could argue now that

26

by exploiting the autosegmental machinery to treat harmony we make a number of predictions that are not all borne out. The autosegmental approach toward harmony predicts that it is possible to find all the phenomena that we also

find

in the

study of tone.

If these phenomena

are

absent

explanations for this absence must be given. A striking difference between tonal and harmonic autosegments is the fact that whereas on the tonal tier we typically find "melodies", i.e. sequences of different tones, we never find melodies on the harmonic tier, excluding the cases where opaque segments are involved. That harmony always involves "degenerate melodies" (i.e. melodies consisting of one unit) has as a consequence that one-to-one mapping (as part of the universal set of AC's)

is hardly

relevant,

whereas

spreading

is crucial. From

this

perspective tones and harmonic features appear to have little in common, considering the claim, discussed in the previous section, that spreading is not a universal convention in the area of tone. One might even go as far as to suggest that the theory of tone and the theory of harmony make use of independent, principles. For tone the crucial convention

is mapping.

Spreading comes in as a language-particular rule (at least according to some phonologists) . For harmony spreading is the essential universal convention since mapping is only required to link the harmonic feature to the leftmost vowel. As for the third possibility i.e. dumping, both areas are again very different. In vowel harmony systems it is never the case that two autosegments are associated to one P-bearing unit, i.e. multiple association is forbidden. This means that dumping is not possible, either as a universal convention or as a language-specific rule. On the other hand, one tone-bearing unit may be associated to two tones and this is typically the result of dumping rules. This fact even plays a crucial role in the argumentation in favor of the autosegmental theory of tone. Summing up, we arrive at the following pictures (32)

Tone

Harmony

a. b.

melodies multiple ass

yes yes

no no

c.

mapping

yes

no

d.

dumping

e.

spreading

yes ?

no yes

These five differences are not logically independent, of course, but that does not take away the impression that harmony is unlike tone in important

27

ways, despite the fact that we can exploit the same formalism

(i.e.

independent tiers and association lines) in both cases. The differences require an explanation, which autosegmentalists have not given so far. In sections 1.3.2.-3 I discussed several diagnostic features for an autosegmental

treatment. We have

seen

that

some of these

features

(melodies, multiple association) are absent in the case of vowel harmony. Two other important diagnostic phenomena involved defective morphemes and stability. The question that naturally arises is whether or not we find these diagnostics in vowel harmony systems? One type of defective morpheme has already been mentioned implicitly. Affixes that harmonize with roots can be represented as defective in the sense that they have no segment on the harmonic tier. Now do we also find suffixes consisting solely of the harmonic tier? One can think of several examples here that fit the description of such a type of defective morpheme. In their discussion of African vowel harmony systems Hall et al. (1974, 250) refer to: "... "Ablaut" . . .a change of vowel series for which there is no overt conditioning factor - and "Reversed Category Shift" - the change of a [+ATR] vowel to a [-ATR] vowel. In both of these cases we believe that we are not dealing with autonomous morphophonemic processes but rather that these represent simply special cases of the Vowel Harmony rules which we have already seen. In the case of [ . . .] "Ablaut" it is possible to speculate that historically there was a Dominant suffix present which has since been lost. The "Dominant suffix" that is referred to can synchronically be analyzed as a suffix consisting solely of the autosegmental tier on which the feature [ATR] is represented. A second example has been discussed in Van der Hulst and Smith (1982a, 22-23). In Terena, as described by Bendor-Samuel (1960), the first person singular of the verbal forms and the possessive of nouns is expressed by nasalizing all vowels and sonorant consonants from left to right up to the first obstruent, which appears as prenasalized. We are dealing here with a prefix consisting solely of the feature [+nasal]: (33) a.

b.

[+nas] N ayo 'my brother'

t+nas]

I

[-nas] "owuku

'my house'

[+nas]

[-nas]

V

piho

'I went'

28

The example coming from Hall et al. (1974) also shows us the phenomenon of stability (in the diachronic sense), if Hall et al. 's speculation turns out to be correct. I know of no examples where stability plays a role in the synchronic derivation, i.e. no case in which for instance a vowel is deleted and its ATR value appears on another vowel. Stability, in the diachronic sense, is probably behind the following examples as well. The first example is given by Clements (1981) in his analysis of Akan vowel harmony. A set of roots starting with a sequence consisting of one out of a specific set of consonants and the opaque low [-ATR] vowel a selects prefixes that are [+ATR]. The analysis of such roots offered by Clements is as follows: (34)

+A w 1 j A n I

What we find here is that an autosegment introduced by one morpheme is realized on another. Since two of the examples, discussed here, involve the feature [ATR] it is instructive that we can also give an example involving the feature [back]. In Hungarian we find backness harmony.

Suffixes

harmonize with roots, i.e. the system is root-controlled. The front unrounded vowels are neutral, i.e. they may cooccur in the same root with back and front rounded vowels. Roots that contain only neutral vowels normally take front suffixes. There is a set of roots, however, containing only neutral vowels that select suffixes with back vowels. We can account for

the

behaviour

of

these

roots

by

giving

them

the

following

representation: —

+B viz

\

nAk

I refer to Van der Hulst (1984) for a detailed discussion of Hungarian vowel harmony. The conclusion that we may draw from the discussion so far is twofold. Firstly, the autosegmental model is perfectly capable of handling the phenomenon of vowel (and consonant) harmony in both an elegant and explanatory fashion, so there is sufficient reason for adopting this theory, even ignoring the fact that it is (to some extent at least) independently motivated for the analysis of tone. Secondly, an autosegmental analysis entails certain predictions involving autosegmental

29

melodies, multiple

linking

and floating

features. Not all of these

predictions are borne out and this entails that at least some principles of autosegmental phonology are true for tone but not for harmony and vice versa. I have ignored here several important and interesting aspects of the autosegmental approach toward vowel harmony, such as the proper treatment of neutral vowels or harmony systems in which root-control is absent (so called dominant harmony systems). The treatment of neutral segments is discussed in Van der Hulst (1984). I trust, however, that the present discussion is sufficiently detailed to give the reader a good idea of the principles of autosegmental phonology. In the remainder of this book the focus will not be on the treatment of tone, nor harmony. Yet in many places reference will be made to the principles of autosegmental phonology, which justifies the foregoing discussion.

1.4. Metrical phonology

My sketch of metrical phonology will be short, because the principles of this theory will be discussed in great detail in the remaining chapters of this book. In the broad sense metrical phonology is concerned with phonological domains and the prominence relations that hold

within these domains. The

theory of metrical phonology has arisen from Liberman's work on English stress and intonation (Liberman 1975). The treatment of English stress within this framework was elaborated in Liberman and Prince (1977), Selkirk (1980) and Hayes (1981, 1982). A general typology of stress systems, based on the metrical theory, was proposed in Halle and Vergnaud (1978) and Hayes (1981) . The basic idea behind metrical phonology is that prominence relations of an utterance can be characterized in terms of a constituent structure that is augmented with an S/W labelling. The S/w labelling expresses the fundamental claim that, within a particular constituent, one daughter is relatively strong with respect to her sister(s) . A second, independent idea is that nodes in the constituent structure are maximally binary branching. Hence the S/w labelling indicates, for each pair of sister nodes, which one is more prominent. The basic building blocks of the theory are then:

30

(36)

a.

^ S

b. W

^ W

S

The labels S and W have no fixed phonetic interpretation. In this sense metrical trees are abstract and uninterpreted. It is a wellknown fact that prominence

can be phonetically

realized in a number of ways. For a

typological survey of the ways in which prominence can be realized in natural languages I refer to Greenberg and Kaschube (1978). Binary branching trees, thus labelled, have one and only one terminal element that is exclusively dominated by nodes that are labelled with S (excluding the top node). This property makes them suitable to express properties of speech units that occur at one place in these units only. One such culminative

property is word stress (cf. Hyman 1977, Greenberg and

Kaschube 1978) . Suppose we want to say that in a language L all words have main stress on the final syllable. The labelled trees in (37b), taking syllables as their terminal elements express this fact adequately. The stress rule for L could be formulated as in (37a): (37)

a. Assign a uniformly right branching tree to each word Label each right node with S and each left node with W b.

Monosyllabic words (i.e. the first case in 37b) receive a tree structure according to the rule in (37a), but we cannot assign a label S, because this label indicates relative strength. Within a constituent that has one daughter only, it makes no sense to say that this daughter has such a relational property. Yet, taking stress to imply the potential of being associated

to an

intonational

pitch movement, monosyllables

can be

stressed. If they cannot be associated to a pitch moment they are clitics. This implies that metrical trees are interpreted according to the following rule: (38)

Within a constituent C main stress falls on the only syllable or on the syllable that is exclusively dominated by nodes labelled S

31 The idea of labelling nodes presupposes the presence of these nodes. Liberman's point of departure is that the constituent structure is given in the form of the morpho-syntactic constituent structure. This can be illustrated clearly with reference to compounds. On the assumption that the relevant

constituent

structure

is the one

that corresponds

to the

morphological structure the compound stress rule of English could consist of the following labelling rule: (39)

Label right node S iff it branches

S law

w degree

s'

w

requirement changes

A very similar proposal was advanced in Rischel (1972), who uses the labels +/- instead of S/W. Rischel also points out that in certain cases the labelled tree must be restructured somewhat to arrive at a satisfying characterization of the stress pattern of compounds, a point that he discusses in more detail in Rischel (1983) . This may be a reason to say that the hierarchical morphological structure of compounds is not so relevant at all. Below the compound level matters are slightly more complicated. Either we are dealing with words that have no morphological structure at all or the morphological structure is simply not the one that is required. I trust that the reader is familiar with the fact that in many languages one may distinguish between affixes that behave prosodically like stems, in which case the morphological structure can be used to assign the S/W labels to, and affixes that are integrated into the prosodic structure of their base, in which case the morphological structure appears to be irrelevant. The distinction will be discussed with reference to Dutch in section 2.2.4. The constituent structure that is to be labelled in complex words of the second type must be built up as part of the stress assignment procedure, as is the case when we are dealing with words that have no morphological structure at all. This is in fact what I did in the schematic example given above in (37) . But also above the compound level the syntactic structure is not the appropriate structure. It has been pointed out in various places that higher level prosodic tree structure is not isomorphic to the syntactic structure, although the former can be derived from the latter by means of a function. Nespor and Vogel (1982), following Selkirk (1982c) discuss this mapping function. They show that the constituent structure thus created not only provides a basis for assigning a relative prominence pattern to an

32

utterance, but also functions as a theory of phonological domains to which phonological rules appear to be sensitive (cf. Selkirk 1980b). Rischel (1983) argues differently and claims that above the compound level, we do take the syntactic structure as a starting point. The resulting trees are labelled and then altered by transformational rules. A similar proposal is advanced in Giegerich (1983). It is clear that in both alternatives it is acknowledged that prosodic structure is not isomorphic to morphological or syntactic structure, but can be derived from it (directly or indirectly). The Liberman and Prince theory embodies one other crucial innovation with regard to the treatment of stress. Prominence

patterns typically

involve more than just a single strong element surrounded by equally weak elements. If sufficient weak elements are present we will usually encounter a rhythmical pattern. Liberman and Prince propose two ways of dealing with rhythmical

patterns.

Within

words

such

patterns

arise

by

grouping

syllables together into feet. A foot is defined as a relatively strong syllable followed or preceded by a sequence of relatively weak syllables. To designate the strongest syllable of the word we must now assume that feet rather than

syllables are gathered in a uniformly branching tree:

(40)

A w A w /\w A s S W S W S W S W aaaaaaaa To account for rhythmical patterns in phrases Liberman and Prince propose a different strategy, i.e. a procedure to create such patterns that does not depend

on

the binary branching constituent

structure, but

crucially

involves another hierarchical structure, distinct from the tree, called the grid. A recent development of metrical phonology involves the claim that grid structure can be used to represent all aspects

of prominence

patterns and that one can do without binary branching trees and S/W labelling (Prince 1983, Selkirk 1984). In chapter 4 these variants of the metrical theory are discussed in great detail. Just like the autosegmental theory, metrical theory, developed to deal with stress, has been extended to other phenomena. In particular it led to a revival of some traditional ideas about syllable structure. It will be evident that the metrical notation is adequate to represent the wellknown fact that for syllables it is also true that one element (usually a vowel) is

33

more prominent than all other elements within that same syllable. In chapter 2 I will discuss the various views on syllable structure that have been developed since.

1.4. Contents of this study

In the next four chapters I will be concerned with the theory of syllable structure and stress. With respect to both areas the strategy will be to discuss various competing theories and to argue in favor of particular variants. The variants that are adopted are then applied to Dutch. Chapters 2 and 3 deal with syllable structure, chapters 4 and 5 with stress.

Chapter

2

The Representation of Syllable Structure

2.1. Introduction

In this chapter I will be concerned with the representation of syllable structure. Section 2.2.1. offers an introduction to the central issues that a theory of syllable structure has to address, indicating some of the current controversies, although no attempt has been made to give an historical account of all the syllable theories that have been offered in the past few years. In 2.2.2. and following sections I will then discuss three specific issues, viz. the role of the sonority hierarchy, syllable internal structure and finally the issue of syllabification. Throughout this chapter I am concerned with developing a model of syllable structure that will be applied to Dutch in chapter 3.

2.2. Theories of syllable structure

2.2.1. Introduction It is generally assumed that in developing a theory of syllable structure we face two tasks. Firstly, we must specify the notion "wellformed syllable (of language L)", and secondly, we must specify how an arbitrary string of language L is syllabified, i.e. parsed into wellformed syllables. Two separate issues can be distinguished when addressing the first task. We have to characterize the sequences of segments that constitute a wellformed syllable (the linear structure) and we have to determine whether and, if so, how these segments are grouped into subsyllabic units (the hierarchical structure). Performing these tasks within an autosegmental framework

implies

that we must

recognize

a third

issue,

viz.

the

38

association between the autosegmental tiers and the central tier. In chapter 1 this central tier was identified as consisting of the terminal symbols of the syllable. I refer to the relation between the autosegmental tiers and the central tier as the aut.osegment.al structure. We can then distinguish the following common core in many proposals advanced in the recent past for the analysis of syllables (cf. Selkirk 1982b): (1)

i.

the linear structure

ii.

the hierarchical structure

iii. the autosegmental structure (2)

(ii) c

]

c

:

c

:

[

]

c

:

c

:

(iii) (i) With regard to all three aspects a characterization comprises two parts, a universal and a language-specific part. I will briefly indicate what is involved in these two parts, referring to some of the different points of view that exist in the literature in order to provide some general background for the subject matter of this and the following chapter.

2.2.1.1. Linear structure: universal It has been widely observed that the shape of melodies (i.e. sequences of segments that must be associated to the central tier) that constitute a wellformed syllable is partly determined by the sonority of the segments: segments are more and more sonorous as we go from the edge of the syllable toward the center, which is by definition the most sonorous segment. On the basis of this particular sequencing of segments we can order segments on a scale, which, phonologists have claimed, can be independently motivated with evidence from language change, language acquisition etc. Following Selkirk

(1982b),

I will

refer

to this observation

as the Sonority

Sequencing Generalization (SSG). Although the SSG occurs frequently in the literature on syllable structure, it is not always clear in what sense it plays a role in the theory. Some refrain from giving a theoretical status to a scale on which

39

segments are ranked because of certain ordering problems that seem to arise. Thus

the SSG is said to be invalid because there are counter-

examples such as syllable-initial clusters with s followed by a stop in the Germanic languages or clusters like syllable-initial mgl or syllable-final pi in Russian. In the majority of cases, however, the linear structure of syllables is in conformity with the SSG. I will therefore adopt the point of view that the particular order in which segments occur in the syllable need only be specified where it deviates

from the SSG. Given a sonority scale (e.g.

obstruent piss and stretch, as well as those in which the "moved" part corresponds to the rhyme. The

fourth

and

final

argument

that CK discuss

concerns

rhyming

58

traditions. It is often said that the rhyme is that part of the syllable that "rhymes". But in the strict sense this is only true of masculine rhyme, where we can indeed observe that precisely that part of the syllable that concerns us here must be identical for the rhyme to be successful. But, as CK observe, there are also other forms of rhyme involving two syllables ("feminine rhyme") or three syllables (28)

("extended rhyme"):

Masculine Rhyme

Feminine Rhyme

Extended Rhyme

hoog - droog kaal - schraal

hoge - droge kale - schrale

hogere - drogere handelen - wandelen

There is one common aspect to these three forms of rhyme. The rhyming unit is a foot minus the onset of the head of the foot. In rhyming then one cannot say that the onset in general is irrelevant. It is only the onset of the head of the foot that is irrelevant: (29)

informed feminine & extended rhymes harten - parken kinderen - tintelen

These rhyming pairs are informed due to different 'non-head' onsets. Rhyme then is not a matter of "rhymes", but of feet minus the onset of the strongest syllable. The latter argument is not entirely unproblematical. One might argue that whatever syllable happens to be the head of the rhyming foot must be parsed according to the

onset/rhyme distinction and that therefore each

syllable must in principle then be parsed according to the onset/rhyme division. For CK it would be "wiser" therefore to regard rhyming as one of the types of external evidence, which, as we saw above, do not point in one direction. I conclude that, although CK are right on several points, there is some reason to be sceptical with respect to their claim that the nucleus gives us all the syllable internal structure that we want. It is important to realize that if CK were not convinced of the superfluousness of the categories rhyme, onset or coda (or any other category, except for the nucleus), they could easily have added another structural tier to their theory (CK, p. 19) . This brings me to a different, more principled objection to their approach. Apart from supplying us with the arguments against the metrical theory of the syllable that I have

just discussed

CK also claim that the

autosegmental theory of the syllable is more restrictive than the metrical

59

theory which makes use of binary branching trees. They argue that for any sequence longer than two segments ambiguity exists with respect to the binary branching tree that goes with it. In the case of three segments we have

two possibilities,

in the case of four segments we have

five

possibilities, and so on. It will be obvious that whatever the number of segments there is always only one flat structure. I claim that the autosegmental approach that CK advocate suffers from an equally damaging richness, however. Adding the possibility of simultaneous representations leads to a significant increase of the descriptive power of the theory: in essence it allows us to analyze the same string in almost as many ways as there are conceivable binary branching trees. To see this we only have to realize that CK allow the possibility to have structural tiers that are not piled on top of each other, but that are in a sense competing. Nothing precludes the possibility then to add a structural tier with a unit called A to which all segments up to and including the first segment of the nucleus are linked or any other conceivable substring of the syllable. One might wonder why CK do not simply accept that the structure of syllables is not flat (whatever the details of the non-flat structure may be) . The reason is, I think, that non-flat structure seems to clash with the autosegmental mode of representation. A flat structure allows us to employ the quite elegant autosegmental conventions in building up syllabic structure in the process of syllabifying a string of segments. This is true (if the segments are grouped in flat syllables), but apart from that point I see

no

evidence

"autosegmental".

for

saying

that

the

structure

of

syllables

is

The relation between nodes that represent prosodic

categories (going from the category segment to the category utterance) is fundamentally different from the relation between, for instance, tones and tone bearing units and I see no advantage in claiming that we capture a significant generalization by expressing the relation that holds in the former and the latter case in terms of the same formalism. In the case of prosodic

constituent

structure

the

"association

lines"

express

a

syntagmatic relation holding between units of a certain prosodic level, and a part-of relation with respect to units of the next prosodic level, whereas in the case of tones the association lines express a paradigmatic relation of cooccurrence between phonological features. So, although the notation may be the same, the interpretation of it is drastically different. The formal properties of autosegmental structure (expressed in the Wellformedness Condition) have no bearing on the question what internal structure we must assume for syllables (or any other prosodic category), because autosegmental structure was assumed to characterize a quite

60

different phenomenon. Without feeling an a priori commitment to binary branching trees, I claim then that there is no reason to allow a similar commitment with respect to flat structures. I would now like to discuss a third possible mode of representing syllable structure. The discussion so far has shown that despite the differences that exist between the metrical and the autosegmental theory they have in common that the irrelevance of onset material with respect to syllable weight is an unexplained fact. I will now turn to an approach to this problem that bears a strong resemblance to

(and derives

from)

proposals that are advanced in Hyman (1983) . The theory that I will discuss here differs in some respects from the one Hyman proposes, but I do not think that these differences do damage to the central idea that Hyman wants to express. The essence of Hyman"s idea is that syllables consist of constituents that he

calls weight

units.

These weight

units

correspond

to

the

traditional notion of a mora. I will therefore refer to this third theory as the mora theory. Within the mora theory, heavy syllables consist of two morae whereas light syllables consist of only one mora. In a language that makes the heavy-light distinction, a heavy syllable's first mora consists of the first vowel plus all preceding segments. The second mora consists of all remaining segments: (30)

Heavy syllable a

A

Light syllable a

I c V V |

level In the following section I will show how one could handle the two different kinds of systems in which syllable weight plays a role. Let me point out here what the mora theory has to offer, when we compare it to the other available theories. Firstly, the distinction between heavy and light syllables falls out as a distinction between branching and non-branching syllables. This implies that the relation between stress (i.e. foot structure) assignment and syllable weight can be formulated in strictly local terms. Foot assignment is sensitive to branching of syllables and not to properties of some node that is more deeply embedded.

61

Secondly, the "riddle" of the irrelevant onset

(and post-nuclear

material) is no longer there. There are no irrelevant constituents or parts of the syllable that must be ignored for the purpose of stress assignment. All segments of the syllable belong to some mora. A direct consequence of this is that syllables are no longer anomalous as part of the prosodic hierarchy. All prosodic constituents seem to consist of one or more units that

are

lower

intonational

on

a prosodic

phrases,

hierarchy:

intonational

an

phrases

utterance

consist

of

consists

of

phonological

phrases, phonological phrases consist of phonological words, phonological words consist of feet and feet consist of syllables. But then quite unexpectedly syllables (at least in the metrical theory) consist of onsets and rhymes and to make this more suspect onsets could just as well be absent. They do not participate in the hierarchy. In the mora theory syllables behave like other prosodic constituents. One might even say that morae too have the appropriate structure, since (at least in the view of some) we rewrite them as a sequence of X 1 s. Hence the whole prosodic hierarchy can be characterized by rules of the type A —> B*. One might ask whether the mora theory is "autosegmental" or "metrical", referring to the question what the internal structure of morae looks like in cases where more than two segments form part of it:

m

m

AA

c c v v c c

m

m

AA

c c v v c c

It seems to me that the mora theory is compatible with both views. If the structure in (31b) is adopted one might say that the mora-theory does not exclude the possibility of defining a category 'onset' or 'coda'. At this point let me compare for the sake of clarity

the mora theory and

the metrical theory, assuming here a variant of the latter which allows us to focus on the essential difference. For the metrical theory, I assume a category nucleus, that can be used to define the difference between light and heavy syllables, and furthermore the categories onset and coda, but not the category rhyme. In the former theory I will adopt variant (31b) and also assume the categories onset and coda. Looking at the figures in (32), we can see that the essential difference between the two approaches lies in the fact that in the mora theory we have "lifted up" the distinction that is made at the nucleus level in the metrical theory:

62

(32)

a

a.

a

b. m

Dn

N

AAA

C

C

V

m

Cd

V

c c

C

C

V V

c

c

A possible objection to the mora theory might come from the fact that it is not directly supported by collocational restrictions. The mora theory shares

this

property

with

the

autosegmental

theory

that

CK

defend.

Proponents of both theories must largely deny the relevance of what was called above "the immediate constituent structure principle of phonotactics" or claim, as CK do, that these restrictions do not support a particular structure. I will return to this issue in chapter 3, section 3.4., with reference to the situation in Dutch. The present sketch of views concerning intrasyllabic structure has made clear that some amount of internal structure is absolutely necessary. The crucial

and

compelling

evidence

comes

from

the

interaction

between

stress/tone and syllable weight, whereas the relevance of the evidence coming from phonotactic constraints may be subject to some debate. Since a certain amount of structure must be assumed, I reject CK's view on syllable structure as being syllable

weight

flat. The solution they propose

(simultaneous

structure)

follows

for

from

representing a

desire

to

represent syllable structure in an autosegmental fashion. I have pointed out above that there is no reason to follow CK in this respect. We have now discussed some of the prevalent views on the representation of syllable structure. In chapter 3 I will return to the issue of syllable internal structure, and the various alternatives we have seen in this section.

I will

confront

the

metrical

representation

and

the

mora

representation with data from Dutch and show that both theories provide an adequate basis for structuring Dutch syllables.

2.2.4. Syllabification rules

The problem of syllabification can be stated as follows. Given a string of arbitrary length S consisting of segments of a language L, we have to be able to decide whether S is a possible sequence of syllables of L. As a first approximation to this issue, we will say that S is a possible sequence of syllables if and only if it can be exhaustively parsed into a sequence of

63

wellformed syllables, i.e. a string of syllables that conform to the syllabic template(s) of L. It turns out that this solution does not lead to a unique parsing in all cases. Suppose that L is Dutch. For the following string of segments two parsings are possible, such that the resulting syllables are wellformed: (33)

(a) lamparm lam.parm

(b)

The problem is of course that laraparm "part of a lamp, litt. : lamp arm" is a compound. In Dutch, as in many other languages, a syllable boundary coincides with the break between members of a compound. Hence, only solution (a) is correct in this case. On the other hand, if lampann was not a compound,

then most native

speakers of Dutch would

agree that

(b)

represents the correct syllabification. What this means then is that two types of cases must be distinguished. In one case, the syllable boundary is determined by morphological principles, whereas if no compound(like) structure is involved, other principles determine the syllable break. Let us refer to the 'other principles' as syllabification rules proper. With respect to the characterization of the domain of syllabification rules proper we might argue that this domain

is a particular morphological

category such as the stem or the word, the interpretation of which depends on the morphological theory that we adopt. For Dutch I take the category word to apply to inflected lexical items, whereas the category stem is split into two categories, i.e. the inflectional stem (input for inflectional rules) and the derivational stem (input for word formation rules). One might argue that in Dutch each derivational stem is also an inflectional stem, but this depends on one's analysis of the non-native vocabulary. Since I will not go into this here, I will simply use the term stem and neglect the distinction made above. Uninflected compounds constitute a stem that itself consists of two stems: (34)

stem stem

stem

lamp

arm

Still pursuing the same line of reasoning we might say that the domain of

64

syllabification is a stem that does not itself dominate another stem. The following examples show, however, that there is more to it: (35)

vijand.achtig

(a)

vijan.dachtig

(b)

vijandachtig

In the case of vi jandachtig 'enemylike' syllabification (a) is correct; yet -achtig is not a stem but a suffix. Now consider the following example: (36)

vijand.ig

(a)

^vijan.dig

(b)

vijandig

In the case of vi jandig 'hostile' syllabification (b) is the correct one whereas again -ig is a suffix. What we have to say, apparently, is that not only stems, but also certain suffixes constitute a domain for syllabification rules. There are other suffixes like -achtig. Booij (1977) points out that suffixes like -ling and -loos also constitute a domain for syllabification: (37)

half.ling

not

hal.fling

verf.loos

not

ver.floos

In various places such suffixes have been referred to as "compoundlike". It is only with respect to syllabification, however, that they behave like subparts of compounds. It appears to be the case then that not only stems but also certain suffixes constitute the domain for syllabification rules. In addition all prefixes

(e.g.

ont-,

ver-,

her-

etc.)

constitute

a

barrier

for

syllabification rules proper: (38)

ont.erven

not

on.terven

ver.edelen

not

ve.redelen

her.eiken

not

he.reiken

The question which arises is how this property, shared by all stems, all prefixes and certain suffixes, should be expressed in the grammar. It is quite generally accepted that utterances of a language are organized hierarchically in two dimensions; cf. Fudge (1969), Selkirk

65 (1982c), Nespor and Vogel (1982). In one dimension, the morpho-syntactic dimension, an utterance is associated with a constituent structure tree that reveals how the utterance is built up out of the minimal grammatical units of the language, i.e. stems, affixes, words, etc. In the other dimension, the prosodic dimension, an utterance is associated with a constituent structure which indicates how it is built up in terms of phonological units, i.e. segments, morae, syllables, etc. In this view there is a unit that is built up out of syllables which is usually referred to as the phonological (or prosodic) word. One of the tasks we face in specifying the grammar of a language L is to describe the relation between the morpho-syntactic and the prosodic hierarchy. Our examples above have revealed that the two hierarchical structures are not unrelated. In at least one respect the prosodic organization is determined by the morpho-syntactic organization, i.e. the partitioning of a string into syllables depends on the morphological structure. Assuming now that the syllables of a language are organized into a unit called the prosodic word, we might express the property of morphological units that appear to be the domain for syllabification in terms of the following rule: (39)

X

>

prosodic word

1. where X is of the category stem and X does not dominate another category labelled stem 2. where X is a prefix 3. where X is a suffix belonging to the set |-ling, (i.e. specified as [+W]) The rule given under (39) gives us a partial specification of the function that relates the morpho-syntactic and the prosodic hierarchy. Much more must be said before we have specified how the complete utterance is organized prosodically. I will limit myself here to a single remark about the further organization up to the level of words. Let us return to our example vijandig. -ig is not specified as C+W], hence the function in (39) will give us only a partial mapping: (40)

W vijand ig stem stem

affix

66

We therefore have to adopt the following rule: (41)

[-W] suffixes are integrated into ("cliticized" to) the preceding phonological word

Applying (41) to (40) will give us (42) (42)

W vijand ig stem

affix

stem At this point I have supplied enough information to return to the subject of this section, viz. the issue of the domain of the syllabification rules proper. Above I said that syllabification rules proper specify how a string is parsed into a string of wellformed syllables. We can now say that the domain of these rules is the phonological word. Within this domain syllabification rules proper give us a parsing that is possible on purely phonological grounds, since it has been shown that the ambiguity that was found in examples like lampann involves the delimitation in terms of morphological information of a higher prosodie constituent, called the phonological word. Saying that the domain of syllabification is the phonological word entails that the notion wellformed syllable of Dutch must also be defined with respect to this domain. Apparently this approach is not indisputable, since Trommelen (1983) defines the notion wellformed syllable with respect to

morphologically

unstructured

stems.

There

appear

to be

certain

differences between syllable structure at the level of basic stems and the level of phonological words. I adopt the point of view, however, that these differences constitute a morphological and not a phonological fact. The set of syllables that occur in underived and uncompounded stems is a proper subset of the

set of

syllables

that

occur

in

(complex) words.

By

characterizing the larger set we characterize the smaller set as well. I see no reason to restrict the discussion of a phonological unit like the syllable to a subset occurring only in a particular morphological category.

67

Given that the domain

of

syllabification

rules

proper has

been

identified, I can proceed with formulating these rules . I said that a string of a language L constitutes a phonological word of L if and only if it can be exhaustively parsed into a sequence of wellformed syllables, i.e. no segments may be left unassociated to a syllable node. The following set of syllabification rules count as an algorithm to construct syllables out of an arbitrary string of segments. Although my purpose in this section is to explore the process of syllabification, I have to make a decision with respect to the particular syllable internal structure that I want to arrive at. I will formulate in detail here rules that create syllables consisting of morae

and

then

indicate

shortly how metrical

syllables

can

be

constructed. Furthermore I refer to Steriade (1982) where a theory of syllabification is developed that is compatible with the metrical theory of the syllable. The first rule of the algorithm locates sonority peaks and assigns to them a node m (for mora). A sonority peak is defined as follows: (43)

Sonority peak A segment p is a sonority peak if neither the preceding segment (if present) nor the following segment (if present) is higher in sonority.

To be able to express "peakhood" in formal terms, we could

express the

sonority of segments in terms of a grid structure in which the sonority values are translated into columns of X's. To give an example, let us assume the following sonority scale for a language L: (44)

stops

|

1

fricatives

nasals

liquids

2

3

4

|

|

|

increasing sonority

vowels

| >

5 >

A grid alignment can be said to be wellformed if a segment that is more sonorous than its neighbour(s) also has a higher grid column than its neighbour(s) and if no segment that is equally sonorous as its neighbour(s) has a lower grid column than its neighbour(s). To make this concrete let us suppose that L is Spanish. The word microbio gets the following grid:

68

(45)

a b c d e f g h *

*

* *

* *

*

* *

* *

*

* *

m i k r o b i o The definition for sonority peak can now be stated as follows: (46)

Sonority peak A segment constitutes a sonority peak if neither the preceding segment (if present) nor the following segment (if present) has a higher grid column

In the above example, b, e, g and h satisfy the definition for "peakhood" . Hence, our first rule (47) assigns node m, as in (48): (47)

m

I

X (48)

> m

X m

where X is a sonority peak. m m

m i k r o b i o Harris

(1983) shows that in Spanish closed

syllables and

syllables

containing a vowel sequence (or a vowel-glide combination) count as heavy. Hence the first segment following a peak must be assigned a node m too: (49)

m X

>

X

m /

X

There is of course a rule that must apply before rule (49) . In a word like Panama we do not want the penultimate syllable to be bimoric: (50)

*

m

m m m

p a n a m a

69

It will be obvious that we must first create "maximal onsets". Here the familiar maximal onset principle comes in, formulated as follows: (51)

Maximal onset principle Given two possible parsings, choose the one where onsets are maximalized, such that no informed syllables arise.

Maximal onsets can be created by formulating a separate rule or by reformulating rule (47) that creates the first mora: (52) (X

1

X ) X n n+1

> (X . . .X )x , 1 n n+1

where X

is a sonority peak, and X ...X is a n+1 1 n maximal sequence of segments such that m is a wellformed mora.

It would be just as easy to incorporate the introduction of an onset node in rule (52). In order not to complicate the issue, I have chosen here to ignore the category onset. In rule (52) reference is made to the notion 'wellformed mora*.

It

is of

course

assumed

here

that

the

grammar

contains

a

specification of what counts as a wellformed syllable, which includes a specification of what counts as a wellformed mora. Let us now turn back to the second mora rule. We saw that in Spanish any segment following the sonority peak qualifies as a second mora. This means that in Spanish light syllables are of the type CV and heavy syllables of the type CVC or CVV. In the previous section we have seen, however, that there are also languages where CVC syllables are classified as light. Hence in such languages only vowels qualify as a second mora. A consonant following the sonority peak is analyzed as part of the first and only mora. Intermediate cases arise too. In Lithuanian only sonorants (whether a vowel or a consonant) qualify as a second mora and there may be languages in which only voiced segments count as morae. An important observation is that, if a language tolerates a voiced consonant as a second mora than all segments that are more sonorous count as a mora too (in this case sonorant consonants and vowels) . We could say then that rule (49) is parametrized in terms of a 'sonority threshold'. The post peak segment ('X') must pass a certain sonority threshold to count as a mora. Rule (49) is further parametrized in the sense that it may be absent

70

altogether. This is the case when the language in question does not make a weight difference among syllables in the sense just described. So far we have discussed two rules. The next Spanish word

shows us that

we need a third rule. After application of (51) and (49) we will get (53): (53)

111

k r u e l d a d m m

A

-(52)->

til

A

k r u e l d a d

-(49)->

m m

A I AI

k r u e l d a d f\

The consonant /l/ must still be assigned to the second mora of the first syllable. The rule in question can be formalized as follows: (54) X

n

(X n+1

X ) m

>

X

n

(X . . .X ) n+1 m

where m is a wellformed mora. If we adopt the "autosegmental" view on syllabic organization we might say that this rule is essentially identical to a subpart of the WFC (i.e. the spreading clause). In other words, given the WFC, rightward association need never be formulated explicitly if we assume that it is a general requirement that the WFC may not introduce association lines that violate language-specific wellformedness conditions. In the previous section I have cast doubt on the significance of applying autosegmental principles to syllabic organization, so I will not press this point. Let me summarize what has been said so far. I adopt below the dashed-line notation to indicate the structural change of the rules: (55)

I.

m (X ...x ) n 1

X

, n+1

where X is a sonority peak and XA ...X n+1 . constitutes a maximal sequence such that m is a wellformed mora

71 II.

X

(X

n

where X III.

n+1 n+1

) has a minimal sonority value V 1

rA--.

X

(X

n

n+1

...X ) m

where m constitutes a wellformed mora One may distinguish between languages in which syllables bear stress and languages

in which morae bear stress and say that in a mora-counting

language the syllable can be dispensed with as a prosodic category, at least for the purpose of stress assignment. I will not speculate here on the question whether there are indeed languages that do not have the syllable as a prosodic category. For a discussion of a possible case in point I refer to Hyman

(1983).

Restricting ourselves to the "normal case" where the category syllable is present we must extend rule (51) slightly in the following way: (56)

a I m

.--A

(X . . .X ) 1 n

X n+1

To summarize let us consider the following example: (57)

siempre 0 1 m

A

a I m

A

s i e m p r e a I m m

by (56)

a I m

AI A

s i e m p r e

by (49)

72

ill 111

ANA

s i e m p r e

by (54)

Assuming a general convention we can say that the second m is adjoined to the preceding o-node:

m

m m

ANA

s i e m p r e One final problem deserves our attention. Consider again the previous example : (59)

*

°

f

ni

mm

a

1 m

AI A I

s i e m p r e I

have

assumed

that

glides

in

Spanish

are

represented

as

vowels

underlyingly. Hence a sequence of two vowels is ambiguous. It may appear on the surface as a sequence of two vowels or as a single syllable containing a diphthong. In itself this would not be a real problem. We would simply have to allow that certain strings can be parsed in two ways. The question is how we will derive more than one parsing. A possible solution is to say that the syllabification rules are optional and persistent.. Recall our requirement that a string is only accepted as properly syllabified if the parsing is exhaustive. There may not be any remaining segments left unassociated. This means then that, although optional, the rules will continue to apply until there is no possibility left for them to apply successfully. Given a string VV, we derive

either (60a) or (60b) after applying the rule that creates the

first mora: (60)

a

m I V

V

a

I I m m I i V

V

73

Applying the rule that creates the second mora, (60a) can be changed into (61): (6D

N

m m I I v v This completes the discussion of syllabification, assuming the moratheory. An algorithm to construct metrical syllable structure could employ the concept of a threshold too. A first rule locates sonority peaks and assigns to them a nucleus node. The post-peak segment can then become a sister of a sonority peak if its sonority value has the threshold value, thus creating a branching nucleus. Following the creation of the nucleus, maximal onsets are created and then the category coda is assigned to remaining consonants. Onsets, nuclei and codas are combined into syllables either directly or after combining nuclei and codas into rhymes. 2.3. Conclusion

In this chapter I have discussed various current theories dealing with syllable structure. I have argued in favor of a "three dimensional" approach in which we distinguish linear, hierarchical and autosegmental structure. To characterize the linear structure I have adopted the point of view that a multivalued feature [sonority] should be employed. With respect to the hierarchical proposals

and

argued

structure against

I have discussed

various

the

theory

autosegmental

syllable

a procedure

to assign

syllable

to a row of segments. Both proposals

concerning

syllable

structure. Finally I have discussed structure

alternative of

structure and syllabification will be discussed with reference to Dutch in the next chapter where I offer a detailed discussion of syllable structure and syllabification.

Chapter

3

Dutch Syllable Structure

3.1. Introduction

In this chapter I will offer a detailed analysis of syllable structure in Dutch. In sections 3.3. and 3.4. I will discuss the linear and hierarchical structure respectively. In a recent monograph Trommelen (1983) offers an analysis of the Dutch syllable employing a somewhat different model, which I will discuss in section 3.5.1. One of Troiturtelen1 s claims is that her analysis of the Dutch syllable provides a fruitful basis for an analysis of the allomorphy of the Dutch diminutive suffix. I must show that my approach can do the same. In section 3.5.2 I will therefore present an analysis of the diminutive allomorphy and suggest that it compares favorably with the analysis offered by Trommelen.

3.2. Preliminary remarks

For the analysis of Dutch syllable structure presented here I will describe my own speech. Since, unlike some other speakers of Dutch, I do not have an apical /r/, the uvular variant, usually transcribed as [a], will be used. Furthermore I do not make a distinction between a voiced velar fricative and a voiceless velar fricative. In my speech there is only /x/, the voiceless variant. Adopting for the moment a feature system like the one proposed in SPE, the consonants that we must take into account then are the following:

76

(1)

p t k b d f s x v z m n ^ l r

jwh

consonantal syllabic

+ + + + + + + + + + + + + + + + + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

sonorant continuant nasal

- - - - - - - - - - + + + + + + + + - - - - - + + + + + - - - + + + + + - _ _ _ _ _ _ _ _ _ + + + _ _ _ _ _

coronal anterior voice

- + - - + - + - - + - + - + - - - + + - + + + + - + + + + _ + _ - + _ - - - + + + -- + + + + + + + + + _

The vowel system will be discussed below in section 3.3.3. A further consequence of taking my own speech as the basis for the analysis is that the consonant cluster /sxr/ as a realization of written is not accounted for, since I have [sr] here. Like almost every language that comes into contact with other languages, Dutch has segments and segment combinations of non-native origin. For instance: (2)

Foreign vowels [6:] [ot: ] Co:] [y:] [u:] [i:]

serre f reule roze centrifuge rouge analyse

Foreign consonants [dz] jeep [tl] chips [g] [g]

aoal charme

[2]

¡jury

(Significantly shorter versions of these six vowels occur in the native sound system, which will be discussed in section 3.3.3.) In the analysis offered here such foreign segments will be ignored. With respect to consonant clusters as well I will ignore a number of cases. In an attempt to say more about clusters (or syllable types in general) than just listing the possibilities, it is necessary to leave foreign (or marginal) segments or segment combinations out of consideration or, at least, to treat them as special in some way. In the framework adopted here, these special cases, i.e. foreign clusters as well as clusters that somehow fall outside the core syllable structure because they "disturb" a consistent and elegant account of by far the largest part of Dutch syllable structure, must be accounted for by means of a number of auxiliary templates, that I will not formulate here. A main template accounts for the core of Dutch syllable structure. In addition I

77

use two types of statements to rule out sequences that are allowed by the quite general main template. Firstly, I will use filters to express certain gaps (either in the form of negative conditions or in the form of if-then conditions) and secondly I will use dissimilarity conditions that express whole sets of cooccurrence restrictions in terms of the notion

1

sonority

1

distance . Before proceeding to a detailed description of the Dutch syllable, let me mention some general facts. It will be shown that at the phonological level, word initial and word medial syllables in Dutch may consist of up to five segments, the minimum being two. Monosyllabic words and word final syllables may consist of up to 9 segments, as in the word under (3), which, except for the X -slot, is a real word promptst 'superlative of prompt'. 0 Apparently no existing Dutch words exploit the full range of possibilities. Since words may begin with the cluster /spr/ I assume, however, that (3) represents a possible word of Dutch: (3)

X + X X ,0 ,1,2

X ,3

X ,4

X + X X X„ ,5 ,6,7,8

I will accept the now common idea, suggested in Fudge (1969) for English, that word-final syllables may be followed by a "syllabic suffix", called the appendix in the recent literature; Fudge uses the term termination. Consider in this respect the X -X slots in (3) and below in (4) (the 6 8 superlative of the adjective bespraakt. 'eloquent'): (4)

X + X X |0 |1 ,2 (be) s

X

X 3 / 4

X ,5

+ X X ,6 ,7

X ,8

p

With respect to the X -slot in both examples it may be argued that this /s/ 0 represents a "syllabic prefix". Apart from the "extras", the sequence of the segments within a syllable will be explained in terms of their sonority value. I will now give a tentative scaling of the segments of Dutch, which can serve as a startingpoint. On this scale the classes of segments are given consecutive values for the multivalued feature [sonority]. The linear structure of the Dutch syllable will then be accounted for by assigning a sonority value to each slot:

78

(5)

obstruent

nasal

1

2

|

|

liquid

|

3

glide

vowel

4

5

|

|

One will notice that the approximants (called 'glides' here, i.e. /j/ and /w/) are regarded as a separate class and not as a class of segments that underlies both these approximants and the high vowels. At this point I will not take a definite position on the question as to whether glides should be identified with a subclass of the high vowels. This matter will be discussed below where the question arises naturally.

3.3. Linear structure

3.3.1. X

1

and X

2

Let us first consider syllables that have a one-consonantal "onset". As in chapter two, I use the term onset to indicate the prevocalic part of the syllable, without actually claiming, that there is a subsyllabic node that is labelled "onset". I will return to this matter in section 3.4. In a oneconsonantal onset any consonant (except the velar nasal) may occupy the first slot. We indicate this as follows. Double square brackets around a slot indicate its sonority value. Thus we have for the first syllable position: (6)

[[X ]] < 5 i.e. the sonority value of the first slot is less than 5

Thus to the first slot I assign an upper value, rather than a (set of) value(s) that refers to an interval on the scale. The non-occurrence of the velar nasal will be accounted for in terms of a filter: (7)

* X I'

I am aware of proposals to derive /f¡/ from an underlying cluster, i.e. /ng/ or /nx/. Both clusters do not appear at the surface. It is then explained, as

79

has been claimed, that /rj/ cannot occur in the onset, since the underlying cluster is ruled out in that position by an independently needed filter. I refer to Trommelen (1983), chapter 5, where she argues convincingly that proposals of this type run into serious problems. The reader may have noticed that the provisional scale has given no place to /h/. I would like to propose here that /h/ is the realization of an empty X^ slot. I will assume that /h/ is the default consonant, as schwa is the default vowel. Its sonority value is zero. Turning now to the two-consonantal onsets I will first give some tables that show the possibilities for syllable-initial clusters consisting of two segments. I refer to Trommelen (1983, ch. 3) for a discussion of the foreign or marginal clusters that are not considered. (8)

Clusters with liquids in X^ p

(9)

t

k

b

d

f

s

x

v

z

1

+

-

+

+

-

+

+

+

+

-

r

+

+

+

+

+

+

+

+

+

-

Clusters with nasals in X^ p

t

k

n - m

-

b

+

-

-

d

f

- -

-

s -

-

x

v

+

- -

+

-

z -

-

-

not considered: /pn/, /fn/, /xn/ (10)

Clusters with glides in X^

j

p t k b d f s x v z - - - - - - ( +) - -

w

—

+

+

—

+

—

—

—

—

+

not considered: /pj/,/tj/,/fj/ Bakker

(1971) offers a detailed and useful study of the structure of

monosyllabic words. Another important source has been Cohen et al. (1959) . The clusters that I leave out of consideration either do not occur at all in Bakker's monosyllabic words or occur with an extremely low frequency. I refer to Trommelen (1983) who offers some discussion of these, and some

80

other cases that are not mentioned here. Clusters with /j/ deserve our attention. Word initially clusters with /j/ are rare, except perhaps the combination /sj/. It is possible to argue that what Bakker treats as /sj/ is in fact one segment /s/. However, in one type of complex words, viz. those formed with the diminutive suffix, we find what appear to be clusters of obstruent followed by /j/.

/ N

r a a m p j

/ K 3

p a a l t j a

h a a x j a b o e f j

3

d o o s j a h a a k j

3

Below I will return to these cases. I will now proceed with an analysis of complex onsets. Looking in general at the possibility of /s/ or /z/ to occur before sonorants we note the following complementary

distributions the voiced

/z/ may only occur before /w/ whereas the voiceless /s/ occurs elsewhere:

1

s +

r

+

m

+

n

+

w

-

I express this distribution in terms of an if-then condition: (13)

+cont

+ant

+cor

-cor -nas

[+voice] It is not necessary to refer to specific positions in the syllable, as long as we specify somehow that the domain of this filter is the syllable. Among the obstruent-liquid clusters we find that /tl/ and /dl/ do not occur. The following filter accounts for this gap:

81

(14)

X

X

I1

[+cor]

I2

[+cor]

As formulated, this filter blocks the we 11 formed cluster /si/ as well. We do not have to complicate the filter, however, because this cluster can also be analyzed as arising from a one-consonantal onset /1/ that is preceded by the prefixal /s/. The "price" we pay is that the filter must refer to specific positions, since (15a) is allowed, but (15b) is not: (15)

a.

X -X, |0 |1 s i

b.

*

X -X„ |1 12 s i

Let us investigate which sonority value must be attached to X^. In (16) I list the value assignments that are possible at first sight: (16)

a. < 5 (i.e. all consonants) b. 2-3 (i.e. nasals and liquids) c. 2-4 (i.e. nasals, liquids and glides) d. 3-4 (i.e. liquids and glides) e. 3

(i.e. only glides)

I will now show that the first choice, (16a), is the correct one. In all choices (except the second) we include glides and we must therefore rule ?ut in that case all combinations with / j/, at least if the syllable occurs in word-initial position. * (17)

X

(in word initial position)

I

/j/

A number of clusters with /w/ must be ruled out as well. (18)

a. pw, bw, fw, vw, xw b. sw

(18b) is handled by the condition (13) given above. The other informed clusters are handled by the following filter;

82 *

(19)

X I1

X I2

-cor

-cor

+ant

+ant

(p,b,f,v)

(w)

[+cont] (f,v,x,s) On the other hand, if we exclude the glides (choice 16b) this implies that /tw/, /dw/ and /kw/, as well as medial clusters with /j/, must be accounted for by means of auxiliary templates. Including nasals (choices a,b,c) also leads to the necessity of several filters (that I will not bother to formulate here) . On the other hand if we exclude all nasals both /sn/ and /sm/ can be analyzed in terms of a pref ixal /s/. In this approach only /kn/ requires an auxiliary template. Before deciding the issue let me point out that, apart from the fact that a sonority value must be assigned to X^, we must account for the fact that associating a segment to X^ restricts the possibility of associating a segment to X^. To give an example, if X^ is filled by a liquid /l/, X^ cannot also be filled by another /l/, nor by /r/. As Harris (1983) states in his analysis of Spanish syllable structure, where a comparable situation exists, the latter type of sequence, i.e. /rl/, is ruled out automatically because it violates the sonority sequencing generalization (SSG). Harris then goes on to take this line of explanation one step further. He suggests that sequences consisting of segments that are too much alike in terms of their sonority value, are forbidden as well. This rules out sequences of identical segments (e.g. /ll/), but also those that consist of segments that are adjacent on the sonority scale (e.g. /nl/, /mr/) . The conclusion is that a whole class of clusters is correctly ruled out if we formulate the following dissimilarity condition: (20)

segments associated to X on the sonority scale

To put it formally: (2D

C C x

2

^

- cc^]: > 2

and X

must be non-adjacent

83

I.e. the minimal difference between the sonority value of X^ and X^ is 2. If this condition is accepted as part of the analysis then we do not have to exclude obstruents and nasals from X^ and we can assign a sonority value to X^ that allows obstruents, nasals and liquids. The only combination that is actually allowed to occur is obstruent + liquid. Condition (21) rules out combinations of classes that are identical or adjacent on the scale, i.e. obs.+obs., nas.+nas., liq.+liq., obs.+nas., nas.+liq., and classes like nas.+obs. , liq.+nas. are ruled out by the SSG (although technically filter (21) takes care of these as well, since [[1]]-[[2]]=-l in those cases). (22)

informed Obs I-

Nas - - I -

Liq •I

wellformed This leaves us with the required class of obstruents and liquids. On the assumption then that glides are excluded from X the following values are assigned to the first two slots: (23)

[[X ]] < 5 C[X 2 ]] < 4

It now becomes more attractive not to exclude glides fromX^, because we can then simplify the analysis by saying that X

and X

have the same sonority

value. The consequences of including glides are that we have to adopt filter (19) to exclude certain combinations of obstruents and /w/. Also we need the filter in (17) to rule out /Cj/, word initially. There is one complication, however. On the scale assumed so far nasals and glides are non-adjacent. This means that we wrongly allow combinations of these two classes. In a situation of this type, I have no choice but to modify the proposed sonority scale. There are two ways to modify it. Either we give liquids and glides the same sonority value, i.e. make them one class with respect to the sonority scale (cf. 24), or we assume a scale in (25) and say that the minimal sonority distance between X^ and X^ is 1.5, rather than referring to nonadjacency:

84

(24)

obs 1

(25)

obs | 1

nas

liq/gl vow

2

3

4

nas liq gl I—I—|

vow |

2 2.5 3

4

Making either assumption we can assign the same sonority value to both X^ and X . On the second view, however, it is required that the dissimilarity condition refers to a sonority value rather than to non-adjacency, i.e. the distance must be minimally 1. 5. In chapter 2, section 2 . 2 . 2. I claimed that the analysis of Dutch syllable structure was going to provide support for the position that the value assignment for sonority is absolute rather than relative, because it would be necessary to assume that not all classes are equally spaced. The theoretical consequence of this would be that the sonority scaling cannot be derived from a traditional binary feature system but must itself be a primitive feature of the system. The question is therefore which of the two scales given above is the appropriate one to account for the remaining positions of the syllable. I will argue that it is indeed the second scale, given in (25), that we need. In (26) one finds a summary of the analysis that has been proposed so far: (26)

a.

[[X

]] < 4 J., z

b.

Dissimilarity condition for X i and

c.

x

2

=

1

(°r

1

1

S

W

1

2

Conditions: 1. Constituent 2 does not contain the DTE of an intonational phrase (i.e. the syllable which is exclusively dominated by nodes labelled S) 2. Constituent 2 is not an unstressed syllable (i.e. syllables which are exclusively dominated by nodes labelled W) The crucial role attributed to the grid as part of the phonological representation is to locate environments, defined in terms of the grid, in

14 1 which Iambic Reversal will apply. The LP theory then uses both trees and grids. A proponent of this grid-cum-tree theory is Hayes (1984), who offers a defence of the original LP conception, thus providing an argument against the two alternative theories we are about to discuss. Kiparsky (1979) points out that, strictly speaking, we do not need the grid to find the circumstances under which Iambic Reversal may apply. If we slightly complicate the rule it applies precisely where we want it to apply, without reference to the grid.

Kiparsky"s proposal allows us to eliminate grids from the theory, or at least to reduce their interest, leading to a tree-only theory. A recent defence of the tree-only approach can be found in Giegerich

(1983).

Giegerich completes the argument in favor of a tree-only theory by arguing that we do not need the grid to account for rhythmic phenomena. He argues that these phenomena can be accounted for in terms of other metrical transformations, rules that restructure the underlying prosodie constituent structure. Instead of producing rhythmic patterns by adding beats to the grid and leaving the constituent structure as it is, Giegerich proposes

to

modify

the

tree

structure,

by

a

series

of

metrical

transformations, which perform operations of the following kind (p. 299). The rule W-pairing changes (12a) into either (12b) or (12c):

(12)

W-Pairing / W

S W

. S

W S

\ Vf

s

A s

A w

w

s

seven pretty little girls

142

seven pretty little

girls

In Giegerich's model the set of metrical transformations maps a prosodic tree that is a copy of the syntactic tree plus S/W labelling onto a surface tree that expresses the surface rhythmical pattern of an

expression.

A different route is also possible, h o w e v e r . Prince (1983) raises the question whether trees rather than grids should be abandoned. His position is supported argument,

(1984). Offering the counterpart of

Kiparsky's

both attempt to show that the Rhythm Rule can be

in Selkirk

formulated

entirely in terms of the grid. The rule that

Selkirk proposes roughly is as

follows s (13)

* * * * *

* * > * * *

Prince chooses an even simpler (14)

formulation:

Move *

To produce the prominence pattern, including main stress, an alternating pattern or an initial beat Prince introduces rules that add beats to the grid directly without intervention of the trees. To a certain extent this was also allowed in LP's standard theory. The major difference between the theory offered by Prince and the standard theory offered by LP is that the strongest stress, too, results from adding a beat to the grid, both at the level of words and at the level of phrasal

stress.

Prince's point of departure is that the RPPR gives us a satisfactory interpretation of metrical trees

(p. 24):

Grid (9a) [our 2a, repeated here as 1 5 ] is distinguished by being minimal in the obvious w a y : it has less structure than any other interpretation of the tree [ w w w with

a

natural

principle

of

s]. The RPPR can be supplemented

minimality

to

pick

out

(9a)

as

143

fundamental. Divergences from the 'flattest' interpretation will arise from subtle variations of emphasis consistent with overall s/w- structure (...), as well as from the pressures of eurhythmicity and phrasal demarcation. For example, i n a [ w w w s ] treelike (9a) the first w is often felt to be more strongly stressed than the others. This fact might be recorded as a supplementary principle of prosodic realization, based on constituent

structure and

linear

order,

distinct from the primary interpretation of the stress pattern imposed by the metrical tree.

w

W a

W 0

S

a *

*

*

* *

a

Prince then observes the following (p. 25): Because the grid carries over so little of the information in the tree, there is another, more direct route to the match-up. We can deal just with

terminals

according

to

a

rule

like

this:

"in any

constituent C, the rightmost terminal is strongest". To put it in more formal and general terms (p. 25): End rule. In a constituent C, the leftmost/rightmost terminal in C is associated to a stronger grid position than any other terminal in C. It is undoubtedly true that the End Rule gives us the same information about relative prominence as the tree + RPPR. The question whether one of the hierarchical structures is

superfluous is therefore legitimate. In the

next section I will discuss the alternative theories in more detail, limiting myself to discussing the two theories with respect to word stress systems.

144

4.3. A comparison of grid and tree theory

In this section I will offer a discussion of the types of word stress systems that I know of, analyzed

in both tree theory and grid theory. This

discussion serves two purposes. Firstly it is my intention to show that both theories actually have the same descriptive coverage. Not only, however, do both theories succeed in accounting for the known variety, they also make use of a highly comparable "machinery", to an extent that one is tempted to speak of "notational variants" (ignoring the issue of constituency). The second goal of this discussion is to provide the appropriate background for the proposals that will be put forward in section 4.3.2. concerning the relation between main stress assignment and alternating stress assignment.

4.3.1. Word stress systems Within the metrical theory of stress a particular

stress system is

characterized by specifying a universally defined set of parameters. Two fundamental parameters involve the shape of feet. First the head of a foot may or may not be allowed to branch. If no branching is allowed the feet are said to be bounded; otherwise they are called unbounded. Secondly the terminal nodes may or may not be sensitive to properties of the syllables they dominate. If the non-head terminals are restricted to positions where they dominate a light syllable, feet are said to be quantity sensitive. In addition it may be required that the head terminal does not dominate a light syllable.

I will

call

such

feet

quantity

determined.

The

logical

possibilities for stress systems according to these parameters are then the following:

145

(16)

Types of feet unbounded quantity insensitive bounded unbounded quantity sensitive bounded unbounded quantity determined bounded

Other important parameters involve the directionality of foot assignment, labelling (SW or WS) and the (non)-iterativity of foot assignment. In the following subsections I will systematically discuss the various types of systems.

I will refer to languages that are cited

in the

literature, for illustrative purposes. I must add that the claims made in the literature about the stress systems of the various languages have been accepted without question. My sources have been Garde (1968), Hyman (1977), Greenberg and Kaschube (1978) and Hayes (1981) . The purpose of this section is to point out how the various stress systems that have been mentioned in the literature are dealt with in tree theory (as developed in Hayes 1981) and grid theory (as developed in Prince 1983).

4.3.1.1. Quantity insensitive systems: bounded In metrical theory the alternating pattern that is present in a Dutch word like eldorado is generated in the tree theory by assigning bounded left dominant feet. Feet in turn are gathered in a word tree: (17) w

w

S

s

s w s w

s w s w

eldorado

*

axioma paradigma

*

* *

*

146

The corresponding grid (cf. 17b) is derived from the tree according to the RPPR. To generate alternating patterns of this type directly

Prince

introduces as a primitive the Perfect Grid (PG): (18)

* *

*

* *

*

* *

*

* *

*

*

Prince assumes that the PG can be "associated" with the string of syllables from Left-to-Right (LR) or from Right-to-Left (RL), either starting with a Peak

(pk) or a Trough

(tr) . The grid is organized in a maximal way.

Maximality implies that "lapses", i.e. sequences of two equally strong positions, are avoided. Apart from the "no lapse" requirement the perfect grid is also subject to a "no clash" requirement, although Prince ends up saying that this second requirement is not absolute. The two parameters (direction

and peak/trough

priority, henceforth

referred

to as the

dominance-parameter) allow four types of systems: (19)

a. LR;tr

b. LR;pk

. . .# c. RL;tr

# ...

o

#

d. RL;pk

# ..• a Hayes (1981) reports that all four types are attested. We can generate all the elementary stress systems that are binary and quantity insensitive if we apply the End Rule (ER) to such "perfect" grids. The ER can apply

147 initially (20)

a.

(I) or finally

(F):

3

*

2

*

*

* a

*

*

*

a

a

a

1

*

b. * a

*

*

* *

*

a

a

*

a

*

*

a

* *

a

*

a

a

The two systems just given are generated by the following stress rules: (21)

a.

b.

PG(LR;pk) ER(I)

PG(LR;tr) ER(F)

As a principle, the ER applies to the highest grid level, although we will see below that Prince makes use of the possibility of restricting application of the ER to the second grid

the

level.

The ER promotes a peripheral unit within a domain D to the most prominent unit of D. Prince proposes to label levels in the grid with names

that

correspond to the domain with respect to which the rules that build the grid are applied. So in (20a) level 1 is called the syllable level, because at that level a grid mark is added to each syllable. Level 3 is called the word level because at that level a beat is added to the most prominent unit of the w o r d . Level 2 is called the foot

level, although

in this case.

Prince

claims, the name for this level cannot be derived from a domain that independently defined as a morpho-syntactic or prosodic

is

constituent.

So far we notice little difference between tree and grid theory except for the obvious fact that trees provide more information than grids, viz. the grouping of units into

constituents.

An advantage of grid theory that Prince does not mention is that tree theoretical analyses must often be supplemented by rules that

"destress"

monosyllabic feet. An example of this can be found in Hayes' analysis of stress in W a r a o . In H a r a o stress falls on the penultimate syllable. Every other syllable to the left of the main stress bears a secondary stress. In words

with

monosyllabic

an

uneven

foot:

number

of

syllables

this

leads

to

an

initial

148

(22)

enahoroahakuta i

However, initial syllables of words with an uneven number of syllables are not reported to be stressed. To account for this Hayes formulates the following destressing rule: (23)

W

0

->

F

/

_L

The important point is that in the theory that Hayes proposes the presence of a foot implies stress, even if the foot is degenerate, i.e. monosyllabic. Hence, Hayes

(and others as well) implicitly

assumes a partly non-

relational interpretation of the trees, that is different from the RPPR. In the grid-only approach we do not get an initial beat to begin with: (24)

enahoroahakutai * * * ** * * ** *

*

*

* *

In this respect the standard tree-cum-grid theory is more successful than the tree-only theory. The grid produced in (24) is exactly the one we get by using the RPPR to derive a grid from the tree produced in (22) . Of course in (24) the grid can be produced without intervention of tree structure. The point

Prince

makes

is

that

the

grid-only

theory

is

simpler

in a

straightforward way and should therefore be preferred, especially when we attach

no

importance

to

the

grouping

of

syllables

and

feet

into

constituents. A possible hint that constituency is not always taken seriously comes from the following example. Given the device of extrametricality there is a way to avoid the destressing rule for Warao, as is pointed out in Halle (1982). If we mark the final syllable in Warao as extrametrical and assign right dominant feet instead of left dominant feet no clashing of stresses is produced at the beginning of words with an odd number of syllables:

149

(25)

enahoroahakuta(i)

It is noteworthy that the ease with which Halle changes the metrical structure from that in (22) to that in (25) demonstrates that the particular grouping of syllables into feet seems to be of little importance. In any case, since no arguments are provided by either Hayes or Halle for one particular grouping Prince's proposal gains credibility. To show how the system works in a slightly more complicated case I will give

Prince's

treatment

of Hawaiian

stress. The data

(deliberately

simplified by Prince) are as follows: (26)

Hawaiian stress (Prince 1983, 49-51) Main stress: final if the final syllable has a long vowel, penultimate otherwise Sec. stress: alternating to the left Extra: uneven syllabled words have an initial ternary foot

(27)

(*) ER(F) *

*

a

a

> ER(I)

> EF(F)

*

*

>

* * * * * * * a a a a a a a

* a

* *

a

*

a

*

*

a

*

PG(RL)* a

* *

a

* a

* a

Note that in the input the final syllable is marked as

extrametrical. The

rule of PG can only assign one grid mark because the second one would clash with the initial mark: PG creates no clash, although Prince will ultimately regard this as a parameter (cf. below). Stated in full the stress system of Hawaiian requires the following rules:

150 (28)

a. EM(F;M) b. ER(F;F) c. PG(RLytr) d. ER(I;F) e. ER{F;W)

To familiarize ourselves with P. 's notation let me spell out what is meant here: a. mark as ExtraMetrical the Final Mora b. End Rule applies adding a grid mark word Finally at the Foot level c. Perfect Gridding applies from Right to Left, trough first d. End Rule applies adding a grid mark Initially at the Foot level e. End Rule applies adding a grid mark Finally at the Word level Prince is the first to note that the given grid based account of Hawaiian stress is somewhat cumbersome. He mentions a few

alternatives, but he does

not make a definite choice. Firstly, it will strike the reader as somewhat inelegant that the End Rule must be stated three times. An obvious alternative would be to apply first the ER(I), then PG(RL;tr), without using

extrametricality,

and

finally

ER(F).

An

even

more

promising

alternative is to apply the ER(F) as the first rule, promoting the final syllable to the word level directly. Then we apply the ER(I), adding a foot level beat. Since PG is subject to the "no clash/no lapse" requirement nothing needs to be said now beyond saying that "there is perfect gridding" . If we furthermore assume that the rhythmic wave always starts at the edge where main stress is assigned we have eliminated the two PG parameters (dominance and direction). Prince considers this possibility, but rejects it for reasons that I will mention shortly hereafter. In section 4.3.2. I will argue in favor of this "main stress first" theory, however. When compared to the system of Warao, the novelty shown by the Hawaiian system is the presence of an initial beat. Prince points out that one can find many more examples of this kind. Instead of using foot-level ER( I), he suggests the possibility of adding a rule Move * to the system, applying after all the other rules, whichever of the alternatives we choose.

151

(30)

* *

*

* *

*

* *

*

*

*

* *

= >

a a a a a a a

*

*

*

*

*

*

*

a a a o a a a

Either way of characterizing the initial beat requires an extra statement. Tree theory (with or without grids) fares neither better nor worse on this point. In a tree based account we need a relabelling rule or a defooting rule depending on the way in which we group syllables into feet:

A A A — > w s w s w s 0 o a a a a(a)

II sAw sAw sAw

A A A s w w s w s a a a a a a(a)

A

— >

A

A

s w w s w s w

a a a a a a a

a a a a a a a

Apparently both theories have to say something extra if alternating stress deviates from a strict binary pattern. A third example that illustrates Prince's framework is stress in winnebago. (32)

Winnebago stress (Hayes 1981, 68) Main stress: third syllable from the left Sec. stress: alternating rightward

(33) # (*) * a

PG(LR,-tr)

*

a

a

*

*

*

*

a

a

a *

*

a

*

a

*

JJR( J )

*

a

a

*

a

*

a

*

*

a

a

*

*

a

*

o

*

*

a

*

a

*

*

o

This system, as Prince points out, shows that we cannot do without the dominance parameter. We need both extrametricality and the dominance parameter to account for stress located on the third-from-edge syllable. Given the limited "window" of end rules there is no other way in which we can get a main stress that is three syllables away from the word boundary.

152

4.3.1.2. Quantity insensitive systems: unbounded As Hayes (1981) observes, a language having final or initial stress, without having

secondary

stresses, can be described

in terms of an

elementary right branching or left branching tree. Languages with simple penultimate or "peninitial" stress use the same trees and extrametricality as well: (34)

Final

Initial

W

W

W

S

o

o

a

a

Penultimate

s a

w a

w

a

w a

Postinitial

We must note that there are at least two other ways to describe the systems with penultimate or postinitial stress in arboreal terms. We can make use of a special labelling rule that is sensitive to branching: (35)

Lexical Category Prominence Rule (LCPR) Label the right node strong iff it branches

This labelling rule is proposed by LP in their analysis of the English stress system. Given this rule a branching node will always be labelled S. Only if two sister nodes are both non-branching is an option possible. If the rest of the tree is labelled WS the unmarked case is to label the left sister W and the right sister S. The LCPR expresses the second logical possibility, the marked option according to Hayes 1981, chapter 3. A second alternative involves assigning feet non-iteratively, i.e. we might assign a left dominant bounded foot at the right edge to account for stress on the penultimate syllable and a right dominant foot at the left edge to account for peninitial stress. A word tree combines the peripheral foot and other syllables of the word:

153

(36)

LCPR

w a

w

a

s w a

w

a

a

s w a

a

w Ns

w

w

a

Non-iterative foot

W a

W a

F A? S W a a

F a

a

a

w

a

As for antepenultimate systems (without secondary stress) at least two descriptions

are

possible.

Either

we

use

a non-iterative

foot

cum

extrametricality or we assign no feet at all and combine the descriptive power of extrametricality and the LCPR. The various descriptive possibilities

that are available

for

the

systems just mentioned only involve differences at the level of constituent structure. Grid theory characterizes the first four systems mentioned (final/initial and penultimate/postinitial syllable) with the end rule and extrametricality. For the antepenultimate type of system it would be necessary to say that perfect gridding can apply non-iteratively, just like foot assignment, which leads to the fact that grid theory makes also available two descriptions for the penult/second syllable type of system. The extra power present in the tree theory is found in the presence of the special labelling rule, the LCPR. Prince argues against the LCPR. Within arboreal theory it can be dispensed with, he claims. As we have seen here the

near-peripheral

stresses caused by the LCPR can also be explained in other ways, i.e. by using feet with a labelling that is the opposite of the word tree labelling or by using extrametricality. Stress on the third syllable can be handled by combining these two devices. Apart from the question whether the LCPR is necessary or not we must realize that the grid-only theory is not so restricted that we could not mimick the result of the LCPR. In passing we mentioned the fact that it may be necessary to assume rules that move * 's around. It seems to me that "move *" can do what the LCPR does and even more. Consider the following example.

154

We want to describe near-peripheral stress. After application of the end rule we now apply "move *": (37)

Move *

In this type of analysis we must stipulate that movement is restricted in the sense that the star cannot be moved beyond the near-peripheral syllable. It seems to me that the LCPR represents a more restricted type of mechanism, because its "local" effect derives from the fact that trees are uniformly branching. The possibility of moving stars around opens the door for a wide range of analyses for every stress system we can think of, existing or not. Here then lies a potential piece of machinery that differentiates the two theories. If the LCPR can be shown to be needed (crucially, not as one of the descriptive possibilities) a problem arises for grid theory because it is forced to allow a mechanism of unlimited descriptive power and must be complemented by a theory of "domains" and "landing sites". But if the LCPR is not proved to be necessary the special labelling rule (in fact any kind of labelling rule) can be eliminated from tree theory and both theories are again "notational variants" (ignoring the constituency issue) . In chapter 5 two metrical analyses of Dutch stress are discussed, one of which uses the LCPR.

4.3.1.3. Quantity sensitive systems: bounded In quantity sensitive systems the internal make-up of syllables influences stress assignment. Prince proposes to account for the difference between heavy and light syllables in terms of the grid. The proposal is that a heavy syllable is associated to a grid column of two asterisks. In addition a heavy syllable may or may not be associated to two grid positions. In this way it is possible to create a three way contrast:

155

(38)

Superheavy

Heavy

* *

Light.

* *

*

*

The distinction between heavy and superheavy syllables is the result of a parameter that Prince calls mora sluicing. The assumption is that all heavy syllables are associated to two grid positions and that the second mora may be "parenthesized". As an example of a Q-sensitive binary system Prince

gives a grid-only

analysis of stress in Tiibatulabal: (39)

Tubatulabal stress (Hayes 1981, 56,60; Prince 1983, 63-67) All stressed syllables are equally strong Stress is on: final syllables long vowels certain morphemes alternating rightward

Tubatulabal is claimed to lack a syllable bearing main stress. Consider the following two examples: (40) (a)

(b) * *

*

*

*

*

* *

ti ka pi ga na yin

*

*

*

*

a na nj^

*

*

*

ni ni mut

The extra mark in the first word is lexical, i.e. it is brought in by the morpheme pigana, while that in the second word results from a heavy syllable: the two underlined i's form one syllable. The surface prominence pattern results from applying PG(RL,pk): (41) PG(RL)

V

In full the stress rule is as follows: (42)

QS(t+voc]) PG(RL;pk;FCO)

156

Two new abbreviations occur h e r e . QS means that the system is Quantity Sensitive

(hence heavy

syllables are

represented

as bipositional

with

their first beat strong) and the addition of [+voc] means that morae must be vocalic. PG is specified as Forward C l a s h Override, i.e. PG may produce a clash. To illustrate what this means take the following example. PG is specified as (LR;tr). In (43a) application is blocked because adding a mark to the grid would produce a clash. In (43b) application is not blocked, despite the clash: (43)

a.

N o FCO PG -x->

*

*

***...

***...

a a a b.

a a a

FCO PG *

_ >

*

***...

*

***...

a a a

a a a

Turning to the examples from Tulatulabal we see that in the first example a beat is added to the final syllable producing a clash with the beat on the penultimate

syllable.

antepenultimate

In

the

second

example

a beat

is

added

syllable although the ante-antepenultimate

to

syllable

the is

heavy. In both cases then PG produces a clash. It is not beyond doubt that we need this marked option of FCO. In the case of adding a beat next to heavy syllables there is no clash since the second mora of the heavy syllable intervenes. There would not be mora sluicing in Tiibatulabal. Only the assignment of a PG to the left of the pregiven stress in the first word produces a clash. It is tempting to suggest that PG never produces a clash, and that what we have here is an application of the ER(F) . The fact that this final "main stress" is not more prominent than another stressed

syllable

should

then

be

seen

as

a

matter

of

phonetic

interpretation or by complicating the ER: ER(F;F). One case that Prince uses as crucial evidence in favor of the parameter FCO

is

the

Passamaquoddy following:

stress

system

is described

of by

Passamaquoddy. Stowell

The

(1979). The

stress basic

system

data

are

of the

157

(44)

Passamaquoddy stress (Hayes 1981, 55,71-2) a. penultimate syllables with a full vowel get main stress b. if the penultimate syllable contains a reduced vowel and the antepenult a full vowel the latter gets main stress c. if both the penultimate and antepenultimate syllable have a reduced vowel that one is mainstressed that is separated by an odd number of syllables from the rightmost full vowel or, if there is no full vowel, from the left word boundary.

Full vowels are i, e, a and u and only the schwa counts as reduced. Stowell represents the difference between full and reduced vowels in terms of nucleus structure. Full vowels correspond to a branching nucleus and the schwa to a nonbranching nucleus. The key to Stowell's treatment of this stress system is the fact that feet (WS, Q-sensitive) iterate from the left, whereas main stress is located to the right. Of course the final syllable is marked as extrametrical. Here are some of the data that Stowell accounts for: (45) a.

b. c. d. e. f.

akanutamakan

"story"

k atawehkanennul \ / luhkewinawak

"we use them"

sak amak

"chiefs" "he was big"

elakilakapan

"workers"

"those who must have been chosen "

g-

mekanutasapanik wamasanamanal

h.

atalasakwe

"he's checking traps'

i.

mat ayaqa s a ka nlk e

j-

wamakasewalaklkwehtahal

"he is heard walking in the grass" "he gave him a black eye"

"he gets them"

These expressions are associated to the following grids (46) a.

*

158 b .

* *

* *

*

*

*

0, *

* *

* *

* *

s

a

e

a

e

*

d.

* ( * * )

u

*

*

*

* *

* *

* *

* *

u

e

i

3

* *

( * )

g

*

* *

*

a

a

( * * )

a

* * *

(*) a

f .

( ** i

159

© * *

**

a

a

**

**

a

(**) e

i

* ** **

**

*

a

i

e

*

* *

a

(**) a

As we can see, two layers of rhythmic structure have been added, one filling up the rhythmless space between heavy syllables and the other one grouping "feet" from right to left. The clashing beats have been encircled. The present system has several interesting properties to which I will turn in the discussion offered in section 4.3.2. The main reason for discussing it here was to observe that Perfect Gridding, operating from left-to-right, produces a clash in examples b. and i. Prince (p.70) therefore points out that the rule has to be marked as FCO since in these examples a star is placed immediately before a heavy syllable. Compare this to the Hawaiian system discussed earlier where PG failed to apply if a beat clashed with the "initial beat". This system shows also, as Prince (p. 80) observes, that the trough that meets the SD of PG(LR;tr) is not the weak beat of a heavy syllable, but the beat of a following light syllable. Were this not the case then the star laid down by PG in example i would have been on c rather than d: (47) ab c d ef *

*

**

*

*

**

a

a a i

PG applies to the string cd and not to be. The pattern of secondary stresses in Passamaquoddy shows then two significant properties of PG. In full the grid based account looks like this:

160

(48)

EM(F;Syll) QS([voc]) PG(LR;tr) PG(RL;pk) ER(F;Wd)

Prince discusses other similar systems in some detail (Creek, Cairene Arabic) that also have quantity sensitive bounded feet and shows that his grid-only theory is capable of generating these systems as well. I refer to his article for some discussion of the tree theoretical accounts of these systems. Before proceeding with the other types of stress systems let me draw attention to a third interesting property of the system of Passamaquoddy. The position of the final main stress in this language is determined by counting syllables from left-to-right. Both Creek and Cairene Arabic have this property too. As in the system of Passamaquoddy main stress in Creek falls on the last syllable in the word that is separated from a preceding heavy syllable or word boundary (if there are no heavies) by an even number of syllables. In Cairene Arabic the rule is the same except for the fact that the number of syllables must be odd. The reason for calling attention to this fact is that in by far the majority of stress systems that I know of main stress is located at the edge where PG (or foot assignment) starts. Consider again Warao. In this language PG runs from left to right and main stress is also located at the left edge. In section 4.3.3. I will argue that this is not accidental, but rather something that our theory must explain.

4.3.1.4. Quantity sensitive systems: unbounded In Q-sensitive systems all heavy syllables are already gridded at the foot level. As no additional alternating stress is reported we must assume that such systems simply lack the rule of Perfect Gridding. There appear to be two important

subtypes of quantity

sensitive

unbounded systems ("E1 means edge, '-E' means opposite edge): (49)

a. E/-E Main stress falls on the last heavy or on the first syllable (if there are no heavies) Main stress falls on the first heavy or on the last syllable (id. )

161

b.

E/E Main stress falls on the last, heavy or on the last syllable (id. ) Main stress falls on the first heavy or on the first syllable (id.)

Hayes (1981) reports that for each type several cases have been attested. Let us look somewhat closer at these two types. The characteristic property of type (49a) is that the words that lack heavy syllables are stressed at "the other side" : stress the last heavy of the word or the first syllable of the word. One of the crucial contributions of the arboreal theory has been that such systems and systems where the default case is precisely reversed (type b: E/E) could be handled by simply postulating constraints on the positions that the head of the foot may occupy. To handle the E/E case we assume that the head of the unbounded foot must dominate a heavy syllable. Above we called this type of foot quantitydetermined. To pick out the final heavy as locus for main stress we must furthermore assume that the word tree is rightbranching (case 50a below). When there is no heavy syllable in the word, we cannot place a foot in that word; there will only be a word tree (case 50b): (50)

a.

W

b.

W

a a a a a a a

ai ia iq *a q q a

"g" indicates a heavy syllable, "o" a light one. To handle the E/-E case we do not require that the head of the foot dominates a heavy syllable, i.e. the head terminal is free. Hence if a word has no heavies the whole word will be dominated by one foot: (51)

a

W

b

w F

ai ai ai a a * ai ai i

a a a a a a a

Prince says that it will come as a "mild shock" that grid theory cannot make

162

"simple" statements about these two cases. Let us go through the two systems. E/E is relatively simples (52)

QS ER(F;F) ER(F;wd)

This rule can be simplified if we assume that the ER will always apply to the highest grid level, unless we explicitly demand otherwise. This reduces (47b) to: (53)

QS ER(F)

The 'mild shock' comes when we consider the E/-E case: (54)

QS ER(I;F) ER(F;wd)

To derive the default (initial) stress. Prince assumes that the first syllable will always get a strong beat. The analysis is indeed suspect. One could argue that the order of the two statements is incorrect, i.e. we should not assign the first syllable a foot level beat in each case, but only if heavies are absent, i.e. the initial stress must be assigned by a "default rule" that only comes into action if there is no heavy syllable in the word. In a standard SPE type of description the E/-E type required the use of essential variables (the Q-variable): (55)

a -> 5 /

- Q#

where Q is a maximal sequence of syllables not containing a heavy syllable In the other type of case (E/E) a different type of variable must be used:

163

(56)

a -> o /

_x#

where X does not contain a heavy syllable Instead of giving a somewhat suspect treatment of the last/first type Prince could have used the Q-variable in the formulation of his end rule, but he did not. Halle (1982), granting the claim that the various word stress systems can be characterized by assigning grids to words directly, without the intermediate step of creating a metrical tree, also observes that for some systems this will entail that rules must make use of essential variables. Halle then says (p. 2): Thus, we now face the question as to whether anything has been gained by eliminating metrical trees if this elimination forces us to introduce variables, which are devices of great descriptive power.

In

fact,

it

would

appear

that

metrical

trees

are

considerably more restricted descriptive devices than variables and should, on those grounds alone, be preferred to variables. By

avoiding

the use of essential variables Prince bypasses Halle's

argument, but despite this it is clear that in accounting for this type of systems grid theory does not count as a clear improvement. The last/first type necessitates adding an extra parameter in tree theory

(quantity

determined feet), but it creates some "pressure" to adjust the grid theory as well. Prince is forced to introduce a category of end rules that introduce foot level beats, which are only taken into account if there are no other foot level beats (due to heavy syllables). In the next section I will discuss the last group of stress systems: limited stress systems. In such systems stress falls on the peripheral (final or first) or near-peripheral syllable (penult or second). This type is interesting because it leads us to propose that grid theory must make use of a domain parameter, which brings the two competing theories

again one

step closer.

4.3.1.5. Limited stress systems and syllable weight. In a number of stress systems stress vacillates between the peripheral and near-peripheral syllable, depending on the weight of either both or one of the syllables involved. In this section I will show how such systems are or

164

can be handled in both theories. In the previous section we have seen two types of bounded Q-sensitive feet. In one type the dominant terminal could dominate a light or a heavy syllable (i.e. the terminal node is free), in the other type the dominant node had to dominate a heavy syllable. One will recall that these two possibilities implied different default, options for words containing no heavy syllables. It appears that both unbounded systems have a "local" counterpart. I will start with the first type, i.e. the local counterpart of the type of stress system where the default option leads to stress at the opposite end of the word. (57)

Rotuman stress Stress is final if the final syllable is heavy, prefinal otherwise Latin stress Stress is penultimate if the penultimate is heavy, antepenultimate otherwise Ossetic stress (Hayes 1981, 78-79) Stress is initial if the first syllable is heavy, on the second syllable otherwise

Rotuman and Latin are of the type "last heavy or first", whereas Ossetic is of the type

"first heavy or last". To make this explicit we could

reformulate the two cases as follows: (58)

Latin Stress

falls

on

the

last heavy

or the

first

syllable,

last

syllable,

considering only the final two syllables Ossetic Stress

falls

on

the

first heavy

or

the

considering only the first two syllables The crucial difference as compared with for example Eastern Chererais is that the choice for "last" or "first" is limited to the two final syllables (in Latin the real final syllable is extrametrical) . In tree theory such systems are accounted for by assigning bounded feet non-iteratively. Consider Rotuman as an example:

165

(59)

taka

(60)

"lie down"

hununuka

"grasp for breath"

maroo

"to be taut"

karakaa

"snore"

a. Assign a left dominant Q-sensitive foot at the right edge

(non-iterative)

b. Word tree is right dominant

(labelled

WS)

(61)

karakaa In the grid-only theory I can think of the following way to handle cases like Rotuman or Latin. Either we stipulate that only the final two syllables may be considered and assume a rule of the form ER(I;F) or we apply PG noniteratively, peak

first. In both alternatives the ER(F) completes the

analysis.

there

Clearly

is

little

difference

in

iterative PG and an ER applying to a domain of two

effect

between

non-

syllables.

The point that is really relevant here is that the equivalence between the unbounded systems discussed in the previous section and the bounded systems discussed here can only be brought to the surface in the grid only theory by adding a domain parameter to the theory. This parameter has two values: the two peripheral syllables or the whole word. It needs little emphasizing that such a parameter is highly comparable to the boundedunbounded parameter in the tree

theory.

On the assumption then that only the final two syllables are taken into account the stress system of Rotuman can be derived as follows (assuming a monopositional representation of heavy (62)

a.

*

...**# 0 0

ER(F;wd) ER(F;F)

b.

*

...**# 0 0

c.

syllables): * *

...**# O O

d.

...**# 0 0

*

*

*

*

*

*

* *

*

* *

* *

* *

*

0 0

00

0 0

*

0 0

Let us now consider the local counterpart of the E/E type of system. In this

166

case the weight of both peripheral syllables matters: (63)

Yapese stress (Hayes 1981, 65) Stress is prefinal if the final syllable is light and the penultimate heavy, final otherwise Malayalam stress (Hayes 1981, 66) Stress is post-initial if the first syllable is light and the second heavy, initial otherwise

In tree theory the device employed to handle these cases is similar to that employed for the unbounded "last, last" and "first, first" cases, i.e. a foot whose head must dominate a heavy syllable. This can be made clearer if we again reformulate the above statements: (64)

Yapese stress Stress the final heavy syllable or the final syllable (if there is no heavy), considering only the final two syllables. Malayalam stress Stress the first heavy syllable or the first syllable (if there is no heavy), considering only the first two syllables.

Comparing this with (42b) we realize that Yapese is the bounded counterpart of Acuacatec Mayan, whereas Malayalam is the bounded counterpart of Khalkha Mongolian. Let us consider the tree treatment of Yapese: at the right edge of the word we assign a leftdominant Q-determined foot (i.e. a foot whose head must dominate a heavy syllable) . Feet are gathered in a right dominant word tree: 165)

A

s

w

a. saalap

J

A

|

b. magpaa

w s

c.

pa?ag

Since the dominant node must branch no foot can be assigned to the third word. The word tree produces final stress. In Malayalam we get the reverse: we assign a left dominant Q-sensitive foot at the left edge. The word tree is left dominant.

167

Of the limited stress systems that we discuss in this section Prince only mentions Malayalam, for which he suggests the use of two end rules, i.e. ER(I?F) and ER(I;Wd) . Had Prince considered the Latin/Rotuman type of case, he would have realized that using two end rules necessitates a domain parameter, limiting the domain of application to two peripheral syllables. In Malayalam the four possible cases at the beginning of the word come out as follows: (66)

a.

*

b.

#**...

*

c.

#**...

o o

a

* *

d.

#**...

a

o

#**...

o

a

a

ER(I;wd)

*

*

*

*

ER(I;F)

*

*

* *

* *

* *

* *

*

0 0

00

00

*

*

00

The ER(I;F) cannot assign the star to the first case where the word starts with a light-heavy syllable sequence and therefore ER(I;wd) will stress the second syllable. The same type of approach can also be used in cases where "retraction" occurs at the right edge. In that case we arrive at the following analysis for Yapese: (67) i.

ER(F;F) ER(F;wd)

ii. a.

*

b.

...**# 00

ER(F;wd) ER(F;F)

*

c.

...**# OO

*

*

*

*

* *

...**# 00

d.

...**# 0 0

*

*

* *

* *

*

*

0 0

0 0

*

*

*

* *

0 0

0 0

The following table makes explicit the similarities between the local and non-local QS systems:

non-loc. local

last/first

last/last

ER(I;F)

ER(F;F)

ER(F;wd)

ER (F;wd)

ER(I;F)

ER(F;F)

E R(F;wd)

ER(F;wd)

168

Let us now consider a third type of peripheral stress system: (69)

Type three Stress is near-peripheral if the near-peripheral syllable is heavy, peripheral otherwise

We find this type both in the word-initial and the word-final variety: (70)

Creek stress, Aklan stress Stress

is

prefinal

if

the

prefinal

syllable

is

heavy,

otherwise stress is final Capanahua stress Stress is on the second syllable if it is heavy, otherwise stress is initial In Aklan and Creek we only get a penultimate stress if the penult is heavy (closed). The Aklan stress system is described by Hayes (1981). To account for this type Hayes employs the LCPR. Given this labelling rule the Aklan system can be generated as follows: (71)

a. assign right dominant Q-sensitive binary feet, RL b. assign a right dominant word tree, LCPR

Consider the following examples, taken from Hayes (1981, chapter 2): (72)

a.

,

A

WS pitu

b.

c.

A

^

"A A" A'

F

F

W A S

WS bisahi

F w

F .s

W S W £ kinaputus

d

-

e A F F s w

F

^

A

s

F w

/ \ W S gasta

asirtar

In examples d and e it is not possible to combine the two peripheral syllables into one foot, because the penultimate syllable is heavy (i.e. closed) ; it cannot be the weak daughter of a foot. Another possible analysis would be to say that the two peripheral syllables are always combined into one foot that is labelled by the LCPR (left node is strong iff it branches) . In fact this is the type of analysis that Hayes proposes for the mirror image case in Capanahua. A grid theoretic account of this type of system must be as follows. After

169

application of an ER(F) we apply a rule of the form "move *" (cf. Halle 1982): (73)

b.

d. a a

a a

a a

ER(F,wd) ER(F,F) *

*

*

a a

*

a a

*

*

a a

Move *

In section 4.3.1.2. we noted that tree theory can resort to a special labelling rule, the LCPR, and that the counterpart of this LCPR in grid theory would be "move *". It now appears to be the case that adding "move *" to grid theory was not just a remote possibility. To describe this third type of limited stress system such a movement rule is required. So far we have seen three types of limited Q-sensitive systems in this section. Before ending this survey of stress systems I want to raise the question whether a fourth type occurs as well, and if so, what consequences that will have for our two competing theories: (74)

Latin

Yápese

Aklan

b.

a a i i

Case d is the (so far) unattested type. Notice what the three attested types have in common. If there is a difference in syllable weight the heavier syllable always bears main stress (case a and b). However, when the syllables are of equal weight we find three out of the four logically possible systems (case c and d) . Grid theory has no problems in generating

170

the fourth type cof system: (75)

ER(I,F) ER(F,Wd) move *

(76)

a.

...o* a *#

ER(F,wd) ER(I, F)

*

c.

*

.0 *0 *#

* *

...**# O O

*

* * *

*

a a

0 0

*

*

d.

...**# 0 0

*

*

*

*

*

*

*

OC

*

*

0 0

*

move *

*

*

*

*

a a

It is interesting that tree theory offers no means to characterize this type of system. If we accept the second analysis suggested for Aklan, this theory offers only three possibilities: (77)

A W# Aw# A

i.e. Quantity-sensitive foot i.e. Quantity-determined foot. i.e. Foot labelled by the LCPR (i.e. left node strong iff it branches)

These three possibilities correspond exactly with the attested cases. The question of the "missing fourth type" will be taken up in chapter 5 where I analyze the stress system of Dutch. I will propose two analyses which both suggest that Dutch represents the fourth type. The type of foot that is required for Dutch deviates from the three types in (77) in that the weak daughter(s) are allowed to contain both light and heavy syllables. 4.3.2 A conclusion and a proposal In the previous sections we have seen that both the grid and the tree theory are capable of accounting for the variety of stress systems and that both

171

face the problem that some systems can be generated in more than one way. We have also seen that Prince (p. 20) may have been somewhat too optimistic in saying that a theory has emerged that is "significantly simpler" than the arboreal theory. It has been made clear here that applying grid theory to the full range of attested cases forces one to add more and more machinery until one ends up in a situation in which there is an almost perfect one-toone correspondence between the grid and tree theory: (78)

Tree theory

Grid theory

Foot assignment -direction:LR/RL

Perfect Gridding -direction:LR/RL

-dominance:SW/WS -Quantity SensitivesYes/No

-dominance:pk/tr -QS:Y/N

-Quantity determined:Y/N -shape:bounded/unbounded

-Foot level end rule -domainsbounded/unbounded

Word Tree -branching:left/right -labelling:uniform/LCPR

End Rule -Edge:l/F -Move *

Extrametricality

Extrametricality

We have seen that in a few cases the theories make slightly different predictions, but this is only the case if we disregard the fact that both theories can be slightly adjusted here or there to eliminate or add empirical claims. I think that the conclusion is justified that the single discriminating property is the absence or presence of a detailed word internal prosodic structure: (79)

Tree theory

Grid theory

I have assumed here that, although Prince makes no use of foot assignment or uniformly branching trees, it is not his intention to deny the existence of the prosodic hierarchy as distinct from the morpho-syntactic hierarchy. Now the two structures embody two rather extreme views and we must not exclude the possibility that an intermediate position may turn out to be preferable to either of the two positions considered so far.

172

Rischel (1982, 201) presents such an intermediate position: [...], there seems to be more of a break between the pretonic part and the remainder than between the posttonic part and the syllables preceding it; this appears in that it is possible to hesitate between the (last) pretonic syllable and the stress-syllable rather than elsewhere, and that there may be an extremely sharp intonational break

here.

I

therefore

venture

to

suggest

a

hierarchical

arrangement as follows, [...]

"

./tx a

la

bas

ter

vi o lin ist ind e The examples are from Danish. Leben (1982) offers suggestions that are similar in certain respects. He argues that there are no compelling reasons for assuming that prosodic constituent structure is binary branching. His proposal is to replace structures as in (81a) with structures as in (81b):

s w w w

s w w w

If Leben 1 s proposal is adopted the word violinistinde could be represented as follows: (82)

w w

w

w

s

w

vi o lin ist ind e

173

In Halle and Clements (1983), finally, the same approach is found. They assume no labelling whatsoever. Instead the leftmost or rightmost daughter is dominated by a branch that ends in a dot, thus indicating that a particular syllable is the head of a constituent. Clearly, there is no significant difference between using + and -, S and W or any other conceivable way to indicate that within a particular constituent one daughter is the head, i.e. the strongest element. It is furthermore irrelevant

to ask whether we are going

to call multiple

branching

structures "metrical" or "autosegmental". The important point is that there is some plausibility in Rischel's proposal to differentiate main stress, expressed in terms of a labelled constituent structure, and other, more detailed aspects of the prominence pattern of words. In chapter 1 I pointed out that standard metrical theory is based on two fundamental

ideas. The first idea is that prominence

relations

are

expressed in terms of a SW labelling imposed on a constituent structure, and the second idea is that the constituent structure is binary branching. The two ideas are logically independent. We can say then that in the proposals advanced by Rischel and Leben only one of the fundamental ideas behind standard metrical theory is dropped. Rischel goes one step further than Leben, however, and also questions the use of constituent structure (i.e. feet) to characterize non-primary stresses (p. 201): If some syllables among the unstressed ones are felt to be more prominent than others, this is probably ascribable to two factors, viz. (1) that each syllable has an inherent degree of prominence, which is a function of its phonological make-up (closed syllables having more prominence than open syllables, and syllables with a full vowel more prominence than syllables with schwa), and (2) that the pitch contour associated with a full stress [ . . s u p p l i e s each syllable with a tone level (F level) which contributes to the impression of more prominence o or less prominence. I have not considered it useful to build such considerations into the assignment of hierarchical structure to a clustering of zero-syllables around a full-stress syllable (and my model therefore comes to look somewhat different from those proposed for English in recent work such as Liberman and Prince 1977 and Selkirk 1980). Rischel does not mention the natural tendency to assign a rhythmic structure as a possible (third) factor. It seems to me that rhythmic structure is an important factor, and both impressionistic and phonetic

174

evidence (involving Swedish data) will be discussed in section 4.3.4.2. We can still agree with Rischel, however, and argue that this rhythmic structure too does not have to be expressed in the constituent structure. The interesting aspect of structures like (80) and (82) is that a fruitful partnership between arboreal theory and grid theory is emerging here. The implication of assuming structures as in (80) and

(82) is that

main stress is characterized in terms of a labelled constituent structure and

that

as

a consequence

secondary

accents must be

characterized

differently. I propose the following theory of stress assignment. I will assume that main stress arises by assigning a foot at one of the edges of a word. Since the sole goal is to assign main stress, foot assignment need never be an iterative rule. To distinguish the foot assigned by a non-iterative stress rule from feet in standard metrical theory, I will refer to the former as a stress foot. As a general convention we may adopt Rischel"s proposal that syllables outside the stress foot are combined as equal sisters of the stress foot into a constituent labelled (prosodic) word. Part of this general

convention would be that the

labelling of the word tree is always such that the stress foot is S (or +) . This is no different from a convention proposed by Liberman and Prince (1977) and Hayes (1981) in connection to the Stray Syllable Adjunction Rule. In Hayes' proposal too, syllables that are adjoined to existing structure by means of this rule are always weak. The resulting arboreal theory is considerably simpler and more constrained than standard metrical theory, yet it makes use of some of the important parameters that have been proved to be useful. In particular it takes over the array of foot types that were discussed at the end of the previous section (plus their mirror image cases), with one difference: feet that contain more than two syllable need not be binary branching. As in the standard LP theory a grid is projected from the tree, although the projection rule is even more trivial than the RPPR: (83)

Assign an asterisk to the strongest daughter of the stress foot

Leaving out the first grid level as somewhat redundant the following Dutch words have been assigned prosodic structure according to our rudimentary) tree-cum-grid theory:

(still

175 (83)

w s w w

W S WW Utopia

WWW

w

parallellogram

rmoceros

It h a s b e e n o b s e r v e d m a n y times that in l a n g u a g e s h a v i n g final stress, the i n i t i a l s y l l a b l e o f a word, if s e p a r a t e d by at least o n e s y l l a b l e from the m a i n s t r e s s e d s y l l a b l e , b e a r s a strong s e c o n d a r y s t r e s s . T h i s is a l s o the c a s e in D u t c h . P r o c e e d i n g o n the a s s u m p t i o n that such a strong

secondary

s t r e s s is not p r e s e n t in all l a n g u a g e s that h a v e final stress, let us see h o w we might differentiate between

l a n g u a g e s w i t h and

languages without

i n i t i a l b e a t . W e m i g h t say that in l a n g u a g e s of the former type

an

syllables

o u t s i d e the stress foot are g a t h e r e d in a second foot that is a sister of t h e (main s t r e s s ) foot. I w i l l use the term a n t i p o l e s t r e s s foot for t h i s second foot, w h i c h , I e m p h a s i z e , d o e s not r e s u l t from i t e r a t i v e a p p l i c a t i o n of the s t r e s s rule, b u t by m e a n s o f a s e p a r a t e c o n v e n t i o n .

Since Dutch has

an

i n i t i a l b e a t of the type r e f e r r e d to t h e s t r u c t u r e o f p a r a l l e l l o g r a m could be represented

as in

(84):

(84)

^

^

w s w w

s w

parallellogram * *

A l i m i t e d t r e e s t r u c t u r e of this type h a s the a d v a n t a g e that it e x p r e s s e s adequately

the

fact

that

longer w o r d s h a v e

two

suitable

locations

for

a n c h o r i n g p i t c h a c c e n t s . T h i s is the case b o t h in D u t c h and in E n g l i s h . I d o not w a n t to p u r s u e this t o p i c in g r e a t d e t a i l , b u t it is i n t e r e s t i n g

to

d e v o t e some a t t e n t i o n to it in o r d e r to m o t i v a t e a c o n s t i t u e n t s t r u c t u r e as in

(84). I q u o t e B o l i n g e r

[. . . ]

it

is

not

(1981,26) h e r e :

necessary

-

or

even

accurate

-

to

insist

that

" p r i m a r y " is m o r e p r o m i n e n t than " s e c o n d a r y " [...]. T o show t h i s w e need o n l y d e m o n s t r a t e that the " p r o n u n c i a t i o n " of a w o r d c o n s t i t u t e s an i n t o n a t i o n a l p a t t e r n ,

and that p a t t e r n

is its c i t a t i o n

form.

176

With "secondary" Bolinger refers to the strongest non-primary stress, located, in English, at the beginning of the word. Bolinger points out that in English answering a question leads to an intonational pattern that he expresses as follows (p. 26): Q: Where did you go? Bos went A: I

to ton

He continues: Similarly a citation form is an answer to a question: What: word did you use? or How do you say this word?, and the same question-answering contour is used, with the high second peak: poe

ga

On

a

A1 omato

ten le

ia

At tion

u tion

In Bolinger's view the syllable carrying main stress is simply that syllable which is associated to the highest pitch peak present in the intonational pattern that is characteristic of the citation form. If prominence is defined in intonational terms, we have here the simplest explanation of why the full syllable at the cutoff point has been called "more prominent". The analyst reads into the lexical representation of word prosody the intonation of the citation form that is ringing in his ear. (p.28) There are cases, however, where the syllable with the so-called secondary accent is the most prominent syllable of the word. Bolinger mentions examples where isolated words receive the highest pitch peak on the first syllable and also mentions the familiar cases of "stress shift" found in phrases of the type thirteen men or academic discipline. If we take the two syllables that have been associated to a grid mark as equal then it is more appropriate to speak of selection rather than shift:

177

In things academic versus academic things the difference is not so much a shift as a SELECTION. Counting leftward

beginning with the

cutoff point there are two long syllables, two positions that can bear an accent, and the one is chosen that is more convenient for the pitch turn. If Bolinger is right and a correct analysis of so-called shifts involves selection then it may be better to represent the structure in (84) without the higher level labelling:

*

*

A possible objection to a structure like in (84) or (85) might be that it does not do justice to the "rhythmic structure" of the word parallellogram that is claimed to have a strong beat

on its third syllable. I would like to

argue that the presence of a beat on the third syllable does not force us to abandon our restricted arboreal theory. I propose that the rhythmic structure for which tree theory has introduced iterative foot assignment and grid theory the perfect grid

is represented separately from those

aspects of the rhythmic structure that in the present proposal have been characterized in terms of a labelled constituent structure. I suggest that the alternation of strong and weak

syllables

is characterized by a

"rhythmic melody" that is placed on a separate tier (Prince suggests the term rhythmic melody) . An argument in favor of this proposal is the fact that rhythmic beats are not suitable locations for placing a pitch accent. The following example illustrates this point. In the phrase Apalachicola falls it is claimed (e.g. in Hayes 1984) that the main stress can shift back due to a stress clash (employing current terminology) . It cannot shift back to the third syllable, however. The only possible landing site is the first syllable. The proposal to separate two aspects of the rhythmic structure conforms to the view expressed by Bolinger, who argues in favor of separating what he calls accentual rhythm and syllabic rhythm. Accentual rhythm involves the tendency to avoid situations in which syllables associated to a pitch accent are too close to each other. Syllabic rhythm involves an alternation

178 of strong and weak syllables. This syllabic rhythm may be due to intrinsic properties of the syllables (such as the quality of the vowel or another property that determines syllable weight) or to an extrinsic

rhythmic

pattern (i.e. a perfect grid) or both. In this section I have advanced two proposals. The first proposal is to separate two components that together make up the prominence pattern of utterances and the second proposal is to deal with the one in terms of a labelled constituent structure and the other in terms of a rhythmic melody. The two proposals are logically independent, i. e. it would be possible to separate the two components and maintain at the same time that both aspects of the prominence pattern are jointly characterized by either a grid or a tree structure. In the next two sections I will give additional support to the view that main stress assignment (and possibly assignment of the strongest secondary stress) must be separated from the assignment of alternating stress. Then I will return to the issue of constituent structure.

4.3.3. The main stress first, theory In this section I want to reconsider certain aspects of the analyses that have been offered in section 4.3.1. I will propose that a certain redundancy present in the analyses leads to the conclusion that we must alter our view of the relation between main stress and alternating stress surrounding it. An appropriate starting point will be to consider in detail the view on characterizing rhythmic structure that was discussed at the end of the preceding section. Instead of building the grid going through the word from left to right or from right to left I will regard the perfect grid as a pregiven autosegmental melody, i.e. a rhythmic melody. Stress systems of the type we discussed in section 4.3.1.1. (bounded, Q-insensitive) could then be generated as follows: (86)

Warao Initial Stress Association Rule (ISAR): Associate the last Stress Bearing Unit with a trough

After application of this ISAR the Wellformedness Condition (WFC) applies thus

completing the derivation. The reader will notice that the present

alternative

comes

very

close

to the way

in which tone patterns

are

associated to words in tone languages (cf. chapter 1). Indeed, I think that

179 what we have here is more than suggestive terminology. (87)

enahoroahakutai ISAR

Ii *

* * *

*

* *

*

*

*

*

* *

enahoroahakutai WFC

w w w ww w ws w enahoroahakutai *

* *

*

** *

* *

*

** *

Main stress has been accounted for by assigning a binary SW foot at the right edge and a multiple branching word tree labelled W*S. The description just given contains a redundancy, however. The fact that the ISAR and the Main Stress Rule refer to the same edge suggests a simpler analysis. Suppose the first move is to apply the main stress rule, assigning (via the tree) a star to the penultimate syllable. The rhythmic melody is then associated by another type of ISAR that requires that the star is linked first, to a peak on the melodic tier. Again the WFC completes the association between melody and text: (88)

enahoroahakutai *

ISAR

\ *

* *

*

**

*

*

*

*

** *

180

enahoroahakutai WFC *

*

**

*

*

**

It seems to me that the relative ordering of first assigning main stress and then the rhythmic melody is attractive, not only because it eliminates the redundancy, but also because it reveals quite straightforwardly to what extent pitch-accent systems and stress-accent systems are the same. The difference between the two arises because in the former type of system a tonal melody is anchored to the star (as, for example, in Japanese), whereas in the latter type a rhythmic melody is anchored to the star. Let us now look at "standard" tree and grid theory. In both grid and tree theory main stress arises by assigning a

peripheral foot head status in

the word tree or by adding an extra beat to the first assigned perfect grid. In other words, we first assign "alternating stresses" and then promote one of them to "primary stress". The above discussion suggests another way of looking at alternating stresses, viz. making them dependent on the main stress. Garde (1968, 53) speaks about alternating stresses (of the kind we are considering here) as "echo de 1'accent". Here I want to take this metaphor seriously and propose that much can be gained if we say that alternating stress results simply as a "by-product" of main stress. To speak in grid terminology: Perfect Griddin^ is determined by the End Rule. In section 4.3.1.1. I mentioned this conception of the relation between main stress and alternating stress, referring to it as the main-stressfirst-theory. In this section I will explore the possibilities and problems that cross our path if we adopt it. Let us observe that, with a few exceptions (Passamaquoddy, Creek, Cairene Arabic, Garawa), systems with alternating stress (with PG) locate the main stress at the edge where iteration of PG starts. So if PG goes from left to right the ER applies word initially and vice versa. If we take this to be the unmarked case, it should be reflected in the formulation of our rules. To take the simplest example first, let us look at the stress system of Maranungku. Here main stress is located on the first syllable and counting from that syllable rightward a secondary stress falls on every other syllable. In standard grid theory (as in standard tree theory) we first assign alternating stress and on the basis of this we determine the main stress. Suppose that we reverse the order and first assign main stress to

181

the first syllable and, based on this, then determine alternating stresses. Under this proposal alternating stresses can properly be called echos of the main stress. Above we saw that this proposal easily translates into an autosegmental grid theory. Of course the order in which main stress and alternating stress is assigned can also be reversed if we adopt Prince 1 s view in which the grid is built up, bottom-to-top. Within this approach Maranungku would have the following analysis: (89)

a. End Rule (l;wd) b. Perfect Gridding

If we assign main stress first we must assume that the ER (which is our main stress rule) as formulated will promote the final stress-bearing unit directly to the word level: (90) * * *

a

*

*

a

*

a

*

a

*

a

*

a

a

Another way to obtain the same result is to say that the end rule is formulated as ER(I) and simply adds one beat and that this beat survives perfect gridding by a general convention: (91)

ER(F) =>

* PG *

=>

In the approach Prince takes PG must be specified for at least two parameters (direction, dominance). Selkirk (1984) follows Prince in this respect. Adopting the main stress first theory proposed here it is obvious that a PG rule, whose parameters have not been specified, cannot apply before the ER has applied, just as the echo can only come after and never

182

before your coin hits the bottom of the well. I suggest here that (unless specified otherwise) PG will always move away from the main stress and, since no clash will be produced, that it will lead to a trough on the syllable next to the main stressed syllable. If sufficient syllables both precede and follow the main stress PG may go in two directions, as seems to be the case in Cahuilla, referred to by Prince as being problematical for a standard grid theory. The following rules generate the stress systems with peripheral stress discussed in sect. 4.3.1.1.-2.: a.

ER(F) + PG

b.

ER (I) + PG ER(F) + EM + PG ER (I) + EM + PG

c. d. e. d.

Recall

ER(F)

e.

ER( I) ER (I ) + EM

f.

ER(F) + EM

that

EM

stands

for

'extrametricality.

The

first

four

(+PG)

correspond to the insensitive bounded systems, whereas the latter four correspond to the insensitive unbounded systems, i.e. those that do not have alternating stress. Let us now investigate how this extremely simple approach can be extended to handle all the other stress systems, without losing its attractiveness.

I will

discuss here the cases of Hawaiian,

Garawa,

Winnebago, Passamaquoddy and Tiibatulabal. I will first consider the Hawaiian stress system. Recall that this system was like elementary systems with two additional properties. The main stress is penultimate, rather than final and in words with five or more syllables the initial syllable has a secondary stress. Let us assume that the final syllable is extrametrical. How are we going to generate the beat at the left edge? Prince's solution was to formulate another ER, operating before PG. After that PG, moving away from main stress, fills in the rhythmic space without creating a clash: (93)

* * *

*

*

*

*

*

a

a

a

a

a

a

* (*)

=

>

*

* *

*

*

*

*

*

a

a

a

a

a

a

=

a

>

183 * *

*

*

*

*

a

*

*

*

*

*

a

a

a

a

*

a

a

The difference between Hawaiian and Maranungku is then that the latter lacks an ER, which causes what I called in the previous section an antipole stress. In the previous section I have suggested that the presence of an antipole stress can be accounted for by assuming that syllables outside the stress foot are gathered into a second foot. The rhythmic structure of Hawaiian words is accounted for by associating a rhythmic melody to the final main stress: (94)

^ ^ ^ w

s

^K Aw

s w w w w s a *

a

*

a

*

a

*

a

*

a *

(a)

i i i I *

It is not evident that we must prevent a peak from associating to the second syllable. We might

adopt

as

a matter

of

convention

that

stresses

corresponding to trees are always stronger than strong beats on the rhythmic melody. Another possibility would be to say that the ISAR requires that each syllable that is the head of a foot must be associated to a peaks (95)

W

Aw

s w w w w s a

a

a

a

a

a

(a)

184

Garawa has initial main stress and a prefinal secondary stress. In addition there is an alternating pattern. As described in Hayes (1981) the leftmost foot is ternary if there is an even number of syllables between the main stress and the prefinal secondary stress. We can explain this pattern by assuming that the rhythmic melody is associated first to the secondary stress: (96)

w

w

w

w

s

The stress system of Garawa is special (and perhaps suspect) because the ISAR refers to the strongest syllable of the weak foot instead of referring to the strongest syllable of the strong foot. Prince notes that Winnebago is problematical for a theory in which PG is "subservient to the ER and rather characterless", filling out "those portions of the grid that the End Rule cannot reach, in an entirely predictable way" (p. 51), in short the theory we are defending here. The reason is that main stress is located on the third syllable. In Prince's conception of the End Rule we can reach at most the second syllable, with the help of extrametricality. Things are different, however, in the theory of end rules (i.e. stress rules) that I have adopted. End rules in my view are foot assignment rules, feet being bounded or unbounded. We can therefore reach the third syllable by marking the first syllable extrametrical and assigning a bounded WS foot, quantity insensitive:

185

The implicit claim behind the present proposal is that alternating stress in all the examples given is a mere phonetic phenomenon,

completely

determined by the main stress. The question might be raised whether alternating stress assignment cannot also be a rule of the phonology proper,

ordered

at an

earlier

level. Without

even considering

the

implications one would expect such a situation to be possible. Phenomena that are phonetic ("low-level") in one language may be phonological in others. This is a very common state of affairs. Suitable candidates are cases where the location of the main stress depends on counting syllables from the word edge that is opposite to the main stress location. I would like to argue that in such cases application of alternating stress assignment has gained the status of an iterative phonological rule, whose structural description is limited to two grid positions: (98)

phonological PG *

a>

* *

> **

0 0

0 0 *

k#

An

* * o o

important

>

* * a a

consequence

of

this proposal

is

that

the

systems

of

Passamaquoddy, Creek and Cairene Arabic can thus be handled without assuming FCO. And since these languages provided the crucial evidence for that parameter it can be dispensed with altogether. Take the system of Passamaquoddy as an example: (99)

* 0 0 0 0 0 0

PG(LR;tr) >

*

*

*

0 o o o o o a

b

It is not necessary to say that PG (LR, tr) is FCO, because PG is an iterative rule whose SD is limited to two positions. In the schematic example given above, it finds two such substrings, a and b. The rule, applying to the first two grid positions cannot see the third position, which happens to be heavy. Note furthermore that assuming a heavy syllable to be monopositional circumvents the problem that it would otherwise not be clear why the second

186

application disregards the alleged second grid position of the heavy syllable. Mora sluicing is unnecessary. The second layer of perfect gridding in Passamaquoddy is of a phonetic nature. Here we find no clashes. There is only an alternating pattern moving away from the main stress. Passamaquoddy

is a language that has both phonetic and phonological

gridding. The interaction between heavy syllable stressing and assignment of alternating stress can also be handled within the present proposal. Consider the case of Tiibatulabal. The stress system of this language requires the following set of rules: (100)

a. Assign main stress to the final syllable b. Assign stress to every heavy syllable c. Associate the rhythmic melody peak-to-peak

(101)

a a a a a a a a

I A *

*

* *

l

*

i

*

l

* *

A

l

*

*

* *

• *

* *

The difference between phonological and phonetic gridding lies in the fact that the second really produces ideal rhythm. That phonetic gridding serves the "laws of eurhythmy" better than phonological gridding is not different from the phenomenon that lower-level rules generally serve the "laws of pronunciation and perception" much better than phonological rules proper. I

emphasize

that the distinction between phonetic

and

phonological

gridding is by no means an unsurprising move. To make this clear I will briefly discuss the completely parallel case of syllable weight. There is presumably no language in which all syllables are precisely the same in terms of length, intensity, sonority or whatever phonetic property that may be intrinsically connected to stressing. In such languages, we might say, syllable weight is a phonetic phenomenon, taken into account after stress has applied. However, in a language in which syllable weight does count for stress purposes syllable weight has been "phonologized", which means that the phonetic rules that spell out syllable weight are now ordered before the stress rules. All this simply involves the wellknown phenomenon of phonologization. What I have proposed above boils down then to the (perhaps trivial) claim that imposing a rhythmical pattern on words is a phonetic property that has, in some languages, been phonologized.

187

I realize that the proposals advanced in this section are programmatic and in need of further elaboration. The point that I want to make here is that main stress assignment must take priority over alternating stress assignment. It has furthermore been suggested that such a move allows us to combine (a slightly modified) metrical and autosegmental theory in an interesting way. In the next section I want to support the priority of main stress assignment with a quite simple argument, involving the treatment of free stress systems.

4.3.4.

Fixed stress and secondary stress

4.3.4.1. Introduction So far we have discussed stress systems in which the location of main stress was predictable on phonological grounds. Systems like these are called fixed stress systems. Opposed to fixed systems are free stress systems. In the latter type of system the location of the main stress is not predictable on phonological grounds and morphological

information.

must therefore be regarded as lexical or Lexical

information

means

that

certain

segments or morphemes are provided with a stress mark in the lexicon. It is known, however, that in most of the so-called fixed stress systems certain irregularities occur, whereas in systems that have been described as free subregularities can be discerned in many cases. It is therefore impossible to draw a sharp line between fixed and free systems, not least because the status of being irregular is itself dependent on the analysis. Within the SPE framework it is for example possible, if not required, to establish

phonological

regularities by giving

forms that appear

as

exceptions at the surface a different underlying representation to which a regular phonological rule may apply. Stress rules refer to specific lexical items, affixes, word classes or other non-phonological information only as a last resource, i.e. in those cases where any phonological generalization requires

the

analyst to assume

underlying

representations

that

are

completely ad hoc. Within this framework then the dichotomy between fixed and free stress systems appears in a different way. Fixed stress systems have stress rules that are "surface true", i.e. that express generalizations about the surface representation of the words of the language. Typically such rules refer to phonological properties, such as edges of prosodic domains, syllable weight properties and the like. Free stress systems,

on

the

other

hand,

have

rules

that

refer

to

abstract

representations, i.e. to phonological properties that are subject to

188

transformation

in

the

subsequent

derivation

or

to

non-phonological

features. Within the framework of lexical phonology the distinction between "late" and "early" stress rules has been given a clear status. The fact that early rules are often restricted to apply in words derived by particular classes of affixes is no longer attributed to abstract phonological properties of these affixes, but rather to a classification of affixes. In this sense lexical phonology meets some of the objections that have been raised against the SPE "phonologism", though some argue that even the generalizations made within one morphological level are not really of a phonological nature and should refer to classes of affixes directly (e.g. Strauss 1982) . I will not go into this issue here. Within lexical phonology early rules are referred to as cyclic (or lexical) and late rules are referred to as post-cyclic (or post-lexical). In this section I will demonstrate that the main stress first theory, proposed in the preceding section, finds support in the fact that the system of secondary stresses in free and fixed stress systems is completely governed by the same rules. A second issue that I wish to discuss is related to the first. In the previous sections we have seen many stress systems in which secondary stresses (except for the antipole secondary stress) are determined on phonological grounds: (102)

a. location of main (or antipole) stress b. syllable weight

I will investigate in this section whether the rules that assign secondary stress in fixed stress systems are sufficient to account for secondary stress in free stress systems. For those who are familiar with the literature on "cyclic stress rules" the obvious answer will be no. A certain syllable of a morphologically complex word may carry a secondary stress because the "same syllable" of a related word from which the complex word is (indirectly) derived carries main

stress.

Since

such

secondary

stresses

cannot be predicted

on

phonological grounds alone, they support the view that among non-primary stresses the distinction between fixed and free exists too. This brings the number of non-primary (or secondary) stress types to three (ignoring the antipole secondary stress):

189

(103)

Secondary stress -types a. Sec. stress due to syllable weight b. Sec. stress due to echo c. Sec. stress due to cyclicity

It is logically possible that all three types play a role in one and the same language. In fact in most analyses of English stress it has been claimed that English is such a language. Recall that although echo stresses are determined on phonological grounds a language having free stress can still have echo stresses. This is so because the reference point for echo stressing (i.e. the main stress or antipole stress) is a phonological property, no matter how this reference point is determined. It would a priori not be unreasonable to suppose that cyclic stress can only occur in free stress systems, i.e. that a necessary (though not sufficient) condition for the presence of all three types of secondary stresses, and in particular the presence of cyclic stresses, is that the language in question has free (i.e. lexical) stress.

This is not a point of

logic, but an empirical claim. I will return to this point in more detail in the following section, but here I will try to indicate why the relation between free stress and cyclic stress should exist at all. Suppose that a language has fixed stress. Within the model of lexical phonology rules that assign stress will apply at the post-cyclical level in such a language. It is assumed in this model that at that point information about word

internal morphological

structure is no longer

"visible".

Suppose now that in such a language cyclic stresses must be assigned. To be able to do this we must be able to see whether a certain word has a smaller part that can be equated to another word in order to carry over stress information from the latter to the former. But this boils down to restoring the word internal grammatical structure and this nullifies the claim that postcyclic rules are "blind" to word internal structure. In other words we lose the possibility of postulating a severe restriction on the way in which different components of the grammar interact, viz. the constraint that the maximal domain of a component A, the cyclic component, which directly feeds into a component B is the minimal domain for B, the non-cyclic component. On the other hand, if a language has free stress this means that morphemes form part of complex words with their stress properties, either because they are idiosyncratic properties of the morphemes or because they are assigned by lexical rules. At the postcyclical level complex words will therefore already contain their cyclic stress, as well as their main

190

stress. Postlexical stress rules in such a language can assign prosodic structure and a rhythmic melody. In the next section I will discuss secondary stress in Italian. The system of secondary stress in Italian has recently been described in some detail in Vogel & Scalise (1982). The purpose of giving a reanalysis here is not to criticize their analysis which is carried out in a standard segmental framework. The purpose is rather to study the interaction of echo and cyclic stress and to show how the facts that Vogel & Scalise describe are handled in a different framework, i.e. the (revised) tree-cum-grid theory.

4.3.4.2. Secondary stress in Italian Vogel and Scalise hold the view that main stress in underived words is a lexical property (p. 214): It is generally agreed that primary stress is not predictable in Italian, at least on the basis of purely phonological criteria, given such minimal (or near minimal) pairs as ancora

1

anchor'/ancora

'again', parlo 'I speak'/parlo 'he spoke' and dutttile 'ductible'/ sottile 'thin*. The placement of main stress in Italian is, despite the facts just mentioned, governed by a number of regularities that make it possible to arrive at a rule-based analysis of Italian stress. This is not inconsistent with Vogel and Scalise' s view, because the rules that assign stress would be lexical in the sense of Kiparsky (1982, 1983) and Mohanan (1982). Hence, although rule-governed, stress can still be seen as a lexical property because the stress rules are lexical. In line with the proposals in the foregoing sections I assume that main stress assignment involves creating word internal prosodic structure and the projection of a degenerate grid. If main stress assignment is a lexical rule, I will assume that on each subsequent cycle the word internal prosodic structure assigned on the previous cycle is erased, while the grid that was projected remains. The focus here is not on main stress, however, but on secondary stress. We will see that the placement of secondary stress in Italian is predictable by the same kind of rules that were motivated in the previous sections. Such a state of affairs strongly suggests that the "main stress first" approach that I discussed in the previous section is correct. We can say that the difference between fixed and free stress systems is caused by a difference in the way main stress is established, i.e. cyclic or post-cyclic. Once the

191

location of main stress has been determined, the derivation in both types of systems is governed by the same post-cyclic introduction and association of a rhythmic melody. Vogel and Scalise (henceforth V&S) first consider the

(secondary)

stress pattern of underived words in which they attest the following stress patterns: (104) i. 6 •••

gru, cane, lampada

ii. oo ...

metà, lavóro, catastrofe

iii.òao ...

colibrì, òbilisco, Aristotele

iv. Sacra . . .

mercoledì, temperatura

To describe this it is sufficient for V&S to postulate the following rule: (105)

Initial Stress Rule (ISR) a

->

[+str]

/

#

a

a [+str]

The first syllable is stressed if it is separated from the mainstressed syllable by minimally one syllable. V&S proceed with the stress pattern of derived words and they establish that the following generalizations, which were found to be true for underived words, are also valid for the majority of the derived vocabulary: (106)

a. There are no stress clashes (i.e. no sequences of stressed syllables) b. Words begin with a stressed syllable (unless this would lead to a clash) c. There are no sequences of more than two unstressed syllables.

The set of stress patterns that conform to these generalizations as far as derived words are concerned shows more types than we have seen for the underived words, which is due to the fact that the former may be much longer: (107)

i-iv. see above v. òaòaa

. ..

elegantemente

vi o oo a a o ...

matemàticamente

vii. oooooo ... . N0000000 * ' . . . vili.

effervescentemente dolorosissimamente

192

ix. acraaaaa . X. òoaòaòao . àaaàaàacraa .

precipitevolissimevolmente

Most of these patterns (all except vi and viii) can be derived by applying, after the ISR, a rule assigning alternating stress applying from right to left. V&S formulate this rule as follows: (108)

Stress Insertion Rule (SIR) °

->

C+str]

/ a -

o

(right-to-left)

What is lacking so far is an account of the remaining patterns: vi and vii. When we compare vi with vii and viii with ix we see that longer sequences of syllables can satisfy the above generalizations in two ways. A third alternative: abahaa. .. is blocked by the presence of the ISR. The question then is whether in these cases the choice of a particular rhythmic pattern is arbitrary or whether it is determined by particular factors. As for the second possibility there are two candidates: syllable weight or cyclicity. V&S dismiss syllable weight as a possible factor by simply giving a list of the relevant counterexamples. That syllable weight is irrelevant is not very surprising if it means anything at all to say that Italian is a syllable-timed language. By definition all syllables are of equal weight in a syllable-timed

language. They

choose

for the

second

possibility:

"secondary stress in complex words depends on the stress of the smaller components

contained

within

them."

(p.277).

According

to

the

view

expressed above concerning the relation between cyclic stress and lexical stress

the

presence

predictability

of

of

main

cyclic stress

stress in

is possible,

Italian.

The

given

following

the

un-

examples

illustrate the dependency of cyclic stress on the stress pattern of embedded words: (109

impossibilita intenzionalita

-

impossibile intenzionale

I will now survey how V&S propose to derive the stress pattern of complex words and then account for the same data using the grid-theory. Consider the following four examples:

193

(110) a. vino + àio

->

vinaio

b. spigolo + oso + issimo '

->

spigolosissimo

/

v febbrilmente y infedeltà

*

c. febbre + ile + mente ' , , d. in + fedele + ta ->

->

Note that there is a rule of vowel deletion, which I will not discuss here. V&S argue that forms such as these motivate the following rule: (111)

Clash Avoidance Rule o _> [-str] /

o [+str]

[+str]

ri W

(right-to-left)

This rule must operate from right to left because of the form in c. Left to right application would delete the stress on the first syllable. Of course one could assume that this stress is restored by the ISR. But this leads to less economical derivations, something that, according to the authors, should be avoided. Now consider the following examples: (112) a. beato + itudine b. dolóre + óèo •

->

beatitudine

-> doloróso f

c. dolore + oso + issimo

->

'

dolorosissimo

The secondary stress that we expect to find on the second syllable occurs on the first. To account for this V&S propose another rule: (113)

Stress Reversal Rule a _> a a C-str][+str]

a

/ # - . . .

[+str][-str]

a C+str]

Again the correct result cannot be obtained by the ISR since after application of this rule the CAR would remove the initial stress in each case. Recall that this rule applies from right to left. At this point one might wonder why V&S do not assume that all stresses but the rightmost one are simply deleted, after which ISR will assign initial stress in each case. The reason is of course that if the stress of the embedded word does not clash with a stress on the next morphological layer and is not located on the second syllable it survives as a cyclic stress. The ISR does apply to derived words but after CAR: (114)

meraviglia + oso

->

meraviglióso

194

V&S demonstrate finally that the rules they have proposed must be partiall} ordered as in (115). Other orderings lead to longer derivation or to the wrong output: (115) CAR

. . jlSR { 1 ISSR)

-

-

SIR

They furthermore assume that all these rules operate at the post-cyclic level. Let me now show how the facts reported by Vogel & Scalise will be accounted for in the (revised) tree-cum-grid framework. First of all I assume that main stress assignment is cyclic. I will also assume that CAR is a cyclic rule. In this way I do not have to say anything about its direction of application: (107)

CAR (cyclic)

On the last cycle a complete prosodie structure will be assigned, in which syllables outside the stress foot are gathered into a foot to account for the initial beat: (108)

W

S

I

S w w w w I impossibilita

. , # * * * * * * *... *

*

*

The star on the third syllable then is a cyclic stress. The corresponding stress foot assigned at the inner cycle [impossibile] has been erased. Clearly, the above account makes no use of rule order except for the fact that we assign CAR to the lexical phonology. Let me give here one example of a derivation involving the application of the CAR:

1 95 dolore] * * *

(109)

oso] * *

issimo] * * *

1st cycle 2nd cycle

CAR

3rd cycle

CAR

post-cyclic

A A

S W W S

WW

[dolorosissimo]

*

* *

*

*

*

* *

*

*

* *

* _

Let me add a critical note regarding the cylic stresses. Some speakers of Italian (including linguists) stressed

syllable and

claim that the syllables between the main

the initial syllable are subject to variable

stressing. This does not imply that cyclic stresses do not occur, not even if no trace of an alternating pattern could be found phonetically. But it is an indication of the fact that the occurrence of cyclic stresses cannot be predicted

by

hard-and-fast

rules

but

rather

follows

from

certain

tendencies that may overrule the effect of PG, which between the main stress and polar initial stress may apparently move in either direction in Italian. As for initial stressing, it has been claimed (e.g. in Selkirk 1984) that a secondary stress may fall on the first or second syllable, depending on the preceding environment. V&S mention a limited set of examples that indeed have an alternation between secondary stress on the first or the second syllable. As a possible reason (apart from dialectal difference) V&S mention the preceding context:

(110)

* * * * * * * * * * *

*

* * *

società oceanografica * *

* *

*

*

*

*

* *

*

* *

*

*

*

*

*

associazione oceanografica

196

Similar

observations

concerning

English

(Ticonderoga

versus

Fort

Ticonderoga) and Dutch are found in Prince (1983) and in van der Hulst and Moortgat

(1981) respectively.

If it is true that initial stressing is only regular if the word has no lefthand

environment

it

is

possible to assume either that prosodie

structure is erased when going from the lexical to the post-lexical, i.e. phrasal level, or that the complete prosodie structure is assigned at the phrasal level, and not earlier: (111)

S WW

S WW w

w w s w w w w

s w w

[società] [oceanografica] -> After

(re)grouping,

strings

of

[società syllables

s w w

oceanografica] between

main

stresses

are

associated to the rhythmic melody, and it may be the case that there is a conflict here between associating

left-to-right

or right-to-left.

The

shift from the secondary stress in Italian suggests that association is from right-to-left in this

language.

To conclude this section, let me draw attention to the following. In Bruce (1983) we find an extremely interesting report on rhythm in Swedish, in which it is shown that a pattern of strong and weak syllables is found in between main stresses according to the following

rule:

Generally, the unstressed syllables will be alternately weak and strong starting from the upcoming stress and counting backwards. Phonetically

this

is

reflected

as

an

durations between successive syllables,

alternation (p. 35)

Consider the following examples taken from Bruce (112)

soldater me kapital

in

(1983, 40):

soldater me kapital

stress

- + -

- - - +

- + -

- - - +

strength

- + -

- - - +

- + +

- + - +

attentat me soldater stress

attendat me soldater

-

- +

-

- +

-

- -

+

-

- +

-

strength -

- +

-

- +

-

+ -

+

+

- +

-

relative

197

The level called "stress" by Bruce is much like our degenerate grid level (except for the fact that Bruce does not report an antipole stress), while the

"strength"

level

corresponds

closely

to

our

rhythmic

melody,

associated from right-to-left. Observe that, according to Bruce, a strong beat can be added on the syllable that immediately follows a syllable with main stress. This suggests that clash avoidance may

not be a real

phenomenon, when dealing with beats that occur on different levels, i.e. the stress level and the strength level (in Bruce's terminology) or the degenerate grid and the rhythmic melody (in the terminology employed here) .

4.4. Metrical feet

The position that has been defended in the previous sections entails that word internal prosodic structure does not comprise the type of foot structure that is familiar from the standard arboreal theory. The type of foot structure that occurs in the arboreal theory discussed in the previous sections always has a head syllable which has main stress, or, when a lefthand environment is missing, the strongest non-primary stress. The notion foot defended here stands much closer then to more traditional usages of the term as found, for example, in the work of Abercrombie (1965, 1967). In this section I will look at the typical argument in favor of using the type of foot that characterizes alternating stress. I will show that the argument is not compelling by discussing a reanalysis of the type of data on which the argumentation is based. Finally I will refer to a case discussed in Lozano (1982) where it is clear that the reanalysis that is suggested for the type of data at issue is correct. Selkirk (1978), Kiparsky (1979) and Hayes (1982) have discussed a number of phenomena that are typically used to argue in favor of standard metrical foot structure.

Selkirk discusses the phenomenon of closed

syllable

adjustment in French and Kiparsky uses the phenomenon of flapping. Here I will discuss one of Hayes' examples. Consider the following SPE-type of rule:

(113)

0

->

t

/

n

-

s

f# IV

This rule accounts for the fact that in a word like prince or pincer an

198

epenthetic t may appear between the n and the s. The SD of the rule is motivated by the fact that no t is inserted in a word like consort or pincet. In the latter case the sequence ns is followed by a stressed vowel, whereas in the former case this sequence is word final or followed by an unstressed vowel. Hayes points out that the disjunction found in the rule appears in other rules as well. I quotes The recurrent, mysterious disjunction found in these rules has been clarified by Kiparsky (1979), who suggested that they be restricted to apply within the metrical

foot, where

feet can be defined

independently on the basis of the English stress system (cf. Selkirk 1980) . For example, in prince and pincer, the [n] and [s] are within the same foot, thus permitting [t] insertion, whereas in consort the intervening foot boundary blocks the rule: prlns

plnsr

kon sort

s w

VF

The question must now be answered how we are going to avoid the recurrent disjunction

mentioned

above

if

we

abandon

standard

metrical

foot

structure. The only possibility is to invoke the notion of ambisyllabicity. Let us assume that there is a quite general rule that creates ambisyllabic consonants: (114)

* *

a

*

c

Y

I.e. a stressed syllable attracts a following consonant if the next syllable is unstressed. Given such a rule we can formulate the epenthesis rule as a simple syllable domain rules (115)

a

A

prlns

a

a

//M

plnsr

o

a

AA

konsort

199

0

->

t

/ n - S)q

Given the rule that creates ambisyllabic consonants, we can reformulate all rules of this type as syllable domain rules, which means that we no longer have a recurrent disjunction in our rules. The

above

argument

shows

that,

although

standard

metrical

foot

structure can be used to simplify the formulation of rules, it is not absolutely required. All examples that I know of (as offered by Hayes, Kiparsky and Selkirk) can be analyzed in essentially the same way. The availability of a foot-based and a syllable-based analysis for segmental phenomena has recently been discussed in Lozano (1982). Lozano analyzes data from Spanish and shows that a particular process allows both a foot-based and a syllable-based analysis. She then continues by showing that the syllable-based analysis can be independently motivated. The process at issue is called 1-velarization. An initial glance at the data suggests that an 1 is velarized when occurring in the rhyme (e.g. (116a)), but Lozano supplies additional data showing that a velarized 1 can also be in onset position, albeit in the onset of an unstressed syllable (e.g. (116b)): (116)

a. azul

palo

"stick"

normal "normal"

"blue"

pelo

"hair"

sutil

fila

"line"

"subtle"

b.

A possible analysis of such facts would be to say that the 1 is velarized if it occurs inside a foot in non-initial position. According to Lozano, however, the additional data do not force one to adopt the view that the velarization rule has to refer to other than the rhyme position. To explain velarization in the cases under b. we only have to assume that the intervocalic 1 is ambisyllabic, i.e. that it is both part of a rhyme and a following onset. We see then that the same strategy used above can also be followed here in order to avoid reference to foot structure. Lozano assumes that the rule creating an ambisyllabic 1 is itself a footbased rule, but since my goal is to avoid reference to foot structure I will not follow her

in this respect. As long as no other rule refers to "foot

structure" the 1-ambisyllabicity rule might just as well be formulated along the lines of the more general rule that we proposed for English. The crucial point of Lozano's paper is that she continues to show that there is an independent reason for representing the 1 as ambisyllabic. This reason involves a process of vowel tensing. Among the non-low vowels there

200

is an allophonic variation between tense (closed) vowels and lax (open) vowels that is determined by syllable structure. The former occur in open syllables, the latter in closed syllables. The important observation is that the lax vowels appear not only in syllables that are clearly closed, but also in those cases where they are followed by an 1 that is ambisyllabic according to the rule just mentioned. From this we may conclude that the ambisyllabicity is independently motivated. It is of course not possible to conclude that the syllable-based approach to the type of data discussed in this section has been validated in general. The previous example does suggest, however, that such an approach has some credibility, and is certainly not less likely than the foot-based analysis. I conclude that standard metrical foot structure, to the extent that it differs from the kind of foot structure that is assumed in the theory proposed in the previous sections, has no strong motivation apart from its primary motivation, which is to characterize alternating stress patterns. We have seen that alternating stress patterns can be characterized in terms of a rhythmic melody that does not impose a constituent structure on strings of syllables. Hence standard metrical foot structure can be eliminated from the theory.

4.5. Conclusions

In this chapter I have discussed several issues concerning the representation and analysis of stress. We compared in detail two current theories referred to as the grid-only and tree-only theory and concluded that there are no significant differences between these two theories, apart from the fact that in the latter we assign a detailed tree structure to the text. I then argued in favor of a compromise regarding the prosodic constituent structure. This compromise tied in well with the point of view, adopted in section 4.3.2., that main stress assignment should be separated from assignment of rhythmic structure. The former can be formalized in terms of the tree format, the latter in terms of an autosegmental theory. In the final section of this chapter it has been shown that argumentation in favor of standard metrical foot structure is inconclusive.

Chapter

5

Dutch Stress

5.1. Introduction

In this chapter the focus is on Dutch stress and its relation to syllable structure. In section 5.2. I offer a discussion of the various ways in which stress assignment is sensitive to syllable structure. In section 5.3. an analysis is presented of the stress pattern of Dutch, focussing on monomorphemic words. I will offer two analyses that differ crucially with respect to the type of foot that is chosen. The first analysis is carried out within the standard metrical (arboreal) theory and the second within the tree-cum-grid theory that has been proposed in the previous chapter. I will argue in favor of the latter analysis.

5.2. Stress and syllable weight in Dutch

In section 3.4. I briefly discussed two cases in which there is a clear correlation between syllable structure and stress in Dutch. Simplex (i.e. morphologically unstructured) words with a final syllable of the shape (C)VVC (and to a lesser extent (C)VCC) are stressed on the final syllable almost without exception: pistool

legioen

anakoloet

konijn

kapitaal

calamiteit

banaan

fenomeen ceramiek avontuur

locomotief karikatuur

fontein

tamboerijn

valeriaan

juweel

pelikaan

capaciteit

konvooi kameel

salamandrijn

202

In terms of the mora theory discussed in chapters 2 and 3, we can correlate this with the fact that such syllables are superheavy, consisting of three morae. An even more compelling generalization that was mentioned in connection with the issue of representing syllables with a schwa is the fact that syllables with a schwa cannot bear main stress. In van der Hulst & Moortgat (1981) the distinction between syllables with a schwa and syllables with a full vowel with respect to stress placement was treated as an instance of the syllable weight distinction. The analysis of syllable structure in Dutch that I offered in chapter 3 showed that syllables with schwa can be analyzed as non-branching (i.e. having one mora), and syllables with a full vowel as "»ranching (i.e. having two morae). The two ways in which syllable structure influences stress placement both fall within the mora theory under a single geometrical property of syllables, viz. the number of morae that a syllable contains. A third property of syllables that has been claimed to be of relevance is the distinction between syllables of the type VC and of the type W

(most

explicitly in Kager & Visch 1983). The relevance of this distinction is much less clear, because a generalization based on this distinction faces many exceptions, but the claim is that VC syllables count as heavier than syllables. At

first

sight one might be inclined

W

to think that the

distinction between VC syllables and VV syllables is also of a geometrical nature involving a distinction between branching versus non-branching of some syllabic node, e.g. the rhyme. Yet I will argue that this is not the case. The main reason is that such a geometrical interpretation is at odds with my analysis of the Dutch syllable. The decision to analyze the openclosed

distinction

in non-geometrical

terms is backed

up by cross-

linguistic observations concerning the marginal role played by this distinction in stress systems. I will argue

that the third distinction

involves greater heaviness of closed as opposed to open syllables, where closed will be defined as "ending in a consonant" and furthermore that the distinction plays a marginal role in the stress system. Before I analyze the stress pattern of monomorphemic Dutch words in detail I will discuss some ideas concerning the representation of syllable weight that appear in the literature. The factors that are claimed to determine "syllable weight" in Dutch do not exhaust the list of possible relevant factors in other languages. There are other factors that may play a role and for these it is quite certain that no plausible geometrical interpretation can be given. Given then that syllable weight is determined both by geometrical and non-geometrical

203 properties

of

syllables,

my

decision

to

interpret

the

open-closed

distinction in non-geometrical terms does not involve an ad hoc move. Newman (1972) is one of the first to point out the relevance of the dichotomy heavy-light as distinct from the more familiar dichotomy openclosed. The cases of syllable weight that he mentions all involve the following opposition: (2)

heavy: W ,

VC

light: V

Interestingly, Newman adds that there may be circumstances in which one wants

to

distinguish

between

W

and VC

in terms of the

open-closed

distinction, but he does not mention such a case. The distinction given in (2) occurs in Dutch with the extra condition that the light syllable must contain a schwa and if the distinction W / V C is relevant Dutch would also be an example where heavy syllables fall in two classes. Hayes

(1981), in his discussion of syllable weight,

starts out by

discussing the opposition between W , VC versus V, i.e. the instantiation that Newman deals with. In the theory that Hayes adopts the distinction is represented as follows: (3)

heavy

light a

a

a

A A 1 A IA

0

R

c v v

0

A II

R

O R

c v c

c v

Small c's and v's represent segments, rather than slots here. Hayes adds: "It is interesting

that virtually

no rules of tree

construction

are

sensitive to the structure (or even presence) of the syllable onset. To account for this, I will assume that all unmarked stress rules apply on the rime projection. " (p. 41). In terms of the mora theory of syllable structure that I have adopted

in chapters

represented as follows:

2 and

3 this type of distinction

is

204

(4)

heavy

light

a

a

A A AI AI m

m

c v v

nt

m

c v c

a

I A m

cv

The sonority threshold for morahood is such that any segment is allowed to occur as a second mora. I have shown in chapters 2 and 3 that the mora representation of syllable structure is adequate to deal with the facts that have been considered in these chapters, but I also pointed out that the same is true for a different type of representation which is in line with another more common view on syllable structure, involving subsyllabic categories such as onset, nucleus, coda and, perhaps, rhyme. I referred to this second approach as the metrical theory of the syllable. In this chapter I will further explore the type of representation involving the category mora, but I add that by making use of the category nucleus, in the way suggested in chapters 2 and 3, we can also make the distinctions among syllable types that are relevant in Dutch (cf. 22a below). At the relevant places I will refer to the possible alternatives in this version of the metrical theory. The second common type of weight distinction involves the opposition between long and short vowels, disregarding the presence or absence of postvocalic consonants. Newman does not discuss this type, but Hyman (1977, 75) mentions the fact that "many languages have stress-placement rules which refer to the distinction between long and short vowels." As Hayes observes, "the proper representation of this opposition is not agreed on in the literature." (Hayes 1981, 41). The solution that Hayes adopts involves the notion of projection. We are allowed to project segments that share a certain feature. Hence we can project from the complete sequence of segments the substrings consisting of vowels: (5)

heavy

light

syllable -> V-projection

A IA a

O R

c v v

A

syllable -> V-projection a

A II

R

O R

- > v v

c v

I

R ->

v

205 a

a

/\R

O

I

A

R

A

C V V C

->

A

V V

Under the mora theory the distinction branching versus non-branching

/

\

A A

In this theory

R

C

V C

A

->

V

is represented again in terms of

syllables:

IAn

m

A1

c v v c

0

A

C V V

CVC

the distinction

A

rn

falls out by

cv

saying

that

the

sonority

threshold is such that only vowels can be a second mora. I refer the reader to chapter 2, sect. 2.2.4. for a discussion of the use of the mora threshold. The two types discussed so far are unified in Hayes 1 metrical account in terms of the category rhyme. In the metrical approach that I advocated, both are unified in terms of the category nucleus. In the mora account, finally, both types are unified under the heading of branching versus non-branching syllables and there is no need for the projection device, let alone to project different

levels.

Another instantiation of the weight distinction that Hayes discusses is the opposition between syllables with a full and syllables with a reduced vowel

(Hayes 1981,

57):

I assume that this distinction is underlyingly one of vowel length, i.e. of gemination . This assumption can be motivated in two ways: first, the full vowels are phonetically longer than the reduced ones; and second, there are apparently no languages having an underlying three-way distinction of the type reduced : full short vowel : full long vowel. This would follow automatically from the assumption that both

the

full-reduced

represented

and

the

long-short

distinction

must

be

underlyingly as gemination.

The situation in Dutch, where the full/reduced contrast plays a crucial role, is perhaps not exactly of the type that Hayes has in mind, because Dutch has a contrast between full long and full short vowels. It is the case, however, that full short vowels and full long vowels share the property that

206

they must all occur in a branching (i.e. heavy) syllable. Hence with respect to stress assignment all full vowels function as a natural class as opposed to the reduced vowel, i.e. the schwa: (7)

Heavy

a

A

m

Light

m

v

c|

v

v )

Hayes continues to discuss a "motley group of rarer distinctions of prominence which stress rules appear to treat in essentially the same manner."

(p. 43). The first is diacritically marked prominence. This

involves cases where certain syllables behave as if they were heavy, i.e. they attract stress, although they do not show a syllabic structure that makes them distinct from light syllables. Hayes proposes to mark such syllables with a diacritic. A second type involves a distinction between syllables with a high tone versus syllables with a low tone, where the former function

as heavy. Hayes offers no attempt to reduce these two

instantiations of the weight difference to a distinction in branching. I refer to Anderson (1983), however, for an argument in favor of representing diacritic marking in terms of branching. So vowels that bear lexical stress are marked as "branching". In this way Anderson attempts to explain an observation made by Jakobson that languages having lexical stress cannot have distinctive vowel length. I will not criticize Anderson's proposal here, but it will be clear that his line of thought is incompatible with the one that I have adopted. Finally Hayes mentions yet another weight distinction, viz. the one between closed and open syllables. He knows of two cases where this opposition appears to play a role: Tiberian Hebrew (as described in McCarthy 1979) and Seneca (as described by Stowell 1979). The treatment that Hayes proposes for such cases is to say that in such languages syllables contain a constituent called the nucleus which may only contain vowels:

207

(8)

heavy

Hence

the

open-closed

dichotomy

is

reduced

to

a

distinction

in

branchingness. Within the mora framework that I have adopted it is not possible to represent the distinction between open and closed syllables in geometrical terms. It goes against the logic of this theory to say that postvocalic consonants can count as a second mora, whereas the second part of a long vowel or diphthong, which is more sonorous, cannot. Hence we cannot create structures as in (9): (9)

heavy

m

C

A V

light

m (v)

C

C

V

(v)

Turning to Dutch for a moment, it can be shown that structures as in (9) cannot represent the correct way of representing the weight distinction between closed and open syllables. The structures in (9) embody the claim that syllables with long vowels or diphthongs count as light, whereas I have argued that syllables of this type are branching, just like VC syllables (cf. 7). I have shown in chapter 3 that the rule creating the first mora in Dutch is the one repeated here in (10): (10)

° m (v X , . . . X

) X

1 n n+1 where X is a sonority peak n+1

208

In this way a distinction was created between branching and non-branching syllables, where W

and VC both came out as branching. Hence I cannot at the

same time propose that they differ among each other in precisely the same way that they differ, as a pair, from syllables with a schwa, which are non-branching. This does not mean that the distinction between closed and open syllables (if proved useful) cannot be characterized formally. The distinction

between VV

and

VC

syllables

can be

characterized

non-

geometrically as follows: (11)

Heavy

Light

(...O o

(...V)o

A final way in which syllables may be of different weight involves the sonority of its segments. Prince (1983)

speculates on the existence of

stress systems in which syllables with low vowels count as heavier than syllables with non-low vowels. In the next section I will suggest that in Dutch the long and short vowels differ with respect to stress assignment in the sense that long vowels, apparently, count as heavier than short vowels. I know of no other good examples but, if found, they would give additional support to the idea that there are several different ways in which syllable weight can be manifested. To summarize, I give in (12) the instantiations of syllable weight that we have discussed: Heavy

(12)

Light VV or VC

V

a.

i.

b.

V or VC ii. VV closed syllable open syllable

c.

marked [+F]

unmarked

d.

high tone

low tone

e.

i.

tense V

ii.

low V

lax V high V

iii. ... Returning to Dutch, I will now discuss the following problem. As we have seen already, in the Dutch stress system the syllable weight distinction does not lead to a simple dichotomy. On the assumption that the W / V C distinction is relevant, there are at least three dichotomies (ignoring for

209

the moment my remark concerning long and short vowels): Syllable Weight Classification

(13)

lights schwa

heavy: full vowel light: VX

light:

W

heavy: VXC

heavy: VC

The question that must be answered is how this classification of syllables must be incorporated into the formal analysis of the stress placement rule. In the first analysis of Dutch stress that I will propose in the next section I will make use of a syllable weight scale for the purpose of foot assignment, i.e. I will adopt the proposal put forward by various Dutch phonologists to rank syllables on a scale on the basis of their propensity to be stressed, much as segments can be ranked on a scale in terms of their sonority. Indeed Dijkstra (1982) and Van Nes (1982) refer to the scale as a "sonority scale". The idea of such a scale is also employed in Kager and Visch (1983). Rather than speaking of a sonority scale I will speak of a weight scale. At the two poles of the syllable weight scale we find syllables with a schwa (having the lowest propensity to receive stress) and the superheavy syllables, VVC and VCC (having the highest propensity to be stressed and appearing in that order). In between we find the heavy syllables (VV, VC), of which the latter type is claimed to have a greater propensity to receive stress. As an initial hypothesis we might adopt therefore the following scale: (14) a

VV

VC

VCC

—increasing weight

WC >

Let us now investigate how, in the publications just mentioned, the weight scale is used in a theory of stress assignment that employs feet. In chapter 4 I distinguished the following types of limited quantity systems:

sensitive

210

(15)

Latin

Yápese

Aklan

There exists general agreement on the claim that Dutch stress is of the limited Q-sensitive type. This is based on the observation that stress occurs on one of the

final three syllables

in monomorphemic

words,

predominantly on the penultimate syllable. Although they show no awareness of this fact Dijkstra (1982) and Kager & Visch (1983) analyze the Dutch stress system as if it constitutes a fourth type, given in (15) in the fourth column. This fourth type may perhaps be expected to exist, given that it fills a gap in a classification that is based on the surface stress patterns, rather than on the way in which these systems can be analyzed within the metrical framework. One of the purposes of this section is to find out whether this is indeed the proper way to handle the Dutch stress system. The analyses proposed by Dijkstra (1982) and Kager & Visch (1983) imply the claim that if stress is located on one of the final two syllables stress goes to the heavier of these two (this is no different from the other types), but they also claim that stress goes to the penultimate syllable if both syllables are equally heavy. This is only a rough statement and we will go into more detail below. Let me focus here on the consequences of such a state of affairs. The type of foot that is needed must surely be quantity sensitive. As I argued in chapter 4, section 4.3.1.5., standard metrical theory employs three types of QS feet: (16)

A r

S X

I

jj •

W

A r

S

A

W

I

•

(+LCPR)

211 None of these three foot types can be used for Dutch, since these feet give us the first three systems in (15), leaving the fourth one unaccounted for. We need a new type of QS foot: (17)

F

A

s X

w

X

This foot type must be interpreted as follows. Both the left and the right node may dominate any type of syllable, but it may not be the case that the syllable which is dominated by S is placed lower on the weight scale than the syllable which is dominated by W. Let us refer to the configuration that is ruled out as a mismatch. The condition for successful foot-assignment can be stated as follows: (18)

F S X

W

K.

Mismatch Condition: o

s

£ a

w

The mere fact that an analysis of Dutch along the lines discussed here requires a new foot type is of course not an argument against such an analysis. It does make one wonder, however, whether the analysis is in fact adequate. This is one of the reasons for considering an analysis that does not make use of this type of foot in section 5.3.1.5. Returning to the weight scale on which this type of foot-assignment is crucially based I want to answer the question whether, and if so, how a relation is established between a scale on which syllables are ranked, and properties of the syllables themselves. In the proposal advanced by Van Nes the scale is primitive in the sense that it is simply postulated. Dijkstra (1982) and Kager & Visch (1983), however, argue that the ranking of syllables can be derived from geometrical properties of syllables. These authors assume the following types of syllable structure for Dutch:

212

(19)

a.

R

b.

R

c.

R

d.

R

N

N

Cd

N

Cd

V V

V

c

V

C C

A

I A

N

A

V V

Cd c

This is essentially the analysis that is proposed by Trommelen (1983). In chapter 3, section 3.5.1. we have seen that the decision to assign vowels to the nucleus node and consonants to the coda node consistently necessitates the adoption of a special convention to explain that a branching coda is allowed after a non-branching nucleus only. In short, the structures in (19) make it impossible to designate one constituent as the obligatory part of the syllable. I also argued that the need to adopt such a convention points to the missing of a

generalization, rather than the making of one.

However, ignoring that point, we can say that the scaling of syllables can be derived from the geometrical properties of syllables if we adopt the following principles: (20)

a.

A branching node attributes more to heaviness than a non-branching node

b.

A branching rhyme attributes more to heaviness than a branching nucleus and a branching nucleus attributes more to heaviness than a branching coda

Given these principles we derive the weight scale adopted above, repeated here for convenience:

(21)

| 3

| w

| vc

| vcc

| wc

Let us now see how a scaling of syllables can be derived from the syllable structure

analysis,

giving

here

corresponding metrical analysis:

both

the

mora

analysis

and

the

2 13

(22)

i. Mora theory a.

a

b.

a

c.

a

d.

A\ A\

m m

A

m m

A

m m m

m m m

v v

v c

v c c

v v c

ii. Metrical theory a.

a

I

N

I

b.

a

c.

I

N

A

a

IN

A

d.

e.

a

N

cd

A I

N

A

cd

In my account there is, in both analyses, no structural difference between b and c, nor between d and e. On the assumption that we want to make the fiveway distinction that is embodied in (21), I have no choice but to say that next to geometrical properties there are also other properties that determine the ranking of syllables. The other relevant factors are syllable closure and vowel quality. Hence a scale can be derived if we adopt the following principles: (23)

a.

Syllables are ranked according

to the number of morae

that they contain b. c.

Closed syllables count as heavier than open syllables Syllables with long vowels count as heavier than syllables with short vowels

Apparently principle b takes precedence over c where a conflict arises. This is the case with respect to VV and VC syllables. We must conclude that by adopting the syllable structure analysis that I have proposed we cannot reduce the weight scale to a single property of syllables. I do not consider this to be a disadvantage for two reasons. Firstly, as 1 have argued above, there are other examples where a weight difference cannot be related to geometrical properties of syllables. The second reason is that by "factoring out" syllable closure as a separate factor we explain why it is possible that this factor (as well as the factor vowel quality) is much less important than the factor that involves the

214

number of morae (see Hyman 1977). For these reasons I maintain that the syllable structure analysis that I have proposed provides an adequate basis for approaching stress assignment in Dutch. Before I proceed with the analysis of Dutch stress I will make a brief remark about the device of extrametricality. If we limit this device to phonological units, as is proposed in Selkirk downgrade

final

syllables

on

the weight

(1984), it allows us to

scale

or

to

disregard

them

altogether, i.e. we may mark extrametrical a final mora or a final syllable. Observe that by adopting the mora-theory we can limit the extrametricality device to stress-bearing units. It is argued in the literature that units are marked as extrametrical either by rule or on an item-to-item basis, i.e. lexically. In the latter case

extrametricality

is

a

straightforward

exception

device,

to

be

interpreted on a par with the diacritic [+F] which is assigned to syllables that receive primary stress contrary to rule. Analyzing a language like Dutch,

for which certain

generalizations

concerning stress placement face a considerable number of exceptions, we have to answer the question when a generalization concerning the stress pattern

of

monomorphemic

words

is

significant

and

when

not.

The

availability of the extrametricality device and the diacritic [+F] do not make it easy to reject a rule as invalid. Quite generally speaking anything is allowed to occur. Consider the following example. Suppose we want to establish what happens when the two peripheral syllables in Dutch are of equal weight. Words consisting of two superheavy syllables are very rare, so in order to have sufficient data on which to base our choice, we must look at sequences of the type W - V V and VC-VC. Strictly speaking,

disyllabic

words with initial stress tell us nothing: (24)

giro

moslim

baboe

atlas

These words have initial stress. Yet this may be because their

final

syllable or final segment is extrametrical. In the former case these words would be like monosyllables and in the latter case (illustrated below) like words having a second syllable with a schwa: a

(25)

a

A m m g

i

i

m r

a

a

A m m o

(o)

m

o

s

m 1

i

(m)

215

a

a

a m m

a

a

m

/I I /I

m o o d s

z o n d a

On the other hand bisyllabic words of this type with final stress are equally unrevealing: (26)

kado

kordon

idee

kristal

In those cases the final syllable maybe specified as C + F ] . Given that both extrametricality and [+F] are exception devices the mildest position is to claim

that

a

stress

rule

is

less

likely to express a

significant

generalization if the number of exceptions is greater than the number of regular cases, whether exceptionality is encoded in terms of extrametricality (to handle negative exceptions) or in terms of a diacritic mark (to handle positive exceptions). This is the position that I will adopt here. This implies that I will in principle propose a stress rule that accounts for the patterns that are in the majority, the dominant patterns. I realize that this position perhaps implies a simplification of the issues that are involved. Firstly, it may happen that a particular "dominant pattern" can only

be captured by adding ad hoc devices to the theory, in

which case one might want to refrain from accounting for this pattern as following from the rule. Secondly, we cannot be sure, that patterns which are in the majority, in fact represent what native speakers regard as rule governed patterns. There are several ways to check whether this is the case or not. One might investigate "wrong pronunciations" of foreign words or historical stress shifts as is done in Van Marie (1980). Another route is taken by Kager & Visch (1983), who claim that synchronic shifts (due to stress clash) are more likely to occur if the resulting stress pattern is predicted by the analysis. Finally, in Van der Hulst & Langeweg (forthc.) results will be reported from an experiment in which speakers of Dutch have been asked to pronounce nonsense words. As far as I have been able to judge from the results obtained, most of the patterns that are in the majority in the vocabulary of Dutch, are also chosen in pronouncing nonsense words. The analysis that is offered in the next section is based on a corpus that contains the vast majority of Dutch monomorphemic nouns, adjectives and verbs, excluding highly specialized vocabulary and words that are clearly felt as foreign by native speakers of Dutch. Of course, monosyllabic words

216

and all polysyllabic words in which only one syllable contains a full vowel have been excluded. It goes without saying that it is impossible to gather a "complete list", among other things because it is not possible to draw a sharp line between foreign and non-foreign words or specialized and nonspecialized words. I claim, however, that the collection (comprising about 3500 items) on which I base my analysis is representative and that the conclusions concerning the dominant, patterns will not be affected by possible oversights on my part. In van der Hulst and Langeweg (forthc.) "complete" lists will be made available. Here I limit myself to giving incomplete lists that serve an illustrative purpose. Because the borders of the corpus are to some extent "fuzzy" I will also speak here in terms of rough totals when comparing the various subsets of data.

5.3. The stress pattern of Dutch monomorphemic words

5.3.1. Nouns 5.3.1.1. Bisyllabic nouns Limiting ourselves to bisyllabic words for the moment let us first look at cases with a final syllable of the structure W C and VCC, i.e. with a final syllable that contains a long vowel or diphthong followed by one consonant, or a short vowel followed by two consonants. For a discussion of the Dutch vowel system I refer to chapter 3, section 3.3.3. There are about 450 cases with finally stressed VVC (the first syllable being either VC or W ) and only about 25 cases with stress on the first syllable:

w-wc paraaf paniek

tarief muziek

oli jf kanaal

lawaai metaal viool

pleidooi kameel probleem

juweel

profiel

idool

atoom

banaan

domein

aluin

stramien mihoen konijn ravijn

siroop

sigaar

poelier

papier

figuur

paleis limiet

ani js kajuit

matroos piloot

klimaat

dieet minuut

debuut

citroen patroon

217

VC-WC

(29)

octaaf

archief

octrooi

portiek

sandaal

garnaal

penseel

kasteel

reptiel

pistool

embleem

kostuum

vulkaan

orkaan

sardien

kalkoen

fontein

fortuin

dolfijn

persoon

tartaar

scharnier tamboer

cultuur

markies

fornuis parkiet

appiaus marmiet

atleet despoot

bandiet

soldaat bandiet

dispuut

EXCEPTIONS

vc-wc

w-wc sieraad

deemoed

potlood

arbeid

weemoed

kleinood

blocnote

minstreel

kalief

koekoek

lichaam

altaar

kroepoek

wierook

wierook

zwaluw

vampier arthur

mammoet cocktail

zenuw Turning to VCC cases we find a less clear picture, i.e. there are relatively more exceptions than in the previous case (regular: 110, exceptional: 70) : ( 30 )

(31)

(32)

(33)

W-VCC smaragd

triomf

spelonk

augurk

kiosk

alarm

katern

balans

eclips

traject

relict

basalt

tumult

gigant

fazant

client

talent

adept

protest

reflex

EXCEPTIONS vijand

eland

eiland

avond

agens

biceps

bruiloft

kobalt

climax

codex

miljard

concern

impuls

forens

ellips

karwats

contract aspect

insekt

consult

pendant

fondant

fragment

segment

concept

biljart

concert

rapport

contrast

orkest

arrest

attest

compost

nonsens

ambacht

asfalt

yoghurt

vc-vcc respons district

EXCEPTIONS herberg

218

albast

ballast

asbest

index

affix

Many of the initially stressed words end in /ks/ (written ) while this cluster occurs only a few times in the finally stressed cases. Facts like these can be expressed in subrules that express subregularities. As we go on I will mention other subregularities of this type, but I will not attempt to formulate explicit rules. In all cases such rules would tell us that the fact that a syllable is either marked [+F] or marked as extrametrical is regarded as "expected" given the segmental make-up of these syllables. I have not been able to find clear phonological grounds for the fact that certain endings deviate from the dominant patterns. In this type of case we find a fair number of exceptions to the generalization that final superheavy syllables are stressed. To illustrate the point that not all exceptions can be predicted consider the following pairs, which have similar endings: miljard

Oswald

tendens sarcast

nonsens albast

orkest alarm

asbest napalm

gigant

vijand

basalt

kobalt

eclips spelonk

biceps

fazant

eland

avond

Let us now turn to the stress pattern of bisyllabic nouns that end in VC and VV. We may distinguish four types: INITIAL

FINAL

VC-VC VV-VV

140

90

170

60

VC-W

170

30

VV-VC

190

120

On the basis of these totals one would be tempted to say that stress is predominantly initial (or prefinal) and hence that bisyllabic words show no evidence for a significant weight difference between VC and VV. Let us first take a close look at the bisyllables in which both syllables

21 9

have the same skeletal structure: (36)

VC-VC: INITIAL STRESS plastic

(37)

samba1

mustang consul

tangram

diftong

hertog wigwam

moslim

pelgrim

album

balkan

sultan

kremlin

perzik

molton

nectar

condor

mentor

atlas

harnas

kermis

kosmos

circus

cactus

sabbat

sorbet

sowjet

cursus cockpit

VC-VC: FINAL STRESS kornak

kristal

appel

kartel

ampul

bonbon

kordon

balkon

chanson

karkas

kompas

succès

bordes

kompres

pincet

portret

trompet

kwartet

servet

komplot

The clearest tendency is for words in to receive final stress (I use "" to indicate spelling) . There are only a few exceptions to that generalization.

I give here what

seem

to be

other

generalizations

("computed" on the complete corpus): (38)

FINAL

INITIAL

30 X (4 exc.)

14 x V+

5 x

19 x V+

6 x (2 exc.)

9 x (1 exc.)

2 x (1 exc.)

20 x (1 exc.)

16 x (6 exc.)

10 x (1 exc.)

3 x (1 exc.)

7 x +[+nas]

There are no words ending in , so if we regard the only two ending in , both having initial stress, as exceptions one might say that a sequence of a mid short [«.back, »ground] vowel and a dental consonant attracts final stress, thereby "disturbing" the predominant initial or prefinal pattern. The phonological

status of this generalization is

suspect, however. The prefinal pattern surfaces even clearer in the case of W - V V bisyllables; (39)

W - W tuba

: INITIAL STRESS yuca

sofa

noga

trema

220

(40)

specie

olie

linie

motie

ruzie

baboe

goeroe

bami

nazi

dodo

kilo

kano

provo

toto

* baby

magie

kopie

W - W

: FINAL STRESS

idee

taugee

taboe

revue

galei

livrei

etui

kopij

depot

kado

buro

reçu

bijou

In the following table we see the correlation between the nature of the final vowel and bisyllables

stress placement. I have added the score for VC-VV

(to be discussed hereafter) for comparison

(again these

numbers refer to the complete corpus): (41)

VV-VV /s/ /s/ /5/

INITIAL 71 2

/5/ /y/ /I/ /5/ /ei/ /ou/ /ui/

41 -

VC-VV FINAL 5 13

INITIAL

12

36

-

-

-

66 2

FINAL -

10 1

2

2

1

48

11

63

6

3

8

-

6

-

1

-

2

-

-

-

-

-

5 -

12

By using the notation /v/, I neglect here the autosegmental representation of length. The notation of diphthongs is also informal. For an analysis of the Dutch vowel system I refer to Chapter 3, section 3.3.3. Clearly the "disturbing" factors are the diphthong /ei/ and the mid unrounded vowel /e/. The same two endings account for the great majority of finally stressed VV-VV and VC-VV words: (42)

V C - W : INITIAL STRESS pinda

wodka

dogma

firma

junta

versie

fractie

kwestie

bamboe

armoe

tosti

kombo

saldo

shampoo

salvo

accu

moskou

whisky

dummy

panty

22 1

(43)

VC-W

: FINAL STRESS

toffee

trofee

moskee

pygmee

vallei

pastei

karwei

abdij

soldij

kandij

partij

circuit

bordeaux

essay

elpee

The facts considered so far lend support to the idea that stress is predominantly prefinal if the final syllable is not superheavy and that the difference between VC and VV syllables is of little significance. The crucial case is going to be the W - V C type. If VC is heavier than W

then we

expect to find here a predominant final pattern, but in fact we do not, as the table in (35) has already revealed: W - V C : INITIAL STRESS

(45)

wa j ang

pisang

klewang

viking

smoking

reling

koning bivak

sarong

goelasj

honing ka jak

havik

eunuch sesam

slalom

forum

datum

divan

neon

python

satan nylon

bizon

ozon fluor

radar

sonar

humor

motor

pias

sinas

lavas

polis

crisis

chaos

bios

kokos

epos

rebus

kubus

fiat

kievit

robot

rotan

W - V C : FINAL STRESS karaf

giraf

tabak

barak

kozak

barok

vazal

tabel

libel

hotel

bacil

pupil

program

kolom

roman

tiran

katrol flacon

wagon

spion

japon

baron

gazon

galop

decor

moeras

matras

procès

adres

kolos

débat

fregat

patat

raket

boeket

tablet

floret

rozet

fagot

cachot

The majority of the finally stressed cases end in the sequences that we mentioned above as attracting stress in the VC-VC cases. The picture is less neat, however, and in some classes there are more exceptions than regular cases.

222

(46)

31

X

(1 exc.)

12

X

2

X

(3 exc.)

6

X

(1 exc.)

6

X

2

X

(6 exc.)

20

X

(22 exc.

5

X

5

X

(1 exc.)

This concludes our discussion of the bisyllabic nouns. So far the safest generalization is that the penultimate syllable is stressed, unless the final syllable is superheavy; in that case we find final stress. The number of exceptions is large, however, and although a significant subclass can be identified on the basis of the segmental make-up of the final syllable the conclusion is inevitable that stress in Dutch is lexical. By this I mean that whatever stress rule we propose it makes more sense to interpret this rule as a lexical redundancy rule than as a rule that actually applies to items that have not been specified for stress. Before proposing an analysis I will first go through the trisyllabic nouns (not containing a syllable with schwa). Here we will find some evidence for the claim that the W / V C difference is relevant for the purpose of stress placement.

5.3.1.2. Trisyllabic nouns The generalization that final superheavy syllables are stressed is not refuted by trisyllabic words. The following list is overwhelming, compared to the almost total absence of exceptions. (47) X - X - W C s FINAL parachute

paragraaf

negatief

sarcofaag

mozaiek

kannibaal

kapitaal

ritueel

gladiool

diadeem

idioom

vaticaan

pelikaan

fenomeen

legioen

paviljoen

kapitein

karabijn

epigoon

kaviaar

atmosfeer

formulier

meteoor

amateur

avontuur

speculaas

paradijs

abrikoos

apparaat

dynamiet

parasiet

kwaliteit

kosmonaut

farmaceut

instituut

(48) X-X-VCC s FINAL dividend

arabesk

ordonnans

dialect

architect

adjudant

foliant

ledikant

diamant

dissident

perkament

testament

hyacint

labyrint

cineast

Turning then to trisyllabic nouns consisting of VC and W

syllables we can

have the following logically possible types. After each type the number of cases that occur in the corpus has been indicated: (49)

INITIAL

PREFINAL

FINAL

a.

V C -- V C -- V C

(-)

V C -- v c - - v c

(12)

VC-- v c -- v c

(3)

b.

V C -- V C -- V V

(-)

V C -- v c - - v v

(22)

VC-- v c -- V V

(6)

c.

V C -- w - - w

(21)

V C -- w - - V V

(43)

V C -- w - - w

(8)

d.

w - - v v -- V V

(59)

V V -- V V -- w

(80)

v v - - v v -- V V

(43)

e.

w - - v v - -VC ( 6 0 )

w -

(20)

v v - - v v - -VC ( 3 5 )

f.

w - - v c -- V C

(-)

w - - V C -- V C

(14)

V V - - V C -- V C

(5)

w - - V C -- w

(-)

w - - v c -- w

(45)

w - - V C -- w

(9)

VC-- v v - - V C

(40)

V C -- v v - - V C

(9)

v c - - w - -VC ( 1 7 )

h.

-VV--VC

In six out of eight cases the prefinal pattern is dominant. The two exceptional cases involve a sequence of W - V C # , but contrary to what one might perhaps expect we find as dominant initial stress and not final stress in these cases. To study the cases more closely let us look at the relevant examples. For the first two types we find no examples with initial stress: Type a: VC-VC-VC (50) VC-VC-VC : PREFINAL abstractum

perfectum

badminton

transistor

syntaxis

wilhelmus

consensus

prospectus

atlantis

(51) VC-VC-VC : FINAL postiljon

compagnon

castagnet

Type b: V C - V C - W (52) V C - V C - W : PREFINAL walhalla

chinchilla

gorilla

vendetta

confessie

224

commissie

percussie

attractie

injectie

restrictie

spaghetti

parlando

infectie

confectie

inductie

instructie

confetti

commando

embargo

espresso

compagnie

industrie

bellettrie

(53) V C - V C - W : FINAL chimpansee

employee

amnestie Proceeding on the assumption that prefinal stress is regular, we mark the words with final stress as exceptional (i.e. [+F]). Observe that these words have the same endings that we also found among the bisyllabic exceptional words, i.e. , and (cf. 38, 41 and 46). The remaining cases end in . Among the bisyllabic words too words ending in constitute a relatively large group of exceptions (see the table in (41), entry /I/). Type c: V C - W - W For the third class we find the three logical possibilities, although prefinal stress is dominant. (54) V C - W - W

r INITIAL

formica

marcia

maffia

raffia

dahlia

hernia

fuchsia

razzia

tombola

pergola

hospita

belgrado

indigo

piccolo

eskimo

embryo

bungalow

It seems that prefinal (/i/ ) is easily "neglected" , a fact that has been noticed by several students of Dutch stress. It is possible to suggest that an in prefinal position gives rise to a mismatch when placed in S position. Kager & Visch (1983) suggest that we acknowledge the fact that segment sonority in general is relevant for stress assignment, but the data indicate that the only convincing instance involves prefinal . This does not warrant such a general claim. I therefore prefer to regard the stressrejecting behavior of prefinal as an arbitrary property of this segment. In the next set of examples we see that there are in addition a few exceptions to this subregularity: (55) V C - W - W alpaca

: PREFINAL armada

collega

nirvana

passiva

225

concilie

communie

victorie

historie

illusie

inflatie

bombarie invasie ambitie

alkali

tornado

torpedo

meccano

bambino

alpino

gestapo

sombrero

(56) V C - W - W

bacterie adhesie conditie

mysterie diffusie andijvie albino allegro

: FINAL

orchidee

attaché

harmonie

fantasie

amfibie

farmacie

rapsodie

The cases with final stress again have typical stress-attracting endings, viz. or (both stand for /e/) and . Type d:

W-W-W

The fourth class is the largest one. Stress is predominantly on the penultimate syllable. In the majority of cases with initial stress we find the "unstressable" and most finally stressed words end in one of the stress-attracting sequences , or . (57) W - W - W

: INITIAL

canada

papoea

omega

aria

paria

varia

gloria

fresia

cavia

paprika

primula

pagina

platina

opera

dominee

kolibrie

maraboe

alibi

farao

risico

mikado

video

stereo

radio

studio

regio

folio

polio

patio

ratio

animo

domino

rotary

W-W-W

:: PREFINAL

spirea

bodega

maria

judoka

koala

akela hyena

pyjama maizena

diploma angina

aroma mascara

arena

angora valuta

natura

mimosa

jehova

fiducie

valeta tragedie

komedie kolonie

religie kanarie

familie materie

opinie memorie

mahonie cohesie

precisie

provisie

erosie

legatie

editie

militie

politie

munitie

negatie positie

notitie

emotie

promotie

devotie

solutie

Sahara peseta remedie

226

spinazie

salami

bikini

okapi

maori

rodeo

imago

dynamo

piano

casino

kimono

bolero

firato

(59) W - W - W

: FINAL

chocola

procedé

scarabee

coryfee

matinee

diarree

protegé

melodie

parodie

prosodie

strategie

litanie

reunie

therapie

theorie

calorie

afasie

relikwie

jaloezie

poezie

defilé

canapé

comité

cichorei

specerij

negorij

curacao

residu

paraplu

Type e: W - W - V C This class requires our special attention. (60) W - W - V C :INITIAL cineac

decibel

abraham

requiem

unicum

stadium

medium

odium

jodium

podium

valium

helium

opium

natrium

vacuum

rataplan

specimen

colofon

odeon

marathon

stadion

or ion

Salomon

libanon

bariton

horizon

jaguar

lucifer

jupiter

senior

junior

monitor

ananas

litotes

genius

tetanus

ramadan

socrates

syfilis

nucleus

lazarus

habitus

horizon

(61) W - W - V C

PREFINAL

museum

supinum

decorum

futurum

senator

curator

pluralis

dualis

adonis

vicaris

salaris

notaris

clematis

pleuritis

(62) W - W - V C

pronomen

FINAL

maniak

frikadel

citadel

karamel

kolonel

aquarel

bagatel

krokodil

protocol

parasol

diagram

colofon

capuchon

picador

etiket

violet

amulet

kabinet

klarinet

bajonet

cabaret

sigaret

menuet

patriot

This type is interesting because here the dominant pattern is not to have stress on the penultimate syllable, but rather on the antepenultimate, i.e.

227

initial syllable. With respect to trisyllabic words having initial stress we have observed above that in most cases the prefinal syllable contains . In this class a significant number of words with initial stress do not fall under the generalization that a prefinal is skipped: vacuum

ramadan

rataplan

odeon

co lo fon

oregon

marathon

Salomon

jaguar

libanon

oberon

equator

ananas

litotes

nucleus

modulus

tetanus

lazarus

The stress rule that I want to formulate must account for the fact that trisyllabic words of this type have predominantly initial stress. At this point we can conclude

for the first time that a rule that assigns

penultimate stress in all cases (unless the final syllable is superheavy) is not going to be successful in accounting for all the dominant patterns. Admittedly, my definition of the notion 'dominant pattern' (involving the definiens

'majority') must be followed to the letter, because we are

dealing with small numbers of words here. But I see no reason so far to think that we are on the wrong track. For completeness let us observe that the cases with final stress typically end in a stress-attracting sequence (cf. the table in 38). Type f: W - V C - V C In this class and the following one as well, we find predominantly prefinal stress and most cases with

final stress end

in a

stress-attracting

syllable. (64) W - V C - V C : PREFINAL pacific

kanunnik

elektron

professor reactor

reflector

detector

synopsis

hibiscus

alumnus

olympus

abortus

augustus

(65) W - V C - V C : FINAL apostrof

parallel

tarantel

meniscus

228

Type g: W - V C - W (66) W - V C - W

s PREFINAL

waranda

agenda

mazurka

programma

dilemma

havanna

madonna

elektra

placenta

aorta

siesta

provincie

parochie

emulsie

defensie

processie

professie

emissie

redactie

retractie

kwitantie

licentie

proportie

chianti

fiasco

flamingo

stiletto

libretto

risotto

autopsie

travestie

(67) W - V C - W

: FINAL

odyssee

neuralgie

démasqué

fricandeau

sacristie

As in class a and b, these two classes lack cases with initial stress. We can make an interesting generalization here: in all types with a penultimate VC syllable initial stress is lacking. Type h: V C - W - V C Like type e, this type shows a majority of words having initial stress: (68) V C - W - V C : INITIAL tomahawk

carnaval

archipel

practicum

calcium

cadmium

maximum

astrakan

charlatan

mocassin

lexicon

pantheon

pentagon

epsilon

handicap

zanzibar

pancreas

hercules

albatros

syllabus

omnibus

exodus

nuntius

dactylus

terminus

octopus

alfabet

almanak

alcohol

(69) V C - W - V C : PREFINAL erratum

passivum

dictator

messias

brandaris

enclisis (70) V C - W - V C : FINAL salmiak

fontanel

carrousel

lampion

pension

pantalon

matador

sjibbolet

pistolet

silhouet

Observe that this type shares a structural property with type e. Both end in the sequence W - V C # . Filtering out the cases where we find a prefinal < i> we are left with the following set of words:

229 (71)

almanak alcohol

carnaval charlatan pentagon

pancreas

tomahawk astrakan pantheon hercules

syllabus

exodus

octopus

mocassin

albatros

alfabet We have now gathered sufficient data to turn to the formal analysis.

5.3.1.3. A metrical analysis In the previous two sections I have discussed bi- and trisyllabic words that consist of syllables without a schwa. We saw that an initial hypothesis saying that prefinal stress is predominant in all cases meets with the problem that words ending in the sequence W - V C # have initial stress as the predominant pattern. In this section I will propose an analysis that explains the occurrence of all the patterns that were found to be dominant. This analysis will then be confronted with a wider range of data, including trisyllabic words with a schwa and words consisting of four syllables. The first analysis is embodied in the following stress rule: (72)

Dutch stress a. Assign feet from right to left that are -binary -Q-sensitive -labelled SW b. Assign a word tree that is -right branching -labelled by the LCPR c. LCPR In the configuration [AB] B is Strong iff i.

it branches

ii.

it dominates a superheavy syllable

iii. it dominates a marked syllable ([+F]) The type of foot on which this analysis is based has been discussed above:

230

(73)

F

A

s

X

w X

Mismatch Condition:

a > a s w a is not lower on the weight scale than s

(i.e. (74)

I

I

I

|

3

vv

vc

vcc

a ) w

I wc

An extra stipulation must be made for words with exceptional final stress. We must prevent a syllable marked as [+F] from becoming the weak syllable in a foot. To achieve this we might give such syllables a place on the scale as high as superheavy syllables. For bi- and trisyllabic words the analysis predicts the following metrical structures: (75)

Bisyllabic words

F

F S

W

s

VC

vc

w

F /\ s w

w w

vc

w

/ X F F w s VV (76)

VC

Trisyllabic words a

b.

"A vc

s

w

vc

vc

r A' vc

vc

vv

A' vc

vv

vv

I w

As

s w

w

w

231

f.

f A' I

VV

s

w

VC

VC

w

s W

Ä

w

VC

VV

h.

w

s

w

VV

W

.W

VC

s

w

VC

W

VC

The interesting aspect of this analysis is that the unexpected behaviour of three types of words (VV-VC in 75d, W - W - V C and V C - W - V C in 76g and 76h respectively) all ending in the same sequence W - V C # is straightforwardly explained. The crucial point is that a sequence W - V C # cannot be combined into a single foot, since that would violate the constraint that syllabic weight differences cannot be contradicted by the S/W labelling: (77)

A

S VV

VC

mismatch

->

To illustrate the working of the LCPR compare the following examples: (78)

a.

^ F

,W

b . ^ F

F

c. ^

F w ,s

F F

A |A A as

s w ballade

s

w

Odessa

s w

kapitein

d.

^

e.

F F

A

s

w

orchidee

[+F]

^ F F

A w

S

s

w

almanak

Given the formulation of the LCPR, words with final superheavy or marked syllables always receive final stress. The special merits of this labelling rule become clear in example (78e) where it predicts antepenultimate stress, whereas a "simple" WS labelling would give final stress.

232

As was mentioned in chapter 3, some find it attractive to argue that superheavy syllables are in fact branching feet themselves: (79)

C V X c The identical behavior of superheavy syllables and branching feet would then be explained and clause ii of the LCPR could be eliminated. Proposals along these lines can be found in McCarthy's analysis of stress in Arabic dialects (McCarthy 1979). In the following figures "X" stands for either

W

or VC syllables: (80)

X

X

cvcc

di

a

mant

I will not adopt this point of view, however. In the next section I will discuss a method of explaining the equivalence of a superheavy syllable and a "real" branching foot that is not based on questionable assumptions concerning constituent structure.

5.3.1.4. Extension of the data base The next type of case that I will consider are words with schwa. The present analysis predicts that if a schwa occurs in the final syllable, stress will be on the penultimate syllable.

Because syllables with a schwa are lower on the scale than any other type of syllable they can be combined into a foot with the preceding syllable. The

233

LCPR labels the final foot with S.

The number of exceptions that I know of

is small, compared to the list of regular cases. (82)

X-X-a

:

PREFINAL

allure

amandel

antenne

apache

asperge

bagage cassette

ballade cilinder

benzine

capsule

colonne

brigade commode

controle

corsage

diaken

discipel

douane

egypte

ellende examen

enquete extase

eskader

etage formule

eksamen etappe horloge

hypnose

jenever

kabouter

kadaver

kalender

kaliber

kamille

karakter

kastanje

kazerne

kazuifel

komkommer

lavendel

lawine

legende

machine

malaise

manoeuvre

maquette molukken

marine raorfine

mascotte narcose

méthode notulen

mirakel novelle

november orakel

nuance oranje

oase pantoffel

obstakel papaver

parabel

parade

parterre

plantage

principe

prognose

psychose

rabarber

rancune

ravage

rotonde

computer

fanfare

oktober

suede

routine

salade

taverne

tentakel

satire tentamen

seconde theater

tirade

trapeze

tribune

vedette

vehikel

visite

vitrage

vitrine

volume

zigeuner

(83)

X-X-s : EXCEPTIONS manchester

oorkonde

armoede

betuwe

nijmegen

veluwe

weduwe The analysis proposed in the preceding section predicts that if the schwa is in second position and a full vowel in final position we will find initial stress. This is so because it is not possible to assign a foot to the last two syllables

in those

cases:

the

final

syllable

is heavier

penultimate syllable and a mismatch would be the result. final syllable is superheavy we predict final stress:

than

the

However, if the

234

b.

(84)

Âw

S

a

X

S X

XC

Both predictions are borne out by the facts, but there are a number of exceptions: (85)

X-3-X ï INITIAL

X-'-VXC : FINAL

hottentot

intellect

parlement

interim

interval

anjelier

appelsien

kakkerlak

rammenas

arsenaal

atletiek

ukkepuk

zeppelin

harlekijn

hermelijn

algebra

selderie

kanselier

kastelein

aceton

bruidegom

porselein

schorseneer

krakeling

oelewap

element

luitenant

ulevel

zwezerik

souvenir

pierement

Charleston

camera

cholera

anemoon

generaal

cinema

kaketoe

molecuul

papegaai

EXCEPTIONS X-a-VXC : INITIAL

X-a-X s FINAL banderol

bataljon

asterisk

habbekrats

carillon

envelop

interest

bellefleur

jarretel

musketon

elzevier

lessenaar

atelier

parvenu

marsepein

ooievaar

hagedis

kotelet

omelet

paperas

energie We see that the number of exceptions is rather high and one might want to question the significance of the generalization that I make here. It cannot be denied, however, that the stress rule that has been established above, makes the correct prediction, assuming, as I have done all along, that the dominant patterns must be treated as regular. As long as the alternative is to treat all words of the type considered here as having stress, I see no reason to abandon the present

unpredictable

analysis.

In the next section I will return to the prosodic characteristics of

235

syllables with a schwa, but first I will discuss the stress pattern of words consisting

of

four

syllables.

Our

stress

rule makes

the

following

predictions s (87)

1. no initial stress 2. final stress if the final syllable is superheavy 3. antepenultimate stress if the final syllable is heavier than the penultimate syllable (VV-VC# a n d a - V X # ) 4. penultimate stress in the remaining cases

As for the first prediction it can be said that the number of words with initial stress is limited to two types of cases. First there are a few '

^

r

grammatical terms like infinitief, accusatief, nomi.nat.ief and secondly there are place names like Scheveningen, Amerongen with schwa in the second and fourth syllable. Both categories must be marked as exceptions. Proceeding with the second prediction, I have not found any exception: (88) X-X-X-WC/VCC : FINAL affiniteit

alexandrijn anakoloet

aperitief

apocalyps

capaciteit

cinemascoop coefficient deodorant

egelantier

experiment

idiolect

ingredient

initiaal

karikatuur

literatuur

locomotief

materiaal

meridiaan

prerogatief

rozemarijn

salamandrijn

The third prediction is borne out, although there are exceptions. Let us first consider the regular cases. Clearly if the third syllable has a schwa and the fourth a full vowel we cannot construct a foot over the last two syllables because this would result in a mismatch: (89)

ri

no

ce

ros

There are not many examples of this type:

236

(90)

rinoceros epitheton asyndeton

More frequent is the type where the final syllable is closed and the penultimate open. In those cases it is also impossible to combine the final sequence W - V C into one foot, because of the mismatch condition. Such words have the same structure as the one assigned to rinoceros: (91) X - X - W - V C : ANTEPENULTIMATE accordéon

acrostichon adagium

adverbium

alluvium

ammonium

anonymus

aquarium

arsenicum

basilicum

catalogus

colloquium

compendium

compositum criterium

decennium

delirium

emeritus

geranium

gymnasium

harmonium

hospitium

imperium

jeruzalem

magnesium

millennium

napoleon

plutonium

politicus

Stipendium

symposion But there are quite a few exceptions: (92) X-X-W-VC : PREFINAL alligator

archivaris

curiosum

indicator

irrealis

jubileum

mausoleum

navigator

oerangoetan radiator

relativum

In addition there are also words having antepenultimate stress that do not conform to the generalization that in such cases the prefinal syllable has less weight than the final syllable: (93) X-X-X-X : ANTEPENULTIMATE acasia

adagio

alinea

ammonia

apocope

azalea

batavia

begonia

deposito

facsimile

forsythia

gardenia

grammatika

harmonika

magnolia

majolica

malaria

penelope

petunia

scenario

utopia We should not fail to observe, however, that nearly all of them show a prefinal , a vowel that, as we have seen before, usually rejects stress. A small but interesting group involves words that have a schwa in the second syllable and a final sequence VV-VC#. Above we saw that such words

237

get antepenultimate stress if the second syllable has a full vowel. Our rule predicts the following structure for such words:

X

a

X

X

And indeed words of this type have penultimate stress: (95)

difteritis atheneum dromedaris elevator pikketanis

For the remaining group of words ending in two heavy syllables our stress rule predicts penultimate stress. (96) X-X-X-X : PREFINAL academie axioma

caballero

allegretto

alliantie catharina

apriori audientie desperado diabolo

diafragma evangelie

domicilie exercitie

eldorado februari

esperanto eucalyptus hacienda harakiri

januari niagara

leontina macaroni oppositie palestina pianola piccalilli

influenza

intermezzo

maharadja panorama

monopolie paradigma

referendum

repercussie rhododendron seminarie theorema

paranoia

The following words have exceptional final stress: (97) X-X-X-X : FINAL allegorie

ammoniak

anatomie

cabriolet

categorie

cavalerie

ceremonie

epidemie

epilepsie

eucharistie

filosofie

hegemonie

individu

marionet

ontogenie

portemonnee

toreador

varieté

238

Note that some of them end in , some in and some in . We have seen before that these endings receive "irregular stress". The final group of words with four syllables to be considered are those with a schwa in the final syllable. Here the rule predicts penultimate stress and there are, as far as I know, no exceptions: (98) X-X-X-3: PREFINAL accolade balustrade chocolade grenadine

amazone barricade

ambassade

ambulance

camouflage

catastrofe cellulose

anecdote

discipline entourage helikopter heroine

episode

garderobe

hypothese

karbonade marmelade

limonade

limousine

logaritme

margarine

mayonaise

nicotine

pantomime

periode

piramide

pirouette

polonaise

polyester

privilege

procedure

salamander

serenade

serpentine

synagoge

tamarinde

terpentine

vacature

vaseline

vestibule

vitamine

This completes our overview of the facts concerning monomorphemic nouns. An analysis has been presented that makes the correct predictions with respect to the occurrence of dominant patterns. It has been pointed out that many of the exceptions fall under a relatively small set of subregularities that all refer to the segmental make-up of the final syllable, and, in one case (involving ), to the segmental make-up of the prefinal syllable. The reader may have noted that I have not discussed bi- or trisyllabic words having superheavy syllables in non-final position. One can think of many logically possible

combinations of heavy and superheavy syllables,

and make interesting predictions with regard to the stress pattern of such words. Unfortunately, the number of monomorphemic words having superheavy syllables in non-final position is too small to test the predictions that follow from the analysis that was proposed here.

5.3.1.5. Syllables with a schwa and an alternative analysis In this section I will start by pointing out that the previous analysis is inadequate in one respect and that precisely on this point a more adequate analysis is conceivable. I will investigate the consequences of adopting the alternative analysis, taking into account what was proposed in the previous chapter, and show that it compares favorably to the analysis that has been presented first. There are two facts involving the schwa that the preceding analysis does

239

not explain. Firstly, schwas can never bear main stress. There are no exceptions to this generalization. This is in sharp contrast to the influence of all other syllable types on stress placement, where we find some or many exceptions in each case, as we have seen in the previous sections. Secondly, because schwas are absolutely unstressable and the foot type used in the metrical analysis is bounded, we cannot assign prosodic structure in those cases where two schwas occur in a sequence: (99)

F evenement ->

evenement

-?->

....

It would seem that we must add a constraint to the preceding analysis (as in 100a), saying that a schwa may not appear in a strong or only syllable of a foot, and, in addition, a special foot type as in (100b) , that is assigned to a word if two schwas occur in a rows (100) a. * ( ... a, ,...)„ j (s) F

b

-

F

m s X

I 8

w a

w a

In van der Hulst and Moortgat (1981) an analysis of Dutch stress was proposed that was designed to capture these two facts more straightforwardly.

The type of foot that is adopted in van der Hulst and Moortgat (henceforth HM) is

quantity determined, left dominant and unbounded. The foot type is

called "determined", because the head of the foot is required to dominate a full vowel. This property, and the property of unboundedness, was invoked to explain the fact that syllables with a schwa can never bear stress. Identifying

the

notion

"head

of

foot"

with

"stress"

we

explain

straightforwardly that syllables with schwa may never be stressed since the type of foot that is chosen requires schwas to occur in W positions of the foot and a full vowel in

S position:

240 (101)

F

s w ... w As

a

Stated in full the stress rule on which this second analysis is based looks as follows. Differences with respect to the preceding analysis have been marked w i t h a star:

(102)

Dutch stress a. Assign feet that

are

*-unbounded •-quantity-determined -labelled

SW

b . Assign a word tree that is -right

branching

-labelled by the LCPR c. LCPR In the configuration

[AB] B is Strong

i.

it branches

ii.

it dominates a superheavy

syllable

iii. it dominates a marked syllable A minor point is that to assign unbounded feet

iff

([+F])

it is unnecessary to specify

a direction. Given this alternative analysis, Dutch no longer

presents

evidence for the "fourth foot type" . What we have used here is the unbounded counterpart

of

the

second

type

in

(16).

Another

consequence

of

this

proposal is that the analysis can quite easily be extended to the derived vocabulary since suffixes (both derivational and inflectional) with schwas are numerous in Dutch. In (103) I give some examples of both derived and simple words. In giving these examples, I introduce another difference with respect to the preceding analysis. In line w i t h the proposals advanced in the previous chapter, I will no longer assume that metrical structure is binary

branching:

241

(103)

a.

A

s

b.

w

A

S

c.

WW

d.

s w w

^TV

w

s

w w

w w

handel

wandelen

adelijke

makkelijkere

slapen

diepere

eigenlijke

stekeligere

onder

veertiger

negentiger

lelijk

bruinige

noordelijken

w s w

begeleiden

77K

w s w w

WW

w

bezoeken

behandelen

gezamelijke

verdenken

belendende

verwereldlijken

geleden

gezapige

vermoedelijke

IA

w w s w

W S

k.

W

W S

WW

W

vergemakkelijken

W

S

W W

W W

verschrikkelijkere

begenadigende In those cases where more than two schwas occur in a row it is possible to apply several rules that delete certain schwas. For example gezameli jke can easily be pronounced as gezam01i. jke. In the case of versclirikkeli jkere we find an alternative form verschrikk01ijker0. I have not investigated rules of schwa deletion systematically, but the facts just mentioned suggest that a foot comprising three syllables, of which there are two with a schwa, constitutes an upper limit. Apart from suffixes with a schwa, Dutch has many suffixes with the syllabic structure VVC or VCC. It is clear that words that are derived with these suffixes are equally well accounted for in both this and the preceding analysis. It is not my intention to discuss the stress pattern of complex words in detail here, so I will not discuss these cases any further. The second and the third row of examples in (103) show schwas that precede the main stress. Following Rischel (1982), I assumed in chapter 4

242

that such syllables are adjoined to the stress foot by means of a stray syllable convention. I also argued in chapter 4 that, instead of the structure that I proposed for these words in (103), it is possible to say that syllables that precede the stress foot are grouped into a second foot that is adjoined to the stress foot by means of the convention just mentioned (see 80 and 81 in chapter 4): (104)

A A

s w s

S

w

W S

W W

W

vergemakkelijken

begeleiden

s

w w

w w

verschrikkelijkere

begenadigende

Structures of this type, I argued, might be used to characterize an antipole stress, i.e. a secondary stress which is located on the syllable that is furthest away from the syllable that has main stress. It is true that syllables with schwas in initial position, separated by one syllable from the syllable that has main stress, are relatively prominent. In terms of rhythm, I feel no difference between begeleiden (with an initial schwa) and fenomenen (with full vowels in every syllable, except the last). On the other hand, it is traditionally claimed that syllables with a schwa are not suitable locations to anchor a pitch accent (ignoring emphatic speech) . If this is so, it would be undesirable that a syllable with a schwa is the head of the antipole stress foot. When the syllables that precede the main stressed syllable contain full vowels, however, there is no objection against grouping these syllables into a second foot, of which the leftmost daughter is labelled S. Let us now investigate the further consequences of adopting the HM analysis. The choice of a quantity determined unbounded foot entails that each full vowel requires a separate foot. This also leads to a significant difference between this analysis and the preceding one. Consider the following examples: (105)

a.

F F Is |w | | bisam

b.

c.

F S

/\ W

bezem

d,

F F Aw . s S W I kapelaan

kapitein

243

In the previous analysis each pair will be

assigned the same prosodic

structure (i.e. that of bezem and kapelaan). The following examples seem to suggest that the LCPR does relevant work in this analysis too: (106)

a.

b.

A A

F F . w

c.

/a

FFF

,s

W

ballade

W

S

kapitein

d.

/A

FFF , W

W

/A

FFF

S

, W,

orchidee

[+F]

S

|W

Odessa

With respect to bisyllabic words we predict in all cases the structure that was assigned to bisam (in 105a), i.e. two monosyllabic feet, labelled SW by the LCPR. A tacid assumption on which this alternative analysis is based, is the following. To account for words like Odessa, with penultimate stress, foot assignment must be iterative, i.e. it is crucial that, after assigning a foot to the final syllable, preceding syllables with a full vowel are assigned a foot as well. This implies that the alternative analysis does not conform to the theory proposed in chapter 4, because I argued there that foot assignment need not be an iterative rule in that theory. However, if we apply

foot

assignment

non-iteratively,

we

arrive

at

the

following

structure for a word like Odessa: (107)

/\F

F

Aw s w

1,

s

Odessa

I.e. the stress rule, if applied non-iteratively, assigns a stress foot to the final syllable and the preceding syllables are grouped into an antipole stress foot, according to the procedure discussed above (and in chapter 4 in more detail). I will now suggest a revision of the HM analysis that will allow me to maintain the claim that stress rules apply non-iteratively. Let us recall that the type of foot employed by HM differs in two ways from the type of foot that was employed in the first analysis:

244

(108)

1st analysis

2nd analysis

S W X X

S ^

W a

The first difference involves the number of syllables that may be grouped into one foot and the second difference involves the type of syllable that is allowed to occur in weak position, i.e. in the HM type of foot the syllables that occupy the weak position must be light (i.e. contain a schwa), whereas in the other foot type both light and heavy syllables are allowed to occur in weak position. It is possible then to distinguish still another foot type, i.e. one that differs in one respect only from the foot type used in the first analysis: (110)

F S VI ... A X

W

X

I.e. a foot type that is unbounded and that allows only heavy syllables in strong position and both heavy and light syllables in weak position. With the analysis proposed in the previous section, this revised HM proposal shares the fact that we adopt a foot type that does not occur in the standard inventory. Without further adjustments, this foot type is not going to be useful, however, because it predicts that stress will fall on the first syllable of the word that contains a full vowel: (111)

F

F

s w w

A

sw

A

Odessa

bisam

F

F

A

s w w w

s w w

anakoloet

orakel

F

A\

s w w w piramide

There is a very simple way to formulate a condition that prevents too many syllables with full vowels to be grouped into one foot. Suppose we take the idea of "syllable weight" seriously and say that each prosodic category is assigned an index that specifies its weight. We might say that the weight index can be derived compositionally from the number of morae that a syllable contains:

245

[1]

[2]

Assuming the mora theory, the scaling of syllables can thus be derived and need not be stipulated by means of an independent weight scale. Let us furthermore assume that the compositional computation of weight values is carried through above the syllable level, which means that feet will be assigned values as well. It is now possible to say that feet have an upper weight limit, being [4] . In this way we "lose" the possibility to make full use of the fact that the foot is unbounded. There can be at most two schwas following a full vowel (2+1+1=4), but this may not be a bad result because, as I showed above, sequences

of more than two schwas are

subject to various

deletion

processes. We only allow the following feet: (113)

a.

F

F

I

A

F

A

VC

VC a

VC a s

VV

VV 3

W

VVC

VVC a

VCC

VCC 3

3

3

A

A

VC VC W

VV

VC VV W

VC

In (114) it is demonstrated what structures we now assign to the various classes of words:

246

(114)

/-x F F w s [2] [3]

A F F w s [4] [3]

I A

A I

[2][2][1]

[2][2][3]

ballade

kapitein

F F w s [2] [4]

I F [4]

/A

A

[2][2][2]

[2][2]

Odessa

bisam

The final two syllables of ballade can be combined into one foot, but it would not be possible to include the first syllable as well. In the case of kapitein, ending in a superheavy syllable, the foot can only comprise the final syllable. In the case of Odessa and bisam it is possible to combine the two final syllables into one foot, and this is where the difference with respect to the HM analysis is revealed. Making use of weight indices and the notion of an upper weight limit for feet allows me to retain the idea that foot-assignment is non-iterative. In addition, weight indices have another advantage. In section 5.3.1.3. I promised to come back to the (unexplained) fact that superheavy syllables (i.e. VXC) behave just like branching feet. In that section I rejected the "easy way out" of simply labelling such syllables with a node label F. Observe now that within the present analysis the equivalence of superheavy syllables and branching

feet has been explained without

questionable

assumptions about constituent structure. A superheavy syllable acts like a foot, because its weight index is almost the same as that of a maximal foot. With respect to words like orchidee, with exceptional final stress, I will assume that the final syllable is lexically assigned a weight index [3], thus accounting for the fact that these syllables cannot be combined into a foot with the preceding syllable: (115)

c

F F w s [4] [3]

A I

[2][2][3] orchidee [3]

Alternatively, we might say that such words are stored in the lexicon with a final monosyllabic foot. Another consequence of the present analysis is that the LCPR becomes virtually superfluous. In the original HM analysis, this labelling rule is

247

crucially invoked to handle words like Odessa,which are assigned three monosyllabic feet, of which the final two must be labelled SW (see 116a) . In the present analysis this is no longer necessary, since the final two syllables are grouped into a single foot, labelled SW (cf. 116b)s (116)

a. F F w .s

S

A

F F ^F IW 1S Iw

S W odessa

Odessa

It is not the case, however, that the analysis that was presented in section 5.3.1.3. and the HM analysis, both in its original and in its revised form, have the same descriptive coverage. A notable difference between the two analyses involves the W / V C distinction. The reader will have noticed that this distinction is ignored in the analyses that have been presented in this section. Within the original HM analysis, this follows from the fact that each syllable with a full vowel constitutes a foot. It is therefore not possible to arrive at a mismatch (i.e. W - V C ) , because we cannot combine a W

syllable and a VC syllable into one foot in the first place, even in the

order VC-VV. A direct consequence of this difference is that a certain class of words having antepenultimate stress moves from the class of "regular" cases to the class of

"exceptions". Recall that the first analysis

correctly predicts that we find initial stress in trisyllabic words ending in the sequence W - V C # . In the original HM analysis we can only account for antepenultimate stress by saying that the final syllable is extrametrical: (117)

First analysis

F

Second analysis

F Iw

s'

W

al

maa

nak

F F IS A w

s w

al

maa

(nak)

->

almanak

To capture the fact that a certain generalization can be made with respect to words ending in the sequence W - V C , we might formulate a redundancy rule saying that a final VC is extrametrical if it is preceded by a W

syllable.

This move, however, involves the claim that extrametricality rules can be context-sensitive, and thus increases the power of this device considerably.

248

In the revised HM analysis two syllables with a full vowel can be combined into one foot, but since both VV and VC syllables have the same weight index, we do not block the possibility of a foot in which a W and a VC syllable cooccurs (118)

W

F

F

[4]

[4]

A

VC

VC VV

Let us investigate the "damage" by making precise here what was gained by treating VV and VC as being of different weight. The class of cases with antepenultimate stress was split up into a class of real exceptions (for which we invoke extrametricality) and a class of words that is "revealed" as regular. However the generalization that words ending in W - V C # receive antepenultimate stress faces a class of exceptions containing words of this type that bear penultimate stress. The number of exceptions of this type must be subtracted from the number of words that have moved into the class of regular cases. Performing the required calculation leads to the conclusion that the first analysis has slightly fewer exceptions than the second analysis. All things being equal one might therefore hold the opinion that there must be a weak preference for acknowledging the W / V C distinction and hence for the first analysis. The price we pay if we neglect the W / V C distinction is that words of the type almanak with initial stress must be stored with an extrametrical final syllable. 1 mention here a general consideration which suggests that we gain a more constrained theory of stress assignment, if we are prepared to neglect the W / V C distinction. In chapter 4 we discussed all types of stress systems that are known to exist. Two large classes were distinguished. In one type of system main stress could be located anywhere in the word (the unbounded type), whereas in the other type of system main stress was assigned to one of the two or three peripheral syllables. The first analysis that was offered suggests that the three peripheral syllables are relevant, but there is a wrinkle. In the cases where stress is on one of the three peripheral syllables (e.g. the Latin system) it is always the case that the final syllable is "invisible". This explains why its composition is of no relevance. In the analysis, presented in section 5.3.1.3. however, we crucially must know how the final syllable is composed. This suggests that this analysis is on the wrong track. A second consideration that is in favor of the alternative analysis, in

249

which we ignore the VV/VC distinction, is that the LCPR is no longer necessary. The labelling of word trees is always WS, and this, as I argued above, follows from the fact that syllables preceding the stress foot, whether or not they are combined into a second foot themselves, are adjoined to the stress foot by means of a general convention. A third advantage of the alternative analysis involves the use of a weight scale. A positive consequence of ignoring the W / V C distinction is that the use of a weight scale becomes unnecessary. The weight differences of light, heavy and superheavy syllables can be derived compositionally from their internal make-up, i.e. from the number of morae that they contain. In this respect the alternative analysis is more economical since no use is made of an ad hoc device (a weight scale), invented to explain the facts of Dutch stress. On the basis of these considerations I believe that the revised HM analysis represents an improvement of the analysis that was presented in the previous section. The former analysis gives a more principled account of the prosodic behavior of syllables with a schwa and makes no use of an independently stated weight scale. All this compensates for the fact that this analysis misses what seems to be a rather marginal generalization.

5.3.2. Extension of the analysis to adjectives and verbs The stress pattern of adjectives and verbs is not different from what we have found for nouns, although verbs are not very interesting because Dutch has only a handful of underived verbs with more than one full vowel. Adjectives that have more than one full vowel end in practically all cases in a superheavy syllable. These adjectives have final stress. The few remaining cases have final stress as well. Consider the following data:

abrupt

absurd

acuut

affreus banaal

apart bankroet

apert brutaal

clandestien

compleet

copieus

curieus

decent

delicieus

direct egaal

divers enorm

dociel exact

extern

extreem

fameus

feodaal

feudaal

fideel

250

frivool genereus

funest immuun

infaam

intens

futiel inert intern

intiem

jaloers

joviaal

kaduuk

kardinaal

kordaat

labiel

lateraal

legaal

lucratief melaats

markant miniem

massief minuscuul

mobiel

modern

naief

obsceen

obscuur

opaak

oraal

ovaal

paraat

parmant

permanent

pervers

pikant

pittoresk

pompeus

precair

precies

profaan

recent

reciprok

riant

robuust

scabreus

secuur

seniel

sereen

serieus

sonoor

spontaan

steriel subtiel

subiet uniek

stabiel subliem

viriel

virtueel

vicieus

Adjectives ending in a sequence full vowel-schwa also conform to the analysis. They have penultimate stress: (120)

formidabel

illuster

integer

macaber

liquide

luguber

morbide

sinister

I know of only three adjectives ending in VC (all having final stress) and no examples ending in VV: (121)

bizar

kapot

koket

There is no objection then to saying that the analysis motivated on the basis of a corpus of underived nouns is valid for the whole underived vocabulary.

251

5.4. Conclusions In this chapter I have compared several analyses of Dutch stress, and finally argued in favor of an analysis that conforms to the proposals that were advanced in chapter 4. It has been shown that stress placement is sensitive to syllable structure, and in particular to the number of morae that a syllable may contain. The particular proposals with respect to syllable structure in Dutch that were advanced in chapter 3 are also supported then by the analysis of stress in Dutch monomorphemic words. The present analysis is incomplete in the sense that I have said little about the placement of stress in the derived vocabulary. This is of course an important issue but I believe a proper treatment requires a separate monograph.

References

Abercrombie, D. (1965), Studies in phonetics and linguistics. London, Oxford University Press. Abercrombie, D. (1967), Elements Edinburgh University Press.

of

general

phonetics.

Edinburgh,

Anderson, S. (1974), The organization of phonology. New York, Academic Press. Anderson, S. (1976), Nasal consonants and the internal structure of segments. Language 52, 326-344. Anderson, S. (1982), Differences in rule type and their structural basis. In SPR I, 1-25. Anderson, S. (1983), A metrical interpretation of some traditional claims about quantity and stress. Manuscript. Awedyk, W. (1975) , The syllable theory and old English phonology. Wroclaw. Bakker, D. (1968), Constant en variabel. De fonematische structuur van de Nederlandse woordvorm. Asten. Bendor-Samuel, J. (1960), Some problems of segmentation in the phonological analysis of Terena. Word 16, 348-355. Bolinger, D. (1981), Two kinds of vowels, two kinds of rhythm. Indiana University Linguistics Club. Booij, G. (1977), Dutch morphology. Dordrecht, Foris. Booij, G. (1984), Syllabestructuur Nederlands. GLOT 7,2.

en

verkleinwoordsvorming

in het

Borowski, T., J. Ito and R. Mester (1983), The formal representation of ambisyllabicityr evidence from Danish. Manuscript. [Paper presented at the NELS conference; to appear in the proceedings]. Bruce, G. (1983), On rhythmic accentuation. Working papers 25, Linguistics—phonetics, Lund university, 35-52. Chomsky, N. (1955), The logical structure of linguistic theory. [Published in 1975, New York, Plenum] Chomsky, N. and M. Halle (1968), The sound pattern of English. New York, Harper and Row. Clements, G.N. (1976) , Vowel harmony in nonlinear generative phonology: an autosegmental model, [distributed by Indiana University Linguistics Club in 1980].

254

Clements, G.N. (1981), Akan vowel harmony: a non-linear analysis. In G.N. Clements (ed.). Harvard Studies in Phonology II, 108-177. Clements, G.N. (1984), Vowel harmony in Akan: A consideration of Stewart's word structure conditions. Ms. Clements, G.N. and E. Sezer (1982), Vowel and consonant disharmony in Turkish. In SPR II, 213-255. Clements, G.N. and K. Ford (1979), Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry 10, 179-210. Clements, G.N. and S.J. Keyser (1983), CV phonology. A generative theory of the syllable. Cambridge, MIT Press. Cohen, A. (1958), Het Nederlands diminutiefsuffix: een morfofonologische proeve. De Hieuwe Taalgids 51, 40-45. Cohen, A. et al (1959), Fonologie van het Nederlands en het Fries. DenHaag, Nijhoff. Conteh, P. et al (1983),A reanalysis of tone in Mende. In J. Kaye et al. (eds.). Current approaches to African linguistics (vol.2). Dordrecht, Foris, 127-137. Crothers, J. and M. Shibatani (1980), Issues in the description of Turkish vowel harmony. In Vago (1980)(éd.), 63-88. Davis, S. (1982), On the internal organization of syllable constituents. Coyote Working Papers in Linguistics 3, 3-22. Dell, F. (1982), L'accentuation dans les phrases en Français. Manuscript CNRS Paris. Den Os, E. (1982), Italiaanse klemtoon en extrametricaliteit. Doktoraal scriptie, Universiteit van Amsterdam. Dijkstra, M. (1982), De basiskenmerken van een ' stress scale ' . TABU 12, 4954. Dresher, B. (1978), Old English and the theory of phonology. University of Massachusetts/Amherst.

Dissertation

Dwyer, D. (1978), What sort of tone language is Mende. Studies in African Linguistics 9, 167-208. Edmunson, T and J. Bendor-Samuel (1966), Tone pattern of Etung. Journal of African Languages 5, 1-6. Elimelech, B. (1976), A tonal grammar of Etsako. UCLA Working Papers in Phonetics 35. Ewen, C. (1982), The internal structure of complex segments. In SPR II, 2767. Farmer, A.L. (1979), Phonological markedness and the sonority hierarchy. In Safir, K. (1979), 172-177. Fudge, E. (1969), Syllables. Journal of Linguistics 5, 253-287.

255

Garde, P. (1968), L'accent. Paris. Giegerich, H. (1983), Metrical structure. German and English. Dissertation, University of Edinburgh. Giljiamse, H. (1982), Morfologische theorievorming. In: R. Stuip and W. Zwaneburg (eds.), Handelingen van het 37e Nederlands filologencongres. Amsterdam, 179-183. Goldsmith, J. (1976), Linguistics Club.

Autosegmental

phonology.

Indiana

University

Greenberg, J. and D. Kaschube (1978), Word prosodic systems. A preliminary report. Working Papers on Language Dniversals 20, 1-18. Hall, B. et al. (1974), African vowel harmony systems from the vantage point of Kalenjin. Afrika und Ubersee 57, 241-267. Halle, M. (1982), Metrical trees and metrical grids. [Text of lecture presented at GLOW, Paris]. Halle, M. and G.N. Clements, Problem book in phonology. Cambridge, MIT Press. Halle, M. and J.R. Vergnaud (1978), Metrical structures in phonology. A fragment of a draft. Manuscript. Halle, M. and J.R. Vergnaud (1980a), Metrical Phonology. Glow Newsletter 4, 33-35. Halle, M. and J.R. Vergnaud (1980b), Three dimensional phonology. Journal of Linguistic Research 1, 83-105. Halle, M. and J.R. Vergnaud (1981), Harmony processes. In: W. Klein and W. Levelt (eds.). Crossing the boundaries in linguistics. Dordrecht, Reidel, 1-23. Halle, M. and J.R. Vergnaud (1982), On the framework of autosegmental phonology. In SPR I, 65-82. Hankamer, J. and J. Aissen (1974), The sonority hierarchy. In A. Bruck, R.A. Fox and M.W. La Galy (eds.). Papers from the parasession on natural phonology. Chicago Linguistic society, Chigaco, 131-145 Haraguchi, S. (1977), The tone pattern of Japanese: an autosegmental theory of tonology. Tokyo, Kaitakusha. Harris, J. (1983), Syllable structure and stress in Spanish: a nonlinear analysis. Cambridge, MIT Press. Hart, G.W. (1981), Nasality and the organization phonology. Indiana University Linguistics Club.

of

autosegmental

Hayes, B. (1981), A metrical theory of stress rules. Indiana University Linguistics Club. Hayes, B. (1982a), Extrametricality and English stress. Linguistic Inquiry 13, 227-276. Hayes, B. (1982b), Metrical structure as the organizing principle of Yidin y phonology. In SPR I, 97-110.

256 Hayes, B. (1984), The phonology of rhythm in English. Linguistic Inquiry 15. Hooper, J. (1976), Introduction to natural generative phonology. New York, Academic Press. Hülst, H.G. van der (1980), On the formulation of phonological rules . In: S . Daalder and M. Gerritsen, Linguistics in the Netherlands 1980. Amsterdam, 1-10. Hülst, H.G. van der (1981), A lexical-prosodic approach toward stress in Dutch. Manuscript. [Paper presented at the annual meeting of the Dutch Linguistic society 1981]. Hülst, H. G. van der (1984), Vowel harmony in Hungarian: a comparison of segmental and autosegmental treatments. In H. van der Hülst and N. Smith (eds.), Advances in Nonlinear Phonology. Dordrecht, Foris. [Forthcoming] Hülst, H.G. van der (forthc.), The autosegmental analysis of pitch accent: the case of Japanese. Ms. Hülst, H.G. van der and S.J. Langeweg (forthc.), Nederlandse klemtoon. Hülst, H.G. van der and M. Moortgat (1981), Prosodische fonologie en de accentuatie van Nederlandse woorden of: leeft het Nederlandse op grote voet? Verslag van de 150e vergadering van de Nederlandse vereniging van fonetische wetenschappen, 1-25. Hülst, H.G. van der and N. Smith (1982a), An overview of autosegmental and metrical phonology. In SPR I, 1-45. Hülst, H.G. van der and N. Smith (1982b), Prosodic domains and opaque segments in autosegmental phonology. In SPR II, 311-336. Hülst, H.G. van der and N. Smith (1982) (eds.), The structure of phonological representations. PART I & PART II. Dordrecht, Foris. [SPR]. Hülst, H.G. van der and N. Smith (1984) (eds.), Advances in nonlinear phonology. Dordrecht, Foris. Hyman, L. (1977), On the nature of linguistic stress. In L. Hyman (ed.), Studies in stress and accent. Southern Occasional Papers in Linguistics 4, 37-82. Hyman, L. (1982), Globality and the accentual analysis of Luganda tone. Journal of Linguistic Research 2, 1-40. Hyman, L. (1983), A theory of phonological weight. Manuscript, University of California/San Diego. Kager, R. and E. Visch (1983), Een metrische analyse van ritmische klemtoonverschijnselen. Doktoraalskriptie, Instituut de Vooys voor Nederlandse taal- en letterkunde. Rijksuniversiteit Utrecht. Kahn, D. (1976), Syllable-based generalizations Indiana University Linguistics Club.

in English phonology.

Kaye, J. (1982), Harmony processes in Vata. In SPR II, 385-452.

257

Kaye, J. and J. R. Vergnaud (1984), Complex segments. [Paper presented at the GLOW conference in Copenhagen.] Kean, M.L. (1975), Dissertation, MIT.

The

theory

of markedness

in generative

grammar.

Kenstowicz, M. (1970), On the notation of vowel length in Lithuanian. Papers in Linguistics 3, 1-43. Kiparsky, P. (1979), Metrical structure is cyclic. Linguistic Inquiry 8, 421-442. Kiparsky, P. (1982a), From cyclic to lexical phonology. In SPR I, 131-175. Kiparsky, P. (1982b), Remarks on the metrical structure of the syllable. In W. Dressier, O. Pfeiffer and J. Rennison (eds.), Phonologica 1980. Innsbruck 1981, 245-256. Kiparsky, P. (1983), Some consequences of lexical phonology. MS, MIT. Kooij, J. (1978), Aksentregels en syllabestruktuur in het Nederlands. De Nieuwe Taalgids 71, 543-548. Kooij, J. (1982), Epenthetische schwa: processen, regels en domeinen. Spektator 11, 315-325. Kooij, J. (forthc.). Stress, rhythm, and morphological structure. In Proceedings of the symposium on Rhythm in Language, Literature, and Music, Bielefeld 1983. Leben, W. (1971), Suprasegmental and segmental representation of tone. Studies in African Linguistics. Supplement 2, 183-200. Leben, W. (1973), Suprasegmental phonology. Indiana University Linguistics Club. Leben, W. (1982), Metrical or autosegmental. In SPR I, 177-190. Levin, J. (1983), Dependent levels of representation: the skeletal tier and syllabic projections. Glow Newsletter 10, 52-54. Liberman, M. (1975), The intonational structure of English. [Reproduced by the Indiana University Linguistics Club in 1978]. Liberman, M. and A. Prince (1977), On Linguistic Inquiry 8, 249-336.

stress and

Lowenstamm, J. (1979), Topics in syllabic University of Massachusetts/Amherst. Lozano, C. (1983), Choosing between analysis: Evidence from Spanish. Ms.

linguistic

phonology.

foot-based

and

rhythm.

Dissertation, syllable-based

Marie, J.van (1980), The stress pattern of Dutch simplex words: A first approximation. In: W. Zonneveld, F. van Coetsem and 0. Robinson (eds.). Studies in Dutch phonology. The Hague, Nijhoff, 79-121. McCarthy, J. (1979), Formal problems in Semitic phonology and morphology. Dissertation, MIT. [Reproduced by the Indiana University Linguistics Club in 1983].

258

McCarthy, J. (1982), Metrical and autosegmental phonology: an overview. Glow Newsletter 7. McCawley, J. (1968), The phonological component of a grammar of Japanese. The Hague, Mouton. McCawley, J. (1977), Accent in Japanese. In L. Hyman (ed.). Studies in Stress and Accent. Southern California Occasional Papers in Linguistics 4, 261-302. McCawley, J. (1978), What is a tone language. In V. Fromkin (ed.) Tone:A linguistic survey. New York, Academic Press, 113-131. Mohanan, K.P. (1981), Lexical phonology. Indiana University Linguistics Club. Moulton, W. G. (1962), The vowels of Dutch: phonetic and distributional classes. Lingua 11, 294-313. Nes, J. van (1982), Een klemtoonpatroon voor het Nederlands. TABU 12, 5567. Nespor, M and I. Vogel (1982) , Prosodic domains of external sandhi rules. In SPR I, 225-255. Newman, P. (1972), Syllable weight as a phonological variable. The nature and function of the contrast between "heavy" and "light" syllables. Studies in African Linguistics 3, 30*1-323. Odden, D. (1980), Associative tone in Shona. Journal of Linguistic Research 1, 37-51. Odden, D. (1984), Stem tone assignment in Shona. In G. N. Clements and J. Goldsmith (eds.), Autosegmental studies in Bantu Tone. Dordrecht, Foris. Pierrehumbert, J. (1980), The intonation. Dissertation, MIT.

phonology

and

phonetics

of

English

Pike, K. and E.V. Pike (1947), Immediate constituents of Mazateco syllables. International Journal of American Linguistics 13, 78-91 Prince, A. (1983), Relating to the grid. Linguistic Inquiry 14, 19-100. Pulleyblank, D. (1982), Tone in lexical phonology. Dissertation, MIT. Ringen, C.O. (1984), Vowel harmony: linear or non-linear. Ms., University of Iowa. Rischel, J. (1972), Compound stress in Danish without a cycle. ARIPUC 6, University of Copenhagen. Rischel, J. (1982), On unit accentuation in Danish and the distinction between deep and surface phonology. ARIPUC 16, 191-239. Also published in Folia Lingüistica XVII (1983). Safir, K. (1979) (ed.). Papers on syllable structure, metrical structure and harmony processes. MIT Working Papers in Linguistics 1. Saporta S. and H. Contreras (1962), A phonological grammar of Spanish. Washington, University of Wash. Press.

259

Selkirk, L. (1978), The French foots on the status of "mute" e. Studies in French Linguistics 1, 141-150. Selkirk, L. (1980a), The role of prosodic categories in English word stress. Linguistic Inquiry 11, 563-605. Selkirk, L. (1980b), Prosodic domains in phonology: Sanskrit revisited. In: M. Aronoff and M.L. Kean (eds.). Juncture. Saratoga, C.A. Anma libri. Selkirk, L. (1982a), The syllable. In SPR II, 337-383. Selkirk, L. (1982b), On the major class features and syllable theory. Manuscript University of Massachusetts, Amherst. Selkirk, L. (1982c), On prosodic structure and its relation to syntactic structure. Indiana University Linguistics Club. Selkirk, L. (1984), Phonology and syntax: the relation between sound and structure. Cambridge, Mass., MIT press. SPE, see Chomsky, N. and M. Halle (1968). SPR, see Hulst, H.G. van der and N. Smith (1982). Steriade, D. (1982), Greek prosodies and the nature of syllabification. Disseration, MIT. Stewart, J. (1983), Akan vowel harmony: the word structure conditions and the floating vowels. Studies in African Linguistics 14. Stowell, T. (1979), Stress systems of the world, unite. In: Safir, K. (1979), 51-76. Strauss, S. (1982), Lexicalist phonology of English and German. Dordrecht, Foris. Trommelen, M. (1983), The syllable in Dutch: with special reference to diminutive formation. Dordrecht, Foris. Underhill, R. (1973), Major class features. [Paper presented at the winter meeting of the LSA, San Diego]. Vago, R. (1980a), Introduction. In R. Vago (ed.). Issues in vowel harmony. Amsterdam, Benjamins. Vago, R. (1980b), A critique of suprasegmental theories of vowel harmony. In R. Vago (ed.), Issues in vowel harmony. Amsterdam, Benjamins. Vago, R. (forthc.). Morpheme-level harmony in a multi—levelled autosegmental framework. Ms. Vennemann, T. (1972), On the theory of syllabic phonology. Linguistische Berichte 18, 1-18. Vogel, I. (1977), The syllable in phonological theory: with special reference to Italian. Dissertation, Stanford University. Vogel, I. and S. Scalise, (1982), Secondary stress in Italian. Lingua 58, 213-242.

260

Voorst, J. Van (1983), A metrical treatment of a-epenthesis in Dutch. Cahiers Linguistiques d'Ottawa, 75-102. Williams, E. (1971), Underlying' tone in Margi and Igbo. [Published in Linguistic Inquiry in 1976]. Woo, N. (1969), Prosody and phonology. Indiana University Linguistics Club. Zonneveld, W. and M. Trommelen (1980). Egg, onion and ouchl On the representation of Dutch diphthongs. In: W. Zonneveld, F. van Coetsem and O. Robinson (eds.), Studies in Dutch phonology. The Hague, Nijhoff, 265-292.