Autosegmental Studies on Pitch Accent 9783110874266, 9783110130324

Citation preview

Autosegmental Studies on Pitch Accent

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 Hülst

Other books in this series: 1

Michael Moortgat, Harry van der Hülst and Teun Hoekstra (eds.) The Scope of Lexical Rules

2

Harry van der Hülst and Norval Smith (eds.) The Structure of Phonological Representations. Part I.

3

Harry van der Hülst and Norval Smith (eds.) The Structure of Phonological Representations. Part II.

4

Gerald Gazdar, Ewan Klein and Geoffrey K. Pullum (eds.) Order, Concord and Constituency

5

W. de Geest and . Putseys (eds.) Sentential Complementation

6

Teun Hoekstra Transitivity. Grammatical Relations in Government-Binding Theory

7

Harry van der Hülst and Norval Smith (eds.) Advances in Nonlinear Phonology

8

Harry van der Hülst Syllable Structure and Stress in Dutch

9

Hans Bennis Gaps and Dummies

10

Ian G. Roberts The Representation of Implicit and Dethematized Subjects

Autosegmental Studies on Pitch Accent Edited by Harry van der Hülst Department of General Linguistics University of Leiden

Norval Smith Department of General Linguistics University of Amsterdam

1988 FORIS PUBLICATIONS Dordrecht - Holland/Providence, RI - U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Distributor for the U.S.A. and Canada: Foris Publications USA, Inc. P.O. Box 5904 Providence RI 02903 U.S.A. Sole distributor for Japan: Toppan Company, Ltd. Shufunotomo Bldg. 1-6, Kanda Surugadai Chiyoda-ku Tokyo 101, Japan

CIP-DATA KONINKLIJKE BIBLIOTHEEK, DEN HAAG Autosegmental Autosegmental studies on pitch accent / ed. by Harry van der Hülst, Norval Smith. - Dordrecht [etc.]: Foris. - (Linguistic models ; 11) With bibliogr., index. ISBN 90-6765-303-9 bound ISBN 90-6765-304-7 paper SISO 805.2 UDC 801.4 Subject heading: pitch accent.

ISBN 90 6765 303 9 (Bound) ISBN 90 6765 304 7 (Paper) © 1988 Foris Publications - Dordrecht No part of this publication may be 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.

Table of Contents

Harry van der Hülst & Norval Smith The Variety of Pitch Accent Systems: Introduction Zrinka Babic Accent Systems in Croatian Dialects

ix

1

Giorgio Banti Two Cushitic Systems: Somali and Oromo Nouns

11

Mary Clark An Accentual Analysis of the Zulu Noun

51

John Goldsmith Prosodic Trends in the Bantu Languages

81

Carlos Gussenhoven Adequacy in Intonation Analysis: The Case of Dutch

95

Shosuke Haraguchi Pitch Accent and Intonation in Japanese

123

Daniel Hirst Tonal Units as Constituents of Prosodic Structure: The Evidence from English and French Intonation 151 Barbara E. Hollenbach The Asymmetrical Distribution of Tone in Copala Trique

167

Andras Kornai & Laszlo Kaiman Hungarian Sentence Intonation

183

Carol C. Mock Pitch Accent and Stress in Isthmus Zapotec

197

David Odden Predictable Tone Systems in Bantu

225

Kay Williamson Tone and Accent in Ijo

253

Meg Withgott & Per-Kristian Halvorsen Phonetic and Phonological Considerations bearing on Representation of East Norwegian Accent

279

Martha Wright A Level-based Model for Pitch Accent Languages

295

Acknowledgements

The idea for this volume grew out of a workshop on Pitch Accent held at the 1984 International Phonology Conference in Eisenstadt, Austria. Some of the articles included here started life as presentations at the workshop. Most of the contributions in this collection were written between 1985 and 1987, although a few date from 1988 in their present form. The editors take full responsibility for the asynchronous nature of the volume. We would like to express our gratitude to John van Lit and Simone Langeweg for correcting the proofs of the manuscripts. Also to the earlier authors for their patience. Lastly, we would like to thank the publishers for their rapid publication of the volume.

Harry van der Hülst Norval Smith

The Variety of Pitch Accent Systems: Introduction Harry van der H lst & Norval Smith University of Leiden & University of Amsterdam

The question of the precise definition of Pitch Accent, and the related question of the difference between a Pitch Accent Language and a Tone Language on the one hand, and a Stress Accent Language on the other, is one that has exercised phonologists for a long time. The first generative phonologist to seriously address the problem however was McCawley (McCawley 1978) . What we will suggest here is that a simple division among the Non-stress cases into two categories (Pitch Accent Language - Tone Language) represents a gross over-simplification of the facts. Rather, it seems to be the case that languages employing non-intonational pitch distinctions make up a continuum that from a theoretical point of view should be approached in terms of a set of parameters that seem to define systems as being more typically "Tone Languages" or more typically "Pitch Accent Languages". Our intention then is not merely to try and provide a definition of "Pitch Accent Language", but rather to discuss the variety of systems that occupy the "space" between pure Tone Languages - if indeed any such exist - and pure Stress Languages, and the parameters that lie behind this variation. There is of course no language that does not at some level employ phonological pitch (i.e. tonal) distinctions. The most important break, we feel, is between those languages that employ phonological pitch distinctions within the Word (including cases where there is spreading of tonal patterns outside the strict boundaries of the word) - whether these are marked in lexical entries, or only introduced via lexical rules - and those languages that ONLY employ phonological pitch distinctions which are spread over larger stretches of the utterance, and which are not typically introduced as parts of the constituent word or stem morphemes. This last use of pitch is usually referred to as ΙΝΤΟΝATIONAL, and is rather typical of languages which have so-called Stress Accent systems, such as Dutch and English. In one approach to intonational systems of this kind, tonal contours are decomposed into a limited set of pitch-movements, which are called pitch accents. These pitch accents are combined and aligned with the "text". The alignment or

Harry van der Hülst

& Norval Smith

association rules do this with reference to stressed/unstressed syllables in focussed constituents, and syntactic and/or prosodic structures. Note then that what intonational pitch accent systems and (one type of) word-internal pitch accent system have in common is that they both involve associations between tonal patterns and accented syllables. In this introduction however we will not be concerned with intonational pitch accent, the study of which within the nonlinear framework was more or less initiated by Pierrehumbert (1980). We refer the reader to the articles dealing with Intonation by Gussenhoven, Hirst and Kornai and Kaiman, as well to a lesser extent to that by Haraguchi. Let us -now proceed by defining the margins of the "space" we have referred to above, between true tone languages and stress accent languages.

1.

TONE AND STRESS LANGUAGES

1.1.

What is a Tone or Stress (accent) Language?

Many definitions of a tone language have been given. We can take the classical definition provided by Pike (1948:3) : "A tone language may be defined as a language having lexically significant, contrastive, but relative pitch on each syllable." There is one obvious problem with P i k e ' s definition. Most (and perhaps all)

tone

languages have at least some syllables (i.e. one or more syllables of certain morphemes) for which tone is NOT contrastive (phonologically distinctive), but is PREDICTABLE from the environment. A rather interesting definition

is provided by Welmers ( 1 9 5 9 : 2 ) :

"A tone languages is a language in which both pitch phonemes and segmental phonemes enter into the composition of at least some morphemes." Clearly this definition is extremely liberal. It includes everything from tone to the most marginal pitch accent system, but excludes stress accent languages such as English and (Standard) Dutch. Let us at this point introduce the concept of "restricted" tone language, employed by Voorhoeve (1973) and Schadeberg ( 1 9 7 3 ) . We might take Pike's definition as referring to cases where tones play much the same role as vowel qualities in languages like Italian, in which every syllable may contain a distinctive vowel - irrespective of the vowels in the other syllables. The vocalic aspect of the individual syllables would be completely "paradigmatic" i.e.

the full paradigm of options is available for every syllable. The analogous

Introduction

xi

tonal case would then be an "unrestricted" tone language making a paradigmatic use of tone. Must such an ideal tone language be restricted to bearing one tone per tone bearing unit (TBU) or may contour tones occur? The fact that more than one tone occurs with a single TBU in some languages may be compared with the occurrence of vocalic diphthongs, which some languages allow, while others do not. Since Welmer's definition draws a line between those languages possessing morphemes with tonal properties, and those languages not availing themselves of this option, it is clear that in his terms we have a dichotomy between tone languages and stress/non-tone languages. Does this dichotomy correspond to the realities of the situation? It has been claimed by some researchers (for example Fry 1955, 1958) that, in so-called stress languages such as English, the most important cue to accent is not in fact amplitude, but a combination of this with duration and intonational pitch, with the last being the most significant as an indicator of accent. However, recent work by Beckman has suggested that the most important phonetic correlate of English "stress" is not in fact pitch at all,

but what

she terms "total amplitude" - a function of amplitude and duration (Beckman 1986). In any event it must be made explicit whether one is speaking about stresses which function as the anchor point of intonational pitch accents, and those which do not. Clearly, the first type will have tonal information associated with them, but this is irrelevant to the definition of stress accent as such insofar as this information is derived from the intonation pattern. We believe that instead of defining STRESS in terms of some physical property or properties, it makes more sense to say that stressed syllables are designated syllables in terms of prosodic constituent structure, which by themselves do not convey any tonal information. This is not to say - as we have just made clear that stress cannot be aligned with tonal information, or for that matter, that stressed syllables cannot have tonal information added to them at the morpheme or word level, as is the case in the Scandinavian languages ( c f . section 2 . 1 . , and the article by Withgott and Halvorsen). As will become clear, only the minimal syllable may be the bearer of stress. Hence we claim that the "mora" is never the domain of stress. This is not to say that "morae" play no role in stress assignment. As is well-known stress assignment may be sensitive to syllable weight, where weight is usually calculated in terms of the number of morae contained in the syllable.

xii

Harry van der H lst

& Norval Smith

Cases which have been claimed to involve a distinction:

*

(m m)

.

σ

=f

*

(m m)

σ

do not then in our opinion represent STRESS distincions but involve rather the direct association of tonal information to either the first or second mora (cf.

the mora-based systems in Croatian discussed by Babic, or the Somali

system dealt with by Banti). Finally let us mention a four-way classification made by Clark (this volume). Note that in this classification what are referred to as pitch accent systems do not fall into a single category, but are split up between classes IB and

IIB. I.

TONAL SYSTEMS

A.

Free Tone

B.

eg. Ewe

Restricted Tone (limited number of tone melodies) incl. Tonal Pitch Accent (one tone melody) eg. Mende/Japanese

II. METRICAL ACCENT

A.

Stress Accent eg. English

B.

Metrical Pitch Accent (realized as H) eg. Ancient Greek

The "metricality" of type IIB systems lies in their recognition of "strong" syllables as the basis for tone association. It is possible for languages to possess two cooccurent systems, eg. free tone and stress accent. And presumably Clark would categorize Western European "pitch accent" systems as a combination of Stress Accent and Metrical Pitch Accent, and Central Carrier (see below) as a combination of Stress Accent and Restricted Tone. We are slightly suspicious of this classification. Our main criticism concerns its nonexhaustive nature. If it is not possible to categorize all languages as belonging to a single category, there is clearly something wrong, since the categories are clearly already combinatory in nature. One qualification would concern category IB) which ought perhaps to be subdivided. There are several possible parameters:

Introduction

xiii

a)

whether the underlying tones spread to cover the whole word

b)

whether the underlying tone ( p a t t e r n ( s ) ) have a fixed position in the word

c)

if the language has a stress accent system in addition; what the consequences of this are for the placement of the tones - i.e. all tonal information on the stressed syllable, tonal information starting from the stressed syllable, or no relationship (note that this subcategorization of (c) already causes problems, since the first two types might be better regarded as combinations of IIA and IIB, while the third type might be IIA and I B ) .

This suggests that rather than a classification into exclusive categories, some kind of feature system might be more adequate. A number of articles in this collection are concerned with developments producing new types of prosodic system (e.g. the articles by Goldsmith, Odden, and Clark, in particular arguing from the facts of Bantu languages). Taking these articles together it appears that one developmental path that is followed starts from Free Tone and proceeds round the four categories mentioned here clockwise, ending up with Stress Accent. In the conclusion a number of such developmental paths will be mentioned briefly.

2.

INTERACTIONS BETWEEN TONE AND STRESS

Of great importance for the consideration of PITCH ACCENT is the set of possible relationships between tone and stress (within the individual lexical item). This may be represented in terms of the following two types:

2.1.

a)

tone attracted by stress/tone rejected by absence of stress

b)

stress attracted by tone

Tone Dependent on Stress

An example of this would be West European languages such as the

Scandinavian

languages (see the article on Norwegian by Withgott and Halvorsen), and Rhenish dialects of Dutch and German (Hermans 1982) where segmental processes such as vowel syncope led to the development of "tones". Here morphemes contain tonal information (either lexical or supplied by default r u l e s ) . This tonal information is restricted to an H (High Tone) or L (Low Tone), which is associated with reference to the stressed syllable. In these cases the stresses play their "usual" role in functioning as potential anchor points for intonational pitch accents. The syllable-based accent systems of Croatian (Babic) appear to fall into this type too. The systems with opposed H and L behave as above, while

xiv

Harry van der H lst

& Norval Smith

there is a second type of system lacking this contrast. Here a redundancy rule associates an H to the stressed syllable. A rather different case is discussed in the article by Hollenbach - that of Trique. Here there is not a total rejection of distinctive tonal information on non-primary stressed syllables - a few such syllables possess tonal information that is phonologically significant.

In fact we can identify among Otomanguean

languages a whole continuum of cases, involving the gradual reduction of the distinctive occurrence of tonal information on non (-primary) stressed syllables, with complete - or near-complete (as in the case of Trique) - rejection as the logical end-point , with a corresponding increase in the complexity in the types of tonal information expressed on the stressed syllable. Note that we abstract away from cases where a reduction in the number of morae in unstressed syllables has the concomitant effect of reducing the number of contour patterns allowed in such syllables. This is an additional factor to be reckoned with. A case where stress only slightly influences the range of tonal options is to be found in Cajonos Zapotec (Nelis and Hollenbach 1980) . Here we have a system of four underlying tones - H, L, HL, and M (Mid) . M however is disallowed in unstressed syllables. Note that despite the fact that stressed vowels are long, and unstressed vowels short, the contour HL does occur in unstressed syllables. As we have said above, a case illustrating the virtual rejection of tonal information on non-primary stressed syllables is that of (Copala) Trique (Hollenbach) . In this language lexical tones normally only occur on stressed (final) syllables, with the occasional occurrence of one tone (pattern) on an unstressed syllable

The set of tonal possibilities is:

a)

stressed syllables:

simple tones: contour tones:

3, 4, 5 21, 32, 53, 34, 35

b)

unstressed syllables:

simple tones: contour tones:

3, 21

(5=High)

4

The comparison of cognates from this dialect and the Chicahuatla dialect (Hollenbach 1977) is highly instructive.

Chicahuatla QJ

_ο ο

ga Ci ga 3 ne 2 3 wa 3 ?a 2 3

Copala ~ο ο

gaCe qaneT32 wa?a 3 2

Gloss to pass to chew to spin

Introduction

ni 3 ti L | 3 a3ne43 ni 4 Ca 3

rete 3 4 ane74 niCa 5 3

vegetable pear to bathe full

mml 31 * 3 wwe

yume_31* yuwe

sweet potato century plant

Here we can observe a rightwards shift of tonal information to the stressed syllable, with the additional limitation that Copala contours are limited to two tones. Another case where we have historical evidence on tone reduction - Mandarin Chinese - exhibits a different

strategy. Tones on unstressed elements are not

shifted towards the stressed syllable, but simply reduced, in a similar fashion to what happens with unstressed vowels. A point that should be emphasized here is the fact that the association of tonal information solely or principally with the stressed syllables of words has as a consequence that it

is not necessary to specify the location of the

tonal information lexically - it is sufficient to include in the lexical representation a statement concerning the tone pattern (or absence of such).

2.2.

Stress dependent on Tone

A good example of a case where the location of the stressed syllable is completely dependent on tonal information is that of Ayutla Mixtec (Pankratz and Pike 1 9 6 7 ) . The usual case in such languages is that H has a particular attraction for stress. In Ayutla Mixtec stress is assigned as follows:

a) b) c) d)

on on on on

the the the the

first first first first

HL sequence if present; if none then, ML sequence if present; if none then, H if present; if none then, syllable

(NB, HM is absent)

Note that here every syllable will possess a tone. Another case illustrating the attraction H tones have for stress would be that of Kimatuumbi (Odden 1984; this volume). A rather d i f f e r e n t type of case would seem to involve the Croatian morabased systems (Babic). Unlike the syllable-based systems these do not allow the determination of the location of the tones on the basis of the stress pattern because here we have an opposition within long stressed syllables of H on the first mora versus H on the second mora. There are basically two types

xvi

Harry van der Hülst

s

Norval Smith

of system: only lexical H; and lexical H versus L. In either case the single tone present - whose location must be lexically specified - will define the position of the stressed syllable. 3.

NO INTERACTION BETWEEN TONE AND STRESS

3.1.

Systems with Stress

Note that there does not require to be any relationship between the location of the stressed syllable and the location of any particular tonal information. Central Carrier (Pike 1986) has consistent word-final stress. However as far as tonal information is concerned there seem to be three types of word: a)

words containing a step down between two syllables: H L mandah

"canvas"

H H L L nanaditso

b)

words that are H; the following item is usually lowered in tone: H y9s

c)

"dew"

(L) (X..)

"wolf"

words that are H; the following item is never lowered: H H hunliz

H (X..)

"skunk"

We may assume that these three types have lexical representations whereby the last H before an L is lexically marked. As we shall see this type resembles the kind of tone pattern encountered in Japanese (section 3 . 2 . and the article by Haraguchi) : H

I

mandah

H

I

nanaditso

H

y9s

hunliz

The surface forms may be simply derived by assuming a rule of leftwards Hspreading, and default L for the for the unmarked items.

-marked items, and a rule of final H-insertion

Introduction

xvi ι

Another case that might at first appear to be a combination of free tone and stress is interpreted by Clark as representing a tonal pitch accent language. Stress placement in Zulu is apparently a matter of some controversy (cf. footnote 16 in Clark's article) but this is not relevant for the content of her article since stress is independent of tonal information. At first sight Zulu might appear to be a free tone language - in nouns (maximally bisyllabic) we have HH, HL, LH and LL. The facts are more complex than this bald statement suggests - Laughren (1984) analyses the tone patterns as respectively HHL, HL, LHL and L. Clark presents evidence for interpreting these oppositions in terms of the following underlying representations: H / \

cvcv 3.2.

H

H

CVCV

cvcv

CVCV

Languages without Stress

Obviously languages lacking lexical stress also have no correlation between (lexical) stress and lexical tone. Such a language is Japanese, where stress appears to be absent altogether (Beckman 1986). The tonal behaviour of most Japanese dialects (Haraquchi 1977; this volume) appears to be classifiable in terms of the following two parameters: a)

location of marked H: fixed/free

b)

spreading of H: none/right/left

(>1 choice possible)

This results at least in the following types: Dialect

a-1 fixed

a-2 free

b-1 none

b-2 right

b-3 left

Old Kyoto Osaka Tokyo Hirosaki Nakamura Shimigawa Sendai In the case of Shimigawa the H obviously does not require lexical specification

xvlii

Harry van der Hülst

& Norval Smith

as its position is forecastable. Sendai does not exhibit any tonal information specific to individual lexical items. Its patterns should perhaps be treated under the heading of intonation. The systems with a free linked H also all

have patterns without any linked

H ( c f . Central Carrier). There are currently two main types of analysis for Japanese. Haraguchi works with a * designating syllables/morae that form the anchor point for particular melodies. This suggests of course a parallel with stress accent. The other main type of analysis employs linked H tones (Archangeli and Pulleyblank 1984; Clark (in p r e s s ) ) . A number of advantages could be listed in favour of the second analysis:

a)

the occurrence of long syllables with differing "mora accents" in some dialects. * * m m versus m m This possibility is rejected by a number of linguists.

b)

the existence of "accentless" patterns - in both syllable systems and mora systems

c)

the existence of Japanese-like systems such as Central Carrier, with independent stress marking

Similar to Japanese are two Cushitic systems discussed by Banti in that one mora may be associated to a tonal pattern. In Somali the tone is H, and basically other morae are L (although a preceding mora within the same syllable may be raised). In Oromo there are two melodies H and HL (as in one possible interpretation of Kagoshima Japanese (Haraguchi 1977; this v o l u m e ) ) . Unlike Kagoshima, the location of these melodies is not restricted to final position, and as in most Japanese dialects we also have spreading. In Oromo the

-tone spreads to the

right. Rather similar to Japanese too are the tonal systems of the Ijo languages (Williamson, this volume). These systems are very complex, forcing Williamson to distinguish between three types of tonal information: a)

linked H ' s

b)

so-called Domain Tones (i.e.

c)

floating (non-spreading) H/L

spreading) H/L

Introduction

xix

To elucidate the distinction between the last two types, these could be represented by floating tones, and initial tone association rules (ITAR's) respectively. To elucidate the distinction between the first two types we compare an analysis utilizing Domain Tones with Haraguchi's analysis of Tokyo Japanese: Haraguchi :

H

H

|

/ ι*

*

kabuto

Domain Tone Analysis:

kokoro

H

// 1

H /

/

1

* kagami

sakura

1

1 I

1 1

1 /

H H

H H

H H

H

(DT)

(DT)

(DT)

(DT)

Note that these representations take no account of the initial lowering that applies to the last three words. In Haraguchi's account the H is first associated to the starred syllable, then spread leftwards; in the DT account there would be a linked H associated to the same syllable, while a Domain (floating in the more usual sense of this term) tone would associate to all syllables to the left. Individual lexical entries may contain all three types of tonal information in some cases in some languages. Another language treated in this book which is claimed not to possess lexical stress is Isthmus Zapotec (Mock). This represents quite a different type of language, however. No tonal information has to be associated with a particular syllable,

rather the tonal information is associated with the

morpheme as a whole. Four tone melodies occur: H, L, HL, LH. Stress as a lexical category is non-existent; it is however employed to indicate focussed elements in the sentence, whereby stress is assigned to the first syllable of such a focussed element. Yet another kind of stressless sytem is encountered in Kinga (Odden). This however is,

as Odden remarks, very stresslike in its distribution of tones.

The two most normal patterns are Pre-Stem-Initial (PSI) and Antepenultimate (APU), corresponding to underlying Η-toned and toneless morphemes (which acquire the APU pattern by default). From Odden's remarks about this system it is clear that we must place it among Clark's Metrical Pitch Accent systems.

xx

Harry van der Hülst

4.

S

Norval Smith

LANGUAGES MAKING MARGINAL USE OF TONE

In this section we wish to mention two languages making a highly marginal use of lexically specified tonal information - Huave and Jeh. 4.1.

Huave

In Central American Huave (Pike and Warkentin 1975) we find four tonal patterns whose occurrence is nearly exclusively determined by the syntactic make-up of the sentence or phrase. In other words tonal information in this languages can be interpreted as intonational, although the burden of "true" intonational contrast is borne, as in many Central American languages, to the last mora or two of the sentence. There is however a very restricted set of items for which lexical tonal information is required. This is only apparent in one of the afore-mentioned four patterns. When a lexical item appears with the part-intonational pattern LHL where the H occupies the first mora of the final syllable, we find a number of items - with a bimoraic final syllable - where the second mora is also H.

Normal Pattern L

Abnormal Pattern

H

olam

"sugar cane" HH

HL

peac

"tortilla"

L

L HL

apiis

tiAk

"dress"

"hill"

HH

ornbiAm

"house"

L HLLL ateaaig

"he breaks'

Note that stress is on the final syllabic in Huave so that this system is reminiscent of Western European pitch accent systems with their opposition between -marked and unmarked morphemes.

4.2.

Jeh

This language is of interest in that its

lets us observe how a restricted tone

system may arise. Jeh is a Mon-Khmer language of Vietnam described in Graden (1966).

Introduction

χχί

Northern Jeh has a limited tonal opposition occurring on the second mora of final open syllables. A number of items have a H tone in this position, e . g . :

Η

tEE

"to scythe"

When we compare the corresponding Southern Jeh form, the source of this Η tone becomes obvious. Southern Jeh has a final /h/

instead of a H tone.

tEh

What we have here is then a former allophonic tonal feature of /h/ which has been phonologized by the loss of the segment /h/. In the Northern Jeh dialect of Dak Trap H tone has a wider distribution, occurring also on some final nasals. This is again an innovation as can be seen by a comparison of these forms with those of other forms of Northern Jeh.

Dak Trap

Other N. Jeh

Gloss

trap

muddy

wak

(boy's name)

H

I

tram H

waN (N = ' n g ' )

This conceivably illustrates the development of a restricted tone language as a step on the way to a free tone language.

5.

CONCLUSION

We would like to conclude by briefly reconsidering the possible developmental routes illustrated in this collection. Taking as a departure point the pure tone language, we can see from the articles dealing with Bantu languages (Goldsmith, Odden, Clark) and also from Wright's contribution, that there are a number of things that can happen - a restriction on the combination of tones, i.e.

the development of a limited

number of tone patterns; the replacement of the opposition H - L by the opposit-

xxii

Harry van der H lst

S

Norval Snr'.th

ion Η - I\gabdha ·.·,·. - ba> ·,.-}

ma

(same meaning as l l . b ) arag ^

girls bä not didn't-see ' I didn't see any girls'

Somali and Oromo Nouns

e.

19

ma _,_ ,, arkeen?

I\gabdho T-JI. 1}

INT they-saw 'did they see some girls?' (yet cf. also example 12.1) The genitive has final

-tone, but for the m. plurals in -o which have the

suffixes -ood or -aäd, whose distribution is governed lexically: gabdho 'girls', mindiyo 'knives' and most other nouns have -odd-, halo 'she-camels', riyo

'goats

and a few other nouns of domestic animals have instead halaäd, riyaäd, etc. ~* The nominative can have final

-tone or be L-toned in all contexts, as shown

in (12.a) and ( 1 2 . b ) . As an exception to what was stated in section 2 . 2 . 1 , D4 nouns can optionally be L-toned when a nominative long possessive is suffixed to them. (But they always have final absolutive case).

(12)

a.

-tone when the long possessive is in the

6

.gabdho

ma arkeen'·' \gabdho' 'did some girls see him?' ( c f . also example l l . e : the sentence is ambiguous) - eey

shäley

yimaadeen

yesterday FOC they came 'some women came yesterday' c.

shäley

heeygu>

lumeen were-lost

-tone, as shown in (13) below:

Fourth declension

Abs igajbdhoi ' gabdhoI

,odayaal\ {lodayaal' , ,} .gaallii laaallil 2.2.2.5

-

'hala - heeygu' she-camels my 'yesterday my she-camels were lost 1

The vocative has initial (13)

halä

b-eey

ha ween, *haeen ' women

The fifth

Gen

Nom

Voc

gaidhood

, gabdho \ ( ,,, } gabdho'

..,, gabdho

, _, odayaal

/odayaal I , ,} Odayaal 1

., , odayaal 3

_ ·,.·,.-, , f. pi. 'elders'

.aaalli\ \igaalli' ·, ·, -I

, ^qaalli

m. sg.

,,, qaaUl

pi. 'girls'

'judge 1

declension. Some m. singular nouns like gabyäa 'poet' and,

optionally, dukaänle 'shopkeeper 1 (which otherwise is a D3 noun), and several feminine sg. and collective nouns like gelley or gelleey 'corn' and, optionally,

20

Giorgio Banti

the Dl feminines like haweeney 'woman', biyooley

'female water-seller' or 'water

sellers' (in general), etc., are inflected according to the fifth

declension.

Its characteristic features are penultimate Η-tone on the absolutive and nominative, ultimate Η-tone in the genitive, and initial Η-tone in the vocative.

(14)

Fifth declension Abs gaby a haweeney

2.3

Gen gabya haweeney

Norn gaby a haweeney

Voc g byaa h veeney

m. sg. f. sg.

'poet' 'woman'

Analysis of the system

2.3.1

The absolutive

As I already stated in section 2 . 1 . , I will maintain here in its broad outlines Hyman's (1981) autosegmental analysis of the Somali prosodic system, which posits a set of accent rules feeding into a tonal component. It is convenient to regard a form similar to the absolutive as the primary form from which the other inflected forms are derived. In the case of D3 and D4 nouns, which have two forms for their absolutive, the more general form - ie. the form with the less specific distribution - will be regarded as primary. In this manner, all classes of nouns lack inflectional suffixes in this form. However, D2, D3 and D4 nouns have ultimate Η-tone, while Dl and D5 nouns have penultimate Η-tone. Monomoraic Dl nouns such as rag 'males' or dhul

'earth' are

a sort of borderline case, because they lack a penultimate mora. The simplest solution, then, is to treat ultimate Η-tone as the general case, which includes also monomoraic Dl nouns, and penultimate Η-tone as the special case that applies only to polymoraic Dl and D5 nouns. Since we already stated that Η-tones correspond to underlying accents (ie. stars) in the framework that is being adopted here, we have the following pattern:

(15)

a.

xaas

(Dl)

* hooyooyim * gabyaa

(Dl) (D5)

b.

rag * hal * gool * socoto

(Dl) (D2) (D2) (D3)

gabdho ( D 4 ) *o We can now state that in (15.a) there is the accent pattern (AP) on the last

two moras, while in (15.b) there is the AP . on the last mora. We introduce in this manner objects called accent patterns, which consist of accents (stars) and

Somali and Oromo Nouns ο empty accent slots ( -signs), and occur on the rightmost moras of nouns. It

21 is

now possible to write the following accent association convention: (16)

Accent Association Convention (AAC): An accent pattern AP is assigned from right to left starting from the last mora, one accentual item to each raora.

Having introduced the AAC, the forms in (15) can be described in the following manner:

(17)

a.

b.

/ X/ * ο / X /,

if Dl polymoraic and D5 noun

Note that (17.a) and (17.b) apply disjunctively, and since (17.b) is the special case, it takes precedence over the more general ( 1 7 . a ) . Some version of Kiparsky's (eg. 1982) Elsewhere Condition can account for this. As we will see later, there are some reasons for regarding the distributional rules (17) as pertaining to the lexical entries, since they may be needed as inputs for some inflectional rules. The absolutive rule will thus have simply to specify that there are no inflectional suffixes, and that in the general case the lexical AP does not change: (18)

Absolutive case / X/

The following two context-sensitive rules are needed, however, in order to account for the special forms of D3 and D4 absolutives in some contexts: (19)

a.

*

-+·

*o /

I

# #

J , if

D3 noun

Ua '

{

baa ι b \ , optionally in D4 nouns (Y)vJ

Rule (19.a) accounts for the behaviour

observed in ( 9 . i i ) , rule (19.b) for

(ll.c-e).7 Let us turn now to the derivation of actually occurring forms, like the following citation forms:

[x as], [hooyo] ^ [hooyo]

(in other contexts the absolutive

22

Giorgio Banti

would be [hooyo] , cf. 2 . 2 . 2 . 3 ) , and [gabdho] . We have already seen that the * * * lexical entries would be xaas, hooyo and, respectively, gabdho. Being citation forms, ie. words used in isolation in the context # #

# #, they require the

absolutive case. Rule (18) applies thus, and the representations don't change. * In the case of the D3 noun # # hooyo # #, however, rule (19.a) applies and the *o AP is changed into . The AAC, ie. ( 1 6 ) , now applies and we remain with the * representation hooyo. In Banti (1984) I claimed that there were tone-accent melodies (TA-melodies) * * such as HL, LH, LL, etc. in Somali. Even if this was descriptively accurate, I think that a simpler solution can be developed by elaborating Hyman's (1981) * original suggestion. Indeed, it is sufficient to posit only the melody H, which will associate to the starred BU according to principle ( 2 0 ) , commonly accepted in autosegmental analyses of accent systems (cf. eg. Clements and Goldsmith, 1984b, p . 1 3 ) :

(20)

*

In addition to principle ( 2 0 ) , however, two more specific operations are needed. Firstly, Η-tones which are associated with a given mora by applying (20) do not spread to contiguous moras, except in the case of the optional process already described in ( 3 ) . Secondly, unassociated moras are assigned L-tones. All this can be expressed by means of the following two rules. Notice that ( 2 1 . a ) , which is a revised version of ( 3 ) , must apply before ( 2 1 . b ) : (21)

a.

A 1

V

b.

V

-*-

V

V , optionally

\l* H

Moras not associated with a Η-tone are assigned L-tones.

The derivation of the surface citation forms [x as], [hooyo] 'v [hooyo], and [gabdho] can now be described in the following manner:

Somali and Oromo Nouns

(22)

# xaas # #

# hooyo # # , *° hooyo , * ο hooyo ^ * ° nooyo

*o xaas *o xaas

I Η

* H ,

ff gabdho # ff (19.a) gabdho

(16)

gabdho

(20)

ΜΗ

* 0

hooyo

Η

*o xaas

, * ο nooyo

HL

23

(21.a) * gabdho

H L

L

(21.b)

H

If rule ( 2 1 . a ) fails to apply, we have the following variant derivation: * ο hooyo '*

(23)

* 0

hooyo

III

LH L

H

2.3.2

The genitive

In accordance with the model which has been developed so far, the genitive forms can be viewed as a specific accentual profile which is imposed on the word. The only exceptions are D4 m. plurals in -ό, which add also the suffixes -ood or -a d, eg., gabdhood and hala d from gabdho 'girls' and, respectively, halo 'shecamels', cf. 2 . 2 . 2 . 4 . The rule is thus the following one: (24)

Genitive case

a.

/ X/

b.

/ X+

*, ood } 'aad

/,

if m. D4 pi. in -o

Before the /ood/ and /aad/ suffixes a final vowel is elided according to a general rule which is needed also in other areas of the morphology of Somali ( c f . section 2 . 2 . 2 . 3 , as well as Banti, 1984; eg. labi Norn of 'two' from /laba + i/, tuklyo lib ax 'a crow and a lion' from /tuke iyo lib ax/, ma abees a? 'is it a snake?" from /ma abeeso

a/, e t c . ) .

In D2, D3 and D4 nouns not included in ( 2 4 . b ) the genitive is identical to the absolutive - but never undergoes either of (19.a) or (19.b) -, whereas in Dl nouns, in D4 nouns with genitives in -ood or -aad, and in D5 nouns it is differ*o * ent. Indeed, the inputs of rules ( 2 4 ) are respectively /hooyooyin/, /gabdho/,

24

Giorgio Banti

*o and /haweeney/ in these cases, while the AP that surfaces is only the pattern

specified in rule ( 2 4 ) . This means that before the application of AAC the original APs are deleted. This can be captured through rule ( 2 5 ) :

(25)

When a new AP is assigned to a word, the old one is deleted.

Notice that since the inflectional rule of the absolutive case, ie. rule ( 1 8 ) , does not contain any specification about a new AP, rule ( 2 5 ) does not apply and the forms retain their lexical AP. 2.3.3

The nominative

This case has a wider range of forms than the two cases discussed till now. Indeed, four separate sub-rules have to be distinguished:

(26)

Nominative case a.

/ §/

i. ii. iii. iv.

in Dl nouns and, optionally, in D4 nouns; in genitives; in the context [ Poss]; optionally in D4 nouns in the context [ Poss Def]

b.

/ X/

in D4 nouns optionally, and in D5 nouns

c.

/ X + i /

in D2 nouns

d.

* ο / X/

in D3 nouns

It should be pointed out that ( 2 6 . a . i i ) accounts for the nominative-of-genitive forms mentioned in section 2 . 2 . 1 , eg. waraabe from waraab£ 'a hyaena's", or gabdhood from gabdhood Of girls', carruur from carruur Of children 1 , etc. On the other hand, (26.a.iii) and (26.a.iv) describe the facts of note 6 and ( 1 2 . c ) . The cases of D4 nouns that optionally have final Η-tone in all

contexts, and of

D5 nouns with penultimate Η-tone are treated in (26.b) as instances of / X /, ie. of an inflectional

rule that doesn't specify any AP and consequently lets

through the word's lexical AP. 2.3.4

The vocative

All the vocative forms listed in 2.2.2.1-5 uniformly have a Η-tone on their first mora, and L-tones on their other moras. This contrasts with what has been seen to happen till now, ie. with the fact that the APs we encountered involved only the last moras of words. It is necessary, then, to posit a separate type of AP, which applies from left to right rather than from right to left, and can be

Somali and Oromo Nouns

25

distinguished by means of a superscript arrow (ie. *). Convention (16) has now to be integrated with a second part in the following manner:

(27)

Accent Association Convention (AAC), revised: i. ii.

An AP is assigned from right to left starting from the last mora, one accentual item to each mora. ->A left-to-right AP (eg. *) is assigned from left to right starting from the first mora, one accentual item to each mora.

The rule for the vocative case will then be ( 2 8 . a ) , and an illustration of its operation is provided in ( 2 8 . b ) . (28)

a.

Vocative case ->· *

/ X/

b.

* ο hooyooyin * ο *

hooyooyin * hooyooyin ->* hooyooyin

(28. a) (25) ( 2 7 . ii)

-V

* hooyooyin

(20)

Η

->· hooyooyin

(21. b)

HL LL L

3.

The Oromo System

3.1

General features of Oromo tone

Andrzejewski (1957, 1970) and Owens (1980, 1982) recognise three tones in Boorana Oromo: high, low and high-low falling tone, both on long and on short vowels. In all the cases I could check with Abdulhaliim and with other informants who spoke varieties of Eastern Oromo, however, I often found Η-tones rather than falling tones. For instance, Boorana inni n m 'he is a person 1 (Owens, 1980, p.146) or wor an

'spear' (Andrzejewski, 1970, p.92; actually 'it is a spear'.

26

Giorgio Banti

because Andrzejewski often uses predicative forms as citation forms, cf. also Owens, 1980, p.181) ^ were realised as Inni (or inni) narna and, respectively, waräana by my informants. In the few cases where I actually heard falling tones, there was no evidence for their having a phonemic status. For instance, the imperative form läali 'look at! 1 could be realised either with a level H-tone or with a HL-tone on its

first syllable in a sentence like gurbäa läali^läali

'look at the b o y l ' . Since imperatives of this class of verbs distinctly have a -tone on their first syllable if the vowel is short (eg. dhugi 'drink!' or biti ' b u y l ' ) and usually a

-tone also when it is long (eg. fuudhi

'take!' or yäami

1

' c a l l . ' ) , it is plausible to think that in these varieties the alternants läali and läali are only two realisations of a single phonemic läali. However, if in Arsii and Eastern Oromo there is a contrast between H-tones and L-tones both on short and on long vowels, with no evidence of tonal units that occur only on long vowels, and in Boorana Oromo falling tones occur both on short and on long vowels (cf. the above forms nam and woräan ) , it

is possible

to think that the units that bear tones (and accents) in Oromo are not moras as in Somali, but rather syllables. It will be seen indeed that there is never any real need to distinguish short vs. long vowels even at an abstract level in this language, as far as accentual and tonal phenomena are concerned. 3.2

The inflectional

3.2.1

categories of Oromo nouns

The case system

The Oromo nominal inflectional system is more complex than the Somali system, since it involves a higher number of cases and of categories liable to inflection. A good survey of the system, at least as far as Boorana is concerned, can be found in Owens (1980, 1982). Therefore, only its

broad outlines will be

sketched here, yet with some modifications. As in Somali and most other East Cushitic languages, an absolutive (Abs) case has to be distinguished. It is the case form which was termed "forma assoluta" by Moreno (1939, p . 3 3 ) , and "simple case" by Owens (1980, p.148; but already "absolutive" in Owens, 1982 p . 4 5 ) . Its range of functions doesn't differ greatly from Somali. Since in the absolutive all the classes of nouns are fully distinguished from one another, we follow here Owen's (1980) practice of using the absolutive form for citation, even though Oromo speakers usually use the positive predicative for citing nouns. Since there are often complex processes of tone sandhi, whose behaviour has not been fully investigated yet, I use for citation the absolutive forms that occur before imperative or negative verb forms, which have invariable tone (cf. Owens, 1980, p.170 f f . ) and don't trigger

Somali and Oromo Nouns

27

tone sandhi in the nouns that precede them. There is a genitive (Gen)

that marks only the last word in a genitive phrase.

As in Somali ( c f . example 5 . d ) , if the genitive phrase contains more than one word, its other words have the form they have in unmarked contexts, ie.

the

absolutive. For instance, in examples (29.a) and (29.b) only the last words of the genitive phrases niitittii gäldho'oo Of the skinny woman 1 and sä'ä obboleettli isäa Of the cow of his sister 1 are in the genitive case, while niitittii , sä'ä and oJbboleettii are in the absolutive.

(29)

a.

mana niitittii gäldho'oo läali house woman skinny look 1 look at the house of the skinny woman'

b.

aanän sä'ä obboleettii isäa dhügi milk cow sister his drink 'drink the milk of his sister's cow'

The nominative (Norn) marks only non-focussed subjects also in Oromo. Focussed subjects have a special subject-focus form ( c f . 30.a vs. 3 0 . b ) . 9 While genitive and other cases are marked only on the last word in their phrase, nominative inflections occur on all

the sister words of the head noun of the subject phrase.

For instance, in example (30.a) intälti, bareedduun and tun are all nominatives, their absolutive forms being intala, bareedduu and, respectively, tana.

(30)

a.

intälti girl

bareedduun tun beautiful this

isa in - jäalanne him not loved-not

'this beautiful girl didn't love him 1 b.

intala

bareedduu

tanä-tu isä in - jäalanne this FOC "it was this beatiful gill who didn't love him"

c.

eegeen jaldeysäa dhertuu tail monkey long 'the tail (eegeen, Norn of eegee) of the monkey Gen of jaldeysa] is long 1

(jaldeysäa,

There are, however, no nominative-of-genitive forms as in Somali, and the genitive which depends from a nominative subject is inflected like a normal genitive, as shown in example ( 3 0 . c ) . In copular clauses not marked for tense there is no verb, and the predicate phrase has a special morphology. As shown in examples (31) , its

last word is in

what I suggest calling the (positive) predicative case. It is marked only tonally

28

Giorgio Banti

in most nouns and adjectives, but also segmentally in the interrogative word mäal 'what?' (that has mäali 'what is it? 1 ) and, optionally, in genitives. If the predicate phrase contains more than one word, the first one is L-toned, while the intermediate words have their unmarked absolutive shape. 1( ^ (See also note 9 for similar behaviour in focussed subject phrases). (31)

a.

kun mäali? this what 'what is this? 1

b.

kun gurbaa gädhee this boy wicked 'this is a wicked (Abs gädhee) boy (Abs gurbäa)'

c.

kun billawa eysümä isäati this knife uncle his 'this is the knife (Abs billawa) of his (isäati, Pred of the Gen isaa, Abs isa) uncle (Abs eysumä)'

The negative counterparts of copular clauses such as those in examples (31.b) and (31.c) contain the invariable particles miti or mii, while the word that precedes them has a special negative predicative (Neg Pred) form:

(32)

a.

kun gurbaa gädhe'ee

mltl

'this is not a wicked boy' b.

kun billawa eysumä isäatli

l .. |

'this is not his uncle's knife'

The L-toned forms gurbaa and billawa in (32.a) and, respectively, (32.b) show that the first word in the negative predicative phrase must be L-toned as in positive predicative phrases and focussed subject phrases; see also notes 9 and 10.

There is a case which marks only the last word in its phrase, and is called 'dative' in Owens (1980, 1982). Yet it is conceded already in Owens (1982, p.57) that it may indicate a 'locative source' in addition to a 'direct beneficiary'. Gragg (1976, p.184) includes its meaning ' f r o m ' , as in manaa dhufe

suffix among the ablatives and assignee it the 'he came from the house' or mana namatii

dhufe

'he came from the man's house' (tones are not indicated here because Gragg doesn't mark them). It appears thus that the label dative-ablative (DAb) may be more appropriate for this case. In ( 3 3 ) there are two more examples of

it.

Somali and Oromo Nouns (33)

a. he

karäa way

29

dheera'aa in - dhufne long not came-not

'he didn't come a long (Abs dheeraa) way (Abs karäa) ' b.

aanän säree xiggo'oo kenni milk dog small bring 'bring milk for the small (Abs xiqqoo) dog (Abs säree)'

There are several other oblique cases in Oromo, which are discussed, eg., in Gragg (1976, p.183 f . ) , and especially in Owens (1980, 1982). They will not be taken into account here for lack of space. Suffice it to remember the instrumental -

and the dative -

(an identical morpheme is used in some varieties

also for conjoining N P s ) , which suffix to forms whose tonal pattern and segmental shape are similar to the dative-ablative, -Vtti 3.2.2

'to, in, a t ' , etc.

The main noun classes

Several categories can be inflected for case in Oromo: (i) nouns; (ii) adjectives; (iii) modifiers like bir Other 1 and dur 'former 1

(which have the variant

shapes biräa and duräa, formally genitives of the related forms bira and d u r a ) ; (iv)

personal pronouns; (v) numerals from 1 to 10 (the other numerals are adject-

ives) ; (vi) the demonstrative san 'that',· (vii) definitives, which are a semantically heterogeneous group of morphemes characterised by the alternance k- m. vs. t-

f.

( c f . the Somali definitives, eg. articles, possessives and demonstratives),

such as kana "V tana ' t h i s ' , kiyya ^ tiyya ' m y ' and other first and second person possessives, käan ^ täan Other', kam ·

tarn 'which?', kä ^ tä that provides

dummy heads in several types of NPs; (viii) genitives; ( i x ) dependent verbal forms ( c f . in-argatinliti, predicative of the genitive of in~argatin ' ( w h o ) didn't receive', in examples i.a and i.b in note 1 0 ) . Only the behaviour of nouns will be taken into account here, because it can be more directly compared with the inflectional behaviour of Somali. Owens (1980, p.149 f f . ) and Voigt (1985) list respectively five and four morphological classes of nouns, based upon the tonal patterns and the number of syllables. However, the inflectional behaviour of nouns is distinctly linked with their final phonemes, as Andrzejewski (1970, p.89 f.)

already observed.

The tonal patterns, indeed, only distinguish two subtypes in two of the main inflectional classes, while the number of syllables doesn't influence the morphological behaviour in any way (with the only exception of monosyllabic nouns, of section 3 . 3 . 4 ) . In my opinion, it is not necessary to distinguish more than three main inflectional classes of nouns on this basis. They correspond to the classification

30

Giorgio Banti

suggested by Andrzejewski (1970, p.89 f f . ) , with the only difference that I prefer to collapse his first and second classes into my first declension ( D l ) , because they d i f f e r only in their nominative forms, which change according to gender. His Class 3 is my D2, while his Class 4 is my D3. 3.2.2.1

The first declension. All the nouns that end in short -a inflect accord-

ing to the first declension. They can be either masculine or feminine, and are L-toned throughout ( D l a ) , or have

-tones on their last two syllables ( D l b ) ,

1

like nama m. ' m a n , aria m. 'elephant', hilleysa m. 'rabbit', intala f. haadha f.

'mother', adda f.

'morning', billäwä

'girl 1 ,

'forehead, front' - all Dla nouns -, or ganarna m.

m. ' k n i f e 1 , nagäyä m. 'peace 1 , heeruw.ä f.

'peace' - all Dlb

nouns. Also plurals in -ootä like nam-ootä from nama 'man' follow this declension. Their genitive always has long

-toned -äa. The nominative has

-tones on

the last two syllables in the variety examined here. Segmentally there is the suffix -I if there are two consonants before the stem-formative -a in the absolutive, but -ni or -ti in masculine and, respectively, feminine nouns with just a single consonant before -a. The regular forms are thus the following ones: 11

(35)

Dla

Dlb

arbi äddl buddeenni intältl

arba adda buddeenä intala

m. f. m. f.

'elephant' 'forehead' 'injera bread' 'girl'

The predicative has no segmental change, but the tonal pattern is of the type Intala in Dla nouns, and buddeena in Dlb nouns. The negative predicative is identical with the genitive in Dlb nouns, but is

-toned throughout in Dla nouns.

The dative-ablative, finally, has final long -aa in both Dla and Dlb nouns, but the tonal patterns change, as shown in (35) below:

(35)

Dla

Abs Gen Norn Pred Neg Pred DAb

3.2.2.2

Dlb

intala f. intaläa intältl intala intaläa intaläa

'girl·'

buddeenä m. 'injera bread' buddeenäa buddeenni buddeena buddeenäa buddeenäa

The second declension. D2 nouns all

of them also have a

end in a

-toned long vowel; some

-tone on their penultimate syllable. It is thus possible to

Somali and Oromo Nouns distinguish D2a nouns like gurbäa m. ' b o y ' , ijoollee f. 'word' or oogruu f.

31

'children', dubbii f.

1

' f i e l d , from D2b nouns like daräaräa m. ' f l o w e r ' , härree

m. 'donkey', illllii f.

' f l o w e r ' , or akäakoo m. 'grand-father'. Also several

types of plurals are D2, like muc-oolli from mucäa 'child 1 , soor-eeyuii from sooreysa ' r i c h ' , etc. All D2 nouns end in - V ' V V where Dl nouns have inflectional endings in -aa. The genitive of gurbäa is thus gurba'äa, of illllii, illlll'li, etc. The nominative has always -n after the long vowel, while the tonal pattern is similar to the absolutive. In the predicative, negative predicative and dative-ablative, apart from the -V'W ending the tonal pattern of D2a nouns parallels Dla nouns, while D2b nouns are similar to Dlb nouns.

(36) Abs Gen Nom Pred Neg Pred DAb

3.2.2.3

D2a

D2b

gurbäa m. 'boy' gurba'äa gurbäan gurbaa gurba'äa gurba'äa

illllli f. ililli'li illlliin illllii illlll'li ililli'ii

'flower'

The third declension. All the nouns ending in -n are inflected accord-

ing to this declension. They are partly old pluralia tantum containing the stemformative -äan ^ -an like bishäan 'water or aanän ' m i l k ' . (In foon 'meat' and loon 'cattle' the stem-formative and 10'

coalesced with the root vowel, cf. So. so' 'meat

'cattle'). Also has plurals like mukk-een from muka 'ties', ilm-äan

from ilma ' s o n ' , or saa-wan from sä'ä

'cow 1 follow this declension.

The nominative of D3 nouns is identical with the absolutive, while in their other inflected forms they add -ii when Dl and D2 nouns end in long vowels.

(37)

D3

Abs Gen Nom Pred Neg Pred DAb

bishäan m./pl. 'water' bishaanii bishäan blshaan bishaanii bishaanii

I did not find D3 nouns with a

-tone on their stem parallel to Dlb and D2b

nouns. Futher research is needed, however, especially as far as plurals are concerned.

32

Giorgio Banti

3.2.2.4

Reinflected

genitives. It has been stated in 3 . 2 . 2 that genitive forms

can be reinflected for other cases. In section 3 . 2 . 1 ( c f . example 30.c) we saw that this does not happen when the genitive depends on a nominative head noun. Even in predicative phrases a genitive may remain unchanged, although there is often the enclitic L-toned particle -ti. For instance, in example ( 3 1 . c ) the predicative phrase can be billawa eysuma isäa rather than billawa eysumä isäati. In negative predicative and dative-ablative phrases, however, genitives are always reinflected. The stems of these reinflected genitives don't end in —W like the simple genitives, but in -Wt, to which -ii is added as in D3 nouns. The tonal patterns have few peculiarities, as shown in (38) below. Examples of reinflected genitives are, beside isäati in ( 3 1 . c ) , isaatli in (32.b) and inargatlniiti in examples ( i . a ) and ( i . b ) under note 10.

(38)

intaläa Gen of intala ( D l a )

Neg Pred DAb

.intaläa ·. ^intalaatii intalaatii intalaatii

ililli'ii Gen of illllii (D2b) .ililli'ii \ililli'iiti' ililli'iitli ililli'iitii

It should be pointed out that, in a historical perspective, ililli'iiti is probably not to be analysed as ililli'li-ti, as we suggested above, but rather as ililli'iit-i, just like ililli'iit-ii and ililli'iit-ii. (For final -i, cf. maal-i, Pred of mäal ' w h a t ? ' ) . The tonal behaviour shows that there has been a morphological reanalysis, and that -ti

is synchronically a separate - even though

an enclitic - word. Otherwise the tonal pattern would be *ililli'iiti, as in the negative predicative form. 3.3 3.3.1

Analysis of the system The absolutive

Oromo noun forms always contain a stem ( S t ) , which consists of a root at least ie., when it doesn't contain also derivational suffixes -, and a stem-formative ( F o r ) , cf. Gragg (1976, p . l 9 4 ) . i 2 Stems can be accented or accentless: Dla and D2a stems are accentless (intal, gurb) , while Dlb and D2b stems are accented * * (buddeen, ilj.ll). Stem-formatives also can be accentless like a or accented like * * * * * * * W - ie. aa, ee, ii, oo, uu -, aan, etc. Since stems can be accented on their last syllable only, there will be but one lexical AP, namely ( 1 7 . a ) , associated from right to left as in Somali, yet only in Dlb and D l a , D2a and D3 nouns have no lexical AP. (Alternatively, they have null lexical AP, ie.

).

Somali and Oromo Nouns

33

The units that bear the accents in Oromo, however, are syllables and not moras as in Somali. As a consequence, our AAC will have to be revised as in ( 3 9 ) , which applies cyclically at each morphological level. (39)

Accent Association Convention (AAC), revised (first part o n l y ) : An AP is assigned from right to left starting from the last BU, one accentual item to each BU.

We thus have the following derivations:

(40)

a.

Dla

Dlb

intal a

buddeen a * buddeen a

intal a b.

D2a

*

gurb aa * gurb aa * gurb aa

*

D2b * Hill * Hill * Hill

* ii * ii * ii

stem level, ( 3 9 ) D3

*

bish aan * bish aan * bish aan

stem level, ( 3 9 ) word level, ( 3 9 )

If we let apart D2b nouns like ilillii for the time being, the absolutive forms can be derived quite straightforwardly by positing for this case a rule like * ( 4 1 ) , which states that the absolutive is marked by the tone-melody H. (41)

Absolutive case / X/ * H

If it applies to D2a and D3 nouns, rule (41) associates a

-tone with the starred

final syllable. The other syllables remain unassociated and rule (42) will apply. (Notice that 42 is rule 21.b expressed in a manner that can apply both to Somali and Oromo). (42)

BUs not associated with any tone are assigned L-tones.

Rule ( 4 2 ) can also account for Dla nouns like intala. Since they contain no accents, the melody H of rule ( 4 1 ) cannot be assigned to any BU, and consequently rule (42) must assign L-tones to every syllable. Dlb nouns like buddeenä show that tone spread applies rightwards across syllable boundaries in Oromo. This can be captured through rule (43) , which is a fairly general case in tone langu-

34

Giorgio Banti

ages, but does not exist in Somali.

(43)

If there are more BUs than tones, the last tone in the melody spreads rightwards to unassociated BUs.

D2b nouns like ilillii have two underlying accents according to ( 4 0 . b ) . This may be a problem because rule ( 4 1 ) is formulated so as to entail that the melody * H applies to the whole word and not to each of its underlying accents. In the next section we will see that a simple solution is to delete any accent that occurs on the right of a given accent. After such a rule applied, D2b nouns can be treated like Dlb nouns. 3.3.2

The genitive

Genitive forms end distinctively in a

-toned long vowel. Within the framework

developed here this can be captured by stating that the rule for (no -reinflected) genitives is the following one:

(44)

Genitive case / X + VV / * H

The AAC ( 3 9 ) associates the accent with the last syllable of the form. Rule ( 4 4 ) presupposes that also segmental phonemes can be treated as autosegments (for an overview of such an approach see, eg., Van der Hülst and Smith, 1982b). In particular, it is necessary to posit a rule of vowel spread very similar to ( 4 3 ) , The well-known Well Formedness Condition ( W F C ) , in so far as it prohibits association lines to cross ( c f . eg. Clements and Goldsmith, 1984b, p . 1 0 ) , will ensure that a vowel doesn't spread to empty vowel slots across consonants. As a consequence, the W slots remain empty in D3 nouns.

(45)

If there are more V-slots than vowels, a vowel spreads rightwards to unassociated V-slots.

The V-slots that remain empty after the application of ( 4 5 ) are assigned the vowel i. This is a sort of default case which parallels rule ( 4 2 ) :

(46)

V-slots that cannot be associated with other vowels are assigned the vowel i.

Somali and Oromo Nouns

35

We can thus have the following derivations:

D2b

D3

intala

Hi 1 i

bi s ban

vccvcv w

vcvccw vv

CV (:

intala

Hi 1 i

bi s h a n

vccvcv w

VCVCCW VV

CV

b

0 -* ' /

-

[vv]

0 / v

suf

W suf

Rule (48.a) applies to Dl and D2 nouns yielding intal aa and, respectively, ililli ii. Rule (48.b) applies only to D2 nouns, yielding ililli'ii from ililli ii. D2a and D3 nouns clearly show that the accent of the formative is deleted before the genitive suffix. This can be stated through rule ( 4 9 ) . 3.3.6

(49)

(See section

for a revision of this r u l e ) .

->

/ For Suf

Gen

In Dlb and D2b nouns, even after rule ( 4 9 ) applied to the last ones, we are left * * * * with two underlying stars: buddeenaa and ililli'ii. This creates a problem in our framework, just as the absolutive of D2b nouns did (cf. section 3 . 3 . 1 ) . * * Indeed, since in Oromo we meet with at least two tonal melodies, ie. H and HL (see

sections 3 . 3 . 4 and 3.3.6 for the second o n e ) , the tonal melodies we found

in rules ( 4 1 ) and ( 4 4 ) cannot be treated as basic tone melodies that are associated whenever a star is met within the domain of a word. Rather, they apply only once in such a domain, and accent rules must ensure that their context is proper-

36

Giorgio Banti

ly specified. A simple solution is to posit a rule like ( 5 0 ) , that deletes any accent to the right of a given accent in a word. ο

(50)

/

*

Rule ( 5 0 ) must be ordered after rule ( 4 9 ) , otherwise its

output would remove all

the possible inputs of rule ( 4 9 ) . This can be obtained if

(50) is regarded as a

basic accentuation rule ( B A R ) , which applies after the operation of the accentual rules at each level. In this manner, we can have the following derivations:

(51)

Dla

* intal a W

intal a VV

intal

aa

D2a

* * gurb aa VV * gurb aa

it W

* * gurb aa W

Dlb

* * buddeen a W * * buddeen a W * buddeen a W

stem level and inflected word level, (39)

buddeen

segmental rules

aa

D2b * * * Hill ii VV * Λ Hill ii VV * * * Hill ii VV * * Hill ii W * * Hill ii W

*

(50)

D3

* * bish aan W *

*

bish aan W

t * bish aan W *

bish aan W

gurb aa W Hill ii W gurb a' aa

Hill i' ii

stem level, ( 3 9 ) lexical word level, (39) (50) inflected word level, ( 3 9 ) (49) (50)

bish aan ii

segmental rules

Tone association and tone spread apply subsequently to the intermediate represent* * * ations intalaa, buddeenaa, ililli'ii, etc., as we already saw in section 3 . 3 . 1 , and yield intalaa, buddeenaa, Hilli'ii, etc. 3.3.3

The nominative

It has been shown in sections 3.2.1-3 that nominative forms vary considerably according to the inflectional class of the noun. D2 nouns add a final -n but have the same tonal patterns as their absolutives, while D3 nouns don't change at all. This can be captured through rule ( 5 2 ) .

Somali and Oromo Nouns (52)

37

Nominative case (i) a.

/ Χ + η /, if D2 noun * Η

b.

/ Χ /,

if D3 noun

* Η

Dl nouns, instead, have a more complex behaviour. There are three suffixes, -i, -ni and -ti,

1Lt

ie.

whose distribution is governed by the phonological context

and by gender. They are not added to the full lexical form, but replace the Dl stem-formative -a. in addition to this, the tonal pattern is different

from the

absolutive in Dla nouns: while this is always L-toned, their nominative has Htones on the two final syllables. It is not difficult to derive the surface forms if we regard the Dl nominatives as characterised by an AP ** or *o. Since BAR ( 5 0 ) reduces anyhow an AP ** to *o, and there is no positive evidence to discriminate between these two APs, the second one will be chosen here, because of its

similarity with the AP of Somali D3 nominatives. Our accent association

convention (39) ensures that this AP associates with the last two syllables in the form.

(See section 4. for the possibility of treating the AP *° as a pre-

accent). In Dlb nouns it is reasonable to postulate that the accent of the AP *o reassociates vacuously with the already starred penultimate syllable. If now we remember that we indicated with the symbol St in section 3 . 3 . 1 the part of a noun that precedes the stem-formative, ie. that we regard any singular or plural noun as consisting of [St For] , the rule for the nominative of Dl nouns can be represented as follows:

(53)

Nominative case a.

* 9 / St + ι /,

(ii) if Dl noun with stem in -CC

* H b.

* 0

/ St + ni /,

if m. Dl noun with stem in -C

H

c.

* ο / St + ti /, if f. Dl noun with stem in -C * Η

38

Giorgio Banti

3.3.4

The predicative

The final L-tones in the predicative forms of Dlb and D2b nouns, ie.

in the

types buddeena and illllii, show that we have here a different tonal melody than in their absolutives buddeena and illllii. The most obvious candidate is the ittonal melody HL. The predicative forms of Dla, D2a and D3 nouns, which all

have accentless

stems, have a -tone on their first syllable. If the predicative tonal melody * is HL, these forms must have an accent on their initial syllable. As in Somali vocatives, this must be an accent that associates from left to right rather than from right to left. We can thus describe the predicative forms with the following statements:

(54)

Predicative case a.

->* / X /, if the stem is accentless * HL

b.

/ X/ * HL

Just as the first part of our AAC was reworded in ( 3 9 ) by replacing 'mora 1 with 'bearing unit 1 (BU) in order to accomodate also the Oromo facts, the second part of the AAC, when properly reworded, can provide for the left-to-right associat->ion of the AP * to D l a , D2a and D3 nouns. In D2a and D3 nouns such as gurbäa * * * * and bishäan the output of ( 5 4 ) has two underlying accents: gurbaa and bishaan respectively. The BAR (50) ensures that only the first accent survives. Genitives must be excluded from both ( 5 4 . a ) and ( 5 4 . b ) , otherwise we would have forms like *lntalaa or *ililli'ii. We thus need a further statement like ( 5 5 ) , where the brackets show the optional character of ti.

(55)

Predicative case / X (ti) /,

if X = genitive noun form

The predicative forms of monosyllabic D3 words like foon 'meat 1 and loon 'cattle 1 which are identical to their absolutive forms, show that if there are more tones than BUs in Oromo, the tones that cannot be associated are deleted. This can be captured through the following statement:

Somali and Oromo Nouns (56)

39

If there are more tones than BUs, the tones that cannot be associated are deleted.

In other words, when the predicative tone melody HL is associated, eg., with foon only the 3.3.5

-tone can associate, while the L-tone is deleted.

The negative predicative

The framework that has been developed so far suffices to describe the negative predicative forms through rule ( 5 7 ) :

(57)

Negative predicative case a.

/ X + W /, t

if the stem is accentless and if X ^ genitive noun form

H

b.

/ X + W/ * H

The segmental rules (45) , ( 4 6 ) and (48) , which were developed for the genitive forms, account for the processing of the underlying suffix W until it reaches its

surface form. The proviso in ( 5 7 . a ) had to be added because the accent

pattern * does not occur with reinflected genitives of Dla, D2a or D3 nouns, which are respectively intalaatli, gurba'aatli and bishaaniitli rather than *intäläatii, *gurba'äatii or *blshäanlitii. The internal -t-

in the reinflected genitives shows that rule ( 4 4 ) is not

enough, but has to be integrated with ( 5 8 ) :

(58)

Genitive case (continued) * / X + Wt / , in the context

Suf

Rule (58) lacks any indication of the tone melody, because this is provided by the final case rule, eg. for negative predicative forms by rule ( 5 7 . b ) . 3.3.6

The dative-ablative

Dlb and D2b forms like buddeenaa and, respectively, ililli'ii show that there is a W-suffix also in this case, like in the genitive and negative predicative, * while the tonal melody is HL. In Dla nouns there is no underlying accent, and * the tonal melody HL cannot be associated. The default rule ( 4 2 ) will then apply, and correctly produce L-toned forms like intalaa.

40

Giorgio Banti However, D2a and D3 forms like gurba'aa or bishaanii, as well as reinflected

genitives like intalaatii or illlli'iitii show that there is something more involved in these forms. Indeed, what happens is that any accent on a stemformative or on the genitive suffix is deleted before the dative-ablative suffix. This means that a rule like (59) applies in these forms:

(59)

*

->- ο

/ stem x" Suf

It is easy to see that rule ( 5 9 ) is just a more general formulation of rule ( 4 9 ) , which applies in genitive forms. Suppose then that we call all the inflectional suffixes that trigger rule ( 5 9 ) deaccenting suffixes

( S u f ) . The dative-ablative

case can thus be described as in ( 6 0 ) , while (61) is a revised formulation of the genitive rules ( 4 4 ) and ( 5 8 ) .

(60)

Dative-ablative case / X + VV / * HL

(61)

Genitive case (revised) a.

/ X+ W / * H

b.

4.

*

/ X + Wt / , in the context

Suf

Conclusions

In discussing the Somali and Oromo systems in sections 2.3 and, respectively, 3 . 3 , we saw that these two languages have very simple sets of tone rules which * * itinvolve the Basic Tone Melody H for Somali, and the two tone melodies H and HL for Oromo. The actually occurring surface forms can be easily derived with the help of general association principles and rules that have often been found to operate in tone languages. In this respect the main difference between Somali and Oromo lies in the fact that Somali lacks tone spread across syllables - the only kind of tone spread observed in Somali involved Η-tones spreading leftwards within syllable boundaries, under one possible interpretation of rule ( 2 1 . a ) - ,

Somali and Oromo Nouns

41

while in Oromo tones associate freely rightwards. This makes the Somali system similar, at least superficially, to typical pitch accent systems such as those of Indo-European languages like Ancient Greek or Lithuanian. Interestingly, however, the Somali facts can also be accounted for by positing a Basic Tone Melody * HL, with the L-tone freely associating with BUs to its right, provided that a rule like (56) deletes L-tones that would remain unassociated in forms with the Η-tone on their final mora. In a synchronic description of Somali this is unnecessary, since if tones cannot spread across syllable boundaries, a default rule like (21.b) or ( 4 2 ) , which is independently necessary, can account for all the forms with L-tones that occur to the right of the Η-toned mora. It is clear, however, that diachronically it is not difficult for the one system to change into the other. A more elaborate system of rules is required in both languages in order to account for the 'binding-site' of the tonal melodies, ie.

for the choice of the

BU the Η-tone in the melody associates with. To this end we have designated one of the tones in the melody as accented, and invoked the general principle ( 2 0 ) that associates accented tones with underlying accents, indicated by means of an asterisk as usual, cf. eg. Clements and Goldsmith (1984b, pp.13 f f . ) . In addition to this, four kinds of accent patterns were posited, which associate to their bearing units (BUs) with the help of an accent association convention, formulated in ( 2 7 ) for Somali and partly revised in (39) in order to accomodate also the Oromo facts

( c f . section 3 . 3 . 4 as w e l l ) . In this manner, accent patterns

are treated as autosegments that bind to their BUs from right to left in both languages, with the exception of AP *, which binds from left to right. The APs ->that were found to occur in both Somali and Oromo are (i) *, (ii) *, (iii) * o, and (iv) ο , ie.

the lack of any accent. The third AP, which occurs lexically in

polymoraic Dl and D5 nouns, and in the absolutive - in some contexts only, however - and nominative of D3 nouns in Somali, and in the nominative of Dl nouns in our variety of Oromo, is the only AP that involves two accentual items. Since its result is that an accent is placed immediately to the left of a designated BU, it is identical to a pre-accent, ie. to an accent that shifts to the BU that precedes it.

Goldsmith (1984, p . 2 6 ) introduced the symbol -> for the

opposite notion of post-accent in Tonga, a Bantu language, and consequently a pre-accent could be represented as ·-, are in my opinion almost equivalent, and I choose here the first one simply because this does not involve an additional symbol. Accent patterns are assigned lexically to nouns in both Somali and Oromo.

42

Giorgio Banti

In Somali the two APs * and * associate to whole words, general case, cf.

AP * being the more

( 1 7 ) . In Oromo, instead, the AP * is assigned both to stems

and to stem-formatives, but some stems and stem-formative s are accentless. If a word has lexically two underlying accents, the second one is deleted by the Basic Accentuation Rule (50) , a sort of mopping-up rule that applies after the other accentual rules of each level or cycle applied. The Somali and Oromo systems of case inflections involve both segmental suffixes and accents. In Oromo also tone melodies have to be specified for each case. The general format that was used here for case inflections is not transformational, but simply describes the relevant features of a given case form. Typical instances are (18) and (26. c) for Somali, and (41) and (53. a) for Oromo, repeated here in (62) : (62)

a.

Somali absolutive case (= 18) / X/

b.

Nominative case in Somali D2 nouns (= 26. c)

c.

Oromo absolutive case (= 41) / X/ * H

d.

Nominative case in Oromo Dl nouns with stem in -CC (= 53. a) /St *+° i /

Rule (18) states that the absolutive case in Somali does not require special affixes or APs; as a consequence, a noun will retain its

lexical AP. Also in

Oromo the absolutive case ( c f . 41 = 62.c) does not require special affixes or * APs. Yet, it is characterised by the tonal melody H, which binds with the lexic * ally accented BU. (In Dla nouns the tonal melody H cannot associate with any BU because there are no underlying accents. The default rule then applies and the form surfaces as L-toned). In the nominative of Somali D2 nouns like go61 'lioness 1 , an -i is added to the lexical form of the word, and the AP o is imposed on the form. The lexical AP * is then deleted according to the basic accentuation principle (25) , which we found necessary to posit for Somali but not for Oromo. This null AP can

Somali and Oromo Nouns

43

then be associated with the last BU by the AAC, and the form will surface as L-toned, ie. gool-i. Rule (53.a = 6 2 . d ) , instead, states that in the nominative case form of this class of Oromo nouns the suffix -i is added not to the lexical form of the word, but rather to its *.

stem only. The AP *o is then associated by ο

*

o

the AAC yielding forms like ibidd-i from Dla ibidd-a m. ' f i r e ' or jabeeny-i from Dlb jabeeny-a m. 'strength'

(realised as [jabeejipa]). When the melody H is

associated and the tone rules apply, the Η-tone spreads to the last syllable and the final forms are respectively iblddi and jabeenyi. It should be pointed out, however, that rule (18 = 6 2 . a ) cannot account for all

the absolutive forms of Somali nouns. In a few contexts, in fact, D3 nouns

and- optionally - D4 nouns undergo sepecial rules that change their APs from * into *o and, respectively, o, as stated in (19.a) and (19.b). In note 7 it was already observed that there are parallels to the occurrence of plural D4 nouns with AP o in the context

C?)V in Arbore, a language that belongs

to the same sub-group as Somali within East Cushitic. Accordingly, at least part of the special rule (19.b) may be a relic of an older stage of the language.

It

often happens, indeed, that synchronically idiosyncratic and anti-systemic behaviours are reflexes of systems that were fully operative in the past. It is thus possible that also rule (19.a) should be explained diachronically. Finally, it is necessary to remember that we found a peculiar accentual behaviour in the Oromo genitive and dative-ablative forms, which we preferred to regard not as the reflex of a different AP, but as the output of rule (59). This rule deletes an accent on the morphemes that occur between the stem and any of these two suffixes, which were therefore called deaccenting suffixes. I hope that the data discussed in this paper may shed some light on the typology of the prosodic systems of East Cushitic languages. Even though several of the facts we dealt with can be analysed along different lines, it

seems to

be clear that both Somali and Oromo have prosodic systems that involve a somewhat elaborate set of accent rules and a very simple set of tone rules.

NOTES

*The Oromo and Somali data in this paper were collected both in Italy with the support of research funds from the Dipartimento di Studi Glottoantropologici of the University of Rome, and in Somalia while the author was conducting research with funds from the Italian Ministry for Foreign Affairs within the framework of the Project for the development of the Somali language, in co-operation with the Somali National University and the Somali Academy of Science, Arts and Literature.

44

Giorgio Banti The following abbreviations should be explained: Ar. Da. Di. Or. Re. So.

Arbore, Omo-Tana, East Cushitic; Dasenech, Omo-Tana, East Cushitic; Dirayta, Oromoid, East Cushitic; Oromo, Oromoid, East Cushitic; Rendille, Omo-Tana, East Cushitic; Somali, Omo-Tana, East Cushitic.

ART FOC INT

article focus particle interrogative particle.

1.

The Somali data are based mainly on the variety spoken by Dr. Cabdinaasir M. Abuukar, an educated Somali born in Gaalkacyo (Mudug) like Hyman's (1981) main informant. However, reference is made also to the more Southern varieties spoken by Prof. Ciise M. Siyaad, born in Ceel Buur (Galguduud), and Prof. Maxamed Muuse Axmed. Broadly speaking, it can thus be claimed that these data fairly represent central varieties of Standard Somali as it is spoken by educated speakers. The Oromo data are instead based mainly on the speech of Abdulhaliim Mohammed Sheekh Abdisalaam, who was born in the Bale province of Ethiopia in an area inhabited mainly by Arsii (Arusi) Oromos, and has now been several years one of the speakers in the Oromo programs of radio Mogadishu. His variety is thus different both from Boorana, upon which Andrzejewski's (1957, 1970) and Owens's (1980, 1982) analyses are based, and from the Mecca (Mecha) and Tuulama (Tulema) varieties spoken in Gojjam, Wellegga, Ilubabor, Kefa, Shewa and Wello, which were described eg. by Moreno (1939) and Gragg (1976, 1982), and are the main basis for literary varieties of Oromo as a result of several factors ranging from the traditionally high level of literacy among Western, ie. Mecca, Oromos to the fact that most of the Oromo speakers of Radio Addis Ababa are from Mecca and Tuulama areas. Data on Eastern varieties of Oromo, however, are from Abdulhaamid Ahmed and Amiin Rashaad, both born in Dire Dawa.

2.

For an explanation of the term 'Dl noun 1 see section 2 . 2 . 2 .

3.

Somali is transcribed here according to the Somali national orthography. Its main peculiarities are: ' c dh Ah

is is is is is

a a a a a

glottal stop, [ 7 ] ; voiced pharyngeal fricative, [·>] ; voiced retroflex stop, [d] : voiceless velar or uvular fricative, [x] or [ ] ; voiceless pharyngeal fricative, [h].

Somali tones are indicated throughout according to Hyman's (1981) analysis, for which see section 2 . 1 . 4.

A number of minor types such as läba 'two' (Abs läba ^ labä, Gen labä, Nora läba ^ l a u ) , etc., are not taken into account here, for obvious reasons of space. It should be pointed out that Hyman (1981), who was able to collapse a number of types that Andrzejewski (1964, 1979) had been forced to keep apart, oversimplified the system. Indeed, he put together in his first declension ( D l ) Andrzejewski's first three declensions (all masculine) and his fourth declension (all feminine), which have very different inflectional behaviours. I prefer instead to split Hyman's first declension into a Dl (the type Abs orgi, Gen orgl, Norn orgi 'billy-goat') and a D2 (the type Abs mind!, Gen mind!, Norn mindiyi ' k n i f e ' ) . While D2 is followed only by

Somali and Oromo Nouns

45

feminine nouns, Dl includes several masculine nouns as well as a group of feminines (haweeney 'woman 1 , säbti 'Saturday 1 , midig 'right side 1 , e t c . ) . Optionally, however, these feminine Dl nouns have a different behaviour, in so far as they may follow the fifth declension. 5.

Some they have One

6.

There are short possessive suffixes in Somali, eg. Is. -käy, 2s. -kaä, etc., which are used mainly with terms referring to close relatives (aäbbe ' f a t h e r 1 , hooyo 'mother 1 , walaäl 'brother', etc.) or to God. Otherwise the possessive suffixes also contain an article or demonstrative and are then called "long possessives', eg. ratl-gaa-ga 'your camel 1 , rati-gäa-gän 'this camel of yours", etc. With short possessives a noun is always L-toned in the nominative. There are thus Abs D3 aabbä-haä 'your f a t h e r ' , D4 walaäl-kaä 'your brother 1 vs. Nom aabba-häa, walaal-käa. With long possessives, instead, the behaviours diverge: Abs D3 furä-häa-ga 'your key 1 , D4 halä-häa-ga 'your she-camel' vs. D3 furä-häa-gu but D4 halä-häa-gu ^ hala-haa-gu ( c f . also 12.c).

7.

Andrzejewski posits for D4 nouns (his sixth declension) a special accentual unit, ie. AU5, which is realised before a pause like his AU1 in this position (ie. mid tone with secondary stress), but before another syllable like his AU3 in this position (ie. mid tone with absence of stress). Since his AUl corresponds to Hyman's and my accent and -tone, and his AU3 to our lack of accent and L-tone, Andrzejewski's claim about this class of nouns is that they are accented only before a pause, but are unaccented elsewhere. Hyman (1981, p . 1 1 9 ) , however, already observed that D4 nouns can optionally be accented or unaccented when they occur as non-focussed objects, ie. in the absolutive in the context illustrated by ( 1 1 . e ) . Since this optional lack of accent also occurs before the particles bäa and bä independently from an intervening pause, I think that the most proper treatment should not be positing a separate accentual unit or tönerne, but rather invoking an optional accent deletion rule like (19.b). Interestingly, at least for the context (Y) V this rule that deletes the final accent and the -tone in D4 nouns may derive from an older stage of the language. Indeed, in Arbore, a related language spoken in Southern Ethiopia between Lake Stephanie and the River Omo, plural nouns in -o, -ä, and -e like kobo 'sandals' (So. kabo), ilko 'teeth' (So. ilico), ind'ä 'eyes' (So. indho), bice 'water' (So. biyo), etc., lose their final H-tone in exactly this position, as shown in example (i) below, drawn from Hayward (1984, p.158). (The spelling is adapted to the general system suggested by Sasse, 1982, for East Cushitic languages; c stands in Arbore forms for [ t / ] ) .

(i)

8.

pluralia tantum of this group may have a genitive without -odd when have a semantically singular meaning. For instance, caano 'milk 1 can the genitive caano with the numeral häl One' in the phrase häl caano cup of m i l k ' .

bice

me

kar>

wärbata?

water what with you-draw 'what is it you draw water (Abs bice) witn?' a a Raised vowels such as in woräan indicate 'vowel coloured breaths', ie. final voiceless vowels, in Oromo studies starting from Andrzejewski (1957). They are retained here mainly for Boorana forms. In all other respects, however, I use the 'new Oromo script' which has been developed in recent years by several groups of Oromos who are publishing books and journals in their mother-tongue with Latin characters. It is similar to the Somali system in so far as it doesn't use diacritics and can be typed

46

Giorgio Banti with any standard keyboard. Its main peculiarities are the following ones: (i)

' c ch dh kh ny ph g

is is is is is is is is is

a a a a a a a a a

glottal stop, ['] ; voiceless ejective alveolar affricate, [ t / 1 ] ; voiceless alveolar affricate, [t/] ,· voiced implosive alveolar stop, [d 1 ] ; voiceless velar fricative, [x]; palatal nasal, (p]; voiceless ejective bilabial stop, [p 1 ] ,· voiceless ejective velar or uvular stop, [k 1 ] or [q 1 ] ; voiceless ejective alveolar stop, [t 1 ] .

Tones are indicated according to Andrzejewski 1 s and Owens's system, ie. V or W for H-tones, V or W for L-tones, and V or VV for falling tones (in Boorana). Notice that in this manner VV indicates a level -tone in Oromo, but a falling tone in Somali! 9.

Subject focus inflections occur only on the last word in the subject phrase, while its first word has to be L-toned. Intermediate words are in the absolutive: (i)

a.

mucaa dheeraa isi'lit - tu aanän dhuge child tall her FOC milk drank 'it was her (isi'lit-tu, subject focus form of Gen isi'ii) tall (Abs dheeraa) child (Abs mucäa) who drank the milk'

b.

qaallu känä - tt dänsa priest this FOC good 'THIS PRIEST is good' (from Owens, 1982, p.56)

10. There is often a pause after the first and even after the second word of a long predicate phrase in a slow and accurate style. A word with (at least) a -tone in its absolutive shows up in its absolutive form in such cases, while a L-toned word has all its syllables raised to H, as shown in the examples below: (i)

a.

tun oJbJbol§ettii ## nänttichä ## qooda isaa in - argatiniiti this sister man share his not received-not 'this is the sister (Abs obboleettii) of the man (Abs namticha) who didn't receive his share'

b.

kun obboleysä ## nämtichä ## qooda isäa in - argatiniiti this brother 'this is the brother (Abs obboleysa) of the man who didn't receive his share'

Generally speaking, it emerges from Owens (1980) and from my data as well that there are heavy restrictions upon the occurrence of L-toned words in a number of contexts, and that there are different kinds of processes that create -tones in such cases. One of them is illustrated in (i.a) and ( i . b ) above. The Alternating Tone Rule described by Owens (1980, pp.171 f f . ) can be viewed in this perspective, if its action is limited to creating a -tone on the first syllable of some classes of basically L-toned words. Anyhow, this is a field which deserves further investigation. The last word of predicative phrases is marked only tonally also in other Southern varieties of Oromo; eg. for Boorana see Owens (1980, p.178 f f . ; 1982, p.48 f . ) . In Eastern varieties this is true only for words in short -a, eg. tuni intala "this is a girl 1 . D2 nouns add -dha like several Northern varieties, eg., inni jäbaadha 'he is strong (Abs jäbäa)', kun

Somali and Oromo Nouns

47

gurbaa gädheedha 'this is a bad boy 1 . -Instead, D3 nouns add -i , like maali 'what is it? 1 (Abs mäal ' w h a t ? ' ) in Abdulhaliim's variety, eg. kuni bishaani 'this is water'. 11. Several sandhi processes often involve the final phonemes of the stem and the initial or t of the nominative endings. For instance, final d and t assimilate to n: areenni from areedä 'beard 1 , namoonni from namoota ' m e n 1 , with following n yields either of fiixnl, fiindhi, or fiin'i from fiixä 'top, point'; ay with n yields either nageyni or nageenyi from nagäyä 'peace' in Abdulhaliim's variety, etc. It should also be pointed out that while the nominative D2 and D3 forms are generally the same in all varieties of Oromo, Dl nominatives change considerably. For instance, Boorana generalises -1 to all the Dl masculines (-nl is retained only in forms like nageenni from nagä 'peace', where the stem had a final y, cf. nagäyä in other varieties), and -ti to all the Dl feminines, introducing it also in cases like additi from add 'forehead' ( c f . add! in Arsii, addl in Eastern Oromo, e t c . ) . Waata, the Southernmost Oromo variety described by Heine ( 1 9 8 1 ) , has -fin and -tlin with the same distribution as -i and -ti in Boorana, eg. muklin i-5 f9 from muk 'tree 1 (elsewhere mukni, mukni), hintaltlin from hintal ' g i r l ' , etc. Mecca Oromo, instead, generalises the masculine forms in -ni to most of the words that have -ti in Arsii and Eastern Oromo, and retains the old feminine ending only in haati from haadha 'mother' and, optionally, in a few other nouns like lafa 'ground, earth 1 . (Mecca forms lack tone marks because no tones are marked in the sources). 12. No Oromo noun can exist in its bare root. Yet the actual distribution and functions of the different stem-formatives, as well as their relationship with the derivational suffixes, have still to be worked out. 13. Rules (48.a) and (48.b) are specific to some Arsii varieties. Eastern varieties spoken, eg., in the Dire Dawa area, and more Southern ones like Boorana behave differently and require different rules, as shown in (i) below. (The Boorana data are from Owens, 1980, pp.156, 161). Dl

D2

Abs Gen DAb

nama nam&a namaa

gurbäa gurba'äa gurba'aa

Arsii

Abs Gen DAb

nama namäa namaa

gurbäa gurbaadhäa ^ gurbäa gurbaadhäa "*> gurbaa

Eastern

Abs Gen DAb

näzn namä nama

gurbä gurbäa gurbaa

Boorana

(i)

14. Gragg (1976, p.183) regards -i as derived from -ni after a consonant cluster, eg. harkni ->· harki from harka 'hand'. This, however, requires an ad hoc phonological rule for nouns, because the normal treatment in other areas of the morphology of this language is i-epenthesis, eg. ar^ita and argina from /arg + ta/ and, respectively, /arg + na/, 2s. and Ip. of the present tense of ärg-uu 'to see'. Also Boorana forms like additi from add 'forehead' ( c f . note 11) vs. Arsii äddi may show i-epenthesis from underlying /add + tf/. As a consequence, I prefer to treat synchronically forms like m. ärbl and f. äddi as containing a separate suffix -I, which occurs only after geminate consonants or consonant clusters.

48

Giorgio Banti

References Andrzejewski, B . W . ( 1 9 5 6 ) , Accentual Patterns in Verbal Forms in the Isaaq Dialect of Somali. Bulletin of the School of Oriental and African Studies 18, 103-29. Andrzejewski, B . W . ( 1 9 5 7 ) , Some Preliminary Observations on the Borana Dialect of Galla. Bulletin of the School of Oriental and African Studies 19, 357-74. Andrzejewski, B.W. (1965). The Declensions of Somali Nouns. School of Oriental and African Studies, London. Andrzejewski, B . W . ( 1 9 7 0 ) , The Role of Tone in the Borana Dialect of Galla. In: Proceedings of the III International Conference of Ethiopian Studies, 88-98. Institute of Ethiopian Studies, Addis Ababa. Andrzejewski, B . W . ( 1 9 7 9 ) , The Case System in Somali. School of Oriental and African Studies, London. Armstrong, Lilias E. ( 1 9 3 4 ) , The Phonetic Structure of Somali. Mitteilungen des Seminars für Orientalischen Sprachen zu Berlin 37.3, 116-61. Banti, Giorgio ( 1 9 8 4 ) , The Morphology of the Nominative Case in Somali. In: Dressler, Pfeiffer and Rennison ( e d s . ) , 27-31. Banti, Giorgio (in p r i n t ) , Lineamenti di fonologia, morfologia e sintassi del somalo e dei suoi dialetti . Bender, M.L. ed. (1976), The Non-Semitic Languages of Ethiopia. African Studies Centre, Michigan State University, East Lansing. Cardona, Giorgio R. and Francesco Agostini, eds. (1981), Fonologia e lessico. Studi Somali 1. Ministero per gli Affari Esteri e Comitato tecnico linguistico per 1'Universita nazionale somala, Rome. Clements, George N . , and John Goldsmith, eds. (1984a), Autosegmental Studies in Bantu Tone. Foris Publications, Dordrecht. Clements, George ., and John Goldsmith, (1984b), Autosegmental Studies in Bantu Tone: Introduction. In: Clements and Goldsmith ( e d s . ) , 1984a, 1-17. Dressier, Wolfgang U. ·, Oskar E. P f e i f f e r , and John R. Rennison, eds. (1984), Discussion Papers for the Fifth International Phonology Meeting. Wiener Linguistische Gazette, Supplement 3. Goldsmith, John (1984), Tone and Accent in Tonga. In: Clements and Goldsmith ( e d s . ) , 1984a, 19-51. Gragg, Gene B. ( 1 9 7 6 ) , Oromo of Wellegga. In: Bender ( e d . ) , 166-95. Gragg, Gene B. ed. (1982), Oromo Dictionary. African Studies Center, Michigan State University, East Lansing. Hayward, Dick ( 1 9 8 4 ) , The Arbore Language: A First Investigation. Helmut

Buske Verlag, Hamburg. Heine, Bernd (1981), The Waata Dialect of Oromo. Language and Dialect Atlas of Kenya, vol. 4. Dietrich Reimer Verlag, Berlin. Heine, Bernd, Thilo C. Schadeberg and Ekkehard W o l f f , eds. (1981), Die Sprachen Afrikas, vol. 2. Helmut Buske Verlag, Hamburg. Hulst, Harry van der and Norval Smith, eds. (1982a), The Structure of Phonological Representations, part I. Foris Publications, Dordrecht. Hulst, Harry van der and Norval Smith (1982b) , An Overview of Autosegmental and Metrical Phonology. In: Van der Hulst and Smith ( e d s . ) , 1-45. Hyman, Larry M. (1981), L'accento tonale in somalo. In: Cardona and Agostini ( e d s . ) , 109-39. Kiparsky, Paul ( 1 9 8 2 ) , From Cyclic Phonology to Lexical Phonology. In: Van der Hulst and Smith ( e d s . ) , 131-175. Klingenheben, A. ( 1 9 4 9 ) , 1st das Somali eine Tonsprache? Zeitschrift für Phonetik und allgemeine Sprachwissenschaft 3. 5-6, 289-303. Moreno, Martino M. ( 1 9 3 9 ) , Grammatica teorico-pratica della lingua galla.

Somali and Oromo Nouns A. Mondadori, Milan. Owens, Jonathan (1980), Observations on Tone in the Booran Dialect of Oromo (Galla). African Language Studies 17, 141-96. Owens, Jonathan (1982), Case in the Booran Dialect of Oromo. Afrika und Übersee 65, 43-74. Puglielli, Annarita, ed. (1984), Aspetti morfologici, lessicali e della focalizzazione. Studi Somali 5. Ministero per gli Affari Esteri e Comitato tecnico linguistico per l'Universitä nazionale somala, Roraa. Puglielli, Annarita, and Ciise M. Siyaad ( 1 9 8 4 ) , La flessione de! nome. In: Puglielli ( e d . ) , 53-112. Sasse, Hans-Jürgen (1981), Die Kuschitischen Sprachen. In: Heine, Schadeberg and Wolff ( e d s . ) , 187-215. Sasse, Hans-Jürgen (1982), An Etymological Dictionary of Burji. Helmut Buske Verlag, Hamburg. Voigt, Rainier M. (1985), Tone Types of Nouns in Borana. Journal of African Languages and Linguistics 7 . 1 , 59-62. Vycichl, W. (1956), Zur Tonologie des Somali: zum Verhältnis zwischen musikalischem Ton und dynamischem Akzent in afrikanischen Sprachen und zur Bildung des Femininums im Somali. Rassegna di Studi Orientali 31, 221-227.

49

An Accentual Analysis of the Zulu Noun Mary Clark University of New Hampshire

0.

Introduction*

In this paper, I will argue that Zulu, a Bantu language of Southern Africa, is a tonal pitch-accent language, rather than a non-accentual tone language or a metrical pitch-accent system. Before beginning the argument, let me first explain what I mean by these terms. First of all,

by a "tone language" I mean a language in which tone is assign-

ed in lexical representation. Tone languages fall into two classes: (i) "free" tone languages, like Fuzhou Chinese (Wright ( 1 9 8 3 ) ) , where any tone-bearing unit may bear any tone, and (ii)

"restricted" tone languages, like Mende (Leben

( 1 9 7 3 , 1 9 7 8 ) ) , in which lexical tone melodies are chosen from among a restricted set,

with each lexical item marked as to the identity of its

melody, along

with any idiosyncratic information about the mapping of the melody onto the segmental string. By a "metrical" language, I mean a language in which one or more syllables of each word are singled out for special prominence by a metrical "tree" (Liberman and Prince ( 1 9 7 7 ) , Hayes (1980) or "grid" (Prince ( 1 9 8 3 ) ) . Metrical prominence is indicated phonetically by increased intensity and/or duration, and also, often, by high pitch. In singing, there is a tendency to align the metrically-prominent

syllable with the strong beat of the musical phrase.

Rules which assign the position of the metrical accent are frequently sensitive to syllable weight. Now consider the notion of "pitch accent". Suppose, following McCawley (1968, 1977,1978), we define a pitch-accent language as a language with just one lexical tone melody (usually ( L ) H ( D ) , in which lexical items differ only in the presence/absence of the tonal melody and its

location with respect to the

segmental string. Given a language which meets this definition, it would be possible, in the absence of other evidence, to analyze it in either of two ways: (i) as a restricted tone language which has just one lexical melody, or (ii) as a metrical language in which the accent is realized as a high tone. Apparently, both types of pitch-accent languages exist. That is, there are metrical pitch-accent languages, including Vedic Sanskrit

(Kiparsky ( m s ) ) ,

52

Mary Clark

Ancient Greek (Steriade ( m s ) ) , and Malayalam (Mohanan ( 1 9 8 2 ) ) , in which the pitch accent is

(part of) the phonetic realization of an underlying metrical

accent, and, also, tonal pitch-accent systems, including Japanese (Clark, in press), Haya (Hyman and Byarushengo ( 1 9 8 4 ) ) , and Tonga (Pulleyblank ( 1 9 8 4 ) ) , in which the accent is simply a tone at every level of phonological representation. Metrical pitch-accent systems have the same characteristic properties as other metrical systems — that is,

every word is accented, the

location of the accent may be affected by syllable weight, and the accent is often realized phonetically by intensity or duration as well as pitch. Tonal pitch-accent systems, in contrast, lack these metrical properties — that is, the accent is generally not marked by intensity or duration, its position is usually unaffected by syllable weight, there is no tendency to align it with the strong beat of a musical phrase, and some words may carry no accent at

all.

Note that this description treats the tonal class as the unmarked pitch-accent class,· a pitch-accent system which lacks the characteristic properties of a metrical system is analyzed, by default, as a tonal system. The four types of prosodic systems which I am assuming here are summarized below, with examples of each type: TONAL SYSTEMS

Free Tone

Restricted tone, including tonal pitch accent

Chinese Ewe

Mende Japanese Tonga Haya

METRICAL ACCENT SYSTEMS

Stress Accent

Metrical pitch accent

English Latin Modern Greek Chinese

Vedic Sanskrit Ancient Greek Malayalam

Note that it is possible for a language to have two prosodic systems — one tonal and one metrical. For example, according to Wright ( 1 9 8 3 ) , Fuzhou Chinese has both (free) tone and metrical stress; except for the fact that some of the tone rules are sensitive to stress, the two systems operate independently of one another. We will find a similar situation in Zulu. In the remainder of this paper, I will present an analysis of the Zulu noun

The Zulu Noun

53

with three aims in mind: first, to show that Zulu is a pitch-accent system of the tonal variety, with the underlying lexical melody "H"; secondly, to develop enough of the tonal grammar of Zulu to give a clear picture of how languages of this type may be expected to work; and finally, to make some suggestions, based on evidence from Zulu, as to how prosodic systems change from one category to another. This developmental question is of particular interest for the Bantu languages, where there appears to have been a great deal of migration from one category to another. In particular, although Proto-Bantu is thought to have been a free tonal language (Clements and Goldsmith ( 1 9 8 4 : 3 - 6 ) ) , many of the modern Bantu languages, including Tonga, Luganda, and Haya, have developed tonal or metrical pitch-accent systems, while others, including many of the languages of Guthrie's Zone G and scattered examples from other zones, have lost their tone altogether in favor of a system of metrical stress.

1.

Zulu Nominal Tone

A Zulu noun consists of a noun class prefix plus a stem. The stem, in turn,

is

made up of a mono- or bi-syllabic root, followed by any number (including zero) of extensional suffixes; for example, 1 ' 2

(1)

a.

[

aba

[ S(TEM)

C1.2 b.

[

aba

äbäntQ

[ ntu ] ] ] R(OOT)/ "human being'

[ S

[ R

]

ntu

ana

'people'

]]

—>

abäntwänä

diminutive c.

[

ema

C1.6

[ S

[ R

khosi 'chief

]

ana

'children' ]

eni

]]

locative

-->

emäkhösänenl 'elder sons' (LOG)

The tonal melody of the stem depends on (i) the tone class of the root, and (ii) the overall length of the stem. (Suffixes are toneless, and so contribute nothing to the melody.) In all ( 1 9 7 0 ) , Rycroft

the analyses with which I am familiar

(Cope

(1963,1980), and, most recently, Laughren ( 1 9 8 1 , 1 9 8 4 ) ) , Zulu

is treated as a restricted tone language. In Laughren's analysis, which is the simplest of the analyses in this model, noun roots are divided into four tonal classes, which d i f f e r in the identity of their lexical melodies and in the mapping of the melody onto the segmental string, as shown below:

54

Mary Clark

(2)

a. L class:

b. HL class:

ntü

kole

"person"

fo

kole 1

H L 1

c. Special HL class, with 1st tone linked to 1st two V ' s :

khosi

H

khathi (L)

'chief

«*

H

I

L

H 'fellow'

khosi I

khathi

r

H (L;

H L school'

H (L)_

d. LHL class:

fo

I

'time 1

I

L

H

Except for Class ( c ) , which has special association properties, the lexical melody is mapped onto the stem by a rule of Initial Tone Association which we will state as follows:

(3)

Initial Association: Link free tones to free V-positions, one-to-one, from left to right

The parenthesized L tones of ( 2 b ) ,

( 2 c ) and ( 2 d ) show up overtly just in case

there are suffixes to carry them: äbä-fo 1

isi-khathi 'time / Isi-khäsh(i)-änä

'fellows'/ abä-f(o)-änä 1

'small time ,

'boys';

etc.

While the identity of the stem melody is determined entirely by the root, its

position may be affected by the presence of suffixes. In particular,

according to Laughen (1984:194) "if

...

the number of tone-bearing units

[in th°. stem, including suffixes ] exceeds the number of tones ... by more than one",

then the "regular" association rules ( ( 2 c ) , ( 3 ) ) fail to apply, and are

replaced by a rule which links the high tone of the melody to the ante-penultimate syllable:

(4)

Final High Tone Association (Laughren's rule ( 2 4 ) )

t t t (where "t" = a tone-bearing unit)

H L

Examples of the application of this rule are given below:

The Zulu Noun (5)

a.

fo-any-any-ana

\ *

of prefix;

H L

universal conventions

H L

b.

Addition ·;.-->

fanyanyana y

khosi-ana-eni

>

Addition of prefix; universal conventions

khosaneni

H L

H L

55

.. .. „ , . . aba-fanyanyana y y

; ;/

\

|/

H L H L 'tiny boys'

emä-khös inen} |

/

1 /

H t'' H t 'elder s(ans (LOG) '

The noun-class prefixes, which are an obligatory part of every noun, fall into two tonal classes — "long" prefixes, with the surface melody LH or HL, and "short" prefixes, with the melody L; for example:

(6)

a. with "long" prefix:

äbä-fänä 'boys'

äbä-ntwänä 'children'

b. with "short" prefix: bä-ntwänä 'Children!'

bä-fänä 'Boys!'

The choice between the "long" melodies, LH and HL, depends on the tone class of the root; the melody is LH before classes ( 2 a ) and ( 2 d ) and HL before classes

(2b) and ( 2 c ) :

(7)

a. äbä-ntu

I I LI

'people' ( 2 a )

L H

b. Isl-khathi

I I I I

L H

L

'time' ( 2 d )

H

I

c.

äbä-fo

I I I

H L

'fellows' ( 2 b )

d. in-kosl

I

V

H(L) H

H

'chief ( 2 c ) ("'." = downstep)

Cope (1970) gives a morphological treatment of this alternation, with each root class subcategorized for one of the two tonal "prefixes" HL or LH. Laughren gives a phonological account in which the LH melody ( 7 a , b ) is derived by a metathesis rule which reverses the (underlying) HL before stems whose melody begins with "L":

(8)

HL —>

LH /

[ pref

[

L

(Laughren's rule ( 1 4 ) )

Mary Clark

56

2.

Zulu as a Tonal Pitch Accent Language

In this section, I will argue, contra Cope, Rycroft, and Laughren, that Zulu is a tonal pitch-accent language with the tonal melody "H". In the analysis which I wish to propose, noun roots are divided into two tonal classes —an "accented"

(="toned") class and an "unaccented" (="untoned") class — with

some roots marked, also, for the position of the tonal accent in relation to the segmental string, as shown below:

(9)

a. Accented (=toned) roots 1

kole

(i)

With unspecified linking:

(ii)

With the H linked to both

(iii)

With the H linked to the 2nd

's:

:

[v

school '

khosi

' chief

khathi

'time'

fo 'fellow 1

1I b. Unaccented (=toneless) roots:

[ntu]

'person'

Prefixes, like roots, are either "accented" or "unaccented":

(10)

Class 2 Accented

(=long) prefixes

Unaccented (=short) prefixes

Class &

aba

ama

[ba]

[ma]

Class 7

isi

[si]

In roots of class ( 9 . a . i ) and in prefixes, the association of the tonal accent is not specified in lexical representation, but is established by Initial Association

( 3 ) , which links the

tone to the leftmost V-position. Low tones

are introduced by rule during the course of the derivation. The pitch-accent analysis not only simplifies the lexical representations (in that accented roots no longer have to be specified as to which lexical melody they carry), but also allows a simple, unified account of what, in Laughren's analysis, had to be treated as two distinct phenomena — the rightward shift of the stem accent in long stemmed-nouns like 'tiny boys' ( 5 a ) and (emä-)khös(i)-an (a)-eni

(äbä-)f(o)-äny-äny-änä

'elder sons' (5b) and the change

of the prefix melody to LH in forms like äbä-ntü and isi-khathi

(7a,b).

The Zulu Noun

57

In the present analysis, both these changes can be accounted for at once, by means of the following three rules: (11)

a. Extratonality Marking (Stem Level) Mark the final syllable of a multi-syllabic stem as "extratonal" 4 b. Spreading V

V V

(Word Level, iterative) (where "V" = a free V-position)

c. Degemination (Word Level, iterative)

These rules are illustrated in the derivations below, first for the "short"stemmed nouns äbä-fänä

'boys' , ämä-khosl ' c h i e f s ' , ba-ntu 'People;' and äbä-ntü

'People', and then for the "long"-stemmed nouns äbä-fänyänyänä

'tiny boys' and

£mä-khösän£nl 'elder sons':

(12)

LEXICAL COMPONENT STEM LEVEL

a.

fa(na) i H

b.

kho(si) \y H

c.

kha(thi) | H

Underlying rep. after Initial Assoc. and Extratonality 5

isi-kha(thi) ! I H H

Attachment of prefix, with Initial Assoc.

isi-kha(thi) T/ I

Spreading ( l i b ) ; Degemination ( H e )

WORD LEVEL

aba-fä(na) ; I b H

ama-kho(si) · V/ H H

n.a.

n.a.

POST-LEXICAL COMPONENT

aba-fana

ama-khosi

|

I

|

isi-khathi

V

I

I

äbä-fänä - n

ämä-khosi m - s

Isi-khathi

H L H L

HL

L

I I I !

M V H

l l l l H

L

H

Removal of Extratonality

L-Insertion Rules

58

Mary Clark

(13)

LEXICAL COMPONENT WORD LEVEL

b. aba-ntu

a. ba-ntu

Underlying representation after attachment of prefix; 6 Initial Association

aba-ntu

n.a.

Spreading ( l i b ) ; Degemination ( H e )

V H

POST-LEXICAL COMPONENT

ba-ntu I

I

i

t,

L-Insertioh rules

abä-ntü

i I I

L H

L

LEXICAL COMPONENT

(14)

STEM LEVEL

a. ntu-any-any-ana

b. fo-any-any-ana I

H

ntwanyanya(na)

fanyanya (na) H

Underlying repr. after Initial Association Coalescence 7 , Extratonality Marking

WORD LEVEL

aba-ntwanyanya(na)

aba-fanyanya(na)

H

H

aba-ntwanyanya(na)

LH

' ,·'.-' *'--

\

H

aba-fanyanya(na)

I

H

aba-ntwanyanya(na)

I

H

l·'

aba-fanyanya(na)

Attachment of prefix; Initial Association Spreading ( l i b )

Degemination ( l i e )

H

POST-LEXICAL COMPONENT äb^-ntwänyänyänfi

jL Li Li lHl L- ' 3.

äbä-fänyänyänä

MM y H L L H L

Removal of Extratonality; L-Insertion rules

Independent Evidence for Extratonality Harking and Degemination

The rules of Extratonality Marking ( l l a ) , Spreading ( l i b ) and Degemination ( l i e ) , which are illustrated in the derivations above, are not only formally simpler than the rules they replace — that is, Laughren's Final High Tone Association

The Zulu Noun (4)

(with the very complex condition which governs its

59

application) and

Metathesis (7) — but also, to make the argument stronger, they have independent motivation in the grammar. For example, Degemination ( l i e ) is needed independently to account for the degemination of the stem tone in forms like ämä-khösänä 'elder sons'

(from underlying äma-khosi-ana), while Extratonality Marking ( l l a )

accounts for the fact that this tone does not degeminate in the related form ämä-khosi?. Derivations for these contrasting forms are given below:

(15)

LEXICAL COMPONENT

STEM LEVEL

v

y

khosi

Underlying representation

khosi-ana H

kho(si)

khosa(na)

V

Coalescence and Extratonality Marking

H

E

WORD LEVEL ama-kho(si)

anta-khosa (na)

H

H

}

H

n.a.

Attachment of prefix; Initial Association

VH n.a.

n.a.

Spreading ( l i b )

ama-khosa(na)

H

Degemination ( l i e )

H

POST-LEXICAL COMPONENT ämä-khosl

ämä-khösänä

H L

H L

l l

V

H

Because the final

Removal of Extratonality; L-lnsertion rules

L H L

of ämä-khosl is still extratonal when Degemination comes up

for application, it is not available to satisfy the structural description of the rule, and, consequently, the rule cannot apply. Note, however, that khosi does undergo Degemination when it

is in first position in a compound; for

_

example,

(16)

I

H

-khösi-käzi

I I V

L H

L

'chief w i f e '

from

-khosi-

H

fi (Doke ( 1 9 2 6 : 1 3 4 ) )

9

Here, because khosi is non-final in the word, its final syllable loses its Extratonality by the Peripherality Condition of Harris ( 1 9 8 3 ) 1 0 , and Degemination

60

Mary Clark

is therefore able to apply. This account of the otherwise puzzling distribution of Degemination in these forms provides strong independent support for our rule of Extratonality Marking. Independent evidence for Spreading ( l i b ) will be given in the

following

section.

4.

Post-lexical Spreading

In the derivations of sections 2 and 3, syllables which are still untoned at the end of the lexical component are shown to surface, in all

cases, with low

tone. While this is correct for slow, careful speech, there is a tendency, at conversational speed, for non-final toneless syllables to attach to the H tone to their left, as in the examples below, from Cope: I

(17)

äbä-fänä

'boys'

(Cope ( 1 9 7 0 : 1 1 6 ) ) 1

ämä-khösänä

or

ämä-khosänä

(cf. ämä-khösänä

'elder sons'

(Cope (1970:119))

( ( 1 5 ) above))

If we adopt the view of Odden ( 1 9 8 2 ) , Carlson (1983), Leung (1985), and Clark ( m s ) that downstep is the phonetic reflex of a sequence of high tones, then the forms of (17) have the tonal representations I

(18)

I

äbä-fänä

ämä-khösänä

ämä-khosänä

H L

L H L

H H L

7

which can be derived from the lexical output forms äba-fäna

( 1 2 a ) and äma-

khosäna ( 1 5 ) by the addition of one lexical and two post-lexical rules, as follows:

(19)

Extratonality Removal (Lexical) Remove all extratonality markings

(20)

Post-lexical Rules a. Spreading (optional):

V V l·''

T

b. Default Tone Assignment:

T

V —> L

(word level)

The Zulu Noun

61

With the addition of these rules, the forms of (18) can now be derived as follows, where the two pronunciations of ämä-khösänä d i f f e r in whether Spreading (20a)

is invoked once or twice 11

(21)

LEXICAL COMPONENT aba-fa(na) | | H H

ama-khosa (na) | | H H

ama-khosa(na) I | H H

Underlying repr. after Spreading ( l i b ) and Degemination ( c f . (15) above)

aba-fana | |

ama-khosana ( I

ama-khosana ( I

Removal of Extratonality

H

H

H

H

H

H

POST-LEXICAL COMPONENT aba-fana

ama-khosana

ama-khosana

H

H

H

H

H

t

H

H

H l

äbä-fänä I | | J

V

Spreading (20a)

L

ämä-khnsanä l J l l l [/ ! l !

ämä-khosänä \ / *** l l \^^^ \ '

H

H

L H L

Default Tone Assignment

HL

Note that post-lexical Spreading (20a) is a variation of lexical Spreading ( l i b ) , and hence provides the independent evidence which I promised above for that rule. 1 2 Our rules for Zulu are now as follows:

(22)

LEXICAL COMPONENT

STEM LEVEL

Initial Association (3) Coalescence ( f n . 7) Extratonality Marking ( l l a )

WORD LEVEL

Spreading ( l i b ) Degemination ( l i e ) Removal of Extratonality (19a)

POST-LEXICAL COMPONENT

Spreading (20a) Default Tone Assignment (20b)

We now turn to a potential problem with our rule of Spreading (20a) — namely, that the condition which prevents this rule from applying to the final tone of the word (that is,

the second "T" in the structural description of the rule)

appears to be contradicted by the data of Cope ( 1 9 7 0 ) , who marks high tone on the string of syllables at the end of a phrase-internal word, with an "upstep" before the first H of the following word:

(23)

:

äbä-fänä

„

ä-bä-yl-'funl

(where "" = upstep)

...

'The boys do not want it' (Cope ( 1 9 7 0 : 1 1 6 ) )

62

Mary Clark

In response to this potential objection, I would like to suggest that, at the I

phonological level, abä-iänä does have the final L tone which our grammar predicts; that is, the phonological tone melody of ( 2 3 ) is (24)

aoa-rana aba-fana

v IV H

a-Da-yi-rur ^a-ba-yi-funi

H L

H L

H

The phonetic output ( 2 3 ) is derived from this representation by two phonetic realization rules, which we will state (roughly and informally) as follows: (25)

a. Upstep (optional): 13 The pitch registers (that is, the realization pitches for high and low tone) are raised in the environment H L b. Downstep (disjunctively ordered with ( a ) ) : The pitch registers ll are lowered in the environment H *

The application of these rules to ( 2 4 ) is shown below, where the top line of each box represents the high pitch register, and the bottom line the low pitch register:

(26)

abafana H

11/

H L

abayifuni H

11

HL

Here the pitch registers are lowered by (25b) after the syllables bä, yi, and fu and raised by (25a) after the syllable fä. Because (25a) raises the low pitch register to the level which was formerly occupied by high, the high pitch register now occupies a higher level still — hence the "upstep" on the bä of abayifuni.

(25a) may also apply before a low tone which has been intro-

duced by Default Tone Assignment inside a word, yielding the alternate pronunciation emä-khosänenl (Cope ( 1 9 7 0 : 1 1 9 ) ) for emä-khösänenl 'elder sons' (LOG)

5.

(Ic).

Phrase Level Rules

In addition to the rules of ( 2 0 ) , Zulu also has several other postlexical rules, including a rule which lengthens the penultimate syllable of a phrase, as in

The Zulu Voun (27)

Non-phrase-final form

63

Phrase-final form

bantu äbäfo

bääntti äbääfo

'People! 1 (VOC) •fellows'

a rule of Final H Dissociation which dissociates a geminate H from the

final

syllable of the phrase, as in (28)

Non-phrase-final form

Phrase-final form Amakho&si

ämäkhosi

and a rule which lowers the final L to

(29)

'chief'

L ( " ) , as in ( 2 7 ) , ( 2 8 ) and ( 2 9 ) Phrase-final form

Non-phrase-final form

äbääntVt äbäntw££n& äbäfäänä

äbäntü äbäntwänä äbäfänä isikolä

'people' 'children' 'boys' 'school'

The rules of Phrase-final H Dissociation and Final Lowering, which are illustrated in ( 2 8 ) and ( 2 9 ) , can be stated straightforwardly as follows:

(30)

a. Final H Dissociation:

V V H

b. Final Lowering:

L

PHRASE

/

PHRASE 15

As for penultimate lengthening, Doke (1926) claims that this is a consequence of primary stress, which, in his transcriptions, falls on the penultimate syllable of the word or phrase, with secondary stress on alternate syllables beginning with the first syllable of each word 16 ·.

(31)

a.

* * * um-fzlzu.-kaazi

'widow' 1 7

b.

* * * um-ftlzu-kazaana

'widow' (dim.)

c.

*

* * * - ~ . ;2 -kazanyaana

«·

ir

*

d. um-fzlau-kazanyanyaana

'widow 1 (double dim.) 'widow' (triple dim.)

64

Mary Clark

* * e. iinyaattO zengaane 'feet 1 'baby'

->-

* * unyawdzzngaanc

,„ "baby's feet 1 1 0 (Doke (1926:184-5))

If this is true, then penultimate lengthening can be stated simply as a rule which geminates the vowel of the primary-stressed syllable, as follows:

(32)

Stressed Syllable Lengthening: . 0

-»-

V

/

f\ ^

where V. is a copy of V . , and

. is the primary-stressed

of the

phrase.

The geminate vowels then "share" their identical features, by a rule which we will state as follows:

(33)

Sharing:

The puzzling contrast in the position of the final pitch-fall in the forms isi-koole ( c f . isl-köle) and ämä-khoSs" ( c f . ämä-khosi) follows automatically from this analysis, in the following way: When Stressed Syllable Lengthening applies to ämä-khosi the epenthetic vowel cannot copy the tone of the host vowel without violating the autosegmental condition which prohibits the crossing of association lines (the Crossed Lines Constraint):

(34)

ama khosi

I I I I III

VCV-C VCV H

*ama khoosi —X—>

H

MM I ' l l

VCV-C WCV H

HH

Consequently, in this case, it copies only the segmental features of the host V, and not its

tone, as shown below:

The Zulu Noun

(35)

65

ama khoosi

III I N i l

vcv-c wcv

V

Final H Dissociation then dissociates the final H:

(36)

ama khoosi

UN I l l l

VCV-C WCV H

H

and the exposed vowels receive low tone by Default Tone Assignment. The Phrase Level derivations for the constrasting forms isi-koole and ämä-khoosi are given below:

(37)

isi-kole isi-ko.

ama-khosi

H

H

I

H

V

H

n. a.

n. a.

isi-kole

ama-khosi

H L

HL

isi-koole

ama-khoosi

H L

HL

1

Default Tone Assignment (20b)

H

Stressed Syllable Lengthening (32)

H

n. a.

isi-koole ^ H

V

1 1 IX

1 1 1 H:H 1L

H L

Spreading (20a)

1 i

l i lH Li

Output of Lexical Component

Sharing ( 3 3 )

L

. a.

ama-khoosi

1 1 FT!

Final H Dissociation (30a) with Default Tone Assignment (20b) 2 0

Isi-koole

ämä-khoösl

Merger (38) below; Final Lowering (30b)

H L

H L

HL

II VI H

\\

L

HL L

IV

HL

Note that, before Final Lowering applies, the string of L tones at the end of ämä-khoosi is merged into a single L, by a rule of the form

(38)

Merger:

V V

L-iL

L

66

Mary Clark Our post-lexical rules so far are as follows:

(39)

WORD LEVEL:

Spreading (20a)

PHRASE LEVEL:

Default Tone Assignment (20b) Stressed Syllable Lengthening ( 3 1 ) Sharing ( 3 2 ) Final H Dissociation (30a) Merger (38) Final Lowering (3 Ob)

I follow Pulleyblank (1983) and Archangel! (1984) in assuming (i) that redundancy rules like Default Tone Assignment must apply at the beginning of the phonological level at which the redundant feature is first referred to in the structural description of a rule, and (ii) that once a redundancy rule has come into effect, it re-applies, by convention, whenever its structural description is met.

One consequence of these assumptions for Zulu is that the postlexical

component must be divided into two levels, with Spreading (20a) , which applies before Default Tone Assignment, in one level, and the end-of -phrase rules, which follow Default Tone Assignment, in another.

6.

Depressor Lowering

In addition to its advantages for the tonal system proper, our pitch-accent analysis also allows a considerable simplification of the rules for "depressor consonants" — voiced, breathy consonants (represented here as "C" ) which have a lowering effect on the tonal melody, as in

(40)

a. zäzeela LH L

'try to walk 1

b. sizSku izk

w L

'we shall go1

c. c. Sngifunl ngi

\

L

H L

' I do not want to'

The analysis of Depressor Lowering which I will present in this section is based on the proposal of Laughren (1984) that depressor consonants project tonal features — the features for extra-low tone — onto the tonal tier. Evidence that Depressor Lowering involves the insertion of a tonal autosegment, rather than simply a phonetic influence from the consonant (as proposed by Lieber (ms) ) , is provided by the interaction of this rule with other tonal phenomena — for example, by the fact that the rule is blocked in nouns such as the following, which belong to the tonal class of ämä-khosi ' c h i e f :

The Zulu Noun

(41)

a. um-fäzl

b.

lH / VH

-JcunzI

c.

K V H H

ubu-'hlunqu

VH VH

67

d. ll-'khändä

l/H V H

Both in Laughren's analysis and in mine, the double linking of the stem tone in words of this class is specified in lexical representation, and so is already present when Depressor Lowering comes into operation. Consequently, if Depressor Lowering is a phonological rule, as we are claiming, then its

failure to apply

in words of this class follows automatically, by the Crossed Lines Constraint. (See the discussion in connection with ( 3 4 ) above.) As for the rule itself, one would like, if possible, to state it as a simple redundancy rule of the form:

(42)

Depressor Lowering:

C_

-*·

However, as is pointed out by Lieber

L_

( m s ) , this is precluded in Laughren's

analysis by the fact that low tones are already present in lexical representation and subject to Spreading, so that, from the very earliest level of the derivation, the forms of ( 40) have the representations

(43)

a. zazela

b. sizoku

W L

TV H L

c.

and rule ( 4 2 ) would be blocked (incorrectly) in

angifuni

\fL HI IL (b) and (c) by the Crossed

Lines Constraint. To account for these forms, then, Laughren is forced to introduce a second Depressor Lowering rule, which links a depressor consonant to an L-tone to either side of it,

(44)

C V

->

C V

and lowers that L to L_:

(mirror-image)

(Laughren's rule ( 6 1 ) )

This rule accounts for the extra-low tone of (40b) (from ( 4 3 b ) ) and ( 4 0 c ) (from ( 4 3 c ) ) , as well as the second L_ of (40a) (from ( 4 3 a ) ) ; however, it not account for the L_ which appears on the first is needed as well.

21

does

z_ of (40a) , and so rule ( 4 2 )

In other words, Laughren's analysis requires two Depressor

Lowering rules — ( 4 2 ) and ( 4 4 ) . Our pitch-accent analysis eliminates the need for a second Depressor Lowering

68

Mary Clark

rule. If Depressor Lowering is assigned to the post-lexical Word-level component then low tones will not yet be present at the time it applies; thus the forms of ( 4 0 ) and ( 4 1 ) will enter the rule with the representations (45)

a. za~zela

b. si-zpAu

c. an-gi-funi

d.

H

H

e. ίη-kunzi

f. ubu-hlungu

I V

Η

um-fazi

I

Η

V

Η

H

σ. ii-khanda

I V

Η

Η

Η

and rule ( 4 2 ) will apply, correctly, in ( a ) - ( c ) , but not ( d ) - ( g ) . It is a strong advantage of our analysis that we are able to account for all Depressor Lowering with this one simple rule.

instances of

22

The L_ tone which is introduced by ( 4 2 ) feeds into a series of rules which we will state as follows (closely following the formulation of Laughren (1984: 211-12)): (46)

a. L Flop:

C

(bi-directional) 23

V

IX L_ b. H Shift:

V,

V

(where V. is unstressed)

A·"

c. Contour Simplification:

V

H [-HIGH] The overall effect of these rules is to spread the L_ tone locally to either side of the depressor consonant, as illustrated in the following derivations for the forms IzI-hl

ld/'Izl-hl

' lci 'seats', Izl-fundo "izl-fuundo

'lessons',

and ubu-hlungu/ubu-hluungu "pain":

(47)

LEXICAL COMPONENT WORD LEVEL

izi-hla(lo)

izi-fu(ndp)

ubu-hlu(ncfu)

Underlying rep. after attachment of prefix, Initial Association

The Zulu Noun

r \

izi-hlalo — ι

izi-fundo

I

Η

H

H

uJbu-hlungu

l

Υ

69

Spreading ( l i b ) , Degemination ( l i e ) , Removal of Extraton. (19a)

POSTLEXICAL COMPONENT WORD LEVEL

izi-hlalo

izi-fundo

\?l LH

HL

n.a.

Depressor Lowering (42) ,- L Flop (46a)

H1

n.a.

-bu-hlungu V H H

n.a.

Ϊ-'

Spreading (20a)

PHRASE LEVEL (phrase-final form)

n.a.

n.a.

izi-hlaalo

izi-fuundo

ubu-hluungu

ΨL HA\LL \L

H L

H

izi-hlaalo

izi-fuundo

izi-hlalo

ν λ\ L

H L L

N \\Ψ HH L

Ψ ΛΙ

H L

izi-hlaalo

izi-fuundo

L

H L

L

ψ l\ 1 L

H L L

n.a.

TT n

n.a.

Sharing ( 3 3 )

n.a.

Contour Simplification (46c)

n

J_i

rl

Jj

ui>u-hluungu

11 !1 ': '·ι

VI/

H

Ψ

vacuous

IV

L

Stressed Syllable Lengthening (32) 2 1 *

w HV vL

n.a.

izi-hlaalo

Default Tone Assignment ( 2 0 b ) ; H Shift (46b)

H L

HL LL

ubu-hluungu

V

Nil

H

Final H Dissociation ( 3 0 a ) , with Depressor Lowering and Default Tone Merger (38) ; Final Lowering (30b)

HL LL 1

IzT-hl alo

izT-fuund PHRASE LEVEL

izi-hlalo

\fl ! i LH

L_

Phonetic Interp. ^

(non-phrase-final f

n.a.

Default Tone Assignment (2 Ob)

n.a.

H Shift ( 4 6 b ) ; Contour Simplification (46c)

L L

izi-hlalo ι—/

n.a.

uiu-hluungu"

'

H

TzT-hl l

L

izi-fundo

IV HIVL_

H L_

izl-f nd

ubu-hlungu

Phonetic Interp.

70

Mary Clark As we could expect, H Shift (46b) is blocked by the Crossed Lines Constraint

if there is a depressor consonant to the right of the target H, as in zazala 'try to walk'

(48)

zazela pi !/

Output of Depressor Lowering and L_ Flop ( 4 6 a )

LH£ zazela

Default Tone Assignment (20b)

blocked

H Shift (46b)

zäzelä

Phonetic Interpretation

The blocking of H Shift in this form provides additional support for our contention that Depressor Lowering involves the projection of a tonal feature onto the tonal tier. A complete list of our rules for Zulu is given below:

(49)

LEXICAL COMPONENT

POSTLEXICAL COMPONENT

STEM LEVEL

Initial Association (3) Extratonality Marking ( l l a ) Coalescence ( f n . 7)

WORD LEVEL

Spreading ( l i b ) Degemination ( H e ) Removal of Extratonality (19)

WORD LEVEL

Depressor Lowering ( 4 2 ) L_ Flop (46a) Spreading (20a)

PHRASE LEVEL

Default Tone Assignment (20b) H Shift (46b) Stressed Syllable Lengthening ( 3 2 ) Sharing ( 3 3 ) Contour Simplification (46c) Final H Dissociation (30a) Merger (38) Final Lowering (30b)

Initial Association, Depressor Lowering, and Default Tone Assignment act as conventions, in that, once they enter the grammar, they remain in effect and re-apply whenever their structural description is met. Space does not permit me to justify every aspect of this rule ordering; however, I would like to point out that the classification of Depressor Lowering as a post-lexical Word Level rule is justified by the fact that the depressor

The Zulu Noun L_ blocks the rightward spreading of the prefixal H (that is, (20a) ) in forms like

it blocks Spreading

'-khwämä ' b a g s ' : t

(50)

izi-khwama

\1^ I

HI,

71

'bags'

H

I

(*lzj.-khw$mä)

( c f . lsj.-khyßinä

7

I --'" I

H

l->"' I

H

H

'bag')

H

(Laughren (1984:203-204)

However, because Depressor Lowering is a postlexical rule, it follows lexical Spreading and Degemination, as can be seen from forms like Izl-hlaalo 'seats' (47)

(from izi-hlalo), in which the prefixal H crucially undergoes Spreading

and Degemination before Depressor Lowering applies. The ordering of H-Shift before Stressed Syllable Lengthening is established by forms like in-tombazaane 'maid'

(from in-tombazane), while "izl-hläälo 'seats' (from izi-hlalo) establish-

es the ordering of Stressed Syllable Lengthening before Contour Simplification.

7.

Conclusion

In summary, I have argued here that the tonal grammar of Zulu — particularly as regards the rightward shift of the H tone in prefixes and long stems (sections 1 , 2 ) and the pitch-lowering effects of depressor consonants — is greatly simplified by treating Zulu as a pitch-accent language with just one lexical tone melody — H. Note that, by the criteria which were established at the beginning of this paper, Zulu is a tonal rather than a metrical pitch-accent system, in that the position of the tonal accent is determined not by metrical structure, but by lexical stipulation, along with association and dissociation rules of the type which are found in tone languages. I will conclude with a brief discussion of the implications of this analysis for our understanding of the historical process by which tonal systems have tended to develop into accentual systems in the Bantu languages. In the story of this process which is presented by Goldsmith (1984:48) and Clements and Goldsmith ( 1 9 8 4 : 7 ) , Proto-Bantu is said to have been a free tone language, on the grounds that bi-syllabic noun stems exhibited all possible combinations of H and L:

(51)

a. CVCV

b. CVCV

c. CVCV

d. CVCV

The change to an accentual system took place, in this view, as a result of the

72

Mary Clark

introduction of "Meeussen's Rule" (H -»· L / H

(Goldsmith (1984:50)), which

collapsed classes (a) and ( b ) , thereby reducing the number of tonal classes to three, and leading to a wholesale re-analysis of the language as an accentual system. Note that it is the development from free tone to pitch-accent which, in this view, is seen as the major typological change; the subsequent change from pitch-accent to stress-accent in languages like Swahili is a (comparatively minor) alteration in the mode of realization of the underlying accent. The analysis of Zulu which has been presented here brings every step of this story into question. First of all,

as we have seen, Zulu has the same

four root tone classes ( ( 2 ) , (9) above) as are postulated for Proto-Bantu in ( 5 1 ) ; however, if my arguments are correct, Zulu is not a free tone language, but a tonal pitch-accent language, with just one lexical tone melody — H. It seems likely, then, that this may also have been true of Proto-Bantu; in other words, Proto-Bantu was probably not a free tone language, but a tonal accent language of the same type as modern Zulu. Secondly, under the view of tonal and metrical accent which has been presented here, the introduction of Meeussen's rule and consequent loss of the contrast between tone classes (a) and (b) (the kole and khosi classes of Zulu) would surely not have caused a major re-analysis of the tonal system, but only a slight simplification of the lexical representation, in that both root classes could now be represented simply as

(52)

cvcv H

with no specification of the association between the tonal accent and the segmental string. What, then, is the source of the change from (tonal) pitch accent to metrical (pitch or stress) accent in Bantu languages like Swahili? What Ί. would like to suggest is that this change took place not because of any tendency for tonal pitch-accent systems to be re-analyzed as metrical accent systems — for example, to my knowledge, such a development has never occurred in Japanese, which has a tonal-accent system very much like that which we have proposed here for Zulu (Clark (in p r e s s ) ) . What happens in Bantu, rather, is that the pitchaccent system may merge with the metrical stress system which already exists in the language. To see how easily such a merger might occur, let us return to Zulu, this

The Zulu Noun

73

time to the class of accented verbs, where the H stem tone typically appears on the penultimate syllable, rather than on the ante-penult as in nouns:

(53)

a. b. c. d. e.

bODnä bönllsä bönäkäälä" bönäkällisa bönäkällsiisä

'see' 'show' 'be visible' 'cause to appear" 'be clearly visible'

(Doke ( 1 9 2 6 : 1 8 3 ) )

In other words, in verb stems of this class, the tonal accent and the stress accent

(indicated by the lengthened vowel) coincide.

The overlap between the two accent-systems in these words initially leads one to wonder whether verbs of this class may already have been re-analyzed as a metrical pitch-accent system — that is,

whether the tonal accent in stems

of this class may be introduced by a rule of the form "Assign high tone to the most heavily-stressed syllable of the word". However, further investigation shows that this is not the case. The relevant evidence comes from examples like the following, in which the penultimate syllable begins with a depressor consonant, and the tonal accent appears on the ante-penult:

(54)

a. ukwendziiwä 'to give' (passive)

b. ukumbiTwa c. ukugeez_a~ 'to dig' (passive) 'to wash' (Doke ( 1 9 2 6 : 1 4 7 ) )

These forms show that the verb emerges from the rules of Spreading ( 2 0 a ) and Degemination ( l i e ) with its tonal accent still linked to the ante-penultimate syllable, just as in nouns. The penultimate position of the tonal accent in the forms of ( 5 3 ) is then obtained by a rule of Rightward Shift which belongs to the Postlexical Word Level and so is blocked in ( 5 4 ) by the presence of a depressor consonant. However, were it not for this blocking effect, the tonal accent of accented verb stems would coincide in every case with the stress accent, and this part of the lexicon would indeed be eligible for reanalysis as a metrical pitqh-accent system. If this system were to spread throughout the language, then Zulu would be a metrical pitch accent language. From there, the further change to a metrical stress system, a change which has been made in Swahili and many of the other languages of Guthrie's Zone G, could be accomplished by very small further alteration in the grammar -- the loss of the

(now redundant) accent tone.

74

Mary Clark

Notes

*I would like to thank Mary Laughren, Donca Steriade, and Rochelle Lieber for helpful criticism of earlier versions of this paper. Most of the work on this paper was done while I was a Visiting Scholar in the Department of Linguistics and Philosophy at M . I . T . , and I am grateful for their kind hospitality and the use of their facilities. 1.

Abrreviatory notation which I will use in this paper is as follows: H = "'" = high tone; L = "'" = low tone; and L_ = "~ " = extra-low tone. "V" = a vowel position; "T" = a tone; "V" = a V-position which is not associated to a tone. Association and dissociation will be represented by a dotted line and by crossed lines, respectively: (i)

a. V T

b.

Ϊ

2.

These and subsequent examples are given in the form which they would take in non-phrase-final position. The final word of the phrase is subject to special rules which, among other things, lengthen the penultimate vowel and lower the final low tone. (See section 5 below.)

3.

The statement of Initial Association which is given here, following Pulleyblank (1984), is the one which I will be assuming in the analysis of the following sections. Laughren's own formulation follows that of Clements and Ford (1979), in which this rule/convention affects only the first tone of the melody, and links it to the first tone-bearing unit. The remaining elements are then linked up by Universal Association Conventions which insure that every tone is linked to a tone-bearing unit and every tone-bearing unit to a tone. In the framework which I will be assuming here, Initial Association is the only association convention; tonebearing units which do not receive a tone by Initial Association remain toneless until they are assigned a tone by later rules or conventions.

4.

The device of "Extratonality Marking", adopted here from Pulleyblank (1984), based on the "extrametricality" device of Nanni (1977) and Hayes ( 1 9 8 2 ) , is a device for removing a peripheral element from the domain of subsequent tone rules and conventions. The fact that the rule applies only to polysyllables is apparently a universal condition on rules of this type; for example, monosyllabic nouns in English are not deprived of stress by the rule which assigns extrametricality to the final syllable of a noun (Hayes (1982)).

5.

Extratonality is indicated here by parentheses. Note that "extratonal" does not mean "non-tone-bearing"; the extratonal syllable may have been assigned a tone in lexical representation or earlier in the derivation, before the assignment of extratonality. This is equivalent to the comparable situation in stress systems, where, for example, the assignment of extrametricality to the final syllable or the verb execute at the word level (so that it does not receive the primary stress) does not deprive it of the stress which it received earlier, at the σ level:

(ii)

Wd: * e : * (*) σ : * *(*) execute

(Prince (1983:30))

The Zulu Noun

75

Note that Extratonality Marking does not apply in these forms because the stem is monosyllabic. Coalescence is a rule which merges a syllable sequence of the form "CV.. V " into a single syllable of the form "CV." or "CGV.". If V is u , 1it is retained as a glide, as in ( 1 4 a ) ; otherwise, it is deleted. In the Natal dialect, as described by Khumalo (1981) (cited by Laughren ( 1 9 8 4 : 2 0 6 ) ) , Degemination does apply to stems like khosi, so that we get, in phrase-final position, (iii) ίή-k in-koosi osi

ϊ Π

This can be accounted for, in the present analysis, by ordering Degemination after Extratonality Removal ( ( 1 9 ) below) in Natal. 9.

Doke gives a phonetic rather than a phonological representation, in which relative pitch is indicated by means of numbers, ranging from 1 to 9. In this and subsequent citations from Doke, I have replaced his pitch numbers with what I take to be the corresponding tonal representation. I have also abstracted away from a late glide-formation rule, whereby, in Doke 1 s transcription, transitions from H to L take place not at the syllable boundary where I have placed them, but in the middle of the following syllable. Note that in the dialect which Doke is describing, the tonal accent of the second stem of a compound is deleted, so that the compound in-khoslkSzT (16) surfaces with just one tonal accent and not two. This rule apparently does not apply in the KwaZulu dialect (cf. the compounds cited by Laughren (1984:208) from Khumalo (1981)).

10. The Peripherality Condition is a universal convention which removes the Extrametricality/tonality marking from a unit which becomes non-peripheral in its domain by the application of a morphological rule. 11. And in the pronunciation

m -khos n , Spreading does not apply at

all.

12. This argument assumes, following Mohanan (1982) and Halle and Mohanan (1985), that a single rule (or variations on a single rule) may remain in force across two or more contiguous phonological levels. 13. This treatment of upstep as the consequence of a phonetic rule of LowRaising is based on a proposal by Khumalo (1981), as described by Laughren (1984:225) . 14. Note that the first H of a phrase is not lowered, even if it follows L, as can be seen in the following representations by Doke (1926), cited by Laughren (1984:223): (iv)

3 2 5 4 isi-khaathi

ι ι ρL Hι L H

'time'

cf.

2 4 3 9 isi-koole

ιι ν ι H L H L _

'school'

This follows from our formulation of Downstep ( 2 5 b ) , but not from the usual formulation, in which H is said to lower after L. 15. I am using the term "phrase" somewhat loosely here. The texts of Doke (1926) show that words are first grouped into "minor" phrases, and that these, in turn, are grouped into a "major" phrase, as in the following

76

Mary Clark excerpt from "Zulu Text No. 3", pg. 257: (v)

6 3 66 5 zahamb_aakc they travelled

63 5 zaayt until

6 33 5 zafiika they reached

55 4 Ahoona there

6 3 36 6 zabebuula they it stripped

6 3 38 9 zabebuula stripped

What we see here is (i) that penultimate lengthening applies to every phrase, and (ii) that the ends of the minor drop only to 5 or 6, with a drop to 9 at the end of the major phrase. Depressor consonants (C) are associated with a pitch of 6. If we interpret Doke's 5 as a simple L, and his 6-9 as L_, then it is unclear whether Final Lowering should be stated as a rule which applies at the end of all phrases, or at the ends of major phrases only. I have chosen the more general environment for the rule, on the grounds that this accounts for the fact that a phrase-final L is not subject to upstep ( ( 2 5 a ) above), but perhaps that fact could be accounted for in some other way. 16. There is apparently some disagreement among Zuluists as to the placement of stress; for example, Rycroft and Ngcobo (1979) [cited by Laughren (1984:231)] state that Dynamic stress is potentially associated with the first syllable of each root or stem. In addition, certain words and formatives ... have special inherent stress ... which may take priority. In practice, actual realization of stress appears to be conditioned by speech rhythm requirements. In Zulu this seems to involve the principle of isochronous stress-timing which may condition the neutralisation of a potential stress when an adjacent syllable has priority, or when (as in rapid speech) more syllables are crammed in, between the stress. Insofar as Doke's transcription is correct, it can be captured formally by the following system of stress rules, stated here in the framework of Prince (1983): (vi)

WORD LEVEL:

a. Mark the final syllable as extrametrical b. Stress the final (remaining) syllable c. Stress alternate syllables across the remainder of the word, beginning at the left, and avoiding stress "clash" PHRASE LEVEL: Re-stress the final stress

17. Note that the m is syllabic. 18. The system of stress rules stated in fn. 16 above would assign a stress also to the first σ of zengaanz. 19. I use [oo ] here as an abbreviation for

ο V

V

20. Note that once Default Tone Assignment has come into effect, it re-applies, by convention, whenever its structural description is met. 21. Laughren (personal communication) now believes that imperative verbs like z zSSlS (40a) have low tonal prefixes. If that is true, then the lowered tone at the beginning of the first syllable could be accounted for by rule (44) , and rule ( 4 0 ) might not be necessary. However, even if a floating low prefixal tone could be motivated in this case, it seems

The Zulu Noun

77

doubtful that this could be done for all cases of word-initial CV; such cases arise in all parts of the lexicon — not only in imperative verbs like zäzSSIS, but also in (unprefixed) vocative nouns such as (vii)

Ogre: 1

zTimu

(VOC)

( c f . nwäSbS

'Chamelion!') (Doke 1926:153)

and quantifying expressions such as "all" and Only": (viii) zpdwlzTmbuuzT

Only goats'

( c f . oodwämähtfäsT

(Doke 1926:163)

Only horses')

22. The rule will have to be generalized to account for the lowering which sometimes occurs in connection with breathy voice features which are introduced by morphological rule and linked to a word-initial vowel or sonorant consonant. For example, hhe copulative prefix 'it is 1 can be expressed simply by breathy voice, as in (ix)

a. wit hi

'it is a stick'

(cf. uutM

'stick')

b. Sm&ändlS

'it is strength'

( c f . ämäändlä

'strength')

As can be seen from these examples, morphologically-assigned depressor features also act as an environment for Depressor Lowering, with the following exception: In the KwaZulu dialect, non-nasal sonorant consonants either do not accept the depressor features at all, as in (a) below, or, alternatively, accept the breathy voice but not the lowered tone, as in (b): I

(x)

a. lukhulu

'it is big'

b. lethä

'bring'

(Laughren (1984:216-17), citing data from Khumalo ( 1 9 8 1 ) ) However, there is apparently some dialect variation in this regard, since Doke:180 gives IFethä ' b r i n g ' , with both breathy voice and lowered tone on [1] . 23. Note that the depressor L_ links only to the immediately adjacent syllabic (or, possibly, sonorant) segments. For example, in the form (xi)

umsdmbuluuko

^ ^ v L L H L

'Monday'

( c f . umSombuluuko

y> ^ L L H

'it is Monday 1 )

L (Rycroft, personal communication to Mary Laughren) the depressor L_ links only to the [m] and [u] to either side of the [b], and not to the preceding vowels. This represents another potential problem for Laughren's analysis of Depressor Lowering, since, under her analysis ümSömbuluSkS would emerge from the lexical component as

(xii)

and be converted by ( 4 4 ) to (xiii,

.... H L

78

Mary Clark When there is space available, as in (50) below, the depressor L_ may subsequently spread to the right by Spreading ( 2 0 a ) . 1

I

24. Note that our rules derive ubu-hluungu rather than ubu-hluungu, as indicated by Laughren ( 1 9 8 4 : 2 2 0 ) . Laughren's representation could be derived from ours by the addition of a simple, low-level rule; however, I doubt that this is necessary, since the two representations are presumably very difficult to distinguish in pronunciation. 25. Note that our rules derive zaz&lä rather than zazelS for the non-phrasefinal form. I do not know which pronunciation is correct. If it is the latter, then our grammar would require an additional rule of the form (xiv) L

->

L_

/

L_ —

] WORD

References Archangel!, Diana ( 1 9 8 4 ) , Underspecification in Yawelmani Phonology and Morphology. M . I . T . Ph. D. Dissertation. Carlson, Robert ( 1 9 8 3 ) , Downstep in Supyire. Studies in African Linguistics 14.1, 35-45. Clark, Mary M. ( m s ) , Downstep Without Floating Tones. Department of English, University of New Hampshire. Clark, Mary M. (in press), Japanese as a Tone Language. In: Takashi Imai and Mamoru Saito ( e d s ) , Issues in Japanese Linguistics. Dordrecht: Foris Publications. Clark, Mary M. (1978a), A Dynamic Treatment of Tone, with Special Attention to the Tonal System of Igbo. Doctoral Dissertation, University of Massachusetts at Amherst. Clark, Mary M. (1978b), On the Distinction Between 'Pitch-Accent' and 'Tone' Languages. In: Mark J. Stein ( e d . ) . Proceedings of the Eighth Annual Meeting of the North Eastern Linguistic Society. University of Massachusetts at Amherst. Clements, G . N . and John Goldsmith (eds.) ( 1 9 8 4 ) , Autosegmental Studies in Bantu Tone. Dordrecht: Foris Publications. Clements, G . N . and Kevin C. Ford ( 1 9 7 9 ) , Kikuyu Tone Shift and Its Synchronic Consequences. Linguistic Inquiry 10.2, 179-210. Cope, A.T. ( 1 9 7 0 ) , Zulu Tonal Morphology. Journal of African Languages 9.3, 111-152. Doke, C . M . ( 1 9 2 6 ) , The Phonetics of the Zulu Language. Johannesburg: Witwatersrand University Press. Goldsmith, John (1984), Tone and Accent in Tonga. In: G . N . Clements and John Goldsmith ( e d s . ) . Guthrie, Malcolm ( 1 9 6 7 ) , The Classification of the Bantu Languages. London: Dawson's of Pall Mall. Halle, Morris and K . P . Mohanan ( 1 9 8 5 ) , The Segmental Phonology of Modern English. Linguistic Inquiry 16:1, 57-116. Harris, James (1983), Syllable Structure and Stress in Spanish: A Nonlinear Analysis. Cambridge: M . I . T . Press. Hayes, B. (1980), A metrical theory of stress rules. PhD thesis, MIT. Hayes, Bruce ( 1 9 8 2 ) , Extrametricality and English Stress. Linguistic Inquiry 13:2, 227-276.

The Zulu Noun

79

H/man, L . M . and E . R . Byarushengo, A model of Ha/a tonology, In: G.d. Clements and J. Goldsmith ( e d s . ) . Khumalo, J . S . M . (1981), Zulu Tonology. Unpublished Master of Arts Dissertation. University of the Witwatersrand, Johannesburg. Kiparsky, Paul ( m s ) , The Vedic and Paninian Accent Systems. Department of Linguistics, Stanford University. Laughren, Mary ( 1 9 8 4 ) , Tone in Zulu Nouns. In: G . N . Clements and John Goldsmith (eds.). Laughren, Mary (1981), An Autosegmental Account of Tone in Zulu. In: G . N . Clements ( e d . ) , Harvard Studies in Phonology II. Indiana University Linguistics Club. Leben, Wra. ( 1 9 7 8 ) , The Representation of Tone. In: Victoria A. Fromkin ( e d . ) , Tone: A Linguistic Survey. Academic Press. Leben, Wm. ( 1 9 7 3 ) , Suprasegmental Phonology. Doctoral Dissertation, M . I . T . Leung, Elizabeth (1985) , Tones in Llogoori Verbs. Paper delivered to the 16th Annual Conference on African Linguistics, Yale University. Liberman, Mark and Alan Prince ( 1 9 7 7 ) , On Stress and Linguistic Rhythm. Linguistic Inquiry 8, 249-336. Lieber, Rochelle ( m s ) , Απ Integrated Theory of Autosegmental Processes. Department of English, University of New Hampshire. McCawley, James D. (1978), What Is a Tone Language? In: Victoria A. Fromkin ( e d . ) , Tone: A Linguistic Survey. Academic Press. McCawley, James D. ( 1 9 7 7 ) , Accent in Japanese. In: Larry M. Hyman ( e d . ) , Studies in Stress and Accent, Southern California Occasional Papers in Linguistic, 4. McCawley, James D. (1968), The Phonological Component of a Grammar of Japanese. The Hague: Mouton. Mohanan, K . P . ( 1 9 8 2 ) , Lexical Phonology. Doctoral Dissertation, M.I.T. Nanni, Deborah ( 1 9 7 7 ) , Stressing Words in -ative. Linguistic Inquiry B, 752-763. Odden, David ( 1 9 8 2 ) , Tonal Phenomena in Kishamba. Studies in African Linguistics 13:2, 177-208. Prince, Alan (1983), Relating to the Grid. Linguistic Inquiry 14:1, 19-100. Pulleyblank, Douglas ( 1 9 8 4 ) . Autosegmental Accent. In: Charles Jones and Peter Sells ( e d s . ) , Proceedings of NELS 14, GSLA, University of Massachusetts at Amherst. Rycroft, O.K. (1980), Nguni Tonal Typology and Common Bantu. African Language Studies 4, 43-68. Rycroft, D.K. (1963), Tone in Zulu Nouns. African Language Studies 4, 43-68. Rycroft, D.K. and A.B. Ncobo ( 1 9 7 9 ) , Say It in Zulu. Manual and Tapes. School of Oriental and African Studies, London. Steriade, Donca ( m s ) , Degenerate Syllables and the Rules of Accent Assignment in Greek. Department of Linguistics and Philosophy, M . I . T . Wright, Martha S. ( 1 9 8 3 ) , A Metrical Approach to Tone Sandhi in Chinese Dialects. Ph.D. Dissertation, University of Massachusetts at Amherst.

Prosodic Trends in the Bantu Languages John Goldsmith University of Chicago 1.

Introduction *

My purpose in this paper is to give a brief and summary sketch of the typical characteristics, and of the evolution, of the Bantu tone and accent systems as we know them today. This area is one in which there is a great deal of ongoing work that promises to deepen and alter our picture of the Bantu systems, and so our conclusions today are necessarily tentative. Nonetheless, a number of trends and general properties can be discerned that are certain to be of interest both to the Africanist and to the general linguist. Putting the family in perspective, it is useful to recall that the Bantu group has a time depth on the order of 2.500 to 3.500 years. (For general discussions, see Bouquiaux et al., (1980), Philippson ( 1 9 7 7 ) ) . Using the now traditional classification of Greenberg (1963), we may note that Bantu is a member of the Bantoid group, itself a part of the Benue-Congo branch of NigerCongo. The Bantoid group includes such languages as Tiv as well, while the next closest group in Niger-Congo is the Kwa group, which includes such well-known languages as Igbo, Yoruba, and the Akan family. (Recently, however, the accuracy of a Kwa vs. Benue-Congo split per se has been the subject of consideration and debate; see Bennett and Sterk ( 1 9 7 7 ) , Williamson ( 1 9 8 2 ) ) . In our remarks on Bantu, we will be concerned only with the tonal properties of nouns and verbs, and more with verbs than with nouns, for the verbal system is where the greatest amount of tonal and accentual action is to be found. The reader will no doubt be aware that all lexical nouns in Bantu languages are marked as belonging to one of between 15 and 21 or so noun classes, the exact number depending on the language, with the classes organized roughly into pairs of singular and plural classes (cl. 1 is singular and class 2 plural (human), class 3 singular and class 4 plural/ and so forth). In addition, the 1st and 2nd person, singular and plural, pronominal forms add in effect four more noun classes to the Bantu system. All nouns have a prefix marking their nominal class, and inflected verbs are marked with a Subject Marker (SM) agreeing with their subject, and under certain conditions with one or more Object Markers (OM) marking agreement with their grammatical object ( s ) .

82

John Goldsmith

The structure of the verb, then,

(1)

is:

Bantu verb: subject marker

tense marker

object marker

radical

extension*2 °

Final Vowel

.stem.

Proto-Bantu was a language with seven contrastive vowels, and each could be long or short in a lexical root. There were two tones, High and Low, and each could associate with any mora of a lexical root, though no single raora was lexically associated with more than one tone. Among the tonal characteristics reconstructable for the Bantu family are: (a) the suffixation of a High tone after the tone of the radical in certain past tenses, and (b) a rule lowering High toned verb stems to Low tone after a High toned Object Marker, perhaps applying in more general phonological contexts. We will call this rule Meeussen's Rule (cf. Goldsmith 1984a). In the evolution of the five hundred or so Bantu languages, we can discern the following trends: (1) Tonal properties, in the broadest sense, are shifting from being properties of moras to properties of morphemes as a whole at the deepest level ( c f . Clements 1983).

(2) The position of surface association of

a tone is shifting from the vowels of the morpheme that the tone is initially associated with to vowels either to the left, or, more frequently, to the right. (This is not in contradiction with the preceding point; the initial association may be an association of the tone to a predictable position in the morpheme.) In some cases this shift is quite general and across the board; in others, shifting of this sort is restricted to etymologically short vowels, and the original vowel length contrast is lost. (3) The association of tones and vowels is becoming more sensitive to metrical structure, either at the word-level or at the phrase-level, with a tendency for High tones to reassociate from a metrically Weak position to a Strong position. (4) In some cases, a "one High tone per word" restriction has arisen, where this restriction can result in either one High toned vowel per word, or one High-toned sequence of vowels per word. (5) Finally, one of the most pervasive trends has been the shift from an "equipollent" treatment of the High tone/Low tone contrast to a "privative" treatment, to use Trübetzkoy's terminology. Thus the evolution has been from what one might think of as a prototypically tonal system, in which each syllable is associated with either a High or a Low tone (with only a limited set of cases where there was no underlying tone associated — primarily the verbal extentions), towards

Prosodic Trends in Bantu

83

systems with what Hyman has aptly called "incomplete tonification" (Hyman 1982). While the primary contrast in the first kind of system is between vowels associated with a High tone and vowels associated with a Low tone, in the new kind of system the primary contrast is between vowels associated with no tone at

all,

on the one hand (typically the reflex of Proto Bantu Low tone), and vowels (the reflexes of Proto Bantu High tones) associated with one of the tones of a fixed tonal pattern or melody. This melody — called the "Basic Accentual Melody" in Goldsmith (1982) — could be High, or Low High, or High Low; quite a few distinct systems of this sort can be distinguished. Thus, in Sukuma, for example, the evolution was as in (2) (see Goldsmith (in press a ) ) . (2)

a.

Proto Bantu:

V

b.

Proto Bantu:

L

Sukuma:

V

V H

Sukuma:

V L

H

This aspect of the Bantu evolution is the characteristic which Goldsmith (1982) suggested was criterial, by definition, for an accent system, a suggestion to which we will return below. It is likely that there is a close connection between this notion of accent, which triggers the insertion of a specified tone melody and its association to a designated vowel, and the Strong metrical positions mentioned just above, which do not trigger insertion of tones, but in quite parallel manner act as attractors with respect to tones already present on the tonal tier. An additional area that has not until recently been studied is the rise of tonal systems with no (or with minimal) lexical contrast. Odden's study in this volume presents our current knowledge of this area.

2.

Shift of orientation from mora to morpheme

The verb radical contains only a single syllable, though it may contain two moras. As we noted above, it is to the radical that extensions are added to form the stem. In contrast to what we find with nouns, with verbs it is only on the first mora of the radical, however, that a lexically distinctive tone may be found. In many Bantu languages various extensions have become syncronically attached to the radicals, forming semantically unanalyzable units ("extended

84

John Goldsmith

radicals"; cf. Guthrie ( 1 9 6 2 ) ) composed frequently of several syllables (and thus moras). Nonetheless, it remains the case that the first mora is the only one that can associate with a lexical tone. Three observations are in order.

(1) If this restriction is just that there

are only two tones in the language and that no lexical entry may have more than one tone on its underlying tonal tier, then the association of that single tone to the first vowel of the radical (or extended radical) follows from association conventions of autosegmental phonology. This in turn argues for a cyclic application of the association conventions, at least to the extent of allowing a layer of the morphology composed of the stem, if the normal rule of tone-tovowel association is one which links the leftmost tone and the leftmost vowel of the domain in question; if there were no stem-cycle, then in languages where the inflectional prefixes have no tones of their own, the radical's tone would be likely to associate with a verbal prefix (such as a subject marker), a most unusual occurrence, in fact. This conclusion is independently an attractive one, corresponding essentially to the derivational morphology. (2) Many of the Bantu languages have today a three-way contrast in the tone (or accent) of the verb radicals. This, however, is a secondary development in languages with tone shifts resulting from a merger of long and short vowels, whereby High tone on a short vowel continued to shift, but High tone on a long vowel did not. This is discussed below in section 3. (3) There are attested cases where the lexical tone appears to associate with a later vowel in the verb stem, as in Digo (Kisseberth ( 1 9 8 4 ) ) . Even here, though, it can be demonstrated that there is a deeper association between the tone and the first vowel of the stem. Noun stems are typically of the form CV(V)CV. Holding aside the case of the long vowel in the first syllable, one would expect four tone patterns in a twotone language for nouns of the form CVCV, and indeed one does find HH, HL, LH, and LL in reconstructions of Proto Bantu (Guthrie, Meeussen) and a good number of present-day Bantu languages. However, a major effect across the Bantu languages has been the loss of contrast of HH and HL ( c f . Carter ( 1 9 7 3 ) ) , with the merger in favor of HL. This effect appears to be intimately related to the application of Meeussen's Rule (High becomes Low after High) across morpheme boundary cited above. Even in languages where a four way contrast is maintained (such as Punu (Nkutsi 1980)), the contrast between HH and HL is

frequently

far more problematic than the other tonal contrasts, both from the point of view of synchronic analysis, and from that of determining straightforward

etymologic-

al derivations; in Punu, for example, the contrast between HL and HH nouns corresponds poorly to the tones reconstructed by Guthrie (1969).

Prosodic Trends in Bantu

3.

85

Tone shift and consequent rise of 3-way contrast

The typology of Van Spaandonck (1967) emphasizes this most remarkable fact about the tone systems of the Bantu languages: It is the rare Bantu tone/accent system in which the underlying tone/accent characteristics of a given vowel or morpheme are realized on that vowel or morpheme (KiHunde, described in Goldsmith (in press b ) , is perhaps as close to being such a case as will be found). Instead, what one may find are systems like the following: 1. CiRuri (Massamba (1982, 1 9 8 4 ) ) . In CiRuri, an underlying High tone

(found,

for example, on all object markers) shifts one syllable to the right, unless it is in word-penultimate position, in which case, it remains on its vowel. Thus:

(3)

o ο o ο

ku ku ku ku

gul a gu gul a tern a tern er a

'buy' 'to buy it 1 'to cut' (same underlying form) 'to cut for someone' (with shift)

2. Digo (Kisseberth ( 1 9 8 4 ) ) . In Digo, a single High tone found on any syllable of a word shifts to the final vowel of the word. If the word is not final in its

syntactic phrase, the High tone shifts again to the final vowel of the

following word. Thus, in (4a) we have a Low-toned infinitive, while in (4b) the Final High tone (actually pronounced as a Rise-Fall pattern over the last two syllables) is associated with the 3rd person person singular Subject Marker underlyingly. Finally in (4c) , we see the High tone continuing its

traveling

down to the following object; the object nguwo would have a Low tone in other contexts (as it does underlyingly).

(4)

a.

ku-gur-a

"to buy"

(Low toned stem)

b.

a ka gur SM Tense Root FV

"He has bought"

c.

a ka gur a nguwo.

"He has bought clothes"

3. In northeastern languages such as Kikuyu (Clements ( 1 9 8 4 ) ) and Chaga (McHugh ( 1 9 8 4 ) )

(in Guthrie's Group E50 and E60, respectively) we find phrase-

level displacement of the entire sequence of High and Low tones, in which the first tone of the phrase seems to associate with both the first and second mora, leaving the rest of the tones to associate in a one-to-one fashion.

86

John Goldsmith

(See Clements (1984) for a detailed analysis of the intricate Kikuyu system). Thus, to take a simple example, the word mo-te "tree" has its

surface pattern

derived as in ( 5 ) ; the Rising pattern on the final syllable shows, upon reflection, that Low tones function as full-fledged

segments on a par with High tones

in this operation. In a system like that of CiRuri, by contrast, the shifting never creates a sequence of High and Low (on the last syllable, for example) because there are only High tones, no Low tones, until the very end of the derivation.

(5)

mo

te

L

H

In several distinct languages a ternary contrast has arisen out of a splitting of etymologically High toned syllables. In Rimi, for example, Olsen (1964) has noted that vowels divide into three groups: low-toned vowels, High toned vowels, and those that place a High tone on the next syllable. Schadeberg (1979) has noted that the contrast between High toned vowels and those placing a High on the next syllable corresponds precisely to the etymological distinction of long versus short vowel, respectively. A similar contrast is found in Digo (Kisseberth ( 1 9 8 4 ) ) , where it is normally the case that lexical High tones found in wordpenultimate position become Low (since — as opposed to grammatical tones — lexical tones in the verb can only appear on the first vowel of the stem, this rule will only apply to monosyllabic stems). However, High tone stems which were historically long are synchronically exceptions to this rule. A similar three-way contrast is found in the more conservative Nguni languages

(see

Rycroft ( 1 9 8 0 ) ) . It appears that the typology of tone shifts established by Van Spaandonck (and utilized by Carter ( 1 9 7 3 ) ) have in a sense been undermined by theoretical advances, in the sense that there are several quite different ways that a Bantu tone system can grammaticalize the superficial effect of a tone shift. There are at the very least three kinds of systems: first, those like Sukuma, where the original High toned syllables are now associated with the Low tone of a Low-High melody, where the following High associates to the following syllable (as in (1) above). Second, there are those like Kikuyu or Chaga, where a postlexical rule in effect pushes the entire phrase-level sequence of Low and High tones to the right by one mora. Third, there are systems like CiRuri, where there is no evidence of Low tones playing any role at all

in the system.

Prosodic Trends in Bantu

4.

87

Metrical structure

While there is no evidence that metrical structure played a role in early Bantu, the picture has changed in many of the present-day languages. Metrical structure has arisen in two distinct ways in the present-day languages, ways which we shall discuss briefly in this section. In both cases, the strong positions of the metrical structure act as points of attraction for the High tones of the tonal pattern.

(1) In several languages, metrical structure has arisen

at the end of the word, making the penultimate and/or the antepenult syllable attractive to the tone or accent of the word, inducing the High tone to shift to the end of the word. (2) In other languages, a metrical structure has arisen from the middle of the word. In the Middle Lake family of KiRundi and Kinyarwanda, there is evidence that a different kind of metrical structure has arisen in which alternating Weak and Strong syllable positions radiate outward from a High tone (in nouns) or certain Tense Markers (in verbs) — both due, we may infer, to a single principle at an earlier stage (Furere and Rialland ( 1 9 8 3 ) , Goldsmith in press c ) . A similar process has been observed in Kimatuumbi (Kisseberth and Odden ( 1 9 8 0 ) , Pulleyblank ( 1 9 8 2 ) , Odden ( 1 9 8 2 ) ) and in Makua (Cheng and Kisseberth (1979, 1980, 1981)). Why this should be — why "rhythmogeny" should be a special Bantu phenomenon among Niger-Congo languages — is unclear. The most straightforward hypothesis, and the one that this writer suspects, is that the enriched morphology of the Bantu languages, both derivational and inflectional, is the single most important factor. The richer segmental morphology led directly to longer words, and this, in turn, is a natural necessary condition for rhythmic structure. Perhaps it is a sufficient condition as well. It was noted above that metrical structure at the end of words has led to systems where the penultimate and/or antepenultimate syllables are singled out for special prosodic attention, affecting both syllable length and tone. Bantu grammars (especially of the southern languages) have long noted a process of penultimate syllable lengthening ( e . g . , in Zulu and Xhosa). In other Bantu languages, there are restrictions on the surface realization of underlying length contrasts. For example, in Chi-Mwiini (Kenstowicz and Kisseberth ( 1 9 7 7 ) ) vowel length is underlyingly contrastive in any position, but may surface only in the penultimate or ultimate syllable. In at least one dialect of KiHunde (Goldsmith in press b ) , length is also contrastive underlyingly, but may surface only in the phrase-penultimate position. Such a system is no doubt likely to lose its

vowel length contrast; indeed, Mateene (1967) reports that his dialect

88

John Goldsmith

of KiHunde has no length contrast, though he does not discuss the many and various tonal repercussions that such a loss, if it has indeed occurred, would entail, since the length contrast holds in the word-level phonology in the first dialect, and plays an important role in the tonal rules there. In both the very southern Nguni languages (including Zulu, Ndebele, and Xhosa), and the northern Digo (Kisseberth ( 1 9 8 4 ) ) , there is a very striking tone shift rule that shifts a High tone to a metrically strong position in the final foot (three syllables) of the word when that final foot has no High tone associated with it. As reported in unpublished work by Drogo, Peterson and this writer (reported now in Peterson

( 1 9 8 5 ) ) , tone in Xhosa is underlyingly associated with the first

mora of the tone-providing morpheme, as in other tonal Bantu languages. However, there is a rule that assigns accent on the verb stem to the antepenultimate syllable, (and a minor rule that shifts this accent one syllable rightward in a subclass of High toned stems marked [ +Accent S h i f t ] ) , followed by a rule that spreads High tone to the right by attraction to this accent. At least two indications weigh in favor of viewing this as tonal attraction to accent, rather than as attraction to the antepenult position. First, if the stem is shorter than three syllables, the attraction is to the first position of the stem, suggesting a pattern familiar from accent assignment systems (assuming accent is assigned on the stem cycle). Second, the specially marked [+Accent Shift] stems shift the accent to the penult if it is assigned first to the antepenult, and to the ultima if it is assigned to the penult. The intricacy of this pattern is difficult, if not impossible, to describe in a set of rules if no recourse is available to an explicit element in the representation which can be shifted by rule, and which later in the derivation serves as the point of attraction to the tone. A rather different metrical structure is found in the Middle Lakes languages of KiRundi and Kinyarwanda. In Kinyarwanda, Furere and Rialland (1983) point out that in the nominal system, an underlying High tone (on the stem) is realized on the surface with a sequence of either two or four High toned vowels in a sequence leftward from the underlying position, spreading far enough to cover the first vowel of the stem (and, if the underlying position is on an odd-numbered vowel of the stem, also the vowel of the noun-prefix). In the neighboring KiRundi verbal system (and almost certainly in Kinyarwanda as w e l l ) , an abstract metrical system is required to account for the fact that several tone rules shifting High tones one mora to the left or right do so in a manner that consistently shifts the High to an odd-numbered mora counting from after

Prosodic Trends in Bantu

89

the Tense-Focus Marker (Goldsmith (in press c ) , Goldsmith and Sabimana ( 1 9 8 5 ) ) , which can no doubt be best analyzed as the result of metrical structure already established on the word. A similar rhythmic echoing of High tone on alternating syllable to the left of the underlying position of accent or High tone (which depending on the analysis) is found in Kimatuumbi, as noted by Odden (1982), and further discussed by Pulleyblank (1982).

5.

"Culminative" character, or one High tone per word

There is in general no restriction on the number of High tones per word, either underlyingly or on the surface, in the Bantu group as a whole. However, a number of processes do point in the direction of such restrictions arising. A clear example is found in KiRundi, where in focused (main clause) verbs, more than one High tone may appear, but where in non-focused verb forms (a category including negative and subordinate verb forms), only the leftmost of the morphologically assigned High tones surfaces. This effect can be seen in the following sort of interaction. In non-focus verb forms, a High tone is assigned to the second mora of the stem. A High tone is,

furthermore, assigned

by a general rule to the second Object Marker whenever there are two Object Markers in the verb, and, in addition, the Reflexive Object Marker always bears a High tone underlyingly. These three principles all operate together in the subordinate form, and whichever assigns the leftmost High tone will have

its

effects manifested on the surface; all other High tones will be lost (Goldsmith and Sabimana ( 1 9 8 5 ) ) .

6.

The domination of Basic Tone Melodies

In a wide range of languages a trend is found towards replacement of an original Low tone by no tone at all

(one finds in Stevick (1969) , in fact, the hypothesis

that this was the original situation in Proto Bantu). This shift is most noticeable in its effects on encouraging tone spreading, that is, the spreading of one tone over a span of several vowels. Recent discussions in the literature on autosegmental phonology leave it an open question as to precisely in which sense it is predictable that tone spreading will occur when vowels (or tone-bearing units) are unassociated and other tones stand nearby, ready to associate.

90

John Goldsmith

(Cf. Pulleyblank (1983) for the strongest set of arguments against totally automatic and universal reassociations under these conditions, arguing against Goldsmith ( 1 9 7 6 ) ) . Whatever the upshot of the ongoing debate turns out to be, it is clear that toneless vowels represent an especially attractive domain for a tone to associate to. In Tonga, for example )see Goldsmith (1982, 1984b) and also Carter ( 1 9 6 2 ) ) the yerb and the following object may form a phonological word under several conditions, including focus on the object. In this case,

all

toneless vowels at the end of the verb associate with a High tone that may spread leftward from the following object, as in ( 6 ) .

(6)

V C V C V

C

Very similar processes occur, inter alia, in Shi (Polak-Bynon ( 1 9 7 5 ) ) , Sukuma (Goldsmith in press a ) , and Luganda ( c f . , e . g . , Hyman ( 1 9 8 2 ) ) , and the kind of restructuring suggested in (2) above is a wide-spread phenomenon. The significance of this point is discussed at greater length in the introductory chapter of Clements and Goldsmith (1984).

7.

Concluding observations

Ending on a speculative note, it

is tempting to consider the parallelisms that

appear to exist between the prosodic systems of conservative Bantu languages and certain characteristics of the vowels in the verb stem in the Bantu languages. The first vowel of the stem of the verb (as defined in (1) above) originally had two characteristics which set it apart: it could bear its own lexically contrastive tone, and it could be underlyingly any of the seven vowels of the language (as it is presently reconstructed). The extensions, by contrast, showed greater restrictions as far as both tone and vowel possibilities were concerned, at the underlyingly level, and correspondingly were more subject to the assimilatory influence of the phonological elements in their environment. In more concrete terms., this means that in the seven vowel systems (with the vowels j , i, e, a, o, u, and y ) , no extensions are reconstructed with the underlying vowels /e/

or

/o/. Similarly, in the languages where the seven vowels have reduced to five, no mid vowels are found underlyingly in the extensions. On the other hand, concerning these vowels "vulnerability" (as one might put i t ) , there is a widespread process of lowering of the vowels /i/

and /u/ to mid /e/ and /o/ when

Prosodic Trends in Bantu

91

the first vowel of the stem is mid (in some languages, when it is mid or low). Though the general range of the phenomenon is less clear, there is also a general process by which the vocalism of the Final Vowel (which is not an extension, of course) influences the extensions regressively. This is discussed generally in Gregoire (1979) , and other aspects are developed in Bastin (1983). One cannot but be struck by the fact that the two positions of the verb stem — the first vowel and the last -- which can independently bear tone are those whose vowels are least immune to assimilatory processes (or, in the autosegmental terms discussed in Goldsmith (1985), spreading of the features on the vocalic tiers), while those vowels without inherent tone -- the extensions — are also those with fewer underlying contrasts and those subject to more processes of assimilation. There appear to be (at a reasonably abstract level, to be sure) certain parallels in these respects to the Altaic languages, and it may well be the case that a deeper understanding of these tonal and vocalic restrictions will have to be based upon a metrical prominence of the first and last vowel of the stem at an early stage of the language. This prominence of the peripheral vowels over the internal ones may well have led in turn to the differences sketched in this paragraph. As I noted in the first paragraph, there is a great deal of ongoing work in this area today, and no doubt each of the points touched on in this paper will be far better understood in a short time. It is perhaps a matter of good fortune for students of Bantu prosody that the current problems in this area tie in so directly with theoretical issues that are currently central to generative phonology.

References Bastin, Υ. (1983), La finale verbale -ide et 1'imbrication en bantou. Tervuren: Mus'ee Royal de l'Afrique centrale. No. 114. Bennett, P.R. and J.P. Sterk ( 1 9 7 7 ) , South-Central Niger-Congo: a reclassification. Studies in African Linguistics 8, 241-273. Bouquiaux, Luc, Larry Hyman, and Jan Voorhoeve, eds. (1980), L'expansion bantoue: actes du collogue international du CNRS, Viviers, 4-16 avril 1977. Paris: SELAF. Carter, Hazel ( 1 9 6 2 ) , notes on the Tonal System of Nothern r.hodesian rlateau Tonga. London: Her Majesty's Stationery Office. Carter, Hazel ( 1 9 6 2 ) , Tonal data. In:Comparative Bantu. African Language Studies 14, 36-52. Cheng, C.-C. and C . W . Kisseberth ( 1 9 7 9 ) , Ikorovere Makua Tonology (Part 1). Studies in the Linguistic Sciences 9.1. Cheng, C.-C. and C.W. Kisseberth ( 1 9 8 0 ) , Ikorovere Makua Tonology (Part 2).

92

John Goldsmith

Studies in the Linguistic Sciences 10.1. Cheng, C.-C. and C.W. Kisseberth ( 1 9 8 1 ) , Ikorovere Makua Tonology (Part 3 ) . Studies in the Linguistic Sciences 11.1. Claughton, John (1983), The Tones of Xhosa Inflection. Communication no. 13. Grahamstown: Rhodes University. Clements, G . N . ( 1 9 8 3 ) , On some parameters of variation in tone languages. MS. Clements, G . N . ( 1 9 8 4 ) , Principles of Tone Assignment in Kikuyu. In: Clements and Goldsmith. Clements, G . N . and Kevin Ford ( 1 9 7 9 ) , Kikuyu tone shift and its Syncronic Consequences. Linguistic Inquiry 10.2. Clements, G . N . and John Goldsmith, eds. ( 1 9 8 4 ) , Autosegmental Studies in Bantu Tone. Dordrecht: Foris Publications. Furere, Rutinywa and Annie Rialland ( 1 9 8 3 ) , Accent tonal en Kinyarwanda. In: Current Approaches to African Linguistics (vol. 2 ) , ed. Jonathan Kaye et al. Dordrecht: Foris Publications. Goldsmith, John ( 1 9 7 6 ) , Autosegmental Phonology. Doctoral dissertation, MIT. (Published 1979, New York: Garland Press.). Goldsmith, John ( 1 9 8 2 ) , Accent Systems. In: The Structure of Phonological Representations, edited by H. van der Hülst and Norval Smith. Dordrecht: Foris Publications. Goldsmith, John (1984a). Meeussen's Rule. In: Language Sound Structure, ed. Mark Aronoff and Richard Oehrle. Cambridge: MIT Press. Goldsmith, John (1984b), Tone and Accent in Tonga. In: Clements and Goldsmith. Goldsmith, John (1985), Vowel Harmony in Khalkha Mongolian, Yaka, Finnish, and Hungarian. In: Phonology Yearbook 1985, Cambridge University Press. Goldsmith, John. In press a. Tone in Sukuma. In: Studies in African and General Linguistics, ed. Didier Goyvaerts. Amsterdam: John Benjamins. Goldsmith, John. In press b. Tone in KiHunde. In:Phonologica Africana 1984, ed. John Rennison. Goldsmith, John. In press c. Rise of Rhythmic Structure in Bantu. In: Phonologica 1984, ed. W. Dressler. Goldsmith, John and Firmard Sabimana. The KiRundi Verb. MS. Greenberg, J.H. ( 1 9 6 3 ) , The Languages of Africa. The Hague: Mouton. Gregoire, Claire (1979), Les Voyelles finales alternantes dans la conjugaison affirmative des langues bantoues centrales. Journal of African Languages and Linguistics volume 1, number 2. Guthrie, Malcolm ( 1 9 6 2 ) , The status of radical extensions in Bantu languages. Journal of African Languages 1, Part 3. Guthrie, Malcolm (1967-1971), Comparative B a n t u . Four volumes. Gregg International Publishers, Farnborough, Hants. Hyman, Larry M. (1982) , Globality and the Accentual Analysis of LuGanda Tone. Journal of Linguistic Research 2 . 3 . Kenstowicz, Michael and Charles Kisseberth ( 1 9 7 7 ) , Topics in Phonological Theory. New York: Academic Press. Kisseberth, Charles ( 1 9 8 4 ) , Digo Tonology. In: Clements and Goldsmith. Kisseberth, Charles and David Odden (1980), Aspects of Tone Assignment in Kimatuumbi. Studies in the Linguistic Sciences 10.1. Massamba, David (1982), Aspects of Tone and Accent in Ci-Ruri. Doctoral dissertation, Indiana University. Massamba, David (1984), Tone and Accent in Ci-Ruri. In: Clements and Goldsmith. Mateene, Kahombo (1967) , Le changement des phonemes du hunde ä partir du Systeme phonologique du proto-bantou. Africana Linguistics III. McCawley, James (1978). What is a Tone Language? In: Victoria Fromkin, ed., Tone: a Linguistic Survey. New York: Academic Press. McHugh, Brian ( 1 9 8 4 ) , Phrasal Tone Rules in Kirua (Vunjo) Chaga. Ms., UCLA.

Prosodic Trends in Bantu

93

Nsuka-Nkutsi, F . , ed. ( I 9 6 0 ) , Elements de description du punu. Lyon: Centre de recherches linguistiques et semiologiques. Odden, David ( 1 9 8 2 ) , An accentual approach to tone in Kimatuumbi. In: Studies in African and General Linguistics, ed. Didier Goyvaerts. Amsterdam: John Benjamins. Olsen, H. ( 1 9 6 4 ) , The Phonology and Morphology of Rimi. Hartford: Hartford Seminary. Peterson, Karen (1985). Tone and Accent in the Xhosa infinitive and present indicative. University of Chicago Department of Linguistics Working Papers NO. 1, Philippson, David ( 1 9 7 7 ) , The Spread of the Bantu Languages. Scientific American 236, 106-114. Polak-Bynon, Louise ( 1 9 7 5 ) , A Shi Grammar. Tervuren: Musoe royal de l'Afrique Centrale. No. 86. Pulleyblank, Douglas ( 1 9 8 2 ) , Accent in Kimatuumbi. In: Current Approaches to African Linguistics (vol. 2), ed. J. Kaye, H. Koopman, D. Sportiche, and A. Dugas. Dordrecht: Foris Publications. Pulleyblank, Douglas (1983), Tone in Lexical Phonology. Doctoral dissertation, MIT. Rycroft, David (1980), Nguni Tonal Typology and Common Bantu. African Language Studies 4, 43-68. Schadeberg, Thilo (1978), Ueber die Toene der nominalen and pronominalen Formen im Rimi. Afrika und Uebersee LXI/3, 189-209. Schadeberg, Thilo ( 1 9 7 9 ) , Ueber die Toene der Verbalen Formen im Rimi. Afrika unde Uebersee LXII/4, 288-313. Stevick, Karl (1969), Tone in Bantu. IJAL 35, 330-341. Spaandonck, M. van ( 1 9 6 7 ) , Morfotonologische Analyse in Bantutalen. Acaderaische Proefschrift, Rijksuniversiteit te Ghent. French translation: L'analyse morphotonologique dans les langues bantoues. Paris: SELAF 23-24, 1971. Williamson, Kay (1982). The Classification of East Niger-Congo. JOLAN 1 [Journal of the Linguistics Association of Nigeria].

*This material is based on work supported by the National Science Foundation under grant no. BNS-8421245.

Adequacy in Intonation Analysis: The Case of Dutch Carlos Gussenhoven University ofNijmegen 1.

Introduction

Linguistic form does not reflect linguistic content in a direct fashion. Languages frequently possess forms that are phonetically similar or identical, but semantically very dissimilar,

like English box - books and tacks and tax

respectively. This lack of direct correspondence has given rise to descriptive models that separate the concrete, phonetic level of description from a more abstract one, which purports to capture linguistic structure on the basis of one form per one morpheme, and thus to represent linguistic content. However, the necessity for this separation of levels in the description of intonation is not always as fully appreciated as it is in the description of the segmental and tonal phonology of languages.

Yet in intonation, too, the sort of phenomena

that have led to the development of models that have rules mediating between more and less abstract representations of linguistic forms, are fairly easy to find. Consider, for example, the Dutch contours (1) and ( 2 ) .

(1)

Heeft u de jafazelen gehad*? Have you the measles had

(2)

Heeft u de ma/elen qehad?

In ( 1 ) , the rise occurs early in the accented syllable ma-, and the end of the contour tends to trail down a little. In ( 2 ) , the rise occurs late in, or after accented ma-, while there is no sagging at the end. Pragmatically, these two reasonably similar pitch contours are very different. Ex ( 1 ) sounds as if

it

was taken from a story read to children; indeed, one would expect the example to be followed by some such reporting clause as said Mr. Owl to Rabbit. Second, (1)

is a question that reveals genuine interest on the speaker's part. It

is

easy to imagine a hearer showing a typical post-measles rash, say. By contrast, (2)

is not reminiscent of story-telling at all.

Also, (2) sounds likn a routine

question, possibly asked by a speaker filling in a form recording the medical history of the umpteenth applicant for some physically sensitive job. On the

96

Carlos Gussenhoven

basis of these indications, (1) and (2) may be expected to be structurally different to an extent not suggested by their phonetic appearance. There is, to be sure, also a formal indication that this is in fact the case. When we add a reporting clause to the examples, we find that the accentless addition has a downward trajectory in ( 1 ) , but continues the level pitch in ( 2 ) . The purpose of this paper is to show that the descriptive model developed for English in Gussenhoven (1983, 1985), in which two levels, one phonetic and one morphological, are recognized, can successfully be applied to Dutch. The analysis is presented in section 2. In section 3 it is shown that this description is in fact capable of accounting for the data in the most extensive description of Dutch intonation to date (Collier & 't Hart 1975) in sometimes illuminating ways, and that it effortlessly characterises intonational

contrasts

which are not adequately represented in 't Hart & Collier's description. This section thus amounts to a reanalysis of their data, one which is far more comprehensive than the summary reanalysis of Ladd (1983), from which it also differs in certain ways.

2.

The analysis

Although we will note certain differences between the two languages, the intonation of Dutch can be described in terms of the same phonological units as that of British English. Most importantly, the morpheme inventories would appear to be the same. At the most abstract level of representation there would have to * * * be three morphemes (also called ' t o n e s ' ) , HL, HLH and LH, one of which is obligatorily chosen at each accented position in the segmental string. There is therefore a one-to-one correspondence between accents on the tonal tier and accents on the segmental tier (both indicated by *). All other contours of Dutch are derived from these tones by phonological operations, to be dealt with below. If no operations apply, these tones surface in a straightforward way. In spirit, the association of tones conforms to Goldsmith's (1976) WellFormedness Condition (WFC), here interpreted as in ( 3 ) .

(3)

Well Formedness Condition a. Association is exhaustive; i.e. no segments on either tier are left unassociated. b. Association lines do not cross.

Adequacy in Intonation Analysis

97

Unaccented stretches before an accent can have a variety of pitch configurations, low-pitched being the unmarked situation

(unlike the situation in British

English, which tends to have mid pitch, de Pijper 1983:101). If we want to * * avoid a situation in which HL and HLH are always preceded by high pitch and * LH always by low pitch, we need to stipulate that the accented position acts like a solid wall that cannot be penetrated by the tone. A condition to the iteffect that T cannot spread left, would of course achieve this. But in view of the fact that no tone elements appear to the left of the accented element in any of the three tones, such a convention and ( 3 a ) would be contradictory to the extent that any segmental material to the left of a first accent is left unassociated (by the convention), in violation of ( 3 a ) . Therefore, it

seems

preferable to revise the first clause of (3) as in ( 3 ' ) . (31)

a. Association is exhaustive, but only from * onwards to the domain end

With respect to the pre-tonal stretch, we will simply assume that there is a variable ONSET that specifies the pitch of this stretch (also known as the 'prehead 1 , Crystal 1969:233). No attempt will be made to define the domain of a tone, referred to in ( 3 1 ) - In the few examples in this article in which a domain ends in the middle of a sentence, it conveniently coincides with a syntactic phrase. Convention ( 3 1 ) is operative whatever options (rules or modifications) are applied. They do not, however, suffice to give us a fully associated contour. The association of intonational tonal elements is not, unfortunately, governed by tone-bearing units in the way the associations of tones in tone languages are measured out in units like the syllable or the mora. The absence of tonebearing units where intonational tonal elements are concerned is clearly suggested by the work of 't Hart s Collier, who found that both rises and falls could be classified on the basis of (among other things) their timing with respect to the syllable (arbitrarily measured with respect to the onset of voicing). For example, they found that a particular type of fall

( ' B ' ) typically comes

very early in the syllable or ' . . . even so early that it can be considered to fall in between the (voiced parts of) two syllables' ('t

Hart & Cohen 1 9 7 3 ) ,

or that one type of rise ( ' 3 ' ) 'is located typically late in the syllable' (as opposed to another type, which occurs typically rather early)

( ' t Hart &

Collier 1975). These facts suggest that tonal elements (H, L, etc.) are associated with points in the time domain that cannot be equated with phono-

98

Carlos Gussenhoven

logical units in the segmental string. Another indication that there are no tone-bearing units is the absence of any restrictions on the number of tonal elements that associate with a syllable or mora: restrictions of this sort are a fairly common situation in tone languages (Maddieson 1978). That is, a * Dutch one-mora syllable like dit ( ' t h i s ' ) can be associated with HLH as easily as can any initially-accented three-syllable word. Clearly, this situation makes it impossible to formulate an association convention in the usual sense, since while we can refer to units on the tonal tier, we cannot refer to any units on the segmental tier with which the tonal units could be paired. On the other hand, two reference points are available. First, the accented syllable - or rather some as yet unspecified timing point in the starred syllable - and, second, the end of the domain onto which the tone is mapped. Spreading, moreover, remains a meaningful concept. We could provisionally stipulate ( 4 ) , which formulation ensures correct association of the three tones to arbitrary segmental strings.

(4)

Only the second tone element of a tone can spread

* In (5) we illustrate the three morphemes with De ooievaar is er ('The stork has arrived") as the segmental· string. For the onset parameter low pitch is assumed.

HLH A second, perhaps self-evident difference between the way our intonational tonal elements are associated with the segmental· string and tone associations in tone languages is that in the former case the sentence contour is associated wholesale, after assembly of both the segmental string and the tonal string in separate phonological· compartments (Gussenhoven 1985). This means that rules of intonational· phonology cannot distinguish between associated and free tonal· el·ements. The next three subsections are devoted to three mutually exclusive operations. These are the Tone Linking Rule ( 2 . 1 ) , Downstep ( 2 . 2 ) and Narration ( 2 . 3 ) . The first two rules necessarily apply to sequences of tones, while the third can also apply to single tones. Subsection 2.4 discusses three modifications, Half-completion, Delay and Stylisation. The difference between the three rules

Adequacy in Intonation Analysis

99

and the modifications would appear to be that a rule always causes a domainend boundary to be deleted, while a modification never does. The rules would thus appear to be related to the formation of larger intonational domains, like the tone unit (Crystal 1969:204). The modifications, however, are concerned with single tones only, and are responsible for the wide array of nuclear tones (or tone variants) that exist in Dutch. 2.1.

The Tone Linking Rule

Final accents tend to have phonetically richer realisations than non-final ones. If it is agreed that the phonological status of an accent is unaffected by whether it is followed by other accents or not (Pierrehumbert 1980, Selkirk 1984:264), then either the realisation of non-final accents will have to be curtailed, or that of final accents enriched. Pierrehumbert (1980) opted for the latter solution, and designed a grammar of intonation in which a final accent, but not a non-final accent, is optionally followed by a 'phrase-tone', while the end of the contour is in addition marked by a boundary tone. Leben (1976) achieved a similar effect by allowing the nuclear-accented tone element to spread to non-nuclear accents. In our own description of the intonation of English, the former solution was chosen. In particular, a Tone Linking Rule (TLR) was proposed. TLR is an operation which can not only be stated in general terms, regardless of the particular combination of tonal morphemes to be linked, but also correctly expresses the intuitively apparent relatedness of different contours. In Dutch, too, this rule, here given as ( 6 ) , can be shown to be operative. The (a) and (b) contours in (7) to (10) are examples of pairs of related contours. More precisely, the (b)-contours would appear to be equivalent to the (a)-contours, except that they are 'said more quickly'

(and will also liter-

1

ally be shorter) than the somewhat 'plodding (a)-contours. Example (7b) links * * * * * * two H L ' s , (8b) links HLH and HL, (9b) two L H ' s , and ( l O b ) LH and HL.

->-

(6 /

(6)

T

T

(7)

a. Ze Verk\ # ,fiard\

('She works hard 1 )

b. Ze yferkt har\

(8)

a. In de ^oiner # hoef je niet te yroften ~~

( ' In the summer need you not to h e a t ' )

100

Carlos Gussenhoven

(9)

a. Moet_j|/# in diey^t? * * / b. Moet Jan in dienst?

('Must Jan in (military) service?')

(10)

a. In dat oeval # moet je even >5el\.en

( ' I n that case must you (particle) phone 1 )

b. In dat geval__n»et—J'e"~even beiden

Note that the pre-nuclear accent in (9b) retains a minor depression at the accented position, corresponding to I. It is assumed that low pitch specified * by L is more precise, and in practice lower, than low pitch in the onset and the connecting stretches between accents. A similar assumption about prenuclear 'downskipped' accents is made by Leben ( 1 9 7 6 ) . ^ The operation of TLR is very straightforward. The final tone element of the left-hand tone, together with its domain-end boundary, is deleted, causing the two tone elements that are now adjacent to be connected directly. In ( l l a ) the dotted line represents the contour before application of TLR, the solid line the contour produced by TLR. The rule accounts in a natural way for the relations noted above between the (a) and (b)-contours: the less elaborate pitch configuration of the prenuclear accent is literally due to its being linked to the nuclear accent. The naturalness of this solution is further underlined by the possibility of in-between positions. When the deletable tone element is shifted

rightwards

('partial l i n k i n g ' ) , rather than deleted outright ('complete linking), the contour illustrated in ( l i b ) is produced: the domain-boundary is deleted.

(11)

(boundary deletion and TLR)

a. HL

(boundary deletion without TLR)

2.2.

Downstep

Downstep, described for English in Pierrehumbert (1980), Ladd (1983), Liberman S Pierrehumbert ( 1 9 8 4 ) , also occurs in Dutch, and is exemplified in ( 1 3 ) . The contour should not be taken to illustrate declination. As stressed by Ladd (1984), 'the ...

point is also that there may be phonological rules that lower

pitch in specific phonological environments (e.g. 'downstep' . . . ) , which means that trend lines may drop more steeply than the frame of reference itself.

Adequacy in Intonation Analysis That is,

101

* without Downstep, the peaks in a sequence of H L ' s already display

a downward trend, which is, however, considerably weaker than the effect of Downstep. While the weaker effect can arguably be attributed to a contextindependent, gradually falling reference line,

Downstep is a phonological

operation applying in a particular context. The rule applies optionally over * sequences of HL, creating a terrace-shaped contour, with the steps at the positions of the accents. The rule would appear to consist of two separate * * operations. First, a H following a HL is downstepped (the lowering effect is * customarily symbolised as " ! " ) , and, second, every H except the final one spreads. 5 The rule could be formulated as in ( 1 2 ) .

(12)

DOWNSTEP

a.

0 -> I / H L

* b. Spread H /

H

* LH

* * The L after spread H is pushed on to the following H, where its effect may in * fact be masked by the step down to J H . An illustration, taken from Collier &

't Hart (1981), is given in ( 1 3 ) . As in the case of TLR, any tone domain boundaries, e.g. that between regelingen and zijn afgeschaft,

(13)

are deleted.

* * * v Al die ingewikkelde reqelingen zijn Afgeschaft

! HL

! HL

all those complicated regulations have-been abolished 2.3.

Narration

One of the more interesting phenomena in the intonation of Dutch is the existence of a stylistic rule, which we will call NARRATION, as application of the rule would appear to be confined to reading. Thus, it may be expected to occur in story telling as well as in news bulletins. The rule can simply be stated as in ( 1 4 ) .

(14)

Spread T

* * As an example, consider contour ( 1 5 a ) , an unlinked HL and LH. Application of

(14) would give us ( 1 5 b ) , a phonetically very different, yet

interchangeable

contour. With respect to the application of ( 1 4 ) , two things should be noted. * First, as was the case in Downstep, H spreads right through the domain boundary, here between alle zakken and met graan gevuld waren. Thus, it would generally

102

Carlos Gussennoven

appear to be the case that application of any of the three operations discussed here (TLR, Downstep, Narration) implies the deletion of the boundary separating the domains for any two adjacent tones subject to an operation. Second, as in * the case of Downstep, we assume that the tone element after T is displaced as * * a result of the spreading of T, not deleted. Indeed, in (15b) the H of LH shows up at the end. The preservation of such a displaced tone element in a nonfinal tone can be illustrated with ( 1 6 ) , where the displaced tone element is unlike the first tone element of the next tone.

(15)

a. En toen/aJj.e /aJj. zakken # met graan

HE

LH

and when all sacks with grain filled were (there was enough left in the barn to last them the whole winter)

b. En t o e n l l e zakken met qraan, gqvuld wargn

LH

(16)

En hoewel /niemand de finish' i\ge^aa\d hee/t

'

'

NARRATION cannot be applied to HLH, there being no contour ( 1 7 ) . Clearly, too much of the characteristic pitch configuration would be displaced to an unaccented position in the contour. The existence of rules like ( 1 2 b ) and ( 1 4 ) shows that association convention (4) does not hold for all tonal constructs. The solution to this problem is in fact suggested by the ungraramaticality of ( 1 7 ) .

(17)

Moeten we allemaal daar naartoe?

must we

all

there

(Ill-formed )

to-to?

The oddity in ( 1 7 ) would appear to be created by the fact that there two tone elements assume the function of final boundary tone. If we define a boundary tone as a tone that associates with the domain end and that will only spread if it has to, then (4) could be reformulated as ( 4 1 ) .

Adequacy in Intonation Analysis (4')

103

Only the last tone element of a tone can be a boundary tone

This revised formulation not only allows (12b) and (14) to be rules, it also correctly characterises ( 1 7 ) as ungrammatical, since here a non-final

* tone element (L) is forced onto the domain end. The assumption is that T does

not spread unless special provision is made for this by rule. 6 In view of the * a-clause of ( 3 1 ) such spreading of T is necessarily rightwards. In the new * formulation, the spreading of L in an underived HLH is now achieved through ιίthe default non-spreading of H, as well as by the fact that final H, because of the presence of the preceding L, does not have to spread in order to flesh itout the contour. In LH, by contrast H will associate with the boundary and spread left. Two further points must be made with respect to NARRATION. First, the spread* ing of T does not extend to comment clauses, reporting clauses, vocatives and the like. The rule thus provides evidence for some sort of boundary between what Bing (1979) calls O-domains 1 and syntactically more integrated unaccented material. I do not concur with Sing's view that these domains are served by a special paradigm of boundary tones, as such a paradigm, combined with the nuclear tones assumed by her, overgenerates. Rather, the O-domains are served by a copy of the unstarred tone element(s) of the nuclear tone. Second, the * * ONSET parameter is obligatorily low with HL, and obligatorily high with LH. Without this provision, the accented syllable would presumably not

sufficiently

stand out. Both aspects are illustrated in ( 1 8 ) . Note that the pitch trajectories of the reporting clauses confirm the analysis of (18a) as HL and that of * (18b) as LH.

(18)

a. 'Dat zal de/ooievaar wel weten',

HL

L

that will the stork (particle) know said the woodpecker

b. 'Moetpij die zakken dan niet vulleX?' , vroeg Jan aan Annelies LH H Must you those bags (particle) not fill asked Jan of Annelies

2.4.

Modifications

Modifications are cross-classified with tones. Each of the three tonal morphemes

204

Carlos Gussenhoven

is optionally affixed with a modification. The phonetic effect of such ation' is integrative, i.e.

'affix-

the surface manifestation of a modified tone cannot

be divided into a segment that represents the tone and a segment that represents the modification. Phonologically, the characteristics of modifications are fairly simple. Thus, HALF-COMPLETION can be represented as in ( 1 9 ) .

(19)

M/ T

That is, the modification changes the second tone element to M, a formulation that captures the phonetic generalisation that tones so modified do not cross the midline ( c f . ( 2 0 ) ) .

(20)

HMH

DELAY shifts the association target of the starred tone element to the right. Observe that it is not the accent in the segmental string that is shifted. Delay is also distinct from spreading: there is still only a single timing * point with which T associates. The segmental stretch immediately preceding the association target (corresponding to the accented syllable) is not specified * by T, but rather seems to be treated as a tonal gap, and is lowish both before * * H and before L. Illustrations are given in ( 2 1 ) .

(21)

* De oo

!vaar is er

HL

De ooievac

HLH

Although it would not seem uncommon in the speech of adult speakers to children, delay is much less frequently used in Dutch than in British English. In 't Hart & Collier's 1975 corpus of 1359 utterances contours like those in ( 2 2 ) belong to the 6% of utterances that their grammar does not account for, though delayed * * HL, together with narrated HL, is singled out for separate mention: "... some pitch movements occurred earlier or later in the syllable than is specified by their definition

( . . . ) to the effect that blocks had to be transcribed as e.g. * * 1C or 3A < i . e . a delayed HL> where the grammar invariably

( ' t Hart & Collier 1975). The contour labelled ' 3 C 1 in * their grammar, here given as ( 2 2 ) , is, in our terms, a HL to which both delay

predicts 1A and 3C"

Adequacy in Intonation Analysis

105

and spreading have applied: the H appears late in or after the accented syllable, * while the high plateau reveals spreading of T, pushing L to the end of the domain.

(22)

*

I

~

^

En waarom /heeft h i j net niet gedaan?

While ( 2 2 ) is thus both phonetically and phonologically akin to a narrated HL, it is not at all suggestive of story-telling. The communicative impact of this rather distinct, and easily recognizable contour is difficult to pin down, but a connotation of 'self-evidence' seems clearly present, which may account for the fact that the contour is not uncommon in TV commercials. As it happens, ( 2 2 ) is the normal realisation of fUL in the local accent of Amsterdam, and seems similar to the fall-variant described by Knowles (1984) for northern British English. But other than in, say, Liverpudlian English, in Amsterdam Dutch it 'carries a certain measure o f , perhaps covert, prestige. On the basis of the above considerations, we assume that ' 3 C ' is a marginal tone of the standard language, taken over from Amsterdam Dutch. This explains its

isolated position

in the tonal system of the language.

STYLISATION (the term and the concept are from Ladd 1978) is implemented somewhat differently

from tone to tone. In general, the modification consists

of a lengthening, at least in sentence-final positron, of all the syllables in the domain of the tone, starting from the accented one, and plateau formation. The interesting tone here is HL. For this tone, plateaus are formed like a descending series of steps, beginning at the accented syllable and further at every strong syllable within the word and at every word-accent position in a new word, while the last syllable, regardless of its prosodic status, always has a plateau to itself. If the last syllable is also the accented one, it

is

broken into two sections by the extra, final plateau. The pitch of the first plateau would appear to fall below that of the topline (i.e. be lower than the peak of an unmodified HL, assuming equivalent prominences), and that of the last plateau to be higher than the base line. In longer tails, a proliferation of plateaus may be formed, a feature that distinguishes the Dutch stylised fall from the English one, which always has only two plateaus. For the purposes of the stylised f a l l , weak syllables include the syllable immediately after the accented one, and any reduced syllables after that. In ( 2 3 ) , the modification is illustrated on (a) a monosyllable, (b) a four-syllable word with a strong third syllable ( / v a : r / ) , and

106

Carlos Gussenhoven

(c) a four-syllable word with a weak third syllable ( / 3 m / ) . DIM stands for the diminutive suffix.

(23)

* a. Ja-än John

* ~~~~ -,—·· b. ooie-vaar-TfJe stork+DIM .

* c. valium-p^e valium+DIM it

Because of the word boundary, the compound mee-duwen (/me:dyw3/ 'with-push', i.e.

'help pushing') has three plateaus, while the segmentally-identical monoit· morphemic weduwe (/we:dyw3/ 'widow') has two. A number of closed-class words (articles, auxiliaries, modal adverbs like maar, toch, niet)

do not count as

separate words, and share the preceding plateau. Thus in ( 2 4 ) , niet shares the plateau of the accented syllable moet, meteen has the word accent on the last syllable, and the article het continues the plateau started on that syllable; antwoord has the word accent on the first syllable, as does geven. As a result, five plateaus are formed. Only two are formed in ( 2 5 ) : the auxiliary zijn continues the first plateau, and the strong but unaccented klaar gets the last.

(24)

(25)

* —— yife moet niet meteen het antwoord ge-ven! you must not at-once the answer give

Dejöatatten zijn klaarl the french-fries are ready

* * Stylised HLH is identical to stylised HL, except that after the first plateau

a single, low plateau is formed, the final H associating with the domain-end, as illustrated in ( 2 6 ) . Thus, monosyllables are broken up again; if there is a strong syllable ( b ) , the plateau division will occur before it;

if there is

not ( c ) , the second plateau goes to the last syllable. Example (26c) does not violate ( 3 1 ) : it forms a plateau on / t j 3 / , and refuses to be forced onto the domain end.

(26)

* / a. Ja-an/

* ./ b. ooie-vaar^e

* / c. valium-pje./

* * Observe that if HLH was merely a variant of HL to which a boundary H was added, * * an analysis favoured by Keijsper (1984), we would expect HL and HLH to behave

identically under stylisation. The fact that they do not, supports the view * * that HL and HLH are separate tonal morphemes. * * Stylised LH, like English stylised LH, is simply a mid level tone, extending

Adequacy in Intonation Analysis

107

from the accented syllable to the end of the domain.

3.

Two analyses compared

The most important research technique used in the development of the grammar of Dutch intonation (henceforth GDI; Cohen & r t Hart 1967, 't Hart & Cohen 1973, 't Hart & Collier 1975) is the perceptual evaluation of 'stylised 1 contours, either informally by the researchers or in experimental situations by naive subjects ( c f . de Pijper 1983:ch2). 'Stylisation' here refers to the ironing out of involuntary fluctuations in the pitch contour by means of analysis-resynthesis techniques. 7 When the contour was maximally 'stylised' while still being 'perceptually equivalent 1 to the original contour, and therefore all of its pitch characteristics could be considered 'voluntary' and hence linguistically relevant, it was analysed in terms of the monotonic pitch segments it consisted of. These segments were classified on the basis of rate-of-change and timing criteria ( e . g . a gradual fall over a number of syllables, a steep rise late in the syllable, e t c . ) , and labelled alphabetically (rises) or numerically ( f a l l s ) , while level high and level low stretches were referred to as 0 and Ο respectively. The slowly declining 0 and Ο are used as the topline and the baseline delimiting the pitch span within which the rises and falls are contained. GDI accounts for the order in which the pitch movements can occur by giving formulas for 'pitch blocks'

(fairly integrated stretches of contour

containing one or two accents), while blocks combine to form 'contours'. The various contours produced by the grammar are regarded as belonging to superordinate contour-types or 'basic intonation patterns' (e.g. the hat-pattern). Thus, the choice of pitch movements to realise a particular accent in a sentence is seen as being in part determined by the basic pattern chosen, the variation within a pattern being ' f r e e 1

( ' t Hart & Cohen 1973, 't Hart & Collier 1979).

Table 1 lists the pitch movements in the GDI (from Collier & 't Hart 1981). A box represents a syllable, the larger, middle section representing the syllable nucleus and the flanking narrower sections representing the syllable margins. The dotted top and bottom lines of the boxes for ' 4 ' and 'D' indicate sequences of syllables ( i . e . these movements are gradual movements over longer segmental stretches). Dotted continuations of movements either indicate usual preceding or following pitch, as in the case of ' 5 ' , or potentially absent pitch during voiceless obstruents, as in the case of '!'. An ampersand indicates that both movements occur on a single syllable. There are of course many differences

108

Carlos Gussenhoven accent-marking rise, steep and early

non-accent-marking rise, steep and very late

accent-marking rise, steep and in the middle of the syllable

gradual rise, marking more than one syllable

short, extra rise, sometimes after 4

accent-marking fall, steep and in the middle of the syllable

non-accent-marking fall, at syntactic boundary, steep, between syllables

non-accent-marking fall, steep and very late in the syllable

gradual fall, usually not distinct from A in l&A

half-fall, weakly accent-marking if occurring in isolation, steep and early in the syllable. It is sometimes displaced rightwards.

Table 1: Relevant pitch movements in GDI

between GDI and the phonological analysis given in section 2 , both with regard to research technique and with regard to the conceptualisation of the structure of intonation. Perhaps the most important difference, however, lies in their respective theoretical pretensions. GDI stresses perceptually validated descriptions of attested contours. The phonological analysis tries to identify the structural units that lie at the root of these contours, and to explain the data in terms of these units. There are two ways in which the difference in approach can be brought out. We could first of all concentrate on the explanatory power of the phonological description, and demonstrate that information that GDI needs to specify, naturally follows from the phonological description. For instance, the information GDI lists in Table 1 for movement ' D ' , viz. that it

Adequacy in Intonation Analysis

109

is usually not distinct from Ά' in Ί & Α 1 , follows from TLR in the phonological 1»

*

description. As illustrated in ( l i b ) , partial linking of H L . . . H L (or ' l & A . . . l & A ' ) , produces such a slope segment at the position of the first Ά', and since 'partial linking' is not very different from 'no linking', the perceptual effect of the substitution is predicted to be fairly small. The other way in which the two approaches can be compared is by looking at how they characterise * * contrasts. For example, the sequence H L . . . H L may undergo either TLR or Downstep, * * which contours would both be natural on Voor niks gaat de zon op ('for nothing goes the sun u p ' , i.e. be clearly different.

'You don't get something for n o t h i n g " ) , but nevertheless If we abstract away from the effect of declination, the

fall on zon has the same range as the rise on niks in the TLR-ed contour ( c f . ( 1 1 ) ) , but only about half that in the downstepped contour. But in this contour, the step down from the top line to the beginning of the downstepped fall on zon, together with that fall itself, occupy the same span as does the fall in the TRL-contour. Unless GDI forces movement Ά' to do double duty here, giving it slightly different timings, or envisages the possibility of coupling two Έ1 movements, the downstepped contour cannot be characterised as different

from

the TLR-ed contour. In the following section, three examples will be given that make the difference in explanatory effect between the two approaches clearer. In section 3.2. it will be shown, again with the help of three examples, that the phonological analysis is capable of characterising intonational contrasts that GDI cannot adequately represent. 3.2.

Explaining the data * * Example 1. When LH is followed by H L ( H ) and TLR is not applied, there will necessarily have to be a pitch drop at the end of the domain of the first tone: we continue with an unmarked low onset for the stretch preceding the following * H of the second tone. While this return to the baseline is a descriptively "free" feature in the phonological analysis, GDI must encode it as a 'relevant pitch movement' (the ' B ' ) . 8 Observe that the phonological analysis explains, in a real sense, why ' B ' is not accent-marking, and why it will occur 'at a syntactic boundary between syllables', features that GDI was forced to just list. An illustration is

( 7 a ) , here repeated for convenience. GDI analysis is

given above the contour, the phonological analysis below

it.

110

(7)

Carlos Gussenhoven

a.

l&A

moet je even

i/

HL

Example 2. A configuration ' l & A ' (also known as a pointed hat) on an utteranceinitial syllable often lacks the rising half, the '!'. That is, Antwerpen, analysed as ' l & A ' , is in fact often pronounced just Ά' ( c f . ( 2 7 ) ) .

(27)

l&A

l&A

GDI correctly analyses both contours identically. Indeed most speakers of the language would not be able to hear the difference. The phonological analysis makes it clear WHY the rising part is often missing. Treating the contours as Λ

instances of HL, it predicts that if there is no appreciable segmental stretch to realise the onset-variable on, and no delay is applied, the contour begins * with H. The phonetic difference between the two contours in ( 2 7 ) is the result a difference in association timing which to all intents and purposes is negligible. GDI postulates a rise in such cases on the assumption that the (pointed) hat-pattern, because of its unmarked nature, is a basic intonational unit of Dutch. In the phonological analysis, the configuration ' l & A ' arises as a result of employing the least marked options for onset (which is * (which is H L ) .

' l o w ' ) , and tone

Example 3. The third example concerns movement ' 4 ' . According to 't Hart & Collier (1975) it may occur in Ί Β 4 5 & Α ' , here illustrated in ( 2 8 ) , and in Collier & 't Hart (1981) it frequently occurs in Ί&Α45&Α 1 , as illustrated in ( 2 9 ) . What would appear to require an explanation here is why ' 4 ' marks an accent in ( 2 8 ) , but does not do so in ( 2 9 ) .

(28)

l B

Ik hoef die neef LH

4-

nooit

-4 5&A.

over de vloer\te hebben

I

L

I need that nephew never more about the floor to have

Adeguacy in Intonation Analysis

(29)

l&A

111

-4 5&A

arvoor met belanfrrnk

Your age is

that-for

not important

The phonological analysis makes it

clear why ' 4 ' behaves differently in differ-

ent contexts. Crucially, in this analysis, ' 4 ' is not a unit. It arises in ( 2 8 ) JT

*

*

as a result of the application of TLR to LH HL, the ' 4 ' linking L and H. In * * * ( 2 9 ) , however, TLR has applied to HLH HL: ' 4 ' links L and H, L being the tone * element that has come to be adjacent to H after the deletion of H to its right. The comparison of ( 2 8 ) and ( 2 9 ) brings up a further point. Observe that in * (28) the '!' is used to represent the rise of LH, but that in ( 2 9 ) the '!' is * * used to represent the pitch movement from the baseline to the H of HL. GDI considers these movements to be the same. The phonological analysis predicts that, averaged over a suitable number of tokens, the '!' in (28) will occur later than the '!' in ( 2 9 ) . This is because in ( 2 8 ) the accented syllable is * associated with L (marking the beginning point of the pitch rise), but in ( 2 9 ) * with H (marking the end point of the pitch rise). If we keep the association time point constant, then obviously the rise in ( 2 8 ) should come as much later than in ( 2 9 ) as it takes for the rising pitch movement to be performed.

Interest-

ingly, the tape accompanying Collier & 't Hart (1981) contains a minimal pair contrasting ' I B 1 and ' 1 A ' , and in it

the prediction is borne out. In ( 3 0 ) the * syllables -na and ie- are associated with HL (the second accent has undergone * TLR), while in ( 3 1 ) these syllables are each associated with LH. In the pitch curves given in the diagrams,

the portions corresponding to the vowels < a . >

in vanavond and in ieder have been marked o f f . As is clear from the diagrams, the above prediction is borne out. It is stressed that the minimal pair is illustrative of the point at issue, not probative: more tokens would obviously be required to show that the difference is systematic.

(30)

ISA

1

λ

We gaan va/(av\nd in /i< ieder geval naar de schou ouftfcmrg

I/

HL

we go

(

HZ.

this-evening in any case to the

HL

theatre

212

Carlos Gassenhoven

Hz 120-

JH

804O

80

120

200

160

CSeC

(31)

We gaan vaiicfvond in lejaer geval naar de sc/nouwburg

\l

I

LH

Hz 120-

80-

\r

JH

r*v

a.

40

3.2.

HL

LH

80

120

160

200 CSec

Characterising contrasts

Inasmuch as the phonological analysis appears to be capable of explaining a number of intonational phenomena described in GDI, the approach taken seems a promising one. In this section it will be demonstrated that the phonological analysis quite naturally characterises the phonological difference between a number of phonetically different contours for which GDI has only got one description available. It is understood, of course, that a description that encodes all phonetic differences in the phonology of the language should be treated with suspicion, as we may expect a good deal of meaningless phonetic fluctuation. If the phonetic differences noted below could be shown to be of this meaningless sort, then the fact that the phonological analysis treats them as d i f f e r e n t should count against that analysis. In trying to decide this issue, we have adopted the GDI technique of analysis-resynthesis and perceptual evaluation. More particularly, for every example of a phonetic difference dealt with below, we created a minimal pair. One member of each pair represented an imitation by the author of an utterance taken from the tape accompanying Collier & 't Hart ( 1 9 8 1 ) . The imitation was analysed, and minimally altered so as to

113

Adequacy in Intonation Analysis

create the other member of the minimal pair. Both the original and the manipulated version were then resynthesised and auditorily evaluated. In all three cases, the two members of the pair were nowhere near being 'perceptually equivalent' and appeared to represent categorically different contours. The pitch curves labelled JH are the ones taken from Collier & 't Hart, and the ones labelled CG are the imitations (solid curve) and manipulations (dotted section) . Example 1. A contour described as just '!'

(redundantly Ό10') can represent

either of two constructs in the phonological analysis:

. A half-completed LH t . Ά level tone, i.e. a stylised LH

In the first situation, the rise represents the movement from L to Μ (recall * * that stylised LH -> LM) and will take place inside the accented syllable. In the second situation, the '!' represents the rising movement from the baseline to the mid-level tone, and since the accented syllable itself is associated * with the M, '!' will take place before it. The two constructs are illustrated in ( 3 2 ) and ( 3 3 ) respectively.

(32)

v—

lorgen De verwachting voor morgen

the forecast

*' * LM (half-completed LH) for tomorrow HZ 160 ·

(33)

40

80

1 2 O 1 6 0 CS*c

"~

De verwachting vooi/ morgen Hz

M

(stylised LH)

160

120

0

«

Ο

Μ

120

160 CS«

114

Carlos Gussenhoven

Use of the stylised LH instead of the half-completed one causes the routine nature of the communication to increase, in line with Ladd's (1978) semantic characterisation of stylisation. The semantic difference is particularly clear * * in the case of ( 2 ) , a half-completed LH: a mid level tone (stylised LH) would be quite inappropriate in the situation described there. Example 2. A contour ' 0A2 ' can arise in either of two ways:

*

. LH to which NARRATION has applied * . HLH with high onset

* The first case is illustrated in ( 3 4 ) . Recall that for narrated LH the onset is high. The pattern occurs quite frequently in Collier & 't Hart (1981). However, unlike what GDI claims to be the case, the pitch drop represented by Ά' 10 occurs before, not in, the accented syllable. In the phonological description, this pitch movement is created when the pitch comes down from the top line in order to realise the L. Interestingly, a contour in which the Ά' occurs actually inside the accented syllable is perfectly well-formed, but very clearly different from ( 3 4 ) . Note that, again, this difference simply cannot be expressed in GDI without extending the inventory of relevant pitch movements. The phonological analysis identifies the difference as one of tone choice, ( 3 5 ) * * having HLH. Note that the domain-final rise is occasioned by spreading of L * in ( 3 4 ) , while that in ( 3 5 ) is the result of straightforward mapping of HLH.

(34)

0

Hz

LH

JH

160 120 80 80

CSec

Adequacy in Intonation Analysis

115

(35) Ga je mee vanavond

Hz

CG

160 120 80

40

80

120 CSec

The ambiguity of Ά' not only occurs after a high onset, but also after narrated * * * HI,. In ( 3 6 ) , weinig has narrated HL, and wind follows with LH. If we replace * * * LH with HLH and apply TLR to the preceding HL, and thus cause the pitch fall represented by Ά' to occur inside the syllable wind, we derive a phonetically similar, but obviously very different contour (cf. ( 3 7 ) ) .

(36)

0 (en hogere middagtemperaturen)

Little wind (and higher afternoon-temperatures) Hz 160

JH

120-

80 40

(37)

A /*. 'we in ig wi'

/

* HL

A

HLH

2

80 CSec

CG

Hz 160

120

8040

80 CSec

216

Carlos Gussenhoven

Example 3. In terms of the phonological analysis, pitch movement Έ' can arise in no fewer than four ways: * . The step-down from the topline to a level tone (stylised LH) * * . The realisation of a half-completed HL (=HM)

. The step-down from a higher plateau to a lower plateau in a * stylised HL * . The realisation of HL in a downstepped contour

Situations 1 and 2 get collapsed in a description ' I E ' . In ( 3 8 ) the topline is * * * * derived by spreading of H in HL, the following tone being M (stylised L H ) . In * * ( 3 9 ) , we have two H L ' s linked by TLR, with the second HL half-completed. As predicted by the phonological analysis, the Έ' necessarily comes before the accented syllable in ( 3 8 ) , but inside it in ( 3 9 ) . Hz 160-

(38)

JH

12O-

80

80 CSec

/ "

n

HL

M (Stylised LH)

Het/regende hard

it rained

hard

(39)

(maar de kinderen bleven toch buiten spelen)

(but the kids stayed out nevertheless)

Hz 160-

CG

120-

I*

*\

Het /regende hard

8040

HM (half-completed HL)

80 CSec

Adequacy in Intonation Analysis Construct

117

( 3 8 ) is restricted to non-final clauses, and seems typical of reading

(observe that Narration has applied to the first tone). Contour ( 3 9 ) can easily * * occur finally, and would be quite natural on a reserved Ze zeiden van wel (roughly "They said they w o u l d ' ) . The pragmatic effects of the two contours are clearly different. The notation ' I E ' is also used to represent the third situation listed above. Observe that in this function ' I E 1 only marks a single accent, as opposed to two in ( 3 8 ) and ( 3 9 ) . For this reason,

't Hart & Cohen (1973) noted that the

pitch movement is not "necessarily associated with prominence'. Collier & 't Hart (1981) gloss Έ' with 'It is sometimes displaced to the right L 1 ( c f . Table 1 ) , suggesting that ' I E ' here is really ' l & E ' , with the Έ' straying o f f . The phonological analysis makes it clear why this situation should arise. Έ' * here marks the step-down from the first to the second plateau of a stylised HL, a tone variant that in Dutch can also be used in non-final clauses. Example (40) is from Collier & 't Hart ( 1 9 8 1 ) . 1 1 It is observed that if we replaced * * Als u ons us adres geeft with Als u op tijd bestelt ( ' I f you order on t i m e ' ) , keeping the intonation the same, the f i n a l , accented syllable is broken into * two plateaus, a feature of stylised HL. In ( 4 0 ) the two plateaus are symbolised * * * M+ and Μ-, ad-hoc notations for the phonetic reflex of stylised Hi.

(40) Als u ons uw qidres geeft

(kunnen we u de rekening sturen)

M+ MIf you us your address give (we can send you the bill)

The fourth situation, in which Έ1 functions in a downstepped contour, has been illustrated in ( 1 3 ) and could be compared to the multi-plateau contour illustrat* ed in ( 2 4 ) , an instantiation of a stylised HL. Contours like ( 2 4 ) do not in fact occur in GDI, but if they did, they would have to be analysed as having a sequence of Έ' movements. It is clear that the phonetic similarity between ( 1 3 ) and ( 2 4 ) in no way reflects linguistic similarity, the Έ1 movements marking accents in ( 1 3 ) , but not in ( 2 4 ) .

4.

CONCLUSION

In this paper, it has been suggested that a description of the intonation of

118

Carlos Gussenhoven

Dutch which incorporates the conventional separation of the systematic phonetic level from a morphological one, is better equipped to explain intonational phenomena and characterise contrasts than a description which is based on the categorisation of surface phonetic forms. It has proved to be possible to describe the wide variety of Dutch intonation contours on the basis of three tonal morphemes and a restricted number of phonological operations. While the phonological analysis has thus proved to be a fruitful

framework

within which to investigate the structure of the intonation of Dutch, at the same time it makes it apparent that there are a number of important lacunae in our knowledge. One of these concerns the question of the domains to which intonational constructs are confined. First, we have seen that for the purposes * * of stylised HL and HLH we need to be able to refer to a domain resembling, but not coinciding with, the foot. Second, examples like (8) - (10) demonstrate that sequences of unlinked tones often divide the segmental string up in ways that seem to obey syntactic phrasing. Operations like TLR and Downstep, too, are not freely applicable, their domains being smaller than that of the grammat* * * * ical category 'sentence'. For example, in Wat is de hoogste berg in Nederland? ( ' W h a t ' s the highest mountain in Holland? 1 ) TLR cannot apply to any sequence of paired accents. Downstep, which can be applied here (assuming HL's are used), * in its turn appears to be blocked across the clause boundary in De uitkomst is * * tien, wat gek is ('The result is ten, which is f u n n y ' ) . The second issue relates to the theoretical status - or lack thereof - of the phonetic representations used in this article. These only serve an expository purpose: they have been used to make it clear how the use of intonational options affect the general configuration of the surface contour. But it should be clear, for example, that the number of phonetically different

'mid' pitches that figure in our descrip-

tion may be quite large. It would be wrong as well as impracticable to attempt to define them, treating them as if they were entities in some phonetic inventory of Dutch. It is assumed, with Liberman & Pierrehumbert (1984), that the generation of the actual FO values proceeds directly from the intonational options themselves. As their research shows, these can be related to the FO contour through phonetic implementation rules that take values for the corresponding phonetic parameters ("reference line 1 , 'downstep 1 , 'range', etc.) as input. However, for such an approach to be feasible more information about the characteristics about Dutch contours, particularly as a function of pitch range, is needed than is at present available. This also applies to our knowledge about the relation between intonational choices like stylisation and TLR on the one hand, and duration on the other. It is hoped that this article will contribute

Adequacy in Intonation Analysis

119

towards a clearer definition of such research goals.

Notes 1.

2.

I should like to thank Rene Collier and Hans 't Hart for their kind and useful discussion of many of the points raised in this article. It is pointed out that the phonological interpretations of the contours taken from their work are my own, and that there is no agreement about the extent to which their notations may obscure intonational contrasts. Thanks are due to the staff of the Phonetics Department of Nijmegen University for helping me to carry out the FO measurements reported in this article. This work was supported by the Netherlands Organization for Advancement of Pure Research.

Thus, for instance, Pierrehumbert ( 1 9 8 0 ) , where the phonological level • of representation is directly generated, rather than derived from some more abstract representation

3.

Leben derives the prenuclear dips in contours like (9b) from the stipulation that starred tones can spread leftwards to prenuclear accents. Note that this makes the phonetics of the prenuclear accent dependent on the tone choice for the nuclear accent, and that contours like ( 8 b ) , which combines £H and HL, cannot be generated.

4.

For a discussion of this issue see Cohen, Collier & 't Hart ( 1 9 8 2 ) .

5.

The formulation would appear to predict that the first operation can occur without the second, i.e. that peaked downstepped contours occur. Although no such contours are given in 't Hart & Collier ( 1 9 7 5 ) , they would not seem illformed in Dutch.

6.

Cf. Goldsmith (1975) who rules out spreading of starred tones in tone languages.

7.

The term stylisation was unfortunately also used by Ladd (1978) to refer to the modification of the falling nucleus to the call contour, and the rising nucleus to the level tone. I assume that no confusion will result if I continue to refer to the modification concerned as 'stylisation 1 .

8.

Compare also Ladd's (1983:747) comment on the status of ' B 1 : 'Type B Fall is simply a prosodic boundary ... followed by a return to the normal pitch of phrase-initial unaccented syllables'.

9.

I follow GDI usage in referring to actually measured contours as pitch curves.

10. In the contours analysed as ' 0 A 1 , however, the Ά' occurs inside the accented syllable, e.g. Collier & 't Hart ( 1 9 8 1 : 3 9 ) . 11. Stylised i$L and half-completed HL are both also illustrated on single words in Collier & 't Hart (1981). However, on non-final accented syllables (e.g. Johanna] ' I E ' appears to denote half-completed HL, but on final accented syllables (e.g. Amerikaan) ΊΕ 1 denotes the stylised HL ( ' t h e syllable -kaan is in a sense broken into two portions by Έ', so that we seem to hear "Amerikaahaan". By thus coraplementarily distributing

120

Carlos Gussenhoven these two tonal constructs, GDI manages to present them as variants of the same contour, but as a result renders itself incapable of characterising Johanna with a stylised $L, or Amerikaan with half-completed HL.

References

Bing, J. ( 1 9 7 9 ) , Aspects of English intonation. Distributed by Indiana University Linguistics Club. Cohen, A . , R. Collier & J. 't Hart ( 1 9 8 2 ) , Declination: Construct or intrinsic feature of speech pitch?. Phonetics 39, 254-73. Cohen, A. & J. 't Hart ( 1 9 6 7 ) , On the anatomy of intonation. Lingua 19, 17792. Collier, R. & J. 't Hart ( 1 9 8 1 ) , Cursus Nederlandse Intonatie. Leuven: Acco. Crystal, D. ( 1 9 6 9 ) , Prosodic Systems and Intonation in English. Cambridge: Cambridge University Press. Hart, J. 't & A. Cohen ( 1 9 7 3 ) , Intonation by rule: A perceptual quest. Journal of Phonetics 1, 309-27. Hart, J. 't & R. Collier ( 1 9 7 5 ) , Integrating different levels of intonation analysis. Journal of Phonetics 3, 235-55. Hart, J. 't & R. Collier ( 1 9 7 9 ) , On the interaction of accentuation and intonation in Dutch. In: E. Fischer-Jjirgensen, J. Rischel & N. Thorsen ( e d s . ) , Proceedings of the Ninth International Congress of Phonetic Sciences 1979. University of Copenhagen, 395-401. Goldsmith, J. ( 1 9 7 5 ) , Tone melodies and autosegments. Proceedings of the Sixth Conference on African Linguistics, 135-47. Gussenhoven, C. ( 1 9 8 3 ) , A semantic analysis of the nuclear tones of English. Distributed by Indiana University Linguistics Club. Published as chapter 6 in Gussenhoven, C. ( 1 9 8 4 ) , On the grammar and semantics of sentence accents. Dordrecht: Foris. Gussenhoven, C. ( 1 9 8 5 ) , Intonation: A whole autosegmental language. In: H. v.d. Hulst & N. Smith ( e d s . ) , Advances in nonlinear phonology. Dordrecht: Foris, 117-31. Keijsper, C . E . ( 1 9 8 4 ) , Vorm en betekenis in Nederlandse toonhoogtecontouren. Forum der Letteren 25, 20-37 (Part I ) , 113-26 (Part I I ) . Knowles, G. ( 1 9 8 4 ) , Variable strategies in intonation. In: D. Gibbon & H. Richter ( e d s . ) . Intonation, accent, rhythm. Berlin: de Gruyter, 226-42. Ladd, D . R . ( 1 9 7 8 ) , Stylized intonation. Language 54, 517-40. Ladd, D . R . ( 1 9 8 3 ) , Phonological features of intonational peaks. Language 59, 721-59. Ladd, D . R . ( 1 9 8 4 ) , Declination: A review and some hypotheses. Phonology Yearbook 1, 53-74. Leben, W.R. ( 1 9 7 6 ) , The tones in English intonation. Linguistic Analysis 2, 69-107. Liberman, M. & J. Pierrehumbert ( 1 9 8 4 ) , Intonational invariance under changes in pitch range and length. In: A. Aronoff & R.T. Oehrle ( e d s . ) , Language sound structure: Studies in phonology presented to Morris Halle by his teacher and students. Cambridge: MIT Press, 157-233. Maddieson, I. ( 1 9 7 8 ) , Universals of tone. In: J . H . Greenberg, C . A . Ferguson & E . A . Moravcsik ( e d s . ) , Universals of· Human Language. Volume 2: Phonology. Stanford: Stanford University Press, 335-56. Pierrehumbert, J. ( 1 9 8 0 ) , The phonology and phonetics of English intonation. MIT dissertation.

Adequacy in Intonation Analysis Pijper, J.R. de (1983), Modelling British English intonation: An analysis by resynthesis of British English intonation. Dordrecht: Foris. Selkirk, E . O . (1984), Phonology and syntax: The relation between sound and structure. Cambridge: MIT Press.

121

Pitch Accent and Intonation in Japanese Shosuke Haraguchi University of Tsukuba 0.

Introduction*

Natural languages fall into two categories: those with an accentual system and those with a nonaccentual one. The former can in turn be classified into stress languages and pitch-accent languages. Stress languages are further subclassified into those with fixed stress and those with movable stress, while pitch-accent languages consist of one-melody types, such as Tokyo Japanese and Hisosaki Japanese, and two-melody types, such as many Kansai Japanese dialects ( e . g . , Osaka, Kyoto, Takamatsu, e t c . ) . Nonaccentual languages can be classified into the so-called tone languages, such as Mandarin Chinese, Mende, etc., and unaccented languages such as Kagoshima, Miyakonojo, Sendai Japanese,

etc.

This can be graphically represented as in ( 1 ) :

(1)

natural languages accentual·"""''''^ stress' fixed

Old

I English

--- nonaccentual·

pitch-accent

tone"

movable one BTM two BTM's

I

English

Tokyo

Osaka

Chinese Kagoshima Miyakonojo Sendai

This article is mainly concerned with a survey of tonal aspects of various Japanese dialects, which belong either to a pitch-accent system or to an unaccented system, as well as a discussion of their theoretical implications. Stress systems and tone systems will be discussed in so far as they are helpful in clarifying some of the characteristics of pitch-accent systems and unaccented systems of Japanese dialects. In section 1, I will begin by discussing briefly several general properties of Japanese pitch-accent. Then in the subsequent sections, we will· turn to a discussion of several· topics of Japanese tonal systems and their relevance to current theoretical issues. As suggested in ( 1 ) , Japanese dialects fall into two types: pitch-accent

124

Shosuke Haraguchi

systems and unaccented systems. Each of the two systems will be divided into a number of subtypes in terms of the number and types of the basic tone melody (BTM for short), to which we will return in section 2. Section 3 is devoted to a reexamination of tone the association rules and universal tone association conventions discussed in Haraguchi (1977, 1979a,b) and a discussion of some alternative proposals. These processes are reformulated into a number of universal principles with some parameters. Some typological remarks on parametric variations are also included. Section 4 is concerned with the justification of the accentual analysis (diacritic * analysis) of Haraguchi (1977, 1979a,b) of Japanese pitch-accent systems. In order to show the necessity for an accentual analysis, I will analyze the tonal system of the Owase dialect and discuss its theoretical implications in section 5. Section 6 discusses a number of special tone melodies which are regarded as specific proposals concerning some intonational phenomena in Japanese. In the final section, we make our concluding remarks on Japanese pitchaccent and intonation.

1.

General Properties of Japanese Pitch-Accent

Let us briefly survey some of the general properties of Japanese pitch-accent, which are more or less direct reflections of characteristics of pitch-accent in general. First of all, Japanese pitch-accent (in fact, pitch-accents in general) is independent of the kind of rhythmic effects that are observable in true stress. As discussed in Haraguchi (to appear), the rhythmic structure of Japanese is analyzed as having an underlying 2/4 time structure. Thus, the most favored seven-five mora-timed rhythm is analyzed as the combination of quadrimoraic and octo-moraic rhythmic structure, as illustrated in ( 2 ) : (2)

s s s s I s s s > j s s s s I s > > > I D D D D D D D 4 D D D D D 4 4 4

(where "s" represents "mora", and " > " , "rest".)

Pitch Accent in Japanese

125

This rhythmic structure is determined solely in terms of the number of moras and "rests", there being no room left for pitch-accent to play a role in this respect. Second, strength or prominence is not a distinctive characteristic of pitchaccent in Japanese nor in any other pitch-accent languages. Third, the accented mora is relatively longer than the unaccented moras of a word or a phrase, although the difference in length is usually inaudible to the ordinary ear (Saikawa, et al. ( 1 9 7 0 ) J . 1 Fourth, a large number of pitch-accent systems in Japanese permit some domains to remain "unaccented", which is clearly different from stress-accented systems, where every phrase carries a primary stress (Clark (to appear}). This is counted as one of the salient differences between stress-accent systems and pitch-accent systems, though it must be kept in mind that some Japanese dialects such as the Fukuoka dialect have accented words only. Fifth, some pitch-accent dialects in Japanese have an underlying accentual system which is different from its

surface accent system. Consider, for example,

the Tsuruoka dialect. The three mora words of this dialect, as shown by Haraguchi (1979a,b), have the underlying accent system as indicated in ( 3 ) :

(3)

Underlying accent system (Tsuruoka): a. unaccented

ο ο ο

b. initial-accented

ο ο ο

c. medial-accented

* ο ο ο

d. final-accented

ο ο ο

Due to the Star Shift rule in ( 4 ) , Tsuruoka has the five derived accent classes indicated in ( 5 ) :

V C° (5)

* V

V

(4)

V C0

-high

a.

ο ο ο (-ο)

b. (i)

ο ο ο (-ο)

-high

(ϋ) ο ο ο (-ο) C.

ο ο ο

d.

ο οοο ο

(-ο) οοο-ο

(where "-ο" indicates post-positional particle of one mora.)

126

ShosuAe Haraguchi

To put it differently, a dialect with Star Shift which affects the underlying accent system has a different and derived accent system. Notice incidentally that Star Shift in Japanese is restricted to rightward movement only. Sixth, some Japanese pitch-accent dialects have (a) process (es) dependent on vowel height. Tsuruoka is one of them. For some other dialects, see Haraguchi (1977, 1979a,b). In constrast, there seems to be no stress language in which stress phenomena are dependent on vowel height. Seventh, there seem to be no systems with fixed pitch-accent, at least in Japanese. This is because pitch-accent is specified in the lexicon only in cases where the position to which pitch-accent is assigned is unpredictable. Eighth, some Japanese pitch-accent systems have turned into nonaccentual systems or so-called "collapsed" systems. There seem to be no stress languages which have converted into a "collapsed" system. These are counted as the central part, if not all,

of the salient properties

of Japanese pitch-accent. Most of these are no doubt shared by other pitchaccent languages as well.

2.

On BTM

Before turning to BTM 1 s in Japanese dialects, let us begin with a brief discussion of assumptions on BTM's. I assume that universal grammar (UG for

short)

permits the following tone melodies as BTM's:

(6)

a. H, L, HL, LH, LHL b. M, MH, ML, LM, HM, LML

Of these melodies, those in (6a) are unmarked in that most languages contain at least one of them as their BTM. Those in ( 6 b ) , in contrast, seem to be marked since they are fairly restricted in their use. Take for example the melody M. The only possible candidates which might be characterized as having this melody would be "monotonous" or "collapsed" dialects. Unfortunately, however, investigation on tonal phenomena of these dialects is scarce. Thus, I must regrettably admit that nothing conclusive concerning this melody can be said at present. I also assume that each grammar selects one or some of these melodies as its

BTM. How many melodies a language has, and which melodies are chosen, is

parametrized, and the values for the parameters are determined on the basis

Pitch Accent in Japanese

127

of experience. Taking into consideration the fact that children acquire tonal melodies amazingly fast, I assume that central parts of the tonal processes are predetermined and thus universal with some parametric variation permitted. I will return to the discussion relating to this point in section 3. Let us for the moment focus attention on the BTM's of Japanese dialects. As I have mentioned above, Japanese dialects are divided into accentural and nonaccentual types. There seem to be no accentual dialects with the BTM H or L in Japanese. Though Shibatani (1979), Archangeli and Pulleyblank (1984) and Clark (to appear) claim that the Tokyo dialect has H as its

BTM and that the

Osaka dialect (or the Kansai dialects)in general) has H and L as its

BTM's, I

think that their claim is not well-supported, thus lacking in persuasive force. Haraguchi (to appear) notes that the BTM of Tokyo Japanese should be HL on the basis of the fact that the lowering of the tone after the accented syllable is distinctive. Unless this characteristic of the Tokyo melody is directly captured in terms of the BTM, we fail to capture a significant melodic property of the dialect in question. To see this, compare Tokyo melodies with Hirosaki melodies:

(7)

Tokyo

c. ko

saku Ira

sakura -mo

kltu

kitune -mo

ne

ko \ ro(-qa)

d. ka I garni \ (-ga) The initial L tone in the Tokyo dialect appears, when these words or phrases are uttered in isolation, and the final L tone in the Hirosaki dialect appears following and preceding a pause. Thus, in the Tokyo dialect the initial lowering does not take place when the word or phrase follows another word or phrase without a pause. In contrast to the Tokyo dialect, in which it is not possible to predict where the melody will change from H to .L, the Hirosaki dialect is unpredictable as to where the melody will turn from L to H. This difference in the Tokyo and Hirosaki melodies can be captured directly if we take the Tokyo melody as HL and the Hirosaki melody as LH. Suppose, for the sake of argument, that the Tokyo dialect has a Η-tone as its

BTM. Then we would also have to set up a Η-tone as the Hirosaki BTM.

128

Shosuke Haraguchi

Thus, the melodic differences between the Tokyo and Hirosaki dialects would not be captured in terras of BTM's. Their differences could be captured only in terms of the differences in the direction of Η-spread. That is, Η-spread would apply from right to left in the Tokyo dialect and from left to right in the Hirosaki dialect, as indicated in (8) and ( 9 ) : (8)

Tokyo

a. sakura

Xvl•*j H

(9)

b. kabuto

Iι H

c. kokoro

d. kagam*

c. usagi l '' V H

d. oto o l \ H n/a

\lj H

n/a

x xJ jH

Hirosaki a. sakura l l n/a H

b. kitune l ' -*' V'' H

Notice in passing that left to right Η-spread in Hirosaki is contradictory to Archangeli and Pulleyblank's claim thet Η-spread in Japanese applies backwards only. By supposing that Η-spread can apply either backwards or forwards, we are able to capture the mirror image property of the Tokyo and Hirosaki melodies, but only indirectly. In contrast, under the assumption that the BTM of the Tokyo dialect is HL and that of the Hirosaki dialect is LH, we can capture both the distinctive properties of each BTM and their mirror image property in a direct and simple way. Notice that in Archangeli and Pulleyblank's

theory, which has H or L as

BTM, each dialect would require a specific stipulation as to whether or not it incorporates backward spreading or forward spreading. In the present theoretical framework, however, no stipulation is required as to the directionality of spreading. This is because the very fact that the Tokyo dialect spreads the H tone only backwards and that the Hirosaki dialect spreads the H tone only forwards is an automatic consequence of the interaction of the BTM, HL or LH, and a version of the universal tone association principles to be discussed in the next section. The observations above will be sufficient

to show that the BTM's of the

Tokyo and Hirosaki dialects are respectively HL and LH. As I have already shown in Haraguchi (1977, 1979a, b ) , the Osaka and other Kansai dialects have two BTM's HL and LHL, which are, roughly speaking regarded as a combination of the BTM of, dialect (LHL).

say, the Tokyo dialect (HL) and the Fukuoka

Pitch Accent in Japanese

129

As far as I know, the only accentual language with three BTM's is the Old Kyoto dialect ( c f . the Ruijumyogisho, 1100 A D ) . Haraguchi (1979b) shows conclusively that this language is to be analyzed as having HL, LH, and LHL melodies As an illustration of this, consider the following two-mora words of the Old Kyoto dialect: (10)

HL Melody a. o o . * b. ο ο * c. ο ο

unaccented

kuti

'mouth'

initial-accented

'stone'

second-accented

'ditch'

LHL Melody d. ο ο

unaccented

1

e. ο ο

initial-accented

•leg·

f. ο ο

second-accented

'rain'

flower'

LH Melody g. o o

unaccented

h. ο ο

initial-accented

'lily'

i. ο ο

final-accented

"pine tree 1

In addition to the above, Old Kyoto is said to contain another surface melody, a rather puzzling one, as indicated in ( 1 1 ) :

(11)

1

rainbow'

The falling tone on the initial mora might be nothing but an exceptional phonetic realization of the Η tone or it might be due to the fact that 'nizi' has a long initial vowel. Whichever may turn out to be the case, this word will be analyzed as belonging to the initial-accented HL melody class. With regard to the analysis in (10) , I must caution the reader about the fact that because of a lack of sufficient data, the analysis of ( l O d ) , (10g), and (lOi) are rather tentative, and thus that there is ample room for revision of, for example, ( l O i ) to either (10g) or ( l O d ) . But irrespective of the final solution, our claim that the Old Kyoto dialect has three BTM's will not be affected. This discussion of Old Kyoto shows beyond any reasonable doubt that Clark (to appear)'s claim that 'the Japanese of that period [= Old Kyoto Japanese —SH] had the typical nominal tone melodies of an unrestricted tone language, including virtually every possible comination of H and L·' does not hold. Notice that this observation gets rid of one of the 'empirical supports'

130

Shosuke Haraguchi

for Clark's move to re-analyze pitch-accent languages as tone languages. Notice also that her second argument based on the observation that 'Japanese differs from true accentual systems in that they permit some domains to remain "unaccented", unlike stress-accented systems, where every phrase carries a primary stress 1 has no force, because the difference is sue to the very difference between the notions of stress and pitch-accent. Let us now return to the main stream of our discussion. We can safely claim that there is no pitch-accented system with four or more BTM's. The only languages with four or more BTM's are so-called tone languages like Mandarin Chinese, which incorporates four BTM's namely, L, H, LH, and HL, and Mende, which incorporates five BTM's, namely, L, H, HL, LH, and LHL. The Kagoshima dialect, which is an example of an unaccentual system in Japanese, has two BTM's, LHL and LH. Tone association in this dialect proceeds from right to left. In contrast, most of the other unaccentual systems in Japanese have just one BTM. For example, the Miyakonojo dialect has the LH melody with a right-to-left tone association process. The Sendai dialect contains the H melody, which is simultaneously associated with all the tonebearing units in a tonal domain by one of the principles of tone association to be discussed in the next section. The observation above will be summarized in the following table: (12)

accentual system

nonaccentual system

L

?

?

H

?

Sendai

HL

Tokyo

?

LH

Hirosaki

Miyakonojo

LHL

Narada; Kumi

?

HL, LHL

Osaka, Kyoto

?

LH, LHL

?

Kagoshima

LH, HL

?

?

HL, LH, LHL

Old Kyoto

?

L, H, LH, HL

_ _ _ _ _ _

L, H, HL, LH, LHL

_ _ _ _ _ _

Mandarin Chinese Mende

The question-marked slots simply indicate that suitable candidates among Japanese dialects are not available at present. Note that the dashes in ( 1 2 ) indicate that there is no accentual system with four or more BTM's. Note also that this table is intended only for (combinations of) BTM's which are possible.

Pitch Accent in Japanese

131

Here, we will not be concerned with the special tone melodies, which will fill some of the question-marked slots. It goes without saying that most of these gaps will be filled on the basis of more extensive research into various other accentual and nonaccentual systems. In closing this brief survey of the BTM's of accentual and nonaccentual systems of Japanese, I would like to stress again that the distinctive or typical melodic characteristics of a dialect or a language should be identified as its BTM.

3.

On Tone Association Processes

In my previous works, I have basically adopted the WFC for the tone proposed by Goldsmith (1976) with some revisions: (13)

WFC

a. All vowels are associated with at least one tone; All tones are associated with at least one vowel. b. Association lines do not cross. Haraguchi (1977) suggests that (13b) may be turned into a more general condition based on the notion of the temporal linearity of speech and other processes. In conformity with Halle and Vergnaud (1982) and others, I will dispense with (13a), giving a number of priciples with certain parameters, if necessary. The principles to be discussed here are basically revisions and extensions of the universal tone association conventions presented in Haraguchi (1977). First of all, (14)

UG must contain the Principle of One-to-One Association:

Principle of One-to-One Association: Associate a sequence of tones with a sequence of tone-bearing units in a one-to-one fashion. Parameter: (a) from left to right or (b) from right to left.

This principle operates in non-accentual systems as illustrated in (15): (15)

a . VI VI VI V1 V 1 ι

I ι

I ι

I I

3

b . V VI VI VI VI ' I

I I

I I

I I

132

Shosuke Haraguchi

It should be pointed out that this principle is not restricted to tonal association only, but is applicable to almost all cases which require two or more tiered structures. Second, UG will have to contain the Principle of the Association of Designated Elements: (16)

Principle of the Association of Designated Elements: Associate the designated tone with the designated tone-oearing unit. Parameter 1: Designated elements are: a. either (i) the first accented V or, (ii) the last accented V, if the domain contains an accent, or b. either (i) the first tone-bearing unit or, (ii) the last tone-bearing unit, if the domain contains no accent. Parameter 2: The designated tone is either (a) H or (b) L.

For example, the values for parameter 1 will be fixed as (ai) and (bii) in the Tokyo dialect, and that for parameter 2 will be fixed as ( a ) . This is because the Tokyo dialect associates the H tone to the first accented V of the accented domain, or the H tone to the last V if the domain lacks accent. Notice that there are 12 possible choices of parametric values, which are summarized in table ( 1 7 ) :

(17)

-\ (a) H

(i)

first *

(ii)

last

(iii)

(b) L

Fukuoka

?

?

?

first *;first V

Kumi

?

(iv)

first *;last V

Tokyo; Osaka

?

(v)

last *;first V

?

?

(vi)

last *,-last V

Hirosaki

?

*

In this table (i) and (ii) are for dialects with accented words only, and (iii-vi) are for those with accented and unaccented words. ( 1 7 ) suggests that Japanese seems not choose the value (b) for parameter 2. With regard to table ( 1 7 ) , what is particularly noteworthy is that Haraguchi (1977,1979a,b) claims that the Hirosaki dialect has the same association rule as the 7"°-i'o dialect. But the findings of Kobayashi (1985) show that this is

Pitch Accent in Japanese

133

not the case. In fact, in this dialect, the H tone is associated with the last, and not the first, accent of the tonal domain. This finding is rather welcome because it fills one of the two gaps in the theoretically possible tonal association processes. As far as I can see, there seems to be no Japanese dialect which contains accented words only, and in which the H tone is associated with the last accented V. And interestingly enough, Japanese seems to have no dialect in which H is associated with the last starred V and with the first V if the domain is accentless. This gap is rather strange, when one considers that all four possible cases of stress languages corresponding to (17iii-vi) are observable, as shown in the following table:1* (18)

first designated V

last designated V

first V (iii)

Vedic Sanskrit;Khalkha Mongolian

(v)

last

Komi

(vi) Aguacatec Mayan

V

(ii)

Eastern Cheremis

Thus, I presume that an extensive examination of the range of accentual languages will eventually fill the gap in (17va). Third, UG will have to contain the Principle of One-to-Many Free Tone Association in the following manner: (19)

Principle of One-to-Many Free Tone Association: If the tonal domain contains one free tone and a number of free tonebearing units ( V ' s ) , then the free tone must be associated with all the free V's.

The effect of this principle is indicated by the broken lines in ( 2 0 ) :

(20)

V

V V V V

V

v >/ ' \ > if/

T

T

l

2

T 3

As a concrete example, take the Sendai dialect. In this dialect, the BTM H is simultaneously associated with all illustrated in ( 2 1 ) :

(21) V H

the tone-bearing units of a phrase, as

134

Shosuke Haraguchi

We will also see the effect of this principle later in section 5. Fourth, UG will have to contain the Principle of Spreading:

(22)

Principle of Spreading: Spread the bound tone to the free V s . Parameter: (a) backwards only, (b) forwards only, (c) both forwards and backwards.

The unmarked value for this parameter seems to be ( c ) . This principle works as illustrated in ( 2 3 ) :

(23)

a.

V

;

V V V V

ι - --\ιτ

τ

b.

V V V V

V

iX- ιτ

τ

Fifth, UG seems to incorporate the Principle of the Association of Free Tone(s) with a bound V: (24)

Principle of the Association of Free Tone(s) with a Bound V: Associate the free tone(s) with a neighboring bound B. Parameter 1: (a) utilize this principle or (b) not. Parameter 2: ( 2 4 ) operates under a condition of (a) the existence of accent or (b) irrespective of the existence of accent.

Tokyo chooses (b) as the value for parameter 1, which means that this principle does not operate in this dialect. Thus, the L tone will remain unassociated with any V in the Tokyo dialect:

(25)

sakura N. \ I H

L

I assume that the unassociated tone simply remains unassociated, not realized at the surface, or is deleted by a general principle. As far as Japanese dialects are concerned, the latter seems generally to be the case because they have no downstep, although this is typically found in a number of African languages. The Osaka dialect chooses (a) for the value of parameter 1 as well as for the value for parameter 2. Thus in this dialect, the L tone is associated with the final accented mora, as illustrated in ( 2 6 ) :

Pitch Accent in Japanese (26)

ame

135

'rain'

l K

L H L

while the L tone remains unassociated in the case of unaccented words such as 'sora 1 in ( 2 7 ) :

(27)

sora

I /

'sky'

L H L

This unassociated L tone will ultimately be deleted. Finally, UG must incorporate the Principle of Peripheral Dissimilation:

(28)

Principle of Peripheral Dissimilation: Insert [a high-pitch, 3 low-pitch] to the adjacent position of the peripheral tone [-a high-pitch, -β low-pitch] and associate the former with the peripheral V. Parameter 1: (a) applicable in all cases, without peripheral-accent or accent, or (c) applicable in Parameter 2: Insert a tone (a) before the the rightmost tone.

(b) applicable in cases second (or penultimate) none of the cases. leftmost tone or (b) after

The function of this principle is to insert a tone with the feature value opposite to the peripheral tone. It corresponds to the former processes of initialraising, initial-lowering, final-raising and final-lowering, all of which are observed in Japanese dialects (see Haraguchi (1977,1978, and 1 9 7 9 a , b ) ) . This principle accounts for the correlation between the BTM and raising/ lowering. For example, initial-lowering can occur in cases with BTM HL or H, while initial-raising is restricted to cases with BTM L, LH or LHL. The Tokyo dialect fixes the value for parameter 1 as ( b ) , and that for parameter 2 as ( a ) . Thus, (28) operates on 'saJcura' (cherry tree) and 'kokoro' (heart), but not on 'kabuto'

(29)

a. sakura

ίΑΙ

L H L

(helmet):

b. kokoro

Ν I

L H L

c. kabuto

IN

H Ln / a

I assume, following Halle and Vergnaud (1982) and Pulleyblank (1983), that if a universal principle or a language specific rule associates a tone with an already associated V, then the previously associated line will be deleted

136

ShosuJce Haraguchi

automatically. This principle, which I will call the Principle of Automatic Dissociation, operates in (29a,b) and turns them into ( 3 0 ) : (30)

a. sakura

I \ JH L L

b. kokoro

I I I L H L

c. n/a

Notice here that there appear to be some difficult problems with regard to the principle of automatic dissociation. Consider, for example, the Hirosaki dialect. After principle (28) has operated, we have the following cases: (31)

a. usagi

j/j B L

b. otoko

\lL HHKL

The principle of automatic dissociation applies only to (3 la) , turning it into ( 3 2 ) , while the application of the principle to (31b) should be blocked. (32)

a. usagi

I I I

b. n/a

L H L

Thus, the contour tone on the accented V remains in the Hirosaki dialect. This observation suggests that this principle is not without problems. One possible way to evade this problem, however, will be to parametrize the principle. If this is proven to be the case, the Hirosaki case ceases to be a difficulty. Let us examine one more problematic case. As a case in point, consider the flop rules in the Takamatsu dialect. In Haraguchi (1977), the case is discussed of the Takamatsu Japanese Η-flop rule which applies when the accented vowel is followed by a mora with a non-high vowel or when it is followed by a mora with a high vowel /i/or/u/ which, in turn, is followed by a mora with another high vowel:

In (33a) , Η-Flop is blocked because the accented vowel is followed by a high vowel / u / , but in (33b) the rule applies to produce the HL contour tone.

Pitch Accent in Japanese

137

Notice here that the principle of automatic dissociation does not operate, which suggests that the Takamatsu dialect somehow blocks the operation of this principle. However, O s t l e r ( 1 9 7 7 ) ' s revision of my original analysis applies to the Η-flopped LHL melody words only, and this L-Flop brings about an automatic dissociation of the association line between the H tone and the accented V:

(34)

L-Flop:

u-v

X L

V

V

H

This suggests that the principle under consideration must operate on the output of L-Flop. This is incompatible with what we have observed with respect to the Η-Flop above. How can we explain this paradoxical situation? One possible solution will be to assume that the Takamatsu dialect has the following constraint:

(35)

The LHL melody words permit only one Η-toned mora per word.

This constraint induces the deletion of the original association line drawn between the H tone and the accented V when two Η-toned moras emerge as a result of Η-Flop. Thus, consider the following derivation:

(36)

a. harisigoto

Η-Flop

VL HI VL (23b)

b. harisigoto

VL Ht-'VL

Constraint (35)

c. harisigoto

VL

H

NJL

d. harisigoto

Ψ' Ν L

H L

The L tone is spread to the now unassociated mora by the principle in ( 2 3 b ) . This solution, supposing that it

is correct, makes both the L-Flop rule and the

operation of the principle of automatic dissociation unnecessary in this case. Thus, the two problematic cases discussed above can be shown to be only apparent, and not real. The principles outlined above will be enough to handle most of the tonal phenomena of Japanese dialects. In concluding this section, I would like to stress that our new analysis turns most of the tone association processes into

138

Shosuke Haraguchi

universal principles, thus leaving to the language-particular grammar the task of fixing the values for the parameters and of adding only a limited number of marked processes, if necessary. This will make it easier for us to account for the fact that tonal melodies are acquired easily and speedily, compared with the acquisition of the manipulation of segmental sounds.

4.

Arguments for the Diacritic * Analysis

Let us now turn to our other topic, i.e., the validity of the accentual diacritic analysis of Japanese accentual systems. In an attempt to develop an explanatory theory within a framework of lexical phonology, Pulleyblank (1983) has argued that in some African languages such as Luganda and Tonga, the introduction of accentual diacritics into the tonal theory is not justifiable; he then proposed a version of the theory which admits pre-linking of tones as the indicator of 'accent'. Along similar lines»Clark (to appear) and Archangeli and Pulleyblank (1984) claimed that accentual diacritics can be dispensed with in the tonal theory of Japanese. As I have not yet attempted a detailed examination of Pulleyblank's analysis of Luganda and Tonga, I have no comment on his ingenious analyses of these languages and accept his claim for the time being. Here, I will restrict my attention to the articles by Clark, and Archangeli and Pulleyblank. As I have already discussed the inappropriateness of the BTM proposed by them, I will focus on the issues relating to accentual diacritics. Clark (to appear) claims that my analysis, which incorporates accentual diacritics, 'has a serious theoretical flaw in that it provides competing analyses — an accentual analysis and a tonal one — for virtually every phenomenon. This indeterminacy arises from the fact that Japanese "accent" (represented in Haraguchi's analysis by the feature *) has no phonetic realization other than tone.' This argument, based on Chomsky's claim that a fully adequate theory must severely reduce the class of possible grammars, has no force at all.

Why?

This is because the introduction of the 'accent' marker into the theory does not necessarily bring about the indeterminacy which Clark points out. We can easily rid ourselves of the indeterminacy in question if we postulate the following principle in our tonal theory:

Pitch Accent in Japanese (37)

139

In the absence of counterevidence, select the tonal analysis over the accentual analysis.

This principle will limit the number of possible grammars considerably, solving the problems adduced to by Clark. Notice that this principle is rather natural because its basic idea is that if there is no counterevidence, we should select the least abstract analysis. With regard to this principle, consider for example the Kagoshima dialect. In Haraguchi ( 1 9 7 7 ) , it is suggested that there are two alternative analyses: one is an accentual analysis and the other a nonaccentual one. As far as I can see, there seems to be no conclusive evidence to choose one over the other. However, the principle in ( 3 7 ) compels us to choose the nonaccentual analysis of tonal phenomena in the Kagoshima dialect. Thus, it will be clear that the theory incorporating accentual diacritics can reduce the possible analyses of tonal phenomena in quite a natural way. This observation will be sufficient to virtually nullify the force of Clark's argument. Let us turn next to the discussion of some evidence to show that accentual diacritics are necessary. Consider first the following example of the Nakamura dialect, which is discussed in detail in Haraguchi ( 1 9 7 7 ) - . 6

(38)

two-mora word (+Nom particle)

a. u fsJ

three-mora word (+Nom particle)

'cow'

'Jcä] su finl

'haze'

'cow-Norn'

ka\ sumi\ -qa

'haze-Norn'

b. * Ito-qra

'string-Norn'

ka \rasu-ga

'crow-Norn 1

c. i |s* [gra

'stone-Norn'

ti \ka \ra-ga

' strength-Nom'

a [zu \k*|-ga

'red beans-Norn'

u I si \-ga

d.

(38) suggests that the BTM of the dialect is LHL and that it has an initialraising rule. In our present framework, this dialect will be analyzed as follows: (39)

a. Accentual system; BTM: LHL b. Principle ( 1 6 ) ; Parameter 1: (ai) and (bii) Parameter 2 : (a) c. Principle ( 2 4 ) ; Parameter 1: (b) d. Principle ( 2 8 ) ; Parameter 1: (b) Parameter 2: (a) e. Principle of Automatic Dissociation f. Principle ( 3 5 )

140

Shosuke Haraguchi

The sample derivations based on this analysis will be as follows:

(40)

a.

usi

usi-ga

tikara-ga

azuki-ga

(39a)

LHL

LHL

LHL

LHL

b.

usi I I

usi-ga I ^ I

tikara-ga · I \ ^·

azuki-ga \ ' I I

39b

< >

«,(19)

i I

LHL

>

ML H LI

: I v

LHL

\iL IH

L

c.

usi-ga

azuki-ga

& (39e)

H L H L

H L H L

(39d)

| -|

|| | |

In (40) , the L tone left unassociated is assumed to be deleted ultimately by a convention. Thus, it will not be realized at the surface. What is interesting in the Nakamura dialect is that the H tone associated with the unaccented moras of words of three or more moras is lowered to a M(id) tone. Regardless of what the precise nature of this process may turn out to be, we will need information as to whether or not the mora is accented and the phrase is longer than two moras. It is clear that the pre-linking theory or any other alternatives which do not introduce an accentual marker cannot handle this process with the same ease and elegance as the accentual diacritic theory. Thus, we are led to the conclusion that an adequate tonal theory of Japanese needs an accentual marker.

5.

Theoretical Implications of the Owase Tonal system

For another argument to show the need for an accentual marker, let us now turn to the analysis of the Owase dialect. This dialect is extremely interesting, both factually and theoretically. It is factually interesting in that its BTM's HL and LHL are approximately the reverse of the standard Kansai dialects (e.g., Osaka, Kyoto, e t c . ) : (41)

Standard Kansai

Owase

H L L H L

L H L H L

As an illustration of this, consider the following cases, which are cited from Kindaichi (1975) with some simplification:

Pitch Accent in Japanese

(42)

Standard Kansai (A)

0 0

* (B) ο

(C) ο

0

0

(D) ο ο

(E) ο ο ο

(F) ο ο ο

Owase HL

HL

LHL

LHL

ο ο *

0

0

ο ο

ο ο

HL

0 0 0

HL

* 0 Ο

ο

LHL LHL

HL HL

LHL

LHL

Example

Gloss

ti[T

1

141

blood '

u \si

'cow'

~n \ a

1

Tin u

'dog 1

Π7

'tree 1

Tt "

1

string*

aki (ga)

1

autumn '

kake

•write (Imp) '

kodo \mo

'children'

agaTru'

•arise '

a J ta [ ma

'head'

a [ gra[ re

'arise (Imp) '

name '

* (G) ο ο ο

HL

ο ο ο

LHL

ha] sira

'pillar 1

(H)

LHL

ο ο ο

HL

ki i -wa

•tree-top"

(I) ο ο ο

LHL

ο ο ο

HL

kabuto

•helmet'

(J) ο ο ο ο

HL

ο Ο 0Ο

LHL

sakana \-a

' fish-Top'

* (K) ο ο ο ο

HL

ο υ ο ο

LHL

saku \ra\ -mo

'cherry-also'

* Ο Ο 0

HL

ο Ο Ο 0

LHL

ta [ ~|reru

•fall'

HL

Λ Ο

LHL

ha] tatyaa

•20 years old-Top1

(N) ο ο ο ο

LHL

Ο

(0) ο Ο Ο 0

LHL

(P) ο ο ο ο

LHL

0 0 0

Λ

(L) ο (M)

s

Ο Ο 0

ο ο ο

HL

tee-ka \ ra

'hand- from 1

Ο

οδό οδό

HL

tee-del mo

'hand-even'

0

ο ο ο

HL

kabuta-a

'helmet-Top'

Note that minor irregularities or exceptions are ignored here to make matters simpler and easier to understand. It should be kept in mind that the situation is slightly more complicated, but by no means affects the main argument of this article. Notice, in addition to this, that there are some gaps which are rather accidental and not systematic. Thus, there are no final accented HL melody words in most of the Kansai dialects, which in turn brings about the lack of the final-accented LHL melody in the Owase dialect. The main characteristics of the tonal system of the Owase dialect, which are crucially relevant to the central concerns of this article, will be summarized as follows: First, as we have just seen, the HL and LHL BTM's of the Owase dialect can be said to correspond respectively to the LHL and HL BTM's of the Standard Kansai dialects.

142

Shosuke Haraguchi

Second, the Owase dialect has an initial-accented LHL melody class. For evidence showing that initial-accented words of this dialect belong to the LHL melody class, see the discussion below. The existence of the initial-accented LHL melody class shows that the fact that most of the Kansai dialects lack this class (at least in nouns) is purely accidental. In view of this, it is worthwhile to point out that Clark's claim that Haraguchi's analysis of the Osaka dialect is inadequate because I 'was forced to say that in the LHL class the accent could occur on any mora except the first, while in the HL class, it could occur on any mora except the last'

is

beside the point. Notice, by the way, that the fact that the Osaka dialect has no initialaccented LHL melody nouns is just accidental. This is demonstrated by the fact that adjectives in this dialect have words of such a class, as observed in Haraguchi (1977). Furthermore, the very fact that the dialect in question lacks a finalaccented HL melody class is also accidental, in view of the fact that the Kameyama dialect, which is one of the standard Kansai dialects, has a number of such nouns. Thus, it is quite natural and legitimate to say that the LHL class nouns of the Osaka dialect can have the accent on any mora except the first, while the HL class nouns can have it on any mora except the last. Thus, contrary to Clark's claim, the need for special stipulations of this sort does not reveal the weakness of my analysis. It actually confirms the legitimacy of my analysis. This is because it correctly reflects the accidental distributional properties of the tonal system of the Osaka nouns. Third, the so-called pre-accenting morpheme '-mo 1 , which is a postposition meaning 'also', cannot be analyzed as extratonal as indicated in:

(43)

(mo)

nor as pre-linked with a L tone:

(44)

mo L

We will have to analyze this morpheme as having a floating pre-accent or to set up an accent assignment rule of the following form:

Pitch Accent in Japanese

143

(45)

This will be clear from the consideration of the phrasal melody, to which we will turn below. This will be considered to be another piece of evidence to show that the accent mark is indispensable in an appropriate tonal theory. Let us now proceed to the fourth characteristic of the Owase tonal system. In the Owase dialect, when two phrases A and B are adjacent and connected closely with no pause intervening between them, we have two possibilities. The first possibility is that if both A and B belong to the LHL melody class and A is accentless, then the phrase A becomes extratonal, as illustrated in (46) :

(46)

(A) B

Thus, the accentual and melodic information of B will determine the tonal melody of the whole sequence A B. As illustrations of this point, consider the following cases:

(47)

(A) B

B

us \syo 'cow-Obj'

unaccented

ussyo ka\ wu LHL

unaccented LHL

LHL

accented LHL

σ \ru

us]sya 'cow-Top' unaccented

LHL 'is here'

LHL

initial-accented

ha |sira

hi | si r a 'pillar 1

unaccented

LHL

initial-accented ka j ga | mi 'mirror'

unaccented

LHL

medial-accented

LHL

accented LHL kono ka \ ga \mi

LHL accented LHL

In the examples in (47b) and ( 4 7 c ) , the initial-accented words behave the same way as the other LHL melody words. This shows that the initial-accented words in the Owase dialect have the LHL melody, as observed in the discussion above. What happens to other cases of the sequence AB? The answer to this question is that elsewhere, the phrase B is interpreted as extratonal, as indicated in:

(48)

A (B)

144

Shosuke Haraguchi

Thus, consider the following cases:

A

(49)

na |

na \i inoti initial-accented LHL

'no 1

initial-accented 1

LHL

initial-accented c. kawu 'keep' HL

'fall' HL

71

, ι|noo kawu

LHL

unaccented HL

initial-accented LHL

u \si

kawu I usi

unaccented LHL

unaccented

a/ne 1 rain'

hum | ante

unaccented

unaccented

HL

HL

HL

hune\mo

e. n o l r u 'embark'

1

ship-also'

unaccented LHL

second-accented

•this 1

ito 'string'

unaccented

LHL

' cow 1

d. huru unaccented

initial-accented kawu 'keep 1

dog-Obj'

unaccented

(B)

LHL

unaccented

HL

unaccented LHL koIno I ito

HL

unaccented LHL

The accentual and melodic information of A extends over the whole phrase AB/ and the phrase B is changed to L-toned, because phrase B is analyzed as extratonal, with the final L of the BTM spreading automatically to B. The discussion above shows that the notion 'extratonal', which has been introduced into tonal theory by Pulleyblank (1983), plays a crucial role in determining the surface tonal melody of the sequence of phrases. In addition to this ( 4 6 ) and (48) show that the Owase dialect has a tonal process which is dependent on the difference in the BTM and that it

needs information concerning

accent. Now, let us consider what will happen if a pause intervenes between A and B. To see this, consider the following cases:

(50)

a. kaz \zya 'wind-Top 1

+

unaccented LHL

huita 'blow-Past' unaccented

kazzua ι ι i'

HL

LHL second-accented

H L

L *

,

siroi

ito \mo •string-also'

initial-accented

[+pause] huita

HL

r

i

L+pauseJ

· *

ito-mo

I \; .---' ' ' II'·"" LHL"

Pitch Accent in Japanese

145

If these two phrases are connected loosely, the surface tone melody of the second member of the phrase AB will be realized in a somewhat different way. Thus, in ( 5 0 a ) , the first two Η-toned moras of 'huita' appear to be

turned

into L-toned moras. Similarly, the first Η-toned mora of 'itomo' in (50b) appears to be turned into a L-toned mora. The apparent changes of this sort in tonal realization will be explained automatically if we simply assume the principle of one-to-many association in ( 1 9 ) . This very fact, in turn, indicates that the principle in (19) is well-motivated. Consider now the phrase with '-mo'. Suppose that this is analyzed as extratonal. Then, the following phrase sequence should undergo ( 4 6 ) : (51)

kodo[mo\-(mo) + 'children-also 1

ta[ o |reru 'tumble'

unaccented

second-accented LHL

LHL

This is because both A and Β are LHL melody words and A is supposed to be accentless in ( 5 1 ) . Thus, the extratonal analysis of 'mo' would predict that (51) is turned into ( 5 2 ) :

(52)

kodomo-mo ta\o[reru

However, judging by Kindaichi's description, this prediction is incorrect, and the well-formed surface melody of (51) should be: (53)

kodo\mo\-mo taoreru

This shows that 'kodomo-mo'

should be penultimately accented, behaving in the

same way as the other accented words. Consequently, we are justified in concluding that the morpheme 'mo 1 is either pre-accenting or assigned the accent by rule ( 4 5 ) . On the basis of the discussion above, I conclude that the Owase tonal system has a number of theoretical implications which are highly crucial in deciding which competing theories are correct.

6.

On Special Tone Melodies

The preceding sections were devoted solely to the problems concerning the BTM

146

Shosuke Haraguchi

and its

association with tone-bearing units. Let us now turn to a number of

problems concerning special tone melodies, which are specific analyses of intonational aspects of Japanese. Special tone melodies in our language can be divided into two types: those that are added to the outputs of the BTM and those that replace the BTM. As for the former type, consider first the question intonation, which usually accompanies both yes-no and wh-questions in a polite style:

(54)

a.

(i)

Tookyo\o-ni ki\no \o "Tokyo-Dir1 'yesterday 1

(ii) b.

ko\re-wa na\n desu\ka? 'this-Top 1 ' w h a t ' ' i s ' ' Q '

c.

a\nata-wa 'you-Top'

do\na \ ta? ' who '

The final rise in question intonation, which usually accompanies the interrogative mood 'ka' or its

(55)

empty category, will be analyzed as in ( 5 5 ) :

Interrogative mood

(Aa) H

If the interrogative mood has a phonetic form 'ka', the H tone is associated with it

later by an appropriate principle; if it has no phonetic form, the H

tone is associated with the final mora of the preceding phrase, as illustrated in

(54aii) and , ( 5 4 c ) . Thus, this special tone melody respects the BTM and

affects the sentence-final mora only. The fact that this melody is associated with the sentence-final mora follows from the fact that the interrogative mood is located in sentence-final position, which seems to be a universal property of the mood in question with some possible parametric variations. As another case of the special tone melody, consider the following case:

(56)

tu|g*|-wa|a 'next-Top'

ko\n \do TiIba\ken \-no a \ru\ u ο \hanasi 'this time' 'Tiba Prefecture-Gen 1 'certain' 'story'

ojhyakusyo \sanj-no 'farmer-Gen'

ο\uti-de 'house-Loc 1

go \boo-cja 'burdock-Sub'

Pitch Accent in Japanese

147

Kindaichi (1951) calls this melody Muraoka Hanako Tune after her name. The characteristic of this tune is clear and simple enough: it raises every phrasefinal mora to a H tone and the original H tone of the unaccented words are slightly lowered, becoming lower than the phrase-final H tone. It should be pointed out here that this phonetic behavior of unaccented words could be handled easily if we introduced accent marks into the theory. This tune will be characterized as follows: The speaker has a special rule which inserts the H tone in every phrase-final position, which is associated with the final mora and the principle of automatic dissociation mentioned in section 3 erases the association line drawn originally between the final mora and the final tone of the BTM. These special tone melodies of the first type are added to the sentences or phrases after the association of the BTM, probably by appropriate postlexical processes. Now let us go on to the second type of special tone melodies, which completely replace the BTM. As a typical example of this melody, consider the following melody, which I name the Zengakuren Agitation Tone after the radical student agitators: (57)

Wareware\-wa\a 'we-Top 1

zettai\ni [i konoyoo\-na\a 'definitely 1 'this kind'

America\no\o ... 'American'

An examination of ( 5 7 ) immediately shows that this tune lengthens the final syllable to two moras and that the tone melody in question is LHL. This melody is associated with each phrase in a one-to-one fashion from right to left. This means that the association of this tune with the text is due to the principle in ( 1 4 ) with the value of the parameter fixed as (b) ( i . e . , from right to l e f t ) . Notice that this special tone melody uses a different association principle from the one required with respect to the BTM. To put it differently,

this tune

is governed by a principle entirely different from the one which governs the association of the BTM. It seems that every special tone melody of this type is accompanied by a favorite association principle. This point will be confirmed by an examination of the following special tone melody: (58)

ti[η omoo ni 'His Majesty thinks 1

wa\ga kooso kooso ... Our great ancestors'

148

Shosuke Haraguchi This tune, which is named 'Chokugo Hoodoku Tune 1 (Recitation Tune of the

Code of Education) by Kindaichi (1951), consists of the repetition of a rather monotonous LH melody. It was popular in Japan before World War II,

but

it

is not in use anymore. (58) shows that this LH melody is associated with each phrase in a one-toone fashion, from left to right. This means that this tune, in contrast with the Zengakuren Agitation Tune, combines the principles in (14) with the value for the parameter fixed as ( a ) . The observation above will be sufficient

to show that the special tone

melody of the type under consideration is specified with the particular tone association principle with values fixed appropriately. As far as I can see, these two types of special tone melodies are the only possible types found in Japanese intonation. If it is proven to be correct to restrict the intonational varieties of Japanese to these two types, then it will be a step toward the establishment of a fully-fledged theory of intonation.

7.

Concluding Remarks

The discussion so far has almost exclusively been concerned with and based on certain tonal phenomena of Japanese. I hope to have demonstrated some of the central characteristics of Japanese accentual and nonaccentual systems. In particular, we have seen that the selection of the BTM's from the set of possible BTM's of UG is made by fixing values for the parameter, and that various tone association processes are reformulated in terms of a number of universal principles with a few parameters. It is suggested that the surface tonal melodies of Japanese dialects are determined by an interaction of a number of these principles with the values fixed for the parameters. I have also presented a few arauments to defend the position adopted by the proponents of the standard theory of autosecimental phonoloqy such as Goldsmith (1976) and Haraguchi ( 1 9 7 7 ) , amonq others, in which an accentual diacritic (*) is lexically specified on a particular syllable in accented words and phrases. As far as I can determine, the alternative theory involving pre-linking of the BTM to each accented lexical item is less adequate than the accentual theory, both factually and conceptually. It is less adequate conceptually because the pre-linking theory specifies the BTM in each lexical item. This is extremely redundant, thus making the lexicon complicated. In addition to this, while the accentual theory claims that the BTM of the Tokyo dialect is restricted to a

Pitch Accent in Japanese

149

single type; the prelinking theory leaves the possibility that some lexical items might have a number of additional BTM's; thus the fact that Tokyo has just one BTM would become clear only after we have examined all

the lexical

items in the lexicon. Since that is totally counter-intuitive, the pre-linkinq theory will be the last choice to make. The discussion in this article has, I hope, clarified a number of issues concerning Japanese tonal phenomena and has shown that the revised version of Haraguchi (1977) presented in section 3 is more adequate than the alternative proposals recently advanced by Clark (to appear), Archangel! and Pulleyblank (1984) and others.

Notes

*I am grateful to Noriko Tanaka and Wayne P. Lawrence for many helpful comments and suggestions on an earlier draft of this article. 1.

In contrast, Clark (to appear) observes that ' [p]honetic studies such as that of Weizmann (1969) [Japanese Accent.· An Analysis Based on AcousticPhonetic Data. Doctoral dissertation, University of Southern California at Los Angeles —SG] have shown that the "accented" mora is not consistently longer or more intense than other moras . . ' As Weizmanns dissertation is not available to me at present, I cannot, unfortunately, evaluate the validity of her claim.

2.

For the full set of facts under consideration, see Haraguchi ( 1 9 7 9 b ) , Uwano (1977) and references therein.

3.

Herr and everywhere else, V stands for a tone-bearing unit, which is considered to be a mora in most Japanese dialects.

4.

For related discussion, see Kiparsky ( 1 9 7 2 ) , Leben ( 1 9 7 3 ) , Hayes (1980), Prince ( 1 9 8 3 ) , and Halle and Mohanan (1985), among others.

5.

There seem to be some unsolved problems concerning the principle of the automatic dissociation. Thus, in the Kumi dialect, the principle does not operate on the contour tone introduced by the Final Rising Contour Formation rule, while it operates elsewhere. See Haraguchi (1977,1978) for relevant data and discussion. This dialect seems to pose a problem, the solution of which is unclear to me at present.

6.

The dotted horizontal line in (38) stands for the M tone.

References Archangeli, D. and D.G. Pulleyblank ( 1 9 8 4 ) , Extratonality and Japanese Accent. Unpublished paper, MIT. Clark, M . M . (to appear), Japanese as a Tone Language. In: T. Imai and M. Saito ( e d s . ) , Issues in Japanese Linguistics. Dordrecht. Foris.

150

Shosuke Haraguchi

Clements, G . N . ( 1 9 7 5 ) , Tone as Speech Melody. Unpublished paper, MIT. Goldsmith, J. ( 1 9 7 6 ) , Autosegmental Phonology. Indiana University Linguistics Club. [Published by Garland, 1979.] Halle, M. and K.P. Mohanan ( 1 9 8 5 ) , Segmental Phonology of Modern English. Linguistic Inquiry 16:1, 57-116. Halle, M. and J . R . Vergnaud ( 1 9 8 2 ) , On the Framework of Autosegmental Phonology. In: H. van der Hülst and N. Smith ( e d s . ) , The Structure of Phonological Representations (Part I ) , 65-82.Dordrecht. Foris. Haraguchi, S. ( 1 9 7 7 ) , The Tone Pattern of Japanese: An Autosegmental Theory of Tonology. Tokyo: Kaitakusha. Haraguchi, S. ( 1 9 7 8 ) , The Tone System of the Kumi Dialect: An Autosegmental Analysis. Descriptive and Applied Linguistics 11, 65-85. Haraguchi, S. ( 1 9 7 9 a ) , Nihongo Oncho no Shoso (Aspects of the Tone systems of Japanese Dialects). Gengo no Kagaku (Science of Language) 7, 21-69. Haraguchi, S. (1979b), Tonology of Japanese Dialects. In: Bedell, G. , E. Kobayashi, and M. Muraki ( e d s . ) , Explorations in Linguistics: Papers in Honor of Kazuko Inoue, 125-46. Haraguchi, S. (to appear), The Multi-dimensional Grammatical Theory. In: Imai, T. and M. Saito ( e d s . ) , Issues in Japanese Linguistics. Dordrecht: Foris. Hayes, B. ( 1 9 8 0 ) , A Metrical Theory of Stress Rules. Doctoral dissertation, MIT. [Revised version (1981) is available from the Indiana University Linguistics Club.] Kindaichi, H. ( 1 9 5 1 ) , Kotoba no Senritsu (Melodies of Language). Kokugogaku (Studies in the Japanese Language) 5, 37-59. Reprinted in Nihongo On'in no Kenkyu, 78-110 (1967). Kindaichi, H. ( 1 9 7 5 ) , Nihon no Hogen (Japanese Dialects). Tokyo: Kyoiku Shuppan. Kiparsky, P. ( 1 9 7 2 ) , Explanation in Phonology. In: S. Peters ( e d . ) , Goals of Linguistic Theory, 189-227. Kobayashi, Y. ( 1 9 8 5 ) , Hirosaki Hogen no Oncho Merodii (Tone Melody in the Hirosaki Dialect). Gengokenkyu (Journal of the Linguistic Society of Japan) 87, 40-67. Leben, W . R . ( 1 9 7 3 ) , Suprasegmental Phonology. Indiana University Linguistics Club. Ostler, N. (1977) , High Vowels and Accentuation in Two Dialects. Unpublished paper, MIT. Ostler, N. ( 1 9 7 8 ) , Autosegmental Theory and Japanese Tone-Spread. NELS 8, 141-53. Prince, A.S. (1983), Relating to the Grid. Lingruistic Inquiry 14.1, 19-100. Pulleyblank, D.G. (1983), Tone in Lexical Phonology. Doctoral dissertation, MIT. Saikawa, T . , Y. Wakahara, S. Makini, S. Itabashi, and K. Kido ( 1 9 7 0 ) , 4 Mora Muimi Tango chu no Onso. Mora no Iti oyobi Akusento to Zizoku Zikan no Kankei (On the Relation between Accent, Place and Duration of Morae or Phonemes in Nonsense Words of 4 Morae). Denki Onkyoo Kenkyukai Shiryo, 1-22. Shibatani, M. ( 1 9 7 9 ) , Review of the Tone Pattern of Japanese: An autosegmental Theory of Tonology. By Shosuke Haraguchi. Tokyo: Kaitakusha, 1977. Language 55.4, 928-36. Uwano, Z. ( 1 9 7 7 ) , Nihongo no Akusento (Accent in Japanese). Nihongo 5 On'in, 281-321. Zubizarreta, M . L . ( 1 9 8 2 ) , The Formal Interpretation of Harmony and Accent: The Tone Pattern of Japanese. In: H. van der Hülst and N. Smith ( e d s . ) , The Structure of Phonological Representations (Part I I ) , 159-212.

Tonal Units as Constituents of Prosodic Structure: The Evidence from English and French Intonation Daniel Hirst CNRS, Institute de phonetique d'Aix en Provence 0.

Introduction*

One of the earliest criticisms of distinctive feature theory was that the descriptive apparatus was insufficient to handle the facts of intonation: There is some discussion of this problem but it is very sketchy, and it remains to be seen whether the proposals made can do justice to the full range of intonational phenomena. (Chomsky 1957 p . 2 4 2 ) . With the development of generative phonology within the next ten years, the situation hardly improved on this point since as the authors of Sound Pattern of English themselves admitted: Our investigations of these features have not progressed to a point where a discussion in print would be useful. (Chomsky and Halle 1968 p . 3 2 9 ) . It is perhaps not surprising that both of the recent major theoretical innovations in the general framework of generative phonology, that of Metrical Phonology (Liberman 1975) and that of Autosegmental Phonology (Goldsmith 1976) have been brought forward partly as an attempt to incorporate these "intonational phenomena" into linguistic descriptions. In fact the development of the descriptive framework has been such, in recent years, that it now appears that we are, if anything, faced with the opposite problem: phonological representations can be constructed using several separate segmental tiers, hierarchical structure, phonological category labels, prominence markings, as well as metrical grids. It is obvious that representations incorporating all these are highly over-redundant. We consequently need to ask which are the primitive elements of phonological representations, and which (to use Alan Prince's formula) are merely "epiphenomenal projections" of the phonological primitives. The question is,

as linguists are wont to say,

an empirical one, and one that we can expect to be very much in the

forefront

152

Daniel Hirst

of phonological research in the near future. Halle & Vergnaud (forthcoming) have argued, for example, that both a metrical grid and a tree-structure are necessary parts of a phonological representation. Prince (1983) and Selkirk (1984), on the other hand, have claimed that given grids, constituency is redundant. In this paper I argue that the linguistic description of English and French intonation seems at first sight to favour a constituency model using phonological categories, but that in fact this is because there is a crucial interaction between the way in which we model fundamental frequency curves and our choice of phonological model.

1.

The acoustic domain approach to intonation

Unlike phonology which deals with discrete, discontinuous elements, phonetics deals essentially with continuous phenomena. This can easily be illustrated if we compare for example a phonological representation of a sentence like "Why were you away?":

(1)

/waiw9juw3wei/

with a spectrogram of the same sentence. In order to derive a set of continuously varying parameters from a finite set of phonological elements, we need to specify a function with a finite number of parameters, and then hope to be able to derive the values of the parameters from the phonological elements. In the case of the vowels, we might assume that each phonological segment of ( 1 ) corresponds to a target-value for the formants, and then associate an appropriate time-value to each phonological element. We would further need to assume that the transitions between the target-values are entirely determined by the phonological/acoustic

function.

Our choice of phonological/acoustic function will, consequently, have repercussions for the phonological model. Note that this is particularly true in the case of intonation where there is less general consensus than in the case of vowels as to the nature of the phonological representation. Turning now to the representation of intonation, a fundamental frequency curve such as that in Figure 1 can be analysed as the interaction between two curves which we can think of as the underlying fundamental frequency on the one hand ( c f . Figure 2) and the residual "microprosodic profile" on the other hand (Figure 3 ) . Such an analysis allows us effectively to factor out an

Constituents of Prosodic Structure

153

intonation contour into the "tune-component" on the one hand, and the "wordcomponent" on the other. The same tune can of course be used with different words and the same words with different tunes.

f

f

^| ··

M

·

·

·

·

•

·

•

250~ - · · · · · £ t

0

"*

"

*

·

·

·

·

tD

™ ™

·

·

·

·

· **·— _10)

· ·

.

.

·

100"™· f

^J V

^

.·

·

·

•

·

·

·

· ^J |3 ^

M

·

·

f

^J Mi

MM

·

·

·

· Irt

·

·

·

·

·

*

»

«

·

·

·

·

·

.

·

·

!

.

·

·

·

·

·

.

. _

.

IV

rf». · . .BB 6 * * *

·

·

·

D

·
a 32

5 3 2

Aara" zu

nara"°a 3

Tone Epenthesis

'used 1

1 3

I

179

3 2 1 3

1 I V

'nara'^a

5

3 2

y

ka ' ra "zu

3

5

1

f\j

5

3 2

.fca'ra 3

3 2

1 1 Aa'ra "zu

In this analysis, the tone break in words like (28c) does not have metrical consequences, but is simply a result of rule ordering. Because the tone raising rule applies after the association processes shown in ( 2 9 ) , a form like 'got married 1 in column two already has tone 2_ associated with the penultimate syllable when it undergoes the raising rule, even though tone 2_ is not present in the underlying form. The replacement of 13 by 4_ is a simple process that yields

3

2

4

'na ra "'a

5.

"zo

l~>

'you got married'.

The Place of Copala Trique in a Typology of Tone Systems

The above description shows that CTr has a tone system with a considerable degree of complexity, but that not all

syllables carry contrastive tone.

(This should be clear from the data presented, even to those who would prefer a different analysis of those data.) The question that needs to be asked, therefore, is:

in what category does such a language fall? And the answer obviously

depends on the categories that are chosen, and the way in which each is defined. A familiar dichotomy that has been used for decades is the opposition between tone languages and pitch-accent languages. It is far from clear just how these terms are to be defined, but the term pitch-accent languages has been consistently used for languages that are in some sense less pervasively tonal than tone languages, and the definition of pitch-accent language therefore depends to some degree on the definition of tone language. For example, if K . L . Pike's original definition of a tone language is employed (= a language in which each syllable carries contrastive t o n e ) , then the term pitch-accent language can be used for languages in which the domain of tone

180

Barbara Hollenbach

is something larger than a single syllable, such as a word or foot, with tone contrasts occurring only on the accented syllable of this domain. Suppose, however, that we choose a less restricted definition of a tone language, such as the presence of tonal information, i.e., paradigmatic contrast in a single position, in lexical representations. The term pitch-accent language can then be used for languages which do not have tonal information in their lexical representations, but which do have significant nonintonational tone in surface forms. This tone is usually the principal realization of a syntagmatic contrast, namely, accentual placement. This is the position taken by McCawley (1978). Note that, if accent is permitted to fall not only on entire syllables, but on morae within syllables, many so-called tone contrasts can be subsumed under accent. See, for example, McCawley's discussion of Lithuanian (pp. 129-30). I return now to the case of CTr. If the second pair of definitions is chosen, CTr is clearly a tone language because paradigmatic tone contrasts must be specified in lexical representations. If, however, the first pair of definitions is chosen, then CTr is somewhere in limbo. It is not a tone language because it does not have contrastive tone on every syllable, but it

is not a pitch-

accent language either because tone contrasts are not limited to accented syllables. It seems clear, therefore,

that there are at least four categories of (non-

intonational) tonal phenomena. There are (1) syllable tone languages, like San Miguel El Grande Mixtec and Huautla Mazatec, that follow Pike's definition. There are (2) restricted tone languages, like CTr, Mazahua, and Kikuyu, in which some, but not all, unaccented syllables are toneless. There are (3) toneaccent languages, like Fame, in which tonal contrasts are limited to accented syllables. And there are (4) accent languages, like Standard Japanese and Lithuanian, in which tone contrasts do not occur in lexical representations, but in which tone realizes the placement of accent. The theoretical stance that linguists adopt may affect their perception of the significance of the above distinctions. For example, the development of autosegmental phonology has made it easy to handle languages in which there are fewer contrasting tones than there are syllables, and a linguist working in this model will probably consider the distinction between types (1) and (2) to be fairly trivial. This is the position I currently hold. Furthermore, a type (1) language can become a type (2) language via a fairly simple historical change. For example, Clements and Ford (1979:187-92) have shown that Kikuyu, a language of type ( 2 ) , developed from a protolanguage of type (1) via movement

Tone in Copal a Trique

181

of tone to the right. The same process of rightward tone movement was also a factor in the development of CTr from a type (1) system (Longacre 1957:100101,

103-110). For these reasons, I consider CTr to be more closely related to

languages of type (1) than to languages of types (3) and ( 4 ) , and I therefore prefer to call it a tone language, rather than a pitch-accent language. At the present state of our knowledge, however, it seems fruitless to argue about how the two traditional terms should be applied to individual languages. Researchers should instead address themselves to two more important tasks. The first of these is to study more languages with tonal phenomena in order to determine the variety of ways in which tone is distributed in words. And the second one is to search for a typological scheme into which all languages can be fit

without doing violence to the facts. It is highly unlikely that my

four-type scheme will prove adequate for such a task, let alone a simple dichotomy between tone and pitch-accent. I predict, however, that an adequate typology will have to recognize both tone (involving paradigmatic contrast) and accent (involving syntagmatic contrast) as independent parameters. In my tentative scheme, tone is relevant to languages of types ( 1 ) , ( 2 ) , and ( 3 ) ; and accent is relevant to at least types (3) and ( 4 ) .

Notes *I would like to express my appreciation to Doris Bartholomew, Carol Mock, Deirdre Wheeler, and the editors for reading an earlier draft of this article and offering many helpful suggestions. 1.

Copala Trique is spoken by about 8.000 people living in the districts of Juxtlahuaca and Putla, Oaxaca, Mexico. The data on which this article is based are from my unpublished field notes, gathered on trips to the Trique area from 1962 to 1980 under the auspices of the Summer Institute of Linguistics. The analysis presented here is a summary of portions of my University of Arizona doctoral dissertation (Hollenbach 1984), and the reader is referred to this longer work for a fuller treatment.

2.

Further details about the acoustic study of Copala Trique tone, including a sample narrow-band spectrogram for each of the five tone levels, are given in Hollenbach (1984:72-85). These spectrograms were produced on a Kay Digital Sona-Graph 7800 and Kay Sona-Graph Printer 7900, which were purchased under Grant BNS 8309112 of the National Science Foundation, whose help is gratefully acknowledged.

3.

Some current approaches to linguistic theory, such as lexical phonology and morphology, developed by Kiparsky (1982) and Mohanan ( 1 9 8 2 ) , claim that such postlexical, word-external morphological rules should not be permitted. Clear examples of such rules are, however, found in both CTr and San Miguel El Grande Mixtec, described in 1948 by K.L. Pike (pp. 7781). A detailed defense of word-external morphology in CTr is given in

182

Barbara Hollenbach Hollenbach (1984:260-379)

References Clements, George N. and Kevin C. Ford (1979) , Kikuyu Tone Shift and its Synchronic Consequences. Linguistic Inquiry 10, 179-210. Gibson, Lorna F. (1956) , Fame (Otomi) Phonemics and Morphophonemics. International Journal of American Linguistics 22, 242-65. Goldsmith, John ( 1 9 7 6 ) , An Overview of Autosegmental Phonology. Linguistic Analysis 2, 22-68. Goldsmith, John (1979), The Aims of Autosegmental Phonology. In: D.A. Dinnsen ( e d . ) , Current Approaches to Phonological Theory. Bloomington: Indiana University Press, 202-222. Halle, Morris and Jean-Roger Vergnaud (1982) , On the Framework of Autosegmental Phonology. In: Harry van der Hülst and Norval Smith ( e d s . ) , The Structure of Phonological Representations (Part I). Dordrecht: Foris Publications, 65-82. Haraguchi, Sh6suke (1977) , The Tone Patterns of Japanese: An Autosegmental Theory of Tonology. Tokyo: Kaitakusha. Hollenbach, Barbara E. ( 1 9 8 4 ) , The Phonology and Morphology of Tone and Laryngeals in Copala Trique. University of Arizona doctoral dissertation. Hollenbach, Barbara E. (1985), Copala Trique Tone and Universal Tone Features. In: Stuart Davis ( e d . ) , Coyote Papers 5. University of Arizona, Department of Linguistics, 96-119. Kiparsky, Paul ( 1 9 8 2 ) , Lexical Morphology and Phonology. In: Linguistics in the Morning Calm: Selected Papers from SICOL-1981. Seoul: Hanshin Publishing Company, 3-91. Longacre, Robert E. ( 1 9 5 7 ) , Proto-Mixtecan. Indiana University Research Center in Anthropology, Folklore, and Linguistics, Publ. 5, Indiana University, Bloomington. McCarthy, John (1981), A Prosodic Theory of Nonconcatenative Morphology. Linguistic Inquiry 12, 373-418. McCawley, James D. (1978), What is a Tone Language? In: Victoria A. Fromkin ( e d . ) , Tone: A Linguistic Survey, New York: Academic Press, 113-31. Mohanan, Karuvannur Eathanveettil (1982), Lexical Phonology. Reproduced by Indiana University Linguistics Club. (Originally an M.I.T. doctoral dissertation, 1981.) Pike, Eunice V. ( 1 9 5 1 ) , Tonemic-Intonemic Correlation in Mazahua (Otomi). International Journal of American Linguistics 17, 37-41. Pike, Kenneth L. (1948), Tone Languages. Ann Arbor: University of Michigan Press. Williams, Edwin (1976), Underlying Tone in Margi and Igbo. Linguistic Inquiry 7, 463-84.

Hungarian Sentence Intonation Andräs Kornai & Läszlo Kaiman Hungarian Academy of Sciences 0.

Introduction*

The present paper describes the patterns of Hungarian sentence intonation using the model developed by Goldsmith ( 1 9 8 2 ) . In Part 1, we attempt to motivate this particular choice of framework, and to outline the interaction of our rule system with the rest of the grammar. Here certain basic principles of lexical phonology will be assumed, rather than argued. The rule system itself will be given in Part 2, and in Part 3 we offer our conclusions. The intonation patterns generated by the rule system are based on the descriptive work of Varga (1981): some minor differences are discussed in the text.

1.

The organization of the grammar

Since intonation contours belong to the sentence as a whole, the rules generating them must be in the postlexical component of the grammar. Thus, in the lexical phonology framework, we must suppose that the rule system operates on lexically fully specified representations. Because of the traditional organization of generative grammars such a representation is not available until lexical insertion. This means that the rules can make use of information provided by the syntax. In Section 2 we will see that only a diacritic f and the category labels will in fact be used (the latter only in rules of cliticization). Our rule system is not sensitive to the segmental material of lexical items: the underlying representation will borrow only the inherent stresses and the syllable structure from the lexicon. The rules are governed by pragmatic information we presume that the semantic component can only interpret sentences with fully specified intonation contours. Although Hungarian is not a pitch-accent language even under the most liberal interpretation of this term, the following phenomena show clearly that an autosegmental treatment of sentence intonation is called for:

184

Andräs Kornai & Läszlo kälmän

1) Contour tones on short vowels. Bort.

'wine-Ace'

l\

Bort?

/V,

HL

LHL

Bort,...

A\

HLH

2) Tonal melodies. Hungarian has four distinct tonal melodies: The 'rising 1 and the 'falling 1 contours of Varga (1981) are conflated here: see 2.4 below. Megfeszitettek. H

L

L L

'They crucified him.'

L

'Did they crucify him? 1

Megfeszitettek? L

L

L H

L

Megfeszitettek... H

H

H H

'They crucified him, (and t h e n . . . ) '

H

'They did crucify hum, ( b u t . . . ) 1

Megfeszitettek,... H

L

L L

H

3) Spreading. Faj?

Fajta?

A

L H L 1

1

race

IN

L

H L

Fajtalan?

IM

L H L

' species' ' sodomite'

Fajtalanok?

VN L H L

' sodomites'

Faitalanokat?

\y \ \

L H L

' sodomites-ACC

Since on the postlexical level no segmental material is ever deleted, the remaining diagnostics (i.e. stability and floating tones cf. e.g. V . d . Hulst and Smith 1982:2.2) are simply not applicable. The well-formedness of autosegmental representations is usually maintained with the aid of an Association Convention that can reapply at every step of the derivation ( c f . e.g. Clements and Sezer 1982:219). At the postlexical· level, however, no rule can be applied more than once (see Kiparsky 1983). The Association Algorith (AA) employed in this paper differs from the familiar mechanisms like Tone Mapping (Williams 1 9 7 1 ) , Well-Formedness Conditions (Goldsmith 1976), etc. essentially only in having this Once only' property. (1)

Association Algorithm 1 (i)

The domain of association is the phonological word.

(ii)

Associate the nearest right-hand neighbors of the previously associated elements with each other.

(iii)

Apply (ii) as many times as possible, then repeat this process with left-hand neighbors.

(iv)

Associate the remaining unassociated syllables (P-bearing elements) with the tone (P-segment) associated to their neighbors.

After ( i v ) , the representation cannot contain unassociated syllables, but un-

Hungarian Sentence Intonation

185

associated tones might still be present. We suppose that in general these are simply deleted; in Hungarian, however, we also have (v)

Associate word-final unassociated tones to the last syllable.

The first association line (the starting point of the AA) is provided by the Basic Tone Melody Association (BTMA) rule of Goldsmith 1982: Associate the accented element of the Basic Tone Melody to the accented element of the word. Thus, the organization of the grammar can be schematized as follows:

(2)

Lexical Phonology and Morphology

Syntax

1 Lexical Insertion (Surface) Syntactic Representation [cliticization Rules! Postlexical Underlying Representation |Accent Rules| Underlying Tone Level BTMA Surface Accent Level

Superficial Representation Low-Level Rules Surface Phonetic Form

The portion of (2) starting with Postlexical Underlying Representation is essentially the same as the schema proposed by Goldsmith (1982:48): the intermediate levels here termed 'Surface Accent Level' and 'Superficial Representation' are introduced solely for the sake of expository convenience. Cliticization rules and low-level rules are tangential to our main purpose; they will be discussed briefly in Section 2.1 and Section 2 . 4 , respectively. The term 'Surface Accent Level·' is justified by the fact that the ruie assigning

186

Andräs Kornai S Läszlo Kaiman

surface stress can be collapsed with the BTMA: for details see Section 2.2 where the accent rules of Hungarian are described in full.

2. 2.1

The rules Rules of cliticization

In most cases the definite article is generated by the syntax under a separate node (for exceptions see Kornai 1984), and in complex NPs the article will not be a sister of the element following it.

Nevertheless, the article is in the

same phonological word as the subsequent item. Since 'lexical' word boundaries and phonological word boundaries do not always coincide in Hungarian, 2 it

is

necessary to apply certain "readjustment rules" in the sense of Chomsky and Halle ( 1 9 6 8 : 9 f f , 3 7 1 f f ) .

These rules are trivial for connectives like ha 'if

(or is 'too') which are always proclitic (enclitic). Personal pronouns, however, behave proclitically only if they precede the main verb of the clause: otherwise, they are enclitic. If we suppose that the usual word-initial stress is supplied by a lexical redundancy rule, and proclitics (enclitics) are marked as such in the lexicon, then the remaining cases can be handled 'with rules that are sensitive only to the categorial status of the adjacent elements. One such rule will delete the (lexical) stress from verbs preceded by some VM (Verbal Modifier, see Ackerman and Komlosy ( 1 9 8 3 ) ) . Ackerman (1984) demonstrates that VM+V constructions can be lexical: in such cases the required stress pattern comes simply from the lexical redundancy rule mentioned above.

# [ Jozsef ] [ VM fat ] [ V vägott ] [ az erdon ] # * * * „ # [ Jozsef ] [ fät vägott ] [ az erdon ] # Joseph wood-ACC cut the forest-in'

If such rules are part of the grammar, the relatively open nature of the V class can readily be explained: the Derivational Simplicity Criterion of Kiparsky (1982) forces us to reanalyze VM+V constructions as Vs.

In a 'per-

colation' theory of word-formation (such as Lieber 1980), no reanalysis is called f o r , since the trees resulting from compounding, affixation, and clitic ization are identical.

Hungarian Sentence Intonation 2.2

187

Accent rules

Certain words appear in the syntactic representation with a diacritic f (read: Foykes). Words in Focus position (See E. Kiss, 1981) or having particular communicative functions (contrastive topic, see Szabolcsi, 1981; contrafocus, see Varga 1982; etc.) are marked by this diacritic. The role of f in the formation of the underlying phonological representation is that elements marked by it cannot cliticize. The rules generating the intonation patterns of corrective sentences

(phrases) are also triggered by f. If a word is marked by f in

the (postlexical) underlying representation, the Eradication Rule (ER) will delete the accent and word boundaries of the subsequent words up to the next f or phrase boundary ( # ) .

E

SO

S

3-1

[f

so s

SO

SO

S

SO

(SO

S

SO)

#

J

Therefore, the segments from one f to the next (or to the end of the phrase) will form one phonological word possessing only one accented syllable. This 'eradicating 1 stress (see Kaiman et al. 1984) is generally believed to be stronger than the normal stress. Minimal pair tests, however, do not show a linguistically significant contrast: in our opinion the quantitative difference between normal and eradicating stress is a secondary phenomenon due to the fact that eradicating stress generally appears on substantially longer phonological words (see also Kaiman et al.

1984). Nevertheless, it is possible to

reformulate the BTMA in such a manner that the surface stresses of the accented syllables depend on the presence or absence of f :

SO

SO

SO

' s

so

Τ

τ Τ

το

S

S

SO

m.i

3.

(only for Hungarian)

ι-· ο

Τ

TO

SO

)0

f

'S

I

LHL

SO

195

Pitch Accent and Stress in Isthmus Zapotec Carol C. Mock

1.

Tone / Accent Typologies*

Goldsmith 1982 notes that the choice of underlying tone melodies in language is governed by (a) lexical considerations, (b) grammatical factors,

(c) both

(a) and ( b ) , or (d) neither (a) nor ( b ) . Under ( a ) , individual morphemes or classes of morphemes bear several different tone melodies, and this fact must be registered in the lexicon; under (b) tone melodies are manipulated as expressions of syntactic structure, as in the KiNtandu dialect of KiKoongo (Daeleman 1983), or else a word's inflectional potential is carried in part by tone, as in the verb systems of Nzema (Mock 1969) and many Bantu languages (Clements and Goldsmith (eds.) 1984, Odden this volume) and Copala Trique {Hollenbach 1984). As Goldsmith points out, cases (a) through (c) are what normally happens in TONE languages such as KiNtandu, Nzema and Trigue, while (d) is what he terms ACCENTUAL. Goldsmith's schema considers the array of sources for the tonal melodies of a language, or where in the formal description of the language tone must first be represented. He claims that only if the surface tonal melodies are completely predictable by phonological rules, rather than being represented lexically of grammatically, is the system accentual. I shall argue that this is too strong a constraint on the term PITCH ACCENT. What is not clear from Goldsmith's presentation is the possible range of accent systems; on the one hand, what the nonlexical and nongrammatical sources of melodies can be, and on the other, how many variations of formal patterning are best analyzed as accentual. The languages he discusses, CiRuri and CiTonga, are said to have accent rather than lexical tone, but they have many similarities to tonal systems: several morphemes in the same word can be pitch accented and not all

words of the same major category are accented; e . g . ,

some verbs bear an accent while others do not. Thus the accentual status of each morpheme is a free variable here, just as the tone of each tone-bearing unit is,

in a tone language; and the presence or absence of accent must be

indicated lexically in these languages. If the location of the accent in polysyllabic words varies freely, this must also be represented in the lexicon. What Goldsmith means by stating that accentual systems do not have lexical

198

Carol Mock

and/or grammatical sources for tonal melodies is that there is only a single tonal melody (or ' a c c e n t ' ) in use. The major difference between accent languages and purely tonal ones, according to Goldsmith, is that there is but one tonal melody available in accent systems, or in other words, a simple binary contrast in some domain between the presence and absence of a single melody, which normally contains a H tone. 1 However, I shall argue in this article that there are other criteria for pitch accent which allow us to speak of pitch accent languages that make use of more than a single accented melody, and that Isthmus Zapotec is one such language. Accent languages like CiRuri and CiTonga can be viewed as having simple pitch accent systems with only one tonal melody. But the term pitch accent has also been used quite differently, in relation to intonation and the focus struct ure of a sentence in its 1961,

discourse context (for English, cf. Bolinger 1958,

Pierrehumbert 1980, Selkirk 1984). As an element of intonation, a pitch

accent is not tied to specific lexical items and seldom occurs more than once in a single word, nor is it constrained to a single tonal melody; Pierrehumbert lists more than five for English. It functions as part of the informational packaging of a text and is therefore likely to covary with other highlighting mechanisms such as marked word order and 'contrastive stress' (the contrastive pairing of grammatically analogous elements, e . g . , if HE does it, then You can TOO). We would do well to have two separate terms for lexical and intonational pitch accents, perhaps reserving PITCH ACCENT for the lexicogrammatical case, and inventing a new term such as TEXTUAL ACCENT for the broader intonational case. Of course, the world of real languages is diverse enough that there remain examples that fit

into several typological categories at once, such as Mazahua,

in which certain syllables bear tone and others bear intonation (E.V. Pike 1951). However, lexicogrammatical pitch accents and their differences from lexical tone systems are obviously of most concern in the articles of this book, rather than textual accents. The difference between (lexicogrammatical) pitch accent and tone is a formal one: a matter of how complex the lexically specified tonal information has to be. To do justice to the complex array of natural pitch accent systems, we will have to work out a measure of formal complexity which includes at least the parameters of tonal distribution and the number of melodies available, if not also the type and complexity of tonal rules. As for distribution, distinctive tones are possible on each syllable of every morpheme in the classical case of a pure tone language; in a restricted tone language only some syllables bear distinctive tone and the pitch of other syllables is

Pitch Accent in Isthmus Zapotec

199

either irrelevant or determined by intonation; and in pitch accent languages the distribution of tones is predictable once the location of the accent is specified as occurring on a certain accent-bearing unit (usually a specific syllable or mora of a morpheme). As for the number of tonal contrasts, if a single tonal melody is used, then only the presence of the accent needs to be represented; if more than one is possible beyond the simple choice of accented or unaccented melodies ( e . g . , contour accents), the identity of each melody must also be indicated, which multiplies the complexity of the system. A related issue is the number of tone levels: the logical cut-off point between pitch accent and restricted tone systems would seem to be where more than two levels of tone are necessary, in that when only two levels of pitch prominence are in use, the less prominent of the two is in many cases analyzable as the absence of the more prominent one, in a binary private opposition. Isthmus Zapotec fits into this sort of schema as a language in which the location of the pitch accent is fixed rather than free, but four different pitch accents are available—H, L, LH and HL. What I suggest as important criteria that together define a range of pitch accent systems are the following: a. Tonal melodies that are characteristic of morphemes instead of purely phonological tone-bearing units, and which are recognizably the same melodies whether part of long or short morphemes.

(Note that this criterion is as true

of textual accents as of pitch accents.) b. Either fixed or free Initial Tone Association Rules (ITARs); the case of free ITARs requires the star convention for indicating which tone-bearing unit the pitch accent is linked to, but fixed accents do not. c. Only two tonal levels, high (H) and low ( L ) , in paradigmatic opposition to each other. d. A finite number of tonal melodies not limited to only two (i.e., an accented melody versus an unaccented one) ,· those systems using more than two tonal melodies require lexical representation of each one. A concern not often clearly addressed is the relationship between pitch accent and the other major means of achieving phonetic prominence, STRESS. Insofar as it is possible to distinguish among types of 'accent' (Hyman 1977b), pitch accent and stress accent can cooccur in a single language, and one of the aims of this paper is to demonstrate that both are significant in Isthmus Zapotec. Yet the issue is complex: the phonetic manifestations of stress usually include pitch movement as well as increased duration and heightened intensity (Hyman 1977a), and thus the possible interactions between stress and pitch

200

Carol Mock

accent (or more generally, lexical tone) demand careful study. There do seem to be pitch accent languages which do not have stress as a separate phenomenon. Japanese appears to have the sort of system in which there is no need to posit both pitch accent and stress; its

accentual melodies can be described as tonal

manifestations of word stress (Zubizarreta 1982). Stress is basically a syntagmatic matter, as are the simplest pitch accent systems (those with only one melody), in that the placement of stress / pitch accent is more crucial to its formal status than is its phonetic realization, whereas lexical tone is quintessentially a matter of paradigmatic choices in more complex pitch accent languages such as Isthmus Zapotec. In the languages with which I am familiar, whenever both stress and pitch accent are present and capable of being analyzed separately, they both are borne by similar phonological units, which complicates the task of determining the phonological status of each. The purpose of this article is to describe both pitch accent and stress in Isthmus Zapotec. I will argue that they are independent of each other in this language, despite the fact that both focus on the same syllable of a morpheme: pitch accent is an inherent part of the lexical identity of each major-category morpheme, whereas stress is assigned according to discourse-related requirements and is therefore strongly constrained by syntactic relations. The description is couched in terms of SYSTEMIC PHONOLOGY, a multidimensional generative theory which builds on the tradition of British prosodic analysis (e.g.

Mock 1969, 1985, Prakasam 1972, 1976, 1977, cf. Martin 1976). In a system-

ic description, phonological oppositions (or distinctive features) are organized into a system network appropriate to their paradigmatic relations (their 'contrastive distributions', in American terms), so that selection from each of the binary features generates the universe of possible phonological choices for a language. Partially ordered realization rules detail the superficial characteristics by which each phonological feature is manifested. With respect to prosodic phenomena such as stress and tone, the realization rules define how each is manifested ( e . g . , either by autosegments or by modification of the CV t i e r ) , how they are tied to specific syllables and how this underlying structure relates to the phonetic level. 2 In this article I will use mainly autosegmental realization rules for the convenience of the readers, but systemic phonology is not ideologically committed to series of abstract derivations (Mock 1985 has less abstract realization rules for essentially the same prosodic phenomena). The crucial difference between systemic phonology and other current generative theories such as autosegmental phonology is that systemic theory formalizes the phonological oppositions to a greater extent, utilizing not

Pitch Accent in Isthmus Zapotec

201

only the concept of binary oppositions, but also portraying the paradigmatic relations among sets of oppositions. All regular gaps in the matrix of combinations of distinctive features are captured via the system network in this approach, rather than being explained away ' a f t e r the fact' by means of filters. Thus systemic phonological descriptions begin 'before' any statements about underlying representations, by characterizing precisely what choices are possible.

2.

Prosodic Oppositions in Isthmus Zapotec

Typologically, Isthmus Zapotec lies on the border between pitch accent languages and restricted tone languages. Two major facts support its pitch accent status: 1. there are only two tone levels, H (H) and low ( L ) , and 2. tonal melodies associate automatically with the moras of each morpheme, starting from

its

initial boundary regardless of the number of moras it has. At the same time, however, the language is deeply tonal, as are most other Otomanguean languages: unstressed

syllabic morphemes such as affixes and clitics bear distinctive

tone independently from the tonal specification of the roots on which they depend rhythmically,

tone sandhi affects any low tone that follows the LH pitch

accent, there are a few purely tonal morphemes 3 , and the realization rules modify the underlying melodies substantially, neutralizing some of the basic oppositions in certain contexts. Thus it would be difficult to argue that Isthmus Zapotec is a typical pitch accent language. Like other pitch accent languages, however, it allows only one distinctive choice of tone per morpheme. There are four d i f f e r e n t tonal melodies for root morphemes, two of which are contours: low ( L ) , high ( H ) , high-falling (HL) and low-rising ( L H ) . These basic options can be represented as a network of binary systems ( 1 ) :

(1)

.-High [-Single Tone

5 -Low

Root Morpheme pHigh-Low -Double Tone -Low-High

202

Carol Mock

The justification for the system Single/Double Tone is not only that it distinguishes between simple tones and contours (a phonetic matter), but also that it enables us to refer concisely to either of the simple tones, as opposed to the double ones. We shall see below that it is useful to do so, in order to specify the tonal conditions under which the voice quality modifications Glottalized and Laryngealized are distinctively opposed. The three tonal oppositions generate four tonal classes, each of which is characterized by one of the four basic pitch accents ( 2 ) . The tonal diacritics in (2) represent the surface tones in prepause position.1*

(2)

ACCENT bü-cä 'that (person)

baza 'arrow'

gu-xharu 'locust 1

'different'

blzu 'bee'

bl-xhözä 'father'

'iguana'

nä-gül •ripe1"

'Oaxacan'

nä-xädä 'short'

di_ 1 curse'

nä-da 'fragile'

blnnl 'people1

bi-rungu 'tree stump'

nqa 'that (thing)' 1

HL LH

powder'

guiu [gyu:]

The disparity between the pitch accent class and the surface tones of words in the H and LH accentual classes of (2) is due to realization rules which modify the basic accents in particular contexts (see 3. below). The underlined syllables in (2) are stressed, as they would be if each word were spoken in isolation. However, in ordinary speech there is usually just one stress on each major syntactic element ( e . g . , NP, V ) , so that many potentially stressable morphemes often occur unstressed (see Mock forthcoming for details). The major phonetic characteristic of stress is increased segmental duration; neither vocalic nor consonantal length is distinctive independently of stress. The vowel of the stressed syllable is lengthened to two moras as shown in ( 3 ) , unless the vowel is distinctively glottalized, ( 4 ) , or the next syllable of the morpheme begins with a fortis consonant (/'p t k c s s m n: 1: y w/), in which case the stress geminates the fortis consonant, ( 5 ) , in a process of resyllabification to be described below. Fortis consonants are strongly articulated and if obstruents, invariably voiceless; their lenis counterparts are variably voiced and articulated less forcefully. This same distribution of lengthening as a manifestation of stress is characteristic of certain other Otomanguean languages as well, e.g., Cajonos Zapotec (Nellis and

Pitch Accent in Isthmus Zapotec

203

Hollenbach 1980:95,98) and Chocho (Mock 1977). (3)

Mzu

[birzu]

'bee'

(4)

rä'de' ra ' ta '

[ra^de^] [ra-ta v ]

'receives a gift' 'lies down 1

(5)

rati

[rat:i]

'dies'

Stress placement within the word is superficially distinctive, as shown in ( 6 ) , but if we admit morpheme boundaries into the phonological description, stress always falls on the first syllable of the root, for any word that bears stress. The verb in (6b) has an aspectual prefix rl-, whereas the noun in (6a) has no syllabic prefix; therefore the stress falls on the first syllable of the noun but the second syllable of the nearly homophonous verb. (6)

a. rigä b. riga

[ri:ga] [riga:]

'husk' 'gets cut 1

(rl-gä)

Because the location of stress within a stressed word is predictable, earlier articles on Isthmus Zapotec prosodies (Mock 1983,1985) argued that stress was an entirely culminative feature that merely added prominence to the pitchaccented syllable of a word (for an earlier and different analysis of Isthmus Zapotec stress, see Pickett 1951, but also see Pickett 1967). A recent investigation of the distribution of stressed and unstressed words in oral texts has revealed, however, that stress serves to highlight the major constituents of sentences (Mock forthcoming). It therefore must be accorded a place within the phonology, despite the fact that it does not differentiate one lexical item from another. That is,

whether a word is to be stressed or not is a phonologic-

ally significant choice, as shown in ( 7 ) , but it is a choice which is made for syntactic (and ultimately discourse-related) reasons rather than as part of the basic phonological information that must be represented in the lexicon. Stress is a feature which is added to a morpheme only in its situational context, at the point where syntactic and textual information is relevant. (7)

a. lu diuxl lu diuxi

'before God' "the face of God'

b. dxl que

'in those days," (as a new temporal frame, e.g. at the beginning of a sentence) 'back then' (anaphoric reference, e.g. at the end of a sentence)

dxl que

204

Carol Mock

Because the decision to stress a morpheme is governed by syntactic and pragmatic factors, the stress feature would not be part of a strictly lexical phonology. But it strongly affects the realizations of both tone and voice quality features, both of which are lexically distinctive. Its importance to tone and voice quality features suggests that stress belongs among the prosodic systems of Istmus Zapotec ( 8 ) : •Stressed

(8)

•Unstressed

-High Root Morpheme

•Single Tone •Low High-Low •Double Tone

.Low-High To understand the interrelations among the various prosodic phenomena in Isthmus Zapotec, there is one more topic we need to address: the behavior of the two modifications of voice quality that are distinctive for the vowels of stressed syllables— glottalization and laryngealization. These are prosodic features just as certainly as accent or stress, having the morpheme as their domain and the stressed syllable of the morpheme as the focus for their realization.

Glottalization is manifested as a postvocalic glottal stop in the stressed

syllable of a glottalized morpheme (except in disyllabic morphemes bearing a H pitch accent), and simultaneously as a word-final glottal stop in prepause position, as shown in ( 9 ) : (HL)

'alligator 1

(L)

'mud'

rl-düVbä '

(L)

'gets carried 1

rl-düil

(L)

'gets used up'

damä'

(H)

Owl 1

(9)

Laryngealization, which is creaky vowel quality and a double pulse to the syllable (as if a very weak glottal constriction interrupted the vocalic nucleus) primarily affects the vowel of the stressed syllable, although in prepause position it also adds a final glottal stop to disyllabic stems that bear a L pitch

Pitch Accent in Isthmus Zapotec accent,

(10).

(10)

döo du riäanä riänä ri-lüubä'

(L) (L) (LH) (L) (L)

205

'rope' 'maize tassel' 'stays' 'burns' 'gets swept'

The manifestations of these voice quality modifications (VQM) are tied to stress. When a word having one of the VOM occurs without stress, it glottalization or laryngealization, ( 1 1 ) ; that is,

loses

its

the underlying VOM are not

manifested.

(11)

bedä ne

(LH, LH)

guldxü xüuni'

(L,L)

'came with, brought' (bäedä 'came') 'herpes zoster" (gui'dxu1 'sore')

The phonological opposition here would seem to be a three-member voice quality system ( 1 2 ) .

(12) Root Morpheme

-Glottalized -Laryngealized

But once we take into account the cooccurrence of pitch accents and the two modified voice qualities, it becomes clear that the situation is not that simple. Glottalized morphemes take only three of the pitch accents: L, H and HL, as shown in ( 1 3 a ) ; and laryngealized morphemes also only take three, but a slightly different set: L, H and LH (13b). (13)

a.

L H HL

MONOSYLLABIC

DISYLLABIC

ri-nda ' rl-ndä' ri-ndä'

nä-yä 'gui ' nä-yänä ' nä-yä ' nl '

' stinks' 'gets bitter' 'gets hot'

DISYLLABIC

MONOSYLLABIC

b. L H LH

chii nun büu

' burnt ' 'picante ' 'clear 1

1

'ten 'there is' 'charcoal '

nä-dxlibl ' nä-dxlibl rl-dxlichi

1

smooth ' fearful ' 'be angry' 1

206

Carol Mock

This skewing reveals constraints against cooccurrence of glottalization and the LH accent, and against laryngealization with the HL accent, with the result that there are only two accentual contexts in which glottalization and laryngealization are directly in opposition with each other: L and H. The system network can capture this fact by recognizing two systems for voice quality, of which only the first is at the same degree of generality as the first accent system:

(14)

/-»

-Plain VO rGlottalized

Root Morpheme

KLaryngealized ingle Tone •Double Tone

The first voice quality system here merely distinguishes between plain voice quality and either of the marked voice qualities, which we have seen are in complementary distribution when the pitch accent is a contour tone, HL or LH. This complementarity is captured by a double entry condition to the second VQ system, ( 1 5 ) ; thus the two marked voice qualities are not truly opposed to each other except when the pitch accent is either H or L, because Modified plus HL is predictably glottalization, whereas Modified plus LH is always laryngealization. Stressed

(15)

Unstressed pPlain VO -Glottalized Root Morpheme

-Modified VO-

t—*

-Laryngealized r-H

-Single T

"-Double T

Pitch Accent in Isthmus Zapotec

207

Thus the single-tone accents are the only feature specifications for which there is a true opposition between glottalizing a morpheme and laryngealizing it.

The dual entry condition for the second VOM system reflects this fact:

Modified VQ and Single Tone are both prerequisite features for entering the system Glottalized / Laryngealized. Because a systemic description forces the investigator to consider the full range of combinations among features in order to identify the paradigmatic as well as syntagmatic contexts in which real phonological choices are possible, some of the feature systems are abstract, purely formal rather than phonetic. So far in this description we have seen two such systems, the first accent system and the first VQM system. The validity of these particular abstract oppositions is supported by morphological alternations between laryngealized and glottalized vowels. When certain tonal morphemes are superimposed on the verb, they replace its

lexical accent

with a grammatical HL, forcing a simultaneous change from laryngealized to glottalized voice quality. One such case is inflection for first person plural, which adds one of the relevant dependent pronouns and changes the pitch accent of the verb to from L to HL. If the verb happens to be laryngealized in

its

basic form, this change to a double tone triggers a switch to glottalized VQ as well ( 1 6 ) .

(16)

a.

-huäa (L) 'standing, stopped' pä m& -huä' ( H L ) dxiichi nu la if already standing firm 1-2 PARTICLE 'if we (inclusive) are already standing firm'

b. gul-düuyä' ti gul-du'yä

(L) (HL) du

'may see' 'until we (exclusive) see'

In terms of the system network, the tonal change blocks entry into the entire second system of voice quality modification, because its tonal entry condition is no longer satisfied. The initial VQM system then becomes the terminal one, with its context-sensitive realization of Modified as glottalization in the context of the HL feature. 5 This skewing of the prosodic combinations reflects Isthmus Zapotec's version of the nearly universal attraction between glottal stops and high tone on the one hand, and between breathiness or laryngealization and low tone on the other (cf.

Hombert 1978 for a discussion of the physiological basis for this). Here,

VQM features are clearly subordinate to tonal features: when an incompatible tone is superimposed on a morpheme having modified voice quality, it is the voice quality that changes rather than the tone. Why this should be the case

208

Carol Mock

in Isthmus Zapotec may be related to either tonogenesis or to 'laryngeogenesis ' , both of which are ancient features in the Otomanguean stock, but a satisfactory answer must await further study.

3.

Realization Rules

The systems detailed above reflect the paradigmatic oppositions of tone, stress and voice quality modification in Isthmus Zapotec, but they say nothing about the syntagmatic or structural axis of language: how each feature is manifested in various contexts. This aspect of the description is handled by the realization rules, which construe formal structures from the phonological features and modify them according to specific syntagmatic contexts. The first rule for any feature converts the feature representation into a structural one and uses the realizational arrow, \.

Subsequent realization rules modify this basic

structural representation and use the rewrite arrow, ->·. Some of the realization rules put elements on the tonal tier, where they are automatically associated with elements on a separate segmental tier according to the Well-Formedness Conditions spelled out below. Other rules add prosodic elements directly to the segmental tier, e . g . , the manifestations of stress and VQM in Isthmus Zapotec. Association rules as such are seldom necessary, but when they are used they are considered a type of realization rule. The rules in this section are presented in their proper order, unless stated otherwise. The first stress rule (Vowel Lengthening) and the four Tone Structure rules are not ordered among themselves, but apply in a block before Tone Sandhi and later tone rules which convert the basic pitch-accents into surface tones. The rules for the realization of voice quality modifications (VQMs) are not ordered relative to Tone Sandhi, but do depend on prior selection of the feature Stressed (and its primary realization rule, Vowel Lengthening). The last realization rule of all is Resyllabif ication. 3.1.

Stress

We have seen that the presence of stress generally adds a mora to the stressed syllable. This is a segmental change which can be stated as Vowel Lengthening:

(17)

Vowel Lengthening Stressed \

σ / C V

(σ) "

(C V)

Pitch Accent in Isthmus Zapotec

209

This formulation is preferred to one which abstracts just the rhythmic aspect of the syllables, as in ( 1 8 ) , because two later rules also make reference to the consonantal and vocalic segments: Glottal Insertion and Resyllabification.

(18)

Stressed \

σ

m /

Λ mj

(σ)

Ι

(m)

The examples in ( 2 ) , (3) and (6) above are accounted for by the Vowel Lengthening rule, while those of (4) and (5) result from the ordered application of Vowel Lengthening and one of two later rules, Glottal Insertion or Resyllabification (see below). The feature system Stressed/Unstressed is privative, and therefore there is no realization rule attached to its negative pole. We shall see, however, that Unstressed is formally necessary because it blocks realization of VQM features. 3.2.

Glottalization and Laryngealization

Let us next consider the realizations of the voice quality modification features. It is important to remember that VQMs are not manifested at all without the concurrent presence of stress; see the examples in (11) above. But there is a problem with their structural representation even when they are manifested: they cannot be put on the same tier with tone without making association lines cross, despite Yip's general suggestion to lump such phenomena together (Yip 1982), because both VQM and tone may affect more than a single syllable of disyllabic morphemes:

(19)

'alligator 1

be' 'ne' CVV CV

IIXj

""HL

'

na-dxii

bl'

CV

CV

I

L

CW

'smooth'

MX]

* L

(where " represents both laryngealization in a stressed morpheme and concomitant glottalization of a final L-toned unstressed syllable.) There are several good reasons for arguing that modifications of voice quality change structure on the segmental tier, despite their independent behavior. One is that glottalized morphemes are short, even when stressed, unless

210

Carol Mock

they also bear underlying H tone, which becomes a superficial LH. Another is that the resyllabification caused by gemination of fortis consonants after a stressed syllable does not take place in either glotallized or laryngealized morphemes. A third reason is that the HL pitch accent is not a phonetic fall in a stressed glottalized syllable, but a level high tone. All this suggests that the basic realization of stress as CW undergoes modification when VQM is present. Let us look first at glottalization. Recall that before pause, glottalization is manifested as post vocalic stops in both syllables of a disyllabic morpheme, but that disyllabic H morphemes bear only a final glottal stop. Although only one subtype of glottalized morpheme displays a single glottal stop, another fact lends support to the idea that glottalization

is realized primarily in

word-final position: when a glottalized suffix is added to a root morpheme that has plain vowels, the entire stem becomes glottalized:

(20)

'your home1

lldxl home

+

-u' 2

->-

li'dxu'

lldxl home

+ H-a' 1

->

lldxe' 'my home 1 (disyllabic H morpheme)

Such data suggest that a glottal stop is first inserted into the segmental tier at the end of a morpheme:

(21)

Glottal Insertion Glottalized \

~>

/ C V V ( C V )

#

In disyllabic morphemes, it then is epenthesized to a second position at the end of the stressed syllable:

(22)

Glottal Epenthesis 0

-v

·>

/

CW

CV?

However, word-final glottal stops are actually manifested only before pause. Monosyllabic glottalized morphemes, however, do not invariably lose their final glottal stop unless the next morpheme begins with a sonorant consonant:

Pitch Accent in Isthmus Zapotec (23)

a. ba'te'

'alligator'

rä_'tä ber·

0 / C V

? (C V) '

(except LH)

The exclusion of rising pitch contours from this shortening rule may have an auditory justification. Phoneticians have noted that rising contours are more salient than falling ones (Candour 1978), and this auditory fact may have protected the stress-induced length of LH glottalized syllables, given the one-to-one correlation between individual tones and moras in Isthmus Zapotec. (The basic LH pitch accent is not affected by this rule, because LH and Glottalized cannot be selected as phonological features of the same morpheme; see 2. above). When the second mora is deleted by rule ( 3 9 ) , the tone associated with it has nowhere to reassociate and simply 'disappears' unless it is also associated with a following syllable ( i . e . , in a disyllabic morpheme), since at this point in the realization seguence each mora already has a tone associated with

218

Carol Mock

it:

(40)

rl-ndä_' CV

'gets hot'

CV

I

I

HL

L

cf. rl-ndä'

'stinks'

CV CW

I NH

L

( ->- LH by L Insertion)

nä-y£Vni'

'clear'

CV CVCV H

The class of glottalized, disyllabic H morphemes requires two further realization rules: Glottal Deletion and H Shift, which must be ordered to follow L Insertion. The first of these deletes the glottal stop from the stressed syllable of such morphemes:

(41)

Glottal Deletion ·>->

0 / C V V

The second rule shifts the H of such morphemes one mora to the right with the result that the stressed syllable bears only L tone: (42)

H Shift C V V C V ·>

I A/

L

H

Once these two rules have taken effect, disyllabic glottalized H morphemes have almost reached their surface form, being further modified only by Resyllabification and the low-level process of Glottal Elision ( 3 . 2 . above). (43)

biQ_xl' CW :w cv CV

V I

L

H

'mosquito'

nä-nrö

'

cv CV C cw

I V

L

L

H

'timid'

Pitch Accent in Isthmus Zapotec 3.6.

219

Final H Modifications

Two low-level realization rules affect H when it

is associated with any un-

stressed syllable in prepause position. Final H Lowering and Final V lengthening. These rules are restricted to syllables that are not glottalized

(i.e.,

glottalization prevents its application).

(44)

Final H Lowering CV

-»·

C V

H

/

L H

The output of the Final H Lowering rule feeds a Final Lengthening rule which, incidentally, also allows the lexically floating H of monosyllabic unstressed LH morphemes and the floating H created by H Launching (rule 37) to be attached:

(45)

Final Lengthening C V

->·

C V V

L H

/

L H

The floating H tone 'docks' automatically at this point, now that it has a mora with which to associate. 8

(46)

'whistle'

bl-gadxe

'Pleiades'

'monkey'

ri-beza

'dwells'

This rule sequence neutralizes the distinction between H and LH pitch accents in monosyllabic morphemes, but only before pause:

(47)

3.7.

PREPAUSE

H rä ca. LH rä que

rä cä . . . rä que . '. .

' over there ' ' over yonder '

H dxl ca. LH dxl que

dxl cä . . . dxl que . '. .

'that day' 'the other day'

Resyllabification

After all these processes of feature manifestation are complete, Resyllabification further modifies the segmental structure of disyllabic morphemes, if

it

has not already been changed by insertion of glotallization or laryngealization,

220

Carol Mock

and if the consonant which begins the second syllable is

(48)

fortis:

Resyllabification C V V

C

VC 5 )

->

C V C

C

VC3)

The examples in ( 4 9 ) below show the effects of Resyllabification. It might be asked what happens to a contour tone when it is associated with a stressed syllable in such a context: does the association line get broken, with the result that the stressed syllable no longer bears two tones? I have no definitive answer to this question, but my impression is that HL is heard as a solitary H whereas superficial LH (which derives from a deeper H) is heard as a shortened or 'squeezed' rising tone. If it turns out that the rising tone preserves the double mora structure of the stressed syllable as it does the length of a glottalized syllable, rule ( 4 8 ) will have to be divided into two separate rules.

L H

Msi döpa nSxh£ ' HL xtuxhü LH näpa

(49)

4.

' ghost ' 'rolled up' ' in place' 1 sunray ' 'has'

nä-xhocä bä-cuel :ä gul-däpä ' gul-däpä nä-titl

'pock-marked' "corn husk 1 'the four of them' 'fourth (täpä ' 4 ' ) 'shaky, weak'

Conclusions

The typological question with which we began this article, and which also is raised in several other articles in this volume, demands the most complete data possible. To decide what the limiting criteria for pitch accent systems are and to delineate their formal and phonetic parameters requires detailed information about the behaviour of pitch in many different languages, and about how pitch interacts with other prosodic phenomena such as stress and voice quality modification. The fairly complex set of prosodic features in Isthmus Zapotec illustrate the importance of looking beyond pitch to elucidate its behaviour; the intricate interplay in Zapotec among pitch accent (on morphemes), surface tone (on moras), stress (on syllables) and voice quality modification (of syllable structure) highlights how interdependent are the issues which an adequate characterization of pitch accent must take into account.

Pitch Accent in Isthmus Zapotec

221

Notes

*I am grateful for comments on an earlier draft of this article by Jan Daeleman, John Goldsmith, Barbara Hollenbach, David Odden and Velma Pickett. 1.

The low tone accentuation of CiTonga would seem to contradict this last statement, but even there there is an obligatory (string of) high tone(s) preceding the starred low of accented morphemes, as opposed to the uniform low tones of unaccented ones.

2.

The presentation here d i f f e r s from Mock (1985) and earlier analyses by its deeper understanding of the role of stress and by a formalization of the difference between syllable and mora, such that identical double tones LL and HH are no longer necessary to the description.

3.

A very few morphemes are purely tonal: 1. the lexical pitch accent of verbs in completive or potential aspect changes from LH and L to H and HL respectively in most subordinate clauses (examples given in Mock (1983); 2. a few nominal compounds bear a HL on their second element which does not correspond to the lexical pitch accent of the word and is not the result of Tone Sandhi, e . g . , yöo (L) 'house' in igue yoo 'roof (literally 'head-of h o u s e ' ) , where igue has L as its lexical pitch accent; 3. certain nouns derived from verbal adjectives bear pitch accents distinct from their adjectival forms, e.g. , yo'xhö' (HL) Old person 1 from nä-yöoxhö' (L) Old 1 .

4.

The Isthmus Zapotec data in this article are written in a modified standard orthography, which I have enriched by adding surface tones and italics for stressed syllables. In general the symbols are equivalent to the phonemic symbols of Pickett (1960), but a few may need explanation: glottalized vowels (vowels plus glottal stop) are a ' , e', i ' , ', u'; laryngealized vowels (creaky and rearticulated) are aa., ee, ii, oo, uu; digraphs ch and dx are /£/ and /5/ respectively; xh is /s/, x is /z/, nn is fortis / ·./ and 1: is added to the orthography here for fortis /!:/. As in Spanish, c and gu are simply /k/, g and gu (before i, e) are /g/, hu is /w/ and gu is /gw/.

5.

For a fuller discussion of this issue in a cross-linguistic perspective, see Mock (1985:358-359).

6.

nüu H is one of the few exceptions to this statement and may well be derived from an underlying LH morpheme perturbed to H by its n- prefix.

7.

It is necessary to divide the tonal spreading process into two rules, Tone Sandhi plus H Launching, because of the way in which LH perturbs to H (instead of HL) and is subsequently lowered by the L Insertion rule. If it were not for this fact, Tone Sandhi could be seen as a simple spread of H away from LH to the right.

8.

A long final syllable with rising tone is not phonetically identical to a stressed LH syllable despite its two moras, because the rise is lowerpitched; e.g. nga [35] nga 113] 'That's it' (where the numbers in the brackets indicate pitch levels according to the system used first by Chao for Chinese, with [1] lowest and [5] highest). A low-level realization rule would specify that VV deriving from Stressed bears higher-pitched tones.

222

Carol Mock

References

Solinger, D. (1958), A Theory of Pitch Accent in English. Word 14, 109-149. Solinger, D. ( 1 9 6 1 ) , Contrastive Accent and Contrastive Stress. Language 37, 83-96. Clements, G . N . and K. Ford ( 1 9 7 9 ) , Kikuyu Tone Shift and its Synchronic Consequences. Linguistic Inquiry 10, 179-210. Clements, G . N . and J. Goldsmith ( e d s . ) . ( 1 9 8 4 ) , Autosegmental Studies in Bantu Tonology. Dordrecht: Foris Publications. Daeleman, J. (1983), Tone-Groups and Tone-Cases in a Bantu Tone Language [KiNtandu dialect of KiKoongo] . LTL, Review of Applied Linguistics 1983, 60-61, 131-141. Candour, J.T. ( 1 9 7 8 ) , The Perception of Tone. In: Fromkin, V.A. ( e d . ) , Tone, A Linguistic Survey. New York: Academic Press. Goldsmith, J. ( 1 9 7 6 ) , Autosegmental Phonology. Doctoral dissertation reproduced by the Indiana University Linguistics Club. Goldsmith, J. (1982), Accent Systems. In: H. van der Hülst and N. Smith ( e d s . ) , The Structure of Phonological Representations (Part I ) , 47-63. Dordrecht: Foris Publications. Hollenbach, B. ( 1 9 8 4 ) , The Phonology and Morphology of Tone and Laryngeals in Copala Trigue. Doctoral dissertation, University of Arizona. Hombert, J.-M. (1978), Consonant Types, Vowel Quality and Tone. In: Fromkin, V.A. ( e d . ) , Tone, A Linguistic Survey. New York: Academic Press. Hyman, L. ( 1 9 7 7 a ) , On the Nature of Linguistic Stress. In: L. Hyman ( e d . ) , Studies in Stress and Accent (SCOPIL 4 ) , 37-82. Los Angeles: University of Southern California. Hyman, L. (1977b), Tone and/or Accent. In: D.J. Napoli ( e d . ) . Elements of Tone, Stress and Accent. Washington, D . C . : Georgetown University Press. Martin, J . R . ( 1 9 7 6 ) , The Meaning of Features in Systemic Linguistics. Mimeo, Department of Linguistics, University of Sydney, Australia. Mock, C . C . ( 1 9 6 9 ) , The Grammatical Units of the Nzema Language: A Systemic Analysis. Doctoral dissertation, University of London (UCL). Mock, C.C. ( 1 9 7 7 ) , Chocho de Santa Catarina Ocotlan. Archive de Lenguas Indigenas de Mexico, vol. 4. Mexico City: Centro de Investigacion para la Integracion Social. Mock, C . C . (1983), Tone Sandhi in Isthmus Zapotec: an Autosegmental Account. Linguistic Analysis 12, 91-139. Mock, C.C. (1985), A Systemic Phonology of Isthmus Zapotec Prosodies. In: W.S. Greaves and J.D. Benson ( e d s . ) , Systemic Perspectives on Discourse: Selected Theoretical Papers from the Ninth International Systemic Workshop, 349-372. Norwood, N . J . : Ablex. Mock, C . C . (forthcoming), Relations between Pitch Accent and Stress. Proceedings of the Chicago Linguistic Society 21 (.1985). Nellis, D . E . and B. Hollenbach ( 1 9 8 0 ) , Fortis versus lenis in Cajonos Zapotec phonology. IJAL 46.92-105. Odden, David (no d a t e ) , On the Role of the Obligatory Contour Principle in Phonological Theory. Unpublished ms. Pickett, V.B. (-1951), Nonphonemic stress: a Problem in Stress Placement in Isthmus Zapotec. Word 7.60-65. Pickett, V.B. ( 1 9 6 0 ) , The Grammatical Hierarchy of Isthmus Zapotec. Language 36.1, Part 2. Pickett, V . B . ( 1 9 6 7 ) , Isthmus Zapotec. In: R. Wauchope and N. McQuown ( e d s . ) , Handbook of Middle American Indians, Vol. 5, 291-310. Austin: University of Texas Press.

Pitch Accent in Isthmus Zapotec

Pierrehumbert, J. (1980) , The Phonology and Phonetics of English Intonation. Doctoral dissertation, MIT. Pike, E . V . ( 1 9 5 1 ) , Tonemic-Intonemic Correlation in Mazahua (Otomi). International Journal of American Linguistics 17, 37-41. Prakasam, V. ( 1 9 7 2 ) , A Systemic Treatment of Certain Aspects of Telegu Phonology. Doctoral dissertation, University of York, England. Prakasam, V. (1976), A Functional View of Phonological Features. Acta Linguistica Academiae Scientiarum Hungaricae 26 ( 1 - 2 ) , 77-88. Prakasam, V. ( 1 9 7 7 ) , An Outline of the Theory of Systemic Phonology. International Journal of Dravidian Linguistics 6, 24-42. Selkirk, E . O . ( 1 9 8 4 ) , Phonology and Syntax: The Relation between Sound and Structure. Cambridge, Mass.: MIT Press. Yip, M. (1982), Against a Segmental Analysis of Zahao and Thai: a Laryngeal Tier Proposal. Linguistic Analysis 9.79-94. Zubizarreta, M . L . (1982), The Formal Interaction of Harmony and Accent: The Tone Pattern of Japanese. In: H. van der Hülst and N. Smith ( e d s . ) , The Structure of Phonological Representations (Part I I ) , 159-212. Dordrecht: Foris Publications.

223

Predictable Tone Systems in Bantu David Odden Ohio State University

1. Introduction* The question whether a description of the prosodic system of some undescribed language should be framed in terms of "stress", "accent" or "tone" (and the justification for a particular choice in a language with a descriptive tradition)

has plagued the workaday linguist for years: unlike notions like "bilabial"

there are few clear principles guiding the trichotomization of languages into these three types. There are certainly no acoustic or physiological properties of these linguistic constructs which justify the separation of languages into two or three types. More productive have been attempts to reduce the difference between tone, stress and accent to differences in the distributional and behavioral properties of these entities. Taking as representatives of the poles languages like Quechua with penultimate stress and Cantonese with unpredictable contrasts in the level and direction of F modulations, we can abstract away a cluster of typical stress versus tonal parameters. In stress languages the position of the prosodic peak is predictable by rule (assigning stress to the penultimate, or antepenultimate, or initial u n i t ) , there is precisely one syllable in the word with that contrast, and that peak is best treated privatively. In a "highly tonal" language, the position of the prosodic contrast is not predictable, there may be more than one prosodic peak (if any), the contrast may be unpredictably realised at any one of a number of pitch levels, or as transitions from any pitch to any other pitch, and relations between contrastive pitches are best treated equipollently. If this split between tone and stress is taken to describe the "ideal" distinction, there are probably no ideal languages. Under any analysis, the location of main stress in English is partly unpredictable using strictly phonological criteria; more than one stress per word is allowed; there are secondary and tertiary stresses. Similarly, there are tone languages which allow only one prosodic peak per word, or which exploit only a fraction of the possible contrasts in the word, or where certain phonetic contrasts are rule governed. A strict division into tone languages versus stress languages is not possible, since some rather clear tonal systems (such as the Bantu languages Shona,

226

David Odden

Sesotho, Zulu and Shambaa) also have regular nonphonemic stress. The Bantu language family provides a rather rich testing ground for theories of tone and accent, since much variation in typological characteristics of tone versus accent can be found in this family. As typological extremes the Vunjo dialect of Chaga has four phonetic levels of pitch and Swahili has penultimate stress. My focus here will be on a typologically intermediate class of languages within Bantu which exhibit one of the typical properties of stress systems, namely the languages with "predictable" tone systems. The fundamental property of these systems is that, at least in the verbal system, the location of H is predictable by reference to the phonological shape of the verb and the morphological (tense-aspect)

category of the verb. In these languages there are no sub-

classifications of verbs into H toned stems versus L toned stems, which is historically anomalous since most Bantu languages divide stems lexically into H versus L stems. A historical understanding of the development of such systems in Bantu is largely hampered by a lack of data, and it is consequently impossible to trace the historical development of these systems with much accuracy. Nonetheless, some progress can be made, and I hope to contribute here both to our understanding of the conversion from tone to accent systems, and of the mechanics of predictable tone systems. I will be dealing here with two rather different subgroups of Bantu predictable tone systems. On the one hand, I will review the structural properties of the tone systems of Kinga, Safwa and HibenaKihehe, languages spoken in the southwest of Tanzania near Lake Nyasa. In these languages, generally one and only one H tone can be found per word, and the rules for tone assignment put tone in places such as on the first stem syllable, or the antepenultimate stem syllable - positions familiar from the study of stress. On the other hand, we encounter the principles of tone assignment in the languages Kimatuumbi, Makua and Kikuria, which allow one or two underlying H tones per word (and multiple surface H tones), and which assign tone on the basis of a "count" of vowels in the stem, such as to the fourth vowel of the stem - such positions are not the central locations for assignment of stress. I argue that the tonal systems of the Nyasa-area languages show stress-like characteristics, and that some of the peculiarities of these languages are the result of the interplay of metrical and tonal structure: Goldsmith (forthcoming) shows that incursion of metrical structure in the Bantu Middle Lake languages explains some otherwise peculiar tonal rules. For the other languages, there is little evidence of metrical structure influencing the tonal grammar, and the system of tonal conjugations appears to have arisen out of earlier

Tone Systems in Bantu

227

tonal properties of verbal affixes. Yet the basic tonal core of the southerly languages also reflects these same tonal properties of affixes, underscoring the observation that both types of languages have taken the same tone system from proto-Bantu and have subjected it to reinterpretation in rather different ways.

2.

The Lake Nyasa Area Languages

2.1.

Safwa

The tonal system of Safwa has been laid out in Voorhoeve ( 1 9 7 3 ) , where it

is

observed that precisely one syllable per word bears a H. This one-per-word constraint is one of the distinguishing characteristics of well-known pitchaccent systems such as Luganda or Japanese. H tone in Safwa is restricted to a few locations in the word: it may occur on any of the final three vowels of the word, or on a prefix vowel two syllables to the left of the stem. A few words have a penultimate H tone ( P U ) , and these may be suspected of being loan words from a language with penultimate stress (e.g. ishitala

r

b e d " ) . Others

may have antepenultimate H tone, but also undergo a final vowel shortening rule (e.g. piliiplli, from pilii-pilii 'pepper'). We may set aside penultimate tone as being extraneous. The tone pattern "ante-pre-stem-initial" seen in forms like

mi-ίηο

'teeth'

is a predictable variant of word-final H tone, due to a tone throwback rule that shifts final H to the second vowel before the stem: when there is no vowel in that position, final H tone remains in place. (1)

mi-ίηο garni-ino mi-ino in-koombe iim-bisi in-koombe m-bisi

'teeth' 'the very teeth" 'it is teeth' 'uncooked beans' 'the beans are uncooked'

Voorhoeve's analysis of this alternation is stated as a pair of rules, one placing a H on the pre-prefix if there is a final H, and a second deleting the final H if there is a preceding H . 3 (2)

[+syl] => [+tone] / [+tone] => [-tone] /

[+syl] + X [+tone] # [+tone]

[-tone]

#

228

David Odden

While these rules are a bit strange synchronically (even stated as a single autosegmental r u l e ) , there is a diachronic explanation for this alternation. First, it is fairly clear that the pre-prefix syllable (a of

mi-ino) was H

toned in Bantu. Secondly, a number of Bantu languages have developed absolute restrictions against word-final H tone, or else in some manner weaken final H ' s Haya, for example (Byarushengo, Hyman and Tenenbaum ( 1 9 7 6 ) ) disallows wordfinal H ' s . Thirdly, we need some explanation for the more general "one H per word" principle of Safwa; the metrical explanation I propose for this restriction is that a right-dominant tree is constructed on H tones, and all H ' s except the strongest H of the word are suppressed. Tone shift can then be explained in the following way. I first assume that the restriction against final H tone derives from making word (or utterance) final position prosodically weak or extrametrical. Earlier forms such as amiin(o) would thus appear as amiin(o) due to the weakening of the final H (and later loss, due to the H of the preprefix) , while the extrametrical final H is retained in mi-ίηό due to the lack of any preceding stronger H (thus, a weak H is strong enough to survive when there is no stronger H ) . This then leaves us with two prosodic contrasts in noun stems, which appear to be equally distributed in nouns,- nouns may have final H (WF) , or they may have antepenultimate H (APU)

(which may occur on the prefix vowel), examples

of the latter being seen in ( 3 ) .

(3)

ama-paanga am -fuko umw-ehale shllihu

'swords' 'bags' 'another one' 'which one'

Η-final nouns could be accounted for by linking a H with the final vowel in the lexicon (subject to later throwback). But the representation of the antepenultimate H nouns is not so simple - the H may show up within the stem, or it may show up on the preceding prefix (ama-fuko

versus ama-p anga), and there-

fore a lexical association analysis fails, since a tonal property of the stem can be realised outside of the stem bearing it.

Along with Voorhoeve, I assume

that the antepenultimate H pattern reflects the application of a default rule assigning H to the antepenultimate vowel of any word bearing no H (all words in Safwa have a H ) .

(4) I

V V V #

Tone Systems in Bantu

229

APU reflects the treatment of underlying toneless nouns, WF reflects the regular treatment of nouns with a H tone (which may be anchored to the final syllable by an early rule and need not be lexically linked), and the nouns in the rare pattern PU may be provided with a lexical link between the H and the penultimate vowel.1* This split into H stems and toneless stems agrees with Voorhoeve's observation that the tone pattern APU correlates with proto-Bantu final L tone, and final H correlates with historical final H: given that L tone in most Bantu languages is best represented as no tone (L being filled in by a default rule), the stem tone of a Safwa noun is merely its

final proto-Bantu tone (or lack

of tone). Verbal tone patterns are also highly restricted - unlike nouns, verb stems are never subcategorized into classes. Instead, the tone pattern of a verb is a function of the tense-aspect morphology of the verb. Stems in Safwa are then basically toneless, and the surface tone is assigned either by some morphologically-conditioned tone rule, or the default rule ( 4 ) . Since assignment of H tone to the penultimate syllable is done by a special rule, added at some cost, the surface tone pattern which we should find for the majority of the verb tenses should be APU, since that pattern represents the default tone assignment. Indeed, most verbal tenses select an antepenultimate H.

(5)

uha-jeendile inhayl-bala baä-bala tuu-bale

'you walked' (remote perfective) will not go' (future) 'and they went' (recent consecutive) 'let us go' (subjunctive)

The penultimate H tone pattern is found only in two verb tenses - the negative recent perfect and the negative recent consecutive.

(6)

tusiga-bajila tusaa-yula

'we could not' 'and we did not yawn'

The verbal equivalent of the nominal final H pattern appears to be found in a limited number of tenses (nearly all descendents of the Bantu subjunctive plus object prefix construction)

which put a H on the pre-stem vowel. 5 Related

to this fact is the fact that imperatives with object prefixes and plural imperatives put a H on the final stem vowel.

230

David Odden

(7)

ä-hwoogope tu-zifise n-daanje balajl

'to fear" 'let us hide them' 'show me' 'go (pi.) continuously'

These two surface patterns can be treated as having a morphologically induced H tone which docks to the final vowel, and which is then thrown back to the prestem position by the tone throwback rule. Note that when there are no prefixes before the stem in the imperative, tone throwback is inapplicable. Voorhoeve mentions a curious rightward shift of tones in verbs which have the continuative marker -agr- or -ang- with a long vowel. As the data below show, whatever the underlying place for tone in the basic verbal form may be, the tone shifts one vowel right in the presence of -aag- or -aang-. (8)

asa-baläga asa-bonesyaagä u-leetage tu-buzyaaje inga-bälaga ziga-baanga

'and he was not going' ( P U ) 'and he was not showing' 'you may bring' (APU) 'let us be asking' 'I may go" (APU) 'they may be'

It would be mistaken (in fact, impossible) to revise the rules of tone assignment in Safwa so that they do not skip over the affixes -aag- and -aang- when they are long. What is called for is a rule of tone shift, which refers to "long -aag-" once and only once. But how is such a rule to be formulated? Clearly the rule must refer to the morpheme -a(n)g- and to the CV tier, in such a fashion that it identifies the allomorph of -a(n)g- as being long (a requirement for the shift rule). The rule must specify the H tone which is shifted, and the rule must indicate that the H is freed from that vowel and is associated with the following vowel. We might begin the construction of such a rule with at least the elements specified in ( 9 ) : note that this rule is informed since it does not tell us what the H is linked to or what to do with it.

(9)

H V V -a (n) g-

The H might be underlyingly associated with the second vowel of -aa(n)g-, as in asabonesyaäga which becomes asaionesyaaga, or it might be associated with

Tone Systems in Bantu the first vowel of the a f f i x , as in tubuzyäaje which becomes The paradox is,

231

tubuzyäaje.

then, that we need to be vague about the association of the

H tone with respect to the conditioning vowels of the a f f i x -aa (n)g-, but we cannot be vague about the H ' s association, since the rule's function is to shift H one vowel to the right. While a strictly tonal analysis of continuative shift is impossible, by refering to the extrametricality of the final syllable we can account for this tone shift. I assume, then, that the default antepenultimate H tone assignment rule does not count three vowels (as was suggested tentatively in ( 4 ) ) - as one might propose for a language with antepenultimate stress, the final syllable of the word is extrametrical: the default H tone ("stress") comes about by construction of a foot labeled S W at the end of the word, and assigning H tone (when there is none) to the strong vowel. Thus the antepenultimate position becomes, analytically, penultimate plus an extrametrical syllable. The rightward shift of tone seen in the context of -aa(n)g- then is the result of a prior morphophonologically conditioned rule of reaffiliation whereby the final extrametrical syllable is reinstated to full prosodic status when preceded by -aa(n)g-; since the placement of H tone is stated in terms of the righthand edge of the foot, but the righthand edge of the foot shifts, then the tone will also shift. The assumption that the final syllable is extrametrical in Safwa accounts for a number of things: it accounts for the otherwise unusual position of the default tone in toneless structures; it accounts for the tonal weakening of the final H tone which resulted in the final / pre-prefixal tone alternations in nouns; it helps to account for the rightward shift of the H in the environment of -aagr-; it may also be the explanation for the fact that monosyllabic noun steins lengthen their (antepenultimate) preprefix vowel (aama-ji

'eggs of

lice', aama-bo ' w a r s ' ) , where stress and lengthening are generally found on the penultimate vowels, rather than the antepenultimate vowels in other Bantu languages. Related to this lengthening is a shortening process mentioned by Voorhoeve - long vowels cannot stand before the antepenult, or finally; that is to say, long vowels must be members of the strongest foot of the word. A number of Bantu languages show a similar pattern of restricting long vowels to penultimate and antepenultimate positions. In Chimwini (Kenstowicz and Kisseberth ( 1 9 7 7 ) ) long vowels are shortened before the antepenultimate syllable, and in Kihunde (Goldsmith ( m s ) ) long vowels cannot precede the penultimate syllable; in these languages, the final syllable is extrametrical and long vowels can only appear in strong feet

(with the further restriction

232

David Odden

in Kihunde that long vowels may only appear in the strong syllable of strong f e e t ) ; as in Safwa, 1 assume a final foot labeled W S. An alternative description for Kihunde would be to forego final extrametricality and assume S W labeling: Kimatuumbi appears to be undertaking such a reanalysis. In Kimatuumbi long vowels can only appear in penultimate and antepenultimate syllables.· while vowels are usually long before NC clusters they never are if the cluster stands after a pre-antepenultimate syllable (viz. changal we "gravel"). Due to the lengthening before NC there are almost no length contrasts in penultimate position before NC clusters. However, in antepenultimate syllables, there are length contrasts before NC, viz. lu-kongobe "wood from log' versus lu-koongowe 'marble piece". The distribution of length contrasts in Kimatuumbi can be explained by assuming that long vowels must appear in the strong foot, and by assigning lexical extrametricality to certain final syllables, hence lu-kongobe versus lu-koongo(we). Since extrametricality is lexically assigned to some final syllables, the grouping together of penultimate and antepenultimate cannot be accomplished by W S labeling and regular final extrametricality - hence Kimatuumbi must have switched to S W labeling. 2.2.

Kinga

Spoken in a geographically close part of southern Tanzania, but by the Guthrie (1971) classification of Bantu languages not particularly closely related to Safwa, Kinga as described by Schadeberg (1973) also exhibits a restrictive tone system, where H tones can appear in a limited number of positions in the word and verbal tone is wholely predictable by reference to tense-aspect distinctions. As Schadeberg described it,

the H tone in nouns may appear on the

pre-stem-initial position, antepenultimate position, or rarely in penultimate position. This distribution of tones is strikingly like that found in Safwa (pre-stem-initial replacing ante-pre-stem-initial).

(10)

\li-kuval\lo \li-deede \)kv-gublkila ϋη-galagala \k\-p\l\l\ \k\-tala

"big road" (APU) 'bee' (APU) "to cover for" (APU) "liar 1 (PSI) "flute" (PSI) "bed" (PU)

I assume that the pattern APU represents the default assignment of H to otherwise toneless stems (again, like Safwa, every word in Kinga bears one and only one H tone). The pre-stem-initial pattern (PSI) represents the surface tone

Tone systems in Bantu

233

pattern of nouns with a H tone - Kinga will then require a rule to assign that H tone to the vowel before the stem (whereas in Safwa the H is docked to the final vowel and is thrown back to the pre-prefix vowel by a partially restricted rule). Schadeberg notes that nouns in the tone class APU correspond to proto-Bantu nouns with final L, and PSI corresponds to final H. Since we find the same correspondence between historical final tone and synchronic tone class in Safwa, the historical explanation for the prefixal H in Kinga is undoubtedly the same as it is in Safwa - Kinga constructed right-dominant trees, and presumably had prefix / final tone alternations. The lack of synchronic alternations in nouns between final H and prefixal H is due to the fact that Kinga nouns appear to always have the (historically) H toned VCV form of the noun class prefixes, whereas H toned VCV in Safwa alternates synchronically with toneless CV prefixes. While the position of tone in verbs is predictable from the morphological shape of the stem, there are more tone patterns found in Kinga verbs than appear in Safwa. The tone pattern APU is the most general in its distribution, as should be the case if antepenultimate H is the default pattern in Kinga. Tenses selecting the pattern APU are seen in (11). (11)

nd\-pvlikaga nand\-pOl\ka aava-pulfkaga tvkava-vonaga ndaava-p like nd\ka-be£lile

Ί am hearing' (present) Ί do not hear 1 (general negative) 'he always heard them1 (durative past) 'we saw them continuously' (narrative) Ί heard them' (remote past) Ί refused 1 (narrative perfect)

The tone pattern PSI is found in a number of verbal tenses, which with one exception select the "final vowel" -e.^ (12)

alaava-pvlika \)kava-pvl\ke natuv -ρυΐ\h\ce nalaava-pvl\h\cage \jngava-pvl\h\cage

'he will hear them' (future) 'you may hear them1 (remote adhortative) 'we can not listen to them' (negative subjunctive) 'we won't listen to them cont.' (neg. future) 'if you listened to them cont.' (conditional)

It is rather striking that along with the selection of a different morphological final vowel in subordinate clause verb tenses we also find a different tone pattern in the verb - a similar correlation between selection of the final vowel affix, stem tone pattern and main-clause versus subordinate clause tense has been observed in other Bantu languages. For example, in Shona, subordinate verbs select a tone melody composed of the tone HHLB (where B is a copy of the

234

David Odden

stem tone). Hyman and Byarushengo (1984) observe that the final vowel in Haya has tense-determined tonal properties which in turn affect the tones of the stem. Six tenses take penultimate H - strikingly, these tenses are all and only the nonfuture negative forms, which with two exceptions select the final vowel -i (the exceptional forms selecting -e, which is likely to be a contraction of a-i).

(13)

nat\)kava-pvl\k ge natwaava-pvltke natuk\)va-p\)l\kagi natvva-puliki nand\ka-geendili nandaa-geendlli

"we did not hear them 1 (neg. narrative) 'we were not listening to them 1 (neg.dur. past) 'we are not hearing them' (neg. present) 'we have not heard them' (neg. perfective) Ί did not go" (neg. narrative perfect) Ί was not going' (neg. remote past)

The analysis proposed by Schadeberg is that these verb tenses select the final vowels -e or -a plus the vowel -i which marks the nonfuture negative - a rule reduces ei to i and ai to e after tone assignment, in which case the surface penultimate tone pattern in verbs is a predictable subcase of the pattern APU. There are some tenses which do not straightforwardly fit

into either APU

or PSI patterns, which are quite revealing of the historical development of the Kinga tone system. One of these "opaque" tenses is the subjunctive, which generally has the tone pattern APU, but takes PSI tone if the verb has an object prefix.

(14)

a-pul\ke ci-ha£m\)ke a-va-p Floating > Prelinked, as dialects move from a tonal to a pitch-accent type. Finally, the fact that systems of the Ijo type are apparently rare demands some explanation. The first prerequisite for this type of system appears to be a large domain; if,

for example, the domain in a given tone language is the

syllable, or a monosyllabic morpheme, then it will not be possible to distinguish Domain from Prelinked tones, and the scenario for this type of development will not arise. The unusually large domain of Ijo, the group, has apparently given scope for the development from tonal to pitch accent systems. In its turn, the large domain is possibly the result of a syntactic need to clearly delimit subject from object in an SOV language which lacks case suffixes.

Table 7.

Dialect differences quantified Kalahari

1.

Bumo

a

.

Domain

4

4

0

0

b

.

Floating

6

5

1

2

5_ 1

3.

Nembe

Number of underlying tones:

c . Prelinked

2.

Kolokuma

1 5

1

1

1 6

1 5

Rules:

FTPlace

1

1

1

1

DTSpread

1

1

0

0

Raising

0

0

0

1

Retraction

0

1

0

0

Default

l_

L

3

4

L 2

L 3

Cycle:

i Total:

1

l 9

1

6

2.

P-

8

8

Tone in Ijo

277

Notes *An earlier version of this paper was presented at the Meeting on Non-Linear phonology held at the University of Tilburg on Feb. 1, 1985. I am grateful to the participants there, especially T.L. Cook and Harry van der Hülst, for helpful discussion. 1.

Ijo belongs to the Niger-Congo family, within which it was classified as Kwa by Greenberg (1963). More recent work indicates that its relationship to the languages classified as Kwa by Greenberg is a much more remote one, and together with the tiny Defaka language (Jenewari 1983) it is now tentatively classified in the Ijoid branch of Atlantic-Congo, which is one of the branches of Niger-Congo (Bendor-Samuel, forthcoming). IJ9 has a large number of dialects which are sub-classified into four clusters. Each cluster may be regarded as a language in that there is generally mutual intelligibility between the dialects within the cluster. Adjacent dialects between different clusters are sometimes also mutually intelligible, while intelligibility is rapidly lost away from the borders. The dialects are here numbered according to a code in which the first digit indicates the language and the later ones indicate subgrouping within the language. The dialect names are abbreviated as follows: OR BI OA NK IB KA OK

= = = = = = =

Oruma Biseni Okordia Nkoro Ibani Kalahari Okrika '

GB = KO = IK = ET = ON = KB = KU =

Gbanrain Kolokuma Ikibiri East Tarakiri Ogboin Kabou Kumbo

WT = ME = AP = BA = OL = BU = OP =

West Tarakiri Mein Apoi Basan Olodiama Bumo Oporoma

01 = ID = OG = OZ = AR = AK = NE =

Oiyakiri Iduwini Ogulagha Oporoza Arogbo Akassa Nembe

2.

Except, in some dialects, an initial V syllable or (in dialects such as Mein, not further considered here) any group-initial syllable.

3.

A group-initial V in Kolokuma is unaffected by DTSpread and therefore receives a Default L, except that some (not all) morphemes of the shape VrV are affected: äru 'canoe' but ere 'woman'.

4.

Both here and in similar cases where the configuration V V results H+L from the rules, an optional surface assimilation of L to H is possible.

5.

The definite markers -ma 'the (feminine singular)' and -mo 'the ( p l u r a l ) ' accept the L, yielding -maä, -moo; but the simple past marker -mi does not and remains -mi.

6.

For classes B and E, the same realizations occur when the next morpheme begins with a vowel.

7.

As in Kolokuma, the initial V is raised in some but not all morphemes of shape VrV: äru but ere.

8.

The final tone is realized H sentence-finally in a statement. Since all groups end H sentence-finally in statements and L in questions, this H is interpreted as a marker of statement intonation.

9.

The configuration H+L results in an obligatory surface assimilation of V L to H.

275

Kay Williamson

References Bendor-Samuel, J.T. (ed.) (Forthcoming), Niger-Congo. Amsterdam: North-Holland. Clements, G . N . , and S.J. Keyser (1983), CV phonology: a generative theory of the syllable. Cambridge: MIT Press. Cook, T.L. (1985), An integrated phonology of Efik, Volume I. Doctoral thesis, University of Amsterdam. Dordrecht: ICG Printing. Edmondson, T . , and J.T. Bendor-Samuel (1966), Tone patterns of Etung. Journal of African Languages 5, 1-6. Greenberg, J.H. (1963), The languages of Africa. The Hague: Mouton. Efere, E . E . (1981), Pitch accent in the Bumo and Nembe dialects of Ijo: a comparison. B.A. long essay, University of Port Harcourt. Jenewari, C.E.W. ( 1 9 7 7 ) , Studies in Kalahari syntax. Doctoral thesis, University of Ibadan. Jenewari, C.E.W. (1983), Defaka: Ijo's closest relative. University of Port Harcourt Press. Kio-Iyalla, P. (1984), A. comparative analysis of the suprasegmentals in Bumo and Kalahari. B.A. long essay, University of Port Harcourt. Leben, W.R. (1971), Suprasegmental and autosegmental representation of tone. Studies in African Linguistics, supplement 2, 183-200. Le Saout, J. ( 1 9 7 9 ) , Notes sur la phonologie du Gouro. C.E.P.L.A.N. 1. Universite de Nice, U . E . R . Lettres et Sciences Humaines. Pulleyblank, D.G. (1983), Tone in lexical phonology. Doctoral thesis, MIT. Rowlands, E.G. (1960), Tone and intonation systems in Brass-Nembe Ijaw. African Language Studies 1, 137-154. Voorhoeve, J. A.E. Meeussen, and K.F. de Blois (1969), New proposals for the description of tone sequences in the Igbo completive phrase. Journal of West African Languages 6, 79-84. Williamson, K. (1965) , A grammar of the Kolokuma dialect of Ijo. (West African Language Monographs, 2 . ) C.U.P. Williamson, K. (1979) , Sentence tone in some Southern Nigerian languages. Proceedings of the 9th International Congress of Phonetic Sciences, II, 424-430. Copenhagen: Institute of Phonetics.

Phonetic and Phonological Considerations bearing on the Representation of East Norwegian Accent Meg Withgott & Per-Kristian Halvorsen CSLI, Stanford University 1.

Introduction

In this paper, we explore some phonetic and phonological predictions that arise from a lexical-autosegmental analysis of East Norwegian prosody. We find support for a postlexical/lexical distinction in tone assignment, and we report on evidence against a "delayed accent" analysis. We also demonstrate that an autosegmental account correctly predicts hitherto unobserved phonetic differences between accents 1 and 2. Many analyses link the appearance of accent 2 with polysyllabic words. We have arcrued previously that East Norwegian tone accent depends on the placement of stress, morphological composition, and on the lexical phonology, and not on a syllabic-counting rule (Withgott and Halvorsen 1984). This point will remain controversial so long as exception mechanisms are posited to account for polysyllabic accent 1 words, and thus we endeavor to show that such mechanisms are unmotivated in this domain. We also provide further support for the lexical theory through an extensive study of compounding and accent assignment. 1.1

Summary of the lexical autosegmental analysis

The analysis proposed in Withgott and Halvorsen (1984) is based on the theories of autosegmental and lexical phonology. (We will assume familiarity with the general history and central ideas of these two theories on part of readers of this volume. An overview may be found in Van der Hülst and Smith 1982.) Tone accents are a pitch/stress composite. In the Oslo dialect, an accent 1 word pronounced in isolation or embedded in a sentence exhibits a contour with a low Fo point on the stressed syllable, which may continue to fall. Tone accent 2 results in a contour with a high FQ point on the stressed syllable which then sharply falls. In either case there can also be initial and final peaks, which are not lexically distinctive. The lexical phonological specification of an accent 1 word, e.g. nunmer 'nummer' and of an accent 2 word, e.g. sonaner 'summer', are L (read "low") and H (read "high"), respectively. The L tone for nunmer is not listed in the

280

Meg Withgott S Per Kristian Halvorsen

lexicon but is introduced by rule. Both tones are associated with the stressed syllable by tone-linking rules. These simple lexical representations are augmented by phrasal tones which are not lexically distinctive. Such postlexical tonal autosegments include an optional initial word-boundary M ("mid"), 1 a final L word accent, and an additional final H accent if the word is in focus position in the discourse, which includes the case of a word pronounced in isolation. Assuming that adjacent identical word tones are merged, the postlexical specifications for citation forms are thus (M)HL(H) for accent 2 and ( M ) L ( H ) for accent 1. Figure 1 contains FQ tracings of three utterances of the phrase lange nummer ('long numbers') differing in the presence of the final optional H. 2

H (for emphasis)

H

A.

H (for emphasis)

C.

A. LANGE nummer B. lange nummer C.

lange NUMMER

Figure 1. Repetitons of the phrase'long numbers' (accent 2 accent 1).Words in uppercase indicate emphasis; boldface Ή' and Ί' indicate lexical tonal autosegments.

Bast Norwegian Accent

281

The assignment and linking of tones in Norwegian, a tone accent language, resemble these processes in tone languages, but there are certain differences. Instead of aligning the first tone with the first syllable in the word, we link the lexical H or L tone to the

(primary) stressed syllable. Monosyllabic

words are unstressed lexically since the manifestation of stress assumes a prominence contrast between syllables. The postlexical grammar fills in the melody on either side of the stressed syllable. Note that underlying forms are radically under specified for tone, unlike in tone languages where all or most vowels have a tonal specification. Words such as Itererinne

'teacher 1

and alene 'alone 1 show that the tone is associated with the stressed syllable, in these cases the penult. (Such medially stressed words are examined in detail in section 2) . Those stressed syllables not linked to a tonal autosegment are subject to a rule of Default Accent Specification which will supply a L tone. The mapping between segments and tones is accomplished by a single rule, Tone Association. Turning to the morphology, prefixes and suffixes are categorized according to which tone they assign and in terms of dominance. Affixes such as u- misand -lig, -dom, and -else cause accent 2 (H) to appear on the accented syllable of words they form, as in:

(1)

uoppklart

misgrep

g j φι lig

sykdom

[u [opp [klart] ]]

[mis [grep] ] n w

[[gjizSr] lig] ADJ

ί [syk] dom] N

NEG-up-clear

mis-grasp

do -able

sick-NOMLZR

1

'mistake '

'doable'

'illness'

'unresolved

Other affixes cause L tone (accent 1) such as the superlative ending -(e)st or the nominalizer -sei :

(2)

vakrest

kjtfrsel

[vakker] est]

[ [kjjzir] vsel] N

beautiful-SVPERL 'most beautiful

1

drive-NOMLZR 'driving'

Neutral affixes do not influence the tone of the final form.

(The tone accent

is indicated to the left of the stressed syllable.) The following forms demonstrate that there is no tonal change when -aktig is added to polysyllabic words. We also maintain that the lexical tonal property of the monosyllabic stems is similarly preserved.

252

(3)

Meg Withgott S Per Kristian Halvorsen TJ

L

fabelaktig

fabel

dameaktig

'fabel 1

' fabulous '

H, dame

' lady- like '

' lady '

FT

vattaktig

vatt

' cottony '

vannaktig

'cotton

1

vann

'water-like'

'water'

Among the tone-inducing affixes, we distinguish between weakly dominant affixes which contribute tone when combined with a non-derived base and strongly dominant affixes which also impose tone on derived forms. -lig, -else, and -sei are members of the class of weakly dominant suffixes. II

An example such as

utferdsel

Outgoing t r a f f i c 1 illustrates that the tone

provided by the class of weakly dominant affixes can be overwritten by subsequent affixation,

H

utferdsel is formed from

TJ

ferdsel

'traffic',

(from the stem ferd-

'travel'). Prefixation of ut~ overrides the L tone supplied by the weakly dominant inner suffix -sei. ut- exemplifies the second category of tone inducing affixes —strongly dominant affixes— which determine tone even if the base form is derived. Other examples include the prefixes om-, pä- and u-. Finally, category information helps determine accent. For example, L tone is found when om- is used to derive a verb, but H tone appears when om- functions as a nominal prefix:

(4)

a. Nouns with H tone accent omtale

omhug

omsikt

'review'

'concern'

'regard'

ornrade

omradeplan

•area'

1

'area map

omrSdeplanlegging 'area planning1

b. Verbs with L tone accent ozntale

omarJbeide

'talk about

'rework'

The description of tone contours depends on the lexical tone of stems, or default L specification, and morphology. At every morphophonological level, any primary stressed syllable not linked to a tone receives a L autosegmental specification by the default rule. Thus the assignment of accent 1 is lexical, but provided either by the rule of Default Accent Specification or through word-formation involving a L-inducing affix. While for stems it suffices to indicate whether or not there is a H tonal autosegment, affixes must be specified as H- or Linducing, or else unmarked as neutral. These simple tonal representations and tone-linking conventions further serve as the basis for a description of the assignment of tones in words derived from

East Norwegian Accent

283

polysyllabic and monosyllabic stems when they enter into compounds. We will pursue the argument in section 3 that only a monosyllabic stem without a linked tone—an unstressed stem — is affected by the affixation of L- or H- inducing inflectional affixes. While non-neutral inflectional affixes supply the tone when combined with the monosyllabic, unstressed stem, the stem determines tone when forming the first member of a compound. The underlying stem tone is similarly revealed when the unstressed stem is combined with a neutral affix. We will now turn to a discussion of the sound of accents 1 and 2 in various polysyllabic environments, and show how lexical and postlexical tonal specifications interact to yield a phonetic representation. We then emphasize the regularity of monosyllabic and polysyllabic phonological representations, and show that recognizing this regularity, in turn, reduces the need for exceptions in an account of compounding in East Norwegian.

2.

How to interpret the phonetic evidence: Why a [ + delayed accent] analysis is not tenable

Bruce (1977) shows that intonation depends on several sources of accent (e.g. word accents, sentence accents). However, it is essential to distinguish the compositional claim, which says that intonation is determined by several kinds of accent, from the claim that a single tonal contour is represented in the grammar, and that rules of temporal alignment distinguish accent 1 and accent 2. Interestingly, the latter claim can be shown to be false. Recall that the tone-linking process makes crucial reference to "stress". Stress in East Norwegian is similar in many respects to stress in Scandinavian dialects which do not have tone accent, such as the Swedish spoken in Finland. Stress in East Norwegian exhibits both the usual phonetic correlates of stress in these other dialects (for acoustic features of stress see e.g. Engstrand and Nordstrand 1984) and phonological properties commonly associated with stress. For instance, a stressed syllable must contain

(minimally) either a long vowel,

or a short vowel followed by a complex coda, while vowel reduction, or deletion, and a limited vowel inventory characterize stresslessness. We thus feel it is necessary to distinguish "stress" and autosegmental tone. However, this distinction can lead to the reductionist analysis of the two tone accents as one melody with two rules of time alignment, which we would like to avoid because of both phonological and phonetic facts. However, many support such a reductionist move. Haugen and Joos (1983:194) write, "In both accents there is a potential melodic curve of high-low-high,· in accent 1 the stress falls on the low, in 2 it falls between the first high and the low, so that the differ-

284

Meg Withgott S Per Kristian Halvorsen

ence between them is one of phase". Lorentz (1984:175) writes "The variant of the accent hypothesis defended here, is thus formulated in a way that does away with lexical tone specification altogether. The tone in question is intonational tone, the accented part of which is delayed one syllable in words containing morphemes as post-accenting." The overall picture -t^hat a sentence containing accent 1 and 2 words will exhibit a sequence of rises and falls- is hardly controversial; the issue of whether or not there are two accents, as opposed to one accent displaced in time, is. Comparisons of accents 1 and 2 which are limited to words with initial stress turn out to be misleading in examining this question for East Norwegian. Words with medial stress display more variation in the realization of the accents, which is problematic for the delayed realization analysis. By employing a compositional autosegmental account we are able to distinguish lexical and postlexical sources of intonation peaks which makes it possible to explicate this variation and the differences between the contours of words in context. Moreover we predict, that the overall shape of accent 1 and 2 lexical items is not identical contrary to the delayed realization account. In examining this prediction, let us consider penultimately stressed words such as materiale (accent 1) "material 1 , and laererlnne (accent 2) 'teacher 1 . A native speaker can produce, a variety of Fo contours for these words depending on discourse factors such as emphasis. We will show that because of this variability, a delayed accent analysis which either posits H* L H for accent 2 with corresponding H L* H for accent 1 (or cf.

Endresen 1983:380), or L H* L for

accent 2 and L* H L for accent 1, will not account for the East Norwegian data. The third alternative, H L H* and H L* H, is not attested in this dialect. (For the reader's convenience, we employ Goldsmith's asterisks to indicate where the tone melody aligns with the segment string (Goldsmith 1976). We dispense with the star in our analysis since the metrical representation of the word directly identifies the location of the stressed syllable, and the lexical tonal autosegment docks to this designated nucleus.) Let us first consider alternative 1, the H* L H/H L* H analysis. Such an analysis could account for the following possible contours:

H L

l a e r e r i* n n e

accent 2 Figure 2. Analysis of two contours under alternative 1

m a t e r i a * l e

Bast Norwegian Accent

285

However, accent 2 can also be realized as in figure 3 where there is a narrow pitch peak over the stressed syllable. Alternative

(1) can be amended by the

addition of an initial mid tone to account for accent 2 contours like this:

lae r e r i * n n e Figure 3. Analysis of an additional accent 2 contour under revised alternative 1

Yet this would predict the existence of accent 1 contours with similar medial peaks, and such contours are unacceptable:

NOT accepted (accent 1) m a t e r i a * l e Figure 4. Alternative 1 overgenerates

For East Norwegian, the tone melody of an accent 2 word cannot be grafted onto an accent 1 word, given a (M)HLH representation and assuming the accent 1 word's stress is associated with the L, and the L of accent 2 is associated with the post-stressed syllable. Let us now consider alternative 2, the melody LHL. This alternative would again predict the contours analyzed in figures 3 and 4, but with a different temporal alignment and tone melody, as illustrated in figure 5:

(A)

H*

l a e r e r i * n n e accent 2 Figure 5. Alternative 2

(B)

m a t e r i a * l e accent 1

286

Meg Withgott s Per Kristian Halvorsen

The accent 1 contour for materiale Ls preferable to that illustrated in ( 4 ) , however it is incorrect in that it specifies a final L after the H. Worse, alternative

2

cannot account for the quite acceptable contour we have just seen for

laererinne ( c f . Figure 2 ) :

lae r e r i n n e Figure 6. LH*L cannot be associated to a common contour; alternative 2 undergenerates Any appeal to additional boundary tones would not help the LH*L analysis due to the variations permitted for accent 2 contours. Thus, neither alternative , 1 nor alternative .2 describes all and only the permitted contours. Turning now to our theory, we interpret the acceptable contours as (M) E L (H) for accent 2, where underlined ''H" indicates the lexical tone, and the prefaced " ( M ) " is the optional initial boundary tone; and for accent 1, (M) + L + L + H which is simplified to (M) I, (H) through the obligatory contour constraint

(Leben

1973), and where underlined "L" indicates the lexical tone, and the prefaced " (M)" is the optional initial phrasal tone. We cannot generate the ungrammatical accent 1 form shown in figure 4. Another good result is that (5B) is also predicted not to occur. (a)

H

(b)

lae r e r i * n n e

m a t e r i a * l e

(H) L

*

_j_:

X

l a e r e r i * n n e

.

(H)

\ L

m a t e r i a * l e

Figure 7. The lexical-autosegmental analysis In conclusion, it is empirically inadequate to characterize the accentual difference in East Norwegian by means of a low on the stressed syllable in accent 1 words and a delayed low in accent 2 words as in the delayed accent account.

East Norwegian Accent

287

Where accent 1 and 2 contours are identical (abstracting away from temporal alignment as in 7c-d), it is due to the interaction of the lexical and the postlexical tonal specifications. Due to these same factors, the two contours can also be quite distinct.

3.

What to make of the phonological evidence: Why it's wrong to rule out monosyllabic accent 2 forms in the lexicon

Vanvik (1978:162) has remarked that polysyllabic words tend to retain their accentual properties when they become the first member of a compound, e.g. accent 1 panser ' armor1, > accent 1 panservogn ''tank1 and accent 2 konge 'king' > accent 2 kongevogn · royal coach· . Assuming that the tone of the compound reflects the tone of the first member, and since monosyllables in compounds do not always exhibit accent 1 (e.g. accent 2 sporvogn 'tram c a r 1 ) , Vanvik concludes that monosyllabic words should not be analyzed as bearing accent 1 despite some similarity in the pitch contour between monosyllabic words pronounced in isolation and polysyllabic accent 1 words. We think that Vanvik's reasoning is essentially correct for the lexical phonology in that he rules out considering all monosyllables as accent 1. However, in the phrasal phonology, we see no reason for not assigning a L accent to monosyllabic words by the operation of Default Accent Specification, if indeed such an assignment is warranted by the phonetic data. It is a different matter if a monosyllabic stem has undergone lexical derivation. Given our autosegmental/lexical model of phonology, accentual properties for monosyllables may only emerge in specific contexts. Thus we claim that spor is associated with a H autosegment that is linkable after compounding with vogn and stress assignment. While we accept the logic of Vanvik's argument that spor cannot be accent 1 due to its behavior in compounds, we restrict the scope of the conclusion to the lexical phonology. Few seem to accept even that part of Vanvik's argument. Liberman (1982: 23) writes, "An equally plausible rule would prove that spor has ace. 1: disyllabic words tend to retain their accents when they enter into compounds, while the accent of monosyllabic words, under similar circumstances, is subject to metatony. A rule like this would find numerous counterparts all over the Scandinavian area." Without trying to evaluate the plausibility of such a tone-switching rule for other dialects, we can note that it does not work well for East Norwegian. Certain monosyllables exhibit accent 2 when they enter into compounds, or when

288

Meg Nithgott S Per Kristian Halvorsen

they are suffixed with neutral affixes. Other monosyllables exhibit accent 1 when they enter into compounds, or when they are suffixed with the same neutral affixes. The lexical-autosegmental proposal provides independent criteria for determining the lexical tone of monosyllables based on their behavior under affixation. And it correctly predicts that the tonal properties which are determined on this basis will decide the behavior of monosyllables in compounding. Certain inflectional affixes induce accent. The definite singular marker, -en/-et, causes accent 1 (cf.

gutten "the boy";

gullet 'the gold 1 ).

Polysyllabic stems suffixed with these lexical items exhibit either accent since, first, they have a stressed syllable on which the lexical tone docks, and second, such affixes are not dominant. Examples such as pike 'girl 1 II

H

L

piken 'the girl1 and villskap 'wildness' -

L

villskapen 'the wildness'

bear this out. The indefinite plural form of the common gender illustrates the same point (variation in the tone of indef. pi. polysyllabic stems:

villa+er

T3

"single family homes'; monosyllabic stems:

H

rose +er 'roses', but consistently accent 2 in indef. pi.

gutter, 'boys',

to a compound formed from .brann, e.g.

H

branner 'fires' etc.). Compare iranner brannfare,

'firehazard'. Brann is not

marked H in the lexicon, so Default Accent Specification

supplies a L autoseg-

ment once the compound undergoes stress assignment. The word subsequently remains L in the plural,

brannfarer, and we predict that all

compounds formed from

brann will behave identically. Without the theoretical notion of tone linking to stressed syllables, and the notion of a default rule, we would not be able to account for the range of data discussed so far. The metatony alternative would have to mark all compounds formed from, e.g. brann, as exceptions, while inflected forms such as .branner would not be exceptional. We will now show that even such weak, exception-based accounts of forms like

T L

JJ H

Jbrannfarebrannfare- iranner branner are hard to maintain because these

pairs turn out to be the norm in the language.

4.

A lexical study of East Norwegian compounds

We assume that the tone accent of a compound word is determined by the tone accent of the stem which carries the lexical stress in the compound. A number of predictions follow from this claim when taken in conjunction with our analysis of lexical tone, Tone Linking, and Default Accent Specification. In particular, we predict that compounds which share the same stressed stem will all exhibit the lexical tone of the stem. This entails that if a stem α is polysyllabic and

Bast Norwegian Accent

289

bears the lexical stress in a compound β, β will have the same tone as α occurring on its own. If α is monosyllabic, the tone of compounds based on α is predicted to reveal

the lexical tone of a, which may be accent 1 or accent 2, even

though no tonal contrast is possible when α occurs in isolation. We also predict that the tone of the compound formed on a coincides with the tone of forms derived by means of neutral affixes such as -aktig. Since compounds, and especially compounds based on monosyllabic stems, provide a good test of our predictions, we undertook a careful quantitative study of this part of the morphology.3 In particular, we evaluated our claim that monosyllabic words have accentual properties in their lexical representation, although no contrast is manifested when they remain unaffixed or otherwise appear as single syllables in the phrase. Secondly, we examined the common hypothesis, rejected here in our lexical analysis, that monosyllabic words are accent 1. In our study, we used a machine readable diactionary for Norwegian with morphological segmentation and tone accent indication.^ We developed a computer program which was capable of automatically isolating all the dictionary entries which were likely to be compounds based on monosyllabic stems. The search routine resulted in only a small number of false positives which we eliminated by manually perusing the output from the search. We found over five thousand compounds based on monosyllabic stems.

Of these stems, a quarter are accent 1 ( 2 5 . 4 % ;

1310 items) and three-fourths are accent 2 (74.6%; 3852 items). This already was a significant finding for us since it is highly undesirable to treat one-fourth of the words in a language as "exceptional", a consequence of the metatony or other exception-based analyses. Moreover, these compounds are not characteristically loans, or recent coinages, and more important, these lexical items behave in a systematic fashion that the grammar should account for. The compounds listed in the dictionary were based on approximately eighthundred monosyllabic stems. 464 of the stems occurred in at least two compounds. Of these, most of the stems (85.7%; 268 items) formed compounds which all had the same accent. Sixty-six stems ( 1 4 . 2 % ) fluctuated, i.e.

they formed both

accent 1 and accent 2 compounds. Of the consistent stems, 32.7% were accent 1 inducing and 67.3% were accent 2 inducing. 6

Meg Withgott s Per Kristian Halvorsen

290

accent 2

100%

accent 1 50%

mixed 1/2

0%

Consistent Stems

Fluctuating Stems

Figure 8. Monosyllabic stems and their behavior in compounds The ratio of fluctuating stems to consistent stems is small enough that we might be willing to simply consider the sixty-six fluctuating stems as exceptional, but there are in fact further distinctions between the fluctuating stems and the consistent stems which help in explaining the existence of the two categories. Most notably, few compounds are formed from a given fluctuating stem. 7 Linguistic rules in general, and lexical processes in particular, are not exception free. Any systematic investigation of a large corpus of data is a reminder of this. But in the case of the tone accent system of Eastern Norwegian the quantitative evidence clearly supports our hypothesis that monosyllabic stems can induce either of the two tone accents, and that stems have stable lexical tonal properties. Therefore, we reject exception-based accounts of forms τ

like

τι

brannfare- branner, since we have shown that such lexical items behave in

a systematic way, and that our analysis straightforwardly accounts both for such forms and for the general fact that compounds consistently exhibit the tone of their stressed member.

5.

Summary

We have been able to show that the lexical-autosegmental analysis can account for the behavior of both monosyllabic and polysyllabic stems in compounds as well as in derived forms. Moreover, we can account for a larger range of tonal contour data than analyses which assume one complex tone contour in the grammar

Bast Norwegian Accent

291

with two alignment processes. Adopting the lexical-phonological analysis of tone accent obviates the need to treat Scandinavian languages as a prosodic anomaly displaying a special category of "accent". Our reliance on autosegmental tones, and the use of default rules, is reminiscent of analyses proposed for unrelated (African) languages (e.g. Hyman and Byarushengo 1984 and Pulleyblank 1983). Languages differ as to how and when autosegments are mapped to tone-bearing units. In languages such as.English, the association is never lexical; in tone languages, sequences of tones may be associated with sequences of syllables at the start of the morphological derivation. In East Norwegian, the association is both lexical and postlexical, and the language is described naturally in terms of tones and stress.

Acknowledgements We discussed several ideas in this paper with Gösta Bruce, and we wish to express our appreciation for his feedback. All errors are our own. The research for this work has been made possible in part through an award from the System Development Foundation to the Center for the Study of Language and Information.

Footnotes 1.

In our previous work, we assumed that the initial M was actually a high lexical tone. It seems more accurate to characterize it as a word-boundary tone which never seems to reach the extremely high FO values observed for either the accent 2 lexical H or the final emphatic H.

2.

The speaker is the second author. We are grateful to Dr. Mark Liberman of AT & T Bell Labs and Prof. Gösta Bruce of Lund University for providing us with F tracings .

3.

When we say "compound based on a monosyllabic stem" we mean a compound where the stress falls on a stem which is monosyllabic.

4.

The data available to us was the Norwegian half of a Norwegian-German dictionary. We are grateful for to Mr. Knut Lie of Universitetsforlaget, Bergen, for giving us permission to use the dictionary, and to Mr. Knut Hofland of NAVF's Computing Center for the Humanities, for providing us with a magnetic tape of the dictionary.

5.

The following is a classification of stems based on the Oslo dialect pronunciation of the compounds examined. Stems which consistently induce accent 1 in compounds:

292

Meg Witgott s Per Kristian Halvorsen

•rmv^jvji.

f

J k.ji *-*-!- -I-

f

Jfc^j.

f

JVJ, y .J -J

f

/VL4. — /

SVU M^— /·

"· V β"

/

J. αΛΛ, — /

_i. G — /

-J. l^— j

lokk-, mai-, mo-, moll-, mor-, nipp-, no-, nok-, nu-, null-, na-, pakk-, papp-, pass-, plass-, platt-, plukk-, poeng-, pop-, press-, pukk-, rang-, ratt-, revy-, ro-, rom-, russ-, sang-, shet-, sjakk-, sjekk-, sjokk-, ski-, sko-, skott-, skru-, skra-, skyss-, slapp-, sleng-, slum-, slyng-, slass-, smuss-, snitt-, song-, spenn-, spil*·, «i* 1 7 _

f-mi_

r - 4 - n m_

ί? +- ΐ li-L··—

e· j-/^ff—

e· 1-skirls—

.

^uyy

c 4-f~irir)—

,-

- L by one or another tonal rule — a case in the spirit of absolute neutralization. In the present double level system, the complete range of 65 tenses fall out rather naturally. The second point, and the more important point, is the learnability problem. The data for tone languages is almost overwhelming in the raw forms and much of the tonal literature cannot be compared fruitfully,

since, as Dwyer (1983)

makes clear, descriptions done by linguists of various theoretical persuasions obscure similarities of process across languages. The model presented here predicts sharp constraints on possible accentual languages, i.e.

only segmental

morphemes may contain arbitrary boundaries to tone processes, there may or

Pitch Accent Languages

may not exist a higher level 'accenting'; if such accenting occurs it

309

is

limited to end, either # or I; whether # can function as a boundary for level 0 tone assignment is a language-specific parameter. Furthermore, higher-level accenting as done by End Rule is limited to a single element, with a possible late doubling effect. There now exist diagnostics for the language learner within the system. If there exist long spans of marked (usually H) tone, then Level 0 mechanics are involved. If there exist single H tones, then Level 1 machinery is involved. Furthermore, a unified Level-Grid theory predicts that a common non-segmental process will be End Rule accenting, and we predict that such rules can only be of the form: add Level 1 accent to End. The only other logical possibility which a Level-Grid theory predicts would be add Level 2 to first Level 1 H. However, if only H and L are possible, then Extrahigh is meaningless and we expect not to find these rules. The arguments against tonal approaches to pitch accent languages then hinge crucially upon there being two types of tone — a spreading type and a single entry type, and in all the languages examined here H pitch functions as fillin for both types of accent. A purely tonal approach can keep track of H and L differences, but

'spread 1 vs.

'non-spread' H cannot be done by purely tonal

means. We return now to final arguments — and these will be against prior metrical approaches, with Zubizaretta's (1982) system chosen for discussion. Zubizaretta also advocated two types of tone,

'spreading 1 and 'single entry 1 , but in her

system this was accomplished by using a combination of 'harmony 1 or non-polarized trees and polarized (- stress) trees. We claim here, however, that the present system has certain advantages over Zubizaretta's. 1. In terms of learnability criteria: The present analysis claims that Level 1 accent can only be H tone; a labelled polarized tree predicts single accentual L appearing, which does not seem to be the case. 2. Also learnability criteria: The accentual patterns of Marugame and Osaka Japanese dialects are identical for three out of four classes, yet the two dialects receive very different treatments in Zubizaretta's system. In the present system, the differences between the two dialects follows simply from whether the Level 1 accent rule applies or does not apply to forms lacking Level 0 accent. 3. Finally, we present empirical coverage arguments: Narada dialect has been problematic for *-accent analyses as pointed out by Abe (1981). Here we show that Zubizaretta's approach will not cover such a dialect, but it

follows simply from the present approach.

310

4.1.

Martha Wright

Other Japanese Dialects

Zubizaretta

(1982) treats six types of Japanese accent systems, using a

combination of stress and harmony trees. Of these six types, the first

is

treated by stress tree only; the remainder involve both types of trees. We agree that the first type follow from a single prominence, as does Prince ( 1 9 8 3 ) , and thus the first type consisting of four dialects are here analyzed as pure Level 1.

(23)

1. Shimagawa (l H followed by series of L)

X = # 0

2. Miyakonozyo (l H preceeded by series of L)

=

3. Izumi (1 L, then 1 H, then series of L)

0

0

0

...

X 0 #

0

X = #(0) 0

0

4. Kagoshima - two classes

Λ

a. Class 1 = rightmost H, series of L

= ...

b. Class 2 = rightmost L, then H, then series of L

0

0

0

#

0

0 (0) #

Two other dialects are analyzed by the present system as non-mixed, i.e.

pure

Level 0: Koshi, to be discussed shortly, and Tokyo, discussed in part 1.1 -

1.2. Three other systems are here analyzed as mixed Level 0 and Level 1 and two of these are treated as mixed, albeit mixed harmony and stress by Zubizaretta. They are Osaka and Marugame. We compare patters for these in

( 2 4 ) and ( 2 5 ) .

i of the present analysis is shown.

(24)

(25)

Osaka Pitch Patterns Class 1 = Level 1 accent

Class 2 = Level 0 accent

1.

...

Ο ΠΓ" #

3.

#~~0

Γ77~#

2.

...

0 I 0 4- 0...#

4.

#~0

.. . 4- 0... #

Marugame Pitch Patterns Class 1 = Level 1 accent

Class 2 = Level 0 accent

1.

# 0... #

3.

# 0

~#

2.

#...

4.

#

Γ77Ϊ

θΓ~0~Τ·0...#

Ο

0...#

The single difference between the two dialects is in pattern 1, which receives Level 1 accent in Osaka, and will not receive such accent in Marugame. We

Pitch Accent Languages

311

suggest for both dialects that Level 1 accenting precedes H assignment to delimited Level 0 material and that #- and -# can function as tone domain limiters as we have seen before. Osaka Class 1 can be obtained by End Rule with lexical Ψ functioning as rightmost end as ( 2 6 ) makes clear.

(26)

Derivation for Osaka Pattern 2 X

. . .0 Ο

Ο Ψ Ο

0

Marugame Class 1 rule will differ from Osaka only in the form of the accent rule — any Level 1 rule must refer in Marugame to Level 0 entries. Zubizaretta treats Class 2 of both dialects as having a H harmony tree followed by a L harmony tree. The Class 1 analysis differs greatly for the two dialects as schematized in ( 2 7 ) .

(27)

Zubizaretta Class 1 analyses Marugame

Osaka H

(L)

L

(H)

harmony

harmony

She uses the differing directionality of her Class 1 trees to set directionality for Class 2 harmony trees for the differing dialects, and thus her analysis results in differing branching of the Class 2 harmony trees for the two dialects. We suggest that the two level approach advocated here makes available to the language learner only a single analysis of the surface facts of Class 2 for both dialects, i.e.

Level 0 and a single 4- marker.

We show coverage here of the two other dialects treated by Zubizaretta — first Koshi which is here treated as pure Level 0. Koshi words show four pitch patterns as in ( 2 8 ) .

(28)

Koshi Pitch Patterns Class 2

Class 1

# ι. # ο ΠΓ 0 2. # ο Γο Ο...Ιo....#

3.

# 0

4.

# 0 0

0

...

#

. . . lo . . . #

The analysis here is that the Class 1 patterns differ from Class 2 patterns in having a single extrametrical element, no #[·)·] , but rather #. . functioning

312

Martha Wright

as leftmost boundary. 1 and 3 have no lexical Ψ , 2 and 4 have-)· marked, and rightmost # can also function as delimiter of tone domain. Zubizaretta notes that single syllable words of class 4 surface with L pitch and hence treats Class 1 as having tree form as in ( 2 9 ) .

(29) L

#

H

H

H

σ

Note that marking a single syllable # ( 0 ) # might appear to be contrary to stress theory, but that is only if we wish to apply Level 1 stress to this form , which would not be possible. However, if extrametricality is in a Level 0 form, then we are saying: spread H to all material between #...#

for this

dialect except for first syllable of class 1. Note also that the present system prohibits analyses employing Level 1 low pitch — we predict only H pitch can be involved. Zubizaretta also uses a polarized L for Class 2 of Hirosaki which has two pitch patterns as in ( 3 0 ) .

(30)

Hirosaki Pitch Patterns Class 1

Class 2

o #

... |o

Γ771 ο #

Again, analysis for Class 2 follows from extrametricality, with Class 2 being mirror image Tokyo. Class 1 can be treated as receiving Level 1 accent — we saw in Marugame that Level 1 accenting can include reference to Level 0 — here we suggest that Hirosaki stipulation is that if Level 0 'accent 1 has not applied, i.e. if there is no lexical \, then Level 1 will apply. Thus, we argue that the present approach provides at least the same empirical coverage as did Zubizaretta's without recourse to branching trees of indeterminate direction. A final point favoring the present approach concerns the Narada dialect. The data is taken from Abe (1981).

(31)

Narada Pitch Patterns single word 1. ka | bu 1 to 2. ~ko | ko | ro~

with clitic -ga 'helmet

1

' heart'

ka \ bu \ to -ga ko

Pitch Accent Languages single word

313

with clitic -ga

3. ka\ ga mi

'mirror'

4. sa I ku ra

'cherry blossom

ka I ga 1

mi I - g a

sa | ku ra -qa

Note that all four patterns are seen as a single class. We treat these as having underlying forms as in ( 3 2 )

(32)

Narada Underlying 1. 2. 3. 4.

forms -ga

ka 4- bu to ko ko 4 ro ka ga mi 4· sa ku r a

We treat those with clitics first. Let us assume that the first syllable is extrametrical, that Narada is 'post-accenting1 and therefore the PCM after the first syllable is included within the extrametrical marking for Level Ο accent. We assume [4·]# and End Rule applying on right end with Level 0 derivation yielding forms in (33) . Here we assume Η insertion before Level 1 accent.

(33) '

a.

(ka±)

buto [4-]

b.

(ko) ko 4- ro

\ ga

c.

(ka)

4- g a [ 4 ]

d.

(sa) ku r a

era mi

[4·]

qa [4·]

Level 1 accent will now apply to these forms. In (33a) End Rule will choose 4^ over # resulting in forms shown in (34) .

(34) a. ka \ bu

to ga

b. ko

ro

ko

ga

c. ka \ ga mi \ ga

d. sa

I ku ra

ga

One problem remains with the pitch on ka 4 garni 4· without clitic — here we assume that if 4- [4·] comes about it

reduces immediately to +_, and thus the derivation

for ka | garni will be as for sa "| Jcura - ga.

314

Martha Wright

Note that Zubizaretta's tree analysis would have difficulties with Narada, since it assumes a single tree construction level. If all first syllables are extrametrical, then they should receive identical treatment, which they do not if there is only one level; if polarized trees are involved, then there is no single branchingness that will give correct results.

5.

Conclusions

In this paper we have argued from a variety of languages for a two-level 'accent 1 system parallel to the Grid level approach for stress. All languages analyzed can be subsumed under the level accent system. Objections which have been raised from the tonal contingent against accentual analyses are now seen to be relevant only to *-analyses rather than accent analyses per se.

1. In

the present approach there is no complex melody insertion associated with *; there is no interaction with tonal rules since the single tonal rule required is a late Η-doubling rule for Tonga. 2 . A number of arguments raised against accentual analyses now seem to lack validity; for example, cf. Pulleyblank 1 s (1984) arguments against lack of accent subordination in stress languages. We see stress subordination only in a subset of stress languages, i.e.

those with

two X levels, an alternating Grid level and a word-level X-ing. Large numbers of stress languages have only End Rule involved as we see here with pitch accent Level 1 End Rule. 3. Finally, efforts by tonal analyses to replace accent systems have resulted in many cases in apparent simplification of tonal assignment to be followed by a trade-off of vastly increased complexity of tonal derivation. In the present system there is no increase in complexity in tonal derivation; there is no tonal derivation. There is End Rule and there is HAssignment.

Notes

*The research presented in this paper was supported in part by a Fulbright research grant on Autosegmental Phonology for work at Tilburg University, Netherlands, 1984-1985. Parts of the paper were presented at a colloquium at the University of Illinois, Spring 1984 and a phonology workshop at Tilburg, 1985. 1.

The reason why we use 'foot' is to stress similarities to the single 'foot' assignment for Classical Arabic familiar from Hayes (1980). There is no evidence that I know of to suggest a 'binding' effect for Luganda as we argue for Tonga.

Pitch Accent Languages

315

2.

Forms are from Fall 1980 USC Field Methods course on Luganda taught by Larry Hyman.

3.

The two types not discussed here are the Perfect Tense type, and various weak forms. Both of these show a long span of H ' s typical of Level 0. For the Perfect Tense type we follow Carter (1971, 1972) in treating nominals as having a #[·)-] insertion and argue for the same insertion for Perfect and other forms. For the weak forms, we argue for cyclic H assignment, resulting, in certain cases, in two pitch spans. Downstep marks the boundary of H coming from two cycles in these forms.

4.

The arguments for the superiority of a Level 0 analysis for Tonga over *-analyses involve curcially single consonant roots. Cf. Wright (1985) for discussion.

5.

We will follow Carter's Schemas for the concatenation of morphemes. DVP = subject agreement marker. She leaves out object agreement markers in her schema — here they are given in this form. They always occur adjacent to the Root.

6.

In Tonga, taking ta[ or ka[ as End allows scope possibilities for these morphemes. What the ramifications of this are are unclear — we simply here stipulate how End is figured in the phonological rule.

References Abe, Yasuaki (1981) , Tone Harmony Schema: A Theory of Pitch Accent. In: Deirdre Wheeler and Wynn Chao ( e d s . ) , University of Massachusetts Occasional Papers, 1-26. Archangel!, Diana and Douglas Pulleyblank (1984) , Extratonality and Japanese Accent. MIT and McGill. Carter, Hazel ( 1 9 6 2 ) , Notes on the Tonal System of Northern Rhodesian Plateau Tonga. Her Majesty's Stationery Office, London. Carter, Hazel ( 1 9 7 1 ) , Morphotonology of Zambian Tonga: Some Developments of Meeussen's System — I. African Language Studies 12, 1-30. Carter, Hazel ( 1 9 7 2 ) , Morphotonology of Zambian Tonga: Some Developments of Meeussen's System — II. African Language Studies 14, 36-52. Clark, Mary (1978), A Dynamic Treatment of Tone with Special Attention to the Tonal System of Igbo. Ph.D. dissertation, UMass. Clark, Mary ( 1 9 8 4 ) , Japanese as a Tone Language. Ms., University of New Hampshire. Clements, G . N . and John Goldsmith (eds.) ( 1 9 8 4 ) , Autosegmental Studies in Bantu Tone. Dordrecht: Foris Publications. Dwyer, David ( 1 9 8 3 ) , The Analysis of Tone in Acholi, Luo and Lango. In: Ivan R. Dihoff ( e d . ) , Current Approaches to African Linguistics, (vol. I ) . Dordrecht: Foris Publications. Goldsmith, John (1984) , Tone and Accent in Tonga. In: Clements and Goldsmith (1984), 19-52. Halle, Morris and J.R. Vergnaud ( 1 9 8 2 ) , On the Framework of Autosegmental Phonology. In: Harry van der Hülst and Norval Smith ( e d s . ) , The Structure of Phonological Representations, vol. I, Dordrecht: Foris Publications. Haraguchi, Shosuke ( 1 9 7 7 ) , The Tone Pattern of Japanese: an Autosegmental Theory of Tone. Tokyo: Kaitakusha. Hayes, Bruce ( 1 9 8 0 ) , A Metrical Theory of Stress Rules. Ph.D. dissertation, MIT.

316

Martha Wright

Heny, Frank ( 1 9 7 1 ) , Explanatory tone assignment rules in Bantu. In: C.W. Kim and H. Stahlke ( e d s . ) , Papers in African Linguistics. Edmonton, Alberta: Linguistic Research, Inc., 175-199. Hyman, Larry (1982a), Globality and the Accentual Analysis of Luganda Tone. Journal of Linguistic Research 2 . 3 , 1-40. Hyman, Larry (1982b), Against Asterisks, in Luganda and Bantu Tonologyt A Reply to Hyman (19'82a). Ms., USC. Hyman, Larry (1984), Word Domains and Downstep in Bamileke-Dschang. Ms., USC. McCawley, James (1968) , The Phonological Component of a Grammar of Japanese*. The Hague: Mouton. McCawley, James ( 1 9 7 3 ) , Some Tonga Tone Rules. In: Stephen R. Anderson and Paul Kiparsky ( e d s . ) , A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston. Meeussen, A.E. (1963), Morphotonology of the Tonga Verb. Journal of African Languages 2 , 72-92. Prince, Alan (1983), Relating to the Grid. Linguistic Inguiry 14.1, 19-100. Pulleyblank, Douglas (1983), Tone in Lexical Phonology. Ph.D. dissertation, MIT. Pulleyblank, Douglas (1984), Autosegmental Accent. WELS 14, 343-349. Wright, Martha ( 1 9 8 3 ) , A Metrical Approach to Tone Sandhi in Chinese Dialects. Ph.D. dissertation, University of Massachusetts. Wright, Martha (1985), Tonga: A Mixed Level Pitch Accent Language. Ms., Syracuse University. Zubizaretta, Maria Luisa (1982) , The Formal Interaction of Harmony and Accent: The Tone Pattern of Japanese. In: Harry van der Hülst and Norval Smith, ( e d s . ) , The Structure of Phonological Representations. Volume II. Dordrecht: Foris Publications.