Introducing Phonology

Introducing PhonologyDesigned for students with only a basic knowledge of linguistics thisleading textbook provides a clear and practical introduction to phonologythe study of sound patterns in language It teaches in a step-by-stepfashion the logical techniques of phonological analysis and thefundamental theories that underpin it This thoroughly revised andupdated edition teaches students how to analyze phonological data howto think critically about data how to formulate rules and hypothesesand how to test themNew to this edition

bull Improved examples over 60 exercises and 14 new problem setsfrom a wide variety of languages encourage students to practicetheir own analysis of phonological processes and patterns

bull A new and updated reference list of phonetic symbols and an updatedtranscription system making data more accessible to students

bull Additional online material includes pedagogical suggestions andpassword-protected answer keys for instructors

david odden is Professor Emeritus in Linguistics at Ohio StateUniversity

Cambridge Introductions to Language and LinguisticsThis new textbook series provides students and their teachers with accessible introductions to the majorsubjects encountered within the study of language and linguistics Assuming no prior knowledge of thesubject each book is written and designed for ease of use in the classroom or seminar and is ideal foradoption on a modular course as the core recommended textbook Each book offers the ideal introductorymaterial for each subject presenting students with an overview of the main topics encountered in theircourse and features a glossary of useful terms chapter previews and summaries suggestions for furtherreading and helpful exercises Each book is accompanied by a supporting website

Books published in the seriesIntroducing Phonology David OddenIntroducing Speech and Language Processing John ColemanIntroducing Phonetic Science Michael Ashby and John MaidmentIntroducing Second Language Acquisition second edition Muriel Saville-TroikeIntroducing English Linguistics Charles F MeyerIntroducing Morphology Rochelle LieberIntroducing Semantics Nick RiemerIntroducing Language Typology Edith A MoravcsikIntroducing Psycholinguistics Paul WarrenIntroducing Phonology second edition David Odden

IntroducingPhonologySecond Edition

DAVID ODDEN

University Printing House Cambridge CB2 8BS United Kingdom

Published in the United States of America by Cambridge University Press New York

Cambridge University Press is part of the University of Cambridge

It furthers the Universityrsquos mission by disseminating knowledge in the pursuit ofeducation learning and research at the highest international levels of excellence

wwwcambridgeorgInformation on this title wwwcambridgeorg9781107627970

copy David Odden 2005 2013

This publication is in copyright Subject to statutory exceptionand to the provisions of relevant collective licensing agreementsno reproduction of any part may take place without the writtenpermission of Cambridge University Press

First published 2005Second edition 2013

Printed and bound in the United Kingdom by TJ International Ltd Padstow Cornwall

A catalogue record for this publication is available from the British Library

Library of Congress Cataloguing in Publication dataOdden David Arnold 1954ndashIntroducing phonology David Odden ndash Second Edition

pages cm ndash (Cambridge introductions to language and linguistics)ISBN 978-1-107-03144-9 (Hardback) ndash ISBN 978-1-107-62797-0 (Paperback)1 Grammar Comparative and generalndashPhonology I TitleP217O3 2013414ndashdc23 2013017430

ISBN 978-1-107-03144-9 HardbackISBN 978-1-107-62797-0 Paperback

Additional resources for this publication at wwwcambridgeorgodden

Cambridge University Press has no responsibility for the persistence or accuracy ofURLs for external or third-party internet websites referred to in this publicationand does not guarantee that any content on such websites is or will remainaccurate or appropriate

ContentsAbout this book page vii

Acknowledgments viii

A note on languages ix

List of abbreviations xii

1 What is phonology 1

11 Phonetics ndash the manifestation of language sound 2

12 Phonology the symbolic perspective on sound 5

Summary 11

Exercises 12

Suggestions for further reading 13

2 Allophonic relations 15

21 English consonantal allophones 16

22 Allophony in other languages 22

Summary 33

Exercises 34

3 Feature theory 39

31 Scientific questions about speech sounds 40

32 Distinctive feature theory 45

33 Features and classes of segments 61

34 Possible phonemes and rules ndash an answer 64

35 The formulation of phonological rules 67

36 Changing the theory 71

Summary 77

Exercises 77

4 Underlying representations 79

41 The importance of correct underlying forms 80

42 Refining the concept of underlying form 83

43 Finding the underlying form 85

44 Practice at problem solving 93

45 Underlying forms and sentence-level phonology 95

46 Underlying forms and multiple columns in the paradigm 98

Summary 107

Exercises 107

5 Interacting processes 115

51 Separating the effects of different rules 116

52 Different effects of rule ordering 128

Summary 139

Exercises 139

6 Doing an analysis 147

61 Yawelmani 148

62 Hehe 155

63 Fore 163

64 Modern Hebrew 170

65 Japanese 176

Summary 185

Exercises 186

7 Phonological typology and naturalness 205

71 Inventories 206

72 Segmental processes 208

73 Prosodically based processes 224

74 Why do things happen 230

Summary 235

8 Abstractness and psychological reality 237

81 Why limit abstractness 238

82 Independent evidence historical restructuring 254

83 Well-motivated abstractness 257

84 Grammar-external evidence for abstractness 267

85 How abstract is phonology 278

Exercises 279

9 Nonlinear representations 285

91 The autosegmental theory of tone the beginnings

of a change 286

92 Extension to the segmental domain 306

93 Suprasegmental structure 313

Summary 319

Exercises 319

Glossary 323

References 329

Index of languages 335

General index 337

vi Contents

About this bookThis is an introductory textbook on phono-logical analysis and does not assume anyprior exposure to phonological concepts Thecore of the book is intended to be used in afirst course in phonology and the chapterswhich focus specifically on analysis can easilybe covered during a ten-week quarter Insofaras it is a textbook in phonology it is not atextbook in phonetics and it presupposesan elementary knowledge of transcriptionalsymbolsThe main emphasis of this book is developing

the foundational skills needed to analyzephonological data especially systems of phono-logical alternations For this reason there issignificantly less emphasis on presenting thevarious theoretical positions which phonolo-

gists have taken over the years Theory cannotbe entirely avoided indeed it is impossible tostate generalizations about a particular lan-guage without a theory which gives you a basisfor postulating general rules The very questionof what the raw data are must be interpreted inthe context of a theory thus analysis needstheory Equally theories are formal modelswhich impose structure on data ndash theories aretheories about data ndash so theories need datahence analysis The theoretical issues that arediscussed herein are chosen because they rep-resent issues which have come up many timesin phonology because they are fundamentalissues and especially because they allowexploration of the deeper philosophical issuesinvolved in theory construction and testing

AcknowledgmentsA number of colleagues have read and com-mented on versions of this book I wouldlike to thank Lee Bickmore Patrik Bye ChetCreider Lisa Dobrin Kathleen Currie HallSharon Hargus Tsan Huang Beth Hume KeithJohnson Ellen Kaisse Susannah Levi MarcelinoLiphola Mary Paster Charles Reiss RichardWright and especially Mary Bradshaw for theirvaluable comments on earlier drafts I wouldalso like to thank students at the University ofWestern Ontario University of WashingtonUniversity of Tromsoslash Ohio State UniversityKyungpook National University Concordia Uni-versity and the 2003 LSA Summer Institute atMSU for serving as a practical sounding boardfor this book Numerous colleagues have pro-vided valuable input leading to the revised ver-sion of this book and I regret being unable tothank them all personallyData from my own field notes provide the

basis for a number of the examples andI would like to thank my many language con-sultants for the data which they have providedme including Tamwakat Gofwen (Angas)Bassey Irele (Efik) Edward Amo (Gatilde) Jean-PaulLamah (Guerze) John Mtenge and the lateMargaret Fivawo (Hehe) Beatrice Mulala(Kamba) Oben Ako (Kenyang) Deo Tungaraza(Kerewe)MatthewKirui (Kipsigis) Habi (Kotoko)

Rose Kamwesa (Llogoori) Emmanuel Manday(Matuumbi) Patrick Bamwine (Nkore) DavidMndolwa (Shambaa) Kokerai Rugara (Shona)Udin Saud (Sundanese) Nawang Nornang(Tibetan) and Christopher Oruma (Urhobo)

I would like to thank a number of professionalcolleagues for providing or otherwise helpingmewith data used in this book including CharlesMarfo (Akan) Grover Hudson (Amharic) BertVaux (Armenian) David Payne (AxinincaCampa) Hamza Al-Mozainy (Bedouin HijaziArabic) NasiombeMutonyi (Bukusu) the late IlseLehiste (Estonian) Anders Holmberg (Finnish)Georgios Tserdanelis (Modern Greek) Lou Hohu-lin (Keley-i) YoungheeChungNojuKimMiraOhand Misun Seo (Korean) Chacha NyaigottiChacha (Kuria) Martin Haspelmath (Lezgian)Marcelino Liphola (Makonde) Karin Michelson(Mohawk) Ove Lorentz (Norwegian) Berit AnneBals Baal (NorthSaami)NadyaVinokurova (SakhaYakut) Wayles Browne Svetlana Godjevac andAndrea Sims (Serbo-Croatian) and Rose Aziza(Urhobo) all ofwhomareblameless for anymisuseI have made of their languages and data

Finally I would like to acknowledge my debtto authors of various source books in particu-lar Whitley 1978 Halle and Clements 1983Pickett 2002 and especially Kenstowicz andKisseberth 1979

A note on languagesThe languages which provided data for thisbook are listed below The name of the lan-guage is given followed by the genetic affili-ation and location of the language finally thesource of the data (ldquoFNrdquo indicates that the datacome from my own field notes) Genetic affili-ation typically gives the lowest level of the lan-guage tree which is likely to be widely knownso Bantu languages will be cited as ldquoBanturdquoand Tiv will be cited as ldquoBenue-Congordquo eventhough ldquoBanturdquo is a part of Benue-Congo andldquoTivrdquo is a specific language in the Tivoid groupof the Southern languages in Bantoid Locationswill generally list one country but sometimesmore since language boundaries rarely respectnational boundaries it is to be understood thatthe listed country (or countries) is the primarylocation where the language is spoken espe-cially the particular dialect used or this maybe the country the language historically origin-ates from (the Yiddish-speaking population ofthe US appears to be larger than that of any onecountry in Eastern Europe due to recent popu-lation movements)

Akan [Volta-Congo Ghana] Dolphyne 1988Charles Marfo pc

Amharic [Semitic Ethiopia] Whitley 1978Grover Hudson pc

Angas [Chadic Nigeria] FNArabela [Zaparoan Peru] Rich 1963Aramaic (Azerbaijani) [Semitic Azerbaijan]Hoberman 1988

Araucanian [Araucanian Argentina Chile]Echeverriacutea and Contreras 1965 Hayes 1995

Armenian [Indo-European Armenia IranTurkey] Vaux 1998 and pc

Axininca Campa [Arawakan Peru] Payne 1981and pc

Bedouin Hijazi Arabic [Semitic Saudi Arabia]Al-Mozainy 1981 and pc

Bukusu [Bantu Kenya] Nasiombe Mutonyi pcCairene Arabic [Semitic Egypt] Broselow 1979Catalan [Romance Spain] Lleo 1970Kenstowicz and Kisseberth 1979 Wheeler1979 Hualde 1992

Chamorro [Austronesian Guam] Topping1968 Topping and Dungca 1973 Kenstowiczand Kisseberth 1979 Chung 1983

Chukchi [Chukotko-Kamchatkan Russia]Krauss 1981

Digo [Bantu Kenya and Tanzania] Kisseberth1984

Efik [Benue-Congo Nigeria] FNEstonian [Uralic Estonia] Ilse Lehiste pcSaagpakk 1992

Evenki [Tungusic Russia] Konstantinova 1964Nedjalkov 1997 Bulatova and Grenoble 1999

Ewe (Anlo) [Volta-Congo Benin] Clements1978

Farsi [Indo-European Iran] Obolensky Panahand Nouri 1963

Finnish [Uralic Finland Russia] Whitney 1956Lehtinen 1963 Anders Holmberg pc

Fore [Papuan Papua New Guinea] Pickett 2002Fula [West Atlantic West Africa] Paradis 1992Gatilde [Volta-Congo Ghana] FN in collaborationwith Mary Paster

Gen [Kwa Togo] FNGreek [Indo-European Greece] GeorgiosTserdanelis pc

Guerze (Kpelle) [Mande Guinea] FNHebrew [Semitic Israel] Kenstowicz andKisseberth 1979

Hehe [Bantu Tanzania] FN in collaborationwith Mary Odden

Holoholo [Bantu Congo] Coupez 1955Hungarian [Uralic Hungary] Vago 1980

Kenesei Vago and Fenyvesi 1998Isthmus Zapotec [Oto-Manguean Mexico]Pickett 2002

Japanese [ Japanese Japan] Martin 1975

Jita [Bantu Tanzania] Downing 1996Kamba [Bantu Kenya] FN in collaboration withRuth Roberts-Kohno

Karok [Hokan USA] Bright 1957 Kenstowiczand Kisseberth 1979

Keley-i [Austronesian Philippines] Kenstowiczand Kisseberth 1979 Lou Hohulin pc

Kenyang [Bantu Cameroon] FNKera [Chadic Chad] Ebert 1975 Kenstowiczand Kisseberth 1979

Kerewe [Bantu Tanzania] FNKikuyu [Bantu Kenya] Clements 1984Kipsigis [Nilotic Kenya] FNKlamath [Penutian USA] Barker 1963 1964Koasati [Muskogean Louisiana] Kimball 1991Kolami [Dravidian India] Emeneau 1961Korean [Korean Korea] Martin 1992 YoungheeChung Noju Kim Mira Oh andMisun Seo pc

Koromfe [Gur Bourkina Fasso] Rennison 1997Kotoko [Chadic Cameroon] FNKuria [Bantu Kenya] FNLamba [Bantu Zambia] Doke 1938 Kenstowiczand Kisseberth 1979

Lardil [Pama-Nyungan Australia] Klokeid1976

Latin [Indo-European Italy] Allen andGreenough 1983 Hale and Buck 1966

Lezgian [Northeast Caucasian Dagestan andAzerbaijan] Haspelmath 1993 and pc

Lithuanian [Indo-European Lithuania]Dambriunas Klimas and Schmalstieg 1966Ambrazas 1997 Kenstowicz 1972aMathiassen 1996

Llogoori [Bantu Kenya] FN in collaborationwith Michael Marlo

Luganda [Bantu Uganda] Cole 1967 Snoxall1967

Lulubo [Nilo-Saharan Sudan] Andersen 1987Makonde [Bantu Mozambique] MarcelinoLiphola pc

Maltese [Semitic Malta] Aquilina 1965 Borgand Azzopardi-Alexandre 1997 Brame 1972Hume 1996

Manipuri [Sino-Tibetan India MyanmarBangaladesh] Bhat and Ningomba 1997

Maranungku [Australian Australia] Tryon1970 Hayes 1995

Margyi [Chadic Nigeria] Hoffmann 1963

Matuumbi [Bantu Tanzania] FNMbunga [Bantu Tanzania] FNMende [Mande Liberia Sierra Leone] Leben1978

Mixtec [Mixtecan Mexico] Pike 1948Goldsmith 1990a

Mixteco [Oto-Manguean Mexico] Pickett 2002Mohawk [Hokan USA] Postal 1968 Beatty1974 Michelson 1988 and pc

Mongo [Bantu Congo] Hulstaert 1961Mongolian [Altaic Mongolia] Hangin 1968Nkore [Bantu Uganda] FN in collaborationwith Robert Poletto

Norwegian [Germanic Norway] Ove Lorentz pcOsage [Siouan Oklahoma] Gleason 1955Ossetic [Indo-European Georgia Russia] Abaev1964 Whitley 1978

Palauan [Austronesian Palau] Josephs 1975Flora 1974

Polish [Slavic Poland] Kenstowicz andKisseberth 1979

Quechua (Cuzco) [Quechua Peru] Bills Vallejoand Troike 1969 Cusihuamaacuten 1976

Saami [Uralic Saacutepmi (Norway SwedenFinland Russia)] FN in collaboration withCurt Rice and Berit Anne Bals Baal

Sakha (Yakhut) [Altaic Russia] Krueger 1962Nadezhda Vinokurova pc

Samoan [Austronesian Samoa] Milner 1966Serbo-Croatian [Slavic Yugoslavia] Kenstowiczand Kisseberth 1979 Wayles BrowneSvetlana Godjevac and Andrea Sims pc

Shambaa [Bantu Tanzania] FNShona [Bantu Zimbabwe] FNSlave [Athapaskan Canada] Rice 1989Slovak [Slavic Slovakia] Kenstowicz 1972bRubach 1993

Somali [Cushitic Somalia] Andrzejewski 1964Kenstowicz 1994 Saeed 1993 1999

Sundanese [Austronesian Indonesia] FNSwati [Bantu Swaziland] FNSyrian Arabic [Semitic Syria] Cowell 1964Tera [Chadic Nigeria] Newman 1968Thai [Daic Thailand] Halle and Clements1983

Tibetan [Sino-Tibetan Tibet] FNTiv [Benue-Congo Nigeria] Arnott 1964Goldsmith 1976

x A note on languages

Tohono Orsquoodham (Papago) [Uto-Aztecan USA]Saxton 1963 Saxton and Saxton 1969Whitley 1978

Tonkawa [Coahuiltecan USA] Hoijer 1933Tswana [Bantu Botswana] Cole 1955 SnymanShole and Le Roux 1990

Turkish [Altaic Turkey] Lees 1961 Foster 1969Halle and Clements 1983

Ukrainian (Sadžava Standard) [Slavic Ukraine]Carlton 1971 Kenstowicz and Kisseberth1979 Press and Pugh 1994 (Standard)Popova 1972 (Sadžava)

Urhobo [Edoid Nigeria] Aziza 2008 and pc FNVata [Kru Cocircte drsquoIvoire] Kaye 1982Votic [Uralic Russia] Ariste 1968

Warao [Warao Venezuela] Osborn 1966 Hayes1995

Weri [Goilalan New Guinea] Boxwell andBoxwell 1966 Hayes 1995

Wintu [Penutian USA] Pitkin 1984Woleaian [Austronesia Micronesia] Sohn 1975Xavante [ Jeacute Brazil] Pickett 2002Yawelmani [Penutian USA] Newman 1944Kenstowicz and Kisseberth 1979

Yekhee (Etsako) [Edoid Nigeria] Elimelech1978

Yiddish [Germanic Eastern Europe] Neil Jacobspc

Yoruba [Kwa Nigeria] Akinlabi 1984Zoque [Mixe-Zoquean Mexico] Pickett 2002

A note on languages xi

Abbreviationsabl ablativeacc accusativeant anteriorATR advanced tongue rootbk backcg constricted glottiscl classcons consonantalcont continuantcor coronaldat dativedB decibeldelrel delayed releasedim diminutivedistr distributedeo each otherfem femininegen genitivehi highHz Hertzimp imperativeintr intransitivelat laterallo lowloc locative

masc masculinems(c) millisecondnas nasalneut neuternom nominativeobj objectpl pluralposs possessivepres presentrd roundsg sing singularsg spread glottisson sonorantsp speciesstrid stridentsyl syllabictns tensetr transitivevcd voicedvcls voicelessvoi voice1 first person2 second person3 third person

CHAPTER

1 What isphonology

PREVIEW

This chapter introduces phonology the study of the sound

systems of language Its key objective is to

u explain the difference between physical sound and

ldquoa soundrdquo as a discrete element of language

u highlight the tradeoff between accuracy and usefulness in

representing sound

u introduce the notion of ldquosound as cognitive symbolrdquo

u present the phonetic underpinnings of phonology

u introduce the notion of phonological rule

KEY TERMSsound

symbol

transcription

grammar

continuousnature ofspeech

Phonology is one of the core fields that compose the discipline of linguis-tics which is the scientific study of language structure One way tounderstand the subject matter of phonology is to contrast it with otherfields within linguistics A very brief explanation is that phonology is thestudy of sound structure in language which is different from the studyof sentence structure (syntax) word structure (morphology) or how lan-guages change over time (historical linguistics) But this is insufficient Animportant feature of the structure of a sentence is how it is pronounced ndash

its sound structure The pronunciation of a given word is also a funda-mental part of the structure of the word And certainly the principles ofpronunciation in a language are subject to change over time So phon-ology has a relationship to numerous domains of linguisticsAn important question is how phonology differs from the closely

related discipline of phonetics Making a principled separation betweenphonetics and phonology is difficult ndash just as it is difficult to make aprincipled separation between physics and chemistry or sociology andanthropology While phonetics and phonology both deal with languagesound they address different aspects of sound Phonetics deals withldquoactualrdquo physical sounds as they are manifested in human speech andconcentrates on acoustic waveforms formant values measurements ofduration measured in milliseconds of amplitude and frequency Phonet-ics also deals with the physical principles underlying the production ofsounds namely vocal tract resonances and the muscles and otherarticulatory structures used to produce those resonances Phonology onthe other hand is an abstract cognitive system dealing with rules in amental grammar principles of subconscious ldquothoughtrdquo as they relate tolanguage soundYet once we look into the central questions of phonology in greater

depth we will find that the boundaries between the disciplines of phon-etics and phonology are not entirely clear-cut As research in both of thesefields has progressed it has become apparent that a better understandingof many issues in phonology requires that you bring phonetics intoconsideration just as a phonological analysis is a prerequisite for phoneticstudy of language

11 Phonetics ndash the manifestationof language sound

Ashby and Maidment (2005) provide a detailed introduction to the subjectarea of phonetics which you should read for greater detail on the acousticand articulatory properties of language sounds and transcription usingthe International Phonetic Alphabet (IPA) This section provides a basicoverview of phonetics to clarify what phonology is aboutFrom the phonetic perspective ldquosoundrdquo refers to mechanical pressure

waves and the sensations arising when such a pressure wave strikes yourear In a physical sound the wave changes continuously and can be

2 INTRODUCING PHONOLOGY

graphed as a waveform showing the amplitude on the vertical axis andtime on the horizontal axis Figure 1 displays the waveform of a pronunci-ation of the word wall with an expanded view of the details of thewaveform at the center of the vowel between w and llFigure 2 provides an analogous waveform of a pronunciation of the

word lsquowillrsquo which differs from wall just in the choice of the vowelInspection of the expanded view of the vowel part of these waveforms

shows differences in the overall shape of the time-varying waveformswhich is what makes these words sound differentIt is difficult to characterize those physical differences from the wave-

form but an analytical tool of phonetics the spectrogram provides a

Amp l i t u d e

FIGURE 1

Amp l i t u d e

FIGURE 2

What is phonology 3

useful way to describe the differences by reducing the absolute amplitudeproperties of a wave at an exact time to a set of (less precise) amplitudecharacteristics in different frequency and time areas In a spectrogramthe vertical axis represents frequency in Hertz (Hz) and darkness repre-sents amplitude Comparing the spectrograms of wall and will in figure 3you can see that there are especially dark bands in the lower part of thespectrogram and the frequency at which these bands occur ndash known asformants ndash is essential to physically distinguishing the vowels of thesetwo words Formants are numbered from the bottom up so the firstformant is at the very bottomIn wall the first two formants are very close together and occur at 634

Hz and 895 Hz whereas in will they are far apart occurring at 464 Hz and1766 Hz The underlying reason for the difference in these sound qualitiesis that the tongue is in a different position during the articulation of thesetwo vowels In the case of the vowel of wall the tongue is relatively lowand retracted and in the case of will the tongue is relatively fronted andraised These differences in the shape of the vocal tract result in differentphysical sounds coming out of the mouthThe physical sound of a wordrsquos pronunciation is highly variable as we

see when we compare the spectrograms of three pronunciations of wall infigure 4 the three spectrograms are obviously differentThe first two pronunciations are produced at different times by the

same speaker differing slightly in where the first two formants occur(634 Hz and 895 Hz for the first token versus 647 Hz and 873 Hz forthe second) and in numerous other ways such as the greater ampli-tude of the lower formants in the first token In the third tokenproduced by a second (male) speaker of the same dialect the first twoformants are noticeably lower and closer together occurring at 541 Hzand 617 Hz

Frequency

1080 msc 0

willFIGURE 3

5000Frequency 1425 msc0

FIGURE 4

Physical variation in sound also arises because of differences in sur-rounding context Figure 5 gives spectrograms of the words wall tall andlawn with grid lines to identify the portion of each spectrogram in themiddle which corresponds to the vowelIn wall the frequency of the first two formants rapidly rises at the

beginning and falls at the end in tall the formant frequencies start higherand fall slowly in lawn the formants rise slowly and do not fall at the endA further important fact about physical sound is that it is continuous sowhile wall tall and lawn are composed of three sounds where the middlesound in each word is the same one there are no actual physical bound-aries between the vowel and the surrounding consonantsThe tools of phonetic analysis can provide very detailed and precise

information about the amplitude frequency and time characteristics ofan utterance ndash a typical spectrogram of a single-syllable word in Englishcould contain around 100000 bits of information The problem is thatthis is too much information ndash a lot of information needs to be discardedto get at something more general and useful

12 Phonology the symbolic perspectiveon sound

Physical sound is too variable and contains too much information to allowus to make meaningful and general statements about the grammarof language sound We require a way to represent just the essentials oflanguage sounds as mental objects which grammars can manipulateA phonological representation of an utterance reduces this great massof phonetic information to a cognitive minimum namely a sequence ofdiscrete segments

121 Symbolic representation of segmentsThe basic tool for converting the continuous stream of speech sound intodiscrete units is the phonetic transcription The idea behind a transcrip-tion is that the variability and continuity of speech can be reduced tosequences of abstract symbols whose interpretation is predefined asymbol standing for all of the concrete variants of the sound Phonologythen is the study of higher-level patterns of language sound conceived in

1370 msc0Time

5000Frequency

wall tall lawn FIGURE 5

What is phonology 5

terms of discrete mental symbols whereas phonetics is the study of howthose mental symbols are manifested as continuous muscular contrac-tions and acoustic waveforms or how such waveforms are perceived asthe discrete symbols that the grammar acts onThe idea of reducing an information-rich structure such as an acous-

tic waveform to a small repertoire of discrete symbols is based on avery important assumption one which has proven to have immeasur-able utility in phonological research namely that there are systematiclimits on possible speech sounds in human language At a practicallevel this assumption is embodied in systems of symbols and associatedphonetic properties such as the International Phonetic Alphabet offigure 6 Ashby and Maidment (2005) give an extensive introductionto phonetic properties and corresponding IPA symbols which youshould consult for more information on phonetic characteristics oflanguage soundThe IPA chart is arranged to suit the needs of phonetic analysis Stand-

ard phonological terminology and classification differ somewhat fromthis usage Phonetic terminology describes [p] as a ldquoplosiverdquo where thatsound is phonologically termed a ldquostoprdquo the vowel [i] is called a ldquocloserdquovowel in phonetics but a ldquohighrdquo vowel in phonology Figure 7 gives theimportant IPA vowel letters with their phonological descriptions whichare used to stand for the mental symbols of phonological analysisThe three most important properties for defining vowels are height

backness and roundness The height of a vowel refers to the fact that thetongue is higher when producing [i] than it is when producing [e] (which ishigher than when producing [aelig]) and the same holds for the relationbetween [u] [o] and [a]Three primary heights are generally recognized namely highmid and

low augmented with the secondary distinction tenselax for nonlowvowels which distinguishes vowel pairs such as [i] (seed) vs [ɪ] (Sid) [e] (late)vs [ε] (let) or [u] (food) vs [ʊ] (foot) where [i e u] are tense and [ɪ ε ʊ] arelax Tense vowels are higher and articulated further from the center of thevocal tract compared to their lax counterparts It is not clear whether thetenselax distinction extends to low vowelsIndependent of height vowels can differ in relative frontness of the

tongue The vowel [i] is produced with a front tongue position whereas [u]is produced with a back tongue position In addition [u] is produced withrounding of the lips it is common but by no means universal for backvowels to also be produced with lip rounding Three phonetic degrees ofhorizontal tongue positioning are generally recognized front centraland back Finally any vowel can be pronounced with protrusion(rounding) of the lips and thus [o] [u] are rounded vowels whereas [i][aelig] are unrounded vowelsWith these independently controllable phonetic parameters ndash five

degrees of height three degrees of fronting and rounding versusnon-rounding ndash we have the potential for up to thirty vowels which is

FIGURE 6

What is phonology 7

many more vowels than are found in English Many of these vowels arelacking in English but can be found in other languages This yields a fairlysymmetrical system of symbols and articulatory classifications but thereare gaps such as the lack of tenselax distinctions among central highvowelsThe major consonants and their classificatory analysis are given in

figure 8Where the IPA term for consonants like [p b] is ldquoplosiverdquo these are

referred to phonologically as ldquostopsrdquo Lateral and rhotic consonants aretermed ldquoliquidsrdquo and non-lateral ldquoapproximantsrdquo are referred to asldquoglidesrdquo Terminology referring to the symbols for implosives ejectivesdiacritics and suprasegmentals is generally the same in phonological andphonetic usageOther classificatory terminology is used in phonological analysis to

refer to the fact that certain sets of sounds act together for grammaticalpurposes Plain stops and affricates are grouped together by consideringaffricates to be a kind of stop (one with a special fricative-type release)Fricatives and stops commonly act as a group and are termed obstruentswhile glides liquids nasals and vowels likewise act together beingtermed sonorants

122 The concerns of phonologyAs a step towards understanding what phonology is and especially howit differs from phonetics we will consider some specific aspects ofsound structure that would be part of a phonological analysis Thepoint which is most important to appreciate at this moment is that

Nonround

tense i i M M highlax

tense e midlax ε з V

aelig a A lowFront Central Back

tense y u u highlax Utense oslash o midlax œ O

Q low Front Central Back

FIGURE 7

The release ofaffricates will bewritten as asuperscript letteranalogous to IPAconventions fornasal and lateralrelease This makesit clear thataffricates are singlesegments notclusters

the ldquosoundsrdquo which phonology is concerned with are symbolic sounds ndashthey are cognitive abstractions which represent but are not the same asphysical sounds

The sounds of a language One aspect of phonology investigates whatthe ldquosoundsrdquo of a language are We would want to take note in a descrip-tion of the phonology of English that we lack the vowel [oslash] that exists inGerman in words like schoumln lsquobeautifulrsquo a vowel which is also found inFrench (spelled eu as in jeune lsquoyoungrsquo) or Norwegian (oslashl lsquobeerrsquo) Similarlythe consonant [θ] exists in English (spelled th in thing path) as well asIcelandic Modern Greek and North Saami) but not in German or French

Consonant symbols

Consonant manner and voicing

Place ofarticulation

vclsstop

vclsaffricate

vclsfricative

vcdstop

vcdaffricate

vcdfricative nasal

bilabial p (pφ) φ b (bβ) β mlabiodental pf f bv v ɱdental t tθ θ d deth eth nalveolar t ts s d dz z nalveopalatal tʃ ʃ dʒ ʒ ɲretroflex ʈ ʈʂ ʂ ɖ ɖʐ ʐ ɳpalatal c (cccedil) ccedil ɟ ɟʝ ʝ ɲvelar k kx x g gɣ ɣ ŋuvular q qχ χ ɢ ɢʁ ʁ ɴpharyngeal ħ ʕlaryngeal ~glottal

ʔ h ɦ

Glides and liquids

labiovelar palatal labiopalatal velar

Glides w j ɥ ɰ

tap trill glide retroflex uvularflap

Rhotics ɾ r ɹ ɽ ʀ

plain retroflex voiceless voicedfricative fricative

Laterals l ɭ ɬ ɮFIGURE 8

What is phonology 9

and not in Latin American Spanish (but it does occur in ContinentalSpanish in words such as cerveza lsquobeerrsquo)

Sounds in languages are not just isolated atoms they are part of asystem The systems of stops in Hindi and English are given in (1)

The stop systems of these languages differ in three ways English does nothave a series of voiced aspirated stops like Hindi [bh dh ɖh gh] nor does ithave a series of retroflex stops [ʈ ʈh ɖ ɖh] Furthermore the phonologicalstatus of the aspirated sounds [ph th kh] is different in the languages asdiscussed in chapter 2 in that they are basic lexical facts of words inHindi but are the result of applying a rule in English

Rules for combining sounds Another aspect of language sound whicha phonological analysis takes account of is that in any language certaincombinations of sounds are allowed but other combinations are sys-tematically impossible The fact that English has the words [bɹɪk] brick[bɹejk] break [bɹɪdʒ] bridge [bɹɛd] bread is a clear indication that thereis no restriction against having words that begin with the consonantsequence br besides these words one can think of many more wordsbeginning with br such as bribe brow and so on Similarly there aremany words which begin with bl such as [bluw] blue [bleʔnt] blatant[blaeligst] blast [blɛnd] blend [blɪŋk] blink showing that there is no ruleagainst words beginning with bl It is also a fact that there is no word[blɪk]1 in English even though the similar words blink brick do existThe question is why is there no word blick in English The bestexplanation for the nonexistence of this word is simply that it is anaccidental gap ndash not every logically possible combination of soundswhich follows the rules of English phonology is found as an actualword of the languageNative speakers of English have the intuition that while blick is not a

word of English it is a theoretically possible word of English and such aword might easily enter the language for example via the introduction ofa new brand of detergent Sixty years ago the English language did nothave any word pronounced [bɪk] but based on the existence of words likebig and pick that word would certainly have been included in the set ofnonexistent but theoretically allowed words of English ContemporaryEnglish of course actually does have that word ndash spelled Bic ndash which isthe brand name of a ballpoint penWhile the nonexistence of blick in English is accidental the exclusion

from English of many other imaginable but nonexistent words is based on

Hindi stops English stopsp t ʈ k p t kph th ʈh kh ph th kh

b d ɖ g b d gbh dh ɖh gh

1 The asterisk is used to indicate that a given word is nonexistent or wrong

a principled restriction of the language While there are words that beginwith sn like snake snip and snort there are no words beginning with bnand thus bnick bnark bniddle are not words of English There simply areno words in English which begin with bn Moreover native speakersof English have a clear intuition that hypothetical bnick bnark bniddlecould not be words of English Similarly there are no words in Englishwhich are pronounced with pn at the beginning a fact which is not onlydemonstrated by the systematic lack of words such as pnark pnig pnilgebut also by the fact that the word spelled pneumonia which derives fromAncient Greek (a language which does allow such consonant combin-ations) is pronounced [nʌmonjə] without p A description of the phonologyof English would provide a basis for characterizing such restrictions onsequences of sounds

Variations in pronunciation In addition to providing an account ofpossible versus impossible words in a language a phonological analysiswill explain other general patterns in the pronunciation of words Forexample there is a very general rule of English phonology whichdictates that the plural suffix on nouns will be pronounced as [ɨz]represented in spelling as es when the preceding consonant is one ofa certain set of consonants including [ʃ] (spelled sh) as in bushes [tʃ](spelled as ch) as in churches and [dʒ] (spelled j ge dge) as in cagesbridges This pattern of pronunciation is not limited to the plural sodespite the difference in spelling the possessive suffix s2 is also subjectto the same rules of pronunciation thus plural bushes is pronouncedthe same as the possessive bushrsquos and plural churches is pronounced thesame as possessive churchrsquos

This is the sense in which phonology is about the sounds of languageFrom the phonological perspective a ldquosoundrdquo is a specific unit whichcombines with other such specific units and which represents physicalsounds What phonology is concerned with is how sounds behave in agrammar

Summary Phonetics and phonology both study language sound Phonologyexamines language sounds as mental units encapsulated symbolicallyfor example as [aelig] or [g] and focuses on how these units function ingrammars Phonetics examines how symbolic sound is manifested as acontinuous physical phenomenon The conversion from the continu-ous external domain to mental representation requires focusing on theinformation that is important which is possible because not all phys-ical properties of speech sounds are cognitively important One of thegoals of phonology is then to discover exactly what these cognitivelyimportant properties are and how they function in expressing regu-larities about languages

2 This is the ldquoapostrophe srdquo suffix found in the childrsquos shoe meaning lsquothe shoe owned by the childrsquo

What is phonology 11

ExercisesThe first three exercises are intended to be a framework for discussion of thepoints made in this chapter rather than being a test of knowledge and technicalskills

1 Examine the following true statements and decide if each best falls into therealm of phonetics or phonologya The sounds in the word frame change continuouslyb The word frame is composed of four segmentsc Towards the end of the word frame the velum is loweredd The last consonant in the word frame is a bilabial nasal

2 Explain what a ldquosymbolrdquo is how is a symbol different from a letter

3 Why would it be undesirable to use the most precise representation of thephysical properties of a spoken word that can be created under currenttechnology in discussing rules of phonology

The following five questions focus on technical skills

4 How many segments (not letters) are there in the following words (in actualpronunciation)

5 Give the phonetic symbols for the following segmentsvoiced velar fricativevoiceless velarized alveolar affricateinterdental nasalejective uvular stoplow front round vowelback mid unrounded vowellax back high round vowelvoiced palatal fricativesyllabic bilabial nasalvoiced laryngeal fricativevoiceless rounded pharyngeal fricativepalatalized voiceless alveolar stop

6 From the following pairs of symbols select the symbol which matches thearticulatory description

sit judge trap fish bite ball up ox through often

e ɛ front mid lax vowelu u creaky high rounded vowelx χ voiced velar fricativeɪ i lax front high vowelʕ ʔ glottal stopθ tθ dental affricateʒ ʝ alveopalatal fricativej ɥ labio-palatal glide

7 Provide the articulatory description of the following segments Example

8 Name the property shared by each segment in the following sets

ɾ z s ɹ n ɗ d t r ɮ ɬ lɣ x ŋ ɠ g kɑ a ʌ ɪ aelig ɛ ɯ ɤ ɨ ə i eʝ j ɲ ɟ ʎɪ ʊ ɛ ɔu ʉ y

Further readingAshby and Maidment 2005 Isac and Reiss 2008 Johnson 1997 Ladefoged and Johnson 2010 Liberman

1983 Stevens 1998

θ voiceless interdental fricative

ɔ aɱ dʊ yaelig oslashts ʂɟ kx

x ɪbv gw

gɣ ʔ

CHAPTER

2 Allophonicrelations

PREVIEW

This chapter begins the analysis of phonological processes

You will

u learn of predictable variants of basic sounds in English

u learn about the concepts ldquophonemerdquo and ldquoallophonerdquo

u discover that similar relations between sounds exist in

other languages

u begin to learn the general technique for inducing

phonological rules from data that come from a language

which you do not know

u be introduced to writing phonological rules

KEY TERMSallophone

phoneme

complementarydistribution

contrast

distinctive

As explained in the preceding chapter the focus of phonology is themental rules which govern the pronunciation of words in a given lan-guage Certain facts about pronunciation simply cannot be predicted byrule for example that in English the word sick is pronounced [sɪk] and sipis pronounced [sɪp] Hence one fundamental component of a language isa lexicon a list of words (or morphemes ndash parts of words) which mustprovide any information which cannot be predicted by rules of thelanguage However much about the pronunciation of words can bepredicted For example in the word tick the initial voiceless consonantt is phonetically aspirated and is phonetically [thɪk] This aspirationcan be demonstrated visually by dangling a tissue in front of themouth when saying the word notice that when you pronounce t thetissue is blown forward In comparison t in the word stick is notaspirated (thus the tissue is not blown forward) so this word is tran-scribed as [stɪk] This fact can be predicted by rule and we now considerhow this is done

21 English consonantal allophones

While the physical difference between t and th in English is just as real asthe difference between t and d there is a fundamental linguistic differ-ence between these two relationships The selection of t versus d mayconstitute the sole difference between many different words in Englishsuch words where two words are differentiated exclusively by a choicebetween one of two segments are referred to as minimal pairs

The difference between [t] and [d] is contrastive (also termed distinctive)in English since this difference ndash voicing ndash forms the sole basis fordistinguishing different words (and thus [t] and [d] contrast)The choice of a voiceless aspirated stop such as [th] versus a voiceless

unaspirated stop such as [t] on the other hand never defines the sole basisfor differentiating words in English The occurrence of [t] versus [th] (also[k] versus [kh] and [p] versus [ph]) follows a rule that aspirated stops areused in one phonological context and unaspirated stops are used in allother contexts In English [t] and [th] are predictable variants of a singleabstract segment a phoneme which we represent as t Purely predict-able variants are termed allophones ndash the sounds are in complementarydistribution because the context where one variant appears is the com-plement of the context where the other sound appears As we haveemphasized one concern of phonology is determining valid relationsbetween pronounced segments and the abstract mental constructs thatthey derive from the phonemes which represent the unity behind

[d] [t] [d] [t]dire tire do twoDick tick had hatsaid set bend bent

observed [t] and [th] etc The implicit claim is that despite there beingactual differences [t] and [th] (also [k] and [kh] [p] and [ph]) are in afundamental sense ldquothe same thingrdquo We reduce the output sounds [t th

k kh p ph] to just the set of sounds t k p and a rule provides the infor-mation ldquorealized as [t] vs [th]rdquo to account for these regularities

211 AspirationWe will turn our attention to rules of pronunciation in English startingwith aspiration to see what some of these regularities are In the first setof words below the phonemes p t k are aspirated whereas they are notaspirated in the second set of words

The selection of an aspirated versus an unaspirated voiceless stop isdetermined by the context in which the stop appears Aspirated stopsappear at the beginning of a word whereas unaspirated stops appear after[s] aspirated stops appear before a vowel or a sonorant consonantwhereas unaspirated stops appear at the end of a word This collectionof contexts can be expressed succinctly by referring to the position of theconsonant in the syllable aspirated stops appear at the beginning of thesyllable and unaspirated stops appear elsewhereWe assume that the voiceless stops are basically unaspirated in English

and explain where aspirated segments appear by having a rule that assignsaspiration to voiceless stops only when the stop is at the beginning of thesyllable the rule can be stated as ldquovoiceless stops become aspirated at thebeginning of a syllablerdquo We donrsquot need a second special rule to deriveunaspirated stops in other environments because that follows directlyfrom our assumption that the basic or underlying form of the voicelessstops in English is unaspirated and they will therefore be pronounced assuch unless they are specifically changed by a rule We investigate the ideaof underlying representations in greater detail in chapter 4

Aspirated stopspool [phuwl] tooth [thuwθ] coop [khuwp]pit [phɪt] tin [thɪn] kill [khɪl]apply [əphlaj] atomic [əthamɪk] account [əkhaeligwnt]prawn [phɹan] truth [thɹuwθ] crab [khɹaeligb]pueblo [phwεblow] twine [thwajn] quill [khwɪl]play [phlej] clay [khlej]puce [phjuws] cube [khjuwb]

Unaspirated stopsspool [spuwl] stool [stuwl] school [skuwl]spit [spɪt] stick [stɪk] skid [skɪd]sap [saeligp] sat [saeligt] sack [saeligk]spray [spɹej] stray [stɹej] screw [skɹuw]split [splɪt] sclerosis [skləɹowsɪs]spew [spjuw] skew [skjuw]

Allophonic relations 17

Actually the issue of aspiration in English is a bit more complex Noticethat in the following words [p] [t] and [k] in the middle of the word arenot aspirated even though the consonant is between vowels or syllabicsonorants ndash between syllable peaks ndash and therefore is presumably at thebeginning of a syllable

Compare these words with seemingly analogous words where thereis aspiration on the stop between vowels such as [əˈthaeligk] attack[əˈkhjuwmjəˌlejt] accumulate [ˈlejˌthɛks] latex [əˈphɛndɪks] appendix Theimportant difference in these words is the location of stress In all of thewords in (4) where a voiceless consonant is not aspirated in syllable-initialposition the consonant is followed by an unstressed vowel In otherwords these data force us to refine our statement of the rule for assign-ment of aspiration to be ldquovoiceless stops become aspirated at the begin-ning of a stressed syllablerdquo The next chapter introduces the details forformalizing rules but for the present we can express that rule as follows

(5) voiceless stop aspirated [ˈσ __

This statement introduces the method of writing rules which will be usedin the book Rules generally take the form ldquoABC_Drdquo where A C D arevariables that stand for single segments like [l] or [d] or phonetic classes suchas ldquovoiceless stoprdquo and B describes the nature of the change some phoneticparameter such as ldquovoicelessrdquo or ldquonasalrdquo The conditioning context mightinvolve only a preceding element in which case ldquoDrdquo would be missing itmight involve only a following element in which case ldquoCrdquowould bemissingor the applicability of the rule might depend on both what precedesand what follows The arrow means ldquobecomesrdquo the slash means ldquoin theenvironmentrdquo where the context is what follows the slash The notationldquo[ˈσrdquo means ldquobeginning of a stressed syllablerdquo thus ldquovoiceless stops becomeaspirated when they are preceded by the beginning of a stressed syllablerdquoThe final chapter of the book introduces syllables in more detail

Alternations involving aspiration The dependence of aspiration on thelocation of stress leads to discovering further evidence for an aspirationrule Certain word-formation processes in English change the location ofstress for example in atom the stress is on thefirst syllable of the root and inthe related adjective atomic the stress is on the second syllable The pairs ofwords in (6) further illustrate the property of stress shifting where theverbs on the left have stress on the second syllable of the root but the nounsderived from these verbs on the right have no stress on the second syllable

ˈhaeligpij happy ˈkhaeligmpɪŋ camping ˈhɛlpɪŋ helpingˈlʌkij lucky ˈsʌkɹ sucker ˈsaltij salty

[əˈphlaj] apply [ˌaeligpləˈkhejʃn] application[səˈphowz] suppose [ˌsʌpəˈzɪʃn] supposition[əˈkhwajɹ] acquire [ˌaeligkwəˈzɪʃn] acquisition

As predicted by our rule for aspiration the phonetic presence or absenceof aspiration on the medial stop of the root may alternate within a givenroot according to where the stress appears in the rootAnother set of examples involves the word-formation process adding -ee

to a verb to form a noun referring to the direct object of the action Thatsuffix must be stressed unlike the subject-nominalization suffix -er

Again as our rule predicts when the stress shifts to the suffix vowel thepronunciation of the preceding consonant changes to become aspirated

Pronunciation of novel utterances Not only does the existence of thisaspiration rule explain why all voiceless stops are aspirated at the begin-ning of a stressed syllable in English words it also explains facts oflanguage behavior by English speakers outside the domain of pronoun-cing ordinary English words First when English speakers are faced with anew word which they have never heard before for example one comingfrom a foreign language voiceless consonants will be aspirated or unas-pirated according to the general rule for the distribution of aspirationThe pronunciation of unfamiliar foreign place names provides one simpledemonstration The place names Stord (Norway) and Palma (Mozambique)will be pronounced by English speakers as [stɔɹd] and [phalmə] as predictedby the aspiration rule The name Stavanger (Norway) may be pronouncedmany ways ndash [stəˈvaeligŋɹ] [ˈstaeligvəndʒɹ] [stəˈvaeligndʒɹ] [ˈstaeligvəŋɹ] and so on butconsistently throughout this variation the t will remain unaspiratedbecause of its position in the syllable In the English pronunciation ofRapallo (Italy) stress could either be on the first syllable in [ˈɹaeligpəlow]with no aspiration because p is at the beginning of an unstressedsyllable or on the second syllable as in [ɹəˈphalow] ndash again the choice ofaspirated versus unaspirated consonant being determined by the rule ofaspirationSecond when English speakers attempt to learn a language which

does not have the same distribution of aspirated and unaspirated con-sonants as in English they encounter difficulties in pronunciation thatreflect the effect of the rule of aspiration Hindi has both aspirated andunaspirated voiceless stops at the beginning of syllables as well as afters Words such as [phal] lsquofruitrsquo and [stan] lsquobreastrsquo are not difficult forEnglish speakers to pronounce accurate pronunciation of [pal] lsquowantrsquoand [sthal] lsquoplacersquo on the other hand are This is due to the fact thatthe rule of aspiration from English interferes in the pronunciation ofother languages

Verb Subject noun Object noun[ˈgɹaelignt] [ˈgɹaeligntɹ] [ˌgɹaelignˈthi] grant[ˈʃɪft] [ˈʃɪftɹ] [ˌʃɪfˈthi] shift[ˈhɛlp] [ˈhɛlpɹ] [ˌhɛlˈphi] help[ˈtʃhowk] [ˈtʃhowkɹ] [ˌtʃhowˈkhi] choke[ˈstɹajk] [ˈstɹajkɹ] [ˌstɹajˈkhi] strike[əˈthaeligk] [əˈthaeligkɹ] [əˌthaeligˈkhi] attack

Finally even in native English words unaspirated stops can show theeffect of the aspiration rule in hyper-slow syllable-by-syllable pronunci-ation Notice that in the normal pronunciation of happy [ˈhaeligpij] only thefirst syllable is stressed and therefore [p] remains unaspirated However ifthis word is pronounced very slowly drawing out each vowel then bothsyllables become stressed and as predicted the stop p is aspirated ndash

[ˈhaelig] [ˌphij] All of these facts are explained by one simple hypothesisthat in English the occurrence of aspiration on stops derives from apply-ing a rule

212 FlappingWe now turn to another rule A phonetic characteristic of many NorthAmerican dialects of English is ldquoflappingrdquo where t and d become theflap [ɾ] in certain contexts for example in [ˈwaɾɹ] water It is clear that thereis no contrast between the flap [ɾ] and any other consonant of Englishthere are no minimal pairs such as hypothetical [hɪt] and [hɪɾ] or [bʌtɹ]and [bʌɾɹ] whose existence would establish that the flap is a distinctphoneme of English Moreover the contexts where the flap appears inEnglish are quite restricted In our previous examples of nonaspiration inthe context ˈvCv in (4) and (6) no examples included [t] as an intervocalicconsonant Now consider the following words

In (8a) orthographic lttgt is phonetically realized as the flap [ɾ] in thecontext ˈV_V that is when it is followed by a vowel or syllabic sonor-ant ndash represented as V ndash and preceded by a stressed vowel or syllabicsonorant Maybe we have just uncovered an orthographic defect ofEnglish since we have no letter for a flap (just as no letter representsθ vs eth) and some important distinctions in pronunciation are lost inspelling The second set of examples show even more clearly that under-lying t becomes a flap in this context We can convince ourselves thatthe verbs [hɪt] [pʊt] and [sεt] end in [t] simply by looking at theuninflected form of the verb or the third-person-singular forms [hɪts][pʊts] and [sεts] where the consonant is pronounced as [t] Then when weconsider the gerund which combines the root with the suffix -ɪŋ we seethat t has become the flap [ɾ] This provides direct evidence that theremust be a rule deriving flaps from plain t since the pronunciation ofroot morphemes may actually change depending on whether or not therule for flapping applies (which depends on whether a vowel follows theroot)There is analogous evidence for an underlying t in the word [ˈaeligɾm ]

atom since again the alveolar consonant in this root may either appear as

Vowels and syllabicsonorants oftenfunction together inphonology and weunify them with theterm syllabic

The theory ofdistinctive featuresgiven in chapter 3makes it easier todistinguishdifferent notions ofvowel and glide

ˈwaɾɹ water ˈwejɾɹ waiter waderˈaeligɾm atom Adam ˈaeligɾəˌthuwd attitude

ˈhɪt hit ˈhɪɾɪŋ hittingˈpʊt put ˈpʊɾɪŋ puttingˈsɛt set ˈsɛɾɪŋ setting

[th] or [ɾ] depending on the phonetic context where the segment appearsFlapping only takes place before an unstressed vowel and thus in aeligtm the consonant t is pronounced as [ɾ] but in the related form [əˈthamɪk]where stress has shifted to the second syllable of the root we can see thatthe underlying t surfaces phonetically (as an aspirate following thepreviously discussed rule of aspiration)We may state the rule of flapping as follows ldquoan alveolar stop becomes

a flap when it is followed by an unstressed syllabic and is preceded by avowel or gliderdquo You will see how vowels and glides are unified in the nextchapter for the moment we use the term vocoid to refer to the phoneticclass of vowels and glides It is again important to note that the notion ofldquovowelrdquo used in this rule must include syllabic sonorants such as [ɹ] for thepreceding segment and [ɹ] or [m ] for the following segment The rule isformalized in (9)

(9) alveolar stop flap vocoid ___ unstressed syllabic

Flapping is not limited to the voiceless alveolar stop t underlying d alsobecomes [ɾ] in this same context

213 Glottal stopThere is one context where flapping of t does not occur when preceded by avowel and followed by an unstressed syllabic segment (vowel or syllabicsonorant) and that is when t is followed by a syllabic [n] Consider firstexamples such as [ˈbʌʔn] button and [ˈkaʔn] cotton Instead of the flap that weexpect basedonourunderstandingof the contextwhereflapping takesplacewe find glottal stop before syllabic [n] Consider the following pairs of words

The bare roots on the left show the underlying t which has not changedto glottal stop and on the right we observe that the addition of the suffixn conditions the change of t to [ʔ] in the context ˈV_n ie when t ispreceded by a stressed vowel and followed by an alveolar nasal Wordslike [ˈaeligɾm ] atom show that the glottal stop rule does not apply before allnasals just alveolar nasalsFinally notice that in casual speech the gerundive suffix -ɪŋ may be

pronounced as [n] When the verb root ends in t that t becomes [ʔ] justin case the suffix becomes [n] and thus provides the crucial contextrequired for the glottal stop creation rule

Base verbs lsquoOne who V-srsquo lsquoV-ingrsquoˈbɪd ˈbɪɾɹ ˈbɪɾɪŋ bidˈhajd ˈhajɾɹ ˈhajɾɪŋ hideˈwejd ˈwejɾɹ ˈwejɾɪŋ wade

[rat] rot [ˈraʔn] rotten[hajt] height [ˈhajʔn] heighten[lajt] light [ˈlajʔn] lighten[faeligt] fat [ˈfaeligʔn] fatten

In the examples considered so far the environment for appearanceof glottalstop has been a following syllabic [n] Is it crucial that the triggering nasalsegment be specifically a syllabic nasal We also find glottal stop before non-syllabic nasals in words such as Whitney [ʍɪʔnij] and fatness [faeligʔnəs] whichshows that the t-glottalization rule does not care about the syllabicity of thefollowingnasalThepresenceofglottal stop in theseexamplescanbeexplainedby the existence of a rule which turns t into glottal stop before [n] or [n]

(13) alveolar stop glottal stop __ alveolar nasal

Notice that this rule applies before a set of segments but not a randomset it applies before alveolar nasals without mention of syllabicity As wewill repeatedly see the conditioning context of phonological rules isstated in terms of phonetic properties

22 Allophony in other languages

Allophonic rules of pronunciation are found in most human languages ifnot indeed all languages What constitutes a subtle contextual variation inone language may constitute a wholesale radical difference in phonemes inanother The difference between unaspirated and aspirated voiceless stopsin English is a completely predictable allophonic one which speakers arenot aware of but in Hindi the contrast between aspirated and unaspiratedvoiceless consonants forms the basis of phonemic contrasts eg [pal] lsquowantrsquo[phal] lsquofruitrsquo Unlike the situation in English aspiration in Hindi is animportant distinctive property of stops which cannot be supplied by a rule

l and d in Tswana The consonants [l] and [d] are clearly separatephonemesin English givenwords such as lie and die ormill andmid However in Tswana(Botswana) there is no contrast between [l] and [d] Phonetic [l] and [d] arecontextually determined variants of a single phoneme surface [l] appearsbefore nonhigh vowels and [d] appears before high vowels (neither conson-ant may come at the end of a word or before another consonant)

Some speakers have[ʔ] only beforesyllabic [n] so theirrule is different Notall Americandialects have thisrule ndash it is lackingin certain Southerndialects andinstead the flappingrule applies SomeBritish dialects havea rule which appliesin a rather differentcontext eg [lεʔə]letter

Base verb Careful speech Casual speechhɪs hɪsɪŋ hɪsnɹat ɹaɾɪŋ ɹaʔnflowt flowɾɪŋ flowʔn

lefifi lsquodarknessrsquo loleme lsquotonguersquoselεpε lsquoaxersquo molɔmo lsquomouthrsquoxobala lsquoto readrsquo mmadi lsquoreaderrsquolerumɔ lsquospearrsquo xoŋala lsquoto marryrsquoloxadima lsquolightning flashrsquo didʒɔ lsquofoodrsquodumεla lsquogreetingsrsquo feedi lsquosweeperrsquolokwalɔ lsquoletterrsquo khudu lsquotortoisersquomosadi lsquowomanrsquo podi lsquogoatrsquobadisa lsquothe herdrsquo hudi lsquowild duckrsquo

Tswana has a rule which can be stated as ldquol becomes [d] before highvowelsrdquo

(15) l d _ high vowel

An equally accurate and general statement of the distribution or [l] and [d]would be ldquod becomes [l] before nonhigh vowelsrdquo

(16) d l _ nonhigh vowel

There is no evidence to show whether the underlying segment is basicallyl or d in Tswana so we would be equally justified in assuming eitherrule (15) or rule (16) Sometimes a language does not provide enoughevidence to allow us to decide which of two (or more) analyses is correct

Tohono Orsquoodham affricates In the language Tohono Orsquoodham (for-merly known as Papago Arizona and Mexico) there is no contrastbetween [d] and [dʒ] or between [t] and [tʃ] The task is to inspect theexamples in (17) and discover what factor governs the choice betweenplain alveolar [d t] versus the alveopalatal affricates [dʒ tʃ] In theseexamples word-final sonorants are devoiced by a regular rule which wedisregard explaining the devoiced m in examples like [wahtʃum]

We do not know at the outset what factor conditions the choice of [t d]versus [tʃ dʒ] (indeed in the world of actual analysis we do not knowin advance that there is any such relationship but to make yourtask easier we will at least start with the knowledge that thereis a predictable relationship and concentrate on discovering the rulegoverning that choice) To begin solving the problem we explore twopossibilities the triggering context may be the segment which immedi-ately precedes the consonant or it may be the segment which immedi-ately follows itLet us start with the hypothesis that it is the immediately preceding

segment which determines how the consonant is pronounced In order toorganize the data so as to reveal what rule might be at work we can

dʒihsk lsquoauntrsquo dɔʔaʔk lsquomountainrsquotʃuli lsquocornerrsquo tʃɯwaʔgi lsquocloudsrsquowahtʃum lsquodrownrsquo taht lsquofootrsquodʒɯwɯhkɔh lsquocut hairrsquo ʔahidaʔk lsquoyearrsquotɔnɔm lsquobe thirstyrsquo hɯhtahpsptʃu lsquomake it 5rsquohɯdʒɯli lsquoselfrsquo tʃihkpan lsquoworkrsquostahtɔnɔmah lsquothirsty timesrsquo ʔida lsquothisrsquomɯɖɯdam lsquorunnerrsquo tɔhntɔ lsquodegeneratersquotɔdsid lsquofrightenrsquo tʃɯpɔsid lsquobrandrsquogahtwi lsquoto shootrsquo tʃɯhtʃi lsquonamersquogɯʔɯdta lsquoget bigrsquo dʒumali lsquolowrsquo

tobidk lsquoWhite Clayrsquo waʔdʒiwih lsquoswimrsquo

spadmahkam lsquolazy onersquo dʒuʔw lsquorabbitsrsquo

simply list the preceding environments where stops versus affricatesappear so h_ means ldquowhen [h] precedesrdquo ndash here the symbol ldquordquo representsthe beginning or end of a word Looking at the examples in (17) andtaking note of what comes immediately before any [t d] versus [tʃ dʒ] wearrive at the following list of contexts

(18) [t d] _ h_ ɯ_ i _ s_ i_ n_ ɔ_[tʃ dʒ] _ h_ ɯ_ ʔ_ p_

Since both types of consonants appear at the beginning of the wordor when preceded by [h] or [ɯ] it is obvious that the precedingcontext cannot be the crucial determining factor We therefore rejectthe idea that the preceding element determines how the phoneme ispronouncedFocusing next on what follows the consonant the list of contexts

correlated with plain stops versus affricates is much simpler

(19) [t d] _ ɔ _a _a _ _s _t _k _ ɯ _w[tʃ dʒ] _i _i _u _u _ɯ

Only the vowels [i u ɯ] (and their devoiced counterparts) follow [tʃ] and[dʒ] and the vowels [a ɔ] follow [t] and [d] Moreover when no vowelfollows ie at the end of the word or before another consonant the plainalveolar appears (taht tɔdsid) The vowels [i u ɯ] have in common theproperty that they are high vowels which allows us to state the contextfor this rule very simply t and d become alveopalatal affricates beforehigh vowels ie

(20) alveolar stop alveopalatal affricate _high vowel

The retroflex consonant [ɖ ] does not undergo this process as seen in[mɯɖɯdam]This account of the distribution of alveolars versus alveopalatals assumes

that underlyingly the consonants are alveolars and that just in case a highvowel follows the consonant becomes an alveopalatal affricate It is import-ant to also consider the competing hypothesis that underlyingly the conson-ants are alveopalatals and that they become alveolars in a context which iscomplementary to that stated in rule (20) The problemwith that hypothesisis that there is no natural statement of that complementary context whichincludes nonhigh vowels consonants and the end of the word

The brace notation is a device used to force a disjunction of unrelatedcontexts into a single rule so this rule states that alveopalatal affricates

alveopalatal affricate rarr alveolar stop_

nonhigh V

become alveolar stops when they are followed either by a nonhigh vowela consonant or are at the end of the word ie there is no coherentgeneralization Since the alternative hypothesis that the consonants inquestion are underlyingly alveopalatals leads to a much more complicatedand less enlightening statement of the distribution of the consonants wereject the alternative hypothesis and assume that the consonants areunderlyingly alveolar

Obstruent voicing in Kipsigis In the Kipsigis language of Kenya thereis no phonemic contrast between voiced and voiceless obstruents as thereis in English No words are distinguished by the selection of voiced versusvoiceless consonants nevertheless phonetic voiced obstruents do exist inthe language

In these examples we can see that the labial and velar consonants becomevoiced when they are both preceded and followed by vowels liquidsnasals and glides these are all sounds which are voiced

(23) voiceless peripheral consonant voiced voiced _ voiced

In stating the context we do not need to say ldquovoiced vowel liquid nasalor gliderdquo since by saying ldquovoicedrdquo alone we refer to the entire class ofvoiced segments It is only when we need to specifically restrict the rule sothat it applies just between voiced consonants for example that we wouldneed to further specify the conditioning class of segmentsWhile you have been told that there is no contrast between [k] and [g]

or between [p] and [b] in this language children learning the languagedo not use explicit instructions so an important question arises howcan you arrive at the conclusion that the choice [k p] versus [g b] ispredictable Two facts lead to this conclusion First analyzing the dis-tribution of consonants in the language would lead to discovering theregularities that no word begins or ends in [b g] and no word has [b g]in combination with another consonant except in combination with the

kuur lsquocallrsquo ke-guur lsquoto callrsquoŋok-ta lsquodogrsquo ŋog-iik lsquodogsrsquokε-tεp lsquorequestrsquo i-teb-e lsquoyou are requestingrsquoker lsquolook atrsquo ke-ger lsquoto look atrsquoput lsquobreak uprsquo ke-but lsquoto break uprsquopoor lsquothresh maizersquo ke-boor lsquoto thresh maizersquoŋeljep-ta lsquotonguersquo ŋeljeb-wek lsquotonguesrsquokisiptʃi lsquoto follow forrsquo iŋgurwet lsquopigrsquokipkirui (name) ke-baakpaak lsquoto strip repeatedlyrsquopʊnbʊn lsquosoftrsquo tilɑkse lsquoit is cuttablersquokirgit lsquobullrsquo kagjam lsquowe atersquotaaptεεt lsquoflower typersquo kebritɑmεεt lsquoto fall asleeprsquokɪblaŋat (name) peetʃiŋge lsquothey are going for

themselvesrsquo

voiced sonorants We would also discover that [p k] do not appearbetween vowels or more generally between voiced segments If therewere no rule governing the distribution of consonants in this languagethen the distribution is presumed to be random which would meanthat we should find examples of [b g] at the beginning or end of wordsor [p k] between vowelsAnother very important clue in understanding the system is the fact

that the pronunciation of morphemes will actually change accordingto the context that they appear in Notice for example that theimperative form [kuur] lsquocallrsquo has a voiceless stop but the same rootis pronounced as [guur] in the infinitive [ke-guur] lsquoto callrsquo Whenlearning words in the language the child must resolve the changesin pronunciation of word parts in order to know exactly what must belearned Sometimes the root lsquocallrsquo is [kuur] sometimes [guur] ndash whenare you supposed to use the pronunciation [guur] Similarly in tryingto figure out the root for the word lsquodogrsquo a child will observe thatin the singular the root portion of the word is pronounced [ŋok] andin the plural it is pronounced [ŋog] From observing that there is analternation between [k] and [g] or [p] and [b] it is a relatively simplematter to arrive at the hypothesis that there is a systematic relationbetween these sounds which leads to an investigation of when [k p]appear versus [g b]

Implosive and plain voiced stops in Matuumbi The distinctionbetween implosive and plain voiced consonants in Matuumbi (Tanzania)can be predicted by a rule

Upon consideration of consonant distribution in these data you will seethat implosives appear in word-initial position and after vowels whereasplain voiced consonants appear exclusively after nasalsThere is further clinching evidence that this generalization is valid

In this language the first-person-singular form of the verb has anasal consonant prefix (there is also a change in the final vowelwhere you get -a in the infinitive and -ε in the ldquoshouldrdquo form the secondcolumn below)

ɓεɓεεlu lsquomale goatrsquo ɠundumuka lsquobe scaredrsquoɓutuka lsquoflowrsquo ɠaala lsquostorage in roofrsquokɔɓɔkwa lsquounfoldrsquo ɓwʊʊmɪ lsquolifersquokɔɔndwa lsquodig clayrsquo ŋgaambalε lsquofish (sp)rsquoɓalaka lsquoluckrsquo ɠʊlʊja lsquodrive fastrsquolisεεŋgεlε lsquodowryrsquo ɓila lsquowithoutrsquoɠɔlɔja lsquostraightenrsquo ɠuna lsquomurmurrsquokiɓʊla lsquotowards Meccarsquo kitʊʊmbɪ lsquohillrsquokjaaŋgi lsquosandrsquo ɓɔmwaana lsquodestroyrsquolikʊʊŋgwa lsquostorage structurersquo ɓʊʊka lsquoleaversquoɠɔɔmba lsquoshoot a gunrsquo ɠʊlʊka lsquoflyrsquoɓalaaŋga lsquocountrsquo aliɓika lsquobe out of orderrsquo

Thus the pronunciation of the root for the word for lsquoflyrsquo alternatesbetween [ɠʊlʊk] and [gʊlʊk] depending on whether a nasal precedesHaving determined that implosives and plain voiced stops are allo-

phonically related in the grammar of Matuumbi it remains to decidewhether the language has basically only plain voiced consonants withimplosives appearing in a special environment or should we assume thatMatuumbi voiced stops are basically implosive and plain voiced conson-ants appear only in a complementary environment The matter boilsdown to the following question is it easier to state the context whereimposives appear or is it easier to state the context where plain voicedconsonants appear We generally assume that the variant with the mosteasily stated distributional context is the variant derived by applying arule However as we saw with the case of [l] and [d] in Tswana a languagemay not provide empirical evidence which is the correct solutionNow let us compare the two possible rules for Matuumbi ldquoimplosives

appear word initially and after a vowelrdquo

versus ldquoplain consonants appear after a nasalrdquo

(27) C nonimplosive nasal _

It is simpler to state the context where plain consonants appear sincetheir distribution requires a single context ndash after a nasal ndash whereasdescribing the process as replacement of plain consonants by implosiveswould require a more complex disjunction ldquoeither after a vowel or inword-initial positionrdquo A concise description of contexts results if weassume that voiced consonants in Matuumbi are basically implosive andthat the nonimplosive variants which appear after nasals are derived by asimple rule implosives become plain voiced consonants after nasalsIt is worth noting that another statement of the implosive-to-plain process

is possible since sequences of consonants are quite restricted in MatuumbiOnly a nasal may precede another ldquotruerdquo consonant ie a consonant otherthan a glide A different statement of the rule is that plain voiced consonantsappear only after other consonants ndash due to the rules of consonant

to V I should Vɠʊlʊka ŋgʊlʊkε lsquoflyrsquoɠɔɔmba ŋgɔɔmbε lsquoshoot a gunrsquoɠɔlɔja ŋgɔlɔjε lsquostraightenrsquoɠuna ŋgunε lsquomurmurrsquoɓalaaŋga mbalaaŋgε lsquocountrsquoɓutuka mbutukε lsquoflowrsquo

ɓʊʊka mbʊʊkε lsquoleaversquoɗuumu nduumu lsquocontinuersquo

C implosive V

rarr⎧⎨⎩

⎫⎬⎭

combination in the language thefirst of two true consonants is necessarily anasal so it is unnecessary to explicitly state that the preceding consonant inthe implosive-to-plain-C rule is a nasal Phonological theory does not alwaysgive a single solution for any given data set so wemust accept that there areat least two ways of describing this pattern One of the goals of the theorytowards which considerable research energy is being expended is develop-ing a principled basis for making a unique and correct choice in such caseswhere the data themselves cannot show which solution is right

Velar and uvular stops in Kenyang In Kenyang (Cameroon) there is nocontrast between the velar consonant k and uvular q

What determines the selection of k versus q is the nature of the vowelwhich precedes the consonant The uvular consonant q is always precededby one of the back nonhigh vowels o ɔ or ɑ whereas velar k appearsanywhere else

(29) voiceless velar uvular back nonhigh vowel _

This relation between vowels and consonants is phonetically natural Thevowels triggering the change have a common place of articulation theyare produced at the lower back region of the pharynx where q (as opposedto k) is articulatedAn alternative is that the underlying segment is a uvular and velar

consonants are derived by rule But under that assumption the rulewhich derives velars is very complex Velars would be preceded by frontor central vowels by high back vowels by a consonant (ŋ) or by a wordboundary We would then end up with a disjunction of contexts in ourstatement of the rule

enɔq lsquotreersquo enoq lsquodrumrsquo

eket lsquohousersquo ntʃiku lsquoI am buyingrsquonek lsquoropersquo ejwɑrek lsquosweet potatorsquoŋgɑq lsquoknifersquo ekɑq lsquolegrsquomək lsquodirtrsquo nɑq lsquobrother in lawrsquo

ndek lsquoEuropeanrsquo pɔbrik lsquowork projectrsquobetək lsquojobrsquo bepək lsquoto capsizersquotiku (name) ku lsquobuyrsquoɑjuk (name) esikɔŋ lsquopipersquokebwep lsquostammeringrsquo ŋkɔq lsquochickenrsquoŋkɑp lsquomoneyrsquo kɔ lsquowalkrsquo

front Vcentral V

high back VC

q k rarr

⎧⎪⎪⎨⎪⎪⎩

⎫⎪⎪⎬⎪⎪⎭

The considerably more complex rule deriving velars from uvulars leads usto reject the hypothesis that these segments are underlyingly uvularAgain we are faced with one way of capturing the generalization exploit-ing phonetically defined classes and an alternative that involves a dis-junctive list where there is nothing that unifies the contexts we selectthe alternative which allows a rule to be stated that refers to a simplephonetically definable context This decision reflects an important dis-covery regarding the nature of phonogical rules which will be discussed ingreater detail in chapter 3 namely that phonological rules operate interms of phonetic classes of segments

Arabela nasalization Nasalization of vowels and glides is predictable inArabela (Peru)

Scanning the data in (31) we see nothing about the following phoneticcontext that explains occurrence of nasalization both oral and nasalvowels precede glottal stop ([taeligweʔ] lsquoforeignerrsquo versus [notildeotildenũʔ] lsquobepainedrsquo) [k] ([nĩ ĩkjaeligaeligʔ] lsquois pouring outrsquo versus [ʃijokwaʔ] lsquogreasersquo) or [n]([mȷaelignũʔ] lsquoswallowrsquo versus [posunatildeh atildeʔ] lsquoshort personrsquo) A regularitydoes emerge once we look at what precedes oral versus nasal vowelswhen a vowel or glide is preceded by a nasal segment ndash be it a nasalconsonant (including [h ] which is always nasal in this language) vowelor glide ndash then a vowel or glide becomes nasalized The rule for nasal-ization can be stated as ldquoa vowel or glide becomes nasalized after anynasal soundrdquo

(32) vocoid nasal nasal_

The naturalness of this rule should be obvious ndash the essential propertythat defines the conditioning class of segment nasality is the very prop-erty that is added to the vowel such a process where a segment becomesmore like some neighboring segment is known as an assimilationPredictable nasalization of vowels almost always derives from a nasalconsonant somewhere near the vowel

Sundanese a problem for the student to solve Bearing this suggestionin mind where do nasalized vowels appear in Sundanese (Indonesia)given these data

nẽẽkjaeligaeligʔ lsquolying on backrsquo motildenũʔ lsquokillrsquotukuruʔ lsquopalm leafrsquo ʃijokwaʔ lsquogreasersquonȷaeligaeligriʔ lsquohe laid it downrsquo suroʔ lsquomonkeyrsquonĩĩkjaeligaeligʔ lsquois pouring outrsquo suwakaʔ lsquofishrsquoposunatildeh atildeʔ lsquoshort personrsquo kuwɔxoʔ lsquoholersquonotildeotildenũʔ lsquobe painedrsquo hẽẽgiʔ lsquotermitesrsquotaeligweʔ lsquoforeignerrsquo hȷũũʃʃaelignotildeʔ lsquowhere I fishedrsquonĩnȷũʔ lsquoto comersquo mȷaelignũʔ lsquoswallowrsquo

nũwatildeʔ lsquopartridgersquo hũwatildeʔ lsquoa yellow birdrsquo

Since the focus at the moment is on finding phonological regularities andnot on manipulating a particular formalism (which we have not yetpresented completely) you should concentrate on expressing the general-ization in clear EnglishWe can also predict the occurrence of long (double) consonants

in Sundanese using the above data supplemented with the datain (34)

What rule determines the length of consonants in this language

Vowel length in Mohawk The context for predicting some variantof a phoneme may include more than one factor There is nocontrast between long and short vowels in Mohawk (North America)what is the generalization regarding where long versus short vowelsappear

abot lsquoheavyrsquo agɨŋ lsquobigrsquoamĩs lsquosweetrsquo anotildem lsquoyoungrsquohandap lsquolightrsquo luhur lsquohighrsquoawon lsquobadrsquo basɨr lsquowetrsquokonẽŋ lsquoyellowrsquo bɨrɨm lsquoredrsquoeŋgal lsquonewrsquo gədde lsquobigrsquomatildehĩr lsquoskillfulrsquo mĩrɨ lsquouncertainrsquomotildehẽhẽd lsquopoorrsquo bumĩ lsquohousersquomərri lsquoduckrsquo matildehatildesiswa lsquostudentrsquomatildeũŋ lsquotigerrsquo mĩatildesih lsquotrue loversquomĩliar lsquobillionrsquo mĩɲatildek lsquooilrsquomũatildera lsquoconfluencersquo pamotildehatildelan lsquoimpossiblersquomatildeẽn lsquoplayrsquo matildeotildet lsquodiersquonatildeʔatildes lsquoget worsersquo mĩʔĩs lsquoleakrsquomatildeʔatildep lsquoexcuse mersquo matildehĩ lsquoenoughrsquonẽwak lsquocatchrsquo tiʔis lsquocoldrsquo

abuabu lsquogreyrsquo bəddil lsquogunrsquogəttih lsquobloodrsquo akar lsquorootrsquosəddih lsquosadrsquo dʒənnəŋŋatilden lsquonamersquobərrəkkah lsquousefulrsquo bagoŋ lsquowild pigrsquobabi lsquopigrsquo kinatilde lsquoquininersquotəbbih lsquofarrsquo bapa lsquofatherrsquobɨbɨr lsquobeltrsquo ŋəppel lsquosweeprsquobənnər lsquocorrectrsquo sɨkɨt lsquosharprsquopanatildes lsquohotrsquo məddəm lsquodarkrsquohukum lsquolawrsquo sərrat lsquoletterrsquokamẽdʒa lsquoshirtrsquo patʃul lsquoshovelrsquobənnatildeŋ lsquothreadrsquo dada lsquotorsorsquopəttis lsquofish saucersquo dʒaŋkuŋ lsquotallrsquoasəm lsquotamarindrsquo wawəs lsquotoothrsquo

One property which holds true of all long vowels is that they appear instressed syllables there are no unstressed long vowels However it wouldbe incorrect to state the rule as lengthening all stressed vowels becausethere are stressed short vowels as in [ˈwisk] We must find a furtherproperty which distinguishes those stressed vowels which becomelengthened from those which do not Looking only at stressed vowelswe can see that short vowels appear before two consonants and longvowels appear before a consonant-plus-vowel sequence It is the combin-ation of two factors being stressed and being before the sequence CVwhich conditions the appearance of long vowels stressed vowels arelengthened if they precede CV and vowels remain short otherwise Wehypothesize the following rule

(36) stressed V long _CV

Since there is no lexical contrast between long and short vowels inMohawkwe assume that all vowels have the same underlying length all long andshortened in one context or all short and lengthened in the complemen-tary context One hypothesis about underlying forms in a given languageresults in simpler grammars which capture generalizations about the lan-guage more directly than do other hypotheses about underlying forms Ifall vowels in Mohawk are underlyingly long you must devise a rule toderive short vowels No single generalization covers all contexts wheresupposed vowel shortening takes place so your analysis would requiretwo rules one to shorten unstressed vowels and another to shorten vowelsfollowed by two consonants In comparison the single rule that stressedvowels lengthen before CV accounts for vowel length under the hypothesisthat vowels in Mohawk are underlyingly short No other rule is neededshort vowels appear everywhere that they are not lengthened

Aspiration in Ossetic Aspiration of voiceless stops can be predicted inOssetic (Caucasus)

ranaˈhezʌs lsquohe trusts herrsquo raˈgedas lsquohe scrapesrsquoˈigeks lsquoI eat itrsquo oˈdawe lsquoflearsquogaˈdadis lsquoI talkrsquo ʌkhˈninũʔ lsquoI will buy itrsquoˈsduha lsquoa little bitrsquo apˈlam lsquoAbramrsquo

ʌˈgaradeʔ lsquoI lay myself downrsquo ˈdʌgehgweʔ lsquoIrsquoll lift itrsquoraˈjʌ thos lsquohe plantsrsquo ˈjegreks lsquoI push itrsquoˈwisk lsquofiversquo roˈjoʔdeʔ lsquohe worksrsquoawerˈjahsa lsquoheartrsquo ˈjagwaks lsquothey and I eat itrsquoˈisgʌs lsquoyou (sg) see herrsquo gatˈgahthos lsquoI look at itrsquojoˈkekhaʔ lsquoitrsquos burningrsquo ʌˈgidjeʔ lsquoI will fly aroundrsquo

thəχ lsquostrengthrsquo khɔttag lsquolinenrsquoχɔstɔg lsquonearrsquo ɔftən lsquobe addedrsquofadath lsquopossibilityrsquo khastɔn lsquoI lookedrsquotshɔst lsquoeyersquo kharkh lsquohenrsquo

Since aspirated and plain consonants appear at the end of the word ([tshɔst]lsquoeyersquo [tshəth] lsquohonorrsquo) the following context alone cannot govern aspir-ation Focusing on what precedes the consonant aspirates appear word-initially or when preceded by a vowel or [r] (ie a sonorant) at the end ofthe word unaspirated consonants appear when before or after anobstruent It is possible to start with unaspirated consonants (as we didfor English) and predict aspiration but a simpler description emerges ifwe start from the assumption that voiceless stops are basically aspiratedin Ossetic and deaspirate a consonant next to an obstruent The relativesimplicity of the resulting analysis should guide your decisions aboutunderlying forms and not a priori decisions about the phonetic natureof the underlying segments that your analysis results in

Optional rules Some rules of pronunciation are optional often knownas ldquofree variationrdquo In Makonde (Mozambique) the phoneme ʃ can bepronounced as either [s] or [ʃ] by speakers of the language the samespeaker may use [s] one time and [ʃ] another time The verb lsquoreadrsquo is thuspronounced as ʃoomja or as soomja and lsquosellrsquo is pronounced as ʃuluuʃa oras suluusa We will indicate such variation in pronunciation by givingthe examples as ldquoʃuluuʃa ~ suluusardquo meaning that the word is pronounce-able either as ʃuluuʃa or as suluusa as the speaker chooses Such appar-ently unconditioned fluctuations in pronunciation are the result of a rulein Makonde which turns ʃ into [s] this rule is optional The optionalnature of the rule is indicated simply by writing ldquooptionalrdquo to the side ofthe rule

(38) ʃ s optional

Normally any rule in the grammar always applies if its phonologicalconditions are satisfied An optional rule may either apply or not so forany optional rule at least two phonetic outcomes are possible either therule applies or it does not apply Assuming the underlying form ʃoomjathe pronunciation [ʃoomja] results if the rule is not applied and [soomja]results if the rule is appliedOptional rules may have environmental conditions on them In

Matuumbi as we have seen in (24) voiced stops are implosive except aftera nasal The voiced velar stop exhibits a further complication that after avowel (but not initially) underlying ɠ optionally becomes a fricative [γ](the symbol ldquo~rdquo indicates ldquomay also be pronounced asrdquo)

akkag lsquoadequatersquo dəkkag lsquosecondrsquotshəppar lsquofourrsquo tshəth lsquohonorrsquotshəχt lsquocheesersquo khɔm lsquowherersquofɔste lsquobehindrsquo khom lsquomouthrsquophirən lsquocomb woolrsquo zaχta lsquohe toldrsquoχɔskard lsquoscissorsrsquo χɔston lsquomilitaryrsquophɔrrɔst lsquoflutteringrsquo

Hence the optional realization of ɠ as [γ] but only after a vowel can beexplained by the following rule

(40) ɠ γ V _ (optional)

The factors determining which variant is selected are individual andsociological reflecting age ethnicity gender and geography inter aliaPhonology does not try to explain why people make the choices they dothat lies in the domain of sociolinguistics We are also only concernedwith systematic options Some speakers of English vary between [aeligks] and[aeligsk] as their pronunciation of ask This is a quirk of a particular word nospeaker says [maeligks] for mask or [fɪsk] for fixIt would also be mistaken to think that there is one grammar for all

speakers of English (or German or Kimatuumbi) and that dialect variationis expressed via a number of optional rules From the perspective ofgrammars as objects describing the linguistic competence of individualsan optional rule is countenanced only if the speaker can actually pro-nounce words in multiple ways In the case of Makonde some speakersactually pronounce ʃoomja in two different ways

ɓaɠana ~ ɓaγana lsquodividersquoɓiliɠana ~ ɓiliγana lsquowrestlersquoɓulaɠa ~ ɓulaγa lsquokillrsquoɠalaambuka ~ (γalaambuka) lsquochangersquo

Summary Contrastive aspects of pronunciation cannot be predicted by rule butallophonic details can be Allophonic changes are a type of rule-governedphonological behavior and phonology is concerned with the study ofrules The practical concern of this chapter is understanding themethod for discovering those rules The linguist looks for regularitiesin the distribution of one sound versus others and attempts to reducemultiple surface segments to one basic segment a phoneme where therelated segments derive by applying a rule to the underlying phonemein some context Going beyond static distribution of sounds youshould look for cases where the pronunciation of morphemes changesdepending on the presence or absence of prefixes and suffixesAssuming that sounds are in complementary distribution you need

to determine which variant is the ldquobasicrdquo underlying one and whichderives by rule The decision is made by comparing the consequencesof alternative hypotheses Sometimes selecting underlying X resultsin a very simple rule for deriving the surface variant [Y] whereasselecting underlying Y results in very complex rules for deriving [X]from Y in such a case the choice of X over Y is well motivatedSometimes no definitive decision can be made

Exercises1 KuriaProvide rules to explain the distribution of the consonants [β r γ] and [b d g] inthe following data (Note that [r] is a fricative consonant in this language)Accents mark tone acute is high tone and ldquohacekrdquo [ˇ] is rising tone

2 Modern GreekDetermine whether the two segments [k] and [kj] are contrastive or aregoverned by rule similarly determine whether the difference between [x] and[xj] is contrastive or predictable If the distribution is rule-governed what is therule and what do you assume to be the underlying consonants in these cases

3 FarsiDescribe the distribution of the trills [r] [r] and the flap [ɾ]

aβaanto lsquopeoplersquo aβamura lsquoyoung menrsquoamahıındi lsquocorn cobsrsquo amakέέndɔ lsquodate fruitsrsquoeβa lsquoforgetrsquo eeŋgwe lsquoleopardrsquoeγa lsquolearnrsquo ekeβwέ lsquofoxrsquohoora lsquothreshrsquo iβiγuruβe lsquosmall pigsrsquoiβiruuŋguuri lsquosoft porridgesrsquo uγusıri lsquohuge ropersquoβainu lsquoyou (pl)rsquo βorjo lsquoon the rightrsquoitʃiiŋgέna lsquogrinding stonesrsquo itʃiiŋguruβe lsquopigrsquoγaβa lsquosharersquo itʃiiŋguta lsquowallsrsquoβεrεka lsquocarry a childrsquo iγituumbe lsquostoolrsquoγuuka lsquoancestorrsquo rema lsquoweedrsquorεεnta lsquobringrsquo oβoγaaka lsquomale adulthoodrsquooβoteendeeru lsquosmoothnessrsquo okoγeemba lsquoto cause rainrsquookoombara lsquoto count mersquo okoβara lsquoto countrsquookoondɔγa lsquoto bewitch mersquo okorɔ γa lsquoto bewitchrsquoroma lsquobitersquo teγeta lsquobe latersquoukuumbuurja lsquoto ask mersquo uruγuta lsquowallrsquo

kano lsquodorsquo kori lsquodaughterrsquoxano lsquolosersquo xori lsquodancesrsquoxjino lsquopourrsquo kjino lsquomoversquokrima lsquoshamersquo xrima lsquomoneyrsquoxufta lsquohandfulrsquo kufeta lsquobonbonsrsquokali lsquocharmsrsquo xali lsquoplightrsquoxjeli lsquoeelrsquo kjeri lsquocandlersquoxjeri lsquohandrsquo oxji lsquonorsquo

aeligrteʃ lsquoarmyrsquo farsi lsquoPersianrsquoqaeligdri lsquoa little bitrsquo rah lsquoroadrsquorast lsquorightrsquo riʃ lsquobeardrsquoahar lsquostarchrsquo axaeligr lsquolastrsquohaeligrtowr lsquohoweverrsquo ʃir lsquolionrsquoahaɾi lsquostarchedrsquo baeligɾadaeligr lsquobrotherrsquotʃeɾa lsquowhyrsquo daɾid lsquoyou haversquobiɾaeligng lsquopalersquo ʃiɾini lsquopastryrsquo

4 OsageWhat rule governs the distribution of [d] versus [eth] in the following data

5 AmharicIs there a phonemic contrast between the vowels [ə] and [ε] in Amharic If notsay what rule governs the distribution of these vowels and what the underlyingvalue of the vowel is

6 GenDetermine the rule which accounts for the distribution of [r] and [l] in thefollowing data

7 ShambaaDescribe the distribution of voiced versus voiceless nasals (voiceless nasals arewritten with a circle under the letter as in m

˚) and voiceless aspirated voiceless

unaspirated and voiced stops in Shambaa

ˈdabri lsquothreersquo ˈaethikhaʒa lsquohe lay downrsquodatʃˈpe lsquoto eatrsquo ˈtʃʔeethe lsquohe killed itrsquodakˈʔe lsquoto digrsquo ˈetheze lsquotonguersquoˈdali lsquogoodrsquo ˈethie lsquoyoursquodaʃˈtu lsquoto bitersquo ˈethiʃki lsquoto washrsquo

fərəs lsquohorsersquo tənəsa lsquostand uprsquojεlɨdʒlɨdʒ lsquograndchildrsquo majεt lsquoseersquogənzəb lsquomoneyrsquo dʒεgna lsquobraversquonəɲ lsquoI amrsquo məwdəd lsquoto likersquomənnəsat lsquoget uprsquo məmkər lsquoadvisersquoʒεle lsquounarmedrsquo jεlləm lsquonorsquomətʃ lsquowhenrsquo məstrsquoət lsquogiversquofəlləgə lsquohe wantedrsquo agəɲɲε lsquohe foundrsquotəmətʃtʃε lsquoit got comfortablersquo mokkərə lsquohe triedrsquokrsquoaʒʒε lsquohe talked in his sleeprsquo ʒεmmərə lsquohe startedrsquolatʃtʃrsquoε lsquohe shavedrsquo aʃʃε lsquohe rubbedrsquobəkkrsquoələ lsquoit germinatedrsquo ʃεməggələ lsquohe became oldrsquo

agble lsquofarmrsquo agoŋglo lsquolizardrsquoaŋɔli lsquoghostrsquo akplɔ lsquospearrsquosabulε lsquoonionrsquo sra lsquostrainrsquoalɔ lsquohandrsquo atitrwε lsquored-billed wood doversquoavlɔ lsquobaitrsquo blafogbe lsquopineapplersquodrε lsquostretch armsrsquo edrɔ lsquodreamrsquoexlɔ lsquofriendrsquo exle lsquoflearsquohlε lsquoreadrsquo ŋlɔ lsquowritersquotʃrɔ lsquoexterminatersquo ɲra lsquobe uglyrsquoklɔ lsquowashrsquo tre lsquogluersquovlu lsquostretch a ropersquo lɔ lsquolikersquomla lsquopound a drumrsquo pleplelu lsquolaughing doversquowla lsquohidersquo zro lsquoflyrsquoesrɔ lsquospousersquo etro lsquoscalersquoeɲrɔ lsquospitting cobrarsquo dʒro lsquohintrsquo

8 ThaiThe obstruents of Thai are illustrated below Determine what the obstruentphonemes of Thai are ([p t and k] are unreleased stops) Are [p t k]distinct phonemes or can they be treated as positional variants of some otherphoneme If so which ones and what evidence supports your decision Notethat no words begin with [g]

9 PalauanAnalyze the distribution of eth θ and d in the following data Examples of the typeldquoX ~ Yrdquo mean that the word can be pronounced either as X or as Y in freevariation

tagi lsquoeggrsquo kitabu lsquobookrsquo paalika lsquoflyrsquoni lsquoit isrsquo ŋombe lsquocowrsquo matagi lsquoeggsrsquododoa lsquopick uprsquo goʃa lsquosleeprsquo babu lsquoskinrsquondimi lsquotonguesrsquo ŋgoto lsquoheartrsquo mbeu lsquoseedrsquonthumbii lsquomonkeyrsquo ŋkhuŋguni lsquobedbugrsquo mpheho lsquowindrsquo

bil lsquoBillrsquo myy lsquohandrsquorak lsquoloversquo baa lsquocrazyrsquoloŋ lsquogo downrsquo bryy lsquoextremely fastrsquohaa lsquofiversquo plaa lsquofishrsquodii lsquogoodrsquo tʃaan lsquodishrsquothee lsquopourrsquo thruumεεn lsquoTrumanrsquokhεŋ lsquohardrsquo panjaa lsquobrainsrsquoləəj lsquopassrsquo phjaa [title]lyak lsquochoosersquo klaaŋ lsquomiddlersquotʃhat lsquoclearrsquo traa lsquostamprsquoriip lsquohurryrsquo ɔɔk lsquoexitrsquophrεε lsquosilk clothrsquo kiə lsquowooden shoesrsquokhwaa lsquoright sidersquo kεε lsquooldrsquodraj lsquodrive (golf)rsquo dyŋ lsquopullrsquokan lsquoward offrsquo tʃuək lsquopure whitersquophleeŋ lsquosongrsquo tʃhan lsquomersquostaaŋ lsquomoneyrsquo rap lsquotakersquojiisip lsquotwentyrsquo phaa lsquoclothrsquokhaa lsquokillrsquo dam lsquoblackrsquoraaj lsquocasersquo tit lsquoget stuckrsquosip lsquotenrsquo pen lsquoaliversquo

kəethə lsquowe (inclusive)rsquo bəethuk lsquomy stonersquoethiak ~ diak lsquonegative verbrsquo maθ lsquoeyersquotŋoθ lsquotattoo needlersquo ethel ~ del lsquonailrsquoethiosəʔ ~ diosəʔ lsquoplace to bathersquo ethik ~ dik lsquowedgersquokuθ lsquolousersquo ʔoethiŋəl lsquovisitrsquokoaθ lsquovisitrsquo eaŋəθ lsquoskyrsquoŋərarəethə lsquoa villagersquo baθ lsquostonersquoieethl lsquomangorsquo ʔəethip lsquoantrsquokəetheb lsquoshortrsquo məethəŋei lsquoknewrsquouethouθ lsquomoneyrsquo olethak lsquoput togetherrsquo

10 Quechua (Cuzco dialect)Describe the distribution of the following four sets of segments k x q χ ŋ ɴ ie u o Some pairs of these segments are allophones (positional variants) of asingle segment You should state which contrasts are phonemic (unpredictable)and which could be predicted by a rule For segments which you think arepositional variants of a single phoneme state which phoneme you think is theunderlying variant and explain why you think so provide a rule which accountsfor all occurrences of the predictable variant (Reminder ɴ is a uvular nasal)

11 Lhasa TibetanThere is no underlying contrast in this language between velars and uvulars noris there an underlying contrast between voiced and voiceless obstruent norbetween stops or fricatives except s which exists underlyingly State what theunderlying segments are and give rules which account for the surfacedistribution of these consonant types (Notational reminder [ɢ] represents avoiced uvular stop)

qori lsquogoldrsquo tʃoχlu lsquocorn on the cobrsquoqrsquoomir lsquogreenrsquo niŋri lsquoearrsquomoqo lsquoruntrsquo hoqrsquoara lsquodeafrsquophulju lsquoblanketrsquo jujaŋ lsquohe recallsrsquotulju lsquobonersquo api lsquotakersquosuti lsquonamersquo oɴqoj lsquobe sickrsquotʃilwi lsquobaby chickrsquo tʃhitʃiŋ lsquohe whispersrsquotʃhaɴqaj lsquogranulatersquo aɴqosaj lsquotoastrsquoqetʃuŋ lsquohe disputesrsquo prsquoisqo lsquobirdrsquomusoχ lsquonewrsquo tʃuŋka lsquotenrsquojaɴqaŋ lsquofor freersquo tʃulju lsquoicersquoqhelja lsquolazyrsquo qrsquoeɴqo lsquozigzaggedrsquotʃeqaŋ lsquostraightrsquo qaŋ lsquoyoursquonoqa lsquoIrsquo tʃaxra lsquofieldrsquotʃeχniŋ lsquohe hatesrsquo soχta lsquosixrsquoaχna lsquothusrsquo ljixlja lsquosmall shawlrsquoqosa lsquohusbandrsquo qara lsquoskinrsquoalqo lsquodogrsquo seɴqa lsquonosersquokaru lsquofarrsquo atoχ lsquofoxrsquoqaŋkuna lsquoyou (pl)rsquo pusaχ lsquoeightrsquotrsquoeχwaj lsquopluckrsquo tʃrsquoaki lsquodryrsquowateχ lsquoagainrsquo aŋka lsquoeaglersquowaχtaj lsquohitrsquo haku lsquoletrsquos gorsquowaqaj lsquotearsrsquo kaŋka lsquoroastedrsquowaxtʃa lsquopoorrsquo waleχ lsquopoorrsquothakaj lsquodroprsquo reχsisqa lsquoknownrsquo

aŋgu lsquopigeonrsquo aŋʈaa lsquoa numberrsquo aŋba lsquoduckrsquoapsoo lsquoshaggy dogrsquo amtʃɔɔ lsquoearrsquo tukʈyy lsquopoison snakersquoamto lsquoa provincersquo ɨγu lsquounclersquo ɨmtʃi lsquodoctorrsquouʈɨ lsquohairrsquo uβɪɪ lsquoforeheadrsquo eʁa lsquobellsrsquoembo lsquodesertedrsquo ʊʊtsi lsquooh-ohrsquo qa lsquosaddlersquoqaʁa lsquoalphabetrsquo qaŋba lsquofootrsquo qamba lsquopliersrsquo

12 Kirzan ArmenianIn this language certain surface vowels can be predictably derived from otherunderlying vowels Discover what vowels in this language are purely predictableand give the rule which derives the predictable vowels

Further readingCohn 1993 Halle 1959 Harris 1994 Kahn 1976 Sapir 1925

qam lsquoto dryrsquo qamtoo lsquooverlandrsquo sarβo lsquosteeprsquokɨkʈi lsquobelchrsquo kɨβu lsquocrawlrsquo kɨɨŋguu lsquotriprsquokik lsquorubberrsquo kiʈuu lsquostudentrsquo kɪɪcuu lsquotranslatorrsquokɪɪrii lsquoroll overrsquo kiiγuu lsquowindowrsquo ku lsquoninersquokuptʃɨ lsquo900rsquo kuptʃaa lsquochairrsquo kεntʃa lsquocontractrsquokεmbo lsquoheadmanrsquo keγoslashoslash lsquohead monkrsquo kerβa lsquoaristocratrsquoqo lsquoheadrsquo qomba lsquomonasteryrsquo qɔr lsquocoatrsquoqɔɔʁɔɔ lsquoroundrsquo tʃheʁa lsquohalfrsquo tʃhuγum lsquocheesersquotopcaa lsquostairsrsquo thoʁoo lsquotonightrsquo ʈaaʁaa lsquopost officersquoʈuγɨ lsquoharborrsquo ʈuɴɢo lsquoChinarsquo nεɴɢaa lsquoimportantrsquopaɴɢɔɔ lsquochestrsquo pεεβaa lsquofrogrsquo simɢaa lsquobuild a housersquo

tʃiv lsquobaby chickenrsquo dzoslashʁi lsquobarrsquomets lsquobigrsquo biʃt lsquobladderrsquotsits lsquobreastrsquo jeχtshi lsquochurchrsquokov lsquocowrsquo dzaeligkh lsquocubrsquodʒoslashk lsquodistinctionrsquo dzy lsquoeggrsquoəsking lsquofingernailrsquo kɑnɑntʃh lsquogreenrsquozijɑn lsquoharmrsquo gjynd lsquoheaprsquotsɑk lsquoholersquo tɑkh lsquohotrsquotun lsquohousersquo thɑkhɑveɾ lsquokingrsquodeʁin lsquoyellowrsquo daelignɑg lsquoknifersquothəthev lsquolight (adj)rsquo ber lsquoloadrsquomis lsquomeatrsquo χeχtʃ lsquomiserablersquopoɾt lsquonavelrsquo thi lsquooarrsquopɑrɑv lsquoold womanrsquo dys lsquooutsidersquotseɾ lsquopeak (n)rsquo boslashχk lsquoradishrsquogjet lsquoriverrsquo toɾonrsquo lsquoRubiaceae plantrsquotsov lsquosearsquo tʃoɾt lsquoservant boyrsquogjaeligɾ lsquosheeprsquo byɾd lsquosnowstormrsquobaeligh lsquospadersquo gjoslashʁ lsquothiefrsquopuk lsquothroatrsquo χiʒ lsquotree gumrsquodoslashʁdoslashʁɑl lsquotremblersquo dʒyɾ lsquowaterrsquolen lsquowidersquo gjil lsquowolfrsquo

CHAPTER

3 Feature theory

PREVIEW

This chapter explores the theory for representing language

sounds as symbolic units You will

u see that sounds are defined in terms of a fixed set of

universal features

u learn the phonetic definitions of features and how to

assign feature values to segments based on phonetic

properties

u understand how phonological rules are formalized in

terms of these features

u see how these features make predictions about possible

sounds and rules in human language

KEY TERMSobservation

predictions

features

natural classes

We have been casual about what sounds as cognitive units are made ofand just treated them as letters labeled by traditional articulatory descrip-tions It is time now to raise a fundamental question are segments furtheranalyzed into ldquopartsrdquo that define them or are they truly atomic ndash unitswhich are not further divisible or analyzable

31 Scientific questions about speech sounds

One of the scientific questions that need to be asked about language iswhat is a possible speech sound Humans can physically produce manymore kinds of sounds than are used in language No language employshand-clapping finger-snapping or vibrations of air between the handand cheek caused by release of air from the mouth when obstructed bythe palm of the hand (though such a sound can easily communicate anattitude) A goal of a scientific theory of language is to systematize suchfacts and explain them thus we have discovered one limitation onlanguage sound and its modality ndash language sounds are produced exclu-sively within the mouth and nasal passages in the area between the lipsand larynxEven staying within the vocal tract languages also do not for

example use whistles or inhalation to form speech sounds nor is alabiolingual trill (aka ldquothe raspberryrdquo) a speech sound in any lan-guage It is important to understand that even though these variousodd sounds are not language sounds they may still be used in commu-nication The ldquoraspberryrdquo in American culture communicates a con-temptuous attitude in parts of coastal East Africa and Scandinaviainhaling with the tongue in the position for schwa expresses agree-ment Such noises lie outside of language and we never find pluralityindicated with these sounds nor are they surrounded by other soundsto form the word dog General communication has no systematic limi-tations short of anatomical ones but in language only a restrictedrange of sounds are usedThe issue of possible speech sounds is complicated by manual lan-

guages such as American Sign Language ASL is technically not a coun-terexample to a claim about modality framed in terms of ldquospeechsoundsrdquo But it is arbitrary to declare manual language to be outsidethe theory of language and facts from such languages are relevant inprinciple Unfortunately knowledge of the signed languages of theworld is very restricted especially in phonology Signed languagesclearly have syntax what isnrsquot clear is what they have by way ofphonologies Researchers have only just begun to scratch the surfaceof sign language phonologies so unfortunately we can say nothingmore about them hereThe central question is what is the basis for defining possible speech

sounds Do we use our ldquospeech anatomyrdquo in every imaginable way oronly in certain well-defined ways

311 Possible differences in soundsOne way to approach the question is to collect samples of the sounds of allof the languages in the world This search (which has never been con-ducted) would reveal massive repetition and would probably reveal thatthe segment [m] in English is exactly the same as the segment [m] inFrench German Tuumlbatuumllabal Arabic Swahili Chinese and innumerableother languages It would also reveal differences some of them perhaps abit surprising Given the richness of our transcriptional resources fornotating phonetic differences between segments you might expect thatif a collection of languages had the same vowels transcribed as [i] and [ɪ]then these vowels should sound the same This is not so

Varieties of phonetic [i] vs [ɪ] Many languages have this pair of vowelsfor example Matuumbi has [i] and [ɪ] But the actual pronunciation of [i]vs [ɪ] differs between English and Matuumbi Matuumbi [i] is higher thanin English and Matuumbi [ɪ] is a bit lower than English [ɪ] ndash to some peopleit almost sounds like [e] (but is clearly different from [e] even the ldquopurerdquo[e] found in Spanish) This might force us to introduce new symbols sothat we can accurately represent these distinctions (This is done inpublications on Matuumbi where the difference is notated as ldquoextremerdquoi u versus ldquoregularrdquo i u) Before we embark on a program of adding newsymbols we should be sure that we know how many symbols to add Itturns out that the pronunciation of [i] and [ɪ] differs in many languagesthese vowels exist in English Kamba Lomwe Matuumbi Bari KipsigisDidinga and Sotho and their actual pronunciation differs in eachlanguageYou do not have to go very far into exotic languages to find this phonetic

difference for the difference between English [i] and German [i] is alsovery noticeable and is something that a language learner must master todevelop a good German or English accent Although the differencesmay bedifficult for the untrained ear to perceive at first they are consistentphysically measurable and reproducible by speakers If written symbolsare to represent phonetic differences between languages a totally accuratetranscription should represent these differences To represent just thisrange of vowel differences involving [i] and [ɪ] over a dozen new symbolswould need to be introduced Yet we do not introduce large numbers ofnew symbols to express these differences in pronunciations becausephonological symbols do not represent the precise phonetic properties ofthe sounds in a language they only represent the essential contrastbetween sounds

Other variants of sounds Similar variation exists with other phoneticcategories The retroflex consonants of Telugu Hindi and Koti are allpronounced differently Hindi has what might be called ldquomildrdquo retroflex-ion where the tip of the tongue is placed just behind the alveolar ridgewhile in Telugu the tip of the tongue is further back and contact is madebetween the palate and the underside of the tongue (sublaminal) in Kotithe tongue is placed further forward but is also sublaminal Finnish

Feature theory 41

Norwegian and English contrast the vowels [a] and [aelig] but in each ofthese languages the vowels are pronounced in a slightly different way Thevoiced velar fricative [γ] found in Arabic Spanish and the Kurdish lan-guage Hawrami are all phonetically different in subtle but audible ways

The important details of speech Although languages can differ sub-stantially in the details of how their sounds are pronounced there arelimits on the types of sound differences which can be exploited contras-tively ie can form the basis for making differences in meaningLanguage can contrast tense [i] and lax [ɪ] but cannot further contrast ahyper-tense high vowel (like that found in Matuumbi) which we mightwrite as [i+] with plain tense [i] as in English or hyper-lax [ɪ-] as inMatuumbi with plain lax [ɪ] as found in English Within a language youfind at most [i] vs [ɪ] Languages can have one series of retroflex conson-ants and cannot contrast Hindi-style [ʈ] with a Telugu-style phonemewhich we might notate as [ʈ+] The phonology simply has ldquoretroflexrdquoand it is up to the phonetic component of a language to say exactly howa retroflex consonant is pronouncedIt is important to emphasize that such phonetic details are not too subtle

to hear The difference between various types of retroflex consonants isquite audible ndash otherwise people could not learn the typical pronunciationof retroflex consonants in their language ndash and the difference betweenEnglish and German [i] is appreciable Children learning German can hearand reproduce German [i] accurately Speakers can also tell when someonemispronounces a German [i] as an English [i] and bilingual GermanndashEnglish speakers can easily switch between the two phonetic vowelsOne thing that phonological theory wants to know is what is a possible

phoneme How might we answer this We could look at all languages andpublish a list A monumental difficulty with that is that there are nearly7000 languages but useful information on around only 10 percent ofthese languages Worse this could only say what phonemic contrastshappen to exist at the present A scientific account of language does notjust ask what has been actually observed it asks about the fundamentalnature of language including potential sounds which may have existed in alanguage spoken 1000 years ago or some future language which will bespoken 1000 years hence We are not just interested in observation we areinterested in predictionIn this connection consider whether a ldquobilabial clickrdquo is a possible

phoneme We symbolize it as [ʘ] ndash it is like a kiss but with the lips flatas for [m] not protruded as for [w] Virtually all languages have bilabialconsonants and we know of dozens of languages with click consonants(Dahalo Sotho Zulu Xhosa Khoekhoe) so the question is whether thecombination of concepts ldquobilabialrdquo and ldquoclickrdquo can define a phoneme As ithappens we know that such a sound does exist but only in two closelyrelated languages Xoo and Eastern 6frac14Hoan members of the Khoisanlanguage family These languages have under 5000 speakers combinedand given socioeconomic factors where these languages are spoken(Namibia and Botswana) it is likely that the languages will no longer be

spoken in 200 years We are fortunate in this case that we have infor-mation on these languages which allows us to say that this is a phonemebut things could have turned out differently The languages could easilyhave died out without having been recorded and then we would wronglyconclude that a bilabial click is not a possible phoneme because it has notbeen observed We need a principled theoretical basis for saying what wethink might be observed

Predictions versus observations A list of facts is scientifically uninter-esting A basic goal of science is to have knowledge that goes beyond whathas been observed because we believe that the universe obeys generallaws A list might be helpful in building a theory but we would not wantto stop with a list because it would give us no explanation why thatparticular list as opposed to some other arbitrary list should constitutethe possible phonemes of language The question ldquowhat is a possiblephonemerdquo should thus be answered by reference to a general theory ofwhat speech sounds are made of just as a theory of ldquopossible atomsrdquo isbased on a general theory of what makes up atoms and rules for puttingthose bits together Science is not simply the accumulation and sorting offacts but rather the attempt to discover laws that regulate the universeSuch laws make predictions about things that we have yet to observecertain things should be found other things should never be foundThe Law of Gravity predicts that a rock will fall to earth which says

what it will do and by implication what it will not do it also wonrsquot go upor sideways Physicists have observed that subatomic particles decay intoother particles Particles have an electrical charge ndash positive negative orneutral ndash and there is a physical law that the charge of a particle ispreserved when it decays (adding up the charges of the decay products)The particle known as a ldquokaonrdquo (K) can be positive (K+) negative (K-) orneutral (K0) a kaon can decay into other particles known as ldquopionsrdquo (π)which also can be positive (π+) negative (π-) or neutral (π0) Thus a neutralkaon may become a positive pion and a negative pion (K0 π+ + π-) or itmay become one positive one negative and one neutral pion (K0 π+ +π- + π0) because in both cases the positives and negatives cancel out andthe sum of charges is neutral (0) The Law of Conservation of Chargeallows these patterns of decay and prohibits a neutral kaon from becom-ing two positive pions (K0 π+ + π+) In the myriad cases of particle decaywhich have been observed experimentally none violates this law whichpredicts what can happen and what cannotAnalogously phonological theory seeks to discover the laws for build-

ing phonemes which predict what phonemes can be found in languagesWe will see that theory after considering a related question which definesphonology

312 Possible rulesPrevious chapters have focused on rules but we havenrsquot paid muchattention to how they should be formulated English has rules definingallowed clusters of two consonants at the beginning of the word The first

Feature theory 43

set of consonant sequences in (1) is allowed whereas the second set ofsequences is disallowed

(1) pr pl br bl tr dr kr kl gr glrp lp rb lb rt rd rk lk rg lg

This restriction is very natural and exists in many languages ndash but it is notinevitable and does not reflect any insurmountable problems of physi-ology or perception Russian allows many of these clusters for example[rtutj] lsquomercuryrsquo exemplifies the sequence [rt] which is impossible inEnglishWe could list the allowed and disallowed sequences of phonemes and

leave it at that but this does not explainwhy these particular sequences areallowed Why donrsquot we find a language which is like English except thatthe specific sequence [lb] is allowed and the sequence [bl] is disallowed Aninteresting generalization regarding sequencing has emerged after com-paring such rules across languages Some languages (eg Hawaiian) do notallow any clusters of consonants and some (Bella Coola a Salishan languageof British Columbia) allow any combination of two consonants but nolanguage allows initial [lb] without also allowing [bl] This is a more inter-esting and suggestive observation since it indicates that there is somethingabout such sequences that is not accidental in English but it is still just arandom fact from a list of accumulated facts if we have no basis forcharacterizing classes of sounds and view the restrictions as restrictionson letters as sounds with no structureThere is a rule in English which requires that all vowels be nasalized

when they appear before a nasal consonant and thus we have a rulesomething like (2)

If rules just replace one arbitrary list of sounds by another list when theystand in front of a third arbitrary list we have to ask why these particularsets of symbols operate together Could we replace the symbol [n] with thesymbol [tʃ] or the symbol [otilde] with the symbol [oslash] and still have a rule insome language It is not likely to be an accident that these particularsymbols are found in the rule a rule similar to this can be found in quite anumber of languages and we would not expect this particular collectionof letters to assemble themselves into a rule in many languages if thesewere just random collections of lettersWere phonological rules stated in terms of randomly assembled sym-

bols there would be no reason to expect (3a) to have a different statusfrom (3b)

(3) a p t tʃ k m n ɲ ŋ _ m n ɲ ŋb b p d q d q b p _ s x o ɪ

ε e ɪ i ɛ e ɪ ĩa ɔ o ʊ atilde ɔ otilde ʊ _ m n ŋu ə aelig ũ ə aelig

This is not the onlyrule governingconsonantsequences at thebeginning of theword in English sofor example thevoiceless alveolarfricative [s] can befollowed by anynonfricative

Rule (3a) ndash nasalization of stops before nasals ndash is quite common but (3b) isnever found in human language This is not an accident but rather reflectsthe fact that the latter process cannot be characterized in terms of a unifiedphonetic operation applying to a phonetically defined context The insightwhich we have implicitly assumed and make explicit here is that rulesoperate not in terms of specific symbols but in terms of definable classesThe basis for defining those classes is a set of phonetic propertiesAs a final illustration of this point rule (4a) is common in the worldrsquos

languages but (4b) is completely unattested

(4) a k g tʃ dʒ _ i eb p r i b _ o n

The first rule refers to phonetically definable classes of segments (velarstops alveopalatal affricates front vowels) and the nature of the changeis definable in terms of a phonetic difference (velars change place ofarticulation and become alveopalatals) The second rule cannot be charac-terized by phonetic properties the sets p r i b and o n are notdefined by some phonetic property and the change of [p] to [i] and [r] to [b]has no coherent phonetic characterizationThe lack of rules like (4b) is not just an isolated limitation of knowledge ndash

itrsquos not simply that we havenrsquot found the specific rules (4b) but we havefound (4a) ndash but rather these kinds of rules represent large systematicclasses (3b) and (4b) represent a general kind of rule where classes ofsegments are defined arbitrarily Consider the constraint on clusters oftwo consonants in English In terms of phonetic classes this reduces to thesimple rule that the first consonant must be a stop and the second conson-ant must be a liquid The second rule changes vowels into nasalized vowelsbefore nasal consonants The basis for defining these classes will be con-sidered now

32 Distinctive feature theory

Just saying that rules are defined in terms of phonetic properties is too broada claim since it says nothing about the phonetic properties that are relevantConsider a hypothetical rule stated in terms of phonetic properties

all vowels change place of articulation so that the original difference informant frequency between F1 and F3 is reduced to half what it originallywas when the vowel appears before a consonant whose duration rangesfrom 100 to 135 ms

What renders this rule implausible (no language has one vaguely resem-bling it) is that it refers to specific numerical durations and to thedifference in frequency between the first and third formantAn acoustic description considers just physical sound but a perceptual

description factors in the question of how the ear and brain processsound The difference between 100 Hz and 125 Hz is acoustically the same

Feature theory 45

as that between 5100 Hz and 5125 Hz The two sets are perceptually verydifferent the former being perceived as ldquomore separaterdquo and the latter asvirtually indistinguishableThe phonetic properties which are the basis of phonological systems are

general and somewhat abstract such as voicing or rounding and arelargely the categories which we have informally been using already theyare not the same as we will see The hypothesis of distinctive featuretheory is that there is a small set around two dozen of phonetically basedproperties which phonological analysis uses These properties thedistinctive features not only define the possible phonemes of humanlanguages but also define phonological rulesThe classical statement of features derives from Chomsky and Halle

(1968) We will use an adapted set of these features which takes intoconsideration refinements Each feature can have one of two values plusand minus so for each speech sound the segment either has the property(is [+Fi]) or lacks the property (is [-Fi]) In this section we follow Chomskyand Halle (1968) and present the generally accepted articulatory correlatesof the features that is what aspects of production the feature relates toThere are also acoustic and perceptual correlates of features pertaining towhat the segment sounds like which are discussed by Jakobson Fant andHalle (1952) using a somewhat different system of features

321 Phonetic preliminariesBy way of phonetic background to understanding certain features twophonetic points need to be clarified First some features are characterizedin terms of the ldquoneutral positionrdquo which is a configuration that the vocaltract is assumed to have immediately prior to speaking The neutralposition approximately that of the vowel [ε] defines relative movementof the tongueSecond you need to know a bit about how the vocal folds vibrate since

some feature definitions relate to the effect on vocal fold vibration(important because it provides most of the sound energy of speech)The vocal folds vibrate when there is enough air pressure below theglottis (the opening between the vocal folds) to force the vocal foldsapart This opening reduces subglottal pressure which allows the foldsto close and this allows air pressure to rebuild to the critical level wherethe vocal folds are blown apart again The critical factor that causes thefolds to open is that the pressure below the vocal folds is higher than thepressure aboveAir flows from the lungs at a roughly constant rate Whether there is

enough drop in pressure for air to force the vocal folds open is thusdetermined by the positioning and tension of the vocal folds (how hardit is to force them apart) and the pressure above the glottis The pressureabove the glottis depends on how effectively pressure buildup can berelieved and this is determined by the degree of constriction in the vocaltract In short the configuration of the vocal folds and the degree andlocation of constriction above the glottis almost exclusively determinewhether there will be voicing

If the pressure above and below the glottis is nearly equal air stopsflowing and voicing is blocked So if the vocal tract is completelyobstructed (as for the production of a voiceless stop like [k]) air flowingthrough the glottis rapidly equalizes the pressure below and above theglottis which stops voicing On the other hand if the obstruction in thevocal tract is negligible (as it is in the vowel [a]) the pressure differentialneeded for voicing is easily maintained since air passing through theglottis is quickly vented from the vocal tractA voiced stop such as [g] is possible even though it involves a total

obstruction of the vocal tract analogous to that found in [k] because ittakes time for pressure to build up in the oral cavity to the point thatvoicing ceases Production of [g] involves ancillary actions to maintainvoicing The pharynx may be widened which gives the air more room toescape delaying the buildup of pressure The larynx may be loweredwhich also increases the volume of the oral cavity the closure for thestop may be weakened slightly allowing tiny amounts of air to flowthrough the velum may be raised somewhat to increase the size of theair cavity or it may be lowered somewhat to allow small (usually imper-ceptible) amounts of air to pass through the nose The duration of theconsonant can be reduced ndash generally voiced stops are phoneticallyshorter than corresponding voiceless stopsCertain sounds such as vowels lack a radical constriction in the vocal

tract so it is quite easy to maintain voicing during such sounds whereaswith other sounds specifically obstruents voicing is difficult to maintainSome accounts of this distinction especially that of Chomsky and Halle(1968) refer to ldquospontaneous voicingrdquo which is grounded on the assump-tion that voicing occurs automatically simply by positioning the vocalfolds in what we might call the ldquodefaultrdquo position For sounds that involvea significant obstruction of the vocal tract special actions are required forvoicing The features [sonorant] and [consonantal] directly relate to theobstruction in the vocal tract which determines whether the vocal foldsvibrate spontaneously

322 Major class featuresOne of the most intuitive distinctions which feature theory needs tocapture is that between consonants and vowels There are three featuresthe so-called major class features which provide a rough first grouping ofsounds into functional types that includes the consonantvoweldistinction

syllabic (syl) forms a syllable peak (and thus can be stressed)sonorant (son) sounds produced with a vocal tract configuration in whichspontaneous voicing is possible

consonantal (cons) sounds produced with a major obstruction in the oralcavity

The feature [syllabic] is unfortunately simultaneously one of themost important features and one of the hardest to define physicallyIt corresponds intuitively to the notion ldquoconsonantrdquo (where [h] [ j] [m]

Feature theory 47

[s] [t] are ldquoconsonantsrdquo) versus ldquovowelrdquo (such as [a] [i]) indeed the onlydifference between the vowels [i u] and the corresponding glides [ j w] isthat [i u] are [+syllabic] and [ j w] are [-syllabic] The feature [syllabic]goes beyond the intuitive vowelconsonant split English has syllabicsonorants such as [r] [l] [n] The main distinction between the Englishwords (American English pronunciation) ear [ɪr] and your [ jr] resides inwhich segments are [+syllabic] versus [-syllabic] In ear the vowel [ɪ] is[+syllabic] and [r] is [-syllabic] whereas in your [ j] is [-syllabic] and [r]is [+syllabic] The words eel [il] and the reduced form of yoursquoll [ jl] for manyspeakers of American English similarly differ in that [i] is the peak of thesyllable (is [+syllabic]) in eel but [l] is the syllable peak in yoursquoll

Other languages have syllabic sonorants which phonemically contrastwith nonsyllabic sonorants such as Serbo-Croatian which contrastssyllabic [r] with nonsyllabic [r] (cf groze lsquofear (gen)rsquo versus groce lsquolittlethroatrsquo) Swahili distinguishes [mbuni] lsquoostrichrsquo and [mbuni] lsquocoffee plantrsquoin the fact that [mbuni] is a three-syllable word and [m] is the peak(the only segment) of that first syllable but [mbuni] is a two-syllableword whose first syllable peak is [u] Although such segments may bethought of as ldquoconsonantsrdquo in one intuitive sense of the concept theyhave the feature value [+syllabic] This is a reminder that there is adifference between popular concepts about language and technical termsldquoConsonantrdquo is not strictly speaking a technical concept of phonologicaltheory even though it is a term quite frequently used by phonologists ndashalmost always with the meaning ldquononpeakrdquo in the syllable ie a[-syllabic] segmentThe definition of [sonorant] could be changed so that glottal configur-

ation is also included then the laryngeals would be [ndashsonorant] There islittle compelling evidence to show whether this would be correct laterwe discuss how to go about finding such evidence for revising featuredefinitionsThe feature [sonorant] captures the distinction between segments

such as vowels and liquids where the constriction in the vocal tract issmall enough that no special effort is required to maintain voicing asopposed to sounds such as stops and fricatives which have enoughconstriction that effort is needed to maintain voicing In an oral stopair cannot flow through the vocal tract at all so oral stops are[ndashsonorant] In a fricative even though there is some airflow there isso much constriction that pressure builds up with the result thatspontaneous voicing is not possible thus fricatives are [ndashsonorant] Ina vowel or glide the vocal tract is only minimally constricted so air canflow without impedance vowels and glides are therefore [+sonorant]A nasal consonant like [n] has a complete obstruction of airflowthrough the oral cavity but nevertheless the nasal passages are openwhich allows free flow of air Air pressure does not build up during theproduction of nasals so nasals are [+sonorant] In the liquid [l] there isa complete obstruction formed by the tip of the tongue with thealveolar ridge but nevertheless air flows freely over the sides of thetongue so [l] is [+sonorant]

The question whether r is [+sonorant] or [-sonorant] has no simpleanswer since many phonetically different segments are transcribed as rsome are [-sonorant] and some are [+sonorant] depending on theirphonetic properties The so-called fricative r of Czech (spelled ř) has aconsiderable constriction so it is [-sonorant] but the English type [ɹ] isa sonorant since there is very little constriction In other languages theremay be more constriction but it is so brief that it does not allow signifi-cant buildup of air pressure (this would be the case with ldquotappedrdquo rrsquos)Even though spontaneous voicing is impossible for the laryngeal conson-ants [h ʔ] because they are formed by positioning the vocal folds so thatvoicing is precluded they are [+sonorant] since they have no constrictionabove the glottis which is the essential property defining [+sonorant]The feature [consonantal] is very similar to the feature [sonorant] but

specifically addresses the question of whether there is any major constric-tion in the oral cavity This feature groups together obstruents liquids andnasals which are [+consonantal] versus vowels glides and laryngeals([h ʔ]) which are [-consonantal] Vowels and glides have a minor obstruc-tion in the vocal tract compared to that formed by a fricative or a stopGlottal stop is formed with an obstruction at the glottis but none in thevocal tract hence it is [-consonantal] In nasals and liquids there is anobstruction in the oral cavity even though the overall constriction of thewhole vocal tract is not high enough to prevent spontaneous voicingRecent research indicates that this feature may not be necessary since itsfunction is usually covered as well or better by other featuresThe most important phonological use of features is that they identify

classes of segments in rules All speech sounds can be analyzed in terms oftheir values for the set of distinctive features and the set of segments thathave a particular value for some feature (or set of feature values) is a naturalclass Thus the segments [a i r m] are members of the [+syllabic] class and[ j h ʔ r m s p] are members of the [-syllabic] class [a r j ʔ r m] are in the[+sonorant] class and [s z p b] are in the [-sonorant] class [a i w h ʔ] are in the[-consonantal] class and [r m r m s p] are in the [+consonantal] class Naturalclasses can be defined in terms of conjunctions of features such as[+consonantal -syllabic] which refers to the set of segments which aresimultaneously [+consonantal] and [-syllabic]When referring to segments defined by a combination of features the

features are written in a single set of brackets ndash [+cons -syl] refers toa single segment which is both +consonantal and -syllabic while [+cons][ndashsyl] refers to a sequence of segments the first being +consonantal and thesecond being -syllabicAccordingly the three major class features combine to define five

maximally differentiated classes exemplified by the following segmentgroups

(5) a i u r l m j w h ʔ r l m s z p bsyllabic + + - - -sonorant + + + + -consonantal - + - + +

Feature theory 49

Further classes are definable by omitting specifications of one or more ofthese features for example the class [-syllabic +sonorant] includes j wh ʔ r l mOne thing to note is that all [+syllabic] segments ie all syllable peaks

are also [+sonorant] It is unclear whether there are syllabic obstruentsie [s] [k] It has been claimed that such things exist in certain dialects ofBerber but their interpretation remains controversial since the prin-ciples for detection of syllables are controversial Another gap is thecombination [-sonorant -consonantal] which would be a physicalimpossibility A [-sonorant] segment would require a major obstructionin the vocal tract but the specification [-consonantal] entails that theobstruction could not be in the oral cavity The only other possibilitywould be constriction of the nasal passages and nostrils are not suffi-ciently constrictable

323 Place of articulationFeatures to define place of articulation are our next functional set Webegin with the features typically used by vowels specifically the[+syllabic -consonantal +sonorant] segments and then proceed to con-sonant features ending with a discussion of the intersection of thesefeatures

Vowel place features The features which define place of articulationfor vowels are the following

high the body of the tongue is raised from the neutral positionlow the body of the tongue is lowered from the neutral positionback the body of the tongue is retracted from the neutral positionround the lips are protrudedtense sounds requiring deliberate accurate maximally distinct gesturesthat involve considerable muscular effort

advanced tongue root produced by drawing the root of the tongueforward

The main features are [high] [low] [back] and [round] Phonologistsprimarily distinguish just front and back vowels governed by [back] frontvowels are [-back] since they do not involve retraction of the tongue bodyand back vowels are [+back] Phonetic central vowels are usually treatedas phonological back vowels since typically central vowels are unroundedand back vowels are rounded Distinctions such as those between [ɨ] and[ɯ] [ɜ] and [ʌ] [y] and [ʉ] [ʚ] and [œ] or [a] and [ɑ] are usually considered tobe phonologically unimportant over-differentiations of language-specificphonetic values of phonologically back unrounded vowels The phonolo-gically relevant question about a vowel pronounced as [ʉ] is not whetherthe tongue position is intermediate between that of [i] and [u] butwhether it patterns with i e y oslash or with u ɯ o ʌ ndash or does it patternapart from either set In lieu of clear examples of a contrast betweencentral and back rounded vowels or central and back unrounded vowelswe will not at the moment postulate any other feature for the frontndashback

dimension though section 36 considers possible evidence for the phono-logical relevance of the concept ldquocentral vowelrdquo Given the phonologicallyquestionable status of distinctive central vowels no significance should beattributed to the use of the symbol [ɨ] versus [ɯ] and typographic conveni-ence may determine that a [+back -round] high vowel is typically tran-scribed as [ɨ]Two main features are employed to represent vowel height High

vowels are [+high] and [-low] low vowels are [+low] and [-high] Novowel can be simultaneously [+high] and [+low] since the tongue cannotbe raised and lowered simultaneously mid vowels are [-high -low] Inaddition any vowel can be produced with lip rounding using the feature[round] These features allow us to characterize the following vowelcontrasts

Note that [ɑ] is a back low unrounded vowel in contrast to the symbol [ɒ]for a back low rounded vowelVowels with a laxer ldquoless deliberaterdquo and lower articulation such as [ɪ]

in English sit or [ε] in English set would be specified as [-tense]

Korean has a set of so-called ldquotenserdquo consonants but these are phoneticallyldquoglottalrdquo consonantsOne question which has not been resolved is the status of low vowels in

terms of this feature Unlike high and mid vowels there do not seem to beanalogous contrasts in low vowels between tense and lax [aelig] Anotherimportant point about this feature is that while [back] [round] [high] and[low] will also play a role in defining consonants [tense] plays no role inconsonantal contrastsThe difference between i and ɪ or e and εhas also been considered to be one

of vowel height (proposed in alternative models where vowel height isgoverned by a single scalar vowel height feature rather than by the binaryfeatures [high] and [low]) This vowel contrast has also been described interms of the feature ldquoAdvanced Tongue Rootrdquo (ATR) especially in the vowelsystems of languages of Africa and Siberia There has been debate over thephonetic difference between [ATR] and [tense] Typically [+tense] frontvowels are fronter than their lax counterparts and [+tense] back vowels

i y ɨ u e oslash ə o aelig œ ɑ ɒhigh + + + + - - - - - - - -low - - - - - - - - + + + +back - - + + - - + + - - + +round - + - + - + - + - + - +

ɪ ʏ ᵻ ʊ ε œ ʌ ɔhigh + + + + - - - -low - - - - - - - -back - - + + - - + +round - + - + - + - +tense - - - - - - - -

Feature theory 51

are backer than their lax counterparts In comparison [+ATR] vowels aresupposed to be generally fronter than corresponding [-ATR] vowels so that[+ATR] back vowels are phonetically fronter than their [-ATR] counterpartsHowever some articulatory studies have shown that the physical basis forthe tenselax distinction in English is no different from that which ATR isbased onUnfortunately the clearest examples of the feature [ATR] are foundin languages of Africa where very little phonetic research has been doneSince no language contrasts both [ATR] and [tense] vowels it is usuallysupposed that there is a single feature whose precise phonetic realizationvaries somewhat from language to language

Consonant place features The main features used for defining conson-antal place of articulation are the following

coronal produced with the blade or tip of the tongue raised from theneutral position

anterior produced with a major constriction located at or in front of thealveolar ridge

strident produced with greater noisinessdistributed produced with a constriction that extends for a considerabledistance along the direction of airflow

Place of articulation in consonants is primarily described with the fea-tures [coronal] and [anterior] Labials labiodentals dentals and alveolarsare [+anterior] since their primary constriction is at or in front of thealveolar ridge (either at the lips the teeth or just back of the teeth)whereas other consonants (including laryngeals) are [-anterior] sincethey lack this front constriction The best way to understand this featureis to remember that it is the defining difference between [s] and [ ʃ ] where[s] is [+anterior] and [ ʃ ] is [-anterior] Anything produced where [s] isproduced or in front of that position is [+anterior] anything producedwhere [ ʃ ] is or behind [ ʃ ] is [-anterior]

Remember that the two IPA letters lttʃgt represent a single [-anterior]segment not a combination of [+anterior] [t] and [-anterior] [ ʃ ]Consonants which involve the blade or tip of the tongue are [+coronal]

and this covers the dentals alveolars alveopalatals and retroflex conson-ants Consonants at other places of articulation ndash labial velar uvular andlaryngeal ndash are [-coronal] Note that this feature does not encompass thebody (back) of the tongue so while velars and uvulars use the tongue theyuse the body of the tongue rather than the blade or tip and therefore are[-coronal] The division of consonants into classes as defined by [coronal]is illustrated below

[+anterior] [-anterior]f φ p θ s t t ʃ tʃ ʂ ʈ ccedil x k q ʕ h ʔ

[+coronal] [-coronal]t θ t s ʃ n l r ɲ ʈ p φ f k q ʕ

Two other features are important in characterizing the traditional placesof articulation The feature [distributed] is used in coronal sounds todistinguish dental [t] from English alveolar [t] or alveopalatal [ ʃ ] fromretroflex [ʂ] the segments [t ʃ] are [+distributed] and [t ʈ ʂ] are[-distributed] The feature [distributed] as applied to coronal consonantsapproximately corresponds to the traditional phonetic notion ldquoapicalrdquo([-distributed]) versus ldquolaminalrdquo ([+distributed]) This feature is not rele-vant for velar and labial sounds and we will not specify any value of[distributed] for noncoronal segmentsThe feature [strident] distinguishes strident [f s] from nonstrident [φ θ]

otherwise the consonants [f φ] would have the same feature specifica-tions Note that the feature [strident] is defined in terms of the aerody-namic property of greater turbulence (which has the acoustic correlate ofgreater noise) not in terms of the movement of a particular articulator ndashthis defining characteristic is accomplished by different articulatory con-figurations In terms of contrastive usage the feature [strident] onlyserves to distinguish bilabial and labiodentals or interdentals and alveo-lars A sound is [+strident] only if it has greater noisiness and ldquogreaterrdquoimplies a comparison In the case of [φ] vs [f] [β] vs [v] [θ] vs [s] or [eth] vs[z] the second sound in the pair is noisier No specific degree of noisinesshas been proposed which would allow you to determine in isolationwhether a given sound meets the definition of strident or not Thus it isimpossible to determine whether [ ʃ ] is [+strident] since there is nocontrast between strident and nonstrident alveopalatal sounds Thephoneme [ ʃ ] is certainly relatively noisy ndash noisier than [θ] ndash but then [θ]is noisier than [φ] is[Strident] is not strictly necessary for making a distinction between [s]

and [θ] since [distributed] also distinguishes these phonemes Since [stri-dent] is therefore only crucial for distinguishing bilabial and labial frica-tives it seems questionable to postulate a feature with such broadimplications solely to account for the contrast between labiodental andbilabial fricatives Nonetheless we need a way of representing this con-trast The main problem is that there are very few languages (such as EweVenda and Shona) which have both [f] and [φ] or [v] and [β] and thephonological rules of these languages do not give us evidence as to howthis distinction should bemade in terms of featuresWewill therefore onlyinvoke the feature [strident] in connection with the [φ β] vs [f v] contrastUsing these three features consonantal places of articulation can be

partially distinguished as follows

Vowel features on consonants The features [high] [low] [back] and[round] are not reserved exclusively for vowels and these typical vowelfeatures can play a role in defining consonants as well As we see in (10)

p t t tʃ ʈ c k q ʕ ʔanterior + + + - - -coronal - + + + + -distributed + - + - -

Feature theory 53

velar uvular pharyngeal and glottal places of articulation are not yetdistinguished this is where the features [high] [low] and [back] becomeimportant Velar uvular and pharyngeal consonants are [+back] sincethey are produced with a retracted tongue body The difference betweenvelar and uvular consonants is that with velar consonants the tongue bodyis raised whereas with uvular consonants it is not and thus velars are[+high] where uvulars are [-high] Pharyngeal consonants are distin-guished from uvulars in that pharyngeals are [+low] and uvulars are[-low] indicating that the constriction for pharyngeals is even lower thanthat for uvularsOne traditional phonetic place of articulation for consonants is that of

ldquopalatalrdquo consonants The term ldquopalatalrdquo is used in many ways for examplethe postalveolar or alveopalatal (palatoalveolar) consonants [ ʃ ] and [tʃ] mightbe referred to as palatals This is strictly speaking a misnomer and the termldquopalatalrdquo is best used only for the ldquotrue palatalsrdquo transcribed as [c ccedil ɟ] Suchconsonants are found in Hungarian and also in German in words like [iccedil] lsquoIrsquoor inNorwegian [ccediloslashper] lsquobuysrsquo These consonants are producedwith the bodyof the tongue raised and fronted and therefore they have the feature values[+high -back] The classical feature system presented here provides noway to distinguish such palatals from palatalized velars ([kj]) either phonet-ically or phonologically Palatalized (fronted) velars exist as allophonic vari-ants of velars before front vowels in English eg [kjip] lsquokeeprsquo they arearticulatorily and acoustically extremely similar to the palatals of Hungar-ian Very little phonological evidence is available regarding the treatment ofldquopalatalsrdquo versus ldquopalatalized velarsrdquo it is quite possible that [c] and [kj] or [ccedil]and [xj] are simply different symbols chosen on the basis of phonologicalpatterning rather than systematic phonetic differencesWith the addition of these features the traditional places of articula-

tion for consonants can now be fully distinguished

The typical vowel features have an additional function as applied toconsonants namely that they define secondary articulations such as pal-atalization and rounding Palatalization involves superimposing theraised and fronted tongue position of the glide [ j] onto the canonicalarticulation of a consonant thus the features [+high -back] are addedto the primary features that characterize a consonant (those being thefeatures that typify [i j]) So for example the essential feature character-istics of a bilabial are [+anterior -coronal] and they are only incidentally[-high -back] A palatalized bilabial would be [+anterior -coronal+high -back] Velarized consonants have the features [+high +back]

p t t tʃ ʈ c kj k q ʕ ʔanterior + + + - - - - - - -coronal - + + + + - - - - -distributed + - + -high - - - - - + + - - -back - - - - - - + + + -low - - - - - - - - + -

analogous to the features of velar consonants pharyngealized consonantshave the features [+back +low] Consonants may also bear the feature[round] Applying various possible secondary articulations to labial con-sonants results in the following specifications

Labialized (pw) palatalized (pj) velarized (pγ) and pharyngealized (pʕ)variants are the most common categories of secondary articulationUvularized consonants ie pq are rare uvularized clicks are attested inJursquohoansi It is unknown if there is a contrast between rounded conson-ants differing in secondary height symbolized above as pw vs po or pɥ vsposlash Feature theory allows such a contrast so eventually we ought tofind examples If as seems likely after some decades of researchsuch contrasts do not exist where predicted there should be a revisionof the theory so that the predictions of the theory better matchobservationsThis treatment of secondary articulations makes other predictions One

is that there cannot be palatalized uvulars or pharyngeals This followsfrom the fact that the features for palatalization ([+high -back]) conflictwith the features for uvulars ([-high +back]) and pharyngeals ([-high+back +low]) Since such segments do not appear to exist this supportsthe theory otherwise we expect ndash in lieu of a principle that prohibitsthem ndash that they will be found in some language Second in this theorya ldquopurerdquo palatal consonant (such as Hungarian [ ɟ]) is equivalent to apalatalized (ie fronted) velar Again since no language makes a contrastbetween a palatal and a palatalized velar this is a good prediction ofthe theory (unless such a contrast is uncovered in which case it becomesa bad prediction of the theory)

324 Manner of articulationOther features relate to the manner in which a segment is produced apartfrom the location of the segmentrsquos constriction The manner features are

continuant (cont) the primary constriction is not narrowed so much thatairflow through the oral cavity is blocked

delayed release (delrel) release of a total constriction is slowed so that africative is formed after the stop portion

nasal (nas) the velum is lowered which allows air to escape through thenose

lateral (lat) the mid section of the tongue is lowered at the side

The feature [continuant] groups together vowels glides fricatives and [h]as [+continuant] Note that [continuant] is a broader group than the trad-itional notion ldquofricativerdquo which refers to segments such as [s] [ ʃ ] or [θ]

p pj pγ pw pɥ pʕ pq po poslash

high - + + + + - - - -back - - + + - + + + -low - - - - - + - - -round - - - + + - - + +

Feature theory 55

The term ldquofricativerdquo generally refers to nonsonorant continuants ie theclass defined by the conjunction of features [+continuant -sonorant]Since continuants are defined as sounds where air can flow continuouslythrough the oral cavity nasals like [m n ŋ] are [-continuant] even thoughthey allow continuous airflow (through the nose)Affricates such as [tʃ pf] are characterized with the feature [+delayed

release] Necessarily all affricates are [-continuant] since they involvecomplete constriction followed by a period of partial fricative-like con-striction and therefore they behave essentially as a kind of stop Thisfeature is in question since [pf tʃ kx] do not act as a unified phonologicalclass nevertheless some feature is needed to characterize stops versusaffricates Various alternatives have been proposed for example that [kx]might just be the pronunciation of aspirated [kh] since velar [kx] and [kh]never seem to contrast perhaps the feature [strident] defines [ts] vs [t] Theproper representation of affricates is a currently unresolved issue inphonologyThe feature [+nasal] is assigned to sounds where air flows through the

nasal passages for example [n] as well as nasalized vowels like [atilde] Liquidsand fricatives can be nasalized as well but the latter especially are quiterare L-like sounds are characterized with the feature [lateral] Almost all[+lateral] sounds are coronal though there are a few reports of velarlaterals Detailed information on the phonetics and phonology of thesesegments is not availableExamples of the major manners of articulation are illustrated below for

coronal place of articulation

325 Laryngeal featuresThree features characterize the state of the glottis

spread glottis (sg) the vocal folds are spread far apartconstricted glottis (cg) the vocal folds are tightly constrictedvoice (voi) the vocal folds vibrate

Voiced sounds are [+voice] The feature [spread glottis] describes aspiratedobstruents ([ph] [bh]) and breathy sonorants ([m] [a]) [constricted glottis]describes implosives ([ɓ]) ejective obstruents ([prsquo]) and laryngealizedsonorants ([m] [a])How to distinguish implosives from ejectives is not entirely obvious but

the standard answer is that ejectives are [-voice] and implosives are[+voice] There are two problems with this One is that implosives donot generally pattern with other [+voiced] consonants in phonological

t n ts s l l tl

delayed release - - + - - - +continuant - - - + + + +lateral - - - - + + +nasal - + - - - + -

systems especially in how consonants affect tone (voiced consonants buttypically not implosives may lower following tones) The second is thatNgiti and Lendu have both voiced and voiceless implosives The languageslack ejectives which raises the possibility that voiceless implosives arephonologically [-voice +cg] which is exactly the specification given toejective consonants You may wonder how [-voice +cg] can be realized asan ejective in languages like Navajo Tigre or Lushootseed and as avoiceless implosive in Ngiti or Lendu This is possible because featurevalues give approximate phonetic descriptions not exact ones The Koreanldquofortisrdquo consonants found in [krsquoata] lsquopeel (noun)rsquo [akrsquoi] lsquomusical instru-mentrsquo or [altrsquoa] lsquobe illrsquo are often described as glottalized and phoneticstudies have shown that they are produced with glottal constrictions thusthey would be described as [-voice +cg] Nevertheless they are notejectives Similarly Khoekhoe (Nama) has a contrast between plain clicks([agraveḿ] lsquodeeprsquo) and glottalized ones ([rsquoaacutem] lsquokillrsquo) but the glottalized clicksrealize the feature [+cg] as a simple constriction of the glottis notinvolving an ejective releaseThe usual explanation for the difference between ejectives in Navajo

and glottalized nonejective consonants in Korean or Khoekhoe is that theyhave the same phonological specifications [-voice +cg] but realize thefeatures differently due to language-specific differences in principles ofphonetic implementation This is an area of feature theory where moreresearch is requiredThe representations of laryngeal contrasts in consonants are given

326 Prosodic featuresFinally in order to account for the existence of length distinctions and torepresent stressed versus unstressed vowels two other features wereproposed

long has greater durationstress has greater emphasis higher amplitude and pitch longerduration

These are obvious long segments are [+long] and stressed vowels are[+stress]A major lacuna in the Chomsky and Halle (1968) account of features is a

lack of features for tone This is remedied in chapter 9 when we introducenonlinear representations For the moment we can at least assume thattones are governed by a binary feature [high tone] ndash this allows only twolevels of tone but we will not be concerned with languages having morethan two tone levels until chapter 9

p b ɓ prsquo ph bɦ

voice - + + - - +cg - - + + - -sg - - - - + +

Feature theory 57

327 Summary of feature valuesFeatures combine quite freely so we cannot give a complete list Bylearning some specific feature values and applying your knowledge ofthe meaning of features it should be possible to arrive at the featurevalues of other segments This is of course possible only if you knowrelevant phonetic details of the sound that you are considering In order toknow the feature values of [ɭ] you need to know that this is the symbol fora retroflex lateral approximant thus it has the features appropriate for [l]and it also has the features that characterize retroflex consonants whichare [-ant -distr] If you do not know the phonetic characteristics of thesegment symbolized as [ʕ] it is necessary to first understand its phoneticproperties ndash it is a voiced pharyngeal continuant ndash before trying to deduceits feature values In reading descriptions of languages it is also importantto understand that a symbol used in published data on a language is notalways used according to a particular standard of phonetic transcriptionpractices at the moment so read the phonetic descriptions of letters in thegrammar carefullyThe standard feature values for the consonants of (American)

English are given in (15) to help you understand how the entire set offeatures is applied to the sound inventory of a language which you arefamiliar with

(15) p t tʃ k b d dʒ g f v θ ethsyl - - - - - - - - - - - -son - - - - - - - - - - - -cons + + + + + + + + + + + +cont - - - - - - - - + + + +delrel - - + - - - + - - - - -lat - - - - - - - - - - - -nas - - - - - - - - - - - -voi - - - - + + + + - + - +cg - - - - - - - - - - - -sg (- - - -) - - - - - - - -ant + + - - + + - - + + + +cor - + + - - + + - - - + +distr - + - + + +high - - - + - - - + - - - -lo - - - - - - - - - - - -back - - - + - - - + - - - -round - - - - - - - - - - - -

s z ʃ ʒ h ʔ m n ŋ ɹ l j wsyl - - - - - - - - - - - - -son - - - - + + + + + + + + +cons + + + + - - + + + - + - -cont + + + + + - - - - + + + +delrel - - - - - - - - - - - - -lat - - - - - - - - - - + - -

The assignment of [spread glottis] ndash aspiration ndash in English stops variesaccording to context so the value [-sg] is in parenthesis in the chartbecause both values of this feature are found on the surface depending oncontext The value [-sg] represents the underlying value

Vowel feature summary Certain feature values are uniform for allvowels [+syl -cons +son +cont -delrel -ant -lat -distr] Typicallyvowels are also [+voice -sg -cg] There are languages such as Maza-teco and Xoo where breathy voicing and glottalization are used contras-tively so in these languages [+sg] and [+cg] are possible specificationsA number of languages have phonetic voiceless vowels but the phono-logical status of voiceless vowels is not so clear thus it may be that thereare no phonologically [-voice] vowels Values of the main features used todistinguish vowels are given in (16) (Recall that we are not certainwhether [tense] applies to low vowels)

Nasality length breathiness and creaky voice are properties freelyavailable to vowels so any of these vowels can have nasal long sgor cg counterparts

Consonant feature summary Primary place of articulation for conson-ants is summarized in (17) using continuant consonants (voiceless in thefirst row voiced in the second numbers in the third row are keyed to

nas - - - - - - + + + - - - -voi - + - + - - + + + + + + +cg - - - - - + - - - - - - -sg - - - - + - - - - - - - -ant + + - - - - + + - - + - -cor + + + + - - - + - + + - -distr - - + + - - -high - - - - - - - - + + - + +lo - - - - - - - - - - - - -back - - - - - - - - + + - - +round - - - - - - - - - + - - +

i y ɨ u e oslash ə o aelig œ ɑ ɒhigh + + + + - - - - - - - -low - - - - - - - - + + + +back - - + + - - + + - - + +round - + - + - + - + - + - +tense + + + + + + + +

Feature theory 59

traditional place of articulation terms) Continuant consonants are usedhere because they exhibit the maximum number of distinctions forexample there are bilabial and labiodental fricatives but only bilabialstops All of these consonants are [-syl +cont -delrel -nas -lat-cg -tense -round]

Secondary place of articulation is illustrated in (18) here restricted tosecondary articulations on [p t] All of these consonants are [-syl -son+cons -cont -delrel -lat -nas -voice -sg -cg -tense]

Round consonants might simply have the specification [+round] Tongueraising and backing is not necessary in order to achieve rounding whereastongue raising and backing is by definition necessary in order to have avelarized consonantA final important point must be made The twenty-one features dis-

cussed here ndash syllabic sonorant consonantal high low back roundtense (advanced tongue root) coronal anterior strident distributed con-tinuant delayed release nasal lateral spread glottis constricted glottisvoice long stress ndash are specific empirical hypotheses This means thatthey are subject to change in the face of evidence that a change isrequired so they are not immutable On the other hand as scientifichypotheses they must be taken seriously until good evidence is presented

1 bilabial 2 labiodental3 (inter-)dental 4 alveolar5 alveopalatal 6 retroflex7 palatal 8 velar9 uvular 10 pharyngeal

11 glottallaryngeal

φ f θ s ʃ ʂ ccedil x χ ħ hβ v eth z ʒ ʐ ʝ γ ʁ ʕ ɦ1 2 3 4 5 6 7 8 9 10 11

ant + + + + - - - - - - -cor - - + + + + - - - - -distr + - + -high - - - - - - + + - - -lo - - - - - - - - - + -back - - - - - - - + + + -

p pw pγ pj pʕ pɥ t tw tγ tj tʕ tɥ

ant + + + + + + + + + + + +cor - - - - - - + + + + + +distr - - - - - - - - - - - -high - (+) + + - + - (+) + + - +lo - - - - + - - - - - + -back - + + - + - - + + - + -round - + - - - + - + - - - +

that another system of features is better (see section 36 and chapter 9 fordiscussion of such changes) Features should not be invented willy-nillyusing distinctive features is not the same as placing a plus sign in front ofa traditional articulatory description and thus describing sounds as[+mid] [+alveolar] or [+vowel] misconstrues the theoretical claim ofdistinctive features

33 Features and classes of segments

Besides defining phonemes features play a role in formalizing rules sincerules are stated in terms of features Every specification such as [+nasal]or [-voice] defines a class of segments The generality of a class isinversely related to how many features are required to specify the classas illustrated in (19)

The most general class defined by a single feature is [+syllabic]which refers to all vowels The size of that class is determined by thesegments in the language [+syllabic] in Spanish refers to [i e a o u] but inEnglish refers to [i ɪ e ε aelig a ɔ o ʊ u ə ʌ r l] As you add features to adescription you narrow down the class making the class less generalThe usual principle adopted in phonology is that simpler rules which usefewer features are preferable to rules using more featuresOne challenge in formalizing rules with features is recognizing the

features which characterize classes Discovering the features whichdefine a class boils down to seeing which values are the same for allsegments in the set then checking that no other segment in theinventory also has that combination of values The main obstacle isthat you have to think of segments in terms of their feature proper-ties which takes practice to become second nature As an exercisetowards understanding the relation between classes of segments andfeature descriptions we will assume a language with the followingsegments

(20) p t k b d g f s x v γ w j l m n a e i o u y

To assist in solving the problems which we will consider feature matricesof these segments are given below in (21)

+sylfrac12 +syl-nas

+syl+rd

+syl+high-nas

35 +syl

-high-lo-nas-tense

266664

377775

ε e ɪ i ε e ɪ i ε e ɪ i ɔ o ʊ ɔ ɪ i ʌ ε ɔa ɔ o ʊ atilde ɔ otilde ʊ a ɔ o ʊ otilde ʊ u ũ ʊ uu ə aelig ũ ə aelig u ə aelig

Feature theory 61

Each of the following sets of segments can be defined in terms of some setof distinctive features

(22) i p t k f s xii p t b d f s v l m niii w j l m n a e i o u yiv p k b g f x v γv j l m n a e ivi v γ w j a e i o u y

In the first set each segment is a voiceless obstruent and equally import-antly every voiceless obstruent of the language is included in this first setThis set could be specified as [-sonorant -voice] or as [-voice] since allvoiceless segments in the language are [-sonorant] Given that bothspecifications refer to exactly the same segments there is no question ofone solution being wrong in the technical sense (assuming the languagehas the segments of (20) if the language had [h] these two featurespecifications would not describe the segments) However unless thereis a compelling reason to do otherwise the simplest definition of the set ofsegments should be given using only those features which are absolutelynecessary The features which are used to exactly define a set of segmentsdepends very much on what the entire set of segments in the language isIf we were dealing with a language which had in addition the segments

cons son syl voi cont nas lat ant cor high bk low rdp + - - - - - - + - - - - -t + - - - - - - + + - - - -k + - - - - - - - - + + - -b + - - + - - - + - - - - -d + - - + - - - + + - - - -g + - - + - - - - - + + - -f + - - - + - - + - - - - -s + - - - + - - + + - - - -x + - - - + - - - - + + - -v + - - + + - - + - - - - -γ + - - + + - - - - + + - -w - + - + + - - - - + + - +j - + - + + - - - - + - - -l + + - + + - + + + - - - -m + + - + - + - + - - - - -n + + - + - + - + + - - - -a - + + + + - - - - - + + -e - + + + + - - - - - - - -i - + + + + - - - - + - - -o - + + + + - - - - - + - +u - + + + + - - - - + + - +y - + + + + - - - - + - - +

[ph th kh] then in specifying the set [p t k f s x] you would have to alsomention [-sg] in order to achieve a definition of the set which excludes[ph th kh]

The set (22ii) contains only consonants (ie [-syllabic] segments) but itdoes not contain all of the [-syllabic] segments of the language Comparethe segments making up (22ii) with the full set of consonants

This set does not include glides [consonantal] is the essential propertywhichdistinguishes glides (including h and ʔ which are lacking here) from regularconsonants Thus the segments in (ii) are [+consonantal] But not all[+consonantal] segments are included in set (ii) the velars are not includedso we need a further restriction The features typically used to specify velarsare [+high +back] so we can use one of those features Thus you can pickout the segments in (ii) as the class of [+consonantal -high] segments or the[+consonantal -back] segments Rather than refer to [consonantal] youcould try to take advantage of the fact that all glides are [+high] and referto (ii) as the set of [-high] segments without mentioning [consonantal] It istrue that all segments in the set are [-high] but [-high] itself cannot be theentire description of this set since not all [-high] segments of the languageare in the set the vowels aeo are not in set (ii) We conclude that[+consonantal -high] is the correct one for this class of segmentsSet (iii) contains a mixture of vowels and consonants it includes all

vowels plus the nasals the lateral [l] and the glides This class is definedby [+sonorant] Another feature which is constant in this group is[+voice] so you could define the class as [+sonorant +voice] But additionof [+voice] contributes nothing so there is no point in mentioning thatfeature as well Set (iv) on the other hand contains only obstruents but notall obstruents Of the whole set of obstruents what is missing from (iv) isthe group tds which are [+coronal] Therefore we can refer to set (iv) bythe combination [-sonorant -coronal]The fifth set j l m n a e i includes a mixture of vowels and conson-

ants Some properties that members of this set have in common are thatthey are voiced and they are sonorants Given the phoneme inventory allsonorants are voiced but not all voiced segments are sonorants Since thevoiced obstruents b d g v γ are not included in this set it would be lessefficient to concentrate on the feature [+voice] thus we focus on thegeneralization that the segments are sonorants Now compare this set tothe total set of sonorants

(24) j l m n a e iw j l m n a e i o u y

We can see that this set of segments is composed of a subset of sonorantsnamely the sonorants excluding w o u y But that set is the set of [+round]segments therefore the set is the set of [+sonorant -round] segments

p t b d f s v l m n Selected class of segmentsp t k b d g f s x v γ w j l m n Entire set of consonants

This set can also beidentified byreference to a singlefeature what onefeature makes thisdistinction

Feature theory 63

The last set also contains a mixture of consonants and vowels itincludes all of the vowel and glides plus the voiced obstruents v γTherefore the feature [sonorant] cannot be used to pick out this class ofsegments since members of the class can have both values for thatfeature However all of the members of this class are voiced Now com-pare set (vi) against the set of all voiced segments

The fundamental difference between [b] and [v] or between [g] and [γ] isthat b g are stops while v γ are continuants This suggests using[+continuant] as one of the defining features for this class Vowels andglides are all [+continuant] so we have passed the first test namely thatall segments in set (vi) are [+continuant +voice] We must also be surethat this is a sufficient specification for the class are there any[+continuant +voice] segments in the language which are not includedin set (vi) The segments to worry about in this case would be l m nwhich are [+voice] We exclude the nasals via [+continuant] and add[-lateral] to exclude lAs a further exercise in understanding how sets of segments are

grouped by the features assume a language with the following segmentalinventory

(26) p pf t ts tʃ c k b bv v β dz dʒ ɟ g m n ŋ f θ s ʃ d eth z ʒ i y e oslash ə o u a w j

For each group determine what feature(s) define the particular set ofsegments

(27) i tʃ c k dʒ ɟ g ŋ ʃ ʒ i y e oslash ə o u a w jii s i ʃ e f z v β a ʒ o u j oslash θ y ə w ethiii k j g c w i u y ɟ ŋiv k g a ə ŋ

34 Possible phonemes and rules ndash an answer

We now return to the theoretical questions raised at the beginning of thischapter what is a possible phoneme and what is a possible phonologicalrule

341 Possible phonemesThe theory of features answers the question of possible phonemes sayingthat the segments which can be constructed using these features are alland the only possible phonemes This gives a mathematical upper limit of2n segments given n binary features so if there are twenty features(a reasonable number) there are 1048576 logically possible feature

v γ w j a e i o u yb d g v γ w j l m n a e i o u y

specifications and this is quite a lot of segments It also has to be physic-ally possible to realize a segment so the number of possible segments issmaller than this Many segments can be imagined which are phoneticallyuninterpretable such as one which is [+high +low] Such a segment isphysically impossible since the tongue cannot be contradictorily raisedand lowered at the same time so the nonexistence of a large class of suchsegments is independently explained Similarly no segment can be[+cons -high -back -ant -cor] A segment which is [+cons] is not avowel or glide The feature [-back] tells us that the segment would havea place of articulation in front of the velar position [-ant] tells us that itmust have a place of articulation behind the alveolar ridge and [-high]tells us that it cannot be a palatal Everything about this descriptionsuggests the vowel [e] except that it is [+consonantal] whereas vowelsare [-consonantal] No major constriction can be formed with the tonguein the position of [e] hence this combination of features happens to bephysically impossible To be attested in a language a segment must beboth combinatorially possible ie it must use just the features given by thetheory and physically possibleAlthough the set of attested phonemes in human languages is

quite large there are significant limitations on what phonemes are pos-sible Retroflex consonants have the features [-anterior +coronal-distributed] Recall the question whether a language could contrasttwo kinds of retroflex consonants such as apical and sublaminal retroflexas found in Hindi versus Telugu According to this theory of features sucha contrast is impossible since no feature is available to describe such adifference within a language Phonetic differences across languages arepossible because phonetic interpretation is not subject to the limitationsof phonological feature theory Were we to discover such a contrast thetheory of features would be challenged because it has no mechanism forexpressing such a distinction Similarly the differences attested in thephonetics of [u] and [ʊ] across languages are never found within a lan-guage In a single language the maximal contrast is between two suchvowels governed by the feature tense (or ATR) The fact that such differ-ences exist at the phonetic level between languages but are neverexploited within a single language as a way to distinguish words is anexample of the difference between phonetic and phonological propertiesThus one of the main goals of distinctive feature theory is providing a

predictive framework for saying what contrasts will and will not be foundin the phoneme systems of human languages

342 Rule formulation and featuresThe most important function of features is to form the basis for writingrules which is crucial in understanding what defines a possible phono-logical rule A typical rule of vowel nasalization which nasalizes allvowels before a nasal can be formulated very simply if stated in features

(28) [+syllabic] [+nasal] __ [+nasal]

Feature theory 65

Such a rule is common in the languages of the world Very uncommon ifit exists at all is one nasalizing only the lax vowel [ɪ] and only before [m]Formulated with features that rule looks as follows

__ nasal][+nasal+ant-cor

+rarr⎥⎥⎦

⎢⎢⎣

⎥⎥⎥

⎢⎢⎢

⎡ +syl-ATR+high-rd

This rule requires significantly more features than (28) since [ɪ] whichundergoes the rule must be distinguished in features from other highvowels such as [i] or [ʊ] which (in this hypothetical case) do not undergothe rule and [m] which triggers the rule must be distinguished from [n]or [ŋ] which do not

Simplicity in rule writing This relation between generality and simpli-city on the one hand and desirability or commonness on the other hasplayed a very important role in phonology all things being equal simplerrules are preferred both for the intrinsic elegance of simple rules andbecause they correlate with more general classes of segments Maximumgenerality is an essential desideratum of scienceThe idea that rules are stated in terms of the simplest most general

classes of phonetically defined segments has an implication for rule for-mulation Suppose we encounter a rule where high vowels (but not midand low vowels) nasalize before nasal stops (n m ŋ) thus in ĩn uŋ uŋand so on We would formulate such a rule as follows

(30) [ ] _+syl+high

rarr +nasal⎡⎢⎣

⎤⎥⎦

+nasal-cont

⎡⎢⎣

⎤⎥⎦

However we could equally well formalize the rule as

rarr+syl+high-low

⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+syl+high-low+nasal

⎡⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎦

_⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+nasal-cont-low

We could freely add [-low] to the specification of the input segment (sinceno vowel can be +high +lowfrac12 thus high vowels automatically wouldpass that condition) and since the same class of vowels is referencedinclusion of [-low] is empirically harmless Saying that the vowelbecomes [+syl +high -low] is harmless since the vowel that undergoesthe change already has these specifications At the same time the addi-tional features in (31) are useless complications so on the theoreticalgrounds of simplicity we formalize the rule as (30) In writing phono-logical rules we specify only features which are mandatoryA formulation like

(32)+nasal-cont

⎡⎢⎣

⎤⎥⎦

[ ] [ ] _+syl rarr +nasal

would mention fewer features but it would be wrong given the factswhich the rule is supposed to account for since the rule should state thatonly high vowels nasalize but this rule nasalizes all vowelsLikewise we could complicate the rule by adding the retriction that

only non-nasal vowels are subject to (30) in (30) we allow the rule tovacuously apply to high vowels that are already nasal There is (andcould be) no direct evidence which tells us whether ĩn undergoes (30)and surfaces as [ĩn] or ĩn is immune to (30) and surfaces as [ĩn] andthere is no conceptual advantage to complicating the rule to prevent itfrom applying in a context where we do not have definitive proofthat the rule applies The standard approach to rule formalization istherefore to write the rule in the simplest possible way consistentwith the facts

Formalizability The claim that rules are stated in terms of phonetic-ally defined classes is essentially an axiom of phonological theory Whatare the consequences of such a restriction Suppose you encounter alanguage with a phonological rule of the type p r i b _ o nSince the segments being changed (p and r) or conditioning the change (oand n) cannot be defined in terms of any combination of features norcan the changes be expressed via any features the foundation of phono-logical theory would be seriously disrupted Such a rule would refute afundamental claim of the theory that processes must be describable interms of these (or similar) features This is what it means to say thatthe theory makes a prediction if that prediction is wrong the theoryitself is wrongMuch more remains to be said about the notion of ldquopossible rulerdquo in

phonology nevertheless we can see that distinctive feature theory playsa vital role in delimiting possible rules especially in terms of character-izing the classes of segments that can function together for a rule Wenow turn to a discussion of rule formalism in the light of distinctivefeature theory

35 The formulation of phonological rules

Many aspects of rule theory were introduced in our informal approachto rule writing in chapter 2 and they carry over in obvious ways tothe formal theory that uses features The general form of a phono-logical rule is

(33) αFiβFj

γFkδFl

εFmζFn

ηFoθFp

Focus Structuralchange

Trigger

Feature theory 67

where Fi Fj Fk are features and α β γ are plus or minus values Thearrow means ldquobecomesrdquo slash means ldquowhen it is in the contextrdquo andthe dash refers to the position of the focus in that context The matrix tothe left of the arrow is the segment changed by the rule that segment isreferred to as the focus or target of the rule The matrix immediately tothe right of the arrow is the structural change and describes the way inwhich the target segment is changed The remainder of the rule consti-tutes the trigger (also known as the determinant or environment)stating the conditions outside the target segment which are necessaryfor application of the rule Instead of the slash a rule can be formulatedwith the mirror-image symbol ldquordquo which means ldquobefore or afterrdquo thusldquoX Y __Zrdquo means ldquoX becomes Y before or after ZrdquoEach element is given as a matrix which expresses a conjunction of

features The matrices of the target and trigger mean ldquoall segments of thelanguage which have the features [αFi] as well as [βFj] rdquo The matrix ofthe structural change means that when a target segment undergoes arule it receives whatever feature values are specified in that matrixThere are a few special symbols which enter into rule formulation One

which we have encountered is the word boundary symbolized as ldquordquoA rule which lengthens a vowel before a word-final sonorant would bewritten as follows

(34) [+syl] [+long] _ [+son]

A rule which devoices a word-initial consonant would be written as

(35) [-son] [-voice] _

A word boundary can come between the target and the trigger segmentsin which case it means ldquowhen the trigger segment is in the next wordrdquoSuch processes are relatively infrequent but for example there is arule in Sanskrit which voices a consonant at the end of a word whenit is followed by a sonorant in the next word so tataham becomes[tadaham] lsquothat Irsquo voicing does not take place strictly within the wordand thus patami lsquoI flyrsquo does not undergo voicing This rule is formulatedas in (36)

(36) [-son] [+voice] _ [+son]

Another symbol is the null Oslash used in the focus or structural change of arule As the focus it means that the segment described to the right of thearrow is inserted in the stated context and as the structural change itmeans that the specified segment is deleted Thus a rule that deletes aword-final short high vowel which is preceded by a sonorant would bewritten as follows

(37) +syl-high-long

[ ]Oslash _ +sonrarr⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

There are occasions where it is necessary to restrict a rule to apply onlywhen a sequence occurs in different morphemes but not withina morpheme Suppose you find a rule that deletes a consonant after aconsonant but only when the consonants are in separate morphemesthus the bimorphemic word tap-ta with p at the end of one mor-pheme and t at the beginning of another becomes [tapa] but themonomorphemic word tapta does not undergo deletion Analogousto the word boundary there is also a morpheme boundary symbolizedby ldquo+rdquo which can be used in writing rules Thus the rule deleting thesecond of two consonants just in case the consonants are in differentmorphemes (hence a morpheme boundary comes between the conson-ants) is stated as

(38) [-syl] Oslash [-syl] +_

You may encounter other conventions of formalism One such notation isthe brace notation Whereas the standard matrix [ ] refers to a conjunc-tion of properties ndash segments which are A and B and C all at once ndash braces express disjunctions that is segments which are A or B or C One ofthe most frequent uses of braces is exemplified by a rule found in anumber of languages which shortens a long vowel if it is followed byeither two consonants or else one consonant plus a word boundary iefollowed by a consonant that is followed by a consonant or Such a rulecan be written as (39)

(39) [ ]_][ ]+syl rarr [ -long -syl⎧⎨⎩

⎫⎬⎭

Most such rules use the notation to encode syllable-related properties soin this case the generalization can be restated as ldquoshorten a long vowelfollowed by a syllable-final consonantrdquo Using [] as the symbol for asyllable boundary this rule could then be reformulated as

(40) [+syl] [-long] _ [-syl]middot

Although the brace notation has been a part of phonological theory it hasbeen viewed with considerable skepticism partly because it is not wellmotivated for more than a handful of phenomena that may have betterexplanations (eg the syllable) and partly because it is a powerful devicethat undermines the central claim that rules operate in terms of naturalclasses (conjunctions of properties)Some rules need to refer to a variably sized sequence of elements

A typical example is vowel harmony where one vowel assimilates afeature from another vowel and ignores any consonants that comebetween Suppose we have a rule where a vowel becomes round after around vowel ignoring any consonants We could not just write the rule as(41) since that incorrectly states that only vowels strictly next to roundvowels harmonize

Feature theory 69

(41)[+syl] rarr [+rd] ⎥⎦

⎤ ⎢⎣⎡+syl+rd

We can use the subscript-zero notation and formalize the rule as in (42)

(42) +syl[ ] +rd[ ] __-syl[ ]0

⎥⎤⎦⎢⎣

⎡ +syl+rd

The expression ldquo[-syl]0rdquo means ldquoany number of [-syl] segmentsrdquo fromnone to an infinite sequence of themA related notation is the parenthesis which surrounds elements that

may be present but are not required A rule of the form X Y _ (WZ)Qmeans that X becomes Y before Q or before WZQ that is beforeQ ignoring WZ The parenthesis notation essentially serves to groupelements together This notation is used most often for certain kinds ofstress-assignment rules and advancements in the theory of stress haverendered parenthesis unnecessary in many casesOne other very useful bit of notation is the feature variable notation So

far it has actually been impossible to formalize one of the most commonphonological rules in languages the rulewhich assimilates a nasal inplace ofarticulation to the following consonant where mk [ŋk] np [mp] andso onWhile we can write a rule whichmakes any nasal become [+ant +cor]before a [+ant +cor] consonant ndash any nasal becomes [n] before t ndash and wecan write a rule to make any nasal [+ant -cor] before a [+ant -cor]consonant ndash nasals become [m] before [p] ndash we cannot express both changesin one rule

(43) [ ] _+nas+ant+cor

eacuteecirceuml

ugraveuacuteucirc

eacuteecirceuml

ugraveuacuteucirc

+ant+cor

[ ] _+nas+ant-cor

eacuteecirceuml

ugraveuacuteucirc

eacuteecirceuml

ugraveuacuteucirc

+ant-cor

The structural change cannot be ldquo[+cor]rdquo because when a nasal becomes[m] it becomes [-cor] For the same reason the change cannot be ldquo [-cor]rdquo since making a nasal become [n] makes it become [+cor] Onesolution is the introduction of feature variables notated with Greekletters α β γ etc whose meaning is ldquothe same valuerdquo Thus a rule whichmakes a nasal take on whatever values the following consonant has forplace of articulation would be written as follows

(44) __ rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

[+nasal ] αantβcor

αantβcor

Thus when the following consonant has the value [+cor] the nasalbecomes [+cor] and when the following consonant has the value [-cor]the nasal becomes [-cor] We will return to issues surrounding thisnotation in chapter 9There are a couple of commonly used informal shorthand practices

which you need to recognize Many rules refer to ldquoconsonantsrdquo versus

ldquovowelsrdquo meaning [-syllabic] and [+syllabic] segments and the short-hand ldquoCrdquo and ldquoVrdquo are often used in place of [-syllabic] and [+syllabic]Also related to the feature variable notation it is sometimes necessary towrite rules which refer to the entire set of features A typical examplewould be in a rule ldquoinsert a vowel which is a copy of the preceding vowelinto a word-final clusterrdquo Rather than explicitly listing every feature withan associated variable such a rule might be written as

(45) Oslash Vi Vi C_C

meaning ldquoinsert a copy of the preceding vowelrdquo

36 Changing the theory

The theory of features is an empirical hypothesis and is subject to revi-sion in the face of appropriate data It is not handed down by a higherauthority nor is it arbitrarily picked at the whim of the analyst It isimportant to give critical thought to how the set of distinctive featurescan be tested empirically and revised One prediction of the theory whichwe have discussed in section 31 is that the two kinds of phonetic retroflexconsonants found in Hindi and Telugu cannot contrast within a languageWhat would happen if a language were discovered which distinguishedtwo degrees of retroflexion Would we discard features altogetherThis situation has already arisen the theory presented here evolved

from earlier similar theories In an earlier theory proposed by Jakobsonand Halle retroflex consonants were described with the feature [flat] Thisfeature was also used to describe rounding pharyngealization and uvu-larization While it may seem strange to describe so many differentarticulatory characteristics with a single feature the decision was justi-fied by the fact that these articulations share an acoustic consequence adownward shift or weakening of higher frequencies The assumption atthat point was that no language could minimally contrast retroflexionrounding and pharyngealization If a language has both [ʈ ] and [kw] thesurface differences in the realization of [flat] as retroflexion versusrounding would be due to language-specific spell-out rulesThe theory would be falsified if you could show that rounding and

pharyngealization are independent and counterexamples were foundArabic has the vowels [i a u] as well as pharyngealized vowels [iʕ aʕ uʕ]which derive by assimilation from a pharyngealized consonant Ifrounding and pharyngealization are both described by the feature [flat]it is impossible to phonologically distinguish [u] and [uʕ] But this is not atall inappropriate since the goal is to represent phonological contrasts notphonetic differences because the difference between [u] and [uʕ] is a low-level phonetic one The relevance of Arabic ndash whether it falsifies thefeature [flat] ndash depends on what you consider to be the purpose of featuresBadagarsquos three-way vowel contrast challenges the standard theory as

well Little is known about this language the contrast was originally

Feature theory 71

reported by Emeneau (1961) and Ladefoged and Maddieson (1996) reportthat few speakers have a three-way contrast The problem posed by thiscontrast has been acknowledged but so far no studies have explored itsnatureAnother prediction is that since uvular and round consonants are both

[+flat] there should be no contrast between round and nonround uvularsor between round velars and nonround uvulars within a language But anumber of languages of the Pacific Northwest including Lushootseedhave the contrast [k kw q qw] this is a fact which is undeniably in thedomain of phonology The Dravidian language Badaga is reported tocontrast plain and retroflex vowels where any of the vowels [i e a o u]can be plain half-retroflex or fully retroflex If [flat] indicates both retro-flexion and rounding it would be impossible to contrast [u] and [u˞] Suchlanguages forced the abandonment of the feature [flat] in favor of thesystem now usedThe specific feature [flat] was wrong not the theory of features itself

Particular features may be incorrect which will cause us to revise orreplace them but revisions should be undertaken only when strongevidence is presented which forces a revision Features form the founda-tion of phonology and revision of those features may lead to considerablechanges in the predictions of the theory Such changes should be under-taken with caution taking note of unexpected consequences If the theorychanges frequently with new features constantly being added this wouldrightly be taken as evidence that the underlying theory is wrongSuppose we find a language with a contrast between regular and sub-

lingual retroflex consonants We could accommodate this hypotheticallanguage into the theory by adding a new feature [sublingual] defined asforming an obstruction with the underside of the tongue This theorymakes a new set of predictions it predicts other contrasts distinguishedby sublinguality We can presumably restrict the feature to the [+coronal]segments on physical grounds The features which distinguish coronalsubclasses are [anterior] and [distributed] which alone can combine todescribe four varieties of coronal ndash which actually exist in a numberof Australian languages With a new feature [sublingual] eight coronalclasses can be distinguished regular and sublingual alveolars regular andsublingual dentals regular and sublingual alveopalatals and regularand sublingual retroflex consonants Yet no such segments have beenfound Such predictions need to be considered when contemplating achange to the theorySimilarly recall the problem of ldquohyper-tenserdquo ldquoplain tenserdquo ldquoplain laxrdquo

and ldquohyper-laxrdquo high vowels across languages we noted that no morethan two such vowels exist in a language governed by the feature [tense]If a language were discovered with three or four such high vowels wecould add a feature ldquohyperrdquo But this makes the prediction that therecould also be four-way contrasts among mid and low vowels If theseimplications are not correct the modification to the theory is not likelyto be the correct solution to the problem In general addition of newfeatures should be undertaken only when there is compelling evidence for

doing so The limited number of features actually in use is an indication ofthe caution with which features are added to the theory

The case for labial A classical case in point of a feature which wasadded in response to significant problems with the existing feature systemis the feature [labial] It is now accepted that feature theory should includethis feature

[labial] sound produced with the lips

This feature was not part of the set of features proposed in Chomskyand Halle (1968) However problems were noticed in the theory without[labial]The argument for adding [labial] is that it makes rules better formaliz-

able It was noticed that the following types of rules inter alia arefrequently attested (see Campbell 1974 Anderson 1974)

(46) a b w _ Cb w b [+nasal] _c w vd i u p b m w u o _

In the first three rules the change from bilabial obstruent to roundedglide or rounded glide to labiodental obstruent is a seemingly arbitrarychange when written according to the then-prevailing system of featuresThere is so little in common between [b] and [w] given these features thata change of [b] to [r] would be simpler to formulate as in (47b) and yet thechange [b] [r] is unattested

_rarr rarr⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎥⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎡⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎦

⎡⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎦

-ant-cons+high+bk+rd

+ant-cor+voi

+ant-cor-high-rd

ba C-cons+high+rd

In the last rule of (46) no expression covers the class p b m w u orather they correspond to the disjunction [+ant -cor] or [+round]These rules can be expressed quite simply with the feature [labial]

(48) [ ] __rarr⎡⎢⎣

⎤⎥⎦

+labial+voi

ndashcons Ca

[ ] __rarr⎡⎢⎣

⎤⎥⎦

+labial-cons

+cons [ ]+nasalb

rarr⎡⎢⎣

⎤⎥⎦

+labial+rd

⎡⎢⎣

⎤⎥⎦

+cons-rd

__rarr [ ]+labial [ ]+labialid

Feature theory 73

Feature redefinition Even modifying definitions of existing featuresmust be done with caution and should be based on substantialevidence that existing definitions fail to allow classes or changes to beexpressed adequately One feature which might be redefined is [con-tinuant] The standard definition states that a segment is [+continuant]if it is produced with air continuously flowing through the oral cavityAn alternative definition is that a segment is [+continuant] if airflows continuously through the vocal tract How do we decide whichdefinition is correct The difference is that under the first definitionnasals are [-continuant] and under the second definition nasals are[+continuant]If the first definition is correct we expect to find a language where p t tʃ

k m n ŋ b d dʒ g undergo or trigger a rule and f s θ x v z eth γ do notunder the ldquooral cavityrdquo definition [-continuant] refers to the class of seg-ments p t tʃ k m n ŋ b d dʒ g On the other hand if the second hypothesisis correct we shouldfind a languagewhere nm n f s x v x γ undergo ortrigger a rule and the remaining consonants p t tʃ k b d dʒ g donot underthe ldquovocal tractrdquo definition of [continuant] the feature specification[+continuant] would refer to the set n m n f s x v x γJust as important as knowing what sets of segments can be referred to

by one theory or another you need to consider what groupings of seg-ments cannot be expressed in a theory Under either definition of [continu-ant] finding a process which refers to p t k b d g proves nothing sinceeither theory can refer to this class either as [-continuant] in the ldquooralcavityrdquo theory or as [-continuant -nasal] in the ldquovocal tractrdquo theory Theadditional feature needed in the ldquovocal tractrdquo theory does complicate therule but that does not in itself disprove the theory If you find a processreferring to n m n f s x v x γ excluding p t k b d g this woulddefinitively argue for the ldquooral cavityrdquo theory Such a class can be referredto with the specification [+continuant] in the ldquooral cavityrdquo theory butthere is no way to refer to that set under the ldquovocal tractrdquo theory As itstands we have not found such clear cases but at least we can identify thetype of evidence needed to definitively choose between the theories Theimplicit claim of feature theory is that it would be impossible for bothkinds of rules to exist in human languages There can only be one defin-ition of any feature if the theory is to be coherent

Central vowels We will consider another case where the features face aproblem with expressing a natural class relating to the treatment ofcentral versus back vowels In chapter 2 we saw that Kenyang [k] and [q]are in complementary distribution with [q] appearing word-finally afterthe vowels [o] [ɔ] and [ɑ] and [k] appearing elsewhere Representativeexamples are reproduced here

(49) enɔq lsquotreersquo enoq lsquodrumrsquo

ŋgɑq lsquoknifersquo ekɑq lsquolegrsquomək lsquodirtrsquo ndek lsquoEuropeanrsquopɔbrik lsquowork projectrsquo ɑjuk (personrsquos name)

Phonetic descriptions of vowels are not usually based on physiologicaldata such as x-ray studies Tongue positions are often deduced by match-ing sound quality with that of a standardly defined vowel we assume thatKenyang schwa is central because it sounds like schwa which is phonet-ically defined as being centralSchwa does not cause lowering of k to q In the standard account of

vowels [ə] differs from [ɔ] only in rounding though phonetic traditionclaims that these vowels also differ in being back ([ɔ]) versus central ([ə])As previously discussed this difference is attributed to a low-level phono-logically insignificant phonetic factorThe problem which Kenyang poses is that it is impossible to formulate

the rule of k-lowering if schwa is phonologically a mid back unroundedvowel A simple attempt at formalizing the rule would be

(50) [ ] __rarr⎡⎢⎣

⎤⎥⎦

+high+back

⎡⎢⎣

⎤⎥⎦

+back+high

If schwa is [+back -high -round] it would satisfy the requirements ofthe rule so should cause lowering of k but it does not therefore thisformulation cannot be correct Since schwa differs from [ɔ] in being[-round] we might try to exclude [ə] by requiring the trigger vowel tobe [+round]

(51)[ ] rarr

⎡⎢⎣

⎤⎥⎦

+high+back

-high+back-high+round

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

But this formulation is not correct either since it would prevent thenonround low vowel [ɑ] from triggering uvularization which in fact itdoes doThese data are a problem for the theory that there is only a two-way

distinction between front and back vowels not a three-way distinctionbetween front central and back vowels The uvularization rule of Kenyangcan be formulated if we assume an additional feature [front] whichcharacterizes front vowels Under that theory back vowels would be[+back -front] front vowels would be [+front -back] and central vowelswould be [-back -front] Since we must account for this fact aboutKenyang the theory must be changed But before adding anything to thetheory it is important to consider all of the consequences of the proposalA positive consequence is that it allows us to account for Kenyang

Another possible example of the relevance of central vowels to phonologycomes from Norwegian (and Swedish) There are three high round vowelsin Norwegian whereas the standard feature theory countenances theexistence of only two high rounded vowels one front and one backExamples in Norwegian spelling are do lsquoouthousersquo du lsquoyou sgrsquo and dylsquoforbearrsquo The vowel o is phonetically [u] and u and y are distinct nonbackround vowels In many transcriptions of Norwegian these are transcribedas [dʉ] lsquoyou sgrsquo and [dy] lsquoforbearrsquo implying a contrast between frontcentral and back round vowels This is exactly what the standard view

Feature theory 75

of central vowels has claimed should not happen and it would appear thatNorwegian falsifies the theoryThe matter is not so simple The vowels spelled u versus y also differ

in lip configuration The vowel u is ldquoin-roundedrdquo with an inwardnarrowing of the lips whereas y is ldquoout-roundedrdquo with an outward-flanging protrusion of the lips This lip difference is hidden by the selec-tion of the IPA symbols [ʉ] versus [y] While it is clear that the standardtheory does not handle the contrast we cannot tell what the correct basisfor maintaining the contrast is We could treat the difference as a front ~central ~ back distinction and disregard the difference in lip configuration(leaving that to phonetic implementation) or we could treat the labialdistinction as primary and leave the presumed tongue position to phon-etic implementationGiven that the theory of features has also accepted the feature [labial] it

is possible that the distinction lies in [labial] versus [round] where the out-rounded vowel ltygt is [+round +labial] and in-rounded ltugt is [-round+labial] ndash or vice versa Unfortunately nothing in the phonological behav-ior of these vowels gives any clue as to the natural class groupings of thevowels so the problem of representing these differences in Norwegianremains unresolved Thus the case for positing a distinct phonologicalcategory of central vowel does not receive very strong support from thevowel contrasts of NorwegianA negative consequence of adding [front] which would allow the phono-

logical definition of a class of central vowels is that it defines unattestedclasses and segments outside the realm of vowels The classical featurescould distinguish just [k] and [kj] using [back] With the addition of[front] we would have a three-way distinction between k-like consonantswhich are [+front -back] [-front -back] and [-front +back] But noevidence at all has emerged for such a contrast in any language Finallythe addition of the feature [front] defines a natural class [-back] contain-ing front and central vowels but not back vowels such a class is notpossible in the classical theory and also seems to be unattested in phono-logical rules This may indicate that the feature [front] is the wrongfeature ndash at any rate it indicates that further research is necessary inorder to understand all of the ramifications of various possible changes tothe theoryThus the evidence for a change to feature theory made to handle

the problematic status of [ə] in Kenyang phonology would not besufficiently strong to warrant complete acceptance of the new featureWe will suspend further discussion of this proposal until later whennonlinear theories of representation are introduced and answers tosome of the problems such as the unattested three-way contrast invelars can be considered The central point is that changes in thetheory are not made at will they are made only after considerableargumentation and evidence that the existing theory is fundamentallyinadequate

Exercises1 Assume a segmental inventory composed of [ʕ k t d s z n p f b i u e o a w h]

Indicate what feature or features characterize the following classes of soundsi ʕ k u o a wii f p k hiii f p b t s d z niv ʕ u o w a b d z n i e

2 Given the segments [w j h ʔ i ε a o ɔ u m l r m ŋ p t kj k q b eth d dj g γ] describethe following segment classes being as economical as you can with your useof featuresi m l r m ŋ p t kj k q b eth d dj g γii w j i ε a o ɔ u m l r m ŋiii w a o ɔ u ŋ k q g γiv w j h i ε a o ɔ u l r eth γv j i kj dj

vi i ε a o ɔ u m

3 Assume the following segmental inventoryp t tʃ q b ɖ g s ʂ β ʒ γ n ŋ l j i ɨ e o ɛ ɔ aelig

Give the minimal feature description which identifies exactly the followingsubsets of the inventoryi p t b s β n lii t tʃ ɖ s ʂ ʒ n liii p t tʃ q b ɖ s ʂ β ʒ n l e o ɛ ɔ aeligiv q g γ ŋ ɨ o ɔv tʃ ɖ ʂ ʒ j i e ɛ aelig

4 State all of the features which are changed in each of the following rulesi p fii t ŋiii o wiv k sv s tvi a i

Summary Language sounds can be defined in terms of a small set of universalphonetically based features which not only define the basic atoms ofphonological representations but also play a central role in the formalexpression of rules An important theme of this chapter is the natureof scientific theories such as the theory of features which makepredictions both about what can happen and what cannot happen Thefundamental role of feature theory is to make specific predictionsabout the kinds of segments and rules that we should find in humanlanguages One of the main concerns of phonological theory is findingthe correct set of features that define the sounds and rule systems of allhuman languages

Feature theory 77

5 Formalize the following rules using distinctive features (segmentalinventories to be assumed for each language are given after the rule inbrackets) In each case if the inventory includes segments [w x y z] and therule is stated as changing [w] and [x] assume that y z can appear in thespecified context and appear as [y z] after the rule applies

6 Mixtec (San Miguel el Grande)

The causative form of the verb in Mixtec has a prefix underlying s whichchanges before certain consonants Formalize a rule which accounts for thesechanges

7 Review previous solutions to exercises which you have done in the precedingchapter and state the rules according to the features given here discuss anyproblems which you may encounter in reformalizing these rules

Further readingCampbell 1974 Chomsky and Halle 1968 Jakobson and Halle 1956 Jakobson Fant and Halle 1952

Trubetzkoy 1939

s-kaka lsquomake walkrsquo s-haa lsquomake sproutrsquoʃ-dɨbɨ lsquomake enterrsquo s-taka lsquogatherrsquos-tʃaku lsquomake liversquo ʃ-lili lsquotightenrsquos-kunu lsquomake runrsquo ʃ-ndata lsquocrackrsquoʃ-dʒaʔa lsquooverthrowrsquo

i b d g β eth γ V _ [p t k b d g β eth γ m n ŋ r i u a ə]ii Oslash j i e _ o u a [p t k b d n j w i y e aelig o u a]iii t s _i [p t k h v d s r l m n j i y e oslash a o u]iv s r V_V [p t k b d g s r l m n h w j e i o u a]v p t [tʃ] __ i e a o u k tʃ ŋ [p t tʃ k n ŋ i e a o u]vi p t t kφ θ r x i e a o w j

h ___ i e a o r l[p t t k b d g φ θ r x r l w j h r l i e a o]

vii p tʃ k q t __ p b m t d n [p t tʃ k q b d g eth l r m n i u a ɛ]viii k g ŋ kj gj ŋj p pj m b t tj n

d tʃ dʒ ɲ kj gj ŋj f s ʃ xj j ___[p pj m b t tj n d tʃ dʒ ɲ kj gj ŋj k g ŋ

q f s ʃ xj j w i u e o a aelig]

CHAPTER

4 Underlyingrepresentations

PREVIEW

This chapter looks deeper into the nature of underlying

forms by

u introducing contrast-neutralizing rules

u seeing how unpredictable information must be part of

the underlying form

u learning what factors are most important in establishing

an underlying representation

u understanding how underlying forms are different from

actually pronounced words

KEY TERMSalternation

neutralization

predictability

structurepreservation

A fundamental characteristic of the rules discussed up to this point isthat they have described totally predictable allophonic processes such asaspiration in English or vowel nasalization in Sundanese For such rulesthe question of the exact underlying form of a word has not been socrucial and in some cases a clear decision could not be made We sawthat in Sundanese every vowel becomes nasalized after a nasal soundand every phonetic nasal vowel appears after a nasal Nasality of vowelscan always be predicted by a rule in this language all nasal vowelsappear in one predictable context and all vowels are predictably nasalin that context It was therefore not crucial to indicate whether a givenvowel is underlyingly nasal or underlyingly oral If you assume thatvowels are underlyingly oral you can write a rule to derive all of thenasal vowels and if you contrarily assume that vowels are all under-lyingly nasal you could write a rule to derive all of the oral vowels Thechoice of underlying sound may make a considerable difference in termsof simplicity and elegance of the solution and this is an importantconsideration in evaluating a phonological analysis but it is possible tocome up with rules which will grind out the correct forms no matterwhat one assumes about underlying representations in these cases Thisis not always the case

41 The importance of correct underlying forms

Neutralizing rules on the other hand are ones where two or moreunderlyingly distinct segments have the same phonetic realization insome context because a rule changes one phoneme into another ndash thusthe distinction of sounds is neutralized This means that if you look at aword in this neutralized context you cannot tell what the underlyingsegment is Such processes force you to pay close attention to maintainingappropriate distinctions in underlying formsConsider the following examples of nominative and genitive forms of

nouns in Russian focusing on the final consonant found in thenominative

(1) Nominative singular Genitive singularvagon vagona lsquowagonrsquoavtomobilj avtomobilja lsquocarrsquovetʃer vetʃera lsquoeveningrsquomuʃ muʒa lsquohusbandrsquokarandaʃ karandaʃa lsquopencilrsquoglas glaza lsquoeyersquogolos golosa lsquovoicersquoras raza lsquotimersquoles lesa lsquoforestrsquoporok poroga lsquothresholdrsquovrak vraga lsquoenemyrsquourok uroka lsquolessonrsquo

To give an explanation for the phonological processes at work in thesedata you must give a preliminary description of the morphology Whilemorphological analysis is not part of phonology per se it is inescapablethat a phonologist must do a morphological analysis of a language todiscover the underlying formIn each of the examples above the genitive form is nearly the

same as the nominative except that the genitive also has the vowel[a] which is the genitive singular suffix We will therefore assume asour initial hypothesis that the bare root of the noun is used to formthe nominative case and the combination of a root plus the suffix-a forms the genitive Nothing more needs to be said about examplessuch as vagon ~ vagona avtomobilj ~ avtomobilja or vetʃer ~ vetʃerawhere as it happens the root ends with a sonorant consonantThe underlying forms of these noun stems are presumably vagonavtomobilj and vetʃer no facts in the data suggest anything elseThese underlying forms are thus identical to the nominative formWith the addition of the genitive suffix -a this will also give thecorrect form of the genitiveThere are stems where the part of the word corresponding to the root

is the same in all forms of the word karandaʃ ~ karandaʃa golos ~ golosales ~ lesa urok ~ uroka porok ~ poroka tsvet ~ tsveta soldat ~ soldata andtrup ~ trupa However in some stems there are differences in the finalconsonant of the root depending on whether we are considering thenominative or the genitive Thus we find the differences muʃ ~ muʒa~ glas ~ glaza porok ~ poroga vrak ~ vraga prut ~ pruda and xlep ~ xlebaSuch variation in the phonetic content of a morpheme (such as a root) isknown as alternation We can easily recognize the phonetic relationbetween the consonant found in the nominative and the consonantfound in the genitive as involving voicing the consonant found in thenominative is the voiceless counterpart of the consonant found in thegenitive Not all noun stems have such an alternation as we can see bypairs such as karandaʃ ~ karandaʃa les ~ lesa urok ~ uroka soldat ~ soldataand trup ~ trupa We have now identified a phonological problem to besolved why does the final consonant of some stems alternate in voicingAnd why do we find this alternation with some stems but not othersThe next two steps in the analysis are intimately connected we must

devise a rule to explain the alternations in voicing and we must set up

porok poroka lsquovicersquotsvet tsveta lsquocolorrsquoprut pruda lsquopondrsquosoldat soldata lsquosoldierrsquozavot zavoda lsquofactoryrsquoxlep xleba lsquobreadrsquogrip griba lsquomushroomrsquo

trup trupa lsquocorpsersquo

Underlying representations 81

appropriate underlying representations for these nouns In order to deter-mine the correct underlying forms we will consider two competinghypotheses regarding the underlying form and in comparing the predic-tions of those two hypotheses we will see that one of those hypotheses isclearly wrongSuppose first that we decide that the form of the noun stem which we

see in the nominative is also the underlying form Such an assumption isreasonable (it is also not automatically correct) since the nominative isgrammatically speaking a more ldquobasicrdquo form of a noun In that case wewould assume the underlying stems glas lsquoeyersquo golos lsquovoicersquo ras lsquotimersquoand les lsquoforestrsquo The problem with this hypothesis is that we would haveno way to explain the genitive forms glaza golosa raza and lesa thecombination of the assumed underlying roots plus the genitive suffix -awould give us glasa golosa rasa and lesa so we would be right only abouthalf the time The important step here is that we test the hypothesis bycombining the supposed root and the affix in a very literal-minded waywhereupon we discover that the predicted forms and the actual forms aredifferentWe could hypothesize that there is also a rule voicing consonants

between vowels (a rule like one which we have previously seen in Kipsigischapter 2)

(2) C [+voice] V_ V

While applying this rule to the assumed underlying forms glas-agolos-a ras-a and les-a would give the correct forms glaza and raza itwould also give incorrect surface forms such as goloza and leza Thus notonly is our first hypothesis about underlying forms wrong it also cannotbe fixed by positing a rule of consonant voicingYou may be tempted to posit a rule that applies only in certain words

such as eye time and so on but not voice forest etc This misconstrues thenature of phonological rules which are general principles that apply to allwords of a particular class ndash most generally these classes are defined interms of phonological properties such as ldquoobstruentrdquo ldquoin word-finalpositionrdquo Rules which are stated as ldquoonly applying in the followingwordsrdquo are almost always wrongThe ldquonominative is underlyingrdquo hypothesis is fundamentally wrong

our failure to come up with an analysis is not because we cannot discernan obscure rule but lies in the faulty assumption that we start with thenominative That form has a consistent phonetic property that any root-final obstruent (which is therefore word-final) is always voiceless whereasin the genitive form there is no such consistency If you look at thegenitive column the last consonant of the root portion of the word maybe either voiced or voicelessWe now consider a second hypothesis where we set up underlying

representations for roots which distinguish stems which have a finalvoiced obstruent in the genitive versus those with a final voiceless obstru-ent We may instead assume the following underlying roots

Under this hypothesis the genitive form can be derived easily The geni-tive form is the stem hypothesized in (3) followed by the suffix -a No ruleis required to derive voiced versus voiceless consonants in the genitiveThat issue has been resolved by our choice of underlying representationswhere some stems end in voiced consonants and others end in voicelessconsonants By our hypothesis the nominative form is simply the under-lying form of the noun stem with no suffixHowever a phonological rule must apply to the nominative form in

order to derive the correct phonetic outputWe have noted that no word inRussian ends phonetically with a voiced obstruent This regular fact allowsus to posit the following rule which devoices any word-final obstruent

(4) Final devoicing[-son] [-voice] _

By this rule an obstruent is devoiced at the end of the word As thisexample has shown an important first step in doing a phonologicalanalysis for phenomena such as word-final devoicing in Russian is toestablish the correct underlying representations which encode unpredict-able informationWhether a consonant is voiced cannot be predicted in English ([dεd]

dead [tεd] Ted [dεt] debt) and must be part of the underlying formSimilarly in Russian since you cannot predict whether a given root endsin a voiced or a voiceless consonant in the genitive that information mustbe part of the underlying form of the root That is information about theroot which cannot always be determined by looking at the surface formof the word itself it must be discovered by looking at the genitive form ofthe noun where the distinction between voiced and voiceless final con-sonants is not eliminated

42 Refining the concept of underlying form

It is important to understand what underlying forms are and what theyare not The nature of underlying forms can be best appreciated in thecontext of the overall organization of a grammar and how a given word is

Final voiced obstruent Final voiceless obstruentmuʒ lsquohusbandrsquo karandaʃ lsquopencilrsquoglaz lsquoeyersquo golos lsquovoicersquoraz lsquotimersquo les lsquoforestrsquoporog lsquothresholdrsquo porok lsquovicersquovrag lsquoenemyrsquo urok lsquolessonrsquoprud lsquopondrsquo tsvet lsquocolorrsquozavod lsquofactoryrsquo soldat lsquosoldierrsquogrib lsquomushroomrsquo trup lsquocorpsersquoxleb lsquobreadrsquo

generated in a sentence The structure of a grammar can be represented interms of the standard block model

This model implies that the output of one grammatical component formsthe input to the next component so the phonological component startswith whatever the morphological component gives it and applies its ownrules to give the surface representation (which are then subject to prin-ciples of physical interpretation within the phonetic component) Theoutput of the morphological component which is the input to the phon-ology is by definition the underlying form so we need to know a little bitabout what the morphological component does to understand what ispresented to the phonologyThe function of the morphological component is to assemble words in

the sense of stating how roots and affixes combine to form a particularword Thus the morphological component is responsible for combining anoun root [dag] and a plural affix [z] in English to give the word dog-s(ie dag-z) or in Russian the morphology combines a noun root [vagon]with an inflectional ending [a] according to rules of inflection for Russianto give the genitive word vagon-a Each morpheme is assumed to have asingle constant phonetically defined shape coming out of the morphology(there are a few exceptions such as the fact that the third-person-singularform of the verb be in English is [ɪz] and the first-person-singular form ofthat verb is [aeligm]) The phonetic realization of any morpheme is subject torules of phonology so while the morphology provides the plural mor-pheme z (spelled ltsgt) the application of phonological rules will makethat morpheme be pronounced as [s] as in cats or [ɨz] as in bushesIt is very important to understand that the grammar does not formally

derive one word from another (Some languages seem to have specialmorphological processes which we will not be discussing here that deriveone word from another ndash clipping such as Sally Sal would be anexample) Rather oneword derives from a given abstract root plus whateveraffixes are relevant and a related word derives by adding a different set ofaffixes to the same abstract root Accordingly the plural of a noun in Englishdoes not derive from the singular rather both the singular and the pluralforms derive from a common root no suffix is added to the root in thesingular and the suffix z is added to the root in the plural The Russiangenitive [vagona] also does not derive from the nominative nor does thenominative derive from the genitive Rather both derive from the rootvagon where the nominative adds no affix and the genitive adds the affix -aThe underlying form of a word is whatever comes out of the morph-

ology and is fed into the phonology before any phonological rules haveapplied The underlying form of the word [kaeligts] is kaeligt-z since that iswhat results in the morphology by applying the rule that combines a noun

Syntax rarr Morphology rarr Phonology rarr Phonetics

Underlying

representation

Surface

representation

Physical

output

root such as cat with the plural suffix The underlying form of the pluralword [kaeligts] is not kaeligt because the plural word has to have the pluralmorpheme However kaeligt is the underlying form of the singular word[kaeligt] There is no phonological rule which inserts z or s in order to form aplural The principles for combining roots and affixes are not part of thephonology and thus there is no need to include rules such as ldquoinsert [z] inthe pluralrdquo Be explicit about what you assume about morphology ina language ie that there is a plural suffix -z in English or a genitivesuffix -a in Russian As for the mechanics of phonological analysis youshould assume for example that the plural suffix is already present inthe underlying form and therefore do not write a rule to insert the pluralsuffix since that rule is part of morphology A phonological analysis statesthe underlying forms of morphemes and describes changes in the phono-logical shape of the root or suffixWe have concluded that the underlying form of the Russian word [prut]

lsquopondrsquo is prud In arriving at that conclusion we saw how important it isto distinguish the phonological concept of an underlying form from themorphological concept ldquobasic formrdquo where the singular form or anuninflected nominative form would be the morphological ldquobasic formrdquoAn underlying form is a strictly phonological concept and is not necessar-ily equivalent to an actually pronounced word (even disregarding thefundamental fact that underlying forms are discrete symbolic representa-tions whereas actually pronounced words are acoustic waveforms) It is arepresentation that is the foundation for explaining the variety of actualpronounciations found in the morpheme as determined by phonologicalcontextThe morphologically basic form of the Russian word for pond is the

unmarked nominative [prut] composed of just the root with no inflec-tional ending In contrast the phonological underlying form is prud foras we have seen if we assume the underlying form to be prut we cannotpredict the genitive [pruda] The word [prud] with a voiced consonant atthe end of the word does not appear as such in the language and thus thesupposition that the underlying form is prud is an abstraction given that[prud] by itself is never found in the language ndash it must be inferred inorder to explain the actual data The basis for that inference is the genitiveform [pruda] which actually contains the hypothesized underlying formas a subpart It is important to understand however that the underlyingform of a root may not actually be directly attested in this way in anysingle word and we will discuss this point in section 46

43 Finding the underlying form

A similar problem arises in explaining the partitive and nominative formsof nouns in Finnish The first step in understanding the phonologicalalternation seen here is to do a standard preliminary morphologicalanalysis of the data which involves identifying which parts of a wordcorrelate with each aspect of word structure (such as root meaning or

grammatical case) The following examples illustrate that the nominativesingular suffix is Oslash (ie there is no overt suffix in the nominative singular)and the partitive singular suffix is -aelig which alternates with -a if there is aback vowel somewhere before it in the word (we will not be concernedwith that vowel alternation in the partitive suffix)

We might assume that the underlying form of the root is the same as thenominative (which has no suffix) The problem which these data pose isthat in some nouns the partitive appears to be simply the nominative plusthe suffix -aelig ~ -a (for example muuri ~ muuria) but for other nouns thefinal vowel alternates with [i] in the nominative and [e] in the partitive(eg joki ~ jokea) It is obvious that the nature of the following vowel doesnot explain this alternation since the same surface-quality suffix vowelcan appear after either e or i ndash compare jokea nimeaelig where [e] appears

Nominative sg Partitive sgaamu aamua lsquomorningrsquohopea hopeaa lsquosilverrsquokatto kattoa lsquoroofrsquokello kelloa lsquoclockrsquokirja kirjaa lsquobookrsquokylmaelig kylaeligaelig lsquocoldrsquokoulu koulua lsquoschoolrsquolintu lintua lsquobirdrsquohylly hyllyaelig lsquoshelfrsquokoslashmpeloslash koslashmpeloslashaelig lsquoclumsyrsquonaeligkoslash naeligkoslashaelig lsquoappearancersquo

joki jokea lsquoriverrsquokivi kiveaelig lsquostonersquomuuri muuria lsquowallrsquonaapuri naapuria lsquoneighborrsquonimi nimeaelig lsquonamersquokaappi kaappia lsquochest of drawersrsquokaikki kaikkea lsquoallrsquokiirehti kiirehtiaelig lsquohurryrsquolehti lehteaelig lsquoleafrsquomaeligki maeligkeaelig lsquohillrsquoovi ovea lsquodoorrsquoposti postia lsquomailrsquotukki tukkia lsquologrsquoaeligiti aeligitiaelig lsquomotherrsquoenglanti englantia lsquoEnglandrsquojaeligrvi jaeligrveaelig lsquolakersquokoski koskea lsquowaterfallrsquoreki rekeaelig lsquosledgersquovaeligki vaeligkeaelig lsquopeoplersquo

before both [a] and [aelig] versus muuria kiirehtiaelig where [i] appears beforethese same vowels Nor can the preceding consonant be called upon topredict what vowel will appear in the partitive as shown by pairs such astukkia kaikkea versus lehteaelig aeligitiaeligThis is an area where there is (potentially) a difference between

language-learning pedagogy and a formal linguistic analysis Faced withthe problem of learning the inflectional distinction muuri ~ muuria versusjoki ~ jokea a second-language class on Finnish might simply have thestudent memorize a list of words like joki ~ jokea where the vowel changesin the inflectional paradigm From the point of view of linguistic analysisthis is the wrong way to look at the question since it implies that this isnot a rule-governed property of the language However second-languagelearning is not the same as linguistic analysis a class in foreign-languageinstruction has a different goal from a class in analysis and some studentsin a language class may receive greater practical benefit from just mem-orizing a list of words Thus it is important to distinguish the teachingmethod where one learns arbitrary lists and a theoretically based analy-sis One simply cannot predict what vowel will appear in the partitiveform if one only considers the pronunciation of the nominative Thismeans nominative forms are not the same as underlying forms (some-thing that we also know given the previous Russian example) The under-lying representation must in some way contain that information whichdetermines whether there will be a vowel alternation in a given wordIn looking for the phonological basis for this vowel alternation it is

important to realize that the alternation in stem-final vowels is notchaotic for we find precisely two possibilities either i in the nominativepaired with i in the partitive or i in the nominative paired with e in thepartitive ndash never for example i paired with u or i paired with o Moreoveronly the vowel i enters into such a vowel alternation in Finnish so thereare no nouns with o in the nominative which is replaced by u in thepartitive nor is u in the nominative ever replaced by o or any other vowelin the partitive One final fact about the data in (6) suggests exactly howthe right underlying representations can explain this alternation of theeight vowels of Finnish [i y e oslash aelig u o a] all of them appear at the end ofthe word except the vowel e Now since the stem of the word for lsquonamersquowhich appears as nimi in the nominative actually appears on the surfaceas nime- in the partitive it is not at all unreasonable to assume that theunderlying form of the stem is in fact nime It would be a bit bizarre toassume an underlying form such as nima since the vowel [a] neverappears in that position in any form of this word the most naturalassumption to make is that the underlying form of a morpheme is actu-ally composed of segments found in some surface manifestation of themorpheme On the other hand the stem of the word for lsquowallrsquo is pro-nounced muuri in both the nominative and the partitive and thereforethere is no reason to assume that it is underlyingly anything otherthan muuriWe will then assume that the underlying vowel at the end of the stem is

actually reflected by the partitive form and thus we would assume

This is a naturalassumption but notan absolute rule aswe see in chapter 8Underlying formscan containsegments not foundin any form of theword Only whenthere is strongevidence fordeparting from thisassumption are youjustified in settingup underlyingforms with suchabstract elements

underlying representations such as joke nime kive lehte ove andso on as well as muuri naapuri kaappi tukki and so on Theunderlying form of partitive [joke-a] would thus be joke-a that is norule at all is required to explain the partitive Instead a rule is neededto explain the surface form of the nominative [joki] which derives fromjoke A very simple neutralizing rule can explain the surface form of thenominative underlying word-final e is raised to i

(7) Final vowel raising

This rule is neutralizing since the distinction between i and eis neutralized by applying this rule an underlying e becomesphonetic [i]Apart from illustrating how important correct underlying forms are

these two examples have also shown that it is dangerous and incorrect inthese two cases to assume that the ldquomost basicrdquo form of a word accordingto morphological criteria is also the underlying form of the word Toreiterate the underlying form of a morpheme is a hypothesis set forthby the analyst a claim that by assuming such-and-such an underlyingform plus some simple set of rules (which need to be discovered by theanalyst) the observed variation in the shape of morphemes can beexplained

Kerewe To better understand the reasoning that leads to correct under-lying forms we investigate other examples Consider the following datafrom Kerewe (Tanzania)

(8) Infinitive 1sg habitual 3sg habitual Imperativekupaamba mpaamba apaamba paamba lsquoadornrsquokupaaŋga mpaaŋga apaaŋga paaŋga lsquoline uprsquokupima mpima apima pima lsquomeasurersquokupuupa mpuupa apuupa puupa lsquobe lightrsquokupeketʃa mpeketʃa apeketʃa peketʃa lsquomake fire

with stickrsquokupiinda mpiinda apiinda piinda lsquobe bentrsquokuhiiga mpiiga ahiiga hiiga lsquohuntrsquokuheeka mpeeka aheeka heeka lsquocarryrsquokuhaaŋga mpaaŋga ahaaŋga haaŋga lsquocreatersquokuheeba mpeeba aheeba heeba lsquoguidersquokuhiima mpiima ahiima hiima lsquogasprsquokuhuuha mpuuha ahuuha huuha lsquobreathe intorsquo

_ [ ]+high

+sylndashrdminusbackndashlo

We notice that every infinitive begins with ku- which we surmise is theprefix for the infinitive the third-singular habitual form has the prefix a-and the first-singular habitual has the prefix m- the imperative involvesno prefix In addition to segmental prefixes there is a change in the firstconsonant of the stem in some verbs in some contexts The initial con-sonant of the verb meaning lsquoguidersquo alternates between [h] and [p] with [p]appearing in the first-singular habitual after [m] and [h] appearing else-where Since this stem appears in two surface variants [heeba] and[peeba] two plausible hypotheses are immediately possible the stem isunderlyingly peeba or the stem is underlyingly heeba If we assumethat the stem is underlyingly heeba we require a rule to explain thedivergence between the predicted form of the first-singular habitualform ndash we would expect [mheeba] [mhiima] etc ndash and the actual formof the verb [mpeeba] [mpiima] and so on Since in fact we do not see thesequence mh anywhere in the data we might assume the followingneutralizing rule

(9) Postnasal hardening

If on the other hand we assume that the root is underlyingly peebawe would need a rule which changes p into [h] when not preceded by anasal ndash in other words when preceded by a vowel or by nothing There isno single property which groups together word-initial position andvowels Thus the supposed rule changing p to [h] would have to be adisjunction of two separate environments

This suggests that rule (10) is wrongMore important than the greater complexity of the rule entailed by

assuming that the word for lsquoguidersquo is underlyingly peeba it is empiricallywrong rule (10) implicitly claims that p should always become [h] word-initially or after a vowel but this is falsified by forms such as kupaambaapaamba paamba lsquoadornrsquo and kupaaŋga apaaŋga paaŋga lsquoline uprsquo If weassume the stems uniformly begin with p then we cannot predictwhether the imperative or infinitive has [h] (kuhaaŋga) or [p] (kupaaŋga)On the other hand if we assume an underlying contrast between initial hand initial p ndash ie haaŋga lsquocreatersquo paaŋga lsquoarrangersquo ndash then we can cor-rectly distinguish those stems which begin with h from those whichbegin with p when no nasal precedes as well as correctly neutralizingthat distinction just in case the stem is preceded by a nasal (mpaaŋgalsquoI creatersquo lsquoI arrangersquo)

In this examplewe only have directevidence for thechange after m soit would be possibleto restrict our ruleto the more specificcontext ldquoafter mrdquoBut this would runcounter to basicassumptions ofscience that we seekthe most generalexplanationspossible not themost restrictedones

[ ]+sg [ ]+nas-cont+ant

eacuteecirceuml

ugraveuacuteucirc

+antminuscorminusvoi

rarr⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+sg+contminusant

⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

⎧⎨⎩

⎫⎬⎭

The rule formalization in (9) exploits a widely used notion about howrules apply known as structure preservation Notice that the structuralchange specified mentions only that h becomes [ndashcont +ant] which aretwo features that characterize the difference between h and [p] There aretwo other actual changes in feature which are not explicitly mentionednamely that the segment becomes [+cons ndashson] These values can beautomatically predicted from the fact that in this language there isonly one voiceless [+ant ndashcor] stop namely [p] The idea underlyingstructure preservation is that each language defines an inventory ofsegments and the structural change of a rule changes from one soundwithin the inventory to another sound Specifying that change as-cont + ant + cons - sonfrac12 fully specifies what the result of the ruleis but -cont + antfrac12 identifies the same unique segment of the languagemore economically

English plurals A further illustration of how to determine the correctunderlying representation comes from English As the following examplesillustrate the surface form of the plural suffix varies between [s] and [z] (aswell as [ɨz] to be discussed later)

The generalization regarding distribution is straightforward [s] appearsafter a voiceless segment and [z] appears after a voiced one (be it anobstruent a liquid nasal or a vowel)This same alternation can be found in the suffix marking the third-

singular present-tense form of verbs

If we suppose that the underlying form of the affixes for noun plural andthird-singular present verbs is z then we would assume the followingrule to derive the phonetic variant [s]

(13) [-son] [-voice] [-voice] __

kaeligps caps kaeligbz cabs klaeligmz clamskaeligts cats kaeligdz cads kaelignz canskaks cocks kagz cogs kaɹz carspɹuwfs proofs hʊvz hooves gəlz gulls

flijz fleasplaeligwz plowspjɹez purees

slaeligps slaps staeligbz stabs slaeligmz slamshɪts hits hajdz hides kaelignz canspowks pokes dɪgz digs haeligŋz hangslaeligfs laughs θrajvz thrives hijlz healspɪθs piths bejethz bathes hɪɹz hears

flajz flies vijtowz vetoes

On the other hand if we were to assume that these suffixes are under-lyingly s we would assume the following rule

(14) [-son] [+voice] [+voice] __

In terms of the simplicity and generality of these two rules the analysesare comparable Both formulations require the same number of phon-etic specifications to state the rule and both formulations apply togeneral and phonetically natural classes However the two analysesdiffer quite significantly in terms of their overall predictions forEnglish The implicit prediction of the first rule (13) is that there shouldbe no voiced obstruents after voiceless segments in English since thatrule would devoice all such obstruents This generalization seems to becorrect there are no words like [jəkd] [pɪfz] [sdap] The implicit pre-diction of the second rule (14) is different that rule implies that thereshould be no voiceless segments after any voiced segments This ismanifestly incorrect as shown by the existence of words such as [hɪs]hiss [paeligθ] path [daeligns] dance [fals] false We prefer a hypothesis whichmakes the correct prediction about the phonetic structure of the lan-guage as a whole and thus we select the underlying form z and a ruledevoicing obstruents after voiceless segments Looking for such asym-metries plays an important role in determining which of two hypothesesis the correct oneThe alternation z ~ s is not limited to the two affixes -z lsquopluralrsquo and -z

lsquo3sg present tensersquo The rule of devoicing can also be seen applying to thepossessive suffix -z

Moreover certain auxiliary verbs such as has [haeligz] and is [ɪs] undergo areduction in casual speech so that they appear simply as [s] or [z] thechoice between these two being determined by the devoicing rule whichwe have motivated

The devoicing rule (13) automatically explains the alternation in thesurface shape of the consonant here as well

Noun Noun + posskaeligt kaeligts catsləg sləgz slugklaeligm klaeligmz clamsnow snowz snow

Noun + has Reduced Noun + is Reduceddʒaeligk haeligz ijʔn dʒaeligks ijʔn dʒaeligk ɪz ijɾɪŋ dʒaeligks ijɾɪŋ Jackpaeligt haeligz ijʔn paeligts ijʔn paeligt ɪz ijɾɪŋ paeligts ijɾɪŋ Patdʒεn haeligz ijʔn dʒεnz ijʔn dʒεn ɪz ijɾɪŋ dʒεnz ijɾɪŋ Jenbab haeligz ijʔn babz ijʔn bab ɪz ijɾɪŋ babz ijɾɪŋ Bobdʒow haeligz ijʔn dʒowz ijʔn dʒow ɪz ijɾɪŋ dʒowz ijɾɪŋ Joe

Jita tone It is important to look for correlations which may lead tocausal explanations in analyzing data Consider the following data fromJita (Tanzania) concentrating on the tones of morphemes (H or high toneis marked with acute accent L or low-toned syllables are unmarked)

(17) a

We can conclude that there is a prefix oku- perhaps marking the infini-tive a suffix -a appearing at the end of every verb and two suffixes -ir-lsquoforrsquo and -an- lsquoeach otherrsquo There are also root morphemes -βum- lsquohitrsquo-siβ- lsquoblockrsquo as well as -luacutem- lsquobitersquo and -kuacuteβ- lsquofoldrsquo We decide that lsquobitersquoand lsquofoldrsquo underlyingly have H tones in part based on the fact thatthere actually is an H tone on the vowels of these roots in the simplestverb formsIn addition we observe that the suffixes -ir- and -an- have H tone when

they come immediately after these verb roots The suffixes do not haveH tone after the first set of roots appearance of H on the suffix iscorrelated with which morpheme immediately precedes the suffix Sincethis unpredictable property is correlated with the preceding root mor-pheme it must therefore be an aspect of the underlying form of thepreceding morphemeWe thus explain the H tone on these suffix morphemes by positing that

[oku-lum-aacuten-a] derives from underlying oku-luacutem-an-a by applying a ruleof tone shift which shifts an H tone rightward to the following syllable aslong as the syllable is not word-final Because of the restriction that H doesnot shift to a final syllable the underlying H surfaces unchanged in[okuluacutema]Now consider the following data

When the L-toned roots of (17a) stand after the object prefixes -mu- lsquohimherrsquo and -t ʃi- lsquoitrsquo they have an H tone at the beginning of the root Againsince the presence of the H is correlated unpredictably with the prefixes-mu- and -t ʃi- we hypothesize that the tones are part of the underlying

okuβuma lsquoto hitrsquo okusiβa lsquoto blockrsquookuβumira lsquoto hit forrsquo okusiβira lsquoto block forrsquookuβumana lsquoto hit eorsquo okusiβana lsquoto block eorsquookuβumirana lsquoto hit for eorsquo okusiβirana lsquoto block for eorsquo

okuluacutema lsquoto bitersquo okukuacuteβa lsquoto foldrsquookulumiacutera lsquoto bite forrsquo okukuβiacutera lsquoto fold forrsquookulumaacutena lsquoto bite eorsquo okukuβaacutena lsquoto fold eorsquookulumiacuterana lsquoto bite for eorsquo okukuβiacuterana lsquoto fold for eorsquo

okumuβuacutema lsquoto hit himherrsquo okumusiacuteβa lsquoto blockhimherrsquo

okumuβuacutemira lsquoto hit for himherrsquo okumusiacuteβira lsquoto block forhimherrsquo

okutʃiβuacutema lsquoto hit itrsquo okutʃisiacuteβa lsquoto block itrsquookutʃiβuacutemira lsquoto hit for itrsquo okutʃisiacuteβira lsquoto block for itrsquo

representation of the prefixes ndash the prefixes are muacute and t ʃiacute and theH toneshifts to the right by the tone shift rule which we have already posited

44 Practice at problem solving

You should now be able to apply this reasoning to data which poseanalogous problems a series of examples are given in this section forpractice

Chamorro vowel alternations There are alternations in the quality ofvowels in initial syllables in some contexts seen in the following data fromChamorro (Mariana Islands)

What underlying representations and what rule or rules are required toaccount for these data When you answer this question you shouldconsider two hypotheses which differ in terms of what form is taken tobe underlying ndash what are the two most obvious ways of treating thesealternations One of these hypotheses is clearly wrong the other is thecorrect hypothesis

Korean Now consider the following data from Korean The first columnin (20) the imperative seems to involve a vowel suffix One reason tothink that there is an imperative suffix is that every imperative endseither in the vowel a or in ə (the choice between a versus ə is based onthe vowel which precedes that suffix a or o versus other vowels andcan be ignored here) A second reason comes from comparing the

gwiacutehən lsquofishrsquo i gwiacutehən lsquothe fishrsquoguacuteməʔ lsquohousersquo i giacuteməʔ lsquothe housersquokaacutetta lsquoletterrsquo joʔ kaacutetta lsquoa letter (object)rsquo

i kǽtta lsquothe letterrsquotsuacutepa lsquocigarettesrsquo i tsiacutepa lsquothe cigarettesrsquofiacutenoʔ lsquotalkrsquo mi fiacutenoʔ lsquolots of talkrsquotuacutenuʔ lsquoto knowrsquo en tiacutenuʔ lsquoyou knowrsquo

tsuacutegoʔ lsquojuicersquo mi tsiacutegoʔ lsquolots of juicersquosoacuteŋsuŋ lsquovillagersquo i seacuteŋsuŋ lsquothe villagersquohuacuteluʔ lsquouprsquo saelign hiacuteluʔ lsquoupwardrsquopeacutetsu lsquochestrsquo i peacutetsu lsquothe chestrsquotoacutemtu lsquokneersquo i teacutemu lsquothe kneersquooacutetdut lsquoantrsquo mi eacutetdut lsquolots of antsrsquooacuteksuʔ lsquohillrsquo gi eacuteksuʔ lsquoat the hillrsquodaacuteŋkulu lsquobig onersquo i dǽŋkulu lsquothe big onersquolaacutehi lsquomalersquo i lǽhi lsquothe malersquolaacutegu lsquonorthrsquo saelign lǽgu lsquotoward northrsquopuloacutennun lsquotrigger fishrsquo i puloacutennun lsquothe trigger fishrsquomundoacuteŋgu lsquocowrsquos stomachrsquo i mundoacuteŋgu lsquothe cowrsquos stomachrsquoputamoneacuteda lsquowalletrsquo i putamoneacuteda lsquothe walletrsquo

In formalizing therule take note ofthe vowel inventoryobserved in thelanguage and seehow the notion of astructure-preserving effectcan simplify thestatement of yourrule

imperative and the plain present forms Comparing ana and annɨnta or kamaand kamnɨnta we can see that for each verb the portions common to boththe imperative and the plain present are respectively an- and kam- From thiswe deduce that there must be a suffix either -a or -ə which marks theimperative and another suffix -ninta which marks the plain present

What is the underlying form of these verb stems and what phonologicalrule or rules are required to account for the variations that are seen in thesurface shape of the various stems

Koasati What is the underlying formof thefirst-singular possessive prefixinKoasati (Louisiana) andwhat phonological rule applies in these examples

Imperative Plain presentana annɨnta lsquohugrsquokama kamnɨnta lsquowindrsquosinə sinnɨnta lsquowear shoesrsquotrsquoatɨmə trsquoatɨmnɨnta lsquotrimrsquo

nəmə nəmnɨnta lsquooverflowrsquo

nama namnɨnta lsquoremainrsquotʃhama tʃhamnɨnta lsquoendurersquoipə imnɨnta lsquoput onrsquokupə kumnɨnta lsquobendrsquotʃəpə tʃəmnɨnta lsquofoldrsquotata tannɨnta lsquoclosersquoputhə punnɨnta lsquoadherersquotʃotʃha tʃonnɨnta lsquofollowrsquo

məkə məŋnɨnta lsquoeatrsquosəkrsquoə səŋnɨnta lsquomixrsquotakrsquoa taŋnɨnta lsquopolishrsquotʃukə tʃuŋnɨnta lsquodiersquoikə iŋnɨnta lsquoripenrsquo

Noun My Napahtʃaacute amapahtʃaacute lsquoshadowrsquo

asiktʃiacute amasiktʃiacute lsquomusclersquoilkanoacute amilkanoacute lsquoright sidersquoifaacute amifaacute lsquodogrsquoapoacute amapoacute lsquograndmotherrsquoiskiacute amiskiacute lsquomotherrsquopatʃokkoacuteka ampatʃokkoacuteka lsquochairrsquotowaacute antowaacute lsquoonionrsquokastoacute aŋkastoacute lsquoflearsquobajaacutena ambajaacutena lsquostomachrsquotaacuteta antaacuteta lsquofatherrsquotʃofkoniacute aɲtʃofkoniacute lsquobonersquokitiɫkaacute aŋkitiɫkaacute lsquohair bangsrsquotoniacute antoniacute lsquohiprsquo

Matuumbi What phonological rules pertaining to consonants operatein the following examples from Matuumbi What are the underlyingforms of the stems of the words for lsquoropersquo lsquopalmrsquo lsquotonguersquo lsquopiece ofwoodrsquo lsquopolersquo and lsquocoveredrsquo Ignore tonal changes

A certain degree of uncertainty regarding the exact underlying form ofthe plural prefix is expected However the underlying form of the stemshould be clear and should be the focus of your analysis You should beable to explain these alternations with two rules In formalizing the rulespay attention to the concept of structure preservation in rules

45 Underlying forms and sentence-levelphonology

In the examples which we have considered so far we have been compar-ing morphologically related words such as a nominative and a genitiveand we have seen that an underlying distinction may be preserved in oneword in a particular inflected form (because in that inflected form theconditions for applying the phonological rule are not satisfied) butthe difference is neutralized in a related word where the conditions forthe rule are present We now consider two additional cases where under-lying distinctions are neutralized depending on context and the neutral-ization takes place within one and the same word depending on wherethe word appears in a sentence What this shows is that phonology is notjust about variations in pronunciation between words but also includesvariations in the pronunciation of a word in different sentential contexts

Singular Plurallugoacutei ŋgoacutei lsquoropersquolugoloacutekaacute ŋgoloacutekaacute lsquostraightrsquolubaacuteu mbaacuteu lsquoribrsquolubaacutegalo mbagaacutelo lsquolathrsquoludʒiiacuteŋgjaacute ɲdʒiiacuteŋgjaacute lsquoenteredrsquolulaaacutela ndaaacutela lsquopepperrsquolulɪmiacute ndɪmi lsquotonguersquolulɪndɪɪlaacute ndɪndɪɪlaacute lsquoguardedrsquolupalaacuteaiacute mbalaacuteaiacute lsquobald headrsquolupaaacutelaacute mbaaacutelaacute lsquowantedrsquoluteacuteelaacute ndeelaacute lsquopiece of woodrsquolukiacuteligo ŋgiliacutego lsquoplace for initiatesrsquolukiacuteli ŋgiacuteli lsquopalmrsquo

lujɪmaacute ɲdʒiacutema lsquopolersquolujoacuteka ɲdʒoacuteka lsquostomach wormrsquo

lujuacutesiacute ɲdʒuacutesi lsquobeersquolujuacuteweacute ɲdʒuacutewe lsquopumpkin plantrsquoluwɪkɪljaacute ŋgwɪkɪljaacute lsquocoveredrsquo

451 Korean final CsThe first case is a rule of Korean that nasalizes stops before nasal consonants(a rule that we have seen operating within words in the preceding section)Thefirst set of examples shows theword for lsquoricersquowhen said alone orwhen itis followedby variouswordswhich beginwith oral consonants and vowels Inthesedata the last consonant of theword for lsquoricersquo is pronounced as [p] In thesecond set of examples thewordwhich follows lsquoricersquo beginswith a nasal andin that case the final consonant of the word for lsquoricersquo is pronounced as [m]

(23) a

Compare those examples with the following examples with the word forlsquochestnutrsquo

(24) a

pap lsquoricersquopap anməkətrsquoa lsquodidnrsquot eat ricersquorice didnrsquot-eatpap winmoke tuətrsquoa lsquoput rice on the

upper floorrsquorice on-upper-floor putpap saməkətrsquoa lsquoate rice at a storersquorice ate-at-storepap totuki humtʃhəkatrsquoa lsquoa thief stole ricersquorice thief-(subj) stole

pam mani məkəla lsquoeat rice a lotrsquorice lot eatpam mək-imjən lsquoif eats ricersquorice eat-ifpam nəmu masikrsquoe məkətrsquoa lsquoI enjoyed rice quite

a lotrsquorice very deliciously atepam nəkho lsquoadd ricersquorice add

pam lsquochestnutrsquopam anməkətrsquoa lsquodidnrsquot eat chestnutrsquochestnut didnrsquot-eatpam winmoke tuətrsquoa lsquoput chestnut on the

upper floorrsquochestnut on-upper-floor putpam saməkətrsquoa lsquoate chestnut at a storersquochestnut ate-at-storepam totuki humtʃhəkatrsquoa lsquoa thief stole chestnutrsquochestnut thief-(subj) stole

pam mani məkəla lsquoeat chestnut a lotrsquochestnut lot eatpam mək-imjən lsquoif eats chestnutrsquochestnut eat-if

In fact the (b) phrases above are actually ambiguous as to whether theword being pronounced means lsquochestnutrsquo or lsquoricersquoThe last consonant of the word for lsquochestnutrsquo is always [m] so we would

presume that the underlying form of that word is pam Since the word forlsquoricersquo varies between [pap] and [pam] and since we know that the under-lying form cannot be pam (this is the underlying form of lsquochestnutrsquo andlsquochestnutrsquo cannot have the same underlying form as lsquoricersquo since they do notbehave the same) we conclude that the underlying form of the word forlsquoricersquo is pap and that a nasalization rule changes p (in fact all stops) tonasals before a nasalWhether aword undergoes that rule depends onwhatfollows the final consonant One and the same word can be pronounceddifferently depending on the properties of the phrase in which it appears

452 Matuumbi toneIn the Korean case which we just considered it happens that the under-lying form of the word is the same as the way the word is pronouncedwhen it is said alone This situation does not hold in Matuumbi where onehas to know how a word is pronounced when it is not at the end of anutterance in order to determine the underlying form of the word Thewords in (25) have an H tone (marked with an acute accent) on the secondvowel from the beginning of the word when said alone When anotherword follows they seem to lose that H tone

In contrast the words of (26) which also have an H tone on the secondvowel from the beginning of the word when the word is said alone keeptheir H tone when another word follows

pam nəmu masikrsquoe məkətrsquoa lsquoI enjoyed chestnutquite a lotrsquo

chestnut very deliciously atepam nəkho lsquoadd chestnutrsquochestnut add

kiwɪkɪljo lsquocoverrsquo ŋga kiwɪkɪljo liacute lsquoit isnrsquot a coverrsquolubaacutegalo lsquolathrsquo ŋga lubagalo liacute lsquoit isnrsquot a lathrsquomikoacuteta lsquosugar canesrsquo ŋga mikota liacute lsquoit isnrsquot sugar canesrsquoŋguuacuteŋguni lsquobedbugrsquo ŋga ŋguuŋguni liacute lsquoit isnrsquot a bedbugrsquolukoacutelogo lsquobreweryrsquo ŋga lukologo liacute lsquoit isnrsquot a breweryrsquomabaacuteando lsquothighsrsquo ŋga mabaando liacute lsquoit isnrsquot thighsrsquokikoacuteloombe lsquoshellrsquo ŋga kikoloombe liacute lsquoit isnrsquot a shellrsquolipɪtanʊʊŋgʊ lsquorainbowrsquo ŋga lipɪtanʊʊŋgʊ liacute lsquoit isnrsquot a rainbowrsquo

lukoacuteŋgobe lsquowoodrsquo ŋga lukoacuteŋgobe liacute lsquoitrsquos not woodrsquokitʊkʊtʊkʊ lsquoquelea birdrsquo ŋga kitʊkʊtʊkʊ liacute lsquoitrsquos not a quelearsquodiiacutewai lsquowinersquo ŋga diiacutewai liacute lsquoitrsquos not winersquolukoacuteoŋgono lsquochicken legrsquo ŋga lukoacuteoŋgono liacute lsquoitrsquos not a legrsquolukoacuteoŋgowe lsquomarblersquo ŋga lukoacuteoŋgowe liacute lsquoitrsquos not marblersquo

There are no words in Matuumbi which are toneless when said bythemselves thus kitekeljo said by itself is an unattested kind of wordThere is a clear contrast in tonal behavior between the words in (25)where the presence of an H tone on the second vowel depends onwhether the word is said alone or is followed by another word andthose in (26) where the second vowel always has an H tone The solutionto this puzzle is that the words in (26) have an underlying H tone ontheir second vowel and thus nothing happens to that tone but thewords in (25) have no underlying H and instead get an H at the end ofan utterance by a rule that assigns an H tone to the second vowel of atoneless word which comes at the end of an utterance Thus in the caseof Matuumbi tone the contrast between underlyingly toneless words andwords with underlying H is best revealed by looking at the word when itappears not by itself it is the citation form of the word that undergoesthe neutralization rule which is the opposite of the situation we justencountered in Korean

46 Underlying forms and multiple columnsin the paradigm

The following data from Samoan illustrate the very important point thatit is wrong to think of deriving underlying forms by chopping off affixesfrom some single column of data In the first set of examples our initialtask is to deduce the underlying form of each of the verb roots and theaffix for the perfective form

matoacutegolo lsquowaterbucksrsquo ŋga matoacutegolo liacute lsquoitrsquos not waterbucksrsquomiviacuteriiŋgo lsquocirclesrsquo ŋga miviacuteriiŋgo liacute lsquoitrsquos not circlesrsquokijoacutegojo lsquobird (sp)rsquo ŋga kijoacutegojo liacute lsquoitrsquos not a birdrsquokikaacutelaaŋgo lsquopanrsquo ŋga kikaacutelaaŋgo liacute lsquoitrsquos not a panrsquo

Simple Perfectiveolo oloia lsquorubrsquolafo lafoia lsquocastrsquoaŋa aŋaia lsquofacersquousu usuia lsquoget up and go earlyrsquotau tauia lsquoreach a destinationrsquotaui tauia lsquorepayrsquosaʔili saʔilia lsquolook forrsquovaŋai vaŋaia lsquoface each otherrsquopaʔi paʔia lsquotouchrsquonaumati naumatia lsquobe waterlessrsquosauni saunia lsquopreparersquoseŋi seŋia lsquobe shyrsquolele lelea lsquoflyrsquosuʔe suʔea lsquouncoverrsquo

Examples such as oloia aŋaia and usuia suggest that the perfectivesuffix is -ia and the simple form of the verb reflects the underlyingform of the root Examples such as seŋi ~ seŋia or lele ~ lele suggesta phonological rule since the combination of the presumed stemsseŋi and lele with the perfective affix -ia would result in the incorrectforms seŋiia leleia However this problem can be corrected bypositing a phonological rule which deletes a front vowel when it ispreceded by a front vowel In the formalization of the rule we saythat the second front vowel is replaced by zero which means that it isdeleted

(28) Vowel-cluster reduction

An alternative hypothesis would be that [i] is inserted between a backvowel and the vowel [a] if we were to presume that the perfective suffix isunderlyingly a

This latter rule is more complicated than (28) Additional data willshow that in addition this would just be plain wrong We abandonthe idea of inserting the vowel i and conclude that the underlying formof the perfective suffix must be -ia hence there must be a rule deletinga front vowel after a front vowel We would then conclude that theunderlying representation of roots is best revealed in the simple verbrather than the perfective since the simple form of the verb showswhether the stem ends with i a vowel which may be deleted in theperfectiveA rather different conclusion about arriving at underlying forms would

have to be drawn from the following additional Samoan examples

taʔe taʔea lsquosmashrsquotafe tafea lsquoflowrsquo

taupule taupulea lsquoconferrsquopalepale palepalea lsquohold firmrsquo

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+syl-back

Oslash

rarr⎡⎢⎣

⎤⎥⎦

+syl+back

Oslash ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]+low

Simple Perfective Simple Perfectivetu tulia lsquostandrsquo au aulia lsquoflow onrsquotau taulia lsquocostrsquo matau mataulia lsquoobserversquoʔalo ʔalofia lsquoavoidrsquo ili ilifia lsquoblowrsquo

oso osofia lsquojumprsquo ulu ulufia lsquoenterrsquosao saofia lsquocollectrsquo taŋo taŋofia lsquotake holdrsquoasu asuŋia lsquosmokersquo soa soaŋia lsquohave a friendrsquo

Here we see that the perfective form of the verb contains a consonantwhich is not present in the simple form That consonant can be any one ofl f ŋ s t m or ʔ given these data An attempt to predict the nature of thatconsonant by an insertion rule proves fruitless We could attempt toinsert an appropriate consonant on the basis of the preceding vowel butwhile l appears after u so do f ([ulufia]) ŋ ([tofuŋia]) and s ([valusia]) andwhile f appears after o so do ŋ ([ifoŋia]) m ([ŋalomia]) and s ([milosia]) Inshort it is simply impossible to predict from anything in the environmentwhat the consonant of the perfective is going to be if we start with thesimple form as the underlying form that consonant must be part of theunderlying representation of the root Thus the underlying forms of thissecond set of roots would be as follows

pole poleŋia lsquobe anxiousrsquo fesili fesiliŋia lsquoquestionrsquoifo ifoŋia lsquobow downrsquo ʔote ʔoteŋia lsquoscoldrsquoula ulaŋia lsquomockrsquo tofu tofuŋia lsquodiversquomilo milosia lsquotwistrsquo laʔa laʔasia lsquosteprsquovalu valusia lsquoscrapersquo taŋi taŋisia lsquocryrsquovela velasia lsquobe cookedrsquo motu motusia lsquobreakrsquoapi apitia lsquobe lodgedrsquo mataʔu mataʔutia lsquofearrsquoeʔe eʔetia lsquobe raisedrsquo sau sautia lsquofallrsquolava lavatia lsquobe ablersquo oʔo oʔotia lsquoarriversquou utia lsquogriprsquo ufi ufitia lsquocoverrsquopuni punitia lsquobe blockedrsquo tanu tanumia lsquocover uprsquosiʔo siʔomia lsquobe enclosedrsquo moʔo moʔomia lsquoadmirersquoŋalo ŋalomia lsquoforgetrsquo tao taomia lsquocoverrsquosopo sopoʔia lsquogo acrossrsquo fana fanaʔia lsquoshootrsquo

tul lsquostandrsquo aul lsquoflow onrsquotaul lsquocostrsquo mataul lsquoobserversquoʔalof lsquoavoidrsquo ilif lsquoblowrsquo

osof lsquojumprsquo uluf lsquoenterrsquosaof lsquocollectrsquo taŋof lsquotake holdrsquoasuŋ lsquosmokersquo soaŋ lsquohave a friendrsquopoleŋ lsquobe anxiousrsquo fesiliŋ lsquoquestionrsquoifoŋ lsquobow downrsquo ʔoteŋ lsquoscoldrsquoulaŋ lsquomockrsquo tofuŋ lsquodiversquomilos lsquotwistrsquo laʔas lsquosteprsquovalus lsquoscrapersquo taŋis lsquocryrsquovelas lsquobe cookedrsquo motus lsquobreakrsquoapit lsquobe lodgedrsquo mataʔut lsquofearrsquoeʔet lsquobe raisedrsquo saut lsquofallrsquolavat lsquobe ablersquo oʔot lsquoarriversquout lsquogriprsquo ufit lsquocoverrsquopunit lsquobe blockedrsquo tanum lsquocover uprsquosiʔom lsquobe enclosedrsquo moʔom lsquoadmirersquoŋalom lsquoforgetrsquo taom lsquocoverrsquosopoʔ lsquogo acrossrsquo fanaʔ lsquoshootrsquo

The postulation of underlying consonants at the end of these roots entailsthe addition of a phonological rule in order to account for the surfaceform of the simple verb where there is no final consonant Noting that noword ends in a consonant phonetically in these examples we can postu-late the following rule of final consonant deletion

(32) Final consonant deletion

C Oslash _

The underlying forms of these verbs can be heuristically derived byeliminating the perfective affix -ia from the perfective form Howevernotice that we made a different heuristic assumption for the first group ofroots which underlyingly ended in a vowel The point is that an under-lying representation is whatever is required to correctly predict all of thesurface variants of a given morpheme it does not necessarily derive fromany one column in a paradigmIt is also important to understand the difference between saying that

the underlying form is the simple form or is the perfective form andsaying that we may best learn what the underlying form is by looking atthe perfective or simple form or some other form The underlying formof the word for lsquostandrsquo is tul We learn that this is the underlying formby comparing the simple form [tu] and the perfective [tulia] and under-standing that the perfective form preserves important information aboutthe underlying form that is lost in the simple form But the perfectiveform itself is [tulia] ndash this is not the underlying form

Palauan The language Palauan provides a second clear illustration ofthe point that one cannot always arrive at the correct underlying repre-sentation by looking at any single column in the paradigm In this lan-guage the underlying form of the word does not actually surface as suchin any form of a word Consider the following examples

The prefix for the present middle is apparently mə the future suffix(found in the future conservative and the future innovative) is -l and theinnovative suffix is -al The position of stress can be predicted by a simplerule the final syllable is stressed if it ends in two consonants otherwisethe second to last (penultimate) syllable is stressedThe fundamental problem of Palauan is how to predict vowel quality in

the root Notice that the root meaning of the word for lsquocoverrsquo has threesurface realizations daacuteŋəb daacuteŋəb and dəŋoacuteb Looking at all of the data we

Present middle Future innovative Future conservativemədaacuteŋəb dəŋəbaacutell dəŋoacutebl lsquocoverrsquoməteacuteʔəb təʔəbaacutell təʔiacutebl lsquopull outrsquoməŋeacutetəm ŋətəmaacutell ŋətoacuteml lsquolickrsquomətaacutebək təbəkaacutell təbaacutekl lsquopatchrsquoməʔaacuterəm ʔərəmaacutell ʔəroacuteml lsquotastersquoməseacutesəb səsəbaacutell səsoacutebl lsquoburnrsquo

notice that the only full vowel in the word is the stressed vowel whichsuggests that unstressed vowels are neutralized to schwa

(34) Unstressed vowel reduction

Note that this rule has no context it does not matter what precedes orfollows the unstressed vowelIn order to predict that the stressed first vowel in the word for lsquocoverrsquo

is [a] that choice of vowel must be part of the underlying representa-tion giving the partial solution daŋVb In contrast the first vowel ofthe word for lsquopull outrsquo must be specified as [e] since that is the vowelwhich appears in this word when the first vowel is stressed givingteʔVb By the same reasoning the second vowel of the word for lsquocoverrsquomust be [o] since that is the realization which the vowel has when it isstressed and the second vowel of the word for lsquopull outrsquo must be [i]Thus the underlying forms of the stems given above would be thefollowing

The underlying form of a verb in Palauan is a rather abstract objectsomething which is never revealed in any single surface form Ratherit must be deduced by looking at information which is manifested ina number of different morphologically related words derived from asingle stem

English A similar example can be found in English as the followingexamples show We will ignore other alternations and focus only on vowelalternations Thus for example alternations such as the one between kand s can be ignored There are many idiolectal differences in the pronun-ciation of certain words such as economy where some people pronouncethe word as [ijˈkanəmij] and others pronounce it as [əˈkanəmij] only attemptto account for the pronunciations given here

-high-lo+back-rd

+syl-stress

eacuteecirceuml

ugraveuacuteucirc

daŋob lsquocoverrsquo teʔib lsquopull outrsquoŋetom lsquolickrsquo tabak lsquopatchrsquoʔarom lsquotastersquo sesob lsquoburnrsquo

ˈmanətown lsquomonotonersquo məˈnatənij lsquomonotonyrsquoˈtɛləgraeligf lsquotelegraphrsquo təˈlɛgrəfij lsquotelegraphyrsquoˈɛpəgraeligf lsquoepigraphrsquo əˈpɪgrəfij lsquoepigraphyrsquoˈrɛlətɪv lsquorelativersquo rəˈlejʃən lsquorelationrsquoəˈkanəmij lsquoeconomyrsquo ˌɛkəˈnamɪk lsquoeconomicrsquoˈdijfεkt lsquodefect (noun)rsquo dəˈfɛktɪv lsquodefectiversquoˈdɛməkraeligt lsquodemocratrsquo dəˈmakrəsij lsquodemocracyrsquo

As in Palauan there is an alternation between stressed full vowel andunstressed schwa We assume underlying stems with multiple fullvowels eg manatown tεlεgraeligf εpɪgraeligf dεmakraeligt fownaladʒ etcBut not every unstressed vowel is reduced cf for example ˈrɛlətɪv ˈdijfεktˈmanətown where the unstressed vowel is in a closed syllable (followed byone or more consonants within that syllable)

Tonkawa reaching the analysis step-by-step The following exampleswill illustrate the logic that leads to seeing the correct underlying formsin explaining variations found in the verb root in Tonkawa (Texas) Youmust first give a morphological analysis of the data identifying themorphemes for progressive present first-singular object and third-pluralobject you must also set forth initial hypotheses about the underlyingforms of roots The data to be accounted for are as follows

Every word in this set ends with oʔ and the verb forms have a third-person subject suggesting that -oʔmarks third-person subject Comparingthe habitual present and present progressive form we see that the present

ˈɪtəlij lsquoItalyrsquo əˈtaeligljən lsquoItalianrsquoˈhamənɪm lsquohomonymrsquo həˈmanəmij lsquohomonymyrsquofəˈnɛtɪks lsquophoneticsrsquo ˌfownəˈtɪʃən lsquophoneticianrsquostəˈtɪstɪks lsquostatisticsrsquo ˌstaeligtəˈstɪʃən lsquostatisticianrsquorəˈsɪprəkl lsquoreciprocalrsquo ˌrɛsəˈprasətij lsquoreciprocityrsquofəˈnalədʒij lsquophonologyrsquo ˌfownəˈladʒəkl lsquophonologicalrsquoˈladʒɪk lsquologicrsquo ləˈdʒɪʃn lsquologicianrsquoˈsɪnənɪm lsquosynonymrsquo səˈnanəmij lsquosynonymyrsquoəˈrɪstəkraeligt lsquoaristocratrsquo ˌɛrəˈstakrəsij lsquoaristocracyrsquo

picnoʔ lsquohe cutsrsquo picnanoʔ lsquohe is cuttingrsquowepcenoʔ lsquohe cuts themrsquo wepcenanoʔ lsquohe is cutting themrsquo

kepcenoʔ lsquohe cuts mersquo kepcenanoʔ lsquohe is cutting mersquonotxoʔ lsquohe hoesrsquo notxonoʔ lsquohe is hoeingrsquowentoxoʔ lsquohe hoes themrsquo wentoxonoʔ lsquohe is hoeing themrsquo

kentoxoʔ lsquohe hoes mersquo kentoxonoʔ lsquohe is hoeing mersquonetloʔ lsquohe licksrsquo netlenoʔ lsquohe is lickingrsquowentaloʔ lsquohe licks themrsquo wentalenoʔ lsquohe is licking themrsquo

kentaloʔ lsquohe licks mersquo kentalenoʔ lsquohe is licking mersquonaxcoʔ lsquohe makes firersquo naxcenoʔ lsquohe is making firersquowenxacoʔ lsquohe makes them

firersquowenxacenoʔ lsquohe is making them firersquo

kenxacoʔ lsquohe makes me firersquo kenxacenoʔ lsquohe is making me firersquojamxoʔ lsquohe paints a facersquo jamxanoʔ lsquohe is painting a facersquowejmaxoʔ lsquohe paints their

facersquowejmaxanoʔ lsquohe is painting their facersquo

kejmaxoʔ lsquohe paints my facersquo kejmaxanoʔ lsquohe is painting my facersquonawloʔ lsquohe spreadsrsquo nawlenoʔ lsquohe is spreadingrsquowenweloʔ lsquohe spreads themrsquo wenwelenoʔ lsquohe is spreading themrsquo

kenweloʔ lsquohe spreads mersquo kenwelenoʔ lsquohe is spreading mersquo

progressive is marked by a suffix -n- or -Vn- before the suffix -oʔ An objectis marked by a prefix we- for third-plural object and ke- for first-singularobject What remains is the verb rootWe have two unresolved questions whether the suffix for the progres-

sive is -n- or is there a vowel in the suffix and what is the underlyingform of the verb root To resolve the first question we look at verbs withno object

We might think that the vowel before -n- is part of the progressive suffixbut if it were part of that suffix it should have a constant underlying formand all surface variants of that vowel should be derived by some simplerule Clearly the vowel before n ranges over a o and e and there is noreasonable way to predict which vowel is present Since that informationis governed by which root appears before the suffix the vowel is part ofthe underlying form of the verb root Thus we arrive at the followingpartial answer to the question about the underlying forms of the verbroots

The progressive form of the verb can be derived straightforwardly byadding the two affixes -n- and -oʔ The habitual present involves theapplication of a further phonological process Based on our hypothesesregarding the underlying forms of the verb roots we predict the followingunderlying forms for the habitual forms

The underlying form is whatever is given by the morphological compon-ent so in this case it would be the root plus progressive suffix followed by

picnoʔ picnanoʔnotxoʔ notxonoʔnetloʔ netlenoʔnaxcoʔ naxcenoʔjamxoʔ jamxanoʔnawloʔ nawlenoʔ

picna lsquocutrsquonotxo lsquohoersquonetle lsquolickrsquonaxce lsquomake a firersquojamxa lsquopaint a facersquonawle lsquospreadrsquo

Predicted form Actual surface formpicnaoʔ picnoʔ lsquocutrsquonotxooʔ notxoʔ lsquohoersquonetleoʔ netloʔ lsquolickrsquonaxceoʔ naxcoʔ lsquomake a firersquojamxaoʔ jamxoʔ lsquopaint a facersquonawleoʔ nawloʔ lsquospreadrsquo

the suffix -oʔ Our initial hypothesis is that the underlying form should beidentical to the surface form until we have evidence that phonologicalrules change the underlying forms in predictable ways The differencebetween the predicted form and the actual surface realization of the verbis that the underlying form has a cluster of vowels which is not found inthe surface form The data do not provide any examples of surface vowelclusters and this fact allows us to state a very simple rule accounting forthe surface form the first of two consecutive vowels is deleted

(41) Vowel cluster reduction

V Oslash _ V

Now we turn to the alternations in the shape of the stem that arisebetween the plain forms of the verb and the verb with an object prefixVerbs with the prefix ke- behave exactly like verbs with the prefix we-Disregarding the suffixes -n- and -oʔ we arrive at the following surfacevariations in the shape of the stem

In forms without a prefix there is a vowel between the first two conson-ants and none between the second and third consonants in forms with aCV prefix there is no vowel between the first two consonants but there isa vowel between the second and third consonants One alternative is thatthis vowel is epenthetic (inserted) the other is that the vowel is part of theunderlying vowel of the stem and is deleted in some phonological contextJust as there is no way to predict what vowel will appear between the firstand second consonants it is also impossible to predict what vowel willappear between the second and third consonants and therefore the vowelcannot be epenthetic In short the underlying representation must con-tain unpredictable vowels after each consonant

The underlying forms of prefixed and unprefixed forms would thus be asfollows (illustrating with the progressive form of the verb)

Stem without prefix Stem with CV prefixpicna pcena lsquocutrsquonotxo ntoxo lsquohoersquonetle ntale lsquolickrsquonaxce nxace lsquomake a firersquojamxa jmaxa lsquopaint a facersquonawle nwele lsquospreadrsquo

picena lsquocutrsquonotoxo lsquohoersquonetale lsquolickrsquonaxace lsquomake a firersquojamaxa lsquopaint a facersquonawele lsquospreadrsquo

Compare this with the surface form of the verbs

The relation between the underlying forms in (44) and surface forms in(45) is simple Each is subject to a rule deleting the second vowel of theword

(46) V Oslash CVC _

Whether the first or second stem vowel is deleted depends on whether aprefix is presentApart from illustrating the point that underlying forms of words may

not correspond to any single column in a wordrsquos paradigm this discussionof Tonkawa illustrates two important characteristics of a phonologicalanalysis The first is that one analyzes data by advancing an initial hypoth-esis and then refining the hypothesis when it becomes clear that theinitial hypothesis doesnrsquot work perfectly Thus we began with the hypoth-esis that the underlying forms were picna notxo netle and so on andwere able to account for a certain amount of data based on that hypoth-esis but later modified our hypothesis about underlying forms to bepicena notoxo netale and so on In other words although our firsthypothesis turned out to be wrong it was close to right and we were ableto identify the source of the problem in our hypothesis and correct itThe second characteristic of our analyis is that we always seek ways to

test the predictions of our hypotheses The hypothesis that the stems areunderlying picna notxo netle etc makes a prediction that if a vowelwere ever to appear between the second and third consonants (forexample due to a rule of vowel insertion) it would always be a singleconsistent and predictable vowel (since we are saying that it is not in theunderlying form) The fact that a different vowel appears in wepcenoʔwentoxoʔ wentaloʔ and wenxacoʔ shows that the prediction of this hypoth-esis is wrong and this forced us to consider the alternative hypothesisthat the underlying form contains a vowel between the second and third

Unprefixed Prefixedpicenanoʔ kepicenanoʔ lsquocutrsquonotoxonoʔ kenotoxonoʔ lsquohoersquonetalenoʔ kenetalenoʔ lsquolickrsquonaxacenoʔ kenaxacenoʔ lsquomake a firersquojamaxanoʔ kejamaxanoʔ lsquopaint a facersquonawelenoʔ kenawelenoʔ lsquospreadrsquo

Unprefixed Prefixedpicnanoʔ kepcenanoʔ lsquocutrsquonotxonoʔ kentoxonoʔ lsquohoersquonetlenoʔ kentalenoʔ lsquolickrsquonaxcenoʔ kenxacenoʔ lsquomake a firersquojamxanoʔ kejmaxanoʔ lsquopaint a facersquonawlenoʔ kenwelenoʔ lsquospreadrsquo

consonants this hypothesis proved to be correct The most basic form ofhypothesis testing that is done in phonology is combining presumedforms of roots and affixes and mechanically applying the rules whichwe assume in the analysis If the wrong form is produced by this testsomething is wrong with the hypothesis ndash either the underlying forms arewrong or the rules are stated incorrectly (or the rules are being applied inthe wrong order a point we get to in the next chapter)

Exercises1 Axininca CampaProvide underlying representations and a phonological rule which will accountfor the following alternations

2 XavanteWhat is the underlying form of the prefix meaning lsquoyourrsquo Formalize the rulewhich accounts for the change in that prefix

Summary Establishing the correct underlying representation for a morpheme isthe most important first step in giving a phonological analysis of dataA correct underlying representation unifies surface variants of a mor-pheme giving recognition of the basic ldquosamenessrdquo of a morphemeregardless of variations in pronunciation which arise because phono-logical rules have applied The underlying form and the system of rulesare thus connected by making the right choice of underlying formand given the right system of rules the rules will correctly operate onjust those segments which participate in the alternation The key tomaking the right decision about underlying forms is to carefully con-sider different hypotheses if a segment in a morpheme has two ormore surface realizations it is often necessary to consider two or morepossibilities for what is underlying ndash is variant [a] [b] or [c] the rightchoice The main issue relevant to answering this question is knowingwhich variant preserves important distinctions and which neutralizesdistinctions The underlying form may not even be seen directly in anyone pronunciation of a morpheme it may be a form inferred fromconsidering a number of specific instantiations of the morpheme

toniro lsquopalmrsquo notoniroti lsquomy palmrsquojaarato lsquoblack beersquo nojaaratoti lsquomy black beersquokanari lsquowild turkeyrsquo nojanariti lsquomy wild turkeyrsquokosiri lsquowhite monkeyrsquo nojosiriti lsquomy white monkeyrsquopisiro lsquosmall toucanrsquo nowisiroti lsquomy small toucanrsquoporita lsquosmall henrsquo noworitati lsquomy small henrsquo

3 Kuria IWhat are the underlying forms of the prefixes marking the infinitive and theobjects lsquoyoursquo lsquoit (cl 3)rsquo (which refers to one of the noun classes in thelanguage) Give a rule that explains the non-underlying pronunciation of theprefix Tone can be ignored

4 KoreanGive the underlying representations of each of the verb stems found belowstate what phonological rule applies to these data (There is a vowel harmonyrule explaining the variation between final a and ə in the imperative which youdo not need to be concerned with)

N your Ndu ʔaddu lsquostomachrsquoʔra ʔajʔra lsquochildrsquohiʔra ti ʔajhiʔrati lsquokneersquotɔ ʔattɔ lsquoeyersquoʔwa ʔajʔwa lsquotoothrsquobrɔ ʔajbrɔ lsquowifersquoʃɛrɛ ʔaʃʃɛrɛ lsquohairrsquopara ʔajpara lsquofootrsquoba ba ʔajbaba lsquofatherrsquo

lsquoto Vrsquo lsquoto V yoursquo lsquoto V it (cl 3)rsquookoreenda ogokoreenda okogoreenda lsquoguardrsquoogosooka okogosooka okogosooka lsquorespectrsquoogoseekera okogoseekera okogoseekera lsquoclose forrsquoogotaaŋgata okogotaaŋgata okogotaaŋgata lsquoleadrsquoogotatʃora okogotatʃora okogotatʃora lsquotearrsquookoheetoka ogokoheetoka okogoheetoka lsquorememberrsquoogokaraangera okogokaraangera okogokaraangera lsquofry forrsquookobeemera ogokobeemera okogobeemera lsquomeasure forrsquookogɔɔta ogokogɔɔta okogogɔɔta lsquoholdrsquookomaɲa ogokomaɲa okogomaɲa lsquoknowrsquo

Imperative Conjunctiveipə ipkrsquoo lsquowearrsquokupə kupkrsquoo lsquobendrsquokapha kapkrsquoo lsquopay backrsquotʃiphə tʃipkrsquoo lsquofeel the pulsersquotata tatkrsquoo lsquoclosersquoputhə putkrsquoo lsquoadherersquoməkə məkkrsquoo lsquoeatrsquotʃukə tʃukkrsquoo lsquodiersquoikə ikkrsquoo lsquoripenrsquotakrsquoa takkrsquoo lsquopolishrsquokrsquoakrsquoa krsquoakkrsquoo lsquoreduce expensesrsquosəkrsquoə səkkrsquoo lsquomixrsquo

5 ZoqueProvide the necessary phonological rules to describe the alternations found inthe following data Give the underlying representation for each noun stem andstate what the underlying form of the morpheme meaning lsquomyrsquo is

6 HungarianExplain what phonological process affects consonants in the followingdata (a vowel harmony rule makes suffix vowels back after back vowels andfront after front vowels which you do not need to account for) State what theunderlying forms are for all morphemes

7 Kuria IIProvide appropriate underlying representations and phonological rules whichwill account for the following data

N my N N my Nwaka nwaka lsquobasketrsquo huki nhuki lsquocigarettersquodisko ndisko lsquorecordrsquo jomo njomo lsquowomanrsquoburu mburu lsquoburrorsquo gaju ŋgaju lsquoroosterrsquopama mbama lsquoclothesrsquo plato mblato lsquoplatersquokaju ŋgaju lsquohorsersquo tʃoʔŋgoja ɲdʒoʔŋgoja lsquorabbitrsquotatah ndatah lsquofatherrsquo tsima ndzima lsquocalabashrsquofaha faha lsquobeltrsquo sək sək lsquobeansrsquoʃapun ʃapun lsquosoaprsquo lawus lawus lsquonailrsquo

Noun In N From N To Nkalap kalabban kalaptol kalapnak lsquohatrsquokut kudban kuttol kutnak lsquowellrsquoʒak ʒagban ʒaktol ʒaknak lsquosackrsquores rezben restoslashl resnek lsquopartrsquoʃrof ʃrovban ʃroftol ʃrofnak lsquoscrewrsquolakaʃ lakaʒban lakaʃtol lakaʃnak lsquoapartmentrsquoketrets ketredzben ketretstoslashl ketretsnek lsquocagersquotest tezdben testtoslashl testnek lsquobodyrsquorab rabban raptol rabnak lsquoprisonerrsquokad kadban kattol kadnak lsquotubrsquomeleg melegben melektoslashl melegnek lsquowarmrsquoviz vizben vistoslashl viznek lsquowaterrsquovaraʒ varaʒban varaʃtol varaʒnak lsquomagicrsquoagj agjban akjtol agjnak lsquobedrsquosem semben semtoslashl semnek lsquoeyersquobyn bynben byntoslashl bynnek lsquocrimersquotoroɲ toroɲban toroɲtol toroɲnak lsquotowerrsquofal falban faltol falnak lsquowallrsquooslashr oslashrben oslashrtoslashl oslashrnek lsquoguardrsquosaj sajban sajtol sajnak lsquomouthrsquo

Verb Verb forsuraaŋga suraaŋgera lsquopraisersquotaaŋgata taaŋgatera lsquoleadrsquobaamba baambera lsquofit a drum headrsquo

8 Isthmus ZapotecProvide an analysis of the following alternations formalizing your rules usingfeatures and listing the underlying forms of all morphemes

9 LezgianGive underlying forms of all morphemes and formalize the rules whichaccount for the following data

reenda reendera lsquoguardrsquorema remera lsquocultivatersquohoora hoorera lsquothreshrsquoroma romera lsquobitersquosooka sookera lsquorespectrsquotatʃora tatʃorera lsquotearrsquosiika seekera lsquoclosersquotiga tegera lsquoleave behindrsquoruga rogera lsquocookrsquosuka sokera lsquoplaitrsquohuuta hootera lsquoblowrsquoriiŋga reeŋgera lsquofoldrsquosiinda seendera lsquowinrsquo

N his Npan ʃpanbe lsquobreadrsquotaburete ʃtaburetebe lsquochairrsquotanguju ʃtangujube lsquoclay dollrsquokuba ʃkubabe lsquocorn doughrsquoluunaʔ hluunabe lsquobedrsquoneza hnezabe lsquoroadrsquomaniʔ hmanibe lsquohorsersquodiidʒaʔ ʃtiidʒabe lsquowordrsquodaa ʃtaabe lsquomatrsquodoo ʃtoobe lsquoropersquogeta ʃketabe lsquotortillarsquogiɲa ʃkiɲabe lsquotrunkrsquogamiʒaʔ ʃkamiʒabe lsquoshirtrsquobere ʃperebe lsquochickenrsquobiuuzaʔ ʃpiuuʒabe lsquoguestrsquobaʔduʔ ʃpaʔdube lsquochildrsquo

sg pl sg pltar tarar lsquotreersquo pel peler lsquoforeheadrsquoqrsquoel qrsquoeler lsquosaltrsquo qaw qawar lsquoroofrsquoʁyl ʁyler lsquohusbandrsquo luw luwar lsquowindrsquolif lifer lsquopigeonrsquo thur thurar lsquoswordrsquoqyl qyler lsquowheatrsquo buluth buluthar lsquocloudrsquokrsquowatʃh krsquowatʃhar lsquofootrsquo tarkwh tarkwhar lsquopestlersquomarkwh markwhar lsquostackrsquo raŋg raŋgar lsquocolorrsquofend fender lsquotrickrsquo krsquoalub krsquoalubar lsquoformrsquokrsquoarab krsquoarabar lsquobonersquo pheleŋg pheleŋgar lsquotigerrsquoqejd qejder lsquoremarkrsquo rab rapar lsquoneedlersquo

10 EstonianGive the underlying representations of the following nouns and formalize thephonological rule which applies in the following data

rib riper lsquoawlrsquo tʃeb tʃeper lsquoclayrsquomug mukar lsquonestrsquo tsegw tsekwer lsquoantrsquomeg meker lsquohairrsquo jad jatar lsquowaterrsquoqwed qweter lsquopartridgersquo kard kartar lsquofalconrsquonekh nekrsquoer lsquomilkrsquo wirth wirtrsquoer lsquohoneyrsquometh metrsquoer lsquokneersquo neth netrsquoer lsquolousersquowikh wikrsquoer lsquoyokersquo wakh wakrsquoar lsquopigrsquohaqh haqrsquoar lsquotruthrsquo reqh reqrsquoer lsquowayrsquo

nom sg gen plkubu kubude lsquobundlersquotiba tibade lsquodroprsquosadu sadude lsquofallingrsquoləbu ləbude lsquofunrsquoabi abide lsquohelprsquosoga sogade lsquomudrsquonibu nibude lsquopointrsquokivi kivide lsquostonersquolugu lugude lsquostoryrsquosaba sabade lsquotailrsquosula sulade lsquothawedrsquotəde tədede lsquotruthrsquokude kudede lsquowebrsquoviht vihtade lsquobathrsquovakk vakkade lsquobirch bark boxrsquoarg argade lsquocowardlyrsquovəlg vəlgade lsquodebtrsquososlashoslasht soslashoslashtade lsquofodderrsquomets metsade lsquoforestrsquolaisk laiskade lsquolazyrsquohoob hoobade lsquoleverrsquohaav haavade lsquowoundrsquolill lillede lsquoflowerrsquokərb kərbede lsquoforestrsquolaht lahtede lsquogulfrsquoneem neemede lsquoheadlandrsquoleep leepede lsquoslopingrsquoluik luikede lsquoswanrsquojaeligljg jaeligljgede lsquotracersquokaart kaartide lsquocardrsquokərjb kərjbide lsquocopperish redrsquopaar paaride lsquopairrsquokook kookide lsquopothookrsquokepp keppide lsquostickrsquopyks pykside lsquotrousersrsquo

11 TibetanNumbers between 11 and 19 are formed by placing the appropriate digit afterthe number 10 and multiples of 10 are formed by placing the appropriatemultiplier before the number 10 What are the underlying forms of the basicnumerals and what phonological rule is involved in accounting for these data

12 North SaamiPosit appropriate underlying forms and any rules needed to explain thefollowing alternations The emphasis here should be on correctly identifying theunderlying form the exact nature of the changes seen here is a more advancedproblem

lind lindude lsquobirdrsquokoon koonude lsquochinrsquokark karkude lsquocrutchrsquohull hullude lsquolunaticrsquoarv arvude lsquonumeralrsquoorb orbude lsquoorphanrsquoroog roogude lsquoreedrsquo

dʒu lsquo10rsquo dʒig lsquo1rsquo dʒugdʒig lsquo11rsquoʃi lsquo4rsquo dʒubʃi lsquo14rsquo ʃibdʒu lsquo40rsquogu lsquo9rsquo dʒurgu lsquo19rsquo gubdʒu lsquo90rsquoŋa lsquo5rsquo dʒuŋa lsquo15rsquo ŋabdʒu lsquo50rsquo

Nominative sg Essivevarit varihin lsquo2-year-old reindeer buckrsquooahpis oahpisin lsquoacquaintancersquotʃoarvvuʃ tʃoarvvuʃin lsquoantlers and skullcaprsquolottaeligʃ lottaeligdʒin lsquosmall birdrsquotʃuojvvat tʃuojvvagin lsquoyellow-brown reindeerrsquoahhkut ahhkubin lsquograndchild of womanrsquosuohkat suohkaethin lsquothickrsquoheedʒoʃ heedʒodʒin lsquopoor guyrsquoaeligddʒut aeligddʒubin lsquograndchild of manrsquobissobeahtset bissobeahtsehin lsquobutt of gunrsquotʃeahtsit tʃeahtsibin lsquochildren of elder brother of manrsquojaeligʔmin jaeligʔmimin lsquodeathrsquolaeliggeʃ laeliggedʒin lsquomountain birchrsquogahpir gahpirin lsquocaprsquogaeligwhtsis gaeligwhtsisin lsquo8 peoplersquoaeligslat aeligslagin [manrsquos name]baethooʃgaeligttset baethooʃgaeligttsebin lsquotailed titrsquoahhkit ahhkiethin lsquoboringrsquobahaelignaeliglat bahaelignaeliglagin lsquobadly behavedrsquobeʃʃtor beʃʃtorin lsquobird typersquoheevemeahhtun heevemeahhtunin lsquoinappropriatersquobejoot bejoohin lsquowhite reindeerrsquobissomeahtun bissomeahtumin lsquounstablersquo

Further readingInkelas 1989 Kaisse and Shaw 1985 Kenstowicz and Kisseberth 1977 Stanley 1967

laethaeligs laethaeligsin lsquosomething jointedrsquoheaijusmielat heaijusmielagin lsquounhappyrsquoheaŋkkan heaŋkkanin lsquohangerrsquojamaelign jamaelignin lsquosomething that

makes noisersquo

CHAPTER

5 Interactingprocesses

PREVIEW

In this chapter you will broaden your understanding of how

phonological systems work by

u looking at more complex patterns of phonological

alternation

u seeing how complex surface patterns of alternations

result from the interaction of different but related

phonological rules

u understanding the effect of different rule orderings

on how an underlying form is changed into a

surface form

KEY TERMSinteraction

ordering

Phonological systems are not made up of isolated and unrelated phono-logical rules there are usually significant interactions between phonologicalprocesses This chapter concentrates on two related topics First a seeminglycomplex set of alternations can be given a simple explanation if you separatethe effect of different rules which may happen to apply in the same formSecond applying rules in different orders can have a significant effect on theway that a given underlying form is mapped onto a surface form

51 Separating the effects of different rules

Very often when you analyze phonological alternations insights into thenature of these alternations are revealed once you realize that a word maybe subject to more than one rule each of which can affect the samesegment You should not think of a phonology as being just a collectionof direct statements of the relation between underlying segments andtheir surface realization Such a description is likely to be confusing andcomplex and will miss a number of important generalizations Look forways to decompose a problem into separate smaller and independentparts stated in terms of simple and general rules The different effectswhich these rules can have on a segment may accumulate to give aseemingly complex pattern of phonetic change

511 Votic palatalization and raisingfrontingThe following example from Votic (Russia) illustrates one way in whichthe account of phonological alternations can be made tractable by analyz-ing the alternations in terms of the interaction between independentphonological processes In these examples [ɫ] represents a velarized l

(1) a Nominative Partitivevərkko vərkkoa lsquonetrsquotʃako tʃakoa lsquocuckoorsquolintu lintua lsquobirdrsquosaatu saatua lsquogardenrsquojaɫka jaɫkaa lsquofootrsquobotʃka botʃkaa lsquobarrelrsquoeinaelig einaeligaelig lsquohayrsquovaeligvy vaeligvyaelig lsquoson-in-lawrsquo

b siili siiliaelig lsquohedgehogrsquoɫusti ɫustia lsquoprettyrsquo

c jarvi jarvəa lsquolakersquomaeligtʃi maeligtʃeaelig lsquohillrsquotʃivi tʃiveaelig lsquostonersquo

d kurtʃi kurkəa lsquostorkrsquoəɫtʃi əɫkəa lsquostrawrsquo

kahtʃi kahkəa lsquobirchrsquo

116 INTRODUCTION PHONOLOGY

The first group of examples (1a) shows that the nominative has no suffixand the partitive has the suffix [-a] or [-aelig] (the choice depends on thepreceding vowels determined by a vowel harmony rule according towhich a suffix vowel is front if the preceding vowel is front ndash therule skips over the vowel [i] but if there are no vowels other than [i]preceding the harmony rule turns the suffix vowel into a front vowel)The second group of examples (1b) illustrates roots which have i as theunderlying final vowel of the root The nouns in the third group (1c)illustrate a phenomenon of final vowel raising and fronting (whichwe have previously seen in closely related Finnish) whereby e and ǝbecome [i] word-finally

(2) Final FrontingRaising

The essential difference between the examples of (1b) and (1c) is that theforms in (1b) underlyingly end in the vowel i and those in (1c) end in eor ǝ In the last set of examples (1d) the noun root underlyingly ends inthe sequence kǝ which can be seen directly in kurkǝ-a However the finalCV of the root appears as [t ʃi] in the nominative kurt ʃi

It would be unrevealing to posit a rule changing word-final kǝ into[tʃi] in one step A problem with such a rule is that the change of a velar toa palatal conditioned by following word-final schwa is not a process foundin other languages and depends on a very specific conjunction of factsthat is not just schwa but word-final schwa You may not know at thispoint that such a rule is not found in other languages ndash part of learningabout phonology is learning what processes do exist in languages some-thing you will have a better basis for judging by the end of this bookWhat you can see right now is that such a rule treats it as a coincidencethat the underlying final schwa actually becomes [i] on the surface by anindependently necessary rule so that much of the supposed rule applyingto kǝ is not actually specific to kǝ

This alternation makes more sense once it is decomposed into the twoconstituent rules which govern it namely final raising (independentlymotivated by the data in (c)) Applying this rule alone to final kǝ wouldresult in the sequence [ki] However [ki] is not an allowed CV sequence inthis language and a process of palatalization takes place in accordancewith the following rule

(3) Palatalization

We can thus account for the change of underlying kurkǝ and ǝɫkǝ to[kurt ʃi] and [ǝɫt ʃi] by applying these two rules in a specific order where the

+syl-rd-lo

rarr __⎡⎢⎣

⎤⎥⎦

+high-back

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+cons+back

+syl-back

[ ]+cor

Interacting processes 117

rule of vowel raising applies before palatalization so that vowel raising isallowed to create occurrences of the vowel [i] and those derived cases of [i]condition the application of palatalization

You should take note of two points regarding how the palatalization ruleis formalized First by strictly making a velar consonant become [+cor]the result would be a velarized retroflex stop [ʈγ] such sounds simply donot exist in the language in fact the [-ant] coronal sounds of the languageare all alveopalatal and the alveopalatal stops in Votic are all affricatesObserved [t ʃ] is the closest segment of the language to [ʈγ]Second we do not have direct evidence that all front vowels trigger the

change of velars in fact we only have direct evidence that word-final [i]triggers the change At the same time we do not have any direct evidencethat it matters whether the triggering vowel is word-final or not nor dowe have any evidence that the other front vowels [y oslash e aelig] fail to triggerthe change Because there is no evidence for adding restrictions to therule we follow the general scientific principle of stating the rule as simplyas possible consistent with the data

512 Kamba palatalization and glide formationThere is a phonological process in Kamba (Kenya) whereby the combin-ation of a velar consonant plus the glide j fuses into an alveopalatalaffricate This can be seen in (5) which involves the plain and causativeforms of verbs In the examples on the left the verb is composed of theinfinitive prefix ko- (which undergoes a process of glide formation beforeanother vowel becoming [w]) followed by the verb root (eg -kam- lsquomilkrsquo)plus an inflectional suffix -a In the righthand column we can see thecausative of the same verb which is formed by suffixing -j- after the verbroot before the inflectional marker -a

(5) to V to cause to Va kokama kokamja lsquomilkrsquo

kokonagrave kokonjagrave lsquohitrsquokolaaagravembagrave kolaagravembjagrave lsquolaprsquokotala kotalja lsquocountrsquokwaambatagrave kwaambatjagrave lsquogo uprsquokwaagraveethagrave kwaagraveethjagrave lsquogovernrsquokweeta kweetja lsquoanswerrsquokwıımba kwıımbja lsquoswellrsquo

b koβikagrave koβitʃagrave lsquoarriversquokoβaacutelokagrave koβaacutelotʃagrave lsquofallrsquokolikagrave kolitʃagrave lsquoenterrsquokolεὲŋgagrave kolεὲɲdʒagrave lsquoaimrsquo

kurkǝ underlyingkurki vowel raisingkurtʃi palatalization

The examples in (a) illustrate the causative affix following various non-velar consonants of the language In (b) we see the causative of variousroots which end in k or g where by analogy to the data in (a) we predictthe causatives koβikjagrave koβaacutelokjagrave kolεὲŋgjagrave and so on Instead ofthe expected consonant sequences kj gj we find instead that the velarconsonant has been replaced by an alveopalatal affricate due to thefollowing rule

(6) Palatalization

Examples of glide formation are also seen in (5) where the vowel o in theinfinitive prefix becomes [w] before another vowel This process of glideformation is further illustrated in (7) and (8) In (7) you can see across allof the columns that the prefix for the infinitive is ko and appearsphonetically as such when it stands before another consonant The lastthree data columns show that the prefixes marking different classes ofobjects are moacute for class 3 meacute for class 4 and keacute for class 7 (Kambanouns have a dozen grammatical agreement classes analogous to genderin some European languages)

When the verb root begins with a vowel we would predict a sequence ofvowels such as koasja for lsquoto losersquo in lieu of a rule modifying vowelsequences Vowel sequences are avoided in Kamba by the application ofthe rule of glide formation according to which any nonlow vowelbecomes a glide before another vowel

to V to V it (cl 3) to V them (cl 4) to V it (cl 7)koethukagrave komoacuteethukagrave komeacuteethukagrave kokeacuteethukagrave lsquochurnrsquokokaagraveethagrave komoacutekaagraveethagrave komeacutekaagraveethagrave kokeacutekaagraveethagrave lsquopraisersquokoliigravendagrave komoacuteliigravendagrave komeacuteliigravendagrave kokeacuteliigravendagrave lsquocoverrsquokomεna komoacutemεna komeacutemεna kokeacutemεna lsquohatersquokoɲuβagrave komoacuteɲuβagrave komeacuteɲuβagrave kokeacuteɲuβagrave lsquochoosersquo

kwεεŋga kwεεɲdʒa lsquoclear a fieldrsquokwaanekagrave kwaanetʃagrave lsquodryrsquokwɔɔka kwɔɔtʃa lsquogather coalsrsquo

⎡⎢⎣

⎤⎥⎦

+cons+back ⎢

⎡⎢

⎢⎥⎦

⎤⎥

+syl-cons-back

[ ]+corrarr Oslash

Sometimes rulesaffect two segmentssimultaneously thisis an example ofhow such rules areformalized statingthat the firstsegment becomes[+cor] and thesecond is deleted

The stem-initialvowel in theseexamples becomeslong as a side effectof the precedingvowel becoming aglide this is knownas compensatorylengthening

to V to V it (cl 3) to V them (cl 4) to V it (cl 7)kwaasja komwaasja komjaasja kotʃaasja lsquolosersquokwaaka komwaaka komjaaka kotʃaaka lsquobuildrsquokwaagravesagrave komwaagravesagrave komjaagravesagrave kotʃaagravesagrave lsquocarversquokoomba komoomba komjoomba kotʃoomba lsquomoldrsquokookeljagrave komoacuteokeljagrave komjoacuteokeljagrave kotʃoacuteokeljagrave lsquoliftrsquokűűna koműűna komjűűna kotʃűűna lsquofetchrsquo

The Glide Formation rule can be formalized as (9)

(9) Glide Formation

While this rule does not explicitly state that the resulting glide is [+high]that value is predictable via structure preservation given the fact that thelanguage does not have glides that are [-high]

This rule would be expected to apply to underlying forms such askouna lsquoto fetchrsquo and ko-omba lsquoto moldrsquo since those forms have anunderlying sequence of a vowel o followed by another vowel Applyingthat rule would result in [kwűűna] and [kwoomba] but these are not thecorrect forms We can resolve this problem once we observe that the glide[w] never appears before the tense round vowels [u o] (but it can appearbefore the vowel [ɔ] as seen in [kwɔɔna] lsquoto seersquo from ko-ɔna)It does not help to restrict rule (9) so that it does not apply before o u

since the vowel e does actually undergo glide formation before thesevowels (ko-me-okelya becomes [komjoacuteokeljagrave] lsquoto lift themrsquo and ko-meacuteűnabecomes [komjűűna] lsquoto fetch themrsquo) What seems to be a restriction onglide formation is highly specific the tense round vowel fails to surface asa glide only if the following vowel is o or u Furthermore the round voweldoes not merely fail to become a glide it actually deletes therefore wecanrsquot just rewrite (9) so that it doesnrsquot apply before [u o] since that wouldgive [koűna] and [koomba]) Two rules are required to account for thesevowel-plus-vowel combinations A very simple solution to this problem isto allow the most general form of the Glide Formation rule to applyimposing no restrictions on which vowels trigger the rule and derivethe intermediate forms kwűűna and kwoomba Since we have observed thatthe surface sequences [wo] and [wu] are lacking in the language we mayposit the following rule of glide deletion which explains both why suchsequences are lacking and what happened to the expected glide in theintermediate forms

(10) Glide Deletion

Glide Formation first creates a glide and some of the glides so created arethen deleted by (10)

kuumbekagrave komuacuteumbekagrave komjuacuteumbekagrave kotʃuacuteumbekagrave lsquoburyrsquokwεεnza komwεεnza komjεεnza kotʃεεnza lsquoshaversquokwεὲndagrave komwεὲndagrave komjεὲndagrave kotʃεὲndagrave lsquolikersquokwɔɔna komwɔɔna komjɔɔna kotʃɔɔna lsquoseersquokwɔɔsa komwɔɔsa komjɔɔsa kotʃɔɔsa lsquotakersquokwɔɔβa komwɔɔβa komjɔɔβa kotʃɔɔβa lsquotiersquo

[ ]+syl [ ]-syl __rarr V

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

-syl+round

+round+tense

Oslash

Another crucial rule interaction which we observe in (8) is betweenGlide Formation and Palatalization As we have seen Palatalization spe-cifically applies to kj and gj and Glide Formation creates glides fromvowels which can trigger application of Palatalization This is shown inthe derivation of [kotʃaasja] from ko-keacute-aasja

Thus Glide Formation creates phonological structures which are cruciallyreferenced by other phonological rules

513 Bukusu nasal+consonant combinationsThe theme which we have been developing in this chapter is that phono-logical grammars are composed of simple rule elements that interact inwaysthat make the data patterns appear complicated and factoring out of thefundamental processes is an essential part of phonological analysis In theexamples which we have considered above such as vowel raisingfrontingand velar palatalization in Votic or glide formation and palatalization inKamba the phonological processes have been sufficiently different that noone would have problems seeing that these are different rules A languagemay have phonological changes which seem similar in nature or whichapply in similar environments and the question arises whether the alterna-tions in question reflect a single phonological rule Or do the alternationsreflect the operation of more than one independent rule with only acciden-tal partial similarity Such a situation arises in Bukusu (Kenya) where anumber of changes affect sequences of nasal plus consonant

Nasal Place Assimilation and Post-Nasal Voicing In the first set ofexamples in (12) a voicing rule makes all underlyingly voiceless consonantsvoiced when preceded by a nasal in this case after the prefix for the first-singular present-tense subject which is n The underlying consonant at thebeginning of the root is revealed directly when the root is preceded by thethird-plural prefix βa- or when there is no prefix as in the imperative

We can state this voicing rule as follows

ko-keacute-aasja underlyingkokjaasja Glide Formationkotʃaasja Palatalization

Imperative 3pl pres 1sg prestʃa βatʃa ɲdʒa lsquogorsquotʃexa βatʃexa ɲdʒexa lsquolaughrsquotʃutʃuuŋga βatʃutʃuuŋga ɲdʒutʃuuŋga lsquosieversquotalaanda βatalaanda ndalaanda lsquogo aroundrsquoteexa βateexa ndeexa lsquocookrsquotiira βatiira ndiira lsquoget ahold ofrsquopiima βapiima mbiima lsquoweighrsquopakala βapakala mbakala lsquowrithe in painrsquoketulula βaketulula ŋgetulula lsquopour outrsquokona βakona ŋgona lsquopass the nightrsquokula βakula ŋgula lsquobuyrsquokwa βakwa ŋgwa lsquofallrsquo

(13) Post-Nasal Voicing

You will also note that a nasal consonant always agrees in place ofarticulation with the following consonant due to the following rule

(14) Nasal Place Assimilation

The data considered so far have not given clear evidence as to what theunderlying place of articulation of the first-singular subject prefix is sincethat nasal always assimilates to the following consonant To determine thatthe prefix is indeed n we turn to the form of stems which underlyinglybegin with a vowel where there is no assimilation In the imperativewhere no prefix precedes the stem the glide [ j] is inserted before the initialvowel (The data in (17) include examples of underlying initial j which isgenerally retained showing that there cannot be a rule of j-deletion)Whenthe third-plural prefix βa precedes the stem the resulting vowel sequenceis simplified to a single nonhigh vowel No rules apply to the first-singularprefix which we can see surfaces as [n] before all vowels

One question that we ought to consider is the ordering of the rules ofvoicing and place assimilation In this case the ordering of the rules doesnot matter whether you apply voicing first and assimilation second orassimilation first and voicing second the result is the same

The reason why ordering does not matter is that the voicing rule doesnot refer to the place of articulation of the nasal and the assimilationrule does not refer to the voicing of the following consonant Thus infor-mation provided by one rule cannot change whether the other rule applies

Post-Nasal Hardening Another process of consonant hardening turnsvoiced continuants into stops after a nasal l and r become d β becomes band j becomes dʒ

[ ] -voice [ ]+nasal[ ]+voicerarr

rarr⎡⎢⎣

⎤⎥⎦

αantβcor ⎢

⎡⎢

⎢⎥⎦

⎤⎥

+sylαantβcor

[ ]+nas

Imperative 3pl pres 1sg presjiixala βeexala niixala lsquositrsquojaasama βaasama naasama lsquogapersquojoola βoola noola lsquoarriversquojeekesja βeekesja neekesja lsquoshowrsquo

n-kwa n-kwavoicing ngwa assimilation ŋkwaassimilation ŋgwa voicing ŋgwa

These data can be accounted for by the following rule

(18) Post-Nasal Hardening

This formalization exploits the concept of structure preservation toaccount for the changes to r l j By becoming [-cont] a change to [-son]is necessitated since there are no oral sonorant stops in Bukusu Likewisethe lack of lateral stops in the language means that l becomes [-lat] whenit becomes [-cont] Since there is no segment [ ɟ] in Bukusu making jbecome a stop entails a change in place of articulation from palatal toalveopalatal and from plain stop to affricateThe generalizations expressed in rules (13) and (18) can be unified into

one even simpler rule which states that consonants after nasals becomevoiced stops

(19) Post-Nasal Voicing-Hardening

l-deletion A third process affecting sequences of nasal plus consonantcan be seen in the following data

Imperative 3pl pres 1sg preslola βalola ndola lsquolookrsquolasa βalasa ndasa lsquoshoot atrsquoleβa βaleβa ndeβa lsquopushrsquolwaala βalwaala ndwaala lsquobe sickrsquora βara nda lsquoputrsquorara βarara ndara lsquobe stungrsquoroβa βaroβa ndoβa lsquoripenrsquorusja βarusja ndusja lsquovomitrsquorja βarja ndja lsquofearrsquoβakala βaβakala mbakala lsquospreadrsquoβala βaβala mbala lsquocountrsquoβasa βaβasa mbasa lsquoforgersquoβoola βaβoola mboola lsquotellrsquojama βajama ɲdʒama lsquoscoutrsquojaaja βajaaja ɲdʒaaja lsquoscramble withrsquojoola βajoola ɲdʒoola lsquoscooprsquojuula βajuula ɲdʒuula lsquosnatchrsquo

rarr cont +nasal[ ]+voice ][ ][-

+nasC+voice-cont ][rarr

⎡⎢⎣

⎤⎥⎦

Imperative 3pl pres 1sg presa tima βatima ndima lsquorunrsquo

taaɲa βataaɲa ndaaɲa lsquohackrsquotiiŋa βatiiŋa ndiiŋa lsquofilterrsquo

The examples in (a) show the effect of rules of voicing and consonanthardening applying as expected to t and r However the examples in(b) show the deletion of underlying l after a nasal These examples contrastwith the first set of examples in (17) where the root also begins withunderlying l the difference between the two sets of verbs is that in thesecond set where l deletes the following consonant is a nasal whereas inthe first set where l does not delete the next consonant is not a nasalThe significance of the examples in (20a) is that although underlying t

l and r all become [d] after a nasal the deletion of an underlyingconsonant in the environment N_VN only affects underlying l Sincethe voicing-hardening rule (19) neutralizes the distinction betweenthe three consonants after a nasal but in fact l acts differently from tand r in the context N_VN we can conclude that there is a prior ruledeleting l ndash but not t or r ndash in this context

(21) l-deletion[+lat] Oslash [+nasal] _ V0 [+nasal]

This rule clearly must apply before the hardening rule changes l into [d]after a nasal since otherwise there would be no way to restrict this rule toapplying only to underlying l When (19) applies underlying n-liindawould become n-diinda but n-riina would also become n-diina Once thathas happened there would be no way to predict the actual pronunciations[niinda] versus [ndiina]On the other hand if you were to apply the l-deletion rule first the rule

could apply in the case of n-liinda to give [niinda] but would not apply ton-riina because that form does not have an l thus by ordering the rules sothat l-deletion comes first the distinction between l which deletes andr which does not delete is preserved

Nasal Cluster Simplification Another phonological process applies to con-sonants after nasal consonantsWhen the root beginswith a nasal consonantthe expected sequence of nasal consonants simplifies to a single consonant

rema βarema ndema lsquochoprsquoriina βariina ndiina lsquorun awayrsquoruma βaruma nduma lsquosendrsquo

b laanda βalaanda naanda lsquogo aroundrsquolaaŋgwa βalaaŋgwa naaŋgwa lsquobe namedrsquoliinda βaliinda niinda lsquowaitrsquoloma βaloma noma lsquosayrsquoloondelela βaloondelela noondelela lsquofollowrsquo

luma βaluma numa lsquobitersquo

Imperative 3pl pres 1sg presmala βamala mala lsquofinishrsquomaɲa βamaɲa maɲa lsquoknowrsquo

meela βameela meela lsquoget drunkrsquoŋoola βaŋoola ŋoola lsquosee into the spirit worldrsquo

In the case of mala lsquoI finishrsquo the underlying form would be n-mala whichwould undergo the place assimilation rule (14) resulting in mmalaAccording to the data available to us there are no sequences of nasals inthe language so it is reasonable to posit the following rule

(23) Nasal Cluster Simplification[+nas] Oslash __ [+nas]

Nasal Deletion The final process which applies to sequences of nasalplus consonant is one deleting a nasal before a voiceless fricative

The underlying form of fuma lsquoI spreadrsquo is n-fuma since the prefix for 1sgis n- and the root is fuma and this contains a sequence nasal plusvoiceless fricative Our data indicate that this sequence does not appearanywhere in the language so we may presume that such sequences areeliminated by a rule of nasal deletion The formulation in (25) accounts forthe deletion facts of (24)

(25) Nasal Deletion

There can be an important connection between how rules are formulatedand how they are ordered In the analysis presented here we posited therules Nasal Deletion (25) and Post-Nasal Voicing-Hardening (19) repeatedhere where Nasal Deletion applies first

ɲa βaɲa ɲa lsquodefecatersquoɲaaɲa βaɲaaɲa ɲaaɲa lsquochewrsquo

ɲwa βaɲwa ɲwa lsquodrinkrsquo

Imperative 3pl pres 1sg presfuma βafuma fuma lsquospreadrsquofuundixa βafuundixa fuundixa lsquoknotrsquofwa βafwa fwa lsquodiersquoxala βaxala xala lsquocutrsquoxalaaŋga βaxalaaŋga xalaaŋga lsquofryrsquoxweesa βaxweesa xweesa lsquopullrsquoseesa βaseesa seesa lsquowinnowrsquo

siimbwa βasiimbwa siimbwa lsquohave indigestionrsquosomja βasomja somja lsquoteachrsquosukuwa βasukuwa sukuwa lsquorub legsrsquosja βasja sja lsquogrindrsquo

+nasal Oslash+cont-voice][ rarr __

⎡⎢⎣

⎤⎥⎦

+nasC+voice-cont ][rarr⎡⎢⎣

⎤⎥⎦

Since according to (25) only voiceless continuants trigger deletion of afollowing nasal we do not expect n-βala lsquoI countrsquo to lose its nasal How-ever there is the possibility that (19) could apply to n-fwa lsquoI diersquo since(19) does not put any conditions on the kind of consonant that becomes avoiced stop ndash but clearly f does not become a voiced stop in the surfaceform [fwa] This is because Nasal Deletion first eliminates the nasal in n-fwa before (19) has a chance to apply and once the nasal is deleted (19)can no longer applyYou might consider eliminating the specification [-voice] from the

formalization of (25) on the grounds that voiced continuants become stopsby (19) so perhaps by applying (19) first we could simplify (25) Such areordering would fail though since (19) would not only correctly changen-βala to [mbala] but would incorrectly change n-fwa to [mbwa] Theonly way to eliminate the specification [-voice] in (25) would be to split(19) into two rules specifically applying to voiced continuants and voice-less stops ndash a considerable complication that negates the advantage ofsimplifying (25) by one feature specification

Summary We have found in Bukusu that there are a number of phono-logical processes which affect N+C clusters by voicing hardening ordeleting the second consonant or deleting the nasal before a nasal or avoiceless fricative

(25) Nasal Deletion

(21) l-Deletion

(23) Nasal Cluster Simplification

Despite some similarity in these processes which involve a commonenvironment of nasal-plus-consonant there is no reasonable way to statethese processes as one rule

__rarr⎡⎢⎣

⎤⎥⎦

+ cont- voice

Oslash[ ]+nasal

rarr⎡⎢⎣

⎤⎥⎦

αantβcor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+sylαantβcor

[ ]+nas

v0[ ] + lat [ ]+nasal[ ]+nasalrarr Oslash

⎤⎥⎦

rarr Oslash[ ]+nas [ ]+nas

In addition to showing how a complex system of phonological alterna-tions decomposes into simpler independent and partially intersectingrules the preceding analyses reveal an important component of phono-logical analysis which is observing regularities in data such as the factthat Bukusu lacks any consonant sequences composed of a nasal plus africative on the surface

514 MatuumbiThe following data from Matuumbi illustrate the different surface real-izations of the noun-class prefixes (nouns are assigned lexically or syntac-tically to different classes conventionally numbered between 1 and 21)You should be able to discern and formalize the rule that applies inthese data and order those rules correctly What rule applies in thefollowing data

The examples in (27) illustrate three rules First there is an optional ruleapplying in both subsets of (27) which deletes u after m hence in thesewords the prefix mu can be pronounced in two ways one with u and onewithout u You should formalize the optional vowel deletion rule illus-trated by these dataAn independent rule assimilates a nasal to the place of articulation of

the following consonant (we saw this rule in previous Matuumbi data inchapters 2 and 4) This rule applies in both subsets of examples and is theonly other rule besides deletion of u involved in the first subset The thirdrule applies in the second subset of examples and explains the change inthe initial consonant of the stem This rule only applies to a glide precededby a nasal which is separated by a morpheme boundary notated inrules as ldquo+rdquo

Class C-initial noun V-initial noun4 mi-kaaacutete lsquoloavesrsquo mj-ooacutetoacute lsquofiresrsquo5 li-kuŋuuacutenda lsquofiltered beerrsquo lj-oowaacute lsquobeehiversquo7 ki-kaacutelaaŋgo lsquofrying panrsquo kj-uuacutelaacute lsquofrogrsquo8 i-kaacutelaaŋgo lsquofrying pansrsquo j-uuacutelaacute lsquofrogsrsquo14 u-toacutepe lsquomudrsquo w-ɪɪmbɪ lsquobeerrsquo11 lu-tooacutendwa lsquostarrsquo lw-aateacute lsquobanana handrsquo13 tu-toacuteopeacute lsquolittle handlesrsquo tw-aaacutena lsquolittle childrenrsquo15 ku-suuacutele lsquoto schoolrsquo kw-iisiacutewaacute lsquoto the islandsrsquo16 mu-kikuacute lsquoin the navelrsquo mw-iikuacute lsquoin the navelsrsquo

Unreduced form Reduced forma mu-tola n-tola lsquoyou (pl) takersquo

mu-kaacutelaaŋgite ŋ-kaacutelaaŋgite lsquoyou (pl) friedrsquomu-pɪme m-pɪme lsquoyou (pl) should measurersquo

b mu-wesa ŋ-ŋwesa lsquoyou (pl) canrsquomu-jɪkɪtiile ɲ-ɲɪkɪtiile lsquoyou (pl) agreedrsquomu-wuuacutengo ŋ-ŋwuuacutengo lsquoin the civetrsquomu-jɪɪga ɲ-ɲɪɪga lsquoin the bodyrsquo

When a rule isoptional that factis indicated bywriting ldquo(optional)rdquoto the right of therule ndash see chapter 2

(An alternative transcription of these second set of forms would be ŋŋwesaand so on the point of writing this as [ŋŋwesa] is to make clear that there isa change in the nature of the initial segment and not the addition ofanother segment)The examples in (28) illustrate the point that nouns in class 7 in the

singular (marked with the prefix ki-) have their plural in class 8 (with theprefix i-) The plural locative form gives further illustration of a phono-logical rule of the language which we already know

How do you explain the following examples of nouns which also havesingulars in class 7 and plurals in class 8 given that the class prefixes inthese examples are underlyingly ki- and i-

The data in (29) demonstrate a specific conclusion about the ordering oftwo of the rules motivated here what is that conclusion

52 Different effects of rule ordering

Besides showing how greater generality can often be achieved by splittinga process into smaller pieces the preceding examples have illustrated thatthe application of one rule can bring into existence new environmentswhere the second rule can apply an environment which did not exist inthe underlying formWhat we observed happening in these cases was thatboth of the rules applied Not all interactions between phonological pro-cesses have this characteristic ndash sometimes applying one rule prevents asecond rule from applying ndash and in this section we consider some of theeffects of different rule orderings

521 Lamba harmony and palatalizationThe following data illustrate the interaction between a rule of vowelharmony and a palatalization rule in the language Lamba (Zambia)

Singular (cl 7) Plural (cl 8) Plural locativeki-baacuteo i-baacuteo mwii-baacuteo lsquostoolrsquoki-bɪgaacute i-bɪgaacute mwii-bɪgaacute lsquopotrsquoki-biacuteliiacutetu i-biacuteliiacutetu mwii-biacuteliiacutetu lsquobox of matchesrsquoki-boacutegojoacute i-boacutegojoacute mwii-boacutegojoacute lsquotoothless personrsquo

Singular (cl 7) Plural (cl 8) Plural locativekjaacuteaiacute jaacuteaiacute mujaacuteaiacute ~ ɲɲaacuteaiacute lsquosoup potrsquokjaaacuteka jaaacuteka mujaaacuteka ~ ɲɲaaacuteka lsquobushrsquokjɪɪkɪ jɪɪkɪ mujɪɪkɪ ~ ɲɲɪɪkɪ lsquostumprsquokjuuacutebaacute juuacutebaacute mujuuacutebaacute ~ ɲɲuuacutebaacute lsquochestrsquo

Plain Passive Neuter Applied Reciprocaltʃita tʃitwa tʃitika tʃitila tʃitana lsquodorsquotula tulwa tulika tulila tulana lsquodigrsquo

In order to see what these data show we must first understand the mor-phological structure of thesewords a stepwhich leads us to realize that thepronunciation of certainmorphemes changes depending on their phoneticcontext Verbs in Lamba are composed of a root of the form CV(C)C anoptional derivational affix marking passive neuter applied or reciprocaland a word-final suffix -a which marks the form as being a verb Theunderlying forms of the passive and reciprocal suffixes are clearly -w- and-an- since they exhibit no phonetic variations The neuter and applied suf-fixes appear phonetically as -ik- and -ek- -il- and -el- The choice of vowel in thesuffix is determined by the vowel which precedes the suffix if the verb rootcontains the vowel i u or a the suffix has the vowel i and if the root containsthe vowel e or o the suffix has the vowel e The group of vowels i u a is nota natural phonetic class so it is implausible that the suffixes areunderlyingly-el- and -ek- with -il- and -ik- being derived by a rule The class of vowels e o isthe phonetic class of mid vowels it is thus evident that this language has avowel harmony rule which assimilates underlying high vowels (in thesuffixes il and ik) to mid vowels when they are preceded by mid vowels

(31) Height harmony

There is an alternation in the realization of certain root-final consonantsAs shown in examples such as kaka ~ kat ʃika and lasa ~ laʃika the velarconsonants and the alveolar continuant s become alveopalatals when theyare followed by the vowel i by processes of palatalization It is difficult toexpress a change of k and s to alveopalatal by one rule without includingt ndashwhich does not change see [patika] ndash so two separate rules are needed

(32) a Stop Palatalization

tʃeta tʃetwa tʃeteka tʃetela tʃetana lsquospyrsquosoŋka soŋkwa soŋkeka soŋkela soŋkana lsquopay taxrsquopata patwa patika patila patana lsquoscoldrsquofisa fiswa fiʃika fiʃila fisana lsquohidersquotʃesa tʃeswa tʃeseka tʃesela tʃesana lsquocutrsquokosa koswa koseka kosela kosana lsquobe strongrsquolasa laswa laʃika laʃila lasana lsquowoundrsquomasa maswa maʃika maʃila masana lsquoplasterrsquoʃika ʃikwa ʃitʃika ʃitʃila ʃikana lsquoburyrsquoseka sekwa sekeka sekela sekana lsquolaugh atrsquopoka pokwa pokeka pokela pokana lsquoreceiversquokaka kakwa katʃika katʃila kakana lsquotiersquofuka fukwa futʃika futʃila fukana lsquocreeprsquo

rarr ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎥[ ]+syl [ ]-high C0

+syl-high-low

rarr⎡⎢⎣

⎤⎥⎦

+high- syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]+cor

b Fricative Palatalization

The interaction between these processes is seen in words which could inprinciple undergo both of these processes roots with the vowel e or o andthe final consonant k or s The example sekeka lsquolaugh atrsquo from sek-ik-ashows how these processes interact Suppose first that palatalizationwere to apply before vowel harmony Since the underlying representationhas the sequence ki which is required by palatalization that rule wouldapply Subsequently vowel harmony would assimilate i to [e] after egiving the wrong surface result This is illustrated below in a derivationwhich spells out the results of applying first palatalization then heightharmony

Thus applying the rules in this order gives the wrong results this ordercannot be correctOn the other hand if we apply the processes in the other order with

height harmony applying before palatalization then the correct form isgenerated

522 Voicing and epenthesisLithuanian Another example which illustrates how an earlier rule canchange a form in such a way that a later rule can no longer apply is foundin Lithuanian There is a process of voicing assimilation in Lithuanianwhereby obstruents agree in voicing with an immediately followingobstruent This rule applies in the following examples to the verbalprefixes at and ap

(35) a

rarr⎡⎢⎣

⎤⎥⎦

+cont+cor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]- ant

sek-ik-a underlyingsekeka height harmony(not applicable) palatalization

at at-eiti lsquoto arriversquoat-imti lsquoto take awayrsquoat-neʃti lsquoto bringrsquoat-leisti lsquoto forgiversquoat-likti lsquoto completersquoat-kopti lsquoto risersquoat-praʃiti lsquoto askrsquoat-kurti lsquoto reestablishrsquo

ap ap-eiti lsquoto circumventrsquo

sek-ik-a underlyingsetʃika palatalizationsetʃeka height harmony

We would assume that the underlying forms of the prefixes are at andap and that there is a rule which voices obstruents before voicedobstruents

(36) Voicing assimilation

The alternative hypothesis would be that the prefixes are underlyingly adand ab However there is no natural context for describing the process ofdevoicing Although devoicing of voiced obstruents before voicelessobstruents is quite natural assuming that the prefixes have underlyingvoiced obstruents would also require the consonant to be devoiced beforevowels and sonorant consonants in order to account for the supposedderivations ad-eiti [ateiti] ab-eiti [apeiti] ad-neʃti [atneʃti] andab-mokiti [apmokiti] But there is clearly no rule prohibiting voicedobstruents before vowels and sonorants in this language (in fact nolanguage has ever been attested with a rule of consonant devoicing wherethe conditioning environment is a following vowel) On the basis of thisreasoning we conclude that the prefixes have underlying voicelessconsonantsWhen the initial consonant of the root is an alveolar stop the vowel [i]

appears after the prefix at and similarly when the initial consonant is abilabial stop [i] is inserted after the consonant of ap

[ ] -son [ ]+voicerarr-son+voice

⎡⎢⎣

⎤⎥⎦

ap-ieʃkoti lsquoto search everywherersquoap-akti lsquoto become blindrsquoap-mokiti lsquoto trainrsquoap-temditi lsquoto obscurersquoap-ʃaukti lsquoto proclaimrsquo

at ad-bekti lsquoto run uprsquoad-gauti lsquoto get backrsquoad-bukti lsquoto become bluntrsquoad-gimti lsquoto be born againrsquo

ap ab-gauti lsquoto deceiversquoab-ʒjureti lsquoto have a look atrsquoab-ʒelti lsquoto become overgrownrsquoab-dauʒiti lsquoto damagersquoab-draskiti lsquoto tearrsquo

ati-duoti lsquoto give backrsquoati-dariti lsquoto openrsquoati-deti lsquoto delayrsquoati-teisti lsquoto adjudicatersquoapi-berti lsquoto strew all overrsquoapi-begti lsquoto run aroundrsquoapi-puti lsquoto grow rottenrsquo

Given just the voicing assimilation rule you would expect forms such as[adduoti] [abberti] by analogy to [adbekti] and [abdauʒiti] Lithuaniandoes not allow sequences of identical consonants so to prevent such aresult an epenthetic vowel is inserted between homorganic obstruentstops (ones having the same values for the place of articulation features)

(38) Epenthesis

The ordering of these rules is important epenthesis (38) must apply beforevoicing assimilation since otherwise the prefix consonant would assimi-late the voicing of the root-initial consonant and would then be separatedfrom that consonant by the epenthetic vowel The result of applying thevoicing assimilation rule first would be to create [adduoti] [abberti] andthen this would undergo vowel epenthesis to give incorrect [adiduoti][abiberti] If on the other hand epenthesis is the first rule applied thenunderlying at-duoti becomes [atiduoti] and ap-berti becomes [apiberti]Epenthesis eliminates the underlying cluster of obstruents preventingthe voicing rule from applying

Armenian Interestingly a similar pair of rules exists in the New Julfa(Iran) dialect of Armenian but they apply in the opposite order If rulesapply in a particular order you would expect to find languages withessentially the same rules A and B where A precedes B in one languageand B precedes A in another this is what we find in comparing Armenianand LithuanianThe first-singular future prefix is underlyingly k- as shown in (39a)

where the prefix is added to a vowel-initial stem That k assimilatesvoicing and aspiration from an obstruent which immediately follows itunderlyingly (but not across a vowel) In addition initial consonant clus-ters are broken up by an epenthetic schwa As the data in (39b) show theprefix consonant first assimilates to the initial consonant of the root andthen is separated from that consonant by schwa

(39) a

rarrOslash⎡⎢⎣

⎤⎥⎦

-cont-sonαantβcor

⎡⎢⎣

⎤⎥⎦

-cont-sonαantβcor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

k-ertham lsquoI will gorsquok-asiem lsquoI will sayrsquok-aniem lsquoI will dorsquok-akaniem lsquoI will watchrsquok-oxniem lsquoI will blessrsquok-uriem lsquoI will swellrsquo

kə-tam lsquoI will giversquokə-kienam lsquoI will existrsquogə-bəzzam lsquoI will buzzrsquogə-lam lsquoI will cryrsquo

The difference between this dialect of Armenian and Lithuanian is thatvowel epenthesis applies before consonant assimilation in Lithuanianbut after that rule in Armenian so that in Armenian both epenthesisand assimilation can apply to a given word whereas in Lithuanianapplying epenthesis to a word means that assimilation can nolonger apply

523 Mongo B-deletion and resolution of vowel hiatusSometimes what needs to be remarked about the interaction betweenprocesses is the failure of one rule to apply to the output of another ruleThis is illustrated in (40) (41) and (46) with examples from Mongo(Congo) The first four examples demonstrate the shape of the varioussubject prefixes when they stand before a consonant

The underlying forms of the subject prefixes are N (which stands for anasal consonant whose exact place of articulation cannot be determined)o a to lo and ba There is a vowel harmony process assimilating theclosed vowel o to the open vowel [ɔ] when the following syllable containseither of the open vowels [ε] or [ɔ] and the prefix for first-singular subjectassimilates in place of articulation to the following consonantThe examples in (41) show how the subject prefixes are realized if the

verb root begins with a vowel

When the first-singular subject prefix stands before the root it has theshape [ndʒ] which we will treat as being the result of insertion of [dʒ]between the prefix and a vowel-initial root (We might also assume the

gə-zəram lsquoI will brayrsquokhə-thuojniem lsquoI will allowrsquo

khə-tʃhaphiem lsquoI will measurersquoghə-bhieɹiem lsquoI will carryrsquoghə-ghuom lsquoI will comersquoghə-dzhieviem lsquoI will formrsquo

Imp 1sg 2sg 3sg 1pl 2pl 3plsaŋga nsaŋga osaŋga asaŋga tosaŋga losaŋga basaŋga lsquosayrsquokamba ŋkamba okamba akamba tokamba lokamba bakamba lsquoworkrsquokɔta ŋkɔta ɔkɔta akɔta tɔkɔta lɔkɔta bakɔta lsquocutrsquotɛŋga ntɛŋga ɔtɛŋga atɛŋga tɔtɛŋga lɔtɛŋga batɛŋga lsquostraightenrsquomεla mmεla ɔmεla amεla tɔmεla lɔmεla bamεla lsquodrinkrsquodʒila ndʒila odʒila adʒila todʒila lodʒila badʒila lsquowaitrsquo

Imp 1sg 2sg 3sg 1pl 2pl 3plεna ndʒεna wεna εna tswεna dʒwεna bεna lsquoseersquoisa ndʒisa wisa isa tswisa dʒwisa bisa lsquohidersquoimedʒa ndʒimedʒa wimedʒa imedʒa tswimedʒa dʒwimedʒa bimedʒa lsquoconsentrsquousa ndʒusa wusa usa tswusa dʒwusa busa lsquothrowrsquo

ina ndʒina wina ina tswina dʒwina bina lsquohatersquo

prefix ndʒ which simplifies before a consonant since such three-consonant sequences viz ndʒ-saŋga do not exist in the language)

(42) Consonant epenthesis

The vowel a deletes before another vowel as shown by the third-singularand third-plural forms a-εna [εna] and ba-εna [bεna]

(43) Vowel truncation[+low] Oslash _ V

The prefixes o to and lo undergo a process of glide formation where obecomes [w] before a vowel

(44) Glide formation[+round] [ndashsyl] _ V

In the case of to and lo a further process affricates these consonantsbefore a glide

(45) Affrication

This affrication process must apply after glide formation since it appliesto a sequence of consonant plus glide that is created by the application ofglide formation from an underlying consonant-plus-vowel sequenceThe final set of examples illustrates verb roots which underlyingly

begin with the consonant b As these data show when underlying b ispreceded by a vowel it is deleted

Thus surface [oina] derives from obina and [baina] derives from babinavia the following rule

(47) Labial elision

In this case even though deletion of b creates new sequences of o+Vand a+V which could in principle undergo the rules of a-deletion and

rarr ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

-syl-cons-back

[ ]+nas VOslash +

Turning l into anaffricate seemsstrange from afunctionalperspective but isexplained by thefact that l wasoriginally d sothis rule comeshistorically from themore naturalchange t d[ts dz] _ i

+cor-syl+high][ ]+delrel[rarr⎡⎢⎣

⎤⎥⎦

Imp 1sg 2sg 3sg 1pl 2pl 3plbina mbina oina aina toina loina baina lsquodancersquobota mbota oota aota toota loota baota lsquobegetrsquo

rarr Oslash V__V⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+voice+ant-cor

glide formation those rules do not in fact apply In other words in thiscase the grammar must contain some kind of explicit statementregarding the interaction of these processes such as an explicitordering of the rules which guarantees that the output of b-deletiondoes not undergo glide formation or a-deletion By ordering theb-deletion rule so that it applies after the glide formation and voweltruncation rules we explain why those two rules fail to apply justin case the consonant b is deleted intervocalically The ordering whereb-deletion precedes vowel truncation and glide formation illustrated in(48b) results in ungrammatical forms which shows that that orderingof the rules is incorrect (ldquoNArdquo means that the rule cannot applybecause the conditions called for in the rule are not satisfied inthe string)

(48) a

Mongo thus provides an example of the failure of rules ndash especiallyvowel truncation and glide formation ndash to apply to the output of a specificrule ndash b-deletion ndash which we explain by ordering b-deletion after the vowelrules

524 Examples for discussionKarok These data from Karok (California) illustrate three interactingphonological processes Comment on the underlying forms of thefollowing words state what phonological rules are motivated and discussthe order in which these processes apply

o-bina a-bina underlyingNA NA glide formationNA NA vowel truncationoina aina b-deletion

o-bina a-bina underlyingoina aina b-deletionwina NA glide formationNA ina vowel truncation[wina] [ina]

Imperative 1sg 3sgpasip nipasip ʔupasip lsquoshootrsquositva niʃitva ʔusitva lsquostealrsquokifnuk nikifnuk ʔukifnuk lsquostooprsquosuprih niʃuprih ʔusuprih lsquomeasurersquoʔifik niʔifik ʔuʔifik lsquopick uprsquoʔiftih niʔiftih ʔuʔiftih lsquogrowingrsquoʔaktuv niʔaktuv ʔuʔaktuv lsquopluck atrsquoʔakrap niʔakrap ʔuʔakrap lsquoslaprsquoʔarip niʔarip ʔuʔarip lsquocut a striprsquoʔaxjar nixjar ʔuxjar lsquofillrsquoʔiʃkak niʃkak ʔuskak lsquojumprsquo

Shona Often a seemingly complex problem can be significantly simpli-fied by breaking the problem up into a few interacting processes If youlook at the phonetic realizations of the passive suffix in Shona (Zim-babwe) you see that there are seven different manifestations of thissuffix However this considerable range of variation can be explained interms of a much smaller set of very general phonological rules whoseinteraction results in many surface realizations of the suffix

The precise rules which you postulate will depend on what you assume tobe the underlying form of the passive suffix since there are two plausibleunderlying forms for the suffix based on the data above The phonologicalalternations seen in the following examples are relevant to deciding whatthe underlying form of the passive suffix is (and therefore exactly how

Active Passive Active Passiveɓika ɓikwa lsquocookrsquo diba dibγa lsquodiprsquopfugama pfugamŋa lsquokneelrsquo pepa pepxa lsquonursersquoɓuɗa ɓuɗγwa lsquogo outrsquo ruma rumŋa lsquobitersquorova rovγa lsquostay awayrsquo maɲa maɲŋwa lsquorunrsquosuŋga suŋgwa lsquotiersquo kwaʃa kwaʃxwa lsquohuntrsquoteŋga teŋgwa lsquobuyrsquo fuŋga fuŋgwa lsquothinkrsquotamba tambγa lsquoplayrsquo bvunza bvunzγwa lsquoaskrsquoimba imbγa lsquosingrsquo gara garγwa lsquostayrsquosetsa setsxwa lsquoamusersquo reɖʐa reɖʐγwa lsquolengthenrsquotapa tapxa lsquocapturersquo βeza βezγwa lsquocarversquorega regwa lsquoleaversquo ibva ibvγa lsquoripenrsquoʃuʒa ʃuʒγwa lsquostore

grainrsquotaʂa taʂxwa lsquoridersquo

peta petxwa lsquofoldrsquo dana danŋwa lsquocallrsquoona onŋwa lsquoseersquo ita itxwa lsquodorsquodoka dokwa lsquosetrsquo seka sekwa lsquolaughrsquofesa fesxwa lsquoprickrsquo rasa rasxwa lsquothrow

awayrsquorana ranŋwa lsquokickrsquo pema pemŋa lsquobeg foodrsquogotʃa gotʃxwa lsquoroastrsquo ʂika ʂikwa lsquoarriversquodzidza dzidzγwa lsquolearnrsquo fuka fukwa lsquocoverrsquofamba fambγa lsquowalkrsquo nandza nandzγwa lsquolickrsquogada gadγwa lsquomountrsquo ɓata ɓatxwa lsquoholdrsquotuma tumŋa lsquosendrsquo tora torγwa lsquotakersquooŋa oŋwa lsquogrowlrsquo rima rimŋa lsquoplowrsquo

sefa sefxa lsquosieversquo kweʐa kweʐγwa lsquoattractrsquodʒudʒa dʒudʒγwa lsquoleakrsquo guruva guruvγa lsquodeceiversquomaŋga maŋgwa lsquoarrestrsquo miɲa miɲŋwa lsquoswallowrsquo

ʔimniʃ nimniʃ ʔumniʃ lsquocookrsquoʔikʃah nikʃah ʔuksah lsquolaughrsquoʔiʃriv niʃriv ʔusriv lsquoshoot at a targetrsquo

these phonological alternations are to be analyzed) These inflected formsinvolve a prefix marking the subject followed by one of various tensemarkers such as -t ʃa- -no- and -a- or no marker finally followed by theverb stem

A further fact which is relevant to deciding on the correct analysis is that[γ] [x] do not appear after vowels or at the beginning of a word

Klamath The data in (52)ndash(56) from Klamath (Oregon) illustrate twoprocesses The first deaspirates and deglottalizes consonants beforeobstruents before glottalized and voiceless resonants as well as inword-final positions The examples in (52) illustrate plain voiceless obstru-ents which do not undergo any phonetic alternations The data belowinvolve a range of inflectionally and derivationally related word formsthe common root is underlined (the last form in this set also illustrates analternation between i and jrsquo which is not crucial)

The data in (53) provide examples of underlyingly glottalized obstruentswhich become plain voiceless consonants unless they are followed by avowel or plain sonorant

Subjunctive Futureurime lsquothat you (sg) plowrsquo utʃarima lsquoyou (sg) will plowrsquo

murime lsquothat you (pl) plowrsquo mutʃarima lsquoyou (pl) will plowrsquo

turime lsquothat they (tiny)plowrsquo

tutʃarima lsquothey (tiny) willplowrsquo

kunatse lsquothat therebe nicersquo

kutʃanatsa lsquothere will be nicersquo

Habitual Recent pastunorima lsquoyou (sg) plowrsquo warima lsquoyou (sg) plowedrsquomunorima lsquoyou (pl) plowrsquo mŋarima lsquoyou (pl) plowedrsquotunorima lsquothey (tiny) plowrsquo txwarima lsquothey (tiny) plowedrsquokunonatsa lsquothere is nicersquo kwanatsa lsquothere was nicersquo

lap-a lsquotwo (obj)rsquo lap lsquotworsquoskhot-a lsquoputs on a

blanketrsquoskhot-pli lsquoputs on a blanket

againrsquoqrsquolatʃ-aksi lsquoBlueberry Placersquo qrsquolatʃ lsquoblueberry (sp)rsquopoq-a lsquobakes camasrsquo poq-s lsquocamas rootrsquolaqi lsquois richrsquo laqjrsquo-aka lsquolittle chiefrsquo

prsquoakrsquo-a lsquosmashesrsquo prsquoak-ska lsquochips off (intr)rsquoʔetrsquo-a lsquodistributesrsquo se-ʔet-s lsquoSaturdayrsquopoq-poqrsquo-a lsquobecomes dustyrsquo poq-tki lsquobecomes dustyrsquotʃhakrsquo-a lsquomelts (intr)rsquo tʃhak-tki lsquomelts (as butter)rsquoʔi-tʃhitʃrsquo-a lsquomakes shavingsrsquo k-tʃhitʃ-ta lsquoscrapes onersquos foot onrsquotʃhloqrsquo-a lsquois smoothrsquo tʃhloq-tki lsquobecomes slickrsquo

Data in (53) show that aspirated consonants deaspirate in this samecontext

The second process syncope deletes a short vowel from the first syllableof a stem when preceded by a CV prefix and followed by CV

What do these examples show about the interaction of these twoprocesses

litʃh-litʃ-lrsquoi lsquostrongrsquo litʃ-tki lsquobecomesstrongrsquo

ponw-oth-a lsquowhile drinkingrsquo ponw-ot-s lsquosomething todrink withrsquo

sotʃh-a lsquokindles a firersquo sotʃ-tila lsquolights a fireunderrsquo

sijoth-a lsquotrades (pl obj) witheach otherrsquo

sijot-pli lsquotrade back(pl obj)rsquo

nrsquoiqh-owa lsquokeeps putting ahand in waterrsquo

nrsquoiq-tpa lsquoreaches andtouchesrsquo

qitrsquo-lqa lsquopours downrsquo qit-qha lsquopours outrsquo-loprsquo-a lsquoeats souprsquo -lop-s lsquosouprsquo

laqita lsquosuspects sorsquo sa-lqita lsquosuspects eorsquomatʃhat-ka lsquolistensrsquo sna-mtʃhat-ila lsquocauses to hearrsquometʃa lsquomoves camprsquo me-mtʃrsquoa lsquomoves

(distributive)rsquosaqotka lsquoask for strsquo sa-sqotqa lsquoask for st

(distributive)rsquositʃaqhwa lsquowash handsrsquo hi-stʃaq-tha lsquoare angry with eorsquosom lsquomouthrsquo so-smrsquo-ak lsquolittle mouths

(distributive)rsquo

qrsquootʃrsquoa lsquobendsrsquo jo-qtʃrsquoa lsquobends with the feetrsquoqhewrsquoa lsquobreaksrsquo tʃhe-qwrsquoa lsquosit on and breakrsquothewrsquoa lsquosurface cracksrsquo je-twrsquoa lsquosteps on and cracks

surfacersquos-tʃrsquoiqrsquoa lsquosquash with a

pointedinstrumentrsquo

ji-tʃqrsquoa lsquosquash by pressurewith the feetrsquo

w-krsquoalrsquoa lsquocuts with a longinstrumentrsquo

kin-klrsquoa lsquomakes a mark withpointerrsquo

w-prsquoeqrsquoa lsquohits in the facewith a longinstrumentrsquo

hom-pqrsquoa lsquoflies in the facersquo

Exercises1 KereweWhat two tone rules are motivated by the following data Explain what order therules apply in Vowels have no accent with L tone treat H tones as [+H] andL tones as [minusH]

2 MbungaAccount for the phonological alternations in the following data Note that thereare two roots for lsquobeatrsquo lsquocutrsquo lsquorubrsquo also there are derivational relations indicatedwith suffixes (-el- -il- -is- -es- etc) which you need not account for except forthe difference between final -a and final -i which mark different tenses you donot need to be concerned with possible suffixes and alternations caused bysuffixes

Summary Systems of phonological alternations in most languages involve anumber of rules This interaction means that you must discern theeffects of individual rules rather than subsume all alternations underone complex do-everything rule A rule changes a given set of segmentsin a uniform manner in a specified environment So even when alanguage like Bukusu has a number of rules pertaining to sequences ofnasal plus consonant ndash rules which have in common a single contextNC ndash there may be quite a number of specific rules that apply in thatcontext Besides identifying what rules exist in a language you mustalso determine what the proper ordering of those rules is The correctorder of a pair of rules can be determined by applying the rules veryliteral-mindedly in both of the logically possible orders

to V to V eo to V for to V for eokubala kubalana kubalila kubalilana lsquocountrsquokugaja kugajana kugajila kugajilana lsquodespisersquokugula kugulana kugulila kugulilana lsquobuyrsquokubala kubalana kubalıla kubalılana lsquokickrsquokuluma kulumana kulumıla kulumılana lsquobitersquokusuna kusunana kusunıla kusunılana lsquopinchrsquokulaba kulabana kulabıla kulabılana lsquopassrsquo

to V us to V it to V for us to V it for uskutubala kukıbala kutubalila kukıtubalila lsquocountrsquokutugaja kukıgaja kutugajila kukıtugajila lsquodespisersquokutugula kukıgula kutugulila kukıtugulila lsquobuyrsquokutubala kukıbala kutubalila kukıtubalila lsquokickrsquokutuluma kukıluma kutulumila kukıtulumila lsquobitersquokutusuna kukısuna kutusunila kukıtusunila lsquopinchrsquokutulaba kukılaba kutulabila kukıtulabila lsquopassrsquo

3 PolishWhat phonological rules are motivated by the following examples and whatorder do those rules apply in

lsquohe V-ed mersquo lsquohe V-ed usrsquokamvutila katufutila lsquorub forrsquokamvuwila katufuwila lsquowash forrsquokanzeka katuseka lsquolaugh atrsquokanzukumula katusukumula lsquopushrsquokadʒimisila katudʒimisila lsquoextinguish forrsquokabota katubota lsquobeatrsquokababanisa katubabanisa lsquosqueezersquokadaŋgila katudaŋgila lsquothrow forrsquokadetela katudetela lsquosay torsquokaguvila katuguvila lsquofall onrsquokandova katutova lsquobeatrsquokaɲdʒubula katutʃubula lsquoscratchrsquokaŋgamula katukamula lsquograbrsquokambutukila katuputukila lsquorub forrsquokanduvila katutuvila lsquorun forrsquokaŋgetulila katuketulila lsquocut forrsquokandelekela katutelekela lsquocook forrsquo

lsquoI will Vrsquo lsquothey will Vrsquodadumuli davadumuli lsquocutrsquodadʒimisi davadʒimisi lsquoextinguishrsquodaguvi davaguvi lsquofallrsquodadeti davadeti lsquosayrsquodababanisi davababanisi lsquosqueezersquodadaŋgi davadaŋgi lsquothrowrsquodanzeki davaseki lsquolaughrsquodamvuti davafuti lsquorubrsquodandovi davatovi lsquobeatrsquodandeleki davateleki lsquocookrsquodaŋgetuli davaketuli lsquocutrsquodaŋgamuli davakamuli lsquograbrsquodanduvi davatuvi lsquorunrsquodambutuki davaputuki lsquorubrsquodaɲdʒubuli davatʃubuli lsquoscratchrsquo

Singular Plural Singular Pluralklup klubi lsquoclubrsquo trup trupi lsquocorpsersquodom domi lsquohousersquo snop snopi lsquosheafrsquoʒwup ʒwobi lsquocribrsquo trut trudi lsquolaborrsquodzvon dzvoni lsquobellrsquo kot koti lsquocatrsquolut lodi lsquoicersquo grus gruzi lsquorubblersquonos nosi lsquonosersquo vus vozi lsquocartrsquowuk wugi lsquolyersquo wuk wuki lsquobowrsquosok soki lsquojuicersquo ruk rogi lsquohornrsquobur bori lsquoforestrsquo vuw vowi lsquooxrsquosul soli lsquosaltrsquo buj boji lsquofightrsquoʃum ʃumi lsquonoisersquo ʒur ʒuri lsquosouprsquo

4 LogooriAccount for the vowel alternations in the following data Tone may be ignored

The following nouns illustrate a productive pattern for making nouns lsquotool forV-ing withrsquo

lsquothey just Vrsquodrsquo lsquothey just Vrsquod forrsquo lsquothey will V

(rem fut)rsquo

lsquothey will V for

(rem fut)rsquo

vaakataanga vaakataangɪra varakataangɛ varakataangɪrɪ lsquostartrsquo

vaakavʊnjaanja vaakavʊnjaanjɪra varakavʊnjaanjɛ varakavʊnjaanjɪrɪ lsquobreakrsquo

vaakazaazama vaakazaazamɪra varakazaazamɛ varakazaazamɪrɪ lsquotastersquo

vaakavʊrʊganja vaakavʊrʊganjɪra varakavʊrʊganjɛ varakavʊrʊganjɪrɪ lsquostirrsquo

vaakaganaganja vaakaganaganjɪra varakaganaganjɛ varakaganaganjɪrɪ lsquothinkrsquo

vaakarʊʊnga vaakarʊʊngɪra varakarʊʊngɪ varakarʊʊngɪrɪ lsquoseasonrsquo

vaakarɪɪnda vaakarɪɪndɪra varakarɪɪndɪ varakarɪɪndɪrɪ lsquoguardrsquo

vaakatʊma vaakatʊmɪra varakatʊmɪ varakatʊmɪrɪ lsquosendrsquo

vaakasuunga vaakasuungɪra varakasuungɪ varakasuungɪrɪ lsquohang uprsquo

vaakatuuma vaakatuumɪra varakatuumɪ varakatuumɪrɪ lsquojumprsquo

vaakavɪna vaakavɪnɪra varakavɪnɪ varakavɪnɪrɪ lsquodancersquo

vaakavisa vaakavisɪra varakavisɪ varakavisɪrɪ lsquohidersquo

vaakavɪta vaakavɪtɪra varakavɪtɪ varakavɪtɪrɪ lsquopassrsquo

vaakagamʊra vaakagamʊrɪra varakagamʊrɪ varakagamʊrɪrɪ lsquocatchrsquo

vaakahɪɪrɪɪta vaakahɪɪrɪɪtɪra varakahɪɪrɪɪtɪ varakahɪɪrɪɪtɪrɪ lsquosnorersquo

vaakadiginja vaakadiginjɪra varakadiginjɪ varakadiginjɪrɪ lsquoticklersquo

vaakahakiza vaakahakizɪra varakahakizɪ varakahakizɪrɪ lsquoscorchrsquo

vaakavariza vaakavarizɪra varakavarizɪ varakavarizɪrɪ lsquocountrsquo

vaakafʊnjɪɪriza vaakafʊnjɪɪrizɪra varakafʊnjɪɪrizɪ varakafʊnjɪɪrizɪrɪ lsquosmellrsquo

vaakamɪnɪɪka vaakamɪnɪɪkɪra varakamɪnɪɪkɪ varakamɪnɪɪkɪrɪ lsquobe illrsquo

vaakagarʊkiza vaakagarʊkizɪra varakagarʊkizɪ varakagarʊkizɪrɪ lsquoreversersquo

vaakarʊʊngikiza vaakarʊʊngikizɪra varakarʊʊngikizɪ varakarʊʊngikizɪrɪ lsquostraightenrsquo

vaakatʊŋaminja vaakatʊŋaminjɪra varakatʊŋaminjɪ varakatʊŋaminjɪrɪ lsquoinvertrsquo

vaakasjɛɛna vaakasjɛɛnɛra varakasjɛɛnɛ varakasjɛɛnɛrɛ lsquosteprsquo

vaakarɛɛmba vaakarɛɛmbɛra varakarɛɛmbɛ varakarɛɛmbɛrɛ lsquoscoldrsquo

vaakarɔɔta vaakarɔɔtɛra varakarɔɔtɛ varakarɔɔtɛrɛ lsquodreamrsquo

vaakasɛka vaakasɛkɛra varakasɛkɛ varakasɛkɛrɛ lsquolaughrsquo

vaakatɛɛva vaakatɛɛvɛra varakatɛɛvɛ varakatɛɛvɛrɛ lsquoaskrsquo

vaakahɛɛnza vaakahɛɛnzɛra varakahɛɛnzɛ varakahɛɛnzɛrɛ lsquoseekrsquo

vaakarɔɔnda vaakarɔɔndɛra varakarɔɔndɛ varakarɔɔndɛrɛ lsquofollowrsquo

vaakarɛɛta vaakarɛɛtɛra varakarɛɛtɛ varakarɛɛtɛrɛ lsquobringrsquo

vaakadɛɛka vaakadɛɛkɛra varakadɛɛkɛ varakadɛɛkɛrɛ lsquocookrsquo

vaakamɛɲa vaakamɛɲɛra varakamɛɲɛ varakamɛɲɛrɛ lsquoliversquo

vaakasɔɔma vaakasɔɔmɛra varakasɔɔmɛ varakasɔɔmɛrɛ lsquoreadrsquo

vaakavɛga vaakavɛgɛra varakavɛgɛ varakavɛgɛrɛ lsquoshaversquo

vaakamɔrɔma vaakamɔrɔmɛra varakamɔrɔmɛ varakamɔrɔmɛrɛ lsquospeakrsquo

vaakarɛka vaakarɛkɛra varakarɛkɛ varakarɛkɛrɛ lsquostoprsquo

ividujɪrʊ lsquopoundrsquo ivivɛgɛrɔ lsquoshaversquo ivikaragɪrʊ lsquocutrsquoivisɔɔmɛrɔ lsquoreadrsquo ivivɪnɪrʊ lsquodancersquo ivinagɪllʊ lsquocatchrsquoivisɛɛmbɛllɔ lsquocultivatersquo ivirʊʊngɪrʊ lsquoseasonrsquo

5 ShonaAcute accent indicates H tone and unaccented vowels have L tone Given thetwo sets of data immediately below what tone rule do the following datamotivate There are alternations in the form of adjectives eg kurefu karefumarefu all meaning lsquolongrsquo Adjectives have an agreement prefix hence ku-refumarks the form of the adjective in one grammatical class and so on In somecases the agreement is realized purely as a change in the initial consonant ofthe adjective ie guru ~ kuru ~ huru which need not be explained

These data provide further illustration of the operation of this tone rule whichwill help you to state the conditions on the rule correctly

In the examples below a second tone rule applies

bveni lsquobaboonrsquo bveni pfupi lsquoshort baboonrsquotafura lsquotablersquo tafura huru lsquobig tablersquoʃoko lsquowordrsquo ʃoko bvupi lsquoshort wordrsquoɓadza lsquohoersquo ɓadza guru lsquobig hoersquozigomana lsquoboy

(augmentative)rsquozigomana guru lsquobig boy

(augmentative)rsquoimba lsquohousersquo imba tʃena lsquoclean housersquomara lsquogazellersquo mara tʃena lsquoclean gazellersquomarı lsquomoneyrsquo marı tʃena lsquoclean moneyrsquoɓaŋga lsquoknifersquo ɓaŋga guru lsquobig knifersquoɗemo lsquoaxersquo ɗemo bvupi lsquoshort axersquonume lsquomessengerrsquo nume pfupi lsquoshort messengerrsquodʒıra lsquoclothrsquo dʒıra dʒena lsquoclean clothrsquoharı lsquopotrsquo hari huru lsquobig potrsquombundudzı lsquowormsrsquo mbundudzi huru lsquobig wormsrsquofuma lsquowealthrsquo fuma tʃena lsquoclean wealthrsquoɲıka lsquocountryrsquo ɲıka huru lsquobig countryrsquohakata lsquobonesrsquo hakata pfupi lsquoshort bonesrsquodʒekera lsquopumpkinrsquo dʒekera guru lsquobig pumpkinrsquo

guɗo lsquobaboonrsquo guɗo rakafa lsquothe baboon diedrsquoɓaɗza lsquohoersquo ɓadza rakawa lsquothe hoe fellrsquonuŋgu lsquoporcupinersquo nuŋgu jakafa lsquothe porcupine

diedrsquoɓaŋga lsquoknifersquo ɓaŋga rakawa lsquothe knife fellrsquonume lsquomessengerrsquo nume jakafa lsquothe messenger diedrsquobuku lsquobookrsquo buku rakawa lsquothe book fellrsquomapfeni lsquobaboonsrsquo mapfeni makuru lsquobig baboonsrsquomapadza lsquohoesrsquo mapadza makuru lsquobig hoesrsquomapaŋga lsquoknivesrsquo mapaŋga makuru lsquobig knivesrsquonume lsquomessengerrsquo nume ndefu lsquotall messengerrsquomatemo lsquoaxesrsquo matemo mapfupi lsquoshort axesrsquomabuku lsquobooksrsquo mabuku maʒındʒı lsquomany booksrsquotʃitoro lsquostorersquo tʃitoro tʃikuru lsquobig storersquo

What do the following examples show about these tone rules

6 GuerzeAccount for the phonological alternations in the following data from Guerze Besure that you state the order of the rules which you propose and justify yourconclusion about ordering

guɗo lsquobaboonrsquo guɗo refu lsquotall baboonrsquobuku lsquobookrsquo buku refu lsquolong bookrsquoɓadza lsquohoersquo ɓadza refu lsquolong hoersquonuŋgu lsquoporcupinersquo nuŋgu ndefu lsquolong porcupinersquomaʃoko lsquowordsrsquo maʃoko marefu lsquolong wordsrsquokuɲıka lsquoto the landrsquo kuɲıka kurefu lsquoto the long landrsquomapadza lsquohoesrsquo mapadza marefu lsquolong hoesrsquokamara lsquogazelle (dim)rsquo kamara karefu lsquolong gazelle (dim)rsquotunuŋgu lsquoporcupines (dim)rsquo tunuŋgu turefu lsquolong porcupines

(dim)rsquoguɗo lsquobaboonrsquo guɗo gobvu lsquothick baboonrsquobuku lsquobookrsquo buku gobvu lsquothick bookrsquoɓadza lsquohoersquo ɓadza gobvu lsquothick hoersquomakuɗo lsquobaboonsrsquo makuɗo makobvu lsquothick baboonsrsquomapadza lsquohoesrsquo mapadza makobvu lsquothick hoesrsquotsamba lsquoletterrsquo tsamba nete lsquothin letterrsquobuku lsquobookrsquo buku ɗete lsquothin bookrsquoɓadza lsquohoersquo ɓadza ɗete lsquothin hoersquoimba lsquohousersquo imba nete lsquothin housersquo

ɓaŋga lsquoknifersquo ɓaŋga ɗete lsquothin knifersquoɗemo lsquoaxersquo ɗemo ɗete lsquothin axersquomurume lsquopersonrsquo murume mutete lsquothin personrsquokahunı lsquofirewood (dim)rsquo kahunı karefu lsquolong firewoodrsquomatʃıra lsquoclothesrsquo matʃıra marefu lsquolong clothesrsquoharı lsquopotrsquo harı nete lsquothin potrsquo

bamaŋ lsquoharp-drumrsquo bama bo lsquo10 harp-drumsrsquobama dɔnɔ lsquo1 harp-drumrsquo bama nɔɔlu lsquo5 harp-drumsrsquobama ŋudono lsquo100 harp-drumsrsquo bama ŋujaa lsquoheavy harp-drumrsquobʌlʌ lsquoyamrsquo bʌlʌ kujaa lsquolong yamrsquobʌlʌ lɔɔlu lsquo5 yamsrsquo bʌlʌ ɲɔŋɔ lsquobad yamrsquobʌlʌ tɛγʌja lsquoblack yamrsquo bʌlʌ joγo lsquowet yamrsquogbɔŋ lsquowoodrsquo gbɔ naa lsquo4 wood piecesrsquogbɔ ɲoγo lsquowet woodrsquo gbɔ nʌnʌ lsquonew woodrsquohiiŋ lsquodesignrsquo hii dɔnɔ lsquo1 designrsquohii gujaa lsquolong designrsquo hii gʌnʌ lsquobig designrsquohii naa lsquo4 designsrsquo hii ɲɔŋɔ lsquobad designrsquohii nokolo lsquosmall designrsquo hʌŋ lsquothingrsquohʌ dɛγʌja lsquoblack thingrsquo hʌ gʌnʌ lsquobig thingrsquohʌ ɲoγo lsquowet thingrsquo hʌ ŋudono lsquo100 thingsrsquohʌ ŋujaa lsquoheavy thingrsquo kɨhɨ lsquosuitcasersquokɨhɨ kujaa lsquolong suitcasersquo kɨhɨ lokolo lsquosmall suitcasersquo

7 CatalanGive phonological rules which account for the following data and indicate whatordering is necessary between these rules For each adjective stem state whatthe underlying form of the root is Pay attention to the difference betweensurface [b d g] and [β eth γ] in terms of predictability

kɨhɨ naa lsquo4 suitcasesrsquo kɨhɨ tɔnɔ lsquo1 suitcasersquokɨhɨ wujaa lsquoheavy suitcasersquo lii lsquoheartrsquolii kʌnʌ lsquobig heartrsquo lii ɲɔŋɔ lsquobad heartrsquolii ŋudono lsquo100 heartsrsquo ɲɨŋ lsquotoothrsquoɲɨ bo lsquo10 teethrsquo ɲɨ dɛγʌja lsquoblack toothrsquoɲɨ gujaa lsquolong toothrsquo ɲɨ nɔɔlu lsquo5 teethrsquoɲɨ ɲɔŋɔ lsquobad toothrsquo ɲɨ nʌnʌ lsquonew toothrsquoɲɨ ŋujaa lsquoheavy toothrsquo tɛɛ lsquochickenrsquotɛɛ kujaa lsquolong chickenrsquo tɛɛ lɔɔlu lsquo5 chickensrsquotɛɛ lokolo lsquosmall chickenrsquo tɛɛ nʌnʌ lsquonew chickenrsquotɛɛ ŋudono lsquo100 chickensrsquo tɛɛ po lsquo10 chickensrsquotɛɛ tɛγʌja lsquoblack chickenrsquo tɛɛ joγo lsquowet chickenrsquojaba lsquoonionrsquo jaba kʌnʌ lsquobig onionrsquojaba naa lsquo4 onionsrsquo jaba nʌnʌ lsquonew onionrsquojaba po lsquo10 onionsrsquo jaba tɔnɔ lsquo1 onionrsquojaba wujaa lsquoheavy onionrsquo

Masc sg Fem sg Masc sg Fem sgəkelj əkeljə lsquothatrsquo mal malə lsquobadrsquosiβil siβilə lsquocivilrsquo əskerp əskerpə lsquoshyrsquoʃop ʃopə lsquodrenchedrsquo sεk sεkə lsquodryrsquoəspεs əspεsə lsquothickrsquo gros grosə lsquolargersquobaʃ baʃə lsquoshortrsquo koʃ koʃə lsquolamersquotot totə lsquoallrsquo brut brutə lsquodirtyrsquopɔk pɔkə lsquolittlersquo prəsis prəsizə lsquoprecisersquofrənses frənsezə lsquoFrenchrsquo gris grizə lsquogreyrsquokəzat kəzaethə lsquomarriedrsquo bwit bwiethə lsquoemptyrsquorɔtʃ rɔʒə lsquoredrsquo botʃ boʒə lsquocrazyrsquoorp orβə lsquoblindrsquo ljark ljarγə lsquolongrsquosek seγə lsquoblindrsquo fəʃuk fəʃuγə lsquoheavyrsquogrok groγə lsquoyellowrsquo puruk puruγə lsquofearfulrsquokandit kandiethə lsquocandidrsquo frεt frεethə lsquocoldrsquosəγu səγurə lsquosurersquo du durə lsquohardrsquosəγəetho səγəethorə lsquoreaperrsquo kla klarə lsquoclearrsquonu nuə lsquonudersquo kru kruə lsquorawrsquoflɔɲdʒu flɔɲdʒə lsquosoftrsquo dropu dropə lsquolazyrsquoəgzaktə əgzaktə lsquoexactrsquo əlβi əlβinə lsquoalbinorsquosa sanə lsquohealthyrsquo pla planə lsquolevelrsquobo bonə lsquogoodrsquo sərε sərεnə lsquocalmrsquosuβlim suβlimə lsquosublimersquo al altə lsquotallrsquofɔr fɔrtə lsquostrongrsquo kur kurtə lsquoshortrsquosor sorethə lsquodeafrsquo bεr bεrethə lsquogreenrsquo

8 FinnishPropose rules which will account for the following alternations It would be bestnot to write a lot of rules which go directly from underlying forms to surfaceforms in one step instead propose a sequence of rules whose combined effectbrings about the change in the underlying form Pay attention to whatconsonants actually exist in the language

Genitive sg Nom sg Nom pl Ablative sg Essive sgkanadan kanada kanadat kanadalta kanadana lsquoCanadarsquokirjan kirja kirjat kirjalta kirjana lsquobookrsquoaamun aamu aamut aamulta aamuna lsquomorningrsquotalon talo talot talolta talona lsquohousersquokoiran koira koirat koiralta koirana lsquodogrsquohyvaelign hyvaelig hyvaeligt hyvaeligltaelig hyvaelignaelig lsquogoodrsquokuvan kuva kuvat kuvalta kuvana lsquopicturersquolain laki lait lailta lakina lsquoroofrsquonaeliglaelign naeliglkaelig naeliglaeligt naeliglaeligltaelig naeliglkaelignaelig lsquohungerrsquojalan jalka jalat jalalta jalkana lsquolegrsquoleuan leuka leuat leualta leukana lsquochinrsquoparan parka parat paralta parkana lsquopoorrsquoreiaelign reikaelig reiaeligt reiaeligltaelig reikaelignaelig lsquoholersquonahan nahka nahat nahalta nahkana lsquohidersquovihon vihko vihot viholta vihkona lsquonotebookrsquolaihan laiha laihat laihalta laihana lsquoleanrsquoavun apu avut avulta apuna lsquohelprsquohalvan halpa halvat halvalta halpana lsquocheaprsquoorvon orpo orvot orvolta orpona lsquoorphanrsquoleivaelign leipaelig leivaeligt leivaeligltaelig leipaelignaelig lsquobreadrsquopaeligivaelign paeligivaelig paeligivaeligt paeligivaeligltaelig paeligivaelignaelig lsquodayrsquokilvan kilpa kilvat kilvalta kilpana lsquocompetitionrsquokylvyn kylpy kylvyt kylvyltaelig kylpynaelig lsquobathrsquotavan tapa tavat tavalta tapana lsquomannerrsquokorvan korva korvat korvalta korvana lsquoearrsquoaeligidin aeligiti aeligidit aeligidiltaelig aeligitinaelig lsquomotherrsquokodin koti kodit kodilta kotina lsquohomersquomuodon muoto muodot muodolta muotona lsquoformrsquotaeligdin taeligti taeligdit taeligdiltaelig taeligtinaelig lsquoauntrsquokadun katu kadut kadulta katuna lsquostreetrsquomaidon maito maidot maidolta maitona lsquomilkrsquoposlashydaelign poslashytaelig poslashydaeligt poslashydaeligltaelig poslashytaelignaelig lsquotablersquotehdyn tehty tehdyt tehdyltaelig tehtynaelig lsquomadersquolaeligmmoslashn laeligmposlash laeligmmoslasht laeligmmoslashltaelig laeligmposlashnaelig lsquowarmthrsquolaŋŋan laŋka laŋŋat laŋŋalta laŋkana lsquothreadrsquosaeligŋŋyn saeligŋky saeligŋŋyt saeligŋŋyltaelig saeligŋkynaelig lsquobedrsquo

san santə lsquosaintrsquo kəlεn kəlεntə lsquohotrsquoprufun prufundə lsquodeeprsquo fəkun fəkundə lsquofertilersquodəsen dəsentə lsquodecentrsquo dulen dulentə lsquobadrsquoəstuethian əstuethiantə lsquostudentrsquo blaŋ blaŋkə lsquowhitersquo

Further readingAnderson 1974 Chomsky 1967 Goldsmith 1990b Kiparsky 1968a Koutsoudas Sanders and Noll 1974

hinnan hinta hinnat hinnalta hintana lsquopricersquolinnun lintu linnut linnulta lintuna lsquobirdrsquoopinnon opinto opinnot opinnolta opintona lsquostudyrsquorannan ranta rannat rannalta rantana lsquoshorersquoluonnon luonto luonnot luonnolta luontona lsquonaturersquopunnan punta punnat punnalta puntana lsquopoundrsquotunnin tunti tunnit tunnilta tuntina lsquohourrsquokunnon kunto kunnot kunnolta kuntona lsquoconditionrsquokannun kannu kannut kannulta kannuna lsquocanrsquolinnan linna linnat linnalta linnana lsquocastlersquotumman tumma tummat tummalta tummana lsquodarkrsquoauriŋŋon auriŋko auriŋŋot auriŋŋolta auriŋkona lsquosunrsquoreŋŋin reŋki reŋŋit reŋŋiltaelig reŋkinaelig lsquofarm handrsquovaŋŋin vaŋki vaŋŋit vaŋŋilta vaŋkina lsquoprisonerrsquokellon kello kellot kellolta kellona lsquowatchrsquokellan kelta kellat kellalta keltana lsquoyellowrsquosillan silta sillat sillalta siltana lsquobridgersquokullan kulta kullat kullalta kultana lsquogoldrsquovirran virta virrat virralta virtana lsquostreamrsquoparran parta parrat parralta partana lsquobeardrsquo

CHAPTER

6 Doing ananalysis

PREVIEW

This chapter explores a subset of the phonologies of

a number of languages The purpose of this chapter is

to make explicit the reasoning typically applied to the task

of solving a phonology problem By studying models of

problem solving you not only better understand the logic

of problem solving you will also gain experience with

rules and issues regarding underlying representations

encountered in the languages of the world

KEY TERMShypothesisformation andtesting

competinghypotheses

Analyzing a system of phonological alternations is not trivial it requirespractice where you gain experience by solving phonological problems ofincreasing complexity experience which facilitates subsequent problemsolving The wider your experience is with actual phonological processesand problem solving the better able you will be to appreciate whatprocesses are common in the languages of the world and to understandthe dynamics of hypothesis formation testing and revision The firstanalyses given here will be more explicit about the reasoning that goesinto solving data sets of this nature in some cases deliberately goingdown the wrong analytical path so that you have the opportunity torecognize the wrong path and see how to get back on the right path Inpractice many of the calculations that are involved here are done withoutexplicitly thinking about it ndash once you have suitable experience withproblem solving

61 Yawelmani

Our first problem involves alternations in the verb paradigm in theYawelmani dialect of Yokuts (California)

611 The dataThree phonological rules will be motivated by the following examplesvowel epenthesis vowel shortening and vowel harmony It is not obviouswhat the underlying representation of verb roots is so besides finding therules we must make decisions about underlying forms

(1) Nonfuture Imperative Dubitative Passive aoristxathin xatkrsquoa xatal xatit lsquoeatrsquodubhun dubkrsquoa dubal dubut lsquolead by handrsquoxilhin xilkrsquoa xilal xilit lsquotanglersquokrsquooʔhin krsquooʔkrsquoo krsquooʔol krsquooʔit lsquothrowrsquo

doshin doskrsquoo dosol dosit lsquoreportrsquoʂaphin ʂapkrsquoa ʂapal ʂapit lsquoburnrsquolanhin lankrsquoa lanal lanit lsquohearrsquomekrsquohin mekrsquokrsquoa mekrsquoal mekrsquoit lsquoswallowrsquo

wonhin wonkrsquoo wonol wonit lsquohidersquoprsquoaxathin prsquoaxatkrsquoa prsquoaxatal prsquoaxatit lsquomournrsquohiwethin hiwetkrsquoa hiwetal hiwetit lsquowalkrsquoʔopothin ʔopotkrsquoo ʔopotol ʔopotit lsquoarise from bedrsquojawalhin jawalkrsquoa jawalal jawalit lsquofollowrsquo

paʔiʈhin paʔiʈkrsquoa paʔʈal paʔʈit lsquofightrsquoʔilikhin ʔilikkrsquoa ʔilkal ʔilkit lsquosingrsquologiwhin logiwkrsquoa logwol logwit lsquopulverizersquoʔugunhun ʔugunkrsquoa ʔugnal ʔugnut lsquodrinkrsquolihimhin lihimkrsquoa lihmal lihmit lsquorunrsquoʔajijhin ʔajijkrsquoa ʔajjal ʔajjit lsquopole a boatrsquo

612 The first step morphologyFirst we need a morphological analysis of the data In a simple case thisinvolves looking at columns and rows of data and figuring out which sub-parts of words are consistently present with one meaning and whichother subparts are consistently present with other meanings This taskis more complicated when the surface shape of roots and affixes changesdue to phonological rules We cannot provide a definitive morphologicalanalysis of these data without knowing what the phonological system isand certainty as to the phonological rules is impossible without knowingthe morphological analysis We break out of this seeming circle byadopting ndash and constantly revising in the face of new evidence ndash a prelim-inary and less precise analysis of the phonology and morphologyImprovement in the underlying representations should result in betterrules and as we refine the system of rules the nature of the underlyingdistinctions becomes clearerIn this case four suffixes are added to roots -hin ~ -hun lsquononfuturersquo -krsquoa

~ -krsquoo lsquoimperativersquo -al ~ -ol lsquodubitativersquo and -it ~ -ut lsquopassive aoristrsquo Thenotation -hin ~ -hun indicates that the suffix is pronounced either as -hin oras -hun We need to discover when one form versus the other is used andexpress that relation in terms of an underlying form and a rule changingthe underlying form

Stem variants Some stems have only one surface shape xat- lsquoeatrsquo dub-lsquolead by handrsquo xil- lsquotanglersquo and krsquooʔ- lsquothrowrsquo so the most natural assump-tion would be that these are the underlying forms for these particularstems (this assumption may turn out to be wrong but it is a good startingassumption) Most stems in the data set have two surface manifestationsAn important first step in understanding the rules of the language is toidentify the alternations in the data and one way to make the alternationsexplicit is to list the phonetic variants of each stem

(2) dos ~ dos lsquoreportrsquo ʂap ~ ʂap lsquoburnrsquolan ~ lan lsquohearrsquo mekrsquo ~ mekrsquo lsquoswallowrsquo

won ~ won lsquohidersquo prsquoaxat ~ prsquoaxat lsquomournrsquohiwet ~ hiwet lsquowalkrsquo ʔopot ~ ʔopot lsquoarise from bedrsquojawal ~ jawal lsquofollowrsquo paʔiʈ ~ paʔʈ lsquofightrsquoʔilik ~ ʔilk lsquosingrsquo logiw ~ logw lsquopulverizersquoʔugun ~ ʔugn lsquodrinkrsquo lihim ~ lihm lsquorunrsquoʔajij ~ ʔajj lsquopole a boatrsquo trsquoojix ~ trsquoojx lsquogive medicinersquo

trsquoojixhin trsquoojixkrsquoa trsquoojxol trsquoojxit lsquogive medicinersquolukrsquoulhun lukrsquoulkrsquoa lukrsquolal lukrsquolut lsquoburyrsquosonilhin sonilkrsquoa sonlol sonlit lsquoput on backrsquoʔamilhin ʔamilkrsquoa ʔamlal ʔamlit lsquohelprsquomojinhin mojinkrsquoa mojnol mojnit lsquobecome tiredrsquoʂalikrsquohin ʂalikrsquokrsquoa ʂalkrsquoal ʂalkrsquoit lsquowake uprsquo

Doing an analysis 149

In these cases decisions must be made regarding the underlying forms

Suffix variants We must decide what the underlying form of eachsuffix is and they all have two surface variants in terms of their voweleither a nonrounded vowel or a rounded vowel For each suffix we groupthe verbs in terms of which variant of the suffix is used with them

613 Identifying phonological regularitiesVowel harmony Having grouped the examples in this fashion a phono-logical regularity can be detected For the suffix hin ~ hun the vowel uappears when the preceding vowel is u and i appears in the suffix afterany other vowel The suffix it ~ ut obeys this same rule The suffixes krsquoa ~krsquoo and al ~ ol have the vowel o after o This can be explained by positing arule of vowel harmony between the suffix vowel and whatever vowelprecedes it where a assimilates to o and i assimilates to u

The variable notation ndash αhigh αhigh ndash expresses the condition that thevowels must have the same value of [high] ie the harmonizing vowelmust be [+high] after a [+high] round vowel and [-high] after a [-high]round vowel in order for the harmony rule to apply

-hin xat xil krsquooʔ dos ʂap lan mekrsquo won prsquoaxat hiwet ʔopotjawal paʔiʈ ʔilik logiw lihim ʔajij trsquoojix sonil ʔamilmojin ʂalikrsquo

-hun dub ʔugun lukrsquoul-krsquoa xat dub xil ʂap lan mekrsquo prsquoaxat hiwet jawal paʔiʈ

ʔilik logiw ʔugun lihim ʔajij trsquoojix lukrsquoul sonil ʔamilmojin ʂalikrsquo

-krsquoo krsquooʔ dos won ʔopot-al xat dub xil ʂap lan mekrsquo prsquoaxat hiwet jawal paʔʈ

ʔilk ʔugn lihm ʔajj lukrsquol ʔaml ʂalkrsquo-ol krsquooʔ dos won ʔopot logw trsquoojx sonl mojn-it xat xil krsquooʔ dos ʂap lan mekrsquo won prsquoaxat hiwet

ʔopot jawal paʔʈ ʔilk logw lihm ʔajj trsquoojx sonl ʔamlmojn ʂalkrsquo

-ut dub ʔugn lukrsquol

lukrsquoul ~ lukrsquol lsquoburyrsquo sonil ~ sonl lsquoput on backrsquoʔamil ~ ʔaml lsquohelprsquo mojin ~ mojn lsquobecome tiredrsquoʂalikrsquo ~ ʂalkrsquo lsquowake uprsquo

mdashCO

⎡⎢⎣

⎤⎥⎦

Vαhigh ⎢

⎡⎢

⎢⎥⎦

⎤⎥

⎥αhigh+round

[ ]+ roundrarr V

Vowel shortening The next problem to tackle is the variation in theshape of the stem A useful next step in trying to analyze that variation isto see whether the variants can be arranged into a small number ofgroups organized according to the nature of the difference between thetwo stem shapes In looking for such an organization notice that somestems alternate in terms of having long versus short vowels and in termsof having versus lacking a second vowel Accordingly we organize thedata into the following classes of stem alternations (including the class ofstems which have no alternation)

The initial hypothesis is that the invariant CVC stems have the underlyingshape CVC If there is no reason to make the underlying form be differentfrom the surface form the two forms should be assumed to be identicalBuilding on that decision we will now set forth a hypothesis for stemswhich vary in shape between CVC and CVC It is highly unlikely thatthese stems also have the underlying shape CVC since that would make ithard to account for stems such as xat which are invariant CVC We couldnot predict whether a stem vowel is supposed to have a length alternationor not and the reasoning that leads to hypothesizing an underlyingdistinction xat vs dos which is contextually neutralized is exactly thesame as that which leads to hypothesizing that in Russian (discussed inchapter 4) the word for lsquotimersquo is underlyingly raz and for lsquoforestrsquo it is lesGiven the conclusion that stems like dos ~ dos have an underlying CVC

form under what circumstance is the underlyingly long vowel of the stemshortened Taking dos as a representative and mechanically combiningthe assumed underlying stemwithwhat we take to be the underlying formof the suffix we arrive at the following underlying and surface relations

The change of a to [o] is due to vowel harmony There is also a change invowel length before krsquoa and hin and not before -al and -it These suffixesare distinguished by whether they begin with a consonant or a vowel thuswhether combining the stem and suffix would result in the sequenceVCC Scanning the entire data set reveals an important generalizationthat a long vowel is always followed by CV that is a long vowel only

underlying dos-hin dos-krsquoa dos-al dos-itsurface doshin doskrsquoo dosol dosit

CVC - xat dub xil krsquooʔCVC ~ CVC ndash dos ~ dos ʂap ~ ʂap lan ~ lan mekrsquo ~ mekrsquo

won ~ wonCVCVC ~ CVCVC ndash prsquoaxat ~ prsquoaxat hiwet ~ hiwet ʔopot ~ ʔopot

jawal ~ jawalCVCVC ~ CVCC ndash paʔiʈ ~ paʔʈ ʔilik ~ ʔilk logiw ~ logw ʔugun

~ ʔugn lihim ~ lihm ʔajij ~ ʔajj trsquoojix ~ trsquoojxlukrsquoul ~ lukrsquol

CVCVC ~ CVCC ndash sonil ~ sonl ʔamil ~ ʔaml mojin ~ mojnʂalikrsquo ~ ʂalkrsquo

occurs in an open syllable The discovery of this generalization allows usto posit the following vowel shortening rule

(7) V [-long] __ CC

This rule is all that is needed to explain both the invariant CVC stems andthe alternating CVC ~ CVC stems Underlyingly dos-hin undergoes (7)and gives the surface form [doshin] ndash all other forms preserve the under-lying length of the vowel The existence of this rule also explains why wedo not find the surface sequence VCC ndash a long vowel before a cluster oftwo consonants ndash anywhere in the data as such sequences undergo vowelshorteningWe turn next to the stems with the shape CVCVC ~ CVCVC such as prsquoaxat

~ prsquoaxat Since we have already encountered a rule which accounts foralternations in vowel length we should immediately suspect that thislength alternation is the same as the one just accounted for in CVC ~ CVCstems When we inspect the contexts where the long-vowel variant occurswe see that there are long vowels when a vowel-initial suffix is added andshort vowels when a consonant-initial suffix is added In other words thesestems are virtually the same as CVC stems except that they have theunderlying shape CVCVC We initially hypothesized that there was a ruleof vowel shortening based on CVC stems and that rule nicely handledthose data The way we formulated that rule was quite general since it onlysaid ldquoshorten a long vowel before two consonantsrdquo Such a statementpredicts that if there are other stem shapes such as CVCVC they too willundergo that ruleWe have now discovered that such stems do undergo theshortening rule providing independent support for that rule

Epenthesis This reduces the unsolved part of the problem to tworemaining classes of stems In one of those there is an alternation betweenpresence versus absence of a vowel and in the second group there is analternation in vowel length as well as an alternation in the presence versuslack of a vowel in the second syllable this should make us suspect that thevowel shortening rule applies to the second of these sets Concentrating onthe contexts where the stem has the shape CV()CVC as opposed to theshape CVCC we notice that CV()CVC appears before consonant-initialsuffixes and CVCC appears before vowel-initial suffixes We do not knowat this point whether the second vowel is underlyingly part of the stem andis deleted in one context or whether the vowel is inserted in a differentcontext Therefore we will consider both possibilities consideration ofalternative hypotheses is an essential part of problem solvingFirst suppose that the vowel is not part of the underlying representation

of the stem In that case we assume the following representations

(8) underlying ʔilk-hin ʔilk-krsquoa ʔilk-al ʔilk-itsurface ʔilik-hin ʔilik-krsquoa ʔilk-al ʔilk-it

underlying ʂalkrsquo-hin ʂalkrsquo-krsquoa ʂalkrsquo-al ʂalkrsquo-itsurface ʂalikrsquo-hin ʂalikrsquo-krsquoa ʂalkrsquo-al ʂalkrsquo-it

Focusing on the hypothesized underlying representations where a vowelmight be inserted we notice that a vowel appears only where the under-lying representation has a sequence of three consonants Looking at all ofthe data we notice that there are no surface sequences of three or moreconsonants making such an epenthesis approach plausibleIn order for an epenthesis solution to work the actual quality of the

inserted vowel must be completely predictable If we were to discover thatthe quality of the second vowel is unpredictable then it would necessarilybe part of the underlying representation since unpredictble informationmust be in the underlying form The vowel in the second syllable is alwayshigh and is round when the preceding vowel is high and round In otherwords the vowel in question is a high vowel whose backness and round-ness is predictable given the rule of vowel harmony and thus the vowel isfully predictable Given the harmony rule we can assume that the secondvowel is i It is then possible to account for these examples by applying thefollowing rule of epenthesis

Given (9) the underlying form of the CVCiC ~ CVCC stems would beCVCC and the underlying form of the CVCiC ~ CVCC stems wouldbe CVCC For stems like ʔilk epenthesis applies to underlyingCVCC+CV(C) to give surface [CVCiC+CV(C)] ʔilk-hin [ʔilikhin] Thealternant CVCC before VC suffixes ~ [ʔilkal] ~ directly reflects the under-lying formFor CVCC stems like ʂalkrsquo epenthesis will also apply to underlying

CVCC+CV(C) giving the surface form [CVCiC+CV(C)] ʂalk-hin [ʂalikhin] When a VC suffix is added to such stems there is no epenthesisbut we do find shortening of the underlyingly long vowel whichstands before a consonant cluster (ʂalkal [ʂalkal]) The rules of vowelharmony epenthesis and vowel shortening combined with our analy-ses of underlying representations account for all aspects of the datain (1) We conclude that epenthesis is a possible account of thesealternationsThe preceding analysis has assumed a rule of epenthesis based on

underlying representations of the form CVCC and CVCC but we shouldexplore the competing hypothesis that the vowel found in these stems isnot inserted and is part of the underlying representation Under thathypothesis underlying representations of the relevant stems would bethe following

(10) paʔit ʔilik logiw ʔugun lihim ʔajij trsquoojix lukrsquoul sonil ʔamil mojin ʂalikrsquo

Presuming that these are the underlying stems a rule of vowel deletion isrequired to explain the discrepancy between surface and underlyingforms which can be seen in (11)

+high[ ]rarrOslash __V C CC

In forms which involve an alternation between a vowel andempty the contextfor vowel deletion would initially appear to be in an open syllable Thisstatement would produce too general a rule since there are many vowelsin open syllables viz xatal krsquooʔit dosit prsquoaxathin and prsquoaxatal amongothers In some of these deletion of a vowel would lead to a word-initialconsonant cluster ie we would predict xtal krsquoʔit dsit prsquoxathin andprsquoxatal and we see no word-initial clusters of consonants If we are tohave vowel deletion the rule must be restricted from creating suchclusters so one way to enforce that requirement is to require the targetof deletion to be preceded by the sequence VC Thus we might hypothe-size the following syncope rule one found in many languages

(12) V Oslash VC _ CV

This rule still makes incorrect predictions since in fact there are vowels inthe context VC_CV as shown by forms such as prsquoaxatal ʔopotit whichaccording to (12) should be deleted Since all such examples involve longvowels it is a simple matter to restrict the assumed deletion rule to shortvowels

With this rule of vowel syncope the problem of vowel ~ Oslash alternationscan also be accounted for The remaining details of the analysis are exactlythe same as they are under the assumption that there is a rule of vowelinsertion

614 Evaluating alternativesIn terms of simply generating the data both the syncope and epenthesisanalyses work The question then becomes is there a reason to choose oneof these hypotheses over the other It is entirely possible that we will notbe able to come up with any compelling reasons for selecting one analysisover the other in which case we must simply accept the fact that there aretwo equally plausible ways to account for the facts As far as the simpli-city naturalness and generality of the two analyses is concerned neithertheory is superior to the other Processes inserting vowels to break up CCCclusters are very common as are rules of syncope which delete shortvowels in the context VC_CVWe should also consider the factual predictions of the two analyses The

epenthesis analysis predicts that there should be no CCC sequences in thelanguage and this appears to be correct On the other hand the syncope

underlying sonil-hin sonil-krsquoa sonil-ol sonil-itsurface sonil-hin sonil-krsquoa sonl-ol sonl-it

underlying lukrsquoul-hun lukrsquoul-krsquoa lukrsquoul-al lukrsquoul-utsurface lukrsquoul-hun lukrsquoul-krsquoa lukrsquol-al lukrsquol-ut

minus long[ ]rarr __Oslash VCV CV

analysis predicts that there should be no short vowels in the contextVC_CV which also appears to be correct Interestingly neither accountactuallymakes the prediction of the competing analysis ndash so the epenthesisanalysis does not preclude the existence of short vowels in the VC_CVcontext and the syncope analysis does not preclude the existence of CCCsequences If it turns out that there are CCC sequences in the language theepenthesis solution will probably have to be rejected whereas if there areVCVCV sequences in the language the syncope analysis will probably haveto be rejected This would motivate further research into the language todetermine if one of these analyses makes a bad predictionA related issue to consider is the question of ldquocoincidencerdquo in terms of

assumed underlying representations In lieu of a specific rule whichrestricts the occurrence of phonemes in some environment we expectphonemes to combinewithout any constraints Clearly theremust be someconstraints on underlying representations in Yawelmani since forexample we do not find underlying representations such as ioate withsequences of vowels In this case there is nomotivation from phonologicalalternations to suspect that theremight be underlying forms such as ioateAs far as logical possibilities in underlying forms are concerned for theissue at hand ndash epenthesis versus deletion ndash both analyses result in system-atic gaps in the logically possible underlying forms Under the epenthesisanalysis there are apparently no stems of the underlying form CVCVCalthough there are stems of the form CVCVC Under the syncope analysiswe notice that all short second-syllable vowels in disyllabic stems are in facti (surface [u] in some cases in accordance with vowel harmony)At this point it is impossible to give strong arguments in favor of one

analysis over another so we accept this indeterminacy for now Thefundamental point is that each analysis implies a set of predictions aboutpossible and impossible forms in the language and these predictions needto be tested against the available data In this case we have not been ableto determine that one theory is clearly superior to the other The mainresearch problem which we face is that the corpus of data from Yawel-mani available to us at this point is restricted so we cannot know whethergeneralizations which we extract about the language based on this par-ticular corpus are representative of the language as a whole Even if wehad access to a reference grammar for the language there is some chancethat our empirical generalizations based on the data from that grammarwould not hold for the whole language if the author of the grammar werenot aware of all relevant types of examples

62 Hehe

The following data illustrate phonological processes of Hehe (Tanzania)Each noun is in one of fifteen numbered noun classes like genders inFrench or German The class of a noun is marked by a prefix The goal is todetermine the underlying form of stems and prefixes and explain theprocesses at work in these data

621 The dataHere are the relevant data from nouns

(14) Class 1mutesi lsquotrapperrsquo mulagusi lsquosorcererrsquomutelesi lsquocookrsquo muɲwi lsquodrinkerrsquomwiimbi lsquosingerrsquo mweendi lsquoone who likes peoplersquomwaasi lsquobuilderrsquo moogofi lsquoone who is afraidrsquomoofusi lsquoone who washesrsquo muutsi lsquoone who comesrsquo

Class 2vatesi lsquotrappersrsquo valagusi lsquosorcerersrsquovatelesi lsquocooksrsquo vaɲwi lsquodrinkersrsquoviimbi lsquosingersrsquo veendi lsquoones who like peoplersquovaasi lsquobuildersrsquo woogofi lsquoones who are afraidrsquowoofusi lsquoones who washrsquo wuutsi lsquoones who comersquo

Class 3muhoomi lsquocow humprsquo muhogo lsquocassavarsquomufuniko lsquocoverrsquo muvili lsquobodyrsquomwiina lsquoholersquo mwiigiigi lsquoshadowrsquo

mweenda lsquoclothrsquo mooto lsquofirersquomuuɲi lsquosaltrsquo

Class 4mihoomi lsquocow humpsrsquo mihogo lsquocassavasrsquomifuniko lsquocoversrsquo mivili lsquobodiesrsquomiina lsquoholesrsquo miigiigi lsquoshadowsrsquomjeenda lsquoclothsrsquo mjooto lsquofiresrsquomjuuɲi lsquosaltsrsquo

Class 6mavafi lsquohairy caterpillarsrsquo masaasi lsquobulletsrsquomaboga lsquopumpkinsrsquo majaji lsquolegsrsquomiino lsquoteethrsquo miiho lsquoeyesrsquo

Class 7kigidi lsquowaistrsquo kingaamba lsquosweet potatorsquokisogo lsquoback of headrsquo tʃuula lsquofrogrsquotʃuunga lsquowet lowlandrsquo tʃaanga lsquograversquokifuniko lsquotiny coverrsquo kivili lsquotiny bodyrsquokihoomi lsquotiny cow humprsquo kivafi lsquotiny hairy caterpillarrsquotʃooto lsquotiny firersquo tʃeenda lsquotiny clothrsquotʃuuɲu lsquotiny saltrsquo kiiho lsquotiny eyersquokiina lsquotiny holersquo kiigiigi lsquotiny shadowrsquo

Class 8figidi lsquowaistsrsquo fingaamba lsquosweet potatoesrsquofisogo lsquobacks of headrsquo fjuula lsquofrogsrsquofjuunga lsquowet lowlandsrsquo fjaanga lsquogravesrsquofifuniko lsquotiny coversrsquo fivili lsquotiny bodiesrsquofihoomi lsquotiny cow humpsrsquo fivafi lsquotiny hairy caterpillarsrsquofjooto lsquotiny firesrsquo fjeenda lsquotiny clothsrsquo

622 Morphological analysisAs always a preliminary morphological analysis is the first step in solvingthis phonology problem Each noun has some prefix that marks nounclass followed by a stem We also see comparing nouns in various classesthat the same stems can appear in different classes so for example class 3mu-hoomi lsquocow humprsquo is clearly related to class 4 mu-hoomi lsquocow humpsrsquo ndashsingulars and plurals are marked by changes in class class 11 lu-teefu lsquoreedmatrsquo is clearly related to ka-teefu lsquosmall matrsquo and tu-teefu lsquosmall matsrsquo Theclass prefixes have a number of phonetic manifestations so we find mu-mw- and m- for classes 1 and 3 va v- and w- for class 2 mi- mj- and m- forclass 4 ma- and m- for class 6 ki- and tʃ- for class 7 fi- and fj- for class 8 lu-and lw- for class 11 ka- and k- for class 12 tu- and tw- for class 13 and wu-w- for class 14

623 Phonological alternationsNoun stems fall in two groups in terms of phonological processes thosewhich begin with a consonant and those beginning with a vowelExamples of stems which begin with a consonant are -tesi (cf mu-tesi

fjuuɲu lsquotiny saltsrsquo fiiho lsquotiny eyesrsquofiina lsquotiny holesrsquo fiigiigi lsquotiny shadowsrsquo

Class 11luteefu lsquoreed matrsquo lupava lsquostirring stickrsquolutego lsquotraprsquo ludali lsquopowerluhaanga lsquosandrsquo lwiimbo lsquosongrsquolweendo lsquolovingrsquo lwaaniko lsquodry stufflwiifwi lsquochameleonrsquo

Class 12kateefu lsquosmall matrsquo kakoongo lsquosmall woundrsquokafuniko lsquosmall coverrsquo kangaamba lsquosmall sweet potatorsquokaasi lsquosmall builderrsquo kiimbi lsquosmall singerrsquokaanga lsquosmall graversquo kooto lsquosmall filersquokuula lsquosmall frogrsquo kuunga lsquosmall wet lowlandrsquo

Class 13tuteefu lsquosmall matsrsquo tukoongo lsquosmall woundsrsquotufuniko lsquosmall coversrsquo tungaamba lsquosmall sweet potatoesrsquotwaasi lsquosmall buildersrsquo twiimbi lsquosmall singersrsquotwaanga lsquosmall gravesrsquo tooto lsquosmall filesrsquotuula lsquosmall frogsrsquo tuunga lsquosmall wet lowlandsrsquo

Class 14wuvaso lsquosleeping placersquo wulime lsquocultivatingrsquowugali lsquoporridgersquo wutiitu lsquoblacknessrsquoweelu lsquowhitenessrsquo wuumi lsquolifersquowoogofu lsquofearrsquo wiijooga lsquomushroomrsquo

waangufu lsquospeedrsquo

va-tesi) and -lagusi (cf mu-lagusi va-lagusi) examples of stems which beginwith vowels are -iimbi (cf mw-iimbi v-iimbi) and -eendi (mw-eendi v-eendi)The best phonological information about the nature of the prefix isavailable from its form before a consonant so our working hypothesis isthat the underlying form of the noun prefix is that found before aconsonant it preserves more informationAs we try to understand the phonological changes found with vowel-

initial stems it is helpful to look for a general unity behind these changesOne important generalization about the language judging from the datais that there are no vowel sequences (what may seem to be sequences suchas ii ee are not sequences but are the orthographic representation ofsingle long-vowel segments) Given the assumption that the prefixes forclasses 1 and 2 are respectively mu and va the expected underlyingforms of the words for lsquosingerrsquo and lsquosingersrsquo would be muiimbi and va-iimbi These differ from the surface forms [mw-iimbi] and [v-iimbi] in thecase of mu-iimbi underlying u has become [w] and in the case ofunderlying va-iimbi underlying [a] has been deleted In both cases theend result is that an underlying cluster of vowels has been eliminated

Glide formation versus vowel deletion Now we should ask why is avowel deleted in one case but turned into a glide in another case Theanswer lies in the nature of the prefix vowel The vowel u becomes theglide [w] and the only difference between u and w is that the former issyllabic (a vowel) where the latter is nonsyllabic The low vowel a on theother hand does not have a corresponding glide in this language (or inany language) In other words a rule of glide formation simply could notapply to a and result in a segment of the languageTo make progress in solving the problem we need to advance hypoth-

eses and test them against the data We therefore assume the followingrules of glide formation and vowel deletion

(16) V empty_V a-deletion

By ordering (16) after (15) we can make (16) very general since (15) willhave already eliminated other vowel sequences At this point we cansimply go through the data from top to bottom seeing whether we areable to account for the examples with no further rules ndash or we may findthat other rules become necessaryFor nouns in class 1 the examples mw-iimbi mw-eendi and mw-aasi are

straightforward deriving from mu-iimbi mu-eendi and mu-aasi Theforms m-oogofi m-oofusi and m-uuci presumably derive from mu-oogofiand mu-oofusi and mu-uuci The vowel u has been deleted whichseems to run counter to our hypothesis that high vowels become glides

glide formation[ ]+high

v v[ ]-sylrarr

before vowels It is possible that there is another rule that deletes ubefore a round vowel

We could also consider letting the glide formation rule apply and thenexplain the difference mu-aasi mw-aasi vs mu-oofusi m-oofusi bysubjecting derived mw-oofusi to a rule deleting w before a round vowel

(18) w empty_[+round] w-deletion

Thus we must keep in mind two hypotheses regarding u+o and u+usequences

v-rounding Now consider class 2 In stems beginning with a vowel weeasily explain v-iimbi v-eendi and v-aasi from va-iimbi va-eendi and va-aasiwhere a-deletion applies Something else seems to be happening inw-oogofi w-oofusi and w-uuci from va-oogofi va-oofusi and va-uutsi Applica-tion of a-deletion would yield v-oogofi v-oofusi and v-uutsi which differfrom the surface forms only in the replacement of v by w Since thisprocess takes place before a round vowel we conjecture that there maybe an assimilation rule such as the following

If there is such a rule in the language it would eliminate any sequencesvu vo and the data contain no such sequences There is still a problem toaddress that w-deletion (18) should apply to woogofi but it does not ndash thesurface form is not [oogofi] Two explanations are available One is that v-rounding is ordered after w-deletion so at the stage where w-deletionwould apply this word has the shape voogofi and not woogofi (so w-deletioncannot apply) The other is that (18) needs to be revised so that it onlydeletes a postconsonantal w before a round vowel

Our decision-making criteria are not stringent enough that we can defini-tively choose between these solutions so we will leave this question openfor the time being

Moving to other classes the nouns in class 3 present no problems Glideformation applies to this prefix so mu-iina [mw-iina] and before a

[ ]+ round__rarr Oslash Vu

u-deletion

⎥⎥⎥

⎢⎢⎢

⎡+ labial+ cont+ voice

rarr [ cons] __ [+round] v-rounding

C+round-syl

⎡⎢⎣

⎤⎥⎦

[ ]+roundrarrOslash

round vowel derived w deletes so mu-ooto mw-ooto which thenbecomes [m-ooto]

Front vowels and glides The nouns in class 4 generally conform to thepredictions of our analysis Note in particular that underlying mi-uuɲuand mi-ooto undergo glide formation before a round vowel Suchexamples show that it was correct to state the glide formation rule in amore general way so that all high vowels (and not just u) become glidesbefore any vowel (not just nonround vowels)We cannot yet fully explain what happens with noun stems beginning

with the vowel i as in m-iina m-iigiigi Given mi-iina mi-iigiigi wepredict surface mj-iina mj-iigiigi This is reminiscent of the problem ofmu-oogofi and mu-uuci and we might want to generalize the ruledeleting a glide to include deleting a front glide before a front vowel(analogous to deleting a round glide before a round vowel) What preventsus from doing this is that while w deletes before both u and o y onlydeletes before i and not e as we can see from mj-eenda It might be moreelegant or symmetrical for round glides to delete before round vowels ofany height and front glides to delete before front vowels of any height butthe facts say otherwise a front glide only deletes before a front high vowel

Checking other classes discovering a palatalization rule The class 6prefix ma- presents no surprises at all it appears as ma- before a conson-ant and its vowel deletes before another vowel as in m-iino from ma-iinoThe class 7 prefix on the other hand is more complex Before a consonantit appears as ki- and it also appears as k(i)- before i Before other vowels itappears as t ʃ as in t ʃ-uula t ʃ-aanga t ʃ-ooto and t ʃ-eenda Again we continuethe procedure of comparing the underlying and predicted surface forms(predicted by mechanically applying the rules which we have alreadypostulated to the underlying forms we have committed ourselves to) tosee exactly what governs this discrepancy From underlying ki-uula ki-aanga ki-ooto and ki-eenda we would expect kj-uula kj-aanga kj-ooto and kj-eenda given glide formation The discrepancy lies in the fact that thepredicted sequence kj has been fused into t ʃ a process of palatalizationfound in many languages Since kj is nowhere found in the data we canconfidently posit the following rule

Since ki surfaces as [tʃ] when attached to a vowel-initial noun stem thequestion arises as to what has happened in k-iiho k-iina and k-iigiigi Theglide formation rule should apply to ki-iiho ki-iina and ki-iigiigi

__rarr Oslash j-deletion⎥⎥⎥

⎢⎢⎢

⎡+ high- back- syl

⎡⎢⎣

⎤⎥⎦

+ high- back

rarr Oslash[ ]+cor⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+ cons+ back- voice

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

- syl- cons- back

giving kj-iiho kj-iina and kj-iigiigi which we would expect to undergo (22)But there is a rule deleting j before i If j is deleted by that rule it could notcondition the change of k to tʃ so all that is required is the orderingstatement that j-deletion precedes palatalization (22) Thus ki-iinabecomes kj-iina by glide formation and before the palatalization rule canapply the j-deletion rule (21) deletes the glide that is crucial for (22)

Deciding on the form of w-deletion degemination At this point wecan quickly check the examples in classes 8 11 12 and 13 and verify thatour analysis explains all of these forms as well The final set of examplesare those in class 14 which has the prefix wu This prefix raises aquestion in terms of our analysis why do we have the sequence [wu]which is eliminated by a rule elsewhere One explanation is the statementof the rule itself if (20) is the correct rule then this w could not deletebecause it is not preceded by a consonant The other possibility is that [wu]actually comes from vu by applying v-rounding (19) which we assumedapplies after w-deletion While both explanations work the analysiswhere [wu] is underlying vu has the disadvantage of being ratherabstract in positing an underlying segment in the prefix which neverappears as such This issue was presaged in chapter 3 and is discussed inmore detail in chapter 8 for the moment we will simply say that given achoice between a concrete analysis where the underlying form of a mor-pheme is composed only of segments which actually appear as such insome surface manifestation of the morpheme and an abstract form witha segment that never appears on the surface the concrete analysis ispreferable to the abstract one all other things being comparable On thatbasis we decide that the underlying form of the class 14 prefix is wuwhich means that the proper explanation for failure of w-deletion lies inthe statement of w-deletion itself as (20)Still analyzing this class of nouns we now focus on examples where the

prefix precedes a vowel-initial stem eg w-eelu w-uumi w-oogofu w-iijoogaand w-aangufu from underlying wu-eelu wu-uumi wu-oogofu wu-iijooga and wu-aangufu Applying glide formation would give the sur-face forms ww-eelu ww-uumi ww-oogofu ww-iijooga and ww-aangufuwhich differ from the surface form in a simple way that they have twowrsquos where the actual form has only a single w which allows us to posit thefollowing degemination rule

624 Extending the dataVerbs are subject to these same rules as some additional data will showand an analysis of verbs will provide additional support for aspects of thisanalysis Hehe is a tone language and while we have not been concernedwith accounting for tone (and have not marked tones) in the following

Glide Degemination__rarr⎡⎢⎣

⎤⎥⎦

- syl+ rd

⎡⎢⎣

⎤⎥⎦

- syl- rd

Oslash

data tones are marked and can be predicted by rule In analyzing thesedata we want to account for the placement of the high tone (H) which ismarked with an acute accent

The morphology These data indicate that all verbs begin with kuacute orsomething derivable from kuacute by the rules already motivated thus weassume that kuacute- is an inflectional prefix In addition all verbs end withthe vowel a which is probably a morpheme since it is unlikely thatevery root would end in exactly the same vowel The stem of the wordfor lsquomilkrsquo is probably -kam- Various grammatical relations are expressedby suffixes standing between the stem and the suffix -a such as -il- lsquoforrsquo-an- lsquoeach otherrsquo -j- lsquomakersquo -w- lsquopassiversquo the objects lsquousrsquo and lsquothemrsquo aremarked by the prefixes -tu- and -va- between the prefix kuacute and the verbstem

Phonological rules Looking at the last three roots which are vowel-initial the prefixes kuacute- tu- and va- are subject to the rules motivated onthe basis of nouns where u becomes [w] before a vowel but deletes aftera consonant and before a round vowel (so ku-oogopa kwoogopa [ko ogopa]) the sequence vo becomes wo (ku-va-oogopa kuvoogopa [kuwoogo pa]) The change of v to w is also seen in examples such askuacutetowaacute and kuacutelawaacute coming (apparently) from ku-tov-w-a and ku-lav-w-a

V V for V for each make Vkuacutekama kuacutekamiacutela kuacutekamilaacutena kuacutekamjaacutekuacutesana kuacutesaniacutela kuacutesanilaacutena kuacutesanjaacutekuacutetova kuacutetoveacutela kuacutetovelaacutena kuacutetovjaacutekuacutelava kuacutelaviacutela kuacutelavilaacutena kuacutelavjaacutekuacutefwiiacutema kuacutefwiimiacutela kuacutefwiimilaacutena kuacutefwiimjaacutekuacutekalaaacutenga kuacutekalaangiacutela kuacutekalaangilaacutena kuacutekalaangjaacutekuacutekalaacuteva kuacutekalaviacutela kuacutekalavilaacutena kuacutekalavjaacutekweacuteenda kweacuteendeacutela kweacuteendelaacutena kweacuteendjaacutekwiacuteimba kwiacuteimbiacutela kwiacuteimbilaacutena kwiacuteimbjaacutekoacuteogoacutepa koacuteogopeacutela koacuteogopelaacutena koacuteogopjaacute

be Vrsquod V us V themkuacutekamwaacute kuacutetukaacutema kuacutevakaacutema lsquomilkrsquokuacutesanwaacute kuacutetusaacutena kuacutevasaacutena lsquocombrsquokuacutetowaacute kuacutetutoacuteva kuacutevatoacuteva lsquobeatrsquokuacutelawaacute kuacutetulaacuteva kuacutevalaacuteva lsquolook atrsquokuacutefwiimwaacute kuacutetufwiiacutema kuacutevafwiiacutema lsquohuntrsquokuacutekalaangwaacute kuacutetukalaaacutenga kuacutevakalaaacutenga lsquofryrsquokuacutekalawaacute mdash mdash lsquotake bathrsquokweacuteendwaacute kuacutetweeacutenda kuacuteveeacutenda lsquoloversquokwiacuteimbwaacute kuacutetwiiacutemba kuacuteviiacutemba lsquosingrsquokoacuteogopwaacute kuacutetoogoacutepa kuacutewoogoacutepa lsquofearrsquo

The rule of v-rounding would derive kuacutetowwaacute and kuacutelawwaacute and the actualphonetic forms can be accounted for based on that intermediate form byGlide DegeminationOne additional segmental process of vowel harmony is motivated by the

above examples The benefactive suffix retains its underlying high vowelin forms such as kuacutekam-iacutel-a kuacutesan-iacutel-a and kuacutefwiim-iacutel-a but that vowelassimilates in height to a preceding mid vowel in examples such askuacutetov-eacutel-a kweacuteend-eacutel-a and koacuteogop-eacutel-a This motivates the following vowelharmony rule

Regarding tone most examples have an H tone on the second-to-last vowelof the word (this may be the second part of a long vowel in the penulti-mate syllable or the only vowel of a short penultimate syllable) whichcan be accounted for by the following rule

(26) V [+H]_C0 V tone assignment

In some verbs this H is missing ndash see kuacutekama kuacutesana kuacutetova Applying thistone assignment rule to these forms would result in outputs such askuacutekaacutema kuacutesaacutena kuacutetoacuteva with H tones on adjacent vowels Since ourexamples contain no cases of consecutive H-toned vowels we may assumea rule along the following lines

What about the columns with the suffixes -j- lsquomakersquo and -w- lsquopassiversquowhich have word-final H not penult H We expect kuacutekalaaacutengwa But ifthese two suffixes are underlyingly i and u then the underlying form ofkuacutekalaangwaacute would be kuacutekalaang-u-a H tone would be assigned to thepenultimate vowel under that assumption giving kuacutekalaanguacutea Howeverwe already know that there is a rule of glide formation which would turnu and i into w and y before vowels a rule which has obviously applied inthese forms Since only syllabic elements can bear tones the tone on thepenultimate vowel apparently shifts to the final syllable where it can bepronounced

63 Fore

The next problem comes from Fore spoken in Papua New Guinea

Vowel Harmony

[ ]-highrarr V C0__V

⎡⎢⎣

⎤⎥⎦

-high-low

[ ]+ H[ ]-Hrarr V C0__V

Such tone shiftwhere the tone of avowel shifts toanother vowel whenthe original voweldeletes ordesyllabifies iscommon in tonelanguages and isdiscussed inchapter 9

631 The dataThe following data motivate a set of phonological rules that apply incombinations of noun plus personal possessive affix Your final goal is toidentify the underlying forms of all roots and affixes to discover theoperative phonological rules and order those rules

632 Morphological analysisSeparating roots from suffixes in this language is difficult since it is notobvious whether certain segments are part of the root and delete in onecontext or are part of the suffix and delete in another context ndash or are theyepenthetic Thus the root for lsquoaxersquomight be tun or it might be tu ndash if theformer some rule must delete n in [tuka] lsquoyour sg axersquo if the latter wewould conclude that the 1sg and 3sg suffixes are -nte -nkwa If we assumethe suffixes -nte -nkwa thenwewould need to explain why they appear as[-ne -wa] after lsquoclothesrsquo and lsquoonersquo It almost seems that in order to get theanswer to one question you have to know the answer to all other questionsThe first step to solving this problem is to determine howmany significant

behavioral categories there are By comparing the forms of lsquoclothesrsquo and lsquoeyersquowe can conclude that these two roots are identical in terms of behavior thesuffixes have the same shape after these two roots and the following inflec-tional material is the same across the roots We can also see that there aremajordifferences in the formof thesuffixesbetween lsquoeyersquoand lsquoliverrsquoalthoughthe roots look very similar and in half of the forms are exactly the same

1sg 2sg 3sg 1pltunte tuka tunkwa tute lsquoaxersquokajne kajga kajwa kajre lsquoclothesrsquokaʔne kaka kaʔwa kate lsquoone (thing)rsquoawnte awka awnkwa awte lsquoliverrsquoawne awga awwa awre lsquoeyersquopine piga piwa pire lsquoshellrsquomaʔne maka maʔwa mate lsquosnakersquokone koga kowa kore lsquotraprsquoawʔne awka awʔwa awte lsquoskinrsquointe ika inkwa ite lsquobeersquonante naka nankwa nate lsquohousersquoagene agega agewa agere lsquonamersquokoʔne koka koʔwa kote lsquobagrsquomune muga muwa mure lsquovomitrsquoarawnte arawka arawnkwa arawte lsquokneecaprsquokajnte kajka kajnkwa kajte lsquogingerrsquoabeʔne abeka abeʔwa abete lsquonavelrsquo

1sg 2sg 3sg 1plkajne kajga kajwa kajre lsquoclothesrsquoawne awga awwa awre lsquoeyersquoawnte awka awnkwa awte lsquoliverrsquo

The data of (28) can be reordered by roots according to the surfacepatterns of the apparent personal suffixes and this reveals that thereare three behavioral classes of roots

In (a) the invariance of the portion that precedes ne in the 1sg ga in the2sg wa in the 3sg and re in the 1pl suggests that these roots are kaj awpi ko age mu further leading to the conclusion that the suffixes are -nelsquo1sgrsquo -ga lsquo2sgrsquo -wa lsquo3sgrsquo -re lsquo1plrsquo or some phonologically similar formHaving identified the rootndashsuffix boundary we can now proceed with thephonological analysis of underlying forms and rules

633 Phonological alternationsWe concluded that the (a) subset of roots are underlyingly kaj aw pi koage mu because those are the parts of words that invariantly correlatewith the choice of a particular root A further consequence of that conclu-sion is that the roots in (b) and (c) which behave differently should have asignificantly different-looking underlying form The roots in (30b) havethe surface realizations [kaʔ maʔ awʔ koʔ abeʔ] and [ka ma aw koabe] The roots of (30a) underlyingly end in a glide or vowel and since theroots in (30b) behave differently those roots must not end in a vowel orglide which leads to the conclusion that the roots of (30b) are kaʔ maʔawʔ koʔ abeʔ ie these roots end in a glottal stopSimilar reasoning applied to the roots of (30c) leads to the conclusion

that these roots are tun awn in nan arawn kajn Again the roots havetwo types of surface realization and the alternative theory for (30c) thatthe roots are tu aw i na araw kaj can be ruled out on the grounds that

1sg 2sg 3sg 1plkajne kajga kajwa kajre lsquoclothesrsquoawne awga awwa awre lsquoeyersquopine piga piwa pire lsquoshellrsquokone koga kowa kore lsquotraprsquoagene agega agewa agere lsquonamersquomune muga muwa mure lsquovomitrsquo

kaʔne kaka kaʔwa kate lsquoone (thing)rsquomaʔne maka maʔwa mate lsquosnakersquoawʔne awka awʔwa awte lsquoskinrsquokoʔne koka koʔwa kote lsquobagrsquoabeʔne abeka abeʔwa abete lsquonavelrsquo

tunte tuka tunkwa tute lsquoaxersquoawnte awka awnkwa awte lsquoliverrsquointe ika inkwa ite lsquobeersquonante naka nankwa nate lsquohousersquoarawnte arawka arawnkwa arawte lsquokneecaprsquokajnte kajka kajnkwa kajte lsquogingerrsquo

this would incorrectly render the (a) and (c) roots indistinguishable Thedistinguishing feature of the (c) roots is that they all end with a nasalHaving sorted out the underlying forms of the roots we can turn to

the suffixes drawing one representative from each phonological class ofroots

One fact stands out from this organization of data that while both the1sg and 1pl suffixes have the variant [te] somewhere these suffixescannot be the same because they act quite differently A second factwhich can be seen from these examples is that the 1pl and 2sg suffixesare similar in the nature and context of their variation Both alternatebetween a voiceless stop and a voiced consonant ndash we can suspect that [r]is the surface voiced counterpart of [t] And the voiced alternant appearsafter roots which underlyingly end in a glide or a vowel whereas thevoiceless variant appears after an underlying nasal or a glottal stopNasals and glottal stops have in common the fact of being [-continuant]

and glides and vowels have in common the fact of being [+voice -cons]This gives rise to two theories regarding the underlying forms of the 2sgand 1pl and the rules that apply to those suffixes First we could assumega re and the following rule to derive the voiceless variant

(32) [+cons] [-voice] [-cont] __ Devoicing

Alternatively we could assume ka te and the following voicing rule

Either analysis is at this point entirely reasonable so we must leave thechoice between these analyses unresolved for the moment We mightreject (33) on the grounds that it requires specification of an additionalfeature but such a rejection would be valid only in the context of twocompeting complete analyses which are empirically correct and otherwisethe same in simplicityThe 3sg suffix surfaces as [kwa] and [wa] the former after a nasal and

the latter after an oral segment That leads to two pairs of rule andunderlying representation If the underlying form of the suffix is wathen there is a rule inserting [k] between a nasal and w

1sg 2sg 3sg 1plaw-ne aw-ga aw-wa aw-re lsquoeyersquo

awʔ-ne aw-ka awʔ-wa aw-te lsquoskinrsquo

awn-te aw-ka awn-kwa aw-te lsquoliverrsquo

rarr[ ]+cons [ ]+voice ⎡⎢⎣

⎤⎥⎦

+ voice- nas

Voicing

If the suffix is underlyingly kwa a rule deletes k after an oral segmentbefore w

Finally the 1sg suffix might be ne or it might be te As noted above wecould rule out the possibility te if we knew that the 1pl suffix is te Thismeans that a choice of te for the 1s entails that the 1pl suffix is not tetherefore is re If the 1sg suffix is ne on the other hand the 1pl could beeither te or re If the 1sg suffix is te then the following rule is requiredto derive the variant [ne]

If the suffix is ne then the following rule derives the variant [te]

(37) [+nas] [ndashnas] [+nas] __ Denasalization

Besides three rules which affect the initial consonant of the personalsuffixes a rule deletes root-final glottal stop and nasals In comparingroots with deleted consonants we see that both glottal stop and nasalsdelete in the same context before the 2sg and 1pl suffixes (which we havedetermined are ka te or ga re)

What phonological property unifies these two suffixes and distinguishesthem from ne ~ te and kwa ~ wa A simple answer would be that thesesuffixes begin with voiceless stops ndash if we assume that the suffixes are nelsquo1sgrsquo ka lsquo2sgrsquo wa lsquo3sgrsquo and te lsquo1plrsquo We will pursue the consequencesof that concrete decision about suffixesThe choice of underlying forms for suffixes entails certain choices for

rules in this analysis we are committed to Voicing (33) k-insertion (34)and Denasalization (37) The rule deleting root-final stops is as follows

(39) [-cont] Oslash __ [-voice] Stop Deletion

We must determine how these four rules are ordered Although Voicingaffects underlying voiceless stops after voiced oral segments we see from

1sg 2sg 3sg 1plawʔ-ne aw-ka awʔ-wa aw-te lsquoskinrsquoawn-te aw-ka awn-kwa aw-te lsquoliverrsquo

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+high+consndashvoice

⎡⎢⎣

⎤⎥⎦

+ rdminus syl

k-insertionrarrOslash [ ]+nas

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+ high+ consminus voice

rarr Oslash [ ]+nas⎡⎢⎣

⎤⎥⎦

+ rdminus syl

k-deletion

⎡⎢⎣

⎤⎥⎦

+ corminus voice

rarr [ ]minusnas [ ]+nas Nasalization

[awka] lsquoyour skinrsquo from awʔka and [awka] lsquoyour liverrsquo from awn-ka thatVoicing precedes Stop DeletionThe structural description of the latter rule is not satisfied in awnka

awʔka hence Voicing does not apply Subsequently Stop Deletion appliesto eliminate n and ʔ before a voiceless stop

Stop Deletion obscures the Voicing rule because it creates surface coun-terexamples to the prediction of Voicing that [k t] should not follow avowel or glideThe ordering of k-insertion is also a matter of concern since that rule

inserts a voiceless stop but Stop Deletion is not triggered by inserted kUnderlying awn-wa undergoes k-insertion to become [awnkwa] a formwhich satisfies the structural description of Stop Deletion (which woulddelete the nasal) yet the nasal is not deleted This indicates that k-insertion follows Stop Deletion ndash k created by the former rule is notpresent when Stop Deletion appliesWe can also determine that Denasalization follows Stop Deletion since

the former rule creates a sequence of nasal plus stop ndash awn-ne [awn-te]lsquomy liverrsquo ndash and Stop Deletion applies to a sequence of nasal plus stop ndash

awn-te [awte] lsquoour liverrsquo ndash yet Stop Deletion does not apply to theoutput of Denasalization In summary the rules of Fore which we haveproposed with their ordering are as follows

To be sure that our analysis works derivations of relevant examples aregiven in (40)

(41) a

awʔ-ka underlyingNA Voicing[awka] Stop Deletion

rarr ⎡⎢⎣

⎤⎥⎦

+ voice- nas

⎡⎢⎣

⎤⎥⎦

+ rd- syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+ high+ cons- voice

[ ]+voice[ ]+cons(33)

[ ]-cont

Voicing

Stop Deletion

k-insertion

Denasalization

__rarr

Oslash

[ ]-voice

[ ]-nas

[ ]+nas

[ ]+nas[ ]+nas

aw-ne aw-ka aw-wa aw-te underlyingawga awre Voicing

[awne] [awga] [awwa] [awre]

634 Alternative analysisNow that we have one analysis of the data we need to consider alterna-tives to determine if our analysis is the best one Our basis for evaluatingalternatives will be how they mesh into an integrated system ndash theindividual rules themselves are not significantly different in terms oftheir simplicity In constructing an alternative to be compared with ourhypothesized account we must construct the best analysis that we canOne alternative to consider is that the 3sg suffix is underlyingly kwa

not wa an assumption which would mean a rule of k-deletion ratherthan insertion There is a fundamental incompatibility between this pro-posed underlying form and the theory that there is a stop-voicing ruleapplying to the affixes te ka since deletion of root-final stops appliesin the latter case (awn-ka [aw-ka] lsquoyour liverrsquo) but not the former(awn-kwa [awn-kwa] lsquohisher liverrsquo) Under the theory that there is ak-deletion rule we must assume the underlying suffixes ga re meaningthat there is a devoicing rule and Stop Deletion must be suitably reformu-lated so that only ga re trigger the rule and ne (te) kwa do notThe hypothesized consonants that trigger Stop Deletion would be g r

which can be distinguished from the consonants that do not trigger therule in being [+voice -nasal] The added complication of specifying thatthe triggering consonant is [-nasal] is necessary only under the assump-tion that the 1sg suffix is ne we can avoid that complication by assumingthat the suffix is te in which case the following alternative statement ofstop deletion is necessitated by the alternative assumptions about under-lying forms (te ga kwa re)

(42) [-cont] Oslash __ [+voice] Stop Deletion (alternative version)

Given these alternative underlying forms the variant [ne] of the 1sg suffixfound in [aw-ne] lsquomy eyersquo and [awʔ-ne] lsquomy skinrsquo (but not [awn-te] lsquomyliverrsquo) can be accounted for by the following nasalization rule

(43) [-voice] [+nasal] [-nasal] __ V Nasalization

awʔ-ne awʔ-ka awʔ-wa awʔ-te underlyingNA NA NA NA Voicing

awka awte Stop Deletion[awʔne] [awka] [awʔwa] [awte]

awn-ne awn-ka awn-wa awn-te underlyingNA NA NA NA Voicing

awka awte Stop Deletionawnkwa k-insertion

awnte Denasalization[awnte] [awka] [awnkwa] [awte]

Yet anotherpossibility whichpreserves theunderlying suffixeska te is that k-deletion specificallyrequires a followingw therefore kwabecomes [wa] butka remainsunchanged Such acomplication in thek-deletion rule issufficient to causeus to reject thatanalysis

The reason for specifying that a following vowel is required is so that thesuffix kwa does not undergo the ruleTo summarize the alternative analysis we might instead assume the

suffixes te ga kwa re and the following rules

(32) [+cons] [-voice] [-cont] __ Post-stop Devoicing

(39) [-cont] Oslash __ [+voice] Stop Deletion

There is a fatal flaw in the alternative analysis centering around theinteraction of Devoicing and Stop Deletion The suffixes which conditionStop Deletion are underlyingly [+voice] but that consonant is also subjectto Devoicing ndash by the stop which is deleted If Stop Devoicing applies firstthen awn-ga becomes awnka and Stop Deletion cannot apply since onlyvoiced consonants trigger the rule ndash [awnka] rather than [awka] wouldresult On the other hand if Stop Deletion applies first then awn-ga doesundergo Stop Deletion to become awga but then the consonant needed totrigger Devoicing no longer exists and [awga] results Thus the hypothe-sized rules cannot be ordered in a manner that gives the correct outputmeaning that the rules are wrong On those grounds the alternativeanalysis must be rejected

64 Modern Hebrew

The next case study comes from a set of alternations in the conjugation ofverbs in a certain derivational class in Modern Hebrew

641 The dataThe goal of this problem is to determine the underlying representations ofthe verbal prefix and the stems as well as whatever rules are needed toaccount for these phonological alternations In some cases a related wordis provided in order to clarify aspects of the underlying stem The data tobe accounted for are in (44)

__rarr Oslash k-deletion⎥⎥⎥

⎢⎢⎢

⎡+ high+ cons- voice

⎡⎢⎣

⎤⎥⎦

+ rd- syl[ ]-nas

These data are froma nonstandarddialect that haspharyngeals whichwere deleted (inthe case of ʕ) orchanged to x (in thecase of ħ) in thestandard dialect

1sg 2sg masc 3sg fem Related word

itparnasti itparnes itparnesu lsquoearnrsquo

itparsamti itparsem itparsemu lsquobecome famousrsquo

idbalbalti idbalbel idbalbelu lsquobe confusedrsquo

idgalgalti idgalgel idgalgelu lsquorevolversquo

itħamakti itħamek itħamku lsquoturn awayrsquo

itlabaʃti itlabeʃ itlapʃu lsquoget dressedrsquo

642 Morphological analysisEach of these verbs has a prefix which is either it or id and the prefixtransparently surfaces as one of these two variants in most examples Thefirst-person-singular form is marked with a suffix -ti the third-singularfeminine has the suffix -u and the second-singular masculine has nosuffix The vowel in the second stem syllable is underlyingly the samefor all verbs this fact is not entirely obvious from these data but is madeobvious by a more extensive analysis of the morphological structure ofwords in the language An analysis of the phonological factors surround-ing the second vowel will show that these surface variants can be derivedfrom one particular underlying vowel Derivationally related words suchas the root underlying iʃtaparti lsquoimproversquo and ʃipur lsquoimprovementrsquo have incommon a set of consonants but their vowels differ (vowel changes are ameans of indicating derivational relations in Semitic languages which wewill not be concerned with)

643 Phonological alternationsVoicing assimilation As for the choice between an underlying voiced orvoiceless consonant in the prefix scanning the data reveals that a voicedconsonant appears before voiced obstruents and a voiceless consonantappears before voiceless obstruents and sonorants Since sonorants arephonetically voiced it is clear that there is no natural context for deriving

idbadarti idbader idbadru lsquomake funrsquo

idgaraʃti idgareʃ idgarʃu lsquodivorcersquo

itpalalti itpalel itpalelu lsquoprayrsquo

itxamamti itxamem itxamemu lsquowarmrsquo

itmotati itmotet itmotetu lsquoquakersquo

itʔoʃaʃti itʔoʃeʃ itʔoʃeʃu lsquorecoverrsquo

idbodati idboded idbodedu lsquoseclude oneselfrsquo

istaparti istaper istapru lsquoget a haircutrsquo sapar lsquobarberrsquo

istarakti istarek istarku lsquocomb hairrsquo ma-srek lsquocombrsquo

iʃtaparti iʃtaper iʃtapru lsquoimproversquo ʃipur lsquoimprovementrsquo

itstalamti itstalem itstalmu lsquohave photo takenrsquo tsalem lsquophotographerrsquo

izdakanti izdaken izdaknu lsquoagersquo zaken lsquooldrsquo

izdarasti izdarez izdarzu lsquohurryrsquo zariz lsquoalertrsquo

itamamti itamem itamemu lsquofeign innocencersquo tamim lsquoinnocentrsquo

idardarti idarder idarderu lsquodeclinersquo dirdur lsquorollingrsquo

itpataħti itpateaħ itpatħu lsquodeveloprsquo

idgalaħti idgaleaħ idgalħu lsquoshaversquo

itnatsaħti itnatseaħ itnatsħu lsquoarguersquo

iʃtagati iʃtagea iʃtagʕu lsquobecome madrsquo

itparati itparea itparʕu lsquocause disorderrsquo

itmaleti itmale itmalʔu lsquobecome fullrsquo

itpaleti itpale itpalʔu lsquobecome surprisedrsquo

itnaseti itnase itnasʔu lsquofeel superiorrsquo

the voiceless consonant [t] so we assume that the prefix is underlyingly itBefore a voiced obstruent a voiceless obstruent becomes voiced

Alternations in V2 The second vowel of the stem has three phoneticvariants [a] as in itparnasti [e] as in itparnes and Oslash as in idbadru(cf idbader) Deletion of the second stem vowel only takes place beforethe suffix -u so we will first attempt to decide when the vowel is deletedA partial specification of the context for vowel deletion is before C+Vwhich explains why the first- and second-person-singular masculineforms (with the suffixes -it and -Oslash) do not undergo vowel deletion Thenext step in determining when a vowel is deleted is to sort the examplesinto two groups those with vowel deletion and those with no voweldeletion In the following examples the site of vowel deletion (or its lack)is marked with an underscore

Based on this grouping we discover a vowel is deleted when it is precededby just a single consonant if two consonants precede the vowel there isno deletionHowever it is not always the case that a vowel deletes after a single

consonant so our rule cannot simply look for one versus two consonantsThere are cases such as itʔoʃeʃu where there is no vowel deletion despitethe fact that there is only a single consonant before the vowel Inspectingall of those examples we discover that the consonants preceding andfollowing the vowel are the same and in every case where a vowel isdeleted the preceding and following consonants are different Thus avowel deletes only if it is preceded by a single consonant and that conson-ant must be different from the consonant that follows the vowel (which isindicated informally as ldquoCi Cjrdquo in the rule)

(47) e Oslash V Ci _ Cj V

[ ]+voice[ ]-sonorant rarr

⎡⎢⎣

⎤⎥⎦

-son+voi

Vowel deletionitħam_ku itlap_ʃu idbad_ruidgar_ʃu istap_ru istar_kuiʃtap_ru itstal_mu izdak_nuizdar_zu itmal_ʔu itpal_ʔuitnas_ʔu itpat_ħu idgal_ħuitnats_ħu iʃtag_ʕu itpar_ʕu

No vowel deletionitparnesu itparsemu idbalbeluidgalgelu idarderu itpaleluitxamemu itmotetu itʔoʃeʃuidbodedu itamemu

At this point we now clearly recognize this process as a kind of syncope aphonological rule which we have encountered many times before

Closed syllable lowering Now we turn to the alternation between [a]and [e] Concentrating on the first set of examples in the data set we find[a] before CC (itparnasti) and [e] before C or CV (itparnes itparnesu)Assuming that this distribution is generally valid we would thereforeposit the following rule to derive [a] from e

(48) e a _ CC

An attempt to derive [e] from underlying a runs into the difficulty thatthe context ldquowhen followed by C or CVrdquo is not a coherent context but isjust a set of two partially related contexts This motivates the decision toselect underlying eIn four examples the second stem vowel e appears as [a] before a single

consonant namely the first-person-singular forms itmotati idbodati iʃtagatiand itparati These examples fall into two distinct subgroups as shown bylooking at their underlying stems which is revealed in the third-singularfeminine forms (itmotet-u idboded-u and iʃtagʕu itparʕu) In the first twoexamples the stems underlyingly end in a coronal stop t or d and in thesecond two examples the stems underlyingly end in the voiced pharyngealʕ At the underlying level the second stem vowel is followed by twoconsonants (itmotetti itbodedti iʃtageʕti and itpareʕti) Surface [a] isexplained on the basis of the underlying consonant cluster ndash it mustsimply be assured that the rules simplifying these clusters apply after (48)In the first two examples (itmotati and idbodati from itmotat-ti and

idbodad-ti) combination of the first-singular suffix with the root would(after assimilation of voicing) be expected to result in itmotatti and idbo-datti In fact the data provide no examples of geminate consonants andwhere geminates might have been created by vowel syncope in idbodedusyncope is blocked Thus the language seems to be pursuing a strategy ofavoiding the creation of geminate consonants We can account for thissimplification of consonant clusters by the following rule

(49) CiCi Ci

This rule also explains itamem and idarder where the stem begins with tor d The underlying forms would be it-tamem and it-darder thesurface form with a single consonant reflects the application of thisconsonant-degemination process

Stems with final pharyngeals and laryngeals The vowel quality ofʃageʕ and pareʕ will be left aside temporarily We thus turn to the stemsrepresented in itpataħti idgalaħti and itnatsaħti What is problematic aboutthese stems is the appearance of [ea] when no suffix is added viz itpateaħidgaleaħ and itnatsaħ Assuming the underlying forms to be itpataħidgalaħ and itnatseħ (selecting e as the second vowel analogous to itparnes

itlabeʃ and idboded) we would need a rule inserting the vowel [a] Thesestems have in common that their final consonant is the pharyngeal [ħ]suggesting a rule along the following lines

(50) Oslash a e _ ħ

Why does this rule only apply in the suffixless second-singular masculineformWhen the stem is followed by -u (itpateħu [itpatħu]) the vowel e isdeleted by the syncope rule so there is no vowel before ħ Syncope doesnot apply before the suffix -ti in itpateħti [itpataħti] but there is still noepenthetic vowel The reason is that underlying e changes to [a] by rule(48) before a cluster of consonants Since that rule changes e to [a] but(50) applies after e prior application of (50) deprives vowel insertion of achance to applyNow returning to the stems ʃageʕ and pareʕ we can see that this same

process of vowel insertion applies in these stems in the second-singularmasculine Starting from iʃtageʕ and itpareʕ vowel epenthesis obviouslyapplies to give intermediate iʃtageaʕ and itpareaʕ This argues that theepenthesis rule should be generalized so that both of the pharyngealconsonants trigger the process

The forms derived by (51) are close to the actual forms which lack theconsonant ʕ and with an appropriate consonant deletion rule we canfinish the derivation of these forms To formalize this rule we need todetermine where the consonant ʕ appears in the language our dataindicate that it appears only before a vowel never before a consonant orat the end of a word (which is to say it never appears at the end of asyllable) Knowing this generalization we posit the following rule

(52) ʕ Oslash_ C (frac14) ʕ Oslash_

No further rules are needed to account for this set of examples In iʃtagatiand itparati from iʃtageʕti and itpareʕti there is no epenthetic vowel This ispredicted by our analysis since these verbs must undergo the rulelowering e to [a] before CC and as we have just argued vowel loweringprecedes vowel epenthesis (thus preventing epenthesis from applying) Inthis respect iʃtagati and itparati are parallel to itpateah idgaleaħ anditnatseaħ The nonparallelism derives from the fact that syllable-final ʕ isdeleted so predicted iʃtagaʕti and itparaʕti are realized as iʃtagati anditparati thanks to this deletionThe final set of verb stems typified by the verb itmaleti ~ itmale ~ itmalʔu

exhibit a glottal stop in some contexts and Oslash in other contexts The twomost obvious hypotheses regarding underlying form are that the stem ismale or else maleʔ It is difficult to decide between these possibilities so

[+low][+low]rarrOslash V e_ C

we will explore both Suppose first that these stems end in glottal stop Inthat case we need a rule deleting glottal stop syllable-finally ndash a similarrule was required to delete the consonant ʔ A crucial difference betweenstems ending in ʔ and stems presumably ending in ʔ is that the stem vowele does not lower to [a] before -ti in the latter set Thus deletion of ʔ wouldhave to be governed by a different rule than deletion of ʔ since ʔ-deletionprecedes lowering and ʔ-deletion follows lowering

An alternative possibility that we want to consider is that these stemsreally end in a vowel not a glottal stop Assuming this surface [itpaleti]would simply reflect concatenation of the stem pale with the suffix andno phonological rule would apply The problem is that we would also needto explain why the rule of syncope does not apply to [itpaleti] since thephonetic context for that rule is found here The glottal-final hypothesiscan explain failure of syncope rather easily by ordering glottal stopdeletion after syncope ndash when syncope applies the form is itpaleʔtiwhere the consonant cluster blocks syncope

Metathesis The last point regarding the Hebrew data is the position of tin the prefix The consonant of the prefix actually appears after the firstconsonant of the stem in the following examples

We would have expected forms such as [itsaparti] [itʃaparti] [ittsalamti]by just prefixing it- to the stem A metathesis rule is therefore neededwhich moves t after the stem-initial consonant What makes this group ofconsonants ndash [s ʃ ts z] ndash a natural class is that they are all and the onlystrident coronals We can thus formalize this rule as follows a coronalstop followed by a coronal strident switch order

The ordering of this metathesis rule with respect to the voicing assimila-tion rule is crucial Given underlying it-zakanti you might attempt toapply metathesis first which would yield iztakanti where voiceless t isplaced after stem-initial z The voicing assimilation rule (in a general formapplying between all obstruents) might apply to yield istakanti So ifmetathesis applies before voicing assimilation we will derive an incorrectresult either iztakanti if there is no voicing assimilation (assuming thatthe rule only turns voiceless consonants into voiced ones) or istakanti ifthere is voicing assimilation However we will derive the correct output ifwe apply voicing assimilation first itzakanti becomes idzakanti whichsurfaces as [izdakanti] by metathesis With this ordering we have com-pleted our analysis of Modern Hebrew phonology

istaparti lsquoget a haircutrsquo istarakti lsquocomb hairrsquoiʃtaparti lsquoimproversquo itstalamti lsquohave photo takenrsquoizdakanti lsquoagersquo izdarasti lsquohurryrsquo

⎡⎢⎣

⎤⎥⎦

+cor-cont

⎡⎢⎣

⎤⎥⎦

+cor+strid

⎡⎢⎣

⎤⎥⎦

+cor-cont

⎡⎢⎣

⎤⎥⎦

+cor+strid

65 Japanese

The analysis of phonological alternations found in connection with theconjugation of verbs in Japanese provides our final illustration of thekinds of issues that must be considered in coming up with appropriaterules and underlying representations In solving this problem it is par-ticularly important to make the correct assumptions about underlyingrepresentations since the selection of underlying forms goes hand inhand with stating the rules correctly

651 The dataThe relevant data are given in (55)

652 Morphological analysisWe could make an initial guess regarding suffixes which leads to thefollowing hypotheses -u ldquopresentrdquo -nai ldquonegativerdquo -tai ldquovolitionalrdquo -taldquopastrdquo and -jo ldquoinchoativerdquo that analysis seems reasonable given the firsttwo verbs in the data We might also surmise that the root is whatever thepresent-tense form is without the present ending ie underlying ner mirʃin jom job kats kas wak tsug kar and ka In lieu of the application of aphonological rule the surface form of a word should simply be whateverwe hypothesize the underlying form of the root to be plus the underlyingform of added affixes Therefore given our preliminary theory of rootsand suffixes in Japanese we predict the following surface forms withhyphens inserted between morphemes to make the division of words intoroots and suffixes clear it is important to understand the literal predic-tions of your analysis and to compare them with the observed facts

Present Negative Volitional Past Inchoativeneru nenai netai neta nejo lsquosleeprsquomiru minai mitai mita mijo lsquoseersquoʃinu ʃinanai ʃinitai ʃinda ʃino lsquodiersquojomu jomanai jomitai jonda jomo lsquoreadrsquojobu jobanai jobitai jonda jobo lsquocallrsquokatsu katanai katʃitai katta kato lsquowinrsquokasu kasanai kaʃitai kaʃita kaso lsquolendrsquowaku wakanai wakitai waita wako lsquoboilrsquotsugu tsuganai tsugitai tsuida tsugo lsquopourrsquokaru karanai karitai katta karo lsquoshearrsquokau kawanai kaitai katta kao lsquobuyrsquo

Predicted surface formsPresent Negative Volitional Past Inchoativener-u ner-nai ner-tai ner-ta ner-jomir-u mir-nai mir-tai mir-ta mir-joʃin-u ʃin-nai ʃin-tai ʃin-ta ʃin-jo

The forms which are correct as is are underlined as we can see all ofthe present-tense forms are correct and none of the others is It is nosurprise that the present-tense forms would be correct since we decidedthat the underlying form of the root is whatever we find in the presenttense minus the vowel -u It is possible but unlikely that every other wordundergoes some phonological rule

Changing our hypothesis Since our first guess about underlying formsis highly suspect we should consider alternative hypotheses Quite oftenthe cause of analytic problems is incorrect underlying forms One place toconsider revising the assumptions about underlying representationswouldbe those of the affixes It was assumed ndash largely on the basis of the first twoforms nenai and minai ndash that the negative suffix is underlyingly -nai How-ever inmost of the examples this apparent suffix is preceded by the vowela (ʃinanai jomanai jobanai and so on) which suggests the alternative possi-bility that the negative suffix is really -anai Similarly the decision that thevolitional suffix is underlyingly -tai was justified based on the fact that itappears as -tai in the first two examples however the suffix is otherwisealways preceded by the vowel i (ʃinitai jomitai jobitai and so on) so thisvowel might analogously be part of the suffixOne fact strongly suggests that the initial hypothesis about the under-

lying forms of suffixes was incorrect The past-tense suffix which we alsoassumed to be -ta behaves very differently from the volitional suffix andthus we have ʃinitai versus ʃinda jomitai versus yonda kat ʃitai versus kattakaritai versus katta (there are similarities such as kaʃitai and kaʃita whichmust also be accounted for) It is quite unlikely that we can account forthese very different phonological patterns by reasonable phonologicalrules if we assume that the volitional and past-tense suffixes differ solelyby the presence of final iIt is this realization that there is a thorough divergence between the

past-tense and volitional suffixes in terms of how they act phonologicallythat provides the key to identifying the right underlying forms Givenhow similar these two suffixes are in surface forms -(i)tai vs -(i)ta but howdifferently they behave phonologically they must have quite differentunderlying forms Since the past-tense suffix rarely has a vowel and thevolitional suffix usually does we modify our hypothesis so that thevolitional is -itai and the past tense is ta Because the negative acts verymuch like the volitional in terms of where it has a vowel we also adoptthe alternative that the negative is anai

jom-u jom-nai jom-tai jom-ta jom-jojob-u job-nai job-tai job-ta job-jokats-u kats-nai kats-tai kats-ta kats-jokas-u kas-nai kas-tai kas-ta kas-jowak-u wak-nai wak-tai wak-ta wak-jotsug-u tsug-nai tsug-tai tsug-ta tsug-jokar-u kar-nai kar-tai kar-ta kar-joka-u ka-nai ka-tai ka-ta ka-jo

These changed assumptions about underlying representations ofsuffixes yield a significant improvement in the accuracy of ourpredicted surface forms as indicated in (57) with correct surface formsunderlined

Implicitly we know that forms such as predicted [katsanai] (for [katanai])and [kas-itai] (for [kaʃitai]) must be explained either with other changes inunderlying forms or by hypothesizing rulesWe will consider one further significant modification of the underlying

representations inspired by the success that resulted from changing ourassumptions about -itai and -anai in reducing the degree to which under-lying and surface forms differ The original and dubious decision to treatthese suffixes as tai and nai was influenced by the fact that that is howthey appear with the first two verbs It is also possible that our initialhypothesis about the underlying form of these two verb roots was incor-rect There is good reason to believe that those assumptions were indeedalso incorrect Compare the surface form of the three verbs in our data setwhich by hypothesis have roots ending in r

Clearly the supposed roots ner and mir act quite differently fromkar The consonant r surfaces in most of the surface forms of the verbmeaning lsquoshearrsquo whereas r only appears in verbs lsquosleeprsquo and lsquoseersquo in thepresent tense In other words there is little reason to believe that thefirst two roots are really ner and mir rather than ne and mi incontrast there seems to be a much stronger basis for saying that theword for lsquoshearrsquo is underlyingly kar Now suppose we change ourassumption about these two verbs and assume that ne and mi endin vowels

Modified predicted surface formsPresent Negative Volitional Past Inchoativener-u ner-anai ner-itai ner-ta ner-jomir-u mir-anai mir-itai mir-ta mir-joʃin-u ʃin-anai ʃin-itai ʃin-ta ʃin-jojom-u jom-anai jom-itai jom-ta jom-jojob-u job-anai job-itai job-ta job-jokats-u kats-anai kats-itai kats-ta kats-jokas-u kas-anai kas-itai kas-ta kas-jowak-u wak-anai wak-itai wak-ta wak-jotsug-u tsug-anai tsug-itai tsug-ta tsug-jokar-u kar-anai kar-itai kar-ta kar-joka-u ka-anai ka-itai ka-ta ka-jo

Present Negative Volitional Past Inchoativener-u ne-nai ne-tai ne-ta ne-jo lsquosleeprsquomir-u mi-nai mi-tai mi-ta mi-jo lsquoseersquokar-u kar-anai kar-itai katt-a kar-o lsquoshearrsquo

In terms of being able to predict the surface forms of verbs withoutphonological rules this has resulted in a slight improvement of predictivepower (sometimes involving a shuffling of correct and incorrect columnswhere under the current hypothesis we no longer directly predict theform of the present tense but we now can generate the past and incho-ative forms without requiring any further rules) More important is thefact that we now have a principled basis in terms of different types ofunderlying forms for predicting the different behavior of the verbs whichhave the present tense neru miru versus karu which are in the first twocases actually vowel-final roots in contrast to a consonant-final root

653 Phonological rulesSince we have made reasonable progress in solving the problem of under-lying forms we will attempt to discover phonological rules which explainremaining differences between underlying and surface forms ndash though italways remains possible that we will need to change our assumed under-lying forms as our analysis progresses The approach to take is to look atforms which are still not completely explained and construct hypothesesto account for these forms what new rules are needed to get from theunderlying to surface forms One useful way to approach this is to lookfor columns or rows of data where similar things seem to be happeningThe incorrectly predicted forms are re-listed below this time excludingthe forms which are already explained with information about the natureof the problem added If a segment is predicted but does not actuallysurface that segment is placed in parentheses if there is a segment whichappears in the surface form but which does not appear to be present in theunderlying form the segment is placed in square brackets segmentswhose phonetic quality differs from the predicted quality are italicized

Modified predicted surface formsPresent Negative Volitional Past Inchoativene-u ne-anai ne-itai ne-ta ne-jomi-u mi-anai mi-itai mi-ta mi-joʃin-u ʃin-anai ʃin-itai ʃin-ta ʃin-jojom-u jom-anai jom-itai jom-ta jom-jojob-u job-anai job-itai job-ta job-jokats-u kats-anai kats-itai kats-ta kats-jokas-u kas-anai kas-itai kas-ta kas-jowak-u wak-anai wak-itai wak-ta wak-jotsug-u tsug-anai tsug-itai tsug-ta tsug-jokar-u kar-anai kar-itai kar-ta kar-joka-u ka-anai ka-itai ka-ta ka-jo

Present Negative Volitional Past Inchoativene[r]u ne(a)nai ne(i)tai lsquosleeprsquomi[r]u mi(a)nai mi(i)tai lsquoseersquo

ʃinta ʃin(j)o lsquodiersquo

The glide in the inchoative In order to explain most of the problemswhich arise with the inchoative form we will consider the possibility thatthere is a rule deleting consonants after consonants since that is thenature of the problem with the inchoative column Such a consonantdeletion cannot be totally general ie deleting any consonant after anyother consonant since as is evident in the past tense column the conson-ant clusters [tt] and [nd] are possible in the language Nevertheless thesetwo clusters are a rather restricted subset of the imaginable two-consonant combinations which can be formed from the consonants ofthe language and this is a good indication that there may be some processdeleting a consonant after another consonant Thus we might assume arule deleting the glide j after a consonant

The postulation of any such rule immediately makes a prediction aboutpossible surface forms there should be no sequences of consonant plusglide in the data Since there are none in the data at hand our hypothesishas passed an important test Armed with this rule we have accountedfor a very large chunk of otherwise problematic examples in (60) ndash all ofthe inchoative forms except for kao lsquobuyrsquo where the glide deletes butthere seems to be no consonant which would condition deletion of theglide

Vowel deletion Another area where some success is possible in recon-ciling underlying and surface forms by focusing on possible segmentsequences is with the verbs lsquosleeprsquo and lsquoseersquo The difference between thepredicted (neanai mianai neitai miitai) and actual forms (nenai minainetai mitai) of the negative and volitional forms is that the actual formslack the suffix vowel In the predicted forms we find a sequence ofvowels whereas in the actual form only the first of those vowels isfound This raises the question whether we might postulate a rule delet-ing a vowel after another vowel In positing such a rule we want toconsider what V-V sequences are found in the data The sequence [ai]exists in the volitional and negative suffixes and in past-tense waita also[ui] in the past of the word for lsquopourrsquo also the sequences [ao] and [au] inthe verb lsquobuyrsquo We do not find sequences of vowels with the front vowels

jomta jom(j)o lsquoreadrsquojobta job(j)o lsquocallrsquo

katsanai katsta kats(j)o lsquowinrsquokasitai kas[i]ta kas(j)o lsquolendrsquo

wakta wak(j)o lsquoboilrsquotsugta tsug(j)o lsquopourrsquokarta kar(j)o lsquoshearrsquo

ka(w)anai ka[t]ta ka(j)o lsquobuyrsquo

⎡⎢⎣

⎤⎥⎦

minusconsminusback

rarr Oslash C_

If you knowJapanese you mayknow of words withj after a consonanteg [Tokjo] whichcontradict theproposed rule Werestrict ourselves tothe specific data setgiven here but arestriction on therule that the deletedconsonant must besuffix-initial solvesthis problem

[e] or [i] plus a vowel ([ia] [ii] [ea] and [ei]) Therefore we posit thefollowing rule of vowel deletion

This resolves many problematic forms of the verbs lsquosleeprsquo and lsquodiersquo suchas the change ne-itai [netai] but there are still examples that wecannot explain In the present tense we find [neru] and [miru] whichwe presume derive from ne-u and mi-u The vowel deletion rule (62)should apply to these underlying forms resulting in incorrect [ne] and[mi] We might try to resolve this by assuming that the vowel [u] cannotbe deleted by (62) ndash we would then need to restrict the rule to excluderound vowels from deletion Alternatively u fails to be deleted in ne-uperhaps a consonant is inserted thereby eliminating the cluster of vowels

Armed with these new rules we will have actually accounted for all formsof the verbs lsquosleeprsquo and lsquoseersquo

Nasal + consonant The remaining problems have been reduced to avery small set A comparison of presumed underlying and surface pastforms is given below

The problem posed by the past-tense form is that by combining the rootwith the suffix -ta underlying clusters of consonants would be createdbut there are very severe restrictions on what consonant clusters exist inJapanese The simplest problem is that presented by [ʃinda] from sintawhere t becomes voiced after a nasal A process of postnasal voicing israther common in the languages of the world so we may hypothesize thatthere is such a process in Japanese

(65) C [+voice] [+nasal] _

The data further suggest that the rule applies in other examples sincewe see that in the past tense [jonda] of the roots jom and job thefinal consonant of the root is a nasal on the surface and t becomesvoiced

__rarr Oslash V+syl-back

⎡⎢⎣

⎤⎥⎦

We will consideranother possibilitylater that thepresent suffix isru so rather thaninserting it inneru we delete itin [jomu]

Oslashrarr r ⎤⎥⎦

⎤⎥⎦

+syl-back

ʃinta [ʃinda] jomta [jonda]jobta [jonda] katsta [katta]kasta [kaʃita] wakta [waita]tsugta [tsuida] karta [katta]kata [katta]

We account for the stems job and jom by noting that the finalconsonant in these roots becomes [n] which is part of the change fromthe nonexistent sequences mt and bt to the actually occurring [nd]Thus these consonants become [n] before t (and subsequently t voicesafter the derived [n])

Although the data only illustrate nasalization before t (66) is stated asgenerally as possible predicting that k or d would nasalize as wellWatching for contexts where a phenomenon seems to be relevant to

more than one form we also notice that the surface forms [waita] and[tsuida] differ from their underlying forms wakta and tsugta byreplacing the preconsonantal velar with the vowel [i] suggesting a vocal-ization rule such as the following

This rule accounts for [waita] and almost accounts for [tsuida] but we stillneed to explain why the suffix consonant is voiced The underlying repre-sentation itself provides a reason for this voicing since underlyingly t ispreceded by a voiced consonant in tsugta We know that t voices inanother context after a nasal so we could account for voicing in [tsuida]by restating the rule so that it applies not just after nasals (which arevoiced) but after all voiced consonants By applying the voicing rulewhich is sensitive to underlying consonant voicing before the velar vocal-ization rule we can explain the opaque surface difference [waita] versus[tsuida] as deriving from the voicing of the consonant which precedes itunderlyingly We also want to be sure to apply rule (67) before rule (66)given the way we have formulated these rules We did not explicitlyrestrict (66) which changes noncoronals to [n] before a consonant toapplying only to labials Therefore the more specific rule (67) must applyfirst otherwise velars would also be incorrectly turned into [n] before aconsonant

654 Taking stockWe should review the analysis to be sure there are no loose ends We havesix rules ndash j-deletion vowel deletion r-insertion consonant voicing velarvocalization and labial nasalization ndash which given our assumptionsregarding roots and suffixes account for most of the forms in the dataset It is important to recheck the full data set against our rules to becertain that our analysis does handle all of the data A few forms remainwhich we cannot fully explain

[ ]- coronal __C+ coronal+ nasal

rarr⎡⎢⎣

⎤⎥⎦

+high[ [

-back[ [

The forms which we have not yet explained are the following Firstwe have not explained the variation in the root-final consonant seen inthe verb meaning lsquowinrsquo (kats-u kat-anai-anai kat ʃ-itai kat-ta kat-o)Second we have not accounted for the variation between s and ʃ in theverb lsquoshearrsquo nor have we explained the presence of the vowel [i] in thepast tense of this verb Finally in the verb lsquobuyrsquo we have not explainedthe presence of [w] in the negative the appearance of a second [t] in thepast-tense form and why in the inchoative form [kao] the suffix con-sonant j deletes

Correcting the final consonant The first problem to tackle is thevariation in the final consonant of the verb lsquowinrsquo Looking at the correl-ation between the phonetic realization of the consonant and the followingsegment we see that [ts] appears before [u] [tʃ] appears before [i] and [t]appears elsewhere It was a mistake to assume that the underlying formof this root contains the consonant ts instead we will assume that theunderlying consonant is t (so nothing more needs to be said about thesurface forms kat-anai kat-ta and kat-o) Looking more generally atthe distribution of [tʃ] and [ts] in the data [t ʃ] only appears before [i] and[ts] only appears before [u] allowing us to posit the following rules

(68) t [+delrel] _ u

Moving to the word for lsquolendrsquowe find a related problem that s appears as[ʃ] before [i] This is reminiscent of the process which we assumed turning tinto t ʃ before i In fact we can decompose the process t t ʃ into two morebasic steps t becomes an affricate before [i] and s and ts become alveo-palatal [ʃ] and [t ʃ] before the vowel [i]

i-epenthesis All that remains to be explained about the word for lsquolendrsquois why [i] appears in the past tense ie why does kasta become kasita(whence [kaʃita]) This is simple we see that [st] does not exist in thelanguage and no assimilations turn it into an existing cluster so [i] isinserted to separate these two consonants

r-assimilation and final w Turning now to the form [katta] lsquoshear (past)rsquofrom kar-ta a simple assimilation is needed to explain this form

(71) r Ci _ Ci

rarrt __ i+delrel-ant

⎡⎢⎣

⎤⎥⎦

rarrOslash __⎥⎥⎥

⎢⎢⎢

⎡+ syl- high- back

+ cont- son

⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+ cor- cont

The last remaining problems are in the verb lsquobuyrsquo where we must explainthe extra [t] in [katta] the presence of [w] in [kawanai] and the loss of jin the inchoative form [kao] We might explain the form [kawanai] by arule of w-insertion inserting w between two occurrences of the vowel [a]more puzzling is the form [katta] whichwe presume derives from ka-ta Itwould be very unusual for a consonant to spontaneously double betweenvowels Since there are so many problems associated with this one rootperhaps the problem lies in our assumptions about the underlying form ofthis root Perhaps thew in [kawanai] is part of the root itselfWhatwould bethe benefit of assuming that this root is really kaw First it explains thepresence of w in [kawanai] Second it provides a basis for the extra [t] in[katta] w assimilates to following [t] Such an assimilation is implicit inour analysis namely rule (71) assimilating r to t We can generalize thisrule to applying to both r and w which are oral sonorants Finallypositing underlying kaw helps to resolve the mystery of why j deletes inthe inchoative form [kao] when otherwise j only deletes when it ispreceded by a consonant If we start with ka-jo there is no reason for jto delete but if we start with kaw-jo j is underlyingly preceded by aconsonant w which causes deletion of j and then w itself is deletedThe cost of this analysis ndash a small cost ndash is that we must explain why [w]

does not appear more widely in the root specifically why we do not findsurface [w] in ka-u ka-itai and ka-o The answer lies in the context where [w]appears [w] only appears before a low vowel suggesting the following rule

At this point we have a complete analysis of the data The rules (inshorthand versions) and underlying forms are recapitulated below

(73) Roots ne lsquosleeprsquo mi lsquoseersquo ʃin lsquodiersquo jom lsquoreadrsquo job lsquocallrsquo katlsquowinrsquo kas lsquolendrsquo wak lsquoboilrsquo tsug lsquopourrsquo kar lsquoshearrsquo kaw lsquobuyrsquo

Suffixes -u lsquopresentrsquo -anai lsquonegativersquo -itai lsquovolitionalrsquo -ta lsquopastrsquo -jolsquoinchoativersquo

Progress by hypothesis forming and testing Three important pointshave emerged as our analysis developed First analysis proceeds step-by-step by forming specific hypotheses which we then check against the

Rulesj Oslash C _ V e i _Oslash r e i _V [-round]b m n _t k g i _tt ts _u i ts s tʃ ʃ _iOslash i s _t r w t _tw Oslash _V t d C _[-lo] [+voi]

w rarr Oslash __⎡⎢⎣

⎤⎥⎦

+syl-low

data revising those hypotheses should they prove to be wrong Second itis vital to consider more than one hypothesis if we had only pursued thefirst hypothesis that the roots ne mi kar and kaw were reallyunderlying ner mir kar and ka we would never have been able tomake sense of the data The most important skill that you can bring to thetask of problem solving is the ability to create and evaluate competinghypotheses intended to explain some fact Finally it is particularlyimportant to remember that assumptions about underlying representa-tions go hand-in-hand with the phonological rules which you postulate fora language When you check your solution the problem may not be thatyour rules are wrong but that your underlying forms are wrong Bycontinuously reviewing the analysis and making sure that the rules workand your assumptions about underlying forms are consistent you shouldarrive at the stage that no further improvements to the analysis arepossible given the data available to youIt might occur to you that there are aspects of the underlying represen-

tation which could still be questioned Consider the present-tense formwhich we assumed was u An alternative may be considered the suffixmight be ru The presence of underlying r in this suffix is made plausibleby the fact that r actually appears in the formsmiru neru We assumed thatr is epenthetic but perhaps it is part of the present suffix That would allowus to eliminate the rule of r-epenthesis which is needed only to account for[neru] and [miru] At the same time we can also simplify the rule of voweldeletion by removing the restriction that only nonround vowels deleteafter [e] and [i] we made that assumption only because ne-u and mi-uapparently did not undergo the process of vowel deletionAny change in assumed underlying forms requires a reconsideration of

those parts of the analysis relevant to that morpheme We would thenassume the underlying forms ʃin-ru jom-ru kat-ru and so on with theroot-final consonant being followed by r This r must be deleted butnotice that we already have a rule which stated in a more general formwould delete this r namely the rule deleting j after a consonant

(74) [+sonor] Oslash C _

If we generalize that rule to apply to any sonorant consonant after aconsonant we eliminate the rule of r-insertion and generalize the rulesj-deletion and vowel deletion which results in a better analysis

Summary Analyzing a complex set of data into a consistent system of underlyingrepresentations and rules requires you to pay attention to detailsA solution to a problem requires that you formulate reasoned hypoth-eses and test them against the data The most important skill needed totest a hypothesis is that you must apply your rules completely literallyDo what the rule says must be done and if that does not give you thecorrect result you must change your underlying representationsrules or rule ordering The ability to conceive of and evaluate multiplehypotheses is one of the most important skills in problem solving

Exercises1 Serbo-CroatianThese data from Serbo-Croatian have been simplified in two ways tomake the problem more manageable Vowel length is omitted and someaccents or stresses are omitted The language has both underlyingstresses whose position cannot be predicted ndash these are not marked inthe transcriptions ndash and a predictable ldquomobilerdquo stress which is assigned byrule ndash these are the stresses indicated here Your analysis should accountfor how stress is assigned in those words marked with a rule-governedstress you should not try to write a rule that predicts whether a word hasa stress assigned by rule versus an underlying stress Ignore the stressof words with no stress mark (other parts of the phonology of suchwords must be accounted for) Past-tense verbs all have the same generalpast-tense suffix and the difference between masculine feminine andneuter past-tense involves the same suffixes as are used to mark genderin adjectives

AdjectivesMasc Fem Neut Plmlad mlada mlado mladı lsquoyoungrsquotup tupa tupo tupı lsquobluntrsquoblag blaga blago blagı lsquomildrsquogrub gruba grubo grubı lsquocoarsersquobeo bela belo belı lsquowhitersquoveseo vesela veselo veseli lsquogayrsquodebeo debela debelo debelı lsquofatrsquomıo mila milo milı lsquodearrsquozelen zelena zeleno zelenı lsquogreenrsquokraden kradena kradeno kradenı lsquostolenrsquodalek daleka daleko dalekı lsquofarrsquovisok visoka visoko visokı lsquohighrsquodubok duboka duboko dubokı lsquodeeprsquokriʒan kriʒana kriʒano kriʒani lsquocrossrsquosuntʃan suntʃana suntʃano suntʃani lsquosunnyrsquosvetʃan svetʃana svetʃano svetʃani lsquoformalrsquobogat bogata bogato bogati lsquorichrsquorapav rapava rapavo rapavi lsquoroughrsquojasan jasna jasno yasnı lsquoclearrsquovaʒan vaʒna vaʒno vaʒnı lsquoimportantrsquosıtan sitna sitno sitnı lsquotinyrsquoledan ledna ledno ledni lsquofrozenrsquotanak tanka tanko tankı lsquoslimrsquokratak kratka kratko kratkı lsquoshortrsquoblızak bliska blisko bliskı lsquoclosersquouzak uska usko uskı lsquonarrowrsquodobar dobra dobro dobrı lsquokindrsquooʃtar oʃtra oʃtro oʃtrı lsquosharprsquobodar bodra bodro bodri lsquoalertrsquo

2 Standard UkrainianStandard Ukrainian has palatalized and nonpalatalized consonants but onlynonpalatalized consonants before e Consonants are generally palatalizedbefore i with some apparent exceptions such as bil j lsquoachersquo which need not beseen as exceptions given the right analysis Give ordered rules to account for thealternations of the following nouns The alternation between o and e is limitedto suffixes Also for masculine nouns referring to persons ovev is insertedbetween the root and the case suffix in the locative singular (see words for lsquoson-in-lawrsquo lsquograndfatherrsquo) The data are initially ambiguous as to whether or not thealternations between o and i and between e and i are to be implemented by thesame rule Consider both possibilities give an argument for selecting one ofthese solutions

ustao ustala ustalo ustali lsquotiredrsquomukao mukla muklo muklı lsquohoarsersquoobao obla oblo oblı lsquoplumprsquopodao podla podlo podlı lsquobasersquo

Verbs1sg pres Masc past Fem past Neut pasttepem tepao tepla teplo lsquowanderrsquoskubem skubao skubla skublo lsquotearrsquotresem tresao tresla treslo lsquoshakersquovezem vezao vezla vezlo lsquoleadrsquo

Masculine nounsNom sg Dat pl Dat sɡ Loc sɡzub zubam zubovji zubji lsquotoothrsquosvjit svjitam svjitovji svjitji lsquolightrsquozjatj zjatjam zjatevji zjatevji lsquoson-in-lawrsquokoʃjilj koʃeljam koʃelevji koʃelji lsquobasketrsquozlodjij zlodjijam zlodjijevji zlodjijevji lsquothiefrsquomjisjatsj mjisjatsjam mjisjatsevji mjisjatsji lsquomonthrsquokorovaj korovajam korovajevji korovaji lsquoround loafrsquokamjinj kamenjam kamenevji kamenji lsquostonersquomjidj mjidjam mjidevji mjidji lsquocopperrsquoxljiw xljivam xljivovji xljivji lsquostablersquoholub holubam holubovji holubji lsquodoversquosjin sjinam sjinovji sjinovji lsquosonrsquolebjidj lebedjam lebedevji lebedji lsquoswanrsquosusjid susjidam susjidovji susjidovji lsquoneighborrsquotʃolovjik tʃolovjikam tʃolovjikovji tʃolovjikovji lsquomanrsquoljid ledam ledovji ledji lsquoicersquobilj boljam bolevji bolji lsquoachersquoriw rovam rovovji rovji lsquoditchrsquostiw stolam stolovji stolji lsquotablersquodjid djidam djidovji djidovji lsquograndfatherrsquoljit ljotam ljotovji ljotji lsquoflightrsquomist mostam mostovji mostji lsquobridgersquovetʃir vetʃoram vetʃorovji vetʃorji lsquoeveningrsquo

3 SomaliAccount for all phonological alternations in these data In your discussion ofthese forms be sure to make it clear what you assume the underlyingrepresentations of relevant morphemes are Your discussion should also make itclear what motivates your underlying representations and rules For instance ifyou could analyze some alternation by assuming underlying X and rule Y saywhy (or whether) that choice is preferable to the alternative of assumingunderlying P and rule Q

Neuter nounsNom sg Gen sg Dat sg Loc sg Gen pltjilo tjila tjilu tjilji tjiw lsquobodyrsquokoleso kolesa kolesu kolesji koljis lsquowheelrsquoozero ozera ozeru ozerji ozjir lsquolakersquoselo sela selu selji sjiw lsquovillagersquopole polja polju polji pilj lsquofieldrsquoslovo slova slovu slovji sliw lsquowordrsquomore morja morju morji mirj lsquosearsquo

Singular Sing definite Pluraldaar daarta daaro lsquohousersquogees geesta geeso lsquosidersquolaf lafta lafo lsquobonersquolug lugta luγo lsquolegrsquonaag naagta naaγo lsquowomanrsquotib tibta tiβo lsquopestlersquosab sabta saβo lsquooutcastrsquobad bada baetho lsquosearsquodʒid dʒida dʒietho lsquopersonrsquofeeɖ feeɖa feeʐo lsquoribrsquoʕiir ʕiirta ʕiiro lsquobuttermilkrsquoʔul ʔuʃa ʔulo lsquostickrsquobil biʃa bilo lsquomonthrsquomeel meeʃa meelo lsquoplacersquokaliil kaliiʃa kaliilo lsquosummerrsquonajl najʃa najlo lsquofemale lambrsquosun sunta sumo lsquopoisonrsquolaan laanta laamo lsquobranchrsquosin sinta simo lsquohiprsquodan danta dano lsquoaffairrsquodaan daanta daano lsquoriver bankrsquosaan saanta saano lsquohidersquonirig nirigta nirgo lsquobaby female camelrsquogaβaɖ gaβaɖa gabɖo lsquogirlrsquohoγol hoγoʃa hoglo lsquodownpourrsquobaγal baγaʃa baglo lsquomulersquowaħar waħarta waħaro lsquofemale kidrsquo

4 LatinProvide a complete account of the following phonological alternations in Latinincluding underlying forms for noun stems

irbad irbada irbaetho lsquoneedlersquokefed kefeda kefeetho lsquopanrsquodʒilin dʒilinta dʒilino lsquofemale dwarfrsquobohol bohoʃa boholo lsquoholersquodʒirid dʒirida dʒirdo lsquotrunkrsquoʔaajad ʔaajada ʔaajaetho lsquomiraclersquogaʕan gaʕanta gaʕmo lsquohandrsquoʔinan ʔinanta ʔinano lsquodaughterrsquo

3sg masc 3sg fern 1pl pastpast pastsuγaj sugtaj sugnaj lsquowaitrsquokaβaj kabtaj kabnaj lsquofixrsquosiethaj sidaj sidnaj lsquocarryrsquodilaj diʃaj dillaj lsquokillrsquoganaj gantaj gannaj lsquoaimrsquotumaj tuntaj tunnaj lsquohammerrsquoargaj aragtaj aragnaj lsquoseersquogudbaj guethubtaj guethubnaj lsquocross a riverrsquoqoslaj qosoʃaj qosollaj lsquolaughrsquohadlaj haethaʃaj haethallaj lsquotalkrsquo

Nominative Genitivearks arkis lsquofortressrsquoduks dukis lsquoleaderrsquodaps dapis lsquofeastrsquoreks regis lsquokingrsquofalanks falangis lsquophalanxrsquofiliks filikis lsquofernrsquolapis lapidis lsquostonersquolis litis lsquostrifersquofraws frawdis lsquodeceitrsquonoks noktis lsquonightrsquofrons frontis lsquobrowrsquofrons frondis lsquoleafrsquoinkus inkudis lsquoanvilrsquosors sortis lsquolotrsquofur furis lsquothiefrsquomurmur murmuris lsquomurmurrsquoaugur auguris lsquoaugurrsquoarbor arboris lsquotreersquopugil pugilis lsquoboxerrsquosal salis lsquosaltrsquoadeps adipis lsquofatrsquoapeks apikis lsquotoprsquo

The following six nouns and adjectives select a different genitive suffix -ias opposed to is You cannot predict on phonological grounds what nounstake this suffix but otherwise these words follow the rules motivated in thelanguage

What other phonological rule or rules are needed to account for the followingdata

5 TurkishProvide a phonological analysis of the following data from Turkish Note thatlong vowels like [a] are phonetically distinct from identical vowel clusterslike [aa]

prinkeps prinkipis lsquochiefrsquoekwes ekwitis lsquohorsemanrsquomiles militis lsquosoldierrsquonomen nominis lsquonamersquokarmen karminis lsquosongrsquolumen luminis lsquolightrsquowenter wentris lsquobellyrsquopater patris lsquofatherrsquokadawer kadaweris lsquocorpsersquotuber tuberis lsquoswellingrsquopiper piperis lsquopepperrsquokarker karkeris lsquoprisonrsquo

dies diei lsquodayrsquoliber liberi lsquofreersquomiser miseri lsquowretchedrsquoager agri lsquofieldrsquosinister sinistri lsquoleftrsquoliber libri lsquobookrsquo

as assis lsquowholersquoos ossis lsquobonersquofar farris lsquospellrsquomel mellis lsquohoneyrsquoos oris lsquomouthrsquoflos floris lsquoflowerrsquomus muris lsquomousersquokrus kruris lsquolegrsquokinis kineris lsquoashrsquopulvis pulveris lsquodustrsquo

Nom Poss Dɑt Abl Nom ploda odasɨ odaja odadan odalar lsquoroomrsquodere deresi dereje dereden dereler lsquoriverrsquoyty ytysy ytyje ytyden ytyler lsquoironrsquobalo balosu baloja balodan balolar lsquoballrsquoarɨ arɨsɨ arɨja arɨdan arɨlar lsquobeersquo

la lasɨ laja ladan lalar lsquola (note)rsquobina binasɨ binaja binadan binalar lsquobuildingrsquoimla imlasɨ imlaja imladan imlalar lsquospellingrsquobe besi beje beden beler lsquoB (letter)rsquokep kepi kepe kepten kepler lsquocaprsquoat atɨ ata attan atlar lsquohorsersquoek eki eke ekten ekler lsquoaffixrsquook oku oka oktan oklar lsquoarrowrsquogytʃ gydʒy gydʒe gytʃten gytʃler lsquopowerrsquoahmet ahmedi ahmede ahmetten ahmetler lsquoAhmedrsquokurt kurdu kurda kurttan kurtlar lsquowormrsquotyrk tyrky tyrke tyrkten tyrkler lsquoTurkrsquogentʃ gentʃi gentʃe gentʃten gentʃler lsquoyoungrsquohalk halkɨ halka halktan halklar lsquofolkrsquoyst ysty yste ystten ystler lsquoupper planersquosarp sarpɨ sarpa sarptan sarplar lsquosteeprsquoharp harbɨ harba harptan harplar lsquowarrsquoalt altɨ alta alttan altlar lsquobottomrsquorenk rengi renge renkten renkler lsquocolorrsquohis hissi hisse histen hisler lsquofeelingrsquohyr hyrry hyrre hyrden hyrler lsquofreersquomahal mahallɨ mahalla mahaldan mahallar lsquoplacersquohak hakkɨ hakka haktan haklar lsquorightrsquozam zammɨ zamma zamdan zamlar lsquoinflationrsquoaf affɨ affa aftan aflar lsquoexcusersquoarap arabɨ araba araptan araplar lsquoArabrsquokojun kojunu kojuna kojundan kojunlar lsquosheeprsquopilot pilotu pilota pilottan pilotlar lsquopilotrsquokitap kitabɨ kitaba kitaptan kitaplar lsquobookrsquodomuz domuzu domuza domuzdan domuzlar lsquopigrsquodavul davulu davula davuldan davullar lsquodrumrsquobajɨr bajɨrɨ bajɨra bajɨrdan bajɨrlar lsquoslopersquosomun somunu somuna somundan somunlar lsquoloafrsquofikir fikri fikre fikirden fikirler lsquoidearsquoisim ismi isme isimden isimler lsquonamersquobojun bojnu bojna bojundan bojunlar lsquoneckrsquotʃevir tʃevri tʃevre tʃevirden tʃevirler lsquoinjusticersquodevir devri devre devirden devirler lsquotransferrsquokojun kojnu kojna kojundan kojunlar lsquobosomrsquokarɨn karnɨ karna karɨndan karɨnlar lsquothoraxrsquoburun burnu burna burundan burunlar lsquonosersquoakɨl aklɨ akla akɨldan akɨllar lsquointelligencersquoʃehir ʃehri ʃehre ʃehirden ʃehirler lsquocityrsquonamaz namazɨ namaza namazdan namazlar lsquoworshiprsquozaman zamanɨ zamana zamandan zamanlar lsquotimersquoharap harabɨ haraba haraptan haraplar lsquoruinedrsquoikaz ikazɨ ikaza ikazdan ikazlar lsquowarningrsquohajat hajatɨ hajata hajattan hajatlar lsquolifersquoispat ispatɨ ispata ispattan ispatlar lsquoproofrsquo

6 KeraPropose rules to account for the following alternations It will prove useful tothink about Kera vowels in terms of high versus nonhigh vowels Also in thislanguage it would be convenient to assume that [h] and [ʔ] are specified as[+low] Pay attention to all verb forms like bɨlan lsquowant mersquo balnan lsquowanted mersquoand balla lsquoyou must wantrsquo ie there are present past and imperative formsinvolved certain tenses being marked by suffixes Finally pay attention to whatmight look like a coincidence in the distribution of vowels in the underlyingforms of verb roots there are no coincidences

inek inei inee inekten inekler lsquocowrsquomantɨk mantɨɨ mantɨa mantɨktan mantɨklar lsquologicrsquoajak ajaɨ ajaa ajaktan ajaklar lsquofootrsquotʃabuk tʃabuu tʃabua tʃabuktan tʃabuklar lsquoquickrsquodakik dakii dakie dakikten dakikler lsquopunctualrsquomerak merakɨ meraka meraktan meraklar lsquocuriosityrsquotebrik tebriki tebrike tebrikten tebrikler lsquogreetingsrsquohukuk hukuku hukuka hukuktan hukuklar lsquolawrsquo

haman lsquoeat mersquo senen lsquomy brotherrsquohamam lsquoeat you (masc)rsquo senem lsquoyour (masc) brotherrsquohɨmi lsquoeat you (fem)rsquo sini lsquoyour (fem) brotherrsquohɨmu lsquoeat himrsquo sinu lsquohis brotherrsquohama lsquoeat herrsquo sena lsquoher brotherrsquohamaŋ lsquoeat you (pl)rsquo seneŋ lsquoyour (pl) brotherrsquo

kolon lsquochange mersquo gidin lsquomy bellyrsquokolom lsquochange you (masc)rsquo gidim lsquoyour (masc) bellyrsquokuli lsquochange you (fem)rsquo gidi lsquoyour (fem) bellyrsquokulu lsquochange himrsquo gidu lsquohis bellyrsquokola lsquochange herrsquo gidɨ lsquoher bellyrsquokoloŋ lsquochange you (pl)rsquo gidiŋ lsquoyour (pl) bellyrsquo

cɨrɨn lsquomy headrsquo gunun lsquowake mersquocɨrɨm lsquoyour (masc) headrsquo gunum lsquowake you (masc)rsquociri lsquoyour (fem) headrsquo guni lsquowake you (fem)rsquocuru lsquohis headrsquo gunu lsquowake himrsquocɨrɨ lsquoher headrsquo gunɨ lsquowake herrsquocɨrɨŋ lsquoyour (pl) headrsquo gunuŋ lsquowake you (pl)rsquo

bɨlan lsquowant mersquo ŋɨfan lsquomeet mersquobɨlam lsquowant you (masc)rsquo ŋɨfam lsquomeet you (masc)rsquobɨli lsquowant you (fem)rsquo ŋɨfi lsquomeet you (fem)rsquobɨlu lsquowant himrsquo ŋɨfu lsquomeet himrsquobɨla lsquowant herrsquo ŋɨfa lsquomeet herrsquobɨlaŋ lsquowant you (pl)rsquo ŋɨfaŋ lsquomeet you (pl)rsquo

7 Keley-iAccount for the alternations in the following verbs The different forms relate towhether the action is in the past or future and which element in the sentence isemphasized (subject object instrument) Roots underlyingly have the shapeCVC(C)VC and certain forms such as the subject focus future require changes inthe stem that result in a CVCCVC shape This may be accomplished byreduplicating the initial CV- for stems whose first vowel is [e] (ʔum-bebhat behat) or doubling the middle consonant (ʔum-buŋŋet ndash buŋet) Thecontrastive identification imperfective form conditions lengthening of theconsonant in the middle of the stem when the first vowel is not [e](memajjuʔ bajuʔ) These changes are part of the morphology so do notattempt to write phonological rules to double consonants or reduplicatesyllables Be sure to explicitly state the underlying form of each root and affixUnderstanding the status of [s] and [h] in this language is important in solvingthis problem It is also important to consider exactly what underlying nasalconsonant is present in these various prefixes and infixes ndash there is evidence inthe data which shows that the underlying nature of the nasal explains certainobserved differences in phonological behavior

ʔasan lsquoknow mersquo ʔapan lsquofind mersquoʔasam lsquoknow you (masc)rsquo ʔapam lsquofind you (masc)rsquoʔɨsi lsquoknow you (fem)rsquo ʔɨpi lsquofind you (fem)rsquoʔɨsu lsquoknow himrsquo ʔɨpu lsquofind himrsquoʔasa lsquoknow herrsquo ʔapa lsquofind herrsquoʔasaŋ lsquoknow you (pl)rsquo ʔapaŋ lsquofind you (pl)rsquo

haran lsquogive me backrsquoharam lsquogive you (masc) backrsquohɨri lsquogive you (fem) backrsquohɨru lsquogive him backrsquohara lsquogive her backrsquoharaŋ lsquogive you (pl) backrsquo

balnan lsquowanted mersquo ŋafnan lsquomet mersquobalnam lsquowanted you (masc)rsquo ŋafnam lsquomet you (masc)rsquobɨlni lsquowanted you (fem)rsquo ŋɨfni lsquomet you (fem)rsquobɨlnu lsquowanted himrsquo ŋɨfnu lsquomet himrsquobalna lsquowanted herrsquo ŋafna lsquomet herrsquobalnaŋ lsquowanted you (pl)rsquo ŋafnaŋ lsquomet you (pl)rsquoballa lsquoyou must wantrsquo ŋafla lsquoyou must meetrsquo

ba lsquonotrsquo pa lsquoagainrsquo bɨpa lsquono morersquo

Subject focus Direct object Instrumental focusfuture focus past pastʔumduntuk dinuntuk ʔinduntuk lsquopunchrsquoʔumbajjuʔ binajuʔ ʔimbajuʔ lsquopound ricersquo

The following past subject clausal focus forms involve a different prefix usingsome of the roots found above A number of roots require reduplication of thefirst root syllable

ʔumdillag dinilag ʔindilag lsquolight lamprsquoʔumgubbat ginubat ʔiŋjgubat lsquofightrsquoʔumhullat hinulat ʔinhulat lsquocoverrsquoʔumbuŋŋet binuŋet ʔimbuŋet lsquoscoldrsquoʔumgalgal ginalgal ʔiŋgalgal lsquochewrsquoʔumʔagtuʔ ʔinagtuʔ ʔinʔagtuʔ lsquocarry on headrsquoʔumʔehneŋ ʔinehneŋ ʔinʔehneŋ lsquostandrsquoʔumbebhat binhat ʔimbehat lsquocut rattanrsquoʔumdedʔek dinʔek ʔindeʔek lsquoaccusersquoʔumtuggun sinugun ʔintugun lsquoadvisersquoʔumtetpen simpen ʔintepen lsquomeasurersquoʔumpeptut pintut ʔimpetut lsquodamrsquoʔumhehpuŋ himpuŋ ʔinhepuŋ lsquobreak a stickrsquoʔumtetkuk siŋkuk ʔintekuk lsquoshoutrsquoʔumkekbet kimbet ʔiŋkebet lsquoscratchrsquoʔumbebdad bindad ʔimbedad lsquountiersquoʔumdedgeh diŋgeh ʔindegeh lsquosickrsquo

Instrumental Contrastive Contrastivepast focus id imperfective id perfectiveʔinduntuk menuntuk nenuntuk lsquopunchrsquoʔimbajuʔ memajjuʔ nemajuʔ lsquopound ricersquoʔindilag menillag nenilag lsquolight lamprsquoʔiŋgubat meŋubbat neŋubat lsquofightrsquoʔinhulat menullat nenulat lsquocoverrsquoʔintanem menannem nenanem lsquoplantrsquoʔimpedug memdug nemdug lsquochasersquoʔimbedad memdad nemdad lsquountiersquoʔiŋkebet meŋbet neŋbet lsquoscratchrsquoʔimbekaʔ memkaʔ nemkaʔ lsquodigrsquoʔintepen mempen nempen lsquomeasurersquoʔintebaʔ membaʔ nembaʔ lsquokill a pigrsquoʔintekuk meŋkuk neŋkuk lsquoshoutrsquoʔindegeh meŋgeh neŋgeh lsquosickrsquoʔinhepaw mempaw nempaw lsquopossessrsquoʔinteled menled nenled lsquostingrsquoʔindeʔek menʔek nenʔek lsquoaccusersquoʔinʔebaʔ meŋbaʔ neŋbaʔ lsquocarry on backrsquoʔinʔinum meŋinnum neŋinum lsquodrinkrsquoʔinʔagtuʔ meŋagtuʔ neŋagtuʔ lsquocarry on headrsquoʔinʔalaʔ meŋallaʔ neŋalaʔ lsquogetrsquoʔinʔawit meŋawwit neŋawit lsquogetrsquo

nandunduntuk lsquopunchrsquo nampepedug lsquochasersquonaŋkekebet lsquoscratchrsquo nambebekaʔ lsquodigrsquonantetekuk lsquoshoutrsquo nandedeʔek lsquoaccuse

8 KuriaIn some (but not all) of the examples below morpheme boundaries havebeen introduced to assist in the analysis Pronouns are assigned to agrammatical class depending on the noun which they refer to conventionallygiven a number (1ndash20) Tone may be disregarded (however it is predictable inthe infinitive) It is important to pay attention to interaction between processesin this problem

ogo-taanga lsquoto beginrsquo oko-gέsa lsquoto harvestrsquooko-rɔ ga lsquoto witchrsquo oko-rema lsquoto plowrsquooko-hoora lsquoto threshrsquo ugu-sııka lsquoto close a doorrsquougu-suraanga lsquoto sing praisersquo uku-gıınga lsquoto shaversquougutuuha lsquoto be bluntrsquo

ogo-ko-barǎ lsquoto count you (sg)rsquo uku-gu-suraanga lsquoto praise you (sg)rsquooko-mo-barǎ lsquoto count himrsquo uku-mu-suraanga lsquoto praise himrsquoogo-to-barǎ lsquoto count usrsquo ugu-tu-suraanga lsquoto praise usrsquooko-ge-barǎ lsquoto count them (4)rsquo uku-gı-suraanga lsquoto praise it (4)rsquooko-re-barǎ lsquoto count it (5)rsquo uku-rı-suraanga lsquoto praise it (5)rsquouku-bı-barǎ lsquoto count it (8)rsquo uku-bı-suraanga lsquoto praise it (8)rsquougu-tʃı-barǎ lsquoto count it (10)rsquo ugu-tʃ ı-suraanga lsquoto praise it (10)rsquo

oko-mo-go-gεsέra lsquoto harvest it (3) for himrsquouku-mu-gu-siıkja lsquoto make him close it (3)rsquouku-mu-gu-siındja lsquoto make him win it (3)rsquooko-ba-suraanga lsquoto praise themrsquooko-mo-ba-suraangera lsquoto praise them for himrsquooko-ba-mu-suraangera lsquoto praise him for themrsquo

To V To make to V To V for To make V forokorema ukurımja okoremerǎ ukurımırja lsquoweedrsquookoroma ukurumja okoromerǎ ukurumırja lsquobitersquookohoora ukuhuurja okohoore ra ukuhuurırja lsquothreshrsquookoheetoka ukuhııtukja okoheetokera ukuhııtukirja lsquorememberrsquookogeemba ukugıımbja okogeembera ukugıımbırja lsquomake rainrsquoogosooka ugusuukja ogosookera ugusuukırja lsquorespectrsquoogotegetǎ ugutıgıtjǎ ogotegetera ugutıgıtirja lsquobe latersquookorɔga okorogja okorɔgέrǎ okorogerja lsquobewitchrsquookogɔɔ ga okogoogja okogɔɔ gέra okogoogerja lsquoslaughterrsquookogɔɔ ta okogootja okogɔɔ tέra okogooterja lsquoholdrsquoogosɔka ogosokja ogosɔkέrǎ ogosokerja lsquopokersquoogotέrέkǎ ogoterekja ogotέrέkεra ogoterekerja lsquobrewrsquookogέsa okogesja okogέsέrǎ okogeserja lsquoharvestrsquoogosέέnsa ogoseensja ogosέέnsέra ogoseenserja lsquowinnowrsquo

nanʔeʔebaʔ lsquocarry on backrsquo nanʔiʔinum lsquodrinkrsquonantanem lsquoplantrsquo

9 LardilAccount for the phonological alternations seen in the data below

To V To make to V To V for To make V forugusııka ugusııkja ogoseekera ugusııkırja lsquoto closersquoukuruga ukurugja okorogerǎ ukurugırja lsquoto cookrsquougusuka ugusukja ogosokerǎ ugusukırja lsquoto plaitrsquoukurıınga ukurııngja okoreengera ukurııngırja lsquoto foldrsquougusıında ugusıındja ogoseendera ugusıındırja lsquoto winrsquo

Imperative Infinitive They will V Then will V forremǎ okorema mbarerema mbareremera lsquocultivatersquobarǎ okobara mbarebara mbarebarera lsquocountrsquoatǎ ogɔɔ ta mbarεέta mbarεέtέra lsquobe splitrsquoahǎ okɔɔ ha mbarεέha mbarεέhέra lsquopick greensrsquoagǎ okɔɔ ga mbarεέga mbarεέgέra lsquoweedrsquoaanga okɔɔ nga mbarεέnga mbarεέngέra lsquorefusersquoandeka okɔɔ ndέkǎ mbarεέndέka mbarεέndέkεra lsquowritersquo

Imperative 3ɡ subjunctive 3sɡ subjunctive forremǎ aremε aremerέ lsquocultivatersquotεrεka atεrεkέ atεrεkέrε lsquobrewrsquoebǎ εεbε εεbεrέ lsquoforgetrsquoegǎ εεgε εεgεrέ lsquolearnrsquoogǎ ɔɔgε ɔɔgεrέ lsquobe sharprsquoεjǎ εεjε εεjεrέ lsquosweeprsquoɔrɔka ɔɔrɔkέ ɔɔrɔkέrε lsquocome outrsquo

Bare N Accusative Nonfuture Futurekentapal kentapalin kentapalŋaɾ kentapaluɽ lsquodugongrsquoketaɾ ketaɾin ketaɾŋaɾ ketaɾuɽ lsquoriverrsquomijaɽ mijaɽin mijaɽŋaɾ mijaɽuɽ lsquospearrsquojupuɾ jupuɾin jupuɾŋaɾ jupuɾuɽ lsquored rock codrsquotaŋuɾ taŋuɾin taŋuɾŋaɾ taŋuɾuɽ lsquocrab (sp)rsquojaɾaman jaɾamanin jaɾamanaɾ jaɾamankuɽ lsquohorsersquomaan maanin maanaɾ maankuɽ lsquospearrsquopiɾŋen piɾŋenin piɾŋenaɾ piɾŋenkuɽ lsquowomanrsquomela melan melaŋaɾ melaɽ lsquosearsquotawa tawan tawaŋaɾ tawaɽ lsquoratrsquowanka wankan wankaŋaɾ wankaɽ lsquoarmrsquokuŋka kuŋkan kuŋkaŋaɾ kuŋkaɽ lsquogroinrsquotaɾŋka taɾŋkan taɾŋkaŋaɾ taɾŋkaɽ lsquobarracudarsquoŋuka ŋukun ŋukuŋaɾ ŋukuɽ lsquowaterrsquoŋuɽa ŋuɽun ŋuɽuŋaɾ ŋuɽuɽ lsquoforeheadrsquokaʈa kaʈun kaʈuŋaɾ kaʈuɽ lsquochildrsquomuna munun munuŋaɾ munuɽ lsquoelbowrsquoŋawa ŋawun ŋawuŋaɾ ŋawuɽ lsquodogrsquokenʈe kenʈin kenʈiŋaɾ kenʈiwuɽ lsquowifersquo

10 Sakha (Yakut)Give a phonological analysis of the following case-marking paradigms of nounsin Sakha

tjimpe tjimpin tjimpiŋaɾ tjimpiwuɽ lsquotailrsquoŋine ŋinin ŋiniŋaɾ ŋiniwuɽ lsquoskinrsquopape papin papiŋaɾ papiwuɽ lsquofatherrsquos motherrsquotjempe tjempen tjempeŋaɾ tjempeɽ lsquomotherrsquos fatherrsquowiʈe wiʈen wiʈeŋaɾ wiʈeɽ lsquointeriorrsquowaŋal waŋalkin waŋalkaɾ waŋalkuɽ lsquoboomerangrsquomenjel menjelkin menjelkaɾ menjelkuɽ lsquodogfish (sp)rsquomakaɾ makaɾkin makaɾkaɾ makaɾkuɽ lsquoanthillrsquojalul jalulun jaluluŋaɾ jaluluɽ lsquoflamersquomajaɾ majaɾan majaɾaŋaɾ majaɾaɽ lsquorainbowrsquotalkuɾ talkuɾan talkuɾaŋaɾ talkuɾaɽ lsquokookaburrarsquowiwal wiwalan wiwalaŋaɾ wiwalaɽ lsquobush mangorsquokaɾikaɾ kaɾikaɾin kaɾikaɾiŋaɾ kaɾikaɾiwuɽ lsquobutter-fishrsquojilijil jilijilin jilijiliŋaɾ jilijiliwuɽ lsquooyster (sp)rsquojukaɾ jukaɾpan jukaɾpaŋaɾ jukaɾpaɽ lsquohusbandrsquopulŋaɾ pulŋaɾpan pulŋaɾpaŋaɾ pulŋaɾpaɽ lsquohugersquowulun wulunkan wulunkaŋaɾ wulunkaɽ lsquofruit (sp)rsquowuʈal wuʈaltjin wuʈaltjiŋaɾ wuʈaltjiwuɽ lsquomeatrsquokantukan kantukantun kantukantuŋaɾ kantukantuɽ lsquoredrsquokaɾwakaɾ kaɾwakaɾwan kaɾwakaɾwaŋaɾ kaɾwakaɾwaɽ lsquowattle (sp)rsquotuɾaɾa tuɾaɾaŋin tuɾaɾaŋaɾ tuɾaɾaŋkuɽ lsquosharkrsquoŋalu ŋalukin ŋalukaɾ ŋalukuɽ lsquostoryrsquokuɾka kuɾkaŋin kuɾkaŋaɾ kuɾkaŋkuɽ lsquopandjarsquotaŋku taŋkuŋin taŋkuŋaɾ taŋkuŋkuɽ lsquooyster (sp)rsquokuɾpuɽu kuɾpuɽuŋin kuɾpuɽuŋaɾ kuɾpuɽuŋkuɽ lsquolancewoodrsquoputu putukan putukaŋaɾ putukaɽ lsquoshortrsquomaali maalijan maalijaŋaɾ maalijaɽ lsquoswamp turtlersquotjintiɾpu tjintiɾpuwan tjintiɾpuwaŋaɾ tjintiɾpuwaɽ lsquowillie wagtailrsquopukatji pukatjijan pukatjijaŋaɾ pukatjijaɽ lsquohawk (sp)rsquomuɾkuni muɾkuniman muɾkunimaŋaɾ muɾkunimaɽ lsquonullahrsquoŋawuŋa ŋawuŋawun ŋawuŋawuŋaɾ ŋawuŋawuɽ lsquotermitersquotipiti tipitipin tipitipiŋaɾ tipitipiwuɽ lsquorock-cod (sp)rsquotapu taputjin taputjiŋaɾ taputjiwuɽ lsquoolder brotherrsquomuŋkumu muŋkumuŋkun muŋkumuŋkuŋaɾ muŋkumuŋkuɽ lsquowooden axersquotjumputju tjumputjumpun tjumputjumpuŋaɾ tjumputjumpuɽ lsquodragonflyrsquo

Noun Plural Associativeaγa aγalar aγalɨɨn lsquofatherrsquopaarta paartalar paartalɨɨn lsquoschool deskrsquotɨa tɨalar tɨalɨɨn lsquoforestrsquokinige kinigeler kinigeliin lsquobookrsquodʒie dʒieler dʒieliin lsquohousersquoije ijeler ijeliin lsquomotherrsquokini kiniler kiniliin lsquo3rd personrsquo

bie bieler bieliin lsquomarersquooγo oγolor oγoluun lsquochildrsquoχopto χoptolor χoptoluun lsquogullrsquoboslashroslash boslashroslashloslashr boslashroslashlyyn lsquowolfrsquotɨal tɨallar tɨallɨɨn lsquowindrsquoɨal ɨallar ɨallɨɨn lsquoneighborrsquokuul kuullar kuulluun lsquosackrsquoat attar attɨɨn lsquohorsersquobalɨk balɨktar balɨktɨɨn lsquofishrsquoɨskaap ɨskaaptar ɨskaaptɨɨn lsquocabinetrsquooγus oγustar oγustuun lsquobullrsquokus kustar kustuun lsquoduckrsquotynnyk tynnykter tynnyktyyn lsquowindowrsquosep septer septiin lsquotoolrsquoet etter ettiin lsquomeatrsquooslashrys oslashryster oslashrystyyn lsquoriverrsquotiis tiister tiistiin lsquotoothrsquosoroχ soroχtor soroχtuun lsquosome personrsquooχ oχtor oχtuun lsquoarrowrsquooloppos oloppostor oloppostuun lsquochairrsquooslashtoslashχ oslashtoslashχtoslashr oslashtoslashχtyyn lsquoabandoned farmrsquoubaj ubajdar ubajdɨɨn lsquoelder brotherrsquosaraj sarajdar sarajdɨɨn lsquobarnrsquotɨj tɨjdar tɨjdɨɨn lsquofoalrsquoatɨɨr atɨɨrdar atɨɨrdɨɨn lsquostallionrsquoojuur ojuurdar ojuurduun lsquoforestrsquoytʃygej ytʃygejder ytʃygejdiin lsquogood personrsquoedʒiij edʒiijder edʒiijdiin lsquoelder sisterrsquotomtor tomtordor tomtorduun lsquoknobrsquomoγotoj moγotojdor moγotojduun lsquochipmunkrsquokoslashtoslashr koslashtoslashrdoslashr koslashtoslashrdyyn lsquobirdrsquoboslashlkoslashj boslashlkoslashjdoslashr boslashlkoslashjdyyn lsquoisletrsquoχatɨŋ χatɨŋnar χatɨŋnɨɨn lsquobirchrsquoaan aannar aannɨɨn lsquodoorrsquotiiŋ tiiŋner tiiŋniin lsquosquirrelrsquosordoŋ sordoŋnor sordoŋnuun lsquopikersquoolom olomnor olomnuun lsquofordrsquooron oronnor oronnuun lsquobedrsquoboslashdoslashŋ boslashdoslashŋnoslashr boslashdoslashŋnyyn lsquostrong onersquo

Noun Partitive Comparative Ablativeaγa aγata aγataaγar aγattan lsquofatherrsquopaarta paartata paartataaγar paartattan lsquoschool deskrsquotɨa tɨata tɨataaγar tɨattan lsquoforestrsquokinige kinigete kinigeteeγer kinigetten lsquobookrsquodʒie dʒiete dʒieteeγer dʒietten lsquohousersquoije ijete ijeteeγer ijetten lsquomotherrsquokini kinite kiniteeγer kinitten lsquo3rd personrsquobie biete bieteeγer bietten lsquomarersquooγo oγoto oγotooγor oγotton lsquochildrsquo

χopto χoptoto χoptotooγor χoptotton lsquogullrsquoboslashroslash boslashroslashtoslash boslashroslashtoslashoslashγoslashr boslashroslashttoslashn lsquowolfrsquotɨal tɨalla tɨallaaγar tɨaltan lsquowindrsquoɨal ɨalla ɨallaaγar ɨaltan lsquoneighborrsquokuul kuulla kuullaaγar kuultan lsquosackrsquomoχsoγol moχsoγollo moχsoγollooγor moχsoγolton lsquofalconrsquoat atta attaaγar attan lsquohorsersquobalɨk balɨkta balɨktaaγar balɨktan lsquofishrsquoɨskaap ɨskaapta ɨskaaptaaγar ɨskaaptan lsquocabinetrsquooγus oγusta oγustaaγar oγustan lsquobullrsquokus kusta kustaaγar kustan lsquoduckrsquotynnyk tynnykte tynnykteeγer tynnykten lsquowindowrsquosep septe septeeγer septen lsquotoolrsquoet ette etteeγer etten lsquomeatrsquooslashrys oslashryste oslashrysteeγer oslashrysten lsquoriverrsquotiis tiiste tiisteeγer tiisten lsquotoothrsquosoroχ soroχto soroχtooγor soroχton lsquosome personrsquooslashtoslashχ oslashtoslashχtoslash oslashtoslashχtoslashoslashγoslashr oslashtoslashχtoslashn lsquoabandoned

farmrsquoubaj ubajda ubajdaaγar ubajtan lsquoelder brotherrsquosaraj sarajda sarajdaaγar sarajtan lsquobarnrsquotɨj tɨjda tɨjdaaγar tɨjtan lsquofoalrsquoatɨɨr atɨɨrda atɨɨrdaaγar atɨɨrtan lsquostallionrsquoχirur χirurda χirurdaaγar χirurtan lsquosurgeonrsquoytʃygej ytʃygejde ytʃygejdeeγer ytʃygejten lsquogood personrsquotomtor tomtordo tomtordooγor tomtorton lsquoknobrsquomoγotoj moγotojdo moγotojdooγor moγotojton lsquochipmunkrsquokoslashtoslashr koslashtoslashrdoslash koslashtoslashrdoslashoslashγoslashr koslashtoslashrtoslashn lsquobirdrsquosuorγan suorγanna suorγannaaγar suorγantan lsquoblanketrsquoχatɨŋ χatɨŋna χatɨŋnaaγar χatɨŋtan lsquobirchrsquoaan aanna aannaaγar aantan lsquodoorrsquotiiŋ tiiŋne tiiŋneeγer tiiŋten lsquosquirrelrsquosordoŋ sordoŋno sordoŋnooγor sordoŋton lsquopikersquoolom olomno olomnooγor olomton lsquofordrsquoboslashdoslashŋ boslashdoslashŋnoslash boslashdoslashŋnoslashoslashγoslashr boslashdoslashŋtoslashn lsquostrong onersquo

Noun Dative Accusativeaγa aγaγa aγanɨ lsquofatherrsquodʒie dʒieγe dʒieni lsquohousersquoije ijeγe ijeni lsquomotherrsquooγo oγoγo oγonu lsquochildrsquoboslashroslash boslashroslashγoslash boslashroslashny lsquowolfrsquotɨal tɨalga tɨalɨ lsquowindrsquokuul kuulga kuulu lsquosackrsquoat akka atɨ lsquohorsersquobalɨk balɨkka balɨgɨ lsquofishrsquoɨskaap ɨskaapka ɨskaabɨ lsquocabinetrsquooγus oγuska oγuhu lsquobullrsquokus kuska kuhu lsquoduckrsquo

11 Sadzhava UkrainianGive a phonological analysis of the following data Assume that all surfaceoccurrences of kj and gj in this language are derived by rule Also assume thatstress is located on the proper vowel in the underlying representation the rulesfor shifting stress are too complex to be considered here Nouns in declension IIdepalatalize a consonant before the locative suffix and nouns in declension IIIdepalatalize in the genitive The variation in the genitive and locative singularsuffix in declension I (-i or -a versus -u) is lexically governed do not write ruleswhich select between these suffixes Concentrate on establishing the correctunderlying representations for the noun stem

sep sepke sebi lsquotoolrsquoet ekke eti lsquomeatrsquotiis tiiske tiihi lsquotoothrsquoot okko otu lsquograssrsquosoroχ soroχχo soroγu lsquosome personrsquooslashtoslashχ oslashtoslashχχoslash oslashtoslashγy lsquoabandoned farmrsquooχ oχχo oγu lsquoarrowrsquosaraj sarajga sarajɨ lsquobarnrsquotɨj tɨjga tɨjɨ lsquofoalrsquokoslashtoslashr koslashtoslashrgoslash koslashtoslashry lsquobirdrsquoojuun ojuuŋŋa ojuunu lsquoshamanrsquoχatɨŋ χatɨŋŋa χatɨŋɨ lsquobirchrsquoaan aaŋŋa aanɨ lsquodoorrsquoolom olomŋo olomu lsquofordrsquo

Declension INom sɡ Gen sɡ Loc sɡˈplast plasˈta plasjˈkji lsquolayerrsquoskoˈrux skoruˈxa skoruˈsji lsquomountain ashrsquoˈγjrjix γjrjixa γjrjiˈsji lsquosinrsquopasˈtux pastuˈxa pastuˈsji lsquoherdsmanrsquoˈmjnjux ˈmjnjuxa ˈmjnjusji lsquofish (sp)rsquoˈpluγ ˈpluγa ˈpluzji lsquoplowrsquoˈsjtjiγ ˈstoγa ˈstozji lsquostackrsquoˈsak ˈsaka ˈsatsji lsquofishnetrsquo

Noun Our N Noun Our Naγa aγabɨt lsquofatherrsquo ije ijebit lsquomotherrsquouol uolbut lsquosonrsquo koslashtoslashr koslashtoslashrbyt lsquobirdrsquokɨlaas kɨlaaspɨt lsquoclassroomrsquo ɨskaap ɨskaappɨt lsquocabinetrsquokuorat kuorappɨt lsquotownrsquo tiis tiispit lsquotoothrsquoohoχ ohoχput lsquostoversquo tynnyk tynnykpyt lsquowindowrsquoaan aammɨt lsquodoorrsquo kapitan kapitammɨt lsquocaptainrsquotiiŋ tiiŋmit lsquosquirrelrsquo oron orommut lsquobedrsquokyn kymmyt lsquodayrsquo

ˈbek bəˈka bəˈtsji lsquobullrsquoˈlest ləsˈta ləsjˈkji lsquoletterrsquoˈlest ˈlesta ˈlesjkji lsquoleafrsquoˈpjljit ˈplota ˈplokji lsquowicker fencersquoˈsjmjrjid ˈsmroda ˈsmrogji lsquostenchrsquoˈfjist fosˈta fosjˈkji lsquotailrsquoˈmjist ˈmosta ˈmosjkji lsquobridgersquoˈljid ˈlaeligdu ləˈdu lsquoicersquoˈdjrjit ˈdrota ˈdrokji lsquothick wirersquoˈmjid ˈmaeligdu məˈdu lsquohoneyrsquoˈvjil voˈla voˈlji lsquooxrsquoˈvjiz ˈvoza ˈvozji lsquocartrsquoˈser ˈsera ˈserji lsquocottage cheesersquoˈsjnjip snoˈpa snoˈpji lsquosheafrsquoˈγreb γrəˈba γrəˈbji lsquomushroomrsquoˈlaeligbjid ˈlaeligbəda ˈlaeligbəgji lsquoswanrsquoˈbaeligrjiγ ˈbaeligrəγa ˈbaeligrəzji lsquoshorersquopəˈrjiγ pəˈroγa pəˈrozji lsquodumplingrsquoˈporjiγ poˈroγa poˈrozji lsquothresholdrsquoboˈljek boljəˈka boljəˈtsji lsquoabcessrsquoˈvorjiγ ˈvoroγa ˈvorozji lsquoenemyrsquoˈkonək ˈkonəka ˈkonətsji lsquograsshopperrsquoˈpotjik poˈtoka poˈtotsji lsquostreamrsquoˈtjik ˈtoka ˈtotsji lsquocurrentrsquoˈkjil koˈla koˈlji lsquostakersquo

Declension IINom sɡ Gen sɡ Loc sɡkoˈvalj kovaˈlje kovaˈle lsquoblacksmithrsquoˈdʒmjilj dʒmjiˈlje dʒmjiˈle lsquobumblebeersquoˈkjrjilj kjrjiˈlje kjrjiˈle lsquorabbitrsquouˈtʃetəlj uˈtʃetəljə uˈtʃetələ lsquoteacherrsquoˈgraeligbjinj ˈgraeligbənjə ˈgraeligbənə lsquocombrsquoˈolənj ˈolənjə ˈolənə lsquodeerrsquoyatʃjˈmjinj yatʃˈmaelignjə yatʃˈmaelignə lsquobarleyrsquoˈyasjinj ˈyasənjə ˈyasənə lsquoash treersquoˈzjekj ˈzjekjə ˈzjetə lsquoson-in-lawrsquo

Declension IIINom sɡ Gen sɡˈmasjkj ˈmastə lsquofatrsquoˈsjmjirjkj ˈsmaeligrtə lsquodeathrsquoˈvjisjkj ˈvjistə lsquonewsrsquoˈsjilj ˈsolə lsquosaltrsquoˈpoʃjisjkj ˈpoʃəstə lsquoepidemicrsquoˈzamjikj ˈzamətə lsquosnowstormrsquoˈskatərjkj ˈskatərtə lsquotableclothrsquoˈkjisjkj ˈkostə lsquobonersquo

12 KoromfeKoromfe has two kinds of vowels [-ATR] ɪ ʊ ε ɔ a and [+ATR] i u e o ʌ Providean analysis of the alternations in the following data which involve singular andplural forms of nouns and different tense-inflections for verbs

Singular Pluralgɪbrε gɪba lsquohatchetrsquohubre hubʌ lsquoditchrsquonεbrε nεba lsquopearsquodĩŋgre dĩŋgʌ lsquobush typersquozoŋgre zoŋgʌ lsquowingrsquolɔŋgrε lɔŋga lsquoshoersquohullre hullʌ lsquogutterrsquosεkrε sεka lsquohalfrsquotεfrε tεfa lsquocotton fiberrsquodabεεrε dabεεja lsquocamprsquodɔɔrε dɔɔja lsquolongrsquogɪgaarε gɪgaaja lsquovulturersquopʊpaarε pʊpaaja lsquograss typersquokoire kojʌ lsquobraceletrsquodʊmdε dʊma lsquolionrsquohulomde hulomʌ lsquomarrowrsquotεmdε tεma lsquobeardrsquologomde logomʌ lsquocamelrsquobɪndε bɪna lsquoheartrsquohɔndε hɔna lsquohoersquohonde honʌ lsquobeanrsquogeŋde geŋʌ lsquopebblersquozεŋdε zεŋa lsquoupper armrsquobεllε bεla lsquobackrsquojɪllε jɪla lsquohornrsquoselle selʌ lsquospacersquopallε pala lsquostretcherrsquodeŋgele deŋgelʌ lsquoopen arearsquosembele sembelʌ lsquopiecersquodaɪnε daja lsquowoodrsquohʊ ɪnε hʊ ja lsquocaterpillarrsquokɔ ɪnε kɔ ja lsquosquirrelrsquokɔ ɔnε kɔ ɔja lsquooldrsquosɔ ɔnε sɔɔ ja lsquoperiodrsquobεtε bεra lsquomale animalrsquodatε dara lsquochestrsquogete gerʌ lsquoforked stickrsquogote gorʌ lsquostreamrsquobɪtε bɪra lsquofrogrsquodɔtε dɔra lsquocloudrsquo

Neutral Past Progressiveta taε taraa lsquoshootrsquogɔ gɔε gɔraa lsquogo backrsquokʊ kɔε kʊraa lsquokillrsquo

Further readingKenstowicz and Kisseberth 1979 Zwicky 1973 1974 1975 Pullum 1976

tu toe turʌʌ lsquocoatrsquoli lee lirʌʌ lsquoforgetrsquodɪ dε dɪraa lsquoeatrsquota taε tanaa lsquocontradictrsquonε nε nεnaa lsquodefecatersquosaɪ sajε saɪraa lsquoseparatersquojɛɪ jɛjɛ jɛɪraa lsquowastersquosɔɪ sɔjε sɔɪraa lsquosplitrsquojεɪ jεjε jεɪnaa lsquocatchrsquodɔɪ dɔ jε dɔɪnaa lsquodreamrsquokεndɪ kεndε kεndraa lsquofinishrsquokεsɪ kεsε kεsraa lsquosurpassrsquokεtɪ kεtε kεtraa lsquoopenrsquotεŋgɪ tεŋgε tεŋgraa lsquoaccompanyrsquojisi jise jisrʌʌ lsquosufficersquojɪsɪ jɪsε jɪsraa lsquodraw waterrsquobirgi birge birgrʌʌ lsquoblackenrsquopasgɪ pasgε pasgraa lsquosplitrsquomεntɪ mεntε mεntraa lsquoassemblersquogondu gonde gondrʌʌ lsquodepartrsquohɔŋgʊ hɔ ŋgε hɔŋgraa lsquopointrsquosʊrgʊ sʊrgε sʊrgraa lsquodroprsquohɔkʊ hɔ kε hɔkraa lsquoscratchrsquozullu zulle zullrʌʌ lsquobowrsquosɪbʊ sɪbε sɪbraa lsquodiersquozambʊ zambε zambraa lsquodeceiversquowufu wufe wufrʌʌ lsquoborrowrsquozɪgamsʊ zɪgamsε zɪgamsraa lsquobe dirtyrsquohεmsʊ hεmsε hεmsraa lsquomeetrsquoleli lele lellʌʌ lsquosingrsquopɪlɪ pɪlε pɪllaa lsquotrample flatrsquotarɪ tarε tataa lsquoplasterrsquofεrɪ fεrε fεtaa lsquocultivatersquotʊrʊ tʊrε tʊtaa lsquointroducersquo

CHAPTER

7 Phonologicaltypology andnaturalness

PREVIEW

One of the main goals of many phonologists is to explain

why certain phonological patterns are found in numerous

languages while other patterns are found in few or no

languages This chapter looks at phonological typology ndash the

study of common versus uncommon phonological rules

KEY TERMStypology

crosslinguisticcomparison

markedness

functionalexplanation

A widely invoked criterion in deciding between analyses of a language iswhether the rules of one analysis aremore natural usually judged in termsofwhether the rules occur frequently across languages As a prerequisite toexplaining why some processes are common uncommon or evenunattested you need an idea of what these common patterns are andproviding such survey information is the domain of typology While onlya very small fraction of the roughly 7000 languages spoken in the worldhave been studied in a way that yields useful information for phonologicaltypology crosslinguistic studies have revealed many recurrent patternswhich form the basis for theorizing about the reason for these patterns

71 Inventories

A comparative typological approach is often employed in the study ofphonological segment inventories It has been observed that certain kindsof segments occur in verymany languages while others occur in only a fewThis observation is embodied in the study ofmarkedness which is the ideathat not all segments or sets of segments or rules have equal status inphonological systems For example many languages have the stop conson-ants [p t k] a system that is said to be unmarked but relatively few have theuvular [q] which is said to bemarked Markedness is a comparative conceptso [q] ismoremarked than [k] but lessmarked than [ʕ] Many languages havethe voiced approximant [l] but few have the voiceless lateral fricative [ɬ] andeven fewer have the voiced lateral fricative [ɮ] Very many languages havethe vowels [i e a o u] not many have the vowels [ɨ œ ʊ ɪ]Related to frequency of segment types across languages is the concept of

implicational relation An example of an implicational relation is thatholding between oral and nasal vowels Many languages have only oralvowels (Spanish German) and many languages have both oral and nasalvowels (French Portuguese) but no language has only nasal vowels thatis the existence of nasal vowels implies the existence of oral vowels Alllanguages have voiced sonorant consonants and some additionally havevoiceless sonorants no language has only voiceless sonorants Or manylanguages have only a voiceless series of obstruents others have bothvoiced and voiceless obstruents but none has only voiced obstruentsThe method of comparing inventories Three methodological issues

need to be borne in mind when conducting such typological studies Firstdetermining what is more common versus less common requires a good-sized random sample of the languages of the world However informationon phonological structure is not easily available for many of the languagesof the world and existing documentation tends to favor certain languages(for example the Indo-European languages) over other languages (those ofNew Guinea)Second it is often difficult to determine the true phonetic values of

segments in a language which you do not know so interpreting a symbolin a grammar may result in error The consonants spelled ltp t kgt may infact be ejective [prsquo trsquo krsquo] but ltp t kgt are used in the spelling system

because p t k are ldquomore basicrdquo segments and the author of a grammarmay notate ejectives with ldquomore basicrdquo symbols if no plain nonejectivevoiceless stops exist in the language This is the case in many Bantulanguages of Southern Africa such as Gitonga and Zulu which contrastphonetically voiceless aspirated and ejective stops ndash there are no plainunaspirated voiceless stops Therefore the ejectives are simply writtenltp t kgt because there is no need to distinguish [p] and [prsquo] This phoneticdetail is noted in some grammars but not in all and if you do not haveexperience with the language and do not read a grammar that mentionsthat ltpgt is ejective you might not notice that these languages have noplain voiceless stopsThird many typological claims are statistical rather than absolute ndash

they are statements about what happens most often and thereforeencountering a language that does not work that way does not falsifythe claim It is very difficult to refute a claim of the form ldquoX is morecommon than Yrdquo unless a very detailed numerical study is undertakenTypical inventories With these caveats here are some general tenden-

cies of phoneme inventories In the realm of consonantal place of articu-lation and using voiceless consonants to represent all obstruents at thatplace of articulation the places represented by [p t k] are the most basicoccurring in almost all languages of the world The next most commonplace would be alveopalatal less common are uvulars dentals and retro-flex coronals least common are pharyngeal All languages have a series ofsimple consonants lacking secondary vocalic articulations The mostcommon secondary articulation is rounding applied to velars then palat-alization relatively uncommon is rounding of labial consonants leastcommon would be distinctive velarization or pharyngealization of con-sonants Among consonants with multiple closures labiovelars like [kp]are the most common clicks though rare seem to be more common thanlinguolabialsIn terms of manners of consonant articulation stops are found in all

languages Most language have at least one fricative (but many Australianlanguages have no fricatives) and the most common fricative is ʃ followedby f and ʃ then x then θ and other fricatives Themost common affricates arethe alveopalatals then the other coronal affricates pf and kx are noticeablyless frequent In terms of laryngeal properties of consonants all languageshave voiceless consonants (inmany the voice onset time of stops is relativelylongand the voiceless stops could be considered tobephonetically aspirated)Plain voiced consonants are also common as is a contrast between voicelessunaspirated and voiceless aspirated stops Ejectives implosives and breathy-voiced consonants aremuch less frequent Among fricatives voicing distinc-tions are not unusual but aspiration breathy voicing and ejection are quitemarkedNearly all languages have at least one nasal consonant but languageswith

a rich system of place contrasts among obstruents may frequently have asmaller set of contrasts among nasals Most languages also have at least oneof [r] or [l] and typically have the glides [w j] Modal voicing is the unmarkedcase for liquids nasals and glides with distinctive laryngealization or

But when alanguage has onlyone variety ofcoronal thatvariety may well bephonetically dentalor postalveolar

Phonological typology and naturalness 207

devoicingaspiration being uncommon Among laryngeal glides [h] is themost common then [ʔ] followed by the relatively infrequent [ɦ]The optimal vowel system would seem to be [i e a o u] and while the

mid vowels [e o] are considered to be more marked than the high vowels[i u] for various reasons having to do with the operation of phonologicalrules (context-free rules raising mid vowels to high are much morecommon than context-free rules lowering high vowels to mid) there arefewer languages with just the vowels [i u a] than with the full set [i u e o a]The commonness of front rounded and back unrounded vowels is correl-ated with vowel height so a number of languages have [y] and not [oslash] butvery few have [oslash] and not [y] Full exploitation of the possibilities for lowback and round vowels [ae ɶ a ɒ] is quite rare but it is not hard to findlanguages with [i y ɨ u] As noted earlier oral vowels are more commonthan nasal vowels and modal voiced vowels are more common thancreaky-voiced or breathy vowels

72 Segmental processes

Recurrent patterns are also found in rules themselves We begin ourtypological survey of processes with segmental processes and proceed toprosodic ones Put roughly segmental phonology deals with how thefeatures of one segment affect the features of another segment andprosodic processes are those that pertain to the structure of syllablesstress and the rhythmic structure of words and phenomena whichrelate to the position of segments in a phonological string This divisionof processes is at this point strictly heuristic but research has shownthat there are important representational differences between segmen-tal ie featural representations and syllabic or rhythmic representa-tions ndash further questions regarding representations are taken up inchapter 9

721 AssimilationsThe most common phonological process in language is assimilationwhere two segments become more alike by having one segment take onvalues for one or more features from a neighboring segment

Vowel harmony An example of assimilation is vowel harmony and thearchetypical example of vowel harmony is the frontndashback vowel harmonyprocess of Turkish In this language vowels within a word are (generally)all front or all back and suffixes alternate according to the frontnessof the preceding vowel The genitive suffix accordingly varies between-in and -ɨn as does the plural suffix lar ~ ler

(1) Nom sg Gen sg Nom pl Gen plip ip-in ip-ler ip-ler-in lsquoropersquotʃɨkiʃ tʃɨkɨʃ-ɨn tʃɨkɨʃ-lar tʃɨkɨʃ-lar-ɨn lsquoexitrsquokɨz kɨz-in kɨz-lar kɨz-lar-ɨn lsquogirlrsquo

This process can be stated formally as (2)

A second kind of vowel harmony found in Turkish is rounding harmonyIn Turkish a rule assimilates any high vowel to the roundness of thepreceding vowel Consider the following data involving stems which endin round vowels

The genitive suffix which has a high vowel becomes rounded when thepreceding vowel is round but the plural suffix which has a nonhigh voweldoes not assimilate in roundness Thus the data in (3) can be accounted forby the following rule

A problem that arises in many vowel harmony systems is that it is diffi-cult if not impossible to be certain what the underlying vowel of the suffixis For the plural suffix we can surmise that the underlying vowel isnonround since it is never phonetically round so the most probablehypotheses are a or e For the genitive suffix any of i ɨ y u would beplausible since from any of these vowels the correct output would resultby applying these rulesIt is sometimes assumed that if all other factors are the same for

selecting between competing hypotheses about the underlying form aless marked (crosslinguistically frequent) segment should be selectedover a more marked segment By that reasoning you might narrow thechoice to i u since ɨ y are significantly more marked than i u Thesame reasoning might lead you to specifically conclude that alternatinghigh vowels are i on the assumption that i is less marked than uhowever that conclusion regarding markedness is not certain Thevalidity of invoking segmental markedness for chosing underlyingforms is a theoretical assumption and does not have clear empirical

ev ev-in ev-ler ev-ler-in lsquohousersquobiber biber-in biber-ler biber-ler-in lsquopepperrsquosap sap-ɨn sap-lar sap-lar-ɨn lsquostalkrsquoadam adam-ɨn adam-lar adam-lar-ɨn lsquomanrsquo

Nom sg Gen sg Nom pl Gen pljyz jyz-yn jyz-ler jyz-ler-in lsquofacersquopul pul-un pul-lar pul-lar-ɨn lsquostamprsquook ok-un ok-lar ok-lar-ɨn lsquoarrowrsquo

son son-un son-lar son-lar-ɨn lsquoendrsquokoslashj koslashj-yn koslashj-ler koslashj-ler-in lsquovillagersquo

rarrV [αback] V [αback]

[+ high][αround] V [αround]

support A further solution to the problem of picking between under-lying forms is that [+high] suffix vowels in Turkish are not specified atall for backness or roundness and thus could be represented with thesymbol I which is not an actual and pronounceable vowel but repre-sents a so-called archiphoneme having the properties of being a voweland being high but being indeterminate for the properties [round] and[back] There are a number of theoretical issues which surround thepossibility of having partially specified segments which we will not gointo hereMongolian also has rounding harmony in this language only nonhigh

vowels undergo the assimilation and only nonhigh vowels trigger theprocess

This rule can be forumlated as in (6)

Typological research has revealed a considerable range of variation inthe conditions that can be put on a rounding harmony rule In Sakhahigh vowels assimilate in roundness to round high and nonhighvowels (cf aγa-lɨɨn lsquofather (associative)rsquo sep-tiin lsquotool (associative)rsquovs oγo-luun lsquochild (associative)rsquo boslashroslash-lyyn lsquowolf (associative)rsquo tynnyk-tyyn lsquowindow (associative)rsquo) but nonhigh vowels only assimilate inroundness to a preceding nonhigh vowel (cf aγa-lar lsquofathersrsquo sep-terlsquotoolsrsquo tynnyk-ter lsquowindowsrsquo kus-tar lsquoducksrsquo vs oγo-lor lsquochildrenrsquo boslashroslash-loslashrlsquowolvesrsquo) As seen in chapter 6 in Yawelmani vowels assimilaterounding from a preceding vowel of the same height (thus highvowels assimilate to high vowels low vowels assimilate to low vowels)As seen in (7) Kirghiz vowels generally assimilate in roundness to anypreceding vowel except that a nonhigh vowel does not assimilate to aback high round vowel (though it will assimilate rounding from afront high round vowel)

Nominative Instrumental Accusativedel del-er del-ig lsquocoatrsquogal gal-ar gal-ig lsquofirersquody dy-ger dy-g lsquoyounger brotherrsquonoslashxoslashr noslashxoslashr-oslashr noslashxoslashr-ig lsquocomradersquodoro doro-gor doro-g lsquostirruprsquo

rarrV [αround] V C0 _

⎡⎢⎣

⎤⎥⎦

-highαround

[- high]

Accusative Dativetaʃ-tɨ taʃ-ka lsquostonersquoiʃ-ti iʃ-ke lsquojobrsquoutʃ-tu utʃ-ka lsquotiprsquo

This survey raises the question whether you might find a language whereroundness harmony only takes place between vowels of different heightsrather than the same height as we have seen Although such examplesare not known to exist we must be cautious about inferring too muchfrom that fact since the vast majority of languages with rounding har-mony are genetically or areally related (Mongolian Kirghiz TurkishSakha) The existence of these kinds of rounding harmony means thatphonological theory must provide the tools to describe them what we donot know is whether other types of rounding harmony also exist Nor is itsafe given our limited database on variation within rounding harmonysystems to make very strong pronouncements about what constitutesldquocommonrdquo versus ldquorarerdquo patterns of rounding harmonyAnother type of vowel harmony is vowel-height harmony Suchharmony

exists in Kuria where the tense mid vowels e o become i u before a highvowel Consider (8) illustrating variations in noun prefixes (omo ~ umu eme~ imi eke ~ ege ~ iki ~ igi ogo ~ ugu) conditioned by the vowel to the right

These examples show that tense mid vowels appear before the low vowel aand the tense and lax mid vowels e ε o ɔ which are [-high] and highvowels appear before high vowels so based just on the phonetic environ-ment where each variant appears we cannot decide what the underlyingvalue of the prefix is [-high] or [+high] Additional data show that theprefixes must underlyingly contain mid vowels there are also prefixeswhich contain invariantly [+high] vowels

Thus the alternations in (8) can be described with the rule (10)

konok-tu konok-ko lsquoguestrsquokoslashz-ty koslashz-goslash lsquoeyersquoyj-ty yj-goslash lsquohousersquo

omooacute-nto lsquopersonrsquo omo-saacuteaacutetʃaacute lsquomalersquoomo-teacute lsquotreersquo omo-goacuteoacutendo lsquoplowed fieldrsquoumu-riacuteiacutesja lsquoboyrsquo umu-muacutera lsquoyoung manrsquoeme-teacute lsquotreesrsquo imi-siacute lsquosugar canesrsquoege-saacuteka lsquostreamrsquo ege-teacute lsquochairrsquoegeeacute-nto lsquothingrsquo igi-tuacuteuacutembe lsquostoolrsquoiki-ruacuteuacuteŋguacuteuri lsquosoft porridgersquo iki-muacuteuacutenέ lsquodeerrsquoogo-gaacutebo lsquohuge basketrsquo ogo-taacutebo lsquohuge bookrsquoogo-seacuteeacutendaacuteno lsquohuge needlersquo ogo-gέna lsquohuge stonersquougu-siacuteri lsquohuge ropersquo

iri-tɔɔkε lsquobananarsquo iri-kέέndɔ lsquodate fruitrsquoiri-hiacuteiacutendi lsquocorn cobrsquo iri-toacutero lsquobuttockrsquoibi-goacuteoacutendo lsquosmall fieldsrsquo ibi-gaacuteaacutetε lsquosmall breadsrsquoibi-guacuteruacutebe lsquosmall pigsrsquo ibi-teacute lsquochairsrsquoitʃi-seacuteeacuteseacute lsquodogrsquo itʃi-ŋaacuteaacutemwi lsquocatrsquoitʃi-ŋɔɔmbε lsquocowrsquo itʃii-ŋguacuteruacutebe lsquopigrsquo

Another variety of vowel-height harmony is complete height harmony anexample of which is found in Matuumbi This language distinguishes fourphonological vowel heights exemplified by the vowels a ε ɪ and i Thevowels of the passive suffix -ilw- and the causative suffix -ij- assimilatecompletely to the height of the preceding nonlow vowel [ε ɪ i]

This process involves the complete assimilation of suffix vowels to the valuesof [high] and [tense] (or [ATR]) from the preceding nonlow vowel Since thelow vowel a does not trigger assimilation the context after a reveals theunderlying nature of harmonizing vowels which we can see are high andtense The following rule will account for the harmonic alternations in (11)

Akan exemplifies a type of vowel harmony which is common especiallyamong the languages of Africa which is assimilation of the feature ATRIn Akan vowels within the word all agree in their value for [ATR] In (13a)the prefix vowels are [+ATR] before the [+ATR] vowel of the word for lsquoeatrsquoand [-ATR] before the [-ATR] vowel of lsquobe calledrsquo (13b) shows this sameharmony affecting other tensendashaspect prefixes

(13) a

aacutesim-a lsquoborrowrsquo aacutesim-ilw-a lsquobe borrowedrsquoiacuten-a lsquodancersquo ɪ n-ilw-a lsquobe dancedrsquokuacuten-a lsquograte coconutrsquo kuacuten-ilw-a lsquobe gratedrsquoʊʊg-a lsquobathersquo ʊʊg-ɪlw-a lsquobe bathedrsquotwɪ ɪk-a lsquolift a loadrsquo twɪɪk-ɪlw-a lsquobe liftedrsquobɔ ɔl-a lsquotear bark

off a treersquobɔɔl-εlw-a lsquobe de-barkedrsquo

kέεŋgεεmb-a

lsquouproot tubersrsquo kέεŋgεεmb-εlw-a

lsquobe uprootedrsquo

tʃaacuteag-a lsquogrindrsquo tʃaacuteag-ij-a lsquomake grindrsquotʃiacuteindʒ-a lsquoslaughterrsquo tʃiacuteindʒ-ij-a lsquomake slaughterrsquoʊʊg-a lsquobathersquo ʊʊg-ɪj-a lsquomake bathersquobɔ ɔl-a lsquode-barkrsquo bɔɔl-εj-a lsquomake de-barkrsquotʃέεŋg-a lsquobuildrsquo tʃέεŋg-εj-a lsquomake buildrsquo

rarr [+ high] _C0 V V[+ tense] [+ high]

⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

αhighβtense

rarr C0 _ ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+lowαhighβtense

lsquoeatrsquo lsquobe calledrsquo1sg mi-di mɪ-dɪ2sg wu-di wʊ-dɪ3sg o-di ɔ-dɪ1pl je-di jε-dɪ

Vowel nasalization is also a common assimilatory process affectingvowels and can be seen in the data of (14) from Gatilde These data illustratenasalization affecting the plural suffix which is underlyingly i andassimilates nasality from the immediately preceding vowel

Another kind of vowel harmony one affecting multiple features is some-times termed ldquoplace harmonyrdquo an example of which comes from Efik InEfik the prefix vowel ε (but not e) becomes [a] before [a] [ɔ] before [ɔ] [ε]before [ε] [e] before [e] and [i] and [o] before [o] and [u]

This process involves assimilation of all features from the followingvowel except the feature [high]

Finally complete vowel harmony where one vowel takes on all featuresfrom a neighboring vowel is found in some languages such as KolamiThis language has a rule of vowel epenthesis which breaks up final

2pl mu-di mʊ-dɪ3pl wo-di wɔ-dɪ

o-be-di lsquohe will eatrsquo ɔ-bε-dɪ lsquohersquoll be calledrsquoo-di-i lsquohe atersquo ɔ-dɪ-ɪ lsquohe was calledrsquoo-ko-di lsquohe goes and eatsrsquo ɔ-kɔ-dɪ lsquohe goes and is calledrsquo

mlεεbo mlεεbo-i lsquoliverrsquonatildene natildene-i lsquolegrsquotʃĩĩsi tʃĩĩsi-i lsquoplatersquoakplɔ akplɔ-i lsquospearrsquogbε gbε-i lsquopathrsquomĩ mĩ-ĩ lsquodrumrsquo

sε sε-ĩ lsquothroatrsquotũ tũ-ĩ lsquogunrsquoŋmɔ ŋmɔ-ĩ lsquofarmrsquo

lematilde lematilde-ĩ lsquoaxrsquo

3sg 3ple-di e-di lsquocomersquoε-bεri e-bεri lsquoshutrsquoa-kaŋ e-kaŋ lsquodenyrsquoɔ-bɔ e-bɔ lsquotakersquoo-kop e-kop lsquohearrsquoo-kut e-kut lsquoseersquo

ε rarr C0 ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

αroundβtenseγback

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

consonant clusters and medial clusters of more than two consonants Theinserted vowel harmonizes with the preceding vowel

Another example of complete vowel harmony is seen in the followingexamples of the causative prefix of Klamath whose vowel completelyassimilates to the following vowel

Complete harmony is unlikely to ever be completely general - all of theseexamples are restricted in application to specific contexts such as epen-thetic vowels as in Kolami or vowels of specific affixal morphemes as inKlamath Another context where total harmony is common is betweenvowels separated only by laryngeal glides h and ʔ a phenomenon referredto as translaryngeal harmony as illustrated in Nenets by the alternationin the locative forms to-hona lsquolakersquo pi-hina lsquostreetrsquo pja-hana lsquotreersquo pe-henalsquostonersquo tu-huna lsquofirersquo The consequences of a completely unrestrictedvowel harmony would be rather drastic - any word could only have onekind of vowel in it were such a rule to be totally general

Consonant assimilations One of the most common processes affectingconsonants is the assimilation of a nasal to the place of articulation of thefollowing consonant An example of this process comes from Matuumbiseen in (19) where the plural prefix ɲ takes on the place of assimilation ofthe following consonant

Stem 1sg pres 1sg past Imperativetum tum-atun tum-tan tum lsquosneezersquoagul agul-atun agul-tan agul lsquodigrsquodakap dakap-atun dakap-tan dakap lsquopushrsquokatk katk-atun katak-tan katak lsquostrikersquomelg melg-atun meleg-tan meleg lsquoshakersquokink kink-atun kinik-tan kinik lsquobreakrsquo

sna-batgal lsquogets someone up from bedrsquosne-lrsquoemlrsquoema lsquomakes someone dizzyrsquosno-bostgi lsquocauses something to turn blackrsquosni-nklilkrsquoa lsquomakes dustyrsquo

Singular Plurallwɪɪ mo ɲɪ mo lsquoland being weededrsquolwaaacutembo ɲaaacutembo lsquobeadrsquolweeacutembe ɲeeacutembe lsquoshaving knifersquolugoloacutekaacute ŋgoloacutekaacute lsquostraightrsquolubaacuteu mbaacuteu lsquoribrsquoludʒiiacuteŋgjaacute ɲdʒiiacuteŋgjaacute lsquoenteredrsquolulaaacutela ndaaacutela lsquopepperrsquolupalaacuteaiacute mbalaacuteaiacute lsquobald headrsquoluteacuteelaacute ndeelaacute lsquopiece of woodrsquolutʃwiiacutetʃwi ɲdʒwiiacutetʃwi lsquotomatorsquo

Place assimilation of nasals in Matuumbi affects all nasals so the data in(20a) illustrate assimilation of preconsonantal n resulting from anoptional vowel deletion rule and (20b) illustrates assimilation of m

(20) a

Sometimes a language with place assimilation of nasals will restrict theprocess to a specific place of articulation For instance Chukchi assi-milates ŋ to a following consonant but does not assimilate n or mThus the stem teŋ lsquogoodrsquo retains underlying ŋ before a vowel andotherwise assimilates to the following consonant however as thelast two examples show n and m do not assimilate to a followingconsonant

A common assimilation affecting consonants after nasals is postvocalicvoicing illustrated by Matuumbi in (22) The data in (22a) illustratevoicing of an underlyingly voiceless consonant at the beginning of a stemafter the prefix ɲ The data in (22b) show voicing of a consonant in a verbafter the reduced form of the subject prefix ni In these examples thevowel i in the prefix optionally deletes and when it does it voices aninitial stop

(22) a

lukiacuteligo ŋgiliacutego lsquoplace for initiatesrsquolukiacuteli ŋgiacuteli lsquopalmrsquo

ni-baacutelaaŋgite m-baacutelaaŋgite lsquoI countedrsquoni-dʒiacuteiŋgiile ɲ-dʒiacuteiŋgiile lsquoI enteredrsquoni-goacuteoɲdʒite ŋ-goacuteoɲdʒite lsquoI sleptrsquo

mu-paacutealite m-paacutealite lsquoyou (pl) wantedrsquomu-teacuteliike n-teacuteliike lsquoyou (pl) cookedrsquomu-tʃaacuteawiile ɲ-tʃaacuteawiile lsquoyou (pl) groundrsquomu-kaacuteatite ŋ-kaacuteatite lsquoyou (pl) cutrsquo

teŋ-əɬʔ-ən lsquogoodrsquotam-waγərγ-ən lsquogood lifersquotam-pera-k lsquoto look goodrsquotan-tʃottʃot lsquogood pillowrsquo

tan-ɬəmŋəɬ lsquogood storyrsquotan-rʔarqə lsquogood breastbandrsquonə-mkə-kin lsquooftenrsquoγa-n-pera-w-ɬen lsquodecoratedrsquo

Singular Plurallu-palaacuteaiacute m-balaacuteaiacute lsquobald headrsquolu-tʃwiiacutetʃwi ɲ-dʒwiiacutetʃwi lsquotomato plantrsquolu-teacuteelaacute n-deelaacute lsquopiece of woodrsquolu-kiacuteligo ŋ-giliacutego lsquoinitiatersquos placersquolu-temaacuteaacute n-demaacuteaacute lsquochoppedrsquolu-tʃapiacuteitʃaacute ɲ-dʒapiacuteitʃaacute lsquocleanrsquo

Not allpreconsonantalnasals conditionthis voicing processin Matuumbi onlynasals which arenonsyllabic in theintermediaterepresentation doHence [mp]sequences such asfound in (20)are possible sincethe process thatdeletes the vowel uresults in a syllabicnasal in theintermediaterepresentation

Stop consonants frequently nasalize before nasal consonants and anexample of this process is found in Korean The examples in (23a) arestems with final nasal consonants those in (23b) have oral consonantsrevealed before the infinitive suffix a ~ ə and undergo nasalization of thatconsonant before the past-tense suffix -nɨnta

(23) a

Matuumbi presents the mirror-image process of postnasal nasalization(this process is only triggered by nasals which are moraic in the intermedi-ate representation) On the left in (24a) the underlying consonant isrevealed when a vowel-final noun-class prefix stands before the stemand on the right a nasal prefix stands before the stem causing the initialconsonant to become nasalized In (24b) nasalization applies to theexample in the second column which undergoes an optional rule deletingthe vowel u from the prefix mu

(24) a

Many languages have a process of voicing assimilation especially inclusters of obstruents which must agree in voicing Most often obstruents

1sg past Optional pronunciationni-paacutealite m-baacutealite lsquoI wanted (recent)rsquoni-teacuteliike n-deacuteliike lsquoI cooked (recent)rsquoni-tʃoacutenite ɲ-dʒoacutenite lsquoI sewed (recent)rsquoni-kʊbiile ŋ-gʊbiile lsquoI hit on legs (recent)rsquo

Infinitive Pastan-a an-nɨnta lsquohugrsquotrsquoatɨm-ə trsquoatɨm-nɨntə lsquotrimrsquo

nəm-ə nəm-nɨnta lsquooverflowrsquo

tʃhəm-a tʃham-nɨnta lsquoendurersquo

ip-ə im-nɨnta lsquowearrsquotat-ə tan-nɨnta lsquoclosersquoputh-ə pun-nɨnta lsquoadherersquotʃotʃh-a tʃon-nɨnta lsquofollowrsquo

mək-ə məŋ-nɨnta lsquoeatrsquotakrsquo-a taŋ-nɨnta lsquopolishik-ə iŋ-nɨnta lsquoripenrsquo

a-baaacutenda lsquoslavesrsquo m-maaacutenda lsquoslavesrsquoa-laacutealo lsquofoolsrsquo n-naacutealo lsquofoolrsquoa-guacutendumuacuteji lsquoscarersrsquo ŋ-ŋuacutendumuacuteji lsquoscarerrsquomi-butuacuteka lsquocarsrsquo m-mutuacuteka lsquocarrsquomi-daacutelaaacutenzi lsquobitter orangesrsquo n-naacutelaaacutenzi lsquobitter orangersquomi-lipuacute lsquotrees (sp)rsquo n-nipuacute lsquotree (sp)rsquomi-guacuteunda lsquofieldsrsquo ŋ-ŋuacuteunda lsquofieldrsquo

mu-buundiacuteke m-muundiacuteke lsquoyou should storersquomu-laabuacuteke n-naabuacuteke lsquoyou should breakfastrsquomu-dʒiiŋgiacute ɲ-ɲiiŋgiacute lsquoyou should enterrsquo

assimilate regressively to the last obstruent in the cluster For example inSanskrit a stem-final consonant reveals its underlying voicing when thefollowing affix begins with a sonorant but assimilates in voicing to afollowing obstruent

Other languages with regressive voicing assimilation are Hungarian andRussianProgressive voicing harmony is also possible though less common than

regressive voicing One example of progressive assimilation is found inNorwegian The (regular) past-tense suffix is -te and it shows up as suchwhen attached to a stem ending in a sonorant or voiceless consonant butafter a voiced obstruent the suffix appears as -de

Another example of progressive voicing harmony is found in Evenki whereanunderlyingly voiced suffix-initial consonant becomesdevoiced after a voice-less obstruent this is illustated below with the accusative case suffix ba

Complete assimilation of a consonant to a following consonant is found inArabic In the data of (28) from the Syrian dialect the consonant l of thedefinite article assimilates completely to a following coronal consonantExamples in (a) show nonassimilation when the following consonant is non-coronal and those in (b) provide stems that begin with coronal consonants

krɳʈ-mas bhind-mas 1pl indicative activekrɳʈ-e bhind-e 1sg indicative middlekrɳʈ-ʈha bhint-tha 2pl indicative activekrɳʈ-ʈe bhint-te 3sg indicative middlekrɳɖ-ɖhve bhind-dhve 2pl indicative middlelsquoweaversquo lsquobindrsquo

smil-e smil-te lsquosmilersquo svoslashm-e svoslashm-te lsquoswimrsquo

hoslashr-e hoslashr-te lsquoheardrsquo lon-e lon-te lsquoborrowrsquo

les-e les-te lsquoreadrsquo spis-e spis-te lsquoeatrsquoreis-e reis-te lsquotravelrsquo ccediloslashp-e ccediloslashp-te lsquobuyrsquotenk-e tenk-te lsquothinkrsquo behoslashv-e behoslashv-de lsquobelongrsquolev-e lev-de lsquolivedrsquo proslashv-e proslashv-de lsquotryrsquobygg-e byg-de lsquobuildrsquo hugg-e hug-de lsquochoprsquognag-e gnag-de lsquognawrsquo krev-e krev-de lsquorequestrsquosag-e sag-de lsquosawrsquo plag-e plag-de lsquoafflictrsquo

asi-ba lsquowomanrsquo ŋami-ba lsquofemale deerrsquopalatka-ba lsquotentrsquo tolgolkil-ba lsquosledsrsquober-be lsquoonionrsquo hunat-pa lsquogirlrsquodet-pe lsquotundrarsquo mit-pe lsquo1pl inclusiversquo

Indefinite Definite Indefinite Definitehawa lhawa lsquoairrsquo bared lbared lsquocoldrsquoʔadham lʔadham lsquoblackrsquo madine lmadine lsquocityrsquo

Consonants are also often susceptible to assimilation of features from aneighboring vowel especially place features of a following vowel Oneprocess is palatalization found in Russian A consonant followed by afront vowel takes on a palatal secondary articulation from the vowel asthe following data show

A second kind of palatalization is found in many languages where typic-ally velar but in some languages also alveolar consonants become alveo-palatals to avoid confusion with the preceding type of palatalization assecondary articulation this latter process is often referred to as corona-lization This process is found in Russian it is triggered by some deriv-ational suffixes with front vowels but not all suffixes

ʕade lʕade lsquocustomrsquo hara lhara lsquoquarterrsquowaħʃ lwaħʃ lsquobeastrsquo jaʔs ljaʔs lsquodespairrsquokalb lkalb lsquodogrsquo xadd lxadd lsquocheekrsquofajj lfajj lsquoshadowrsquo γada lγada lsquolunchrsquo

sʕaff sʕsʕaff lsquorowrsquo talet ttalet lsquothirdrsquotaxt ttaxt lsquobedrsquo raʔbe rraʔbe lsquoneckrsquonəde nnəde lsquodewrsquo life llife lsquoloofahrsquodifaʕ ddifaʕ lsquodefensersquo smike ssmike lsquothickʃoraba ʃʃoraba lsquosouprsquo ʒamil ʒʒamil lsquoprettyrsquozaki zzaki lsquobrightrsquo tʕaleb tʕtʕaleb lsquostudentrsquozʕabet zʕzʕabet lsquoofficerrsquo dʕahuk dʕdʕahuk lsquojollyrsquo

vkus lsquotastersquo vkusj-en lsquotastyrsquoum lsquointellectrsquo umj-en lsquocleverrsquogolot d lsquohungerrsquo golodj-en lsquohungryrsquostol lsquotablersquo stolj-e lsquotable (loc)rsquoguba lsquoliprsquo gubj-e lsquolip (loc)rsquomesto lsquoplacersquo mestj-e lsquoplace (loc)rsquoglub-ok lsquodeeprsquo glubj-ina lsquodepthrsquoton-ok lsquothinrsquo tonj-ina lsquothinnessrsquovor lsquothief vorj-iska lsquothief (pejorative)rsquodom lsquohousersquo domj-iska lsquohouse (pejorative)rsquogorot d lsquotownrsquo gorodj-iska lsquotown (pejorative)rsquo

druk g lsquofriendrsquo druʒ-itj lsquoto be friends withrsquomuka lsquotormentrsquo mutʃj-itj lsquoto tormentrsquogrjex lsquosinrsquo grjeʃ-itj lsquoto sinrsquostrok g lsquostrictrsquo stroʒ-e lsquostricterrsquodjik lsquowildrsquo djitʃj-e lsquowilderrsquosux lsquodryrsquo suʃ-e lsquostricterrsquokrut lsquosteeprsquo krutʃj-e lsquosteeperrsquogad-ok lsquofoulrsquo gaʒ-e lsquofoulerrsquovɨs-ok lsquotallrsquo vɨʃ-e lsquotallerrsquonjiz-ok lsquolowrsquo njiʒ-e lsquolowerrsquo

The alveopalatalfricatives ʃ ʒ arenot phoneticallypalatalizable inRussian whereasthe alveopalatalaffricate is alwayspalatalized

Another common vowel-to-consonant effect is affrication of coronalobstruents before high vowels An example of this is found in Japanesewhere t becomes [ts] before [u] and [tʃ] before [i]

Outside the domain of assimilations in place of articulation the mostcommon segmental interaction between consonants and vowels (or some-times other sonorants) is lenition or weakening Typical examples oflenition involve either the voicing of voiceless stops or the voicing andspirantization of stops the conditioning context is a preceding vowelsometimes a preceding and following vowel An example of the spiranti-zation type of lenition is found in Spanish where the voiced stops b d gbecome voiced spirants [β eth γ] after vocoids

This can be seen as assimilation of the value [continuant] from a precedingvocoidAn example of combined voicing and spirantization is found in

Tibetan where voiceless noncoronal stops become voiced spirantsbetween vowels

In some cases the result of lenition is a glide so in Axininca Campa stem-initial k p become [j w] after a vowel

The converse process whereby spirants sonorants or glides become obstru-ent stops after consonants is also found in a number of languages ndash thisprocess is generally referred to as hardening In Matuumbi sonorantsbecome voiced stops after a nasal The data in (35) illustrate this

Negative Provisional Infinitive Volitionalmat-anai mat-eba mats-u matʃ-itai lsquowaitrsquotat-anai tat-eba tats-u tatʃ-itai lsquostandrsquokat-anai kat-eba kats-u katʃ-itai lsquowinrsquo

N with N there are Nrsquosburro kom burro aj βurros lsquodonkeyrsquodeetho kon deetho aj etheethos lsquofingerrsquogato koŋ gato aj γatos lsquocatrsquo

Past affirmative Past negativetʃaa-βəree ma-tʃaa-βəree lsquogorsquopaa-βəree ma-βaa-βəree lsquolightrsquopii-βaree mə-βii-βəree lsquorenouncersquokuu-βəree mə-γuu-βəree lsquowaitrsquokə-βəaree mə-γə-βəree lsquohidersquoqɔɔ-βəree ma-ʁɔɔ-βəree lsquotake time outrsquo

jaarato lsquoblack beersquo no-jaaratoti lsquomy black beersquokanari lsquowild turkeyrsquo no-janariti lsquomy wild turkeyrsquoporita lsquosmall henrsquo no-woritati lsquomy small henrsquo

phenomenon with the alternation in stem-initial consonant found betweenthe singular and plural

Another context where hardening is common is when the consonant isgeminate One example is found in Fula where geminate spirants becomestops In (36) plural forms have a medial geminate (this derives by anassimilation to a following ɗ so that [tʃabbi] derives from tʃaw- ɗ i via theintermediate stage t ʃawwi)

Geminate hardening also occurs in Ganda In the data of (37) the singularform of nouns in this particular class is formed by geminating the initialconsonant the underlying consonant is revealed in the plural

In this language only sonorants harden to stops

722 DissimilationLess common in the languages of the world are processes of dissimilationwhereby one of two similar consonants changes to become less like theother An example of such a process is lateral dissimilation as found in

lu-laaacutela lsquopepper plantrsquo n-daaacutela lsquopepper plantsrsquolu-jɪ maacute lsquopolersquo ɲ-dʒɪ ma lsquopolesrsquojuacutekuta lsquoto be fullrsquo ɲ-dʒukuacutetaacute lsquofullrsquowaacute lsquoto diersquo ŋ-gwaaacuteaacute lsquodeadrsquolɪ ndɪɪla lsquoto guardrsquo n-dɪndɪ ɪlaacute lsquoguardedrsquo

Plural Diminutive singulartʃabbi tʃawel lsquostickrsquolebbi lewel lsquomonthrsquopobbi powel lsquohyenarsquoɲεbbε ɲewel lsquobeanrsquoleppi lefel lsquoribbonrsquokoppi kofel lsquoearrsquotʃoppi tʃofel lsquochickrsquo

Singular Pluralggi ma-gi lsquoeggrsquoddaala ma-daala lsquoladderrsquoddʒuba ma-juba lsquodoversquoggwaanga ma-waanga lsquonationrsquoddaanga ma-laanga lsquolilyrsquo

Singular Pluralffumu ma-fumu lsquospearrsquoffuumbe ma-fuumbe lsquocivetrsquossaandʒa ma-saandʒa lsquodry plantain leafrsquozzike ma-zike lsquochimpanzeersquozziga ma-ziga lsquotearrsquovviivi ma-viivi lsquokneersquo

Sundanese In this language the plural is formed by infixing -ar- after theinitial consonant as seen in (39a) When another r follows within thestem the r of the infix dissimilates to l

A similar process affects the adjectival suffix -alis in Latin where ldissimilates to [r] if the preceding stem contains another l

Dissimilation of aspiration is attested in other languages such as ManipuriIn (41) the first consonant of the directional suffixes -thok and -khət deaspi-rates if preceded by another aspirate or h (and if the immediately precedingsegment is a vowel or sonorant the consonant becomes voiced)

Many Bantu languages such as Kuria have a voicing dissimilation pro-cess whereby k becomes g when the following syllable has a voicelessconsonant (excluding h) This results in alternations in the form of theinfinitive prefix which is underlyingly oko as well as the second-singular object prefix ko and the (diminutive) object prefix ka Thedata in (42a) motivate the underlying prefix oko and (42b) showsapplication of dissimilation to the prefix (42c) shows the object prefixesko and ka which also dissimilate and (42d) shows the contrastingprefixes go and ga which have underlyingly voiced consonants anddo not assimilate

Singular Pluralkusut k-ar-usut lsquomessyrsquopoho p-ar-oho lsquoforgetrsquogətol g-ar-ətol lsquodiligentrsquoŋoplok ŋ-ar-oplok lsquoflop downrsquoŋuliat ŋ-ar-uliat lsquostretchrsquotuwaŋ t-ar-uwaŋ lsquoeatrsquomasak m-ar-asak lsquocookrsquo

ŋɨrɨt ŋ-al-ɨrɨt lsquocutrsquonugar n-al-ugar lsquodig uprsquotʃombrek tʃ-al-ombrek lsquocoldrsquobotʃor b-al-otʃor lsquoleakingrsquobɨŋhar b-al-ɨŋhar lsquorichrsquohormat h-al-ormat lsquorespectrsquo

navalis lsquonavalrsquo episcopalis lsquoepiscopalrsquosolaris lsquosolarrsquo militaris lsquomilitaryrsquolupanaris lsquowhorishrsquo

pi-thok lsquogive outrsquo pi-khət lsquogive upwardsrsquocət-thok lsquogo outrsquo cət-khət lsquogo upwardsrsquokhik-tok lsquosprinkle outrsquo khik-kət lsquosprinkle upwardsrsquohut-tok lsquobore outrsquo hut-kət lsquobore upwardsrsquokhoj-dok lsquotrim outrsquo khoj-gət lsquotrim upwardsrsquothin-dok lsquopierce outrsquo thin-gət lsquopierce upwardsrsquo

(42) a

The language Chukchi has a number of dissimilatory processes One ofthese dissimilates nasality by changing ŋ to γ before a nasal

A second dissimilation in the language changes the first in a sequence ofidentical fricatives to a stop

An important feature of this rule is that only homorganic clusters dissimi-late Other combinations such as yγ wɬ or ɬγ remain unchanged

Finally the glide j dissimilates to γ before a coronal consonant

oko-reacutema lsquoto cultivatersquo uku-ɲaacutehaareacuteka lsquoto be hurtrsquouku-miacuteɲooŋgoacutera lsquoto crush uku-giacuteiacuteŋgiacuteraacute lsquoto shaversquooko-gɔ ɔgaacute lsquoto slaughterrsquo uku-buacutena lsquoto breakrsquooko-bɔtʃa lsquoto vomitrsquo oko-hoacuteoacutera lsquoto threshrsquo

ogo-taacuteaacuteŋgaacute lsquoto beginrsquo ugu-tuacuteuacutehaacute lsquoto be bluntrsquougu-suacuteraaŋgaacute lsquoto sing praisersquo ogo-sέέnsaacute lsquoto winnowrsquo

ugu-kjaacute lsquoto dawnrsquo ogo-kέɲa lsquoto runrsquo

ogo-koacute-baacuterǎ lsquoto count you sgrsquouku-guacute-suacuteraaacuteŋga lsquoto praise you sgrsquoogo-kaacute-baacuterǎ lsquoto count itrsquooko-gaacute-suacuteraaacuteŋga lsquoto praise itrsquo

oko-goacute-baacuterǎ lsquoto count itrsquo uku-guacute-suacuteraaacuteŋga lsquoto praise itrsquooko-gaacute-baacuterǎ lsquoto count themrsquo oko-gaacute-suacuteraaacuteŋga lsquoto praise themrsquo

taraŋ-ək lsquobuild a dwellingrsquo nə-tarəγ-more lsquowe built a dwellingrsquomətɬəŋ-ən lsquofiversquo mətɬəγ-more lsquowe fiversquoenawrəŋ-ək lsquoto give as a giftrsquo enawrəγ-nen lsquohe gave itrsquopetʔiŋ lsquocoldrsquo petʔiγ-ŋinqey lsquoboy with a coldrsquo

meniγ lsquoclothrsquo manek-γəpə lsquofrom clothrsquoətɬəγ-ən lsquofatherrsquo ətɬək-γəjiwq-ew lsquopaternal markingrsquorəγrəγ lsquowoolrsquo rəγrək-γəpə lsquofrom woolrsquojeγteɬ-ək lsquoto liversquo γe-jeγtet-ɬin lsquohe livedrsquoɬəmŋəɬteɬ-ək lsquotell storiesrsquo γa-ɬəmŋəɬtet-ɬen lsquotold storiesrsquoŋew-ʔen lsquowomanrsquo ŋak-waŋe-γərγ-ən lsquowomanrsquos sewingrsquoiɬγətew-ək lsquoto washrsquo iɬγətew-wʔi lsquohe washedrsquo

kətəjγat-ək lsquoblowrsquo γa-n-pera-w-ɬen lsquodecoratedrsquoʔiw-pipiq-əɬγ-ən lsquowolf mousersquo

wʔej-ək lsquograssrsquo wʔeγ-ti lsquograssesrsquoŋin-qej lsquoboyrsquo ŋen-qaγ-tʃəŋ-ən lsquobig boyrsquotʃaj lsquotearsquo tʃaγ-naɬk-ək lsquoto make tearsquoqej-we lsquocorrectrsquo qeγ-ɬənanγet lsquotruthqəjəqej lsquonestlingrsquo qaγ-jaʔjaq lsquoyoung seagullrsquo

Dissimilation between vowels is also found in languages One case comesfrom Woleiaian where the low back vowel a becomes [e] before thelow back vowels a and ɒ This process affects the causative prefix gaseen below

In Wintu the vowels e o become [i u] before a by a similar kind ofdissimilation

Examples of low vowel dissimilating to nonlow vowels before low vowels arealso found in Kera and Southern Russian Interestingly most examples ofdissimilation between vowels are precisely of this nature we do not seem tofind cases of high vowels dissimilating to nonhigh near other high vowels

723 Other segmental processesThere are other segmental processeswhich donot neatlyfit into the categoryof assimilation or dissimilation One such example is neutralizationwhereby a phonetic contrast is deleted in some context which consonantsare particularly susceptible to One case is the neutralization of laryngealcontrasts in consonants at the end of the syllable as exemplified by Korean

Another kind of neutralization is place neutralization which can beexemplified by Saami Saami restricts word-final consonants to the set tn r l s ʃ ie the voiceless coronal nonaffricates The data in (50) show

ga-repa lsquoapproach itrsquo ga-beʃi lsquoheat itrsquoga-sɨwe lsquomake it standrsquo ga-sere lsquomake it hitrsquoge-bbaro lsquobend itrsquo ge-makɨ lsquogive birth to himrsquo

ge-mɒwe lsquoerase itrsquo ge-tɒtɒwe lsquosupport itrsquoge-wasɨr lsquohurt itrsquo ge-tɒla lsquomake it bloomrsquo

lel-a lila lsquoto transformrsquo

lel-u lelu lsquotransformrsquolel-it lelit lsquotransformedrsquodek-a dika lsquoto climbrsquodek dek lsquoclimbrsquodek-na dekna lsquoto steprsquodoj-a duja lsquoto giversquodoj-u doju lsquogiversquodoj-i doji lsquogiftrsquo

Infinitive Conjunctiveip-ə ip-krsquoo lsquowearrsquokaph-a kap-krsquoo lsquopay backtat-ə tat-krsquoo lsquoclosersquoputh-ə put-krsquoo lsquoadherersquotʃotʃh-a tʃot-krsquoo lsquofollowrsquo

mək-ə mək-krsquoo lsquoeatrsquotakrsquo-a tak-krsquoo lsquopolish

that noun stems can end in an array of consonants as revealed by theessive form of the noun which takes the suffix -(i)n but in the nominativewhich has no suffix all places of articulation are neutralized to coronal

It is interesting that Saami also neutralizes laryngeal contrasts finally sovoiced stops become voiceless it is unknown whether a language mayexhibit neutralization of place contrasts without also having neutraliza-tion of laryngeal contrasts

73 Prosodically based processes

A second major class of phonological processes can be termed ldquoprosodi-cally motivated processesrdquo Such processes have an effect on the structureof the syllable (or higher prosodic units such as the ldquofootrdquo) usually byinserting or deleting a consonant or changing the status of a segmentfrom vowel to consonant or vice versa

Vowel sequences A very common set of prosodic processes is the classof processes which eliminate V+V sequences Many languages disallowsequences of vowels and when such sequences would arise by the com-bination of morphemes one of the vowels is often changed One of themost common such changes is glide formation whereby a high vowelbecomes a glide before another vowel Quite often this process is accom-panied with a lengthening of the surviving vowel a phenomenon knownas compensatory lengthening For example in Matuumbi high vowelsbecome glides before other vowels as shown by the data in (51) Theexamples on the left show that the noun prefixes have underlying vowelsand those on the right illustrate application of glide formation

Nominative sg Essiveoahpis oahpis-in lsquoacquaintancersquotʃoarvvuʃ tʃoarvvuʃ-in lsquoantlers and skullcaprsquogahpir gahpir-in lsquocaprsquoheevemĕahhtun heevemĕahhtun-in lsquoinappropriatersquovarit varih-in lsquo2-year-old reindeer buckrsquotʃuojvvat tʃuojvvag-in lsquoyellow-brown reindeerrsquoahhkut ahhkub-in lsquograndchild of womanrsquolottaeligʃ lottaeligdʒ-in lsquosmall birdrsquosuohkat suohkaeth-in lsquothickjaeligʔmin jaeligʔmim-in lsquodeathrsquo

mi-kaaacutete lsquoloavesrsquo mj-ooacutetoacute lsquofiresrsquoli-kuŋuuacutenda lsquofiltered beerrsquo lj-oowaacute lsquobeehiversquoki-kaacutelaaŋgo lsquofrying panrsquo kj-uuacutelaacute lsquofrogrsquoi-kaacutelaaŋgo lsquofrying pansrsquo j-uuacutelaacute lsquofrogsrsquolu-tooacutendwa lsquostarrsquo lw-aateacute lsquobanana handrsquoku-suuacutele lsquoto schoolrsquo kw-iisiacutewaacute lsquoto the islandsrsquomu-kikaacutelaaŋgo lsquoin the frying panrsquo mw-iikaacutelaaŋgo lsquoin the frying pansrsquo

The foot is roughlya grouping oftwo syllables intoa rhythmic unitwhich is primarilyrelevant inphonology forthe description ofstress assignment

Although the stem-initial vowel is long on the surface in these examplesunderlyingly the vowel is short as shown when the stem has no prefix orwhen the prefix vowel is a Thus compare ka-oacutetoacute lsquolittle firersquo ma-owaacutelsquobeehivesrsquo ka-uacutelaacute little frogrsquo ateacute lsquobanana handsrsquo ipʊkʊ lsquoratsrsquoVowel sequences can also be eliminated by coalescing the two vowels

into a single vowel often one which preserves characteristics of theindividual vowel This happens in Matuumbi as well where the combin-ations au and ai become [oo] and [ee] This rule is optional in Matuumbiso the uncoalesced vowel sequence can also be pronounced (thus motivat-ing the underlying representation)

The change of au and ai to [oo] and [ee] can be seen as creatinga compromise vowel one which preserves the height of the initial vowela and the backness and roundness of the second vowelSometimes vowel sequences are avoided simply by deleting one of the

vowels with no compensatory lengthening Thus at the phrasal level inMakonde word-final a deletes before an initial vowel cf lipeeta engaanga lipeet engaanga lsquothe knapsack cut itrsquo likuka engaanga likuk engaangalsquothe trunk cut itrsquo nneemba idanaao nneemb idanaao lsquothe boy bring himrsquo

Vowel epenthesis The converse process of vowel epenthesis is alsoquite common One context that often results in epenthesis is when anunderlying form has too many consonants in a row given the syllablestructure of the language Insertion of a vowel then reduces the size of theconsonant cluster An example of such epenthesis is found in Fula In thislanguage no more than two consonants are allowed in a row As the dataof (53) show when the causative suffix -na is added to a stem ending intwo consonants the vowel i is inserted thus avoiding three consecutiveconsonants

In Matuumbicoalescence onlyapplies in a specificgrammaticaldomain betweenvowels ofprefixes andthus one does notfind this sameprocess affectingthe prefix-plus-stem combinationfound in ka-uacutelaacutelsquolittle frogrsquo

a-i-teacuteliike ee-teacuteliike lsquohe cooked themrsquo

pa-uacute-kaaacutetiteacute pooacute-kaaacutetiteacute lsquowhen you cutrsquopa-baacute-i-kaacuteatiteacute pa-beacutee-kaacuteatiteacute lsquowhen they cut themrsquo

a-u-kaacuteatite oo-kaacuteatite lsquohe cut itrsquoka-u-tʊʊmbʊka koo-tʊʊmbʊka lsquowhen it was fallingrsquopa-i-taaacutebu pee-taaacutebu lsquowhere the books arersquopa-u-tiacutetili poo-tiacutetili lsquowhere the chicken louse isrsquoka-u-meacutejaacute koomeacutejaacute lsquolittle white antrsquona-u-tʃaaacutepu noo-tʃaaacutepu lsquowith dirtrsquo

Continuous Causativehula hulna lsquolaughrsquojara jarna lsquodrinkrsquowoja wojna lsquocryrsquodʒula dʒulna lsquobe Muslimrsquo

wurto wurtina lsquocome outrsquowuddʒa wuddʒina lsquostealrsquojotto jottina lsquoarriversquo

Another form of vowel epenthesis is one that eliminates certain kinds ofconsonants in a particular position The only consonants at the end of theword in Kotoko are sonorants so while the past tense of the verbs in (54a)is formed with just the stem the verbs in (54b) require final epentheticschwa

Another factor motivating epenthesis is word size viz the need to avoidmonosyllabic words One example is seen in the following data fromMohawk where the first-singular prefix is preceded by the vowel iacute onlywhen it is attached to a monosyllabic stem

The adaptation of loanwords into North Saami from Scandinavian lan-guages (Norwegian or Swedish) illustrates a variant on the Mohawk-typeminimal-word motivation for epenthesis In this case a vowel is insertedto prevent a monosyllabic stress foot ndash though interestingly this require-ment is determined on the basis of the Norwegian source whereas in theSaami word stress is (predictably) on the first syllable Except for a smallset of ldquospecialrdquo words (pronouns grammatical words) words in Saamimust be at least two syllables long Thus the appearance of a final epen-thetic vowel in the following loanwords is not surprising

In contrast in the following loanwords there is no epenthetic vowel Thelocation of stress which is the key to understanding this problem is

Infinitive Past Infinitive Pasthagravem-agrave haacutem lsquoyawnrsquo ɗagraven-agrave ɗagraven lsquotiersquoskwagravel-agrave skwaacutel lsquowantrsquo vegraver-agrave vegraver lsquoflyrsquoləhagravej-agrave ləhagravej lsquofearrsquo lagravew-agrave lagravew lsquofightrsquo

gəɓ-agrave gəɓə lsquoanswerrsquo kagraveɗ-agrave kaacuteɗ ə lsquocrossrsquolagraveb-agrave lagravebə lsquotellrsquo dʒagraveg-agrave dʒagravegə lsquocookrsquogigravetʃ-agrave gigravetʃə lsquosweeprsquo ʔə k-agrave ʔəkə lsquotake by forcersquosagravep-agrave sapə lsquochasersquo vigravet-agrave vigravetə lsquoblow on a firersquovənagraveh-agrave vənagravehə lsquovomitrsquo hə s-agrave həsə lsquospillrsquoɗ əv-agrave ɗə və lsquoputrsquo bagraveγ-agrave bagraveγə lsquosplit woodrsquo

k-atiruacutet-haʔ lsquoI pull itrsquok-ataʔkeraacutehkwaʔ lsquoI floatrsquok-keacutetskw-as lsquoI raise itrsquok-hniacutenus lsquoI buyrsquok-tat-s iacutektats lsquoI offer itrsquok-jʌ-s iacutekjʌs lsquoI put itrsquok-ket-s iacutekkets lsquoI scrape itrsquo

Saami Norwegiandaeligjgi deig lsquodoughrsquonijbi kniv lsquoknifersquovowʔnɑ vogn lsquowagonrsquomuwrɑ mur lsquowallrsquo

marked on the Norwegian source though stress is not marked in theorthography

The above examples are ambiguous in analysis since the source word isboth polysyllabic and has a nonfinal stress The examples in (58) on theother hand show epenthesis when the stress-foot in the source word ismonosyllabic even though the overall word is polysyllabic

Onset creation Consonants can also be inserted The main cause ofconsonant insertion is the avoidance of initial vowels or vowel sequencesIn Arabic all syllables begin with a consonant and if a word has no under-lying initial consonant a glottal stop is inserted thus al-walad [ʔalwalad]lsquothe boyrsquo In the Hare and Bearlake dialects of Slave words cannot beginwith a vowel so when a vowel-initial root stands at the beginning of a word(including in a compound) the consonant h is inserted

In Axininca Campa t is inserted between vowels ndash this language does nothave a glottal stop phoneme Thus i-N-koma-i [inkomati] lsquohe willpaddlersquo

Saami Norwegiandisdɑt ˈtirsdag lsquoTuesdayrsquokaeligwrret ˈkavring lsquoruskrsquoakaethemihkɑr akaˈdemiker lsquoacademicrsquoministɑr miˈnister lsquoministerrsquoteahter teˈater lsquotheaterrsquotemhpel ˈtempel lsquotemplersquoorgel ˈorgel lsquoorganrsquoprofessor proˈfessor lsquoprofessorrsquoplaeligstɑr ˈplaster lsquoplasterrsquokaelighkɑl ˈkakkel lsquoglazed tilersquo

hotellɑ hoˈtel lsquohotelrsquomɑrɑtonɑ maraˈton lsquomarathonrsquouniversitehtɑ universiˈtet lsquouniversityrsquotɑbeallɑ taˈbell lsquo(time-)tablersquoprivaelightɑ priˈvat lsquoprivatersquokɑmelɑ kaˈmel lsquocamelrsquopolaeligrɑ poˈlar lsquopolarrsquo

s-otildedee lsquomy older brotherrsquodene-[h]otildedee lsquoBrother (in church)rsquon-anaj lsquoyour (sg) sister-in-law (man speaking)rsquo[h]anaj lsquosister-in-lawrsquo

b-ekrsquoeacutehdiacute lsquoI take care of himherrsquobebiacute [h]ekrsquoeacutehdiacute lsquoI take care of the babyrsquoku-edehfe kuacutedehfe lsquoI chased themrsquo

sah [h]edeacutehfe lsquoshe chased the bearrsquo

Cluster reduction Deletion of consonants can be found in languagesThe most common factor motivating consonant deletion is the avoidanceof certain kinds of consonant clusters ndash a factor which also can motivatevowel epenthesis Consonant cluster simplification is found in Korean

Another cause of cluster simplification is the avoidance of certain specifictypes of consonant clusters Shona avoids clusters of the form Cj althoughCw is perfectly acceptable The deletion of j after a consonant affects theform of possessive pronouns in various noun classes Demonstratives andpossessive pronouns are formed with an agreement prefix reflecting theclass of the noun plus a stem -no for lsquothisrsquo and -angu for lsquomyrsquo Before thestem -angu a high vowel becomes a glide Where this would result in a Cysequence the glide is deleted

Since i-angu becomes jangu it is evident that the vowel i does become aglide before a vowel rather than uniformly deleting

Stress lengthening and reduction Processes lengthening stressedvowels are also rather common An example of stress-induced vowellengthening is found in Makonde where the penultimate syllable isstressed and the stressed vowel is always lengthened

Imperative Conjunctive Indicativepalp-a pal-krsquoo pal-trsquoa lsquotread onrsquoulph-ə ul-krsquoo ul-trsquoa lsquochantrsquoilk-ə il-krsquoo il-trsquoa lsquoreadrsquohalth-a hal-krsquoo hal-trsquoa lsquotastersquotalm-a tam-krsquoo tam-trsquoa lsquoresemblersquoanc-a an-krsquoo an-trsquoa lsquosit downrsquo

lsquothisrsquo lsquomyrsquo Classu-no w-angu 3mu-no mw-angu 18ku-no kw-angu 17ru-no rw-angu 11i-no j-angu 9ri-no r-angu 6tʃi-no tʃ-angu 7ʐwi-no ʐw-angu 8dzi-no dz-angu 10

kuacute-ˈliacuteiacutem-a lsquoto cultivatersquokuacute-liacuteˈm-iacuteiacutel-a lsquoto cultivate forrsquokuacute-liacuteˈm-aacuteaacuten-a lsquoto cultivate each otherrsquokuacute-liacutem-aacuteˈn-iacuteiacutel-a lsquoto cultivate for each otherrsquokuacute-liacutem-aacuten-iacutel-aacute-liacutem-aacuteˈn-iacuteiacutel-a lsquoto cultivate for each other

continuouslyrsquo

A related process is the reduction of unstressed vowels as found inEnglish From alternations like bəˈrɔmətr ~ ˌbɛrəˈmɛtrιk ˈmɔnəpowl ~ məˈnɔpəlijwe know that unstressed vowels in English are reduced to schwa Russianalso reduces unstressed nonhigh vowels so that a o become [ə] or [a] inthe syllable immediately before the stress

Reduction of unstressed vowels can go all the way to deletion so inPalestinian Arabic unstressed high vowels in an open syllable are deleted

Syllable weight limits Many languages disallow long vowels in syllablesclosed by consonants and the following examples from Yawelmani showthat this language enforces such a prohibition against VVC syllables byshortening the underlying long vowel

A typical explanation for this pattern is that long vowels contribute extraldquoweightrdquo to a syllable (often expressed as the mora) and syllable-final con-sonants also contribute weight Languages with restrictions such as thosefound in Yawelmani are subject to limits on the weight of their syllables

Stress patterns Stress assignment has been the subject of intensivetypological study and has proven a fruitful area for decomposing phono-logical parameters See Hayes (1995) for a survey of different stresssystems One very common stress assignment pattern is the alternatingpattern where every other syllable is assigned a stress Maranungkuexemplifies this pattern where the main stress is on the first syllableand secondary stresses are on all subsequent odd-numbered syllables

goroˈd-ok [gəraˈdok] lsquocitiesrsquo ˈgorod [ˈgorəd] lsquocityrsquoˈpoda-l [ˈpodəl] lsquohe gaversquo po-ˈda-tj [paˈdatj] lsquoto giversquo

Palestinian Arabic3sg masc 3sg fem 1sgˈħamal ˈħamalat ħaˈmalt lsquocarryrsquoˈkatab ˈkatabat kaˈtabt lsquowritersquoˈdaras ˈdarasat daˈrast lsquostudyrsquoˈʃirib ˈʃirbat ˈʃribt lsquodrinkrsquoˈnizil ˈnizlat ˈnzilt lsquodescendrsquoˈfihim ˈfihmat ˈfhimt lsquounderstandrsquo

Nonfuture Imperative Dubitative Passive aoristCVC xathin xatkrsquoa xatal xatit lsquoeatrsquo

doshin doskrsquoo dosol dosit lsquoreportrsquoCVVC ʂaphin ʂapkrsquoa ʂapal ʂapit lsquoburnrsquo

wonhin wonkrsquoo wonol wonit lsquohidersquo

ˈtiralk lsquosalivarsquo ˈmereˌpet lsquobeardrsquoˈjangarˌmata lsquothe Pleaiadesrsquo ˈlangkaˌrateˌi lsquoprawnrsquoˈweleˌpeneˌmanta lsquoduck (sp)rsquo

A variant of this pattern occurs in Araucanian where the main stressappears on the second syllable and secondary stresses appear on everyeven-numbered syllable following

The mirror image of the Maranugku pattern is found in Weri where thelast syllable has the main stress and every other syllable preceding hassecondary stress

Finally Warao places the main stress on the penultimate syllable and hassecondary stresses on alternating syllables before

Another property exhibited by many stress systems is quantity-sensitivitywhere stress is assigned based on the weight of a syllable PalestinianArabic has such a stress system where stress is assigned to the finalsyllable if that syllable is heavy to the penult if the penult is heavy andthe final syllable is light and to the antepenult otherwise The typicaldefinition of a heavy syllable is one with either a long vowel or a finalconsonant however it should be noted that in Arabic final syllables havea special definition for ldquoheavyrdquo which is that a single consonant does notmake the syllable heavy but two consonants do

74 Why do things happen

Two of the central questions which phonological theory has soughtanswers to are ldquowhy does rule X existrdquo and ldquocan rule Y existrdquo Very manylanguages have a process changing velars into alveopalatals (k tʃ) before

wuˈle lsquotomorrowrsquo

tiˈpanto lsquoyearrsquoeˈlumuˌju lsquogive usrsquoeˈluaˌenew lsquohe will give mersquokiˈmubaˌluwuˌlaj lsquohe pretended not to knowrsquo

ŋinˈtip lsquobeersquoˌkuliˈpu lsquohair of armrsquo

uˌluaˈmit lsquomistrsquoˌakuˌneteˈpal lsquotimesrsquo

jiˌwaraˈnae lsquohe finished itrsquoˌjapuˌrukiˌtaneˈhase lsquoverily to climbrsquoeˌnahoˌroaˌhakuˈtai lsquothe one who caused him to eatrsquo

radˈjoo lsquoradiorsquo qaˈreet lsquoI readrsquokaˈtabt lsquoI wrotersquo ˈqara lsquohe readrsquoˈqarat lsquoshe readrsquo kaˈtabna lsquowe wrotersquoqaˈreethum lsquoI read themrsquo ˈkatabu lsquothey wrotersquoˈkatabat lsquoshe wrotersquo ma kataˈbatʃ lsquoshe didnrsquot writersquo

front vowels and a rule voicing voiceless stops after nasals (mp mb) isalso quite common It is natural to wonder why such rules would occur inmany languages and a number of theoretical explanations have beenoffered to explain this It is also important to also ask about imaginablerules we want to know for example if any language has a rule turning alabial into an alveopalatal before a front vowel one devoicing a voicedstop after a nasal or one turning s m into l k before w ʃ Only bycontrasting attested with imaginable but unattested phenomena do the-ories become of scientific interest

Impossible rules There is a clear and justified belief among phonolo-gists that the rule s m l k _w ʃ is ldquounnaturalrdquo and any theorywhich predicts that such a rule is on a par with regressive voicing assimi-lation would not be a useful theory We have seen in chapter 3 that it isactually impossible to formulate such a process given the theory of dis-tinctive features since the classes of segments defining target and triggerand the nature of the structural change cannot be expressed in thetheory The fact that neither this rule nor any of the innumerable otherconceivable random pairings of segments into rules has ever been attestedin any language gives us a basis for believing that phonological rulesshould at least be ldquopossiblerdquo in the very simple technical sense expressedby feature theory Whether a rule is possible or impossible must bedetermined in the context of a specific theoryAnother pair of rules which we might wonder about are those in (71)

(71) a

The pattern of alternation in (a) is quite common and was exemplifiedearlier in this chapter as nasal place assimilation The second pattern ofalternation in (b) on the other hand is not attested in any language Giventhe nonexistence of the pattern (b) we may ask ldquowhy is this pattern notattestedrdquoThe easy answer to this question is that pattern (b) is not phonetically

natural This begs the question of how we know what is a phoneticallynatural versus an unnatural pattern and unfortunately the connectionbetween ldquoactually attested phonological rulerdquo and ldquophonetically naturalrdquois so close that some people may assume that commonly occurringrules are by definition phonetically natural and unattested rules areunnatural This is circular if we are to preclude a pattern such as (b)as phonetically unnatural there must be an independent metric of

mtʃ ɲtʃ ŋtʃ ɲtʃ

ɲp mp np mpɲk ŋk nk ŋkɲt nt ntʃ ɲtʃ

mtʃ ntʃ (not ɲtʃ) ŋtʃ ɲtʃ

ɲp ŋp np mpɲk ŋk nk mkɲt ɲt ntʃ ntʃ

phonetic naturalness Otherwise we would simply be saying ldquosuch-and-such rule is unattested because it is unattestedrdquo which is a pointlesstautologyAnother answer to the question of why pattern (b) is not attested but

pattern (a) is would appeal to a formal property of phonological theoryWe will temporarily forgo a detailed analysis of how these processes canbe formulated ndash this is taken up in chapter 9 ndash but in one theory the so-called linear theory practiced in the 1960s and 1970s there was also noformal explanation for this difference and the rules in (b) were possibleusing feature variable notation By contrast the nonlinear theory intro-duced in the late 1970s has a different answer formalizing such rules istechnically impossible just as writing a rule s m l k _w ʃis impossible in classical feature theory Themechanism for processeswherethe output has a variable value (ie the result can be either [+anterior] or[ndashanterior]) requires the target segment to take the same values for thefeatures and to take on all valueswithin certain feature sets The alternationin (b) does not have this property (for example the change of ɲp to [ŋp] doesnot copy the feature [labial]) and therefore according to the nonlinear theorythis is an unformalizable rule The process is (correctly) predicted to beunattested in human language

Unlikely rules Now consider a rule p tʃ _i e which seems hardlydifferent from k tʃ _i e except the latter is common and the formeris apparently not found in any language Since we donrsquot know ofexamples we must wonder why there is such a gap in what is attestedPerhaps if we had the ldquoright theoryrdquo every rule that is possible under atheory would actually be attested in some language In both the linear andnonlinear theories these are both technically possible rulesOne legitimate strategy is to assume that this is an accidental gap and

hope that further research will eventually turn up such a rule Given thatonly a tiny fraction of the worldrsquos languages have been surveyed this isreasonable There is a bit of danger in assuming that the apparent non-existence of labial coronalization is an accidental gap because we donrsquotwant to mistakenly ignore the nonexistence of the imaginary rule s m[l k]_[w ʃ ] as another accidental gapThe difference between these two kinds of rules lies in an implicit

estimation of how big the gap is between prediction and observationA number of rules would fall under the rubric ldquolabial coronalizationrdquowhich would be formalizable under standard feature theories

If the rules p [tʃ] _[i] p [tʃ] _[i e] and p f b [tʃ ʃ dʒ] _[i e] wereall attested and only the rule p b [tʃ dʒ] _[i] were missing there wouldbe no question that this is an accidental gap The number of rules whichcan be formulated in standard theories is large running in the millions orbillions If we canrsquot find one or some dozen particular rules in the

p tʃ _i p b tʃ dʒ _ip tʃ _i e etc p f b tʃ ʃ dʒ _i e etc

This number hasnever beencalculated partlybecause the natureof the theory(hence thecharacterizationldquotheoreticallypossible rulerdquo)changes ratherrapidly andpartly becausephonologists arenrsquotusually concernedwith combinatorics

hundred or so languages that we have looked at this shouldnrsquot causeserious concern because the chance of finding any one rule out of the setof theoretically possible rules is fairly low and this one gap is of no moresignificance than a failure to toss a million-sided coin a few hundred timesand not have the coin land with side number 957219 on topWe should be a bit more concerned when we identify a somewhat

large class ndash hundreds or perhaps even a thousand ndash of possible ruleswhich are all unattested and which seem to follow a discernable pattern(ie ldquoalveopalatalization of labialsrdquo) Remember though that we are deal-ing with a million-sided coin and only a few hundred tosses of the coinThe unattested set of rules represents perhaps a tenth of a percent ofthe logically possible set and given the small size of the sample ofphonological rules actually available to us the chances of actually findingsuch a rule are still not very highThe situation with the rule s m [l k] _[w ʃ ] is quite different This

rule is a representative of an immense class of imaginable rules formed byarbitrarily combining sounds in lists If rules are unstructured collectionsof segments changing randomly in arbitrary contexts then given a mere8192 (frac14213) imaginable language sounds there are around 1045000 differ-ent ways to arrange those segments into rules of the type _ in comparison to around a billion ways with standard ruletheory Almost every rule which is theoretically predicted underthe ldquorandom segmentrdquo theory falls into the class of rules of the types m [l k] _[w ʃ] and yet not a single one of these rules has beenattested Probability theory says that virtually every attested rule shouldbe of this type given how many of the imaginable arbitrary rules thereare This is why the lack of rules of the type s m [l k] _[w ʃ] issignificant ndash it represents the tip of a mammoth iceberg of failed predic-tions of the ldquorandom phonemerdquo theory of rulesAnother way to cope with this gap is to seek an explanation outside

phonological theory itself An analog would be the explanation for whyArctic mammals have small furry ears and desert mammals have largernaked ears proportionate to the size of the animal There is no independ-ent ldquolaw of biologyrdquo that states that ear size should be directly correlatedwith average temperature but this observation makes sense given a littleknowledge of the physics of heat radiation and the basic structure of earsIn a nutshell you lose a lot of body heat from big ears which is a goodthing in the desert and a bad thing in the Arctic Perhaps there is anexplanation outside the domain of phonological theory itself for the lackof labial coronalization in the set of attested rulesWhat might be the functional explanation for the lack of such a pro-

cess We first need to understand what might be a theory-externalfunctional explanation for the common change k tʃ _i e In a vastnumber of languages there is some degree of fronting of velar consonantsto [kj] before front vowels The reason for this is not hard to see canonicalvelars have a further back tongue position and front vowels have afurther front tongue position To produce [ki] with a truly back [k] anda truly front [i] the tongue body would have to move forward a

considerable distance essentially instantaneously This is impossible andsome compromise is required The compromise reached in most lan-guages is that the tongue advances in anticipation of the vowel [i] duringproduction of [k] resulting in a palatalized velar ie the output [kji]which is virtually the same as [ci] with a ldquotrue palatalrdquo stopThe actual amount of consonantal fronting before front vowels that is

found in a language may vary from the barely perceivable to the reason-ably evident (as in English) to the blatantly obvious (as in Russian) Thisrelatively small physiological change of tongue fronting has a dispropor-tionately more profound effect on the actual acoustic output Essentially aplain [k] sounds more like a [p] than like [c] ([k] has a lower formantfrequency for the consonant release burst) and [tʃ] sounds more like [t]or [tʃ] (in having a higher burst frequency) than like [k] which it isphysiologically more similar to The acoustic similarity of alveopalatalslike [tʃ] and palatals like [tʃ] is great enough that it is easy to confuse onefor the other Thus a child learning a language might (mis)interpret aphonetic alternation [k] ~ [tʃ] as the alternation [k] ~ [tʃ]Explaining why k tʃ _i e does exist is a first step in understanding

the lack of labial coronalization before front vowels The next question iswhether there are analogous circumstances under which our unattestedrule might also come into existence Since the production of [p] and theproduction of [i] involve totally different articulators a bit of tongueadvancement for the production of [i] will have a relatively negligibleeffect on the acoustics of the release burst for the labial and especiallywill not produce a sound that is likely to be confused with [tʃ] Theconstriction in the palatal region will be more open for i after the releaseof p because the tongue does not already produce a complete obstructionin that region (a maximally small constriction) as it does with k It ispossible to radically advance the tongue towards the [i]-position and makeenough of a palatal constriction during the production of a [p] so that amore [tʃ]-like release will result but this will not happen simply as aresponse to a small physically motivated change as it does with k Thusthe probability of such a change ndash p tʃ ndash coming about by phoneticmechanisms is very small and to the extent that phonological rules gettheir initial impetus from the grammaticalization of phonetic variantsthe chances of ever encountering labial coronalization are slimAnother approach which might be explored focuses on articulatory

consequences of velar coronalization versus labial coronalization Velarsand alveolars involve the tongue as their major articulator as does [tʃ]whereas labials do not involve the tongue at all We might then conjecturethat there is some physiological constraint that prevents switching majorarticulators even in phonological rules But we canrsquot just say that labialsnever become linguals they typically do in nasal assimilation In factthere is a process in the Nguni subgroup of Bantu languages (Zulu XhosaSwati Ndebele) where at least historically labials become alveopalatalsbefore w which is very close to the unattested process which we have beenlooking for By this process a labial consonant becomes a palatal beforethe passive suffix -w- as in the following data from Swati

This is a clear counterexample to any claim that labials cannot switch majorarticulator and is a rather odd rule from a phonetic perspective (as pointedout by Ohala 1978) Rather than just leave it at that we should ask how suchan odd rule could have come into existence In a number of Bantu languagesespecially those spoken in southern Africa there is a low-level phoneticprocess of velarization and unrounding where sequences of labial consonantplus [w] are pronounced with decreased lip rounding and increased velarconstriction so that underlying pw is pronounced as [pɯ] with [ɯ] notating asemi-rounded partial velar constriction The degree of velar constrictionvaries from dialect to dialect and language to language and the degree ofphonetic constriction increases as one progresses further south among theBantu languages of the area so in Karanga Shona pw is pronounced with anoticeable obstruent-like velar fricative release and no rounding as [px] Theplace of articulation of the velar release shifts further forward depending onthe language and dialect being realized as [pccedil] in Pedi or as [pʃ] in Sotho andfinally as [tʃ] in Nguni So what seems like a quite radical change given justthe underlying-to-surface relation p [tʃ] in Nguni is actually just theaccumulated result of a number of fortuitously combined less radical stepsOne of the current debates in phonology ndash a long-standing debate given

new vitality by the increased interest in phonetics ndash is the question of theextent to which phonological theory should explicitly include reference toconcepts rooted in phonetics such as ease of articulation perceptibility andconfusability and issues pertaining to communicative function Virtuallyevery imaginable position on this question has been espoused and it iscertain that the formalistfunctionalist debate will persist unresolved fordecades

Further readingGreenberg 1978 Hale and Reiss 2006 Hayes Kirchner and Steriade 2004 Maddiesson 1984 Odden 2013

Active Passivekuacute-khaacutendiŋ-a kuacute-khaacutendiacuteŋ-w-a lsquodry roastrsquokuacute-kaacuteph-a kuacute-kaacuteʃ-w-a lsquochoprsquokuacute-kxeacutebh-a kuacute-kxeacutedʒw-a lsquoscrapersquokuacute-luacutem-a kuacute-luacuteɲ-w-a lsquobitersquokuacute-nwaacuteb-a kuacute-nwaacutetʃ-w-a lsquoburyrsquo

Summary The distinction between unattested rare and well-known patterns inphonology has been important in the development of theory How dowe distinguish between actually nonexistent patterns and patternsthat we are unaware of Which unattested patterns should the formaltheory preclude Why are certain patterns found in very many lan-guages Should the formal theory try to account for frequency ofoccurrence These questions will remain vital research topics inphonology for many years

CHAPTER

8 Abstractnessandpsychologicalreality

PREVIEW

This chapter explores the extent to which underlying and

surface forms can be different ndash what constraints if

any are tenable within the formal theory what the issues

are in limiting abstractness and how to address these

questions empirically The central question raised in this

chapter is ldquowhat counts as evidence for a phonological

analysisrdquo

KEY TERMSabstractness

absoluteneutralization

psychologicalreality

external evidence

A fundamental question in the theory of phonology has been ldquohowabstract is phonologyrdquo specifically how different can the underlyingand phonetic forms of a word be The essential question is whethergrammars use entities that are not directly observed Related to this isthe question whether a linguistic model requiring elements that cannotbe directly observed reflects what the human mind does The very conceptof a mental representation of speech such as a phonological surface formlike [sɔks] socks which is not itself an observable physical event requiresabstracting away from many specifics of speech Without generalizingbeyond the directly observable it would be impossible to make even themost mundane observations about any language The question is there-fore not whether phonology is abstract at all but rather what degree ofabstractness is requiredIf underlying representations are fully concrete ndash if they are the same as

surface representations ndash the underlying forms of English [khɔrts] courtsand [khowdz] codes would be khɔrt-s and khowd-z Such an extremelysurface-oriented view of phonology would ignore the fact that the wordshave in common the plural morpheme whose pronunciation variesaccording to the environment By hypothesizing that the underlying formof [khɔrts] is khɔrt-z we can say that the plural pronounced s in [khɔrts] andthe plural pronounced z in [khowdz] are one and the same thing Suchabstractness in phonological analysis yields the benefit of explaining thesimilaries in pronunciation of the various realizations of the pluralmorpheme

81 Why limit abstractness

First we must understand what motivates concern over abstractness

811 Limiting possible analysesOne reason to limit the divergence between underlying and surface formsis to constrain the theory of phonology to prevent it from making wrongclaims about how languages work With no constraint on abstractnessevery conceivable derivation from underlying to surface form would inprinciple be allowed by the theory Just as the theory of phonology seeksto constrain the concept of ldquopossible rulerdquo so that an imaginable rulesuch as s p q r m l t v _ s k ə m (unattested in any humanlanguage) can be ruled out on formal grounds so too might we wish torule out a derivation from underlying qoslashɬijʌ to surface [gəˈraʒ] as tooabstract Since a goal of linguistic theory has been to restrict the class oftheoretically possible languages to just the type that is actually observedlimiting abstractness in a well-defined way limits the number of possiblelanguagesAnother reason for concern over abstractness is that it makes a particu-

lar claim about human cognition that the mentally stored units of lan-guage can include things that the speaker has not actually heard butarrives at by inference based on a line of indirect evidence Since first

language acquisition does not proceed by conscious reasoning it cannotbe taken for granted that everyday academic reasoning skills are automati-cally available to children

Mental reality and language acquisition This second considerationwhether abstractness (of some particular degree) is part of human cogni-tive capacity is the most important question arising in this debate this isa fundamental consideration for a theory such as generative grammarthat seeks a model of language in the mind Because the details ofspecific languages are not built into children at birth but must beinduced from the ambient linguistic data aided by general cognitivecapacity and whatever language faculty is universally available to allhumans (ie the theory of grammar) a basic concern regarding thepsychological reality of grammatical constructs ndash for phonology rulesand underlying forms ndash is whether they can be learned from theprimary language dataThe role of a universal grammatical component is to make the job of

language acquisition easier by uncompromisingly removing certain kindsof imaginable descriptions from consideration Distinctive features areone way of making this job easier since they limit the ways of analyzingdata Universal constraints on abstractness might similarly help a childtrying to arrive at underlying representation for a language and therehave been a number of proposals as to the relationship between theunderlying and surface forms Attractive as it might seem to proposeformal constraints on the theory of grammar to prohibit English fromhaving qoslashɬijʌ be the underlying form of [gəˈraʒ] garage we will not actu-ally assume that this is a matter for the formal theory of grammar ratherit is a consequence of how a phonology is learned thus the question ofabstractness is outside the domain of grammatical theoryFaced with a word pronounced [dɔg] a child learning English has no

reason to assume that its underlying form is anything other than dɔg Butfaced with the word atom [ˈaeligɾəm] and the related word atomic [əˈthɔmɪk] thechild needs to arrive at an underlying representation for the root onwhich these two words are based such that rules of English phonologycan apply to derive the phonetic variants [ˈaeligɾəm] and [əˈthɔmɪk] anappropriate representation would be [aeligtɔm] It is in the face of sucha specific motivation for an abstract underlying form that we wouldassume the underlying form isnrsquot simply the surface form The solutionto the so-called problem of abstractness which will be adopted here issimply that abstractness per se is not a problem what really requiresinvestigation is the kind of evidence that properly motivates a phono-logical analysis

Abstractness and phonemic representations One particular degreeof abstractness is widely accepted as self-evident needing no furtherjustification namely that underlying representations do not contain allo-phonic variants of phonemes It is generally assumed that English [stɔp][thɔp] are underlyingly stɔp tɔp without aspiration because there is

Abstractness and psychological reality 239

(by assumption) no underlying aspiration in English Similarly we knowthat the underlying form of [hɪɾɪŋ] hitting is hıtıŋ not only because the flapis an allophone in English but also because of the related word [hɪt] hitwhere the [t] is directly pronounced Thus it is commonly assumed thatunderlying forms are at least as abstract as phonemic representationswith all allophonically predictable features eliminatedThis assumption can lead to problems What is the medial consonant in

the underlying form of a word like [waɾ r] water Assuming that the flap isnot a phoneme in English (there are no minimal or near-minimal pairscontrasting [t] or [d] vs [ɾ]) this forces us to say that it must be somethingother than [ɾ] The word is spelled with t but spelling is not relevant tounderlying representations Children acquire words without knowinghow to spell and most languages of the world are unwritten yet under-lying representations must be acquired for all human languages Spellingis also unreliable and could lead us to the unjustified conclusion thatthe underlying vowels of [tuw] too to two [θruw] through [duw] due and[druw] drew are all differentSince [waɾ r] is not composed of a root plus suffix we cannot look at

related forms to reveal the underlying consonant (as we can in wad-erversus wait-er both [wejɾ r]) Any number of hypotheses could be set forth ndash

waɾ r watr wadr waethr waβ r waγ r and so on Hypotheses like waβ rand waγ r can be rejected on the grounds that they are pointlesslyabstract containing segments which do not occur phonetically in Englishand there is no reason to believe that they exist underlyingly Nothing isgained by positing such underlying representations thus nothing justifiesthese hypotheses Two facts argue decisively against hypothetical waβrwaγ r and their ilk First there is no evidence for a rule in Englisheffecting the change γ [ɾ] or β [ɾ] and addition of such a rulerequired to convert the underlying form into the surface form rulesagainst such an analysis since there exist analyses which at least do notforce the inclusion of otherwise unmotivated rules Second a specificchoice between waβ r and waγ r or waʔ r and innumerable other possi-bilities which also lack an underlying flap is totally arbitrary and leavesthe language analyst ndash student and child alike ndash with the unresolvablepuzzle ldquowhy this underlying form and not some otherrdquo which can onlybe resolved by fiatThe hypothesis waethr is less abstract since it is composed only of

observed segments of English it is however factually wrong because itwould be impossible to craft rules for English to turn eth into a flap in thiscontext (consider father bother weather which indicate that there cannotbe a rule changing eth into a flap in some context) Only three hypothesesremain viable waɾ r watr and wadr None of these hypotheses positssurface nonexistent segments and given the rules of English ndash Flappingspecifically ndash any of these underlying representations would result in thecorrect surface formThere is no standard answer to the question of the underlying form of

water but certain arguments can be marshalled to support differentpositions We initially rejected the theory that the underlying form might

be waɾ r because it posits what we assumed to be a nonexistent underlyingsegment in the language but we should reconsider that decision to atleast explain our argument for rejecting an underlying flap Hypothesiz-ing waɾ r necessitates another phoneme in the inventory of Englishunderlying segments violating an analytic economy principle whichsays that you should select a parsimonious underlying inventory for alanguage This perhaps reflects the basic principle of scientific reasoningthat simpler more economical solutions are better than complicatedsolutions that posit unnecessary machinery But no concrete linguisticarguments indicate that elimination of phonemes is an actual goal ofphonological acquisition Economy of the underlying inventory cannotbe judged in a theoretical vacuum and in at least one contemporarytheory Optimality Theory it is impossible to state generalizations aboutunderlying representations so it is impossible to say that English has nounderlying flapA somewhat stronger argument against allowing an underlying flap is

that the surface distribution of [ɾ] is restricted It only appears betweenvocoids (vowels and glides) and only if the following vowel is unstressedwhich is precisely the context where t d actively are changed into theflap [ɾ] (hit [hɪt] ~ hitting [hɪɾɪŋ] hide [hajd] ~ hiding [hajɾɪŋ]) We can explainthe lack of words in English like [hiɾ] [ɾuwl] [aeligfɾ r] and [əɾǽk] if weassume that the flap [ɾ] is not in the inventory of underlying segmentsof English and only derives from t or d by this specific rule Thisargument recognizes the importance of capturing major generaliza-tions about language which is the central concern of linguistics it saysthat it would be too much of a coincidence if in assuming underlying ɾin water we failed to note that underlying flap only appears in a veryfew contextsThis argument is founded on the presumption that distribution of

segments in underlying forms cannot be restricted otherwise we wouldsimply state a restriction on where underlying flaps appear and let theunderlying form of [waɾ r] be fully concrete Some theories do not haveconditions on underlying forms (Optimality Theory) others do Some-thing like conditions on underlying forms seems inevitable since forexample there cannot be any words in English of the form sCiVCi henceslil sneen spup skuck yet it is uncertain what status such conditionshave in the theory of grammar The assumption that all regularities abouta language must be captured in the grammar has been a fundamentalassumption for many theories of phonology but has also been challenged(see Hale and Rice 2006) so we cannot take it for granted that thegrammar is solely responsible for explaining the distribution of the flapin EnglishStill even if we decide that the underlying form doesnrsquot have a flap

that leaves open the choice between t and d which is purely arbitraryThe choice might be made by appealing to markedness (chapter 7)insofar as [t] is a less marked ie crosslinguistically common segmentthan [d] Whether this reasoning is correct remains to be determinedempirically

812 A principled limit on abstractnessIn connection with our first neutralization rule final devoicing inRussian (chapter 4) we explained the alternation [porok] lsquothreshold(nom sg)rsquo ~ [poroga] lsquothreshold (gen sg)rsquo by saying that underlyinglythe stem ends with g The abstract representation porog for [porok]lsquothreshold (nom sg)rsquo is justified by the fact that [porok] and [poroga]have the same root morpheme and porog is one of the two actuallyoccurring pronunciations of the morpheme In hypothesizing under-lying forms of morphemes we have repeatedly emphasized the utilityof considering any and all of the surface realizations of a given mor-pheme as candidates for being the underlying form One might evenadvance a formal principle regarding abstractness (a principle to thiseffect was proposed in the theory of Natural Generative Phonologysee Vennemann 1974)

(1) The underlying form of a morpheme must actually be pronouncedas such in some surface form containing the morpheme

The underlying cognitive presupposition of such a principle is thathumans only abstract the nature of morphemes by directly selecting fromtokens of perceptual experience with that unitWhen you look at a broad range of phonological analyses it very often

turns out that the supposed underlying form of a morpheme is indeeddirectly observed in some surface form Nonetheless such a principlecannot be an absolute condition on the relation between underlying andsurface forms that is it cannot be a principle in the theory of grammarRecall from chapter 4 that in Palauan all unstressed vowels becomeschwa and underlying forms of roots may contain two full vowels forexample daŋob lsquocoverrsquo teʔib lsquopull outrsquo ŋetom lsquolickrsquo We are justified inconcluding that the first vowel in daŋob is a because it is actuallypronounced as such in [mə-ˈdaŋəb] when the first root vowel is stressedand we are justified in concluding that the second vowel is o because thatis how it is pronounced in [dəˈŋobl] Although each hypothesized under-lying vowel can be pronounced in one surface variant of the root oranother no single surface form actually contains both vowels in theirunreduced form the hypothesized underlying form daŋob is never pro-nounced as such thus our analysis of Palauan is a counterexample to theexcessively restrictive statement (1) Similar examples come from English(cf the underlying stem tεlεgraeligf which explains the surface vowel qual-ities in [ˈtɛləgraeligf ] and [təˈlɛgrəf-ij]) and Tonkawa (cf picena which isjustified based on the surface forms picna-n-oʔ and we-pcen-oʔ) Condition(1) also runs into problems in Yawelmani (chapter 6) which has a ruleshortening a long vowel before a cluster of two consonants and anotherrule inserting i after the first of three consonants The two rules applyin stems such as ʔaml so that epenthesis turns ʔaml-hin into[ʔamil-him] and shortening turns ʔaml-al into [ʔamlal] The problem for(1) is that ʔaml can never be pronounced as such since either the vowel isshortened or else i is inserted

Rather than abandon the enterprise of doing phonology in these lan-guages out of misguided allegiance to an a priori assumption about therelationship between underlying and surface forms we might consider aweaker constraint which allows underlying forms of morphemes to becomposed of segments that are actually pronounced in some attestation ofthe morpheme but disallows representations that are more abstract

(2) The underlying form of a word must contain only segments actuallypronounced as such in some related word containing the morpheme

Even this cannot be an absolute requirement One case that runs afoul ofthis condition is the case of stem-final voiced stops in Catalan (chapter 5problem 7) There is a rule devoicing final obstruents and another rulespirantizing intervocalic voiced stops These rules result in alternationssuch as sεk lsquodry (masc)rsquo ~ sεkə lsquodry (fem)rsquo from sεk versus sek lsquoblind(masc)rsquo ~ seγə lsquoblind (fem)rsquo from seg The underlying voiced stop g isnot directly attested in any form of the stem seg and thus runs afoul ofconstraint (2)Another counterexample to (2) is Hehe (chapter 6) That language has a

rule assigning H tone to a penultimate vowel that is not also immediatelypreceded by an H This rule accounts for the position of the second H tonein words like kuacute-kam-iacute l-a lsquoto milk forrsquo kuacute-kam-il-aacuten-a lsquoto milk for eachotherrsquo and the lack of H tone in kuacute-kam-a lsquoto milkrsquo where the penultimatevowel is preceded by an H-toned vowel Surface forms such as kuacute-kam-y-aacutelsquoto cause to milkrsquo and kuacute-kam-w-aacute lsquoto be milkedrsquo would seem to be excep-tions but actually they follow the general pattern perfectly as long as werecognize that the underlying forms are kuacute-kam-i-a and kuacute-kam-u-aGiven those underlying forms the H is regularly assigned to the penulti-mate vowel giving kuacute-kam-iacute-a and kuacute-kam-uacute-a and then the high vowelsbecome glides before a vowel causing the H tone to be transferred to thefinal vowel The important point about these examples is that theassumed vowels of the causative and passive never surface as vowels theyappear only as glides since by quirks of Hehe morphology the mor-phemes -i- and -u- are always followed by a vowel suffix so they alwaysundergo glide formation

813 Case studies in abstract analysisWe will look in depth at two cases of abstract phonological analysis onefrom Matuumbi and one from Sanskrit where abstract underlying formsare well motivated these are contrasted with some proposals for Englishwhich are not well motivated Our goal is to see that the problem ofabstractness is not about the formal phonetic distance between under-lying and surface forms but rather it involves the question of how strongthe evidence is for positing an abstract underlying representation

Abstract mu in Matuumbi Matuumbi provides an example of anabstract underlying representation involving an underlying vowel whichnever surfaces as such In this language the noun prefix which marks

nouns of lexical class 3 has a number of surface realizations such as [m][n] [ŋ] and [mw] but the underlying representation of this prefix is mudespite the fact that the prefix never actually has that surface manifest-ation with the vowel uWe begin with the effect which nasals have on a following consonant

Sequences of nasal plus consonant are subject to a number of rules inMatuumbi and there are two different patterns depending on the natureof the nasal One such nasal is the prefix ɲ- marking nouns and adjec-tives of grammatical class 9 When this prefix comes before an under-lyingly voiced consonant the nasal assimilates in place of articulation tothat consonant by a general rule that all nasals agree in place of articula-tion with an immediately following consonant

When added to a stem beginning with a nasal consonant the nasaldeletes

The prefix ɲ causes a following voiceless consonant to become voiced

Finally ɲ causes a following glide to become a voiced stop preserving theplace properties of the glide

We know that the prefix is underlyingly ɲ because that is how it surfacesbefore vowel-initial adjectives such as ɲ-epeeacutesi lsquolight (cl 9)rsquo ɲ-iiacutepi lsquoshort(cl 9)rsquoDifferent effects are triggered by the nasal of the prefix mu which

marks second-plural subjects on verbs This prefix has the underlyingform mu and it can surface as such when the following stem beginswith a consonant

Adjective (cl 9) Verbm-bomwaacuteanaacute boacutemwaana lsquopointlessly destroyrsquoŋ-goloacutekaacute goacuteloka lsquobe straightrsquoɲ-dʒiluacutekaacute dʒiacuteluka lsquofall downrsquo

Adjective (cl 9) Verbmamaacuteandwaacute maacutemaandwa lsquonailrsquomimiacutenaacute miacutemina lsquospillrsquonamaacutetaacute naacutemata lsquobe stickyrsquo

Adjective (cl 9) Verbn-dɪnɪkaacute tɪnɪka lsquocutrsquon-demaacuteaacute teacutema lsquochoprsquoɲ-dʒapiacuteitʃaacute tʃaacutepiitʃa lsquobe cleanrsquo

Adjective (cl 9) Verbɲ-dʒukuacutetaacute juacutekuta lsquobe fullrsquoŋ-gwaaacuteaacute waacute lsquodiersquoŋ-gwɪkɪljaacute wɪkɪlja lsquocoverrsquo

A rule deletes the vowel u preceded by m when the vowel precedes aconsonant (you observed this rule in chapter 5) and this rule appliesoptionally in this prefix Before a stem beginning with a voiced consonantdeletion of the vowel results in a cluster of a nasal plus a consonant and mcauses nasalization of the following consonant (compare the examples in(7) where the vowel is not deleted)

This reveals an important difference between the two sets of postnasalprocesses In underlying nasal C sequences such as ɲ-bomwaacuteanaacute m-bomwaacuteanaacute lsquodestroyed (cl 9)rsquo the nasal only assimilates in place of articu-lation to the following C but in nasal + consonant sequences derived bydeletion of u the prefixal nasal causes nasalization of a following voicedconsonantAnother difference between ɲC versus muC is evident when the prefix

mu comes before a stem beginning with a nasal consonant The data in(9) show that when u deletes the resulting cluster of nasals does notundergo nasal deletion (The reason for this is that mu first becomes asyllabic nasal m and nasalization takes place after a syllabic nasal)

In comparison class 9 ɲ-mimiacutenaacute with the prefix ɲ surfaces as mimiacutenaacutelsquospilled (cl 9)rsquo having undergone degeminationA third difference between ɲ + C versus mu+C emerges with stems

that begin with a voiceless consonant As seen in (10) mu simply assimi-lates in place of articulation to the following voiceless consonant

Remember though that ɲ causes a following voiceless consonant tobecome voiced so ɲ-tɪnɪkaacute ndɪnɪkaacute lsquocut (cl 9)rsquo

Finally mu causes a following glide to become a nasal at the sameplace of articulation as the glide

mu-buundiacuteke lsquoyou should storersquomu-laabuacuteke lsquoyou should breakfastrsquomu-dʒiiŋgiacute lsquoyou should enterrsquomu-gooacuteɲdʒe lsquoyou should sleeprsquo

m-muundiacuteke lsquoyou should storersquon-naabuacuteke lsquoyou should breakfastrsquoɲ-ɲiiŋgiacute lsquoyou should enterrsquoŋ-ŋooacuteɲdʒe lsquoyou should sleeprsquo

mu-miacutemiine m-miacutemiine lsquoyou (pl) spilledrsquomu-noacuteolite n-noacuteolite lsquoyou (pl) sharpenedrsquomu-ŋaacuteandite ŋ-ŋaacuteandite lsquoyou (pl) playedrsquo

mu-paaacutende m-paaacutende lsquoyou should plantrsquomu-telekeacute n-telekeacute lsquoyou should cookrsquomu-tʃoneacute ɲ-tʃoneacute lsquoyou should sewrsquo

mu-kalaaacuteŋge ŋ-kalaaacuteŋge lsquoyou should fryrsquo

Underlying ɲ on the other hand causes a following glide to become avoiced stop cf ɲ-wɪkɪljaacute ŋ-gwɪkɪ ljaacute lsquocovered (cl 9)rsquoThe differences between ɲ and mu go beyond just their effects on

following consonants they also have different effects on preceding andfollowing vowels In the case of mu the preceding vowel lengthens whenu deletes

On the other hand ɲ has no effect on the length of a preceding vowel

Finally ɲ surfaces as [ɲ] before a vowel and the length of the followingvowel is not affected But mu surfaces as [mw] before a vowel due to aprocess of glide formation and the following vowel is always lengthened

A number of properties distinguish mu from ɲ Apart from the import-ant fact that positing these different underlying representations providesa phonological basis for distinguishing these effects our choices of under-lying forms are uncontroversial because the posited forms of the prefixesare actually directly attested in some surface variant recall that thesecond-plural verbal subject prefix mu can actually be pronounced as[mu] since deletion of u is optional for this prefixDeletion of u is obligatory in this prefix and optional in the subject

prefix because subject prefixes have a ldquolooserrdquo bond to the following stemthan lexical class prefixes which are joined with the stem to form aspecial phonological domainNow we are in position to discuss a prefix whose underlying representa-

tion can only be inferred indirectly The prefix for class 3 nouns andadjectives is underlyingly mu like the second-plural verbal subjectprefix Unlike the verb prefix the vowel u of the class 3 noun prefixalways deletes and mu never appears as such on the surface ndash itsunderlying presence can only be inferred indirectly A strong indicationthat this prefix is underlyingly mu is the fact that it has exactly the same

mu-wɪkɪliacute ŋ-ŋwɪkɪliacute lsquoyou should coverrsquomu-jɪkɪtiacute ɲ-ɲɪkɪtiacute lsquoyou should agreersquo

iwɪkɪljoacute mu-tooacutele lsquoyou should take coverrsquoiwɪkɪljoacuteo n-tooacutele idɲuuacutemba mu-bomwaaacutene lsquoyou should destroy the housersquoɲuuacutembaa m-momwaaacutene id

iwɪkɪljo m-bwapwaacuteanikaacute lsquobroken coverrsquoɲumbaacute m-bomwaacuteanaacute lsquodestroyed housersquo

Stemɲ iiacutepi ɲ-iiacutepi lsquoshort (cl 9)rsquo

epeeacutesi ɲ-epeeacutesi lsquolight (cl 9)rsquomu ɪɪmb-e mw-ɪɪmb-e lsquoyou should digrsquo

eleeacutew-e mw-eeleeacutew-e lsquoyou should understandrsquo

effect on a following consonant as the reduced form of the subject prefixmu has It causes a voiced consonant to become nasalized

It forms a geminate nasal with a following nasal

It also does not cause a following voiceless consonant to become voiced

Another reason to believe that this prefix is underlyingly mu is thatwhen it comes before a stem beginning with a vowel the prefix showsup as [mw] and the following vowel is lengthened

Under the hypothesis that the class 3 prefix is mu we automaticallypredict that the prefix should have this exact shape before a voweljust as the uncontroversial prefix mu marking second-plural subjecthasFinally the data in (19) show that this prefix has the same effect of

lengthening the preceding vowel as the second-plural subject prefix has

The only reasonable assumption is that this prefix is underlyingly mudespite the fact that the vowel u never actually appears as such

Infinitive Adjective (cl 3)buacuteundika m-muuacutendikaacute lsquostorersquolaacuteabuka n-naaacutebukaacute lsquobreakfastrsquodʒiacuteiŋgja ɲ-ɲiiacuteŋgjaacute lsquoenterrsquogoacuteoɲdʒa ŋ-ŋooacuteɲdʒaacute lsquosleeprsquo

Infinitive Adjective (cl 3)maacuteta m-mataacuteaacute lsquoplasterrsquomuacutelika m-muliacutekaacute lsquoburnrsquonaacutemata n-namaacutetaacute lsquobe stickyrsquo

Infinitive Adjective (cl 3)paacuteanda m-paaacutendaacute lsquoplantrsquoteacuteleka n-teleacutekaacute lsquocookrsquotʃoacutena ɲ-tʃonaacuteaacute lsquosewrsquo

kaacutelaaŋga ŋ-kalaacuteaŋgaacute lsquofryrsquo

Infinitive Adjective (cl 3)aacutelibika mwaaliacutebikaacute lsquobreakrsquoeacutepuka mweepuacutekaacute lsquoavoidrsquoɪɪmba mwɪɪmbaacute lsquodigrsquooacutetoka mwootoacutekaacute lsquopuncturersquo

mwooacutego lsquocassavarsquo mwoogoo m-mouacute lsquorotten cassavarsquompɪlaacute lsquofootballrsquo mpɪlaacutea m-puwaacuteanikaacute lsquobroken footballrsquonkoacuteta lsquosweetsrsquo nkotaa n-nogaacuteaacute lsquogood sweetsrsquonkwaacute lsquospearrsquo nkwaacutea n-kʊlʊ lsquobig spearrsquo

Direct attestation of the hypothesized underlying segment would pro-vide very clear evidence for the segment in an underlying form butunderlying forms can also be established by indirect means such asshowing that one morpheme behaves in a manner parallel to some otherwhich has a known and uncontroversial underlying form Thus the factthat the class 3 prefix behaves in all other respects exactly like prefixeswhich are uncontroversially mu suffices to justify the conclusion thatthe class 3 prefix is indeed mu

Abstract ai and au in Sanskrit A significantly more abstract repre-sentation of the mid vowels [e o] is required for Sanskrit These surfacevowels derive from the diphthongs ai au which are never phoneticallymanifested anywhere in the language The surface vowels (syllabics) anddiphthongs of Sanskrit are in (20)

Two things to be remarked regarding the inventory are that while thelanguage has diphthongs with a long first element ai au it has nodiphthongs with a short first element Second the mid vowels only appearas long never short These two facts turn out to be relatedOne phonological rule of the language fuses identical vowels into

a single long vowel This process operates at the phrasal level soexamples are quite easy to come by simply by combining two wordsin a sentence

A second process combines long or short a with i and u (long or short)giving the long mid vowels e and o

These data point to an explanation for the distribution of vowels noted in(20) which is that underlying ai and au become e and o and that this isthe only source of mid vowels in the language This explains why the midvowels are all long and also explains why there are no diphthongs ai auThere is also a rule shortening a long vowel before another vowel at thephrasal level which is why at the phrasal level a plus i does not form along diphthong [ai]

a i u r l a e i o u r ai au

na lsquonotrsquo + asti lsquoisrsquo nasti lsquois notrsquona lsquonotrsquo + aste lsquohe sitsrsquo naste lsquohe doesnrsquot sitrsquonadi lsquoriverrsquo + iwa lsquolikersquo nadiwa lsquolike a riverrsquojadi lsquoifrsquo + iccedilwarah lsquolordrsquo jadiccedilwarah lsquoif the lordrsquonadi lsquoriverrsquo + iccedilwarah lsquolordrsquo nadiccedilwarah lsquolord riverrsquosadhu lsquowellrsquo + uktam lsquosaidrsquo sadhuktam lsquowell saidrsquo

ca lsquoandrsquo + iha lsquoherersquo ceha lsquoand herersquoca lsquoandrsquo + uktam lsquosaidrsquo coktam lsquoand saidrsquosa lsquoshersquo + uktam lsquosaidrsquo soktam lsquoshe saidrsquosa lsquoshersquo + iccedilwara lsquoO Lordrsquo seccedilwara lsquoshe O Lordrsquo

There is a word-internal context where the short diphthongs ai and auwould be expected to arise by concatenation of morphemes and wherewe find surface e o instead The imperfective tense involves the prefixa-tion of a-

If the stem begins with the vowel a the prefix a- combines with followinga to give a long vowel just as a + a a at the phrasal level

When the root begins with the vowels i u the resulting sequences ai()au() surface as long mid vowels

These alternations exemplify the rule where ai au [e o]We have shown that a + i a + u surface as [e o] so now we will

concentrate on the related conclusion that [e o] derive from underlyingai au One argument supporting this conclusion is a surface generaliza-tion about vowel combinations that when a combines with what wouldsurface as word initial o or e the result is a long diphthong au ai

This fusion process makes sense given the proposal that [e] and [o] derivefrom ai and au The examples in (26b) remind us that initial [eo] inthese examples transparently derive from a + i a + u because in theseexamples a is the imperfective prefix and the root vowels u i can be seendirectly in the present tense Thus the underlying forms of [caukʂat]and [caikʂat] are [caa-ukʂat] and [caa-ikʂat] The surface long diphthongderives from the combination of the sequence of arsquos into one long aThe same pattern holds for all words beginning with mid vowels evenwhen there is no morphological justification for decomposing [e o] intoa+i a+uOther evidence argues for deriving surface [e o] from ai au There is a

general rule where the high vowels i u surface as the glides [ j w] beforeanother vowel which applies at the phrasal level in the following examples

bhar-at-i lsquohe bearsrsquo a-bhar-at lsquohe borersquotuɲɟ-at-i lsquohe urgesrsquo a-tuɲɟ-at lsquohe urgedrsquowardh-at-i lsquohe growsrsquo a-wardh-at lsquohe grewrsquo

aɟ-at-i lsquohe drivesrsquo aɟ-at lsquohe droversquoaɲc-at-i lsquohe bendsrsquo aɲc-at lsquohe bentrsquo

il-at-i lsquohe is quietrsquo el-at lsquohe was quietrsquoikʂ-at-i lsquohe seesrsquo ekʂ-at lsquohe sawrsquo

ukʂ-at-i lsquohe sprinklesrsquo okʂ-at lsquohe sprinkledrsquoubɟ-at-i lsquohe forcesrsquo obɟ-at lsquohe forcedrsquo

a ca lsquoandrsquo + okʂat lsquohe sprinkledrsquo caukʂat lsquoand he sprinkledrsquoca lsquoandrsquo + ekʂat lsquohe sawrsquo caikʂat lsquoand he sawrsquo

b ca lsquoandrsquo + ukʂati lsquohe sprinklesrsquo cokʂati lsquoand he sprinklesrsquoca lsquoandrsquo + ikʂati lsquohe seesrsquo cekʂati lsquoand he seesrsquo

The mid vowels [e o] become [aj aw] before another vowel (an optionalrule most usually applied can delete the glide in this context giving avowel sequence)

This makes perfect sense under the hypothesis that [e o] derive fromai au Under that hypothesis wanaiastai undergoes glide formationbefore another vowel ( just as jadiaham does) giving [wanajaste]

Abstractness in English Now we will consider an abstract analysiswhose legitimacy has been questioned since the main point being madehere is that abstract analyses can be well motivated it is important toconsider what is not sufficient motivation for an abstract analysisA classic case of questionable abstractness is the analysis of English [ɔj]proposed in Chomsky and Halle 1968 (SPE) that [ɔj] derives from œ InSPE English vowels are given a very abstract analysis with approximatelythe following relations between underlying and surface representations ofvowels where ī ū and so forth represent tense vowels in the transcriptionused there

The first step in arguing for this representation is to defend the assump-tion that [aj] [aw] [ij] [uw] [ej] [ow] derive from ī ū ē ō ǣ and ɔThe claim is motivated by the Trisyllabic Laxing alternation in Englishwhich relates the vowels of divine ~ divinity ([aj] ~ [ɪ]) profound ~ profundity([aw] ~ [ə]) serene ~ serenity ([ij] ~ [ε]) verbose ~ verbosity ([ow] ~ [ɔ]) and sane ~sanity ([ej] ~ [aelig]) These word pairs are assumed to be morphologicallyrelated so both words in the pairs would have a common root thequestion is what the underlying vowel of the root is It is assumed thattense vowels undergo a process known as Vowel Shift which rotates atense vowelrsquos height one degree upward ndash low vowels become mid midvowels become high and high vowels become low Another processthat is relevant is Diphthongization which inserts a glide after a tense

eti lsquohe comesrsquo + rʂi lsquoseerrsquo etj rʂijadi lsquoif rsquo + aham lsquoIrsquo jadj ahamjadi lsquoif rsquo + aditjah lsquosons of Aditirsquo jadj aditjaheti lsquoshe comesrsquo + uma lsquoUmarsquo etj umabhawatu lsquolet it bersquo + iccedilwarah lsquoLordrsquo bhawatw iccedilwarahsadhu lsquowellrsquo + eti lsquohe comesrsquo sadhw eti

prabho lsquoO Masterrsquo + eti lsquohe comesrsquo prabhaw etiwane lsquoin the forestrsquo + aste lsquohe sitsrsquo wanaj astewane lsquoin the forestrsquo + eti lsquohe comesrsquo wanaj etiprabho lsquoO Masterrsquo + okʂat lsquohe sprinkledrsquo prabhaw okʂat

ī [aj] ū [aw]ē [ij] ō [uw]ǣ [ej] ɔ [ow]œ [ɔj] ā [ɔʌ]

vowel agreeing in backness with that vowel By those rules (and a fewothers) sǣn becomes [sējn] serēn becomes [sərījn] and divīn becomes[dəvajn] By the Trisyllabic Laxing rule when a tense vowel precedes thepenultimate syllable of the word the vowel becomes lax which preventsthe vowel from shifting in height (shifting only affects tense vowels)Accordingly [dəvajn] and [dəvɪnətij] share the root dəvīn In [dəvajn] thetense vowel diphthongizes to [dəvījn] which undergoes Vowel Shift Indəvīn-iti the vowel ī instead undergoes Trisyllabic Laxing and thereforesurfaces as [ɪ]In this way SPE reduces the underlying vowel inventory of English to ī

ū ē ī ǣ ā ɔ plus the diphthong ɔj Having eliminated most of thediphthongs from underlying representations we are still left with onediphthong In addition there is an asymmetry in the inventory thatEnglish has three out of four of the possible low tense vowels lackinga front round vowel [œ] It is then surmised that this gap in the systemof tense vowels and the remaining diphthong can be explained awaysimultaneously if [ɔj] derives from underlying œ Furthermore given thesystem of rules in SPE if there were an underlying vowel œ it wouldautomatically become [ɔj]Briefly œ undergoes diphthongization to become œj because œ is a

front vowel and the glide inserted by diphthongization has the samebackness as the preceding tense vowel The vowel œ is subject to backnessreadjustment which makes front low vowels [+back] before glides (by thesame process œj which derives from ī by Vowel Shift becomes [ay]) Sincehypothesized œ does not become [oslash] and must remain a low vowel inorder to undergo backness adjustment Vowel Shift must not apply to œThis is accomplished by constraining the rule to not affect a vowel whosevalues of backness and roundness are different

What constitutes a valid motivation This analysis of [ɔj] is typical ofhighly abstract phonological analyses advocated in early generative phon-ology where little concern was given to maintaining a close relationbetween surface and underlying forms The idea of deriving [ɔj] from œis not totally gratuitous since it is motivated by a desire to maintain amore symmetrical system of underlying representations But the goal ofproducing symmetry in underlying representations cannot be maintainedat all costs and whatever merits there are to a symmetrical more elegantunderlying representation must be balanced against the fact that abstractunderlying forms are inherently difficult for a child to learn Put simplythe decision to analyze English vowels abstractly is justified only by anesoteric philosophical consideration ndash symmetry ndash and we have no evi-dence that this philosophical perspective is shared by the child learningthe language If achieving symmetry in the underlying form isnrsquot a suffi-cient reason to claim that [ɔj] comes from œ what would motivate anabstract analysisAbstractness can easily be justified by showing that it helps to account

for phonological alternations as we have seen in Palauan TonkawaMatuumbi Hehe and Sanskrit No such advantage accrues to an abstract

analysis of [ɔj] in English The only potential alternations involving [ɔj] area few word pairs of questionable synchronic relatedness such as joint ~juncture point ~ puncture ointment ~ unctuous boil ~ bouillon joy ~ jubilantsoil ~ sully choice ~ choose voice ~ vociferous royal ~ regal This handful ofwords gives no support to the abstract hypothesis If underlying œ wereto undergo laxing the result should be the phonetically nonexistent vowel[œ] and deriving the mixture of observed vowels [ʌ] [ʊ] [uw] [ow] or [ij]from [œ] would require rather ad hoc rules The hypothesized underlyingvowel system ī ū ē ī ǣ ɔ œ runs afoul of an otherwise valid implicationalrelation in vowel systems across languages that the presence of a lowfront rounded vowel (which is one of the more marked vowels in lan-guages) implies the presence of nonlow front round vowels This typo-logical implicational principle would be violated by this abstract analysisof English which has no underlying y oslash in other words idealizationsabout underlying forms can conflictAn important aspect of the argument for [ɔj] as œ is the issue of

independent motivation for the rules that would derive [ɔj] The argumentfor those rules in particular Vowel Shift is not ironclad Its motivation insynchronic English hinges on alternations of the type divine ~ divinityprofound ~ profundity but these alternations are lexically restricted andtotally unproductive in English (unlike the phonological alternations inthe form of the plural suffix as well as the somewhat productive voicingalternation in life ~ lives) A consequence of the decision to analyze all casesof [aj] as deriving from ī is that many other abstract assumptions had tobe made to explain the presence of tense vowels and diphthongs in unex-pected positions (such as before the penultimate syllable)To account for the contrast between contrite ~ contrition where ī

becomes lax and t [ʃ ] versus right ~ righteous where there is no vowellaxing and t [tʃ] it was claimed that the underlying form of right is rixtand rules are developed whereby ixC [ajC] Abstract x is called on toexplain the failure of Trisyllabic Laxing in the word nightingale claimed toderive from nixtVngǣl To explain the failure of Trisyllabic Laxing inwords like rosary it is assumed that the final segment is j and not iviz rɔsVrj Other examples are that the contrast between veto (with noflapping and a secondary stress on [o]) vs motto (with flapping and nostress on [o]) was predicted by positing different vowels ndash mɔto vs vētɔeven though the vowel qualities are surface identical Words such asrelevance are claimed to contain an abstract nonhigh front glide whosefunction is to trigger assibilation of t and then delete so relevance wouldderive from relevante the symbol e representing a nonsyllabic nonhighfront vocoid (a segment not attested in any language to date)It is not enough to just reject these analyses as being too abstract since

that circularly answers the abstractness controversy by fiat We need topair any such rejection with an alternative analysis that states what we dodo with these words and this reanalysis formed a significant componentof post-SPE research More importantly we need to identify the methodo-logical assumptions that resulted in these excessively abstract analysesOne point which emerged from this debate is that a more conservative

stance on word-relatedness is called for A core assumption in phono-logical analysis is that underlying representations allow related words tobe derived from a unified source by rules The concept ldquorelated wordrdquoneeds to be scrutinized carefully because liberally assuming that ldquorelatedwordsrdquo have common underlying forms can yield very abstract analyses

Word-relatedness Consider word pairs such as happyglad talllong andyoungold Such words are ldquorelatedrdquo in having similar semantic propertiesbut they are not morphologically related and no one would proposederiving happy and glad from a single underlying root Nor would anyonepropose treating such pairs as brainbrandy painpantry graingrant asinvolving a single underlying root since there is no semantic relationbetween members of the pair Pairs such as fivepunch are related historic-ally but the connection is known only to students of the history ofEnglish The words father and paternal are related semantically and phono-logically but this does not mean that we can derive father and paternalfrom a common root in the grammar of English It may be tempting toposit relations between choir and chorus shield and shelter or hole andhollow but these do not represent word-formation processes of modernEnglish grammarThe concept of ldquorelatednessrdquo that matters for phonology is in terms of

morphological derivation if two words are related they must have somemorpheme in common It is uncontroversial that words such as cook andcooked or book and books are morphologically related in a synchronicgrammar the words share common roots cook and book via highly pro-ductive morphological processes which derive plurals of nouns and past-tense forms of verbs An analysis of word formation which failed tocapture this fact would be inadequate The relation between tall andtallness or compute and computability is similarly undeniable In such casesthe syntactic and semantic relations between the words are transparentand the morphological processes represented are regular and productiveSome morphological relations are not so clear -ment attaches to some

verbs such as bereavement achievement detachment deployment paymentplacement allotment but it is not fully productive since we donrsquot havethinkment takement allowment intervenement computement givementThere are a number of verbnoun pairs like explainexplanation declinedeclination definedefinition impressimpression confuseconfusion whichinvolve affixation of -(Vt)-ion but it is not fully productive as shown bythe nonexistence of pairs like containcontanation refinerefination stressstression imposeimposion abuseabusion Since it is not totally predictablewhich -ion nouns exist or what their exact form is these words may justbe listed in the lexicon If they are there is no reason why the words couldnot have slightly different underlying formsIt is thus legitimate to question whether pairs such as verboseverbosity

profoundprofundity divinedivinity represent cases of synchronic derivationfrom a single root rather than being phonologically and semanticallysimilar pairs of words which are nevertheless entered as separate andformally unrelated lexical items The question of how to judge formal

word-relatedness remains controversial to this day and with it manyissues pertaining to phonological abstractness

82 Independent evidence historicalrestructuring

Paul Kiparskyrsquos seminal 1968 paper ldquoHow abstract is phonologyrdquo raisesthe question whether limits on abstractness are possible and desirableKiparskyrsquos concern is the postulation of segments which are never real-ized where a language is assumed to have an underlying distinctionbetween two segments which are always phonetically merged A classicexample is Hungarian which has a vowel harmony rule where suffixvowels agree with the preceding vowel in backness eg haz-am lsquomyhousersquo fylem lsquomy earrsquo viz-em lsquomy waterrsquo A small number of roots withthe front vowels [i i e] always have back vowels in suffixes eg heɟ-amlsquomy rindrsquo ɲilam lsquomy arrowrsquo The abstract analysis is that these roots haveunderlying back vowels [ɨ ɨ ə] which later become front vowels Thismove makes these roots phonologically regular The reasoning is thatsince these front vowels seem to act as though they are back vowels interms of the vowel harmony system maybe they really are back vowels ata deeper levelKiparsky terms this kind of analysis absolute neutralization to be

distinguished from contextual neutralization In contextual neutraliza-tion the distinction between two underlying segments is neutralized insome contexts but is preserved in others Final devoicing in Russian iscontextual neutralization because in the words porok and porog thedistinction between k and g is neutralized in the nominative singular[porok] but is maintained in genitive [poroka] vs [poroga] With absoluteneutralization the distinction is eliminated in all contexts and thus inHungarian ɨ is always neutralized with i Kiparsky argues that whilecontextual neutralization is common and has demonstrable psychologicalreality absolute neutralization is a theoretically constructed fictionIn arguing against absolute neutralization Kiparsky faces the challenge

that a number of cases of such abstractness had been postulated so goodreasons for rejecting those analyses must be found Kiparsky focuses onthe extent to which the psychological reality of theoretical constructs canbe measured ndash this is an important consideration since linguistic theoriesare usually intended to be models of the psychological processes under-lying linguistic behavior The problem is that it is impossible to directlytest whether linguistic constructs are psychologically valid by any simpleor obvious tests Linguistic properties are highly abstract and not easilytested in the same way that one can experimentally test the ability toperceive touch or distinguish colors or sounds Kiparsky argues that onecan in certain circumstances use the pattern of language change as atheory-external test of grammatical theories It is argued that historicalsound change can provide just such a test

An abstract phonological distinction cannot be justified on the basis ofthe fact that two historically distinct sounds merge in the history of alanguage so even if it were shown that Hungarian heɟ lsquorindrsquo and ɲil lsquomyarrowrsquo derived from earlier həɟ and ɲɨl this would not be evidence foran abstract underlying form in modern Hungarian A child learning thelanguage has no access to this kind of historical information WhatKiparsky points out is that you can inspect a later stage of a language tolearn about the analysis of a language that was actually given at an earlierstage of the language and then adduce general principles about gram-mars based on such independent evidence

821 Yiddish final devoicingThe history of Yiddish devoicing is one example of such evidence In theoldest forms of German represented by Old High German there was norestriction against word-final voiced consonants so Old High German hadwords like tag lsquodayrsquo ~ taga lsquodaysrsquo gab lsquohe gaversquo ~ gābumes lsquowe gaversquo sneidlsquohe cutrsquo ~ snīdan lsquoto cutrsquo hand lsquohandrsquo land lsquolandrsquo Between 900 and 1200 inthe Middle High German period a rule of devoicing was added whichresulted in tac lsquodayrsquo ~ tage lsquodaysrsquo gap lsquohe gaversquo ~ gāben lsquowe gaversquo sneit lsquohecutrsquo ~ snīden lsquoto cutrsquo hant lsquohandrsquo ~ hende lsquohandsrsquo wec lsquoroadrsquo ~ weges lsquoroadsrsquoAround this time Yiddish began to develop as a language separate from

German and would have shared this devoicing rule Devoicing of finalconsonants in Yiddish is attested in manuscripts from the thirteenthcentury where the word for lsquodayrsquo is written lttakgt using the letter kuf[k] and not gimel [g] In some dialects such as Central and WesternYiddish this devoicing persists up to today where you find tak lsquodayrsquo ~tag-n lsquodaysrsquo lant lsquolandrsquo ~ lend-ər lsquolandsrsquo with the stem-final voiced conson-ants of tag and land undergoing final devoicing in the singular In somedialects such as the Northeastern dialect of Yiddish the devoicing rulewas lost from the grammar so that dialect has tog lsquodayrsquo ~ tog-n lsquodaysrsquowhere the originally voiced consonant reappears as voiced This processwhere an earlier sound change is dropped from the grammar is known asreversal of sound change consonants revert to their original state foundbefore the sound change appliedThere are mysterious exceptions to restoration of original voiced con-

sonants One case is the word gelt lsquomoneyrsquowhich derives historically fromgeld with a voiced consonant The reason for the different treatments ofgelt and tag words which both ended with voiced consonants at earlierstages of the language is the difference in the presence or absence ofphonological alternations within the paradigm of a word In the case oftag the plural form had a suffix -n and so while the singular was subjectto devoicing the plural was not this word had the paradigmatic alterna-tions [tak] ~ [tagn] On the basis of these alternations a child learning thelanguage would have no problem discovering that the underlying form ofthe stem is tag It is expected that once the final devoicing rule is lost theunderlying form tag resurfaces since there is no longer a devoicing ruleIn the word gelt the situation was different There was no inflectional

ending which followed this particular noun At the earliest stages of the

language a child learning the language only encounters [geld] and therewould be no basis for assuming that the underlying form is anythingother than geld When the devoicing rule was added to the grammarthe pronunciation of the word changed to [gelt] Since this particularconsonant was always word-final the devoicing rule would have alwaysapplied to it so the stem only had the phonetic form [gelt] Althougheither geld or gelt as underlying form would yield the surface form[gelt] there is no reason to assume that the surface and underlying formsare different A priori criteria may support one decision or the other butwhat we need to know is what independent test tells us that ourreasoning is correct The loss of the devoicing rule provides exactly theneeded empirical test it allows us to know what underlying form Yiddish-learning children must have assumed at this earlier stage Knowing theactual underlying form provides an important insight into the learningstrategies that children make during language acquisitionWhen the devoicing rule was added there were no alternations in gelt

so a child would have no reason to assume that the underlying form of theword is anything other than gelt The child never hears geld and has noreason to think that the underlying form is different from gelt At aneven later stage the rule of final devoicing is dropped from the grammarof certain dialects This allows the underlying and historically originalvoiced consonant of tag to be pronounced again since it is no longersubject to devoicing and thanks to the paradigmatic k ~ g alternation theunderlying form was established as being tag This rule loss has no effecton gelt since despite being derived historically from a voiced consonantthe final consonant of the stem had been reanalyzed as t ndash a reanalysispredicted by the presumption that an underlying form is different fromthe surface form only if there is good reason for assuming so Becausethere are no alternations for this word there was no reason to assume anabstract underlying formAnother important kind of exception to the reversal of devoicing is seen

in the adverb avek lsquoawayrsquo This word was originally aveg with a voicedconsonant This adverb also had no inflected relatives which allowed theunderlying voicing of the final consonant to be unambiguously deter-mined so once the devoicing rule was added to the grammar it wasimpossible to determine whether the underlying form was avek oraveg Again starting from the assumption that underlying forms donot deviate from surface forms without reason there is no reason toassume that phonetic [avek] derives from anything other than avek sincethe word is actually pronounced [avek] The fact that the underlying formis directly revealed as avek in the dialects which dropped devoicing sup-ports this decisionThe example also reveals something interesting about what might (but

does not) constitute a ldquoreasonrdquo for abstractness The adverb avek is histor-ically related to the noun veg lsquowayrsquo The voicing of the last consonant inthe noun stem can be recovered within the paradigm given the earlieralternations vek lsquowayrsquo ~ vegn lsquowaysrsquo because the singular and plural formsof the noun are clearly related to each other The evidence from the plural

noun had no impact on the childrsquos selection of the underlying form for theadverb since there is no synchronic connection between the adverb andthe noun ndash no process derives nouns and adverbs from a unified source sonothing connects the words for lsquowayrsquo and lsquoawayrsquo The divergence of vegand avek in Yiddish points out that you cannot freely assume that any twophonetically and semantically similar words are actually derived from asingle underlying form

822 Historical evidence and the treatmentof absolute neutralization

Kiparsky draws two main conclusions from this and similar cases First hepoints out that in lieu of alternations supporting abstractness the surfaceand underlying forms should be assumed to be identical alternations arecentral to supporting an abstract underlying form Second and more con-troversially these examples are used in an argument against the psycho-logical reality of absolute neutralization The argument is as follows Casessuch as Yiddish show the psychological reality of contextual neutralizationsince it can be reversed However there is no known case where absoluteneutralization has been historically reversed if absolute neutralization hadthe psychological reality of contextual neutralization we would expect tofind a reversal of absolute neutralization and we have not Thereforeputative cases of absolute neutralization lack psychological realityKiparsky proposes that morphemes which seem to motivate abstract

segments are simply lexical exceptions to the rule in question they failto undergo or trigger a rule For the problematic roots of Hungarianwhere front vowels seem to trigger back harmony such as heɟ-am lsquomyrindrsquo ɲil-am lsquomy arrowrsquo the proposal is that these roots are marked asexceptions to vowel harmony On the assumption that harmonizing suf-fixes all contain underlying back vowels the fact that back vowels appearin suffixes after these roots boils down to the fact that the suffixes haveunderlying back vowels and since these roots do not trigger vowel har-mony the underlying vowel quality is preserved on the surface

83 Well-motivated abstractness

While it is certainly true that some putative processes of absoluteneutralization are not well supported and the abstract property onlydiacritically marks a root as an exception to one rule there are internallywell-supported cases of absolute neutralization Two famous cases areYawelmani discussed by Kisseberth (1969) and Maltese discussed byBrame (1972)

831 Yawelmani uAspects of Yawelmani have been discussed in chapter 6 Two of themost important processes are vowel harmony and vowel shortening Theexamples in (30) demonstrate the basics of vowel harmony a suffix vowelbecomes rounded if it is preceded by a round vowel of the same height

Thus the root vowel o has no effect on the suffixes hin and it but causesrounding of krsquoa and al mdash and the converse holds of the vowel uThe data in (31) show that long vowels cannot appear before two

consonants These stems have underlying long vowels and when followedby a consonant-initial affix the vowel shortens

Another class of verb roots has the surface pattern CVCVC ndash the peculiarfact about these roots is that the first vowel is always a short version of thesecond vowel

In [woʔuj-hun] [dolul-hun] the second vowel is epenthetic so these rootsunderlyingly have the shape CVCC parallel to [ʔamil-hin] ~ [ʔamlal]lsquohelprsquoThere are problematic roots in (33) Although the stem vowel is a mid

vowel a following nonhigh vowel does not harmonize ndash they seem to beexceptions Worse a high vowel does harmonize with the root vowel eventhough it does not even satisfy the basic phonological requirement forharmony (the vowels must be of the same height)

A noteworthy property of such roots is that their vowels are alwayslongThere is another irregularity connected with certain surface mid

vowels The data in (34) illustrate a set of CVCVV(C) roots where as wenoticed before the two vowels are otherwise identical In these verbs thesecond long vowel is a nonhigh version of the first vowel

Nonfuture Imperative Dubitative Passive aoristxat-hin xat-krsquoa xat-al xat-it lsquoeatrsquodub-hun dub-krsquoa dub-al dub-ut lsquolead by handrsquoxil-hin xil-krsquoa xil-al xil-it lsquotanglersquokrsquooʔ-hin krsquooʔ-krsquoo krsquooʔ-ol krsquooʔ-it lsquothrowrsquo

Nonfuture Imperative Dubitative Passive aoristdos-hin dos-krsquoo dos-ol dos-it lsquoreportrsquoʂap-hin ʂap-krsquoa ʂap-al ʂap-it lsquoburnrsquomekrsquo-hin mekrsquo-krsquoa mekrsquo-al mekrsquo-it lsquoswallowrsquo

Nonfuture Imperative Dubitative Passive aoristprsquoaxat-hin prsquoaxat-krsquoa prsquoaxat-al prsquoaxat-it lsquomournrsquoʔopot-hin ʔopot-krsquoo ʔopot-ol ʔopot-it lsquoarise from bedrsquojawal-hin jawal-krsquoa jawal-al jawal-it lsquofollowrsquo

Nonfuture Imperative Dubitative Passive aoristcrsquoom-hun crsquoom-krsquoa crsquoom-al crsquoom-ut lsquodestroyrsquoʂog-hun ʂog-krsquoa ʂog-al ʂog-ut lsquouncorkrsquowoʔuj-hun woʔuj-krsquoa woʔj-al woʔj-ut lsquofall asleeprsquodolul-hun dolul-krsquoa doll-al doll-ut lsquoclimbrsquo

The surface mid vowels of these stems act irregularly for harmony ndash theydo not trigger harmony in mid vowels so they do not act like other midvowels They also exceptionally trigger harmony in high vowels as onlyhigh vowels otherwise doWhen you consider the vowels of Yawelmani ndash [i e a o u e o a] ndash you see

that long high vowels are lacking in the language The preceding myster-ies are solved if you assume for instance that the underlying stem of theverb lsquoscorchrsquo is tunuj As such the root would obey the canonicalrestriction on the vowels of a bivocalic stem ndash they are the same vowel ndashand you expect u to trigger harmony on high vowels but not on midvowels as is the case A subsequent rule lowers u to [o] merging thedistinction between underlying o and uThe assumption that u becomes [o] and therefore some instances of

[o] derive from u explains other puzzling alternations There is a vowel-shortening process which applies in certain morphological contexts Onecontext is the causative which adds the suffix -ala and shortens thepreceding stem vowel

We have seen in (33) that the root [crsquoom] has the phonological character-istics of an abstract vowel so given the surface-irregular pattern of vowelharmony in crsquoom-hun crsquoom-krsquoa we can see that the underlying vowel mustbe a high vowel The fact that the vowel actually shows up as a high vowelas a result of the morphologically conditioned shortening rule givesfurther support to the hypothesized abstract underlying vowelThe approach which Kiparsky advocates for absolute neutralization

does not work for Yawelmani these words are not exceptions Being anexception has a specific meaning that a given morpheme fails to undergoor trigger a rule which it otherwise would undergo The fact that vowelharmony does not apply in crsquoom-al can be treated as exceptionality Butthis root does actually trigger vowel harmony as shown by crsquoom-ut andsuch application is problematic since the rule is applying when the formalconditions of the rule are not even satisfied on the surface Marking a rootas an exception says that although the root would be expected to undergoa rule it simply fails to undergo the rule What we have in Yawelmani is

Nonfuture Imperative Dubitative Passive aoristhiwet-hin hiwet-krsquoa hiwet-al hiwet-it lsquowalkrsquoʔile-hin ʔile-krsquo ʔile-l ʔile-t lsquofanrsquoʂudokrsquo-hun ʂudokrsquo-krsquoa ʂudokrsquo-al ʂudokrsquo-ut lsquoremoversquotrsquounoj-hun trsquounoj-krsquoa trsquounoj-al trsquounoj-ut lsquoscorchrsquocrsquoujo-hun crsquoujo-krsquo crsquoujo-l crsquoujo-t lsquourinatersquo

Nonfuture plain Nonfuture causativetis-hin tis-ala-hin lsquocome outrsquohojo-hin hoj-olo-hin lsquohave a namersquomekrsquo-hin mikrsquo-ala-hin lsquoeatrsquocrsquoom-hun crsquoum-ala-hin lsquodestroyrsquo

something different ndash a form is triggering a rule even though it shouldnot The exceptionality analysis also offers no account of stems such ascrsquoujo-hun where the first vowel should have been a copy of the secondvowel but instead shows up as a high vowel nor does the exceptionalityaccount have any way to explain why the ldquoexceptionalrdquo roots show upwith high vowels when the root is subject to morphological vowelshortening as in crsquoom-hun ~ crsquoum-ala-hinAlthough the specific segment u is not pronounced as such in the

language concern over the fact that pronunciations do not include thatparticular segment would be misguided from the generative perspectivewhich holds that language sounds are defined in terms of features and theprimary unit of representation is the feature not the segment All of thefeatures comprising u ndash vowel height roundness length ndash are observedin the surface manifestations of the abstract vowels

832 Maltese ʕAnother well-supported case of absolute neutralization comes from Mal-tese We will just outline the basics of the argument you should readBrame (1972) to understand the full argument After outlining some basicphonological processes we consider examples which seem superficiallyinexplicable but which can be explained easily if we posit an abstractunderlying consonant ʕ

8321 Basic Maltese phonologyStress and apocope (36) examplifies two central processes of the lan-guage namely stress assignment and apocope Disregarding one conson-ant at the end of the word the generalization is that stress is assigned tothe last heavy syllable ndash one that ends in a (nonfinal) consonant or onewith a long vowel

The second group illustrates apocope which deletes an unstressed vowelfollowed by CV The underlying stem of the word for lsquograbbedrsquo is ħatafseen in the third-singular masculine form After stress is assigned in third-singular feminine ħaacutetaf-et (37) gives surface [ħatf-et]

In ħataf-t stress is assigned to thefinal syllable since that syllable is heavy (onlyone final consonant is disregarded in making the determination whethera syllable is heavy) and therefore the initial vowel is deleted giving [ħtaacuteft]

seacutena lsquoyearrsquo sultaacutean lsquokingrsquoʔattuacuteus lsquocatrsquo ħduacuteura lsquogreennessrsquoħaacutetaf lsquohe grabbedrsquo beacutezaʔ lsquohe spatrsquoħaacutetf-et lsquoshe grabbedrsquo beacutezʔ-et lsquoshe spatrsquoħtaacutef-t lsquoI grabbedrsquo bzaacuteʔ-t lsquoI spatrsquoħtaacutef-na lsquowe grabbedrsquo bzaacuteʔ-na lsquowe spatrsquo

V Oslash _ CV Apocope[ndashstress]

Unstressed reduction and harmony Two other rules are unstressed-vowel reduction and vowel harmony By the former process motivated in(38) unstressed i reduces to e The third-singular feminine suffix is under-lyingly -it which you can see directly when it is stressed The underlyingform of kiacuteteb is kitib When stress falls on the first syllable of this rootthe second syllable reduces to e but when stress is final the secondsyllable has i

Thus the following rule is motivated

By vowel harmony i becomes [o] when preceded by o

Surface koacuterb-ot derives from korob-it by applying stress assignment thevowel harmony in (41) and apocope

Epenthesis The data in (42) illustrate another rule which inserts [i]before a word-initial sonorant that is followed by a consonant

Stress assignment and apocope predict laʔat-na lʔaacutet-na the resultingconsonant cluster sonorant plus obstruent sequence is eliminated by thefollowing rule

(43) Oslash i _ [+ sonor] C Epenthesis

Regressive harmony and precoronal fronting These rules apply in theimperfective conjugation which has a prefix ni- lsquo1st personrsquo ti- lsquo2nd personrsquo

haacutetf-et lsquoshe grabbedrsquo ħatf-iacutet-kom lsquoshe grabbed you (pl)rsquobeacutezʔ-et lsquoshe spatrsquo bezʔ-iacutet-l-ek lsquoshe spat at yoursquokiacuteteb lsquohe wrotersquo ktiacuteb-t lsquoI wrotersquo

i [-high] Unstressed V-reduction[-stress]

koacuterob lsquohe groanedrsquo koacuterb-ot lsquoshe groanedrsquoʃoacuterob lsquohe drankrsquo ʃoacuterb-ot lsquoshe drankrsquo

i [+round] V C0 _ Harmony[+round]

laacuteʔat lsquohe hitrsquo roacuteħos lsquoit (masc) became cheaprsquolaacuteʔt-et lsquoshe hitrsquo roacuteħs-ot lsquoit (fem) became cheaprsquoilʔaacutet-t lsquoI hitrsquo irħoacutes-t lsquoI became cheaprsquoilʔaacutet-na lsquowe hitrsquo irħoacutes-na lsquowe became cheaprsquomaacuterad lsquohe became sickrsquo neacutefaħ lsquohe blewrsquo

maacuterd-et lsquoshe became sickrsquo neacutefħ-et lsquoshe blewrsquo

imraacuted-t lsquoI became sickrsquo infaacuteħ-t lsquoI blewrsquo

imraacuted-na lsquowe became sickrsquo infaacuteħ-na lsquowe blewrsquo

or ji- lsquo3rd personrsquo plus a suffix -u lsquopluralrsquo for plural subjects The under-lying prefix vowel i is seen in the following data

When the first stem vowel is o the prefix vowel harmonizes to o

This can be explained by generalizing harmony (41) so that it appliesbefore or after a round vowel The nature of the stem-initial consonantis important in determining whether there is surface harmony if the firstconsonant is a coronal obstruent there appears to be no harmony

Examples such as noacute-bzoʔ show that if the coronal obstruent is not imme-diately after the prefix vowel harmony applies The explanation forapparent failure of harmony is simply that there is a rule fronting o whena coronal obstruent follows

(47) _o -back[ [

rarr⎡⎢⎣

⎤⎥⎦

+ cor- son

Guttural lowering Another process lowers i to a before the ldquogutturalrdquoconsonants ʔ and ħTreating glottal stop as [+low] is controversial since that contradicts the

standard definition of [+low] involving tongue lowering Recent researchin feature theory shows the need for a feature that includes laryngealglides in a class with low vowels and pharyngeal consonants

niacute-msaħ lsquoI wipersquo tiacute-msaħ lsquoyou wipersquoniacute-ʃbaħ lsquoI resemblersquo tiacute-ʃbaħ lsquoyou resemblersquoniacute-kteb lsquoI writersquo tiacute-kteb lsquoyou writersquoniacute-tlef lsquoI losersquo tiacute-tlef lsquoyou losersquo

noacute-bzoʔ lsquoI spitrsquo toacute-bzoʔ lsquoyou spitrsquonoacute-krob lsquoI groanrsquo toacute-krob lsquoyou groanrsquonoacute-ħlom lsquoI dreamrsquo toacute-ħlom lsquoyou dreamrsquo

noacute-ʔtol lsquoI killrsquo toacute-ʔtol lsquoyou killrsquonoacute-rbot lsquoI tiersquo toacute-rbot lsquoyou tiersquonoacute-lʔot lsquoI hitrsquo toacute-lʔot lsquoyou hitrsquo

niacute-drob lsquoI woundrsquo tiacute-drob lsquoyou woundrsquoniacute-tlob lsquoI prayrsquo tiacute-tlob lsquoyou prayrsquoniacute-skot lsquoI become silentrsquo tiacute-skot lsquoyou become silentrsquoniacute-zloʔ lsquoI sliprsquo tiacute-zloʔ lsquoyou sliprsquoniacute-ʃrob lsquoI drinkrsquo tiacute-ʃrob lsquoyou drinkrsquo

naacute-ʔsam lsquoI dividersquo taacute-ʔsam lsquoyou dividersquonaacute-ʔbel lsquoI agreersquo taacute-ʔbel lsquoyou agreersquonaacute-ħrab lsquoI fleersquo taacute-ħrab lsquoyou fleersquonaacute-ħleb lsquoI milkrsquo taacute-ħleb lsquoyou milkrsquo

This motivates the following rule

Metathesis (50) and (51) illustrate another process When the stem hasa medial obstruent the prefix vowel is stressed and the stem vowel deletesbefore -u

This is as expected underlying ni-msaħ-u is stressed on the first syllableand the medial unstressed vowel deletes because it is followed by CVThe example [noacutebzʔu] from ni-bzoʔ-u shows that harmony must precedeapocope since otherwise apocope would have deleted the stem vowelwhich triggers harmonyWhen the second stem consonant is a sonorant in the presence of the

suffix -u the prefix has no stress and the stem retains its underlyingvowel which is stressed Unstressed i reduces to [e] so [niacute-dneb] derivesfrom ni-dnib The underlying high vowel is revealed when the stemvowel is stressed as in [nidiacutenbu]

Based solely on stress assignment and apocope as illustrated in (50)wewouldpredict niacutednbu noacutetlbu This again would result in an unattested consonantcluster in the syllable onset ndash a sonorant followed by an obstruent ndashwhich isavoided by a process of vocalic metathesis whereby niacute-tlif-u ni-tiacutelf-u

In some stems which undergo (52) the vowel alternates between i and a

i [+ low] _ C Guttural lowering[+ low]

niacute-msaħ lsquoI wipersquo niacute-msħ-u lsquowe wipersquonoacute-bzoʔ lsquoI spitrsquo noacute-bzʔ-u lsquowe spitrsquoniacute-dħol lsquoI enterrsquo niacute-dħl-u lsquowe enterrsquonaacute-ʔsam lsquoI dividersquo naacute-ʔsm-u lsquowe dividersquonaacute-ħdem lsquoI workrsquo naacute-ħdm-u lsquowe workrsquo

niacute-dneb lsquoI sinrsquo ni-diacutenb-u lsquowe sinrsquoniacute-tlef lsquoI losersquo ni-tiacutelf-u lsquowe losersquoniacute-tlob lsquoI prayrsquo ni-toacutelb-u lsquowe prayrsquonoacute-krob lsquoI groanrsquo no-koacuterb-u lsquowe groanrsquonoacute-ʔmos lsquoI kickrsquo no-ʔoacutems-u lsquowe kickrsquonaacute-ħrab lsquoI fleersquo na-ħaacuterb-u lsquowe fleersquonaacute-ħraʔ lsquoI burnrsquo na-ħaacuterʔ-u lsquowe burnrsquonaacute-ʔleb lsquoI overturnrsquo na-ʔiacutelb-u lsquowe overturnrsquo

V C C Vi C V V C Vi C C V Metathesis[+ son]

niacute-fraħ lsquoI rejoicersquo ni-fiacuterħ-u lsquowe rejoicersquoniacute-tlaʔ lsquoI leaversquo ni-tiacutelʔ-u lsquowe leaversquoniacute-sraʔ lsquoI stealrsquo ni-siacuterʔ-u lsquowe stealrsquo

Theunderlying stemvowel is i in these casesWhenno vowel suffix is addedunderlying ni-friħ becomes [niacute-fraħ] by Guttural Lowering (49) When -u isadded metathesis moves underlying i away from the guttural consonantwhich triggered lowering hence the underlying vowel is directly revealed

Stems with long vowels The stems which we have considered previ-ously are of the underlying shape CVCVC There are also stems with theshape CVVC illustrated in the perfective aspect in (54)

These stems exhibit a process of vowel shortening where aa becomes o or i(the choice is lexically determined) before a CC cluster

(55) aa io _ CC

When the imperfective prefixes ni- ti- are added to stems beginning with along vowel stress is assigned to that vowel and the prefix vowel is deletedIn the case of the first-person prefix ni this results in an initial nCcluster which is repaired by inserting the vowel i

From ni-duur you expect stress to be assigned to thefinal syllable because ofthe long vowel Since the vowel of ni is unstressed and in an open syllable itshould delete giving nduacuteur The resulting cluster then undergoes epenthesis

8322 Apparent irregularities A number of verbs seem to be irregularand yet they are systematic in their irregularity the irregularity is only interms of the surface form which can be made perfectly regular bypositing an abstract underlying consonant ʕ One set of examples is seenin the data in (57) where the stem contains a surface long vowel This longvowel is unexpectedly skipped over by stress assignment unlike verbswith underlying long vowels such as in-duacuteur lsquoI turnrsquo seen in (54)

daacutear lsquohe turnedrsquo saacutear lsquoit (masc) grew ripersquodaacutear-et lsquoshe turnedrsquo saacutear-et lsquoit (fem) grew ripersquodaacutear-u lsquothey turnedrsquo saacutear-u lsquothey grew ripersquodoacuter-t lsquoI turnedrsquo siacuter-t lsquoI became ripersquodoacuter-na lsquowe turnedrsquo siacuter-na lsquowe became ripersquodoacuter-tu lsquoyou turnedrsquo siacuter-tu lsquoyou became ripersquo

in-duacuteur lsquoI turnrsquo in-siacuteir lsquoI become ripersquot-duacuteur lsquoyou turnrsquo t-siacuteir lsquoyou become ripersquoin-suacuteuʔ lsquoI driversquo in-ziacuteid lsquoI addrsquot-suacuteuʔ lsquoyou driversquo t-ziacuteid lsquoyou addrsquo

niacute-sool lsquoI coughrsquo ni-soacuteol-u lsquowe coughrsquoniacute-laab lsquoI playrsquo ni-laacuteab-u lsquowe playrsquoniacute-baat lsquoI sendrsquo ni-baacuteat-u lsquowe sendrsquonoacute-ʔood lsquoI stayrsquo no-ʔoacuteod-u lsquowe stayrsquonoacute-bood lsquoI hatersquo no-boacuteod-u lsquowe hatersquo

The location of stress and the retention of the prefix vowel in noacute-ʔood isparallel to the retention of the prefix vowel in other tri-consonantal stemsin (44)ndash(48) such as niacute-msaħ lsquoI wipersquo If the underlying stem of niacute-sool had aconsonant ie were sXol where X is some consonant yet to be fullyidentified the parallelism with ni-msaħ and the divergence from in-duacuteurwould be explained The surface long vowel in niacutesool would derive by acompensatory lengthening side effect coming from the deletion of theconsonant X in niacute-sXolAnother unexpected property of the stems in (57) is that when the

plural suffix -u is added the prefix vowel is stressless and unelided in anopen syllable and the stress shifts to the stem eg ni-soacuteol-u lsquowe coughrsquoThus contrast ni-soacuteol-u with niacute-msħ-u lsquowe wipersquo which differ in thisrespect and compare ni-soacuteol-u to ni-ʃoacuterb-u lsquowe drinkrsquo which are closelyparallel Recall that if the medial stem consonant is a sonorant expectedV-CRC-V instead undergoes metathesis of the stem vowel around themedial consonant so ni-ʃrob-u becomes ni-ʃoacuterb-u (creating a closed syllablewhich attracts stress) If we hypothesize that the underlying stem is sXolthen the change of ni-sXol-u to ni-soacuteXl-u (phonetic nisoacuteolu) would makesense andwould further show that X is a sonorant consonant ʕ qualifies asa sonorant (it involves minimal constriction in the vocal tract)Another pecularity is that these long vowels resist shortening before CC

In contrast to examples in (54) such as daacutear lsquohe turnedrsquo doacuter-t lsquoI turnedrsquowith vowel shortening before CC these long vowels do not shortenContinuing with the hypothesis of an abstract consonant in soXol weexplain the preservation of the long vowel in [soacuteolt] if this form derivesfrom sXol-t where deletion of X (which we suspect is specifically ʕ)lengthens the vowel and does so after vowel shortening has appliedThere is a further anomaly in a subset of stems with the consonant X in

the middle of the root if the initial stem consonant is a sonorant epen-thetic i appears when a consonant-initial suffix is added Compare (59a)where the first consonant is not a sonorant with (59b) where the firstconsonant is a sonorant

The verbs in (59b) behave like those in (42) eg laacuteʔat lsquohe hitrsquo ~ ilʔaacutet-tlsquoI hitrsquo where the initial sonorant + C cluster undergoes epenthesis of i

soacuteol lsquohe coughedrsquo soacuteolt lsquoI coughedrsquo soacuteolna lsquowe coughedrsquosoacuteob lsquohe lamentedrsquo soacuteobt lsquoI lamentedrsquo soacuteobna lsquowe lamentedrsquoʔaacutead lsquohe stayedrsquo ʔaacuteadt lsquoI stayedrsquo ʔaacuteadna lsquowe stayedrsquobaacutead lsquohe hatedrsquo baacuteadt lsquoI hatedrsquo baacuteadna lsquowe hatedrsquo

a ʔaacutead lsquohe stayedrsquo ʔaacuteadt lsquoI stayedrsquobaacutead lsquohe hatedrsquo baacuteadt lsquoI hatedrsquosoacuteol lsquohe coughedrsquo soacuteolt lsquoI coughedrsquo

b maacutead lsquohe chewedrsquo imaacuteadt lsquoI chewedrsquonaacuteas lsquohe dozedrsquo inaacuteast lsquoI dozedrsquolaacuteaʔ lsquohe lickedrsquo ilaacuteaʔt lsquoI lickedrsquo

The forms in (59b) make sense on the basis of the abstract forms maacuteʕad ~mʕaacutedt where the latter form undergoes vowel epenthesis and then theconsonant ʕ deletes lengthening the neighboring vowel Before ʕ isdeleted it forms a cluster with the preceding sonorant which triggersthe rule of epenthesisOther mysteries are solved by positing this consonant in underlying

forms In (60) the first stem consonant appears to be a coronal obstruentWe have previously seen that when the stem-initial consonant is a cor-onal obstruent vowel harmony is undone (niacute-tlob lsquoI prayrsquo) so (60) isexceptional on the surface In addition the prefix vowel is unexpectedlylong whereas otherwise it has always been short

These forms are unexceptional if we assume that the initial consonantof the stem is not d dʒ t but the abstract consonant ʕ thus ʕdos ʕdʒobʕtor ʕ is not a coronal obstruent so it does not cause fronting of theprefix vowelOther examples provide crucial evidence regarding the nature of this

abstract consonant The data in (61) show a lengthened prefix vowelwhich argues that the stems underlyingly have the initial abstract con-sonant that deletes and causes vowel lengthening [naacutealaʔ] comes fromni-ʕlaʔ

In addition the quality of the prefix vowel has changed from i to [aa]even though in these examples the consonant which follows on thesurface is a coronal If the abstract consonant is a pharyngeal as we havehypothesized then the vowel change is automatically explained by theGuttural Lowering ruleWe have considered stems where the first and second root consonants

are the consonant ʕ now we consider root-final ʕ The data in (62) showexamples of verbs whose true underlying imperfective stems are CCV

The plural suffix u becomes [w] after final a Although the second con-sonant is a sonorant the metathesis rule does not apply in naacuteʔraw becauseno cluster of consonants containing a sonorant in the middle would resultNow compare verbs with a medial sonorant where the final consonant

is hypothesized ʕ The singular columns do not have any striking irregu-larities which distinguish them from true CVCV stems

noacuteodos lsquoI diversquo toacuteodos lsquoyou diversquonoacuteodʒob lsquoI pleasersquo toacuteodʒob lsquoyou pleasersquonoacuteotor lsquoI stumblersquo toacuteotor lsquoyou stumblersquo

naacutealaʔ lsquoI closersquo taacutealaʔ lsquoyou closersquonaacuteasar lsquoI squeezersquo taacuteasar lsquoyou squeezersquonaacutearaʃ lsquoI ticklersquo taacutearaʃ lsquoyou ticklersquo

naacute-ʔra lsquoI readrsquo naacute-ʔra-w lsquowe readrsquoniacute-mla lsquoI fillrsquo niacute-mla-w lsquowe fillrsquo

The prefix vowel is unstressed and in an open syllable which is found onlyin connection with metathesis but metathesis is invoked only to avoidclusters with a medial sonorant which would not exist in hypothetical[niacuteblau] This is explained if the stem ends with ʕ Thus ni-smiʕ-ushould surface as nisiacutemʕu by analogy to ni-tlob-u [nitoacutelbu] lsquowe askrsquoThe consonant ʕ induces lowering of the vowel i and ʕ itself becomes agiving the surface formA final set of examples provides additional motivation for assuming

underlying ʕ Participles are formed by giving the stem the shape CCVVCselecting either ii or uu As the data in (64) show stems ending in theconsonant ʕ realize that consonant as [ħ] after long high vowels

These data provide evidence bearing on the underlying status of theabstract consonant since it actually appears on the surface as a voicelesspharyngeal in (64) Although the forms of the participials [ftiacuteiħ] and [tfiacuteiħ]are analogous we can tell from the inflected forms [feacutetaħ] lsquohe openedrsquoversus [teacutefa] lsquohe threwrsquo that the stems must end in different consonantsThe most reasonable assumption is that the final consonant in the case of[teacutefa] is some pharyngeal other than [ħ] whichwould be [ʕ] Thus at least forverb stems ending in ʕ the underlying pharyngeal status of the conso-nant can be seen directly even though it is voiceless Since the abstractconsonant can be pinned down rather precisely in this context we reasonthat in all other contexts the abstract consonant must be ʕ as wellThe crucial difference between these examples of abstractness and cases

such as putative ɨ and ə in Hungarian or deriving [ɔj] from œ in Englishis that there is strong language-internal evidence for the abstract distinc-tion u vs o in Yawelmani or for the abstract consonant ʕ in Maltese

84 Grammar-external evidencefor abstractness

Yawelmani and Maltese provide well-motivated abstract analyses basedon patterns of alternation in the grammar We would still like to findgrammar-external evidence that abstract analyses can be psychologically

niacute-sma lsquoI hearrsquo ni-siacutema-w lsquowe hearrsquoniacute-zra lsquoI sowrsquo ni-ziacutera-w lsquowe sowrsquo

niacute-bla lsquoI swallowrsquo ni-biacutela-w lsquowe swallowrsquo

naacute-ʔla lsquoI earnrsquo na-ʔiacutela-w lsquowe earnrsquo

ʔaacutetel lsquohe killedrsquo ʔtiacuteil lsquokillingrsquo maʔtuacuteul lsquokilledrsquoħaacutetaf lsquohe grabbedrsquo ħtiacuteif lsquograbbingrsquo maħtuacuteuf lsquograbbedrsquofeacutetaħ lsquohe openedrsquo ftiacuteiħ lsquoopeningrsquo miftuacuteuħ lsquoopeningrsquoteacutefa lsquohe threwrsquo tfiacuteiħ lsquothrowingrsquo mitfuacuteuħ lsquothrownrsquobaacutela lsquohe swallowedrsquo bliacuteiħ lsquoswallowingrsquo mibluacuteuħ lsquoswallowedrsquoʔaacutela lsquohe earnedrsquo ʔliacuteiħ lsquoearningrsquo maʔluacuteuħ lsquoearnedrsquo

valid analogous to the historical arguments which Kiparsky adducedfrom the history of Yiddish and other languages in support of the moresurface-oriented approach to phonology

841 Abstract analysis and historical change TeraOne such argument for the psychological reality of abstract analysiscomes from Tera Newman 1968 provides a synchronic and diachronicargument for abstract phonology where similar surface forms have dif-ferent underlying forms

The synchronic argument Data in (65) illustrate a basic alternationSome nouns ending in [i] in their citation forms lack that vowel in phrasemedial contexts

Not all words ending in [i] prepausally engage in this alternation as thedata in (66) demonstrate

Given a vowel ~ Oslash alternation plus a set of stems which are invariantlyi-final in (66) we might be led to surmise that the stems in (65) are C-finaland take an epenthetic vowel [i] phrase-finally This can be ruled out given(67) where the stem ends in a consonant both phrase-medially andphrase-finally

A completely surface-oriented account where the underlying form mustbe one of the surface variants is untenable the nouns in (65) have avariant with the vowel [i] but selecting i for the underlying form failsto distinguish (65) from (66) which always have [i] and the nouns of (65)also have a variant with no final vowel but the nouns in (67) always lack afinal vowel

na seɗi lsquothis is a snakersquo na seɗ ɓa lsquothis is nota snakersquo

na deɓi lsquothis is gumrsquo na deɓ ɓa lsquothis is notgumrsquo

dala wa wuɗi lsquoDala pointedrsquodala wa wuɗ koro lsquoDala pointed at

the donkeyrsquodala wa mbuki lsquoDala threwrsquo

dala wa mbuk koro lsquoDala threw atthe donkeyrsquo

na wuɗi lsquothis is milkrsquo na wuɗi ɓa lsquothis is not milkrsquoa saɓi lsquothis is a stickrsquo na saɓi ɓa lsquothis is not a stickrsquo

na ruf lsquothis is a baboonrsquo na ruf ɓa lsquothis is not a baboonrsquotin zoɓ lsquoshe is a slobrsquo tin zoɓ ɓa lsquoshe is not a slobrsquona ɓoŋ lsquothis is whitersquo na ɓoŋ ɓa lsquothis is not whitersquo

Other roots of the variable-final type give evidence that the problematicstems in (65) underlyingly end in schwa The data in (68) provide mono-syllabic words which have the shape Ci prepausally and Cə phrasemedially

These words contrast with ones that have invariant [i] in both contexts

For the stems in (68) an obvious nonabstract solution is available thestems end with ə and there is a rule turning schwa into [i] prepausally

(70) ə i _

This applies in dala wa ɗi lsquoDala wentrsquo from dala wa ɗə but final schwa isunaffected in dala wa ɗə goma lsquoDala went to the marketrsquo The stems in (69)do not alternate since they end in the vowel i This solution is nonab-stract since the underlying form ɗə is one of the observed surfacevariantsThere are other stems with final [i] prepausally and [ə] phrase medially

These stems either have the shape [CVCCə] phrase-medially or else [CVZə]where Z is a voiced consonantThis gives the following groups of stems with an underlying final schwa

For most of these stems postulating underlying schwa is quite concretesince schwa actually surfaces in phrase-medial context However in poly-syllabic stems such as deɓi ~ deɓ with a single voiceless consonant before

dala wa ɮi lsquoDala receivedrsquodala wa ɮə sule lsquoDala received a shillingrsquodala wa ɗi lsquoDala wentrsquodala wa ɗə goma lsquoDala went to the marketrsquo

dala wa ɮi lsquoDala paidrsquodala wa ɮi sule lsquoDala paid a shillingrsquodala wa vi lsquoDala roastedrsquodala wa vi ɮu lsquoDala roasted meatrsquo

na pərsi lsquothis is a horsersquona pərsə ɓa lsquothis is not a horsersquodala wa kədi lsquoDala pulledrsquodala wa kədə koro lsquoDala pulled a donkeyrsquo

Stem shape Medial PrepausalCə Cə CiCVCCə CVCCə CVCCiCVZə CVZə CVZiCVCə CVC CVCi

final schwa the analysis is abstract because schwa is never phoneticallymanifested in the morpheme The decision that the vowel in question isschwa is based on analogy with a known behavior of schwa it becomes [i]prepausallyOur analysis requires a rule that deletes word-final phrase-medial

schwa providing the stem is polysyllabic and ends only in a single voice-less consonant

More evidence supports abstract schwa in certain words The examples in(74a) show that when a vowel -a marking definite nouns is suffixed to astem such as pərsə which ends in schwa schwa deletes whereas under-lying i is not deleted The data in (74b) show the same thing with theimperative suffix u

This motivates a rule of prevocalic schwa deletion which providesanother diagnostic that differentiates schwa from i

(75) ə Oslash _ V

Although lsquothrowrsquo only has the surface variants [mbuki] ~ [mbuk] itbehaves exactly like stems such as kədə where schwa is phoneticallyrealized and acts unlike vi in losing its final vowel before another vowelFinally there is an allomorphic variation in the form of the adjectivesuffix -kandi which shows up as -kandi when the stem ends in a vowel(saɓir taɗa-kandi lsquoheavy stickrsquo) and as -ndi when the stem ends in a conson-ant (saɓir teɓer-ndi lsquostraight stickrsquo) The stem of the word for lsquolongrsquo ends inabstract schwa since it alternates between final [i] (saɓira kəri lsquothe stick islongrsquo) and medial Oslash (saɓira kər ɓa lsquothe stick is not longrsquo) Furthermore thestem selects the postvocalic variant of the adjective suffix (saɓir kər-kandilsquolong stickrsquo) even though on the surface the stem ends with a consonantand not a vowel This anomaly is explained by the hypothesis that thestem does in fact end in a vowel namely schwa Thus multiple lines ofargument establish the presence of an abstract vowel schwa in a numberof words in the synchronic grammar of Tera

The diachronic argument A recent sound change in Tera provides agrammar-external test of the abstract hypothesis In one dialect of Teraspoken in the town of Zambuk a rule was added which palatalized t d and

ə Oslash V C _

[- voice]

a pərsi pərsə lsquohorsersquo pərs-a lsquothe horsersquowuɗi lsquomilkrsquo wuɗi-a lsquothe milkrsquo

b vi lsquoto roastrsquo vi-u lsquoroastrsquoɗi ɗə lsquoto gorsquo ɗ-u lsquogorsquokədi kədə lsquoto pullrsquo kəd-u lsquopullrsquombuki mbukə lsquoto throwrsquo mbuk-u lsquothrowrsquo

ɗ to tʃ dʒ and drsquoʒ before i The dialect of Tera spoken in Wuyo is represen-tative of the rest of Tera in retaining the original alveolars Thus we findWuyo da Zambuk da lsquoonersquo with no palatalization but Wuyo di Zambukdʒi lsquoto get uprsquo where d palatalizes There are synchronic alternationswhich further motivate this palatalization process in the contemporarygrammar of the Zambuk dialect so where the Wuyo dialect has xat-a lsquomybrotherrsquo xat-in lsquohis brotherrsquo the Zambuk dialect has xat-a xatʃ-in In Wuyoone findswuɗi lsquomilkrsquo and in Zambuk one findswudrsquoʒi deriving from wuɗi ndashthat the final vowel is i and not ə is shown by the phrase medial formwuɗiWhile palatalization is active in the Zambuk dialect it does not affect

all surface sequences of alveolar plus [i] in particular it does not affect [i]which derives from schwa In the Wuyo dialect lsquoto pullrsquo is kədi beforepause kədə medially (cf dala wa kədə koro lsquoDala pulled a donkeyrsquo) andtherefore we know that the stem is kədə In the Zambuk dialect themedial form is also kədə showing that the stem ends in schwa in thatdialect and the prepausal form is kədi Thus palatalization does not applyto the output of final schwa-fronting the failure of palatalization to applyto this derived [di] sequence provides another diagnostic of the distinctionbetween i and [i] derived from əFurther confirming our hypothesis about abstract schwa the stem

wuɗə lsquoto pointrsquo which appears in the Wuyo dialect as wuɗi prepausallyand as wuɗ medially (dala wa wuɗ koro lsquoDala pointed at a donkeyrsquo)appears as wuɗi in the Zambuk dialect without palatalization as isregularly the case with the vowel [i] derived from ə The fact that theinnovative sound change of palatalization found in the Zambuk dialectis sensitive to the sometimes abstract distinction between underlying iversus ones derived from schwas especially when the schwa neversurfaces supports the claim that abstract underlying forms can bepsychologically real

842 Abstract reanalysis in Matuumbi NC sequencesOther evidence for abstract phonology comes from a historical reanalysisof postnasal consonants in the Bantu language Matuumbi Nouns in Bantuare composed of a prefix plus stem and the prefix changes betweensingular and plural For example proto-Bantu mu-ntu lsquopersonrsquo containsthe class 1 prefix mu- marking certain singular nouns and the pluralba-ntu lsquopeoplersquo contains the class 2 prefix ba- Different nouns take differ-ent noun-class prefixes (following the tradition of historical linguisticsreconstructed forms are marked with an asterisk)

(76) Proto-Bantu sg Class Proto-Bantu pl Classmʊ-ntʊ 1 ba-ntʊ 2 lsquopersonrsquomʊ-gʊnda 3 mɪ-gʊnda 4 lsquofieldrsquoli-tako 5 ma-tako 6 lsquobuttockrsquom-paka 9 dim-paka 10 lsquocatrsquolʊ-badu 11 dim-badu 10 lsquoribrsquo

A postnasal voicing rule was added in the proto-Rufiji-Ruvuma subgroupof Bantu (a subgroup which includes Matuumbi) so that original mpakalsquocatrsquo came to be pronounced mbaka in this subgroup

Another inconsequential change is that the class 10 prefix originally din-lost di so the class 10 prefix became completely homophonous with theclass 9 prefixIn the Nkongo dialect of Matuumbi there was a change in the morpho-

logical system so that nouns which were originally assigned to classes9ndash10 now form their plurals in class 6 with the prefixma- Earlier ŋaambolsquosnake ~ snakesrsquo now has the forms ŋaacuteambo lsquosnakersquo ma-ŋaacuteambo lsquosnakesrsquoGiven surface [mbwa] lsquodogrsquo (proto-Bantu m-bʊa) originally in classes 9ndash

10 the concrete analysis is that the underlying form in proto-Rufiji ism-bwa It was always pronounced as [mbwa] since the root was alwayspreceded by a nasal prefix The absence of alternations in the phoneticrealization of the initial consonant would give reason to think that phon-etic [b] derives from underlying b By the same reasoning we predict thatearlier mpaka lsquocatrsquo is reanalyzed as b once the word came to be pro-nounced as mbaka in all contexts compare Yiddish gelt

The restructuring of the morphological system of Nkongo Matuumbiwhere the original class pairing 9ndash10 is reanalyzed as 9ndash6 allows us to testthis prediction since nouns with their singulars in class 9 no longer havea nasal final prefix in all forms the plural has the prefix ma- As thefollowing data show the concrete approach is wrong

While the distinction mp ~ mb was neutralized it was neutralized infavor of a phonetically more abstract consonant p rather than the con-crete consonant b

Proto-Bantu Matuumbimpaka mbaka lsquocatrsquoŋkaŋga ŋgaaŋga lsquoguinea fowlrsquontembo ndeembo lsquoelephantrsquomʊntʊ muundu lsquopersonrsquoŋkʊŋgʊnɪ ŋguuŋguni lsquobedbugrsquo

cf mbabada mbabala lsquobushbuckrsquombʊdi mbwi lsquogoatrsquombʊa mbwa lsquodogrsquo

Proto-Bantu Matuumbi sg Original pl Innovative plm-pembe m-beembe m-beembe ma-peembe lsquohornrsquoŋ-kʊkʊ ŋ-guku ŋ-gʊkʊ ma-kuku lsquochickenrsquom-bʊa m-bwa m-bwa ma-pwa lsquodogrsquom-babada m-babala m-babala ma-pabala lsquobushbuckrsquom-bʊdi m-bwi m-bwi ma-pwi lsquogoatrsquom-baŋgo m-baaŋgo m-baaŋgo ma-paaŋgo lsquowarthogrsquom-bʊtʊka m-bʊtʊka m-bʊtʊka ma-pʊtʊka lsquoantelopersquo

This reanalysis did not affect all nouns which had a singular or plural inclasses 9ndash10 it affected only nouns which originally had both theirsingulars and plurals in this class ie only those nouns lacking alterna-tion Nouns with a singular in class 11 and a plural in class 10 preserve theoriginal voicing of the consonant

A word such as lsquoribrsquo always had a morphological variant which transpar-ently revealed the underlying consonant so the contrast betweenn-toondwa [ndoondwa] and n-goi [ŋgoi] was made obvious by thesingulars [lu-toondwa] and [lu-goi]While it is totally expected that there should be a neutralization of mp

and mb in words likembakambwa ndash there would have been no evidence tosupport a distinction between surface [mb] deriving from mb versus [mb]deriving from mp ndash surprisingly from the viewpoint of concrete phon-ology the direction of neutralization where [mb] is reanalyzed as mp isunexpected One explanation for this surprising reanalysis regards thequestion of markedness of different consonants Given a choice betweenunderlying m + b and m + p where either choice would independentlyresult in [mb] one can make a phonetically conservative choice andassume m + b or make a choice which selects a less marked consonantie m + p In this case it is evident that the less marked choice is selectedwhere the choice of consonants is empirically arbitrarySuch examples illustrating phonetically concrete versus abstract rean-

alyses motivated by considerations such as markedness are not wellenough studied that we can explain why language change works oneway in some cases and another way in other cases In the case of Yiddishavek from historically prior aveg there would be no advantage at all inassuming underlying aveg from the perspective of markedness or phon-etic conservatism

843 Language games and Bedouin ArabicLanguage games can also provide evidence for the mental reality of under-lying representations Their relevance is that language game modifica-tions are not always performed on the surface form so by modifying thephonetic environment in which segments appear in the language gamesmay cause rules to apply when they would not normally (providingevidence for the reality of the phonological process) or prevent a rulefrom applying when it normally would (revealing the abstract underlyingform) An example of such evidence comes from Bedouin Arabic spoken inSaudi Arabia discussed by Al-Mozainy (1981) A number of verbs have theunderlying form CaCaC but this analysis is abstract in that for these

Proto-Bantu Matuumbi sg Matuumbi plm-badu lu-bau m-bau lsquoribrsquon-godi lu-goi ŋ-goi lsquoropersquon-dɪmi lu-lɪmi n-dɪmi lsquotonguersquoŋ-kʊŋgʊnɪ lu-kuuŋguni ŋ-guuŋguni lsquobedbugrsquon-tondʊa lu-toondwa n-doondwa lsquostarrsquo

verbs the first vowel sequence is never found on the surface and the rootsurfaces as [CiCaC]

8431 Regular language phonology We begin by motivating aspects ofthe phonology of the language especially underlying representationsusing regular language data Verb stems may have different underlyingvowels but the passive is formed by systematically replacing all under-lying vowels with i Underlying i deletes in an open syllable as shown bythe following data

Taking underlying ħizim and ħizim-t as examples the vowel i in the firstsyllable is in an open syllable so the rule of high-vowel deletion appliesgiving [ħzim] and [ħzimt] In the case of ħizim-at both vowels i are in anopen syllable the second i deletes which makes the first syllable closedso the first vowel does not delete resulting in [ħizmat] The following ruleis motivated by (80)

(81) i Oslash _ CV High-vowel deletion

Now we consider another class of nonpassive verbs where the underlyingstem shape is CaCiC In these stems the second vowel shows up as i whenthere is no vowel after the stem The first vowel of the stem alternatesbetween [i] and [a] surfacing as [i] when the second vowel appears as [i]otherwise surfacing as [a] Examples of verbs with this vocalic pattern areseen in (82)

In underlying samiʕ-at the vowel i is in an open syllable so it deletesgiving [samʕat] In samiʕ and samiʕ-t final i does not delete since it is notin an open syllable and a assimilates to [i] before [i] by the followingharmony rule

(83) a i _ C i

3sg masc 3sg fem 1sgħzim ħizm-at ħzim-t lsquobe tiedrsquoħfir ħifr-at ħfir-t lsquobe dugrsquoʃrib ʃirb-at ʃrib-t lsquobe drunkrsquoʕzim ʕizm-at ʕzim-t lsquobe invitedrsquolbis libs-at lbis-t lsquobe wornrsquo

3sg masc 3sg fem 1sgsimiʕ samʕ-at simiʕ-t lsquohearrsquolibis labs-at libis-t lsquowearrsquoʃirib ʃarb-at ʃirib-t lsquodrinkrsquojibis jabs-at jibis-t lsquobecome dryrsquosilim salm-at silim-t lsquosaversquoliʕib laʕb-at liʕib-t lsquoplayrsquoħilim ħalm-at ħilim-t lsquodreamrsquo

This creates a surface [i] in an open syllable which does not undergodeletionNow we turn to stems with the underlying shape CaCaC In a number

of such verbs this representation is uncontroversial since that is how itsurfaces

Examples such as [gʕadat] from gaʕad-at illustrate the application ofanother rule one deleting a when followed by CVCV

(85) a Oslash _ CVCV

An important fact about the stems in (84) is that the second consonant is aguttural (x γ ħ h ʕ or ʔ) There is a dissimilative process in the languageturning a into [i] in an open syllable if the next vowel is a providing thatthe vowel is neither preceded nor followed by a guttural consonant In theabove examples the consonant in the middle of the stem is a guttural soneither the first nor the second vowel can undergo the dissimilativeraising rule Now consider the data in (86) where the first consonant isa guttural but the second is not

This verbal restriction on the consonant next to the target vowel goesbeyond what is allowed in the version of the formal theory presentedhere How such conditions are to be incorporated into an analysis hasbeen the subject of debateHere the first vowel of the stem cannot become [i] because of the

preceding consonant but the second vowel does dissimilate to [i] whenfollowed by a and thus ʕazam-at becomes [ʕzimat] (with deletion of thefirst vowel by (85)) This rule is separate from the harmony rule that turnsa into [i] before [i] because harmony applies irrespective of the flankingconsonants cf [ħilim] lsquohe dreamtrsquo

(87) a i _ C a (target is not adjacent to a guttural consonant)

In [ʕazam] and [ʕazamt] there is no dissimilation because the first consonant isguttural which prevents the following a from undergoing dissimilation

3sg masc 3sg fem 1sggaʕad gʕad-at gaʕad-t lsquositrsquowaʕad wʕad-at waʕad-t lsquopromisersquotʕaʕan tʕʕan-at tʕaʕan-t lsquostabrsquosaħab sħab-at saħab-t lsquopullrsquotʕaħan tʕħan-at tʕaħan-t lsquogrindrsquodaxal dxal-at daxal-t lsquoenterrsquonaxal nxal-at naxal-t lsquosiftrsquo

3sg masc 3sg fem 1sgʕazam ʕzim-at ʕazam-t lsquoinvitersquoħazam ħzim-at ħazam-t lsquotiersquohakam hkim-at hakam-t lsquorulersquo

Examples in (88) show the same restriction on dissimilation of thesecond vowel a which does not become [i] when the last consonant is aguttural

Another consonantal property inhibiting dissimilation is a coronal sonor-ant In this case if the two vowels are separated by any of n r l there isno dissimilation In the examples of (89) the first vowel is prevented fromdissimilating because it is preceded by a guttural In addition the secondstem vowel is prevented from dissimilating because it is separated fromsuffixal a by a coronal sonorant Therefore both underlying stem vowelsremain unchanged

In the examples of (90) the first vowel is followed by a consonant otherthan a coronal sonorant and is neither preceded nor followed by aguttural so it dissimilates to [i] The second vowel is followed by a coronalsonorant so there is no dissimilation in the second syllable

In (91) we find verbs with a coronal sonorant as the second consonant Thesecond vowel a dissimilates before a since the intervening consonant isneither guttural nor a coronal sonorant The preceding coronal sonoranthas no effect on dissimilation since unlike the effect of gutturals coronalsonorants only have an effect if they stand after the target vowel

3sg masc 3sg fem 1sgdifaʕ dfaʕ-at difaʕ-t lsquopushrsquorʕikaʕ rʕkaʕ-at rʕikaʕ-t lsquobendrsquoxadaʕ xdaʕ-at xadaʕ-t lsquocheatrsquo

3sg masc 3sg fem 1sgħafar ħfar-at ħafar-t lsquodigrsquoħamal ħmal-at ħamal-t lsquocarryrsquoγasal γsal-at γasal-t lsquowashrsquo

3sg masc 3sg fem 1sgnizal nzal-at nizal-t lsquoget downrsquosikan skan-at sikan-t lsquooccupyrsquokisar ksar-at kisar-t lsquobreakrsquodifan dfan-at difan-t lsquoburyrsquonital ntal-at nital-t lsquostealrsquoʃitar ʃtar-at ʃitar-t lsquodividersquo

3sg masc 3sg fem 1sgdʒalas dʒlis-at dʒalas-t lsquositrsquogarʕasʕ grʕisʕ-at garʕasʕ-t lsquostingrsquogarʕatʕ grʕitʕ-at garʕatʕ-t lsquothrowrsquo

sarag srig-at sarag-t lsquostealrsquobalas blis-at balas-t lsquodenouncersquoʃanag ʃnig-at ʃanag-t lsquohangrsquodaras dris-at daras-t lsquostudyrsquo

Finally verbs with no gutturals or coronal sonorants are given in (92)

By the deletion rule (85) underlying katabat becomes ktabat whichbecomes [ktibat] by dissimilation In katab-t since the first vowel is notfollowed by CVCV it cannot elide and it dissimilates to [i] before [a] in thesecond syllableThe vowel a in the second syllable of verbs like [kitab] is only mildly

abstract since it does surface as [a] as long as the syllable is not open Theinitial a the syllable on the other hand is fully abstract since there is nocontext in this verb where the underlying a appears as such in theseverbs and instead the vowel only appears as [i] However we know thatthe initial vowel cannot be i since if it were that vowel would delete inan open syllable ndash contrast active [kitab] and [kitabt] from katab andkatab-t with the passives [ktib] and [ktibt] from kitib and kitib-tThe occurrence of initial nondeleting [i] in an open syllable is entirely

predictable It appears when neither the first nor second stem consonantis a guttural and when the second stem consonant is not a coronalsonorant This nondeleting [a] is thus in complementary distribution withsurface [a] (which nonabstractly derives from underlying a) which onlyappears when one of the first two consonants is a guttural or the secondconsonant is a coronal sonorantHence there is strong language-internal motivation for claiming that

the initial vowel of stems such as [kitab] is underlyingly a and is subjectto dissimilation to [i] or deletion

8432 Language game evidence There is a language game used byspeakers of Arabic which provides independent evidence for the mentalreality of these rules and underlying representations The rule for thelanguage game is very simple permute the order of consonants withinthe root Now let us consider the various phonetic results of permuta-tion on the verb forms ħazam lsquohe tiedrsquo and ħzim-at lsquoshe tiedrsquo In ħazamthe first vowel does not dissimilate because of the preceding gutturalin ħzimat the second stem vowel dissimilates because it is neitherpreceded nor followed by a gutural and it is not followed by a coronalsonorant

(93) lsquohe tiedrsquo lsquoshe tiedrsquoħamaz ħmizat ~zaħam zħamat ~zimaħ zmaħat

3sg masc 3sg fem 1sgkitab ktib-at kitab-t lsquowritersquomisak msik-at misak-t lsquocatchrsquosikat skit-at sikat-t lsquostop talkingrsquonitaf ntif-at nitaf-t lsquopluckrsquogisam gsim-at gisam-t lsquodividersquogiethab gethib-at giethab-t lsquocatchrsquonikas nkis-at nikas-t lsquoretainrsquo

In the permuted forms ħamaz and ħmizat where the second and thirdconsonants have exchanged place the vocalic pattern remains the samebecause the transposition has not crucially changed the consonantalenvironmentNow consider the forms zimaħ ~ zmaħat This pattern of transposition

has two effects on the vowel pattern First because the first consonant isnow not a guttural the dissimilation rule can apply in the first syllabledemonstrating the reality of the dissimilation rule Second because thefinal consonant is now a guttural the dissimilation rule cannot apply inthe second syllable demonstrating the reality of the blocking condition ondissimilation Finally in the case of zaħam ~ zħamat because the medialconsonant is a guttural neither vowel can dissimilateA crucial example in terms of testing the validity of the proposed

CaCaC underlying form for surface [CiCaC] stems is a stem such asdafaʕ lsquopushrsquo which surfaces as [difaʕ] Such a supposed underlying repre-sentation is abstract since the vowel of the first syllable always surfaces as[i] or Oslash cf difaʕ lsquohe pushedrsquo dfaʕat lsquoshe pushedrsquo never as a This stemcontains a final pharyngeal consonant and therefore movement of thatconsonant to first or second position will put the first vowel in contactwith a pharyngeal This should then block dissimilation and will directlyreveal the hypothesized underlying vowel to be [a]

The fact that this vowel actually surfaces as [a] under the circumstancespredicted by the abstract hypothesis gives strong support to the claim foran abstract representation of such stems as having the vowel patternCaCaC

85 How abstract is phonology

On the one hand we have argued for abstract analyses of MatuumbiYawelmani Maltese and other languages but we have argued againstabstract analyses of English The reason for this apparently inconsistentview of abstractness is that abstractness per se is not the issue the properquestion to be focusing on is what motivates an analysis Thus we con-clude that the formal theory of grammar imposes no constraints on therelation between underlying and surface forms though the theory doesstate what kinds of elements can exist in underlying representationsphonetically interpretable combinations of features ie segmentsThis does not mean that highly abstract underlying representations

can be gratuitously assumed Underlying representations require

lsquohe pushedrsquo lsquoshe pushedrsquofidaʕ fdaʕat ~daʕaf dʕafat ~ʕadaf ʕdifat ~ʕafad ʕfidat

motivation they must be acquired by children learning the languageand the best assumption to make is that in lieu of evidence to thecontrary underlying and surface forms are identical The question thatneeds further investigation is what constitutes valid ldquoevidence to thecontraryrdquo Phonological alternations in the shape of a morpheme pro-vide very powerful evidence for abstractness It remains an open ques-tion whether other considerations are also valid in constructing anunderlying formAlthough we have focused on the relation between underlying and

surface forms the larger question which this debate raises is what countsas valid evidence for testing a phonological theory It has provenextremely difficult to resolve questions about the psychological realityof theorized linguistic constructs Two approaches both valid have beentaken One is the ldquodomain-internalrdquo approach where formal constraintsare proposed to the effect that (for example) underlying forms should be asubpart of an actually pronounced word in the language or underlyingforms should only contain segments actually pronounced in the languageWe cannot show that these claims are literally ldquowrongrdquo what we can do isshow that such a position renders us incapable of capturing importantgeneralizations about the phonologies of Maltese and Yawelmani forexampleThe other approach the ldquodomain-externalrdquo approach seeks evidence

from outside the domain of synchronic phonological grammars them-selves in an attempt to find independent evidence that answers thequestion of what is actually in the mind of the speaker Any number ofsuch approaches can be imagined ndash neurosurgery psycholinguistictesting language games historical change the study of language acquisi-tion and so on Such evidence is extremely hard to find in the first placevirtually all relevant experimental work is conducted on a tiny handful ofcommonly spoken languages which typically do not have internally well-motivated abstractness Additionally the experimental methodologymust be critically evaluated which is usually very difficult to do outsideonersquos own discipline Finally the evidence must be interpreted against ageneral theory of for example child developmental psychology Thequestion of how to empirically validate theory-internal hypothesesremains very much an open question in phonology as it is in all scientificdomains

Exercises1 SlovakThe focus of this problem is the underlying representation of diphthongsDiscuss the underlying status of diphthongs in Slovak based on these dataNouns in Slovak come in three genders which determine what suffix if any isused in the nominative singular masculines have no suffix feminines have -aand neuters have -o

A There is a process of lengthening which takes place in certain morphologicalcontexts including the genitive plural and the diminutive

B There is also a shortening rule that applies in certain morphological contextsincluding the imperfective of verbs and the comparative of adjectives

Nom sg Gen pllipa lip lsquolinden treersquomuxa mux lsquoflyrsquolopata lopat lsquoshovelrsquosrna srn lsquodeerrsquoʒena ʒien lsquowomanrsquokazeta kaziet lsquoboxrsquohora huor lsquoforestrsquosirota siruot lsquoorphanrsquopaeligta piat lsquoheelrsquomaeligta miat lsquomintrsquokopito kopit lsquohoofrsquobruxo brux lsquobellyrsquoblato blat lsquomudrsquosalto salt lsquosomersaultrsquoembargo embarg lsquoembargorsquojablko jablk lsquoapplersquokoleso kolies lsquowheelrsquolono luon lsquolaprsquohovaeligdo hoviad lsquobeastrsquovlada vlad lsquogovernmentrsquobluza bluz lsquoblousersquodlato dlat lsquochiselrsquovino vin lsquovinersquotʃiara tʃiar lsquolinersquohniezdo hniezd lsquonestrsquo

Noun Diminutivehrad hradok lsquocastlersquolist listok lsquoleafrsquoxlp xlpok lsquohairrsquokvet kvietok lsquoflowerrsquohovaeligdo hoviadok lsquobeastrsquo

Perfective Imperfectiveodlisitj odlisovatj lsquoto distinguishrsquokupitj kupovatj lsquoto buyrsquoohlasitj ohlasovatj lsquoto announcersquopredlʒitj predlzovatj lsquoto extendrsquooblietatj obletovatj lsquoto fly aroundrsquouviazatj uvaeligzovatj lsquoto bindrsquo

Adjective Comparativebliski bliʃʃi lsquonearrsquouski uʃʃi lsquonarrowrsquokratki kratʃi lsquoshortrsquobieli belʃi lsquowhitersquorietki retʃi lsquorarersquo

C There is an alternation in the form of case suffixes which is governed byproperties of the stem which precedes

D The rule that explains the alternations in C also explains why a rulemotivated by the data in A seems not to have applied

E Some stems underlyingly end with consonant clusters and undergo aprocess of vowel epenthesis that eliminates certain kinds of consonantclusters

UrhoboAccount for the phonological alternations in the following data Tone can beignored The diacritic underneath a vowel indicates that the vowel is [+ATR](ldquoAdvanced Tongue Rootrdquo) and vowels without the diacritic are [-ATR]

Nom sg Gen sg Nom pl Dat pl Loc plmesto mesta mesta mestam mestax lsquotownrsquoblato blata blata blatam blatax lsquomudrsquohovaeligdo hovaeligda hovaeligda hovaeligdam hovaeligdax lsquotownrsquopismeno pismena pismena pismenam pismenax lsquoletterrsquozameno zamena zamena zamenam zamenax lsquopronounrsquodlato dlata dlata dlatam dlatax lsquotownrsquovino vina vina vinam vinax lsquowinersquohniezdo hniezda hniezda hniezdam hniezdax lsquonestrsquo

Nom sg Gen plzahrada zahrad lsquogardenrsquoniʒina niʒin lsquohollowrsquozatoka zatok lsquoinletrsquopismeno pismen lsquoletterrsquozameno zamen lsquopronounrsquolietʃivo lietʃiv lsquodrugrsquo

Nom sg Gen plikra ikier lsquoroersquo (cf also ikernati lsquoabounding in roersquo)ihla ihiel lsquoneedlersquodogma dogiem lsquodogmarsquososna sosien lsquopine treersquobedro bedier lsquohiprsquoradlo radiel lsquoplowrsquohradba hradieb lsquorampartrsquodoska dosiek lsquoboardrsquokridlo kridel lsquowingrsquotʃislo tʃisel lsquonumberrsquopasmo pasem lsquozonersquovlakno vlaken lsquofiberrsquoplatno platen lsquolinenrsquo

sı lsquopullrsquo esjo lsquoto pullrsquo uɾuhɾe lsquoropersquo sj uɾuhɾe lsquopull a ropersquofı lsquosprayrsquo efjo lsquoto sprayrsquo ewu lsquoclothesrsquo fj ewu lsquospray

clothesrsquoku lsquopourrsquo ekwo lsquoto pourrsquo eβɾı lsquooilrsquo kw eβɾı lsquopour oilrsquo

ru lsquodorsquo erwo lsquoto dorsquo ezeke lsquodedicationrsquo rw ezeke lsquodo adedicationrsquo

se lsquocallrsquo ese lsquoto callrsquo oʃaɾe lsquomanrsquo s oʃaɾe lsquocall a manrsquome lsquoplaitrsquo eme lsquoto plaitrsquo eco lsquohairrsquo m eco lsquoplait hairrsquoco lsquostealrsquo eco lsquoto stealrsquo ekpu lsquobagrsquo c ekpu lsquosteal a bagrsquoφe lsquourinatersquo eφe lsquoto urinatersquo ego lsquobottlersquo φ ego lsquofill a bottlersquoʃe lsquosellrsquo eʃe lsquoto sellrsquo eŋma lsquoclothesrsquo ʃ eŋma lsquosell clothesrsquohwe lsquolaughrsquo ehwe lsquoto laughrsquo omo lsquochildrsquo hw omo lsquolaugh at a

childrsquove lsquoexposersquo eve lsquoto exposersquo v omo lsquoexpose a

childrsquogbe lsquoclearrsquo egbe lsquoto clearrsquo aγwa lsquoforestrsquo gb aγwa lsquoclear a

forestrsquote lsquobe

worthlessrsquoete lsquoto be

worthlessrsquoko lsquoplantrsquo eko lsquoto plantrsquo ırıbo lsquopepperrsquo k ĩrıbo lsquoplant

pepperrsquoγo lsquoworshiprsquo eγo lsquoto worshiprsquo ını lsquoelephantrsquo γ ını lsquoworship

elephantrsquosa lsquoshootrsquo esa lsquoto shootrsquo ohwo lsquopersonrsquo s ohwo lsquoshoot a

personrsquohwa lsquopayrsquo ehwa lsquoto payrsquo hw ohwo lsquopay a

personrsquoγe lsquobe foolishrsquo eγe lsquoto be

foolishrsquoφe lsquobe wide eφe lsquoto be widersquoβje lsquobearrsquo eβje lsquoto bearrsquo omo lsquochildrsquo βj omo lsquobear a

childrsquore lsquoeatrsquo erjo lsquoto eatrsquo one lsquoyamrsquo rj one lsquoeat yamrsquose lsquorejectrsquo esȷo lsquoto rejectrsquo efe lsquowealthrsquo sȷ e fe lsquoreject

wealthrsquoco lsquotradersquo ecwo lsquoto tradersquo ere lsquomatrsquo cw ere lsquotrade a

matrsquoso lsquosingrsquo eswo lsquoto singrsquo une lsquosongrsquo sw une lsquosing a

songrsquo

ldquosprayrdquo refers to lavish gift-giving

mısıwe osıβe mısıɾı osıɾı mıʒısje oʒısje mısıɾo lsquopullrsquo

mıfıwe ofıβe mıfıɾı ofıɾı mıʒıfje oʒıfje mıfıɾo lsquosprayrsquo

mıkuwe okuβe mıkuɾu okuɾu mıʒıkwe oʒıkwe mıkuɾo lsquopourrsquo

mıruwe oruβe mıruɾu oruɾu mıʒırwe oʒırwe mıruɾo lsquodorsquo

mısewe oseβe mıseɾı oseɾı mıʒıse oʒıse mıseɾo lsquocallrsquo

mımewe omeβe mımeɾı omeɾı mıʒıme oʒıme mımeɾo lsquoplaitrsquo

mıcowe ocoβe mıcoɾı ocoɾı mıʒıco oʒıco mıcoɾo lsquostealrsquo

meφewe oφeβe meφeɾe oφeɾe meʒeφe oʒeφe meφeɾo lsquourinate onrsquo

meʃewe oʃeβe meʃeɾe oʃeɾe meʒeʃe oʒeʃe meʃeɾo lsquosellrsquo

mehwewe ohweβe mehweɾe ohweɾe meʒehwe oʒehwe mehweɾo lsquolaughrsquo

meve we ove βe meve ɾe ove ɾe meʒeve oʒeve meve ɾo lsquoexposersquo

megbe we ogbeβe megbe ɾe ogbe ɾe meʒegbe oʒegbe megbeɾo lsquoclearrsquo

lsquoI V (you)rsquo lsquoshe Vs

(me)rsquo

lsquoI V-edrsquo lsquoshe

V-edrsquo

lsquoI am still

V-ingrsquo

lsquoshe is still

V-ingrsquo

lsquoI have

V-ed himrsquo

Further readingChomsky and Halle 1968 Hudson 1974 Hyman 1970 Kiparsky 1968b Sapir 1933

mete ote mete ɾe ote ɾe meʒete oʒete lsquobe worthlessrsquo

meko we okoβe mekoɾe okoɾe meʒeko oʒeko mekoɾo lsquoplantrsquo

meγowe oγoβe meγoɾe oγoɾe meʒeγo oʒeγo meγoɾo lsquoworshiprsquo

mesa we osaβe mesaɾe osa ɾe meʒesa oʒesa mesaɾo lsquoshootrsquo

mehwa we ohwa βe mehwa ɾe ohwa ɾe meʒehwa oʒehwa mehwa ɾo lsquopayrsquo

mıγe oγe mıγeɾı oγe ɾı mıʒıγe oʒıγe lsquobe foolishrsquo

mıφe oφe mıφeɾı oφe ɾı mıʒıφe oʒıφe lsquobe widersquo

mıβjewe oβjeβe mıβjeɾı oβjeɾı mıʒıβje oʒıβje mıβjeɾo lsquobearrsquo

mere we ore βe mereɾe ore ɾe meʒerja oʒerja mereɾo lsquoeatrsquo

mese we ose βe mese ɾe ose ɾe meʒesȷa oʒesȷa mese ɾo lsquorejectrsquo

mecowe ocoβe mecoɾo ocoɾo meʒecwa oʒecwa mecoɾo lsquotradersquo

mesowe osoβe mesoɾo osoɾo meʒeswa oʒeswa mesoɾo lsquosingrsquo

(me)rsquo

V-edrsquo

lsquoI am still

V-ingrsquo

lsquoshe is still

V-ingrsquo

lsquoI have

V-ed himrsquo

CHAPTER

9 Nonlinearrepresentations

PREVIEW

This final chapter introduces an alternative model of how

sounds are represented the nonlinear theory The purpose

of this chapter is to show how troublesome facts can lead to

a reconceptualization of a domain which seemed to be

understood leading to an even better understanding of

the nature of language sounds This will also help you to

understand how and why theories change

KEY TERMSautosegmentalphonology

tone stability

floating tone

across-the-boardeffects

feature geometry

syllable

The theoretical model we have been assuming ndash known as the lineartheory of representation ndash was quite successful in explaining anumber of facts about sound systems An essential characteristic ofthe theory is that segments are matrices of feature values whereevery segment has a specification for each of the two dozen distinctivefeatures There was one phonological realm which the theoryhad largely ignored and that was tone and that had significantrepercussions

91 The autosegmental theory of tonethe beginnings of a change

There were a few proposals regarding tone features but they did not reachthe degree of acceptance that those for other features reached One of theprimary problems regarding tone was how to represent contour tonessuch as rising and falling

911 The problem of contoursOne possibility is that contour tones are simply H (high) or L (low) toneswith a positive specification for a feature ldquocontourrdquo We could take thepitch at the beginning of a vowel as representing the ldquobasicrdquo tone valueand if the pitch changes from that point (either up or down) then thevowel is [+contour] This gives us the following representations of H LR (rising) and F (falling) tones

(1)⎡⎢⎣

⎤⎥⎦

+H-contour

⎡⎢⎣

⎤⎥⎦

-H-contour

⎡⎢⎣

⎤⎥⎦

-H+contour

⎡⎢⎣

⎤⎥⎦

+H+contour

Such a theory is ultimately insufficient since it ignores tone levels (MidSuperlow Superhigh) but we can pursue this theory to see what progresscan be made Perhaps if this theory works it can be modified to accountfor other tone levelsAn essential test of a theory of features is how it accounts for

phonological processes This theory of tone makes predictions itpredicts that R and F will be a natural class because they are[+contour] and it predicts that L and R are a natural class becausethey are [ndashH] As it happens some relevant typological work had beendone on natural tone rules most notably Hyman and Schuh 1974Such research has shown that the following are fairly common tonalprocesses

(2) a H RfLFg_ b L FfHRg_c H F_fLRg_ d L R_fHFg

The problem is that the ldquo[contour]rdquo theory does not provide any naturalway to express all of these processes The last two processes can beformulated

(3) c [+H] [+contour]_[-H]

d [-H] [+contour]_[+H]

However the first two processes cannot be formalized since LF or HRare not a definable class using this theory L tone is ex hypothesii [-H]whereas F is [+H] so the class of progressive tone assimilations one of themost common tone rules is unformalizableThis theory also predicts the following rules which are simply the

rules in (3) with the conditioning environment on the left rather thanthe right

Unlike the common rules in (2) such rules are totally nonexistent in thelanguages of the world The ldquo[contour]rdquo theory thus makes a bad predic-tion that certain processes should exist when they do not and in additionthe theory provides no way to express certain very natural processes inparticular processes where the conditioning environment is on the leftFinally even for the two processes which the theory can formalize in (3)there is an unexplained element of arbitrariness ndash why should an H tonebecome a falling tone before [ndashH] Those processes are formally just assimple to express as the rules in (5) and should therefore be found ascommonly as the former set of rules but in fact this latter set of rules iscompletely unattested

It is obvious that this theory of tone is wrong but what is the alternativeThere was a long-standing intuition that contour tones were in some sensecomposite tones so that R is simply a combination of an L followed by anH and F is a combination of an H followed by an L falling and rising pitchis simply the continuous transition between the higher and lower pitchlevels that H and L define An example of the kind of phonologicalpatterns which were responsible for this intuition is the pattern of tonechanges that result from merging vowels between words in Yekheeillustrated below

[+H] [+contour] [-H] __ (H F LR __ )[+H] [+contour] [-H] __ (L R HF __ )

c [+H] [+contour]_[+H] (H F_HF)d [-H] [+contour]_[-H] (L R_LR)

iacutedzeacute eacutelagrave iacutedzeacutelagrave lsquothree axesrsquoegravekeacute eacutelagrave egravekeacutelagrave lsquothree ramsrsquouacutedzeacute ogravekpaacute uacutedzocirckpaacute lsquoone axersquoogravekeacute ogravekpaacute ogravekocirckpaacute lsquoone ramrsquo

oacutewagrave oacutewagrave oacutewǒwagrave lsquoevery housersquo

Nonlinear representations 287

The combination of H+L results in a falling tone and L+H results in arising tone How can the intuition that fall is H+L and rise is L+H beexpressed in the theoryThere is little problem in doing this for contour tones on long vowels

since long vowels can be represented as a sequence of identical vowels sotreating a long rising tone as being a sequence of tones is easy

(7)⎡⎢⎣

⎤⎥⎦

+ syllabic+ back- rd- H tone

⎡⎢⎣

⎤⎥⎦

+ syllabic+ back- rd+ H tone

= =a agraveaacute

The problem is short contour tones A single vowel cannot be both [ndashH tone]and [+H tone] and feature values cannot be ordered within a segment butthat is what is needed to represent short rising and falling tones

912 Autosegmental contoursA resolution of this problem was set forth in Goldsmith 1976 who pro-posed that tones be given an autonomous representation from the rest ofthe segment so that regular segments would be represented at one leveland tones would be at another level with the two levels of representationbeing synchronized via association lines This theory known as autoseg-mental phonology posited representations such as those in (8)

(8) agrave = H

a = L H

acirc = H L

The representation of [aacute] simply says that when the rest of the vocal tractis in the configuration for the vowel [a] the vocal folds should be vibratingat a high rate as befits an H tone The representation for [a] on the otherhand says that while the rest of the vocal tract is producing the shortvowel [a] the larynx should start vibrating slowly (produce an L tone) andthen change to a higher rate of vibration to match that specified for anH tone ndash this produces the smooth increase in pitch which we hear as arising tone The representation of [acirc] simply reverses the order of the tonalspecificationsThe view which autosegmental phonology takes of rules is different

from that taken in the classical segmental theory Rather than viewing theprocesses in (2) as being random changes in feature values autosegmentaltheory views these operations as being adjustments in the temporal rela-tions between the segmental tier and the tonal tier Thus the change in(2a) where H becomes rising after L and fall can be expressed as (9)

(9) (H) L H

(H) L H

By simply adding an association between the L tone element on the leftand the vowel which stands to the right we are able to express this tonal

change without changing the intrinsic feature content of the string wechange only the timing relation between tones and vowels This is notatedas in (10) where the dashed association line means ldquoinsert an associationlinerdquo

(10) L H

Two other notational conventions are needed to understand the formula-tion of autosegmental rules First the deletion of an association line isindicated by crossing out the line

Second an element (tone or vowel) which has no corresponding associ-ation on the other tier (vowel or tone) is indicated with the mark [ˊ]thus Vˊ indicates a toneless vowel and Hˊ indicates an H not linked to avowelOne striking advantage of the autosegmental model is that it allows

us to express this common tonal process in a very simple way Thetheory also allows each of the remaining processes in (2) to beexpressed equally simply ndash in fact essentially identically as involvingan expansion of the temporal domain of a tone either to the left or tothe right

(12) H L (=2b))

H L (=2c))

L H (=2d))

The problem of the natural classes formed by contour tones and leveltones was particularly vexing for the linear theory Most striking was thefact that what constitutes a natural class for contour tones depends on thelinear order of the target and conditioning tones If the conditioning tonesstand on the left then the natural classes observed are LF and HRand if the conditioning tones stand on the right then the natural group-ings are LR and HF In all other cases the groupings of elements intonatural classes are independent of whether the target is to the right or theleft of the trigger The autosegmental representation of contour tonesthus provides a very natural explanation of what is otherwise a quitebizarre quirk in the concept ldquonatural classrdquoThe autosegmental model also provides a principled explanation for the

nonexistence of rules such as (4) ie the rules H F LR _ and L R HF _ The change of H to F after L would involve not just an adjustmentin the temporal organization of an L-H sequence but would necessitatethe insertion of a separate L to the right of the H tone which would haveno connection with the preceding L the change of H to F after R is evenworse in that the change involves insertion of L when H is remotelypreceded by a L Thus the closest that one could come to formalizingsuch a rule in the autosegmental approach would be as in (13)

(13) L (H)

As we will discuss in this chapter autosegmental theory resulted in aconsiderable reconceptualization of phonological processes and the ideathat rules should be stated as insertions and deletions of associationrelationships made it impossible to express certain kinds of arbitraryactions such as that of (13)In addition to the fact that the theory provides a much-needed account

of contour tones quite a number of other arguments can be given for theautosegmental theory of tone The essential claim of the theory is thatthere is not a one-to-one relation between the number of tones in anutterance and the number of vowels a single tone can be associated withmultiple vowels or a single vowel can have multiple tones Moreover anoperation on one tier such as the deletion of a vowel does not entail acorresponding deletion on the other tier We will look at a number ofarguments for the autonomy of tones and the vowels which phoneticallybear them in the following sections

913 Tone preservationOne very common property exhibited by tones is stability where thedeletion of a vowel does not result in the deletion of the tone borne bythe vowel Very commonly the tone of a deleted vowel is transferred to theneighboring vowel often resulting in a contour tone We have seen anexample of this phenomenon in Yekhee where the combination of anL vowel plus H vowel results in a rising-toned vowel and H+L gives afalling-toned vowel

In the autosegmental theory deletion of a vowel does not directly affectthe tone which was associated with it and as a result after deletion of thevowel the tone simply remains on the tonal tier with no association withthe segmental tier ndash such an unassociated tone is referred to as a floatingtone

(15) L H

o akp o k

o o aw

One of the principles proposed in this theory is that all vowels must(eventually) bear some tone and all tones must be borne by somevowel ndash this condition is known as the Well-formedness ConditionAccordingly the unassociated tones which resulted from the deletion ofa vowel would then be associated with the following vowel resulting in afalling or rising tone

(16) LH L

ogravekeacute ogravekpaacute ogravekocirckpaacute lsquoone ramrsquo

The combination of two like-toned vowels as in the case of egravekeacute eacutelagrave egravekeacutelagravelsquothree ramsrsquo brings out another principle of the theory By the operationof vowel deletion and reassociation of the floating tone one would expectthe following representation

(17) H HL L

e k e l a

This would not be distinct from the simple tone melody LHL (17) says thatthe vowel e should be produced at high pitch at the beginning and at theend with no other pitches being produced The Twin Sister Conventionwas proposed as a constraint on the theory so that such a phoneticallyindistinguishable representation is formally disallowed

(18) Twin Sister ConventionAdjacent identical tones on one vowel are automatically simplified

Another illustration of the autosegmental treatment of tone preservationcomes from Mongo When vowels are brought together either directly inthe underlying representation or as the result of deleting certain conson-ants the vowel sequence is reduced to a single vowel which preserves allof the component tones of the two vowels This can result not just inthe simple contours R and F but also in the complex three-tone contoursfallndashrise (FR) and risendashfall (RF)

The derivation of the last example illustrates how the autosegmentaltheory explains the pattern elegantly In this case the first vowel deletescausing its two tones to become floating Those tones are associated withthe following vowel by the Well-formedness Conditions This results intwo adjacent H tones on one vowel which by the Twin Sister Conventionreduce to one H giving the phonetic output

(20) LH LH HL L LH LH HL L

c ccm m e mb e e mb e

LH LH HL L LH L

m m e mb e mb e e

H+H H begravetaacutembaacute beacutefeacute begravetaacutembeacutefeacute lsquotwo treesrsquoL+L L lagrave igravetoacutekograve ligravetoacutekograve lsquowith the forkrsquoH+L F mpugraveluacute igravenέ mpugravedʒwicircnέ lsquothese birdsrsquoL+H R lagrave bɔnagrave lɔnagrave lsquowith the babyrsquoH+F F soacutengoacuteloacute ɔtswὲ soacutengoacutelɔtswὲ lsquomay S enterrsquoH+R FR bagraveloacutengaacute bakaacuteeacute bagraveloacutenga kaacuteeacute lsquohis bloodrsquoL+F RF fagravekagravelagrave ɔtswagrave fagravekagravelɔ tswagrave lsquoF comes inrsquoL+R R bǎnkograve bǎmɔ bǎnkǎmɔ lsquothose othersrsquoR+F RF ɔmɔ ecircmbegrave ɔme mbegrave lsquomay someone else

singrsquo

The fact that the theory effortlessly handles three-tone contours whenthe linear theory struggled to handle even two-tone contours is clearevidence that autosegmental theory is the better theory

914 Across-the-board effectsAnother phenomenon which argues for the autosegmental representationof tone is across-the-board tone change An illustration of such a tonaleffect can be found in Shona The examples in (21) show that if a nounbegins with some number of H tones those Hrsquos become L when precededby one of the prefixes neacute- seacute- and cheacute

As shown in (22) and by the last example of (21) an H tone which isnot part of an initial string of Hrsquos will not undergo this loweringprocess

The problem is that if we look at a word such asmbuacutenduacutedziacute as having threeH tones then there is no way to apply the lowering rule to the word andget the right results Suppose we apply the following rule to a standardsegmental representation of this word

Beginning from neacute-mbuacutenduacutedziacute this rule would apply to the first H-tonedvowel giving neacute-mbugravenduacutendziacute However the rule could not apply again sincethe vowel of the second syllable is not immediately preceded by the prefixwhich triggers the rule And recall from examples such as neacute-mugraveruacutemeacute thatthe rule does not apply to noninitial H tonesThis problem has a simple solution in autosegmental theory where we

are not required to represent a string of n H-toned vowels as having nH tones Instead these words can have a single H tone which is associatedwith a number of vowels

(24) H L

benzibvunza mbundudzi

N with N like N of Nmbwaacute neacute-mbwagrave seacute-mbwagrave cheacute-mbwagrave lsquodogrsquohoacuteveacute neacute-hogravevegrave seacute-hogravevegrave cheacute-hogravevegrave lsquofishrsquombuacutenduacutedziacute neacute-mbugravendugravedzigrave seacute-mbugravendugravedzigrave cheacute-mbugravendugravedzigrave lsquoarmy wormrsquo

haacutekaacutetagrave neacute-hagravekagravetagrave seacute-hagravekagravetagrave cheacute-hagravekagravetagrave lsquobonesrsquobeacutenziacutebvugravenzaacute neacute-begravenzigravebvugravenzaacute seacute-begravenzigravebvugravenzaacute cheacute-begravenzigravebvugravenzaacute lsquofoolrsquo

N with N like N of Nmugraveruacutemeacute neacute-mugraveruacutemeacute seacute-mugraveruacutemeacute cheacute-mugraveruacutemeacute lsquomanrsquobagravedzaacute neacute-bagravedzaacute seacute-bagravedzaacute cheacute-bagravedzaacute lsquohoersquo

V [ndashH] se ne che _[+H] [+H]

Given these representations the tone-lowering process will only operateon a single tone the initial tone of the noun but this may be translatedinto an effect on a number of adjacent vowels

(25) L L

benzibvunza

mbundudzi

There is a complication in this rule which gives further support to theautosegmental account of this process Although this process lowers astring of H tones at the beginning of a noun when one of these prefixesprecedes a prefixed structure lowering does not affect every initialH tone When one prefix precedes another prefix which precedes a nounwith initial Hrsquos the second prefix has an L tone and the noun keeps itsH tones

However if there are three of these prefixes the second prefix has an L toneand lowering also affects the first (apparent) string of tones in the noun

A simple statement like ldquolower a sequence of adjacent Hrsquosrdquo after anH prefix would be wrong as these data show What we see here is analternating pattern which follows automatically from the rule that wehave posited and the autosegmental theory of representations Considerthe derivation of a form with two prefixes

(28) H H

se-che-mbunduzi

The lowering ofH on che gives that prefix an L tone and therefore that prefixcannot then cause lowering of the Hrsquos of the noun On the other hand ifthere are three such prefixes the first H-toned prefix causes the secondprefix to become L and that prevents prefix 2 from lowering prefix 3 Sinceprefix 3 keeps its H tone it therefore can cause lowering of H in the noun

(29) HH H

se-ne-che-mbundudzi

H LH H

se-ne-che-mbundudzi

Thus it is not simply a matter of lowering the tones of any number ofvowels Unlike the traditional segmental theory the autosegmental modelprovides a very simple and principled characterization of these patterns oftone lowering

N of N like of Nmbuacutenduacutedziacute cheacute-mbugravendugravedzigrave seacute-chegrave-mbuacutenduacutedziacute lsquoarmy wormrsquo

haacutekaacutetagrave cheacute-hagravekagravetagrave seacute-chegrave-haacutekaacutetagrave lsquobonesrsquo

seacute-negrave-cheacute-mbugravendugravedzigrave lsquolike with of army wormrsquo

seacute-negrave-cheacute-hagravekagravetagrave lsquolike with of bonesrsquo

915 Melodic patternsAnother phenomenon which supports the autonomy of tones and seg-ments is the phenomenon of melodic tonal restriction In some lan-guages there are restrictions on the possible tones of wordsirrespective of the number of vowels in the word Mende is an exampleof such a language Although this language has H L rising falling andrisendashfalling tones the distribution of those tones in words is quiterestricted Words can be analyzed as falling into one of five tone melod-ies illustrated in (30)

If tones were completely unrestricted then given five surface tones onewould predict twenty-five patterns for bisyllabic words and 125 patternsfor trisyllabic words Instead one finds five patterns no matter how manyvowels there are

(31) LHL L H L

nikili nyaha

This distribution can be explained if the restriction is simply stated at thelevel of the tonal representation the tone pattern must be one of H L LHHL or LHL As seen in (31) given an autosegmental representation of tonenigravekiacuteligrave nyagravehacirc and mba all have the same tonal representation

916 Floating tonesAnother tonal phenomenon which confounds the segmental approach totone but is handled quite easily with autosegmental representations isthe phenomenon of floating tones which are tones not linked to a vowel

Anlo tone The Anlo dialect of Ewe provides one example The data in(32) illustrate some general tone rules of Ewe Underlyingly the nounlsquobuffalorsquo is ētō with M tone on its two vowels However it surfaces as[egravetograve] with L tones either phrase-finally or when the following word has anL tone

These alternations are explained by two rules one rule lowers M (mid) toL at the end of a phrase and the second assimilates M to a following L

(33) ML_ ML_L

H haacutewaacutemaacute lsquowaistrsquo pέlέ lsquohousersquo kɔ lsquowarrsquoL kpagravekagraveligrave lsquothree-legged chairrsquo bὲlὲ lsquotrousersrsquo kpagrave lsquodebtrsquoHL feacutelagravemagrave lsquojunctionrsquo keacutenyagrave lsquounclersquo mbucirc lsquoowlrsquoLH ndagravevulaacute lsquoslingrsquo fagravendeacute lsquocottonrsquo mbǎ lsquoricersquoLHL nigravekiacuteligrave lsquogroundnutrsquo nyagravehacirc lsquowomanrsquo mba lsquocompanionrsquo

egravetograve lsquobuffalorsquo egravetograve megrave lsquoin a buffalorsquoētō φēφlē lsquobuffalo-buyingrsquo ētō djiacute lsquoon a buffalorsquoētō mēgbeacute lsquobehind a buffalorsquo

Thus in the citation form ētō first becomes ētograve then [egravetograve]Two other tone rules are exemplified by the data in (34)

Here we see a process which raises M to Superhigh tone (SH) when it issurrounded by H tones subsequently a nonfinal H tone assimilates to apreceding or following SH tone

(35) MSHH_H HSH SH_

We know from ētō mēgbeacute lsquobehind a buffalorsquo that mēgbeacute has the tones MHTherefore the underlying form of egravető megbeacute lsquobehind a mortarrsquo is egravetoacute mēgbeacuteThe underlying form is subject to the rule raising M to SH since the M issurrounded by H tones giving egravetoacute megbeacute This then undergoes the SHassimilation rule Another set of examples illustrating these tone pro-cesses is (36) where the noun agravetjiacutekē ends in the underlying sequenceHM When followed by mēgbeacute the sequence HMMH results so thiscannot undergo the M-raising rule However when followed by dyiacute theM-raising rule applies to kē giving an SH tone and the preceding syllablethen assimilates this SH

There are some apparently problematic nouns which seem to have a verydifferent surface pattern In the citation form the final M tone does notlower when followed by the MM-toned participle φēφlē the initial toneof the participle mysteriously changes to H the following L-toned post-position megrave inexplicably has a falling tone the postposition mēgbeacute mys-teriously has an initial SH tone

All of these mysteries are resolved once we recognize that this nounactually does not end with an M tone but rather ends with an H tonethat is not associated with a vowel thus the underlying form of the nounlsquomortarrsquo is (38)

Because this noun ends in a (floating) H tone and not an M tone the rulelowering prepausal M to L does not apply which explains why the finaltone does not lower The floating H at the end of the noun associates withthe next vowel if possible which explains the appearance of an H on the

egravetoacute lsquomountainrsquo egravetoacute djiacute lsquoon a mountainrsquoegravető megbeacute lsquobehind a mountainrsquo

agravetjiacutekegrave lsquorootrsquo agravetjiacutekē φēφlē lsquoroot-buyingrsquoagravetjiacutekē mēgbeacute lsquobehind a rootrsquo agravetjiacuteke djiacute lsquoon a rootrsquo

ētō lsquomortarrsquo ētō φēφlē lsquomortar-buyingrsquoētō mecirc lsquoin a mortarrsquo ētō djiacute lsquoon a mortarrsquoētō megbeacute lsquobehind a mortarrsquo

following postposition as a falling tone (when the postposition is mono-syllabic) or level H (when the next word is polysyllabic) Finally thefloating H serves as one of the triggering tones for the rule turningM into SH as seen in ētō megbeacute The hypothesis that this word (and otherswhich behave like it) ends in a floating H tone thus provides a unifiedexplanation for a range of facts that would otherwise be inexplicableHowever the postulation of such a thing as a ldquofloating tonerdquo is possibleonly assuming the autosegmental framework where tones and featuresare not necessarily in a one-to-one relation

Mixtec Another example of floating tones can be seen in the languageMixtec As (39) indicates some words such as kēē lsquowill eatrsquo have no effecton the tone of the following word but other words such as the apparentlyhomophonous verb meaning lsquowill go awayrsquo cause the initial tone tobecome H

A similar effect is seen in (40) where tagravekaacute lsquoallrsquo has no effect on thefollowing word but maacuteaacute lsquothatrsquo causes raising of the initial tone of thenext word

These data can be explained very easily if we assume the following under-lying representations

(41) MM MMH L H H

m a alsquothatrsquo

k e e k e e t a k alsquowill eatrsquo lsquowill go awayrsquo lsquoallrsquo

When a word ending in a floating H tone such as lsquowill go awayrsquo or lsquothatrsquois followed by another word that H associates to the first vowel of thenext word and replaces the initial lexical tone When there is no followingword the floating tone simply deletes

Gatilde Other evidence forfloating tones comes fromGatilde Some of the evidenceforfloating L tone in this language involves the phenomenon of ldquodownsteprdquowhich is the contrastive partial lowering of the pitch level of tones at aspecified position Downstep is exemplified in Gatilde with the words [kɔtɔkɔ]lsquoporcupinersquo [ogravenũf ũ] lsquosnakersquo and [aacutetaacutetuacute] lsquocloudrsquo In lsquoporcupinersquo the syllable

sugravetʃiacute lsquochildrsquo kēē lsquowill go awayrsquokōograve lsquosnakersquokēē lsquowill eatrsquokēē sugravetʃiacute lsquothe child will eatrsquo kēē suacutetʃiacute lsquothe child will go awayrsquokēē kōograve lsquothe snake will eatrsquo kēē koacuteograve lsquothe snake will go awayrsquo

tagravekaacute sugravetʃiacute lsquoall the childrenrsquo maacuteaacute suacutetʃiacute lsquothat childrsquotagravekaacute bēʔē lsquoall the housesrsquo maacuteaacute beacuteʔē lsquothat housersquotagravekaacute kōograve lsquoall the snakesrsquo maacuteaacute koacuteograve lsquothat snakersquotagravekaacute migravenī lsquoall the puddlesrsquo maacuteaacute miacutenī lsquothat puddlersquo

[tɔ] has H and the following syllable [kɔ] has L ndash the physical pitches aremaximally separate The second and third syllables of lsquosnakersquo are bothH and are not physically distinct ndash they are produced at the same pitchat the top of the voice range In the third example the syllable [taacute] hasthe same high pitch that all of the second syllables of these words haveand the following syllable which is phonologically H-toned has a pitchphysically between that of the L-toned syllable of [kɔtɔkɔ] and the H-toned syllable of [ogravenũ fũ ] What happens here is that the pitch range ofall tones is lowered after the second syllable of [aacutetaacutetuacute] even those of afollowing word This lowering of pitch range notated with ldquordquo is knownas ldquodownsteprdquo A floating L between H tones is what in fact generallycauses downstepIn Gatilde there is a rule changing the tone sequence HL before pause into

HH The operation of this rule can be seen in the data of (42) where thepresence of the future tense prefix -bagraveaacute- causes a change in the tone offinal L-toned verbs with the shape CV (the unmodified tone of the root isseen in the 3sg past form)

The necessity of restricting this rule to HL before pause is demonstratedby examples such as egravebagraveaacutegbegrave Agravekograve lsquohe will kill Akorsquo egravebagraveaacutekpὲ agravetagraveagravedeacutelsquohe will sew a shirtrsquo egravebagraveaacuteʃɔ kpagraveŋ lsquohe will pull a ropersquo In such examplesthe tone sequence is not prepausal and the underlying L is retained inphrase-medial position whereas the verb has H tone in prepausal pos-ition in (42)The restriction to applying just to prepausal HL also explains why verbs

with long vowels or two syllables do not undergo this alternation the L-toned syllable that comes after the H is not also at the end of the phrasesince another L tone follows it

A further restriction is that this rule does not apply to tense-inflections onverbs for example the plural imperative -agrave ( ɲɛ-heacute-agrave lsquobuy (pl)rsquo) or thehabitual -ɔ (egrave-matildedʒeacute-ɔ lsquohe sendsrsquo)

3sg past 3sg futureegrave-tʃagrave egrave-bagraveaacute-tʃaacute lsquodigrsquoegrave-dʒograve egrave-bagraveaacute-dʒoacute lsquodancersquoegrave-gbegrave egrave-bagraveaacute-gbeacute lsquokillrsquoegrave-kpὲ egrave-bagraveaacute-kpέ lsquosewrsquo

egrave-ʃɔ egrave-bagraveaacute-ʃɔ lsquopullrsquoegrave-tũ egrave-bagraveaacute-tũ lsquojumprsquoegrave-wograve egrave-bagraveaacute-woacute lsquowearrsquo

3sg past 3sg futureegrave-gbɔ egrave-bagraveaacute-gbɔɔ lsquohuntrsquoegrave-hagraveograve egrave-bagraveaacute-hagraveograve lsquoworryrsquoegrave-sɔɔ egrave-bagraveaacute-sɔɔ lsquocatchrsquoegrave-sɔlegrave egrave-bagraveaacute-sɔlegrave lsquoprayrsquoegrave-hagravelagrave egrave-bagraveaacute-hagravelagrave lsquochosersquo

A second relevant rule of Gatilde is Plateauing whereby HLH becomes HHHThis can be seen in (44) involving verbs with final HL If the followingword begins with L tone the final L of the verb is unchanged When thefollowing object begins with an H tone the resulting HLH sequencebecomes HHH by the Plateauing rule

This rule also applies within words when the verb stem has the underlyingtonepatternLHand isprecededbyanH-tonedprefix suchas the futureprefix

Again by the Plateauing rule egrave-bagraveaacute-hugraveluacute becomes [egrave-bagraveaacute-huacuteluacute]There are a number of areas in the language where floating tones can be

motivated The perfective tense provides one relevant example Considerthe data in (46) which contrasts the form of the subjunctive and theperfective Segmentally these tenses are identical their difference liesin their tone In both tenses the subject prefix has an H tone In theperfective the rule affecting prepausal HL exceptionally fails to apply toan L-toned CV stem but in the subjunctive that rule applies as expected

You might think that the perfective is an exception to the general ruleturning HL into HH but there is more to itAnother anomaly of the perfective is that the Plateauing rule fails to

apply between the verbs of (46) and the initial H tone of a following wordeven though the requisite tone sequence is found

ɲɛ -heacute-agrave lsquobuy (pl)rsquoɲɛ -heacute-

aacute tũ lsquobuy (pl) a gunrsquoɲɛ -heacute-agrave f ɔ lsquobuy (pl) oilrsquoegrave-matildedʒeacute-ɔ agravekograve lsquohe sends Akorsquoegrave-matildedʒeacuteɔ aacutekuacute lsquohe sends Akursquomĩ ŋgbegrave kwagravekwέ lsquoI am killing a mousersquomĩ ŋgbeacute foacutetegrave lsquoI am killing a termitersquo

In these examplesthe rule changingprepausal HLto HH does notapply to theverb in citationform because theL tone is in atense suffix

3sg past 3sg futureegrave-hugraveluacute egrave-bagraveaacute-huacuteluacute lsquojumprsquoegrave-kagraveseacute egrave-bagraveaacute-kaacuteseacute lsquolearnrsquoegrave-kogravedʒoacute egrave-bagraveaacute-koacutedʒoacute lsquojudgersquoegrave-matildedʒeacute egrave-bagraveaacute-matilde dʒeacute lsquosendrsquo

3sg subjunctive 3sg perfectiveeacute-tʃaacute eacute-tʃagrave lsquodigrsquoeacute-dʒoacute eacute-dʒograve lsquodancersquoeacute-gbeacute eacute-gbegrave lsquokillrsquoeacute-kpέ eacute-kpὲ lsquosewrsquo

eacute-ʃ ɔ eacute-ʃɔ lsquopullrsquoeacute-woacute eacute-wograve lsquowearrsquo

eacute-gbegrave aacutekuacute lsquohe has killed Akursquoeacute-ʃɔ guacute

gɔ lsquohe has pulled a nosersquoeacute-wograve dʒwέέ lsquohe has worn grassrsquo

The failure of both the HL HH rule and the Plateauing rule can beexplained by positing that the perfective tense is marked by a floatingL tone which comes between the subject prefix and the verb stem thusthe phonological representation of perfective eacute-wo would be (48) and wecan identify a L tone which has no assciated vowel as being the morphememarking the perfective

(48) H L L

e - wo

The floating L between the H and the L of the root means that the H is notnext to the prepausal L and therefore the rule changing HL intoHH cannot apply In addition the presence of this floating L explainswhy this verb form does not undergo Plateauing Thus two anomalies areexplained by the postulation of a floating L toneOther examples of the failure of the Plateauing rule in this tense can

be seen below The examples from the simple past show that these verbroots underlyingly have the tone pattern LH which surfaces unchangedafter the L-toned subject prefix used in the simple past The subjunctivedata show that these stems do otherwise undergo Plateauing after an H-toned prefix the perfective data show that in the perfective tense Pla-teauing fails to apply within the word because of the floating L of theperfective

Again these facts can be explained by positing a floating L tone in theperfective tense that L means that the actual tone sequence is HLLH notHLH so Plateauing would simply not be applicable to that tone sequence

(50) H

e hulu-

Finally the postulation of a floating L as the marker of the perfectiveexplains why a downstep spontaneously emerges between the subjectprefix and a stem-initial H tone in the perfective but not in thesubjunctive

3sg past 3sg subjunctive 3sg perfectiveegrave-hugraveluacute eacute-huacuteluacute eacute-hugraveluacute lsquojumprsquoegrave-kagraveseacute eacute-kaacuteseacute eacute-kagraveseacute lsquolearnrsquoegrave-kogravedʒoacute eacute-koacutedʒoacute eacute-kogravedʒoacute lsquojudgersquoegrave-matildedʒeacute eacute-matildedʒeacute eacute-matildedʒeacute lsquosendrsquo

3sg past 3sg subjunctive 3sg perfectiveegrave-beacute eacute-beacute eacute-beacute lsquoquarrelrsquoegrave-tʃũ eacute-tʃũ eacute-tʃũ lsquosendrsquoegrave-dũ eacute-dũ eacute-dũ lsquocultivatersquoegrave-foacute eacute-foacute eacute-foacute lsquoweeprsquoegrave-fɔteacute eacute-fɔ teacute eacute-fɔteacute lsquopourrsquoegrave-dʒaacuteleacute eacute-dʒaacuteleacute eacute-dʒaacuteleacute lsquorinsersquo

Thus the postulation of a floating tone as the marker of the perfectiveexplains a number of anomalies insofar as floating tones have a coherenttheoretical status in autosegmental phonology but not in the linear theorythey provide strong support for the correctness of the autosegmentalmodel

917 Tonal morphemesAnother example of the kind of dissynchrony between tones and vowelswhich is explained by the autosegmental model is the tonal morphemewhere a particular morpheme is expressed solely as a tone ndash this is avariant of the problem of floating tones One such example is the expres-sion of case marking and the marking of modified nouns in Angas Whena noun is case marked in Angas (when it is at the end of the subject orobject NP for example) case marking is indicated with a suffixed floatingH which links to the final vowel forming a rising tone if the final tone ofthe noun is M or L When a noun is followed by an adjective in its phrasethat fact is marked by the suffixation of a floating L tone which forms afalling contour tone when the last tone is M or H

Tiv is another language with morphemes being marked by tone in thiscase verbal tense-aspect Verb roots in Tiv lexically have either an H toneor an L tone on the first syllable of the root The general past tense ismarked with a floating L tone the past habitual with an H the recent pastwith the tone sequence HL

teacuteŋ lsquoropersquo teacuteŋ lsquorope (case)rsquo tecircŋ lsquorope (modified)rsquomuacutes lsquocatrsquo muacutes lsquocat (case)rsquo mucircs lsquocat (mod)rsquotʃeacuten lsquohoersquo tʃeacuten lsquohoe (case)rsquo tʃecircn lsquohoe (mod)rsquoɲiacute lsquoelephantrsquo ɲiacute lsquoelephant (case)rsquo ɲicirc lsquoelephant (mod)rsquoʔās lsquodogrsquo ʔas lsquodog (case)rsquo ʔas lsquodog (mod)rsquoʒwāl lsquoboyrsquo ʒwa l lsquoboy (case)rsquo ʒwa l lsquoboy (mod)rsquoɟēm lsquochildrsquo ɟem lsquochild (case)rsquo ɟe m lsquochild (mod)rsquomagraves lsquolocust beanrsquo mǎs lsquobean (case)rsquo magraves lsquobean (mod)rsquopugravek lsquosouprsquo pǔk lsquosoup (case)rsquo pugravek lsquosoup (mod)rsquoʔagraves lsquotooth ʔǎs lsquotooth (case)rsquo ʔagraves lsquotooth (mod)rsquodʒoacuteligrave lsquoapersquo dʒoacutelǐ lsquoape (case)rsquo dʒoacuteligrave lsquoape (mod)rsquo

H verbs L verbsGeneral past (L)vaacute lsquocomersquo dzagrave lsquogorsquouacutengwagrave lsquohearrsquo vegravendegrave lsquorefusersquojeacutevegravesegrave lsquofleersquo ngogravehograverograve lsquoacceptrsquo

Past habitual (H)vaacute dzaacuteuacutengwaacute vegravendeacutejeacuteveacuteseacute ngogravehoacuteroacute

Recent past (HL)vaacute dzaacuteuacutengwaacute vegravendeacutejeacuteveacutesegrave ngogravehoacuterograve

In addition to showing the effects of various floating tone morphemeswhich mark tense-aspect these data illustrate the application of acontour-simplification rule We now consider how representative formsare derived The concatenation of the L root ngohoro and the recent pastmorpheme gives the following underlying form

(54) L

ngohor

These tones must be assigned to the vowels of the stem we can see thatthe first tone links to the first free vowel and the second tone links to thesecond free vowel This is an instance of one-to-one left-to-rightmapping

(55) Link free tones to free vowels one-to-one from left to right

This process is so common that it had been thought that it is actually auniversal convention on free tones ndash we now know since languages havebeen discovered which do not obey this condition ndash that it is a language-specific rule though a very common one Application of this rule to (54)gives the surface formNow consider the disyllabic L root vegravendeacute This root has two vowels but

three tones If all of the tones were to be associated with the vowels of theroot this would force the final syllable to bear the tone sequence HL ie itwould have a falling toneWe can see that there are no contour tones in thedata This leaves us with two possibilities in accounting for vegravendeacute either therule associating floating tones with vowels simply does not link a floatingtonewith a vowel that already has a tone orfloating tones do associatewithvowels that already bear an H and then some later rule eliminates tonalcontour tones If we assume that floating tones are all initially associatedwith a vowel and contours are later eliminated we will require thefollowing rule which deletes the L-tone component of a falling tone

(56) H L

Oslash

Finally we come to dzagrave which has H if one of the floating tone patternsH or HL is added to the root This can be explained if floating tones areassociated with root vowels even when this would result in a contourtone Linking the melodic tones to this root would result in the followingrepresentation

(57) H LL

Rule (56) applies in a mirror-image fashion it deletes L in combinationwith an H on one vowel standing before or after the H This explains whythe lexical L is replaced with an H Under the alternative account thatfloating tones only link to vowels which do not have any other tone we

would be unable to explain why the lexical L is replaced by H when amelodic pattern with an H tone is added

918 Toneless vowelsAnother phenomenondemonstrating the independence of tones andvowelsis the existence of underlyingly toneless vowels This can be illustratedwithdata from Margyi There are two tones in Margyi H and L but there arethree underlying types of vowels in terms of tonal behavior namely H Land toneless Examples of underlyingly toneless morphemes are ɗəl lsquobuyrsquoskə lsquowaitrsquo and na lsquoawayrsquo When two morphemes with underlying tonesare combined there are no surface tone changes However when one of thetoneless morphemes is combined with a morpheme with tone the tonelessmorpheme takes on the tone of the tone-bearing morpheme

As (59) indicates this can be accounted for by spreading tone (ie addingassociations between tone and vowels) to toneless vowels

(59) H H

ta + na

ndal + naba+el

The form ɗ əl-nagrave lsquoto sellrsquo which combines two toneless morphemes illus-trates another property of tone systems Since all vowels must on thesurface have some tonal specification the following question arises ifthere is no tone present in the string which could spread to tonelessvowels how do toneless vowels get their surface tone The answer is thatthere are also rules of default tone assignment which guarantee that if avowel does not otherwise have a tone value one is automatically assignedSuch a rule can be formalized as (60)

Generally in languages with two levels of tone the default value assignedto otherwise toneless vowels is L in languages with three tone levelsthe default tone specification is usually M tone Yoruba is a languagewith three tone levels where it can be argued that M-toned vowels areactually underlyingly toneless and M tones are assigned by a defaulttone-assignment rule The examples in (61) illustrate a very general

taacute + baacute taacutebaacute lsquoto cook allrsquondagravel + baacute ndagravelbaacute lsquoto throw outrsquoɗəl + baacute ɗ əlbaacute lsquoto buyrsquonaacute + ɗagrave naacuteɗagrave lsquogive mersquohərigrave + ɗagrave h ərɗagrave lsquobring mersquoskə + ɗagrave sk əɗagrave lsquowait for mersquotaacute + na taacutenaacute lsquoto cook and put asidersquondagravel + na ndagravelnagrave lsquoto throw awayrsquoɗəl + na ɗ əlnagrave lsquoto sellrsquo

tone-spreading rule whereby L tone becomes falling after H and H tonebecomes rising after L However M is unchanged after either L or H andM also has no effect on a following L or H

The question is how to exclude M tone from being targeted by this ruleand how to prevent M tone from spreading If we assume that tonallyunspecified vowels are assigned an M tone by default and that M tones inYoruba derive only from application of this default specification rule thenwe can explain these patterns rather simply We can assume the followingtone-spreading rule where T represents any tone

(62) T T

The fact that contours are not formed with M tone follows from the factthat a contour is two tone specifications on one vowel plus the hypothesisthat M tone is only assigned if there is no tonal specification on a vowel

919 Tonal mobilityThe final demonstration of the autonomy of tone from segments is thetone mobility which is the fact that tones can move about from vowel tovowel quite easily in a fashion not shared with segmental properties Oneexample of tonal mobility comes from Nkore seen in (63) This languagehas an underlying contrast between words whose last syllable is H tonedand those whose penultimate syllable is H toned In prepausal positionunderlyingly final H tones shift to the penultimate syllable thus neutral-izing with nouns having an underlyingly penult H When some wordfollows the noun the underlying position of the H tone is clearly revealed

kograve pɔ lsquoit is not plentifulrsquo kograve dũ lsquoit is not sweetrsquooacute pɔ lsquoit is plentifulrsquo oacute dũ lsquoit is sweetrsquoὲkɔ lsquolessonrsquo ɔbɔ lsquomonkeyrsquoɔf ɔ lsquomourningrsquo giacutegā lsquoheightrsquoi ʃeacute lsquowork ēdʒograve lsquosnakersquo

Nouns with penult Hogravekugraveguacuterugrave lsquolegrsquo ogravekugraveguacuterugrave kugraverugraveugravenjigrave lsquogood legrsquoogravemugravekoacutezigrave lsquoworkerrsquo ogravemugravekoacutezigrave mugraverugraveugravenjigrave lsquogood workerrsquoegravembuacutezigrave lsquogoatrsquo egravembuacutezigrave nugraveugravenjigrave lsquogood goatrsquoegravechigravekoacutepograve lsquocuprsquo egravechigravekoacutepograve chigraverugraveugravenjigrave lsquogood cuprsquoegravembiacutebograve lsquoseedsrsquo egravembiacutebograve nugraveugravenjigrave lsquogood seedsrsquo

Nouns with final Hogravemugraveguacutezigrave lsquobuyerrsquo ogravemugravegugraveziacute mugraverugraveugravenjigrave lsquogood buyerrsquoogravemugravekaacutemagrave lsquochief ogravemugravekagravemaacute mugraverugraveugravenjigrave lsquogood chiefegraveeacutembwagrave lsquodogrsquo egraveegravembwaacute nugraveugravenjigrave lsquogood dogrsquoogravebuacuterograve lsquomilletrsquo ogravebugraveroacute bugraverugraveugravenjigrave lsquogood milletrsquokagravesuacutekugrave lsquoparrotrsquo kagravesugravekuacute nugraveugravenjigrave lsquogood parrotrsquo

There are a number of reasons internal to the grammar of Nkore fortreating L tone as the default tone and for only specifying H tones inthe phonology so that phonetically L-toned vowels are actually tonelessThis alternation can be accounted for by the following rule of tone-throwback

(64)xH

Another example of tone shift can be seen in Kikuyu Like Nkore there aregood reasons to analyze this language phonologically solely in terms ofthe position of H tones with vowels not otherwise specified as H beingrealized phonetically with a default L tone We will follow the conventionadopted in such cases as marking H-toned vowels with an acute accentand not marking toneless (default L) vowelsConsider the Kikuyu data in (65) illustrating the current habitual tense

The first two examples in (65a) would indicate that the morphemes to--rɔr- -aγ- and -a are all toneless The third example however shows theroot rɔr with an H tone this happens only when the root is preceded bythe object prefix ma In (65b) we see that ndash in contrast to what we see in(65a) ndash the habitual suffix -aγ- has an H tone when it is preceded by theroot tom (which is itself toneless on the surface) As with (65a) the syllablethat follows ma has an H tone

(65) a

It is clear then that certain syllables have the property of causing thefollowing syllable to have a surface H tone This is further demonstratedin (66) where the derivational suffixes -er- and -an- follow the roots -rɔr-and -tom- we can see that the syllable after -tom always receives anH tone

Further examples of this phenomenon are seen in the examples of therecent past in (67) In (67a) the root rɔr (which generally has no H tone)

to -rɔr-aγ -a lsquowe look atrsquowe-look at-hab-tenseto -mo -rɔr -aγ -a lsquowe look at himrsquo

we-him-look at-hab-tenseto -ma -rɔr -aγ -a lsquowe look at themrsquo

we-them-look at-hab-tense

to-tom-aacuteγ-a lsquowe sendrsquoto-mo-tom-aacuteγ-a lsquowe send himrsquo

to-ma-toacutem-aacuteγ-a lsquowe send themrsquo

to-rɔr-er-aγ-a lsquowe look forrsquoto-tom-eacuter-aγ-a lsquowe send forrsquoto-rɔr-an-aγ-a lsquowe look at each otherrsquoto-tom-aacuten-aγ-a lsquowe send each otherrsquoto-rɔr-er-an-aγ-a lsquowe look for each otherrsquoto-tom-eacuter-an-aγ-a lsquowe send for each otherrsquo

has an H tone when it stands immediately after the recent-past-tense prefix -a- or the object prefix that follows -a- will have a surfaceH tone The examples in (67b) show the same thing with the root -tom-which we have seen has the property of assigning an H tone to thefollowing vowel

(67) a

We would assume that the root -toacutem- has an H as do the object prefix -maacute-and the tense prefix -a- and this H tone is subject to the following rule oftone shift which moves every H tone one vowel to the right

(68) Hx

Thus to-toacutem-er-aγ-a becomes totomeacuteraγa to-maacute-rɔr-aγ-a becomes tomarɔ-raγa and to-aacute-maacute-toacutem-a becomes toamaacutetoacutemaacute

(69) H H

t o a m a t o m ax x x

An even more dramatic example of tone shifting comes from Digo In thislanguage the last H tone of a word shifts to the end of the word The rootvugura is toneless as is the object prefix ni but the object prefix a lsquothemrsquo

has an underlying H tone which is phonetically realized on the last vowelof the word Similarly the root togora is toneless as is the subject prefix nibut the third-singular subject prefix a has an H tone which shifts to theend of the word Lastly the root tsukura is toneless as is the tense-aspectprefix -na- but the perfective prefix ka has an H tone which shifts to thelast vowel of the word

(70) a

These data can be accounted for by a rule of tone shift which is essentiallythe same as the Kikuyu rule differing only in that the tone shifts all theway to the end of the word

(71) Hx

to-a-rɔr-a lsquowe looked atrsquoto-a-moacute-rɔr-a lsquowe looked at himrsquo

to-a-maacute-rɔ r-a lsquowe looked at themrsquo

to-a-toacutem-aacute lsquowe sentrsquoto-a-moacute-tom-aacute lsquowe sent himrsquo

to-a-maacute-toacutem-aacute lsquowe sent themrsquo

ku-vugura lsquoto untiersquo ku-vugurira lsquoto untie forrsquoku-ni-vugurira lsquoto untie for mersquo ku-a-vuguriraacute lsquoto untie for themrsquo

ku-togora lsquoto praisersquo ni-na-togora lsquoIrsquom praisingrsquoa-na-togoraacute lsquohersquos praisingrsquo

ku-tsukura lsquoto carryrsquo ni-na-tsukura lsquoIrsquom carryingrsquoa-na-tsukuraacute lsquohersquos carryingrsquo ni-ka-tsukuraacute lsquoI have carriedrsquo

92 Extension to the segmental domain

The foregoing modification of phonological theory had the obviousgood consequence that tonal phenomena could be accounted for verynicely whereas previously tone was largely outside the grasp of thetheory The impact of autosegmental phonology was much more pro-found than that however The obvious thing to wonder is if tonesare separate from the rest of the segment then perhaps segmentsthemselves are not such monolithic unstructured entities And soinvestigators looked for evidence for a similar separation of segmentalfeatures

921 The autonomy of all featuresAn example of segmental phenomena which are reminiscent of autoseg-mental tonal properties is floating segmental features as morphemesOne such case is seen in Vata where the past-tense marker can be arguedto be simply the specification [+high] which is suffixed to the stem and isrealized phonetically on the last vowel

A second example comes from Fula where a particular agreement pattern(ldquopattern Brdquo below) is marked by a prefix composed of the segmentalspecification [ndash continuant] which causes an initial continuant to becomea stop

Aramaic CP Azerbaijani Aramaic provides evidence for treating thefeature [constricted pharynx] ([CP]) autosegmentally This dialect has acontrast between pharyngealized or emphatic vowels (A E I U O) specifiedas [+CP] and plain vowels (a e i u o) In most words either all of the vowelsare emphatic or none of them is

n le lsquoI eatrsquo n li lsquoI atersquon ple lsquoI passrsquo n plɪ lsquoI passedrsquon mlε lsquoI gorsquo n mlɪ lsquoI wentrsquon no lsquoI hearrsquo n nu lsquoI heardrsquon zɔ lsquoI placersquo n zʊ lsquoI placedrsquon wɔlɔ lsquoI washrsquo n wɔlʊ lsquoI washedrsquo

Pattern A Pattern Bwecco becce lsquoribrsquowibdʒo bibdʒe lsquowingrsquoruulde duule lsquocloudrsquosekko cekke lsquomatrsquohello kelle lsquoslaprsquojeɓre dʒeɓel lsquoseedrsquojimre dʒimel lsquopoemrsquo

jontere dʒonte lsquoweek

CP has beenproposed as afeature used todescribepharyngealization

Some words may have nonemphatic vowels followed by emphatic vowelsIn such a case the first emphatic vowel is always a low vowel

These distributional properties will play an important role in arguing foran autosegmental treatment of [CP]In line with the fact that all vowels in a word generally agree in the

feature [CP] (76) shows that suffixes harmonize in [CP] with the precedingvowel

[CP] will spread through a whole sequence of suffixes

We will assume that the only value underlyingly marked for this featureis [+CP] and that [+CP] spreads to the right by the following rule

(78) [+CP]

This rule thus explains why [+CP] vowels are always followed by [+CP]vowels However we also need to explain why roots with a [+CP] specifi-cation (generally) have [+CP] beginning with the first vowel We canassume that in the general case the specification [+CP] is not associatedwith any particular vowel but is just floating and an unassociated [+CP]specification is associated with the first vowel of the word by the followingrule

AmrA lsquowoolrsquo brata lsquodaughterrsquozArʔA lsquoseedrsquo bela lsquohousersquoqUlOx lsquostand uprsquo nŭdʒum lsquosorceryrsquo

ʃarAw lsquocorn growing wildrsquo riswAj lsquounmannerly speechrsquosejfullAh lsquoa great dealrsquo fandbAz lsquotricksterrsquoniʃAn lsquosignrsquo peʃtAmAl lsquotowelrsquomilAqE lsquohung grapesrsquo elijAhU lsquonamersquogalimbAdʒI lsquobrotherrsquos wifersquo silAhlAmIʃ lsquosupplied with weaponsrsquo

lixma lsquobreadrsquo lixm-e plpirtʃaxwar-a lsquoold womanrsquo pirtʃaxwar-e plnOhr-A lsquomirrorrsquo nOhr-E pldIqnAxwAr-A lsquoold manrsquo dIqnAxwAr-E plklu lsquowrite (sg)rsquo klu-mun plbilbul lsquoseekrsquo bilbul-un plqU lsquorisersquo qU-mUn plmIʃltUn lsquomake a kingrsquo mIʃltUn-Un pl

mĭr-a lsquoshe saidrsquo xIt-lAx lsquoyou (fem sg) sewedrsquomir-wa-la lsquoshe had saidrsquo xIt-wA-lAx lsquoyou had sewnrsquomir-wa-la-la lsquoshe had said itrsquo xIt-wA-lAx-U lsquoyou had sewn themrsquo

(79) [+CP]

The derivation of mIʃItUn-Un lsquomake a king (pl)rsquo shows these rules

(80) [+CP] [+CP] [+CP]

mifitun-unmifitun-unmifitun-un

(rule 79) (rule 78)

There are some suffixes whose vowels are invariably emphatic that vowelis always the vowel [A] No suffixes are invariably plain

These suffixes will be assumed to have underlying [CP] specificationsin contrast to most other suffixes which are unspecified for [CP] Since thesuffix vowel is lexically associated with [+CP] it does not associate with thefirst vowel of the word and since it does not associate with the first vowel ofthe word [+CP] does not spread to any vowels before that of the suffixWe also find spreading of [+CP] between members of a compound In

the examples of (82) [+CP] spreads from the first compound to the second

This is the expected pattern [+CP] spreads rightward from the firstmember of the compound to the secondIf the second member of the compound has [+CP] vowels [+CP] spreads

through the second member of the compound

This apparent exceptional leftward spreading of [+CP] is nothing of thesort Rather the second member of the compound has a floating [+CP]

qalăma lsquopenrsquo qalam-dAn lsquocase for scribersquos utensilsrsquoqand lsquosugarrsquo qand-dAn lsquosugarbowlrsquoʃakăr lsquosugarrsquo ʃakăr-dAn lsquosugarbowlrsquodukana lsquostorersquo dukan-dAr lsquoshopkeeperrsquomewana lsquoguestrsquo mewan-dAr lsquohospitablersquodʒut lsquoplowrsquo dʒut-kAr lsquoplowerrsquonŭdʒum lsquosorceryrsquo nŭdʒum-kAr lsquosorcererrsquonaqʃ lsquoengravingrsquo naqʃ-kAr lsquoengraverrsquo

tAhA lsquo3rsquo imme lsquo100rsquotAhA-mmE lsquo300rsquodIqnA lsquobeardrsquo xwara lsquowhitersquodIqnA-xwArA lsquoold manrsquo

xwara lsquowhitersquo dIqnA lsquobeardrsquoxwArA-dIqnA lsquoold manrsquobe lsquowithoutrsquo hAd lsquolimitrsquobEhAd lsquoexceedinglyrsquoqahwa lsquocoffeersquo xAnA lsquoshelterrsquoqAhwA-xAnA lsquocoffee-roomrsquo

specification in a compound that feature links to the first vowel of theword by rule (79) and then spreads to the right

(84) [+CP] [+CP] [+CP]

xwara diqna xwara diqna xwara diqna

Another case of [+CP] appearing to the left of the morpheme where itoriginates is seen in (85) where a prefix is added to a root with a floating[+CP] specification

Given the assumption that a root specification of [+CP] is not generallyassociated in the underlying form (except in roots such as (75) where [+CP]is unpredictably associatedwith a noninitial low vowel) our analysis predictsthat the [+CP] specification will link to the first vowel of the word which willbe the prefix vowel in this case and spreads to the right thereafterThe locational suffix -istan has the interesting property that it causes all

vowels in the word to which it is attached to become [+CP]

This makes sense if the suffix -istan also has a floating specification [+CP]which automatically associates with the first vowel of the stem and thenspreads rightward

(87) [+CP] [+CP] [+CP]

xarab - istan xarab - istan xarab - istan

922 Feature geometryIt was realized that all features are autonomous from all other featuresand exhibit the kind of behavior which motivated the autosegmentaltreatment of tone The question then arises as to exactly how featuresare arranged and what they associate with if the ldquosegmentrdquo has had all ofits features removed The generally accepted theory of how features relateto each other is expressed in terms of a feature-tree such as (88) Thistree ndash known as a feature geometry ndash expresses the idea that while allfeatures express a degree of autonomy certain subsets of the features

xoʃ lsquogoodrsquo na-xoʃ lsquoillrsquohAq lsquorightrsquo nA-hAq lsquowrongrsquorAzI lsquosatisfiedrsquo nA-rAzI lsquounsatisfiedrsquopjala lsquofallrsquo ma-pole lsquocause to fallrsquoʃatoe lsquodrinkrsquo ma-stoe lsquogive drinkrsquomjAsA lsquosuckrsquo mA-mOsE lsquogive the suckrsquorAdOxE lsquoboil (intr)rsquo mA-rdOxE lsquoboil (tr)rsquo

xaraba lsquoruinedrsquo xArAb-IstAn lsquoruined placersquotʃol lsquouninhabited landrsquo tʃOl-IstAn lsquowildernessrsquohind lsquoIndiarsquo hInd-IstAn lsquoIndiarsquo

form coherent phonological groups as expressed by their being groupedtogether into constituents such as ldquoLaryngealrdquo and ldquoPlacerdquo

The organization of features into such a structure went hand-in-handwith the realization that the theory of rules could be constrained in veryimportant ways A long-standing problem in phonological theory was thequestion of how to express rules of multiple-feature assimilation We havediscussed rules of nasal place assimilation in previous chapters and notedin chapter 6 that such rules necessitate a special notation the featurevariable notation using α β γ and so on The notation makes some verybad predictions First notice that complete place assimilation requiresspecification of ten features in total

acoronalbanteriorgbackdhighqdistributed

mdashmdash

This is less simple and by the simplicity metric used in that theoryshould occur less frequently than (90)

(90) C [αcoronal] ___ [αcoronal]

This prediction is totally wrong (90) is not just uncommon it is com-pletely unattested Were there to be such a rule that assimilates only thespecification of coronal we would expect to find sets of assimilations suchas the following

nasalcontinuantconsonantal

lateral

Laryngeal Place

sonorantstrident

Coronal

anteriordistributed

Labial

roundlow

Dorsal

voiceconstricted

glottis

spreadglottis

The fact that the feature-variable theory allows us to formulate such anunnatural process at all and assigns a much higher probability of occur-rence to such a rule is a sign that something is wrong with the theoryThe theory says that there is only a minor difference in naturalness

between (92) and (89) since the rules are the same except that (92) doesnot include assimilation of the feature [anterior]

acoronalgbackdhighqdistributed

mdashmdash

There is a huge empirical difference between these rules (89) is verycommon (92) is unattested Rule (92) is almost complete place assimila-tion but [anterior] is not assimilated so np ɲk and mt become [mp][ŋk] and [nt] as expected but ɲt and ntʃ do not assimilate (as they wouldunder complete place assimilation) similarly ŋtʃ becomes [ɲtʃ] asexpected (and as well attested) but ŋp and ŋt become [np] and [nt] sincethe underlying value [ndash anterior] from ŋ would not be changed Thus theinclusion of feature variables in the theory incorrectly predicts the possi-bility of many types of rules which do not exist in human languageThe variable-feature theory gives no special status to a rule where both

occurrences of α occur on the same feature

(93) C

qcoronalaanteriorbbackghighddistributed

mdashmdash

This rule describes an equally unnatural and unattested process wherebya consonant becomes [t] before [pj] [p] before [q] and [pj] before [k] Rulessuch as (93) do not exist in human language which indicates that thelinear theory which uses this notation as a means of expressing assimila-tions makes poor predictions regarding the nature of phonological rulesThe variable notation allows us to refer to legions of unnatural classes

by randomly linking two unrelated features with a single variable

(94) a b c dahigharound

adistributedanasal

acoronalaanterior

avoicealateral

Class (a) applied to vowels refers to [y u e ə a] (b) refers to [n ɲ p ʈ k] butexcludes [mɳ t tʃ ŋ] (c) groups together [t k] andexcludes [p tʃ] (d) refers to [l]plus voiceless consonants Such groupings are not attested in any languageWith the advent of a theory of feature geometry such as in (88) this

problem disappeared In that theory the process of place assimilation isformulated not as the change of one feature value into another but isexpressed as the spreading of one node ndash in this case the Place node ndash atthe expense of another Place node Thus the change ɲ [m] _ [p] is seenas working as in (95)

(95) root root

Place Place

Labial

x[nasal]

+distrib-anterior

Coronal

Just as tone assimilation is the rightward or leftward expansion of thedomain of a tone feature this process of place assimilation is expansion ofthe domain of one set of place specifications to the exclusion of anotherWhen one Place node spreads and replaces the Place node of a neighboringsegment that means that all of the original place features are deleted andthe segment then comes to bear the entire set of place features that theneighboring segment hasWhat the feature-variable notation was able to do was express multiple-

feature assimilations but given this alternative theory multiple featureassimilations will be recast as spreading some node such as Place Thefeature-variable notation can be entirely eliminated since its one usefulfunction is expressed by different means The theory of feature geometryenables a simple hypothesis regarding the form of phonological ruleswhich radically constrains the power of phonological theory The hypoth-esis is that phonological rules can perform one simple operation (such asspreading inserting or deletion) on a single element (a feature or organiz-ing node in the feature tree)The thrust of much work on the organization of phonological representa-

tions has been to show that this theory indeed predicts all and only the kindsof assimilations found in human languages (specific details of the structureof the feature tree have been refined so thatwe nowknow for example thatthe featureswhich characterize vowel height formanode in the feature treeas do the features for the frontback distinction in vowels) The nonlinearaccount of assimilations precludes the unnatural classes constructed by theexpressions in (94) since the theory has no way to tie a specific value for afeature to the value of another feature The theory does not allow a rule like(92) which involves spreading of only some features under the place nodeThe nature of a tree like (88) dictates that when a rule operates on a highernode all nodes underneath it are affected equally Unattested ldquoassimila-tionsrdquo typified by (93) cannot be described at all in the feature-geometrictheory since in that theory the concept ldquoassimilationrdquonecessarilymeans ldquoofthe same unitrdquo which was not the case in the variable-feature theoryThe theory of features in (88) makes other claims pertaining to how

place of articulation is specified which has some interesting conse-quences In the linear model of features every segment had a completeset of plus or minus values for all features at all levels This is not the casewith the theory of (88) In this theory a well-formed consonant simplyrequires specification of one of the articulator nodes Labial Coronal orDorsal While a coronal consonant may have a specification under the

Dorsal node for a secondary vocalic articulation such as palatalization orvelarization plain coronals will not have any specification for [back] or[high] similarly consonants have no specification for [round] or Labialunless they are labial consonants or secondarily rounded In other wordssegments are specified in terms of positive characteristic propertiesThis has a significant implication in terms of natural classes Whereas

labials coronals and dorsals are natural classes in this theory (each has acommon property) ndash and in actual phonological processes these seg-ments do function as natural classes ndash the complements of these sets donot function as units in processes and the theory in (88) provides no wayto refer to the complement of those classes Thus there is no natural classof [ndashcoronal] segments ([p k] excluding [t tʃ]) in this theory Coronal is notseen as a binary feature in the theory but is a single-valued or privativeproperty and thus there is no way to refer to the noncoronals sincenatural classes are defined in terms of properties which they share notproperties that they donrsquot share (just as one would not class rocks andinsects together as a natural group to the exclusion of flowers by termingthe group ldquothe class of nonflowersrdquo) Importantly phonological rules donot ever seem to refer to the group [ndashcoronal] even though the class[+coronal] is well attested as a phonological class The model in (88)explains why we do not find languages referring to the set [p k] It alsoexplains something that was unexplained in the earlier model the con-sonantal groupings [p t] versus [tʃ k] are unattested in phonological rulesThe earlier model predicted these classes which are based on assignmentof the feature [anterior] In the model (88) the feature [anterior] is adependent of the Coronal node and thus labials and velars do not have aspecification of [anterior] so there is no basis for grouping [p t] or [tʃ k]together

93 Suprasegmental structure

Another aspect of nonlinear representational theory is the claim thatthere are phonologically significant structures above the level of thesegment ie units that encompass multiple segments Such structuresare referred to as ldquoprosodicrdquo a term which refers to poetic meter rhythmand singing which are aspects of language use that involve ldquohow stringsof segments are performedrdquo The best-known unit of prosody is the trad-itional concept of the syllable The term itself is one of the oldest inlinguistics originating from Ancient Greek sullabe but the nature of thesyllable and arguments for it have been elusive At various points incontemporary linguistics scholars have rejected or embraced the syllableand the syllable was not part of standard generative phonological theoryuntil 1976 when Kahn produced strong arguments for it within autoseg-mental theoryThe intuitive concept of ldquosyllablerdquo is not particularly difficult to under-

stand it is a string of segments which centers around one or more vowelsand includes some consonants to the left and to the right The problem

Laryngealconsonants likeh and ʔ howevermay lack anyplace specificationsthe featurestructure oflaryngeals remainsa topic forinvestigation

resides in justifying the addition of this concept to our arsenal of analyt-ical devices In segmental representations there are audible consequencesof features for example you can hear voicing nasality and glottalizationon segments even though relating features to phonetic properties isdifficult The problem of the syllable is that it has no audible definingproperty thus it cannot be justified as a prima facie transcriptional factno amount of ear training will enable you to ldquohearrdquo how many syllablesthere are in a word of the form [CVVVVC] in some unfamiliar languageand in [VCCCCV] you cannot ldquohearrdquo where one syllable begins and theother ends The evidence for the syllable is indirect in that groupingsequences of segments into a unit can lead to a simpler account of certainphonological processes in numerous languages

Possible consonant clusters One of the most widely invoked argu-ments of this nature regards the rules for possible consonant clusterswhich reflect the fact that sequences of segments have to be organizedinto definable syllables and languages impose various restrictions on howsyllables can be formed We will start with possible word-beginnings andword-ends in English and see how these relate to syllable structure Initialclusters may have the form sC (Cfrac14consonant) as in stick spit skunk alsosnow smite slay or they may be of the type OR (Ofrac14obstruent Rfrac14glide orliquid) as in fray through fly bleed breed pray clue The longest possibleinitial cluster has the shape sCR (sprint sklerotic strip splice) which reflectsthe interaction of the two rules pertaining to possible initial consonantclustersWords which violate these rules cannot be words of English thus

consonant plus stop clusters other than sC are nonexistent and are judgedby native speakers as being impossible (bnick pnort ptack dbonk fnilge)Likewise there are no stop+fricative clusters (kfimp ksunk pthing) Sonor-ants as the first member of a cluster are also excluded mbop rtot lfayyluck wnurge There are additional more specific restrictions on thepattern of allowed initial clusters For example coronal plus l is excluded(tluth dlifficult thlash chlort) except for [sl] (sleep) thanks to the specialrule allowing sC clusters Sequences of labial+w are also disallowed(pwang bwint mwerge fwet)Clusters of consonants at the end of English words are also subject to

restrictions Any consonant except h can stand at the end Consonantclusters can be of the form sonorant+consonant Thus words can endwith glide+consonant (height clown mouse leaf) liquid plus consonant(halt harp hart bilk false film born farm carl) or nasal+consonant (dancerunt punk brand lamp lymph lense) There are certain restrictions on suchfinal clusters One is that in a nasal plus voiced stop cluster the stop mustbe noncoronal thus fringe hand are allowed and [laeligmb] [haeligŋg] withpronounced final [b] [g] are disallowed The consonants [r j w] cannot bethe second consonant in a cluster [l] can follow [r j w] but not a nasal andnasals can only follow [r j w l]Certain sequences of voiceless obstruents are also allowed as long as

either the second consonant is [+anterior +coronal] (apt act depth apse

raft) or else the first consonant is s (cast cask clasp) Obstruent sequencesending in a noncoronal or nonanterior consonant are excluded (atp atclupsh ratf) as are clusters of fricative+obstruent where the fricative isnot s (cashk lithprafk) Clusters ending with voiced obstruents are alsodisallowed (abd abz) Notice that all of these rules involve allowed ordisallowed sequences of two consonants ndash no rules of combination specif-ically apply to just three-member clusters or four-member clusters andobserved limits on initial and final clusters all reduce to a chain of limitson two-consonant sequences It is also important to note that certainotherwise excluded clusters do arise when inflectional affixes are addedfor example the final cluster [bz] exists in the plural cabs and [gd] exists inpast tense flagged but such clusters only exist as combinatins of root plussuffixThe importance of the syllable in understanding these restrictions

comes from the fact that these are not just restrictions on how wordscan begin or end they are restrictions on how syllables can begin and endTaken together the preceding rules for syllable beginnings and endingsdefine possible word-medial clusters Some examples of allowed word-medial clusters are [tm] in atmosphere [mb] in camber [ʃr] in mushroom[rt] in barter [sb] in asbestos [bn] in Abney [md] in Camden [db] in Ledbetter[ʃk] in ashcan and [kf] in breakfast Note that these are not possible initial orfinal clusters except that [rt] is a possible final cluster In such cases thefirst consonant is the final consonant of one syllable and the second isthe initial consonant of the next syllable ndash [kaeligmbr] [bartr] [aeligbnij] [lɛdbɛtr] [brɛkfʌst] Three-consonant clusters are possible for examplebolster Andrew hamster translate electron costly which can be arrangedinto a possible syllable-final sequence followed by a possible sylable-initialsequence viz [bolstr] [aeligndruw] [haeligmstr] [traelignslejt] [ʌlɛktran][kastlij]Now consider illicit three-consonant medial clusters exemplified by

catmbop ([tmb]) fishrtot [ʃrt] gasbnick ([sbn]) lamdbonk ([mdb]) gushk-fimp ([ʃkf]) We have seen that the individual consonant pairs arepossible ndash [tm] [mb] [ʃr] [rt] [sb] [bn] [md] [db] [ʃk] and [kf] ndash but onlybecause the first member is a syllable-final consonant and the second issyllable-initial The three-consonant cluster [tmb] is ruled out because tmis not a possible syllable-final cluster and mb is not a possible syllable-initial cluster thus m cannot be assigned to any syllable ndash neither catmbopnor catmbop follows the rules for syllabification of consonants in EnglishSimilarly sb is not a possible syllable-final cluster and bn is not a possibleinitial cluster thus the cluster in gasbnick cannot be syllabifiedA syllable-based analysis of possible clusters automatically predicts therestrictions on word-medial three-consonant clusters Without the syl-lable as an organizing unit over segments a very complex set of additionalrules would be required to account for the restrictions on medial clusters

Phonological rules Rules of English consonant allophony discussed inchapter 2 also support the postulation of the syllable insofar as thoserules are best stated with reference to the syllable The best-known such

rule is the aspiration rule As is commonly recognized and explicitlyassumed in our previous discussion of the aspiration rule voiceless stopsare aspirated at the beginning of a syllable explaining the aspiration in[phɪt phlat əˈphɪr ʌˈphlaj] but not in [spɪt splɪt ʌˈspɛrəgəs slaeligp aeligpt]Another rule of American English which refers to the syllable is the one

glottalizing syllable-final voiceless stops where p t k become unreleasedglottalized [p˺ t˺ k˺] after a vocoid in the same syllable There is dialectalvariation in the extent to which all voiceless consonants undergo this rulebut examples involving t (which is the most susceptible to glottalization)include hit heart catkin Atkins light clout heights hearts atlas atlantic andWatneyrsquos By contrast there is no glottalization of t in stem apt beltmattress atrocious In the word stem t is clearly not preceded by a vocoidat all so the conditions of the rule are not satisfied likewise in apt andbelt In mattress atrocious the cluster tr is a cluster at the beginning of thesecond syllable so while t is preceded by a vocoid it is not in the samesyllable Consequently there is no glottalization in these examples On theother hand there is glottalization in atlas atlantic since tl is not a permit-ted initial cluster in English these words are syllabified as atlas atlanticLikewise tn is not an allowed cluster at the beginning of the syllable soWatneyrsquos is syllabified Watneyrsquos Since t is in the same syllable as thepreceding vocoid the consonant becomes glottalizedThe rule of glottalization provides important evidence regarding the

nature of the syllable The required relationship between the target conson-ant and the triggering vocoid is that they must be in the same syllable ndash theconsonant does not have to be at the end of the syllable see [kwaɹt˺s] lsquoquartzrsquoThis means that the ldquosyllablerdquo is not just a boundary ordered betweensegments ndash the phonological significance of the syllable goes beyond encod-ing the concepts ldquosyllable-initialrdquo and ldquosyllable-finalrdquo Being in a syllable is aproperty shared by a span of segments Analogous to the autosegmentalrepresentation of H linked to multiple vowels in Shona seen in (24) thesegments of [kwaɹt˺s] are linked to one syllable entity notated as σ

k w a r t s

The rule deriving glottalized consonants can accordingly be formulatedas (97)

[-voice-cont] [+sg][-cons]

r-unrounding A third rule of English phonology providing evidencefor the syllable is the one which pertains to rounding of r In somedialects r is realized both as a rounded and an unrounded rhotic approx-imant [ɹ] and [ɹw] following the rule that ɹw unrounds after a nonroundvowel in the same syllable Thus r is round in [ɹwejɲdʒ] range [thɹwej] tray[stɹwej] stray [fɹwej] fray also in [kɔɹw] core [tʊɹw] tour where the vowelpreceding r in the syllable is round and in [ʌˈɹwej] array where the

preceding vowel is in a separate syllable but r is unrounded in [kaɹ] car[kaɹt] cart [ˈbɪɹ] beer [hɛɹld] Harold The following rule unrounds ɹw after atautosyllabic nonround vowel

(98) s

[+cor-cons]

Vowel reduction Vowel reduction provides another argument forthe syllable in English The data below show as we have observed inchapter 4 that unstressed vowels reduce to schwa

A simple statement like ldquoan unstressed vowel becomes schwardquo forms thecore of the correct generalization but the following data indicate that thematter is more complex since the nature of the following consonantsmatters In some cases a CC cluster can stand between the target ofreduction and the next vowel but in other cases a CC cluster blocksreduction

If we take cognizance of syllable boundaries especially the ends of con-sonant clusters that are allowed in the beginning of the syllable then thegeneralization becomes much clearer unstressed vowels reduce to schwain English when they are at the end of the syllable

Other phenomena referring to the syllable Across languages therehas been a recurring puzzle regarding the expression of natural classesvia features and the role of word boundaries The problem is that thereexist many rules which treat a consonant and a word boundary alike but

Reduced Unreduced[əˈlaeligw] lsquoallowrsquo [ˈaeliglow] lsquoaloersquo[əˈnɔj] lsquoannoyrsquo [ˈaelignəlɪst] lsquoanalystrsquo[təˈlɛgrʌfij] lsquotelegraphyrsquo [ˌtɛləˈgraeligfɪk] lsquotelegraphicrsquo

Reduced Unreduced[əˈbrʌpt] lsquoabruptrsquo [aeligdˈmanɪʃ] lsquoadmonishrsquo[əˈtrowʃəs] lsquoatrociousrsquo [aeligtˈlaeligntɪk] lsquoatlanticrsquo[əˈstranəmij] lsquoastronomyrsquo [aelignˈdijən] lsquoAndean[əˈfrejd] lsquoafraidrsquo [arˈtɪstɪk] lsquoartisticrsquo

[aeliglˈpaeligkə] lsquoalpacarsquo

only for a specific set of rules Many dialects of Arabic have such a ruleone of vowel epenthesis which inserts [i] after a consonant which isfollowed by either two consonants or one consonant and a word bound-ary Thus in many dialects of Eastern Arabic underlying katab-t becomes[katabit] lsquoI wrotersquo and katab-l-kum becomes [katabilkum] lsquohe wrote toyou plrsquo The following rule seems to be required in a theory which doesnot have recourse to the syllable

(102) Oslash [i] C mdashmdash CC

Similarly a number of languages such as Yawelmani (chapter 6) haverules shortening long vowels when followed by two consonants orby a word-final consonant (thus taxakrsquoa taxak [taxak] lsquobringrsquodos-hin [doshin] lsquoreport (nonfuture)rsquo) which would be formalizedas follows

(103) [-long] mdashmdash C[+syl]C

The problem is that these rules crucially depend on the brace notation(ldquo rdquo) which joins together sets of elements which have nothing incommon a notation which has generally been viewed with extremeskepticism But what alternative is there since we cannot deny the exist-ence of these phenomenaThe concept of syllable provides an alternative way to account for such

facts What clusters of consonants and word-final consonants have incommon is that in many languages syllables have the maximal structureCVX therefore in taxak and doshin where there is shortening thelong vowels have in common the fact that the long vowel is followed by aconsonant ndash the syllable is ldquoclosedrdquo In contrast in [dosol] lsquoreport (dubi-tative)rsquo no consonant follows the long vowel Expressed in terms ofsyllable structure the vowel-shortening rule of Yawelmani (and manyother languages) can be expressed quite simply without requiring refer-ence to the questionable brace notation

V [-long] C

Another type of argument for the syllable is the domain argumentexamples being the arguments from English glottalization and r-unrounding where the fact of being in the same syllable is a crucialcondition on the rule One example comes from Cairene Arabic wherepharyngealization spreads to all segments in the syllable (originating fromsome coronal sonsonant ndash t and tʕ are contrastive phonemes in Arabiclikewise d and dʕ s and sʕ and in some dialects r and rʕ) Pharyngealizationalso affects vowels via this pharyngealization-spreading rule Examples ofthis distribution are [rʕaʕbʕ] lsquoLordrsquo from rʕabvs [rab] lsquoit sproutedrsquo [tʕiʕnʕ]lsquomudrsquo from tʕin] vs [tin] lsquofigsrsquo see especially the alternation [lʕaʕtʕiʕfʕ]lsquopleasant (m)rsquo ~ [lʕaʕtʕiʕfa] lsquopleasant (f)rsquo from lʕatʕif The addition of thefeminine affix -a has the consequence that the root-final consonant is

syllable final in the masculine but initial in the following syllable in thefeminine The rule of pharyngealization is formalized in (105)

(105) (mirror-image)

Because of the syllabification differences between lʕatʕif and lʕatʕifa fis subject to the rule only in the masculine despite the fact that theconditioning factor a vowel with the pharyngealization feature (derivedby spreading pharyngealization from the syllable-initial consonant) isimmediately adjacent to the consonant in both cases

Other suprasegmental units In addition to the syllable research hasprovided evidence for a number of other prosodic units First the syllableitself may have structure ndash the initial cluster of consonants form an onsetconstituent the final cluster of consonants form a coda constituent thevowel or vowels which form the heart of the syllable are the nucleus thenucleus and coda together may constitute a rhyme constituent Anotherprosodic unit related to the expression of syllabicity length syllable-weight and tone-bearingness is the mora Groups of syllables may them-selves be organized into a higher-level unit relevant to rhythm and stressknown as the foot and finally there may be a panoply of word- andphrase-level constituents such as the prosodic word phonologicalphrase and intonational phrase Such matters are part of the ongoingresearch program of phonological theory

Exercises1 LuluboNote on tone marks [v ]frac14 rising from L to M [v ]frac14 falling from M to L [v ]frac14 risingfrom M to H and [v ] frac14 falling from H to M Give the underlying form of the noun

Summary Answering a simple problem namely how to represent contour tonesled to ideas which not only solved the problem of contours but alsosolved a whole array of problems related to tone Since there is noreason to think that there should be a special theory just for tone anatural development of these changes applied to tone was a generalapplication of the autosegmental idea to all of phonology This resultedin sweeping changes to the theory of phonology and has resolvedmany earlier problems in how to state rules in a constrained mannerThis generalization of the results in one area to an entire subdisciplineis typical of the progression of scientific theories

roots and whatever morphemes mark the four case forms in the following databriefly discuss what theoretically interesting property these data illustrate Theword [anɖὲ] is the verb lsquoI seersquo in different tenses

2 ShambaaPropose autosegmental rules to account for the following tone alternationsNote that all infinitives have the final suffix -a

3 HoloholoVerbs have an infinitive prefix or a subject marker an optional negative prefixthen an optional object pronoun and lastly the verb stem The stem iscomposed of a root a number of optional derivational suffixes plus themorpheme -a which means lsquononpast verbrsquo or -ile meaning lsquopastrsquo Consonantmutation rules can be ignored (eg il in) as well as some of the segmentalallomorphs (kuhuuleena from kuhuulilana or kumweena from kumonila)What is important is tone and rules relating to vowel sequences Assume aprinciple of compensatory lengthening for the language where glide formationand vowel fusion applying to an underlying V+V sequence lengthen thevowel -i+o becomes [joo]

There are regularities regarding vowel length to consider There are no surfacerepresentations such as [kuponka] with a short vowel followed by thesequence nasal plus consonant also no forms like [kufjaka] with short vowelafter a glide Furthermore no words end in a long vowel

The data are divided into conceptually related groups illustrating a particularpoint such as a rule a particular restriction on a rule or the surface tonepattern of words of a particular syllabic structure It is important to integratethe whole data set and for example to relate kumonana lsquoto see each otherrsquoto kumona lsquoto seersquo and also to kulolana lsquoto look at eorsquo since kumonanahas morphemes in common with both words

Bare noun Unfocused object Focused object Proper nameSubjunctive Past Past

ebı anɖὲ bı anɖὲ ebǐ anɖὲ ebı lsquolionrsquoarɪ anɖὲ arı anɖὲ arı anɖε ar ɪ lsquobirdrsquotı anɖὲ tı anɖὲ tı anɖὲ t ı lsquocowrsquo

lsquoto Vrsquo lsquoto V forrsquo lsquoto V eorsquo lsquoto V for eorsquo lsquoto V itrsquo lsquoto V it forrsquokudika kudikia kudikana kudikiana kutʃıdıka kutʃıdıkıa lsquocookrsquokutoa kutoea kutoana kutoeana kutʃıtoa kutʃıtoea lsquobeatrsquokuʃuntha kuʃunthia kuʃunthana kuʃunthiana kutʃıʃuntha kutʃıʃunthıa lsquobathersquo

lsquoto Vrsquo lsquoto V forrsquo lsquoto V eatʃ otherrsquo lsquoto V for eatʃ otherrsquokukoma kukomea kukomana kukomeana lsquokillrsquokufua kufuıa kufuana kufuıana lsquolaunderrsquokuʃ ıʃa kuʃ ıʃ ıa kuʃ ıʃana kuʃ ıʃ ıana lsquosmearrsquokufumbatıʃa kufumbatıʃ ıa kufumbatıʃana kufumbatıʃ ıana lsquopackrsquo

kumona lsquoto seersquo kusila lsquoto forgersquokulola lsquoto look atrsquo kubula lsquoto drawrsquo

kumonana lsquoto see eorsquo kusilıla lsquoto forge forrsquokulolana lsquoto look at eorsquo kubulila lsquoto draw forrsquo

kusilılana lsquoto forge for eorsquo kubulilana lsquoto draw for eorsquokutegelela lsquoto listenrsquo kutegelesja lsquoto make listenrsquokutegelelana lsquoto listen to eorsquo kusololana lsquoto choose eorsquo

kulja lsquoto eatrsquo kuhja lsquoto carryrsquokuliila lsquoto eat forrsquo kuhiila lsquoto carry forrsquokubuusja lsquoto askrsquo kukwaata lsquoto ownrsquokubiiha lsquoto be badrsquo kuhiita lsquoto be blackrsquokutuuta lsquoto hitrsquo kusjiika lsquoto buryrsquo

kubiika lsquoto putrsquo kubiikılila lsquoto put forrsquokuliilıla lsquoto eat for st for st elsersquo kukwaatana lsquoto own eorsquokusjiikana lsquoto bury eorsquo kutuutila lsquoto hit forrsquo

kwiita lsquoto callrsquo kwiitana lsquoto call eorsquokweema lsquoto sufferrsquo kwaatıka lsquoto splitrsquokweelela lsquoto clean uprsquo kweelelana lsquoto clean eo uprsquokwiihaga lsquoto killrsquo kwiihagana lsquoto kill eorsquokooja lsquoto restrsquo kuula lsquoto buyrsquokooga lsquoto washrsquo koogela lsquoto wash forrsquokoogelela lsquoto wash for st for st elsersquo koogelelana lsquoto wash for eorsquo

kutoonta lsquoto fillrsquo kutoontamana lsquoto be fullrsquokuloomba lsquoto requestrsquo kuloombela lsquoto request forrsquokuloombelana lsquoto request for eorsquo kusiindala lsquoto make disappearrsquokusiingına lsquoto put acrossrsquo kusiingınına lsquoto put across forrsquokwiimba lsquoto singrsquo kwiimbıla lsquoto sing forrsquo

kunjwiisa lsquoto make drinkrsquo kunjwiisııbwa lsquoto be made to drinkrsquokuhuuleena lsquoto hit for eorsquo kutimwıına lsquoto break forrsquo

kuhima lsquoto leaversquo kuhimja lsquoto make leaversquokukwaata lsquoto ownrsquo kukwaatja lsquoto make ownrsquokoonka lsquoto suckrsquo koonkja lsquoto make suckrsquokubusa lsquoto missrsquo kubusja lsquoto make missrsquo

kukoloma lsquoto irritatersquo kumukoloma lsquoto irritate himrsquokubakoloma lsquoto irritate themrsquo kulola lsquoto look atrsquokumulola lsquoto look at himrsquo kubalola lsquoto look at themrsquokumumona lsquoto see himrsquo kubamona lsquoto see themrsquokutegelela lsquoto listen torsquo kumutegelela lsquoto listen to himrsquokubategelela lsquoto listen to themrsquo

kusimona lsquoto not seersquo kulola lsquoto look atrsquokusilola lsquoto not look atrsquo kusikoloma lsquoto not irritatersquokusimulola lsquoto not look at himrsquo kusibalola lsquoto not look at themrsquo

Further readingClements and Hume 1995 Goldsmith 1990a Hayes 1986 Odden 1995

kusimumona lsquoto not see himrsquo kusibamona lsquoto not see themrsquokutiinına lsquoto worryrsquo kusitıınına lsquoto not worryrsquo

kwiitıınına lsquoto worry oneselfrsquo kumutiinına lsquoto worry himrsquokwiilola lsquoto look at oneselfrsquo kwiimona lsquoto see oneselfrsquokuula lsquoto buyrsquo kusjuula lsquoto not buyrsquokusjuulana lsquoto not buy eorsquo kusimuula lsquoto not buy himrsquokwiita lsquoto callrsquo kusiita lsquoto not callrsquokusiilola lsquoto not look at selfrsquo kusiimona lsquoto not see selfrsquo

Glossaryabsolute neutralization The elimination of an underlying phoneme in all contexts so that it

always merges with some other phonemeacoustics The study of physical vibrations (sounds)affricate A stop with a homorganic fricative release

allomorphs Different surface realizations of a single morpheme traditionallyonly considering nonallophonic differences eg the three variantsof the English plural [-s] [-z] and [-ɨz]

allophone A contentually determined variant of a phoneme the realization ofa phoneme in a specific environment eg [k] [kh] [kj] [khj] inEnglish are allophones of the phoneme k

alveolar ridge The ridge between the back of the teeth and the hard palatealveopalatal A consonant produced by placing the tongue on the hard palate

behind the alveolar ridgeapproximant A sound made with very little constriction where articulators

approximate but do not touch which produces no turbulence inthe airflow

archiphoneme A theoretical segment which is only partially specified for phoneticproperties omitting some properties such as voicing or nasalitywhich may be determined by rule

articulation The contact of two speech organs such as the tongue tip and thehard palate

aspiration Noise produced by air rushing through the open glottis at therelease of a consonant

assimilation Making segments be more similar along some dimensionassociation lines Lines which indicate that two autosegments are in an association

relation thus are produced at the same timebilabial A sound produced with both lipsblade The flat surface of the tongue behind the tip and in front of the root

breathy A sound produced with abducted vocal folds and a high rate ofairflow through the glottis

central A vowel formed with the tongue horizontally positioned in thecenter of the space for vowel articulation between front and back(compare mid for the vertical axis)

click A stop consonant produced by creating a vacuum inside the mouthwith a raisedback of the tongue and tongue tip or closed lips Employedin a limited number of African especially Khoisan languages

coda The final sequence of consonants in a syllablecompensatory lengthening The lengthening of a segment caused by the deletion or desyllabi-

fication of an adjacent segmentcomplementary

distributionDistribution of two or more sounds where the context in which onesound appears is the complement of the contexts where the othersounds appear

complex wave A waveform built from more than one sine wavecontour tone A tone produced by movement from one pitch level to another

contrast A property of pairs of sounds two sounds contrast if they can formthe sole difference between different words in a language

coronalization The change of a noncoronal sound (p k) to a coronal sound (t tʃ)usually in the environment of a front vowel or glide

creaky An irregular mode of vocal fold vibration where only the frontportion vibrates

dental A consonant produced by contact with the teethdeterminant The segment in the environment which causes a phonological

change (also trigger)diphthong A combination of two vocoids within the syllable nucleus

dissimilation Making two segments become less alikedistinctive features A set of phonetic properties hypothesized to be universal and the

basis for all human language soundsdownstep A contrastive lowering of tone register notated with a raised

exclamation mark or down-arrow See upstepejective A stop consonant produced by raising the larynx with the glottis

constricted which creates pressure in the oral cavityenvironment The sounds preceding and following some other sound

epenthesis Insertion of a segmentflap A consonant produced by rapidly striking one articulator with

another Flaps are usually produced with the tonguefloating tone A tone which is not associated with a segment

focus In a rule the segment which undergoes the changefoot A rhythmic grouping of syllables relevant for stress systems

formant An overtone caused by the resonance frequency of the vocaltract a frequency band where there is a concentration ofacoustic energy

free variation A pair of pronunciations either of which can be used the choice isnot governed by grammatical factors

frequency Rate of repetition of a (semi-)periodic functionfricative A sound produced by forcing air through a narrow constriction

which creates turbulencefront A vowel formed with the tongue horizontally positioned in front of

the space for vowel articulation closest to the mouth openingglide A vowel-like consonant produced with minimal constriction

glottis The opening in the larynx between the vocal folds through whichair passes

hacek The diacritic symbol ˇ used to indicate rising tone on vowelshardening The change of a less constricted consonant to a more constricted

one such as the change of a glide to a fricative or a stophigh Sounds produced with a raised tongue body For vowels [i u] as

contrasted with [e o]homorganic Having the same place of articulation

implicational relation The relation where presence of one property in a language is anecessary precondition for the presence of some other property

324 Glossary

implosive A stop consonant formed by creating a vacuum within the mouthby constricting and lowering the larynx

labial A segment involving the lips as an articulatorlarynx The cartilaginous structure that houses the vocal folds

lax Vowel produced with a less deliberate more central or lowerarticulation Comparable to open contrast tense

lenition A change of a consonant to reduce the degree of constriction egthe change from a stop to a fricative or glide

lexicon The collection of morphemes which must be memorized a mentaldictionary

lingual Pertaining to the tongueliquids Consonants of the type [r l]

low Sounds produced with a lowered tongue vowels like [a aelig] andpharyngeals [ħ ʕ]

major class The set of features [sonorant] [syllabic] [consonantal] or theirequivalents

manner of articulation Traditionally the properties of a consonant other than the place ofarticulation and its laryngeal properties

markedness An abstract property referring to the ldquounusualnessrdquo or difficulty of asound or process

mid Vowel sounds such as [e o] produced with the tongue around themidpoint on the vertical axis compare central which pertains tothe midpoint along the horizontal axis

minimal pair A pair of distinct words differing solely in the choice of a singlesegment

mora A unit of prosodic weight related to length a long vowel has twomoras and a short vowel has one The mora may be a property ofboth a particular segment and an entire syllable

morpheme The smallest unit of word analysis such as a root or affixSupposedly the smallest meaning-bearing unit but not allmorphemes have identifiable meanings

morpheme structure rulesconditions

Rules that state the nature of possible underlying forms ofmorphemes

morphophonemics Phonological alternations especially nonallophonic changesnasal A sound produced with air flowing through the nasal passages

natural class A set of segments defined by a particular combination of featurespecifications which act as a group in phonological rules

neutral position The position which the tongue assumes prior to speakingapproximately that of [ε] Used as the reference point to definerelative movements of the tongue

neutralization Eliminating an underlying distinction between phonemes in somecontext

nucleus The vowels or syllabic segments which form the center of a syllableobstruent A nonsonorant consonant such as stops and fricatives

onset The consonants at the beginning of the syllable which precede thevowel

onset The initial sequence of consonants in a syllable

Glossary 325

palatal Referring to the hard or soft palate As a primary articulation aconsonant produced at the boundary between the hard and softpalate

palatalization Either a secondary articulation made by superimposing a j-likearticulation on a consonant or a wholesale change of a consonantrsquosplace of articulation to alveopalatal (see coronalization)

pharynx The lower part of the throatphonation The manner of vibration of the vocal folds (modal breathy creaky)phoneme A mental integration of the different physical properties of the

sounds used in a language abstracting away from specific phoneticproperties which are due to the context where the sound appears

pitch The percept of rate of vibrationprenasalization A sound produced with an initial interval of nasal airflow ndash often

treated as a homorganic cluster of nasal plus consonantprivative A feature having only one value either the feature is present or not

presentprosody Properties ldquoaboverdquo the segment which pertain to syllabification

length stress and rhythmretroflex Consonant articulation involving the tip of the tongue and the back

of the alveolar ridge or palatereversal of sound change The historical loss of a phonological rule which leads to the (par-

tial) restoration of earlier sounds ndash Yiddish and Ukrainian provideclassic examples

rhyme A portion of the syllable encompassing the nucleus and codaround A sound produced with protruded lips

segment A mental division of the continuous stream of speech into signifi-cant permutable units

semi-vowels See glidespectrogram A continuous analytic display of acoustic properties of sound over

time showing which frequencies are emphasized at each momentspontaneous voicing Passive vibration of the vocal folds which results from breathing a

characteristic of sonorants This is brought about by a particularpositioning of the vocal folds combined with a relatively uncon-stricted air passage

stop A sound where the flow of air is completely obstructedstress A form of prosodic prominence typically resulting in greater length

and higher pitch within the syllablestructural change That part of a rule which states in what way a given sound changes

structure preserving The property of rules that outputs are modified to preserve thenature of underlying forms especially in terms of what phonemesexist in the language

syllable A unit of speech claimed to be relevant for the organization ofwords a grouping of consonants and vowels into a C0V1C0

constituentsyllable peak The span within the syllable perceived as (capable of) bearing stress

syncope Deletion of a vowel in a medial syllable especially in a fashion thataffects alternating syllables

326 Glossary

target See focustense Vowel produced with a more deliberate and higher articulation

Comparable to close contrast laxtone A property based on the contrastive use of pitch

translaryngeal harmony Assimilation of vowels which applies only across laryngealconsonants

trigger See determinanttypology The parametric study of crosslinguistic variation in grammatical

structureunderlying Pertaining to the initial state in a phonological derivation the

phonological facts holding of a word or morpheme before phono-logical rules affect changes

upstep A contrastive raising of tone register notated with a raised invertedexclamation mark or an up-arrow See downstep

uvular A consonant formed by constricting the back of the throat near theuvula with the back of the tongue

velar A consonant formed by bringing together the back of the tongueand the soft palate

velarized A secondary articulation formed by approximating the back of thetongue towards the soft palate

velum The soft palatevocal folds Two membranes in the larynx whose vibration provides voicing

and most of the sound energy of speechvocal tract The air passages above the glottis including the oral tract and the

nasal passagesvocoid A vowel-like sound with no major obstruction the class of vowels

and glidesvoicing The presence of vocal fold vibrations during the production of a

sound produces voicingvowel harmony An assimilation between vowels where one vowel takes on the

properties of a neighboring vowelwaveform A display of the time-varying amplitude of sound pressureweakening See lenition

weight A property of syllables which may be divided into light and heavysyllables heavy syllables typically have a long vowel or diphthongor sometimes a short vowel plus consonant See mora

Glossary 327

ReferencesAbaev V I 1964 A Grammatical Sketch of OsseticBloomington Indiana University Press

Akinlabi Akin 1984 ldquoTonal underspecification andYoruba tonerdquo Ibadan University of Ibadan doctoraldissertation

Al-Mozainy Hamza Q 1981 ldquoVowel alternations in aBedouin Hijazi Arabic dialect abstractness andstressrdquo Austin University of Texas doctoraldissertation

Allen Joseph and J B Greenough 1983 Allen andGreenoughrsquos New Latin Grammar for Schools and CollegesFounded on Comparative Grammar New Rochelle NYA D Caratzas

Ambrazas Vytautas 1997 Lithuanian GrammarVilnius Baltos Lankos

Andersen Torben 1987 ldquoAn outline of Lulubophonologyrdquo Studies in African Linguistics 18 39ndash65

Anderson Stephen 1974 The Organization ofPhonology New York Academic Press

Andrzejewski B 1964 The Declensions of Somali NounsLondon School of Oriental and African Studies

Aquilina J 1965 Maltese London Teach YourselfBooks

Ariste Paul 1968 A Grammar of the Votic LanguageBloomington Indiana University Press

Arnott David W 1964 ldquoDownstep in the Tiv verbalsystemrdquo African Language Studies 5 34ndash51

Ashby Michael and John Maidment 2005 IntroducingPhonetic Science Cambridge Cambridge UniversityPress

Aziza Rose 2008 ldquoNeutralization of contrast in thevowel system of Urhobordquo Studies in African Linguistics37 1ndash19

Barker Muhammad A R 1963 Klamath Dictionary Uni-versity of California Publications in Linguistics 31 LosAngeles and Berkeley University of California Press

1964 Klamath Grammar University of CaliforniaPublications in Linguistics 32 Los Angeles andBerkeley University of California Press

Beatty John 1974 Mohawk Morphology OccasionalPublications in Anthropology Linguistic Series 2Greeley CO Museum of Anthropology University ofNorthern Colorado

Bhat D N S and M S Ningomba 1997 ManipuriGrammar Munich Lincom Europa

Bills Garland Bernardo Vallejo and Rudolph Troike1969 An Introduction to Spoken Quechua AustinUniversity of Texas Press

Borg Albert and Marie Azzopardi-Alexandre 1997Maltese London Routledge

Boxwell Helen and Maurice Boxwell 1966 ldquoWeriphonemesrdquo In S A Wurm (ed) Papers in New GuineaLinguistics 5 77ndash93 Australian National UniversityCanberra

Brame Michael 1972 ldquoOn the abstractness ofphonology Maltese rdquo In M Brame (ed) Contributionsto Generative Phonology 22ndash61 Austin University ofTexas Press

Bright William 1957 The Karok Language Universityof California Publications in Linguistics 13 LosAngeles and Berkeley University of California Press

Broselow Ellen 1979 ldquoCairene Arabic syllable struc-turerdquo Linguistic Analysis 5 345ndash82

Bulatova Nadezhda Ja and Lenore Grenoble 1999Evenki Munich Lincom Europa

Campbell Lyle 1974 ldquoPhonological featuresproblems and proposalsrdquo Language 50 52ndash65

Carlton Terence 1971 The Declension of Nouns inUkrainian a Studentrsquos Reference EdmontonDepartment of Slavic Languages University ofAlberta

Chomsky Noam 1967 ldquoSome general properties ofphonological rulesrdquo Language 43 102ndash28

Chomsky Noam and Morris Halle 1968 The SoundPattern of English New York Harper and Row

Chung Sandra 1983 ldquoTransderivational relations inChamorro phonologyrdquo Language 59 35ndash66

Clements G N 1978 ldquoTone and syntax in Ewerdquo InD J Napoli (ed) Elements of Tone Stress and Intonation21ndash99 Washington Georgetown University Press

1984 ldquoPrinciples of tone assignment in KikuyurdquoIn G N Clements and J Goldsmith (eds)Autosegmental Studies in Bantu Tonology 281ndash340Dordrecht Foris

Clements G N and Elizabeth Hume 1995 ldquoTheinternal organization of speech soundsrdquo InJ Goldsmith (ed) Handbook of Phonological Theory245ndash306 Oxford Blackwell

Cohn Abigail 1993 ldquoNasalization in Englishphonology or phoneticsrdquo Phonology 10 43ndash81

Cole Desmond 1955 An Introduction to TswanaCapetown Longman

1967 Some Features of Ganda Linguistic StructureJohannesburg Witwatersrand Press

Coupez Andreacute 1955 Esquisse de la langue holoholoTerveuren Museacutee royale de lrsquoAfrique centrale

Cowell Mark 1964 Reference Grammar of Syrian ArabicWashington Georgetown University Press

Cusihuamaacuten Antonio 1976 Diccionario QuechuaCuzco-Collao Lima Ministerio de educacion institutode estudios peruanos

Dambriunas Leonardas Antanas Klimas andWilliam Schmalstieg 1966 Introduction to ModernLithuanian Brooklyn Franciscan Fathers Press

Doke Clement 1938 Textbook of Lamba GrammarJohannesburg Witwatersrand Press

Dolphyne Florence Abena 1988 The Akan (Twi-Fante)Language Accra Ghana Universities Press

Downing Laura 1996 The Tonal Phonology of JitaMunich Lincom Europa

Ebert Karen 1975 Sprache und Tradition der Kera(Tschad) Marburger Studien zur Afrika undAsienkunde Serie A Bd 6 Berlin Reimer

Echeverriacutea Max and Heles Contreras 1965ldquoAraucanian phonemicsrdquo International Journal ofAmerican Linguistics 31 132ndash5

Elimelech Baruch 1978 A Tonal Grammar of EtsakoBerkeley University of California Press

Emeneau Murray 1961 Kolami a DravidianLanguage Annamalainagar Annamalai UniversityPress

Flora Marie Jo-Ann 1974 ldquoPalauan phonology andmorphologyrdquo San Diego University of Californiadoctoral dissertation

Foster Joseph 1969 ldquoOn some phonological rules inTurkishrdquo Champaign University of Illinois doctoraldissertation

Gleason Henry 1955 An Introduction to DescriptiveLinguistics New York Holt

Goldsmith John 1976 ldquoAutosegmental phonologyrdquoCambridge MA MIT doctoral dissertationDistributed by Indiana University Linguistics ClubBloomington

1990a Autosegmental and Metrical Phonology OxfordBlackwell

1990b ldquoHarmonic phonologyrdquo In J Goldsmith(ed) The Last Phonological Rule 21ndash60 ChicagoUniversity of Chicago Press

Greenberg Joseph 1978 Universals of HumanLanguage Stanford Stanford University Press

Hale Mark and Charles Reiss 2006 The PhonologicalEnterprise Oxford Oxford University Press

Hale William Gardner and Carl Darling Buck 1966A Latin Grammar Tuscaloosa University of AlabamaPress

Halle Morris 1959 The Sound Pattern of RussianThe Hague Mouton

Halle Morris and George N Clements 1983 ProblemBook of Phonology a Workbook for Courses in IntroductoryLinguistics and Modern Phonology Cambridge MA MITPress

Hangin John 1968 Basic Course in MongolianBloomington Indiana University Press

Harris John 1994 English Sound Structure OxfordBlackwell

Haspelmath Martin 1993 A Grammar of LezgianBerlin and New York Mouton de Gruyter

Hayes Bruce 1986 ldquoAssimilation as spreading inToba Batakrdquo Linguistic Inquiry 17 467ndash99

330 References

1995Metrical Stress Theory Principles and Case StudiesChicago University of Chicago Press

Hayes Bruce Robert Kirchner and Donca Steriade(eds) 2004 Phonetically Based Phonology CambridgeCambridge University Press

Hoberman Robert 1988 ldquoEmphasis harmony in aModern Aramaic dialectrdquo Language 64 1ndash26

Hoffmann Carl 1963 A Grammar of the MargiLanguage London Oxford University Press

Hoijer Harry 1933 ldquoTonkawa an Indian language ofTexasrdquo In Franz Boas (ed) Handbook of AmericanIndian Languages vol 3 1ndash148 WashingtonSmithsonian Institution

Hualde Jose 1992 Catalan a Comprehensive GrammarLondon Routledge

Hudson Grover 1974 ldquoThe role of SPCs in naturalgenerative phonologyrdquo In A Bruck R Fox andM LaGaly (eds) Papers from the Parasession on NaturalPhonology 171ndash83 Chicago Chicago LinguisticsSociety

Hulstaert Gustav 1961 Grammaire du lomoacutengoTerveuren Museacutee royale de lrsquoAfrique centrale

Hume Elizabeth 1996 ldquoCoronal consonant frontvowel parallels in Malteserdquo Natural Language andLinguistic Theory 14 163ndash203

Hyman Larry 1970 ldquoHow concrete is phonologyrdquoLanguage 46 58ndash76

Hyman Larry and Russell Schuh 1974 ldquoUniversals oftone rules evidence from West Africardquo LinguisticInquiry 5 81ndash115

Inkelas Sharon 1989 ldquoProsodic constituency in thelexiconrdquo Stanford Stanford University doctoraldissertation

Isac Daniela and Charles Reiss 2008 I-Language AnIntroduction to Linguistics as Cognitive Science OxfordOxford University Press

Jakobson Roman and Morris Halle 1956Fundamentals of Language The Hague Mouton

Jakobson Roman Gunnar Fant and Morris Halle1952 Preliminaries to Speech Analysis Cambridge MAMIT Press

Johnson Keith 1997 Acoustic and Auditory PhoneticsOxford Blackwell

Josephs Lewis 1975 Palauan Reference GrammarHonolulu University of Hawaii Press

Kahn Daniel 1976 ldquoSyllable based generalizationsin English phonologyrdquo Cambridge MIT doctoraldissertation Distributed by Indiana UniversityLinguistics Club Bloomington

Kaisse Ellen and Patricia Shaw 1985 ldquoOn the theoryof lexical phonologyrdquo Phonology 2 1ndash30

Kaye Jonathan 1982 ldquoHarmony processes in VatardquoIn N Smith and H van der Hulst (eds) Structureof Phonological Representations part II 385ndash452Dordrecht Foris

Kenesei Istvaacuten Robert M Vago and Anna Fenyvesi1998 Hungarian London and New York Routledge

Kenstowicz Michael 1972a ldquoLithuanian phonologyrdquoStudies in the Linguistic Sciences 22 1ndash85

1972b ldquoThe morphophonemics of the Slovaknounrdquo Papers in Linguistics 5 550ndash67

1994 Phonology in Generative Grammar OxfordBlackwell

Kenstowicz Michael and Charles Kisseberth 1977Topics in Phonological Theory New York AcademicPress

1979 Generative Phonology Description and TheoryNew York Academic Press

Kimball Geoffrey D 1991 Koasati Grammar LincolnUniversity of Nebraska Press

Kiparsky Paul 1968a ldquoLinguistic universals andlinguistic changerdquo In E Bach and R Harms (eds)Universals of Linguistic Theory 171ndash202 New YorkHolt

1968b ldquoHow abstract is phonologyrdquo Distributedby Indiana University Linguistics Club BloomingtonReprinted in P Kiparsky Explanation in PhonologyDordrecht Foris 1982

Kisseberth Charles 1969 ldquoOn the abstractness ofphonology the evidence from Yawelmanirdquo Papers inLinguistics 1 248ndash82

1984 ldquoDigo tonologyrdquo In G N Clements andJ Goldsmith (eds) Autosegmental Studies in BantuTonology 105ndash82 Dordrecht Foris

Klokeid Terrence 1976 ldquoTopics in Lardil grammarrdquoCambridge MA MIT doctoral dissertation

References 331

Konstantinova Olga A 1964 Evenkijskij Jazyk FonetikaMorfologija Moscow Nauka

Koutsoudas Andreas Gerald Sanders and Craig Noll1974 ldquoOn the application of phonological rulesrdquoLanguage 50 1ndash28

Krauss Scott 1981 ldquoTopics in Chukchee phonologyand morphologyrdquo Urbana University of Illinoisdoctoral dissertation

Krueger John 1962 Yakut Manual Area HandbookGrammar Graded Reader and Glossary BloomingtonIndiana University Press

Ladefoged Peter and Keith Johnson 2010 A Course inPhonetics 6th edition Cengage Learning Boston

Leben William 1978 ldquoThe representation of tonerdquoIn V Fromkin (ed) Tone a Linguistic Survey 177ndash219New York Academic Press

Lees Robert 1961 The Phonology of Modern StandardTurkish Uralic and Altaic Series 6 BloomingtonIndiana University Press

Lehtinen Meri 1963 Basic Course in FinnishBloomington Indiana University Press

Liberman Mark 1983 ldquoUncommon approaches to thestudy of speechrdquo In PMacNeilage (ed) The Production ofSpeech 265ndash74 New York and Berlin Springer

Lleo Concepcioacuten 1970 Problems of Catalan PhonologyStudies in Linguistics and Language Learning 8Seattle University of Washington

Maddieson Ian 1984 Patterns of Sounds CambridgeCambridge University Press

Martin Samuel 1975 A Reference Grammar of JapaneseNew Haven Yale University Press

1992 A Reference Grammar of Korean Rutland VTC E Tuttle

Mathiassen Terje 1996 A Short Grammar ofLithuanian Columbus Slavica

Michelson Karin 1988 A Comparative Study of LakeIroquoian Accent Dordrecht Reidel

Milner G B 1966 Samoan Dictionary London OxfordUniversity Press

Nedjalkov Igor 1997 Evenki Routledge London

Newman Paul 1968 ldquoThe reality of morphophone-micsrdquo Language 44 507ndash15

Newman Stanley 1944 Yokuts Language of CaliforniaNew York Viking Fund Publications inAnthropology 2

Obolensky Serge Kambiz Panah and FereidounNouri 1963 Persian Basic Course Units 1ndash12Washington Foreign Service Institute reprinted byCenter for Applied Linguistics

Obolensky Serge Debebow Zelelie and MulugetaAndualem 1964 Amharic Washington ForeignService Institute

Odden David 1995 ldquoTone African languagesrdquo InJ Goldsmith (ed) The Handbook of Phonological Theory444ndash75 Oxford Blackwell

2013 ldquoFormal phonologyrdquo In S Blaho M Kraumlmerand B Moreacuten-Duolljaacute (eds) Nordlyd 401 A Festschrifton the Occasion of X Years of CASTL Phonology andCurt Ricersquos Lth Birthday 249ndash73

Ohala John 1978 ldquoSouthern Bantu vs the world thecase of palatalization of labialsrdquo Proceedings of theAnnual Meeting of the Berkeley Linguistic Society 4370ndash86

Osborn Henry 1966 ldquoWarao I phonology andmorphophonemicsrdquo International Journal of AmericanLinguistics 32 108ndash23

Paradis Carole 1992 Lexical Phonology and Morphologythe Nominal Classes in Fula New York and LondonGarland

Payne David L 1981 The Phonology and Morphology ofAxininca Campa Arlington Summer Institute ofLinguistics

Pickett Velma 2002 Manual de Morfosintaxis 4th edi-tion (electronic) Mexico City SIL Mexico

Pike Kenneth 1948 Tone Languages a Technique forDetermining the Number and Type of Pitch Contrasts ina Language Ann Arbor University of Michigan Press

Pitkin Harvey 1984 Wintu Grammar Berkeley andLos Angeles University of California Press

Popova Tatiana V 1972 ldquoParadigmatičeskijekonsonantyje rjady čredovany v jugo-zapadnyxukrainskix dialektax (na materiale govoras sadžava)rdquo In G Klepikova (ed) Karpatskajadialektologia i onomastika 179ndash239 Moscow Nauka

Postal Paul 1968 Aspects of Phonological Theory NewYork Harper amp Row

332 References

Press Ian and Stefan Pugh 1994 Colloquial UkrainianLondon and New York Routledge

Pullum Geoff 1976 ldquoThe Duke of York gambitrdquoJournal of Linguistics 12 83ndash102

Rennison John 1997 Koromfe London and New YorkRoutledge

Rice Keren 1989 A Grammar of Slave Berlin DeGruyter

Rich Furne 1963 ldquoArabela phonemes and high-levelphonologyrdquo In B Elson (ed) Studies in Peruvian IndianLanguages 193ndash206 Norman OK Summer Instituteof Linguistics

Rubach Jerzy 1993 The Lexical Phonology of SlovakOxford Clarendon Press

Saagpakk Paul 1992 Eesti-Inglise Sotildenaraamat EstonianndashEnglish Dictionary 2nd edition TallinnKoolibri

Saeed John 1993 Somali Reference Grammar Kensing-ton MD Dunwoody

1999 Somali Amsterdam and PhiladelphiaBenjamins

Sapir Edward 1925 ldquoSound patterns in languagerdquoLanguage 1 37ndash51

1933 ldquoThe psychological reality of phonemesrdquo InDavid Mandelbaum (ed) Selected Writings of EdwardSapir 46ndash60 Berkeley and Los Angeles University ofCalifornia Press

Saxton Dean 1963 ldquoPapago phonemesrdquo InternationalJournal of American Linguistics 29 29ndash35

Saxton Dean and Lucille Saxton 1969 Papago andPima to English Dictionary Tucson University of Ari-zona Press

Siptaacuter Peacuteter and Mikloacutes Toumlrkenczy 2000 The Phon-ology of Hungarian Oxford Clarendon Press

Snoxall R A 1967 LugandandashEnglish DictionaryOxford Oxford University Press

Snyman J J Shole and J Le Roux 1990 Dikišinare yaSetswana English Afrikaans Pretoria Via Afrika

Sohn Hoh-min 1975 Woleaian Reference GrammarHonolulu University of Hawaii Press

Stanley Richard 1967 ldquoRedundancy rules inphonologyrdquo Language 43 393ndash436

Stevens Kenneth 1998 Acoustic Phonetics CambridgeMA MIT Press

Topping Donald 1968 ldquoChamorro vowel harmonyrdquoOceanic Linguistics 7 67ndash79

Topping Donald and Bernadita Dungca 1973Chamorro Reference Grammar Honolulu Universityof Hawaii Press

Trubetzkoy Nicholas S 1939 Principles of PhonologyTranslated by C Baltaxe 1969 Berkeley and LosAngeles University of California Press

Tryon James 1970 An Introduction to MaranungkuPacific Linguistics B15 Canberra Australian NationalUniversity

Vago Robert 1980 The Sound Pattern of HungarianWashington Georgetown University Press

Vaux Bert 1998 The Phonology of Armenian OxfordOxford University Press

Vennemann Theo 1974 ldquoWords and syllables in Nat-ural Generative Grammarrdquo In A Bruck R Fox andM LaGaly (eds) Papers from the Parasession on NaturalPhonology 346ndash74 Chicago Chicago Linguistics Society

Wheeler Max 1979 Phonology of Catalan OxfordBlackwell

Whitley M Stanley 1978 Generative Phonology Work-book Madison University of Wisconsin Press

Whitney Arthur 1956 Teach Yourself Finnish KentHodder amp Stoughton

Zwicky Arnold 1973 ldquoThe analytic leap fromlsquoSome Xs are Ysrsquo to lsquoAll Xs are Ysrsquordquo Chicago LinguisticSociety 9 700ndash9

1974 ldquoTaking a false steprdquo Language 50 215ndash24

1975 ldquoThe strategy of generative phonologyrdquo InW Dressler and F V Mareš (eds) Phonologica 1972151ndash68 Munich Fink

References 333

Index of languagesAkan 212Amharic 35Angas 300Arabela 29Arabic (Bedouin) 273ndash8Arabic (Maltese) 260ndash7Arabic (Palestinian) 229Arabic (Syrian) 217Aramaic 306ndash9Armenian (Kirzan) 38Armenian (New Julfa) 132ndash3Axininca Campa 107 219 227

Bukusu 121ndash7

Catalan 144ndash5Chamorro 93Chukchi 215 222

Digo 305

Efik 213English 16ndash22 90ndash1 102ndash3 250ndash4Estonian 111Evenki 217Ewe (Anlo) 294ndash6

Farsi 34Finnish 85ndash8 145Fore 163ndash70Fula 220 225 306

Gatilde 213 296ndash300Ganda 220Gen 35Greek (Modern) 34Guerze (Guinean Kpelle) 143

Hebrew (Modern) 170ndash5Hehe 155ndash63Hindi 10 22Holoholo 320Hungarian 109 254

Japanese 176ndash85 219Jita 92ndash3

Kamba 118ndash21Karok 135ndash6Keley-i 193ndash4Kenyang 28ndash9Kera 192Kerewe 88ndash90Kikuyu 304ndash5

Kipsigis 25ndash6Kirghiz 210Klamath 137ndash9 214Koasati 94Kolami 214Korean 93ndash4 108 216 223 228Koromfe 202Kotoko 226Kuria 34 108ndash9 195ndash6 211ndash12 221

Lamba 128ndash30Lardil 196Latin 189ndash90 221Lezgian 110Lithuanian 130ndash2Logoori 141Lulubo 320

Makonde 225 228Manipuri 221Margyi 302Matuumbi 26ndash7 95 127ndash8 212 214ndash16 219 271ndash3Mbunga 139Mende 294Mixtec 78 296Mohawk 30ndash1 226Mongo 133ndash5 291ndash2Mongolian 210

Nenets 214Nkore 303Norwegian 217

Osage 35Ossetic 31ndash2

Palauan 36 101ndash2Polish 140

Quechua (Cuzco) 37

Russian 80ndash3 218ndash19 229

Saami (North) 112 224 226ndash7Sakha 197 210Samoan 98ndash101Sanskrit 217 248ndash50Serbo-Croatian 186Shambaa 36 320Shona 136ndash7 142ndash3 228 292ndash3Slave 227Slovak 280ndash1Somali 188Spanish 219

Sundanese 29ndash30 221Swati 234

Tera 268ndash71Thai 36Tibetan 37 112 219Tiv 300ndash2Tohono Orsquoodham 23ndash5Tonkawa 103ndash7Tswana 22ndash3Turkish 190 209

Ukrainian (Sadzhava) 200ndash1Ukrainian (Standard) 187Urhobo 281ndash3

Vata 306Votic 116ndash18

Wintu 223Woleiaian 223

Xavante 108

Yawelmani 210 229 257ndash60Yekhee 287 290Yiddish 255ndash7Yoruba 303

Zapotec (Isthmus) 110Zoque 109

336 Index of languages

General indexabsolute neutralization 257ndash67accuracy 2ndash3across the board effects 292ndash3affricate 8 56allophone 16alternation 18ndash22 81archiphoneme 210assimilation 29 122 131ndash3 172 183

208ndash20association line 288autosegmental phonology 286ndash313

central vowels 74ndash6citation forms 95ndash8compensatory lengthening 119 224 265complementary distribution 16consonants

features 52ndash5 59ndash60phonetic properties 8secondary articulation 53ndash5

contrast 16

determinant 68dissimilation 220ndash3

ejectives 56environment (rule) 68evidence grammar-external 254ndash7 278explanation 230ndash5

feature distinctive 45ndash76feature geometry 309ndash13features

advanced tongue root 50back 50consonantal 47constricted glottis 56continuant 55delayed release 55flat 71high 50labial 73ndash4lateral 55long 57low 50nasal 55round 50sonorant 47spread glottis 56stress 57syllabic 47tense 50voice 56

focus (rule) 68formant 4

free variation 32functional explanation 233ndash5

grammar 2 5 11 31 33 83ndash4 121

hardening 219historical change 254ndash7 268ndash73homorganic 132

implicational relation 206implosives 56inventory 9ndash10 206ndash8

language acquisition 239 255ndash7language games 273ndash8lenition 219liquid 8 48

major class 47ndash50manner of articulation 55ndash6markedness 206minimal pair 16mirror-image notation 68mora 229morpheme 16 20 26 81 84morphology 84ndash5

natural class 44ndash5 49 61ndash4neutralization 80 88 223ndash4neutralization (absolute) 254

obstruent 8optionality 32ndash3 127

phoneme 16phonetic detail 4 41ndash2 50phonetics 2ndash5possible phoneme 64ndash5possible rule 231ndash2predictability 16prediction 42ndash3 45 64ndash5 67 230ndash5privative 313prosody 224ndash30

reversal of sound change 255rule ordering 115ndash39rules 10ndash11 18 82

formalization 67ndash71

segment 5 16 18 40simplicity 25ndash6 29 32 65ndash7sonorant 8spectrogram 3spontaneous voicing 47stress 229ndash30

structural change 68structure preservation 90 95 123syllable 18 313ndash19symbol

mental 5 9transcription 5ndash8

target 68tone 286ndash305

contours 286ndash90default 302ndash3floating 294ndash300melodies 294mobility 303ndash5morphemes 300ndash2

stability 290ndash2trigger 68Twin Sister Convention 291

underlying form 17 80ndash93 95ndash8

vocal fold vibration 46ndash7vocoid 21vowel harmony 133 208ndash14vowels

features 50ndash2 59phonetic properties 6

weakening 219Well-Formedness Condition 290

338 General index

Cover13

Half-title13

Series-page13

Title13

Imprints13

Contents

About this book

Acknowledgments

A note on languages

Abbreviations

Chapter 1 What is phonology

11 Phonetics - the manifestation of language sound

12 Phonology the symbolic perspective on sound

121 Symbolic representation of segments

122 The concerns of phonology

The sounds of a language

Rules for combining sounds

Variations in pronunciation

Summary

Exercises

Further reading

Chapter 2 Allophonic relations


211 Aspiration

Alternations involving aspiration

Pronunciation of novel utterances

212 Flapping

213 Glottal stop


l and d in Tswana

Tohono Oodham affricates

Obstruent voicing in Kipsigis

Implosive and plain voiced stops in Matuumbi

Velar and uvular stops in Kenyang

Arabela nasalization

Sundanese a problem for the student to solve

Vowel length in Mohawk

Aspiration in Ossetic

Optional rules

Summary

Exercises

1 Kuria

2 Modern Greek

3 Farsi

4 Osage

5 Amharic

6 Gen

7 Shambaa

8 Thai

9 Palauan

10 Quechua (Cuzco dialect)

11 Lhasa Tibetan

12 Kirzan Armenian

Further reading

Chapter 3 Feature theory


311 Possible differences in sounds

Varieties of phonetic [i] vs []

Other variants of sounds

The important details of speech

Predictions versus observations

312 Possible rules


321 Phonetic preliminaries

322 Major class features

323 Place of articulation

Vowel place features

Consonant place features

Vowel features on consonants

324 Manner of articulation

325 Laryngeal features

326 Prosodic features

327 Summary of feature values

Vowel feature summary

Consonant feature summary


34 Possible phonemes and rules - an answer

341 Possible phonemes

342 Rule formulation and features

Simplicity in rule writing

Formalizability



The case for labial

Feature redefinition

Central vowels

Summary

Exercises

Further reading

Chapter 4 Underlying representations




Kerewe

English plurals

Jita tone


Chamorro vowel alternations

Korean

Koasati

Matuumbi

45 Underlying forms and sentence-level phonology

451 Korean final Cs

452 Matuumbi tone

46 Underlying forms and multiple columns in the paradigm

Palauan

English

Tonkawa reaching the analysis step-by-step

Summary

Exercises

1 Axininca Campa

2 Xavante

3 Kuria I

4 Korean

5 Zoque

6 Hungarian

7 Kuria II

8 Isthmus Zapotec

9 Lezgian

10 Estonian

11 Tibetan

12 North Saami

Further reading

Chapter 5 Interacting processes


511 Votic palatalization and raisingfronting

512 Kamba palatalization and glide formation

513 Bukusu nasal+consonant combinations

Nasal Place Assimilation and Post-Nasal Voicing

Post-Nasal Hardening

l-deletion

Nasal Cluster Simplification

Nasal Deletion

Summary

514 Matuumbi


521 Lamba harmony and palatalization

522 Voicing and epenthesis

Lithuanian

Armenian

523 Mongo B-deletion and resolution of vowel hiatus

524 Examples for discussion

Karok

Shona

Klamath

Summary

Exercises

1 Kerewe

2 Mbunga

3 Polish

4 Logoori

5 Shona

6 Guerze

7 Catalan

8 Finnish

Further reading

Chapter 6 Doing an analysis

61 Yawelmani

611 The data

612 The first step morphology

Stem variants

Suffix variants

613 Identifying phonological regularities

Vowel harmony

Vowel shortening

Epenthesis

614 Evaluating alternatives

62 Hehe

621 The data

622 Morphological analysis

623 Phonological alternations

Glide formation versus vowel deletion

v-rounding

Front vowels and glides

Checking other classes discovering a palatalization rule

Deciding on the form of w-deletion degemination

624 Extending the data

The morphology

Phonological rules

63 Fore

631 The data



634 Alternative analysis

64 Modern Hebrew

641 The data



Voicing assimilation

Alternations in V2

Closed syllable lowering

Stems with final pharyngeals and laryngeals

Metathesis

65 Japanese

651 The data


Changing our hypothesis

653 Phonological rules

The glide in the inchoative

Vowel deletion

Nasal + consonant

654 Taking stock

Correcting the final consonant

i-epenthesis

r-assimilation and final w

Progress by hypothesis forming and testing

Summary

Exercises

1 Serbo-Croatian

2 Standard Ukrainian

3 Somali

4 Latin

5 Turkish

6 Kera

7 Keley-i

8 Kuria

9 Lardil

10 Sakha (Yakut)

11 Sadzhava Ukrainian

12 Koromfe

Further reading

Chapter 7 Phonological typology and naturalness

71 Inventories


721 Assimilations

Vowel harmony

Consonant assimilations

722 Dissimilation

723 Other segmental processes


Vowel sequences

Vowel epenthesis

Onset creation

Cluster reduction

Stress lengthening and reduction

Syllable weight limits

Stress patterns


Impossible rules

Unlikely rules

Summary

Further reading

Chapter 8 Abstractness and psychological reality


811 Limiting possible analyses

Mental reality and language acquisition

Abstractness and phonemic representations

812 A principled limit on abstractness

813 Case studies in abstract analysis

Abstract mu in Matuumbi

Abstract ai and au in Sanskrit

Abstractness in English

What constitutes a valid motivation

Word-relatedness

82 Independent evidence historical restructuring

821 Yiddish final devoicing

822 Historical evidence and the treatment of absolute neutralization


831 Yawelmani u

832 Maltese

8321 Basic Maltese phonology

Stress and apocope

Unstressed reduction and harmony

Epenthesis

Regressive harmony and precoronal fronting

Guttural lowering

Metathesis

Stems with long vowels

8322 Apparent irregularities

84 Grammar-external evidence for abstractness

841 Abstract analysis and historical change Tera

The synchronic argument

The diachronic argument

842 Abstract reanalysis in Matuumbi NC sequences

843 Language games and Bedouin Arabic

8431 Regular language phonology

8432 Language game evidence


Exercises

1 Slovak

Urhobo

Further reading

Chapter 9 Nonlinear representations

91 The autosegmental theory of tone the beginnings of a change

911 The problem of contours

912 Autosegmental contours

913 Tone preservation

914 Across-the-board effects

915 Melodic patterns

916 Floating tones

Anlo tone

Mixtec

Gatilde

917 Tonal morphemes

918 Toneless vowels

919 Tonal mobility


921 The autonomy of all features

Aramaic CP

922 Feature geometry


Possible consonant clusters

Phonological rules

r-unrounding

Vowel reduction

Other phenomena referring to the syllable

Other suprasegmental units

Summary

Exercises

1 Lulubo

2 Shambaa

3 Holoholo

Further reading

Glossary

References

Index of languages

General index

Cambridge Introductions to Language and LinguisticsThis new textbook series provides students and their teachers with accessible introductions to the majorsubjects encountered within the study of language and linguistics Assuming no prior knowledge of thesubject each book is written and designed for ease of use in the classroom or seminar and is ideal foradoption on a modular course as the core recommended textbook Each book offers the ideal introductorymaterial for each subject presenting students with an overview of the main topics encountered in theircourse and features a glossary of useful terms chapter previews and summaries suggestions for furtherreading and helpful exercises Each book is accompanied by a supporting website

Books published in the seriesIntroducing Phonology David OddenIntroducing Speech and Language Processing John ColemanIntroducing Phonetic Science Michael Ashby and John MaidmentIntroducing Second Language Acquisition second edition Muriel Saville-TroikeIntroducing English Linguistics Charles F MeyerIntroducing Morphology Rochelle LieberIntroducing Semantics Nick RiemerIntroducing Language Typology Edith A MoravcsikIntroducing Psycholinguistics Paul WarrenIntroducing Phonology second edition David Odden

DAVID ODDEN

Summary 11

Exercises 12

Summary 33

Exercises 34

3 Feature theory 39

Summary 77

Exercises 77

Summary 107

Exercises 107

Summary 139

Exercises 139

61 Yawelmani 148

62 Hehe 155

63 Fore 163

65 Japanese 176

Summary 185

Exercises 186

71 Inventories 206

Summary 235

Exercises 279

of a change 286

Summary 319

Exercises 319

Glossary 323

References 329

General index 337

vi Contents

CHAPTER

1 What isphonology

PREVIEW

representing sound

KEY TERMSsound

symbol

transcription

grammar

Amp l i t u d e

FIGURE 1

Amp l i t u d e

FIGURE 2

What is phonology 3

Frequency

1080 msc 0

willFIGURE 3

FIGURE 4

1370 msc0Time

5000Frequency

What is phonology 5

FIGURE 6

What is phonology 7

Nonround

FIGURE 7

Consonant symbols

vclsstop

vclsaffricate

vclsfricative

vcdstop

vcdaffricate

vcdfricative nasal

ʔ h ɦ

Glides and liquids

Glides w j ɥ ɰ

What is phonology 9

1983 Stevens 1998

x ɪbv gw

gɣ ʔ

CHAPTER

PREVIEW

You will

other languages

KEY TERMSallophone

phoneme

contrast

distinctive

nonhigh V

themselvesrsquo

C implosive V

rarr⎧⎨⎩

⎫⎬⎭

front Vcentral V

high back VC

q k rarr

⎧⎪⎪⎨⎪⎪⎩

⎫⎪⎪⎬⎪⎪⎭

(38) ʃ s optional

CHAPTER

3 Feature theory

PREVIEW

universal features

properties

predictions

features

natural classes

Feature theory 41

Feature theory 43

Feature theory 45

Feature theory 47

Feature theory 49

Feature theory 51

Feature theory 53

Feature theory 55

t n ts s l l tl

voice - + + - - +cg - - + + - -sg - - - - + +

Feature theory 57

Feature theory 59

11 glottallaryngeal

+sylfrac12 +syl-nas

+syl+rd

+syl+high-nas

35 +syl

-high-lo-nas-tense

266664

377775

Feature theory 61

Feature theory 63

Feature theory 65

+rarr⎥⎥⎦

⎢⎢⎣

⎥⎥⎥

⎢⎢⎢

(30) [ ] _+syl+high

⎤⎥⎦

+nasal-cont

⎡⎢⎣

⎤⎥⎦

rarr+syl+high-low

⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+syl+high-low+nasal

⎡⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎦

_⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+nasal-cont-low

(32)+nasal-cont

⎡⎢⎣

⎤⎥⎦

(33) αFiβFj

γFkδFl

εFmζFn

ηFoθFp

Trigger

Feature theory 67

(34) [+syl] [+long] _ [+son]

(35) [-son] [-voice] _

(36) [-son] [+voice] _ [+son]

(37) +syl-high-long

⎤⎥⎥⎥⎦

(39) [ ]_][ ]+syl rarr [ -long -syl⎧⎨⎩

⎫⎬⎭

Feature theory 69

(41)[+syl] rarr [+rd] ⎥⎦

(42) +syl[ ] +rd[ ] __-syl[ ]0

⎥⎤⎦⎢⎣

⎡ +syl+rd

eacuteecirceuml

ugraveuacuteucirc

eacuteecirceuml

ugraveuacuteucirc

+ant+cor

[ ] _+nas+ant-cor

eacuteecirceuml

ugraveuacuteucirc

eacuteecirceuml

ugraveuacuteucirc

+ant-cor

(44) __ rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

αantβcor

Feature theory 71

_rarr rarr⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎥⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎡⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎦

⎡⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎦

+ant-cor+voi

+ant-cor-high-rd

ba C-cons+high+rd

(48) [ ] __rarr⎡⎢⎣

⎤⎥⎦

+labial+voi

ndashcons Ca

[ ] __rarr⎡⎢⎣

⎤⎥⎦

+labial-cons

+cons [ ]+nasalb

rarr⎡⎢⎣

⎤⎥⎦

+labial+rd

⎡⎢⎣

⎤⎥⎦

+cons-rd

Feature theory 73

(50) [ ] __rarr⎡⎢⎣

⎤⎥⎦

+high+back

⎡⎢⎣

⎤⎥⎦

+back+high

(51)[ ] rarr

⎡⎢⎣

⎤⎥⎦

+high+back

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

Feature theory 75

vi i ε a o ɔ u m

Feature theory 77

Trubetzkoy 1939

CHAPTER

PREVIEW

forms by

the underlying form

neutralization

predictability

(2) C [+voice] V_ V

Underlying

representation

Surface

representation

Physical

output

_ [ ]+high

eacuteecirceuml

ugraveuacuteucirc

rarr⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+sg+contminusant

⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

⎧⎨⎩

⎫⎬⎭

(17) a

(23) a

(24) a

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+syl-back

Oslash

rarr⎡⎢⎣

⎤⎥⎦

+syl+back

Oslash ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]+low

C Oslash _

-high-lo+back-rd

+syl-stress

eacuteecirceuml

ugraveuacuteucirc

V Oslash _ V

(46) V Oslash CVC _

makes noisersquo

CHAPTER

PREVIEW

alternation

phonological rules

surface form

ordering

(3) Palatalization

+syl-rd-lo

rarr __⎡⎢⎣

⎤⎥⎦

+high-back

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+cons+back

+syl-back

[ ]+cor

(6) Palatalization

⎡⎢⎣

⎤⎥⎦

+cons+back ⎢

⎡⎢

⎢⎥⎦

⎤⎥

+syl-cons-back

[ ]+corrarr Oslash

(9) Glide Formation

(10) Glide Deletion

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

-syl+round

+round+tense

Oslash

rarr⎡⎢⎣

⎤⎥⎦

αantβcor ⎢

⎡⎢

⎢⎥⎦

⎤⎥

+sylαantβcor

[ ]+nas

⎡⎢⎣

⎤⎥⎦

(25) Nasal Deletion

⎡⎢⎣

⎤⎥⎦

(25) Nasal Deletion

(21) l-Deletion

__rarr⎡⎢⎣

⎤⎥⎦

+ cont- voice

Oslash[ ]+nasal

rarr⎡⎢⎣

⎤⎥⎦

αantβcor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+sylαantβcor

[ ]+nas

⎤⎥⎦

(31) Height harmony

rarr ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl-high-low

rarr⎡⎢⎣

⎤⎥⎦

+high- syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]+cor

(35) a

rarr⎡⎢⎣

⎤⎥⎦

+cont+cor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]- ant

⎡⎢⎣

⎤⎥⎦

(38) Epenthesis

(39) a

rarrOslash⎡⎢⎣

⎤⎥⎦

-cont-sonαantβcor

⎡⎢⎣

⎤⎥⎦

-cont-sonαantβcor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

(45) Affrication

(47) Labial elision

rarr ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

-syl-cons-back

[ ]+nas VOslash +

⎤⎥⎦

⎡⎢

⎢⎥⎦

⎤⎥

+voice+ant-cor

(48) a

(distributive)rsquo

(rem fut)rsquo

CHAPTER

6 Doing ananalysis

PREVIEW

competinghypotheses

61 Yawelmani

mdashCO

⎡⎢⎣

⎤⎥⎦

Vαhigh ⎢

⎡⎢

⎢⎥⎦

⎤⎥

⎥αhigh+round

[ ]+ roundrarr V

(7) V [-long] __ CC

62 Hehe

v v[ ]-sylrarr

u-deletion

⎥⎥⎥

⎢⎢⎢

C+round-syl

⎡⎢⎣

⎤⎥⎦

[ ]+roundrarrOslash

⎢⎢⎢

⎡⎢⎣

⎤⎥⎦

+ high- back

⎡⎢

⎢⎥⎦

⎤⎥

+ cons+ back- voice

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

- syl- cons- back

⎤⎥⎦

- syl+ rd

⎡⎢⎣

⎤⎥⎦

- syl- rd

Oslash

63 Fore

Vowel Harmony

⎡⎢⎣

⎤⎥⎦

-high-low

[ ]+ H[ ]-Hrarr V C0__V

⎤⎥⎦

+ voice- nas

Voicing

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎡⎢⎣

⎤⎥⎦

+ rdminus syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎤⎥⎦

+ rdminus syl

k-deletion

⎡⎢⎣

⎤⎥⎦

+ corminus voice

(41) a

rarr ⎡⎢⎣

⎤⎥⎦

+ voice- nas

⎡⎢⎣

⎤⎥⎦

+ rd- syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+ high+ cons- voice

[ ]-cont

Voicing

Stop Deletion

k-insertion

Denasalization

__rarr

Oslash

[ ]-voice

[ ]-nas

[ ]+nas

[ ]+nas[ ]+nas

64 Modern Hebrew

⎢⎢⎢

⎡⎢⎣

⎤⎥⎦

+ rd- syl[ ]-nas

⎡⎢⎣

⎤⎥⎦

-son+voi

(48) e a _ CC

(49) CiCi Ci

⎡⎢⎣

⎤⎥⎦

+cor-cont

⎡⎢⎣

⎤⎥⎦

+cor+strid

⎡⎢⎣

⎤⎥⎦

+cor-cont

⎡⎢⎣

⎤⎥⎦

+cor+strid

65 Japanese

⎡⎢⎣

⎤⎥⎦

minusconsminusback

rarr Oslash C_

⎡⎢⎣

⎤⎥⎦

+syl-back

rarr⎡⎢⎣

⎤⎥⎦

+high[ [

-back[ [

(71) r Ci _ Ci

⎡⎢⎣

⎤⎥⎦

⎢⎢⎢

+ cont- son

⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+ cor- cont

⎤⎥⎦

+syl-low

CHAPTER

PREVIEW

KEY TERMStypology

markedness

71 Inventories

⎡⎢⎣

⎤⎥⎦

-highαround

[- high]

(13) a

kέεŋgεεmb-a

⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

αhighβtense

rarr C0 _ ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+lowαhighβtense

ε rarr C0 ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

(20) a

(22) a

(23) a

(24) a

(42) a

continuouslyrsquo

(71) a

CHAPTER

PREVIEW

analysisrdquo

external evidence

(47) _o -back[ [

rarr⎡⎢⎣

⎤⎥⎦

+ cor- son

(55) aa io _ CC

(70) ə i _

(75) ə Oslash _ V

ə Oslash V C _

[- voice]

(83) a i _ C i

songrsquo

(me)rsquo

V-edrsquo

lsquoI am still

V-ingrsquo

lsquoshe is still

V-ingrsquo

lsquoI have

V-ed himrsquo

(me)rsquo

V-edrsquo

lsquoI am still

V-ingrsquo

lsquoshe is still

V-ingrsquo

lsquoI have

V-ed himrsquo

CHAPTER

PREVIEW

tone stability

floating tone

feature geometry

syllable

(1)⎡⎢⎣

⎤⎥⎦

+H-contour

⎡⎢⎣

⎤⎥⎦

-H-contour

⎡⎢⎣

⎤⎥⎦

-H+contour

⎡⎢⎣

⎤⎥⎦

+H+contour

(3) c [+H] [+contour]_[-H]

(7)⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

= =a agraveaacute

(8) agrave = H

a = L H

acirc = H L

(9) (H) L H

(H) L H

(10) L H

(12) H L (=2b))

H L (=2c))

L H (=2d))

(13) L (H)

(15) L H

o akp o k

o o aw

(16) LH L

(17) H HL L

e k e l a

LH LH HL L LH L

m m e mb e mb e e

singrsquo

(24) H L

(25) L L

benzibvunza

mbundudzi

(28) H H

se-che-mbunduzi

(29) HH H

se-ne-che-mbundudzi

H LH H

se-ne-che-mbundudzi

(31) LHL L H L

nikili nyaha

(33) ML_ ML_L

(35) MSHH_H HSH SH_

(41) MM MMH L H H

m a alsquothatrsquo

(48) H L L

e - wo

(50) H

e hulu-

(54) L

ngohor

(56) H L

Oslash

(57) H LL

(59) H H

ta + na

ndal + naba+el

(62) T T

(64)xH

(65) a

(67) a

(68) Hx

(69) H H

(70) a

(71) Hx

(78) [+CP]

(79) [+CP]

(80) [+CP] [+CP] [+CP]

(rule 79) (rule 78)

(84) [+CP] [+CP] [+CP]

(87) [+CP] [+CP] [+CP]

mdashmdash

lateral

Laryngeal Place

sonorantstrident

Coronal

anteriordistributed

Labial

roundlow

Dorsal

voiceconstricted

glottis

spreadglottis

mdashmdash

(93) C

mdashmdash

adistributedanasal

acoronalaanterior

avoicealateral

(95) root root

Place Place

Labial

x[nasal]

+distrib-anterior

Coronal

k w a r t s

(98) s

[+cor-cons]

V [-long] C

324 Glossary

Glossary 325

326 Glossary

Glossary 327

330 References

References 331

332 References

References 333

Bukusu 121ndash7

Digo 305

Quechua (Cuzco) 37

Xavante 108

contrast 16

grammar 2 5 11 31 33 83ndash4 121

338 General index

Cover13

Half-title13

Series-page13

Title13

Imprints13

Contents

About this book

Acknowledgments

A note on languages

Abbreviations








Summary

Exercises

Further reading



211 Aspiration



212 Flapping

213 Glottal stop


l and d in Tswana









Optional rules

Summary

Exercises

1 Kuria

2 Modern Greek

3 Farsi

4 Osage

5 Amharic

6 Gen

7 Shambaa

8 Thai

9 Palauan


11 Lhasa Tibetan

12 Kirzan Armenian

Further reading








312 Possible rules



















Formalizability



The case for labial


Central vowels

Summary

Exercises

Further reading





Kerewe

English plurals

Jita tone



Korean

Koasati

Matuumbi


451 Korean final Cs

452 Matuumbi tone


Palauan

English


Summary

Exercises

1 Axininca Campa

2 Xavante

3 Kuria I

4 Korean

5 Zoque

6 Hungarian

7 Kuria II

8 Isthmus Zapotec

9 Lezgian

10 Estonian

11 Tibetan

12 North Saami

Further reading








l-deletion


Nasal Deletion

Summary

514 Matuumbi




Lithuanian

Armenian



Karok

Shona

Klamath

Summary

Exercises

1 Kerewe

2 Mbunga

3 Polish

4 Logoori

5 Shona

6 Guerze

7 Catalan

8 Finnish

Further reading


61 Yawelmani

611 The data


Stem variants

Suffix variants


Vowel harmony

Vowel shortening

Epenthesis


62 Hehe

621 The data




v-rounding





The morphology

Phonological rules

63 Fore

631 The data




64 Modern Hebrew

641 The data




Alternations in V2



Metathesis

65 Japanese

651 The data





Vowel deletion

Nasal + consonant

654 Taking stock


i-epenthesis



Summary

Exercises

1 Serbo-Croatian


3 Somali

4 Latin

5 Turkish

6 Kera

7 Keley-i

8 Kuria

9 Lardil

10 Sakha (Yakut)


12 Koromfe

Further reading


71 Inventories


721 Assimilations

Vowel harmony


722 Dissimilation



Vowel sequences

Vowel epenthesis

Onset creation

Cluster reduction



Stress patterns


Impossible rules

Unlikely rules

Summary

Further reading












Word-relatedness





831 Yawelmani u

832 Maltese


Stress and apocope


Epenthesis


Guttural lowering

Metathesis












Exercises

1 Slovak

Urhobo

Further reading








916 Floating tones

Anlo tone

Mixtec

Gatilde

917 Tonal morphemes

918 Toneless vowels

919 Tonal mobility



Aramaic CP




Phonological rules

r-unrounding

Vowel reduction



Summary

Exercises

1 Lulubo

2 Shambaa

3 Holoholo

Further reading

Glossary

References

Index of languages

General index

DAVID ODDEN

Summary 11

Exercises 12

Summary 33

Exercises 34

3 Feature theory 39

Summary 77

Exercises 77

Summary 107

Exercises 107

Summary 139

Exercises 139

61 Yawelmani 148

62 Hehe 155

63 Fore 163

65 Japanese 176

Summary 185

Exercises 186

71 Inventories 206

Summary 235

Exercises 279

of a change 286

Summary 319

Exercises 319

Glossary 323

References 329

General index 337

vi Contents

CHAPTER

1 What isphonology

PREVIEW

representing sound

KEY TERMSsound

symbol

transcription

grammar

Amp l i t u d e

FIGURE 1

Amp l i t u d e

FIGURE 2

What is phonology 3

Frequency

1080 msc 0

willFIGURE 3

FIGURE 4

1370 msc0Time

5000Frequency

What is phonology 5

FIGURE 6

What is phonology 7

Nonround

FIGURE 7

Consonant symbols

vclsstop

vclsaffricate

vclsfricative

vcdstop

vcdaffricate

vcdfricative nasal

ʔ h ɦ

Glides and liquids

Glides w j ɥ ɰ

What is phonology 9

1983 Stevens 1998

x ɪbv gw

gɣ ʔ

CHAPTER

PREVIEW

You will

other languages

KEY TERMSallophone

phoneme

contrast

distinctive

nonhigh V

themselvesrsquo

C implosive V

rarr⎧⎨⎩

⎫⎬⎭

front Vcentral V

high back VC

q k rarr

⎧⎪⎪⎨⎪⎪⎩

⎫⎪⎪⎬⎪⎪⎭

(38) ʃ s optional

CHAPTER

3 Feature theory

PREVIEW

universal features

properties

predictions

features

natural classes

Feature theory 41

Feature theory 43

Feature theory 45

Feature theory 47

Feature theory 49

Feature theory 51

Feature theory 53

Feature theory 55

t n ts s l l tl

voice - + + - - +cg - - + + - -sg - - - - + +

Feature theory 57

Feature theory 59

11 glottallaryngeal

+sylfrac12 +syl-nas

+syl+rd

+syl+high-nas

35 +syl

-high-lo-nas-tense

266664

377775

Feature theory 61

Feature theory 63

Feature theory 65

+rarr⎥⎥⎦

⎢⎢⎣

⎥⎥⎥

⎢⎢⎢

(30) [ ] _+syl+high

⎤⎥⎦

+nasal-cont

⎡⎢⎣

⎤⎥⎦

rarr+syl+high-low

⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+syl+high-low+nasal

⎡⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎦

_⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+nasal-cont-low

(32)+nasal-cont

⎡⎢⎣

⎤⎥⎦

(33) αFiβFj

γFkδFl

εFmζFn

ηFoθFp

Trigger

Feature theory 67

(34) [+syl] [+long] _ [+son]

(35) [-son] [-voice] _

(36) [-son] [+voice] _ [+son]

(37) +syl-high-long

⎤⎥⎥⎥⎦

(39) [ ]_][ ]+syl rarr [ -long -syl⎧⎨⎩

⎫⎬⎭

Feature theory 69

(41)[+syl] rarr [+rd] ⎥⎦

(42) +syl[ ] +rd[ ] __-syl[ ]0

⎥⎤⎦⎢⎣

⎡ +syl+rd

eacuteecirceuml

ugraveuacuteucirc

eacuteecirceuml

ugraveuacuteucirc

+ant+cor

[ ] _+nas+ant-cor

eacuteecirceuml

ugraveuacuteucirc

eacuteecirceuml

ugraveuacuteucirc

+ant-cor

(44) __ rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

αantβcor

Feature theory 71

_rarr rarr⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎥⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎡⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎦

⎡⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎦

+ant-cor+voi

+ant-cor-high-rd

ba C-cons+high+rd

(48) [ ] __rarr⎡⎢⎣

⎤⎥⎦

+labial+voi

ndashcons Ca

[ ] __rarr⎡⎢⎣

⎤⎥⎦

+labial-cons

+cons [ ]+nasalb

rarr⎡⎢⎣

⎤⎥⎦

+labial+rd

⎡⎢⎣

⎤⎥⎦

+cons-rd

Feature theory 73

(50) [ ] __rarr⎡⎢⎣

⎤⎥⎦

+high+back

⎡⎢⎣

⎤⎥⎦

+back+high

(51)[ ] rarr

⎡⎢⎣

⎤⎥⎦

+high+back

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

Feature theory 75

vi i ε a o ɔ u m

Feature theory 77

Trubetzkoy 1939

CHAPTER

PREVIEW

forms by

the underlying form

neutralization

predictability

(2) C [+voice] V_ V

Underlying

representation

Surface

representation

Physical

output

_ [ ]+high

eacuteecirceuml

ugraveuacuteucirc

rarr⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

+sg+contminusant

⎡⎢⎢⎢⎣

⎤⎥⎥⎥⎦

⎧⎨⎩

⎫⎬⎭

(17) a

(23) a

(24) a

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+syl-back

Oslash

rarr⎡⎢⎣

⎤⎥⎦

+syl+back

Oslash ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]+low

C Oslash _

-high-lo+back-rd

+syl-stress

eacuteecirceuml

ugraveuacuteucirc

V Oslash _ V

(46) V Oslash CVC _

makes noisersquo

CHAPTER

PREVIEW

alternation

phonological rules

surface form

ordering

(3) Palatalization

+syl-rd-lo

rarr __⎡⎢⎣

⎤⎥⎦

+high-back

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+cons+back

+syl-back

[ ]+cor

(6) Palatalization

⎡⎢⎣

⎤⎥⎦

+cons+back ⎢

⎡⎢

⎢⎥⎦

⎤⎥

+syl-cons-back

[ ]+corrarr Oslash

(9) Glide Formation

(10) Glide Deletion

__rarr⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

-syl+round

+round+tense

Oslash

rarr⎡⎢⎣

⎤⎥⎦

αantβcor ⎢

⎡⎢

⎢⎥⎦

⎤⎥

+sylαantβcor

[ ]+nas

⎡⎢⎣

⎤⎥⎦

(25) Nasal Deletion

⎡⎢⎣

⎤⎥⎦

(25) Nasal Deletion

(21) l-Deletion

__rarr⎡⎢⎣

⎤⎥⎦

+ cont- voice

Oslash[ ]+nasal

rarr⎡⎢⎣

⎤⎥⎦

αantβcor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+sylαantβcor

[ ]+nas

⎤⎥⎦

(31) Height harmony

rarr ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl-high-low

rarr⎡⎢⎣

⎤⎥⎦

+high- syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]+cor

(35) a

rarr⎡⎢⎣

⎤⎥⎦

+cont+cor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

[ ]- ant

⎡⎢⎣

⎤⎥⎦

(38) Epenthesis

(39) a

rarrOslash⎡⎢⎣

⎤⎥⎦

-cont-sonαantβcor

⎡⎢⎣

⎤⎥⎦

-cont-sonαantβcor

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+syl+high-back

(45) Affrication

(47) Labial elision

rarr ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

-syl-cons-back

[ ]+nas VOslash +

⎤⎥⎦

⎡⎢

⎢⎥⎦

⎤⎥

+voice+ant-cor

(48) a

(distributive)rsquo

(rem fut)rsquo

CHAPTER

6 Doing ananalysis

PREVIEW

competinghypotheses

61 Yawelmani

mdashCO

⎡⎢⎣

⎤⎥⎦

Vαhigh ⎢

⎡⎢

⎢⎥⎦

⎤⎥

⎥αhigh+round

[ ]+ roundrarr V

(7) V [-long] __ CC

62 Hehe

v v[ ]-sylrarr

u-deletion

⎥⎥⎥

⎢⎢⎢

C+round-syl

⎡⎢⎣

⎤⎥⎦

[ ]+roundrarrOslash

⎢⎢⎢

⎡⎢⎣

⎤⎥⎦

+ high- back

⎡⎢

⎢⎥⎦

⎤⎥

+ cons+ back- voice

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

- syl- cons- back

⎤⎥⎦

- syl+ rd

⎡⎢⎣

⎤⎥⎦

- syl- rd

Oslash

63 Fore

Vowel Harmony

⎡⎢⎣

⎤⎥⎦

-high-low

[ ]+ H[ ]-Hrarr V C0__V

⎤⎥⎦

+ voice- nas

Voicing

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎡⎢⎣

⎤⎥⎦

+ rdminus syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎤⎥⎦

+ rdminus syl

k-deletion

⎡⎢⎣

⎤⎥⎦

+ corminus voice

(41) a

rarr ⎡⎢⎣

⎤⎥⎦

+ voice- nas

⎡⎢⎣

⎤⎥⎦

+ rd- syl

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+ high+ cons- voice

[ ]-cont

Voicing

Stop Deletion

k-insertion

Denasalization

__rarr

Oslash

[ ]-voice

[ ]-nas

[ ]+nas

[ ]+nas[ ]+nas

64 Modern Hebrew

⎢⎢⎢

⎡⎢⎣

⎤⎥⎦

+ rd- syl[ ]-nas

⎡⎢⎣

⎤⎥⎦

-son+voi

(48) e a _ CC

(49) CiCi Ci

⎡⎢⎣

⎤⎥⎦

+cor-cont

⎡⎢⎣

⎤⎥⎦

+cor+strid

⎡⎢⎣

⎤⎥⎦

+cor-cont

⎡⎢⎣

⎤⎥⎦

+cor+strid

65 Japanese

⎡⎢⎣

⎤⎥⎦

minusconsminusback

rarr Oslash C_

⎡⎢⎣

⎤⎥⎦

+syl-back

rarr⎡⎢⎣

⎤⎥⎦

+high[ [

-back[ [

(71) r Ci _ Ci

⎡⎢⎣

⎤⎥⎦

⎢⎢⎢

+ cont- son

⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

+ cor- cont

⎤⎥⎦

+syl-low

CHAPTER

PREVIEW

KEY TERMStypology

markedness

71 Inventories

⎡⎢⎣

⎤⎥⎦

-highαround

[- high]

(13) a

kέεŋgεεmb-a

⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

αhighβtense

rarr C0 _ ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

+lowαhighβtense

ε rarr C0 ⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

⎢⎣

⎡⎢

⎢⎥⎦

⎤⎥

(20) a

(22) a

(23) a

(24) a

(42) a

continuouslyrsquo

(71) a

CHAPTER

PREVIEW

analysisrdquo

external evidence

(47) _o -back[ [

rarr⎡⎢⎣

⎤⎥⎦

+ cor- son

(55) aa io _ CC

(70) ə i _

(75) ə Oslash _ V

ə Oslash V C _

[- voice]

(83) a i _ C i

songrsquo

(me)rsquo

V-edrsquo

lsquoI am still

V-ingrsquo

lsquoshe is still

V-ingrsquo

lsquoI have

V-ed himrsquo

(me)rsquo

V-edrsquo

lsquoI am still

V-ingrsquo

lsquoshe is still

V-ingrsquo

lsquoI have

V-ed himrsquo

CHAPTER

PREVIEW

tone stability

floating tone

feature geometry

syllable

(1)⎡⎢⎣

⎤⎥⎦

+H-contour

⎡⎢⎣

⎤⎥⎦

-H-contour

⎡⎢⎣

⎤⎥⎦

-H+contour

⎡⎢⎣

⎤⎥⎦

+H+contour

(3) c [+H] [+contour]_[-H]

(7)⎡⎢⎣

⎤⎥⎦

⎡⎢⎣

⎤⎥⎦

= =a agraveaacute

(8) agrave = H

a = L H

acirc = H L

(9) (H) L H

(H) L H

(10) L H

(12) H L (=2b))

H L (=2c))

L H (=2d))

(13) L (H)

(15) L H

o akp o k

o o aw

(16) LH L

(17) H HL L

e k e l a

LH LH HL L LH L

m m e mb e mb e e

singrsquo

(24) H L

(25) L L

benzibvunza

mbundudzi

(28) H H

se-che-mbunduzi

(29) HH H

se-ne-che-mbundudzi

H LH H

se-ne-che-mbundudzi

(31) LHL L H L

nikili nyaha

(33) ML_ ML_L

(35) MSHH_H HSH SH_

(41) MM MMH L H H

m a alsquothatrsquo

(48) H L L

e - wo

(50) H

e hulu-

(54) L

ngohor

(56) H L

Oslash

(57) H LL

(59) H H

ta + na

ndal + naba+el

(62) T T

(64)xH

(65) a

(67) a

(68) Hx

(69) H H

(70) a

(71) Hx

(78) [+CP]

(79) [+CP]

(80) [+CP] [+CP] [+CP]

(rule 79) (rule 78)

(84) [+CP] [+CP] [+CP]

(87) [+CP] [+CP] [+CP]

mdashmdash

lateral

Laryngeal Place

sonorantstrident

Coronal

anteriordistributed

Labial

roundlow

Dorsal

voiceconstricted

glottis

spreadglottis

mdashmdash

(93) C

mdashmdash

adistributedanasal

acoronalaanterior

avoicealateral

(95) root root

Place Place

Labial

x[nasal]

+distrib-anterior

Coronal

k w a r t s

(98) s

[+cor-cons]

V [-long] C

324 Glossary

Glossary 325

326 Glossary

Glossary 327

330 References

References 331

332 References

References 333

Bukusu 121ndash7

Digo 305

Quechua (Cuzco) 37

Xavante 108

contrast 16

grammar 2 5 11 31 33 83ndash4 121

338 General index

Cover13

Half-title13

Series-page13

Title13

Imprints13

Contents

About this book

Acknowledgments

A note on languages

Abbreviations








Summary

Exercises

Further reading



211 Aspiration



212 Flapping

213 Glottal stop


l and d in Tswana









Optional rules

Summary

Exercises

1 Kuria

2 Modern Greek

3 Farsi

4 Osage

5 Amharic

6 Gen

7 Shambaa

8 Thai

9 Palauan


11 Lhasa Tibetan

12 Kirzan Armenian

Further reading








312 Possible rules



















Formalizability



The case for labial


Central vowels

Summary

Exercises

Further reading





Kerewe

English plurals

Jita tone



Korean

Koasati

Matuumbi


451 Korean final Cs

452 Matuumbi tone


Palauan

English


Summary

Exercises

1 Axininca Campa

2 Xavante

3 Kuria I

4 Korean

5 Zoque

6 Hungarian

7 Kuria II

8 Isthmus Zapotec

9 Lezgian

10 Estonian

11 Tibetan

12 North Saami

Further reading








l-deletion


Nasal Deletion

Summary

514 Matuumbi




Lithuanian

Armenian



Karok

Shona

Klamath

Summary

Exercises

1 Kerewe

2 Mbunga

3 Polish

4 Logoori

5 Shona

6 Guerze

7 Catalan

8 Finnish

Further reading


61 Yawelmani

611 The data


Stem variants

Suffix variants


Vowel harmony

Vowel shortening

Epenthesis


62 Hehe

621 The data




v-rounding





The morphology

Phonological rules

63 Fore

631 The data




64 Modern Hebrew

641 The data




Alternations in V2



Metathesis

65 Japanese

651 The data





Vowel deletion

Nasal + consonant

654 Taking stock


i-epenthesis



Summary

Exercises

1 Serbo-Croatian


3 Somali

4 Latin

5 Turkish

6 Kera

7 Keley-i

8 Kuria

9 Lardil

10 Sakha (Yakut)


12 Koromfe

Further reading


71 Inventories


721 Assimilations

Vowel harmony


722 Dissimilation



Vowel sequences

Vowel epenthesis

Onset creation

Cluster reduction



Stress patterns


Impossible rules

Unlikely rules

Summary

Further reading












Word-relatedness





831 Yawelmani u

832 Maltese


Stress and apocope


Epenthesis


Guttural lowering

Metathesis












Exercises

1 Slovak

Urhobo

Further reading








916 Floating tones

Anlo tone

Mixtec

Gatilde

917 Tonal morphemes

918 Toneless vowels

919 Tonal mobility



Aramaic CP




Phonological rules

r-unrounding

Vowel reduction



Summary

Exercises

1 Lulubo

2 Shambaa

3 Holoholo

Further reading

Glossary

References

Index of languages

General index

Documents

Introducing Phonology