Understanding Phonology

Understanding Phonology, Fourth Edition provides a clear, accessible and broad introduction to Phonology. Introducing basic concepts, it provides a comprehensive account of phonological issues such as segmental contrasts; syllables and moras; quan­tity, tone, intonation and stress; feature geometry; and prosodic constituent structure.

This new edition has been reorganized and revised with key features including:

• AbrandneweResourceatwww.routledge.com/9781138961425,whichcontainsa full answer key for all exercises, and audio recordings of illustrative examples;

• Illustrationsinlanguagesfromallsixcontinentsandallmajorlanguagefami­lies,includingArabic,Mandarin,Finnish,ZuluandHawaiian;

• Over 140 exercises to test understanding, including new exercises involvinglarger data sets;

• Revisedcoverageoftone,stressandopacityinOT.

Understanding Phonology is essential reading for students coming to this topic for the first time.

Carlos Gussenhoven is emeritus professor of general and experimental phonology at RadboudUniversityNijmegen,Netherlands.

Haike Jacobs isprofessorofFrenchlinguisticsatRadboudUniversityNijmegen,Netherlands.

Understanding Language Series

Series Editors:

Bernard Comrie,UniversityofCalifornia,SantaBarbara,USA Greville Corbett,SurreyMorphologyGroup,UniversityofSurrey,UK

The Understanding Language series provides approachable, yet authoritative,introductionstomajortopicsinlinguistics.Idealforstudentswithlittleornopriorknowledge of linguistics, each book carefully explains the basics, emphasizing understanding of the essential notions rather than arguing for a particular theoreti­cal position.

Othertitlesintheseries:

Understanding PragmaticsGunterSenft

Understanding Child Language AcquisitionCarolineRowland

Understanding Semantics, Second EditionSebastianLöbner

Understanding Syntax, Fourth EditionMaggieTallerman

Understanding PhoneticsPatriciaAshby

Understanding Morphology, Second EditionMartinHaspelmathandAndreaD.Sims

Understanding Language TestingDanDouglas

Understanding Second Language AcquisitionLourdesOrtega

Study Skills for LinguisticsJeanetteSakel

Understanding Language ChangeKateBurridgeandAlexanderBergs

Understanding Phonology, Fourth Edition CarlosGussenhovenandHaikeJacobs

Formoreinformationonanyofthesetitles,ortoorder,goto www.routledge.com/series/ULAN

Understanding

PhonologyFourth edition

Carlos Gussenhoven and Haike Jacobs

Fourtheditionpublished2017 byRoutledge 2ParkSquare,MiltonPark,Abingdon,OxonOX144RN

andbyRoutledge 711ThirdAvenue,NewYork,NY10017

Routledge is an imprint of the Taylor & Francis Group, an informa business

©2017CarlosGussenhovenandHaikeJacobs

TherightofCarlosGussenhovenandHaikeJacobstobeidentifiedasauthorsofthisworkhasbeenassertedbytheminaccordancewithsections77and78oftheCopyright,DesignsandPatentsAct 1988.

Allrightsreserved.Nopartofthisbookmaybereprintedorreproducedorutilisedinanyformorbyanyelectronic,mechanical,orothermeans,nowknownorhereafterinvented,includingphotocopyingand recording, or in any information storage or retrieval system, without permission in writing from the publishers.

Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe.

FirsteditionpublishedbyHodderArnold1998

ThirdeditionpublishedbyRoutledge2013

British Library Cataloguing-in-Publication Data AcataloguerecordforthisbookisavailablefromtheBritishLibrary

Library of Congress Cataloging-in-Publication Data Names:Gussenhoven,Carlos,1946-author.|Jacobs,Haike,1961-author. Title:Understandingphonology/CarlosGussenhovenandHaikeJacobs. Description:FourthEdition.|MiltonPark,Abingdon,Oxon;NewYork,NY: Routledge,[2017]|Series:Understandinglanguageseries|Includesbibliographical references and index. Identifiers:LCCN2016036736|ISBN9781138961418(hardback)|ISBN9781138961425(pbk.)| ISBN9781315267982(ebook) Subjects:LCSH:Grammar,Comparativeandgeneral—Phonology. Classification:LCCP217.G8672017|DDC414—dc23 LCrecordavailableat https://lccn.loc.gov/2016036736

ISBN:978-1-138-96141-8(hbk) ISBN:978-1-138-96142-5(pbk) ISBN:978-1-315-26798-2(ebk)

TypesetinMinionProandTimesNewRoman byApexCoVantage,LLC

VisittheeResource:www.routledge.com/9781138961425

Contents

Preface viAcknowledgements viiiThe IPA chart ix

1 Structures in languages 1

2 The production of speech 17

3 Some typology: sameness and difference 36

4 Making the form fit 48

5 Distinctive features 64

6 Describing places of articulation 76

7 Connecting underlying and surface representations 94

8 Levels of representation 109

9 Between the segment and the syllable 126

10 Tones 148

11 Word stress 174

12 Phonology above the word 200

13 Transparency and opacity with rules and constraints 215

14 Feature geometry 232

References 259Language index 273Subject index 281

Preface

Thereareabout7,000 languages in theworld today.Almostcertainly,no twoofthem have the same sound structure: they vary widely in the number of consonants and vowels they have, in their use of tonal contrasts, in their stress patterns, in the shapeoftheirsyllables,andsoon.Atthesametime,alltheselanguagesshowstrik­ing similarities in the way they structure their sound systems. Phonology is a thriv­ing field of linguistic research that strives to understand the structure behind these systems.Howdo these similaritiesarise?Andwhy,at thesame time, is there somuchvariation?Howisourknowledgeofthepronunciationofourlanguagerep­resentedinourbrain?Howcanwedescribethepronunciationofalanguage?Whatdopeopledowhentheyplaylanguagegames?Whydoloanwordsoftensoundsototallydifferentfromthewaytheyarepronouncedinthedonorlanguages?Theseand many other questions are dealt with in this book. In our discussion, we have tried to sketch the development of scientific thinking about the sound structure of languages and to take an unbiased view of the cognitive or physiological nature of theexplanations.Wehopewehavesucceededinthistaskinatleastsomeplacesin the book, and have got close enough to this ideal for it to serve as a reliable and relevant introduction to an important and exciting field.

Thisbookreflectsthethoughtsanddiscoveriesofmanyphonologists.Wehavelearnt to appreciate the value and implications of these theoretical positions not only by reading their publications but also by attending their classes and discussing theissueswiththem.Needlesstosay,ourdebttothemisinestimable.

Anumberofpeopledeservethanksforprovidinguswithdataforthefirstedi­tion(1998).TheseareAojuChen,FredericGaggeri,KeesGroenewoud,JadrankaGvozdanović, Özden Heebink-Mandaci, Eric Kellerman, Joost Kremers, AditiLahiri, Will Leben, Manjari Ohala, Michael Redford, Barbara van den Brekel,WilbervanderBeek,LeoWetzelsandYoung-meeYuCho.

Thesecondedition(2005)ofthisbookbenefitedfromthecommentsmadebyElanDresher,SanDuanmu,EdFlemming,BoHagström,VictoriaRosén,IngmarSteiner,RikvanGijn,JeroenvandeWeijer,LeoWetzels,MariaWoltersandthosewho responded to the publisher’s questionnaire.

In the third edition (2011),wemadeanumberof additions and inpart rear­ranged theold text.Anew chapter 1was added toplace our subject in amoregeneralscientificcontext.Itwasexpandedwiththefirsthalfoftheoldchapter2on the difference between morphosyntactic structure and phonological structure. Chapter 2isessentiallytheoldchapter1,whilechapter3isarevisedversionoftheoldchapter 2.Chapter8wasaddedinordertodealwithresponsestophonological

Preface vii

opacityinOptimalityTheory.TheauthorswouldliketothankJanineBerns,HyongSilCho,BertCranen,AnneCutler,MarindaHagen,RobertKennedy,HikaruOsawaandHenningReetzfortheirhelpwiththisthirdedition.

Themainconceptualchangeinthisfourthedition(2017) isamoreconsistentautosegmental treatment of three classes of segments (vowels, consonants andtones)andoftheirinclusionintheprosodicstructure.Thishasledtoarearrange­ment of the chapters on syllables, tone and stress, as well as a rewriting of the last twoofthesechapters.Whilethefunctionalrolesoftonesareemphasizedthrough­out the discussion, as a class of segments tones are treated in a single chapter on lexical and intonational tones. The chapter on stress has been expanded and deals with typological issues as much as with theoretical accounts. These rewritings have benefitedfromthereviewbyDavidDeterdinginJIPA43(2013).Inaddition,wehavesplit the chapter on distinctive features into two, so that students will experience amoreevenworkloadacrosschapters.Thenewchapter5dealswith themajor-class, laryngeal and manner features, while chapter 6 treats place features andincludes the rule format conventions that in earlier editions were discussed in a chapteronDutchdiminutives.Thatchapterhasbeendiscardedinthisedition.Thechapter on underlying and surface representations is now immediately followed bythechapteronLexicalPhonology.Finally,wehavecollapsedthetwochapterson feature geometry into one, and reduced the discussion on opacity. The chapters onopacityandfeaturegeometryare the lastchapters in thebook.Asaresultofthesechanges,thenumberofchaptershasbeenreducedfrom16to14.Wethank MirjamBroersma,HyongSilCho,BernardComrie,GrevCorbett,CatiaCucchiarini,Gunnar Holmstedt, Judith Hanssen, Beste Kamali, Hamed Rahmani, Makiko Sadakata andLeoWetzels for comments on a draft version andhelpwith data.Finally, it was an unspeakably reassuring experience to findHelen Tredget andKatherineWetzelskilfullyseeingourmanuscriptthroughproduction.

CarlosGussenhovenandHaikeJacobsRadboudUniversityNijmegen,theNetherlands

July2016

Acknowledgements

TheInternationalPhoneticAlphabet(2015)isreproducedbykindpermissionoftheInternationalPhoneticAssociation.

http://taylorandfrancis.com

Structures in languages

1.1 INTRODUCTION

Imagine a biologist wandering in a newly discovered forest inhabited by thousands ofundescribedanimalspecies.Anyresearchheorshetakesonwillyieldcompletelynew scientific data. This privileged situation is very much that of today’s phonol­ogist.At leasthalfof the7,000or so languages in theworldeitherhave sketchyaccountsdevotedtothemorareentirelyundescribed.Notwolanguageshaveeverbeen found that have the same phonology. In fact, varieties of the same language oftendifferintheirphonologicalsystems,andtherearethereforenumerousoppor­tunities to be the first in history to know about some phonological phenomenon. Ourhopeisthatyouwillbeabletoderivesomeexcitementevenfromtheexistinglanguage data that are presented in this book and that you can see how the ways humansorganizetheirvocalresourcesintogrammarsarequiteremarkable.Admit­tedly, seeing those structures, whether in new or existing data, requires more than a quicklook.Ourgeneralaiminthisbookistoguideyouthroughthevariousaspectsof phonological structure and indicate how they vary across languages.

Let’sbeginbyobservingthatallhumanlanguageshavetwoco-existingstructures:a phonological structure and a morphosyntacticstructure.Beforethispointcanbe made, we need to make it clear what it means for languages to have ‘structure’ to beginwith.Insection1.2,wediscusshowlanguagesvaryintheextenttowhichtheyallow particular kinds of structure to be ‘seen’, or observed. Phonological structure is not the same as the orthography in alphabetic writing systems, and we urge you to keep the notions of ‘letter’ and ‘sound’ distinct in your thinking about pronuncia­tion.Insection1.4,weexplainwhatismeantbymorphosyntacticstructureandthenmove on to a thought experiment in which you are invited to imagine a world with­out phonological structure, a mental exercise that is intended to make you see more clearly what it is. Its independence from the morphosyntactic structure is brought out by another thought experiment, in which we imagine a world where all lan­guageshavethesamephonologicalstructure.Finally,wewillmakethepointthatalllanguages have phonological structures, but that sign languages express their phono­logical elements visually, as manual and facial gestures, rather than acoustically.

1.2 OBSERVING LINGUISTIC STRUCTURE

Bythetimeheorsheisfiveyearsold,everychildinthisworldhaslearnttospeakahumanlanguage.Today,thiseventhappensabout16,000timeseveryhour.1 It is hard to say how many languages spoken in the world today are still being used by

1

Structures in languages2

care­givers to communicate with their children, but one thing is certain: those lan­guagesvarygreatlyintheirstructure.Someofthemwillhaveapassiveverbform,andotherswillnot;somewillhave5vowels,some13andyetothers25,andlike­wise the number of consonants will vary greatly; some will use pitch to distinguish words(seesection2.2.4),andsomewillnot,andsoon.Childrenofthatageareusually capable of saying that they have five fingers on each of their two hands, but if you were to ask them how many vowels their language has, your question would be royally ignored. This is not because the child does not necessarily know the mean­ing of the word ‘vowel’, but because the question goes well beyond what humans can naturally know. The structure of our language is not accessible to us in the way the outward shape of our body is; in fact, people are normally not even aware that their language has any structure at all.

1.2.1 Awareness of language structure

How,then,canweeverdevelopanawarenessofthatstructure?Incountriesinwhichchildren are taught to write in an alphabetic orthography, awareness of segments typi­cally arises because to a large extent the letters used to write words stand in some reg­ularrelationshiptothesounds–wewilloftencallthem‘segments’–inthosewords.ItisinfactquiteamentalsteptorealizethatanEnglishwordliketea consists of two segments,a[t]andan[iː],ratherthanbeingasingleunitofsound.(Thesymbol[ː],knownasthelengthmark,indicatesthattheprecedingsoundislong;seechapter2.)Ingeneral,awarenessofstructuralelementswilldependontwofactors.First,besidesyour natural inclination to look into such matters, there are the demands that are made on you to do so, as will happen at school. Illiterates may be unaware of the exis­tence of segments, and it will take more than a little work to reach that understanding (Moraisetal.1979).Second,thelanguageitselfwillrevealsomeelementsofstructuremore readily than others. That is, the structural elements of any one language vary in ‘salience’, and it is understandably easier to become aware of more striking elements thanoflessstrikingelements.Forinstance,thenotion‘lexicalword’naturallydevel­opsfairlyeasilyforspeakersofmostlanguages.Pre-schoolspeakersofEnglishusuallyknow that Johnny shouted! consists of two words, and No! of one, even though they will be less sure in cases like Don’t!, shell shocked or shaggy-dog story.

1.2.2 Language diversity

Butagain,wemustrememberthatlanguagesvary,andthusalsovaryinthesalienceof comparable notions. Inuit, spoken in northernAlaska, northernCanada andwesternGreenland,hasacomplexsystemofsuffixesandincorporatesnounswithverbs: many of its words or sentences are much like Don’t!, shell shocked and shaggy-dog story. Inuit children may therefore have different intuitions about the notion ‘word’thanEnglishchildren.Oragain,ifyouaskaspeakerofGermanhowmanysyllablesthereareinaparticularGermanword,youwilltypicallygetaquickandcorrect answer, but if you pose the same question to a speaker of Japanese about a Japanese word, you may well draw a blank, or get the wrong answer. This is because

Structures in languages 3

syllables are not particularly salient; or, rather, another structural element, smaller than the syllable, is more salient in Japanese. This is the mora (Hyman1985),tobediscussedinchapter9.Soifyouwantaquickanswer,youshouldaskhowmanymoras thereare in someJapaneseword.Ashortvowelandeach ‘half ’ofa longvowel are examples of moras, such that hi ‘day’ is one mora and boo ‘stick’ is two. In addition, a consonant in the coda of the syllable is a mora. In Japanese, this could be thefirst‘half ’ofalongconsonant(a‘geminate’),asoccurringinnattoo ‘fermented soybeans’,oranasalconsonant,like[n]asinkéndoo ‘swordsmanship’.Soinkéndoo therearefourmoras:[e],[n],[o]andanother[o],justasinnattoo,where[a],thefirst[t],[o]andanother[o]arethemoras.

In general, there tend to be various generalizations in a language that make refer­ence to salient structural elements, which may in part explain why these elements are salientandthusopentotheintuitionofspeakers(Kubozono1999).InJapanese,forinstance, there are many ways in which the moraic structure of words is relevant to the wayyoupronouncethemandthewaytheyaretreated.However,thesyllable,too,isastructuralelementinJapanese.Forinstance,itdeterminesthepossiblelocationsoftheword accent. Japanese words are idiosyncratically either accented or unaccented, and if a word is accented, an accent occurs on one of its syllables. In other words, the number ofsyllablesdeterminesthepotentiallocationsoftheJapaneseaccent.Althoughbothkéndoo and bokokugo each have four moras, there are in principle four locations in which the word accent could have occurred in bokokugo (awordthathappenstobeunaccented),butonlytwoinkéndoo (awordthathappenstohaveanaccentonthefirstsyllable).GiventherelativelylowsalienceofJapanesesyllables,itwilltakeaspeakerof Japanese more effort to be able to say how many syllables there are in kéndoo than itwouldtakeaspeakerofGermantodothesameforthecomparableGermanwordHandy [hεndi]‘mobilephone;cellphone’.Conversely,eventhoughthephonologyofGermancontainsanelement‘mora’,theGermanspeakerwouldprobablygiveyouablank stare if you asked him or her how many moras there are in that word.

Q1 The lines of the lyric Do-re-mi from the 1959 musical The Sound of Music (Richard Rodgers and Oscar Hammerstein) have the same number of syllables as the notes in the song. They are given in (1). In (2), a popular Japanese version of this song is given, again with the numbers of syllables in each line.

WhydoonlythefirstandlastlinesoftheJapaneseversionhavesevensyllables?(1) Doe,adeer,afemaledeer 7

Ray,adropofgoldensun 7Mi,anameIcallmyself 7Far,along,longwaytorun 7Sew,aneedlepullingthread 7La,anotetofollowsew 7Tea,Idrinkwithjamandbread 7

Structures in languages4

1.2.3 The role of spelling

Above,wesawthatawarenessofsegmentalstructurecanbe inducedbyculturalfactorslikelearninganalphabeticwritingsystem.Atthesametime,however,alan­guage’s orthography can be a hindrance to understanding the structure of language, becauseoftheirregularitiesintherelationsbetweenlettersandsegments.TheEng­lish word laugh hasfive lettersbut threesegments, [l], [ɑː]or[æ] depending on thedialect,and[f],whiletheFrenchwordtaxi has four letters and five segments, [t],[a],[k],[s]and[i].Speakersofsuchlanguageswilltendtomixuplettersandsegments in their attempts to become aware of the pronunciation of words. This is the reason why segments have been assigned unique symbols by the International PhoneticAssociation(IPA),whicharegenerallyusedbylinguists.Theyarewritteninsquarebrackets,ashasbeendonehere.TheIPAhasawebsitefeaturingallthesesymbols,manyofwhichweretakenfromtheRomanalphabet,andtheirmeanings,with sound files giving examples.2

Q2 In one of his sketches, the American comedian and singer Tom Lehrer referred to an eccentric friend of his called [ˈhεnri], who spelled his name ‘h’, ‘e’, ‘n’, ‘3’, ‘r’, ‘y’. After allowing his audience to ponder the mer-its of the spelling Hen3ry for a couple of seconds, he added: ‘The “3” is silent, you know’.

1 Whatismeantbya‘silentletter’?2 Argueon thebasisof theEnglishword light that ‘silent letter’ is not a

particularly helpful notion.

(2) Dowadonatunodo 7 ‘Do is do of donatu (donuts)’Rewaremonnore 6 ‘Re is re of remon (lemon)’Miwaminnanomi 6 ‘Mi is mi of minna (everyone)’Fawafaitonofa 6 ‘Fa is fa of faito (fight)’Sowaaoisora 6 ‘So is so of the blue sora (sky)’Rawarappanora 6 ‘Ra is ra of rappa (trumpet)’Siwasiawaseyo 7 ‘Si is si of siawase (happy)’

1.3 WHAT LINGUISTS DO

Asagroup,linguiststrytobecomeawareoftheentirestructureofthelanguagesthey study, even if each individual linguist is usually concerned with a specific aspect of those structures. The difference between the incipient awareness that lay languageusershave–whichapartfrombeinghopelesslyincompleteisoftengrossly

Structures in languages 5

mistaken – and the work of linguists is that the latter’s efforts take place in a scien­tific context. Thus, they will typically give evidence why they assume some struc­ture to exist and will hold themselves accountable for mistakes by publishing their hypothesesinpeer-reviewedjournals.

In principle, linguists aim to discover the entire structural system of every lan­guage,motivatinghypothesized structuresas theygoalong.Ofcourse, thisgoalis unrealistic, because languages are dying out at a rapid rate and the number of linguists falls far short of what would be needed for that task. It is hard to say how manylanguageswouldbesufficienttobeabletoanswerthequestionthatdefinesafurther goal of linguistics: what factors allow linguistic structures to exist, or, rather, what factors account for the distribution of those structures across the languages of theworld?Whyaresomeelementscommonandothersrare,andyetotherimag­inableonesapparentlynon-existent?Thosefactorswillcomefrommanydifferentdomains,likeourbrains(cognitivefactors),ourperceptualsystemandourspeechproductionsystem(together,thesearephoneticfactors),aswellassocial factors. Togofurtherstill,howdothesefactorsexplainthewaythatlanguagesareacquiredbyinfants?Andwhatistheexplanationforthefindingthatfromonegenerationtothe next the linguistic structures that are acquired are slightly different, such that over time languages change in a way that makes it hard to see that stages of a lan­guagethatareathousandyearsapartareactually‘thesamelanguage’?

1.3.1 Language-internal evidence

The evidence linguists present is of two kinds: language­internal and language­ external. In the case of language­internal evidence, the assumptions that linguists make about some element in the language structure are backed up by regularities in thelanguagethatrefertothehypothesizedelement.Wehaveseenanexampleofthisabove:thesyllabledetermineswheretheaccentcancomeinJapanese.Torepeatthispoint,adisyllabicstructurelike[hasi]couldbehási, hasí or hasi (recallthataJapa­nesewordmaybeunaccented),andtheseactuallyhappentobewords:‘chopsticks’,‘bridge’and‘edge’,respectively.Bycontrast,thereareonlytwotheoreticallypossiblewordsthatconsistofthesyllable[ree]:rée and ree, which again are real words, ‘a bow in greeting’ and ‘gratitude’, respectively. It doesn’t make sense to ask what reé might mean in Japanese: it couldn’t be a word, because the accent never occurs on half of a syllable.Now,forittobethecasethatthenumberofaccentablepositionsisequaltothe number of syllables it must of course be the case that the syllable exists.

Q3

1 HowmanyJapanesewordscouldbeformedfrom[hi]?Andhowmanywordscouldbeformedfrom[kentoo]?

2 Thereare,infact,twoJapanesewordsthatarepronounced[rée]:‘example’and‘bow,greeting’.Whatdoyoucallwordsthathavethesamepronunciation?

Structures in languages6

1.3.2 Language-external evidence

In addition to the language­internal evidence illustrated with the example of the Japanese syllable, there are many other sources of evidence. Most importantly,linguists can make phonetic measurements. They can measure the acoustic prop­erties of speech, track the articulatory gestures by speakers and register listeners’ responses to speech stimuli in perception experiments. These measurements enable themtounderstandbetterjustwhatisgoingonwhenpeoplecommunicatethroughspokenlanguage.Suchphoneticresearchnotonlyrevealsindetailthephysiologi­cal, physical and psychological events that together make up the ‘speech chain’, but can also be used to provide evidence for or against a particular structural hypoth­esis.Forinstance,toshowthatthemoraisanimportantelementinthestructureofJapanese, we could collect Japanese words varying in the number of syllables and thenumberofmoras.Onecategorywouldhavewordsofonemora,likego ‘word, language’; another words with two moras and one syllable, like kúu ‘bite,eat(byinsects)’;athirdwordswithtwomorasandtwosyllables,likemotí; and so on. If we then measured the durations of the words, we could see whether word duration is better explained by how many moras a word has or by how many syllables a word has. If it turned out that the mora explained better than the syllable how long a word is, we would have phonetic evidence for the greater relevance of the mora in Japanese.

Behaviouralexperimentstrytofindevidenceforlinguisticstructurebyregister­ingsubjects’responsestospecificallydesignedtasks.Suchtasksshould,ofcourse,neverdirectlyasksubjectstogiveanswerstoquestionsaboutlinguisticstructures,or at least never take such data at face value, because this carries the obvious risk ofincludingallsortsofmisconceptionsintheevidence.Toreturntolanguageartforms, we could use the rhythm of poetry to investigate the syllabic structure of words.IfwewantedtoknowhowmanysyllablesthereareinanEnglishwordlikehire, we could ask whether names like Hire Fire and Mo Toe could take the place of Humpty-Dumpty in the nursery rhyme Humpty-Dumpty sat on a wall. If all is well, there would be significantly more votes for Hire Fire than for Mo Toe, which difference could then be argued to be due to the fact that hire and fire are disyl­labic words, like Humpty and Dumpty, while Mo and Toe arenot.Asyoumayhaveguessed, real­life experimental questions and tasks are more complex than these examples suggest.

The research agenda of psychologists, whose interest in language received a boost inthe1960s,wastounravelthementalprocessesthatallowlistenerstounderstandspeechandtheprocessesthatallowspeakerstoproduceit.Youwillbefamiliarwiththe sensation you have that when listening to a foreign language you cannot tell where the words are. The reason for this is of course that, within a coherent phrase, there are no pauses between the words, any more than there are pauses between the syllablesormorasofwords(Cutler2011).Infact,ifby‘pauses’weweretothinkof brief periods of silence, then we will find quite a number of ‘pauses’ in coherent phrases, but these will occur within words as well as at the edges of words with recklessabandon.InanEnglishphraselikepaper napkins, there are two such silent

Structures in languages 7

intervals,onecorrespondingtothesecond[p] in paper andonetothe[pk] com­bination in napkins. The boundary between the two words does not correspond to any silent interval at all. If there are no audible word beginnings, the question arises hownativespeakersofEnglishavoidhearingwordsmadeupofsoundsequencesinside words, like pay in paper or nap in napkin, or even across word boundaries. Figure1.1showsaspeechwaveformofthisphrase,withlinesindicatingthebegin­ning and ending of sounds, as well as the word boundary. In this case, we cannot even say where the word begins. This could be defined as the moment that the lips closefortheformationof[p],butsincethisactionproducesnoacousticenergy,itcannot be distinguished from the silence that occurred before this event.

In fact, psycholinguists have shown that during speech perception, listeners do hear strings of sounds inside words as possible words, but this happens very briefly andbelowourawareness.Wordstringsacrosswordboundariesmayalso‘activate’words that were never said, such as income in mustering compassion (. . . muster income passion . . .),butbecauseoftheparticularwaysinwhichsoundscombineinto words, cross­word strings of sounds are less likely to constitute words. In fact, languages vary in the way their speakers detect word boundaries, depending on the phonologicalregularitiesthatapplytowordboundaries(e.g.allwordsmaybeginwithastressedsyllable,asinFinnish)andthephonetics(e.g.allvowel-initialwordsareprovidedwithaglottalstopbeforethem,asinGerman).

Q4 Why would the English phrase Call the police! never cause a word to be heard made up of a cross-word string of segments, while this may well happen in Wildest endings?

0

p p ph h k hæe

0.5 1.51Time (s)

napkinspaper

Figure 1.1 SpeechwaveformofaBritishEnglishutteranceofpaper napkins.Segmentboundarieswithinwordshave interrupted lines; the one separating the words is uninterrupted.

Otherresearchhasshownthatitiseasiertothinkofaword,asmeasuredbythetimeittakestosaythewordforanobjectinapicture,whenthespeakerhasjust

Structures in languages8

beenalerted to someotherword thatbegins in the sameway.That is, after justhearing checkers,peoplearequickertosay‘Cherry!’whenpresentedwithapictureofacherrythanafterjusthearingbackgammon. Priming research of this sort can go some way towards understanding how speakers retrieve the pronunciation of a wordbeforetheyactuallypronounceit(MeyerandBelke2007).

Psycholinguistic research into the processing of language in the brains of speakers and hearers and, more recently, neurological research into brain activity during lan­guage processing should of course be informed by the results of linguistic research intothestructuresoflanguages.Equally,structurallinguisticresearchshouldtakeplace in the wider context of psychological, neurological and sociological research, something that has not always been the case. In fact, some people would say that linguists are notorious for working out their problems in isolation from other fields, while some linguists may feel that psycholinguistic research has not yet addressed the possible roles of all aspects of the linguistic representation.

1.4 MORPHOSYNTACTIC STRUCTURE

Phonology is concerned with a particular aspect of linguistic structure. In order to see what part of linguistic structure is phonological, we will first briefly consider that part of the linguistic structure which is not phonological, the morphosyntactic structure. The morphosyntactic structure of a language can be seen in the arrange­mentof themeaningfulunitsofany linguisticexpression.Adistinction ismadebetween morphology, which deals with the structure up to the level of the word, and syntax, which deals with the structure above the word.

1.4.1 Morphemes and words

The smallest morphosyntactic unit is known as the morpheme. The word scratch is a single morpheme, but the word pens containstwomorphemes.Amorphemethat can be a word by itself is a free morpheme. The morpheme pen could form a word by itself, and is for that reason an example of a free morpheme, while the morpheme s is a bound morpheme, because it must combine with some other morphemeinordertobepartofaword.Awordconsistingofasingle(necessar­ilyfree)morphemeisasimplex word, other words being complex.Animportantgroup of bound morphemes are affixes, which can be divided into prefixes and suffixes,dependingonwhethertheyareplacedbeforeorafterwhatisknownasthebase. Thus, the s inourexampleisasuffixbecauseitisplacedafterthebasepen, and so is able in the complex word scratchable.

Most words have a category membership, which allows them to be used in certain positions in sentence structure. If we assume that a sentence consists of aNounandaVerb, in thatorder, thenthesentencePens scratch is well­formed, while *Scratch pens is an ill-formed sentence (as indicatedby the*) ifpens is a subject.Affixesaresubcategorized for a category membership, meaning that they onlyattachtobasesthataremembersofaparticularcategory.Thus,thepluralsuffix s

Structures in languages 9

attachestonouns,whilethesuffixable attaches to transitive verbs, i.e. verbs that takeadirectobject,suchasscratch, read and drink (butnotsleep: *sleepable).Affixesare also distinguished on the basis of whether they create new words, in which case they are known as derivational affixes, ormerely createnew formsof the sameword,likethepluralsuffixs orthepasttensesuffixinscratched, in which case they are inflectional.Derivationalsuffixescomewiththeirowncategorymembership,which is frequently different from the category of their base. Thus, able has the cat­egoryAdjective,andeverycomplexwordendinginable isthereforeanAdjective.In(1),thestructuralinformationforthesetwoaffixesisgiveninanotationknownas labelled bracketing.

(1) a [[ ]N z] plural

b [[ ]Vtrans əbl]Adj ‘isabletoVerb’

Observethatabasecanbesimplexorcomplex.Inscratchers,thepluralsuffixwasattached to scratcher, itself composed of scratch and er.Asimplexbasewhichisafree morpheme is also known as a stem, while a simplex base that is not, like the Englishverbalbasemit in transmit, is also known as a root.

Q5 Explain why there are no words *dieable and *seemable in English, and why imaginable is a word.

1.4.2 Syntactic structure

Wordsdonotdirectlyenterintosentencestructure.IfthestructureofoursimplesentenceSweresimply[NV],itwouldonlybepossibletoformsentencesofthetypeKittens scratch, Pens leak and Time flies. In fact, we can also form A good pen mustn’t scratch or The pen I bought in Italy leaks.Here,thestructuralpositionofPens in our original example sentence Pens scratch is occupied by the word groups A good pen and The pen I bought in Italy. This is the reason why, in between the structural level oftheword,illustratedin(1),andthestructurallevelofthesentence(orclause),weneedtorecognizethestructurallevelofthe(syntactic)phrase. That is, instead of[NV],oursimplesentencehasthestructure[NounPhrase–VerbPhrase],or[NPVP].InadditiontotheNPandtheVP,anotherfrequenttypeofphraseisthePrepositionalPhrase,orPP,whichconsistsofaprepositionplusanNP,suchaswith the leaky pen, on the surface or in Italy.

The introduction of a level of phrasal structure allows us to point to two proper­tiesofmorphosyntacticstructure.Oneisthatthisstructureishierarchical. This means that we can distinguish higher and lower levels of structure: a word level, a phrase level and a sentence level. The second property is recursiveness. This means that a unit of a given structural level can be incorporated into a unit of the same structural level, or that a unit of a higher level of structure can be incorporated

Structures in languages10

intoaunitofalowerlevel.Forexample,thesentenceThe pen I bought in Italy leaks consistsoftwophrases,theNPThe pen I bought in Italy andtheVPleaks. The first oftheseismadeupofanNPThe pen andanS(which)I bought in Italy, which sen­tenceinturnconsistsoftheNPI andaVPbought which (i.e.the pen)in Italy. The structurein(2)expressesthis.In(3),wegivethemini-grammarthatwillproducethissentence.Recursivenessisshownin(3b),first,bythefactthatwecanfeedahigher-levelSintoalower-levelNPand,second,bythefactthatwecanfeedanNPintoanNP.Asyouwillhaveguessed,wecaninprincipleapplythesamerulesagainand again. It may be fun to do this, as illustrated by the children’s verse This is the farmer that kissed the girl that chased the cat that killed the mouse, etc.

(2) S[ NP[ NP[Thepen]NP S[ NP[I]NP VP[bought(sc.thepen)inItaly]VP ]S ]NP VP[leaks]VP ]S

(3) a S: NP VP b NP: NP S

There is much that is the same in the syntax of the languages of the world, but thereisalsoagooddealofvariation.SomelanguageshaveaVPinwhichtheVerbprecedestherest(i.e.[VNP]or[VPP]),whileotherlanguagesmaintaintheoppo­siteorder(i.e.[NPV]and[PPV]).Moreover,somelanguageshaveasyntacticrulethatmovestheVerbtoinitialpositioninthesentence.ThistypeofvariationisoftenreferredtoasthatbetweenS(ubject)V(erb)O(bject),SOVandVSO.

Q6 Identify the nouns and verbs in the following words:

1 unscratchable2 road tax increase

Q7 Identify all the NPs in the following sentence:

He wrote the letter with a pen he bought in Italy

1.5 A WORLD WITHOUT PHONOLOGICAL STRUCTURE

Spoken languages use human vocal sound to give shape to their morphemes.Phonology is the branch of linguistics that aims to describe the way in which this medium of human vocal sound is structured, in languages generally as well as in individuallanguages.Toseewhatismeantbythesoundstructureoflanguages,itmay be instructive to pretend for a while that languages do not have it. Imagine thateverymorphemeofalanguagewereassignedsomevocalization.Conceivably,these vocalizations could be quite lengthy in view of the large number that would

Structures in languages 11

beneededtodistinguishallthemorphemesofthelanguage.But,importantly,therewould be no implication that they should have structure in the sense that they are composedofsubparts,anymorethanabstractpaintingsare.Togiveahypotheti­calexample,vocalization(4)mightbethemorphememeaning‘oaktree’insomelanguage.

(4) ‘oaktree’:high-pitchedwheeze,trailingoffintoavoicedcoughwithcentral,nasalvowelquality

The wide variation in morphosyntactic structure that is found in the languages of the world could exist without there being any structure to the human vocal sounds that languagesuse as amedium.Clearly, in thehypothetical situationdescribedabove,languagescouldstillhaveSVOorVSOastheirbasicwordorder.TheycouldeitherhaveanextensiveCasesystem(i.e.bestrongly inflectional)orhavemanyprepositions; they may or may not have articles, they may or may not mark plural inbothSubjectsandVerbs(‘Concord’)andsoon.Whereourhypotheticalsituationis different from the situation in the real world is that nothing would have to be said about the sound structure, because there would not beany.Allthatanyonecoulddoismakealistofdescriptionsliketheonein(4).

Q8 Why, in the hypothetical situation above, would it be impossible for speech errors like [klɪskεə] for kiss Claire to occur?

Now let us turn to real life.The first observation to bemade about the pro­nunciation of morphemes is that there are recurringelements.For instance, thesound patterns used for morphemes can be analysed as strings of segments. This isthebasisonwhichwecansaythatEnglishcat has the same segments as tack or act, althoughthesegmentsoccurindifferentorders.Minimally,then,thetaskofphono logy is to state what these recurring elements are.

The second observation to be made is that the recurring elements do not occur inallpossibleorders.Forexample,whilecat, act, tack are possible combinations in Englishofthesegments[k],[æ]and[t],thesamedoesnotgofor[ætk].Noticethatitisnotthecasethatthesequence[ætk]neveroccursinEnglish.Thissequenceispart of the word Atkins, for instance, and occurs in the sentence The cat killed the mouse. Whatthisshowsisthatthereissomeconstituenthigherthanthesegmentwhich imposes constraints on what sequences of segments it may contain. In the example, that constituent is the syllable. That is, *[ætk]cannotoccurasasequenceof segments inside the syllable. In general, elements at one level of structure com­bineinrestrictedwaystoformelementsofahigherlevelofstructure.Wewillseeinchapter5thatthesegmentistobelookeduponasanelementthatitselfcombineselements of a lower structural level, called distinctive features.Soasecondtaskforphonological theory is to state what the permitted patterns of arrangement of the phonological elements are.

Structures in languages12

Athirdimportantobservationisthatsegmentsmaybepronounceddifferentlydepending on their environment. This is because languages usually have processes that affect (‘change’) segments in particular contexts. In English, for example,[p tk] areaspiratedwhen theyare thefirst consonantof a syllable.Hence, tack ispronounced[thæk],butstack is [stæk].Similarly, theDutch[z] in[zeː] ‘sea’ ispronounced[s]whenitisprecededbyanobstruent,asin[ɔp seː]‘atsea’,butnotelsewhere,asin[aːn zeː]‘attheseaside’.Oragain,inFrenchthefinalconsonantofanadjectivelike[pətit]‘small’ispronouncedinle petit autobus ‘the little bus’, but not in le petit camion ‘the little lorry’, as a result of the difference in the first segment inthefollowingnoun.Astheseexamplesmakeclear,oneandthesamemorphememay have different pronunciations in different contexts as a result of the existence of these processes.

Minimally,then,aphonologicaldescriptionwillhavetoanswerthreequestions:what are the phonological elements, how do they combine into higher structural elements,andwhatgeneralizationsexistabouttheircontextualvariation?

Q9 Do you think that sign languages for the deaf have phonologies?

1.5.1 One phonology for all languages?

Now, itmighthavebeenthecase thatallhuman languageshadthesame sound structure. In order to see what this would mean, let’s once more enter into an imagi­naryworld.Wewillmakeupa‘universalphonology’andthengivepartialdescrip­tions of two – very different – languages. The universal sound structure is given in (5),and(6)givespartialdescriptionsofthetwolanguages.

(5) a Segmentinventory:C V

p t k i u b d g a mn

l b Arrangement: Awordconsistsofone,twoorthreesyllables. AsyllableconsistsofCV. c Process: fricationInitialobstruentsinnon-initialwordsarefricatives.

Becauseoffrication, [ptk]willbe[fθx],and[bdg]willbe[vðɣ],inthecontext specified.Because the fricativesarevariantsof theplosives, theyarenotlisted separately in (5a). Observe that the labial obstruent in this language hastwo pronunciations, [p] and [f]. The distinction between a sound category orphoneme,here/p/,anditsdifferentpronunciationsisusuallyexpressedbywriting

Structures in languages 13

thephonemebetweenslashesandthepronunciationsbetweenbrackets,[p]and[f](seechapter7).

(6) Language I Language IISyntax SVO VSOSubject-VerbConcord yes no‘cow’ pi kapu‘graze’ namu ni‘field’ diku puPlural [[ ]N/Vlu]pl [[ ]Nti]pl

‘in’(Preposition) ma –LocativeCase – [[ ]Nla]loc

Usingthedatainthefirstcolumnof(6),wefindthatinLanguageIthesentencemeaning ‘Cows are grazing in thefield’wouldbe [pilunamulumaðiku].Here,Subject-VerbConcordisexpressedbythedoubleoccurrenceof[lu],whiletheposi­tionofthePP‘inthefield’isthatoftheOinSVO.

Q10

1 Whydoesthewordfor‘field’inthesentenceofLanguageIbeginwith[ð],andnotwith[d]?

2 WhatisthetranslationofthissentenceintoLanguageII?3 Whywould theprocessweassumed in(5c)beaconvenient featureof

humanlanguage,asseenfromthepointofviewofthelistener?

The reason why a single phonology for all languages is so improbable is that the phonologiesoflanguageschangeovertime,justasdoesthemorphosyntax.Onegeneral factor inducing change affects both aspects of structure.When learn­ing their language, infants may make different generalizations based on the data fromwhattheirparentsdidwhentheylearntthelanguage.Changesinducedbysuch generalizations will affect word order as well as the forms of specific words. Tobeginwith a syntactic example, the position of the verb changed from anearlierSOVinProto-GermanictoSVOinEnglish.OtherGermaniclanguagesremainedSOV,exceptthattheymovedthefiniteverbformtothesecondposi­tioninmainclauses.Asaresult,EnglishnowhasSVOinbothdependentclauses(7a)andmainclauses(7b),butDutchhasSOVindependentclausesandapar­tial SVO, known as ‘Verb-Second’, in main clauses: S-Vfinite-O-Vnonfinite (Koster1975).Here, English acquired amore general rule than didDutch and otherGermaniclanguages.

(7) English Dutcha Ithinkshewould like to buy fish Ikdenkdatzeviswil kopenb Shewould like to buy fish Zewil viskopen

Structures in languages14

Amorphological example concerns theDutch3sg verbal suffix [t]. It appearsafterstemslikeken ‘know’ and raak ‘hit’, which thus change to kent and raakt when usedwitha3sgsubjectNP.However,auxiliaryverbslikewil ‘would like’ and kan ‘can’don’ttakethis[t]:De jongen kent het ‘The boy knows it’, but De jongen kan het ‘Theboycan(do)it’.Childrenoftengeneralizet-suffixationtotheauxiliaryverbwil and say De jongen wilt het for the parental De jongen wil het ‘The boy wants it’. In this case, the treatment of wil is probably inspired by the meaning ‘want to’, which is expressedbyanon-auxiliaryverb inmany languages, even thoughDutchwil behavesjustliketheotherauxiliariesinotherrespects.Suchnovelgeneralizationsare usually modified in later stages of the acquisition process, as when children acquiringEnglishstartusingmice instead of the mouses which they may have pro­duced before, but some stay and lead to language change.

Asecondfactorreliesonchanges in thepronunciationofspecificsounds,orofspecificsoundsinspecificcontexts.Suchchangesareverycommon,althoughwemaynot always realize that the variation we observe between speakers may be an indica­tionofchange.Suchchangesaresociologicallydeterminedinthesensethattheyarisewithinasmallergroupofspeakers.Again,someofthesechangeswillbecomemoregeneral,ashappenedinthecaseofthedisappearanceof[r]fromtheendsofsyllablesinanumberofvarietiesofEnglish.IntheEnglishspokeninEngland,thisdisappear­ance caused court to have the same pronunciation as caught, and garter to rhyme with sonata. Asaresult,thephonologyoftheinnovating‘r-less’speakersrepresentsafairlydrastic change relative to that of the older system, for instance in having additional, newvowels,likethediphthong[ɪə]inawordlikebeer. IntheEnglishspokeninEng­land, beer now rhymes with idea, butintheEnglishspokenintheUSAandCanada,which preserves an ‘r­full’ version of the language, they do not rhyme.

1.6 TWO KINDS OF STRUCTURE

The fact that the vocalizations that are used to represent the meaningful units of lan­guage themselves have structure has an important consequence, which is that terms like ‘constituent’, ‘element’ and ‘unit’ are ambiguous. The speaker or writer might be referring to a morphosyntactic unit, such as ‘the morpheme hill’ or ‘the sentence I like it’, ortoaphonologicalunitlike‘thesegment[ε]’or‘thesyllable[kεt]’.

Q11 Divide the following English words up into (a) morphemes and (b) syllables:

elephantspalm oilunsettling

Structures in languages 15

Thus, a linguistic expression always has two structures, a morphosyntactic one, which reflects the meaningful elements in the expression, and a phonological one, which is the structure most immediately relevant to the pronunciation of the expression. The distinction is a very real one, because the morphosyntactic constit­uents do not map one to one onto the phonological constituents; to use a technical term, they are not isomorphic. That is, morphemes do not exclusively correspond tosegments,orexclusivelytosyllables,etc.WhiletheDutchpolitesecondpersonpronounconsistsoftheonesegment[y],themorpheme[ɪk]‘I’consistsoftwo.Andwhile these two pronouns each consist of one syllable, the informal second person plural pronoun consists of two: [jʏ.li]. (Theperiod ‘.’ is used to indicate syllableboundaries.) InEnglish, thesinglesyllable [suːz]canrepresent twomorphemes,Sue and is, as pronounced in the sentence Sue is ticklish. In(8),themorphosyntacticstructure of that sentence is given. The way the sentence is analysed in morphosyn­tacticconstituentsshouldbecomparedwiththe(partial)phonologicalconstituentanalysisofthatsamesentence,givenin(9).Noticethatjustlikethemorphosyn­tactic structure, the phonological structure is hierarchical. There is a layer of seg-ments,whichbuildalayerofsyllables(symbolizedσ),whichbuildalayeroffeet(symbolizedF),whichbuilda layerofphonological words (symbolizedω),andso on. In addition to the different constituent structures for Sue is, for example, notice that ticklish consists of two syllables as well as of two morphemes, but that the syllables and the morphemes do not divide the word up in the same way. The dual structure outlined in this section exists in all languages. It is dealt with in more detailinchapter12.

(8)

Q12

1 InthecaseofEnglish[æ],[k]and[t],allpermittedarrangementsareinfact words: cat, tack, act. But ifwereplace [æ]with [ɪ], combining thethree segments yields only two words: kit and tick. Woulditbecorrecttosaythat[ɪkt]isanill-formedcombinationinEnglish?

2 Canyouexplainwhy[mrɛk]wouldneverbeintroducedasabrandnameinEnglish,while[krɛm]mightwellbe?

Structures in languages16

(9)

1.7 CONCLUSION

This chapter has made the point that linguistic expressions have two parallel hierar­chicalstructures.Oneofthese,themorphosyntacticstructure,reflectsthemeaningof the linguistic expression, while the other, the phonological structure, reflects its pronunciation.Agivenconstituentineitherstructurewilltypicallynotconsistentlymap onto any single constituent in the other structure, a point that will be worked outmoreinchapter12.Beforewemoveontoafurtherdiscussionofthephono­logical structure, we first deal with the way we produce and articulate speech, in chapter2.

NOTES

1 EstimatebasedonMedianCrudeBirthRatefor2015–2020(http://data.un.org),minusUnder-5ChildMortality(http://www.who.int/features/qa/13/en/index.html).

2 http://web.uvic.ca/ling/resources/ipa/charts/IPAlab/IPAlab.htm

The production of speech

2.1 INTRODUCTION

Todescribehowweproducespeechandwhatspeechlookslikeacousticallyinthespace of a single chapter is a tall order, and you would do well to consult other text­booksthatdealmorespecificallywiththephoneticsofspeech,likeCatford(1988),Laver(1994),LadefogedandMaddieson(1996),LadefogedandJohnston(2010),Ashby(2011),ReetzandJongman(2011)orZsiga(2012).Wedescribethespeechproductionprocessintwostages.First,weconsidertheroleofthelungsandthelarynx. This part of the speech organs is responsible for the actions of the vocal folds,whicharelocatedinsidethelarynx.Acommonandspectacularactionofthevocal folds occurs when they vibrate against each other so as to produce a buzz­ingsound,whichcanbevaried inpitchand loudness.Second,wedealwith therole of the channel extending from the larynx onwards, called the vocal tract. It is formed by the pharynx, the mouth and, for nasal and nasalized sounds, the nasal cavity. The vocal tract modifies the buzzing larynx sound. In its unmodified form, it would sound much like a small petrol engine, but this is hard to verify so long as we keepourheadson!Becausethevocaltractcanassumemanydifferentshapes,thesemodifications are highly varied. The most striking effect here is the production of different vowel sounds. The term organs of speech is used to refer to parts of the body in the larynx and the vocal tract that are involved in the production of speech. It is a misleading term in that it suggests that we have special physical organs for speaking. This is not so: all our so­called organs of speech have primary biological functions relating to our respiratory system and the processing of food.

The pronunciation of words is conventionally represented with the help of pho-netic symbols, any such representation being a phonetic transcription. The sym­bolsusedinthisbookarethoseproposedbytheInternationalPhoneticAssociation(IPA),whichcanbefoundonpageix.Aphoneticsymbolstandsforaparticularspeech sound, or segment, which is defined independently of any language. Pho­netic symbols may be accompanied by diacritics, signs which are printed above or below a phonetic symbol or with which the symbol is superscripted, and which specify particular features of pronunciation. For example, in the transcription[khæt],whichindicatesthepronunciationoftheEnglishwordcat, [kh]representsa[k]whichisaccompaniedbyaspiration,abrief[h]-likesoundoccurringbetweenthe[k]properandthefollowing[æ].Itisnotalwaysnecessary,orevendesirable,toindicate all the features of the pronunciation of a word in a transcription: the tran­scription[kæt] isoftensufficiently informative if the readerknows the languageconcerned.Atranscriptionthatincludesagreatdealofdetailiscallednarrow.

2

The production of speech18

2.2 THE LUNGS AND THE LARYNX

A crucial requirement for the production of acoustic energy is amechanism tocreate an air pressure difference in the appropriate locations in the larynx and the vocal tract. There are a number of ways in which air pressure differences for speech production are created, called airstream mechanisms (Abercrombie 1967).Themost commonly used by far is the pulmonic airstream mechanism, and the great majorityofspeechsounds are produced with the help of increased air pressure cre­atedbyourlungs.Beforewebegintospeak,webreathein,takinginsufficientairtoproduce an utterance of reasonable length. Instead of simply letting go of the mus­culartensionandallowingourlungstocollapse,pushingtheairfromthem(whichiswhatwewoulddoifwewerebreathingnormally),weslowlyeaseupontheten­sion, thereby slowing down the exhalation phase. This artificially extended period ofpressurefromourlungsisusedtoproducespeech.Becauseitistheexhalationphase rather than the inhalation phase that is used, these speech sounds are called egressive.Mostlanguagesonlyhavepulmonicegressivesounds.Insection2.8,wewill briefly describe the production mechanisms of three types of nonpulmonic sounds(clicks,implosivesandejectives).

Afterpassingthroughthebronchiandthetrachea,thefirstorgantheairstreamwill meet on its path from the lungs is the larynx. The outward part of this organ canbefelt–and,especiallyinmen,beseen–atthefrontoftheneck(theAdam’sapple).Thelarynxisavalve,whichcanbeopenedandclosedbytwothickishflapsthatrunfrombacktofrontinsidethelarynx(seeFigure2.1).Theseflapsarepri­marily there to prevent food or saliva from entering the lungs, but because they also have a function in speech they are known as the vocal folds or the vocal cords.1 The aperture between them is called the glottis.Noaircanpassthroughtheglottiswhen it is closed, while the air can flow quite freely through an open glottis.

(a)

(c)

(b)

Figure 2.1Schematicdrawingsof(a)aclosedglottis,asduringtheclosurestageofaglottalstoporduringtheclosedphaseofthevibratingglottis;(b)anopenglottis;(c)anarrowedglottis.

The production of speech 19

Thepulmonicallyproducedpressuredifferenceisusedforthreepurposes.First,it can be used to drive the vibratory opening and closing actions of organs like thevocal folds,asexplained in section2.2.1.During thearticulationof [m], forinstance, vocal­fold vibration can easily be felt by placing one’s fingertips on one’s ‘Adam’sapple’.Second,itcanbeusedtogenerateaflowofairthatcanbechannelledthrough a narrow opening to create audible air turbulence, or friction. This happens duringthearticulationof[s],forinstance.Third,itcanbeusedtobuilduppressurebehind a complete blockage of the vocal tract in order to create an explosive sound when the blockage is suddenly removed. This occurs in the articulation of plosive consonantslike[p],suchaswhenwesay[pa].

2.2.1 The vocal folds: the open and vibrating glottis

There are many consonants that are produced with the glottis held open, as in ordi­narybreathing.Suchsoundsarecalledvoiceless, and we hear them because other speech organs, the tongue or the lips, are used to generate fricative or explosive soundsfurtherupinthevocaltract.Examplesofvoicelesssoundsare[f]and[ʃ]infish andthesoundsequence[st]instay.

The vocal folds are exploited in various ways to create sound which can be used as a basis for speech. This is known as phonation. The most important type of pho­nation is voice,whichisproducedwhenthevocalfoldsvibrate.Vocal-foldvibra­tionoccurswhentheclosedglottisissubjectedtoincreasedsubglottalairpressurewhichissufficienttoblowthevocalfoldsapart,butnotenoughtopreventthemfrom falling together again when the air pressure between them drops as a result of theBernoullieffect,aphysicaleffectwhichcausespressuremimimaatpointswheretheflowofgasesorliquidsishigh.Assoonastheyhavebeensuckedtogether,thevocal folds are once more blown apart as a result of the subglottal air pressure. This processtypicallyrepeatsitselfmorethan100timespersecondforthelargerandlaxervocalfoldsofmen,andover200timespersecondforthesmallervocalfoldsofwomen.Consonantslike[m],[l]and[j](whichwewillseearesonorantcon­sonants)andvowelsarenormallyvoiced.Voicedobstruentconsonantsalsoexist,suchas[b]inEnglishabbey and[z]inlazy, and usually contrast with their voiceless counterparts(inthesecaseswith[p]and[s],asinhappy, lacy).

Q13 When you attempt to lift a heavy object, like a table or large stone, you naturally close your vocal folds. You do this in order to prevent your contracting muscles from losing their leverage due to a collapse of the rib cage over which the muscles are stretched. While holding your breath in this way, try to make noises in your mouth. You will find you can make various sounds, by clicking with your tongue or making a popping sound by suddenly opening your lips. Why can’t these sounds be classed as ‘pulmonic’?

The production of speech20

2.2.2 Devoicing and aspiration

Whenaconsonantthatisnormallyvoicedispronouncedwithoutvocal-foldvibra­tion in some context, it is said to be devoiced.Devoicedsegmentsaresymbolizedwithacirclebelowthesymbol(whichmayalsoappearaboveit,ifthereisnospacebelowit).Forexample,devoicingmayfollowvoicelessobstruents,inparticularplo­sives. Inmany languages, thevocal foldsmaybegin tovibrate immediatelyafterthe release of the closure made for the plosive for a following voiced segment, as in[pa],butinotherlanguagesthevocalfoldsmayremainopenforawhile.Inthelatter case, the plosives are said to be aspirated.Asshownabove,aspiratedplosivesaresymbolizedwithasuperscript[h],asin[pha].Englishhasvoicelessaspiratedplosives at the beginning of the syllable, as in tea, pea, key. If a sonorant conso­nant rather than a vowel follows the aspirated plosive, the aspiration is indicated by meansofthedevoicingdiacritic,asinEnglish[plei]play.

The timing relation between plosive releases and the onset of vocal­fold vibration is expressed as voice onset time, orVOT.TheVOTiszerowhentheplosivereleaseand the onset of vocal­fold vibration are simultaneous; when the onset of vocal­fold vibrationisearlierthanthereleaseoftheplosive,VOTisnegative,andwhenitislater,itispositive.TypicalvaluesforaspiratedplosivesarebetweenVOT +50msandVOT +80ms.

Figure2.2showsaspeech waveform of the word pass, spoken by a speaker of BritishEnglishas[pɑːs].Thefirstacousticeventistheburstof[p].(Recallfromchapter1thatwecannottellfromtheacousticswhena[p]beginsifitoccursaftersilence.)Thenthereissomeweakturbulence,whichistheaspirationasindicatedby[h].TheVOTismeasuredfromthebeginningofthebursttothebeginningofthevowel,andis61msinthisexample.Thewaveformduringthevowelshowsthevibratingvocalfolds.Finally,thevoiceless[s]consistsofloudturbulence.Voicedsounds like [m] and [ɑː] are ‘periodic sounds’, because the waveform shows a

0 1Time (s)

p sh ɑː

Figure 2.2SpeechwaveformoftheBritishEnglishwordpass. The vertical lines demarcate the release of the bilabial[p],thebeginningofthevocal-foldvibrationfor[ɑː],thebeginningof[s]andtheendof[s].Thevoiceonsettime(VOT)isgivenbythedurationbetweenthefirsttwodemarcationlines.

The production of speech 21

repeatingpattern.Eachoftherepeatedportionsofthewaveformcorrespondstoanopening-and-closingactionofthevocalfoldsandisknownasaperiod.Figure2.3reproducesasectionfromthewaveforminFigure2.2,withoneperiodmarkedout.Noticethattheshapeofthewaveformduringthisperiodismoreorlessthesameasthoseoftheperiodsbeforeandafter.Thedurationoftheperiod,usuallymea­suredinmilliseconds(ms),dependsonthefrequencywithwhichthevocalfoldscarryouttheiropening-and-closingactions.Iftheseoccur200timesper1,000ms(expressedas200Hz),theperiodwillbeexactly5ms.Thefrequencyofvibrationofthe vocal folds is the fundamental frequency of the speech signal, also referred to asthef0(‘F-zero’).Theshapeofthewaveformduringaperioddeterminesthequal­ityofthesound,inthiscasethatoftheBritishEnglishvowel[ɑː].Panel(b) showsaperiodofaboutthesamedurationforthevowel[iː]inpeace, andpanel(c)doesthe samefor[ʊ]infoot. The different shapes of the waveforms are due to the different shapes the vocal tract assumes during the production of these vowels: the reso­nances that are set up in the air in the vocal tract depend on its length and shape. The relation between these resonance frequencies, known as formants, and vowel qualitiesisbrieflydiscussedinsection2.4.

Theturbulentsignalfor[s]hasnofundamentalfrequency.Itconsistsof‘noise’,a largely random pattern of vibrations in which broad frequency zones are more

0.00498

0.025Time(s)0

Panel a

0.00483

0.025Time(s)0

Panel b

0.00494

Time(s)0

Panel c

0.025

Figure 2.3Waveformsfor[ɑː](panela),[iː](panelb)and[ʊ](panelc)withapproximatelyequallylongperiods.

The production of speech22

emphasized than others. These different emphases in the frequency spectrum deter­minewhethertheturbulencesoundsmorelikeahigh-friction[s]oralow-friction[f],etc.Whenturbulenceandvocal-foldvibrationareproducedsimultaneously,weproducesoundslike[z]and[v].

Q14 What would the speech waveform of the word ceased look like?

Q15 What is the approximate fundamental frequency of the vowel shown in panel c of Figure 2.3?

2.2.3 Special types of phonation

Three special types of phonation are mentioned here: whisper, breathy voice and creaky voice.

1 Thevocalfoldscanbebroughttogethertoformanarrowing which produces friction when air passes through it. This is how people whisper: instead of voice, glottal friction may be used as the acoustic source to be modified. Awhisperedspeechsoundoccurringinotherwisevoice-phonatedspeechis[h],asinEnglish[hæt]hat. Whispercanbeindicatedbythedevoicingsymbol,ashasbeendonein[ææt].

2 Breathy voice occurs when the closing phase of the vibration is not complete, sothatairisallowedtoflowthroughwithfrictionduringphonation.BreathyvoiceisusedinEuropeanlanguagestosignalconfidentiality(Laver1994:200),andissometimesusedtocreatetheeffectofasexyvoice.Voicedaspiratedplo­sives, which occur in many languages spoken on the Indian subcontinent, are produced by allowing breathy voice to be used throughout the plosive and the followingvowel,asinHindi[bɦal]‘forehead’.Breathyvoicecanbeindicated by[],asinDutch[uut]‘hat’,alternativelytranscribedas[ɦut].

3 Creaky voice or laryngealized voice is produced with tight vocal folds, and oftenallows the listener tohear theopeningactionsof thevocal foldsasseparateevents.(Theeffectmayremindyouofthesoundproducedwhenrunningafingernailacrosstheteethofacomb.)BritishEnglishspeakersmay break into creak at the ends of their utterances, when the pitch is low. BothmaleandfemalespeakersofAmericanEnglishmayhaveconsistentcreakduringlongerpartsofutterances,togetherwithfairlylowpitch.ManyNiloticlanguagesuselaryngealvoicecontrastively(Ladefoged1971).Dinkahasasetofvowelswithcreakyvoice,symbolizedbymeansof[ ],whichcontrast with a set with breathy voice, as in [ròːr] ‘forest’ – [ròːr] ‘men’(Andersen1987).

The production of speech 23

2.2.4 Pitch

Variationsinthefrequencyofvibrationareheardbythelistenerasvariationsofpitch: the more frequently the vocal folds open and close, the higher the pitch. In languageslikeEnglish,French,GermanandSpanish,variationsinpitchareusedtosignaldifferentdiscoursalmeanings.For instance, intheutteranceBut I don’t want it!, the syllable want will be higher than the syllable it, but in Want it?, it will be the higher syllable. These two intonation patterns are known as the ‘fall’ and the ‘rise’, and add a ‘declarative’ and ‘interrogative’ meaning to the utterance, respec­tively. In other languages, called tone languages, different pitch patterns are used in the same way vowels and consonants are used in all languages, i.e. to distinguish words fromone another.Dinkahas low tone [ˋ], high tone [´] and falling tone[ˆ],asillustratedby[ròw]‘two’vs [rów]‘thirst’and[yóː w]‘bones’vs [yôː m]‘wind’(Andersen1987).Karenhasthreeleveltones,high,mid[¯]andlow,asillustratedby[tə]‘one’,[tə]‘spoon’and[tə]‘ant’(Jones1961).Notationsfortonesvary.TheIPAofferstheaccentmarksusedintheseexamplesandthe‘toneletters’introducedbyChao(1930),whichconsistofaverticalbarwithagraphicrepresentationofthepitchanchoredonitsleft,printedafterthesyllable.Ahighfallingtoneisindicatedby[a˥˩],amidfallingtoneby[a˧˩],amidleveltoneby[a˧],andsoon.Athirdnota­tion is in fact more widely used than the tone letters for languages spoken outside Africa,includingChinese:superscriptsequencesoftwoorthreedigitsfrom1to5,where5isthehighestpitch.Thethreetonelettersabovecouldthusbereplacedwithdigits,asin[a51],[a31]and[a33].

2.2.5 The glottal stop

It is possible during phonation to suddenly close the glottis, hold that closure briefly and then, equally suddenly, allow the vocal folds to vibrate again. The resulting speech sound is known as the glottal stop,symbolizedby[ʔ].Aglottalstoprequiresvoicingononesideonly;hence,itcanbeinitial,asinHawaiian[ʔu]‘tomoan’,andfinal,as inTsou[suʔ] ‘to fall’.Englishhas theparalinguistic item[ʔm ʔm ],withthefirst[m]higherinpitchthanthesecond,meaning‘no’,usedininformal settings. InGerman, [ʔ] occurs before the second syllable ofBeamter ‘civilservant’([bəʔamtɐ]).

Tosummarizesection2.2,thelarynxproduceseithernosound(openglottis)ora phonated sound. In the former case, the segment is said to be voiceless, and the acoustic energy created for the speech sound lies at a location above the larynx. In the other case, the phonation is either voice or whisper. Three types of voice were distinguished, normal voice, commonly known as modal voice, breathy voice and creakyvoice.Finally,abriefclosurebeforeoraftervoicingrepresentsaglottalstop.In the next section, we consider the ways in which an acoustic source can be created at a location above the larynx, as well as the ways in which the phonated laryngeal sound source can be modified by differences in the shape of the supralaryngeal cavities, which together form the vocal tract.

The production of speech24

2.3 THE VOCAL TRACT

Thevocal tract extendsall theway to the lips (seeFigure2.4). It consistsof thepharynx and the mouth, to which an extra tube extending to the nostrils may be coupled, the nasal cavity.This (potentiallybifurcating) tube acts as a resonator,modifying the sound produced at the glottis. The sound produced by a vibrating glottiscanbemodifiedbychangingthepositionofthetongue,jawandlipssoastoproducearangeofdifferentvowelqualities.Second,insidethevocaltractthereare further opportunities for generating sound, which in their turn will be modi­fiedbytheshapeofthevocaltractinfrontofthesoundsource.Forinstance,thetongue can be brought up against the roof of the mouth to form a constriction that generatesfrictionwhenairispassedthroughit,asduring[s]and[x].Inthissection,wedescribethethreepartsofthevocaltractmentionedabove,andinsection2.4

Figure 2.4Cross-sectionofthevocaltract.

NASAL CAVITY

O RA L C AV IT Y

rootbackfrontcrown

upper teeth

lower teeth

larynx

epiglottis

oesophagus

uvula

soft palate (velum)

alveolar ridge

hard palate

PHA

RYNX

The production of speech 25

wedealwith thepositions that the tongue, jawand lipsmayassumetoproducedifferentvowels.Insection2.5weidentifythevariousplacesatwhichlanguagesmake articulatory constrictions. In the same section, we will classify the types of constriction that are used.

2.3.1 The pharynx

The pharynx is the vertical part of the tube extending up from the larynx to the velum. The forward wall is formed by the root of the tongue, which faces the back wall of the pharynx.

2.3.2 The nasal cavity

The soft palate, or velum, is a valve which closes off the entrance to the nasal cavity when it is pressed up, but opens the cavity when it is allowed to hang down, as in ordinarybreathing.(Whenwehaveacold,theentrancetothenasalcavitymaybeblocked by mucus, which forces us to breathe through the mouth, and hence to pro­duceonlyoralsounds.)Thevelumendsinapear-shapedlittleblobofflesh,whichcanbeseenduringthespeaker’sarticulationof[aː].Itiscalledtheuvula.

2.3.3 The mouth

The mouth is the most important part of the vocal tract because it is here that the mostdrasticmodificationsofitsshapeareachievedandthemajorityofthearticu­latorycontactsaremade.Theroofofthemouthisformedbythesoftpalate,withthe uvula at the extreme end, and the hard palate,whichliestothefrontofthesoftpalate.Withacurled-backtongue,itispossibletofeelthehardpalatearchingbacktowherethesoftpalatebegins.Immediatelybehindthefrontteethisthealveolar ridge, which is touched by the tongue during the pronunciation of dada; then there are the upper front teeth themselves and the upper lip.Belowtheseparts thereare the lower lip, the lower front teeth and the tongue, of which the lower lip and tongue are active articulators. The zone immediately behind the tip of the tongue is called the blade.Youuseitwhenimitatingthesharp,hissingsoundofasnake.Together,tipandbladearecalledthecrown,whichtermisfromClements(1985).The part of the tongue opposite the hard palate is called the front, the part opposite thesoftpalate iscalled theback. The section comprising both front and back is known as the dorsum.

2.4 VOWELS

Forvowels,thevocalfoldsvibrate.Thecrownisheldbehindthelowerteeth,whilethe dorsum is bunched, forming a constriction that allows a frictionless escape of the air.Thelipsmayberounded,causingthevocaltracttobelengthened.Vowelsmayhave a more or less stationary configuration of the speech organs or be characterized byatrajectoryofthetongueorlips.Moreover,theymaybeoralornasalized.

The production of speech26

2.4.1 Monophthongs

Avowelwhosequalityremainsstableduringitsproduction,like[æ]inEnglishhat or[uː]inGerman[guːt]‘good’,isknownasamonophthong.Differentvowelquali­ties are produced by different tongue or lip positions. The location of the tongue bunch can be varied in a vertical dimension as well as a horizontal dimension. IntheBritishEnglishphonetictradition,fourstepsarerecognizedintheverticaldimension: close, close-mid, open-mid and open.IntheAmericanEnglishtradi­tion, the three terms high, mid and low are used to distinguish three vowel heights. Horizontally,thebunchcangofromfront to back, with central being used as an in-betweenvalue.Ifwedisregardtheopenvowel[ɑ],mostlanguagesjusthavefrontunrounded and back rounded vowels. Italian, for instance, has the vowels shown in(1).Thepositionofthelipsisroundedfor[u],[o]and[ɔ],andunroundedfor[i],[e],[ε]and[a].(Notethat,deviatingfromIPAconventions,authorsnormallyusethetypographicallysimplersymbol[a]torepresentacentralorcentraltobackopenunroundedvowellike[ɑ].Wehavefollowedthispracticeinthisbook.)

(1) Frontunrounded Backunrounded Backrounded

High i u

Mid e o

Low ε a ɔ

Ifthetonguepositionsusedfor[i],[e]and[ε]arecombinedwithroundedlips,thevowels[y],asinFrench[lyn]‘moon’;[ø],asin[pø]‘(a)little’;and[œ],asin[sœl] ‘alone’ are produced.A rounded [ɑ] is [ɒ], and may occur in dog as pro­nouncedinvarietiesofEnglishspokeninEnglandandontheeastcoastoftheUSA.Unroundedvowelswiththebunchinthebackorcentreexistalso.Theunroundedcounterpartof[ɔ]is[ʌ],thatof[o]is[ɤ],andthatof[u]is[ɯ].

Roundedfrontvowelsaresomewhatmorecentralthanunroundedones,whileunrounded back vowels are somewhat more central than rounded back vowels. Frequently,of the lowvowelsonly thebackmember isever rounded,andmanylanguagesavoidlowroundvowelsaltogether.ThevowelsofconservativeKorean,givenin(2),illustratetheavoidanceoflowroundvowels.(Inthemodernvarieties,therearesevenvowels.Thevowel[æ]hasmergedwith[ɛ]formanyspeakers,while[y]and[ø]havebecome[wi]and[we]inthespeechofmostspeakers.)Examplewordsaregivenin(3).

(2) Frontunrounded Centralizedfrontrounded

Centralized backunrounded

Backrounded

High i y ɯ uMid ɛ ø oLow (æ) a ʌ

(3) gi ‘era’ gy ‘ear’ gɯ ‘he’ gu ‘sphere’gɛ ‘crab’ kø ‘craftiness’ no ‘paddle’gæ ‘dog’ na ‘I’ nʌ ‘you’

The production of speech 27

Whileacentralpositionofthetonguebunchisthusquitecommon,centralvowelspattern like back vowels if they are unrounded, and like front vowels if they are rounded.

2.4.2 Diphthongs

Whentwodifferentvowelsappearinthesamesyllable,thecombinationisknownas a diphthong.Englishhasdiphthongsin[lai]lie, [naʊ]now and[dʒɔɪ]joy, while Germanhasthediphthongs[ai],[ɑu]and[ɔi]in[tsait]‘time’,[hɑus]‘house’and[ɔiç]‘you(obj pl)’,respectively.

2.4.3 Nasalization

Vowelscanbenasalized:ifduringtheirproductionthesoftpalateislowered(sothatthenasalcavityisopenedup),anasalqualityisaddedtothem.Nasalizedvow­elsoccurinPortugueseandFrench,andaresymbolizedbyplacingatildeoverthesymbol.Thus,Frenchhasthreenasalizedvowels,occurringin[vɛ]‘wine’,[tãt]‘aunt’and[ɔt]‘shame’.

Q16 Are there any close, close-mid or open-mid front vowels in the English phrase I’m not here to make friends? Any velar sounds? Any fricatives whose friction is produced upstream from the larynx?

The acoustic structure of vowels is determined by the resonances of the air in the vocal tract which modify the glottal waveform. These resonance frequencies or for­mantsarecountedfromlowesttohighest(formant1,orF1,isthusthelowest).F1corresponds to the degree of opening of the oral cavity. It is low for close vowels like [iyu]andhighforanopenvowellike[a].F2approximatelydependsonthedistancebetween the lips and the main constriction in the oral cavity: a forward constriction asfor[i]resultsinahighF2,whileavelaroruvularconstrictionresultsinalowF2.TheeffectofF2issimilartotheresonanceheardinabottleunderatap:asitfillsuptowards the beginning of the neck and the column of air gets shorter, the resonance causedbythesplashingwaterishigher.F1isratherrelatedtothesizeoftheopen­ingofthebottle:athinneckcreatesadullereffect(lowF1,asforclosevowels)thanawideneck(highF1,asforopenvowels).Inthenextchapterwewillseehowthedistance between the lips and the constriction is a determining factor in explaining the frequencies of occurrence of vowel sounds in the languages of the world.

Youcanhear theeffectofyoursecond formant independentlyof the laryngealsourcesoundwhenyouwhistle.Highnotesareproducedwith the tongue in thepositionfor[i](or,better,[y],becauseoftheliprounding),lownoteswiththetongueinthepositionfor[u]andintermediatenoteswithintermediatetonguepositionsforhighvowels,like[ʉ].Youcanhearyourfirstformantifyoucloseyourglottis(seeQ13)andflickafingeragainstthesideofyourthroatgoingfrom[i]via[e],[a],[o]to[u],oncepervowelposition.Figure2.5plotsthefirsttwoformantfrequenciesofamalespeakerofthesevenItalianvowelsin(1).ThehorizontalaxisrepresentsF2,withitsoriginontheright,whiletheverticalaxisrepresentsF1,withitsoriginatthe

The production of speech28

top.Bothaxesarenon-linear,withhighervaluesbeingmorecompressedthanlowervalues.Byarrangingtheaxesinthisway,thevowelsgettobespatiallyarrangedasinaconventionalimpressionisticvoweldiagramorvoweltrapezoid(seethevoweltrapezoidintheIPAchartonpageix)orasindiagramslikethosein(1).Incidentally,youmaynowseewhypeoplestartedcallingthefundamentalfrequencythe‘f0’.

Becauseoftheshorterlengthandsmallerwidthofachild’svocaltract,theformantfrequencies of vowels produced by children are higher than those of adults, particu­larlythoseofF2.Likewise,women,whosevocaltractsareapproximately15 cmlongascomparedto17.5cminmen,havehigherformantfrequenciesthanmenforthe‘same’vowels.Listenersnormalizeforthosedifferencesafterextractingtheinforma­tion about the sex and maturation of the speaker from the speech waveform.

2.5 CONSTRICTIONS

Varioustypesofconstriction can be made in different locations in the vowel tract. Wewillfirstgothroughthedifferentplaces,andthendiscussthedifferentkindsofconstrictionthatexistforconsonants.Theexpression‘toarticulatewithX’isusedtomean‘toformaconstrictionatX’.

Figure 2.5 PlotsofF1andF2ofsevenItalianvowelsasspokenbyamalespeakerinaspacewithinverted,logarithmicaxes.BycourtesyofAnttiIivonen,http://www.helsinki.fi/speechsciences/projects/vowelcharts/.

The production of speech 29

2.5.1 Places of articulation

Pharyngeal

Therootofthetonguearticulateswiththebackwallofthepharynx.GulfArabichasavoicelesspharyngealfricative,asin[laħam]‘meat’.

Dorsal

The dorsum articulates with the roof of the mouth. If it is the back that articulates with the softpalate, the termvelar is used.Thisplace is used for [k] inFrenchquand ‘when’,for[ŋ]inEnglishhang, for[g]inEnglishgood andfor[x]inScot­tishEnglishloch. If it is the front which articulates with the hard palate, the term palatal or palatoalveolar isused.Examplesare[j] inSpanishyo ‘I’, [c]inDutch[ˈkacjə]‘cat+dim’or[ʃ]inEnglishship. These types of segments also involve a rais­ingofthecrown,andarethereforetreatedunder‘Coronal’below.Frontedvelars,like[ç]inGermannicht or[k]inFrenchqui, should be distinguished from palatals. Retractionofthedorsumallowsthebackofthetonguetoarticulatewiththeuvula.Avoicelessuvularfricative[χ]occursinWesternDutch,asin[χeːl] ‘yellow’.Theuvularstops[qg]occurinTlingit,forinstance.

Coronal

The crown may articulate with the upper teeth, the alveolar ridge or the forward partofthehardpalateimmediatelybehindthealveolarridge.Whenthecrownarticulates with the upper teeth, dentalsegmentsareproduced,like[θ]inEng­lish thing and[ð]inthis and that;thedentalplosives[td]occur,forinstance,inSinhalese[tadə]‘hard’.Thelabelalveolar is used if the crown articulates with the alveolarridge.Examplesare[t]and[d]inGerman[thuːn]‘do’,[duː]‘you’,[n]in[naxt]‘night’or[l]asin[ˈaləs]‘everything’.Ifthecrownarticulateswiththerearedge of the alveolar ridge, a postalveolarconsonantisproduced.English[ʃ]inshore, [ӡ]inmeasure, [tʃ]inchip and[dӡ]injet are articulated with the crown of thetongue,whilethefrontofthetongueisraisedtowardsthehardpalate.Often,asinDutch,thetipisheldbehindthelowerteethforthistypeofconsonant,inwhich case the contact is alternatively labelled prepalatal.Apostalveolararticu­lationwithjustthetipofthetongueoccursinEnglish[tɹ]asintry. If this type of contact is made with the tongue tip or the underside of the tongue blade with the tongue tip curled back, the term retroflexisused.Likemanylanguagesspo­keninIndia,Hindihasretroflexconsonantslike[ʈ ɖ ɳ],asin[thəɳɖi]‘cold’and[gɦəɳʈa]‘hour’.

Labial

If the lips articulatewitheachother,asinEnglish[pbm]inspot, bell, mad, the place is bilabial. If the lower lip articulates with the upper teeth, the place is labiodental. Itisusedfor[f]inGerman[fiː]‘cattle’orFrench[fεʁ] ‘do’.

The production of speech30

2.5.2 Types of constriction

Afirstsubdivisiondistinguishesbetweentwokindsofconstriction.

1 A constriction that is tight enough to lead to frictionwhena (voicedorvoiceless)airstreamispassedthroughit,asusedwithobstruents.

2 Atypeofconstrictionthatallowsavoicedairstreamtopassthroughwith­out friction, which is used for sonorants.

In the case of obstruents, an acoustic source is actually created at the point of articulation:eitherapoppingsoundisproduced(forplosives),orfrictionispro­ducedat thatspot (for fricativesandaffricatedplosives).Theauditoryqualityofsonorants relies exclusively on the different shapes the vocal tract is given, i.e. on the resulting modifications of the acoustic characteristics of the sound produced by phonation in the larynx.

Obstruentsaresubdividedintoplosives(alsocalledstops),fricatives and affricates.

Plosives

These are formed by creating a complete closure at some point in the speech tract, behind which the air from the lungs is compressed until the closure is abruptly releasedsothattheairexplodesoutwards.Sincethesoftpalateisraised,theaircan­notescapethroughthenasalcavity.ExamplesareFrenchvoiceless[ptk],asin[pip]‘pipe’,[tip]‘type’,[ekip]‘crew’.VoicedplosivesoccurinFrench[bide]‘bidet’and[gã]‘glove’. Plosives have a very brief friction burst when they are released, which is not usually heard as friction but is responsible for the popping quality of plosive releases.

Fricatives

These are formed by narrowing the speech tract to such a degree that audible fric-tion is producedwhen air passes through. English has the voiceless labiodentalfricative[f]infee, thevoicelessdentalfricative[θ]inthigh, thevoicelessalveolar[s]in sigh andthevoicelesspalatoalveolar[ʃ]inshy. Thevoicedcounterparts[vðzӡ]occur in vie, that, zoo and measure, respectively.(Atthebeginningofthesyllable,English[vð]arefrequentlypronouncedwithoutfriction.)

Affricates

Affricates areplosiveswhose release is slow insteadof sudden, causing a longerphaseofturbulence.Theaffricates[pf]and[ts]occurinGerman[ˈpfaifə]‘pipe’and[tsait] ‘time’,andthepalatoalveolaraffricates[tʃ]and[dӡ]occurinEnglishcheer and jeer, respectively.

Sonorants

These divide into nasals and approximants.

The production of speech 31

Nasals

Fornasals thesoftpalate is lowered,andtheoralcavity isblockedcompletelyatsomepoint.Aslow,deliberatepronunciationofmorning will allow us to observe how each of the three nasal consonants in the word has a different place of articu­lation:abilabial[m],analveolar[n]andadorsal[ŋ].(Pre-)palatal[ɲ]occursinDutch,asin[ˈspɑɲjə]‘Spain’,andinFrench,asin[aɲo]‘lamb’.

Approximants

Approximantsderivetheirnamefromtheapproximationofthearticulators,whichgives rise to a light or near-contact, the airstream being so weak that no friction is produced.The(pre-)palatalapproximant[j]occursinEnglishyes, while a bilabial one([β])occursinSouthernDutch[βat]‘what’.(Whencombinedwithasymbolforavoicedfricative,thesubscript[ ]indicatesanapproximant,i.e.frictionlesspronun­ciation.)Frequently,languageshaverhotics,or[r]-typesegments,thatareapprox­imant, such as the postalveolar approximant [ɹ] inEnglish ray. Also, trills occur, during which the uvula is allowed to vibrate against the back of the tongue [ʀ],as inFrench[paʀi]‘Paris’,orthetonguetipagainstthealveolarridge[r],asinSpanish[ˈpεro]‘dog’.When,insteadofaseriesofvibratorytaps,asinglesuchbriefcontactismade,aflapisproduced([ɾ]).InSpanishandCatalanthealveolarflapcontrastswithatrill,aminimalpairinCatalanbeing[ˈpaɾə]‘father’–[ˈparə]‘grapevine’.InArawak,thealveolarflapcontrastswitharetroflexflap,whichinvolvesflickingthecurled­back tongue forward, causing the tip to hit the rear edge of the alveolar ridge, asillustratedin[hoɾòɾo]‘swampy’–[hòɽoro]‘cloud’(Pet1979),whileTodacontraststhreeplacesofarticulationfortrills(orflaps),postdental,alveolarandretroflex,asin[kar]‘borderofcloth’,[kar]‘juice’and[kaɽ]‘penforcalves’(Spajićetal.1996).

For[l],asinGerman[ˈaləs]‘everything’,theairstreamispartlyblockedbythetongue tip contact with the alveolar ridge, but allowed to escape freely on one or both sides. Sounds which have this type of partial occlusion are called lateral. Because the air escapes without friction through the lateral opening, the Ger­manlateral isanapproximant.Thevoiceless lateral fricative([ɬ])hasaturbulentescapeofairalongthelateralopening(s);itoccursinWelsh,asintheplacename[ɬanˈdɪdno](Llandudno).Lateralsareusuallyalveolar:thecrownarticulateswiththealveolar ridge.Apalatal (approximant) lateral [ʎ]occurs in Italian [ˈzbaʎːo]‘mistake’orCatalan[ʎop]‘wolf ’.

2.6 SEGMENTAL DURATION

Inmanylanguages,durationcontrastsexistinthegroupofvowels,asinCzechandHawaiian,orinthegroupofconsonants,asinItalianandTamazight,orinboth,asinFinnishandJapanese.Longconsonantsarealsocalledgeminates. Italian and Japanese have long consonants in intervocalic position, while other languages may additionallyhavetheminword-initialandword-finalpositions,likeSwissGerman

The production of speech32

(Kraehenmann2001).An illustrationof suchaquantitycontrast is Italian[fato]‘fate’versus[fatːo] ‘fact’.Frequently,onlyasubsetof theconsonantsoccursbothlongandshort.Thus,only[ptks]occuraslongconsonantsinthenativevocabu­laryofJapanese.Vocalicquantitycontrasts,suchthatexistinHawaiian,aremorecommon.Asinthecaseofconsonantaldurationcontrasts,sometimesonlyasubsetofthevowelsoccursbothlongandshort.Thus,Chipewyanhasfiveshortvowels ([ieaou]),butonlythreeofthemhavelongvariants([iː aː uː]).Moreover,thequalityofthelongvowelsmaydiffersomewhatfromthatoftheshortvowels.Finn­ish, for instance, has eight monophthongal vowels appearing both short and long, butlong[eː]hasacloserqualitythanshort[ε].

InGerman,thedurationalcontrastcoincideswithadifferenceinthepositionofthe root of the tongue, which is advanced somewhat during the pronunciation of thelongvowels.Suchvowelsareknownastense,whiletheirunmodifiedcounter­partsarelax.Asaresultofthewideningofthepharynx,thetonguebodywilltendto be higher for tense vowels than for lax vowels, at least for vowels that are not fully open.In(4),whichgivesthevowelsystemofStandardGerman,thesymbolontheleftineachboxisalax,shortvowel,whilethesymbolontherightrepresentsalong,tensevowel.Examplesofwordswiththesevowelscanbefoundonpage81.

(4) Front unrounded

Centralizedfrontrounded

Centralizedbackunrounded

Backrounded

High i iː yyː ʊuː

Mid εeː œøː ɔoː

Low (æː) a aː

InmoreconservativeGerman,thereisalsoalonglaxvowel[æː] (bracketedin(4)),whichhasmergedwith[eː]inthespeechofmanyspeakers.

2.7 COMPLEX CONSONANTS

Complexconsonantsaresinglesegmentswhichinsomewayhavetwodistinguish­ablearticulations.Wedistinguishconsonantswithasecondaryarticulation,conso­nants with a double articulation, and manner­contour consonants.

2.7.1 Secondary articulations

The articulation of a consonant does not require the services of the entire tongue as wellasthelips.For[td],forinstance,onlythecrownandthesidesofthetongueareused,andfor[fpm]onlythelips.Infact,thelipsandtonguecanbeusedtoforma vocalic articulation simultaneously with the production of the consonant. The result is what is known as a consonant with secondary articulation. The following types of secondary articulation can be distinguished:

1 Labialization. During the articulation of the consonant, the lips arerounded.Alabializedvelarplosive,forinstance,issymbolized[kw].

The production of speech 33

2 Palatalization.Thefrontofthetongueisraised(asfor[i]or[j])duringthepronunciationofaconsonant.Apalatalizedbilabialplosive,forinstance,issymbolized[pj].

3 Velarization.Thebackof the tongue is raised (as for [u]or [ɯ])duringthepronunciationofaconsonant.Avelarizedalveolarlateralapproximant(‘darkl’),symbolized[ɫ],isusedpostvocalicallyinmostvarietiesofEnglish,as in all.

4 Pharyngealization. The root of the tongue is retracted towards the back wallofthepharynx.Arabic‘emphatic’consonantsarepharyngealized.Thisis indicatedbya superscriptvoicedpharyngeal fricative symbolafter theconsonantsymbol(e.g.[tˤ])orbyadotplacedbelowthephoneticsymbolconcerned(e.g.[s]).

2.7.2 Double articulations

Someconsonantshave two consonant-type constrictions at the same time, suchthat it is not possible to say which is ‘primary’ (i.e. consonant-like) and which‘secondary’(vowel-like).Well-knownexamplesare[kp]and[gb],labial-velarplo­sives,whichoccurinmanyNiger-Congolanguages.TheEnglishconsonant[w]isalabial­velar approximant: it is pronounced with a raising of the back, as well as with roundingofthelips.Suchconsonantsaresaidtoinvolveadouble articulation.

2.7.3 Manner-contour consonants

Someconsonantschangetheirconstriction-typehalf-waythrough.Anexampleisprovided by prenasalized stops,whichoccurforinstanceinBantulanguages.Suchconsonants begin like nasals and end like plosives at the same place of articulation ([mb],[nd],[ŋg]).

2.8 NONPULMONIC CONSONANTS

Webrieflydescribethethreeclassesofnonpulmonicsegmentshere.

1 Clicks.Clicksareproducedwiththehelpofavelarclosure(asfor[k])plusaclosure somewhere further forward. This forward closure may be located at thelips(rare)orat,orimmediatelybehind,thealveolarridge.Thetrappedair in the pocket in front of the velar closure is rarefied by lowering the body ofthetongueor,inthecaseofabilabialclick,thejaw.Whentheforwardclosure is released to allow outside air to rush into the pocket of rarefied air, a clacking noise burst results. This way of creating an air pressure differ­ence is known as the velaric airstream mechanism (Abercrombie1967).Clicksmaybecontrastivelyaccompaniedbyaglottalstopatthebeginningof the postclick vowel, by nasalization on the following vowel, by aspira­tionprecedingthefollowingvowelorbyaslow(affricated)releaseofthevelarclosure.Fivearticulationplacesfortheforwardcontactoccur.Their

The production of speech34

symbolsareincludedunder‘Othersymbols’intheIPAchartonpageix.ThealveolarclickisusedparalinguisticallyinEnglishasasignofdisapproval(renderedtut, tut inBritishEnglishandtsk, tsk inAmericanEnglish),whilerepeated lateral alveolar clicks are sometimes used to imitate the noise a horse’shoovesmakeonthepavement.Clicksonlyoccurinlanguagesspo­keninsouthernAfrica,!Xũ,NamaandXhosabeingwell-knownexamplesof such languages.

2 Implosives. Implosives have a closure as for plosives, as well as a closed glottis.Byloweringthelarynx,theairinthemouthandpharynxisrarefied,sothatanoiseburstwillresultwhentheoralclosureisreleased.Because,onits way down, the speaker relaxes his or her glottal closure, phonation will occurastheglottismeetstheairinthetrachea.SuchspeechsegmentsoccurinVietnamese,aswellas insomeNiger-Congolanguages.Theirsymbolsare[ɓɗɠ].

3 Ejectives.Theinitialarticulatoryconfigurationofejectivesislikethatusedfor implosives, but instead of being pulled down, the larynx is pushed up. The air inside the pharynx and mouth is compressed, so that upon the releaseoftheoralclosure,anegressivenoiseburstoccurs.Ejectivesarerea­sonablycommonandaresymbolized[p’t’k’].Implosivesandejectivesareproduced with the glottalic airsteam mechanism.

2.9 STRESS

In many languages, words consist of rhythmic units called feet, the most common foot typebeingdisyllabic.Oneof the syllablesof the foot ismoreprominentorstrongerthantheothersyllable(s)init,whichforthisreasoniscalledthestressed syllable. In city, this is thefirstsyllable.Aword likecelebration contains the two feet[sɛlə]and[breɪʃn].Oneofthefeetinawordhastheprimary stress or word stress. In celebration, this is the last foot. The other feet have secondary stress. The IPAnotationforprimarystressis[ˈ],andforsecondarystressitis[ˌ],tobeplacedbefore the syllables concerned. The stress pattern of celebration contrasts with that of[ˈæləˌgeɪtə]alligator in the location of the primary stress. In addition to disyl­labic feet, there may also be monosyllabic feet, as in cat, and ternary feet, as in origin [ˈɔrədӡɪn].Inmanypublicationsonthistopic,[′]isusedoverthevowelforprimarystress,and[ ′] for secondary stress. In this book these accent marks are used only for tone(section2.2.4andchapter10).

2.10 CONCLUSION

This chapter has outlined the workings of the speech production mechanism in a waythatwillenableyoutofollowthediscussionintherestofthisbook.Formanyusers of the book, it will have served as a brief refresher course, while for others, who may be new to the topic, it will have served as an introduction.

The production of speech 35

NOTE

1 Thereisanadditionalvalve,calledtheepiglottis,positionedabovethelarynxwheretheroot of the tongue begins. It normally points upwards, but it flaps down to channel food and saliva into the oesophagus – the tube behind the larynx leading to the stomach – when we swallow.

Some typology: sameness and difference

3.1 INTRODUCTION

The phonologies of different languages are in many respects very similar, to the extent that some features appear to be part of every language. These cross­linguistic similaritiesareduetotwosources.Oneisthestructureofthehumanbrain.Eventhough we don’t understand the workings of this complex neurological organ, it seems obvious that the physiological basis for the hierarchical structure of language istobelocatedinthewaythebrainoperates.Ourawarenessofalinguisticexpres­sion like Many pens leak will concern simultaneous structures at different levels in each of two structural hierarchies: it is a sentence, and there are two phrases, threewords,foursyllables,astringof11segments,andaprosodicstructure,whichwillbesimilarlyhierarchical.Whenviewedasstructural layers, linguistichierar­chies must be processed for production and perception purposes with parallel clock times at different frequencies, one for segmental articulations, one for stress, and soon(Poeppel2003).Otheraspectsaremoreclearlyduetotheergonomicsofthespeech process. Language users, and thus languages, prefer distinctions that areeasy toperceiveandeasy toproduce.Thedifferencebetween[t]and [n] isverycleartotheperceiverandnottoodifficulttomakefortheproducer.Thecontactmade by the tongue tip and rims with the upper gums is the same, while the velum isloweredfor[n],openingupthenoseattheback,andraisedfor[t],trappingtheair behind the oral closure. In the open­velum position, very little effort is needed to bring the vocal folds somewhat closer together than during breathing in order for them to start vibrating. This is so easy that voicing during sonorant consonants and vowelshasbeencalled‘spontaneousvibration’.Bycontrast,ifweblocktheegressiveflow by closing off both the mouth and the nose, it will take some effort to create a sufficientlypowerfulairstreamthroughtheglottistomakethemvibrateatall:wearepumpingmoreandmoreairintoasmall,closedpocketofair.Speakerswouldbe well advised therefore not to be too eager to vibrate their vocal folds while their vocal tract is significantly obstructed, because maintaining an air pressure differ­enceacrosstheglottistakessomeeffort.(Youcanincreasethepressurefromthelungs,ormakemoreroombypullingupthevelumfor[dg]orblowingoutyourcheeksfor[b].)Thatiswhyalmostalllanguageshaveadental/alveolar[t]and[n],but only 64%have the voiced counterpart of [t] ([d]), and fewer than 0.5% thevoicelesscounterpartof[n]([n]).1 It would be extremely improbable to find a lan­guagethathadno[tn]butdidhave[dn].Anotherspeakerinterestistoduplicatecontrasts.Ifalanguagehasthevowels[ieaou]andyoufindithasanasalized[õ],itwouldbeagoodbettoassumeitalsohas[ã],whilethepresenceof[e]isalmostas

3

Some typology: sameness and difference 37

likely. The speaker’s phonetic routine of nasalization is exploited so as to maximize contrastswiththesamevelum-loweringgesture.Butheretoothehearerisnotfor­gotten.Forhighvowelslike[iu],nasalizationdoesnothaveawholelotofacousticeffect,andmanylanguages,likeFrench,thereforeleavethemoutoftheirsubsystemofnasalvowels(‘systemgaps’).

Low-costcontrastsarethusfrequent.Somesoundcontrastsaresoeasytomakeand so clear to the ear that they are found in every language that has been described. For instance, no languagehas been foundwithout vowels, andno languagehasbeenfoundwithoutconsonants.Therearetworesponsestothisstateofaffairs.Oneis to assume that whatever languages have in common is universal and that the explanation for the universality is that this is hard­wired in our brains, innate. The other is to assume that the neurological, physiological, physical and social condi­tions under which languages arose and developed are the same across our species and the forces that determine their structure are the same. Those forces must allow for a fair number of degrees of freedom in order to explain the variation that is seen. Apparently,aftergoingfortheobvioussoundcontrasts,whichoccurinthegreatmajorityoflanguages,weighingupthecosttothespeakerandthebenefitforthehearermayleadtoalargenumberofoptions.Also,thereareapparentlyotherfac­tors that explain why some languages have many sound contrasts, and others few. The only factor that we can at this point be certain of is the historical dimension: the phonological complexity of languages changes very little in its transfer from onegenerationtothenext.OldEnglish,forexample,hadbroadlythesamelevelofcomplexityasmostcontemporaryvarietiesofEnglish,eventhoughithaschangedvirtuallybeyondrecognitionaftersome1,500yearsofdevelopment.

Q17 Why is [iea aoõu] an implausible vowel system?

Q18 Assume that all the languages that were ever spoken in the world have at least three vowels. What explanation would be given of this fact by someone who rejects the theory that human brains are genet-ically programmed to have at least three vowels?

3.2 VARYING COMPLEXITY

Languages differ in theirmorphosyntax just as they differ in their phonologies.Tobeginwith,theydiffergreatlyinthenumberofsegmentstheyhave.InUPSID(seenote1),acorpusofsegmentinventoriesof451languages(approximately7%ofallthelanguagesoftheworld),thesmallestnumberappearstobe11(e.g.Roto­kas,spokeninPapuaNewGuinea)andthelargestastaggering141(!Xũ,spokeninNamibiaandAngola)(Maddieson1984:9).Andwhentwolanguageshavethe

Some typology: sameness and difference38

samenumberof segments, theyareunlikely tohave identical sets.Another lan­guagethat,likeRotokas,has11segmentsisPirahã,whichisspokeninColombia(Everett1986).Itshares[ptkgioa]withRotokas,butwhileRotokashas[eu]andtheconsonants[β]andaflap,Pirahãhas[ʔbsh].

Second,differentlanguageswillhavedifferentconstraintsonthewaysegmentsarecombinedtoformsyllables.Theinitialconsonant(s)ofthesyllableareknownas the onset, the vowel is in the peak,andtheclosingconsonant(s)formthecoda. Codaandpeakformaconstituentcalledtherhyme(alsospelledrime).In(1),weshow these constituents in a tree diagram.

(1)

RHYME

CODAPEAKONSET

-CVC-

(1)

AccordingtoBlevins(1995),thelowestdegreeofcomplexityinsyllablestructureisrepresentedbylanguagesthathaveasingle(short)vowelinthepeakandallowmaximally one consonant in the onset. The syllable structure of such languages is (C)V.Furthercomplexitycanbeachievedinanumberofways:

1 Theonsetmaybeobligatory.Languageswithobligatoryonsetsarenothardtofind,likeArabic,DyirbalandKlamath,butlanguagesthatallowonlyCV,i.e. an obligatory onset preceding an obligatory monomoraic vowel, must berare,totheextentthatwehavenoexample.Analyticalissuesmayarisefromphoneticonsetsinonsetlesssyllables.Manylanguages,likeGerman,regularlyhaveaglottalstopinonsetlesssyllables,asin[ˈafə],pronounced[ʔafə],‘monkey’,whileinMbaandNdungaword-initialhighvowelsappeartobevariablyprecededby[ʔ],[ɦ]orahomorganicsemivowel(Pasch1986:32,91).HenanMandarinhasahomorganic semivowelprecedinganon­lowonsetlessvowel,asin[i214],pronounced[ji],‘one,numeral’;[y41],pro­nounced [ɥy], ‘fish’; [u341], pronounced [wu], ‘house’; [ɤ341], pronounced[ɰɤ], ‘hungry’; and a glottal stop before low vowels, as in [æɛ341], pro­nounced [ʔæɛ], ‘love’ (personal field notes, CG, 2014). Phonemic glottalstopsareeasilydiagnosedinlanguageslikeHawaiian,wheretheycontrastwithemptyonsets,asin[aa]‘jaw’and[ʔaa]‘fiery’(ElbertandPukui1979),orArabic,whichhasthesamecontrastinthecodaofthesyllableandhastheglottalstopintheonsetwithinwords,asin[lawʔa]‘sorrow’(ThelwallandSa’adeddin1990).Germanhasnoneofthesethreeproperties.

2 Theremaybeanoptionalcoda.Languagesthatdon’tallowcodasincludeHawaiian,BakwiriandFijian.Ifthelanguageallowsacoda,afurtheroptionis that the coda may be complex, i.e. may be a cluster containing more than oneC. Spanish, Japanese and ItalianallowonlyoneC,butKlamathandFrenchallowtwo.

Some typology: sameness and difference 39

3 Theonsetmaybecomplex.SomelanguagesallowonlyoneCintheonset,likeFinnish,HawaiianandKlamath;othersallowtwoormore,asinDutch,Spanish andTaba (BowdenandHajek1996). Strikingly, onsets aremorecomplexthancodasinallowingmorephonologicalcontrasts.Forthisrea­son, the position of the onset has been characterized as ‘privileged’ relative tothecoda(Beckman1988).

4 Thepeakmaybecomplex,i.e.beVV.Thisisthecaseif,bythesideofshortvowels, the languagehas longvowels (inwhichcase thepeak isfilledbyadoubleoccurrenceof the samevowel,ViVi) and/orhasdiphthongs (inwhichcasethepeakisfilledbytwodifferentvowels,V1V2).Spanishhasthediphthong[ue],butnolongvowels,whileWolofhaslongandshortvowelsbutnodiphthongs.Dutch andFinnishhaveboth long vowels anddiph­thongs.DinkaandEstonianarerareexamplesof languageswitha three-wayquantitycontrastforvowels;i.e.theyhavethepeaksV,VVandVVV.

Somelanguagesallowthesyllablepeak,i.e.thestructuralpositionindicatedby‘V’,tobefilledwithaconsonant.Thus,inCzech,theliquids[rl]appearinthepeak,asinthegeographicalnames[ˈbr.no],[ˈvl.ta.va],Brno, Vltava, andAmericanEng­lish has words like girl [grːl]andmountain [ˈmaʊn.tn].ZwaraTamazightallowsallconsonantstobesyllabic,evenvoicelessones,asshownby[i.ʒm.ˈmu.ʃən]‘raisins’,[a.ˈdf.fu] ‘apple’. In the secondword, thenucleus (‘V’) of the stressed syllable isfilledbyavoiceless fricative (Gussenhoven2017).Where languagesagree in thenumber of consonants they allow in the coda or onset, they will still differ in how theseconsonantpositionscanbefilled.InGerman,[kn-ʃn-]arepossibleonsets,asin[ˈknaː.bə]‘boy’,[ˈʃnε.kə]‘snail’,andsoare[ps-ks-],asin[psy.ço.ˈloː.gɪʃ]‘psy­chological’,[ˈksan.tn]‘Xanten’,allofwhicharelackinginEnglish.(German[ts-]asin[tsaːr]‘czar’isusuallyinterpretedasasinglesegment,anaffricate.)Conversely,Englishhas[lj-]inlure, whichisabsentinGerman.Anunexceptionalimplicationalrelationship is that the presence of a complex constituent implies the permissability of less complex constituents. Thus, a language that allows two consonants in the onset will also allow one consonant in the onset.

Third,differentlanguageswillhavedifferentphonologicalprocesses.AprocessthatoccursinBritishEnglishbutnotinGermanispreglottalization, bywhich[ptk]are[ʔp ʔt ʔk]whentheyoccurattheendofasyllablebeforeanotherconsonant,asin ripped, mats, thickness. AndanexampleofaprocessinGermanthatdoesnotexistinEnglishisfinal devoicing, bywhichallobstruents(i.e.plosivesandfricatives)are voiceless when they occur at the end of a syllable. This rule is responsible for the factthatinGermannovoicedfricativeorplosiveappearsattheendofaword,eventhoughvoicedobstruentsoccurintheinflectedformsofsuchwords.Forinstance,thenominativeformfor‘dog’is[hʊnt],whilethegenitiveformis[hʊn.dəs].Manylanguages have a rule of final devoicing, includingPolish,DutchandCatalan.

Q19 The following words, taken from Huisman et al. (1981), illustrate the syllable structure of Angaatɨha (the period separates the syllables):

Some typology: sameness and difference40

kə.mo.ai ‘him’a.ti.ʔə.ɾə ‘thunder’ai.n.tə.ʔo ‘bird type’ma.nji.njai ‘children(Objective)’ta.m.pwai.ʔo ‘lizard type’

1 GivetheAngaatɨhasyllablestructureasaCVformula,placingoptionalelements in brackets.

2 Doesthelanguagehavesyllabicconsonants?

Q20

1 Canyouthinkof(a)asegmentwhichexistsinEnglishbutnotinyourownlanguage,and(b)asegmentthatexistsinyourownlanguagebutnotinEnglish?IfyournativelanguageisEnglish,answerthequestionforanyforeign language you are familiar with.

2 WhatwouldthefinalconsonantoftheGermanwordforKiev be?Why?

3.3 UNIVERSALS AND IMPLICATIONAL RELATIONS

While it isclear thatdifferent languagesmayhaveverydifferentphonologies,atthe same time it is clear theyhavemany things incommon.Fora start, all lan­guages would appear to have syllables, and all segment inventories can be split into consonantsandvowels.Allconsonantinventoriesincludevoicelessplosives;i.e.alllanguageshaveatleasttwoofthethreeconsonants[ptk].Then,therearenear-universals.Forinstance,onlytwolanguagesinUPSID,RotokasandPiraha, have no sonorantconsonants.Or,again,alllanguagesinUPSIDexceptHawaiian(i.e.99%)havesomekindof[t],and90%have[i].Itisalsostrikingthatthecoronalplaceofarticulation is much commoner generally, and also shows more subdivisions, than either the labial or the dorsal places of articulation.

Whenthegroupofmorecommonsegmentsiscomparedwiththegroupofunu­sual segments, there are two observations to be made:

1 Unusualsegmentstendtooccurinlargersegmentinventories.Forinstance,anunusualsegmentlike[kwh],a[k]withroundingandaspiration,typicallyoccurs in languages with large consonant inventories, like that of Igbo, whichhas20pulmonicegressiveplosivesinadditiontothreeimplosives,orHaida,whichhas46consonantsinall.

2 Unusualsegmentstendtobephonologicallymorecomplexthancommonsegments.Forexample,acommonsegment like[k](99.4%) just involvesacompleteclosurebetweenthebackofthetongueandthesoftpalate.The

Some typology: sameness and difference 41

articulationof[k]allowsorgansofspeechotherthanthebackofthetongueto take the line of least resistance, requiring no accompanying actions of thevocal folds(likeaspiration,orvoicingduringtheclosure,oraglottalclosure),ofthelips(liprounding)orofthefrontofthetongue(palataliza­tion).Ontheotherhand,anunusualsegmentlike[kwh]requiresthesameoralclosureas[k],butinadditionhasaspirationaswellasliprounding.Therelationshipisnotabsolute,however.Although[θ],forinstance,isphono­logically simple, it is nevertheless a rare segment.

Whatdothesefactssuggestaboutthephonologicalstructureoflanguage?Thefirstfact suggests that languages ‘build up’ their phonologies in an ordered fashion. It suggeststhatalanguagewillhavesegmentXonlyifitalreadyhassegmentY.Thereareinfactmanysuchimplicationalrelationships,asRomanJakobsonhadtaughttheworldin1928(Jakobson1990:chs.9,10,18).Thus,nolanguagehasavoicelessnasalwithoutalsohavingitsvoicedcounterpart,andnolanguagehas[z]withoutalsohaving [s].Withvery fewexceptions, languagesdonothave front roundedvowels([yø])iftheydonotalsohavefrontunrounded([ie])andbackrounded([u o])vowels.Aswillbeclear,mostoftheseimplicationsaretendential,andtrueonly in a statistical sense.For instance, thepresenceof [f] generally implies thepresenceof[p],butChuaveisanexampleofalanguageforwhichthisisnottrue.Apparently, there isnosuch thingasanabsoluteorder inwhich languagesavailthemselves of the universal phonological resources.

The second fact suggests that one way in which languages construct their seg­ment inventories is by adding elements to already existing segments. If we continue to use the metaphor of ‘building up’ the segment inventory, it is as if you begin by making some choices from a collection of fairly run­of­the­mill segments, and then, as you require more of them, select further elements which you can use to createmoresegments.Forexample,manylanguageshave[ptk],whileasmallernumberhave[ptk]and[bdg].Theselanguagescanbeseentohaveincreasedtheirinventory by adding the element ‘vibrating vocal cords’ to the plosive segments theyalreadyhad.Then,asmallernumberstillhasthese,plus[ph th kh],forinstance.Here,thelanguagecanbeseentohaveadded‘aspiration’tothe[ptk]italreadyhad.(But,again,insteadof‘aspiration’,languagesmayemployotherelementsthatservetoexpandthesegmentinventory[ptk bdg].)

Q21 The mean number of consonants per language in UPSID is 22.8, with a range of 6–95. The mean number of vowels is 8.7, with a range of 3–46. In what respects are the following languages atypical?

1 !Xũhas95consonantsand46vowels.2 Pawaiahas10consonantsand12vowels.3 Haidahas46consonantsand3vowels.4 Norwegianhas22consonantsand19vowels.

Some typology: sameness and difference42

As Jakobson (1968) emphasized in a famous monograph, children learn thesounds of their language in a particular order, regardless of the language, which reflects the scale of complexity implied in this paragraph: the unusual tends to be rare and tends to be acquired late. Thus, the metaphor of ‘building up’ the phonol­ogy in a sense comes true every time a child learns a language.

3.3.1 Plain or special?

AthirdobservationismadebyMaddieson(1984):thenumberofvowelsandthenumber of consonants are positively correlated. That is, when one language has more consonants than another, it is likely that it also has more vowels. This sug­geststhatlanguagesthataremorecomplexinoneareaofthephonology(inthiscase,theconsonantinventory)alsotendtobecomplexinotherareas(inthiscase,thevowelinventory).Itisapparentlynotthecasethatcomplexityinoneareaissomehow compensated for by simplicity in other areas: phonologies differ in com­plexity.Forinstance,languageswithlargerinventoriestendtohavemorecomplexsyllable structures. That is, languages that allow only simple syllable structures like CVtendtohavesmallerinventoriesthanlanguagesthat,likeEnglishandGerman,allowuptothreeconsonantsat thebeginningof thesyllable(cf.English[strɔː]straw, German [ʃtroː] ‘straw’), aswell asmore thanone at the end (cf.English[læmp]lamp). In tone languages, the number of different pitch patterns typically lies between two and six, but the number of such patterns does not appear to be greater in languages with smaller inventories or in languages with simpler syllable structures.UPSIDdoesnotlistpermittedsyllablestructures,orprosodicdistinc­tions,sothefactsherearenoteasytogive.Maddieson(1984:22)does,however,present the total number of possible syllables in a small corpus of languages. The totalnumberisdeterminedby(a)thenumberofdifferentvowelsandconsonants,(b)thepermittedsyllablestructuresand(c)thenumberoftonal(prosodic)dis­tinctions. If complexity in one area of the phonology goes hand in hand with com­plexity in another, then the greater the number of possible syllables in a language, the larger should be the number of segments, and the larger should be the number of permitted syllable structures, as well as the number of prosodic distinctions. In the representative list of languagesMaddiesongives, this is indeed the case.Togivesomeideaofthemultiplicativeeffectofthesethreefactors:thelanguagewith the largestnumberofdifferent syllables inMaddieson’sminicorpus (Thai)has146timesasmanypossiblesyllablesasthelanguagewiththesmallestnumber(Hawaiian).

Q22 In a corpus of 10 languages, the number of segmental contrasts was found to be related to the mean word length expressed as the number of segments in the word, in the sense that the smaller the segment inventory was, the greater the mean word length was (Nettle 1995). Speculate on the cause of this negative correlation.

Some typology: sameness and difference 43

3.3.2 Avoiding complexity

There is another, somewhat more tenuous, relation worth drawing attention to. Unusual segmentsarenotunusual justbecause theyoccur in relatively few lan­guages; they also tend to be less frequent in the languages that have them. That is, as languages increase their phonological resources, they tend to do less with them. Forinstance,Dutchhasfrontunrounded,backroundedandfrontroundedvowels.First,thesetoffrontunroundedvowelscontainsonesegmentmorethaneitherthefront rounded or the back rounded sets. Then, in almost every case, the frequency of occurrence of the front rounded vowels is lower than those of the correspond­ingfrontunroundedorbackroundedvowels.ThefiguresinTable3.1arebasedonthefrequenciesofthevowelsinthe1,000mostcommonDutchwords,expressedas percentages of the total number of vowels. The frequency of occurrence of these vowelsamountsto20%ofallthesegments.Whatcanalsobeseenbycomparingthefiguresinthesecondrowwiththoseofthefirstandfourthrowsisthatlax(short)vowels are more common than tense vowels.

3.3.3 A word of caution

Youmay by now have gained the erroneous impression that it is in fact easytocount thesegments,orsyllables,orprosodicdistinctions, ina language.Wehasten to dispel this notion: it is not easy to count these things at all. There are tworeasonsforthis.Oneisthatlanguagesfrequentlyhavewhataresometimescalled marginal segmentsorpatternsofarrangements(Moulton1962).Thesearerestricted to onomatopoeic words, in which the phonology echoes the meaning of theword,and(recent)loanwords.Forinstance,Dutchhasanumberofsuchvow­els,amongwhich[εː],whichoccursinloans(e.g.[krεːm]‘cream’)andintheono­matopoeic[ˈblεːrə]‘cry,bawl’.ShoulditbecountedasaDutchvowel?Or,again,noDutchwordsbeginwith[fj-],except[fjɔrt]‘fjord’.Is[fj-]aDutchonset,orisitaNorwegianonsetwhichspeakersofDutchhavetakeninstride?Dependingontheanswerstosuchquestions,thenumberofDutchvowelswillvarybetween16and25.Thesecondreasonwhyitisdifficulttocountthesethingsisthatthecount will depend on the analysis.Supposealanguagehasthefivevowelqualities[ieaou].Supposefurtherthattheycanbeeitherlongorshortand,moreover,

Table 3.1Proportionalfrequenciesofoccurrence(%)withinthegroupofDutchvowelsoffrontunrounded,frontroundedandbackroundedvowels(adaptedfromvandenBroecke1976).

Front unrounded Front rounded Back rounded

Tense,short i 9.70 y 1.89 u 2.71Lax,short ɪ 13.53 ʏ 2.33 ɔ 12.12Lax,short ε 22.02Tense,long eː 12.32 øː 0.29 oː 10.23Diphthong εi 9.36 œy 1.79 ʌu 1.70Total 66.93 6.30 26.77 100

Some typology: sameness and difference44

that every combination occurs as a diphthong. The number of different vowels in asenseis5,butinanothersenseitis30.UPSIDwouldinacaselikethisgivethenumberas5,butif,say,thecombinations[ae],[uu],[eo]hadbeenmissingfromthe language, then, paradoxically, the number of vowels would have been given as 27.Becauseofthelargenumberoflanguagesinvolved,thetrendsnotedinUPSIDareunlikelytodependverymuchonhowthesedecisionsarearrivedat.Whatisimportant is that you should not get the idea that it is easy to say what a segment is, and that counting things is no problem.

3.3.4 Speech ergonomics

Clearly, languages somehow monitor the development of their phonologies, andchecksegmentsandinventoriesoffagainsttwoverygeneralguidelines:‘Don’tmakethingsdifficultforthespeaker’and‘Don’tmakethingsdifficultforthelistener’.Thatis, the best systems are those in which contrasts are maximally distinct with the least amountofarticulatoryeffort(Flemming1995;Boersma1998).Thereasonsforspe­cific statistical tendencies may therefore be either articulatory or perceptual. In some cases, thespeaker’sandthe listener’s interestsmaygohand inhand,butoftentheexplanation of the statistical fact lies in either the speaker’s interest or the listener’s:

1 Plosivesaremorecommonthanfricatives.Plosivesrequireabriefclosureofthe oral tract in order for air pressure to build up behind it. This is all it takes tocreateabrieffrictionburstattherelease.Bycontrast,fricativesrequireconsiderable airflow for the full duration of the consonant to keep the air turbulence going and produce audible friction. The speaker’s interest here favours plosives over fricatives.

2 Voicelessplosivesaremorecommonthanvoicedones.Witheveryopen­ing action of the vibrating vocal folds, the speaker releases some air from the trachea into thevocal tract.Because thevocal tract isclosedoffdur­ing the closure phase of the plosive, the air pressure in it will rise until it equals the air pressure below the vocal folds that drives their vibration. Just as we stop pushing the lever of a bicycle pump when the tube is full, so we will avoid a continuation of the vocal­fold vibration during the closure stage oftheplosive.Nosuchergonomicconflictarisesduringvoicelessplosives,therebeingnorequirementforanyairflowduringanopenglottis.Again,the speaker’s interest is at stake here.

3 Voicelessfricativesaremorecommonthanvoicedfricatives.Therelativelylow airflow that results from the release of a rapid series of small air puffs into the vocal tract during vocal­fold vibration conflicts with the require­mentofasufficientlygenerousairflowthroughthenarrowedpassagewherethefricativeisarticulated.Asaresult,voicedfrictionisweakerthanvoice­lessfriction.Arguably,sincevoicelessfrictionisperceptuallymoredistinctfrom the acoustic properties of sonorant consonants and voiced plosives, speaker and hearer interests go hand in hand here.

Some typology: sameness and difference 45

4 Front rounded and back unrounded vowels are less common than eitherfrontunroundedorbackroundedvowels.Whenyouplaceanemptybottleunderarunningtap,theresonanceinthebottlesetupbythejetofairhittingthe bottom of the bottle or the surface of the water already in it increases from lowpitchedtohighpitched.Startingfrom[u],youcanreproducethiseffectbymovingintothearticulatorypositionfor[i].At[u],thelipsarerounded,pouted even, so as to increase the distance between the raised back of the tongue and the aperture at the lips, creating the longest possible distance betweenthem.Butwhensaying[i],wespreadthelips,soastoshortenthedistancebetweentheraisedfrontofthetongueandthelipaperture.Whilethearticulationof[i]correspondstothefulleststateofthebottleandthatof[u]totheemptiestone,frontrounded[y]andbackunrounded[ɯ]representintermediate positions, which are less likely to be used than the perceptually moreextremevowels.Here,thehearer’sinterestisthedecisivefactor.

Easeofproductionismostclearlyseeninthetendencyforparticulararticula­tions to persist. Prenasalized stops, for instance, always have the same place of articulationfor thenasalandtheoralstop,as in labial,coronalanddorsal [mbnd ŋg]: segments like *[mk] or *[nb] have not been attested. Similarly, nasalconsonants in the coda tend to share the place of articulation of the following plosive,asinJapanese[ram.pʉ]‘lamp’,[jon.da]‘read’and[maŋ.gan]‘manganese’.Andmost languageshave rules ofassimilation, which cause some articulatory featureofonesegmenttobetransferredtoanadjacentone,suchaswhenEnglishin ispronounced[ɪm]inin Paris or[ɪŋ]inin Copenhagen. Whileforthespeakerthere are obvious advantages in extending the scope of an articulatory gesture, for the listener it is better to be able to hear differences. Paradoxically, therefore, thereisalsothetendencyforlanguagestoavoidrepetitionofthesamething.Forinstance, many languages that have labialized consonants lack labialized labials. Thus,Bakairíhas [tw kw dw gw]but lacks [pw bw], even though thenonlabializedplosiveseriesshowsnogaps(Wetzels1997a).Likewise,therearemanylanguagesthatallowcomplexonsetslike[pn-kn-]or[pl-kl-]butdisallow[tn-]and[tl-],whichlastcombinationshaveasingleplaceofarticulation.Inchapter9,wewillintroduce and exemplify the ObligatoryContourPrinciple,which is held to be responsible for this avoidance of repetition.

3.3.5 System gaps

Wehaveseenthatthesegmentinventoriesoflanguagestendtobeconstructedasif languages drew on the stock of phonological resources by adding elements to setsofsegments.Byaddingto[p tk] theelement ‘vocal-foldvibration’,youwillproducetheseries[bdg];furtheradding‘aspiration’willproduce[ph th kh],asinBurmese,forinstance,andsoon.Orifyouhavefivevowels,youmayallowthemto be long as well as short, or you may nasalize them, or provide them with pha­ryngealization.(Thisispartoftheanswertothesecretofthe46vowelsof!Xũ.)If

Some typology: sameness and difference46

phonological resources are typically made available per set of segments rather than perindividualsegment,asindeedshownbyClements(2004),youmaywellwonderwhy languages sooftenhavesystem gaps. For instance,Dutchhas thevoicelessunaspiratedseries[ptk]butonlythevoiced[bd].Sowhereis[g]?Dutchisnotalone in having this gap. It is in fact the most commonly occurring gap among the voicedstops(e.g.Czech,Hixkaryana,Thai).And[p]istheleastcommonamongthevoicelessplosives,beingabsentfromlanguagesasdiverseasArabic,Chuave,Dizi,Hausa,VietnameseandYoruba.Intheexamplesgivenhere,thereappearstobearelationbetweenthegapsandefficiency.Thevoicelessplosive[p]isrelativelyinefficientfromthepointofviewofthelistener,becausethestopburst,whichisoneofthemajorcuestothepresenceofaplosive,isofmuchlowerintensityinthecaseof[p]thaninthecaseofotherplosives,duetothelackofaresonatingcavityinfrontofthepointofrelease,wheretheburstiscreated(Stevens1997:494).Thevoiced[g]isrelativelyinefficientfromthepointofviewofthespeaker,becausetherelatively small air cavity behind the velar closure causes the air to accumulate fast below it, increasing the supraglottal air pressure and diminishing the glottal airflow, therebycausingvoicingtostop(Ohala1989).Thatis,[p]isrelativelyhardtohear,and[g] isrelativelyhardtosay.Again,whilethesestatisticaltendenciesareveryclear, languagesmaydeviatefromthem.Thus,Hawaiian,quiteexceptionally,has[pk]butnot[t].Inthiscase,anearlierstageofthelanguagedidhave[ptk],butsomehow[t]wasreplacedwith[k],after[k]hadbecome[ʔ].

Q23 The Zaza variety of Kurdish has the voiceless plosive series [ptckq], but in the voiced series there is one missing consonant (Todd 2002). Which one would you guess this is?

Q24

1 Onaverage,voicelessplosivesaremorefrequentthanvoicedplosives,andcoronal segments are more frequent than noncoronal segments. This holds true both for the occurrence of these segments in the inventories of the lan­guagesoftheworld,andfortheiroccurrenceinthewords(ortexts)ofanyindividuallanguage.Torepresentthissituationgraphically,drawasetoftwocoordinates, with frequency of occurrence on the y­axis and the three places of articulation labial, coronal and dorsal onthex-axis.Drawtwotheo­retical graphs, one for the voiceless plosives and one for the voiced plosives.

2 English[ptk]havefrequenciesof1.78,6.42and3.09,while[bdg]havefrequenciesof1.97,5.14and1.05(Gimson1989:219).Inasetoftwocoor­dinates, with frequency of occurrence along the y­axis and the three places of articulation on the x­axis, draw two graphs, one for the voiceless plo­sivesandoneforthevoicedplosives.Explainwhythepositionsof[p]and[g]differfromthoseinthe‘theoretical’graphsofthepreviousquestion.

Some typology: sameness and difference 47

3.4 CONCLUSION

Obviously,thefactthatlanguagesshowsomanysimilaritiesintheirsoundstruc­turescannotbeaccidental.Differentperspectiveshavebeentakenonthisfact.Thefirst explicit response has been an assumption that linguistic structure is an inherent, ‘innate’ property of human beings, whereby the shape and form of the innate com­ponentisyettobedetermined.Weareuncertainoftheextenttowhichphonologistsbelieve that certain segments or particular distinctive features are innate, but our guess is that most would maintain that it is contrastiveness which is innate, but that the way contrasts get spelled out in features may not be. The most general statement of the aims of phonology is that it seeks to establish the ‘possible space’ of phonologi­cal structure, and to show that the actual phonological systems we find in languages fit into that space, while non­existent structures do not. There are, for instance, many processes, but the number of possible processes that are never attested is very much larger.Adherentstotheinnatenesspositionwillarguethatinnatenessmayexplainwhychildrenlearnphonologicalstructuressoquickly.Humansmaystartoutwitha certain amount of ‘skeletal’ information, which they fill in with language­specific informationonthebasisofthelanguagetheyareexposedto.Withregardtosyllablestructure, for instance, the innate information might be that there are syllables, and thatsyllableshavepeaksandonsets.Whatthechildwouldwanttoknownextis(1)whether theonsetmayremainemptyand/or (2)whether theremaybeacoda(ClementsandKeyser1983:29;Kaye1989:56;Blevins1995).Attheotherendoftheinnateness spectrum, the assumption will be that phonological structure is ‘emergent’. Giventheinputlanguageandgivenahostofambientfactors,includingthephysiologi­calpropertiesofspeakersandhearersthatwerediscussedinchapter2,phonologicalfeature systems, segment inventories, syllable structures and so on arise and develop. Itseemsevidentthatbothpositionsarecorrect.Theissueistounderstandjusthowweemploy our innate cognitive and physiological structures to learn and use languages, and what the role might be of any more peripheral social and climatic factors.

Acrucialassumptionthatunderliesthenotionofaphonologicalsystemisthatthe pronunciation of a language can be described with the help of a finite set of discrete constituents, i.e. segments, syllables, feet, etc. The pronunciation of every morpheme consists of a particular configuration of those constituents. These pho­nological constituents are meaningless, and distinct from the meaningful, morpho­syntactic constituents of the language, such as morphemes, morphological words, syntacticphrasesandsentences.Inchapter12,wewillseehowphonologicalstruc­ture continues above the level of the word.

NOTE

1 BasedontheUCLAPhonologicalSegmentInventoryDatabase(UPSID),whichcanbeconvenientlyapproachedwiththehelpofHenningReetz’ssearchprogramathttp://web.phonetik.uni-frankfurt.de/upsid_info.html.

Making the form fit

4.1 INTRODUCTION

Inchapters1and3,we sawhowdifferent languageshavedifferentphonologies.Oneoftheclearestillustrationsofthisfactisprovidedbytheadaptationofloan­words to the phonology of the borrowing language. In this process, speakers will interpret the pronunciation of the words of the foreign language in terms of the phonological elements of their own. The way in which they do this can tell us a great dealaboutthephonologyofthespeaker’snativelanguage.Forexample,theFrenchpronunciation[fiˈliŋ]forEnglish[ˈf iːlɪŋ]revealsthatFrenchdoesnotdistinguishtenseandlaxvowels,anduses[i]forboth[iː]and[ɪ].Second,itplacesanaccenton the last syllable, regardless of where the stress was in the original word. In this chapter we will discuss the process of nativization, and illustrate it mainly on the basisofEnglishloansinHawaiianandoneIndonesianloanwordinKonjo.Theselanguageshaveverydifferentphonologies,thephonologyofEnglishbeingmuchmorecomplexthanthatofHawaiian,inparticular.Aftershowinghowthepronun­ciation of foreign words is shaped by the phonological structure of the native lan­guage, it is pointed out that the phonological representation of native morphemes, too,mayneedtobeadjusted.Thisneedmayarisewhenmorphemesarecombined.Ifalanguagewiththesyllablestructure(C)V(C)onlyallowsacodaconsonantinword­final position, something will have to be done whenever a consonant­initial suffixlike[ka]isattachedtoaconsonant-finalbaselike[taf],since*[tafka]wouldbeill-formed.Inordertodescribephonologicaladjustments,twoapproacheshavebeen adopted: rules and constraints. The difference between these two approaches is briefly explained and illustrated.

4.2 HAWAIIAN

First,wegiveabriefoutlineofHawaiianphonology,basedonElbertandPukui(1979).Thesyllablestructureischaracterizedbytheformulain(1).

(1) (C)V(V)

That is, syllables do not have codas, vowels can be long or short, and the onset containsmaximallyoneconsonant.TheHawaiianphonemeinventoryisextremelysmall.Itisgivenintabularformin(2).TherowsintheC-systemstandformannersof articulation, while the columns stand for places of articulation, ordered from

4

Making the form fit 49

labialtoglottal.Forthevowels,therowsstandfordegreesoftongueheight,whilethe columns have the order ‘front unrounded’, ‘back unrounded’ and ‘back rounded’.

(2) V C i u p k ʔ e o h

a m n w l

Consecutive vowels are diphthongs if the second is higher than the first (e.g.[au]or[oi]),andlongvowelsiftheyareidentical(i.e.[aa]is[aː]).Intermsof(1),thesevowelsequencesareVVandaremonosyllabic.Othersequencesofvowelsaredividedovertwosyllables.A[j]isinsertedbetween[ie]andalowervowel,and[w]isinsertedbetween[uo]andalowervowel.Thelatterconsonantisindistinguish­able from theunpredictableoccurrenceof the [w] listed in (2).A sequence like[ua]isthereforeequivalentto[uwa],and[ia]ispronounced[ija].PossibleHawai­ianwordsare[iwa]‘nine’,[niʔihau](geographicalname),[honolulu](geographi­calname),[ala]‘road’,[ʔala]‘fragrant’,[puaʔohi]‘chatter’,[kaukau]‘admonish’and[hoʔolauleʔa]‘celebration’.Impossiblewordsare*[tuʔa],*[plai]and*[kehunanal].

Q25 Explain why the last three items cannot be Hawaiian words.

4.3 ADJUSTMENT PROCESSES

4.3.1 The process of nativization

AspeakerofHawaiianspeakingEnglishmightwellwishyou[melekelikimaka]on25December.Thisutteranceis theresultoftheinterpretationof[mεɹi kɹɪsməs].The situation exemplified here is representative of what happens when speakers of one language decide to speak another language without adopting any of the pho­nologyofthatotherlanguage.Whenfacedwiththetaskofpronouncinganexpres­sion in a foreign language while using only the phonology of their native language, speakersneedto(a)interpreteachofthesegmentsintheforeignwordintermsofthenativesegmentsystem;and(b)makesurethatnostringsarisethatbreakthesyllable structure constraints or any other phonotactic constraints of their language. Thesetwotypesofprocessesshouldbeseenasdifferentparses,accordingtoSilver­man(1992).ThefirstparsetakesplaceatthePerceptual Level: the acoustic input, or the acoustic image that the speaker has of the foreign word, is interpreted as a string of native segments. In the case of Merry Christmas, this process must have resultedinthesegmentsinthesecondcolumnof(3).Thesegmentsinthefirstcol­umnrepresenttheEnglishinterpretationofthisexpression.

Making the form fit50

(3) Input Perceptual Level Operative Levelm m mε e eɹ l li e ek k k

e(cf.(1))ɹ l lɪ i is k k

i(cf.(1))m m mə a as k k

a(cf.(1))

Compared to theEnglish segments,Hawaiianhas phonetically quite similarsegments available inmost cases.The fact that there is no [r]-type consonantaccountsfortheinterpretationof[ɹ]as[l],while[k],beingtheonlylingualnon­sonorantconsonant,isthebestinterpretationof[s].Nownoticethatthisstringof segments (those in the second column) is notwell-formed. In particular, itcannotbe analysed as sequencesof (1),withorwithout theoptional elementsin that formula. The aim of the second parse, referred to as the Operative Level, is tomakethestringofsegmentsperceivedat thePerceptualLevelconformtothephonotacticconstraintsofthelanguage.Mostimportantly,thesegmentswillhave to be accommodated by giving them legitimate positions in syllable struc­ture.Ratherthanthrowingthe[l]inChristmas out, therefore, a vowel is inserted between [k]and [l].Asa result,both [k]and [l] arenowsingleconsonants intheironsets,asrequiredby(1).Similarly,vowelsareinsertedafterthesecondandthird[k].(Thenewvowelwouldappeartobeacopyofanearbyvowel,butwewillignorethisaspect.)

Silverman’s divisionof theprocess of nativization into aPerceptual Level andanOperativeLevelisconvenientandallowsustoseethedistinctionbetweenthephonologicalelements(vowels,consonants)andthephonologicalstructurecon­tainingthesegments.Respectively,theymaybethoughtofasthethingsthatneedtobepackagedand the legitimatewaysofpackaging them.However,wecannotreally separate the two processes. It is not the case that the segments produced by the perceptual parse must be accommodated by the phonological structure at all costs, because it is not always the most likely segmental interpretations that sur­vive.Rather, itwouldappear tobe thephoneticoutputof thenewphonologicalrepresentation which is evaluated for its similarity to the phonetics of the original realization. Instead of supplying a vowel in order to make the segment string con­formtothenativesyllablestructure,thelanguagemightthereforeadjustthestringof segments by replacing a consonant with one that is not too different from it but that can beaccommodatedwithouttheadditionofasyllable.Forinstance,Germandoesnotallowvoicedobstruentsinthecoda,whichdooccurinEnglish.In(4),a

Making the form fit 51

[b]intheEnglishinputforaGermanloanwordisultimatelyreplacedwitha[p],becausethepresenceof[b]isill-formedinthatposition.AsolutionthattheGer­man speaker does notresorttoistosupplyavowelafterthe[b]sothatitcanbepreservedasanonsetconsonant(*[pabə]).

(4) Input Perceptual Level Operative Levelp p pʌ a ab b p(topreservefinalpositionofplosive)

AnothercaseofanalternativechoiceofconsonantcomesfromKonjo.Inthislan­guage,awordmustminimallyconsistofafoot,whichinKonjoisdisyllabicandhasthestressonthefirstsyllable.Ithasthevowels[ieaou],anditssyllablestructureis(C)V(C).Word-finally,thecodacanonlybe[ʔ]or[ŋ].Ithasafullsetofnasals ([mnɲŋ])aswellasalateral(FribergandFriberg1991).

Q26 Why are the phonological representations [meŋ] and [leʔem] not possible Konjo words?

Konjohasadopted the Indonesianword [lεm] ‘glue’ (itself a loan fromDutch[lεim]) as [leʔeŋ].Atfirst sight, thismaybe a little surprising.Note that [leme]wouldbeapossibleword; infact, theform[lame]isanactualword: ‘tuber’.Theform[leʔeŋ]conformstoKonjophonologyinthatitisdisyllabicanddoesnothave[m]word-finally.Here,thePerceptualLevelanalysismustbe[lem],butasinthecaseoftheGermanloanfromEnglish,thefirstdecisionattheOperativeLevelwastoreplaceanimpossiblecodaconsonantwithapossibleonethatissimilar:[m]isreplacedwith[ŋ](see(5)).

(5) Input Perceptual Level Operative Levell l lε e e

ʔ(toserveasCininsertedCVsyllable)e(toserveasVininsertedCVsyllable)

m m ŋ(topreservefinalpositionofthenasal)

Sometimes,asegmentthatispresentintheoriginalformisnottranslatedintheloanword,whichmayhappenifthesegmentisnotparticularlysalient.AnexampleistheCantoneseword[lip]‘lift’,anadaptationofEnglish[lɪft].Here,whilethefirstthreesegmentsintheEnglishoriginalhavebeentranslated,thefinal[t]wassimplyleftuninterpreted.

Aquickwayoflearningsomethingaboutthephonologyofalanguageistolookat loanwords borrowed from languages that you do know the phonology of. The adjustmentsthataremadewillindicatewhatstructuresareungrammaticalintheborrowing language.

Making the form fit52

Q28 To salvage consonants that would otherwise be illegitimate codas, Japanese provides the vowel [ʉ] to allow them to be onsets, as in [tɕiːzʉ] ‘cheese’ and [masʉkʉ] ‘mask’. However, after [t] the vowel [o] is used, as in [toːsʉto] ‘toast’. Referring back to Q27, can you explain this fact?

Q27 Japanese has the following processes, which apply additively.

1 Thecoronalplosive[t]isaffricatedto[ts]before[iʉ].2 Thecoronalobstruent[s]isprepalatal[ɕ]before[i].3 In casual speech, the close vowels [iʉ] are devoiced to [i ʉ] between

voicelessobstruentsorafteravoicelessobstruentatthewordend.

Give the narrow transcriptions of /tʉkemono/ ‘pickled vegetables’, /sitá/‘tongue’and/ótiba/‘fallenleaves’.

Q29 Here are some Japanese loans from Dutch (Vos 1963; de Graaf 1990).

Loanword Dutch originsoːda ˈsoːdaː ‘soda’koːhiː ˈkɔfi ‘coffee’korera ˈxoːləraː ‘cholera’meraŋkoriː meːlɑŋxoːˈli ‘melancholy’hipokonderiː hipoːxɔnˈdri ‘hypochondria’maŋgan mɑŋˈɣaːn ‘manganese’bombon bɔmˈbɔn ‘kind of sweet’gomʉ ɣʏm ‘rubber’kari ˈkaːli ‘potash’kiniːne kiˈninə ‘quinine’mesʉ mεs ‘knife’karʉkʉ ˈkɑlək ‘chalk’masʉto mɑst ‘mast’bʉriki blɪk ‘tin’orʉgoːrʉ ˈɔrɣəl ‘pipe organ’sʉkopːʉ sxɔp ‘spade’

In this section it has been shown that the pronunciation of foreign words is adjustedtothephonologicalstructureofthenativelanguage.Inthenextsectionwewillseethatinthenativephonologysimilaradjustmentprocessesmayoccur.

Making the form fit 53

pompʉ pɔmp ‘pump’pisʉtorʉ piˈstoːl ‘pistol’gipʉsʉ ɣɪps ‘plaster cast’kan kɑn ‘kettle’rampʉ lɑmp ‘lamp’

1 Listtheconsonantsthatappearword-finallyintheJapanesewords.2 Listtheconsonantsthatcanappearinthecodaofnon-finalsyllables.3 Whyisavoweladdedafterthewordfor‘rubber’butnotafterthewordfor

‘manganese’?4 Onthebasisofthesedata,whatwouldyousayarethevoicelessfricatives

ofJapanese?5 WhatevidenceistherethatJapanesedoesnotallowCC-onsets?(Infact,

JapaneseallowsCC-onsets,providedthesecondCis[j],asinthegeo­graphicalname[kjoːto].)

Q30 Study the following Hawaiian loanwords from English carefully.

English Hawaiian English Hawaiianˈælbət ʔalapaki ‘Albert’ waɪn waina ‘wine’ˈtɪkɪt kikiki ‘ticket’ raɪs laiki ‘rice’səʊp kopa ‘soap’ bεl pele ‘bell’bɪə pia ‘beer’ ˈflaʊə palaoa ‘flour’kɪlt kiliki ‘kilt’ ˈzəʊdiæk kokiaka ‘zodiac’brʌʃ palaki ‘brush’ ˈθaʊzənd kaukani ‘thousand’ˈstɔːri kole ‘story’ pɑːm paama ‘palm’skuːl kola ‘school’ ˈεləfənt ʔelepani ‘elephant’

1 ListtheonsetandcodaclustersintheinputformsthatarebrokenupintodifferentsyllablesintheHawaiianoutputforms.

2 Listtheconsonantclustersthatarenotsobrokenup,i.e.thatarenotfullyinterpreted.

3 ListtheEnglishconsonantsforwhichHawaiian[p]isused.Whatarticu­latoryaspectdotheseconsonantshaveincommon?

4 ListtheEnglishconsonantsforwhichHawaiian[k]isused.Whatarticu­latoryaspectdotheseconsonantshaveincommon?

5 CanyoumakeaguessastowhatthenativizedHawaiianformforEnglishfalse mightbe?

6 IfsomeoneaskedyouwhytheHawaiianspeakershouldn’tsimplysay[s]insteadof[k]inwordslikesoap and Christmas, what would your answer be?WoulditbecorrecttoanswerthatthefullynativespeakerofHawai­iancannothearthedifferencebetween[s]and[k]?

Making the form fit54

4.3.2 Adjustments in the native vocabulary

It may seem self­evident that the phonological shape of the morphemes of a language willconformtothestructuralconstraintsholdinginthatlanguage.Byandlarge,thisis true: a language that disallows complex onsets will typically not have words with complexonsets.Yetitfrequentlyhappensthatill-formedstructurespotentiallyariseinnativeforms.Foronething,thephonologicalshapeofmorphemesthatcannotbythemselvesbewords,likeaffixes,neednotconformtotheconstraintsonsyllablestructure.Anaffixmayconsistofasingleconsonant,forinstance,andassuchnotbeagoodsyllable.Usually,well-formedsyllablesareonlyrequiredattheleveloftheword.Moregenerally,whensuffixesareattachedtobasestoformcomplexwords,orwhen words come together to form phrases, there is no guarantee that the phono­logicalshapeofthecombinationiswell-formed,andadjustmentsarethereforefre­quentlycalledfor.Forexample,theunderlyingformoftheEnglishnominalpluralsuffixis[z].Thisparticularphonologicalrepresentationisfinewhenthemorphemeattaches to eye toform[aɪz],butitcannotbeusedinthatsameformincombinationwith either nose or back. In the first case, a vowel is inserted between the stem and thesuffix,toform[nəʊzɪz],astheform*[nəʊzz]isill-formed;inthesecondcase,[bæks]isformed,itbeingimpossibletohavetwoadjacentobstruentsinthesamesyllablethatdifferinvoicing(*[bækz]).Asaresult,thesuffix[z]hasthreedifferentpronunciations, or morpheme alternants,[z],[s]and[ɪz].

4.4 TWO APPROACHES

Therehavebeentwoapproachestothequestionofhowphonologicaladjustmentsof(native)morphemesshouldbedescribed,onebasedonrulesthatchangerepresenta­tions and one based on constraints that require representations to have certain forms.

4.4.1 Rules

Thetraditionalapproach,associatedwiththemonumentalworkbyNoamChom­skyandMorrisHalle,The Sound Pattern of English (SPE) (1968),usesrules thatchange the phonological representation of the morpheme in particular phonologi­cal contexts. These rules are ordered: each rule except the first applies to the output oftheprecedingrule.Afteralltheappropriatechangeshavebeenmade,thecorrectform surfaces. This is a derivational approach, in the sense that the surface form of the expression is derived in a series of structure­changing operations from the underlying form.

InthecaseoftheEnglishpluralsuffix,thederivationalapproachpostulatestworules.One,ɪ­insertion(6),insertsavowelbetweentwosibilants([szʃ ӡ]).Theother,devoicing(7),devoicesanobstruentafteravoicelessobstruentinthesamesyllable.

(6) ɪ-insertion Insert[ɪ]betweentwoadjacentsibilantsinthesameword.(7) devoicing Avoicedobstruentbecomesvoicelessafteravoicelessobstruent.

Making the form fit 55

The adjustments are given in the derivation (8).The first line represents theunderlying forms, and the last line the surface forms. Intervening lines show the work of the phonological rules, in the order in which they apply. Thus, rule (6), ɪ­insertion, doesnotapply to [bæk-z]and [aɪ-z]because these formshaveno instances of adjacent sibilants, but it does apply to [kɪs-z],which contains asequenceof[z]and[s] inasingleword.Non-applicationissometimesexplicitlyindicatedby(n.a.), as has been done here.

(8) Inputrepresentations bæk-z kɪs-z aɪ-zRule(6) (n.a.) ɪ (n.a.)Rule(7) s (n.a.) (n.a.)Outputrepresentations

bæks kɪs-ɪz aɪz

Animportantadvantageofapplyingrulesinsequenceisthattheycanexpressgeneralizationsinsimpleways.Notice,forexample,thatdevoicing can be formu­latedinthesimplewaythatithasbeenin(7),thankstotheworkofɪ­insertion. If we were to apply devoicing before ɪ­insertion, itwouldchangethe[z]after[kɪs]into[s],becauseitappearsafteravoicelessobstruentintheinput.Byfirstinsertingthe[ɪ],the[z]isnolongeradjacentto[s]andisskippedbydevoicing. Ofcourse,in rule­based descriptions not all rules are crucially ordered in this way.

Q31 The following are the underlying forms of a number of words in Tonkawa (Phelps 1975).

netale­oʔwe­netale­oʔnetale­n­oʔwe­netale­n­oʔpicena­oʔwe­picena­oʔpicena­n­oʔwe­picena­n­oʔ

‘he licks it’‘he licks them’‘he is licking it’‘he is licking them’‘he cuts it’‘he cuts them’‘he is cutting it’‘he is cutting them’

1 Identifytheunderlyingformsofthemorphemesfor‘lick’,‘cut’,‘3sg­subj’, ‘3sg­obj’,‘3pl­obj’, and the ‘progressive’(cf.Englishto be [Verb]-ing).(NB:OneofthesedoesnothaveaphonologicalforminTonkawa.)

2 Tworulesproducethesurfaceforms.(i) contraction Delete the second vowel in aword-initial sequence

CVCVCV.(ii)truncationDeletethesecondoftwoadjacentvowels.

Giveunderlyingforms,ruleapplicationsandsurfaceformsof‘helicksit’,‘he is cutting them’ and ‘he cuts them’, using the two rules contraction and truncation.

3 Arethetworulesordered?

Making the form fit56

4.4.2 Constraints

The model of serially ordered rules may be compared to a production line in a factory.Eachstageinthatlineisdevotedtosomespecificoperationwhichisper­formed on the unfinished product, with the last stage delivering the end product. Overthelasttwodecades,adifferentapproachhasbeentakentothephonologicaladjustments.Inthisconstraint-based approach, demands are put on the surface form,andanyformthatdoesnotcomplywiththeseconstraintsisrejectedinfavourof a form that does. The model is therefore more like a bureau for quality control thanafactory.Themostsuccessfulconstraint-basedtheoryisOptimalityTheory,oftenabbreviatedasOT(PrinceandSmolensky1993;Kager1999).Thistheoryholdsthat constraints are universal. There are two important features of the theory that explainwhylanguagesneverthelesshavedifferentphonologies.First,languagesdif­fer in the importance they attach to the various constraints. That is, the phonology of a language is given by the ranking of the set of universal constraints, known as that language’s constraint hierarchy.Second,constraintsmaybecontradictory,and thus may be violated: if two constraints are contradictory, the one that is ranked higher will have priority.

Aconstraint-basedapproachtotheEnglishnominalpluralsmightpostulatecon­straints(9)and(10).Accordingto(9),sequencesofsibilantsareexcludedwithinthesameword.Andaccordingtotheconstraintin(10),wecannothaveasequenceof obstruents that differ in that one is voiced and the other is voiceless. Thus, a form like *[nəʊzz]violatesthefirstconstraint,andaformlike*[bækz]thesecond.

(9) *sibsib: Sequencesofsibilantsareprohibitedwithintheword.(10) *αvoice–αvoice:Sequencesofobstruentswithinthesyllablemustagreeforvoicing.

SohowdoesOptimalityTheorydeterminewhattheoutputformmustbe?Foranygiven input form, there will initially be an unlimited set of output forms. This free generation of potential output forms is taken care of by a function called Gen(for‘Generator’),whichissubjectonlytoverygeneralconstraintsofwell-formedness.Letusillustratethiswiththepluralfor[kɪs].Foraninputform[kɪs-z],someofthegeneratedoutputswillbe[kɪsɪz],[kɪsz],[kɪs],[kɪzz],butalsoformslike[pɛts]or[tæpt].Therearetwogeneralforcesatworkthatdeterminewhichofthesenumer­ouspotentialoutputformsischosenbythelanguage.Oneoftheseforcesiscalledfaithfulness: it is the force that tries to make the output form identical to the input form.Thus,ifEnglishwerecompletelyfaithful,thepluralofkiss wouldbe[kɪsz].The other force might be said to be the unmarked way of pronouncing things. If this force were allowed to have its way, unchecked by any other force, all words in thelanguage,orindeedinalllanguages,wouldendupassomethinglike[ba],orperhaps[tə]:anythingmorethanthiswouldbemore‘marked’inthesenseoflesscommon,morecomplexandmoredifficulttopronounceorperceive.Inreality,theoutcome isdeterminedbyhowthese two forces interact.Eachof the forcesis represented by a set of universal constraints, and every language ranks these constraints in its ownway.Again, if all the Faithfulness constraints are ranked

Making the form fit 57

aboveallthephonologicalconstraints,nophonologicaladjustmentswillbemadeto the input form.However, typicallyoneormorephonological constraintsarerankedaboveoneormoreFaithfulnessconstraints,whichmeansthatinthecaseofaconflict,thephonologicalconstraintwins.Everyconstraintthatisinspectedwill thus throw out a number of candidate forms, and this process goes on until thereisonlyoneformleft.OptimalityTheorythusholdsthattheoutputformisthe optimal form,theformthatisleftastheonlysurvivorofallcandidateformsafteraninspectionoftheconstrainthierarchy.

McCarthy and Prince (1993) propose three important constraints to expressFaithfulness. Max-IO requires that each segment in the input form (‘I’) has acorrespondingsegmentintheoutputform(‘O’).Thatis,theinputis‘maximally’represented in the output, and the constraint is therefore violated if a segment is deleted. Dep-IOrequires that each segment in the output form has a correspond­ing segment in the input form. That is, the output must be entirely ‘dependent’ on the input, and the constraint is violated by any inserted segment. Third, Ident(F)requiresthateveryfeature(‘F’)oftheinputsegmentis‘identical’toeveryfeaturein the output segment. That is, this constraint is violated if a segment changes from voiceless[t]tovoiced[d],say,orfrombilabial[m]todorsal[ŋ].(Thetheoryallowsthese constraints to be split up into detailed subconstraints, and they are therefore betterseenasconstraintfamilies.)Therearemanyphonologicalconstraints,someof which we will present informally below. It will be clear that the output form will be as close as possible to the input form, and that every deviation must be forced by some higher­ranking phonological constraint. It will thus be clear that an output [tæpt]foraninput[kɪsz] isunlikelytosurviveanevaluationbytheFaithfulnessconstraints,whichwillquicklyseetoitthat[tæpt]isdiscardedinfavourofformsthatmakeabetterjobofpreservingtheinput.

(11) Max-IO: Deletionofsegmentsisprohibited.(12) dep-io: Insertionofsegmentsisprohibited.(13) ident(f): Asegmentintheinputisidenticaltothecorrespondingsegmentintheoutput.

The operation of evaluating the collection of possible output forms is called Eval (for ‘Evaluation’).This evaluation is shown in tableaux.The tableau in (14)willserve as an illustration. The constraints are arranged in the columns, and the forms to be evaluated are arranged in the rows. The input form to be evaluated is given in thetopleftcorner.A* in a cell indicates that the form of that row breaks the con­straint in that column, and *!indicatesthatsuchaviolationeliminatesthatformfrom further consideration: the violation is fatal. The optimal form, the winner, is marked .Shadedcells indicatethattheconstraint inthatcolumnhasbecomeirrelevanttothefateoftheformintherowconcerned.Asisshownin(14), it ismoreimportantinEnglishtoobey*SibSib than to obey Dep-IO:in order to pre­venttheadjacencyof[s]and[z],asegment[ɪ]isinsertedbetweenthem.However,asshownintableau(15),thelanguageis lessconcernedabout*αvoice–αvoice: althoughitcouldhavesavedthevoiced[z]from‘becoming’[s]in[bækz]byinsert­ingavoweltoform[bækɪz],itchoosesnottobreakDep-IOforthispurpose.To

Making the form fit58

satisfy *αvoice–αvoice, it is, however, prepared to violate Ident(F)in the matter ofthevoicingof[z]:theinput[z]correspondstoanon-identicaloutput[s].

(14) kɪsz *sibsib dep-io *αvoice–αvoice

kɪsz *! *kɪsɪz * kɪzz *! kɪss *!

(15) bækz *sibsib dep-io *αvoice–αvoice ident(f) bækɪz *! bækz *!bæks *

Incidentally, in addition to (9) and (10), therewould have to be a constraintrequiringthatthephonologicalcontentofthestemmustberetained.Otherwise,*[bægz], which satisfies*αvoice–αvoice, would also be described as a correct form,bythesideof[bæks].

Q32 The form *[kɪs] is an incorrect output form for the plural kisses.

1 Whichconstraint(notlistedintableaux(14)and(15))isresponsibleforrulingthisformout?

2 Drawatableaufortheinputform[kɪsz](i.e.thepluralofkiss), with the twopotentialoutput forms[kɪs]and[kɪsɪz].Thereshouldonlybe twoconstraints in your tableau: Dep-IOand the constraint you gave as the answer to thepreviousquestion.Would you rank the latter constraintabove or below Dep-IO?Motivateyouranswer.

Q33 The consonant inventory of Mauritian Creole contains the coronal fricatives [s z]. Palatoalveolar fricatives are absent. French words containing such fricatives, like [ʃəvø] ‘hair’ and [ӡeneral] ‘general’, are adapted as [seve] and [zeneral], respectively. Assume the two constraints Ident(F) and *ʃ/ʒ. Draw a constraint tableau with the correct ranking of these two constraints which shows the fate of the possible output forms [seve] and [ʃeve] in Mauritian Creole.

4.5 CHOOSING BETWEEN RULES AND CONSTRAINTS

Youmaynowthinkthatthereisnotmuchdifferencebetweenthetwoapproaches.There are,however, two importantdifferences.First, in aderivational approach,

Making the form fit 59

constraints and rules will sometimes both be needed and, moreover, appear to do thesamework.Crucially,arule-basedgrammarwillalsoneedconstraintstochar­acterize the phonological well­formedness of morphemes that have only a single form. For instance, the constraint *αvoice–αvoice (10) is generally needed tocharacterizethewell-formedsyllablesofEnglish:*[æbs],*[æpz],*[ztiː]and*[sbiː]are ungrammatical,while [ɔks] and [ædz] are fine, and are indeedused for thewords ox and adze, aswould be [bæks], the surnameof composerArnoldBax. However, because the plural form [bæks] needs to be derived from [bæk-z], arule of devoicing(7)isrequiredtoeffectthechangefrom[z]to[s].Thatis,therule-basedapproachneedsboth(7)and(10),eventhoughtheywouldappeartobedescribingthesameregularityofEnglish.Thisinherentdrawbackoftherule-based approach is known as the duplication problem(KenstowiczandKisseberth1977:136).Aconstraint-basedapproachdoesnotrunintothisproblem:thecon­straintonEnglishsyllablestructurewillactasaconditiononallforms,regardlessof whether they are morphologically derived or not.

Aseconddifferenceconcernsthephonologicaladjustmentsthatonlyappearinloanwords.Forthepurposesofthenativephonology,Hawaiiandoesnotrequirea rule that breaks up consonant clusters, since all the morphemes of the language conformto the requirementof the structure (C)V(V): there simplyarenomor­phemesthatbeginwithconsonantclusters.Inordertoaccountfortheadjustmentmadewhenawordcontainingacomplexonset like[kr] isadaptedtothenativephonology, the derivational approach will have to add such a rule to the phonol­ogy(inthiscase,onetochange[r]into[l],andonetoinsertavowelbetweenthe[k]andthe[l]).Inaconstraint-baseddescription,nosuch‘extra’grammarwouldneedtobesupplied.TocharacterizethenativeHawaiianforms,aconstraintforbid­ding complex onsets would need to be undominated anyway, meaning that no con­straint ranks above it, since the language allows no exceptions. That is, in general, inanOTdescription,theconstraintswouldberankedsoastobringanynewinputforms in line with the structural demands of the language.

Toillustratethislastpoint,consideragaintheKonjowordfor‘glue’,[leʔeŋ].TheinputformistheIndonesianword[lem].Amongthegeneratedoutputformstherewill be forms like [lem], [lemem], [leʔ], etc.Why isn’t theoutput [lem]?This isbecause the language has phonological constraints that are ranked higher than the Faithfulnessconstraints.Onesuchconstraintisthattheoutputshouldminimallybe a disyllabic foot, a constraint that is undominated in the language, so that there are no exceptions.

(16) MinWord: Awordisminimallya(binary)foot.

SinceMinWord outranks Dep-IO, a disyllabic form will be preferred to the monosyllabic[lem].Letussuppose thatwehaveevaluatedthepotentialoutputsuptoapointwhereallmonosyllabiccandidates(e.g.[leŋ],[lem],[leʔ])havebeendiscarded by MinWord.There are two further constraints that are relevant at this point.One,CodaCondition(17),forbidsthepresenceofconsonantsotherthan[ŋʔ]inthecoda.

Making the form fit60

(17) codacond: Acodaconsonantis[ʔ]or[ŋ].

After inspection of CodaCond, there are two reasonably faithful candidates left:[le.ʔeŋ]and[le.me].(Wewillignorethequestionwhy[ʔ]and[e]arethebestchoiceshereforthe insertedsegments.)Thesecondconstraintthat isrelevantatthis point is Align-Stem-Right(18),whichrequiresthatnosegmentsshouldbeaddedto,orremovedfrom,theendoftheinput.Formally,therequirementisthattheendof[lem]shouldcoincidewiththeendofasyllableintheoutput.AlignmentconstraintsarehighlyserviceableinOTandaregenerallyresponsibleforlocatingbothmorphologicalandphonologicalconstituentsinthelinguisticstructure.Forinstance,aprefixisdescribedasamorphemethatalignsitsleftedgewiththeleftedgeoftheworditisincludedin(McCarthyandPrince1993).

(18) align-steM-right: Thelastsegmentoftheinputcorrespondswiththelastsegmentoftheoutput.

In Konjo, Align-Stem-Right outranks Dep-IO.The form [leʔeŋ] passes thehigher-rankedconstraint,becausetheendofthelastsyllable,[ʔeŋ],coincideswiththeendoftheinputform[lem],and[leme]failsit,becausetheendofthesyllable[me]doesnot.Thesecorrespondencesareshownin(19).(Wecouldofcourse,per­versely, assumedifferent correspondences.For instance, theoutput [m] in (19a)mightbetakentocorrespondtothesecond[e]intheinput,butthatoutputformwould be thrown out on the grounds of multiple violations of specific versions of Ident(F).Noteventheconsonantalnatureof[m]wouldhavebeenpreserved,andan assumption of a high­ranking Ident([consonant])wouldbeenoughtodiscardthatoutputform.)

(19) a. l e ʔ e ŋ b. l e m e ǀ ǀ ǀ ǀ ǀ ǀ l e m l e m

Thefactthat[leʔeŋ]failsthelower-rankingDep-IOis of no importance: since therearenoothercandidatesleft,itisirrelevantwhetherthesuccessfulcandidatebreaks any constraints ranked below the constraint that eliminated the last rival form(s).Neither is itof any relevance that forms that failed topassa constraintsatisfyanylower-rankedconstraints.Thetableauin(20)schematizesthesituation.

(20) lem align-steM-right dep-io leme *! * leʔeŋ **

Inotherwords,theadjustmentofIndonesian[lem]toKonjo[leʔeŋ]isaccountedfor with the help of the same constraint hierarchy as is used for characterizing the nativeformsofthelanguage(Yip1993;JacobsandGussenhoven2000).Itwouldthereforeappearthat,atleastinprinciple,OTscoresoveradescriptionthatusesstructure-changingrulesintwoways.First,OTcancharacterizethephonological

Making the form fit 61

grammar of a language in a way that avoids having to state the same information twice, once as a constraint on monomorphemic forms and once in terms of a rule that changespolymorphemicformssoastoconformtothatconstraint.Second,toaccountforphonologicaladjustmentsthatareonlyobservedintheadaptationofloanwords(i.e.neverinthenativeforms),arule-basedanalysismustaddrulestothenativegrammarthatbringthoseadaptationsabout.Bycontrast,theconstraintsthatinanOTanalysischaracterizethenativeformsshouldequallycharacterizetheoptimal form of any incoming loanword.

Q34 In Japanese words, only a single voiced obstruent may occur. This constraint is known as Lyman’s Law. As a result, [pato], [bato] and [pado] are possible words, but *[bado] is not.

1 Explainwhy*[gazi]isn’t,but[gami]is,apossibleword.

Compoundsareformedfromtwowords,likeEnglisharmchair from arm and chair. EachofthewordsmakingupacompoundformsaseparatedomainforLyman’sLaw. In many compounds, the initial consonant of the second word isvoiced,asshownin(1).ThisisknownasRendaku(ItoandMester1986).

(1)ori ‘coloured’ kami ‘paper’ ori-gami ‘colouredpaper’ take ‘bamboo’ sao ‘pole’ take­zao ‘bamboo pole’ garasʉ ‘glass’ tama ‘beads’ garasʉ­dama ‘glass beads’

Rendakufailssystematicallyinthecompoundsillustratedin(2).

(2)kita ‘north’ kaze ‘wind’ kita-kaze ‘northwind’ tsʉno ‘horn’ tokage ‘lizard’ tsʉno­tokage ‘horned lizard’ siro ‘white’ tabi ‘tabi’ siro­tabi ‘white tabi’

2 WhatmightbethereasonthatRendakufailsin(2)?3 If youwere to give anOTanalysis of compounds, and Rendakuand

Lyman’sLaware twoconstraints,howwouldyourankthem?Put thetwoconstraints ina tableauwith/kita-kaze/as input,and*[kita-gaze]and[kita-kaze]asoutputs.

4.5.1 Gradient violation and unranked constraints

Sometimes,twoformsbreakthesameconstraint,butoneofthemcanbesaidtoviolate the constraint more thantheotherform.Also,itwillfrequentlyhappenthattwo constraints are never contradictory and therefore that their ranking makes no

Making the form fit62

difference.Weillustratethefirstpointwiththehelpofahypotheticallanguage,Un-Konjo,inwhichtherankingofAlign-Stem-Right and Dep-IOis the reverse of thatinKonjo.Thishasbeendoneintableau(21).Noticethatinspectionoftheform[leʔeŋ]leadstotwoviolationsofDep-IO,compared with only one in the case of [leme].Multiple(or‘gradient’)violationisshownbythenumberofstarsintherel­evantcells.Sincetheform[leʔeŋ],withtwoinsertedsegments,isaworsecandidateforthepurposesofthisconstraintthan[leme],whichhasonlyoneinsertedseg­ment,itiseliminated.Thus,[leme]isthewinnerinUn-Konjo,andAlign-Stem-Right,alongwithallotherlower-rankedconstraints,isleftuninspected.

(21) lem dep-io align-steM-right

leme * * leʔeŋ **!

LetusgobacktoKonjo.Observe,again,thatthepotentialoutputform*[leʔem]is non­optimal because it breaks CodaCond.LikeAlign-Stem-Right,this con­straint must therefore be ranked above Ident(F),whichrequiresthat[m]remains[m].However, therankingofCodaCond with respect to the two constraints in tableau(20)isindifferent.Wecouldinsertiteitherbeforeorafterthem.Intableau(22)wehaveplaceditinfirstposition;toindicatethatitsrankingisindifferent,itis separated from Align-Stem-Rightbyadottedline.Likewise,whileIdent(F)is ranked below CodaCond and Align-Stem-Right, it does not interact with Dep-IO.

(22) lem codacond align-steM-right dep-io ident(f) leme *! * leʔem *! ** leʔeŋ ** *

Q35 Draw two tableaux that are identical to tableau (22), except that CodaCond is, respectively, in third and in fourth position. Please motivate your answers to the questions below.

1 IsDep­IOcrucially ranked above CodaCond?2 IsCodaCondcrucially ranked above Ident(F)?

Thesebriefillustrationsoftherule-basedapproach(section4.4.1)andofacon­straint-based approach (section 4.4.2 and this section) give only themain ideasbehind the derivational theory of SPE andofOptimalityTheory.Inlaterchapterswewillillustratebothapproachesmoreextensively.Inchapters7,8and12wepres­ent various descriptions in the derivational framework of SPE,andinchapter11we give an account of the word­stress locations in a number of languages in the frameworkofOptimalityTheory.

Making the form fit 63

4.6 CONCLUSION

In this chapter we have seen that the phonologies of languages actively impose pho­nologicaladjustmentsoninputforms.Suchadjustmentsaremostreadilyobserv­ableinloanwords.However,theneedtomakeadjustmentsalsoarisesinthenativevocabulary of the language when morphemes are combined in words and phrases, since such combination may lead to the creation of phonological representations thatareill-formed.Therearetwowaysinwhichphonologicaladjustmentshavebeendescribed.First,theycanbedescribedwiththehelpofaseriesofrules,whichsuc­cessively change the representation so as to make it conform to the requirements ofthelanguage.Second,theycanbedescribedbymeansofoutputconstraintsthatstatewhat formsmust look like.OptimalityTheory is a constraint-based theorythat postulates that a language can be characterized by a ranking of a set of violable, universal output constraints, and that the correct form is the form that violates the constraints least.

Distinctive features

5.1 INTRODUCTION

Inchapters2and3wesawthatthesegmentinventoriesoflanguagescanbedividedinto subgroups. Thus, we have separated the group of vowels from the group of consonants, and when we discussed the way languages ‘build up’ their inventories, wedistinguishedgroupsofvoicelessobstruents(e.g.[ptk])fromvoicedgroupsthatareotherwisethesame(i.e.[bdg]inourexample).Thiswouldappeartosug­gest that the segment is not the smallest constituent of phonological structure. In thischapter,wewillmotivate thisassumption.Wewill introducethedistinctive features that will in principle enable us to describe the segments in the world’s lan­guages, and to refer to those groups of segments that play a role in their character­istic phonological processes and constraints. The latter consideration will be shown to provide an important motivation for the assumption of distinctive features. In this perspective, these features are the elements by which we can refer to natural segment classes, groups of segments that are treated as groups by languages.

The distinctive features discussed in this chapter are binary, meaning that they alwayshaveeitherapositiveoranegativespecification,like[+voice],asfor[b],or[−voice],asfor[p].Weintroducethreegroupsofthesefeatures.The‘major-class’features compartmentalize segments into four classes by means of two binary fea­tures.Next,wediscussthreebinaryfeaturesspecifyingstatesoftheglottis(‘laryn­geal’features).Finally,therearefourfeatureswithwhichwecanspecifymannersofarticulation(‘manner’features).Wewillseethatnotallsegmentsarespecifiedforallfeatures.Thatis,inadditiontosegmentsthatare[+nasal]and[−nasal],therearesegmentsthathavenospecificationfornasality.Thesegment[ʔ],forinstance,which is produced with the help of a glottal closure, is not specified for any manner feature.

5.2 MOTIVATING DISTINCTIVE FEATURES

Inventories of segments can be seen as composed of intersecting sets of seg­ments, like ‘all the voiceless segments’ or ‘all the segments articulated with lip rounding’ or ‘all the segments with a coronal place of articulation’. This fact does not in itself imply that a segment is represented in the synchronic structure in terms of a number of separate features, rather than as an unanalysable constit­uent. It could be the case that we are simply dealing with a reflex of the way segmentinventoriesdevelopedhistorically.ManycathedralsinEuropecontain

5

Distinctive features 65

elements of earlier buildings, but this does not mean that these elements are in any sense functional today.

The chief motivation for the introduction, and hence definition, of a feature is that it enables us to characterize a natural segment class. It appears to be the case that languages frequently refer to particular groups of segments, while other con­ceivable groupings are never referred to. Thus, languages frequently ban voiced obstruents from the final position in the word or from the syllable coda, but no languagewouldban [mdӡ] from thecodawhileallowing [bnɣ]. Ifparticulargroups of segments figure again and again in phonological generalizations about syllable structure or contextual variation, then evidently those groups must share something. This is what we intend to capture with the help of features by which phonological grammars can recognize them.

Q36

1 Whichconsonantsareaspiratedsyllable-initiallyinEnglish?2 Ifweignoretheordinalsuffix[θ],asoccurringinsixth, which consonants

canformaword-finalcomplexcodainapositionafter[s]inEnglish?3 Whichconsonantscanoccurbetween[s]and[r]inthesyllableonsetof

Englishwords?4 What is the significanceof the fact that thepreceding threequestions

havethesameanswer?

In addition to the requirement that distinctive features should enable us to refer to natural segment classes, we should require of a feature analysis that the distinc­tive features can characterize the segment inventories of the languages of the world. That is, all segments must be characterizable in terms of some unique combination of features. This requirement is quite self­evident: we don’t want to end up with a list offeaturesthatcannotcharacterizethedifferencebetween[m]and[n],say.

There is a third requirementplacedondistinctive features.Considerfirst thatthere is no a priori reason to suppose that natural segment classes should consist ofphoneticallysimilarsegments.Insteadof[ptk],itmighthavebeenthecasethatitwas[ptʃ]whichwereaspiratedinEnglish.Ifmoreandmorenaturalsegmentclasses were found to be phonetically arbitrary, we might be led to believe that a distinctive feature was an abstraction, i.e. that it couldn’t be given a definition other thanintermsofthecollectionofsegmentsthathadit.Tocontinuethehypotheti­calexample,wewouldsaythat[ptʃ]are[+delta],andthat[+delta]consonantsareaspirated in some context. In reality, we see time and again that the segments in natural segment classes are phonetically similar. This has led to the requirement of the Naturalness Condition (Postal1968:73),accordingtowhichdistinctivefea­turesmust have a phonetic (articulatory or acoustic) definition.Notice that therelationshipisnottheotherwayaround.Weneednot,indeedshouldnot,postulate

Distinctive features66

a distinctive feature purely because there is some phonetic property that a group of segmentshasincommon.AsKaye(1989:27n.)putsit:‘Onecouldgroupsoundsaccording to the total energy involved in their production, the number of different muscles involved in their articulation, their length in milliseconds, the distance an articulatormoves fromsomepredefinedneutralposition,and soon.’Obviously,none of those definitions corresponds to a natural segment class. The three require­ments we must impose on a distinctive feature system, therefore, are that:

1 Theyshouldbecapableofcharacterizingnaturalsegmentclasses.2 Theyshouldbecapableofdescribingallsegmentalcontrastsintheworld’s

languages.3 Theyshouldbedefinableinphoneticterms.

5.3 FEATURE VALUES

Featuresmayhavevalues.Abinaryfeaturehaseitherthevalue‘+’orthevalue‘−’.The claim here is that both the group of segments that has the minus value and the groupthathastheplusvalueformnaturalclasses.Forinstance,theassumptionofthebinaryfeature[±voice]impliesthatlanguagesrefertogroupsofvoicelessseg­mentsaswellastogroupsofvoicedsegments.EversincetheywerefirstproposedinJakobsonetal.(1952),distinctivefeatureshavestandardlybeenassumedtobebinary. In recent years, phonologists have proposed univalent features, effectively afeaturewithoutavalue(e.g.Ewen1995).Inthiscase,referencecanonlybemadeto the class of segments that has the feature, not to the collection of segments that doesnotpossessit.Othertermsforunivalentareunary or privative.Inchapter6we will discuss a set of univalent features characterizing the four main articulation zones, like labial. Multivaluedfeaturesarenolongerusedinphonology.Onethatwas common in the heyday of SPE wasthefeature[nstress],wheren was a num­ber to indicate the degree of stress of the vowel that was specified with the feature. Settingupandmotivatingafeaturesetisacoreresearchtaskinthephonologyofsegments.OurfeaturesetlargelyfollowsHalleandClements(1983),butithasalsobeeninformedbySagey(1986).Below,wediscussmajor-class features, which clas­sify segments into segment types like ‘vowel’ and ‘obstruent’; laryngeal features, which specify the glottal properties of the segment; and manner features, which specify the type of constriction, or more generally the manner of articulation.

5.4 MAJOR-CLASS FEATURES

Therearethreemajor-classfeatures,[±consonantal],[±sonorant]and[±approximant].

1 [±consonantal].[+cons]segmentshaveaconstrictionsomewherealongthecentre line in the vocal tract which is at least as narrow as that required for a

Distinctive features 67

fricative;[−cons]segmentslacksuchaconstriction.Thus,[+cons]areplo­sives,affricates,fricatives,nasals,lateralsand[r],while[−cons]arevowels,glideslike[jɥw],and,becausetheirstrictureisinthelarynxratherthanthevocaltract,[hɦ ʔ].

2 [±sonorant].This feature distinguishes obstruents ([−son]) from sonorantconsonantsandvowels([+son]).[+son]segmentsareproducedwithacon­striction in the vocal tract which allows the air pressure behind it and in front ofittoberelativelyequal,whilethisisnotthecasefor[−son]segments.Thatis,either[−son]segmentshaveanoralconstrictionwhichcausesasignificantincreaseintheairpressurebehindit(e.g.[s],[d]),orthereisnoconstrictioninthevocaltract.Sincethevocaltractdoesnotincludethelarynx,[h]and[ʔ]are[−son].So[+son]areallvowels,glideslike[ʋwj],liquidsandnasals,while[−son]areplosives,fricatives,affricatesandlaryngealsegments.1

Q37

1 Howmany(possiblyoverlapping)naturalsegmentclassescanbereferredtowithtwobinaryfeaturesatone’sdisposal?Hint:Anaturalclasscanbecaptured by one feature or by a combination of features.

2 Use[±consonantal]and[±sonorant]tocharacterizefourclasses,givingexamples of segments for each class.

Q39 In the variety of Spanish spoken in the state of Cordoba in Colombia, the first of two adjacent consonants was assimilated to the second, creating a geminate consonant (Charette 1989). For example, the word for ‘door’, which is [pwerta] in Peninsular Spanish, is [pwetta]

Q38 In Dutch, there is a rule that places a [ə] between the noun stem and the diminutive ending [tjə], as when [bɑl] ‘ball’ is affixed with [tjə] and becomes [ˈbɑlətjə]. On the basis of the following data, character-ize the group of segments after which this [ə] is inserted.

bɑl bɑlətjə ‘ball’ dɪŋkitɔj dɪŋkitɔjtjə ‘DinkyToy’kɔm kɔmətjə ‘bowl’ kɑp kɑpjə ‘hood’lɑχ lɑχjə ‘laugh’ bεs bεʃə ‘berry’kɑn kɑnətjə ‘jug’ rɔk rɔkjə ‘skirt’kɑr kɑrətjə ‘cart’ dɪŋ dɪŋətjə ‘thing’pεt pεtjə ‘cap’ sɔk sɔkjə ‘sock’

Distinctive features68

Q40 In Dutch, sequences of identical segments, which arise when the last segment of one morpheme is the same as the first of the next, are degeminated. Characterize the class of sounds that is subject to the rule in terms of distinctive features.

Input Outputχeːl lɑmpjə χeː lɑmpjə ‘yellow lamp’f is sɑpjə f i sɑpjə ‘nasty drink’leːχχaːtjə leːχaːtjə ‘empty hole’fraːjjɑχt fraːjjɑχt ‘beautiful yacht’løːk kɪnt løː kɪnt ‘nice child’slɪmmεiʃjə slɪ mεiʃjə ‘clever girl’ryʋ ʋeːr ryʋ ʋeːr ‘stormy weather’doːffεntjə doːfεntjə ‘deaf fellow’χutkoːp pɑk χutkoː pɑk ‘cheap suit’fεinnøːʃjə fεinøːʃjə ‘fine nose’niʋ ʋɔntjə niʋ ʋɔntjə ‘new wound’fiːr raːmən fiː raːmən ‘four windows’

in the Cordoba variety. The process did not always apply. Character-ize the class of consonants that underwent the process.

Earlier form Later formserdo seddo ‘pork’awto awto ‘car’talko takko ‘talc’doktor dottor ‘doctor’algo aggo ‘something’neptuno nettuno ‘Neptune’fohforo fohforo ‘match’magdalena maddalena ‘Madeleine’ojgo ojgo ‘onion’arma amma ‘weapon’ahno ahno ‘donkey’

3 [±approximant]. [+approx]are thosesegmentswhichhaveaconstrictioninthevocaltractwhichallowsafree(frictionless)escapeofair,whilefor [−approx] segments this is not the case (Ladefoged 1971: 46; Clements1993).Vowelsandnon-nasalsonorants,like[1ɹ ʌ],are[+approx]segments.(Theterm‘lateral’isusedforanyl-typesounds,whiletheterm‘rhotic’referstoanyr-typesound;lateralsandrhoticsareoftenreferredtoas‘liquids’.)

Distinctive features 69

Inmanylanguages,thedifferencebetween[iu]and[jw]canbeinterpretedintermsofsyllableconstituencyratherthanphonologicalcontent.Insuchcases,[u]isdifferentfromalabial-velarapproximant[w]notbecausethesesegmentshavedif­ferent features, but because they occupy different positions in the syllable: the peak for[u],andthemargin,usuallytheonset,for[w].SeealsoQ83.

5.5 LARYNGEAL FEATURES

There are three laryngeal features, [±voice], [±spread glottis] and [±constrictedglottis].

1 [±voice].[+voice]aresegmentsforwhichthevocalfoldsarecloseenoughtogether to allow vibration,while for [−voice] this is not the case.Thus,[+voice]arevowels(e.g. [iʌ ɛ aɪ]),sonorantconsonants(e.g.[mɲ l r ʀw])andvoicedobstruents(e.g.[bzɣ dʒ]and[ɦ]),while[−voice]arevoicelessobstruents(e.g.[pθʃtsh]).

2 [±spreadglottis].[+spread]segmentshaveavocalfoldconfigurationthatproduces audible friction in the glottis, while [−spread] segments lack

Q41 Copy the following hypothetical segment inventory and draw lines that separate (a) [−son] from [+son], (b) [−approx] from [+approx] and (c) [+cons] from [−cons].

p t k ʔb d gf s x hv z ɣm n ŋ

l rw ji a u

Q42 In English, certain segments may appear after [sp st sk] in the same syllable. Characterize them with the help of a single distinctive fea-ture on the basis of the following data.

skɪp spreɪ stjuː əˈspaɪəstɔp skjuːd skləˈrəʊsɪs skræpskwɒd spreɪn spjuː streɪtstæk strɔː ˈsplεndɪd skʌl

Distinctive features70

suchaconfiguration.Thus,aspiratedsegments like[ph kwh]and[hɦ]are[+spread],whileothersegmentsare[−spread].

3 [±constrictedglottis].For[+constr]segmentsthevocalfoldsaretenseanddrawntogether,whilefor[−constr]segmentsthisisnotthecase.Thus,[ʔ],laryngealizedvowels(e.g.[u])andlaryngealizedsonorantconsonants(e.g.[mʔ]), glottalized obstruents (e.g. preglottalized [ˀp] or ejective [p’]) are[+constr].Soareimplosives([ɓ ɗ ɠ]).Othersegmentsare[−constr].

Q43

1 American English [p t k] are accompanied by a glottal closure whenappearing in the syllable coda, as in sit, atlas, popcorn, duckpond. Whatfeaturespecificationdotheseplosivesacquireinthiscontext?

2 InSouthernOromo,aruleofi-epenthesisinserts[i]betweentheejec­tives[t’ tʃ’]anda following[tn],asshownin(a).However, [ʔ ɗ t]donot trigger the rule but undergo other changes that are not relevant here, as shown in (b,c,d) (Lloret 1995).What combination of featuresdistinguishes the [t’] from the three consonants that do not triggeri-epenthesis?

a fit’­na fit’ina ‘we finish’b aʔ­na aːna ‘we push’c feːɗ­ta feːtːa ‘you want’d bit­ta bitːa ‘you buy’

5.6 MANNER FEATURES

Therearefourmannerfeatures,[±continuant],[±nasal],[±strident]and[±lateral].

1 [±continuant].[+cont]segmentslackacentralocclusioninthevocaltract,while[−cont]segmentsareproducedwithsuchanocclusion.Thus,plosives(e.g.[pdg]),nasalconsonants(e.g.[mŋ]),affricates(e.g.[tʃ])andlater­als(e.g.[l])are[−cont],andothersegmentsare[+cont].Somelanguagesapparently treat laterals as [+cont],which isphoneticallyunderstandablein the sense that while these segments have a central occlusion, they have a lateral aperture.

2 [±nasal].[+nas]segments(e.g.[mnŋ])areproducedwiththevelum(‘softpalate’)lowered,and[−nas]segmentshavetheveluminitsclosed(raised)position.Nasalconsonantsandnasalizedvowelsare[+nas];othersegmentsare[−nas].

3 [±strident].[±strident]isrelevantforobstruentsonlyandreferstoatypeoffriction.[+strident]segmentscauseanoisierkindoffrictionthan[−strident]

Distinctive features 71

segments.[+strident]voicelessfricativesare[fsʃχ];[−strident]onesare [θçx].Togetherwith[coronal](section6.2.2), thefeaturecanbeusedtocapturesibilants([szʃ ӡ tʃ dӡ]),neededforacorrectdescriptionofthecontextforEnglish ɪ­insertion. Languagesforwhichacontrastbetween[fv]and[ɸβ]hasbeenreportedincludeEweandVenda(LadefogedandMaddieson1996: 140).ThenameEwe, [εβε], forms aminimalpairwiththewordfor‘two’,[εvε],inthatlanguage,whileEnglishcontrasts[sz]with [θð],asinsigh, xi (theGreekletterξ,[zaɪ]),thigh, thy. The feature’s other taskistodistinguishplosivesfromaffricates,bothofwhichare[−son,−cont].Suchcontrastsarecommon,asinGerman[tɔl]‘mad’vs[tsɔl]‘importduty’,English [taɪm] time vs [tʃaɪm] chime, orCorsican [ˈɟalu] ‘freeze+1sg’ vs [ˈdӡalu]‘yellow’.Therepresentationofaffricatesiscontroversial.Awidelysupportedview,however,isthattheyare[+strident]plosives(Rubach1994;Clements1999).Foronething,itisatleastsuggestivethataffricatestypi­callyhavestridentfrictionafterthereleaseoftheclosure,asin[pftstʃkχ]ratherthan[pɸtθcçkx].

Q44 In Scottish English, [i e a o u ʌ ʌi] are pronounced as long [iː eː aː oː uː ʌː ae] in open syllables. The long vowels (including [ae]) also appear before certain consonants. How can this class of consonants be characterized? This regularity, described by Aitken (1981, 1984), is known as Aitken’s Law.

raeð writhe mʌil milenʌin nine beːӡ beigetiːz tease rod roadlʌːv love kaːr carliθ Leith hom homelʌif life pis peacemel mail raʃ rash

Q45 In Turkish, obstruents are voiceless in the syllable coda, as shown in (1) (after Kim 1997).

(1) Underlying Objective Pluralip ipi ipler ‘rope’dib dibi dipler ‘bottom’at atɯ atlar ‘horse’ad adɯ atlar ‘name’køk køki køkler ‘root’gøg gøgi gøkler ‘sky’

Distinctive features72

4 [±lateral]. [+lat] segmentshave a central tongue contact in the oral cav­ity with one or both sides of the tongue being held away from the roof of themouth,allowingtheairtoescapethere,likealveolar[1]andprepalatal[ʎ].Othersoundsare[−lat].Alateralescapeoftheairisalsopossibleforobstruents,likethelateralfricatives[ɬ](voiceless)and[ɮ](voiced)andthelateralaffricates[tɬ]and[dɮ].

5.7 AMBIGUITY AND NON-SPECIFICATION

In this section we further illustrate the kind of reasoning by which featural analyses ofsegmentsarearrivedat.Asyouwillrecallfromsection5.2,thefirstandfore­most motivation for the definition of a feature is that it allows natural classes to be referred to, as has been illustrated in the exercises so far in this chapter. It is pos­sible, of course, that there is conflicting evidence for the inclusion or exclusion of a particularsegmentinanaturalclass.Thisisthecasefor[l],whichbehavesambigu­ouslyvis-à-visthefeature[±continuant]whendifferentlanguagesareconsidered.

Q46 In Dutch, certain consonants must be voiceless at the end of a syl-lable. Characterize that class of consonants with the help of distinc-tive features.

Singular Plural Singular Pluralliːr liːrən ‘pulley’ kaːrt kaːrtən ‘card’mεp mεpən ‘slap’ rɑnt rɑndən ‘edge’spεlt spεldən ‘pin’ bɑl bɑlən ‘ball’kɑnt kɑntən ‘side’ ӡiˈrɑf ӡiˈrɑfən ‘giraffe’rεis rεizən ‘journey’ ʋεp ʋεbən ‘web’dœyf dœyvən ‘pigeon’ paːrt paːrdən ‘horse’ʋɑŋ ʋɑŋən ‘cheek’ sxun sxunən ‘shoe’kʌus kʌusən ‘sock’ leːʋ leːʋən ‘lion’

diʃ diʃ i diʃler ‘tooth’kɯz kɯzɯ kɯzlar ‘daughter’deniz denizi denizler ‘sea’satʃ satʃɯ satʃlar ‘hair’aːdӡ aːdӡɯ aːtʃlar ‘tree’

1 Whichobstruentsaredevoiced?2 Canthisgroupofobstruentsbecharacterizedasanaturalclass?

Distinctive features 73

Wewillconcludethatlanguagesmaydifferinthefeaturevalueof[continuant]for[l].Asecondtopicdiscussedhereisfeaturalunderspecification.Itistobeexpectedthat certain (classes of) segments are not specified for all features.That is, fea­tures may be irrelevant for certain segments. Thus, vowels will not be specified for [±lateral],simplybecausethosesegmentscannotbedifferentiatedwiththehelpofthisfeature.Lessobviously,itappearsthat[h]doesnotparticipateinrulesreferringto[±continuant].Inthesectionthatfollows,wewillarguethatlaryngealsegmentsare not specified for either manner or place features.

5.7.1 The ambiguous behaviour of [l]

IfyoudidQ44correctly,youwillhavedrawntheconclusionthatinScottishEnglish[l]is[−cont].Inotherlanguages,however,thesamesegmentmayhavetobeana­lysedas [+cont]. InFrisian,vowelsarenasalizedbefore [n] in thesamesyllable,provideda[+cont]consonantfollows.The[n]itselfissubsequentlylost(Tiersma1985).Fromthedatain(1),whichconsistofinfinitivalverbformsprefixedwith[in-]or[oən-],itisclearthatthe[l]mustbe[+cont]:thegroupofconsonantsthatdoesnotallowthechangetogothroughis[ptkgn],whiletheconsonantsthatdoallowitare[sfjʋrl].

(1) in-pɑkə impɑkə ‘towrapup’oən-trεkə oəntrεkə ‘totaketoheart’oən-komə oəŋkomə ‘toarrive’in-gɪən iŋgɪən ‘toenter’oən-nɪmə oənnɪmə ‘toaccept’oən-stiən oəstiən ‘toplease’in-fɔlə ĩfɔlə ‘tofallin’in-jɑːn ĩjɑːn ‘togivein’in-ʋɛjə ĩʋɛjə ‘tolivewithone’sparents’oən-ropə oəropə ‘tocall’in-lɪzə ĩlɪzə ‘topreserve’

5.7.2 Laryngeals have only ‘major-class’ feature specifications

TheFrisian ruleofnasalization may also give us an indication about another question that concerns the feature [continuant]. The laryngeal consonants are [−cons,−son].Ifwetakethedefinitionsofthesefeaturesseriously,thismeansthat[hʔ]haveaconstrictioninthelarynx,butnoconstrictioninthevocaltract,andthusdonothavethesortofconstrictionthatsonorantslike[j]have.Thatis,theydo not have a constriction in the vocal tract that is at least as narrow as used by fricatives. The question therefore arises whether it makes sense to want to specify [hʔ]formannerfeaturesorforplacefeatures,since,afterall,ifthereisnostricture,howcouldwespecifyeitheritsmannerorplace?Whilethismakesgoodphoneticsense, the first question is whether languages make reference to any manner or place features of glottal consonants. For example,wemightfind rules that refertosetsofthetype‘[h]andother(classesof)[+cont]segments’.Onesuchsetis‘all

Distinctive features74

vowels,theapproximants[lrjw]andallfricatives,including[h]’,whichwouldbetheclass[+cont];anotheris[jw],vowelsand[h](butnot[ʔ)],whichwouldbetheclass[−cons,+cont].Aswehavejustseen,Frisiannasalization is a rule referring totheclassofcontinuantconsonants,but,interestingly,[h]isnotamongthem,asshownin(2).If[h]werespecifiedformannerfeatures,thatis,if[h]were[+cont],wewouldexpecttheruleofnasalizationtoapplybeforeit.Sincenasalizationdoesnotapplybefore[h],wehavetoconcludethatthissegmentisnotspecifiedforman­nerfeaturesandhencecannotbeconsidered[+cont].

(2) inhεljə ‘toholdin’oənhiərə ‘tolistento’

ADutchassimilationrulepointstothesameconclusion.progressive devoic­ing devoicesfricativesafterobstruents,asshownin(3).

(3) zeː ‘sea’ ɔpˈseː ‘atsea’ӡurˈnaːl ‘journal’ dɪtʃurˈnaːl ‘thisjournal’vyːr ‘fire’ ˈkɑmpfyːr ‘campfire’ɣɑs ‘gas’ dɪtˈxɑs ‘thisgas’

Theclassofvoicedfricativescouldbereferredtoby[−son,+cont].Clearly, if[h]is[+cont],thisconsonantwouldbeincludedinthatnaturalclass.Asithap­pens,theDutchglottalfricativeisfrequentlyvoiced.Thisvoiced[ɦ]showsupafterobstruents, too: stadhuis ‘town hall’may be pronounced [stɑtˈɦœys] aswell as[stɑtˈhœys](RietveldandLoman1985).Dutch[ɦ]isthusnotsubjecttoprogres­sive devoicing andcannotbe[+cont].Wenowhaveevidencefromtwolanguagesthat[h] isnotspecifiedforcontinuancy.WithMcCarthy(1988)wewillassumetherefore that laryngeal segments are not specified either for manner or for place features.

Q47 In the Limburgish dialect of Geleen, the diminutive suffix [kə] is palatalized to [cə] if the stem ends in one of a group of coronal consonants.

feːs feːskə ‘feast, party’mɪɲʃ mɪɲʃkə ‘man, human’hɛiʃ hɛiʃkə ‘glove’nɔːʎ nœːʎkə ‘needle’oːr øːrkə ‘ear’bɛl bɛlkə ‘bell’leɲc leɲcə ‘ribbon’ɛːɲ ɛːɲcə ‘duck’hɔut hœycə ‘hat’tant tɛɲcə ‘aunt’pɑcʃ pɛcʃkə ‘cap’

Distinctive features 75

5.8 CONCLUSION

Distinctivefeaturesarecapableofcapturingthenaturalclassesthatcanbeidenti­fied in many languages. The theoretical claim they embody is that distributional generalizations and phonological processes arrange the segments they apply to in groups that possess some feature or feature combination. This suggests that phonetic processes which may initially affect specific segments in one generation become focused in a following generation on a group of segments that are defined bysomefeaturespecification.Forinstance,eventhoughtherearenowordsinEng­lishthatbeginwith[spw]or[stw],wemaypredictthatiftheywereintroducedasloanwords, they would fall under the generalization that no aspiration or devoicing ofapproximantsappliestosegmentsaftersyllablesbeginningwith[sC-],likesport or splint, regardless ofwhether lateVOTs (see section 2.2.2) affected [w] in thedonorlanguage.Oragain,whiledegeminationinDutchappliestoallconsonants,it skips [−cons] [ʋ],despite the fact that thisconsonanthasa fairlyfirmcontactbetweentheupperteethandthelowerlipinDutchasspokennorthofthemainriv­ers. The rival theory here is of course that all such groupings are a result of natural phonetic tendencies.

Wehaveseenthatsegmentsmaybeunspecifiedforcertainfeatures,asaresultof which natural classes are more straightforwardly expressed, such that generali­zationsmaybe simpler.For instance, a characterization that all fricatives except[ɦ]arevoiceless insomecontext loses itsexceptionclause ifonlyoral fricatives,i.e.nottheglottalfricatives,havemannerfeatureslike[+cont].Finally,languagesmay apparently develop different specifications for the same phonetic segments, as suggestedbythepositivevalueofthefeature[±continuant]insomelanguagesandthe negative one in other languages. This is an indication that distinctive features emerge during language acquisition rather than being innately listed.

NOTE

1 Ithasearlieralsobeenassumedthat[hʔ]are[+son],asinChomskyandHalle(1968),HalleandClements(1983)andTrommelenandZonneveld(1983).

1 Thediminutiveimposesumlaut on the last vowel of the stem, causing them to be [−back]. It also degeminates consonants, so that [taɲc­cə]appears as [tɛɲcə] in the output. List the consonants after which [ce]appears.

2 Whatmustbethevalueof[±continuant]for[lʎ],ifconsonantsthatarefollowedby[ce]aretobecharacterizedasanaturalclass?

3. Doyouthinkthewordfor‘cap’hasanaffricate?

Describing places of articulation

6.1 INTRODUCTION

In this chapter, we introduce the features that specify the place of articulation of consonantsandthetonguepositionofvowels.Amongtheplacefeatures,therearefour univalent features specifying themajor areas of articulation.These are thefeatures [labial],[coronal],[dorsal]and [radical]:either a segment has the feature,oritdoesnot.Thisimpliesthat,justaslaryngealsegmentsarenotspecifiedforanumberoffeatures(thosespecifyingeithermannerorplaceofarticulation),not all segments will be specified for all the place features: a consonant which is not coronal will therefore not have the feature [coronal].Withinthesemajorzonesofarticulation, binary place features are used to characterize the more detailed articu­latorydistinctions.Thefeature[±round],forexample,willbeusedtospecifyseg­ments that are articulated with the help of the lips, i.e. that are [labial].Segmentswhose articulation does not involve any activity of the lips will thus be neither [+round]nor[−round]:theyhavenospecificationforeither[labial]or[±round].

InthischapterwealsopresentthenotationaldevicesChomskyandHalle(1968)employedintheirformulationofphonologicalrules,whichoftenfigureinphonolog­ical generalizations generally, whether these are expressed as rules or as constraints.

6.2 PLACE FEATURES

6.2.1 Labial

[labial]segmentsarearticulatedwiththelips,like[fpm],orinthecaseofvow­elsareformedwithliprounding,like[yoœ].Segmentsthatare[labial]may be specifiedfor[±round].

1 [±round].[+round]segmentshaveliprounding,like[pw tw o u ɔ];[−round]segments do not. In rare cases, unrounded and rounded labial segments contrast, as inMargi,NambakaengoandKilivila (Senft1986).Labializedsegmentslike[tw]willbediscussedfurtherinchapter14.

6.2.2 Coronal

[coronal] segments are articulated with a raised crown of the tongue, i.e. a raisedtipand/orblade,rangingfromadental[θ]toaprepalatal[j].Examplesof[coronal]segmentsare[tzlθjʃ ɲr].[coronal]segments are further specified

6

Describing places of articulation 77

forthefeatures[±anterior]and[±distributed],andinthecaseofcoronalfricativesandaffricatesalsofor[±strident].

1 [±anterior]. For [+ant] segments, the crown articulateswith the alveolarridgeorsomewherefurtherforward,whilefor[–ant]segments,thecrownarticulateswith apoint behind the alveolar ridge.Thus, [t d s z θ ðn l]are [+ant],whileprepalatalorpostalveolarandretroflexconsonants (e.g. [cɟ ʃ ӡ ɲ]and[ʈ ɖ ʂ ʐ ɳ ɭɽ])are[−ant].

2 [±distributed].Segmentsthatare[+distr]areproducedwithaconstrictionthat extends for a relatively great distance along the vocal tract, while for [−distr]segmentsthisisnotthecase.Thus,consonantsproducedwiththetipofthetongue(apicalconsonantslikeBritishEnglish[tdn])are[−distr],as are [s z] (Clements 1985).Blade-articulated (laminal) consonants like [ʃ tʃ ӡ]are[+distr].Dentalconsonantslike[θðtl]arealso[+distr],becauseeven where it is only the tip that touches the front teeth, the blade is close tothealveolarridgeandinfactcontributestotheacousticeffect.Retroflexconsonantsare[−distr]:thetiparticulateswiththepartofthepalateimme­diately behind the alveolar ridge.Australian languages frequently have afour­way opposition, utilizing the four possibilities given by these features (Butcher 2006). Four coronal stops and nasals contrast inKayardild, forinstance,asshownin(1)(Evans1995).1

(1) (Lamino-)dental

(Apico-)alveolar

(Lamino-)prepalatal

Retroflex

tn tn cɲ ʈɳ

Ant + + – –

Distr + – + –

Q48 In Bengali, there is an optional rule which deletes [r] before certain consonants, allowing the consonant to geminate. Characterize the class of consonants that trigger r-deletion on the basis of the fol-lowing data (Hayes and Lahiri 1991).

Input Outputbarʃa baʃʃa ‘rainy season’miʈhu­r ʃari miʈhu­ʃʃari ‘Mithu’ssari’ram­er bari ram­er bari ‘Ram’shouse’ram­er ʈaka ram­eʈ ʈaka ‘Ram’smoney’fon­korbo fon­korbo ‘will telephone’ʃundor dɔrdʒʒa ʃundod dɔddʒa ‘beautiful door’bɔrdi bɔddi ‘elder sister’bɦorti bɦotti ‘full’

Describing places of articulation78

6.2.3 Dorsal

[dorsal]soundsarearticulatedwithbuncheddorsum:[kg x ɣŋ](velars),aswellas[çk](frontedvelars)anduvulars(e.g.[χq]).Inaddition,allvowelsare[dorsal].[dorsal]segmentsarefurtherspecifiedforasetoffeaturesthatspecifyjustwherethe bunch of the tongue body is located, the tongue body features.

1 [±high].Segmentsthatare[+high]raisethedorsumtoapositionclosetotheroofofthemouth,while[−high]segmentsdonot.Thus,[+high]seg­mentsare[iɪ y ʏ u ʊ ɯ],aswellas[ç],and[kg x ɣŋ],while[χeoa]are[−high],forinstance.

2 [±low].[+low]segmentshavethebuncheddorsumlowinthemouth,while[−low]segmentsdonot.Thus,[+low]segmentsare[aεː ɔː],forinstance.

3 [±back].[+back]segmentshavethebunchofthetonguepositionedinthecentreorfurtherback,while[−back]segmentshavethebunchinthefront.Thus,[+back]segmentsarevelaranduvularconsonants(e.g.[kɣχ])andvowelslike[uə o ʌ ɑ],while[−back]segmentsarefrontedvelarslike[k]and[ç],andvowelslike[iyøε].Although[ç]isclassedwiththepalatalcon­sonants[cjɲ]intheIPAchart(i.e.withthe[coronal]consonants),itisafronted velar, i.e. a [dorsal]consonant.Manylanguageshave[ç]and[x]in complementary distribution depending on the backness of the preceding orfollowingvowel.Forinstance,Greek[ˈçεri]‘hand’beginswiththesamephonemeas[ˈxari]‘charm’.

4 [±tense].[+tense]vowelslike[ieaou]areproducedwithamoreperiph­eralandsomewhatclosertonguepositionthantheir[−tense]counterparts[ɪεɑ ɔ ʊ].Thefeatureisrelevantonlyifthelanguagehasvocalicopposi­tions like[i– ɪ],[y–ʏ],[u–ʊ],etc.It iscommonlyusedinGermaniclanguages,whichhavecontrasts likeEnglish [suːt] suit – [sʊt] soot and German[ˈmiːtə] ‘rentalfee’–[ˈmɪtə] ‘middle’.Thefeatures[±AdvancedTongueRoot;ATR],asused,forinstance,inthedescriptionoftheWestAfrican language Akan (Lindau 1978), and [±Retracted Tongue Root;RTR], used, for instance, in the description of the Tungusic languagesofSiberia(Li1996),maybeseenasphoneticvariantsofthisphonologi­calfeature.[+ATR]involvesaforwardpositionofthetonguebody,withconcomitantenlargementofthepharynx,while[+RTR]involvesaretrac­tion and lowering of the tongue body, with concomitant narrowing of the pharynx.Akanhasfourplainvowelsandfive[+ATR]vowels,threeoftheformertypeoccurringin[−ATR][εbʊɔ] ‘stone’,andthreeof the latter

kor­lo kol­lo ‘do-3sg,fut’kor­t o kot­to ‘do-3sg.past’matʃh­er nɑːk matʃh­en nɑːk ‘fish’s nose’matʃh­er matha maʈʃ­er matha ‘fish head’

Describing places of articulation 79

in[+ATR] [ebuo]‘nest’.TheBaiyinnavarietyofOrochenhasnineplainandnine[+RTR]vowels,and[−RTR][oloː] ‘tocook’, for instance,contrastswith[+RTR][ɔlɔː]‘towade’.Thefeatures[±tense],[±ATR]and[±RTR],while phonetically somewhat different, appear never to co­occur in the samelanguage(cf.HalleandClements1983:7;LadefogedandMaddieson1996:300).

6.2.4 Radical

[radical](also[pharyngeal])sounds are articulated with the root of the tongue. Avoicelessfricative[ħ]occursinmanyvarietiesofArabic,asdoesapharyngealapproximant[ʕ].SeeLadefogedandMaddieson(1996)formoreinformation.

6.2.5 Some examples

Table6.1containsfeaturevaluesforanumberofrepresentativeconsonants.Noticethat many segments have more than one place feature and are technically ‘complex’. A round vowel like [u] is [labial, +round] aswell as [dorsal, +back, +high],whilealabialized[t]isboth[coronal]and [labial].Inchapter14wewillfurtherdiscuss the representation of complex consonants.

Aseven-vowelsystemlikethatofItaliancomesoutasin(2).Itisillustratedin(3).

(2) [–back][–round]

[+back][–round]

[+back][+round]

[+high,–low] i u

[−high,–low] e o

[−high,+low] ε a ɔ

(3) milːe ‘thousand’ puɲːo ‘fist’

seta ‘silk’ sole ‘sun’

sεmpre ‘always’ skala ‘stairs’ fɔka ‘seal’

Asobservedabove,inmanyGermaniclanguages,thevowelsdivideintoalaxsetandatenseset.StandardGermanhasthevowelsystemin(4),wherethelaxvowelisgivenontheleftofeachcell.Examplesaregivenin(5).Thebracketed[æː]hasmergedwith[eː]inthespeechofmanyspeakers.Notethat[εɔ]arethelaxcoun­terpartsof[eː oː]inGermanandDutch.

(4) [–back][–round]

[–back][+round]

[+back][–round]

[+back][+round]

[+high,–low] ɪiː ʏ yː ʊuː

[–high,–low] εeː œ øː ɔoː

[–high,+low] (æː) aɑː

Tabl

e 6.

1Fe

atur

esp

ecifi

catio

nsof2

3re

pres

entativ

eco

nson

antsfo

r16featur

es.B

inar

yfeatur

esare

spec

ified

as+

or−

, whi

le th

e pr

esen

ce o

f a u

nary

feat

ure

is in

dica

ted

by √

. Bl

anks

indi

cateth

atth

eco

nson

anti

snot

spec

ified

fort

hefe

atur

e.Th

eva

lueof

thefeatur

e[c

ontin

uant

]for

[1]v

ariesa

cros

slan

guag

es.

pt

tpf

tʃkh

bf

θs

ʃç

χv

ɣm

ɳŋ

rl

jw

h

cons

++

++

++

++

++

++

++

++

++

++

––

–son

––

––

––

––

––

––

––

–+

++

++

++

–approx

––

––

––

––

––

––

––

––

––

++

++

cont

––

––

––

–+

++

++

++

+–

––

+?

++

nas

––

––

––

––

––

––

––

–+

++

––

––

lat

––

––

––

––

––

––

––

––

––

––

+–

voice

––

––

––

+–

––

––

–+

++

++

++

++

–spread

––

––

–+

––

––

––

––

––

––

––

––

+la

bia

l√

√√

√√

√√

co

r√

√√

√√

√√

√√

√distr

+–

++

–+

––

–+

ant

++

–+

+–

–+

+–

strid

––

–+

+–

–+

–+

+–

++

–d

or

sal

√√

√√

√√

high

++

–+

++

back

+–

++

++

Describing places of articulation 81

(5) Laxmɪtə ‘centre’ hʏlə ‘cover’ fʊks ‘fox’

dεkə ‘blanket’ hœlə ‘hell’ hɔlts ‘wood’

katsə ‘cat’

Tense

diːnst ‘service’ myːdə ‘tired’ guːt ‘good’

meːr ‘more’ ʃøːn ‘nice’ roːt ‘red’

(bæːr) ‘bear’ rɑːm ‘cream’

Historically,theGermanictense-laxcontrastarosefromaquantitydistinction,asevidencedby thecorrelationbetween the features longand tense inGerman.However,inmanyvarietiesofWestGermanictheyareindependent,suchthatlaxvowelsmaybelong,andtensevowelsshort.StandardDutchhasshorttense[+high]vowels,forinstance,asseenin(6).

(6) [–back][−round]

[–back][+round]

[+back][−round]

[+back][+round]

[+high,−low] ɪ,i ʏ,y u[−high,−low] ε,eː øː ɔ,oː[−high,+low] ɑ,aː

There are languages that contrast four vowel heights without employing a tense­laxcontrast,ofwhichDanishandImondaareexamples.Thetongue-heightfeatures[±high]and[±low]cannotcharacterizesuchsystems.Ofthefourtheoreticallypos­sible combinations, the specification *[+high,+low]mustberuledout,becauseitiscontradictory: the body of the tongue cannot simultaneously be raised and lowered. TheImondavowelsystemisgivenin(7)(Seiler1985).ClementsandHume(1995)presentafeatureframeworkthatcanaccountforsuchfour-heightsystems(aswellasforfive-heightsystems,whichhavealsobeenreportedtoexist).

(7) Front unrounded

Back unrounded

Back rounded

High i uHighmid e oLowmid ε ə ɔLow æ a ɒ

Q49 In Corsican, some vowels are nasalized before a nasal consonant in the same syllable (Agostini 1995). How would you characterize the class of vowels that undergo this nasalization?

ˈprintʃipe ‘prince’ˈãŋku ‘also’ˈpõnte ‘bridge’

Describing places of articulation82

ˈfungu ‘mushroom’ˈkõntu ‘account’fundaˈmentu ‘foundation’

Q50 Yanyuwa has four contrastive coronal places of articulation, illus-trated below for voiced plosives. (The lamino-postalveolar one is like [dʒ] without the fricative release.)

Lamino-dental Apico-alveolar Apico-postalveolar Lamino-postalveolard d ɖ d

1 Showhow the features [±anterior] and [±distributed] cancharacterizeeach of these articulation places.

2 Yanyuwahasafurthercontrastbetweenavoicedpalatalandvelarplosive[ɟ]and[g].Assumingtheseareboth[dorsal],how would you charac­terizetheseplacesofarticulation?

6.3 REDUNDANT VS CONTRASTIVE FEATURES

Frequently,featurespecificationsarepredictable.Thispredictabilityisinpartacon­sequenceoftheincompatibilityofparticularfeaturespecifications.Forinstance,aswehaveseenintheprevioussection,thepresenceof[+low]inthespecificationofavowelpredictsthatitwillbe[−high]:thetonguecannotberaisedandloweredatthesametime.Thesamegoesfor[+high]:itpredicts[−low].Inlargemeasure,how­ever, the predictability is language­specific and results from the fact that not all lan­guagesusethephonologicalpossibilitiestothefull.TaketheTurkishvowelsystem,forinstance.Thislanguageemploysallthecombinationsof[±back]and[±round],buthasonlytwodistinctivevowelheights,asshownin(8),withexamplesin(9).

(8) [−back, –round]

[−back, +round]

[+back, –round]

[+back, +round]

[+high] i y ɯ u

[–high] e ø a o

(9) diʃ ‘tooth’ gyl ‘rose’ kɯz ‘daughter’ tuz ‘salt’kep ‘cap’ gøl ‘lake’ at ‘horse’ son ‘end’

In this system the feature [±low] is entirely redundant, as one of its values, [+low], doesnot appear to beused at all. In other inventories, both values of afeature may occur, but one value may be predictable from other features. In the

Describing places of articulation 83

Italianvowelsystem(2),forinstance,[+round]predicts[+back]:allroundvowelsinthislanguageareback.(Thisabsenceoffrontroundedvowels,incidentally,isacharacteristic thatItalianhas incommonwithmost languages intheworld.)Oragain,whileinmanylanguagesobstruentscomeinpairscontrastingfor[±voice],like[pb],[sz],etc.,sonorantsegmentslike[mlr]areusuallyalwaysvoiced,withtheresultthatinthoselanguages[+son]predicts[+voice].Thismeansthatwhileintheclassofobstruentsthefeature[±voice]iscontrastive(ordistinctive),[±voice]isredundantinsonorants:theyarepredictably[+voice].

Alargebodyofwork,oftenreferredtoasUnderspecification Theory, has been devoted to the question of whether redundant features should be included in the underlying representation of morphemes, and, if not, when they should be sup­plied. Just as in the case of the non­specification of manner and place features in glottalconsonants(seesection5.7.2),argumentsfororagainstunderspecificationhave been based on the simplifying effect that the absence of a feature has on the formulationofphonologicalgeneralizations.Wegiveoneexamplehere,fromSteri­ade(1987).Ifweassumethat[l]is[+cont]inLatin,[±lateral]isdistinctivewithintheclassofliquids,[lr],or[+approx,+cons].Intheclassofnasalsandobstruents([−approx]) aswell as in the class of glides and vowels ([−cons]), the feature isredundant.Steriade(1987)proposesthatonlycontrastivefeaturesarespecifiedinunderlyingrepresentations.Thismeansthat[r]ismarkedas[−lat]and[l]as[+lat],butthatinnoothersegmentisthereaspecificationfor[±lateral].Thisassumptionallows for an interesting description of the distribution of the alternants -aris and -alis oftheadjectivalsuffix.Considertheformsin(10).

(10) a nav-alis ‘naval’ crimin-alis ‘criminal b sol-aris ‘solar’ milit-aris ‘military’ c flor-alis ‘floral’

Theformsin(10a,b)suggestthatthealternationdependsonthepresenceofthefeature [+lat] in thebase: if thebasecontains [+lat], the form -aris is used, oth­erwise -alis. Ifweassumetheunderlyingformofthesuffixtobe/alis/,the[+lat][l]isseentochangeinto[−lat][r]afterabasecontaining[+lat].(Suchprocessesare known as dissimilations.)Butnowconsider(9c).Althoughthebasecontains[+lat],duetothepresenceof[l]in[fl-],thesuffixisalis. It is here that the underly­ing[−lat]specificationof[r]comesinuseful.Becauseothernonlateralsounds,like[t]ofmilitaris, arenotspecifiedas[−lat],Steriade’sproposalallowsustosaythatthelastspecificationfor[lat]inthebasedeterminesthespecificationfor[lat]ofthefirstconsonantinthesuffix.Thisisshownin(11).

(11) [+lat] [–lat] | | a mil it-ar is [+lat] [–lat] [+lat] | | | b fl or-a lis

Describing places of articulation84

This particular conception of underspecification, in which features are specified only in segments that contrast for the feature concerned, is known as contrastive underspecification. The topic of underspecification is closely related to one’s view ofhowsegmentsarerepresented.Arivalviewis thatunderlyingrepresentationsonly ever include one of the two feature specifications of contrastive features.

Q51 In Luganda, [r] and [l] occur in complementary distribution (Chess-was 1963).

olubiri ‘palace enclosure’ akalulu ‘vote’liɲɲa ‘climb’ eŋgiri ‘warthog’ssaffaali ‘safari’ eŋkula ‘rhinoceros’eraŋg ‘dye’ akasaale ‘arrow’akasolja ‘roof ’ olumuli ‘reed’kampala ‘Kampala’ liiri ‘silk’omulere ‘flute’ akabonero ‘sign’omulenzi ‘boy’ weeraba ‘goodbye’luma ‘to hurt’ lje ‘my(ClassV)’

1 Whatdeterminestheirdistribution?Listthecontextsinwhicheachallo­phone occurs.

2 Whichofthesetwocontextsisstatableintermsofdistinctivefeatures?3 Whichofthesegmentswouldyouchooseastheunderlyingone?Please

motivate your answer.

Q52 Use minimal numbers of distinctive features to characterize the five natural segment classes within the segment inventory of Telugu.

p ph t th ʈʃ ʈ ʈh k kh

b bɦ d dɦ dʒ ɖ ɖɦ g gɦ

s ʂm n ɳ

l r ɭw ji e a o u1 ph bɦ th dɦ ʈh ɖɦ kh gɦ

2 t th d dɦ ʈʃ dʒ ʈ ʈh ɖ ɖɦ3 b bɦ m4 wjieaou5 a o

Describing places of articulation 85

Q53 Use minimal numbers of distinctive features to characterize the five natural segment classes within the segment inventory of Amharic (Hayward and Hayward 1992).

t t’ ʈw’ c c’ k kw kw’b bw d ɟ g gw

f fw s s’ ʃ h hw

z ӡm mw n ɲ ŋ

l rw ji ɯ ue ɤ o

a1 f fw s s’ z ʃ ӡ2 ʈ’ ʈw’ c’ kw’ s’3 c c’ ɟ ʃ ӡ4 n ɲ5 u o

6.4 WRITING RULES

Phonological rules are formal expressions that describe changes in the phonologi­calrepresentationsofwords.Asaresultoftheapplicationofaphonologicalrule,a segment may be inserted or deleted, or one or more of its feature values may be changed(ChomskyandHalle1968,SPE).Itwillalsobeshownthattheassumptionthat rules are ordered, whereby the output of one rule serves as the input to the next, allows for simple phonological generalizations.

In the representation proposed by SPE, a segment is a list of feature specifications, also referred to as a feature matrix.Althoughintheoriginalproposalallfeatureswerebinary,wegivetherepresentationwiththeunivalentplacefeature[coronal]in(12b).Anexpression like ‘[t]’ is thereforeashorthandnotation for thecorre­sponding feature matrix.

Amorphemeisrepresentedasastringoffeaturematrices.Thebeginningsandends of morphemes are indicated by boundary symbols.Whatmay,withhindsight,beseenaspuzzlingisthattheonlyboundariesweremorphosyntactic.Syllableitywashandledbyafeature,[−syl]orCvs[+syl]orC.Thesymbol+wasincludedin representations at the boundaries of word-internal morphemes (morpheme boundary),while#servedastheword boundary.Thus,(13)istherepresentationof the word pens. In much of the literature, word­internal morpheme boundaries arerepresentedbymeansofadash,asin[pεn-z]pens, ratherthanwith+.

Describing places of articulation86

(12)

(13)

Sentencesare representedas stringsofwords.Theposition inSPE was one in which the morphology and syntax preceded the phonology: only when the words are inserted into the sentence will the phonological rules be called upon to make thenecessaryadjustments.In(14),a‘shorthand’representationisgivenofthesen­tence Pens leak.

(14) #pεn+z##li:k#

6.4.1 The general SPE rule format

The general format of an SPE ruleisasgivenin(15).

Describing places of articulation 87

(15)

Theinformationtotheleftofthearrowisthefocus of the change, that between the arrow and the slash is the structural change (SC),whiletheinformationtotherightofthe slash is the context. The focus plus the context is known as the structural descrip-tion(SD).Rulesareassumedtoapplywithin words, and if a rule is to apply across word boundaries,theSDneedstoincludethesymbol#.Incontrast,rulesareassumedtoapplyacross+-boundaries,evenifno+isspecified.(Ifaruleappliesonly ata+-bound­ary,thismustbeincludedintheSD.)Toillustrate,wegive(16)astherulethatdevoicesvoicedobstruentsatthewordend,asoccurringinGermanandDutch,forinstance.

(16) final devoicing [–son]→ [–voice]/__ #

Therulescansthefeaturematricesfromlefttorightforthepresenceof[−son];then,ifitfindsone,itcheckswhetherthereisa#ontherightand,ifthereis,itspec­ifies‘–’asthevalueofthefeature[voice]inthematrixwith[−son].Inmanycases,the obstruent will already be [−voice].Theruleisthensaidtoapplyvacuously: the SDissatisfied,butapplicationof theruledoesnotbringaboutadifference.Thereason for doing it this way is that the rule can be written with fewer terms than if weweretoaddthefeature[+voice]tothefocusoftherule.

Theobstruentundergoing rule (16) frequentlyappears inword-finalposition.However,sometimesanobstruentneedstobedevoicedthatisnotadjacenttothewordboundary.Forinstance,whenthe3sgsuffix[t]isattachedtotheverbstem[leːz]‘read’,thesurfaceresultisnot*[leːzt]but[leːst].IntheSPE representations, this means that we must express in the rule that other obstruents may intervene between the obstruent to be devoiced and the word end. This is done by adding the term‘C0’beforethe#.TheCitselfisshorthandforasegmentoutsidethenucleusof the syllable, while Vissimilarlyusedtomeanavowelinthesyllablenucleus.The subscripted number n means ‘n ormore’(instancesofthesymbol), justasasuperscripted digit means ‘n orless’.Theterm‘C0’ therefore translates as ‘zero or more consonants’ and has the effect of allowing rules to ignore consonants in the positionsinwhichitisused.Ourrulethereforenowlookslike(17).

(17) final devoicing[−son]→ [–voice]/__C0#

Version (17) would be able to apply twice in the same word.When shiftingthroughtheformfromlefttoright,therulewillfindthatitsSDisinfactmetinaformlike[leːz-t]‘read-3sg’.Similarly,itismettwiceinacaselike[hoːvd]‘head’,

Describing places of articulation88

whosesingularandpluralformsare[hoːft]and[ˈhoːv.dən],respectively.Thefirstfocus is [v],when [d] corresponds toC0, and the second is [d],whenC0 corre­sponds to no consonant.

Q55 The empty-set symbol Ø is used to the left of the arrow in the case of insertion, and to the right of the arrow in the case of deletion. What would be the prose version of (1), for instance? And of (2)?

(1)Ø→ə/V__r#(2)r→Ø/__C

#

Q54 The Dutch past participle is formed by prefixing [χə] and suf-fixing [d] to the verb stem. The suffix shows up in its underlying form when the inflectional suffix [ə] is added, as in [χə+vʏl+d+ə] ‘partic+fill+partic+infl’, but otherwise shows up as [t]. In the case of a verb stem like [sχʏd], the uninflected [χə+sχʏd+d] may be formed, and is pronounced [χəsχʏt]. Does rule (17) apply to both underlying [d]s in this last form? Motivate your answer.

Inretrospect,thenotation‘C0’ was really a way of avoiding reference to syllables. The devoicing rule really applies to the syllable coda, as shown by the way speakers ofDutchtreatEnglishwordslikeSidney [ˈsɪd.ni],business [ˈbɪz.nɪs].Inloans,Englishsyllable-final[dz]arereplacedwith[ts],respectively(Booij1995).Venneman(1972),Hooper(1976)andKahn(1976)(re)introducedthesyllableinphonologicalrepresen­tation.Referencetothesyllableanditsconstituentsmakesitpossibletogiveaformula­tion of final devoicingwhichtakestheseadditionalfactsintoaccount.In(18),theσ-labelledparenthesisisusedtoindicatemembershipinthesyllablecoda.Inchapter 9we will deal more extensively with the syllable in the structural descriptions of rules.

(18) final devoicing [−son]→[ −voice]/__C0)σ

6.4.2 Additional notations

Anotationaldevicethathasbeenwidelydiscreditedasatheoreticalelement,butthat isoftenusedoutofconvenience, is thebrace. The brace notation is used to expressadisjunctionbetweentwoormoreterms(‘either. . .or’),andisthusfoundinrulesthatarepartlyidentical.Again,whennoreferencetothesyllableismade,the syllable-finalcontextcanoftenbecapturedbysaying ‘beforeaconsonantorthewordend’.Rule(19),forinstance,nasalizesvowelsbeforeanasalfollowedbyanotherconsonantorthewordend.ItwouldchangeFrench[bɔn]‘good-masc’ and [bɔnte]‘goodness’into[bɔn]and[bɔnte],respectively.

(19) V→[+nas]/__[+nas]C#

Describing places of articulation 89

Toexpressassimilations,featurevaluesneedtobemadetoagree.Forinstance,inTurkish,vowelharmonyrequiresthathighvowelsinsuffixesagreefor[back]and[round]withtheprecedingvowelintheword,asshownin(20).

(20) Nominative Possessive a køj køjy ‘village’ b kep kepi ‘cap’ c at atɯ ‘horse’ d son sonu ‘end’

Aruletoachievethiswouldusevariable feature values, expressed with the help ofGreekletters.Rule(21)saysthathighvowelsagreeinbacknessandroundnesswith the preceding vowel: α couldbe−or+,andsocouldβ, independently.

(21) V

C+

→

+

highbackround

backround

αβ

αβ

/ 0 C0__

Ifthetargetvowelinthesuffixis[−high],onlythebacknessfeaturesinthestemand suffix agree. To express such conditional dependence, angled brackets havebeenused.In(22),thisfurtherspecificationhasbeenindicated:alltargetvowelsagreeinbacknesswiththelastvowelofthestem,andifthetargetvowelis[+high],it also agrees in roundness with the last vowel of the stem.

(22) turkish voWel harMony Vhigh

backround

backround<+ >

→ < >

< >

αβ

αβ

/ +C C0 0 __

Q56

1 Writeoutthefourreadingsof(21).2 Thepluralsuffixiseither[ler]or[lar].Whichitemsin(20)take[lar]?

The parenthesis notationincludesoptionalelementsinrules.Dutchhasaruleofregressive voicing which applies within words as well as across word boundaries. Withinwords,itappliesto[lif-də]‘dear-ness,love’toform[livdə],andacrosswordsitappliesin[lif##diːr]‘dearanimal’toform[livdiːr].SinceSPE assumes that every wordthatleavesthemorphologyhas#saroundit,two#sneedtobespecifiedintheSD.However,becausetherulealsoapplieswithinwords,theyneedtobeputinparentheses to indicate that they may, but need not, be present in the representa­tion,asshownin(23).

(23) − → + ( )+−−

son voicevoicecontson

/ __ # #

Describing places of articulation90

Thereareprocessesthataffectmorethanonesegment.Forexample,metathesisisaprocessthatswitchesroundtwosegments,ascanbeseenintheOldEnglishwordfor grass, whichvariedbetween[græs]and[gærs].Tobeabletorefertochangesinvolving more than one segment, the transformational rule format was used. Itliststherelevantstringofsegmentsandboundariestotheleftofthearrow,andrepeatsthatstring,withtheSC,totherightofthearrow.Thesegmentsofthecon­textarenot literally reproduced,but identifiedwith thehelpofdigits.Rule (24)says: ‘Deleteacoronalnasalbeforeaconsonantoratthewordend,andnasalizethevowelthatprecedesit’.ItwouldchangeFrench[bɔn]and[bɔnte]into[bɔ]and[bɔte],respectively.

(24) VC

COR

++

→+

consnas

# 13

1 2

nasØ

33

Q57 Assuming the segment inventory given below, give prose statements of rules (1) to (3).

(1) V

−

→ +[ ]

lowhigh / __ #

(2)

VV

−

→

back

highhighlow

αβ

αβ

low

Ø / __

(3) Ø→

−−

+

contvoice

ant

nas

antCOR

COR

αα

/

+−

__ #

contvoice

antCOR

α

p t c k i ub d ɟ g e of s ʃ χ av z ʒ ɣm n ɲ ŋ l r

Describing places of articulation 91

Q58 Assuming the segment inventory given in the previous question, write formal rulesː

1 Obstruentsarevoicelessafterword-internalorword-externalobstruents.2 An[ə]isinsertedbetween[r l]andalabialobstruentinthesameword.3 Suffix-initialvowelsare front if theprecedingsyllablecontains[i],and

backiftheprecedingsyllablecontains[u].

6.5 LINEAR ORDER

Obviously,ifinsomelanguagetherewereonlyonerule,itsinputwouldconsistofthelexicalrepresentationoftheformsthatmeetitsstructuraldescription(theunderly­ingrepresentation,UR),whileitsoutputwouldcorrespondtotheactualpronuncia­tion(thesurfacerepresentation,SR).Languagesgenerallyhavemorethanonerule,and the question therefore arises of how phonological rules apply in a rule­based analysis:insequenceorsimultaneously?Letusassume,forthesakeofargument,thatrulesapplysimultaneouslytotheunderlyingrepresentation.Anyrulewouldalways scan the underlying representation to see if its structural description was met and, if it was, the change would be made. This mode of application is known as simultaneous rule ordering.Anotherpossibilitywouldbethattheoutputofonerule is taken as the input of another rule. In this situation, you would have to know in which order the rules applied. This option is known as linear rule ordering.Abriefconsiderationofsimultaneousruleorderingwillleadtoarejectionofthatoption.Taken­deletion and ə­insertion inDutch.Thefirstdeletesword-final[n]after[ə],whilethesecondinserts[ə]betweenaliquidandaconsonantotherthan[td].In(25),theunderlyingformsofthefirsttwowords,‘calf ’and‘bedsheet’,undergooneoftheserulesonly,butthethirdform,‘fern’,isapotentialtargetforboth.Simul­taneous application would prevent n­deletion from applying, as there is no schwa toitsleft.Togettheoutputform[vaːrə],whichisalsocommon,wemustfirstapplyə­insertion soastocreatetheintermediateform[vaːrən].Nown­deletion can applysoastogivethealternativeform,asin(26).

(25) kɑlf laːkən vaːrn n-deletion - Ø - ə-insertion ə - ə [kɑləf laːkə vaːrən]

(26) kɑlf laːkən vaːrn ə-insertion ə - ə n-deletion - Ø Ø [kɑləf laːkə vaːrə]

Tosalvagethehypothesisthatrulesapplysimultaneously,wewouldhavetopro­vide n­deletionwithanenrichedSD:‘afterschwaor[r]’.Notonlydoesthiscarry

Describing places of articulation92

the disadvantage of an otherwise unnecessary addition to the context, it would also be impossible to capture schwa and [r] in a single feature specificationwithoutspecialassumptionsabout the featurematricesof those twosegments.By insist­ingon serial ruleordering, the rule-based theoryofChomskyandHalle (1968)also achieved that varieties of the same language can be described by reversing rule orders.Asithappens,[vaːrən],theoutputof(25),isthemoreusualpronunciationof that word.

Asonefurtherexampleofhow,inarule-baseddescription,linearorderavoidscomplicated phonological generalizations, take the two rules of English whichapply to theplurals ofnouns.As youwill remember, theplural suffixhas threealternants.After[szʃ ӡ tʃ dӡ],thealternant[ɪz]isused;aftervoicelesssegmentsotherthan[sʃ tʃ],[s]occurs,whileelsewhere[z]isused,whichwearguedinchap­ter4isalsotheunderlyingformofthemorpheme.Thepluralsof[bʌs],[pεn]and[bʊk](bus, pen, book) are, respectively, [bʌsɪz],[pεnz]and[bʊks].Wethusneedonerulethatinserts[ɪ]betweenthestemandthesuffixifthestemendsin[szʃ ӡ](ɪ-insertion),andonethatdevoices[z]aftervoicelesssegments(devoicing).Werepeatthisderivationin(27).

(27) bʌs-z pɛn-z bʊk-z

ɪ-insertion bʌsɪz (n.a.) (n.a.)

devoicing (n.a.) (n.a.) bʊks

bʌsɪz pεnz bʊks

Which theory allowsus towrite the simpler rules, the simultaneous-orderingtheoryor the linear-ordering theory?For the formulationof ɪ­insertion, given in(28),itwouldnotmakeanydifferencewhichtheoryweadopted,becauseundereither theory the rule applies to the underlying form. This is necessarily so under a simultaneous mode of application, and in an ordered mode of application, it would simply be ordered first.

(28) ɪ-insertion Ø → ɪ/[+strident]__+ z

Nowconsiderthesecondrule,devoicing. In a simultaneous­ordering theory, wewouldhavetomakesurethatitdoesnotdevoicethe[z]incaseslike[bʌs-z].That is, theruleneeds toexpress ‘devoice[z]afterallvoicelesssegmentsexcept[sʃ tʃ],i.e.after[ptkfθ]’.However,[ptkfθ]donotformanaturalclass:pho­nological rules do not typically refer to this group of sounds, forcing us to use thesuspect ‘either . . .or’braces.Theformulation in(29)says: ‘afteravoicelessobstruentwhichiseither[−strid](i.e.[θ])or[labial](i.e.[pf])or[−cont](i.e.[ptk])’.Tomakemattersworse,thefeature[−cont]includestheaffricate[tʃ],sothat(29)stillmakesthewrongpredictionthat[z]devoicesinwordslikebatches ([bætʃɪz],from[bætʃ-z]).Inordertosingleout[ptk]totheexclusionof[tʃ]wecouldreplace[−cont]withthefeaturecombination[−cont,−strid],sinceonlytheaffricateis[+strid].

Describing places of articulation 93

(29) devoicing − → −

−

−

−−

son voice

voicestrident

contstrid

LABIAL/

___ #

Nowconsidertheotheroption.Ifdevoicing is allowed to apply to the output of ɪ­insertion, wecouldsimplify(29)to(30).Thisisbecauseatthetimethatdevoic­inggetstoapplyto[bʌsɪz](theoutputofɪ-insertion),the[z]nolongerappearsafteravoicelesssegment,butafter[+voice][ɪ].

(30) devoicing [−son]→[−voice]/[−voice]__#

The theory of linear rule ordering also assumes that the order of rules is constant. Thatis,iftworulesapplyintheorderA-Btooneform,theycannotapplyintheorderB-Atothenext.Theorderingisalsotransitive.ThismeansthatiftherearethreerulesA,BandC,whererulesAandBapplyintheorderA-BandrulesBandCapplyintheorderB-C,thenrulesAandCnecessarilyapplyintheorderA-C.Ofcourse, the ordering of two rules is relevant only if the order in which they apply makes a difference for the forms that are produced. There are many pairs of rules thatdonotinteract.Suchrulesareunordered.

6.6 CONCLUSION

Segmentscanbeanalysedintocollectionsofunivalentandbinaryfeatures.Asuc­cessful feature analysis will characterize the segmental contrasts that are used in the languages of the world, while it will additionally allow natural feature classes, groups ofsegmentsthatfigureinphonologicalgrammars,tobereferredto.Wehaveseenthat segments may be unspecified for certain features, because the feature is irrelevant fortheclassofsegmentsconcerned,eitheruniversally,like[±strident]inthecaseofvowelsor[±lateral]inthecaseof[h],orinsomelanguage,like[±low]inthecaseofthevowelsofTurkish.Univalentfeatureswillbeabsentinallsegmentstowhichtheydo not apply, like [labial]inthecaseof[t].Wehavealsomadeadistinctionbetweencontrastive and redundant features, where the former are features that, in some class ofsegmentsinsomelanguage,characterizecontrasts,like[±voice]inalanguagewithvoicedandvoicelessobstruents,andthelatterarenot,like[±voice]inalanguagethatonlyhasvoicelessobstruentsandvoicedsonorantconsonantsandvowels.Itisoftenassumed that redundant features do not appear in underlying representations.

NOTE

1 Thispredictsthatnolanguagecontrastsdentalstops,whichare[+ant,+distr],withlami­nallyproducedalveolarstops,whicharealso[+ant,+distr],whichwouldseemtobecorrect.

Connecting underlying and surface representations

7.1 INTRODUCTION

Thevariationinthepronunciationofwordsistrulymind-boggling.Biologicaldif­ferences between speakers lead to different acoustic outputs, and independently of whether they are adult or child and male or female, people speaking the same language will have different accents, depending on their social class and the region theygrewupin.Withinthespeechofthesamespeaker,thepronunciationofwordswillvarywiththedegreeofformality.Andevenifwetriedtoremoveallvariationfrom our speech, we would not be able to produce acoustically identical pronuncia­tions of the same word, due to the inevitable variation in the physiological mechan­ics and physical conditions of our environment. In this chapter, we will consider in more detail the type of within­speaker variation which is central to our topic, that which is a function of the phonological context. IntheEnglishspokeninNorthAmerica,thefinalconsonantinanexpressionlikeRight! willbeanunreleased[t˺],but in Right on!, thissame/t/ willbeanalveolarflap[ɾ].ThepronunciationoftheEnglishpasttensesuffixiseither[t]or[d],dependingonthevoicingofthepre­cedingsegment,giving[lʊkt]asthepasttenseoflook [lʊk],but[bɛgd]asthepasttense of beg [bɛg].Andiftheverbstemendsin[t]or[d],wefinditispronounced[ɪd],asin[ˈbʌfətɪd]and[ˈniːdɪd],thepasttenseformsof buffet [ˈbʌfət]andneed [niːd].Variationinasegment’spronunciationwhichisdeterminedbythephono­logical context is known as ‘allophony’, or allophonic variation. It is discussed in section7.2,whereitisdistinguishedfromstylisticvariation,whichisexplainedbythe degree of formality of the speech.

Phonologists have responded to allophony and the segmental variation in the pronunciation of morphemes of the kind illustrated for the English past tensesuffixbyassumingmultiple levelsofrepresentation.Inthecaseofallophony,wearenormallydealingwith‘novel’sounds,like[tʰ]and[t˺],whichdonotfigureinunderlyingrepresentations.InthementallexiconofaspeakerofAmericanEnglish,unreleased final plosives appear in the same shape and form as do aspirated plosives in word­initial position. That is, tight islistedas[taɪt],notas[t aɪt˺].Importantly,too, the existence of two levels of representation will equally make it possible to express the phenomenon that a morpheme has a single phonological form at some cognitive level of representation, but has a number of phonological forms in the sur­facepronunciation,themorphemealternantsofchapter4.Here,thesurfacemor­pheme alternants are represented in terms of segments that also exist in underlying

7

Underlying and surface representations 95

representations. For instance, the surface [s], [z] and [ɪz] of the regularEnglishpluralsuffixalsoappearinunderlyingformslikehiss, fez and the pre­final syllable in socialism. It also happens that what are different sounds underlyingly show up as the same sound on the surface, a phenomenon known as neutralization.

7.2 ALLOPHONIC AND STYLISTIC VARIATION

The ‘segments’ referred to so far in this book typically display a good deal of varia­tion,eveninthespeechofasinglespeaker.Twofactorsarelargelyresponsibleforthis intraspeakervariation.Oneisstyle, or the degree of formality of the speech situation.AnutterancelikeRight! willtypicallyhaveanunreleasedfinalplosive[t˺],asexplainedinsection7.1,butinformalspeechstylesthe[t]maywellbereleased(sothatyoucanhearaweakish[s]-likesoundafterit).Insociolinguisticstudies,the stylistic variants[t˺]and[ts]areaccordinglyseenasthepossiblevaluesofaphonological variable (t),whose frequencies of occurrence are compared acrossdifferent speech styles for different groups of speakers. The second type of variation is due to variation in the phonological context.Suchcontext-dependentvariantsare called allophones. The term phoneme is used to refer to the segment category towhich thevariousallophonesand stylisticvariantsbelong.Todistinguish thephonemeasasegmentcategoryfromtheindividualallophones,itisoftenplacedbetweenslashes://.Forexample,inBritishEnglish,thereexisttworatherdifferentpronunciationsofthephoneme/l/. Whenitoccursintheonset,itispronouncedasan alveolar lateral approximant, perhaps with some slight palatalization, as in leek, follow, a segment which is known as ‘clear l’. In the coda, as in ill, cold, its alveolar contact is accompanied by retraction of the tongue body towards the uvula, a sound symbolizedby[ɫ]andcommonlyknownas‘darkl’.Again,inmanylanguages,coro­nal consonants alternate with palatoalveolar or alveolopalatal consonants, whereby thepalatoalveolaroneappearsbefore[i],orsometimesbeforeotherfrontvowelsaswell.Thus,inKorean,underlying[sipsam]ispronounced[ɕipsam]‘thirteen’,whileIgbohas[ʃ]before[ie]and[s]elsewhere,asshownby[oʃiri]‘hecooked’,[ɔʃere]‘hesaid’vs[osεrε]‘hewrote’and[ɔsara]‘hewashed’(Jones1967:21).Theseexamplesillustratethetwokindsofphonologicalcontextsthatdetermineallophony.Oneisaparticular segment or segment group, and the other is a structural position like the coda or the onset of the syllable.

Althoughallophonicvariationindifferentlanguagesmayshowsomesimilarity,assuggestedbytheexampleof[s]beforefrontvowels,itisimportanttoseethatlanguages will differ in the segmental differences they use to differentiate between words. That is, a given segmental difference may be phonemic in one language and allophonic in another.While the difference between [s] and [ʃ] is allophonic inIgbo,being fully conditionedbywhether [i e]oroneof theothervowelsof thelanguage follows in theword, inEnglish thedifference iscontrastive: it is used todifferentiatebetweenmorphemes.Whenasegmentaldifferenceiscontrastive,there will typically be pairs of words that are distinguished only in that one has one

Underlying and surface representations96

segmentwheretheotherhastheothersegment.Examplesofsuchminimal pairs inEnglishare[sɪp – ʃɪp](sip – ship) and[liːs – liːʃ](lease – leash).

Q59 For each of the three following languages, say whether the difference between [r] and [r] is (a) allophonic, (b) phonemic or (c) stylistic. Motivate your answers.

1 InsometypesofSouthernSwedish,[ʀ]always appears at the beginning ofasyllableand[r]alwaysattheend.

2 In some types of Dutch, [r] is the usual realization in words like[proˈχrɑmaː] ‘programme’. Some speakers of those varieties use thispronunciation in everyday life but use [ʀ], as in [pʀoˈχʀɑma], whenannouncing programmes on national television.

3 In Provençal, the word for ‘evening’ is [sεʀo] and the word for ‘saw (noun)’is[sεro](Jones1967).

Q60 In Tolitoli, an alveolar lateral approximant [l] is in complementary distribution with a retroflex lateral flap, given as [l]. What deter-mines their distribution (Himmelmann 1991)?

mologo ‘wash hands’ tolitoli ‘Tolitoli’ulag ‘snake’ kikilo ‘firefly’membembelan ‘to tremble’ moŋgiulan ‘to choke’labia ‘sago’ lelembalan ‘to carry on the shoulder with a stick’

Allophonicvariationisentirelypredictable.Assuch,thereisnoreasontosupplyallophonic variation in the lexicon. The only lateral segment to appear in the lexicon ofBritishEnglishisjustplain[l].Inarule-baseddescription,aruleofl­velarization would be postulated that adds the velarization to the specification of this consonant in thecontext‘endofsyllable’.Thisisshownin(1)forthewordsleak and ill.

(1) Underlying liːk ɪl1-velarization (n.a.) ɫ

Output liːk ɪɫ

Allophonicdifferencestypicallyarisebecauseparticularcontextsinviteparticu­laradaptationsintheproductionofthesegment.Asaresult,allophonesareincom-plementary distribution.Thephonological context inwhich [ʃ] occurs in Igbo(before[ie]) is thecomplementof thephonologicalcontext inwhich[s]occurs(beforetheremainingvowelsofthelanguage):togethertheymakeupthephono­logicalcontextinwhichtheunderlyingsegment[s]occurs.

Underlying and surface representations 97

Acomplementarydistributionofsegmentscanalsoariseinadifferentway,suchas when one segment happens to be restricted to the onset position and another to the coda position. In such cases, the segments are not historically related in the way thatcontextualallophoneslike[s]and[ʃ]are.Itwillthendependonthesimilar­ity of the two segments whether the phonology treats them as contextual variants ofthesameunderlyingsegment.Awell-knowncaseispresentedby[ŋ]and[h]inEnglish.Because[h]maynotoccurattheendofasyllableand[ŋ]maynotoccurat the beginning of a syllable, the two segments are in complementary distribution. (Thatis,[hæŋ]‘hang’isaword,butneither*[tæh]nor*[ŋæt]couldbeaword.)Thereisgeneralagreementthat[h]and[ŋ]areunderlyingsegments,sincethereisno plausible way in which one can be derived from the other.

Q61

1 InPlautdiitsch,thedistributionof[x]and[ç]isillustratedby[axt]‘eight’,[açt] ‘real’,[laxt] ‘laughs’and[laçt] ‘lays’.Are[x]and[ç]differentpho­nemes,differentallophonesordifferentstylisticvariantsinPlautdiitsch?(Nieuweboer1998)

2 InOldEnglish,[ç]and[x]areincomplementarydistribution.Statethecontexts in which each of these allophones occurs.

kniçt ‘boy’ feːçθ ‘takes’meəx ‘manure’ broːxtə ‘brought’fɑːx ‘hostile’ nεəlæːçtə ‘approached’liːçtɑn ‘alleviate’ bɔxtə ‘bought’uːxt ‘daybreak’ hiːç ‘high’

Q63 French words like réseau [ʀezo], fraises [fʀɛz] and zone [zon] tend to be pronounced as [reso, fres, son], respectively, by Spanish speakers.

Q62 Korean has adopted a number of loanwords from English, as illus-trated below. Are [r] and [l] allophones in Korean?

English Korean[ˈhoʊtεl] [ho.thεl] ‘hotel’[ˈmεlən] [mε.ron] ‘melon’[ˈleizə] [rε.i.sʌ] ‘laser’[ˈskændəl] [sɯ.khεn.dɯl] ‘scandal’[ˈmɪstəri] [mi.sɯ.thε.ri] ‘mystery’

Underlying and surface representations98

7.3 TWO LEVELS OF REPRESENTATION

Three general arguments have been advanced for the assumption of different levels of representation.

1 One argument is economy. Why supply allophonic information in thelexical entries if it can be stated in a set of allophonic rules that are valid forallmorphemesofthelexicon?Thestrengthofthisargumenthasbeencalled intoquestionbyKenstowicz (1994a: 69),whopointsout that it isnot self­evident that the descriptive economy achieved by having allophonic information supplied by rules should be reflected in the actual phonologi­cal representations in the mental lexicon. There is apparently no reason to assume that the brain could not store all that redundant information for each wordinwhichitoccurs.However,theissuehereisnotjuststoragecapac­ity,butalso,orevenmainly,searchtimeinspeechperception(LahiriandMarslen-Wilson1991).Itmaybeexpectedtobemoredifficulttoretrievethe correct phonological form from a set of fully specified, hence complex, representations than from a set of more economical representations.

2 A secondargument for the assumptionof two levelsof representation isthat with a single level it would not be possible to express the phonological relatedness ofmorpheme alternants.We take theEnglishplural suffixasanexample.Suppose that insteadof saying that there is amorpheme [z]which indifferentphonological contexts is adjusted inorder toobey thephonologicalconstraintsofEnglish,weweretosaythatthreeallomorphs are listed in the lexicon, [z], [s]and[ɪz], eachofwhich isused ina spe­cificphonologicalenvironment:[ɪz]aftersibilants,[s]after(other)voice­lesssegmentsand[z]inothercases.Thisdescriptionoftheregularpluralformation would be correct, in that for every noun we can predict the plural form.Whatthedescriptionfailstoexpress,however,isthat,somehow,thethreealternants[z],[s]and[ɪz]arethesamemorpheme,or,moreprecisely,the same morpheme alternant. It is not adequate to say that this identity is expressedinthesemantics.Morphemesthatmeanthesamethingcanhave

Similarly, they tend to pronounce French poison [pwazɔ] ‘poison’ and poisson [pwasɔ] ‘fish’ both with [s]. Even though Spanish doesn’t con-trast [s] and [z], [z] occurs in words like desdén [dezˈden] ‘disdain’, mismo [ˈmizmo] ‘same’ and más vueltas [maz ˈbweltas] ‘more laps’.

1 Are[s]and[z]differentphonemesinSpanish?Motivateyouranswer.2 HowwouldanuntrainedSpanishspeakerofFrenchpronouncese laisse

desirer [sə lɛs deziʀe]‘isverydesirable’:as[selesdesire],[selezdesire],[selesdezire]or[selezdezire]?Motivateyouranswer.

Underlying and surface representations 99

differentphonologicalforms,inwhichcasetheyareallomorphs.TheEng­lishcomparativeisexpressedbythesuffix-er insomeadjectives,likenice, but by means of the ‘periphrastic’ more in the case of others, like esthetic, or, again,theregularpluralsuffix[z]isanallomorphbythesideofthepluralsuffix[ən]inoxen. The availability of two levels of representation makes it possible to state that underlyingly the phonological form of the regular plu­ralsuffixis[z],eventhoughinsurfacerepresentationsitshowsupas[z],[s]and[ɪz](Anderson1974).Phonologicallymotivatedmorphemealternationcan thus be characterized as variation in the shape of the same underlying form in a way that differentiates such variation from cases in which the dif­ferent phonological forms are unrelated, like more and -er, as well as from irregular forms, like went, the past tense form of go, for example, or feet, the plural form of foot. In such cases, the unpredictable allomorphs must be listed in the lexicon, since there is no plausible phonological generalization which could account for them.

WhilethecaseforasingleunderlyingformfortheEnglishpluralisintui­tively very clear, there has been no answer to the general question when dif­ferent forms should be related to a single underlying form and when they mustbe listedasseparateword forms.ChomskyandHalle (1968)derivedwords like sane and sanity fromthesameunderlyingform[sǣn],justaspro-found and profundity arederivedfromacommon[profūnd](wheretheover­barindicatestenseness).Othershavequestionedwhetherthisisrealistic,andhave attempted to develop experimental procedures to answer such questions empirically(McCawley1986;Ohala1986;WangandDerwing1986).

Q64 Which of the following pairs of English words would you say con-tain a common underlying form?

sew – sewage cork – corkageblow – blew conceive – conceptiontalk – talked fraternal – brother

3 Thethirdreasonforthepostulationofanunderlyingrepresentationisthatmany generalizations are only valid at a level other than the surface level. Kenstowicz (1994a: 72), citingMohanan (1992), gives as an example thegeneralizationthatsequencesofsibilantsarebrokenupbyavowelinEng­lish,whichisresponsibleforthefactthatthepluralofEnglishbus is[ˈbʌsɪz]rather than *[bʌsz](or*[bʌss]).ThegeneralizationalsoholdsinSingaporeEnglish:thepluralofkiss is kiss[əz] and that of nose is nos[əz].Inthisvari­etyofEnglish,thereisthefurtherfactthatplosivesaredeletedinthecodaafterfricatives,sothatlift, list, task are[lɪf, lɪs, tɑːs].Whenthesewordsarepluralized, theycomeoutas[lɪfs, lɪss, tɑːss], respectively.Apparently, the

Underlying and surface representations100

generalizationthatadjacentsibilantsmustbeseparatedbyavoweldoesnotholdatthesurfacelevelinSingaporeEnglish,butis true at a level of repre­sentationatwhichthefinalplosivemuststillbepresent.Clearly,ifwetookit to be true at the surface level, the generalization would predict that the plural of list, withitssurfacepronunciation[lɪss],was*[ˈlɪsəz].Ifinsteadweassumethat,underlyingly,theform[lɪst]exists,itquitehappilyallows[z]tobeadded,withouttheneedforaninserted[ə].Thereisinfactindependentevidence that the plosives exist at a deeper level of representation, because they showup in verbal formsbefore vowel-initial suffixes, as in lif[t]ing, lis[t]ing, tas[k]ing. (Theassumptionhereisthatthenounlist and the verb list havethesamephonologicalform.)In(2),thesituationisschematized.Thepluralmorpheme[z] isattached to theunderlying formsofkiss and list. The[ə]isinsertedbetweentheadjacentsibilantsintheformforkisses, but not in that for lists, since in the latter form the sibilants are separated by [t].Onlyafterthevowelhasbeensuppliedcan[t]bedeleted.From(2),itis clear that the generalization ə­insertion is only true for the underlying representation, not for the surface representation.

(2) Underlying kɪs-z lɪst-z lɪst-ɪŋə-insertion ə (n.a.) (n.a.)t-deletion (n.a.) Ø (n.a.)Otherrules (n.a.)  s (n.a.)Surface kɪsəz lɪss lɪstɪŋ

Q65

1 Ifitwereassumedthattheunderlyingformoftheverblist was[lɪs],whatincorrect prediction would be made about the pronunciation of the pre­sentparticipleformofthisverb?

2 ThepastparticipleofregularSingaporeEnglishverbsisformedbysuffix­ing[d]tothestem,asin[saɪd]sighed, [faɪld]filed. Likethepluralsuffix[z],it isdevoicedaftervoicelessobstruents,asin[wɔːkt]walked, while [ə]isinsertedifthestemendsin[td].Iftheverblist, whose past parti­cipleformis[lɪstəd],wereunderlyingly[lɪs],whatincorrectpredictionwouldbemadeaboutitspastparticipleform?Whydoestheassumptionofunderlying[lɪst]makethecorrectprediction?

Surfaceformsthatcontradictaphonologicalgeneralization,like[lɪss]inthelastlineof(2),areopaque,thatis,non-transparent.Takingarule-basedperspective,opacity can arise because some rule has failed to apply and the rule’s structural descriptionismetinthesurfaceform.Thisoccursinthecaseof[lɪss], inwhichə­insertionappearsineffective.Equally,arulemayapplyeventhoughthecontextoftheruleisnot–anylonger!–presentinthesurfaceform.Anexampleofthistype

Underlying and surface representations 101

ofopacitymayoccurinJapanese,whenavowelisdeletedaftercausingapreceding[t]tobe[ts](cf.Q27).Whentheunderlyingform[tʉkemono]ispronounced[tske­mono],the[ʉ]thattriggeredtheaffricationof[t]hasdisappearedbetweenvoice­less consonants.Opacity is typically dealtwith quite adequately in a rule-baseddescription, because rules can be ordered so that segments can be allowed to have active or passive effects before they are deleted. InOptimalityTheory, however,opacity is not easily dealt with, not, that is, if all constraints are by definition valid foroutputforms(seechapter13).

7.4 NEUTRALIZATION

InthecaseofBritishEnglish‘darkl’,therulewepostulatedproducesanovelseg­ment: before the application of the rule, no morpheme contained that segment. However,itisfrequentlythecasethattheoutputofaruleisasegmentthatalreadyexists in the context concerned. This is true, for instance, for the rule of devoicing thatdevoicesthe[z]oftheEnglishpluralsuffixto[s]aftervoicelessobstruents,aswhen[bʊk-z](books)ischangedto[bʊks].Englishalreadyhassuchformsunder­lyingly, like box [bɔks]. It alsoholdsgood forfinal devoicing, which exists in GermanandDutch,amongmanyotherlanguages.Theserulesproducesegments(voicelessobstruents)inpositionswheresuchsegmentsalreadyappear.InaDutchwordlike[pɑd]‘toad’,thefinal[d]willbedevoicedto[t],becausetheconsonantoccursinthecodaofthesyllable.Asaresult,thesingularformispronounced[pɑt],butwhenthepluralsuffix[ən]isattached,theformis[ˈpɑdən].Since[t]appearsinthatpositioninwordslike[kɑt]‘cat’(whosepluralis[ˈkɑtən]),theoppositionbetween [d] and [t] is neutralized in syllable-final position. Rules like EnglishdevoicingandDutchfinal devoicing, therefore, are neutralization rules, while BritishEnglishl­velarization is an allophonic rule.

7.5 CHOOSING THE UNDERLYING FORM

Whenamorphemehasanumberofalternants,oneofthesewillhavetobecho­sen as the underlying form by the phonologist and – if the phonological model reflectsourmentalworld–bytheinfantacquiringhisorherlanguage.Agoodunderlying form satisfies two requirements. First, it should allow you towriterulesthatdonotdestroyasegmentalcontrastofthelanguage.Second,theruleorrules that are needed to produce the allophony or allomorphy will be easy to state. Asforthefirstrequirement,rememberthataneutralizationoccurswheneveraruleproducesanoutputthatalreadyexistedinthecontextconcerned.Aparticu­larlybadchoicethereforewouldbetochooseaneutralizedsegment.‘Don’tlosecontrasts!’istheimportantadvicehere.Takefinal devoicing, discussed earlier inthelastparagraphofsection3.2andinsection6.4.1.RecallfromtheprevioussectionthattheDutchmorphemefor‘toad’hastwoalternants,[pɑd]and[pɑt],occurringintheplural[ˈpɑdən]andthesingular[pɑt],respectively.Ifweassume

Underlying and surface representations102

thattheunderlyingformisindeed[pɑd],wecanderivethetwoalternantswiththehelpof(3),repeatedfrom(18)inchapter6inproseform.

(3) final devoicing Obstruentsarevoicelessincodaposition.

Noticethatthisrulealsoappliesvacuouslyinthecaseof[kɑt]‘cat’,whichhasapluralform[ˈkɑtən].Thederivationisgivenin(4).

(4) ur pɑd pɑ.d-ən kɑt kɑ.t-ənfinal devoicing t (n.a.) vac. (n.a.)Output pɑt pɑ.dən kɑt kɑ.tən

If we were to flout the advice expressed by ‘Don’t lose contrasts!’, we mightassumethatthereisavoicingrulethatvoicesobstruentsintheplural.Let’sassumeforthesakeofargumentthattheunderlyingformof‘toad’is[pɑt].Insteadofaruledevoicing obstruents in the coda, we would need a rule that voices obstruents in the onset,i.e.(5).Asituationwouldarisewhichisshownin(6).Clearly,thisdescrip­tionproducesthewrongresultsinthecaseofthepluralof ‘cat’.Rule(5)isnotacorrectgeneralizationaboutDutch,andthedescriptionin(4)mustbeconsideredsuperior.

(5) onset voicingObstruentsintheonsetarevoiced.

(6) ur pɑt pɑ.t-ən kɑt kɑ.t-ənonset voicing (n.a.) d (n.a.) dOutput pɑt pɑ.dən kɑt *kɑ.dən

The same reasoning applies to the choice between the [s] and [z] allomorphsoftheEnglishpluralsuffix,discussedinsection4.4.1.With/z/astheunderlyingform,weneedadevoicingruletotake/bæk-z/to[bæks].Butifweweretoassume/s/as theunderlying form,avoicingrulewouldapply soas tochange [s] to [z]aftervoicedsegments.Atfirst sight, thismayseemfine:a form like /bʊk-s/cannowsurfaceunchangedas[bʊks],whileformslike/pεn-s/pens and/kaʊ-s/cows arechangedto[pεnz]and[kaʊz]bythevoicingrule.However,[s]and[z]contrastaftervoicedsounds,asshownbydose [dəʊs]versusdoze [dəʊz]andbypence [pεns]versus lens [lεnz],orindeedpens [pεnz].Takingtheneutralized[s]asunderlyingwouldthereforeleadtotheincorrectvoicingof[s]inpence /pεns/to*[pεnz],alongwiththecorrectcreationof[pεnz]outof/pεn-s/.Takingtheunderlyingformtobe/z/meansthat wecansafelydevoiceitto[s],sincethereisnocontrastwith[z]thatwe need to worry about.

The second requirement is simplicity in the formulation of the rules or con­straints.Acorrect choiceof theunderlying form leads to correct linguisticgen­eralizations. Thus, it is correct to say that all obstruents are voiceless in the coda inDutch,and,conversely,itisincorrecttosaythatallobstruentsarevoicedintheonsetinDutch.Similarly,itiscorrecttosaythatallword-finalcoronalfricativesinEnglisharevoicelessaftervoicelesssounds,butitisincorrecttosaythatallword-finalcoronal fricativesarevoicedaftervoicedsounds.1 In order to illustrate that

Underlying and surface representations 103

correct generalizations are simple to state, often simpler than if we took another morpheme alternant as the underlying form, let us consider the choice between /ɪz/ and /z/. The context for the insertion rule is simpler than that of the dele-tion rule. Insertion occurs ‘between sibilants’, deletion between ‘plosives, nasals, approximants, vowels and non-sibilant fricatives on the one hand and sibilants on the other’. We saw in chapter 5 how incongruous groupings are difficult to state in formal notation. The fact that contexts in rules and constraints are simple to state must reflect the fact that infants acquiring the language make generalizing hypoth-eses about the grammar.

The choice between /ɪz/ and /z/ could equally be motivated on the basis of the first requirement, that of no loss of contrasts, but the effect is a little more difficult to see. If we choose /z/ and an insertion rule, we correctly predict that there are no sequences of sibilants at the end of the word. That is, English has no words like *[kɪss], *[sɪʃs] or *[tʃɪzz]. However, if we choose /ɪz/ and a rule of ɪ-deletion, we predict that there are no words ending in non-sibilants and [ɪz], like Los Angeles [lɒs ˈændӡəlɪz] (which exists by the side of [lɒs ˈændӡəliːz] and [lɒs ˈændӡəlɪs], Wells 2008). With ɪ-deletion, we would end up with *[lɒs ændӡəlz] as one of the names of the city. Clearly, we wouldn’t want to give up the contrast between /ɪz/ and /z/ after non-sibilants.

Q66

[hwan ðat aːpril wiθ hiz ʃuːrəz soːtə]when that April with his showers sweet[ðə druxt ɔv martʃ haθ pεrsəd toː ðə roːtə]the drought of March has pressed to the root

are the first two lines of Geoffrey Chaucer’s The Canterbury Tales, with word-by-word glosses. The pronunciation of his words has been changed drastically by the many generations of speakers that have acquired the language since he wrote them some 625 years ago. For instance, the consonant [x] disappeared after making the vowel before it long, and Chaucer’s [uː] corresponds to [aʊ] today (i.e. [druxt] went from [druːt] to [draʊt]), and diphthongs before [r] acquired an extra syllable from the schwa-like transition between [aʊ] and [r] (i.e. [ʃuːr] went from [ʃaʊr] to [ˈʃaʊ.ər]).

1 There are two words that suggest that the present-day morpheme alterna-tion for the plural suffix and for the past/participle suffix did not exist in Chaucer’s time. What are they?

2 Argue that the answer to the previous question shows that synchronic gram-mars do not necessarily reflect the historical processes of language change.

Underlying and surface representations104

In practice, it is usually easier to see whether our choice of underlying form leads to a good generalization than whether it preserves the language’s contrasts. The place of articulation of nasal consonants in the syllable coda alternates in many languages as dictated by the place of articulation of the following consonant. In English,forinstance,theprepositionin is[ɪn]beforethe[t]ofTallinn, [ɪm]beforethe[p]ofParis and[ɪŋ]beforethe[k]ofCopenhagen. The rule is neutralizing in thecaseof[m]and[ŋ]:original[m]existsindi[m] Paris, which is di[m] in isola­tion,andoriginal[ŋ]existsinlo[ŋ]Copenhagen, which is lo[ŋ] inisolation.Noth­ingmergeswith[n],asthereisnoassimilationindi[m] Turin and lo[ŋ] Turin. Onthebasisofthesefacts,wemustavoidincludinganeutralizing[m]or[ŋ]intheunderlyingform,andinsteadtake[n],togivetheUR/ɪn/. Butifwehad incorrectly taken/ɪŋ/,say,asunderlying,wewouldprobablyhavebeenalertedthatsomethingwaswrongoncewegottowritingtheassimilationtocoronal[n].Thiswouldoccurbeforecoronal[tdsznlr],asinin Tallinn, in Dublin, in Stockholm, in Zagreb, in Nicosia, in London, in Rome, but oddly also before vowels, as in in Athens.Anoddgrouping like ‘coronal consonants and vowels’ cannot be properly described with thehelpofdistinctivefeatures.Clearly,aruletakingeither/m/or/ŋ/to[n]willbemorecomplexthatonetaking/n/toeither[m]or[ŋ].

Allophonicvariation,too,shouldbedescribedwith‘correctlinguisticgeneraliza­tions’.InQ60,wesawthatinTolitoli[l]appearsafterfrontvowelsandatthewordbeginning, while [ʟ]occursafterbackvowels.Itisthussimplertostatewhen/l/isrealized as [ʟ]thanitistostatewhen/ʟ/isrealizedas[l].Forthisreason,/l/isthebetter choice for the underlying segment. The more complex context can then be given,orimplicitlyunderstood,as‘elsewhere’.Thisisshownin(7).

(7) Tolitoli l-flapping /l/→ [l] afterbackvowels [l]elsewhere

Q67 Both [tɔn] and [tɔm] are common Dutch first names, derived from Anthonius and Thomas, respectively. If the child’s surname was [ˈbleːkfɛlt] Bleekveld and the choice was between those two names, which would you advise the parents to choose for their new son?

7.6 CONCLUSION

Therecognitionoftwolevelsofrepresentation,asurfacerepresentation(SR)andamoreabstractunderlyingrepresentation(UR),isthecornerstoneofphonologi­cal theory. It makes it possible to describe morpheme alternants as variants of the same morpheme, and opens the way to a description in which the differences in phonological form between the alternants are expressed in terms of general state­mentsaboutcontextuallydefinedphonologicaladjustments.URsandSRsusually

Underlying and surface representations 105

differfromeachotherinthatURsaremoredetailedthanSRs,butSRsmayalsoobliteratedistinctionsthatexistinURs,andthusneutralizecontrasts.Thereisnoeasy algorithm which, given a range of surface morpheme alternants, will lead to the‘correct’UR.URsarechosensothattheresultantgrammaristhesimplestthatcanbeconstructedandno incorrectpredictionsaremade.Anotherquestionthat is hard to answer is whether forms are to be regarded as morpheme alternants thathaveacommonURorasformsthatarelistedseparatelyinthelexicon.Inprac­tice,thesequestionswillnotoftenvexthephonologist,sincetheydonotseemtoodifficulttoanswerinthemajorityofcases.Itmaybeexpectedthatpsycholinguisticresearch will provide new insights here, in particular where the second question is concerned.

Q68 The Balantak nouns in the first column of the first data set have the derived forms in the second column when prefixed with an affix meaning ‘one’. The second data set shows verbs stems and their derivations with a prefix meaning ‘unintentionally’ (Busenitz and Busenitz 1991; some of the data are inferred).

Noun stem Prefixed form Gloss for stema wuras sawuras ‘seed’b bituʔon sambituʔon ‘month’c loloon saloloon ‘thousand’d taːʔ santaːʔ ‘word’e koeŋ saŋkoeŋ ‘head of grain’f utok saŋutok ‘brain’g sumpir sansumpir ‘beard’h apu saŋapu ‘fire’i noa sanoa ‘breath’j gampal saŋgampal ‘underlayer’k ŋoːr saŋoːr ‘nose’l malom samalom ‘night’m roon saroon ‘banana leaf ’

Verb stem Prefixed form Gloss for stemn giok toŋgiok ‘move’o piːleʔ tompiːleʔ ‘see’p joːŋ tojoːŋ ‘shake’q kana toŋkana ‘hit’r wawau towawau ‘do’s tobok tontobok ‘stab’t luaʔ toluaʔ ‘vomit’u sosop tonsosop ‘suck’

Underlying and surface representations106

v ŋoap toŋoap ‘yawn’w dawoʔ tondawoʔ ‘fall’x balo tombalo ‘throw’y tunu tontunu ‘burn’z roŋor toroŋor ‘hear’aa uːs toŋuːs ‘chew’

1 Listthealternantsofeachprefix.2 Foreachalternant,listtheinitialsegmentofeachbasebeforewhichthe

alternant occurs.3 Foreachprefix,decidewhichalternantistheUR.Motivateyourchoice

with reference to forms f, h and aa.4 Tworulesareneededtoderivethesurfaceforms.Oneisaplaceassimila­

tion rule for nasals. The second is either a nasal insertion rule or a nasal deletionrule.Motivateyourchoicebetweenthesetwopossibilities.

5 Giveaformalnotationoftheruleyouhavearguedfor.

Q69 The underlying and surface forms of representative Dutch noun stems and diminutive forms are listed below (e.g. van de Weijer 2002; van der Hulst 2008).

a εi εicə ‘egg’b lɪp lɪpjə ‘lip’c drœyf drœyfjə ‘grape’d kraːχ kraːχjə ‘collar’e rɔk rɔkjə ‘skirt’f koːnəŋ koːnəŋkjə ‘king’g slɑŋ slɑŋəcə ‘snake’h stul stulcə ‘chair’i bɔm bɔməcə ‘bomb’j kɑr kɑrəcə ‘cart’k beːzəm beːzəmpjə ‘broom’l raːm raːmpjə ‘frame’m bɑl bɑləcə ‘ball’n snɔr snɔrəcə ‘moustache’o oːr oːrcə ‘ear’p sχun sχuɲcə ‘shoe’q pɑs pɑʃə ‘step’

1 Thelastruletoaffectthesurfaceformsassimilateslaminalcoronals(i.e.[tdszn])before[j]to[cɟ ʃ ʒ ɲ],respectively.Thecoronalconsonants [rl]areunaffected.Ifwetakeafollow-upruleofj-deletion for granted, how should palatalization beformulated?

Underlying and surface representations 107

2 Listthefivealternantsofthediminutivesuffixintheirshapesbefore pala­talization andj-deletionapplied.IfwetaketheURofthediminutivesuffixtobe[tjə], therearethreerulesthatwillbeneeded:t-deletion, whichdeletesthe[t]inthesuffix;ə­insertion; and place assimilation. Determinethecontextfort-deletion.

3 Determine the context for ə­insertion. Recall thatDutch has sets oftenseandlaxvowels,asshownin(6)inchapter6.

4 Giveaproseversionofplace assimilation.Whymustthisruleapplyafterə­insertion?

5 Underlyingformsofnounsthatendin[t]endupwithout[t]ineitherthestemorthediminutivesuffix,asshownintheformsbelow.Assumethat[t]inthestemisdeletedbydegemination.Foreachoftheformss,wand x, determine a unique pair of ordered rules among degemination, t­deletion, ə­insertion and place assimilation.

r naːlt naːlcə ‘needle’s mɑnt mɑɲcə ‘basket’t fut fucə ‘foot’u bɔrt bɔrcə ‘plate’v kɑst kɑʃə ‘cupboard’w bɔχt bɔχjə ‘bend’x hɛmt hɛmpjə ‘vest; shirt’

Q70 The underlying form of the diminutive suffix in the dialect of Utrecht is [ti]. Noun stems and diminutive forms are listed below (van den Berg 1975).

a εi εiχi ‘egg’b lɪp lɪpi ‘lip’c drœyf drœyfi ‘grape’d kraːχ kraːχi ‘collar’e rɔk rɔki ‘skirt’f koːnəŋ koːnəŋki ‘king’g slɑŋ slɑŋəχi ‘snake’h stul stultsi ‘chair’i bɔm bɔməχi ‘bomb’j kɑr kɑrəχi ‘cart’k beːzəm beːzəmpi ‘broom’l raːm raːmpi ‘frame’m bɑl bɑləχi ‘ball’n snɔr snɔrəχi ‘moustache’o oːr oːrtsi ‘ear’

Underlying and surface representations108

NOTE

1 Infact,theEnglishgeneralizationismuchmoregeneralthanthis.Itholdsthatnovoic­ing differences exist within sequences of obstruents inside the word.

p sχun sχuntsi ‘shoe’q pɑs pɑsi ‘step’r naːlt naːltsi ‘needle’s mɑnt mɑntsi ‘basket’t fut futsi ‘foot’u bɔrt bɔrtsi ‘plate’v kɑst kɑsi ‘cupboard’w bɔχt bɔχi ‘bend’x hɛmt hɛmpi ‘vest; shirt’

1 Listthealternantsofthediminutivesuffix.2 TheUtrechtgrammarhast-deletion, degemination, ə­insertion and

place assimilation, justasinStandardDutch.TwospecificallyUtrechtrulesareneededtoaccountforthealternationsinthediminutivesuffix.Determinethepositionofeachofthesetworulesintheorderedsetofrules.Formulatetheserules,usingsymbolsforsegmentsratherthandis­tinctivefeatures.(Hint:Theirformulationcanbeverysimpleiftheyareorderedright.)

3 Showthederivationofitemsa,e,i,l,t,uandw.

Levels of representation

8.1 INTRODUCTION

In this chapter, we consider the question of how many phonological representations awordhas.Sofarwehavepostulatedtwolevelsofrepresentation,anunderlyingoneandasurfaceone,apositionthatwasmotivatedinchapter7.Oneindicationthat this two­level model is inadequate is that it fails to account for the intuitions of native speakers about the pronunciation of the words of their language. In brief, the underlying representation would seem to be too abstract, while the surface representationappearstobetoodetailed(cf.Schane1971).TheadventofLexicalPhonology put an end to this particular shortcoming. It postulates an intermedi­ate level of representation, the lexical representation(Kiparsky1982a,1985,1993;Mohanan1986).Notonlydoes itcorrespondtonative-speaker intuitions, itwillalsobeshowntohaveanumberofinterestingproperties.Afterdealingwiththisthree­level phonological model, we turn our attention to the relation between the surfacerepresentationandthephysicalpronunciationofthewords.Wewillseethatlanguages typically differ in the way they realize surface representations which must beassumedtobeidentical.Suchlanguage-specificrealizationsofphonologicalele­ments are accounted for by rules of phonetic implementation.

8.2 DEFINING AN INTERMEDIATE LEVEL OF REPRESENTATION

If you look up the pronunciation of a word in a dictionary, you will find it is nor­mally given in phonemic transcription.Inthistypeoftranscription,anEnglishword like pin istranscribed[pɪn],not[phɪn].Thisisbecausethesegment[ph]isanallophonicvariantofthephoneme/p/,andassuchhasnoplaceinaphonemictranscription.Clearly,thedictionary’sphonemictranscriptiondefinesalevelofrepresentationwhich ismore abstract than the surface level.However, it doesnot correspond with the underlying representation either. The pronunciation of looked,forinstance,wouldbegivenas[lʊkt],notas[lʊkd].Theformthatisgivenincorporatestheoutputofadevoicingrulethatmakesobstruentsvoicelessaftervoiceless obstruents. That is, the phonemic transcription apparently corresponds to a level of representation which is somewhere between the underlying level and the surface level.

It might at first sight seem reasonable to suppose that the intermediate level cor­responds to the output of all rules that produce existing segments, which were called ‘neutralizationrules’inchapter7,whileallrulesthatproducenovelsegments,or

8

Levels of representation110

‘allophonic rules’, would then apply to the intermediate representation so as to pro­duce the more detailed surface representation. This would put all the rules that pro­duce phonemes, or existing segments, in a different compartment of the grammar fromalltherulesthatproduceallophones,ornovelsegments.Whilethisassump­tion is almost correct, there is still something not quite right. This is because there areruleswhoseoutputisamixofexistingandnovelsegments.Notallrulesallowthemselves tobecharacterizedaseither ‘neutralizing’or ‘allophonic’.Somerulesare both: depending on the input, they produce either an already existing segment oranovelsegment.Forinstance,Dutchhasarulethatvoicesplosivesbefore[b d],called regressive voicing. (Itappliestoallobstruents,butwewillignorethefric­ativeshere.)Thelanguagehasthethreevoicelessplosives[ptk],eachofwhichcanappearbefore/bd/.regressive voicingthusproducesthevoicedplosives[bdg].Dutchhasoppositionsbetween/p/and/b/,asin[pɑk]‘parcel’–[bɑk]‘tray’,and/t/ and/d/, asin[tɑk]‘branch’–[dɑk]‘roof ’,butthereisnocontrast/k/–/g/. Therefore, the output of regressive voicingispartlyphonemic,viz.when/pt/arevoicedto/bd/,andpartlyallophonic,viz.when/k/ isvoicedto[g].In(1),thefirstcolumngives the underlying representations, while the third column gives the results of regressive voicing.

(1) Underlying Phonemic Surfaceɔpdunœytbraːkzɑkbuk

ɔbdunœydbraːkzɑkbuk

[ɔbdun][œydbraːk][zɑgbuk]

‘puton’‘breakout’‘pocketbook’

It should be clear that it would not be very clever to maintain a distinction between rules that produce existing segments and rules that produce novel seg­ments, for this would mean that we would have to split regressive voicing, givenin(2),intotworules,onetoproducetheexistingsegments/bd/,the‘neu­tralizing’partoftherule,shownin(3a),andasecondtoproducethenovel[g],the ‘allophonic’part, shown in (3b). (We formulate the rules so as to voice allobstruents.)

(2) regressive voicing [–son] → [+voice] / __(##)

−−+

soncontvoice

(3) a regr.p,t-voicing −

→ + ( )−son

voices

COR

LAB

/ __ # #oon

contvoice−+

b regr.k-voicing −

→ + ( )

−−+

sonvoice

soncontvoice

DORSAL/ __ # #

Rule(2)convertstheunderlyingformsinthefirstcolumnin(1)intotheformsinthethirdcolumn,whichseemsjustright.Ifwewantedtomakeaclearcutbetweenaphonemicmoduleandanallophonicmoduleinthephonologicalgrammar,rule (2)

Levels of representation 111

wouldhave tobesplitup into(3a), theneutralizing,phonemicrule(toproduce /bd/)and(3b),theallophonicrule(toproduce[g]).

Halle(1959)usedtheaboveargument,whichhemadeonthebasisofasimi­lar case of undesirable rule duplication in Russian, to argue that the notion ofthe‘StructuralistPhoneme’,whichwasacornerstoneofphonologicaltheoryasitexistedbeforetheadventof‘GenerativePhonology’,wasmisguided.Thetwistliesinwhetherwethinkof[d]asproducedby(2)asbeing‘apronunciationof/d/’or‘apronunciationof /t/’.AmericanStructuralistphonologists likeZelligS.Harris,CharlesF.HockettandW.FreemanTwaddell,whoseworkwassupersededbythatoftheGenerativistsofthelate1950sand1960s,likeMorrisHalle,NoamChomskyand Paul Postal, held that the underlying representation was transferred into an intermediate, phonemic representation, which in its turn was transformed by rules into an allophonic surface representation. The notion phoneme is based on the existence of a surface contrast.The(entirelyreasonableandvalid)ideawasthatwhenever a minimal pair could be formed, i.e. a pair of words whose members differbyonesegmentonly,likeEnglishlock – rock, or beat – boot, the two segments responsibleforthedifferencemustbelongtodifferentphonemes.Sotheabovetwominimalpairs are evidence for the existenceof thephonemecategories /l/– /r/and/iː/–/uː/inEnglish.Whatmadetheirtheoryproblematic,however,wastheadditionalpremissthatasegmentcouldonlybelongtoonephoneme.Sooncea/t/hadbeensetupinalanguageonthebasisofsomeminimalpairinwhichthesegment[t]contrastswithsomeothersegment,anyoccurrenceof[t]inanyotherword, regardless of context, also belonged to thephoneme /t/.Thispremisshasbeenreferredtoas‘Onceaphoneme,alwaysaphoneme’.Adherencetothisprinci­ple leads inevitably to the sort of undesirable splits in generalizations that we have seeninthecaseofDutchregressive voicing . Since[d]belongstothephoneme/d/,aconclusionbasedonminimalpairsinwhichitcontrastswith[t]intheonset,[d]mustalsobeassumedtorepresent/d/whenitoccursinthecodabeforeavoicedplosive,asinthewordfor‘breakout’in(1).Sincethesamereasoningdoesnotapplyto[g],thepositionencapsulatedin‘Onceaphoneme,alwaysaphoneme’impliesthe recognition of an undesirable representational level illustrated in the second columnof(1).

Halle’s argument makes it clear that there is no place for the Structuralist Phoneme in phonological theory. It does not, of course, rule out the existence of any intermediate representation. The answer to the question of what the appropri­ateintermediatelevelishasbeenprovidedbythetheoryofLexicalPhonology.

8.3 LEXICAL PHONOLOGY

ThecrucialassumptionmadebyLexicalPhonologyisthatsomeofthephonologi­cal generalizations of a language are stated in the lexicon, the morphological mod­ule which incorporates the semantic, phonological and morphological information of the language’s morphemes, while others are stated outside it. That is, a distinction is drawn between lexical phonological rules and postlexical phonological rules,

Levels of representation112

the latter applying after thewordshavebeen inserted into thephrase (Kiparsky1982a;Mohanan1986).In(4),anumberofthedistinguishingpropertiesarelisted.

(4) Lexical rules Postlexical rulesa May refer to morphological labels Cannot refer to morphological labelsb May have exceptions Cannot have exceptionsc Structure-preserving Need not be structure-preservingd Accessible to native-speaker intuition Not easily accessible to native-speaker intuitione Cannot apply across word boundaries May apply across word boundariesf Must precede all postlexical rules Must follow all lexical rules

Wediscusseachofthesepropertiesinthefollowingsections.

8.3.1 Reference to morphological labels

Since lexicalrulesapply insidethe lexiconandpostlexicalrulesdonot,onlytheformerhaveaccesstocategorylabelslike‘N(oun)’,‘V(erb)’,etc.Forexample,thereisaruleinDutchthatdeletesword-final[n]after[ə],asshownin(5).

(5) Underlying Surface[[loːp]V + ən]Inf loːpə ‘towalk’[[zaːk]N +ən]Pl zaːkə ‘things’[teːkən]N teːkə ‘sign’[[teːkən]V +ən]Inf teːkənə ‘todraw’[oːpən]Adj oːpə ‘open’

When[ən]occursfinallyinaverbstem,however,nodeletiontakesplace(Koe­foed1979;TrommelenandZonneveld1983).Thisisshownin(6).

(6) Underlying Surface[teːkən]V teːkən ‘draw’[oːpən]V oːpən ‘open’

n­deletion(7)thusdistinguishesbetweenthenoun‘sign’andadjective‘open’in(5)ontheonehandandtheverbstemsfor‘draw’and‘open’in(6)ontheother.Suchaconditioncanonlybeputonalexicalrule.Onceawordhasleftthelexi­con and has been inserted in syntactic structure, category labels are removed, and postlexical rules therefore cannot refer to them.

(7) n-deletion n→Ø /ə__]X Condition:X ≠ Verb

8.3.2 Exceptions

Lexicalrules,butnotpostlexicalrules,haveaccesstothelexicon,andassuchcantellwhichwordtheyaredealingwith.Arulethathasexceptions,therefore,cannotbeapostlexicalrule.Toreturntotheexampleofn­deletion(7)above:[ˈhεidən]

Levels of representation 113

‘heathen’and[ˈkrɪstən]‘Christian’areexceptionalinnotundergoingtherule.Theentriesofthesewordsareassumedtobeprovidedwiththeinformation‘Notsubjectto(7)’.Similarly,Englishhasaruleoftrisyllabiclaxing, which laxes a vowel in theantepenultimatesyllableofwordsderivedwithsuffixeslike-ity. Examplesaregivenin(8).However, thewordsnicety and obesity (cf.nice, obese) exceptionally have[aɪ iː]intheantepenultimatesyllable,ratherthantheexpectedlax[ɪε].

(8) Tense Laxdiv[aɪ]ne div[ɪ]nityv[eɪ]n v[æ]nityser[iː]ne ser[ε]nity

Whilepostlexical rulescannothaveexceptions, lexical rulescouldeitherhaveexceptionsorbeexceptionless.Forinstance,Englishhasaruledeleting[n]after[m]attheendoftheword(Kiparsky1985).Thisrulemustbelexical,becauseitneeds information about the status of the word before inflectional endings are added.Thus,itappliesinthefirstcolumnin(9),andinthesecondcolumn,wherethe words have been provided with inflectional endings, but not in the third col­umn,where[n]isnotfinalintheword.Thislexicaln-deletion rule is exception­less:therearenowordsinEnglishthatendin[mn].

(9) Stem Inflectedform Derived formdam[Ø] dam[Ø]ed, dam [Ø]ing, dam[Ø] dam[n]ationcolum[Ø] colum[Ø]s, colum[Ø] colum[n]arhym[Ø] hym[Ø]s, hym[Ø]ing, hym[Ø] hym[n]al

8.3.3 Structure preservation

Lexicalrulesarestructure-preservinginthesensethattheiroutputisconfinedtosegments that already exist in underlying representations. The idea is that there is a lexical inventory of vowels, consonants and tones which is smaller than the inventoryobservableinsurfacerepresentation.Forexample,sinceintheunderly­ingrepresentationofEnglishwordsthereisnoneedtodistinguishaspiratedfromunaspiratedplosives,thisdistinctionbeingallophonicinEnglish,therulethatcre­atesaspiratedplosivesmustbepostlexical.Thesegments[ph th kh]arenovelseg­ments,i.e.notincludedintheEnglishlexicalsegmentinventory.

Structurepreservationisnotanexceptionlesspropertyoflexicalrules.Anum­berofvarietiesofEnglishhaverulesthatapplybeforetheaffixationofinflectionalendingsandmustforthatreasonbelexical,liketherulethatdeletes[n]inwordslike autumn, discussed above. The point is that many such rules produce novel seg­ments (Harris1994:28).Anexample is theScottishrule lengtheningword-finalvowels(aswellasvowelsbefore[+voice,+cont]segments)(Aitken’sLaw).Unex­pectedly,theinflectionalsuffix[d]ofthepasttenseandpastparticipleisignoredinthecontextoftherule,whichmakestherulealexicalrule.Thisisshownin(10).Buttheproductoftheruleisanovelsegment.

Levels of representation114

(10) Uninflectedstem Past tense formfid feed friːd freedtʌɪd tide taed tiedsʌɪd side saed sighedfud food suːd sued

Interestingly, there is evidence that when a novel sound is produced by a lexical rule, it may be made available for inclusion in lexical representations. Thus, some speakers pronounce concise and scythe with[ae],eventhoughinScottishEnglishthesewordsendin[s]and[θ],respectively(Aitken1984).Clearly,developmentalstages in which all occurrences of a novel segment are produced by a lexical rule must be expected to occur, if it is assumed that at least some lexical rules historically start out as postlexical rules.

8.3.4 Native-speaker intuitions

Native speakers would appear to make reference to the lexical representationwhen determining whether two phonetically different sounds are ‘the same sound’; their judgements refer to the lexical segment inventory. For instance,nativespeakersofEnglishregard thesecondsegment in stop and the first seg­ment of top as the same sound, even though they are phonetically different, which fits with the assumption that aspirationisapostlexicalrule.Likewise,phoneti­cally identical sounds that were neutralized by a postlexical rule will typically belookeduponasdifferentsounds.AsaresultoftheAmericanEnglishprocessof flapping, the intervocalic consonants in Adam and atom are phonetically identicalinallstylesexceptthemostformalones([ˈæɾəm]),butnativespeakersneverthelessconsiderthemdifferentconsonants.Bycontrast,whenalexicalruleneutralizes an underlying opposition, the intuition of the native speaker tends to conform to its output.

8.3.5 Application across word boundaries

Becauselexicalrulesapplyinthelexicon,theirstructuraldescriptioncanneverbedeterminedbyelementstakenfromdifferentwords.Arulethatappliesacrosswordboundaries,therefore,mustbeapostlexicalrule.Dutchregressive voicing (2)isa postlexical rule for this reason.

Q71

1 Whatdoyouthinkisthelexicalrepresentationof‘breakout’in(1)?2 DoyouthinkthatDutchregressive voicing hasexceptions?3 Whatwouldnative speakers ofDutch say is the last consonant of the

prefixin[œyd-braːk]?

Levels of representation 115

8.3.6 Lexical rules apply before postlexical rules

Thefinaldistinguishingpropertylistedin(4)isoncemoreanecessaryconsequenceoftheLexicalPhonologymodel.Wordsgetinsertedintopostlexicalstructuresintheirlexicalrepresentations,i.e.afterall lexicalruleshaveapplied.Itfollowsthatifweknowthata rule ispostlexical, a rule thatmustapplyafter itmustalsobepostlexical,andthusdisplaythepostlexicalpropertieslistedin(4).

8.4 PHONOLOGICAL INFORMATION IN THE LEXICON

Allphonological rulesmay refer tophonological information.Thephonologicalinformationavailableinthelexiconisnotconfinedtosegments.SyllableandfootstructurealsoexistinthelexiconofDutchandEnglish(Booij1988;Inkelas1989).Evidenceforthispositionisprovidedbymorphologicalprocessesthataresensitivetothesyllablestructureorthestressofthebase.Forexample,theEnglishcompara­tiveandsuperlativesuffixes[[]Adj ər]and[[]Adj əst]requirethatthebaseshouldnotexceedabinaryfoot.Therefore,theformationisallowedwiththeadjectivesin(11a)butnotwiththosein(11b).

(11) a white (waɪt)F (waɪtər)F

noble (nəʊbl)F (nəʊblər)Fsilly (sɪli)F (sɪliər)F

b beautiful (bjuːtəfl)F *(bjuːtəflər)manifest (mænə)F (fest)F *(mænə)F (festər)F serene sə(riːn)F *sə(riːnər)F

Apossibleillustrationofthepointthatlexicalrulesmayrefertobothmorphologi­cal structure and phonological structure is provided by final devoicing in two vari­etiesofGerman.InLowGerman,thevarietiesofGermanspokeninthenorthernhalfof the country, this process applies to syllable­final obstruents, as shown in the second columnof(12)(Venneman1972).Therequirementthattheobstruentshouldoccurin thecodaalsoholds forHighGerman,but thisvarietyrequires inadditionthatinpositionsbeforesonorantstheobstruentshouldbemorpheme-final.Asaresult,theexamplesin(12b),inwhichtheobstruentisbothsyllable-andstem-final,showdevoicing in both varieties, but differences appear in the case of such words as Adler ‘eagle’,whereanobstruentissyllable-final,butnotstem-final,asshownin(12c).

(12) Underlying Low German High Germana kɪnd kɪnt kɪnt ‘child’

ˈkɪnd-ɪʃ ˈkɪn.dɪʃ ˈkɪn.dɪʃ ‘childish’b ˈkɪnd-lɪç ˈkɪnt-lɪç ˈkɪnt-lɪç ‘childlike’

ˈtaːg-lɪç ˈteːk-lɪç ˈteːk-lɪç ‘daily’c ˈvaːgnɐʀ ˈvaːk.nɐʀ ˈvaːg.nɐʀ ‘Wagner’

ˈmagmaː ˈmak.maː ˈmag.maː ‘magma’ˈaːdlɐʀ ˈaːt.lɐʀ ˈaːd.lɐʀ ‘eagle’ˈɔrdn-ʊŋ ˈɔrt.nʊŋ ˈɔrd.nʊŋ ‘order’

Levels of representation116

These data have been given different accounts, though. Rubach (1990) provides an account based on resyllabification of the High German consonant sequences ([e.g. ˈvaː.gnər]). Similarly, Giegerich (1992) assumes that the plosive moves to syllable-initial position in some speech styles. A drawback to these solutions is that they run counter to the generalization that clusters which are ill-formed word-internally, like French [ps-] in capsule [kap.syl], *[ka.psyl], may be well-formed word-initially, as in psychologie (Selkirk 1982), but that the opposite case (e.g. *[dl] word-initially, but admissible word-medially in a case like Adler) has not otherwise been attested.

Q72 The Korean lexical consonant inventory is given in (1).

(1) lab cor [+ant] cor [–ant] dorsal larb d dʒ gp t tʃ kph th tʃh kh

s hz

m n ŋl

Before [i], [d t th s z n l] are prepalatal by a rule of palatalization: [ɟ c ch ɕ ʑ ɲ ʎ], as shown in (2), causing surface syllables like *[si] to be ill-formed.

(2) gaps-i gapɕi ‘price+nominative’sigan ɕigan ‘time’kini kiɲi ‘meal’bʌthi bʌchi ‘endure’thi chi ‘dust’gaksi gakɕi ‘bride’madi maɟi ‘knot’

A rule of affrication causes coronal plosives to become affricates before [i], if [i] forms part of a suffix. Thus, [d th] are replaced with [dʒ tʃh] in the contexts shown in (3a). ([t] happens not to occur stem-finally.) The affricates [dʒ tʃh] also appear in underlying representation, as shown in (3b) (after Kiparsky 1993; symbols as in Lee 1993).

(3) a bath-i batʃhi ‘field+nominative’gud-i gudӡi ‘harden+adv’gjʌth-i gjʌtʃhi ‘side+nominative’

b dӡip dӡip ‘house’dӡidӡʌ dӡidӡʌ ‘tear (imp)’dӡʌd-i dӡʌdӡi ‘milk+nominative’

Levels of representation 117

1 Ispalatalizationastructure-preservingrule?2 Isaffrication astructure-preservingrule?3 Mentionthreepropertiesofaffrication that are consistent with its sta­

tus as a lexical rule.

Q73 French has a rule of vowel nasalization, which nasalizes a vowel before [n] in the same syllable (Tranel 1981). After vowel nasal-ization, a rule of n-deletion deletes [n] if it appears in the coda after a nasalized vowel. This is illustrated in (1), where the feminine forms were created by suffixing the stem with [ə]. (The nasal vowel [ɛ] is the reflex of underlying [i y e]; please ignore the variation between [e] and [ε].)

(1) Stem Masc. Fem. Derivationplεn ‘full’ plɛ plεnə plenitydə ‘fullness’bryn ‘brown’ brɛ brynə brynε ‘brownish’fin ‘fine’ fɛ finə finεs ‘fineness’bɔn ‘good’ bɔ bɔnə bɔnifije ‘make good’roman ‘Romance’ romã romanə romanist ‘scholarofRomance’

There is a third rule, ə-deletion, whichdeletesfoot-final[ə]exceptinthemostformalstyles,causingthefeminineformsin(1)tobe[plεn,bryn,fin,bɔn,roman]onthesurface.

1 Give formal notations of the three rules, referring to syllable and footboundariesasinexample(18)insection6.4.1.

2 Showthederivationofthemasculineandfeminineformsfor‘fine’.3 Beforeaconsonant-initialnoun, the feminine indefinitearticle is [yn],

asin[ynfam]‘awoman’;beforeavowel-initialfemininewordlike[e.ro.in] ‘heroine’, the indefinite article [yn] isdividedover two syllables, asin[y.ne.ro.in]‘aheroine’.Themasculineformoftheindefinitearticleis[ɛ]beforeconsonant-initialnouns,asin[ɛ garsɔ]and[ɛn]beforevowel-initialnouns,wherethe[n]issyllabifiedwiththenoun-initialvowel,asin[ɛ.nɔm]‘aman’.Howwouldyouaccountforthepreservationofthe[n]inthemasculineindefinitearticle?

4 Whymustn­deletion bepostlexical?5 Insomecontexts,word-final [ə] ispreserved,as in[bãdədεsine] ‘strip

cartoon’. Is ə­deletion alexicalorapostlexicalrule?6 Citroën ispronounced[si.tʀo.εn].Whycanthisfactbeusedasanargu­

ment for the assumption that vowelnasalizationisalexicalrule?7 There aremanywords like [ɔdə] ‘wave’,which only ever have a nasal­

izedvowel,neveranoralvowelplus[n].Whatwouldyouassumeastheunderlyingformofthewordfor‘wave’?

Levels of representation118

8.5 CONTROVERSIAL PROPERTIES OF LEXICAL RULES

Otherpropertiesthathavebeenclaimedtodistinguishlexicalfrompostlexicalrulesarenotasconvincingasthepropertiesmentionedin(4),andhavethereforebeendisputed(HalleandKenstowicz1991;Kiparsky1993).Oneoftheseisnon-derived environment blocking. This is the phenomenon that many lexical rules would appear toskipunderivedwords, i.e. toapplyonly to forms thatarederived.Forinstance,Englishtrisyllabiclaxingappliestoderivedformslike[ˈvænəti]van-ity, [dəˈvɪnəti]divinity, but never to underived forms like ivory [ˈaɪvəri]*[ˈɪvəri],nightingale [ˈnaɪtɪŋgeɪl]*[ˈnɪtɪŋgeɪl].Accordingly,ithasbeenproposedthatlexicalrulescanonlyapplytoderivedforms.However,manycaseshavebeenpresentedinwhich structure­preserving rules apply to all occurrences of a morpheme, derived or underived.

Q74 In Dutch, word-final [n] is deleted after [ə] (section 8.3.1). Does the rule show the effect of non-derived environment blocking?

8.6 BEYOND THE SURFACE REPRESENTATION

WehaveseenabovethatthephonologicalgrammaradvocatedbyLexicalPhonol­ogycanbeschematizedasin(13).

Q75 In Northern Irish English, all occurrences of [εː] can be derived from [iə] (Harris 1994). Some alternations are shown below.

[iə] [εː] [iə] [εː]fate day station pay themmade stayed same say moreraise rays fail dailybaby playful

1 Istherulethatproduces[εː]alexicalrule?2 Isitaneutralizationrule?3 Isitstructure-preserving?4 Doesitshowtheeffectofnon-derivedenvironmentblocking?

Levels of representation 119

(13) UNDERLYING REPRESENTATION

Lexical rules

Postlexical rules

LEXICAL REPRESENTATION

SURFACE REPRESENTATION

↓

↓

(13)

The surface phonological representation which is produced by the grammar contains all the linguistic information the articulators require to do the work of pronouncing the expressions concerned. In other words, it is a fully specified pho­nological representation.Converting thatcognitive representation into thephysi­ological actions that constitute the articulation of the expression is the task of the phonetic implementation rules.Thediagramin(13)canthereforebecompletedasin(14).

(14) UNDERLYING REPRESENTATION

Phonology

Phonetics

ARTICULATORY PROGRAMME

Lexical rules

Postlexical rules

Phonetic implementation rules

LEXICAL REPRESENTATION

SURFACE REPRESENTATION

↓

↓

↓

(14)

8.6.1 Phonetic implementation

The phonological surface representation which ultimately results from the phono­logical grammar consists of some configuration of phonological features and struc­tures. The translation of this discrete representation into quantitative physiological activity,thearticulatoryactionsdiscussedinchapter2,istakencareofbyrulesofphonetic implementation (Pierrehumbert1980,1990;Keating1990).These takephonological presentations as input and deliver continuous acoustic signals as out­put.Ithasbeenclaimedthatearlierwork,likeChomskyandHalle(1968),assumedthat such rules were universal, but we have not found that assumption expressed in print. In any event, everyday experience tells us that this would be an improbable position. The same phonological syllable is always pronounced somewhat differently byspeakersofdifferentlanguages.Forinstance,asyllablelike[kis]willtypicallybepronouncedwitha somewhatmore fronted [k] inFrench than in eitherSpanishorDutch,while [s]willhave lower-frequency friction inSpanish than inFrenchand perhaps Dutch. Phonetic implementation is not to be thought of as giving

Levels of representation120

speaker­specific invariant outputs. It is not an ‘automatic’ algorithm, but a complex set of articulatory procedures whose execution is under the control of the speaker (KingstonandDiehl1994).Speakerswillmakeadjustmentssoas topreservethedistinctiveness of phonological contrasts, depending on ambient noise or distance to the hearer. In addition, a host of non­linguistic factors will affect the articulation. Speakersofthesamelanguagewillfine-tunethearticulationofcertainsegmentsinordertosignalthesocialgrouptheybelongto.Asaresult,wecanoftenformafairidea of the speaker’s social background on the basis of the precise quality of their vowels, say. Similarly, the speaker’s emotionwill affectpronunciation in complexways, while communicative urgency will determine the precision with which speech is produced.

8.6.2 Models of implementation

In a widely adopted view, phonetic implementation involves the translation of a phonological feature into a target for the designated articulatory parameter. Forinstance,thefeature[–cont]triggerstheformationofanoralclosureatthelocation specifiedby theplace features.A feature like [+nasal]will trigger anopeningofthevelumforthesegmentspecifiedforthatfeature,while[–nasal]will do the opposite. The different targets of the same articulatory parameter will be connected up by transitions known as interpolations. Thus, the movement

Q76 In New York City English, BATH-raising raises and diphthongizes [æ] to [εə] or [ɪə]. The rule applies in a number of phonological con-texts, among which is ‘when a nasal follows in the same syllable’. It thus fails in family [fæ.mə.li] but applies in damn [dɪəm]. Moreover, it applies in underived major-class stems, so that the stressed minor-class auxiliary can has [æ], while the noun can (of beer) has [ɪə]. Like-wise, swam (a derived past tense of swim) has [æ], while ham has [ɪə].

1 IsBATH-raising a rule of phonetic implementation, a postlexical pho­nologicalruleoralexicalphonologicalrule?Motivateyouranswer.

2 Englishswan [swɔn]usedtohaveanunrounded[a]intheoldeststagesofthelanguage.The[w]caused[a]tohaveliprounding,afterwhichitwasinterpreted as [ɔ],unless a velar consonant followed. This also took place in words like quality, whichwereborrowedfromNormanFrenchwithanunrounded[a]aftertheinvasionof1066.However,the[a]inpresent-daywordswith[wa]doesn’tacquirethisrounding,asshownbyaquatic [əˈkwætɪk],andthereisthusnoproductiveruleof[a]-roundingtoday.HowcouldyoushowthatBATH-raisingisproductive?Hint:ThinkofaNewYorkschoolchildlearningaforeignlanguage.

Levels of representation 121

of the velum from a raised to a lowered position constitutes the interpolation between two phonologically specified target positions. This view of phonetic implementationimpliestwothings.First,phoneticimplementationrulesdonotthemselves add or remove phonological features: the only thing they can do is translate features into articulations. Second, segments that have no specifica­tion for a particular feature may be in the path of an interpolation between two segmentsthatdo.Forinstance,Cohn(1990)showsthatthenasalizationofthevowelinAmericanEnglishwordslikeden, room, sung shouldbedescribedinjustthat way. The nasalization starts early in the vowel and progressively increases to reach full velic opening at the end of the vowel. This is shown graphically in (15a),wherethegraphrepresentsthenasalairflow,or–indirectly–thedegreetowhichthevelumislowered.Cohnarguesthatin(15a)thevowelhasnospecifi­cationfor thefeature[±nasal],andthatitsnasalityresultsfromtheinterpolationbetweenthe[–nasal]specificationoftheprecedingconsonantandthe[+nasal]specificationofthefollowingconsonant.Bycontrast,inwordslikecamp, dent, sink, the velum opens fully right at the start of the vowel and closes fairly sud­denly at the moment when the oral contact for the voiceless plosive is made. In fact, the nasal consonant disappears before a voiceless plosive in the coda, so that the nasality of the vowel can be seen as the transfer of that feature from the deleted[mnŋ].Thisisshownin(15b).(Thevoicelessplosiveisaccompaniedby a glottal closure introduced by an independent rule of glottalization.)In(15c),thenasality‘cline’runsintheoppositedirection,inwordslikemood, Ned. Thus,onlyin(15b)doesthenasalityofthevowelresultfroma[+nasal]featureon the vowel, which is supplied by a phonological rule nasalizing vowels before asequenceofatautosyllabic(i.e.occurringinthesamesyllable)nasalandplo­sive.In(15a,c),thenasalityonthevowelresultsfromaninterpolationbetweenoppositevaluesfor[±nasal]oneithersideofthevowel.

(15)

Analternativeviewholdsthatphonologicalfeaturesaretranslatedintogestures.That is, instead of defining the beginning and end points of articulatory movements, theelementsinagesturalmodelarethearticulatorymovementsthemselves(Zsiga1997).Underthisview,thephonologicalfeaturesinthesyllable[dεn]aretranslatedintoalaryngealgesture(voicing),twotonguetipgestures(onefor[d]andonefor[n]),atonguebodygesture(palatalwide,toproducethequalityof[ε]),andavelicclosing and a velic opening gesture. Their timings are governed first by the order of the segments and second by the more detailed, language­specific instructions. If the opening gesture of the velum is initiated at the same time as the tongue tip closing gesture, there will be some nasalization during the last part of the vowel: the velum opens immediately when the gesture is started, while it will take the tongue tip some time actually to reach the alveolar ridge and, with the rims of the tongue,

Levels of representation122

make the alveolar closure. In a language that allows a greater degree of nasalization, the velic opening gesture is extended so that it starts earlier.

The developers of the gestural model had originally intended the gestures to be the phonological features encoding the pronunciation of words in the lexicon (BrowmanandGoldstein 1989).Aproblemwith this ‘phonetic’ viewof phono­logicalrepresentationsisthatitmakesitdifficulttostatethecontrastsalanguageemploys. It would, for instance, be impossible to say whether the nasalization during avowelwascontrastiveorallophonic.Arelatedassumptionwasthatphonologicaladjustmentsarosefromextendingthedurationofgestures,suchthatgesturescometooverlap(asinthehypotheticalcaseofthenasalizationofprenasalvowelsnotedabove).This,however,makesitimpossibletodistinguishbetweenphonological,i.e.categorical,effects(whichtypicallycausegesturestodisappearorbereordered)andgradientphoneticeffects(whichcanbeveryadequatelyexpressedbyadjustingthedurationsofgestures).Thegesturalmodelisthusbettersuitedfordealingwiththephonetic implementation than for encoding underlying forms.

Recently,phoneticimplementationhasbeenthoughtoflessastheconstructionof an acoustic signal on the basis of gesture­based or target­based interpretations of phonological features than as the probabilistic selection of auditory representations rememberedbythespeaker,knownas‘exemplars’.Speakersappeartomodifythepronunciation of specific words on the basis of the pronunciation of other speakers whom they recentlyheardpronounce thosewords.Theseword-by-word adjust­ments suggest that we store detailed phonetic information about individual words. Awell-knownargument for the fact thatwe storeminimal informationabout aword’s phonology is that we cannot usually say what any word of our language, piano, say, sounded like when we first heard it, and are typically unable to say whether the speaker was male or female, adult or child, or what its pitch pattern or speechtempowas(Halle1985).Adherentsofexemplarmodelswillnodoubtgrantthat this is true, but point out that we do remember the last time or couple of times weheardit.Tacitlyorovertly,werememberagreatdealaboutspeakersandtheway they said things, and we may well use that information during our own speech production. There is no conflict between the two facts, however, since one refers to our grammar, our knowledge of contrasts and the structures they are embedded in, whiletheotherreferstohowwepronouncedthem(Pierrehumbert2002).Modelsof phonetic implementation that are based on word­based phonetic knowledge are known as ‘exemplar models’.

8.6.3 Deciding between phonology and phonetic implementation

It is not always obvious whether a particular generalization should be accounted forinthephonologyorinthephoneticimplementation.Areasonableassumptionwould be that if the output of the rule crucially feeds into another rule, the regu­laritymustbephonological.Anexamplemayclarifythispoint.InmanyvarietiesofBritishEnglish,includingReceivedPronunciation(RP),thereisatendencytoinsert a voiceless plosive between a nasal and a following fricative, if they belong to

Levels of representation 123

thesamesyllable.Asaresult,awordlikesense is pronounced as if it was the plural of cent,i.e.as[sεnts]ratherthan[sεns].Atfirstsight,therewouldappeartobetwopossiblewaysofaccountingforthisprocess.First,itmightresultfromthedetailsofphoneticimplementation.For[n],thereisanalveolarclosureandthevelumislow­ered.Togetto[s],thealveolarclosuremustbegraduallyreleasedwhilethevelummustberaised.Theoccurrenceofthe[t]willnowdependontheprecisetimingofthese two articulatory gestures: if the velum is raised before the alveolar closure is released, an oral stop will come to exist between the moment the velic closure is madeandthealveolarclosureisreleased.(Thevocalfoldswilltypicallystopvibrat­ingoncethevelicclosureismade,andthestopthereforeisvoiceless.)Thisaccountisinprinciplequiteplausible,butwouldnotappeartoberightinthecaseofRP.Asithappens,theinsertionof[t]createsthecontextforaphonologicalprocesswhichglottalizes voiceless stops in the coda of the syllable. If the insertion of the voiceless stop in sense is recorded in the phonological representation, we should expect it to trigger preglottalization. The fact that such inserted stops do trigger preglot­talization, justasdounderlyingstops,impliesthatvoiceless stop insertion must be a phonological process.

Q77 Would you expect sense and cents to be homophonous in British English?

Bycontrast,theequivalentprocessinAmericanEnglishisaninstanceofpho­netic implementation (Ohala 1986).Thefirst syllable in aword like teamster or sensitive may well be followed by an intrusive voiceless stop, but it does not feed into the phonology of the language: these syllables appear to be longer than sylla­blesclosedbyavoicelessstopwouldbeinthesamecontext.Also,wordslikesense and cents are distinct. Cents has a categorically nasalized vowel, as well as a glottal stop([sɛʔts](see(15b)),whilesense has a partially nasal vowel and no glottal stop, apronunciationcorrespondingto(15a):[sɛɛn(t)s]).Thus,inanSPE­type descrip­tion,theBritishEnglishsituationiscapturedby(16),whiletheAmericanEnglishsituationisbestdescribedbysomesuchinstructionasin(17).

(16) BrEstop insertionØ→−−

+

−

contvoice

nas so

PLACEPLACE

ασ

α/ [ __

nncont+

] σ

(17) AmE stop insertion (Implementation): When realizing an intraword sequence [+nas]-[–son,+cont]beforeanunstressedsyllable,timethevelicclosurefor[s]justbeforeoratthemomentofthereleaseofthealveolarclosurefor[n].

Levels of representation124

Q78 Before voiced obstruents, vowels are longer than before voiceless obstruents. Languages vary in the amount of lengthening before voiced obstruents. The difference in French ([vid] ‘empty’ vs [vit] ‘quickly’) is much smaller than in English ([biːd] bead vs [biːt] beat), for instance. How would you account for this lengthening before voiced obstruents?

Q79 English [s] before [ʃ], as in stocks shelves, trace shapes, often assimi-lates to a consonant that is indistinguishable from [ʃ], but is also fre-quently pronounced as a fricative that gradually moves from [s] to [ʃ] (Holst and Nolan 1995). How would you characterize these two situations in terms of the rule typology discussed in this chapter?

8.7 CONCLUSION

WehaveshownthatLexicalPhonologyprovidesasatisfactoryanswertotheprob­lem of the definition of a level of representation between the underlying representa­tion and the surface representation. Instead of drawing the dividing line between rules of neutralization and rules of allophony, it assumes that phonology exists in twoseparatecomponentsinthegrammar.Onepartresidesinthelexicon,whereit can refer to information present in the lexicon, which is, first, morphological information and, second, the set of lexical phonological segments and structures. The other part is outside the lexicon, where morphological information is no longer available and segments can be produced that do not form part of the lexical seg­mentinventory.Anumberofpropertiesthatcanbeassociatedwithlexicalrulescontribute to the coherence and explanatory power of this distinction between lexi­cal and postlexical rules.

Postlexical rules must be distinguished not only from lexical rules but also from phoneticimplementationrules.Whilepostlexicalrulesarephonologicalrules,andcan therefore only manipulate the phonological elements that representations con­sist of (features, segments,moras, etc.), phonetic implementation rules canonlytranslate the phonological representation into actions of the articulators. The effect of phonetic implementation rules may resemble the effect of phonological rules, and in practice it is not always easy to tell when a rule is phonological.

It might be thought that if lexical rules can be sensitive to morphological infor­mation, postlexical rules will be able to refer to syntactic information. There is

Levels of representation 125

some evidence against this assumption.Rather, itwould appear that postlexicalrules refer to representations that are part of a phonological constituent structure, known as the prosodic hierarchy. The theory of Prosodic Phonology, to be dis­cussedinchapter12,providesananswertothequestionofwhatthatconstituentstructure is and how it might be related to syntactic structure.

Between the segment and the syllable

9.1 INTRODUCTION

The next three chapters are devoted to phonological elements in the representations of words and sentences that go beyond the strings of consonants and vowels. This chapter introduces the syllable,aconstituentthatgroupssegments.Chapter10willdeal with tones, whether functioning in the representation of words or as part of the intonation,whilechapter11introducesthefootandwordstress.

Asyouwillhavecometoexpectfromchapter3,theallowablesegmentalcompo­sitions of syllables show considerable cross­linguistic similarities, despite their large rangeofcomplexity.WewillseethatthesimilarityislargelyduetotheeffectoftheSonorityProfile, which suggests that the ideal syllable has a burst of acoustic cues(structurallycorrespondingtotheonset)followedbyadrawn-outsonorouspart(structurallycorrespondingtotherhyme)(section9.2).Wewillnextseethatbetween the segments and the syllable an additional level of structure is required. Animportantroleitismeanttofulfilistherepresentationofsegmentalduration.There have been two ways in which this intermediate level has been conceptual­ized.OneistheCV-tier,alevelofstructurebetweenthesegmentsandthesyllablespecifiedasC-slotsandV-slots,introducedinsection9.3,togetherwiththenotionsof autosegment and association. The other approach assumes a rhyme constituent whichdominatesmoras,or lengthunits(section9.4). Ineitherapproach, lengthisrepresentedintermsofthenumberofunits,C-slotsandV-slotsinthefirstcaseandmorasinthesecond.Anotherimportantrolefortheintermediatelevelofrep­resentation is to specify the membership of segments to the subconstituents of the syllable, the onset and rhyme, the latter dividing into peak and coda(earlierintro­ducedinsection3.2).Weconsidertheroleofthesyllableandsyllablemembershipin phonological generalizations about the distribution and realization of segments (section9.5),includingthosethatrequiretheexistenceofambisyllabicconsonants(section9.6).

9.2 SYLLABIFICATION: THE MAXIMUM ONSET PRINCIPLE

Languagesvarygreatly in thecomplexityof theirsyllables.SomeonlyallowCV,quiteafewhave(C)V,stillothersallowacoda,andfurthercomplexityisachievedby allowingmore than one segment in onset, peak or coda (section 3.2). Inde­pendently of this variation, there are striking cross-linguistic similarities. An

9

Between the segment and the syllable 127

intervocalic consonant will occur in the onset of the second syllable, rather than the coda of the first. That is, consonants prefer to form an onset rather than a coda, if they can legitimately do so. The principle responsible for maximizing the onset is (1)(Kahn1976).

(1) MaxiMuM onset principle (MOP) Firstmaketheonsetaslongasitlegitimatelycanbe;thenformalegitimatecoda.

TheMOP requires that a string like [tata] should be syllabified [ta.ta], ratherthan *[tat.a].Andifthelanguageallows[st]onsets,thenastringlike[asta]willbe[a.sta],ratherthan*[as.ta]or*[ast.a].Ofcourse,itisessentialtokeepinmindthat,while theMOPisauniversalprinciple, languagesdiffer in thekindsof syllablesthey allow.For example,Dutch andEnglish allow [st] in theonset, but Spanishdoesnot.Applicationof(1)toaSpanishstring[basta]willthereforeyield[bas.ta],not *[ba.sta].Bythesametoken,Germansyllabifiesextra as[εks.tra],butDutchas[εk.stra],becauseDutchhas[st]onsets,butGermandoesnot,eventhoughitdoeshave[ʃt]onsets.

Languagesdiffer inthesyllabification domain. InWestGermanic languages,the syllabification domain tends to be the word, or, rather, the constituent to be introducedinchapter12asthephonologicalword.Forinstance,theEnglishgrey tiles is not homophonous with great isles, thefirstsyllablebeing[greɪ]inthefirstphraseand[greɪt]inthesecond,whichdifferencewillhaveaneffectonthedura­tionof[eɪ],inparticular.Otherlanguagesmayhavelargerdomains,likeFrench,which syllabifies across words and where petit ami ‘little friend’ and petit tamis ‘littlesieve’areboth[pə.ti.ta.mi],orsmallerdomains,likeJapanese,wherecertainsuffixeswillnotsyllabifywiththestem,asshownby/a.ni/‘brother’and/an.i/‘easygoing’. In the Japanese case, the corresponding phonetic difference is very clear, because the onset nasal is coronal, but the coda nasal more likely to be velar or uvular.Thatis,/a.ni/ispronounced[ani],and/an.i/[aŋʔi].Moregenerally,pre­fixesinGermandonotsyllabifywiththestem,asinVerein [fεrʔain]‘club’,whereVer- is a prefix.

9.2.1 The Sonority Profile

The segmental composition of onsets and codas shows striking similarities across different languages. If you were to bet on the type of consonant that could occupy the secondposition inaCConset in someunknown language, your chancesofwinningwouldbewellservedbyoptingforaglidelike[j]or[w].Andifyourbetconcernedthefirstconsonant,youwouldbewelladvisedtogofor[p],[t]or[k].Venneman(1972)describedthesetendenciesintermsofanumberofsyllable‘laws’.MuchoftheregularityiscapturedbytheSonorityProfile,givenin(2).

(2) sonority profile Thesonorityofasyllableincreasesfromthebeginningofthesyllableonwards,anddecreases

fromthebeginningofthepeakonwards.

Between the segment and the syllable128

Intuitively,sonorityisrelatedtotheoverallacousticenergyofsegments.In(3),the classes of segments that are usually distinguished along this dimension are listed in the order of increasing sonority.

(3) sonority scale Obstruents–Nasals–Liquids([lr],etc.)–Glides([wj],etc.)–Vowels

Thus, the observation is that any onset that reverses the direction of increasing sonority,like[mk-]or[wl-],islesscommonthanonethatdoesnot,like[pn-]or[ml-].Conversely,anyrhymethatincreasesthesonorityfromlefttoright,like[-lj],isdisfa­voured.ViolationsoftheSonorityProfileareindeedrare.Swedishhassyllableslike[bærj]‘mountain’,whileDutchhassyllableslike[ʋrɪŋ]‘wring’,wherethefirstconso­nantisalabiodentalapproximant.Inaddition,Clements(1990)observesthatintheonset largesonoritydifferencesarepreferredoversmallones,making[pj-]abetteronsetthan[lj-],whileintherhymesmallsonoritydifferencesarepreferredoverlargeones,making[-j]abettercodathan[-t].Apparently,syllables,liketheworldinT.S.Eliot’sThe Hollow Men,prefertostartwithabangandendwithawhimper.Becausethe beginning of the syllable is maximally salient as a result, catching the listener’s attention when it should, the reason for this state of affairs is probably perceptual.

ThereisaninterestingconsequenceoftheobservationbyClements(1990)thatsyllables are marked by a sudden increase in sonority followed by a drawn­out sonorouspart.Syllablestendtogroupinwordssothatthesonorityoftheendofone syllable is greater than that of the beginning of the next, favouring a whimper­bang transition over a bang­whimper one. In languages without complex onsets or codas,[al.ka]willthereforebeamorelikelystructurethan[ak.la].Thistendencywas earlier described as the SyllableContactLawbyVenneman(1972).Soalsowhen considering the relation between syllables, the tendency is to maximize the bang at the beginning.

Sonoritycanbedefinedintermsofthefeaturesintroducedinchapter5and6(Clements1990).Thefirstfoursonorityclassesof(3)arecharacterizedbythefea­tures [±consonantal], [±sonorant], [±approximant]. To distinguish vowels fromapproximants, syllabicity needs to be called upon, if these are represented in terms ofthesamefeatures,whichintheanalysispresentedinchapter6isthecasefor[w]and[u],forinstance(seealsoQ145).

Q80 What do the English words stigma, comrade and Daphne have in common?

9.3 EXPANDING REPRESENTATIONS: HIERARCHIES AND AUTOSEGMENTS

In this section, we introduce two ways in which phonological representations go beyond a linear representation, i.e. a single string of phonological elements, which was characteristic of earlier theories. There are two ways in which representations are

Between the segment and the syllable 129

moreelaboratethanasinglestring.Oneisthehierarchicalrepresentationbywhichseg­ments are dominated by constituents like the syllable. The second is less self­ evident. This is the arrangement of different classes of segments as strings that are parallel in time. The next two sections introduce these two enrichments of the representation.

9.3.1 Skeletal slots

There is a lot of evidence that segmental length must be represented separately from thephonological contentof segments.Aparticularly clear example ispro­videdbylanguagegames.Oneofthese,Ludikya,reportedbyClements(1986)forLuganda,involvesareversaloftheorderofthesyllablesintheword.Lugandahasbothavowel-lengthcontrastandacontrastbetweenlongandshortconsonants.Weindicate long segments by doubling the symbol. The striking thing is that although the syllables move round in this game, the durational structure remains intact, as shownin(4).

(4) Luganda Ludikya mukono nokomu ‘arm’ mubinikilo lokinibimu ‘funnel’ baana naaba ‘children’ ɟɟuba bbaɟu ‘dove’ kiwoɟɟolo loɟowwoki ‘butterfly’ kubaɟɟa ɟabakku ‘toworkinwood’

In order to express the mutual independence of segmental length and segmental quality,ClementsandKeyser(1983)assumedthatsegmentsarenotimmediatelyassociated with their syllables, but are dominated by structural positions, known as(skeletal)slots, which encode segmental length, also known as segmental quan­tity.Consonantsandvowels thatareassociatedwith single slotsare short,whilelong vowels and geminate consonants are represented as being doubly linked to two slots.Additionally,theyassumethattherearetwotypes,aCandaV,wheretheCrepresentsasyllablemargin(onsetorcoda)andVasyllablepeak.Thatis,CV-slotsindicate a segment’s membership of subsyllabic constituents, in addition to provid­ing a representation for segmental quantity.

TheCV-slotsaredominatedbythesyllablenodes,eachsyllablebeingrepresentedby σ.Forexample,alanguagewithalengthcontrastforvowelsaswellasforconso­nants,liketheDravidianlanguagesMalayalamandTamil,wouldhaverepresenta­tionslikethosein(5),whichrepresenttheTamilwords[paʈu, paːʈu, paʈːu, paːʈːu],respectively(Firth1957;Mohanan1986:108).All fourwordsconsistof twosyl­lableseachandhavethesamestringofsegments.However,theydifferintheirseg­mentaltimingstructureasexpressedbytheC-slotsandV-slots.

(5) (5)

C b. C C C C C C C Cc. d.| | || | | | | | | | |p a p au u p a u p a u

‘to endure’ ‘to sing’ ‘a song’‘enduring’

| | | | | | | |V V V V V V V Va. CV V

Between the segment and the syllable130

TheCVrepresentationallowsustocharacterizetheLugandawordgameLudikyaasillustratedin(6).InLudikya,onlypartofeachsyllableistransposed,thestringofconsonants and vowel segments. The remainder of the representation, the structure abovethesegments(C-slotsandV-slots),isretained.Thisisshownin(6),wherethetoplinegivestheorderofthesegmentsintheLugandaword,whilethebottomlinegivestheLudikyaversion.Observehowthestringsofskeletalslotsareidenticalin the two versions of each word.

(6) (6) b a n a| | |

C C| | |n b a

||

||V V V

k a a| | | | |C C C C

a b

b

a k

u

u

||

| | | | |||V V V

a

In(5)and(6),segmentsareincludedinhierarchicalstructures(section1.5).Adistinctionissometimesmade,orimplied,betweenthelinesconnectingseg­ments and higher constituents and the lines that connect elements of different ranksinthehierarchy,liketheslotsontheCV-tierandthesyllables.Thosethatconnect segments, elements with featural content, to structural constituents are referred to as association lines, the term phonologists use to indicate the structural position of elements with phonological content relative to ‘packaging’ units.Thus, inLuganda, the [ɟ] in ‘towork inwood’ in (6) isassociated (alsoknownas‘linked’)totwoC-slots,whileinLudikyaitis[k]whichhasthisdoubleassociation.

9.3.2 Autosegments

McCarthy(1985a)usedtheCV-tiertoaccountforthefactthatcertainmorphemesinArabicappeartobespecifiedintermsofstringsofconsonantandvowelposi­tions,referredtoastemplates.ThemorphologyoftheArabicverbincludesanum­ber of derivationalmorphemes (‘conjugations’),which arenot just expressed byparticularaffixes,butalsobydifferenttemplates.Insomecasesthereisnoaffix,sothatconjugationscandifferonlyintheCVtemplatetheyhave.Theverbalrootcon­sists only of consonants, usually three. The vowels in such verbal forms represent thethirdtypeofmorpheme,whichcorrespondstoverbalaspectorvoice.In(7),someofthesemorphemesarelisted.Averbalform,then,consistsminimallyofaverbalroot(twoormoreconsonants),aconjugation(aCVtemplate)andaverbalaspect(oneormorevowels).

(7) ktb ‘write’ ħq ‘betrue’ CVCVC ‘Plain’ CVCCVC ‘Intensive’ CVVCVC ‘Influencing’ a ‘ActivePerfective’ ui ‘PassivePerfective’

Between the segment and the syllable 131

In (8), the six forms (three conjugations by two aspects) for ‘write’ are given as well as five forms for ‘be true’. The final [a] is 3sg;masc, not shown in (9).1

(8) Conjugation Active Passive Plain katab-a ‘he wrote’ kutib-a ‘it was written’ Intensive kattab-a ‘he caused kuttib-a ‘it was made

to write’ to write’ Influencing kaatab-a ‘hecorresponded’ kuutib-a ‘hewas

correspondedwith’ Plain ħaqq-a ‘itistrue’ – Intensive ħaqqaq-a ‘herealizedst.’ ħuqqiq-a ‘ithasbeenrealized’ Influencing ħaaqaq-a ‘hecontested ħuuqiq-a ‘hisrightwas

sb.’sright contested’

To derive these forms, we need to assume that the consonants and vowels form separate, parallel segment strings. When there is more than one string of segments in the representation, the segments in each string are known as ‘autonomous’ seg-ments, or autosegments for short (Goldsmith 1976). Autosegments were first introduced and motivated for tones, as we will see in the next chapter. Because the strings of consonants and vowels are located on different tiers, consonants can associate with C-slots and vowels to V-slots without getting their association lines crossed or entangled. The C-slots and V-slots themselves form a single string, the CV-tier. In (9), the two segmental tiers are shown above and below the CV-tier. Imagine that between each segment tier and the CV-tier there is a plane across which lie the association lines, as if the CV-tier was the spine of a spiral-bound book with two pages. Association lines between a consonant and non-adjacent C-slots will not be able to interfere with association lines between a vowel and one or more V-slots, because they are drawn on different pages.

(9) (9) k b k t b

C V C V C

ia ‘he caused to write’

t

C C C V V V C

u‘he was corresponded with’

When a segment associates with two slots, it has spread from the first to the second slot if the perception is one whereby you work from left to right through the segmental string and create the associations as you go. For the morpheme ‘be true’ we could either assume an underlying ‘triliteral’ root ([ħqq]) or a ‘biliteral’ one ([ħq]). If we assume that the direction of association is left to right, the form [ħaqqaq-a], for instance, can be accounted for by the spreading of [q] in a biliteral root to all available C-slots, as shown in (10): the left-over C-slots are filled by the spreading of the last consonant. If surface repetitions of the same consonant are produced by the left-to-right spreading of a rightmost consonant segment to empty C-slots, as shown in (10), the first two of three surface consonants in verbal and

Between the segment and the syllable132

nominal forms will never be identical. This prediction is by and large correct, a rare exceptionbeing[dadan]‘plaything’(McCarthy1985a:146).Thus,aconstraintontheunderlyingformofArabicrootsisthatthereshouldbenoadjacentidenticalconsonants. It is akin to the ban on sequences of identical tones, to be discussed in chapter10,andknownastheObligatoryContourPrinciple(OCP).

(10) (10) qq

CCCC CC C

a i‘he realized s.t.’ ‘his right was contested’

VV VV V

u

Q81 A secret language based on Amharic uses the forms in the second column for the words in the first (McCarthy 1985b). The apostrophe indicates glottalization of the consonant; geminate glottalized con-sonants are only marked once for glottalization.

Amharic Disguised forma gɪn gainən ‘but’

mətt’a mait’ət’ ‘come’kifu kaifəf ‘cruel’t’ətt’a t’ait’ət’ ‘drink’hed haidəd ‘go’wəddədə waidəd ‘love’

b wərk’ wairk’ək’ ‘gold’təmara taimrər ‘learn’sɪgara saigrər ‘cigarette’səkkərə saikrər ‘drunkard’kəbad kaibdəd ‘difficult’wɪʃət waiʃtət ‘lie’

1 What is the difference between theAmharicwords and the disguisedwords?

2 Whichsegmentsarepreservedinthespeechdisguisedforms,andwhicharenot?

3 Whatdeterminesthenumberof(surface)consonantsinthespeechdis­guisedforms?

4 Giveanaccountoftheformationofthespeechdisguisedforms,employ­ingarepresentationwithaskeletalCV-tier.Consider[ai]asingleshortvowelthatassociateswithasingleV-slot.

5 HowdoesyourdescriptionofthissecretlanguagedifferfromArabicver­balmorphology?

6 Givetherepresentationsofthedisguisedformsfor‘drink’and‘gold’.

Between the segment and the syllable 133

Q82 Scottish Gaelic has a three-way length opposition for vowels, ‘short’, ‘half-long’ and ‘fully long’, as illustrated by [ʃin] ‘we’, [ʃiːnj] ‘venison’, [ʃiːːnj] ‘to sing’, or [tuɫ] ‘to go’, [uːɫ] ‘apple’, [suːːl] ‘eye’. At most, two dif-ferent vowel qualities may occur per syllable, as illustrated in [ian] ‘John’ and [iaːn] ‘bird’. In addition, diphthongs are either half-long or fully long. Moreover, the distribution of the durations of two intra-syllabic vowel segments is ‘short, short’ (as in [ian]) or ‘short, half-long’ (as in [iaːn]), but never ‘half-long, short’ (*[iːan]) (Ternes 1973: 96). How would you account for these facts?

9.3.3 Unfilled and unassociated slots

TheslotsintheCVskeletonarestructuralpositions,whichintheusualcasearedominatedbyasyllablenodeandareassociatedwithsegments.Byassumingthatslots don’t have to be associated with any segments and need not be dominated by a syllable,theirexplanatoryrolecanbeextended.ClementsandKeyser(1983)callonboththesepossibilitiestoaccountforthealternationbetweenFrenchmorpheme-finalconsonantswithØ,knownasliaison, and the phenomenon of h-aspiré.

TheFrenchdefinitearticleis[lə](masc;sg)or[la](fem;sg)forthesingular,and[le](pl),asshownin(11a).Whenthenounbeginswithavowel,thevowelofthesingular definite article is deleted, while the plural is followed by an apparently inserted[z],ascanbeseenin(11b).Infact,manywordsappeartohaveaconsonantthatonlyshowsupbeforeavowel,like[pəti],whichhasa[t]in[pətitami]‘littlefriend’. These potential consonants are called liaison consonants.

(11) Singular Plural a ləbwa lebwa ‘wood’ ləpa lepa ‘step’ lakaraf lekaraf ‘carafe’ lanɥi lenɥi ‘night’ b labe lezabe ‘priest’ lide lezide ‘idea’

The loss of the vowel of the singular definite article is effected by elision (12),whichdeletes thefinalV-slotof thedefinite articlebeforeV, causing its syllablenode to be lost as well.

(12) V

Ø(12) ELISION

[…]DEF

/ V→

Liaisonisexpressedin(13),whichsaysthatanonsetlesssyllablesyllabifiesanyunsyllabified consonant appearing before the vowel, as shown by the dashed line.

(13) (13) LIAISON

C V

Between the segment and the syllable134

These rules produce the correct results if the singular and plural forms of the arti­clearerepresentedasin(14a,b).ThesingularwillloseitsvowelbeforeaV-initialword, while the unsyllabified final consonant in the plural will only be syllabified, andpronounced,beforeaV-initialword.

(14) b.

C C V C| | |l e z

‘DEF; PL’‘DEF; SG; FEM’

V

a.(14) σ σ

|a

|l

Certainvowel-initialnounsappeartobehaveasiftheybeganwithaconsonant:they take the preconsonantal alternants of the definite article, both in the singular andintheplural,asshownin(15).Inordertoaccountforthebehaviourofwordslike ‘hero’ and ‘hatred’, their underlying forms are assumed to begin with an empty C,whichissyllabifiedasanonsetCjustasanyfilledCwouldbe.Thisisshownin(16b),whichrepresentationshouldbecomparedwith(16a).

(15) Singular Plural ləero leero ‘hero’ laεn leεn ‘hatred’(16) (16) a

|V|a

V

e||

C

b

b

C V|e

V|o

|r

C

Thus,thedeletionofpre-V[a]in(17a),aswellasitsretentioninthepre-Ccontextof(17b),arereadilyaccountedfor.Likewise,thedifferencebetweenthepresenceoftheliaison[z]in‘thepriests’(18a)anditsabsencein‘theheroes’(18b)isexplainedbytheinabilityoftheunsyllabified[z]toundergoliaisonin(18b).Thissolutionaccurately expresses the fact that words with h-aspiré behave as if they began with a consonant,eventhoughnosurfaceconsonantisobserved.Also,itcorrectlycharac­terizes both the distribution and the identity of the liaison consonant.

(17) (17) ba

C C| |

C|l

V|e l a

C|n

V|i

C|d

V|ε

V

(18) (18) ba

C|l

V|e

C|z

C

z|

e|V V

|a

C|b

C C|l

V|e

CC|z

V|e

C|r z

|o|V

9.3.4 Compensatory lengthening

Afinalargumentfortheexistenceoftheskeletaltieristhephenomenonofcom-pensatory lengthening.Frequently,thelossofasegmentisincompleteinthesense

Between the segment and the syllable 135

that the time it took before it was deleted is preserved in a neighbouring segment. TheparentlanguageofEnglishandFrisian,whichatonetimeconstitutedaWestGermanic dialect sometimes referred to as Ingwæonic, underwent a process ofnasal loss before fricatives within the word. Prior to the loss of the nasal, the vowel beforeitwasshort,butalongvowelremainstoday.In(19),wegivereconstructed(hypothetical)forms.ThepresenceofthenasalisattestedinGothictexts,aswellasinmodernGerman.(Probably,thenasalizationwaspreservedonthevowelatfirst,andwaslostlater.Weignorethisinthetranscriptions.)

(19) gans gaːs ‘goose’ fimf fiːf ‘five’ tanθ taːθ ‘tooth’ munθ muːθ ‘mouth’

The representation with the skeletal tier allows one to express the change as a retimingofthesegments,asshownin(20).Anadditionalmeasure,onewhichistriggeredbythe[–cons]segmentassociatedwithit,isachangeofthevacatedC-slotintoaV-slot.

(20)

Compensatorylengtheningcanalsoberecognizedinconsonantsthathavecometooccupythesyllablepeak(syllabic consonants).ThesecondsyllableinEnglishbottle [ˈbɔtl]consistspurelyoftheconsonant[l]:thereleaseoftheplosive[t]takesplacethroughaloweringofthesidesofthetongue,thetipretainingitscontact(lat-eral plosion).Assumingthatthedurationofthesyllabic[l]isnodifferentfromthedurationof[əl],alesscommonbutattestedpronunciation,thesyllabicconsonantcouldberepresentedasin(21).

(21)

Retimingofsegmentsisawell-knownfeatureofBantulanguages.Classpre­fixeslike[ki]usuallyonlyshowupinthatformbeforeconsonant-initialstems.Whenthestembeginswithavowel,however,theclosevowelturnsintoaglidewhile thevowel lengthens.Consider theLugandaexamples in (22), fromCle­ments(1986).

(22) li+ato ljaːto ‘boat’ mu+iko mwiːko ‘trowel’ ki+uma kjuːma ‘metalobject’ mi+aka mjaːka ‘year’

Between the segment and the syllable136

If it is assumed that the resulting onset does not consist of two consonants, but of a single, complex consonant which is either labialized (in the case of retimed [u]) or palatalized (in the case of retimed [i]), the process can be described as in (23).

(23) (23)

C V C V C V C V| | | | | |l a t o l j a t o

V|i

V|

Q83 In Catalan, a word-initial high vowel turns into a glide when pre-ceded by a stressed vowel, as in [kuntəˈrajsˈtɔriəs] contará históries ‘he will tell stories’ and [kətəˈlawniβərˈsal] catalá universal ‘universal Catalan’ (Mascaró 1970; Hualde 1992). Assuming that the duration of the vowel plus glide is equivalent to that of a short vowel, how could you describe this process in CV phonology?

9.4 Moras

The skeletal tier fails to account for two phenomena (Hayes 1989a). One is that compensatory lengthening always occurs in the case of segments deleted from the rhyme, never in the case of segments deleted from the onset. This suggests that segments in the rhyme possess something that other segments do not possess. Sec-ond, it appears that many languages distinguish syllables on the basis of quantity, a property of syllables which is determined by the number of segments in the rhyme, again to the exclusion of their onset. As for the first objection, consider the fate of the onsets [kn- gn-] in Middle English. When [k g] were lost, and the English syllable no longer admitted onsets that consisted of [−cont] [+nas], there was no compensatory lengthening of any of the other segments in words like knot and gnat, now pronounced [nɔt] and [næt], not *[n:ɔt] or *[nɔ:t], etc.

Second, the location of word stress frequently appears to be sensitive to the seg-mental composition of the rhyme, while the number of segments in the onset is typi-cally irrelevant (for some such cases, see Topintzi 2010). In Hawaiian, for instance, stress falls on the last syllable if it contains a long vowel, and on the penultimate syl-lable if the last is a short vowel, as illustrated by [naˈnaː] ‘strut’, [ˈnaːna] ‘for him’ and [loːˈʔihi] ‘long’. Significantly, a consonant in the rhyme is often counted as if it was a vowel. In Hopi, the stress falls on the first syllable if the rhyme contains a long vowel or a short vowel plus a coda consonant, but on the second syllable if the first syllable contains a short vowel (Jeanne 1982). That is, the consonant in the rhyme can deter-mine the location of the stress in a way that a consonant in the onset cannot.

To capture the difference between segments in the rhyme and segments in the onset, the mora has been proposed as an intermediate level of structure in the rhyme (Hyman 1985; Hayes 1989a). The prosodically ‘active’ status of a segment can then be expressed by giving it moraic status, while ‘inactive’ segments are non-moraic.

Between the segment and the syllable 137

A language thathasnovowel-lengthdistinction anddoesnot allowa codawillhaveonlymonomoraicsyllableslike(24a).Longvowelslooklike(24b),whilenon-moraicandmoraicconsonantsareshownin(24c,d).Geminateconsonantsmustbemoraicinordertoexpresstheirduration.TheItaliandistinctionbetween[ˈfato]‘fate’and[ˈfatːo]‘fact’isexpressedbyrepresentingthesingletonconsonantasanonset consonant, while the geminate is a consonant that associates with a mora in thefirstsyllableandisa(non-moraic)onsetinthesecond,asshownin(25).

(24) (24)

short V

c.

non-moraic coda moraic coda

b.

long V

d. a.

t t t t tta a aa

(25) (25) ba

f a t o f a t o

Althoughitisnotuncommonforlanguagestoallowthelastsyllableofthewordto have three moras, languages that more generally allow their rhymes to have three moras are rare. If a language has coda consonants, usually only the first consonant afterashortvowelismoraic,whilethenextconsonantoranyconsonantafteralongvowelisnon-moraic.Languagesthathavethreedegreesofvowellength,likeDinka,and languages that allow a geminate to follow a long vowel have trimoraic syllables. Tamil,aswesawin(5),isanexampleofalanguagewithgeminatesafterlongvow­els.Themoraic representationsof those four structuresaregiven in (26),where(26a)hasamonomoraic,(26b,c)abimoraicand(26d)atrimoraicfirstsyllable.

(26)

a a a ap p p pu

c.b.

u u u

d.a.

Trimoraiclanguagesarerare,anditisthereforecommonforlongvowelstobebanned before geminates. For instance, Koya words like [aːnd] ‘female power’,[manasuːrku] ‘men’, [meːnduːli] ‘back’ are unremarkable in the language, as areshort vowels before a geminate, as in [pɪkːa] ‘cup’. In these words, long vowelsappear before a coda consonant, and short vowels before a geminate, but a form like *[puːtːi],inwhichalongvowelappearsbeforeageminate,isill-formed.Whenthisconfiguration arises as a result of morphological concatenation, the vowel is short­ened, as shownby /keː­tː­oːɳɖa/,which surfaces as [ket.toːɳ.ɖa].,where the fullstopindicatesasyllableboundary.TheKoyarhymethushasmaximallytwomoras,where the second can be either the second half of a long vowel or a consonant;

Between the segment and the syllable138

anyadditionalconsonantsarenon-moraic.Sincegeminatescannotbenon-moraic,theyrequirethevowelbeforethemtobeshort(Tyler1969).

Q84 Give the moraic-cum-syllabic representations of the Koya words [pɪkːa], [manasuːrku] and [beske] ‘when’.

Q85 The syllable structure of Tera is CV(X), where X can be V or C. Except at the end of the sentence, word-final [ə] is deleted unless the preceding syllable has a long vowel or when it is immediately preceded by two consonants. Thus, the words in the first column are subject to the rule, but not those in the third (Newman 1970). ([mb] and [nd] are single consonants.)

ə deleted ə retainedsəɗə ‘snake’ pərsə ‘horse’wurə ‘tomorrow’ meːnə ‘today’mbukə ‘to throw at’ xənndə ‘nose’

1 Whatdeterminesif[ə]isdeleted?2 DoesTerahavemoraicconsonants?Motivateyouranswer.

Q86 Tagalog has a denominal formation meaning ‘an imitation of’ (Schachter and Otanes 1972). The formation involves the reduplication of the noun and the suffixation of [an], as shown below. Syllable onsets in Tagalog are C or CC, where – at least in the native vocabulary – the C in second position is either [j] or [w]. The syllable may have a coda, which is obligatory in word-final position. In addition to many other consonants, [h] and [ʔ] can occur at the end of the word, but these con-sonants do not appear word-internally in the coda. Vowel length is in general unpredictable, except in word-final position, where vowels are always long. This can be seen in the forms in both columns. However, vowel length in ‘imitation of’ forms is subject to further restrictions.

Base Derived forma baːhaːj ‘house’ bahajbahaːjaːn ‘doll house’b libruːh ‘book’ libruːlibruːhaːn ‘imitation book’c paːriːʔ ‘priest’ pariːpariːʔaːn ‘pretended priest’

Between the segment and the syllable 139

9.5 SYLLABLE-BASED GENERALIZATIONS

Like all phonological constituents, the syllablemayhave a role in the structuraldescription of phonological generalizations. It is very common for consonants to be pronounced differently depending on their position in the syllable. Thus, as was explainedinchapter7,BritishEnglish[l]isaccompaniedbyvelarizationwheneverit appears in thecoda, as shown in (27).Observe that [lj-] is a legitimateonset,occurring word­initially in a word like lure [ljʊə],forinstance.

(27) In onset In rhyme [laɪ] lie [hɔːɫ] hall [lʊk] look [kɪɫt] kilt [ˈvæli] valley [ˈmɪɫdjuː] mildew [ɪˈtæljən] Italian [ˈmɔrəɫ] moral

TheneedtoincludesyllablestructureinEnglishphonologicalrepresentationswasdemonstratedbyKahn(1976).Heshowedthatinmanycasesthecontextofa rule could only be expressed in segmental terms at the cost of fairly baroque specifications. In the case of l­velarization, we would have to state that it applied ‘before all consonants except [j] andat thewordend’.Obviously, this approachdoes not yield a natural class. It would also not be possible to derive the non­velarizedallophonefromanunderlyingvelarized[ɫ],asinthatcasewewouldhavetostatethecontextas‘beforeallvowelsand[j]’.Again,thisisnotanaturalclass,since[w]isexcluded.Referencetothesyllablemakesitpossibletoformulatethecontextasin(28).

(28)

Manyotherexamplescanbegivenofconsonantshavingdifferentallophonesdependingonwhether theyappear in theonsetor the rhyme.Thus, theDutch

d bulaklaːk ‘flower’ bulaklakbulaklaːkaːn ‘artificial flower’e kabaːjuːh ‘horse’ kabajuːkabajuːhaːn ‘imitation horse’f daːhoːn ‘leaf ’ dahondahoːnaːn ‘imitation leaf ’

1 Therearetwogeneralizationstobemadeaboutvoweldurationin‘imita­tion of ’ formations. The first concerns syllables that are not stem­final. Whatisit?

2 Thesecondgeneralizationconcernsstem-final(butnotword-final)syl­lables.Whatisit?

3 IsthesecondgeneralizationmoreeasilyexpressedintermsofamoraicrepresentationoraCVrepresentation?Motivateyouranswer.

Between the segment and the syllable140

labialapproximantislabiodental[ʋ]intheonsetandroundedbilabial[β]inthecodainthevarietiesspokeninthenorthernhalfoftheDutchlanguagearea,asshownin(29).Again,neitherthegeneralization‘inthecoda’northegeneraliza­tion‘intheonset’canbeexpresseddirectly.SinceDutchallows[ʋr-]onsets,theonsethappenstobeequivalentto‘beforeavowelor[r]’,whilethecodamustbecapturedby‘beforeaconsonantotherthan[r]orawordboundary’.Suchstruc­turaldescriptionsclearlylookarbitraryandfailtoexpressjustwhatitisthatgov­erns the allophony.

(29) In onset In rhyme [leːˈʋɪn] ‘lioness’ [leːβ] ‘lion’ [ʋreːt] ‘cruel’ [χeːβt] ‘yawn+3sg’ [ˈkleːʋaŋ] ‘scimitar’ [ˈlyβtə] ‘lull’

Q87 In Bugotu (Solomon Islands), vowels can be long or short. Any two vowels in either order can form a diphthong, such as [ao], [io], as inferred from the data in Kennedy (2008). It has a reduplicative prefix whose phonological form depends on the base, as illustrated in (1a,b,c,d). The full stop indicates a syllable boundary, and a dash indicates that the unreduplicated form is not used. The language has undominated NoCoda, and a form like *[kal.ka.lu] would therefore be ungrammatical.

(1) a ka.lu – kau.ka.lu ‘to stir up’ka.ve ‘grandmother’ kae.ka.ve ‘to be old, of woman’pa.ŋa – paa.pa.ŋa ‘to pant’lo.po ‘be in a roll’ loo.lo.po ‘to fold’po.no.ti ‘to close’ poo.po.no.ti ‘to patch’

b koa – koa.koa ‘fornication; to sin’mee ‘to be foolish’ mee.mee ‘to be importunate’rei.ŋa ‘to look’ rei.rei.ŋa ‘appearance’

c a.θo ‘cord, rope’ a.θoa.θo ‘to signal’i.li ‘to totter’ i.lii.li ‘to be drunk’u.do.lu ‘to be gathered’ u.dou.do.lu (no gloss available)

d ou – ou.ou ‘to cough’ao – ao.ao ‘crow, bird’iu ‘dog’ (mane)iu-iu ‘loafer’

The prefix has a segmentless structure of two moras which is filled with a copy of the segments from the base, left to right.Unparsed segments aredeleted. Observethatinthisproblemamorphologicaloperationinvolvestheaddition of a specific phonological structure which packages segments. This frequent phenomenon is known as prosodic morphology (McCarthy1998).

Between the segment and the syllable 141

Theunderlyingformof[kau.ka.lu]isthusasin(2a)andthatof[poo.po.no.ti]asin(2b).

(2)(2) a b

kalu + k a l u ponoti + p o n o t i

1 Showtheautosegmentalsurfacerepresentationof[kau.ka.lu].2 Dep(σ)bans the addition of syllable nodes, Max(μ)bans thedeletion

of moras, and Onsetrequiresthatsyllableshaveonsets.Drawatableauwith these three constraints and (2a) as input. Assume moraic-cum- syllabicrepresentationsof[kau.ka.lu],[*au.ka.lu],*[a.lu.ka.lu],*[ka.ka.lu]and *[ka.lu.ka.lu]ascandidateoutputs.Dotheseconstraintshavetoberankedtomake[kau.ka.lu]thewinner?

3 Thetableauyoudrewupforquestion2explainswhy[a.θoa.θo]isbetterthan *[a.θa.θo]and*[a.θo.a.θo],butnotwhyitisbetterthan*[θoo.a.θo].Showthesefourcandidatesinatableauandproposeaconstraintwhichisviolated by *[θoo.a.θo]butnotby[a.θoa.θo].Doesyourconstrainthavetoberankedwiththeotherthree?

4 What would be the reduplicated form of a hypothetical loanword [ko.kao]?

9.6 POST-MOP SYLLABIFICATION RULES

TheMOP(section9.2),togetherwiththespecificationofasyllabificationdomainwithin which it is applicable, suffices to explain the syllable structure of manylanguages.Other languages have additional syllabification rules.Themotivationfor these additional rules is that consonants may behave as if they are in the coda according to one generalization, while at the same time acting as onset consonants of the following syllable according to another generalization. There have been two syllable-basedapproachestothissituation.OneistoredotheinitialMOP-drivensyllabificationafteraderivationorcliticization, so thatwhatappearedasacodaconsonant at one level of representation appears as an onset consonant at another, a description known as resyllabification. If the first syllabification was V.CCV,asshownin(30b), thesecondwouldbe(30c).OtherdescriptionswillgoontheassumptionthattheMOPisapersistentprincipleandthatadditionalsyllabificationrules must respect existing syllabification. The new syllabification is ‘added’ to the old, and as a result a consonant may simultaneously belong to the coda of one syl­lable and to the onset of the next: such consonants are said to be ambisyllabic.AsstressedbyKahn,thelatterpositionismorerestrictive,inthatitlimitsthenumber

Between the segment and the syllable142

of possible syllabifications. In the hypothetical situation (30), the syllabificationrequiredby theMOPwouldbeV.CCV,andonly thefirstCcanbeambisyllabic(30a),sincetheNoCrossingConstraint rules out the ambisyllabicity of a sec­ondC.Bycontrast,inaresyllabificationanalysis,thereisnoprincipledlimittothenumberofconsonantsthatcanberesyllabified.Startingoutwith(30b),wecouldchangethatinto(30c),butthereisnoreasonwhyafurtherresyllabificationcouldnotchangethattoVCC.V(notshown).Inthissection,wewillarguethatambisyl­labicitycanbemotivatedonthebasisofKahn’sdescriptionof(American)Englishflapping.

(30) (30)

VCCV

b.

VCCV

c.

CCV

a.

V

9.6.1 Ambisyllabicity in English

Englishhastwopost-MOPsyllabificationruleswhichcreateambisyllabicconso­nants. liaison applies across word boundaries so as to cause a word­final conso­nanttobeintheonsetofafollowingvowel-initialword.Sinceitdoesn’ttherebyloseitsassociationasacodaconsonant,anambisyllabicconsonantsiscreated.Toreturntotheexample insection9.2, /t/ ingreat isles is both coda in great and onset in isles,while/t/ingrey tiles isexclusivelyanonsetconsonant.Rule(31)onlyappliesacross word boundaries, because word­internally onsets will have been created by theMOP.Somemoreexamplesaregivenin(32).Aftertheapplicationofliaison, the word­initial vowel can no longer be preceded by a glottal stop.

(31)

C V

(31) LIAISON

(32) AmbisyllabicC faɪvˈεgz /v/ fiveeggs oʊldˈɪŋglɪʃ /d/ Old English saʊθˈæfrəkə /θ/ South Africa

The other rule creating ambisyllabic consonants is rightcapture, which causes thefirstconsonantofanunstressedsyllabletoserveasa(final)codaconsonantofaprecedingsyllable,asformulatedinGussenhoven(1986).Thatis,itisasyllabificationrulewhichapplieswithinthedomainofthefoot.TheEnglishfootisleft-dominant,inwhichtheweaksyllable(s)typicallyhave[ə i ʊ]or,especiallyword-finally,[ioʊ].Notethatothervowelsarestrongand,therefore,defineafoot(seechapter11).Foot-basedrightcapture,givenin(33),spreadsanonsetconsonanttoaprecedingsyllable,whereit forms part of its coda.

(33)

C

(33) (… …) FRIGHT CAPTURE

Between the segment and the syllable 143

Asaresultof(33),the[p]inawordlikehappy will be the final consonant of the firstsyllableandthefirstconsonantofthesecondsyllable.AsinthecaseofMOP,wemustmaketheassumptionthat theresultantcoda iswell-formedinEnglish.That is, rightcapture cannot create a syllable *[bεdf]in[ˈbεdfərd]Bedford, even thoughthiswordisasinglefoot.Examplesofwordsinwhichconsonantsbecomeambisyllabicby(33)aregivenin(34).In(34a),theruleappliesafteranopensyl­lable,in(34b,c)itappliesafteraclosedsyllable,while(34d)showsthattheleft-handsyllable may be unstressed.

(34) a ˈsɪti [t] city b ˈkɑnsərt [s] concert c ˈkʌntri [t] country d əˈspεrəgəs [r],[g] asparagus

AspointedoutbyKahn(1976),AmericanEnglishflapping, the rule that causes certaininstancesofintervocalic[td]tobepronouncedasanalveolarflap,providesevidence that ambisyllabic consonants are a natural class. flapping applies in what areatfirstsighttwoentirelyunrelatedcontexts.In(35),thecontextsinwhichtherule applies are compared with a context in which it does not. The curious gener­alization that emerges is that flapping applies when the right­hand vowel is in the next word, or is in a weak syllable in the same word.

(35) flappingapplies Noflapping flappingapplies [ˈleɪtər] later [ˈleɪtεks] latex [ˈleɪtεks] late ex [ˈhɪtɪŋ] hitting [ˈhɪtaɪt] Hittite [hɪtˈaɪk] hit Ike [ˈsɪti] city [ˈsætaɪər] satire [ˈnaɪtaʊl] night owl

The unifying element of the contexts in which flapping applies is ambisyllabic­ity. liaison and rightcapture, which can both be motivated on independent grounds (cf. alsoRubach1996),provide theappropriate representations for rule(36),whichsimplysays‘Flap[td]whentheyareambisyllabic’.

(36)

It has been claimed that languages do not contrast ambisyllabic and geminate consonants,whichsuggeststhattheyberepresentedidenticallywithinwords(vanderHulst1985).Accordingly,anambisyllabicconsonanthasanassociationwiththe last mora of the syllable it closes and with the onset of the next, as shown in (37a)forOld English and(37b)forcity.

(37)

Between the segment and the syllable144

Therearethreealternativesolutionsthatmakenouseofambisyllabicity.First,inadescriptionwithoutsyllabicstructure,wewouldarriveat(38),afterKahn(1976),where the angled brackets indicate mutually dependent terms.

(38) flapping[td]→[ɾ]/[–cons]__<##>a +syl

stress<− >

b

Condition:ifnota,thenb

In addition to the awkward condition ‘within the word, the following vowel must be unstressed’, we also need to stipulate that in slow, deliberate speech styles, in which rules like liaisondonotapply,word-final[td]willnotbeflapped.Thiswillmakeitnecessary to add a second condition: ‘In unconnected speech, not a.’ The second alter­native is a resyllabification account. This theory will have to assume that flapping appliestocodaconsonants(Selkirk1982).Inorderforthisaccounttogothrough,there must be no liaison(inordertokeepfinal[td]incodaposition),whilerightcapture must be replaced with a rule that delinks the initial onset consonant of a weak syllable and moves it into the coda of a preceding syllable, so that get $ Anne, get $ it and gett$ing are syllabified as indicated by $. The two­stage syllabification will havetofaceanumberofsurfacefacts.Oneisthatrealcodaconsonants,like[d]inToo bad! or[t]inThe lot,arenotflapped,butareinfactfrequentlyunreleased.Second, [td]intargetflappingcontextsareneutralized(cf.bidder – bitter), but contrast in the coda(bid – bit).Third,theclaimthatweaksyllablesinwordslikecity or sadder are onsetlessisdifficulttosquarewiththegeneralfindingthatrealonsetlesssyllablesmaybepronouncedwithapreceding[ʔ].Moreover,aswasobservedabove,thereisthegeneral drawback that the resyllabification option will have to be constrained by some independent principle, since it would otherwise freely allow changes of onsets into codas and vice versa. The third solution to the definition of the context of flapping is one which ties the data to the domain of the foot. This is a possible account, but with atwist.Ifflappingisthedefaultimplementationof[td]ratherthanthespecialcase,thecomplementofKahn’sgeneralizationcanbegiven.Atthebeginningofthewordandthefoot,/t/isrealizedas[tʰ],asintomorrow, retain, and in the syllable coda as an unreleased[t˺],asright, atlas.Elsewhere,itisflapped(Harris2004).

Q88 In some dialects of Spanish (Harris 1983), [r] and [l] are realized as [j] in certain positions, as illustrated in (1).

(1) revolver [rεvɔjvεj] ‘revolver’carta [kajta] ‘card’papel [papεj] ‘paper’calor [kalɔj] ‘heat’

Givetworule-basedaccounts,onewithandonewithoutreferencetothesyl­lable, and argue that the first is superior.

Between the segment and the syllable 145

Q89 British English (RP) /r/ can be realized either as a flap [ɾ] or an approximant [ɹ] in the expressions in columns I and III, but in those in column II only an approximant realization is available (Rubach 1996).

I II IIIcourage courageous for examplevery reduce for instancebaron red the other endlaurel bright

1 Canthesedatabedescribedonanassumptionofambisyllabicity?2 Canthesedatabedescribedintermsoffootstructure?

Q90 The structure of the syllable rhyme of Brazilian Portuguese (BP) is V(X), where X can be C or V. BP has the nasals [m n ɲ]. Prepalatal [ɲ] distinguishes itself from the other nasals in a number of respects. First, while [m n] can appear word-initially, as shown in (1a,b), [ɲ] can-not, as shown in (1c), where V stands for any vowel (Wetzels 1997b).

(1) a ˈmɔvel ‘mobile’b ˈnɔrti ‘north’c *ɲV...

Second, although diphthongs are freely formed out of consecutive vowelsbeforeothernasals(2a,b),theydonotoccurbefore[ɲ].Instead,thetwovow­elsaredividedovertwosyllables(2c,d).

(2) a ˈrei.nu ‘kingdom’b an.ˈdai.mu ‘stage’c fu.ˈiɲa *ˈfuiɲa ‘weasel’d raˈiɲa *ˈraiɲa ‘queen’

Third,while[mn]tolerateaconsonantimmediatelybeforethem(3a,b,c,d),[ɲ]doesnot(3e).

(3) a aˈdornu ‘ornament’b ˈarma ‘weapon’c ˈalma ‘soul’d vulneˈravel ‘vulnerable’e *VrɲV,*VlɲV

Between the segment and the syllable146

Thethreedifferencesbetween[ɲ]ontheonehandand[mn]ontheothercanbeexplainedif[ɲ]isincorporateddifferentlyinsyllablestructurefrom[mn].

1 Whatisthisstructuraldifference?

BPhasaruleofvowelnasalization which causes a stressed vowel to be nasalizedbeforeanasal,asshownin(4a,b).Unstressedvowelsremainoralbeforenasalconsonants,asshownin(4c,d).However,before[ɲ],vowelsarealwaysnasalized,asshownin(4e,f).

(4) a ˈsinu ‘bell’b ˈkãma ‘bed’c boˈnεka ‘doll’d kumaˈri ‘chilli’e aˈrãɲa ‘spider’f diˈɲeiru ‘money’

Itispossibletoaccountforthenasalizationofstressedvowelsbefore[mn]andthelackofnasalizationofunstressedvowelsbefore[mn]byassumingthatinstressedsyllablesconsonantslike[mn]areincorporateddifferentlyinsyllable structure than in unstressed syllables.

2 Whatisthatassumption?3 What is under that assumption the correct generalization for vowel

nasalization?4 Howdoesyoursolutionaccountforthefactthatbefore[ɲ]bothstressed

andunstressedvowelsarenasalized?

9.7 CONCLUSION

In this chapter we have seen that segmental duration can be accounted for by assuming that there is an independent tier of skeletal slots which constitutes an intermediate level of structure between the segments and the syllable. In addition, theelementson this tier serve todefine themorphemes forverbaspect inAra­bic.By allowingC-slots tobeunsyllabified, it also appearedpossible to accountforFrenchliaison,whileanemptyC-slotcouldbeusedtoaccountforh-aspiré in French.However,adistinctionmustbemadebetweensegmentsthatdeterminethestatus of the syllable with respect to foot assignment and cause compensatory dura­tional effects, on the one hand, and those that do not, on the other. The mora serves to represent the special prosodic status of these segments. The moraic theory pre­dicts that compensatory effects and sensitivity to stress rules go hand in hand in any one language. It is usually assumed that the mora tier should replace, rather than

Between the segment and the syllable 147

supplement,theCV-tier.WehavealsoseenthatthesyllableplaysanactiveroleintheSDsofphonologicalrules,whichoftenappearsensitivetowhetheraconsonantisinthecodaorintheonset.Additionally,ithasbeenshownthatambisyllabicityis a form of representation required for the description of a number of processes.

Ofcourse,theadvantagesofallowingenrichedrepresentationslikeambisyllabic­ity need to be weighed up against a sparser inventory of representations, which will place more of the descriptive burden on the availability of different levels of repre­sentation,asinSelkirk’s(1982)accountofAmericanEnglishflapping. However,in the next chapter, we will see how multiple linking for a third kind of segments besidesvowelsandconsonants,tones,hasbroughtmanybenefitsinitswake.Manyphenomena discussed there in fact lay at the origin of the idea of autosegmental phonology and multiple linking.

NOTE

1 Theexpectedform*[ħaqaq-a]‘itistrue’isruledoutbecauseofageneralprocessdelet­ing vowels in this context.

Tones

10.1 INTRODUCTION

In addition to consonants and vowels, phonological representations contain tones (section2.2.5).Inmorethanhalfofthelanguagesintheworld,tonesareusedalongwithvowelsandconsonantstokeepmorphemesapart,whethermajor-classwordsor function words, in the same way that languages use vowels and consonants for thispurpose.StandardMandarin,forinstance,hasfourlexicallycontrastivetonepatterns,asshownbytheminimalquadrupletin(1a,b,c,d).In(1e),afifthwordisgiven,whichhasthesametonepatternas(1d)buthasadifferentconsonant,mini­mallycontrastingwith(1d).

(1) (1)

maa‘scold’

baa‘eight’

baa‘pull up (grass)’

baa‘grasp’

baa‘father’

ba edc

Equally,tonesmayhavemorphologicalroles,aswhentheyexpressdefiniteness,number or gender on nouns; person, aspect or tense on verbs; or morphological class.WhereasStandardMandarinonlyhaslexical tones, many other languages additionally, or in a case like Somali even exclusively, have such grammatical tones(Hyman2016).Bythesideoflexicaltones,manyOtomangueanlanguageshave more or less regular correspondences between tones on the verb and their morphological class. For instance, Zenzontepec Chatino has mid tone on thepotentialandprogressiveformsof/-soʔ/‘pick’,aboundmorphemerepresentativeofverbclassA2,buthightoneinthecompletive,/-sóʔ/.1Andwhenusedinverbs,the eight tones of Iau indicate aspectual meanings, like ‘totality of action­durative’, asin[tai34]‘bepullingoff ’and‘resultative-durative’,asin[tai3]‘havebeenpulledoff ’,differentlyfromtheirlexicalfunctionsinnouns,like[te34]‘flooring’and[te3]‘mosquito’(Bateman1990).Languagesthatuseatleastsometoneseitherlexicallyor grammatically are known as tone languages.TheyaremainlyfoundinAsia,particularly in SouthEastAsia, aswell as inAfrica and theAmericas.A thirdfunction of tones is to add a discourse meaning to the expression or to indicate prosodic phrasing.MostAustronesian and Indo-European languages onlyhavetoneswiththesefunctions.InEnglish,forinstance,theexpressionJonathan may soundlikeananswertoaquestionin(2a),likeaquestionin(2b),andlikeeitherareminderoraquestionin(2c).

(2) (2)

JonathanJonathan Jonathanba c

10

Tones 149

Aphrasing function of tone occurs in an English sentence likeOnce we’re in China, we can practise our Chinese, where the last syllable of China is likely to have a hightoneindicatingtheboundary.Whentonesfunctiontosignaldiscoursalmean­ings or phrasing, they are intonational tones. The term ‘intonation language’ is not common, because it would incorrectly imply that tone languages have no intona­tionaltones.Manytonelanguageshaveintonationalboundarytones,forinstance.

Whiletherehavebeenanumberofproposalsforthefeaturalanalysisoftones,like[±highpitch]tocharacterizehighandlowtones(HvsL)(seeBao1999;Hyman2010),inthischapterwerefertotonesasintegralsegments(HML),whereMismid tone. Importantly, tones are autosegments. They are arranged on a separate tier fromthesegmentaltier(ortiers,ifvowelsandconsonantsappearonseparatetiers,as inArabicverbalmorphology, section9.3).Goldsmith(1976)showedthat thetonalpatternsofthewordsofmanyAfricanlanguagescanbeinsightfullydescribedautosegmentally.Werepeathispointthatalinearrepresentationisinadequateforexplainingthosedata(section10.2)andreproducesomeofhisargumentsinsec­tions10.3and10.4.First,whenlanguageshaveasmallsetoflexicaltoneconfigura­tions, or word melodies, distributed over words with different numbers of syllables, it makes sense to keep the tones and the segments as separate strings in the lexicon and associate the tones with structural positions according to a general algorithm. Asecondargumentforhavingaseparatetonetierisbasedonthefactthatvowelsmaybedeletedbuttheirtonepreserved(tone stability),whileathirdargumentisbasedontheexistenceofmorphemesthatconsistonlyoftone(tonal morphemes).Section10.5presentstheObligatoryContourPrinciple(OCP),whilesection10.6introducesthenotionofaccentasaplaceholderfortonalconfigurations.Sec­tion 10.7 deals with the phonetic implementation of tones, distinguishing theirscaling from their alignment. The privative representation of tone is briefly dis­cussedinsection10.8,andsection10.9discussesconsonantal f0 perturbations and some of their effects on tone distributions.

Q91 What is wrong with the following statements?

1 Inatonelanguage,achangeinpitchcanchangethemeaning.2 Inatonelanguage,achangeinpitchcanchangethemeaningofaword.3 Germanisafrontroundedvowellanguage.

10.2 THE INADEQUACY OF A LINEAR MODEL

Until the1970s, itwasassumedthatsegments formasingle layerofstructure,asingle string of self­contained feature matrices, whereby tones were incorporated as pitch features in vowel matrices. This linear conception of segmental structure makes it impossible to represent aspects of pronunciation that characterize more

Tones150

thanonesegmentasasinglefeature.Languagesinfactfrequentlytreatparticularaspects of pronunciation, notably tone, nasality and tongue body features, as if they belongedtolongersequencesofsegments.Forinstance,ifalanguageweretohavewords whose syllables were either all high­toned or all low­toned, then it would makesensetosaythateachwordofthelanguageeitherhadthefeature[+hightone](‘H’)or[–hightone](‘L’).IntheSPE representation of feature matrices, we would not be able to express this generalization: every vowel would have to be specified for[±hightone].Ifwedidthat,thequestionariseswhythewordsofthislanguagehaveeitherconsistentstringsof[+hightone]specificationsorconsistentstringsof[–hightone]specifications,withoutevermixingtheseinthesameword.Wecouldof course construct a grammar that demands this consistency, through the intro­duction of some constraint that requires that all the segments in a word have the same specification for featureX.However, the same effectwould be obtained iffeatureXwaspartof theunderlying representationof thewordandsomealgo­rithm distributed it over all the relevant segments. This would allow us to express directly that words are either high­toned or low­toned. That is, we could represent the tonal features, making up the tones, and the other features, making up the other segments, as two separate strings and then allow different numbers of segments in the two strings, such that a word might consist of six vowels, say, but of only one tone.Weshowthedifferenceforahypotheticalhigh-tonedword[tata]in(3):(3a)gives the linear situation adopted in SPE (where[a]standsfortherestofthedistinc­tivefeaturesinthatmatrix),and(3b),orequivalently(3c),givestherepresentationwithtwoparallelstrings.Equally,thetonalstringmighthavemoretonesthanthesegmental string has vowels, opening up the possibility of more than one tone char­acterizing one vowel.

(3)

Goldsmith(1976)characterizedtheSPE position as a theory that adheres to the Absolute Slicing Hypothesis: the phonological representation of a word is given by making a number of clean cuts along its time axis, each slice so produced being a segment. The representation which he argued for instead is one in which the cuts can be made for subsets of features, such that slices made for the features specifying vowels and consonants leave the features of tones unaffected, and vice versa.

10.3 WORD MELODIES

If, regardless of the number of syllables, the words of a language were either entirely high­toned or entirely low­toned, a situation briefly considered in the previous sec­tion, it would obviously be economical to specify each word once either for low tone or for high tone. Put differently, we would wish to say that the language had two word melodies,HandL,andthateachwordhadeithertheoneortheother.

Tones 151

Etung presents a somewhat more complex case (Edmondson & Bendor-Samuel 1966). A syllable in an Etung word has high, low, falling or rising tone. Importantly, however, there are certain constraints on the distribution of the tone types: not all of them can appear on all syllables. The IPA accent symbols used in the transcription of tone are given in the first column of (4), and the tonal representations in the third.

(4) á high tone H à low tone L â falling tone HL a risingtone LH aVv falling-risingtone HLH avV rising-fallingtone LHL

In (5a,b,c), we list actual words of one syllable, two syllables and three syllables, respectively. Notice that level tones (i.e. a high tone or a low tone) may occur on all syllables, but that contour tones (falling tone or rising tone) only appear on mono-syllabic words and on the final syllable of disyllabic words. Moreover, if a syllable has the same tone as the preceding syllable, all following syllables have that tone. Thus, the words in (5a,b,c) are all fine, while those in (5d,e,f) are all bad. In (5d), we see that contour tones must be final, in (5e) we see that contour tones cannot appear in words of more than two syllables, while in (5f) we see that in words of three syl-lables, the first two cannot have the same tone if the third has a different tone.

(5) a monosyllabicwords kpá,kpè,nâ,no b disyllabicwords óbá,èkát,òbô,ódà,ábo c trisyllabicwords ékúé,èjùrì,ókpùgà,édìmbá d disyllabic *âbó,*abó e trisyllabic *ábòmbâ,*ábômbà,*âbòmbà f trisyllabic *ábómbà,*àbòmbá

Leben (1973) argued that such distributional facts can be captured by assuming that tones are not chosen per syllable, but per word, and that, moreover, these tone patterns are independent of the vowels and consonants. These Etung tone patterns are given in (6). Importantly, the tonal tier and the segmental tier are unassociated in the lexicon: the point is that all and only the well-formed lexical patterns of (5) can be derived by an association algorithm, so that we account for the ungram-maticality of the forms in (5d,e,f ). To achieve this, Goldsmith (1976) provided the Association Convention (7). The phonological constituent with which a tone associates, the tone-bearing unit (TBU), is either the mora or the syllable. In Etung, the TBU is the syllable. For typographical convenience, a tone’s association line is with the vowel of the syllable concerned.

(6) AwordhasoneofthesixpatternsL,H,LH,HL,LHL,HLH.(7) Association convention

a AssociatetonesandTBUslefttoright,onetoone. b Associateleft-overTBUswiththelasttone. c Associateleft-overtoneswiththelastTBU.

Tones152

In(8),anumberofexampleshavebeenworkedout.In(8a),thenumberoftonesandthenumberofTBUsisequal,andone-to-oneassociationisallweneedtodo(clause(7a)).In(8b,c),therearefewertonesthantherearesyllables.Accordingtoclause(7b),wemustnowassociate the last tonewiththe left-oversyllable(s),anoperation known as spreading (seesection9.3.2).NoticethatthiscausesthesametonetobeassociatedwithmorethanoneTBU.In(8d),therearemoretonesthanthereareTBUs.ThelastTBUreceivestheleft-overtone(clause(7c)),causingthisTBUtobeassociatedwithacontourtone.

(8)

Etunghasnofalling-risingorrising-fallingwordsofthetype*navV.Apparently,thereisalimittothenumberoftonesthatcanassociatewithaTBUinthislan­guage. This No CrowdingConstraint, which appears to be quite common in languagesgenerally,isgivenin(9).

(9) no croWding constraint:

The above description accounts elegantly for the distribution of the four surface tonetypes(high,low,falling,rising)ofEtung.Ifweweretoelevatethosesurfacetone types to features inside a segmental feature matrix, a description would result thatwould consist of a series of arbitrary facts. In (10), this alternative ismadeexplicit. This cumbersome description should be compared to the simple autoseg­mentalalternativeconsistingof(6),(7)and(9).

(10) a AsyllablehasoneofthefourtonesHigh,Low,Fall,Rise. b FallandRiseneveroccuronanon-finalsyllable. c WordsofmorethantwosyllablesneverhaveaFalloraRise. d Wordsofmorethantwosyllablesneverhavethesametoneonthefirstn syllablesifthenth

+1syllablehasadifferenttone,wheren >1.

TheTBUmaybedefineddifferentlyfordifferentlanguages.InsomelanguagesthesyllableistheTBU,sothatlongvowelscountasoneTBU.Aswewillseeinsomeof

Tones 153

thequestions,otherlanguageshavethesonorantmoraastheTBU,inwhichcaselongvowelsaretwoTBUs,whilesonorantconsonantsinthesyllablerhymearealsoTBUs.

Q92 The majority of Mende words have the tone patterns illustrated below:

one syllableː kɔ ‘war’, kpà ‘debt’, mbû ‘owl’, mba ‘rice’, mbavV ‘companion’two syllablesː pɛlɛ ‘house’,njàhâ ‘woman’,ŋgílà ‘dog’, fàndé ‘cotton’,bɛlɛ

‘trousers’three syllablesː kpàkàlì ‘tripod chair’, félàmà ‘junction’, ndàvúlá ‘sling’,

nìkílì‘groundnut’,háwáná‘waistline’

1 WhataretheunderlyingtonepatternsofMende?2 DoesMende,likeEtung,obeytheNoCrowdingConstraint?3 Givethetonalrepresentationsofthewordsfor‘woman’and‘tripodchair’.

10.3.1 Language-specific associations

The AssociationConvention (7) is not an obligatory procedure in tone lan­guages.Forone thing, theremaybe language-particular ruleswhichoverride it.InKikuyu,theassociationproceedsfromthesecondTBUintheword,ratherthanthefirst.Thewordin(12),whichmeans‘wayofreleasingoneselfquickly’,ismadeupofthemorphemesin(11).Theverbrootextension[aŋg] ‘quickly’ istoneless.The association of the tones contributed by the various morphemes proceeds as predictedby(7),withthelastHspreadingtotheleft-overTBUs,exceptthatinsteadof associatingwith thefirstTBU, thefirst tone associateswith the secondTBU.Subsequently,thefirstemptyTBUisassociatedwiththefirsttone,asshownin(12)(ClementsandFord1979).

(11) mo e rɛk+aŋg eriɛ L H L H class prefix reflexive allow+quickly noun suffix

(12) mo e r ka ge ri

L H L

(12)

H

Q93 The Kikuyu words for ‘firewood’ and ‘tree’ are [ròko] and [mòto].

1 Whataretheunderlyingtonepatternsofthesewords?2 InTharaka,arelatedBantulanguagespokeninKenya,thesewordshave

thesameunderlyingtonepatternsbutarerealizedas[ròkó]and[mòtó].HowwouldyoudescribethisdifferencebetweenKikuyuandTharaka?

Tones154

Thus, languages may have specific constraints on associations of particular tones. Forinstance,HymanandNgunga(1994)showthattheTBUwithwhichatoneasso­ciatesinCiyaoverbalstemsmaydependonpropertieslike‘positive’(Honthelastmora),‘negative’(Honthesecondmora)or‘infinitive’(Honthefirstmora,withspreading).EveninlanguageslikeEtung,whosewordtonalpatternsareapparentlygoverned by the AssociationConvention(7),theremaybeexceptionalpatternsthatcannotbesoaccountedfor.Likewise,thereisnouniversalrequirementthattonesshouldalwaysspreadtoemptyTBUs.Pulleyblank(1986)showsthatinTiv,non-initialTBUsmayremainunspecifiedfortoneuntilapointinthederivationwhentheyreceiveadefaultL.Neitherneeditbethecasethatleft-overtonesalwaysformcontoursonthelastTBU.Languagesmaynottoleratecontouringatall,obey­inganevenmorerestrictiveconstraintthan(9).

3 TheTharakawordfor‘wayofreleasingoneselfquickly’ismadeupofthesamemorphemesandthesametonesasin(11).HowwillitberealizedinTharaka?Showtherepresentation.

Q94 Zulu (Laughren 1984) obeys a more restrictive constraint than the No Crowding Constraint (9), since no contour tones may appear on short vowels at all. Zulu words behave either like Kikuyu words or like Etung words. That is, every word is lexically specified as to whether association starts with the first or the second TBU. In Natal Coast Zulu, the word for ‘chiefs’ consists of two morphemes, /ama/ and /kosi/, both of which have underlying HL. The first morpheme is specified for association to start with the first TBU, the second for association to start with the second TBU. After this initial associa-tion, the rest of the derivation is taken care of by the Association Convention in (7).

1 Howwillthewordfor‘chiefs’berealized?2 Whyistherealizationnot[ámàkòsî]?

While theAssociationConvention in (7)maywell represent anunmarkedprocedure, evidently languages introduce language­particular instructions or have representations that deviate from its predictions. The only uncontroversially univer­sal aspect of autosegmental representations is the No CrossingConstraint,given in(13).Itisaninalienablefeatureoftheautosegmentalrepresentation:ifitdidnothold, the order of the autosegments on different tiers could change relative to the ordersonothertiers.Theconstraintin(13)istheretopreventsuchorderchanges.

(13) no crossing constraint: Associationlinesdonotcross.

Tones 155

In this section, the autosegmental representation of tones was argued for on the basisofdistributionalrestrictionsthattheyaresubjecttoinvariouslanguages.

Q95 West Greenlandic, a language that only has intonational boundary tones, has a phrase-final melody HL and an utterance-final H, which come together as HLH at the utterance end. The timing of these boundary tones is illustrated in the following data (Rischel 1974; Nagano-Madsen 1993; preceding unmarked syllables have default tone, realized as middish pitch).

a akívàrá ‘I answered him’b atâːsíq ‘one’c akívaː ‘he answered him’d úvàŋá ‘I’e uváŋàlú ‘and I’f uvaŋátːaː ‘I, too’

1 WhatistheTBUinWestGreenlandic?2 Showthestructureoftheexpressionsina,b,c.Showlongvowelswith

doublesymbols(i.e.[aː]as[aa]).3 Whatisthedirectionofassociationofthetones?

Yes-no questions end in the same melody but are distinguished from declaratives by the addition of a mora to the final syllable.

Declarative Interrogativeg takuwâːná ‘he saw me’ takuwáːnaː ‘didheseeme?’h takúwìjúk ‘you saw him’ takuwíjuːk ‘didyouseehim?’i tsigúwaː ‘he takes that’ tsiguwaVv ːː ‘didhetakethat?’j apíràí ‘he asked them’ apirâːí ‘didheaskthem?’

4 Givethetonalrepresentationsofthedeclarativeandinterrogativeformsof ‘he asked them’.

5 Thepenultimatevoweloftheinterrogativeformin(g)isfullyhigh-toned.Whatextraassumptionmustbemadetoaccountforthisfact?

6 Whatisthedirectionofassociationoftheelementsinadiphthong?

10.4 TONAL STABILITY AND TONAL MORPHEMES

Sometimes,avowelisdeleted,butthetoneithadshowsuponanadjacentvowel.In(14),someexamplesaregivenfromEtsako,inwhichanonhighvowelisdeletedbeforeanothervowel(Elimelech1976).Whenthefinalvowelofthestemisdeleted,

Tones156

its tone shows up on the next vowel, where it creates a contour. If the tone of the deleted vowel is the same as the tone on the following vowel, however, nothing hap­pens.InEtsako,reduplicationofthenounhasthemeaning‘every’.

(14) ìkpà ‘cup’ ìkpìkpà ‘everycup’ ówà ‘house’ ówowà ‘everyhouse’ ɔɣɛdɛ ‘banana’ ɔɣɛdɔɣɛdɛ ‘everybanana’

Again,itisofcoursepossible to describe these facts in a theory in which all fea­turesarecontainedinsegment-sizematrices.Wecouldassumeafeatureanalysisofthelow,high,fallingandrisingtonesasin(15)andwrite(16)toaccountforthetonechanges.Inprose,(16)saysthataword-finalvowelisdeletedbeforeaword-initial vowel, and if it has a different tone from the word­initial vowel of the next word, the tone on the word­initial vowel becomes rising if it was high and falling ifitwaslow.Whileitcorrectlydescribesthesituation,itpresentsthefactsasarbi­trary:wecouldjustaseasilywritearulewith[–contour]ratherthan[+contour],but that rule would not describe any actual language data.

(15) Low` High´ Fallˆ RiseHightone – + – +Contour – – + +

(16) +

−

syl

hightone

contour

α #

+

−−

syl

hightone

contour

α → Ø 2 +contour

1 2 3 3

The autosegmental representation allows us to express the fact that it is only the vowel,notitstone,thatisdeleted.Byreassociatingthestrandedtonetothenextvowel,weaccuratelyexpressthetonaladjustments.In(18b)ariseisproduced,in(18c)afall,whilein(18a)notonalchangeispredicted.Associatingthesamefeaturemore than once to a structural element is meaningless, and the two identical tones are automatically reduced to one by a convention known as the TwinSisterCon­vention(see(17);ClementsandKeyser1983).

(17) tWin sister convention

(18)

Tones 157

Q96 Here are some examples of a word game in Bakwiri (Hombert 1986). Bakwiri has prenasalized stops: [mb], [nd] and [ŋg].

Bakwiri Word gamea mɔkɔ kɔmɔ ‘plantain’b lówá wáló ‘excrement’c kóndì ndíkò ‘rice’d iŋɔ ŋɔʔì ‘throat’e ézèè zéʔèè ‘it is not’f lùùŋgá ŋgààlú ‘stomach’g lìòβá βààljó ‘door’

1 Whatisthegeneralruleforformingword-gamewords?Applyyourruletothehypotheticalform[ndàkóó].

2 Bakwiri syllable structure is (C)V(V),whereVVmaybefilled by onevowel,orbytwodifferentvowels.Giverelevantexamples.

3 Whyisn’tthegameformfor‘door’*[βààlíó]?4 Whydoyouthinkthereareglottalstopsinthewordsfor‘itisnot’and‘throat’?

Q97 Kisanga has the syllable structure (C)V(V). In a children’s language game called Kinshingelo, sentences (a) and (b) are pronounced as (c) and (d), respectively (Coupez 1969).

a óbémúkwèːtù twâjáːkúmúkólá‘you, my friend, come with me to the river’

b bâːnábákàɟíbâːmúkóːŋgòbaːtèmwâːkúdímá‘theCongolesewomenlovetocultivatetheland’

c béómútùːkwèjâtwáːkúmúlákód bâːnábáɟìkábâːmúŋgóːkòbaːmwàtêːkúmádí

ArguethatthislanguagegameprovidesevidencefortheexistenceinKisangaof the word, the syllable, the mora and the tonal tier.

Tonesmaybesystematicallydistributedovermorphologicalcategoriesalongwithmorphemes that are also encoded in vowels and consonants. The term ‘grammati­caltone’,oneofthethreefunctionalcategoriesdistinguishedinsection10.1,morestrictly applies to tonal morphemes(Hyman2016).Likeintonationalmorphemes,tonal morphemes lack vowels and consonants in their phonological representation but, unlike intonational morphemes, have meanings that fit into the morphological and syntactic paradigms of the language, instead of expressing discoursal meanings

Tones158

ortheprosodicphrasingstructure.InKalabari,whichhasalargenumberoftonemelodies expressing morphological or syntactic categories, verbs can be made intransitivebyimposingaLHmelodyonthem(HarryandHyman2014).Thisisshownin(19),where!indicatesthattheHinthatsyllableisdownstepped(seesec­tion10.7.1).

(19) Transitive Intransitive a ɗìmà ‘change’ ɗìmá ‘bechanged’ b sá!kí ‘begin’ sàkí ‘begin’ c kíkímà ‘cover’ kìkìmá ‘becovered’ d pákìrí ‘answer’ pàkìrí ‘beanswered’ e gbóló!má ‘mixup’ gbòlòmá ‘bemixedup’ f kán ‘demolish’ kàán ‘bedemolished’

Q98

1 WhatistheTBUinKalabari?2 WhatisthedirectionofassociationoftheLHmelody?

10.5 THE ObligatOry COntOur PrinCiPle

The success of the autosegmental account of the word tone patterns in languages likeEtungdependsinpartonthedecisionnottorepresentthepitchpatternofeachsyllableasaseparatetoneintherepresentation.Crucially,adjacentsyllableswiththe same tone need not each be specified separately for tone, since the same tone canassociatewithmoresyllablesthroughspreading.Considerationslikethesehaveledtotheassumptionthatidenticaltonescannotbeadjacentinthesameword.Insection9.3.2,thisrestrictiononrepresentationswasreferredtoastheObligatoryContourPrinciple (OCP), formulatedbyGoldsmith (1976)with reference toLeben(1973)as(20). It forbids theoccurrenceof identicaladjacent tones inthesame morpheme.

(20) obligatory contour principle: *V V| |Ti Ti

AnotherargumentfortheOCPisthatruleschangingordeletingtonesusuallydelete what on the surface appear to be sequences of identical tones. If we assume that the string of identical surface tones is in reality a single, multiply associated tone,areadyexplanationisfoundforthisphenomenon.Example(21),fromOdden(1986),comesfromShona.The‘associativeprefix’[né]‘with’lowersanyconsecutive

Tones 159

H-tonestolow.Clearly,ifthesewordshaveonlyoneH,rule(22)canchangethatHintoL,thusaccountingforthesedata,asshownin(23).

(21) mbwá ‘dog’ hóvé ‘fish’ mbúndúdzí ‘armyworms’

(22) H → L / H __|

ne

(22)

(23) ne-mbundudzi → ne-mbundudzi|

H H H L

(23)|

In contrast, if each of the three syllables of ‘army worms’ were to have its own H,amorecomplicatedanalysisresults.WemustapplyaruledeletingHafter[né]providedthefollowingtoneisalsoH(whichrulewillapplymorethanoncetothesamewordinacaselike‘armyworms’),thenchangetheremainingH-toneintoLand,third,spreadthatLbacktotheprecedingtonelesssyllables.Clearly,havingasingleH-toneintherepresentationmakeslifealotsimpler!

Q99 How does the derivation of ‘every cup’ proceed if, as argued by Leben (1978), Etsako obeys the OCP? Would there still be a need for the operation of the Twin Sister Convention?

InOptimalityTheory,constraintsareviolable,andtheOCPthereforecannotbeauniversallyobeyedconstraint.Foronething,languageslikeEtungfrequentlyhaveexceptionalpatterns.Forinstance,wordswithsurfacemelodiesHHLandLLHdoexist,like[ŋgárè]‘pepper’.SuchwordsmustbeassumedeithertohaveunderlyingmelodieslikeHHL,contra theOCP,ortohaveasingleHtoneprelinked to two TBUsinthelexicon,contra the AssociationConvention. In the latter option, the surfacerepresentationsof‘pepper’wouldlooklike(24)(Odden1986).Representa­tionsthatviolatetheOCPexistinmanylanguages.

(24) (24) are|

H L

10.6 ACCENT

Tones,whetherlexicalorpostlexical,havealocationandphonologicalcontent.ThelocationofthetonesinanEtungwordisdescribedintermsoftheAssociation

Tones160

Convention, while their phonological content is described in terms of a word melody.Anotherdeviceforspecifyinglocationisaccent.Insection1.2.2,wesawthat Japanese words fall into two lexical classes, accented words and unaccented words.Accentedwords have aHL tonemelody on one of their syllables,whileunaccented words have no such melody. There are many generalizations about the presenceorabsenceof the lexicalHLtonemelody,as in thecaseofcompounds(Kubozono1993;Kubozonoetal.1997).Certainlyininformaldescriptions,itwillmake sense to abstract away from the tones and assume some accent marking, so as to be able to refer to ‘accented word’, ‘accented syllable’, ‘accent deletion’, etc. It is not always obvious whether there is a theoretical advantage to an accentual analysis as opposed to a tonal analysis.Whenwordsareunderlyinglymarkedforthepresenceof some specific tone or tones, such marking is effectively equivalent to the pres­enceoflexicaltone.Hyman’s(2001:1368)definitionofatonelanguageissuitablyneutral between the two analyses: ‘a language with tone is one in which an indica­tionofpitchentersintothelexicalrealisationofatleastsomemorphemes’.Here,‘anindication of pitch’ could refer either to an accent or to a tone.

The case for an accentual analysis is strengthened by the existence of more than one tone or tone melody in a language in combination with generalizations about tonal location.Barasanahas two tonemelodies,HandHL.Theseareassociatedleft-to-right tomoras,with spreading,much as inZulu (Q94). Idiosyncratically,wordsbegintheassociationateitherthefirstorsecondmora(Gomez-ImbertandKenstowicz2000).Atonelessword-initialmorawillbepronouncedwithlowpitch,equivalenttoL.Atonalanalysiswouldspecifyfourtonalmelodies,H,HL,LHandLHL,andassumeatonalassociationalgorithmasforEtung.Representationsinthisanalysisareshownin(25).

(25) a jai b hee c wai d cai

H L H L HLH L‘jaguar’ ‘ancestral’ ‘fish’ ‘catfish’

(25)

An argument for an accentual analysis, effectively equivalent to that given byGomez-ImbertandKenstowicz(2000),canbebasedonthecompoundrule,whichdeletestoneonthesecondconstituent,theheadofthecompound.Totake(25a,b,c)as examples, [héè] ‘ancestral’ combineswith [jáí] to forma compound [héè jàì]‘shaman’,and[wàí]formsacompoundwith[kúmà]‘glutinousstew’tocreate[wàíkúmá]‘glutinousfishstew’.Thissuggeststhatthesecondelementinacompoundloses its accent, such that only the tones on the first word are inserted. In an accen­tualanalysis,lexicalrepresentationswillneedaspecificationoftheaccent(firstorsecondmora),aswellasofatonemelody,HorHL.Thewordsin(25)nowlooklike(26),where‘*’afteravowelindicatestheaccentedmora.

(26) a ja*i b he*e c wai* d cai* H HL H HL

Tones 161

Q100

1 Usingthemodelof(25),givethetonalassociationsofBarasana‘shaman’and ‘glutinous fish stew’.

2 FirstandsecondpersonpronounsalwayshaveaH-melody,whilethirdpersonpronounshaveaHLmelody.Whenprefixedtoanoun,theirtonemelody copies onto the noun, whose original tone melody is deleted, as intheexamplesbelow.Usethisruletoargueforanaccentualanalysis.

a mání ‘1pl’ mínì ‘pet’ – mánímíní ‘ourpet’b ínà ‘3pl’ mínì ‘pet’ – ínàmínì ‘theirpet’c jɨɨ ‘1sg’ mínì ‘pet’ – jɨɨmíní ‘mypet’d mání ‘1pl’ wɨhɨbò ‘tray’ – máníwɨhɨbó ‘ourtray’e ínà ‘3pl’ bàbárá ‘friends’ – ínàbàbárà ‘their friends’f jɨɨ ‘1sg’ wɨhɨbò ‘tray’ – jɨɨ wɨhɨbó ‘mytray’

AllEnglishwordsareinprincipleaccentable,asindeedtheymustbe,sinceanutterance contains minimally one accented word, whose main stressed syllable will host an intonational tone or tone melody. The analytical separation of the locationofthetoneanditsidentityistheonlyoptioninacaselikeEnglish.First,all tone assignments are postlexical, which implies that any lexical specifications of tones are ruled out. If we were nevertheless to insist on specifying tones in the lexicon, our efforts would be frustrated by the fact that there are many tone melodies that can in principle be inserted in an accented syllable, depending on theirdiscoursalmeaning.Also,therearemanygeneralizationsthatcanbemadeabout the assignment or deletion of pitch accents. It would be bizarrely uneco­nomical to list the tones that could occur in an accented syllable as being absent fromit,insteadofjustsayingthatthesyllableisnotaccented.Onedeaccentua­tionruleofEnglish is thecompoundrule,whichis just liketheBarasanarule:deletetheaccentonthesecondconstituent(27).Anotherisadeaccentingrulefor post­focal words, whose effect is seen on the lack of accent on pills in(28b),where the focus is don’t need. The reason for a deaccentuation can on occasion be ambiguous.ThetitleofWalterLang’s1954movieThere’s NO Business Like SHOW Business sounds as if the second mention of business is deaccented because of the previous occurrence of that word in the sentence, but it would of course have been deaccented anyway, as it is in There’s NOTHing like SHOW business, because it is the second constituent in a compound.

(27) a BUSiness,MODel – BUSinessmodel b DRAma,QUEEN – DRAmaqueen(28) A:DidyougetyourPILLS? B:Idon’tNEEDthosepillsanymore.

Tones162

10.7 PHONETIC IMPLEMENTATION OF TONE

The description of the pronunciation of tones follows the target­and­interpolation modeloutlinedinsection8.6.2.Atone’spronunciationisthusatargetwitha‘ver­tical’ and a ‘horizontal’ specification. The ‘vertical’ specification is some value on thef0scale,knownasits‘scaling’(Pierrehumbert1980:47).WhilePierrehumbert(1980) concentrated on the scaling of tonal targets, most of the subsequentresearch has been devoted to the ‘horizontal’ dimension, the synchronization of the tonal target with the segmental string. The phonetic implementation of tones, justasofvowelsandconsonants,isbothcontext-sensitiveandlanguage-specific,asexplainedinsection8.6.1.Toillustrate,considerthephoneticimplementationof theMandarin lexical tonepatterns in (1),whichwillhave tonal representa­tionsmuchasin(29).ThebracketedHintheLowwordtonein(1c)onlyshowsupwhenanotherLfollowsandattheendofaphrase(seeDuanmu2000:221).The end of the syllable provides the approximate location of the last tone’s target (Xu2006).A sequenceof aLow tone (L), aHigh tone (H) and aFalling tone(HL), therefore,will bepronounced as shownby the solid line inpanel (a) ofFigure10.1,wherethebulletsrepresentthetargetsofthetones.ThismeansthattheHigh tonecomesoutphonetically as a risingpitchmovement through thesecondsyllable.However,thispitchcontourisdistinctfromthepitchcontourfortheRisingtone(LH)inthesamecontext,whosetwotargetsareindicatedbytheopencircles.TherisingmovementforLH,whichisdefinedbytheinterpolationbetween a low target and a high target in the same syllable, is located late in the secondsyllable,insteadofcoveringtheentiresyllable.Inpanel(b),theaveragedmeasured contours are shown for the second syllable in emphatic speech. In this speechstyle,thehightargetoftherisingtonemayinfactbelocatedjustoverinthe next syllable. This ‘peak delay’ is a well­known feature of tonal realizations in languages more generally.

L

1st 2nd 3rd

H/LH HL(a)

100

(b)

150

200

F0 (H

z)

250

L H/LH HL

Rising toneHigh tone

Figure 10.1TheoreticalMandarinChinesepitchcontoursofHightoneandRisingtonebetweenLowtoneandFallingtone(panela),andactualpitchcontoursofHightoneandRisingtoneafterLowtoneandbeforeFallingtone,averagedover18tokens(panelb;adaptedfromChenandGussenhoven2008).

Tones 163

(29) a baa b baa c baa d baa

H L H L(H) H L‘eight’ ‘pull up’ ‘grasp’ ‘father’

(29)

Thef0tracksinFigure10.1makeitclearthatwecannotthinkofthephoneticrealization of tones as sequences of high and low pitches. Just as sequences of vow­els will be pronounced with transitions between their acoustic targets, so will the targets of tones be separated by transitional phases, the interpolated stretches of the f0contour.

10.7.1 The vertical dimension: scaling

Factors affecting the scaling of tonal targets which are likely to play a role inmany languagesareupstep,downstep,H-raising,declination,final loweringandundershoot.

Upstep

WhenthesecondoftwoadjacentH-tonesispronouncedatahigherpitchthanthe first, it is upstepped.Upstep is rare for lexical tones, butwas reported forAcatlánMixtecbyPikeandWistrand(1974),wheretheimplementationruleisiterative and thusupsteps aH also relative to a precedingupsteppedH.Non-iterative upstep is common in intonation systems. Intonational tones are inde­pendentmorphemes. A distinction ismade between tones that are associatedwith the accented syllables of some words, and tones that are located at the edges of larger domains. The former are word melodies, or pitch accents (Pierrehum­bert1980),whilethelatterareknownasboundary tones. Their meanings have to do with whether you are giving the listener some information or are asking for some and how the listener is supposed to relate the information with what she or healreadyknows(e.g.PierrehumbertandHirschberg1990;Bartels1999),whileboundarytonesinadditionindicateprosodicphrasing.Likewordmelodies,pitchaccentsmayconsistofoneormoretones,butunlikethewordmelodiesofEtungandBarasana, typicallyonlyoneof themassociateswith theaccented syllable.Becausetheassociatingtonemaybeanyofthetonesinamulti-tonepitchaccent,it needs to be marked for its preferential association, which is done with the help of a following ‘*’.2InWesternEuropeanlanguages,boundarytonesoccurattheedgesoftheintonationalphrase,withapercentsignindicatingtheedge,%LandH%in(30).TheexampleshowstheapplicationofupsteptoH%afteraprecedingH*(Pierrehumbert1980).

(30) Anything else for you?

%L H* H%

(30)

Tones164

Downstep

DownstepamountstoaloweringofaH-tonerelativetoaprecedingH-tone,insomespecificcontext.Therearebroadlytwophonologicalcontextsinwhichitoccurs.OnerequiresaLtooccurbetweenthetriggeringHandtheHthatundergoesdownstep.Theotherrequiresthetriggertobeadjacent–onthetonaltier,i.e.disregardingthesegmentaltier!–totheHundergoingthedownstep(Odden1986).AnexampleofthefirstoccursinKuki-Thaadow,whichhasthreewordmelodies,HL,HandL,andusesthesyllableastheTBU(Hyman2007).HLsurfacesonlyasafallinphrase-finalposition.Whenoccurringbeforeanothertone,alevelhightoneisrealized,whichisaccountedforbydelinkingL.AtonethathasnoassociationwithaTBUissaidto be floating.In(31a),twowordswithHLarefollowedbywordswithL,creatingtwo valleys, whereby the second, third and fourth peaks are downstepped relative to theprecedingone.Observethatin(31a),thefloatingL-tonemaybeassumedtobedeleted.Itspresencehasnoeffectontherealizationofanyothertone.Bycontrast,in(31b),thepresenceofafloatingLisrevealedbythedownstepittriggersonthenextH.ThesecondandthirdwordshaveadownsteppedH,withoutaprecedingvalley,andthepresenceofthefloatingL-tonesisrevealedonlythroughtheloweredpitchofthenextH.TheeffectofthefloatingLisconfirmedbytheformin(31c).Asinmanylanguages,aHspreadsrightinKuki-Thaadow,inthiscasebeforeL.H-spreadingappliedin(31c)soastodisassociatethelexicalL,asecondsourceofafloatingL.Aspredicted,thefollowingHisdownstepped.

(31) a na hu di ka dey ee b c uy tsom giet zoo lien giet

L HL %L H ‘I want you to come’ ‘eight short dogs’ ‘eight big monkeys’

(31)

L HLHL HLHL L HL L HL

Iterativedownstep,asin(31a,b),isinprincipleunbounded.AclassicstudybyLiber­manandPierrehumbert(1984)showshowtheprogressivelydownsteppedH-targetscanbemodelledwiththehelpofafractionoftheportionofthef0aboveareferenceline.Withareferencelineof100Hz,aprecedingtargetof150Hzandadownstepfactorof0.7,thiswillplacethenextH-targetat100+(150–100)×0.7,or135Hz.

DownstepofHafterHoccurs inCilungu (Bickmore2007).Toneless [ful] ‘towash’in(32a)isprecededbytheclassprefix[ku]andthesubject[tú]‘1pl’, the only tonedmorphemein(32a).ItsHspreadstothetonelesssyllablestilltheendoftheverb,leaving[a]toneless,amorphemeknownastheFinalVowel,whichisgenerallypresentonverbalforms.In(32b),aH-tonedverb[sopolol]‘tountie’isusedinthesamecontext.NowtheHof[tú]spreadsuntiltheHoftheverb,whichitselfspreadstillitgetstotheFinalVowel.

(32) a tu ku ful-a b tu ku sopolol-a

H L H H L‘we are washing’ ‘we are untying’

(32)

Tones 165

H-raising

H-raisingiscommoninAfricanlanguagesandhasnotablybeeninvestigatedforYoruba,whichcontrastsH,MandL inthesurfacerepresentation.BeforeL,butnotbeforeM,Hisraised.Theeffectcanbesostrongastocancelouttheeffectofdownstep,meaningthatthepitchofthesecondH-toneinaHLHLsequencemaybarelybelowerthanthefirst(ConnellandLadd1990;LaniranandClements2003).H-raisingisalsosometimesreferredtoas‘upstep’.

Declination

Declination,shownin(33a),appliesoverlargerconstituentsliketheintonationalphrase and amounts to a gradual lowering of tone targets across the domain, an effect which is weaker than that of downstep, if that exists in the language. Thus, eventhoughasentencemaycontainexclusivelyH-tonesorexclusivelyL-tones,adescending trend is typically observed. In questions, declination may be suspended (Grønnum1992).

Final lowering

The last 400msor soof adeclarativeutterance are loweredover andabove theeffects of declination and downstep. The extra lowering of plansin(33a)mayserveas a finality cue, indicating an opportunity for a turn switch to another speaker. In (33b),adownsteppingcontourisshowninwhichthetargetsofthefinalH* and L%liebelowtheratesofdescentestablishedfortheprecedinghighandlowtar­getsinthesentence(LibermanandPierrehumbert1984).Withoutdownstep,theH*­targets may still show a descending pattern due to declination, an effect that is muchsmallerandpossiblylesssystematic,asillustratedin(33a).

(33) a It’s never too late to change your plans

%L H* L%

b It’s never too late to change your plans

(33)

H*LH*L H*

%L H* L%H*LH*L H*

FinalloweringmayresembledownstepwhenitaffectsafinalH-toneinDinka.ThelanguagehaslexicalL,HandHLandemploysthesyllableastheTBU,asin[tìːk]‘woman’,[báɲ]‘chief ’,[tûŋ]‘horns’(Andersen1987).DinkadownstepsHafteraLrelativetoanearlierH.In(34),theword[lɔːːk](Dinkahasathree-wayvowelquantity contrast) ispronouncedwithdownsteppedHbecauseof thepreceding

Tones166

HL.However,thephrase-finalsyllablealsohasaloweredtonewhich,accordingtoAndersen(1987),islikethatofadownsteppedH.Sincethesyllableisphrase-final,afinalL%maybeheldresponsibleforthelowering.Thisboundarytonemaybeassumedtoassociatewiththelastsyllable,sothat,together,HandL%arerealizedasaloweredH,ratherthanasfallingpitch.

(34) ca m t l k n n t n

PERF;2SG child wash;PARTICIPLE when here‘When did you wash the child here?’

(34)

H L%HH L H

Undershoot

Whentimeisatapremium,aswillhappenwhenanumberofdifferenttargetsneedto be reached in a short time, the more extreme targets may be undershot. This will happen for the end point of the raising and lowering of the tongue in a sequence like /aia/,whichmaybepronounced[aea]or[aɛa]asaresult.Similarly,aHLHsequencemaycomeoutas[HMH].AnextremecaseofthisoccursinZeelandDutch,whereH*LH%onafinalsyllableinanintonationalphrasemayberealizedasasequenceoftworiseswithabrieflevelstretchbetweenthem(Hanssen2017).

10.7.2 The horizontal dimension: alignment

AssociatedtonesarepronouncedclosetotheirTBU.ThereisageneraltendencyforH-tonestobedelayed,causingtheirtargettooccurlateinthesyllableorearlyinthenextsyllable.DelayedH-targetsmaycauseafollowingLto‘falloff ’thesegmen­talutterance,causingtruncationoffalls.Grabe(1998)showshowalong-vowelledGermansyllableschief ‘slanted’ will show a fall to low pitch in the same communi­cative context in which short­vowelled Schiff ‘ship’ shows high pitch or a fall from high tomid.The tendency todelayH is reflected in right-spreadingof tones inAfricantonelanguagesinparticular.Forinstance,inYorubaL-HisrestructuredasL-LHandH-LasH-HL(Hyman2007).

Floatingtonesmaynotinfacthaveatarget.Foronething,languagesmayrou­tinelydeletethem.Whentheyareretained,theirpresencemayonlybefeltinthetargetofaneighbouringtone,asdiscussedaboveunder‘Downstep’.Floatinginto­nationaltonesaretypicallyretainedandpronounced.InFigure 10.2,thepitchcon­tour and speech waveform are given of You may MINimize the email window by clicking on that DASH there.TherearetwooccurrencesofthepitchaccentH*L;thesentencebeginswith%LandendswithH%,asshownin(35).Ofthesesixtones,twoareassociated,whilefourarefloating.InEnglish,onlyonetonecanassociatewiththeTBU,theaccentedsyllable,the‘starred’tone,H* in this case. The timing of associatedtoneH* is relatable to a specific syllable, but there is no invariant syllable withwhichthefollowingListimed.Inafinalsyllableintheintonationalphrase(IP,

Tones 167

seechapter12),itoccursonthesamesyllableasH*,butin(36)itmayonlyoccuron can,twosyllablesaftertheaccentedsyllable.Itwouldn’tmakemuchofadiffer­enceifLwaspronouncedduringyou or even see,butasimilardisplacementofH* would be very noticeable and would create the impression of a different intonation pattern. That is, the timing of the pitch accent is determined by the anchoring of the starredtonetotherhymeoftheaccentedsyllable.Unsurprisingly,%LandH%arerealizedatthebeginningandendoftheIP.Boundarytonesillustratepar excellence how tones, like all phonological elements, are aligned somewhere in the linguistic structure(seealsosection11.3.2;McCarthyandPrince1993).Analignmentstate­ment specifies the edges of two constituents which should temporally coincide. In English,therightedgeofH%alignswiththerightedgeoftheIP.Trailingtonesofnon-finalpitchaccents,likeLofH*Lonminimize in(35),alignrightmost,beingpronouncedjustbeforethetargetofH*.Bycontrast, trailingtonesoffinalpitchaccentsarepronouncedleftmost,herecausingasteeperfallinH*L.IfweexpandtheIPwithunaccentedwords,asin(36),thefallfromDASH would still be steep and you can see or can see will more or less have low­level pitch throughout.

(35)

H*L

You may minimize your email window by clicking on that dash there

H*L H%%L

(35)

(36) (36) … clicking on the dash you can see there

H*L H%

0 3.5Time (s)

uoY yam minimize the email window by clicking on that dash there

0

300

100

200

Pitc

h (H

z)

%L

L L

H*

H*

H%

Figure 10.2PitchcurveofaBritishEnglishutteranceofYou may minimize the email window by clicking on that dash there,spokenwithH*Lonmin- and on dashandaninitial%LandfinalH%.

Tones168

Q102 In Kipare, an utterance-final H is downstepped, or in our terms finally lowered, as shown in (1a) (Compare (1b); Odden 1986).

(1) a ìpá!ŋgá ‘machete’b ìpáŋgálédi ‘good machete’

AssumethattheendoftheutterancehasaboundaryL%,whichcausesthedownstep on the final H.When L-tones precede the final H, a frequent,thoughnotobligatory,pronunciationisoneinwhichthefinal(downstepped)Hspreads leftward toall theL-tonedTBUs. (2a) shows thepronunciationwithoutthespreading,(2b)thepronunciationwith.

(2) a nìfínìkìrèmàɣèmbèmàè!ðá ‘I covered long hoes’b nìfí!níkírémáɣémbémáéðá

Althoughitisoptional,thespreadingofthefinalHshownin(2b)isalwaystoallTBUs.Thatis,itisnotpossibletopronouncethesentencewithspreadingofHtoonlysomeofthewordsthatarepronouncedwithlowpitchin(2a).WhatdoesthissuggestabouttherepresentationoftheL-tonedwordsbeforeHspreads?

Q101 Describe the f0 pattern that will occur on John if the sentence Has JOHN? is pronounced with the intonation contour %L H*L H%. Give a representation with tones and targets in the way shown in (35) and (36).

The term ‘alignment’ has been used for the phonological specification of a tone’s location(Pierrehumbert1980;Gussenhoven2000)aswellasforthedetailedpho­netic specification of a tone’s target (‘phonetic alignment’; Arvaniti et al. 2000).Languagesdifferinthetypicallocationofbothassociatedandfloatingtones.OnefactorthathasbeenshowntocauseretractionofH-targetsistheproximityofanupcomingH-targetor IPboundary.Theneed for sufficient space to carryout apitchmovementmaythereforecounteractthetendencytoallowH-targetstodriftrightwards.

10.8 PRIVATIVE TONE

Tonesystemsoftenlendthemselvestoaprivativeanalysis.Therearetwowaysinwhichacontrastcanbeprivative.Mostfrequently,thetonecontrastexistsonthe

Tones 169

surface, but, underlyingly, there may be no need to specify one of the tones involved in thecontrast.For instance, ina languagewithHandL,onlyHmighteverbereferredtobyphonologicalgeneralizations.Whenalltheconstraintsofthegram­mar have been taken into account, any remaining toneless syllables are provided withLtone.

TheSomaliwordhasmaximallyoneH-tone,always in thefinal twosyllables.TBUswithoutaH-tonearepronouncedatlowormidpitch,whichwemayassumeisduetoadefaultLaddedatafinalstagetofreemoras,theTBUsofSomali.Inthegrammar,therearemanyrulespositioningtheH-tone,butnorulesotherthanthedefaultruleeverrefertoL.

Q103 Somali uses tone to encode a gender contrast (Hyman 2009). Mas-culine nouns have falling tone on final long syllables, low tone on final short syllables and, depending on the vowel quantity in the last two syllables, high, low or rising tone on the penult, while femi-nine nouns have high tone on final short vowels, rising tone on final long vowels and low tone on the penult.

a ínàn ‘boy’ ìnán ‘girl’b ʕèèsnáàn ‘young he­goat’ ʕèèsnàán ‘young she­goat’c náʔàs ‘stupid man’ nàʔás ‘stupid woman’d qàálìn ‘young male camel’ qààlán ‘young female camel’e dàrmáàn ‘colt’ dàrmàán ‘filly’f béèr ‘liver masc’ bèér ‘garden fem’

Canyouimproveontheabovedescriptionofthetwotonepatterns?

Theprivativeoppositionmayalsopersistonthesurface.IntheCentralFranco­niandialectsofGermanandDutch,abouthalfthewordshavealexicaltone,otherwords being toneless. The languages have a generous supply of intonational tones, to which one or more lexical tones are added depending on the number of words withlexicaltoneinthesentence.Ifnosuchwordsareincluded,thesentencejusthas intonational tones. InHasseltLimburgish, the lexical tone isL,whileLH* is theonlypitchaccent. In(37a),only intonational tonesoccur,while in(37b) theword for ‘socks’hasa lexicalL-tone in the stressed syllable.Because theTBU isnowoccupied,thepitchaccent,whichisalignedtotherightofL,hasnoTBUtoassociatewith.Phonetically,thepitchpeakduetoHin(37b)islaterthanthatin(37a)(Peters2008).

(37) a ix mut n x n paar ki sk s h b b ix mut n x n paar ki sk s h b

%L L*H L% LL*H‘I still need a few cheeses’ ‘I still need a few socks’

L% L%

(37)

Tones170

10.9 NOT BY f0 ALONE

Vocalfoldsarenotjustemployedforcreatingvariationsintheirrateofvibration.Voicequalitydifferencesareeffectedbydifferentmodesofvibration,whilevoicingcontrasts in consonants provide yet another use in the phonologies of languages. In sections10.9.1and10.9.2,wediscusstheconnectionsbetweenthesetwoaspectsand tone.

10.9.1 Voice quality

Modalvoicemayphonologicallycontrastwitheitherbreathyvoiceorcreakyvoiceorboth(seesection2.2.3).Breathyphonationisassociatedwithlowerratesofvocalfold vibration, since the airflow through the glottis is to a large extent usurped by theneedtocreatefrictionbetweenthevocalfolds.AdepressingeffectofbreathyvoiceisseeninItunyosoTrique(DiCanio2012),wherebreathyvoiceoccursasaresultof[ɦ]inthecoda.Comparedtothef0patternsinopensyllables,thef0pat­ternsofthefourleveltonesdescendinsyllablesclosedby[ɦ],whilethemidfallingcontour descends more steeply. Interestingly, in addition to these five tones, open syllableshavetwofallingcontours,whereassyllablesclosedby[ɦ]havetworisingcontours.Apparently,havingsevencontrastivefallingcontoursgoesbeyondwhatis ergonomically tolerable for the language.

Creaky voice presents a more complex picture. Often, it is associated withlowtones.InStandardMandarin,Tone3,thelowordippingtone,routinelyhascreak,whichhelpstomakeitdistinctfromtherisingtone,Tone2.Creakasanenhancement of low pitch is a frequent feature in languages, a connection which must stem from the ease with which irregular, creaky phonation can be achieved withlowervibrationrates.Atthesametime,tensenessofthevocalfoldsasusedin‘constricted’or‘tight’phonationwillleadtohighervibrationrates.Kingston(2005)presentsthedevelopmentoflaryngealsegmentsintohighandlowtonesindifferentdialectgroupsofAthabaskanasresultingfromtheseoppositeeffects.Co-occurrencesofphonationtypesandlexicaltonesarecommoninvarietiesofVietnamese(Michaud2004;Brunelle2009).Whencreakyvoicederivesfromtheoccurrenceof[ʔ]inthecoda,ashorteningofthesonorantportionoftherhymefrequently occurs. Since pitch shapes are less distinguishable on short vowels,a reduction in thenumberof tonecontrasts typically results.For instance, theseven-tone contrast in non-glottal rhymes in Itunyoso Trique is reduced to athree­way contrast in glottal rhymes.

10.9.2 f0 perturbations and tone distribution

Consonantscan‘perturb’thesmoothphonationwhichischaracteristicofvowelsand sonorant consonants, with their free flow of air from the glottis onwards. Theoralconstrictionofvoicedobstruentslike[b]or[v]willcauseairpressureto

Tones 171

buildupintheoralandpharyngealcavities,asaresultofwhichf0ratesmaybedepressed, because the higher air pressure above the larynx will inhibit the flow ofairthroughtheglottis.Theeffectisoftenvisibleasadipinthef0track.Thedepressingeffectmaycarryovertofollowingvowels,whosef0willbelowerinthefirst20msorsothanaftersonorantconsonantslike[m].Voicedimplosiveconso­nantslike[ɓ ɗ ɠ]mayhaveastrongersucheffect.Vocalfoldvibrationhererelieson the lowering of the larynx, which will result in a simultaneous subglottal air pressure increase and supralaryngeal air pressure decrease. This pressure differ­ence is used to get the vocal folds to vibrate, but the effect is short­lived and weak andleadstoalowvibrationfrequency.Voicelessobstruents,bycontrast,arelikely to have an f0 raising effect due to the tensing of the vocal foldswhich is cre­atedinordertokeeptheglottisopen.Tonesystemsmayreflecttheseeffectsinsegmentdependentdistributions.LanguagesinAfricafrequentlyhavelowtonesafterimplosives,andvarietiesofWuChinesetendtohavethefullsuiteoftonecontrastsonlyafter sonorantconsonants.TheShanghaivarietyhas three toneswhichappearonsonorantrhymes,HL,MHandLM,buttheycontrastonlyaftersonorantonsets.Aftervoicelessobstruents,onlythehighertwotonesoccur,HLandMH,whilevoicedobstruentsonlytolerateLM,the lowest tone.Inrhymesclosed by a glottal stop, which have very short sonorant portions in the rhyme, there isa two-waycontrastbetweenHandLaftersonorants,while there isnocontrastafterobstruents,withHoccurringaftervoicelessonesandLaftervoicedones(ChenandGussenhoven2015).TheYuhuanvarietyhasmoretonecontrastsinsonorantrhymes,asshownin(38),butshowsasimilardistribution.Thatis,aftervoicelessobstruents,H,HLandLHappear,whileL,MLandLHappearafter/bdg dz v z ɦ/,thevoicedobstruentonsets(Gussenhovenetal.2016).

(38) a mi b mi c mi d mi e mi

H L HL ML LH‘glance’ ‘temple’ ‘second

(unit of time)’‘depict’ ‘seedling’

(38)

Q104 Yuhuan Wu syllables can begin with [j] and [w], optionally pre-ceded by a consonant. These glides are either to be analysed as onset consonants or as initial vowels in the rhyme, so that the seg-mental structure of the LH word for ‘rice seedling’, phonetically [jã], is either /jã/ or /iã/. Syllables of this type have a three-way tone contrast.

1 Whatwouldyouconsiderthecorrectanalysisoftheglides?2 Whichthreetonesdoyouexpecttooccurinthesesyllables?3 Andhowmanytonescanappearonasyllablelike/i/?

Tones172

Q105 Giryama has a H vs Ø contrast on words, with default L-insertion in toneless syllables. An underlying H-tone is pronounced on the penultimate syllable of the intonational phrase, regardless of the position of the morpheme it comes with. In (1a) no H-tones occur, while in (1b) the 3sg prefix /á/ occurs in initial position. However, its H-tone associates with the penult, where a somewhat delayed pronunciation causes that syllable to have rising pitch (Volk 2011).

1 Explainthedifferencebetweenassociationand(phonological)alignmenton thebasisof (1b), andbe explicit onwhetherH spreads rightwardsfrom its sponsoring morpheme.

2 WhatmightbetheexplanationoftheassociationofHwiththepenultratherthanthefinalsyllablein(1a,b)?

3 WhentwoH-tonesoccur,thefirstassociateswiththefirstsyllableoftheverb,asin(2a),whichcontainsaverbwithanunderlyingH.Whatmightbe the reason that the firstH-tone in (2b) associateswith the syllableprecedingthe(H-toned)verb?

4 DoestheconstraintontheassociationofthefirstHwiththeverbequallyapplytothefinalHintheIP?

(1)a nìnàmàrìgìzìːkà b à nà màrìgìziːkà 1stprescome­to­an­end ‘Iamcomingtoanend’ ‘Sheorheiscomingtoanend’(2)a ànàkúlìhìriːkà b ànábàmbàhuːlà ‘Sheorheistrustworthy’ ‘Sheorheisseparating’

10.10 CONCLUSION

Thischapterhasdealtwithaconspicuousphonologicaltypeofsegment,tones.WehaveseenhowtoneinAfricanlanguagesgaverisetotheautosegmentalmodelofphonological representation and the main constraints that govern those represen­tations, like the NoCrossingConstraint, the TwinSisterCovention,theNoCrowdingConstraintandtheOCP,aswellasassociationalgorithms like theAssociationConvention.Tonescanbedistinguishedonthebasisoftheirfunc­tion: they may be lexical, grammatical and intonational. Phonologically, they may be associated or floating, but always located somewhere, depending on their pho­nological alignment.Wehave seen thatdescriptionsof thedistributionof tonesmay call on concepts like tone melody and accent. The phonetic implementation of toneswas discussed in terms of their scaling, the placement in the f0 range,and their synchronization with the segmental structure, the phonetic alignment. For each dimension, a number of widely attested factors were identified that

Tones 173

systematicallyaffecttherealizationoftonetargets.Finally,wehavedrawnattentionto the privative nature of tone oppositions and the way tone systems may be shaped by theperturbationsof f0causedbyconsonantalarticulations.Thenextchaptertackles the third aspect of word prosodic structure, stress.

NOTES

1 Seehttp://www.oto-manguean.surrey.ac.uk/forasurveyoflinguisticpropertiesofOto­manguean languages.

2 Additionally,multi-tonepitchaccentshavea‘+’betweentheirtonesinPierrehumbert(1980) and a great deal of laterwork, as in L*+HorL+H*, a convention that is not adopted below.

Word stress

11.1 INTRODUCTION

Besideslexicaltoneandquantity,wordsmayhavestress.Thisisaphonologicalver­sion of an articulatory tendency to pronounce syllables with an alternating strong­weakpattern.Forsomereason,itishardtopronounceasequenceofsyllableswithequalcareforeachofthem.Forinstance,whenrecitingthealphabetinalanguagewith monosyllabic words for the letters, the tendency may reveal itself in a strong­weakpairingoftheletternames:ÁBĆDÉFǴ. In many languages, this tendency is reflected in the phonological structure of words, known as a ‘metrical structure’, a ‘foot structure’ or simply ‘stress’.

Therearephonologicalaswellasphoneticdiagnostics forstress.Hyman(2006)identifies three phonological properties that characterize word prominence as ‘stress’:

1 Thewordprominenceisobligatory.Languageswithwordstresshavestressonalloftheirwords(exceptfunctionalmorphemes,whichmaybeincor­poratedwithotherwords;see12.6.1onclitics).Asanobligatorypropertyofwords, stress differs from tone or bimoricity.

2 Stressisapropertyofasyllable. In particular, there is no such thing as a stressed mora as distinct from a syllable.

3 Stressisculminative, meaning that one syllable in the word has the greatest prominence.

Phonetically, stress has been defined as increased articulatory precision as shown by an increased segmental duration, more peripheral vowel qualities and greater intensity, inparticularforthehigherspectralfrequencies(Fry1955,1958;Beck­man1986;SluijterandvanHeuven1996).Theemploymentoff0featuresisseenasa reflection of the presence of tones on stressed syllables.

Insection11.2,weconsiderhowstressmanifestsitselfinlanguages,keepingtheabovedistinctionbetweenphonologicalandphoneticcriteriainmind.Wewilldis­cuss tone systems that resemble stress systems in a distributional sense, making the point that such systems have sometimes been described as stress systems, as well as languagesthathaveneithertonenorstress.Tones,whetherornotanalysedaspitchaccents, are more frequently deleted postlexically than stress and are as a result more likely tofigure inmorphosyntacticgeneralizations.Section11.3willmakethe point that stress systems always have a default location, meaning that there are noclearexamplesoflanguageswhichdistributestressentirelyunpredictably.Alan­guage need not have a single default location, like the penultimate syllable, but may

11

Word stress 175

vary the default as a function of the segmental make­up of syllables that potentially have stress, a phenomenon known as quantity­sensitivity or weight­sensitivity. Importantly, we will see how the formulation of generalizations about the loca­tionofstresshasbeenanexcitingareaofresearchinthepastdecades.Weidentifyand define the main parameters that figure in these generalizations on the basis of datafromalargenumberoflanguages.Finally,section11.4showshowOptimal­ity Theory has been applied to the problem of deriving default stress distributions.

11.2 MANIFESTATIONS OF STRESS

Syllableswithstressmaydistinguishthemselvesfromothersyllablesinanumberofways. In many languages, stressed syllables allow greater phonological complexity, anothercaseofphonologicalprivilege(‘positionalfaithfulness’),besidestheposi­tionofthesyllableonset(section3.2;Beckman1988).First,theyarethesyllablestowhich tones associate. These tones or combinations of tones are also known as ‘pitch accents’, a somewhat overused term which should be used with care to avoid ambi­guities.Asillustratedinsection10.7,Englishpitchaccentsmakeuptheintonationofsentencestogetherwithboundarytones(Pierrehumbert1980).AcommonpitchaccentofEnglishisarising-fallingpeak,whichmaybeanalysedasafallingmelody(H*L; e.g.Leben1976).Anutterance inEnglishmustminimallyhaveonepitchaccent, and a citation pronunciation of Edinburgh, which has the main stress on the antepenultimatesyllable,[ˈɛdnbrə],will thereforehaveapitchaccent(hereH*L)associatedwiththefirstsyllable,asseenin(1a).BecauseinAlaska [əˈlæskə] thesecondsyllablehasthemainstress,thisiswherethepitchaccentgoes,asin(1b).Aberdeen[ˌæbəˈdiːn]hastwostressedsyllables,thefirstbeinga‘secondarystress’.Whilethemainstressmusthaveapitchaccentwhenthewordissaidbyitself,asecondary stress before the main stress may also have a pitch accent, creating two pitchaccentsinthisword,asshownin(1c).AbouthalfofthestressedsyllablesinanEnglishtextwillbeaccented,i.e.haveapitchaccent.

(1) (1)

||||%L H* %L L L L% %L H*L% H*L L%

AberdéenAláskaÉdinburgha c b

H*

Just as only stressed syllables may have a pitch accent, stressed syllables may be segmentallyprivileged.Theyuniquelyhavebimoraic syllable rhymes inSwedish,where bimoraic quantity for vowels is mutually exclusive with geminate consonants (e.g.*/at/,/aːt/,/atː/,*/aːtː/;Riad2014).Catalanhassevenvowelsinstressedsyl­lablesbutonlythreeinunstressedones(Nadeu2016),andmostvarietiesofEnglishhave large numbers of vowels, but only a small number of vowels can appear in unstressedsyllables(Bolinger1986:347;cf.Gussenhoven2011).Whenstressedsyl­lables are signalled by a pitch accent, but do not differ from unstressed ones in seg­mentalcomplexity,theymaystilldifferindurationorvowelquality.Forinstance,

Word stress176

Spanish unstressed syllables,which don’t obviously distinguish themselves fromstressedsyllablesintheirsegmentalcomposition(Berg1998:112),areshorterthanstressed syllables (Ortega-LlebariaandPrieto2011),while the samewillbe truefor many other languages without enhanced segmental complexity in stressed syl­lables. In all of these cases, the stressed syllables are said to have phonetic stress. It meansthatbesidesf0featureswhichareattributabletothepresenceofstress,thereare observable durational and spectral properties, whether or not these are due to phoneticrealizationsofsegmentallysimilarsyllables,asinSpanish,ortosystem­aticsegmentaldifferencesbetweenstressedandunstressedsyllables,asinEnglish.In fact, the distinction between ‘pure’ phonetic stress and segmentally induced phoneticstressmaybehardtodraw,becauseitmaydependontheanalysis.Forinstance, what one phonologist may regard as the stressed and unstressed allo­phones of the same vowel, another phonologist may take to be two different vowels.

Q106 Up until the 14th edition of Daniel Jones’s English Pronouncing Dictionary (Jones 1977), the words forward and foreword were given as [ˈfɔːwəd] and [ˈfɔːwəːd] (ignoring other differences), while in the 14th edition, which was fully revised by his successor A.C. Gimson, they have been given as [ˈfɔːwəd] and [ˈfɔːwɜːd]. How can this difference be interpreted phonologically?

Stressed syllablesmay equally be the privileged locations of lexical tones. In the ColognedialectofGerman, themainwordstress is the locationofa tonecontrast(GussenhovenandPeters2004),whilethesameistrueforword-finalstressedsyllablesinMa’ya,whichhasathree-waycontrastthere(Remijsen2007),andSwedishnon-finalsyllables,whichlikeCologneGermanhaveatwo-waytonecontrast(Riad2014).

11.2.1 Absence of phonetic stress

Whatifalanguagehasnophoneticstress,i.e.nophoneticpropertiesapartfromthef0effectsofthepitchaccent,butdistributesf0featuresinawaythatisreminiscentofthewaystressisdistributedacrosswords?InPersian,forinstance,everymajor-class word as pronounced in isolation has high pitch on the last syllable, which is evidentlyduetoaH-tone,butbecauseithasnophoneticstress(Abolhasanizadehetal.2012),anymotivationforsayingithas‘stress’liespurelyinthedistributionalpropertyofthetone:everywordhasone,atleastwhensaidinisolation.Beckman(1986) avoided any possible confusion by distinguishing syllableswith phoneticstress as having ‘stress accent’ and syllables without as having ‘non­stress accent’, respectively.Under ‘stress accent’, she explicitly includedcases inwhich stressedsyllablesdifferfromunstressedonesinphonologicalcomplexity,likeEnglish,whileJapanesewasher exampleof a languagewith ‘non-stress accent’. Still, insteadof‘non­stress accent’, many linguists will use the term ‘pitch accent’ here. In this usage of that term, it is important to keep in mind that reference is made to a language

Word stress 177

withoutphoneticstress,andthat‘pitchaccent’heremeans‘exclusivelytone’.RecallthatEnglish,alanguagewithobviousstress,also has pitch accents. Thus, the pres­ence of a pitch accent does not in general exclude the presence of phonetic stress.

Whentonesaredeletedfromasyllableinalanguagewithoutphoneticstress,thatsyllable is phonologically identical to a syllable without tone at any representational level.Toillustratethehomophonyof‘deaccented’syllablesandsyllablesthatneverhad a tone, we turn to Persian again. The generalization that the last syllable of every major-classwordhas aH-tone is true regardless of the attachmenton the rightof functionwords, likeauxiliaryverbsandpossessives.For instance, [tɒbéʃ], theproper name Tabesh,isdistinctfrom[tɒbeʃ]‘myswing’,where[eʃ]isthe1sg pos­sessive.Thelanguagehasanumberofrulesdeletingtones(‘pitchaccents’).Oneoftheseappliesinobjectclausesoffactiveverbs(Rahmanietal.forthcoming).Inthiscontext,thetwoabovestructuresarehomophonous,causing(2)tobeambiguous.

(2) mehdímídunekeuntɒbeʃe a Mehdiknowsthatthat+deMTabesh-is ‘MehdiknowsthatthatisTabesh’ b Mehdiknowsthatthat+deMswing-my-is ‘Mehdiknowsthatthatismyswing’

Oragain,inTokyoJapanese,[káki]‘oyster’,[kakí]‘fence’andunaccented[kaki]‘persimmon’ are homophonous when used as the first constituent in a compound with [zíken] ‘case’,making [kakizíken] ambiguous between ‘the oyster case’, ‘thefence case’ and ‘the persimmon case’.

Q107 Nubi has no phonetic stress. Every word is lexically marked for the location of a H-tone appearing on the antepenult, penult or final syllable, referred to as the ‘accent’. The language has two rules that manipulate accents. First, a gerund with a direct object is deac-cented, as shown by the contrast in (1) and (2).

(1) Ána gúrúa wedí buká I will give books ‘I will give books’(2) Wedi buká séme giving books good ‘Givingbooksisgood’

Second,anadjectivalverbbeforeanounlosesitsaccentinfavourofanaccentonitsfirstsyllable,asillustratedin(3)and(4).

(3) tabán ‘totaunt’(4) tábantóru ‘totauntabull’

1 Whatare theNubi translationsof ‘Iwill tauntabull’, ‘Iwill taunt’and‘Tauntingabullisgood’?

2 GivetwoargumentsforcharacterizingtheNubiaccentastone,notstress.

Word stress178

11.2.2 Phonetic stress as an obligatory feature

EventhoughHyman(2006)defines‘stress’intermsofdistributionalpropertiesandis thus neutral on how it is realized, the languages with ‘obligatory’ stress he dis­cusseswouldallappeartohavephoneticstress.Apparently,phoneticstressisunde­letable,i.e.obligatory,andinvariablyshowsupinsurfacerepresentations.Englishforeword[ˈfɔː(r)wɜː(r)d]andforward[ˈfɔː(r)wə(r)d]aredistinctinanyphonologi­calcontext,evenifnotones(pitchaccents)appearontheirstressedsyllables.Incontrast, tones tend to be manipulable and are thus frequently targets of phonologi­cal rules, including deletion rules.

Languages vary in theway they exploit toneon stressed syllables for express­inglinguisticcontrasts.Untypically,fromanIndo-Europeanpointofview,everyEgyptianArabicword in the sentencehas a pitch accent on its stressed syllable(Hellmuth2007).Bynotbeingdistinctivelypresentorabsent, theonly functionithasistosignalthepresenceofaword.Attheotherextreme,Englishhasmanygrammatical distinctions which are expressed through differences in accentuation (cf.section10.6).Forinstance,The STORKS are RINGED counts as a general rule of practice, while The STORKS are being ringed might be used to report on the actual process of people providing some storks with tracking rings.

11.2.3 Lexical vs postlexical prosody

Discussionsofwordprosodic structures should strictly speakingbe confined tothosethatareincludedinlexicalrepresentations.Fortone,thisistypicallymadeexplicit. Definitions of the notion ‘tone language’ invariably include some suchprovision,andpostlexicaltonesareseparatedoffasintonationaltones.However,postlexicaltones(whetherornottheyarecalled‘pitchaccents’)havebeenreferredtoas‘stress’whentheirdistributionresemblesthatofstress.Forinstance,Frenchwords are sometimes said to have word­final stress as a result of the fact that the language has pitch accents on the initial and final syllables of phonological phrases, and a citation pronunciation of a word will therefore have a pitch accent on its finalsyllable.‘Real’stressispostlexicalinlanguagesthatassignfeettophonologicalwords,withoutanyexceptions.Forinstance,Bengalihasstressonthefirstsyllableofeverymajor-classwordandrhythmicstressoneveryoddsyllableinthewordthereafter(Chatterji1921:19).Theonlyclearlyattestedmanifestationisthepres­ence of a pitch accent on the first syllable of the first such word in the phonologi­calphrase(HayesandLahiri1991;Kahn2010),butthereportofbinaryrhythmicstressbyChatterji (1921)suggests that the languagehaspostlexical stresswithaphrasally determined postlexical pitch accent, the latter aspect being comparable toFrench.Paralleltolexicaltone,stressislexicalifitslocationissensitivetomor­phological operations or if segmentally comparable underived words have different stress locations(seesection11.3.4).Postlexicalquantityoccurs inItalian(bytheside of lexical quantity due to its intervocalic geminate­singleton contrast, illus­tratedby[ˈdi.ta]‘fingers’vs[ˈdit.ta]‘businesscompany’).Ifthelastwordinaninto­nation phrase has penultimate stress on an open syllable, the vowel is lengthened, asin(Vivo a) R[oː]ma ‘IliveinRome’.Thelexicalstatusofprosodicphonological

Word stress 179

features may be detectable in the results of a short­term memory task involving prosodicminimalpairslike[númi–numí],developedbyEmmanuelDupouxandcolleagues(Dupouxetal.2001;Rahmanietal.2015).Inthattask,speakerswhoselanguageincludeslexicalprosody,likeDutchandJapanese,outperformspeakersoflanguageswithout,likeFrench.

Languageswithpostlexicalprosodyaffectingtherepresentationofwords,likeFrench(tone),Bengali(stress)andItalian(longvowels),shouldbedistinguishedfrom languages that have no word prosodic representation either lexically or postlexically,likevarietiesofMalay,includingIndonesian(Maskikit-EssedandGussenhoven2016).AmboneseMalayhasintonationalmelodies,buttheycanonlyberelatedtotheedgesofintonationalphrasesandthefinalword.Declara­tives end in a rising­falling contour, which tends to be completed within the space of the final word, but there is no systematic temporal alignment of the contour with any syllable. The peak will generally be closer to the phrase end if the final word is a monosyllable than if it is polysyllabic. The melody can thus be describedasafloatingLHLboundarytonecomplex,butnotasastress-boundor syllable­bound pitch accent.

11.3 DESCRIBING STRESS PATTERNS

Stressisapropertyofsyllables.However,sincetherewasnorepresentationofthesyllable at the time, the earliest phonological description indicated stress by employ­ing an n-aryfeature[nstress]tobeincludedinthefeaturespecificationofvowels.Apartfromthequestionableattributionofstresstovowelsinsteadofsyllables,therewere twounsatisfactoryaspectsof thisanalysis.First,wordstresswasconceivedofasvaryinglevelsofstressacrossthesyllables(vowels)ofaword,with[1stress]beingthemostprominent(ChomskyandHalle1968).Thismeantthat(avowelin)a syllable with word stress and a pitch accent was seen as possessing a more intense level of stress than a stressed syllable without a pitch accent, something which most phonologists today will see as a conflation of different concepts, stress and tone. The other unsatisfactory aspect was that stress specifications for syllables ignored the fact that the stress systems of languages reflect the alternation tendency noted at thebeginningofthischapter.Stressedsyllablestypicallyalternateinabinaryfash­ion,s-w-s-w-s-w,asopposedto,say,s-s-w-w-s-s.However,aruleassigningstressaftereverytwo,three,fourormoresyllablesisnotmoreorlesscomplexthanaruleassigningstresstoeveryoddsyllable.Whatwouldbeneededhereissomerepresen­tation of stress that imposes, or favours, this ‘counting­by­two’ aspect.

11.3.1 Footing

Thefirstproposal forgatheringsyllables intogroupsof twowasmadebyLiber­manandPrince(1977)andVergnaudandHalle(1978),whointroducedabinarybranching treewith a strong and aweaknode, referred to as a ‘foot’ by Selkirk(1978).Adoptingthisrepresentationaldevice,Hayes(1981)formulatedthethree

Word stress180

parameterslistedin(3)tocharacterizestresssystems.Insteadoftrees,wewillusetwo­layered foot and word structures to indicate the metrical structure.

(3) Dominance: IambvsTrochee Edge: Lefttorightvsrighttoleft Weight: Quantity-sensitivevsquantity-insensitive

Dominance

Languagesmaychoosewhethertherightorleftnodeofabinarytreeisdominant.Aw-sstructureisknownasaniamb(/ˈaɪ.æm(b)/),whileas-wstructureisatro-chee (/ˈtrəʊ.ki/ or/ˈtrəʊ.kiː/).ManyEuropeanlanguagesaretrochaic,whilemanyAmericanlanguagesareiambic(vanderHulstetal.2010).

Edge

Languages differ inwhether themain stress occurs near the beginning or theend of the word, expressed by the choice for the second parameter. The words in left-edge languagesmay, for instance,have either initial orpeninitial stress,while those in right­edge languages may have penultimate or final stress. There aretwosub-parametersunder‘Edge’.First,whenfootstructurecoversthewholeword, the footing is ‘iterative’, and feet are built until the opposite word edge is reached. This potentially creates two or more stressed syllables in the word. Itera­tivefootingisalsoknownas‘rhythmicstress’.Second,wheniterativefootingendsupexcluding justa single syllable, some languages leave that syllableunfooted(a ‘stray’ syllable),whereasothers turn it intoamonosyllabic foot,knownasa‘degenerate’ foot.

Weight

In many languages, the composition of feet is sensitive to the segmental composi­tionofsyllables.Ahighlyfrequentrestrictionisthataweaknodemaynotdominatea bimoraic, i.e. ‘heavy’, syllable. If the language builds trochees from the right edge, thenastructurelike[ta.ta.ta.taː]wouldnottolerateatrocheeoverthelasttwosyl­lables,sinceword-final[taː]wouldendupinitsweakposition.Instead,thefinalsyllable would be a monosyllabic foot, with the disyllabic trochee being built over the penultimate and antepenultimate syllables: [ta.ˈ(ta.ta).ˈ(taː)]. Languages thatare quantity­insensitive build the foot structures over their words without regard for the segmental composition of syllables, even in cases in which the language has both heavy and light syllables.

Q108 English builds quantity-sensitive trochees from the right edge of the word. Determine on the basis of the following words whether English has stray word-initial syllables.

Word stress 181

Wewillillustratethethreeparameterswiththestresssystemsofthreelanguages.First,Icelandicwordsbeginwiththemainstress,andsecondarystressoccursonotheroddsyllablesintheword(Árnason1985).Thisisshownin(4),wherefeetareindicated by parentheses, with the stressed syllable marked ‘x’, while ‘.’ indicates an unstressedsyllable.Toindicatethatthefirststressedsyllablehasthemainstress,anextra layer is added over the foot which represents the word. In it, the stressed syl­lable with the main stress has an extra x placed over it.

(4) (x ) (x .) a tas.ka ‘briefcase’ (x ) (x .)(x .) b a.kva.rel.la ‘watercolour’ (x ) (x .)(x ) c høv.ðiŋ.gja ‘chieftain-gen+pl’ (x ) (x . )(x) d sep.tem.ber ‘September’ (x ) (x .)(x .)(x) e bi.o.gra.vi.a ‘biography’

Intermsofthesefirstthreeparameters,Icelandicbuildstrocheesfromtheleftwordedgeandisquantity-insensitive.Forthislastpoint,noticethatclosedsyllablestoleratebeinginanunstressedposition,asshownby[ðiŋ]in(4c)and[tem]in(4d).

The second and third languages differ from Icelandic and from each other in terms of‘Dominance’,‘Edge’,thesub-parametersof‘Edge’and‘Weight’.Garawashowsthatthe edge which is relevant for the main stress may be different from that for the secondary stresses. This type of language indicates that the main stress may become disassociated from an alternating rhythm, and suggests that it is important to distin­guish‘rhythmicstress’from‘mainwordstress’.Examples(5a,b)showperfectIcelan­dicstructures,butwhileGarawabuildsatrocheeattheleft word edge for the main stress, secondary stresses occur on even syllables, counting from the right, as shown in(5c),whereiterationfromtheleftwouldhavecreatedatrochee[la.rin](Furby1974;Hayes1981).Regardlessoftheedgefromwhichtheyarebuilt,thetrocheesareagainquantity-insensitive,asshownby[cim]in(5d)and[ɾim]in(5e).Exam­ples(5c,e,f)showanotherdifferencewithIcelandic.Left-oversyllablesare‘stray’inGarawa,butarefootedinIcelandic.Thesedegeneratefeetwereshownin(4c,d,e).

kæn.ˈtiːn canteenbə.ˈnæ.nə bananasə.ˈnɑː.trə Sinatrasæn.ˈtæ.nə Santana

Word stress182

(5) (x ) (x .) a ja.mi ‘eye’ (x ) (x .)(x .)(x .) b ja.ka.la.ka.lam.pa ‘loose’ (x ) (x .) (x .) c ka.ma.la.rin.ɉi ‘wrist’ (x ) (x .)(x .) d wa.cim.pa.ŋu ‘armpit’ (x ) (x .) (x .)(x .) e ŋaŋ.ki.ri.ki.rim.pa.ɟi ‘foughtwithboomerangs’ (x ) (x .) f pun.ja.la ‘white’

Hopihasaquantity-sensitiveiambattheleftwordedge.Also,unlikeIcelandicandGarawa,itdoesn’tappeartohaverhythmicstress,andthusbuildsitsfootnon-iteratively.Allsyllablesoutsidetheleft-edgefootarethus‘stray’.

(6) (x ) (x) a les.ta.vi ‘roofbeam’ ( x ) (.x) b ko.jo.ŋo ‘turkey’ ( x ) (. x) c me.loː.ni ‘melon’ (x ) (x) d naː.qi ‘earring’

Q109

1 Whycan’tHopibedescribedashavingaleft-edgeorright-edgetrochee?2 Hopi [ca.qap.ta] ‘disk’has themain stresson the second syllable.This

showsitisquantity-sensitive.Trueorfalse?

AfourthparameterproposedbyHayes(1981),‘Boundedness’,allowsthechar­acterization of stress systems which fail to show the typically observed tendency to confinethelocationofthemainstresstoasmalldomainnearawordedge.Suchunbounded languages don’t have rhythmic stress and are always quantity­sensitive. For instance,mainstress inmonomorphemicwords inAmele isonthe leftmost

Word stress 183

heavy(i.e.closed)syllable,and if therearenoheavysyllables, it ison the initialsyllable.Adiphthonginword-finalpositioncountsasheavy,butaslightelsewhere(Roberts1987:357).Theconditionwithoutheavysyllablesisillustratedin(7a,b,c).Examples(7d,e,f,g)eachcontainoneheavysyllable,whichattractsthestress.Rob­erts(1987)doesn’tgiveanynounsoradjectiveswithtwoheavysyllables;(7h)isaverbillustratingleftmostchoiceoutoftwoheavies.Nofootstructureisindicated,since the pattern was derived without making a choice between trochee and iamb.

(7) a ˈma.la ‘chicken’ b ˈmeu.la ‘righthand’ c ˈni.fu.la ‘k.o.beetle’ d bai.ˈal ‘woodensword’ e ʔo.ʔo.ˈwai ‘prayingmantis’ f ja.ˈwal.ti ‘northwind’ g ˈlaʔ.gbai.a ‘scorpion’ h ˈwol.doʔ ‘surpass’

Amelehasleftmostdefaultstressandstressestheleftmostheavysyllable,butthespecification of the edge of the stress when no heavy syllables are present and the rightmost/leftmostspecificationoftheheavysyllablesareindependent.Thus,fourtypesofunboundedsystemshavebeenreported(Baković2004).Gordon(2000)observes that there is a paucity of phonetic work, in particular on languages with a ‘default­to­opposite­edge’ pattern, and that more research is called for.

11.3.2 Quantity-sensitivity biases for trochees and iambs

Aparametrictheorycarriesanexpectationoforthogonaldimensions,suchthatanycombinationofselectedoptionsisreadilyattested.Unboundedlanguagesgoagainstthis expectation, because of the implication that in such languages quantity­sensitivity andabsenceofrhythmicstressgohandinhand.Hayes(1985,1987)furtherpointedout that iambsareoverwhelminglyquantity-sensitive.He connected thiswith theresults of perception experiments which show that listeners group sequences of stim­uliof equalduration in abinary fashionwith initialprominence, i.e. (tata) (tata)(tata),butsequencesof stimuliwithunequalduration inbinarygroupswithfinalprominence,i.e.(tataa)(tataa)(tataa),theIambic/TrochaicLaw.Thissuggeststhatiambicrhythmisexpresseddurationally.Tocapture the typologicalbias,he intro­ducedthefootinventoryin(8),whereσ is either μμ or μ. The syllabic trochee is meant to cater to quantity­insensitive systems, and cannot be instantiated by a monosyllable. The inherently quantity­sensitive iamb and the moraic trochee should cover quantity­sensitivesystems,andbothofthesecanbemonosyllabic.ForthedisyllabicversionofthemoraictrocheewedeviatefromHayes(1995)inallowingaheavystrongbranch,(μμ μ),bythesideof(μ μ),insteadofjustthe‘even’moraictrochee[μ μ].

(8) a syllabic trochee σ σ b Moraic trochee σ μ ifpossible,orelseμμc iaMb μ σ ifpossible,orelseμμ

Word stress184

Q111 Assuming that the final syllables of the words in (9) are extrametri-cal, do these data require a moraic trochee or an iamb to account for the position of the main stress?

Q110 Cairene Arabic has iterative footing from the left, with main stress on the final foot.

1 Why do the positions of the main stress in items (a) and (b) differ from those in items (c), (d) and (e)?

2 Do these data support Hayes’s position to exclude the ‘uneven’ trochee [μμ μ]?

a (ˈka.ta).ba ‘he wrote’b (ˌʃa.ga).(ˈra.tu).hu ‘his tree nom’c (ˌʃa.ga).(ˌra.tu).(ˈhu.ma) ‘their dual tree nom’d (ˌɁad).(ˌwi.ya).(ˈtu.hu) ‘his drugs nom’e (ˌmar).(ˈta.ba) ‘mattress’

11.3.3 Flexible right edges

Stress systems may behave as if the right edge of the word in fact occurs before the physical word end. It is as if some final element, the last consonant, mora or syllable, is systematically ignored, in which case the ignored element is said to be extramet-rical. A subpart of the Turkish lexicon, among which are underived place names, behaves as if the final syllable is extrametrical. The position of the main stress is on the antepenult if it is heavy (equivalent to ‘closed’, since Turkish lacks vowel quan-tity) and the penult is light, but otherwise is on the penult (Sezer 1983). The fact that only the penult and the antepenult are involved in the generalization suggests we should ignore the final syllable and think of how the two syllables that do matter could be grouped into a binary foot. Examples (9a,b) illustrate how stress goes to the closed antepenult if the penult is open, and (9c,d) show how an open antepenult forces the stress to occur on the penult. The antepenult in (9e,f) is closed, but since the penult is also closed, stress is on the penult. In all cases, the final syllable varies between closed, as in (9a,c,e), and open, as in (b,d,f), without effect.

(9) a ˈer.zu.rum ‘Erzurum’ b ˈan.ka.ra ‘Ankara’ c in.di.a.na.ˈpolis ‘Indianapolis’ d a.la.ˈba.ma ‘Alabama’ e is.ˈtan.bul ‘Istanbul’ f an.ˈtal.ja ‘Antalya’

Another theoretical device that has been proposed is catalexis (Kiparsky 1991). It specifies an empty final syllable (or mora) which is seen by the footing algorithm but is not otherwise observable.

Word stress 185

11.3.4 Morphological effects and exceptions

Morphologicalprocessesfrequentlyimposespecificstresslocations,whichiswhyaccounts are usually based on the evidence of underived words. In English, forinstance,thereisagroupofsuffixeswhichhavethisproperty,like-ee, -eer, -ic, -ity, as illustrated by nomiˈnee (cf.ˈpedigree),mountaiˈneer (cf.ˈWindermere),orˈganic (cf.ˈlimerick)andfriˈgidity (cf.Missiˈssippi).Thereare,broadly,twowaysinwhichsucheffectshavebeenaccounted for.One is touse the regular stressassignmentalgorithmofunderivedwords,butprovide‘stress-changing’suffixeswithafeaturelike[±cyclic],asinHalleandKenstowicz(1991).Theotherwayistospecifythefootstructure they impose on the base in their underlying representation, along with the segments,asstandardlyassumedforSlavic(Stankiewicz1993)andappliedtoEng­lishinGussenhoven(1994).A‘pre-accenting’suffixlike-ity would come with a tro­chee with its weak position over the penultimate syllable and its strong position over an empty syllable before it, while ­ese comesasamonosyllabictrochee.Exceptionscansimilarlybetreatedintwoways.Oneoptionisforthelexicalrepresentationsto include a property that is recognized by the theoretical framework. Italian, for instance, has regular penultimate stress but has a number of underived words with antepenultimate stress as well as words with final stress. Its penultimate regular stress canbeseeninloanwordadaptation,asinthecaseofGerman[ˈhamburk],whichhasbeentakenoveras[amˈburgo](Sluyters1990),wherepenultimatestressisobliga­tory because of the closed syllable. The system is thus readily described with the help ofamoraictrocheealignedontheright.Toaccountfortheexceptionalwordslike[ˈka.li.tʃe]‘chalice’,afinalextrametricalsyllablecanbelistedinthelexicon(cf.regu­lar[kɔrˈni.tʃe]‘frame’),whilewordslike[tʃit.ˈta]‘city’couldhaveafinalcatalecticsyllable. The alternative is to provide their lexical entries with the foot structure con­cerned.TheruleSezer(1983)gaveforthewordgroupexemplifiedin(9)infacthasmanyexceptions,inadditiontovariation(e.g.[ˈys.ky.dar]or[ys.ˈky.dar]‘Üsküdar’).

Q112 A possible proposal for Turkish is to simply prespecify stressed syl-lables or feet in the lexicon (cf. Barak and Vogel 2001; Özçelik 2014).

1 Prespecifying themetrical structureof all ‘Sezer’words goes against awidelyobservedphenomenoninstresssystems.Whatisit?

2 Whichof the followingdataareexceptions to thedescription inSezer(1983)?a [bel.ˈtʃi.ka] ‘Belgium’b [ams.ˈter.dam] ‘Amsterdam’c [ys.ˈky.dar] ‘Üsküdar’d [er.ˈzin.can] ‘Erzincan’e [ˈko.pe.nag] ‘Copenhagen’

3 Apossibleobjectionagainstaspecificationofmetricalstructureinthelexiconforexceptionalwordsoraffixesisthatstructurescanbespecifiedthatdonotexistinthelanguage,likeasuffiximposingword-initialmain

Word stress186

Q113 English [ŋ] obligatorily appears before dorsal consonants in the words in (1), but varies with [n] in the words in (2) (Kiparsky 1979). What is the domain of the obligatory occurrence of [ŋ]? Motivate your answer.

(1) ˈɪŋk inkˈɪŋ.krə.mənt increment

(2) ɪŋ.ˈkriːs or ɪn.ˈkriːs increase (verb)ˈɪŋ.ˌkriːs or ˈɪn.ˌkriːs increase (noun)

Q114

1 DutchhasavocativechantwhichissimilartotheEnglishtune.In[ˈnɑχt.pɔr.ˌtiːr] ‘night porter’, a compound of [ˈnɑχt] and [pɔr.ˈtiːr], the firstpitch level includes the second syllable, while in [ˈnɑχt.ˌʋaː.kər] ‘night

11.3.5 Independent evidence for the foot

Ideally, the reality of the feet that are constructed to account for the stress locations in wordsisalsorevealedbythewayinwhichothergeneralizationsrefertothem.Suchconfirmation is in principle provided by all languages in which different segmental conditionsholdforstressedandunstressedsyllables,likeEnglish(Harris2013).Inaddition, there may be phonological and phonetic processes which refer to the foot. Englishvocativechant,whichconsistsofahigherandalowerpitch,analysedastheH*!Hpitchaccent,spreadsitsfirsttoneuntilitreachesastressedsyllableorthefinalsyllableof the intonationalphrase. In (10a),whichhas a secondary stresson thepenult, the first pitch level covers cater­, while the second covers pillar.In(10b),themain stress is again on the pre­antepenult, but because the secondary stress is word­final, the first level covers the three syllables catama-.Example(10c)showsthatifthere are no secondary stresses, the second level also begins on the last syllable.

(10) a [ˈkæ.təˌpɪ.lə] ‘caterpillar’

b [ˈkæ.tə.mə.ˌræn] ‘catamaran’

c [ˈkɔː.mə.rənt] ‘cormorant’

stress inEnglishwords.Byaccountingforexceptionalityandmorpho­logical effects in terms of theoretical devices that are more restrictive, like catalexisorextrametricality,suchovergenerationcouldbeavoided.Canyouthinkofanargumenttoneutralizethisobjection?

Word stress 187

watchman’,acompoundwithasecondconstituent[ˈʋaː.kər],itdoesnot.Whatmetricalrepresentationdoesthissuggestforthefirstsyllableof[pɔr.ˈtiːr]?

2 Dutch [ɦ] ispronounced in thewords in (1),but,despite the spelling,thereisno[ɦ]inthewordsin(2).Whatdeterminesitsdistribution?

(1) [ˈɦut] hoed ‘hat’[ˈaː.bra.ˌɦɑm] Abraham ‘Abraham’[jo.ˈɦɑ.nəs] Johannes ‘John’[ɦo.ˈtɛl] hotel ‘hotel’

(2) [ˈni.ə] Niehe (propername)[a.ˈloːa] aloha ‘aloha’[ˈtaː.u] tahu ‘tofu’

Q115 Yabem, a tone language, has stress on the last syllable, as shown by the data below (Ross 1993).

a wá ‘mango’ k pá.ˈlíŋ ‘careless’b wà ‘crocodile’ l bà.ˈlìŋ ‘faraway’c ó.ˈlí ‘body’ m sá.ˈkíŋ ‘service’d ò.ˈlì ‘wages’ n sà.ˈgìŋ ‘housepartition’e á.ˈwé ‘outside’ o ká.ká.ˈtɔŋ ‘I made a heap’f à.ˈwè ‘woman’ p ká.lé.ˈtí ‘Iran’g já.ˈó ‘prohibition’ q ká.kí.ˈlí ‘Iprovokedtrouble’h jà.ˈò ‘enmity’ r ká.lè.ˈsù ‘I poked’i kúŋ ‘callout’ s ká.gè.ˈlì ‘I stepped over s.t.’j gùŋ ‘spears.t.’ t ká.mà.ˈdòm ‘Ibrokeintwo’

1 DeterminetheYabemfootintermsoffoottypeandwordedge.2 Tonesdonotfreelyvaryfromsyllabletosyllable.Whatconstrainstheir

freedom?3 Isvoicinginplosivescontrastive?4 Whatisthe1sgmorpheme?5 WhataretheYabemtranslationsof‘Icalledout’and‘Ispearedsomething’?

11.4 STRESS AND OPTIMALITY THEORY

MorerecentdescriptionsofstresssystemsarefrequentlypresentedwithinOptimal­ityTheory.Recallthatthistheoreticalprogrammeiscommittedtoisolatingstruc­tural properties as single constraints. In the case of stress, this commitment may not

Word stress188

seem to sit well with the observation that structural properties frequently correlate, as illustratedbytheIambic/TrochaicLaw,forinstance.Hayes’sIambin(8)combinesadominance property and the property of being quantity­sensitive, but a fine­grained approach would encode dominance and quantity­sensitivity in separate constraints. In this section, we will lean towards fine­grained versions of the various constraints, asthiswillshowbesthowthisworks.AnotherfeatureofOptimalityTheorywhichwill be relevant to stress is its unselective definition of the input. It should readily be capable of dealing with any phonological structure, however overspecified or under­specified.ThisfeatureisknownasRichness-of-the-Base,oftenabbreviatedasROTB,and implies that any input will end up as some well­formed output. This makes the lexical specification of metrical structure for words with exceptional stress unprob­lematicinprinciple,andthusrenderstheinclusionofdiacriticfeatureslike[+cyclic]lessdesirable.AthirdaspectofOptimalityTheorywhichwillbebroughtoutinthediscussion is its working hypothesis that constraints are grounded in phonetics, a goalthatwasearlierexpressedwithrespecttodistinctivefeaturesinPostal’s(1968)NaturalnessCondition and, perhaps less successfully, with respect to phonologi­calrules(‘processes’)inStampe(1980).

11.4.1 Foot shapes: dominance, quantity-sensitivity and degenerate feet

Dominanceistakencareofbyrankingtheconstraints(11a)and(11b).Aswillbeclear,right-dominanceresultsif(11a)outranks(11b),left-dominanceiftherank­ing is the reverse.

(11) a iaMb:Feetareiambic. b trochee:Feetaretrochaic.

Thefootconstraints in(11)couldbemademoreelaborate,soastoreflectthestatisticaltendenciesexpressedin(8),asintheUnevenIamb(Kager1999:151),whichissatisfiedby(ta.ˈtaa),butnotby(ta.ˈta)or(ˈta),whichfailtohaveabimo­raic strongposition,andnotby (ˈtaa),which lacks the initial lightposition.For(11),weassumethefine-grainedinterpretationthattheyjustbanfeetshowingthereversedominance.Thatis,(11a)isviolatedby(ˈta.ta),(ˈta.taa),(ˈtaa.ta)and(ˈtaa.taa),butsatisfiedby(ˈta),(ˈtaa),(ta.ˈta),(taa.ˈta),(ta.ˈtaa)and(taa.ˈtaa)aswellasbytheabsenceofafoot.Whereasthedirectionofdominanceistakencareofbytherelativerankingoftheconstraintsin(11),footconstructionandbinaryrhythmaretakencareofbyconstraints(12)and(13).

(12) parse-σ:Asyllablemustbeparsedintoafoot.(13) foot binarity(ftbin):Afootmustbebinary(eithertwosyllablesortwomoras).

These two constraints will lead to an alternation between weaker and stronger articulationsofsyllables,asillustratedin(14)formarmolada ‘marmelade nom; sg’ inthequantity-insensitivetrochaiclanguagePolish(DogilandWilliams1999).Can­didates(c)and(d)areruledoutbecausenotallsyllableshavebeenparsedintoafoot.Candidates(b)and(e)areruledoutbecausetheycontainfeetthatarenotbinary.

Word stress 189

(14) mar.mo.la.da Parse-σ FtBin

 a (ˈmar.mo)(ˈla.da) b (ˈmar.mo.la.da) *! c (ˈmar.mo)la.da *!* d mar.mo(ˈla.da) *!* e (ˈmar.mo.la)(ˈda) *!*

Quantity-sensitive languages require the presence of stress either on long vow-els or on both long vowels and closed syllables. This property is enforced by (15), which disallows a bimoraic syllable in the weak position of a foot.

(15) Weight-to-stress PrinciPle (WsP): Bimoraicsyllablesarestressed.

The inclusion of closed syllables among the bimoraic syllables of a language is effected by constraint (16). The term ‘Weight-by-Position’ refers to a consonant’s entitlement to moraic status on the grounds of its position in the coda as opposed to the onset (Hayes 1989a).

(16) Weight-By-Position (coda-μ):Acodaconsonantismoraic.

The ranking of Parse-σ relative to WSP and Coda-μ determines whether a lan-guage is quantity-sensitive. Closed syllables in Polish do not attract stress, as shown by the fact that disyllabic words with final open syllables, like cawa [ˈka.va] ‘coffee’, have the same position of main stress as disyllabic words with final closed syllables, like Przemek [ˈpʃɛ.mɛk] ‘boy’s name’. In Spanish, prosodically similar words, like casa [ˈka.sa] ‘house’ and papel [paˈpɛl] ‘paper’, have different stress locations, due to the language’s sensitivity to the bimoricity of closed syllables. The difference is captured by ranking Parse-σ below WSP and Coda-μ for Spanish, as in (17), but above these two constraints for Polish, as in (18).

(17) pa.pɛl WSP Coda-μ Parse-σ a   μ    μμ

pa.(ˈpɛl)*

b    μ   μμ(ˈpa.pɛl)

*!

c μ    μ(ˈpa.pɛl)

*!

(18) pʃɛ.mɛk Parse-σ WSP coda-μ a   μ    μμ

pʃɛ.(ˈmɛk)*!

b     μ  μμ(ˈpʃɛ.mɛk)

*!

 c     μ  μ(ˈpʃɛ.mɛk)

*

Thus, to characterize a quantity-insensitive language like Polish or Icelandic, Parse-σ must rank above WSP and Coda-μ. The ranking of the latter two may seem

Word stress190

immaterial,ascanbeestablishedbyreversingtherankingofWSPandCoda-μ in (18).However,withahigher-rankingParse­σ, therankingWSP» Coda-μchar­acterizes closed syllables as monomoraic rather than bimoraic, and if we wish to suppressunmotivatedbimoricity for closed syllables, as indeedwe should,WSPmust outrank Coda-μ.Thissamerankingequallycharacterizes iambicquantity-insensitive languages, like Shiwilu (Valenzuela and Gussenhoven 2013). Again,Parse­σ » WSPensuresthatadisyllabicfootisformed,ontheleftedgeasithap­pens,whileWSP» Coda-μensuresthatclosedsyllablesaremonomoraic,asshownin(19)forthewordfor‘maniocstick’,wherehigh-rankingIambisassumed.Sincethelanguagehasnoquantitycontrast,anassumptionofbimoricityforShiwiluisnot supported by either segmental length or stress.

(19) kɘn.ɲa.la parse-σ WSP Coda-μ a μ μ μ

(kɘn.ˈɲa).la* *

b μμ μ μ(ˈkɘn).ɲa.la

**!

c μμ μ μ(kɘn.ˈɲa).la

* *!

The constraint Parse­σ interacts not only withWSP (15) and Coda-μ (16) toexpress whether a language is quantity­sensitive or quantity­insensitive, but also with FtBin(13)soastoaccountforalanguage’srejection(Spanish,Shiwilu)oracceptance(Icelandic)ofdegeneratefeet.Topreventthefinalsyllableof[kǝn.ˈɲa.la]fromform­ing a monomoraic foot, FtBinmustoutrankParse-σ.Theeffectisshownin(20).

(20) kǝn.ɲa.la ftbin parse-σ  a μ μ μ

(kǝn.ˈɲa).la*

b μ μ μ(kǝn.ˈɲa).(ˈla)

*!

The reverse ranking is required for rightward trochaic Icelandic as well as for leftward iambic Weri, both of which will parse left-over syllables as degener­atefeet(BoxwellandBoxwell1966).ThisisillustratedfortheWeriword‘hairofarm’in(21).InIcelandic,thesamerankingwillreject[(ˈseptem)ber]infavourof[(ˈseptem)(ber)],assuminghigh-rankingTrochee.

(21) ku.li.pu parse-σ ftbin

 a μ μ μ(ˈku).(li.ˈpu)

*

b μ μ μku.(li.ˈpu)

*!

Quantity­sensitive languages must rank Parse­σ below WSP and Coda-μ.Observe,contraryto(20a)forShiwilu,howaninitialdisyllabiciambfortheHopiword for ‘roofbeam’ (6a) is rejected because it fails to obeyWSP for the initial

Word stress 191

stressedsyllable,asshownin(22).Thesameresultisobtainedforlanguageswith­outclosedsyllablesbutwithaquantitycontrastforvowels,likeHawaiian,whichisquantity-sensitiveforlongvowelsanddiphthongs,asin(23)forthegeographicalname Ni’ihau.

(22) les.ta.vi WSP coda-μ parse-σ a μ μ μ

(les.ˈta).vi*! *

b μμ μ μ(les.ˈta).vi

*! *

c μμ μ μ(ˈles).ta.vi

**

(23) ni.ʔi.hau WSP coda-μ parse-σ a μ μ μμ

ni.(ˈʔi.hau)*!

b μ μ μμni.ʔi.(ˈhau)

*

c μ μ μμ(ˈni.ʔi).hau

*!

The ranking of WSP above Parse­σ thus describes the quantity-sensitivityof long vowels and diphthongs. The relative ranking of Coda-μ and Parse­σisresponsiblefortreatingclosedsyllablesasheavyoraslight.In(22),thebimoraicclosed syllable results from ranking Coda-μaboveParse­σ.Letusnextconsiderhow the fine­grained approach encoding dominance and quantity­sensitivity in separate constraints allows us to describe quantity­sensitive languages which treatclosedsyllablesaslight,likeLenakel.Ithasaright-edgequantity-sensitivetrochee and rhythmic stress from the right, as illustrated in (24), fromHayes(1995).As illustrated in (25), this requires thatWSPdominatesParse­σ, nec­essary for stressing a final syllable containing a long vowel, but where Parse­σitself dominates Coda-μ,inordertopreventstressingafinalclosedsyllable,asillustratedin(26).

(24) a ˈna.puk ‘song’ b tɨ.ˈkom.kom ‘branches’ c tuˌpwa.lu.ˈka.luk ‘lungs’ d ˌrɨ.me.ˈtjaːw ‘hearrived’ e ˌtɨ.na.ˌkam.wap.ˈkeːn ‘youtwowillbejealous’

(25) rɨ.me.tjaːw WSP parse-σ coda-μ a μ μ μμ

(ˈrɨ.me).(ˈtjaːw) b μ μ μμ

rɨ.(ˈme.tjaːw)*! *

c μ μ μμ(ˈrɨ.me).tjaːw

*! *

d μ μ μμrɨ.me.(ˈtjaːw)

*!*

Word stress192

(26) na.puk WSP parse-σ coda-μ a μ μ

(ˈna.puk)*

b μ μna.(ˈpuk)

*! *

c μ μμ(ˈna.puk)

*!

d μ μμna.(ˈpuk)

*!

Q116

1 Therearenoquantity-sensitivelanguagesinwhichlongvowelsarelightandclosedsyllablesheavy.Howdoesthatfollowfromtheanalysis?

2 Bengalihasquantity-insensitiveword-initialstress,withbinaryrhyth­micstressoneveryoddsyllableintheword,rejectingdegeneratefeet(see section 11.2.3). It also has geminates, as in [ˈta.lob.bo] ‘palatal’.Assuminghigh-rankingTrocheeand a moraic representation of gemi­nates,canthislanguagebedescribedbyarankingofWSP,FtBin and Parse­σ?

11.4.2 Edges and iteration

McCarthyandPrince(1993,1998)capturetheleft-to-rightorright-to-leftparsingof feet by appealing to alignment, a device to locate phonological or morphological constituents in thephonological structure.Recall fromchapter4howtheposi­tionofanaffixisdeterminedbyaconstraintwhichalignsoneof itsedgeswiththecorrectedgeoftheword,leftforprefixesandrightforsuffixes.Generally,analignmentconstraint stipulates that therightor leftedgeofaconstituent (root,stem,morphologicalword,mora, syllable, foot,phonologicalword (ω), intona­tionalphrase,etc.)mustcoincidewiththerightorleftedgeofsomeotherconstitu­ent.Alignmentconstraint(27a)requiresthattherightedgesofanyphonologicalwordandafootshouldcoincide,while(27b)doesthesameforthebeginningofthephonologicalword.If(27a)outranks(27b),atrisyllabicwordwillbeparsedasσ(σσ),whiletheoppositerankingleadsto(σσ)σ.Aswillbeclear,insteadofω, languages may also position feet relative to morphological constituents, like the morphological word.

(27) a align(ω,Foot,R):Therightedgeofaphonologicalwordisalignedwiththerightedgeofafoot.

b align(ω,Foot,L):Theleftedgeofaphonologicalwordisalignedwiththeleftedgeofafoot.

Word stress 193

Hopi,which aswe have seen has an iamb at the leftword edge, is describedby ranking the two constraint pairs (11) and (27) as Iamb » Trochee and Align(ω,Foot,L) » Align(ω,Foot,R). Since the iamb is quantity-sensitive, bothWSPandCoda-μmustdominateParse­σ.Violationsof(27)areindicatedbythenumber of syllables each foot is removed from the relevant word edge, the latter specifiedby#,asshownin(28).

(28) les.ta.vi iaMb Al(ω,Ft,L) Wsp coda-μ parse-σ trochee Al(ω,Ft,R) a μμ μ μ

(ˈles).ta.vi** σσ#

b μμ μ   μ(ˈles.ta).vi

*! * σ#

c μμ μ μles.(ta.ˈvi)

#σ*! *! * *

d μμ μ μ(les.ˈta).vi

*! * * σ#

e μ μ μ(les.ˈta).vi

*! * * σ#

Hopidoesn’thaveiterativestress,butmanylanguagesofcoursedo.Iterativefootingcan be enforced by demanding that feet align with a word edge, as opposed to demand­ingthatawordedgealignwithafootedge.Thisconstraintpairisgivenin(29).

(29) a align(Foot,ω,R): The right edge of every foot is aligned with the right edge of thephonologicalword.

b align(Foot,ω,L):Theleftedgeofeveryfootisalignedwiththeleftedgeofthephonologicalword.

Thenumberofviolationsof(29)isequaltothesumofthesyllablesthateachfootis removed fromthe relevantwordedge.Tominimize thoseviolations,Parse­σ(12),whichgenerallypromotesexhaustive footing,willhave toberankedbelow(29),as itmustbe innon-iterativeHopi.Thereverserankinggive iterative foot­ing, as tableau (30) illustrates for the Garawaword for ‘wrist’, after Kenstowicz(1994b).Recallfromsection11.3.1thatGarawaisquantity-insensitive,istrochaic,has rhythmic stress from the right and disallows degenerate feet. Parse­σmustberankedbelowFtBintopreventdegeneratefeet,butabove(29)toenforcerhythmicstress.Thus,candidates(30a,b)eachincurasingleviolationofParse­σ,butonly(30a)correctlyalignsits leftwordboundarywithafoot.Candidate(30c)failstohaveanyrhythmicfooting,while(30d)istoogenerousinitsparsing,incurringaviolation of FtBin.

(30) ka.ma.la.rin.ɟi ftbin parse-σ al(ω,Ft,l) al(Ft,ω,r) al(ω,Ft,r) al(Ft,ω,l) a (ˈka.ma).la.(ˌrin.ɟi) * σσσ# #σσσ b ka.(ˈma.la).(ˌrin.ɟi) * #σ! σσ# #σ,#σσσ c (ˈka.ma).la.rin.ɟi **!* σσσ# σσσ# d (ˈka).(ˌma.la).(ˌrin.ɟi) *! σσ#,σσσσ# #σ,#σσσ

Word stress194

Q118 Cavineña (Key 1968) has quantity-insensitive stress with the main stress on the right.

1 Isthelanguagetrochaicoriambic?2 Draw a tableau with the four constraints Parse­σ, FtBin, Al(Ft,ω,L)

andAl(Ft,ω,R).Yourtableaushouldhavedottedlinesbetweencolumnsrepresenting unranked constraints. Illustrate your analysis with the word [a.ta.ta.wa.ha].

a ˈe.na ‘water’b ki.ˈri.ka ‘paper, book’c ˌa.si.ˈka.da ‘dirty’d a.ˌta.ta.ˈwa.ha ‘kind of bee’e ˌa.ta.ˌta.wa.ˈha.ke ‘kindofbee+poss’

Q117

1 AllsyllablesoftheGarawaword[nim.pa.la.ŋin.mu.ku.nan.ɟi.mi.ra]‘fromourowntwo’areparsedintofeet.Why?

2 Using a tableauwith the constraints as ranked in (30), show the fatesof the following output forms for the Garawa word for ‘fought withboomerangs’:a (ˈŋan.ki)(ˌri.ki)(ˌrim.pa).ɟib (ˈŋan.ki)(ˌri.ki).rim.(ˌpa.ɟi)c (ˈŋan.ki).ri.(ˌki.rim).(ˌpa.ɟi)

In(30),rhythmicstress,duetoAl(Ft,ω,R), represents the secondary stress, with the main stress being due to Al(ω,Ft,L).The implication that rhythmic stress is secondary stress when the main stress and rhythmic stress refer to different edges is notenshrinedinOptimalityTheory.Athirdpairofalignmentconstraintsspecifiestheedgewherethemainstressisfound.In(30),theunderstandingisthat(31b)outranks(31a).

(31) a align(Head,ω,R):Thefootwithmainstressisalignedwiththerightedgeofthephonologicalword.

b align(Head,ω,L):Thefootwithmainstressisalignedwiththeleftedgeofthephonologicalword.

Word stress 195

Q119 Creek has the main stress on the right, preceded by rhythmic stress (Haas 1977; Hayes 1995).

1 IdentifytheformswhichrevealwhetherCreekisiambicortrochaic.2 DrawatableauwiththefiveconstraintsParse­σ,Al(Ft,ω,R),Al(Ft,ω,L),

WSP and Coda-μ, using item (f) with candidates [(ˌalμ).(pa.ˈto).ci](wheresuperscriptμindicatesamoraiccoda),[(ˌal).(pa.ˈto).ci],[(ˌalμ).pa.(to.ˈci)],[(alμ.ˌpa).(to.ˈci)]and[(ˌalμ).pa.(ˈto).ci]asanillustration.

a i.ˈfa ‘dog’b i.ˈfo.ci ‘puppy’c po.ˈcos.wa ‘axe’d a.ˌpa.ta.ˈka ‘pancake’e ˌak.to.ˈpa ‘bridge’f ˌal.pa.ˈto.ci ‘baby alligator’g ˌnaf.ki.ˌti.ka.ˌiː.ji.ˈta ‘hit­pl;obj’

11.4.3 Geminates and weight

Themoraicrepresentationofgeminates(seechapter9andQ116(2))predictsthatquantity­sensitive languages with geminates in postvocalic position, as in Italian [ˈfat.to]‘done’,haveheavyclosedsyllables,inadditiontoheavylongvowelsiftheyhaveavowelquantitycontrast.Somewhatsurprisingly,thisdoesnotappeartobetrueforagreatmanysuchlanguages(Davis2011).Thereareanumberofresponsesto this state of affairs.We could liberalize the representationof geminates, suchthat those making heavy syllables have a moraic status, while those that don’t have a direct double association with two syllable nodes, such that geminates bypass the moralevelinbothsyllables.Anotheroptionistomaketheweightofamoradepen­denton its segmental features.This solutionwouldbeparallel to that forTBUs,which in languages like Japanese are tone­bearing only if the segment they are asso­ciatedwithis[+son],thusexcludingobstruents.

Q120 Wolof has initial stress, but peninitial stress if the peninitial syl-lable has a long vowel and the initial syllable does not, as shown in items (a) to (i) (Bell 2003).

a ˈwo.ne.wu ‘to show off ’b ˈbaa.si ‘couscous’c ˈfee.sal ‘to blacken’d ˈpe.tax ‘pigeon’e ˈɟar.goɲ ‘spider’

Word stress196

11.4.4 Non-finality and clashes

Insection10.7.2,wesawthattonaltargetstendtoberetractedintheproximityofupcomingpitchaccentsorphraseboundaries.Stresssystemsrevealphonologicalreflexesofthisbehaviourinatendencytoavoidword-finalstressaswellasadjacentstresses. The corresponding constraints are NonFinH and NoClash.

The insertion of a pitch accent is more problematic in a word­final syllable than in other syllables. Phrase­final pronunciations, which include citation pronuncia­tions, will have only one syllable available for the pitch accent plus any final bound­ary tones. There are two tendencies in languages which offer relief to such tonal crowdingonfinalsyllables.Oneisthetendencyforword-finalsyllablestobelonger(Zhang2002).English, for instance, liberally allows trimoraic syllables inword-final position (e.g. [ˈkraʊn] crown), butmuch less so inword-internal position(asin[ˈskaʊn.drəl] scoundrel),andsuchrestrictionsmaybeabsoluteinotherlan­guages. The second is for languages to avoid main stress on the word­final syllable. The earlier introduction of extrametricality as a descriptive device thus receives a straightforward motivation. Iambic languages frequently have initial stress in disyl­lables,wherefinalstresswouldbeexpected,asinHopiandShiwilu,asshownforthelatterin(32).

(32) a mi.ˈka.ra.waʔ ‘turkey’ b ˈsi.sɘk ‘porcupine’ c ˈi.kɘr ‘hurt’ d ˈmu.tuʔ ‘head’ e ˈlɘk ‘1sg’ f i.ˈkɘr.mu.tuɁ.lɘk ‘Ihaveaheadache’

Theconstraintwhichcanpreventfinalstressis(33).

(33) non-finality head(nonfinh): Afootheadmaynotbefinal.

As a quantity-insensitive left-edge iambic language, Shiwilu has high-rankingAl(ω,Ft,L)andIamb,soastocreateregularpeninitialstress(34),butbyrankingNonFinH above Al(ω,Ft,R)andParse-σ, initial stress results in disyllables, as in (35),justasitdoesinmonosyllables,thankstoAl(ω,Ft,L),asin(36).Becausethere

f ˈmii.nɛɛl ‘habit’g ga.ˈnaar ‘chicken’h wo.ˈjaa.na.ti ‘tobegoncemore’i sɛj.ˈlɛɛn ‘get married imp’

1 IsWolofquantity-sensitive?Motivateyouranswer.2 IsWoloftrochaicoriambic?3 Thewordfor‘understanding’is[deg.ˈgooμμ].Isthisunexpected?

Word stress 197

is no rhythmic stress, Align(Ft,ω,R) outranks Parse-σ. Form (34c), incidentally, would also be eliminated by FtBin, provided it ranks below NonFinH.

(34) i.kɘr.mu.tuɁ.lɘk Al(ω,Ft,L) IAmb NoNFINH Al(Ft,ω,R) PArse-σ a (i.ˈkɘr).mu.tuɁ.lɘk σσσ# *** b (i.ˈkɘr).(mu.ˈtuɁ).lɘk σ#, σσσ#! * c ˈ(i).kɘr.mu.tuɁ.lɘk σσσσ#! ****

(35) si.sɘk Al(ω,Ft,L) IAmb NoNFINH Al(Ft,ω,R) PArse-σ a (ˈsi).sɘk σ# * b (si.ˈsɘk) *! c (ˈsi. sɘk) *!

(36) lɘk Al(ω,Ft,L) IAmb NoNFINH Al(Ft,ω,R) PArse-σ a (ˈlɘk) * b lɘk *! * *

Although NonFinH, formulated as in (33), directly reflects the phonetic moti-vation behind it, a constraint banning word-final feet altogether is required for a language like Macedonian, where the main stress invariably is on the antepenulti-mate syllable (‘type 4 dialects’; cf. Baerman 1999), as illustrated in (38). By ranking NonFinF high, the final two syllables are unavailable for stress.

(37) NoN-FINAlIty Foot (NoNFINF): A foot may not be final.(38) ˈdo.ve.dam ‘I bring’ do.ˈve.de.te ‘you bring’ do.ˈve.dox.te ‘you brought’ ˈdo.ve.dam ‘I bring’ do.ˈve.de.te ‘you;Pl bring’ do.ˈve.dox.te ‘you;Pl have brought’

Q121 Cavineña belongs with Chama and Tacana to the Tacanan lan-guage family. Cavineña does not have monosyllabic content words. Chama/Tacana words like di ‘mosquito’ and do ‘howler monkey’ are [(ˈdiɁi)] and [(ˈdoɁo)] in Cavineña. Historically, this differ-ence could have arisen in either of two ways. Proto-Tacanan may have been like Cavineña and word-final vowel deletion occurred in Chama/Tacana. Alternatively, Proto-Tacanan was like Chama/Tacana and a vowel was inserted in Cavineña. Which scenario is the more plausible, considering the data below?

Cavineña Tacana Chama ˈbi.na ˈbi.na ˈbi.ña ‘bat’ a.ˈkwi.ha a.ˈki.da a.ˈkwi.sa ‘thorn’ ˌe.do.ˈko.ho ˌe.do.ˈme.so ˌe.do.ˈxo.ho ‘inside’

Word stress198

Dutchisaquantity-sensitivelanguagewhichcategoricallyrejectsadjacentstressedsyllables.Disyllabicwordswithmainstressonthefinalsyllablehaveastrayinitialsyllable,asin[riˈviːr]‘river’,[kɑnˈtoːr]‘office’,whosedurationandvoicequalityaresignificantlyreduced.Disyllableswithmainstressonthefirstsyllableincludethefinalsyllableastheweakbranchofatrochee,asin[ˈpɪnda]‘peanut’,[ˈdɔktɔr]‘holderofdoctorate’(Gussenhoven1993,2009).Sincethebanonadjacentstressesisblindtothe segmental composition of stressless syllables, something needs to be done to cir­cumventtheeffectofWSPandCoda-μ,whichgenerallydemandthatclosedsyllablesare stressed. NoClash(39)servesthispurposebydirectlytargetingtheproblem,asshownin(41).Ittakesthelocationofthemainstressforgranted.Thefactthatinitialbut not final syllables are stray is here attributed to the trochaic nature of the language, but the difference also follows from the right­alignment of trochees.

(39) noclash:Footheadsarenotadjacent.

(40) dɔk.tɔr trochee noclash Wsp parse-σ a (ˈdɔk.tɔr) * b (ˈdɔk).tɔr * *! c (ˈdɔk)(ˌtɔr) *!

(41) kɑn.toːr trochee noclash Wsp parse-σ a kɑn.(ˈtoːr) * * b (ˌkɑn).(ˈtoːr) *! c (kɑn.ˈtoːr) *!

11.5 CONCLUSION

Stress systems represent an exciting area of research. In this chapter, we haveintroduced the concepts that have played a role in typological discussions of word prominence.Wedrewadistinctionbetweenlanguageswithandwithoutwordpro­sodic representations. In the second group, which may be small, no representation beyond a syllabified string of consonants and vowels is present in words, whether lexicallyorpostlexically(Malay).Withinthefirstgroup,wedistinguishedbetweenlanguageswithandwithoutphoneticstress.Wealsoobservedthattonesarepresentin the lexical or postlexical representation regardless of the presence of phonetic stress. In languages with phonetic stress, the distribution of tones is frequently at least in part determined by the metrical structure, for instance because stressed syl­lablesofaccentedwordshavepostlexicaltone(English,Swedish)orbecausealexi­cal tone contrast is only available on the stressed syllable, with possible restrictions onitslocation(Ma’ya,Swedish).Somelanguageswithoutphoneticstresshavetonedistributions that resemble the distributions found for stress systems, in which case thetonesmayhavebeendescribedasstress(Nubi,Persian).Tonesmaybepresenton or be assigned to words in the lexicon, in which case the language is referred to as a ‘tone language’, or postlexically. Postlexical tones are usually referred to as intonational tones.

Word stress 199

Stresswascharacterizedasapropertyofsyllables,notofmorasorfeet,andasalwayshavingadefaultlocation.Stressesareproperlytobeseenasfeet,rhythmi­cally motivated constituents potentially enclosing more than one syllable, which we assume explains the obligatory nature of stress. The implication that stress systems are governed by generalizations, either applying unexceptionally or describing the defaultstructures,hasattractedagreatdealoftheoreticalwork.Section11.3setoutthemainvariables inthosegeneralizationsonthebasisofworkbyHayes(1981,1987,1995),withampleexemplification.Sincestresshasbeenamuch-debatedaswellasrelativelysuccessfultopicinOptimalityTheory,section11.4presentedthemost general constraints which have been used in descriptions of stress systems. These descriptions had a double aim. They not only displayed a wide array of stress systemsbut also showedhowOptimalityTheoryhasbeenapplied tovariedbutcoherent sets of data.

Phonology above the word

12.1 IntroductIon

The phonological structure of languages extends beyond the syllable and the foot. There is a hierarchy of phonological constituents, so that lower, and thus typically smaller, constituents are contained within higher, and thus typically larger, ones. For instance, the foot is contained within, or dominated by, the phonological word, and phonological words are grouped into phonological phrases, etc. Instead of ‘phonological constituent’, the term ‘prosodic constituent’ is often used, particularly for the higher constituents, like the phonological phrase, the intonational phrase and the phonological utterance. The entire structure above the syllable is often referred to as the ‘Prosodic Hierarchy’. This line of research began with Selkirk (1978) and was consolidated by Nespor and Vogel (1986), Hayes (1989b) and, for the phonological word, Booij (1985), among others.

In this chapter, we discuss and illustrate the phonological utterance (U), the into-national phrase (IP), the phonological phrase (φ) and the phonological word (ω). In (1), an example of a sentence is given that has been parsed into these constituents. The next two lower constituents are the foot (chapter 11) and the syllable (chapter 9). Notice, for instance, that weak forms, the function words were and to, are not sepa-rate ωs, but are included with the following form in the same ω. In fact, they aren’t even feet, since they have no stress. Before discussing the prosodic constituents in more detail, we first deal with three general questions. First, how do prosodic con-stituents manifest themselves? Second, what is the general structure of the prosodic hierarchy? And, third, what determines the prosodic structure of specific sentences?

(1)

12.2 GeneralIzatIons InvolvInG prosodIc constItuents

Prosodic constituents manifest themselves in four ways:

1 Boundary strength. There is a greater degree of articulatory integration in lower constituents than in higher ones. For instance, within an English

12

Phonology above the word 201

syllable like [læmp] as inLampton, the articulatory gestures occur rela­tivelyquicklyaftereachother,butwithinhigher-rankingconstituentsthesequenceofmovementsin[læmp]willbelesstightlycoordinated,asinA fu[ll amp]utation or A [lamb p]assed by. That is, prosodic breaks between higher constituents are stronger than those between lower constituents. Boundary strengthwill reveal itself in a number of phoneticmeasures.The extent to which the articulation of a vowel is influenced by that of a vowel or consonant in a neighbouring syllable will depend on the strength of the boundary between the syllables containing the vowels. In general, coarticulationwillcauseaschwatobecloserin[əpiː]thanin[əpɑː],inanticipation of the tongue position of the following vowel, but this effect will be stronger if the two vowels occur within an ω,asin[ə]ppeal, than when they occur in different ωs, as in Emm[ə] Peel (cf.Cho2004).Also,the duration of the last syllable of a prosodic constituent is typically longer asthelengtheningisstronger(or‘higher’,thinkinghierarchically),whichis known as final lengthening, also known as preboundary lengthening. Equally,theinitialsegmentsofaconstituentaremoreclearlypronouncedas the boundary is stronger, which is known as initial strengthening (Keatinget al.2004).

2 Boundary tones. Higher constituents are often characterized by intona­tional boundary tones.When that happens, the boundary is particularlyeasytohear.ExamplesofthiswillbegivenwhenwediscusstheφandtheIP.

3 Postlexicalphonologicalprocesses.Prosodicconstituentsfrequentlydeter­mine the distribution of segments and the application of phonological pro­cesses,asillustratedforthesyllableandthefootinchapters9and11.Theways in which reference is made to phonological constituents have been classedintothreetypes(Selkirk1980).a Domain limit constraints.Reference ismade to the leftor right edge

of a constituent. In many languages, ωs must end in a consonant, as inTagalogforinstance,althoughsyllablesoccurfreelywithoutacodaword­internally.

b Domainspanconstraints.Thecontextandthefocusofsomegeneraliza­tionmustoccurwithasingleconstituentofsomerank.Forinstance,aswe will see below, Italian has a rule of s­voicing, whichcauses[z],never[s],toappearbetweenvowelscontainedwithinanω.

c Domain juncture constraints.The context of the generalizationmayincludetheleftandrightedgesofadjacentconstituentsofsomerank,provided this boundary occurs within some higher constituent. Forinstance, as wewill see, [s] is optionally voiced to [z] inDutch if itoccurs finally in the ω and the next ω begins with a vowel, provided this boundary falls within an IP.

4 Phonology-sensitivesyntax.Lastly,ithasbeenobservedthatsyntacticandmorphologicalrulesmaybesensitivetothesizeofconstituents(ZecandInkelas1990).EnglishhasHeavyNPShift, which allows the movement of theobjectin(3b)toclause-finalposition.However,aconditionontheruleisthattheobjectNPmustconsistofmorethanoneφ.Forthisreason,(3a)

Phonology above the word202

cannot move its NP.Bycontrast,(3b)showsthateitherorderisfineiftheobjectNPcontainsmorethanoneφ.

(2) to NPV NP(2)

(3) a Hegavethebooktoheraunt(*...toherauntthebook). b HegavetoherauntthebookaboutMozart(or:thebookaboutMozarttoheraunt).

12.3 THE STRICT LAYER HYPOTHESIS

Whatdoestheprosodichierarchylooklike?Acommonviewisthattheconstitu­ents obey the StrictLayerHypothesis, which is perfectly obeyed by representa­tion(4).

(4) ( )U( )( )IP( )( )( )( )φ( )( )( )( )( )( )ω

Therehavebeenanumberofformulationsofthisprinciple(Selkirk1984;Ladd2008:291),whichcanbegivenas(5).

(5) strict layer hypothesis (slh):a prosodicconstituentofrankn isimmediatelydominatedbyasingleconstituentofrankn + 1.

Anω,forexample,willbedominatedbyφ.Ifitisn’t,asin(6a),thestructureisnon-exhaustivelyparsed, in this case into constituents of rank2.As it happens,theseriousnessofviolationsofSLHvariesconsiderably.First,non-exhaustivepars­ingofsyllablesintofeetfrequentlyoccurswhenword-peripheralsyllablesareleftunfooted(seesection11.3.1),anuncontroversialcaseofanSLHviolation.Thus,segmentAin(6a)maybeginasyllablewithoutatthesametimebeginningafoot.Second,ifanωweretobedominatedbytwoφs,wewouldcreatetheincoherentoccurrenceofaφ-boundarywithinanω.ThistypeofviolationoftheSLH,shownfortheboundarybetweenAandBin(6b)andknownas‘improperbracketing’,isruled out completely. Third, if a constituent were to dominate a constituent of the samerank,wewouldhaverecursiveness.Astructureisrecursiveifsomeconstitu­entappearswithinaconstituentofthesameoralowerrank.In(6c),forinstance,two constituentsof rank1 are containedwithin a constituentof the same rank.There have been various claims that same­rank recursiveness in prosodic phrasing must in fact sometimes be recognized.

(6) a ( )3 b ( )3 c( )2 ( )( )2 ( )1

(A )( )1 ( )(AB)( )1 ( )(A )1

Phonology above the word 203

Asobservedinsection1.4,recursivenessinthemorphosyntacticstructureisone of the hallmarks of human language. It explains why sentence length is infi­nite, as in This is the cat that caught the rat that stole the cheese that lay on the table that . . . ,whereeveryNPexceptThis hasthestructure[. . .[. . .]S]NP, as in [the cat [the cat stole the cheese]S]NP, inwhichtheShasanNPwhichtakestheform[. . .[. . .]S]NP, and so on, ad infinitum.Similarly,sentencesmayappearaspremodifierswithinNPs, as inan I-couldn’t-care-less attitude. In phonological structure, such spectacular cases of recursiveness are unknown. It will be safe to say, for instance, that we will never come across a language which places a ω in the nucleus of a syllable.

12.4 FACTORS DETERMINING PROSODIC PHRASING

Whatdetermineswheretheseprosodicconstituentsbeginandend?Notsurpris­ingly, an important factor is the morphosyntactic structure. It would, to give an extremeexample,beunexpectedforthemainbreakin(7a)tooccurbetweento and arrive,withsmallerbreaksasindicated.Rather,wewouldexpectsomethingmorelike(7b).Infact,in(7b),thesmallerdomainsareφs,andthelargeronesIPs.Aswillbeclear,theycorrespondwithsyntacticphrases,unlikethebracketedpartsof(7a).Bysomehowmarkingoffthemeaningfulconstituentsinthepronunciation,parsingof the expression will be easier for the listener.

(7) a (Thefirst)(trainto)(arriveisthe)(onefromParis) b (Thefirsttrain)(toarrive)(istheone)(fromParis)

However, a secondary role is played by constituent length. Sincemorphosyn­tactic constituents of a given rank may vary hugely in length, a one­to­one cor­respondence between phonological and morphosyntactic constituents would put unreasonabledemandsonspeakers.Forinstance,thelargeNPaftersee in(8a)issyntactically equivalent to the word her in(8b).Itwouldbequiteastrainonthespeakertoproduceaφthatrunsallthewayfromthe old to road, while it would equally be awkward to produce two in quick succession for I can see and her. In (8a)thereistoomuchphonologicalstructureandin(8b)toolittleforacomfort­ablyrhythmicoccurrenceofprosodicbreaks.Generally,therewouldappeartobeatendencyforconstituentstoconsistoftwolowerconstituents(Selkirk2000).

(8) a I canseetheoldcustomsofficeattheendofthebendintheroad. b Icanseeher.

Notsurprisingly,aφtendstobeproducedforeachoftheNP-internalphrasesin(8a):(the old customs office), (at the end), (of the bend) and (in the road). Andin(8b)theNPher will be incorporated with the preceding see into a single ω, pronounced [siːə],torhymewithMaria. This incorporation of phonologically weak words into adjacentwords iscalledcliticization, and her here is a clitic that attaches to the host see,insteadofformingaφ,whichitwouldhavebeenentitledtoifithadbeen

Phonology above the word204

a noun. Cliticizationofphonologicallyweakwordsinsideaω with a regular word is a specific case of what is called restructuring, the incorporation of phonologically light structureswith adjacentwords into aphonological constituentof the rankwhich it would otherwise have had to itself.

In(8a), thecorrespondencebetweenthesyntacticandthephonologicalstruc­tureisstillinasenseonetoone,sinceinsidethelargeNP,the‘maximalprojection’,therearesmallerphrases,PPsandanNP,orXPsforshort,whichcorrespondtoφs.However,alsowhenthesyntaxis identical,differentprosodicstructuresmaybeimposedonthegroundsoflength.Forinstance,anIPboundaryismorelikelyafterthesubjectNPinHippopotamuses like to swim in the river than in John likes to swim in the river. Languageswillvaryintheirpreferenceforreflectingthemor­phosyntactic structure in the phonology at the expense of an even distribution of phonological constituents.

Athirdfactoristheinformation structure of the sentence. If someone answers the question When was Mozart born? by saying In January 1756, all of the infor­mationexpressedisnewtothespeakerwhoposedthequestion.However, ifthesame expression was said in response to Was Mozart born in January 1756 or in February 1756? only January would be the new information. These different focus constituents are indicated in (9a) and (9b), respectively. Languages that encodesuch differences in information structure, or focus structure, may do so in differ­entways.Frequently,thisisdonethroughcontrastsinprosodicphrasing.WewillseeanexampleofthisinthediscussionoftheφinBengali.Otherlanguages,likeEnglish,usepitchaccentsforthispurpose(section10.6).

(9) a In[January1756]FOC b In[January]FOC 1756

Alignmentconstraintswilltakecareofthecoincidenceofmorphosyntacticcon­stituentsandprosodicconstituents.Selkirk’s(2000)AlignXP,for instance, given in(10),canbeusedtodescribethephonologicalphrasingoftheBasquesentencein(11a).Itexplainswhy(11b)isungrammatical,becausetherightedgeoftheXPmeaning‘toAmaia’sgrandmother’doesnotendaφ.Thecoincidenceofprosodicconstituents and focus constituents can likewise be accounted for by means of alignment constraints.

(10) alignxp: Aligntherightedgeofanxp withtherightedgeofaφ.(11) a φ(Amaienamumari)φφ(liburua)φφ(emondotzo)φ Amaia-gengrandmother-datbookgiveaux

ShegavethebooktoAmaia’sgrandmother b *φ(Amaienamumariliburua)φφ(emondotzo)φ

AnexampleofaconstraintthatconsidersphonologicallengthisBinary(Elordi­eta1997),aspecificformofthegeneralphenomenonthatconstituentsmustn’tbetooshortortoolong.Infact,syntacticallywell-behaved(13b)isungrammatical,because(12)outranks(10).

Phonology above the word 205

(12) binary: Thefirstφofthesentencemustcontainminimallytwoωs.(13) a φ(Amaiariamumenliburua)φφ(emondotzo)φ Amaia-datgrandmother-gen bookgiveaux

Shegavegrandmother’sbooktoAmaia b *φ(Amaiari)φφ(amumenliburua)φφ(emondotzo)φ

Q122 If the syntactic structure of (13a) were to be used as a response to Did she give Joseba’s book to Amaia?, the focus constituent would be amumen, the ‘new’ information. The left edge of this kind of focus constituent must be aligned with a φ-boundary, as expressed in (1).

(1)AlignFOC:AligntheleftedgeofaFOC-constituentwiththeleftedgeofaφ.

Theprosodicstructureofthereply,intranslation‘ShegaveGRANDMOTH­ER’S booktoAmaia’,istheonegivenasungrammaticalin(13b)ofthetext.

1 Howwouldyouaccountforthefactthatthefirstφconsistsofasingleω?2 What would the prosodic structure be of the equivalent of ‘She gave

grandmother’sBOOKtoAmaia’?

Wenow turnbriefly to the fourprosodic constituents.Thesearenot theonlyprosodicconstituentsthatarediscussedintheliterature.Forinstance,an‘accen­tualphrase’andan‘intermediatephrase’areoftenreferredto,bothofwhichwouldrank above the ωandbelowtheIP.Theirpositionrelativetotheφmayvaryacrossdescriptions.

12.5 THE PROSODIC CONSTITUENTS

12.5.1 The phonological utterance

NesporandVogel(1986)illustratethedomainspaneffectoftheU,onr-linkingin the standard variety of English spoken inEngland,RP (ReceivedPronuncia­tion).Likemanyothervarieties,RPdisallowsthe[–cons]consonants[hjwr]inthecoda.Morphemesthatendinnonhighvowels([ə ɪəεə ɔː ɑː ɜː],asinvilla, idea, fair, paw, car, stir) arefollowedby[r],ifthenextmorphemebeginswithanonset­lesssyllable,asillustratedin(14).1In(14a,b),r-linkingis seen to apply within the wordandacrosswords.However,theupperlimitistheU:whileitcanapplyacross

Phonology above the word206

two sentences addressed to the same listener and not separated by a pause, it cannot apply across two sentences addressed to different listeners, even if they are spoken withoutaninterveningpause.Theexamples(14c,d)illustratethattheUisnotnec­essarily isomorphic with a single syntactic sentence, but that there is nevertheless anupperlimittowhatcanbeaccommodatedwithinthesameU.

(14) a stɜː stir U(...sti[r]ing...)U b fεə fair U(Afai[r]idea)U c ˈʃiːlə Sheila U(HiSheila![r]Everythingallright?)U d ˈpiːtə Peter U(HiPeter!)U*[r]U(Openthewindow,Sheila)U

12.5.2 The intonational phrase

TheIP,alsoabbreviatedasι,tendstocorrespondtotherootsentence,i.e.asin­gle[NPVP]structurewithoutextrapositionsor interruptions.Selkirk(1978)gives(15b), inwhichtheextraposition in Pakistan and the restrictive relative clause which is a weekday have been assigned to separate IPs, leaving the root sentence,whichwouldotherwisebeasingleIP(15b),tobedividedovertwoIPs.However,aswasthecasewithU,thedivisionofspeechintoIPsisnotpurelysyntacticallydriven.Inparticular,whenthesubjectislongerthanasinglelexicalwordtherewilltendtobeanIPboundarybetweenthesubjectNPandtheVP,asshownin(15c).

(15) a ι(InPakistan)ι ι(Tuesday)ι ι(whichisaweekday)ι ι(isaholiday)ι b ι(Tuesdayisaholiday)ι c ι(ThesecondTuesdayofeverymonth)ι ι(isaholiday)ι

The domain span effect of the IP can be illustrated with a rhythmic accentuation affectingcertainadverbialsinDutch,like[ɑltεit]altijd ‘always’, which can appear in a variety of sentential positions in the IP. They are accented on the final syllable whennootherpitchaccentedwordfollowsinthesameIP,asillustratedin(16a,b).The retraction of the accent to the first syllable occurs when they are followed by an accentedwordwithintheIP,asillustratedin(16c,d).

(16) a ι(NaardeWAterstandenluistertzealTIJD)ι tothewaterlevelreportslistensshealways ‘Thewaterlevelreportsshewillalwayslistento’ b ι(WaarzealTIJDnaarluistert)ι ι(zijndeWAterstanden)ι c ι(ALtijdluistertzenaardeWAterstanden)ι d ι(ZeluistertALtijdnaardeWAterstanden)ι

Asillustratedinchapter10,theIPisboundedbyintonationalboundarytonesinmanylanguages,adomainlimitphenomenon.Englishnon-finalIPsarefrequentlyclosedbyH%afteraH*Lpitchaccent,whichcausesthefinalsyllableofincident in (17a)tohavehighpitch.InBritishEnglish,thesameH*LH%patternmaybeusedforquestions,asshownin(17b).

Phonology above the word 207

(17)

a (In press reports of the incident) (his name wasn’t mentioned)||

%L

%L

b (Have you read the press reports of the incident?)

H*L

|H*L

L% H*L L%

H%

(17)

H%

ι

ι ι

ι ι ι

12.5.3 the phonological phrase

Although languages vary in the details of the correspondence, the φ tends to cor-respond to the syntactic phrase. Hayes (1989b) shows that the φ defines the domain of the English rhythm rule. In (18a) Chinese is an adjective inside the NP the Chinese dishes, while in (18b) the Chinese is an NP, the indirect object of gives. An adjustment of the accentuation pattern from CHINESE to CHInese under the influ-ence of the following accent on DISHes takes place in (18a), but not in (18b). Simi-larly, the German rhythm rule is sensitive to the German φ in (19a), the headless object NP das hell-blaue forms a φ by itself, while in (19b) it forms part of the NP, and hence the φ, das hell-blaue Bild.

(18) a OnTuesdays,hegivesφ(theCHIneseDISHes)φ b OnTuesdays,hegivesφ(theCHINESE)φφ(DISHes)φ(19) a Ich fand φ(dasHELL-BLAUe)φφ(SCHÖN)φ ‘Ifoundthelight-blueonebeautiful’ b Ich fand φ(dasHELL-blaueBILD)φ ‘Ifoundthelight-bluepicture’

In Bengali, the φ is phonologically marked by a final boundary tone Hφ if it con-tains an intonational pitch accent L*, as shown in (20a). Because the right-hand boundary of the φ is sensitive to the focus of the sentence in Bengali, (20a) con-trasts with (20b), whose focus is confined to the first constituent of the compound word for ‘fishhead’. Bengali also has IP-final boundary tones, like Lι. The Bengali φ equally defines the domain of the rule of r-deletion and a regressive voicing pro-cess (Hayes and Lahiri 1991; see Q48).

(20)

tumi (kon mat her-matha) (ranna-korle) )|L* H

you‘Which FISHHEAD did you cook?’

which fishhead cooked

(20)a

L

L* H

b tumi (kon mat her) (math a) (ranna-korle) )|

‘The head of which FISH did you cook?’

L

ι

ι

ι

ι

tumi (kon mat her-matha) (ranna-korle) )|L* H

you‘Which FISHHEAD did you cook?’

which fishhead cooked

(20)a

L

L* H

b tumi (kon mat her) (math a) (ranna-korle) )|

‘The head of which FISH did you cook?’

L

ι

ι

ι

ι

Phonology above the word208

Again,syntacticconstituencydoesnotprovidetheonlyrelevantinformationforderivingφ-structure.ConstituentstotherightoftheNPheadinEnglish,suchasthe PP of ancient China intheNPthe CHINESE of ANcient CHIna, form their own φ,asshownbytheabsenceofaneffectoftherhythmrule on the word Chinese. However,whenthepostposedPPdoesnotcontainalexicalhead,likethePPon him intheNPthat report on him, asingleφisformedcontainingthewholeNP.Recallfromsection12.2thatZecandInkelas(1990)pointedoutthatthesyntacticruleofheavyNPshift,whichplacesanobjectNPinsentence-finalposition,onlyappliesiftheNPiscomposedofatleasttwoφs.Thisisshownin(21),takenfromInkelas(1989).Intheungrammaticalb-example,theobjectisonlyasingleφ,thankstotheprosodically restructured PP on him.

(21) a MarygavetoSusanφ(thatreport)φφ(onDukakis)φ b *MarygavetoSusanφ(thatreportonhim)φ

Theexamplesin(22a,b)(NesporandVogel1986)illustrateanotherwayinwhichnonsyntactic information is relevant to φ-formation in (American) English, asrevealed by the behaviour of the rhythmrule.In(22a),[ˌriːprəˈduːs]reproduce has the main stress on the last syllable, which pronunciation is as expected when it isthelastwordintheφ.However,in(22b)ithasundergonetherhythmrule, a pronunciation that requires that it should be followed by another accented word in thesameφ.ThisisexplainedbytheassumptionthatiftheAdverbialPhraseconsistsof a single ω,itcanoptionallybeincludedintheφonitsleft,insteadofformingitsownφ.

(22) a Rabbitsφ(REproDUCE)φφ(QUICKlyandDILligently)φ b Rabbitsφ(REproduceQUICKly)φ

Q123 What is the φ-structure of sentences (1) and (2)? Why is (2) ungrammatical?

(1) IwasexplainingtothestudentstheproblemofthedoublenegationinMiddleEnglish.

(2)*I was explaining to the students the problem.

Q124 The English rhythm rule can apply in (1) but not in (2). How would you account for this difference?

(1)ThisisREpresentedinSIX.(i.e.‘in(6)’)(2)ThisisREpreSENtedinSIXA.(i.e.‘in(6a)’)

Phonology above the word 209

12.5.4 The phonological word

Frequently,whenaphenomenonissaidtobeword-based,itisinfactconfinedtothe domain of the ω (alsoknownasthe prosodic word).Crucially,ω does not cor­respond inaone-to-one fashion to themorphologicalword.For instance,whilecompoundsrepresent‘words’inthesenseofmorphologicalcategorieslikeNoun,VerborAdjective,inmanylanguageseachoftheconstituentpartsformsaphono­logicaldomainfor(consonantorvowel)harmony,wordstressandsyllabification.Thus, vowelharmonyinTurkish(cf.section6.4.2)isconfinedtotheconstituentsofthecompound,asshownin(23),wherethevowels inthefirstconstituentare[–back]andthevowelsinthesecondareall[+back].InGerman,theMaximumOnsetPrinciple(section9.2)doesnotapplyacrossthe internalboundaryinacompound,asshownin(24).InGreek,vowelhiatusbetweenthemembersofthecompound is tolerated, but not within non­compound words, while each of the constituentshasitsownwordstress,exactlyasiftheyformedanNPlike[meˈɣalia erɣaˈsia]‘bigworks’,asshownin(25)(NesporandVogel1986;Nespor1998).

(23) a køpekbalɯgɯ(*køpekbeligi) ‘shark(lit.dogfish)’ b eljazɯsɯ(*eljezisi) ‘handwriting’(24) a mʊnt.aːrt(*mʊn.taːrt) ‘dialect(lit.mouthtype)’ b ais.lœfl(*ai.slœfl) ‘icespoon’(25) a oˈmaðaerɣaˈsias(*omaðerɣaˈsias) ‘workteam’ b ˈzoniasfaˈlias(*zonasfaˈlias) ‘safetyzone’

Italian s­voicing, shownin(26),providesanexampleofanω-domain span rule. Itvoicesnongeminate[s]betweenvowels,asillustratedin(27)(NesporandVogel1986).Theruleappliesinasimplexwordin(27a)andinasuffixedwordin(27b),butdoesnotapplyacrossaprefixanditsbase(27c)oracombinationofastemandaword(27d),eventhoughallfouritemsin(27)aresinglemorphologicalwords.The generalization that brings this disparate group of morphosyntactic constituents under one heading is the ω.

(26) italians-voicings→z/

V C V|

(. . .__. . .)ω ω

(27) a ˈkaza ‘house’ b kaˈzina ‘house+DIM’ c asoˈtʃale *azoˈtʃale ‘a-social’ d ˈfilosoˈvjetiko *ˈfilozoˈvjetiko ‘pro-Soviet’

AconstituencyeffectoftheDutchω wasnotedbyBooij(1985).Dutchcoordi­natedNPsallowdeletionofidenticalpartsinthecoordinatedconstituents.Sche­matically, the situation can be represented as AB/ and CB, where the slash marks thedeleteditem.Crucially,thedeletedportionintheleft-handcoordinateisnotamorphological(orsyntactic)constituent.Thisisillustratedin(28).ThedeletedBcorrespondstotheheadnounofanNPin(28a),toanounwithinacompound

Phonology above the word210

nounin(28b),toaverbalstemin(28c)andtoanadjectivalsuffixin(28d).Whilethe deleted portions are quite heterogeneous when viewed from a morphosyntactic perspective, their common phonological characteristic is that they form separate syllabification domains, i.e. ωs.Dutchdoesnot generally syllabify acrosswords,whileallprefixesandcertainsuffixeslike-schap and -achtig do not syllabify together with the base they are attached to, forming separate ωs.

(28) a [[grote]Adj[maten]N]NP en[[kleine]Adj [matenN]NP ω(grote)ω ω(en)ω ω(kleine)ω ω(maten)ω ‘small(sizes)andlargesizes’ b [[minimum]N [maten]N]N en[[maximum]N [maten]N]N

ω(minimum)ω ω(en)ω ω(maximum)ω ω(maten)ω ‘minimum(sizes)andmaximumsizes’ c [in[voer]N]N en[uit[voer]N]N

ω(in)ω ω(en)ω ω(uit)ω(voer)ω ‘im(port)andexport’ d [[zwanger]Adj schap]N en[[moeder]N schap]N

ω(zwanger)ωω(en)ω ω(moeder)ω ω(schap)ω ‘pregnant(hood)andmotherhood’

The assumption that the deleted portion should minimally be an ω is supported bytheimpossibilityofdeletingsuffixesthatdosyllabifywiththebase.Dutchhastwoadjective-formingsuffixesmeaning‘like’.Thesuffix-achtig [ɑχtəχ]islike-schap in(28d),andformsitsownω,butthesuffix-ig [əχ]isincorporatedintotheω of its base.Accordingly,deletionof -achtig is possiblein(29a),whilein(29b)nodeletionis possible.

(29) a [[paars]Adj achtig]Adj en[[groen]Adj achtig]Adj

ω(paars)ω ω(en)ω ω(groen)ω ω(achtig)ω ‘purple(-like)andgreen-like’ b [[paars]Adj ig]Adj en[[groen]Adj ig]Adj

ω(paarsig)ωω(en)ωω(groenig)ω ‘purple-likeandgreen-like’

Q125 In Dutch, prevocalic [s] is often voiced to [z] after voiced segments, as shown in (1). However, no voicing is possible in the examples in (2). What determines when [s] may be voiced?

(1) [[ˈhœys]N [ɑrts]N]N [z] ‘familydoctor’ [[ʋɑs]Aux [ˈaːrdəχ]Adj]VP [z] ‘wasfriendly’ [[ˈmεns]N [aːp]N]N [z] ‘orang-utan’ [ˈmɪs[oːχst]N]N [z] ‘failedharvest’ [[ˈεis]N ɑχtəχ]Adj [z] ‘ice-like’(2) [ˈhεis]V ən]V *[z] ‘hoist+INF’ [ˈkɑns]N ən]N *[z] ‘chance+PL’ [ˈmɑsaː]N *[z] ‘mass’

Phonology above the word 211

12.6 DERIVING PROSODIC CONSTITUENTS

Withtheω we have come to the lowest prosodic constituent which can somehow berelatedtomorphosyntacticconstituency.Whattheω and higher constituents have in common is that at least part of their formation is dependent on the mor­phosyntactic structure of the language. The question of how the relation between the two kinds of constituency is to be expressed has received different answers in the literature. InNesporandVogel (1986), the relation isbasedonavarietyofmorphosyntactic properties.A typical statement of such a relationmight be‘IncludetheheadofthesyntacticconstituentS,togetherwithalltheprosodiccon­stituentsofrankConitsnon-recursiveside,inProsodicConstituentofrankC+1.’ (The non-recursive side is the left side in right-branching structures, and viceversa.)Selkirk(1986),withreferencetoaproposalinanearlierversionofChen(1987),suggestedthattheunifyingelementintherelationbetweenprosodicandmorphosyntactic constituency is reference to edges. This approach lay at the basis of the development of alignment constraints inOptimalityTheory (chapter 4).We illustrateedgealignmentwith theDutchω. It can be derived with the help ofmorphologicalinformationonly(cf.Booij1977:103;vanderHulst1984:85).AsisthecaseinItalian,suffixesaresyllabifiedwiththeirbase,butprefixesneverare. This suggests that the ω co­begins with the beginning of the morphological category‘word’,i.e.anystemorderivedword,asexpressedin(30).Thisexcludessuffixesand,aswewillseeinsection12.6.1,certainfunctionwords,likepronounsandprepositions.Constraint(30)correctlypredictsthatprefixesandconstituentsof the compound form individual ωs.

(30) align(morphologicalword,ω, Left):Theleftedgeofamorphologicalwordalignswiththeleftedgeofω.

In(31a),theprefixbeginsalexicalcategory(thecomplexword),andsodoesthebase.(Theendofanon-finalω isofcoursedefinedbythebeginningofthenext.)Similarly,(31b)illustrateshowseparatedomainsarecreatedfortheconstituentsofcompounds: each of them begins a lexical category, while the first, additionally but redundantly,beginsthecompound,anotherlexicalcategory.In(31c),suffixesareincluded in the ω ontheleft,becausesuffixesdonotbeginlexicalstems.

(31) Morphology ω-structure Syllabification a [ɔnt[εiχən]V]V ω(ɔnt)ω ω(εiχən)ω ɔnt.εi.χən ‘dispossess’ [ɔn[eːvən]Adj]Adj ω(ɔn)ω ω(eːvən)ω ɔn.eː.vən ‘uneven’ b [[rεin]N [aːk]N ]N ω(rεin)ω ω(aːk)ω rεin.aːk ‘Rhinebarge’ [[kεrk]N [œyl]N ]N ω(kεrk)ω ω(œyl)ω kεrk.œyl ‘barnowl’ c [[teːkən]V ɪŋ]N ω(teːkənɪŋ)ω teː.kə.nɪŋ ‘drawing’ [[ʋɑndəl]V aːr]N ω(ʋɑndəlaːr)ω ʋɑn.də.laːr ‘walker’

AspointedoutbyInkelas(1989)andBooij(1996),prosodicstructure,likeanyother aspect of the phonological representation of words or morphemes, can be includedinunderlyingrepresentations.Aswesawin(28d)and(29)above,many

Phonology above the word212

Dutchfull-vowelledsuffixesarenotincludedintheωontheirleft,likethenomi­nalizingsuffix-schap andtheadjectivalsuffix-achtig, which do not syllabify with theirbase.Inthelexicon,thesesuffixeswillthereforebelistedasωs.

Selkirk (2011) re-evaluates the pervasive tendency for syntactically definedconstituents to correspond to prosodic constituents like ω,φand ι. Insteadofrequiringcoincidenceofeitherleftorrightedges,Match Theory claims that the correspondencebetween,say,asyntacticphraseandφisdirect,withedgesonboth sides coinciding.For somerecentworkon thisposition, seeSelkirkandLee(2015).

12.6.1 Clitics

Pronouns,auxiliaryverbs,conjunctionsandthelikecannotbegivenω status by (30)inDutch,sincetheydonotbelongtoamajorwordclass.Butsincetheyarewords,notaffixes, theycannotattachtosomeother iteminthe lexiconthat is a major-class item.Many functionwordsare in fact included inω postlexically. If (30)isalsovalidpostlexically,thepredictionisthatfunctionwordsinDutchshouldbehavelikesuffixes,i.e.beincludedintheωtotheirleft.Thisisindeedwhatwefind.In(32a)thearticle[ən]encliticizesontotheprecedingverbform[rip]‘called’.Thisexplainswhythearticlecannot,innaturalspeech,bepronounced[ʔən]inthiscontext.Thesamegoesforthepreposition[ɪn]andthedefinitearticle[ət]in(32b)(Booij1996).

(32) a [rip]V [ən]Art Hijω(ri.pən)ωkat ‘hecalledacat’ b [χaːt]V [ɪn]Prep [ət]Art Hetω(χaː.tɪ.nət)ωputje ‘itgoesintothedrain’

Otherfunctionwords,alsothosethatlackanonset,haveafullvowel,liketheconjunctions[ɔf]‘or’and[εn]‘and’.Thesewordsareatbestonlyvariablysyllabi­fied with the preceding word, and they will therefore have to be given ω status in the lexicon,alongwith the suffixes that formtheirownω, like -schap. It has alsobeennotedthatcertainfunctionwordsandaffixesbehaveneitheraselementsthat are included in the same ω as their host word nor as elements that form an ωbythemselves.Unlikeotherwords,Italianobjectpronounshavenostress,butthey nevertheless maintain their status as ωs, as shown by their phonology at the boundarieswithstressedwords.Foronething, theydonotchangethe locationofthestressonthe‘real’wordstotheirleft,asin[teˈlefonalo]‘Callhim!’,whosepre-antepenultimatestresswouldbeungrammaticalif[lo]werepartofthesameword.Foranother,radoppiamentosyntattico, which geminates a word­initial consonant after a stressedvowel, as in [ˈdal ˈliːbri] ‘Givebooks!’, as in a listinglike Don’t donate clothes. Give computers! Give pens! Give books!, from[ˈda ˈliːbri],equallyappliestopronouns,asin[ˈdallo]‘Giveit!’,from[ˈdalo].Onthebasisofsuchdata,NesporandVogel (1986)postulate theCliticGroupas a constituentimmediately above the ω, which makes it possible to account for phonological processes that occur between words and such recalcitrant morphemes, but that

Phonology above the word 213

fail tooccur inothermorphemecombinations.Alternative accountshavebeenpresented, however.

12.6.2 The syntactic residue

TheU,ι,φ,ωand,perhaps,theCliticGrouparetheprosodicconstituentswhichdefine the relevant domains of processes that apply above the word level. Prosodic theory thus distinguishes itself from theories that claim that such rules can refer directlytosyntacticstructure,suchasKaisse(1985).Nevertheless,instanceshavebeen found of rules that apply across words which do apparently refer to syntac­tic categories, as would appear to be the case for French liaison (Hayes 1990;Post 2000).Hayes (1990) proposes that such ‘residual’ syntax-sensitivity shouldbe accounted for in the lexicon. That is, the phonological rules that produce the required forms are in fact lexical rules, and the forms they produce are thus avail­able in the lexicon, ready for insertion into syntactic phrases. For instance, thephonological rule that shortensfinal long vowels inHausa verbs is syntacticallyconditioned: it only applies if a direct object that contains a major-class nounimmediately follows.This is illustrated in(33a),whichcontrastswith(33b), twocases in which the morphosyntactic condition is not met.

(33) a náːkáːmàkíːfíː ‘Ihavecaughtafish’ b náːkáːmàːʃí ‘Ihavecaughtit’ náːkáːmàːwàmúːsáːkíːfíː ‘IhavecaughtMusaafish’

Thelexicalruleisgivenin(34);itsmorphosyntacticconditioningisexpressedbythe‘Frame’givenbelowit.(BecausetheonlycategorythatcanoccurinitiallyinaVPbeforeanNPisaV,itisnotnecessarytolabelthewordasaVerbintherule.)Thus, when a verb is to be inserted in a sentence, the more specific form produced by final vowel shortening is chosen if the morphosyntactic condition applies.

(34) final voWel shorteningV→Ø /[...V__]Frame1 Frame1:[__NP...]VP

The assumption that syntax­dependent rules are in fact lexical rules puts such alternations in a comparable position with phrasal allomorphy of the sort that is seenintheEnglishindefinitearticle,whichis[ən]beforevowelsbut[ə]elsewhere.Asthenamesuggests,phrasalallomorphsarerivalphonologicalformswhosedis­tribution is governed by properties of the surrounding words. There are, however, twodifferencesbetweenthesetwocasesworthmentioning.First,theformsin(34)are generated by a rule, because they involve a whole class of words rather than a singlemorpheme,and,second,in(34)theconditioningismorphosyntacticratherthanpurelyphonological.HayesreferstoformslikeEnglish[ən]andHausa[káːmà]as precompiled,theideabeingthattheycomeready-madefromthelexicon.Apre­dictionofthistreatmentisthatruleslike(34)mighthaveexceptions.Infact,inthecontextof(34),arestrictedgroupofHausaverbs,e.g.[sàjáː] ‘bought’,raisetheirfinal[aː],inadditiontoshorteningit,asin[náːsàjíàbíncí]‘Iboughtfood’.

Phonology above the word214

12.7 CONCLUSION

Phonological rules that apply above the level of the word are constrained by pro­sodic constituents that form a hierarchical structure. These constituents are not directlymappableontothemorphosyntacticstructure.Whiletheprosodichierar­chy above the ω is in large part derivable from the syntactic structure, mismatches do occur, mainly as a result of the incommensurate lengths of the phonological constituents(cf.SelkirkandLee2015).Asyntacticconstituentmaybetoolongtofit into the phonological constituent it usually maps onto, with the result that it is divided over a number of such constituents, or too short, in which case it may be includedinaconstituentofthesame(restructuring)oralowerrank(cliticization).

The structural separation of the morphosyntactic and phonological grammars mayappeartobecompromisedintwoways.First,wehaveseenthatprosodiccon­stituentsnotonlydefinethedomainsatorwithinwhoseedgesphonologicaladjust­ments occur, but are arguably also referred to by rules of syntax, as in the case of heavyNPshift.Thismayindicatethat,justasfootstructuremaybeavailableinthelexicon(cf.chapter11),somorphosyntacticstructurewillexistsimultaneouslywith the phonological structure. For the reverse case, phonological ruleswhichappealtosyntacticinformation,Hayes(1990)providedasolutionwhichupholdsthe distinction between lexical rules, which can refer to morphological information and which potentially have exceptions, and postlexical rules, which can only refer tophonologicalrepresentationsandwhichcannothaveexceptions.Heproposedthat phonological rules for which syntactic information is relevant are in fact lexi­calrules.Suchsyntax-sensitiverulesonlyapparentlyapplyabovethelevelofthewordandactuallyproduce theappropriatealternants in the lexicon.During theconstruction of the sentence, these precompiled forms are inserted in the specific contexts instead of their rival, more general alternant.

Prosodic structure appears to be more orderly than syntactic structure. The SLH,which forbids improper bracketing, recursivity and non­exhaustiveness, constrains the prosodic structure from the ω onwards reasonably successfully, since deviations appear to be limited. Phonological constituents below the ω, which are not deriv­able fromthemorphosyntactic structure,deviate fromtheSLH in limitedways.Inparticular,notallsyllablesneedtobeincludedinfootstructure(straysyllables,section11.3.1).Wemaysafelypredictthatprosodicphonologywillcontinuetobean exciting field of research for some time yet.

NOTE

1 Thesegmentisknownas‘linkingr’ if there is an <r> in the spelling and ‘intrusive link­ing r’ifthereisnot.Inthelattercase,speakersmayfeelthatthepronunciationif[r]is‘incorrect’.

Transparency and opacity with rules and constraints

13.1 INTRODUCTION

In earlier chapters, we have seen how phonological representations have changed from linear strings, as in the early SPE model, to non­linear ones with autonomous and parallel strings of vowels, consonants and tones as well as a hierarchical organization ofconstituentsencompassingthesesegments.Wehavealsoseentwowaysinwhichadjustments intheserepresentationsastheyareincorporatedin ‘later’morphosyn­tacticstructureshavebeendescribed.Onewaywastoallowrules tochangerepre­sentations, and the other was a competition among possible output representations as decidedbyaseriesofrankedcriteria,theconstraintsofOptimalityTheory(OT).

Thelasttwochaptersofthisbookreturntoeachofthesebroadtopics.Chapter 14will consider the question whether segments have internal structure. In the SPE model, the segment consisted of a list of distinctive features, without any internal ordering.However,astrongcasehasbeenmadefortheassumptionthatsegmentsare trees whose branches represent specific groups of features, like those defining theplaceofarticulationofasegment.Chapter13willpointoutthatwhileOThasbooked some striking results that are unattainable in a rule­based framework, like the avoidance of the duplication problem, there are things a rule­based analysis can dothatseemunattainableinOT.

Wewillstartbydiscussinganumberofissuesinruleorderingwhichhavepre­occupiedmanyaphonologistovertheyears.First,aretheregeneralprinciplesthatpredicttheorderinwhichrulesapply?Arethereruleordersthataremorenaturalormoreexpectedthanothers?Theworkonthesequestionshasledtotherecog­nition of four types of rule interaction: the feeding, bleeding, counterfeeding and counterbleedingorders.Additionally,wedrawattentiontothenotionofopacity.Asurfaceformmaylookasifsomeruleshouldhaveappliedtoit,butdidn’t,orit may look as if some rule applied to it, but without apparent need, because the appropriatephonologicalcontextisn’tthere.Suchopaqueformsariseasamatterof course from rule ordering, as we will illustrate, but are predicted not to exist by anytheoryinwhichallconstraintsapplytosurfaceforms.Solutionstotheopac­ityprobleminOThavecomeandgone inrapidsuccession. In thischapter,wediscusstwoofthem.OneisinlinewiththetheoryofLexicalPhonologydescribedinchapter8,andaccordinglysplitstheOTgrammarupintominimallyalexicalandapostlexical component,knownasStratalPhonology.Thesecond involvesan enrichment to the evaluation procedure known as ‘chains’, which attempt to

13

Transparency and opacity216

reproducethecreationofintermediateforms,asstandardlyproducedaftereveryruleapplication,inOT.

13.2 EXTRINSIC AND INTRINSIC ORDERING

Is the order in which rules apply predictable from any properties of the rules con­cerned?Ifitis,noorderingstatementwouldbenecessary:theruleorderissaidtobeintrinsic. If the order is not given by the theory, and an explicit ordering statement ofthetype‘RuleXappliesbeforeRuleY’ is necessary, the rule order is extrinsic. The issueof intrinsicruleorderingoccupiedmanyphonologists in the1970s,but thesearch for the principles that exhaustively govern the order in which rules apply is generallyconsideredtohavebeenunsuccessful(KenstowiczandKisseberth1977).Aprinciple thathas stood the testof time is theelsewherecondition. This is really a principle governing the application of rules in general, and has been invoked inmorphologyaswellasphonology.Whatitsaysisthatwhenoneruleappliestoa subset of the forms that another rule applies to, the general rule is blocked from applyingtothatsubset.Amorphologicalexamplewillmakeitclearwhythis isausefulprinciple.TaketheEnglishmorphologicalrulesgivenin(1)and(2).Rule(1)says:‘Attachthesuffix[z]tonounstemsinordertoformtheplural’, andrule(2)says:‘Attachthesuffix[ən]tothenounstem ox in order to form the plural’.

(1) [[ ]n z]pl

(2) [[ɔks]nən]pl

In order to prevent the formation of *oxes, wemusteitherstipulatethat(2) appliesbefore(1),oraddto(1)theclause‘exceptinthecaseof[ɔks]’.Theelsewherecon­ditionmakeseithermoveunnecessary:because(2)appliestoasubsetofthecon­textsspecifiedby(1),itautomaticallyblocks(1).Thisprinciplethussavesusfromhavingtoaddallsortsofexceptionclausestogeneralrules.AFinnishphonologicalexample,fromKiparsky(1973),isgivenin(3).

(3) Underlying Derived menek mene ‘go’ menekalas menealas ‘godown’ menekpois meneppois ‘goaway’ menekkotiin menekkotiin ‘gohome’

Word-final[k]inFinnishisdeleted,unlessaconsonantfollows,inwhichcasethe[k]assimilatestothatconsonant,creatingageminate.Thek-assimilation rule is givenin(4).Theruleatissueisk-deletion.Withouttheelsewherecondition, wewouldhavetostatethisruleas(5),whichexplicitlyspecifiesthecontext‘whenfollowedeitherbyavowelorbyapause’,i.e.‘exceptwhenaCfollows’.

(4) k-assiMilation k # C→3 2 3

1 2 3

Transparency and opacity 217

(5) k-deletion k /__#→

ØV

pause

If we were to formulate k­deletion as (6), theelsewherecondition would ensure that it appliesonly if (4)didnot apply.Rule (4)applies toaproper sub­setoftheformstowhich(6)isapplicable,andthereforeappliesfirst,blocking(6)fromapplyingto[menekpois].Also,(6)willnotapplyto[k]in[menekkotiin],whichafterallwasinputtorule(4).Inthisform,(4)assimilatedthe‘original’[k]of[menek]tothe[k]of[kotiin].Althoughitappliedvacuously,itdidapply,therebyblocking(6).

(6) k-deletionk→ Ø /__ #

Awidelyciteddatasetusedforarguingforextrinsic ordering is based on dialec­talvariationinCanadianEnglish(Joos1942),whichcanbedescribedasresultingfrom different orderings of the same two rules, flapping and pre­fortis clipping (thetermsarefromWells1982).Thefirstrule,givenin(7),causes[td]tobepro­nouncedasanalveolartapbeforereducedsyllables,asin[ˈsɪɾi]city. Notethat[ɾ]isvoiced.Thesecondrule(8)shortensvowelsandsonorantconsonantsprecedingvoicelesssegments.Asaresultofthisrule,the[iː]in[biːd]bead is longer than that in[biːt]beat, the[εn]in[tεnz]tens islongerthanthe[εn]in[tεns]tense, and the [iː]in[ˈtiːzɪŋ]teasing islongerthatin[ˈliːsɪŋ]leasing. (Moreadequatestatementsoftheserulesarepossible,butnotnecessaryforthepointatissue.)Wewilluseanadhocfeature[±long].

(7) flapping [td]→ɾ/[−cons]__ Vstress−

(8) pre-fortis clipping [+voice]→ [−long]/__[−voice]

In some dialects, these rules applied in the order flapping – pre­fortis clip­ping. That is, the words rider (‘someonewho rides’) andwriter (‘someonewhowrites’)arehomophones,bothbeing[ˈraiɾər],whileride and write are[raid]and[rʌɪt](where[ʌɪ]representsashortened[aɪ]).Thisisshownin(9).

(9) raɪd-ər raɪt-ər raɪd raɪt Rule(7) ɾ ɾ (n.a.) (n.a.) Rule(8) (n.a.) (n.a.) (n.a.) ʌɪ Output ˈraɪɾər ˈraɪɾər raɪd rʌɪt

Today,dialectslike(9)haveapparentlybecomeextinct(Kaye1990).InCanadianEnglishtoday,thesewordsarepronounced[raiɾər],[rʌɪɾər],[raɪd]and[rʌɪt].Thissituationisobtainedifwereversetheorderoftherules,asshownin(10).Butifdialects can differ depending on the order in which two rules apply, as indeed they can,itcannotbethecasethatruleorderingispredictable(see(26)insection6.5).

Transparency and opacity218

(10) raid-ər rait-ər raid rait Rule(8) (n.a.) ʌɪ (n.a.) ʌɪ Rule(7) ɾ ɾ (n.a.) (n.a.) Output ˈraiɾər ˈrʌɪɾər raid rʌɪt

13.3 FEEDING, COUNTERFEEDING, BLEEDING, COUNTERBLEEDING

Earlyattemptstofinduniversalprinciplesgoverningtheorderinwhichrulesapplyled to a categorization of rule interactions(Kiparsky1968).Theideawasthatcer­tain rule orderings were more natural or expected than others, and that phonologi­cal change could in part be explained by assuming that in the course of time rules tendtoreordersothattheycometohavenaturalorders.Whiletheprincipleshavebeen abandoned, the terms used to refer to the different types of interaction have acquiredwidecurrency.Fourtypeswillbedistinguished.Ineverycasetworulesareassumed,whichwillbereferredtoasruleAandruleB.

13.3.1 Feeding order

IfruleAincreasesthenumberofformstowhichruleBcanapply,theorderA–Bisafeedingorder.TheBritishEnglishdatain(11)illustratepreglottalization, given in (12),arulewhichglottalizesvoicelessplosivesinthecoda,discussedinsection8.6.3.

(11) Underlying Derived lʊkt lʊɁkt looked kæts kæɁts cats hɪnts hɪnɁts hints kæmpgraʊnd kæmɁpgraʊnd campground(12) preglottalization

−−

→ + + ( )

contvoice

constr voice C/ __ # #

Now consider the data in (13). In the first and second columns, we see thatbetween a nasal and a voiceless fricative in the same syllable a voiceless plosive is inserted,whoseplaceofarticulationisthatoftheprecedingnasal.Asshowninthethird column, this rule of voiceless stop insertion(14)appliesbefore(12),andthus causes words like prince, which underlyingly end in a nasal consonant followed byafricative,tobeinputtorule(12).Thatis,(14)feeds(12).

(13) Underlying Fortis stop Preglottalization prɪns prɪnts prɪnɁts prince lεŋθ lεŋkθ lεŋɁkθ length wɔːmθ wɔːmpθ wɔːmɁpθ warmth(14) voiceless stop insertion

Ø→

−−

+

contvoice

PLACE

Cnas

PLACEα α/

+−

__

Ccont

voice σ

Transparency and opacity 219

13.3.2 Counterfeeding order

Theoppositeof feeding iscounterfeeding. If ruleA increases thenumberof theformstowhichruleBcanapply, theorderB–A isacounterfeedingorder.TheFrenchfeminineadjectivalsuffixis[ə].Thelanguagehasarulethatdeletesword-final[ə],whichisgivenasə­deletionin(15).Thelanguagealsohasfinal con­sonant deletion (16),whichdeletescertainword-final consonants incontextsotherthanbeforeavowelorglide,causing[pətit]tobepronounced[pəti]inthecontext concerned. final consonant deletion applies before ə­deletion. If the rulesappliedintheopposite(feeding)order,ə­deletion would cause the feminine alternant of ‘little’ to be homophonous with the masculine form. In other words, therulesapplyincounterfeedingorder,asillustratedin(17)fortheformsfor‘littlenephew’ and ‘little niece’.

(15) ə-deletion ə→ Ø /__ #(16) final consonant deletion C→Ø/__##[+cons](17) pətitnəvø pətit-ənjεs Rule(16) Ø (n.a.) Rule(15) (n.a.) Ø pətinəvø pətitnjεs

13.3.3 Bleeding order

IfruleAdecreasesthenumberoftheformstowhichruleBcanapply,theorderA–Bisableedingorder.Thistypeofinteraction,inwhichruleApreventsruleBfromapplyingtoparticularforms,occursinthederivationoftheEnglishplural.Inchapter4,itwasshownthatɪ­insertion prevents devoicing from applying to a formlike[bʌs-z]byseparatingthefinal[z]fromthestem-finalobstruent.In(18),this interaction is shown. That is, ɪ­insertion bleeds devoicing.

(18) bæk-z kɪs-z aɪ-z ɪ-insertion (n.a.) ɪ (n.a.) devoicing s (n.a.) (n.a.) bæks kɪsɪz aɪz

Thus,whentworulesAandBareinacounterfeedingorder,theapplicationofruleAdoes not increase thenumberofformstowhichruleBcanapply.Whentheyareinableedingorder,ruleAactuallydecreases the number of forms to which rule Bcanapply.

13.3.4 Counterbleeding order

Theoppositeofbleedingiscounterbleeding.IfruleAdecreasesthenumberoftheformstowhichruleBcanapply,theorderB–Aisacounterbleedingorder.Iftworules can apply to the underlying form, this order enables both rules actually to doso.IntheKaatsheuveldialectofDutch,[ə]isinsertedbetweenthenounstemandthediminutivesuffix,ifthestemendsinalaxvowelfollowedbyasonorantconsonant.Thus,wefinditin(19a),butnotin(19b),wherethestemendsinan

Transparency and opacity220

obstruent,orin(19c),wherethestemhasatensevowel.Theruleofə­insertion is givenin(20).Here,Xisafreevariableandrepresentswhatevershapethediminu­tivesuffixhas.

(19) a snɔr-kə snɔrəkə ‘moustache’ hɑl-kə hɑləkə ‘hall’ kɑm-kə kɑməkə ‘comb’ b lɑp-kə lɑpkə ‘cloth’ mʏs-kə mʏskə ‘sparrow’ c raːm-kə raːmkə ‘window’

(20) ə-insertionØ→ə/[−tense] ++

consson

+__X]diM

Kaatsheuvelalsohasarulewhichinserts[s]betweenthediminutivesuffix[kə]and a stem­final dorsal consonant. Thus, s­insertion (21)breaksupasequenceof two dorsal consonants, as in [bɑkskə], from [bɑk­kə] ‘tray’, [maːxske], from[maːx­kə]‘stomach’.

(21) s-insertionØ→s/+

consDORSAL

+__kə]diM

If we want to know how these two rules interact, we need to consider the diminu­tive formofaword like[slɑŋ] ‘snake’,whichsatisfiesbothrules.Letussuppose,contrary to fact, that it is *[slɑŋəkə].Inordertoobtainthisform,wewouldhavetoapply ə­insertionfirst,soastodestroythecontextofthetwoadjacentdorsalcon­sonants.Thiswouldbeableedingorder.Thecorrectform,however,is[slɑŋəskə].Thatis,weneedtoapplytherulesinacounterbleedingorder:first(21),then(20).

(22) bɑk-kə slɑŋ-kə snɔr-kə Rule(21) s s (n.a.) Rule(20) (n.a.) ə ə bɑkskə slɑŋəskə snɔrəkə

Q126 Mwera has three rules, given as (1), (2) and (3) below. Two noun stems in Mwera are [gomo] ‘lip’ and [kuja] ‘cape bean’. The plural is formed by prefixing a nasal consonant, whose underlying form is [n]. The plural surface forms are [ŋomo] and [ŋguja]. The following three rules derive the surface forms (Kenstowicz and Kisseberth 1977: 157).

(1)−−+

→ +[ ]

soncontvoice

nasØ/ __

Transparency and opacity 221

(2)−−

→ +[ ]

−+

soncont

voicesylnas

/ __

(3)−+

→[ ]

−

syl

nasPLACE

contPLACE

αα

featuresfeatures

/ __

1 Suggestsuitablenamesfortheserules.2 Showthatofthesixpossibleordersthattheserulescouldhave,onlyone

is correct.

Q127 Dutch has a number of rules affecting the feature [±voice] in obstruents. In order to derive the surface forms in the second col-umn from the underlying forms in the first, four rules are required: degemination, final devoicing, progressive devoicing and regressive voicing. The first two can be formalized as follows:

degemination [+cons](#)[+cons] 1 23 → Ø23 Conditionː1= 3final devoicing [−son]→ [−voice]/__C0#

Underlying Derived a laːt#blujər laːdblujər ‘late developer’b bɑd#bruk bɑdbruk ‘bathing trunks’c kʌud#vyːr kʌutfyːr ‘gangrene’d ʋɑnd#teːχəl ʋɑnteːχəl ‘wall tile’e leːz#fʌut leːsfʌut ‘reading error’f lup#zœyvər lupsœyvər ‘very pure’g boːt#tɔχt boːtɔχt ‘boattrip’h kaːz#zaːk kaːsaːk ‘cheese shop’i kɔp#bɑl kɔbɑl ‘header’(football)j χʌud#dεlvər χʌudεlvər ‘prospector’k χʌud#koːrts χʌutkoːrts ‘gold fever’l krɑb#sχaːv krɑpsχaːf ‘scraper’m leːz#brɪl leːzbrɪl ‘reading glasses’n hεis#bɑlk hεizbɑlk ‘hoisting beam’

1 Determinewhattheothertworulesshoulddo,andhowthefourrulesshould be ordered.

2 Give formal notations of progressive devoicing and regressive voicing.

3 Givethederivationsofitems(d),(f),(h),(i)and(m).

Transparency and opacity222

13.4 TRANSPARENCY AND OPACITY: RULES AND CONSTRAINTS

Whichoftheruleorderingsdiscussedintheprevioussectionarethemorenat­ural,ormoreexpected?Theanswerdepends tosomeextenton theperspectivethatistaken.Kiparsky(1968)hasarguedthathistoricalchangecouldinpartbeexplained by assuming that over time rules tend to reorder so as to maximize their application.IfruleAfeedsruleB,thenumberofformstowhichruleBcanapplyis increased,becauseruleAsuppliesadditionalformstoruleB,aswasthecasefor preglottalization,discussedearlier.Asaresult, ruleBcanbemaximallyapplied.Obviously, in a counterfeeding order, rule Bwill not be suppliedwithadditionalformstowhichitcanapply,asillustratedforFrench final consonant deletionin13.3.2.Similarly,ifruleAremoves formstowhichruleBcanpoten­tiallyapply(A–B:ableedingorder),theoppositeorder,counterbleeding,allowsruleB to applymaximally.Thus,on thebasisof theprincipleofmaximal rule application, feeding and counterbleeding are the natural rule orders.Whereasthere was general consensus that feeding was more natural than counterfeeding, there was disagreement as to whether bleeding or counterbleeding should be con­sidered the more natural order.

Butwhatdetermineswhat is ‘natural’or ‘unmarked’?From theperspectiveofrule maximization, feeding and counterbleeding are natural rule orders, but an alternative perspective was that rules are more natural when their application is transparentonthesurface.Fromthepointofviewofmaximal surface transpar-ency,bleedingandfeedingarethenaturalorders.Toseethis,twomorecasesofcounterfeeding and counterbleeding rule orders are presented.

InGranCanarianSpanish(Oftedal1985),thevoicedstops/b d g/arespirantizedto[βðɣ]inintervocalicposition,asshownin(23a).Additionally,Voicing causes intervocalic/ptk/toberealizedas[bdg],asillustratedin(23b).

(23a) Underlying spirantization /roba/ [roβa] ‘he/shesteals’ /nada/ [naða] ‘nothing’ /lagana/ [laɣana] ‘theappetite’

Q128 The river Linge [lɪŋə] flows not far from Kaatsheuvel. What would you expect the diminutive form to be in that dialect?

Q129 In (10), the order in which flapping and pre-fortis clipping are applied to rider and writer in mainstream Canadian English results in different surface forms for these two words. What type of rule order is this?

Transparency and opacity 223

(23b)Underlying voicing /tipiko/ [tibigo] ‘typical’ /lakama/ [lagama] ‘thebed’ /unatjεnda/ [unadjεnda] ‘ashop’

AcounterfeedingapplicationofSpirantization followed by Voicingstraight­forwardlygeneratesthecorrectsurfaceforms.However,fromasurfaceperspective,thevoicedplosivein[lagama]doesnotmakegoodsense,giventheexistenceofSpirantization.Bythisrule,intervocalic/g/surfacesas[ɣ],but[g]survivesinsurfaceformslike[lagama].

Or,again,Slovak(Rubach2000)hasadiminutivesuffix/æ/whichtriggersPala­talization of a preceding stem­final coronal, as in páňa /pan-æ/[paɲæ]‘master(dim)’.Afurtherprocessofæ-Backingbacksunderlying/æ/to[a]afternon-labi­als,bywhich[paɲæ]becomessurface[paɲa].So,whenthediminutivesuffix/æ/ispreceded by a labial consonant, as in holúbä /holub-æ/[holubæ]‘pigeon(dim)’,nei­ther rule applies. This is an example of a counterbleeding rule order: Palataliza­tionappliesbeforeæ-Backingtoprevent/pan+æ/fromsurfacingwithincorrect*[pana].Fromtheperspectiveofmaximal ruleapplication, thiscounterbleedingrule order is natural, since it makes Palatalization maximally applicable, but from the perspective of maximal surface transparency it is not, since it makes Palatalizationanon-transparentrule.Thisisbecausethesurfaceform[paɲa]has a palatalized consonant in the absence of a following front vowel that triggers thepalatalization.Inotherwords,somethinghappens,but it isunclear(opaque,not surface transparent)why itdoes.Foracounterfeedingruleorder, thingsarethe other way around: something does not happen([lagama]notturninginto[laɣama]),butitisnotclear(notsurfacetransparent)whyitdoesnot.Thatis,bothcounterfeedingandcounterbleedingleadtosurfaceopacity.Counterbleedingleadsto forms with rule outputs that have no context and is therefore also referred to as over-application. Counterfeedingleadstoformswithcontextsthatspareatargetand is also referred to as under-application. These two opaque rule interactions are problematicforclassicOT,whoseconstraintsapplytosurfaceforms.

13.5 OPACITY AND OT: TWO SOLUTIONS

13.5.1 Counterfeeding opacity is problematic for OT

OThasnoproblemsinaccountingforfeeding(transparent)ruleorders,butcannothandlecounterfeeding(opaque)ruleorders.ThisisillustratedforGranCanarianSpanishintableau(25),whichcontainstwofaithfulnessconstraintsandtwo marked-nessconstraints,aslistedin(24).

(24) ident(voice): Aconsonantintheoutputhasthesame[voice]specificationasintheinput. ident(cont): Aconsonantintheoutputhasthesame[continuant]specificationasintheinput. voicing: novoicelessintervocalicplosive. spirantization: novoicedintervocalicplosive.

Transparency and opacity224

(25a) /lagana/ voicing ident(voice) spirantization ident(cont)[lagana] *![lakana] *! * [laɣana] *

(25b) /lakama/ voicing ident(voice) spirantization ident(cont)[lagama] * *![lakama] *! [laɣama] * *

Tableau(25a)showsthat, inorderforSpirantization to take place, the con­straint Spirantization must dominate Ident(cont).If it does not, the first out­putcandidatewillbeselectedasoptimal.Similarly,asshownin(25b),toproducevoiced intervocalic plosives, Voicingmust dominate Ident(voice).This ranking, however,incorrectlypredictsthat/lakama/isoptimallyrealizedas[laɣama]ratherthanasactual[lagama].

Q130 Show that the ranking of Voicing » Ident(voice) below Spirant-ization » Ident(cont) does not produce the correct results either. What is the incorrect result of this ranking for the inter-vocalic plosives?

Tableaux(25a)and(25b)showthatOTcannotaccountforacounterfeedingruleinteraction.Asithappens,counterfeedingsituationsinvolvingspirantizationandvoicingarequitecommon.TheGranCanarianSpanishcasecanalsobeobservedinSardinian(Bolognesi1998)andCorsican(Gurevich2004).Kaye(1975)proposeda functional motivation for counterfeeding opacity: keeping intact lexical contrasts and thus avoiding neutralization. Indeed, if spirantization and voicing were to stand inafeedingrelationship,thesurfaceformsforminimalpairslike/nada/[naða]and/nata/[nada]‘cream’,or/roba/[roβa]and/ropa/[roba]‘clothes’,wouldberealizedidentically,as[naða]and[roβa]respectively,neutralizingtheunderlyingcontrast.Infact,Gurevich(2004)observesinherstudyof230similarprocessesinacorpusof153languagesthatinthemajorityofcases(92%)theyarenon-neutralizing.Inci­dentally,thiscontrast-preservingtendencyisnotonlyproblematicforOT,butalso forarule-basedtheory.HaleandReiss(2008:14)state:‘Opaquerulesarenotsurfacetrue, rules that are not surface true are harder to learn, failure to learn aspects of the ambient language constitutes a diachronic change, therefore, languages are more likelytoloseagiveninstanceofopacitythantogainone.’Now,iftheunmarkedruleorderisafeedingruleorder(Kiparsky1982b),itisnoteasytounderstandwhyinthemajorityofcasesacounterfeedingorderingbetweenSpirantizationand Voicingis attested.

Transparency and opacity 225

13.5.2 Counterbleeding opacity is problematic for OT

AcounterbleedingruleorderofPalatalizationandæ-Backing leads to a pala­talizedconsonant inSlovak [paɲa] forunderlying /pan+æ/ ‘master (dim)’.How­ever,apalatalizedconsonanthereappearswithout itscontext /æ/andassuch isnottransparent.Tableau(27),wherethefaithfulnessandmarkednessconstraintsin(26)areused,showsthatcounterbleedingopacityisasproblematicforOTascoun­terfeeding opacity. The third output candidate is incorrectly selected as optimal, insteadoftheactualSlovakform[paɲa].

(26) ident(back): Avowelintheoutputhasthesamevalueforthefeature[back]asintheinput. ident(ant): Aconsonantintheoutputhasthesamevalueforthefeature[anterior]asintheinput. palatalization: nonon-palatalizedcoronalconsonantsbeforeafrontvowel. æ-backing: no[æ]afternon-labials.

(27) /pan+æ/ æ-backing ident(back) palatalization ident(ant)panæ *! *paɲae *! *pana *paɲa * *!

Q131 Show that a ranking æ-Backing » Ident(back) below Palatal-ization » Ident(ant) is not helpful in obtaining a counterbleed-ing relationship between Palatalization and æ-Backing.

TheobviousreasonwhyOTcannotdescribecounterfeedingandcounterbleed­ingopacityistheuseofoutputmarkednessconstraints.Aconstraintsuchasspi­rantization (no voiced intervocalic plosive) cannot differentiate between [lagana] as anoutputof /lagana/ and [lagama]as apossibleoutputof /lakama/,neither of which satisfy spirantization.Arule-basedapproachcanexploit theunderlyingdifferencebetween/g/in/gana/and/k/in/kama/byapplyingSpirant-ization before Voicing,creating[lagama]from/lakama/afterSpirantization hasapplied.InOT,thereisnostraightforwardwaytomakeuseofthatunderlyingdifference, because its output constraints cannot refer to inputs. In the next section, we will briefly discuss the merits of two solutions that have been proposed to allow OTtohandlecounterfeedingandcounterbleedingopacity.

13.5.3 Stratal OT and bringing some derivation back in the OT model

Inchapter8,wehaveseenthattherearegoodreasonstoassumethat,inadditiontothe underlying and the surface representation, there is an intermediate level of rep­resentation:thelexicalrepresentation.Thatideahasalsobeenimplementedin OT

Transparency and opacity226

and is generally known as Stratal OT. The underlying form passes through a lexical constraint hierarchy and yields an optimal intermediate output form. This form is then input to a second constraint hierarchy, potentially different from the first, andproducestheoptimalsurfaceform(Booij1997;Rubach2000;Kiparsky2015;Bermúdez-Oteroforthcoming).Forcounterbleeding,opaqueSlovakPalataliza­tion, Rubach (2000)hasproposed the following analysis. In thefirst constrainthierarchy,theorderoftheconstraintsæ-Backing and Ident(back)is the opposite tothatintableau(27).Thisisillustratedin(28),wheretheoptimaloutputcandi­datenowbecomesthesecondoutputcandidate,[paɲæ],insteadof[pana].

(28)

/pan+æ/ ident(back) æ-backing palatalization ident(ant)panæ * *! paɲæ * *pana *!paɲa *! *

Thisoptimal intermediateoutput form [paɲæ] is then input to a secondcon­strainthierarchy,wheretheorderoftheconstraintsæ-Backing andIdent(back)isinverted,asillustratedin(29).

(29)

/paɲæ/ æ-backing ident(back) palatalization ident(ant)panæ *! * *paɲæ *!pana * *! paɲa *

Theresultofthistwo-levelapproachisthatcorrect[paɲa]isselectedastheopti­maloutputformforunderlying/pan+æ/.

Q132 Provide a Stratal OT analysis of the counterfeeding case in Gran Canarian Spanish, such that Spirantization is satisfied at the first level, but Voicing is not.

Although Stratal OT can describe the problematic non-surface-transparentinteractions, it is not always evident that the different levels are independently motivated. Spirantization and voicing would at first sight seem to belong to the postlexical domain. Spirantization would be considered a postlexical process in themodeloutlinedinchapter8,becauseitproducesanovelsegment,[ɣ].IfthefirstlevelinQ132isapostlexicallevel,asecondpostlexicallevelisrequiredforVoicing to be satisfied by non­spirantized plosives.

Movingonthesecondsolution,wenoteagainthatwhatneedstobeexpressedisthatitisallrightforGranCanarianSpanish/lakama/tochangeinto[lagama],but

Transparency and opacity 227

thatitshouldnotfurtherchangeinto[laɣama].Similarly,Slovak/pan-æ/shouldfirst changeto[paɲæ]andthen to[paɲa],butnotdirectly from/pan-æ/to[pana].Toachievethis,itisnecessarytokeeptrackofthederivationalhistory.StandardOTcannotdothat,asmarkednessconstraintsareoutputconstraints.Assuch,theycanrefer neither to the input nor to the way input and output are connected. The second solution, then, is to make sure that changes in the phonological representation take placeonebyone,ratherthanatonego.Achangeinthephonologicalrepresenta­tion of a word implies that a faithfulness constraint is violated, but a markedness constraint respected.Going from input to output, successive changes produce aseries of intermediate forms that step­wise give up faithfulness in order to improve onmarkedness. In(30),we illustrate this forSlovak/pan+æ/.Theinput isgivenin the first column, and the second column lists the chain of intermediate forms, whereby the faithfulness constraint which an intermediate form violates is listed below the intermediate form in question.

(30)

Input Intermediateforms Output1/pan-æ/ <panæ>

NoFaithfulnessViolations[panæ]

2/pan-æ/ <panæ>,<paɲæ> ident(ant)

[paɲæ]

3/pan-æ/ <panæ>,<pana> ident(back)

[pana]

4/pan-æ/ <panæ>,<paɲæ>, <paɲa> ident(ant) ident(ant)

ident(back)

[paɲa]

Chain 1 just contains the fully faithful form. Chain 2 contains an interme­diate form that violates Ident(ant), while chain 3 contains one that violatesIdent(back).Chain4contains two intermediate forms,<paɲæ>,whichviolatesIdent(ant),and <paɲa>, which violates both Ident(ant) and Ident(back).Howcanwenowimplement the idea that /pan+æ/shouldfirst changeto[paɲæ]andthen to[paɲa],butnotdirectly from/pan+æ/to[pana]?Thiscanbedonebyimpos­inganorderontheviolationsofthefaithfulnessconstraints.Forthis,anewtypeofconstraint is introduced, Prec(edence),whichisgivenin(31),whereAandBarefaithfulness constraints.

(31) prec(a,b): anyformviolatingB isprecededbyaformviolatingAandisnotfollowedbyaformviolatingA.

There are thus twoways of violating (31): by having a B-violatorwithout apreceding A-violator in the chain and by a B-violator which is followed byan A-violator in the chain. To check whether an output candidate violatesPrec(Ident(ant),Ident(back)), say, we scan the list of intermediate forms to see whether a form that violates Ident(back) is preceded by a form that violates Ident(ant).If this is not the case, a violation of Prec(Ident(ant),Ident(back))is incurred.Next, if a form that violates Ident(back) is followed by one that

Transparency and opacity228

violates Ident(ant),anotherviolationisincurred(McCarthy2007).ForSlovak,weillustratethisin(32),whichliststheoutputcandidatesinthefirstcolumn,asusual, but now together with the input form and the chain of intermediate forms. ConstraintPrec(Ident(ant),Ident(back))isirrelevantforcandidate1,whichviolatesneitherofthetwofaithfulnessconstraints,whilecandidate2satisfiesit,because the form with the violation of Ident(ant),<panæ>, is not preceded by a form that violates Ident(back). However, Prec(Ident(ant),Ident(back))isviolatedbycandidate3,whichhasaformviolatingIdent(back)in its chain of intermediate forms which is not preceded by one showing a violation of Ident(ant).Candidate4violatesbothIdent(ant)and Ident(back).Ascanofthe chain of intermediate forms shows that the violations occurred in the right order, and there is thus no violation of Prec(Ident(ant),Ident(back))by can­didate4.

(32)

/pan-æ/ æ-backing ident (back)

prec (ident(ant), ident(back))

palatalization ident (ant)

1/pan-æ/<panæ>[panæ]

*! *

2/pan-æ/<panæ,paɲæ>[paɲæ]

*! *

3/pan-æ/<panæ,pana>[pana]

* *!

4/pan-æ/<panæ,paɲæ,paɲa>[paɲa]

* *

Gran Canarian Spanish can be brought in line with Prec(Ident(cont), Ident(voice)).In(33),thechainsarelistedinthesameformatasin(30).

(33) Input Intermediateforms Output1/lagana/ <lagana>

NoFaithfulnessViolations[lagana]

2/lagana/ <lagana>,<lakana> ident(voice)

[lakana]

3/lagana/ <lagana>,<laɣana> ident(cont)

[laɣana]

4/lakama/ <lakama>,<lagama> ident(voice)

[lagama]

5/lakama/ <lakama> [lakama]6/lakama/ <lakama>,<lagama>, <laɣama>

ident(voice) ident(voice) ident(cont)

[laɣama]

Transparency and opacity 229

In(34a),thetreatmentof/lagana/isshown.Asithappens,thePrecedence con­strainthasnocrucialroleinaccountingfortheexclusionofcandidate2.Whilethechain of intermediate forms actually contains a form violating Ident(voice) without a preceding form with a fricative, the violation of Prec(Ident(cont),Ident(voice))isirrelevant.Thus,OTwithchainsisonlyrelevantto‘harmonic’forms,i.e.candi­dates that are not already eliminated by constraints that outrank the Precedence constraint,likecandidate2in(34a).Candidate1violatesneitherfaithfulnesscon­straint, so that again Prec(Ident(cont),Ident(voice)) has no role to play, but founders on spirantization.Asaresult,theuncontestedwinneriscandidate3.

(34a)

/lagana/ voicing ident (voice)

Prec (ident(cont), ident(voice))

spirant ident(cont)

1/lagana/<lagana>[lagana]

*!

2/lagana/<lagana,lakana>[lakana]

*! * *

3/lagana/<lagana,laɣana> [laɣana]

*

Wethusmoveonto/lakama/.Becausecandidate5iseliminatedbyvoicing, the competitionfor/lakama/isbetweencandidates4and6,asshownintableau(34b).Candidate6violatesPrec(Ident(cont),Ident(voice))twice.First,theviolationofIdent(voice) is not preceded by a violation of Ident(cont)and,second,theviolation ofIdent(voice) is followed by a violation of Ident(cont).Theviolationbywinningcandidate4ofSpirantizationis non­fatal, since its fate as the winner is ensured by obeying the Precedence constraint.

(34b)

/lakama/ voicing ident (voice)

prec(ident(cont), ident(voice))

spirant ident (cont)

4/lakama/<lakama,lagama> [lagama]

* *

5/lakama/<lakama>[lakama]

*!

6/lakama/<lakama,lagama,laɣama>[laɣama]

* *!* *

In this section, we have briefly illustrated two ways in which OT allows us to describe counterfeeding and counterbleeding opacity. The simpler solution was that ofStratalOT,whichassumesdifferentphonologicalgrammarsfordifferentstages

Transparency and opacity230

in the process of the incorporation of phonological forms into higher levels of the morphosyntacticstructure.AdisadvantagewasthatinthecaseofGranCanarianSpanishthetwolevelscouldnotbeindependentlymotivated.Thesecondsolutioninvolved the addition of a mechanism that keeps track of the history of changes to anunderlyingrepresentation.AnevaluationofOTwithchainswillneedtorecog­nize that this theory imports some form of ‘extrinsic’ order into a theory with freely rankable constraints. If the set of constraints is truly universal, as we mentioned in chapter4,thisimpliesthatallthePrecedence constraints required to describe all possibleopaqueinteractionswillhavetobepartoftheconstraintset.Togetherwiththe lack of independent evidence for the mechanics of the solution, this implication may leave one with an expectation of new and better insights.

13.6 CONCLUSION

In this chapter we have discussed and exemplified four possible types of rule inter­action:feeding,counterfeeding,bleedingandcounterbleeding.Weestablishedthatcounterfeeding and counterbleeding lead to opaque phonological surface forms. These contradict some phonological rule of the language, either by showing the effectoftherulewherethereisnolongeranycontextforit(counterbleeding),orbynotshowingtheeffectwherethecontextispresent(counterfeeding).

WehavealsoseenthatOTcannot account for opaque forms if markedness con­straints are there to say something about surface forms. In response to the inherently ‘transparent’natureofOTconstraintswebrieflyreviewedtwosolutions:StratalOTand OTwith chains of intermediate forms connecting inputs to outputs. The basic observationmadeinchapter4wasthatoneofthegoalsofphonologicaltheoryisto account for the fact that morphemes may be pronounced differently depend­ingupontheirenvironment.Whenwecomparedrule-basedandconstraint-baseddescriptions, we argued that OTis to be preferred, given that it does not run into the duplication problem – the necessity of stating the same generalization both in a phonological rule and in a constraint, a problem inherent in rule­based descrip­tions.Also,OTwasarguedtobeabletoexpressdirectlyhowanynewinputformsfrom any donor language are brought in line with the structural demands of the recipient language.SystematicmodificationsinpronunciationthusseemedtobetakencareofbyOT.Inthischapter,wehaveseenthatOTcandealwithopacity,acommon fact of life in the phonologies of languages, but only by introducing addi­tional machinery that does not always seem well motivated.

This is the point where we should step back and briefly evaluate the phono­logical enterprise. SinceSPE, theoretical phonology has had a two­fold research agenda.Oneofthesefocusedonrepresentationsandhas ledtotheintroductionof moras, syllables, feet and more, as we have shown in the preceding chapters. The second has focused on levels of representation and how these are connected. Whilethefirstfocushasledtoexcitinginsights,thesecondfocusonthephonologi­calresearchagendaleavesuswonderingwhatwentwrong.Withoutmuchado,aspeakerofGranCanarianSpanishwilltakeasoundsequence[lagama]torefertoa

Transparency and opacity 231

representation /la kama/, just as he will take a sound sequence [la ɣama] to refer to a representation /la gama/. But none of the theories that we have discussed would appear to convincingly explain how underlying /la kama/ and /pan+æ/ get to be [la gama] and [paɲa] in actual pronunciation. Perhaps it is relevant to note that, whether rule-based or constraint-based, all input-output models appear to have a speaker orientation in common, a perspective that might be too narrow. There have been proposals that take the opposite perspective (Smolensky 1996; Boersma 1998; Kager 1999). Progress in this area could come from research into the way language production and perception are dealt with in real life, in the language user’s brain. Such research is now being conducted both by psycholinguists on the basis of various behavioural tasks and by neurocognitive scientists on the basis of the time course and localization of brain activity associated with producing and perceiving speech. New perspectives on the old problem of connecting inputs and outputs may well therefore be waiting around the corner.

Feature geometry

14.1 INTRODUCTION

In this last chapter, we will present a hierarchical configuration of the distinctive fea­turesintroducedinchapters5and6.Themainadvantagesofthisautosegmentalizedrepresentation are, first, that features can spread to neighbouring segments, thereby representingthepartialidentityofadjacentsegmentsinthemostparsimoniousway,and, second, that specific groups of features can spread, thereby making the claim thatnotallcombinationsoffeaturescanbethefocusofanassimilation.Asforthefirst point, assimilations show a non­arbitrary relationship between the context and the target.Timeandagain, rulesappear to transfera specific featureorgroupoffeaturesfromonesegmenttoaneighbouringsegment.However,therepresentationwe introduced inchapters5and6 is incapableof expressing this.ConsideroncemoretheDutchruleofregressive voicing,whichvoicesobstruentsbefore[bd].It is a natural rule in the sense that voiced segments cause preceding segments to be voiced,asopposedtovoicelessorlabial,say.Anarrangementofsegmentsconsist­ing of self­contained lists of features will not provide a natural way of expressing the transferoffeaturesfromonesegmenttothenext.Neithercanwemakeadistinc­tionbetweenthetransferofanexpectedfeature,[+voice],inregressive voicing, andanunexpected feature.Anelegant solution to thispredicament shouldallowsegments to share the same feature, much in the way successive tone­bearing units canbeassociatedwiththesametone.Asforthesecondpoint,inordertotransferthe right groups of features from one segment to the next, the representation should allow only those feature groups to be shared that actually go together in assimilation rules.Manylanguageshaveprocessesthattransferthefeaturesspecifyingaconso­nant’s place of articulation to a preceding nasal consonant, for instance. If the features are unstructured lists, it is impossible to distinguish between groups of features that are never shared and groups that routinely are. That is, we would not want an unnat­ural feature group like [labial,±sonorant]tobeaseasilytransferrableasanaturalonelike[±voice,±spreadglottis,±constrictedglottis].

In feature geometry (Clements 1985, 1993; Sagey 1986;McCarthy 1988), theproblem of the grouping of features is solved by representing segments as trees, in which the nodes represent features and feature groups. The sharing problem is solvedbyassumingthatasinglenodemaybepartofmorethanonetree.Toshowthemeritsofthisproposal,twomaintopicswillbedealtwith.Oneconcernstheopportunity that is afforded by the underspecification of features – and, hence, the absence of particular nodes in the representations of segments – to characterize long­distance assimilations. Particularly relevant in this context are word­based

14

Feature geometry 233

restrictions on the distribution of specific features, as is seen in vowel harmony. The solution to characterizing these distributional restrictions is reminiscent of the descriptionof the lexical tonepatternsof languages likeEtung andMende,dis­cussed in chapter 10.The second topic concerns the representationof the threetypesofcomplexsegmentsrecognizedinchapter2.Finally,wewillreturntothequestionoftherepresentationofpalatoalveolarandpalatalsegments,like[tʃ]and[j],andarguethatthey,too,arecomplexsegments.

14.2 NATURAL FEATURE CLASSES

Asobservedabove,assimilationsofplaceinvolvethetransferofacollectionoffeatures.Manylanguageshaverulesassimilatingnasalconsonantstotheplaceofarticulationofthefollowingconsonant.Forinstance,Hindi,whichhasbilabial,dental,retroflex,palato-alveolarandvelaroralstopconsonants,alsohasthecorrespondingnasals[m n ɳ ɲŋ],whoseplaceofarticulationisdeterminedbythefollowingconsonantintheword.Word-internally, sequences like *[nk],*[ŋʈ],*[ɳt],*[mc]or*[mt]arethereforeill-formed.Asaresult,aprefixlike[sam]‘together’showsalternationsofthetypeillustratedin(1) (Ohala1983).Clearly,theplacefeaturesmustbeaccessibleasagroupinordertoexpress this phonological process, the transfer of all the place features, including depen­dentoneslike[±anterior],asshownbythewordsfor‘equilibrium’and‘movement’.

(1) aːkaːr ‘shape’ samaːkaːr ‘homophonous’kiːrtan ‘devotionalsinging’ saŋkiːrtan ‘collectivedevotionalsinging’toːl ‘measure’ santoːl ‘equilibrium’calan ‘conduct’ saɲcalan ‘movement’naːd ‘sound’ sannaːd ‘consonance’

Thereare alsoprocesses that transfer a subgroupof theplace features.Englishcoronal assimilationassimilates[tdnl]inplaceofarticulationtothefollowingcoronalconsonant.Thatis,before[θð]theyaredental,before[tdnl]theyremainalveolar,andbefore[ɹ],whichinEnglishhasapostalveolarplaceofarticulation,theyarepostalveolar.In(2),werepeatthefeaturevaluesforthesegmentsconcerned.Thedatain(3)illustratehowthevaluesofthefeatures[anterior]and[distributed]aretogetherpassedontothepreceding[–cont,+cor]segment(Clements1985).

(2) COR Ant Distrθðn + +tdnl + –ɹt – –

(3) ɪnɹəʊm in Romeɔːlðεə all thereɪnθɜːsk in Thirsktɹaɪ trygεtɹεd get redwidθ width

Feature geometry234

Ofcourse, single featuresarealso frequently transferred fromone segment tothenext.ConsidertheOldEnglishdatainchapter7(Q61(2)),whichshowedthatthe[–back][ç]appearedafter[–back]vowels,and[+back][x]after[+back]vowels.This process is understandable if we assume that the vowel transfers the value for thefeature[back](butnootherfeatures)tothefollowingdorsalfricative.Justasphonological processes are seen as evidence for the existence of natural segment classes(anygroupofsegmentsreferredtobyaprocess),soanaturalfeatureclasscanbedefinedasagroupoffeaturesthatismanipulated(i.e.transferred,deletedorinserted)bysomephonologicalprocess.

14.3 BUILDING A TREE

Underatheoryinwhichtherepresentationofasegmentisanunstructuredlistoffeatures,therepresentationof[t]wouldbeasin(4a).WhatruleslikeHindinasalplaceassimilationsuggestisthattherepresentationmustbesomethinglike(4b).Further,onthebasisofprocesseslikethatillustratedin(3)wewouldseemtoneedarepresentationlikethatin(4c).Whatthismeans,ofcourse,isthatthesegmentisnotasingleconstituent,buthasinternalstructure.Suchconstituencyisbestrepre­sented by means of a tree diagram.

(4) a+cons

sonvoice

spreadconstr

naslat

cont

+antdistr

COR

−−−−−−−

−

b

+consson

voicespreadconstr

naslat

cont

−−−−−−−

−

COR

+antdistr

c

+consson

voicespreadconstr

naslat

cont

−−−−−−−

−

COR

+antdistr

14.3.1 The PlaCe node

The place node dominates articulator nodes corresponding to the univalent fea­tures [labial],[coronal],[dorsal]and[radical],asseenin(5). Eachofthesedominates subconstituents corresponding to their relevant features. Thus, [labial]dominates [±round], [coronal] dominates [±anterior] and [±distributed], and[dorsal] dominates [±back], [±high] and [±low].With the help of this repre­sentation, processes can refer to the constituent [labial,distr,round],butnotto

Feature geometry 235

thenonconstituent [high,ant], for instance.Wewill assume that thepositionof[±tense]willbeunder[rad],assumingitisequivalentto[±ATR].

(5) PLACE(5)

LABIAL

distrant back lowhigh tense

RADDORSALCOR

round

14.3.2 The laryngeal node

What other natural feature classes are there besides place features? Lass (1976:145ff.)hasarguedthatthelaryngealsegments[hʔ]havenosupralaryngealspeci­fication at all. He based his case on the frequently observed phenomenon thatobstruentssometimesappeartolosealltheirsupralaryngealinformation.ConsidertheLondonEnglishdata in(6),whicharerepresentativeof thiskindofprocess.Incertaincontexts,preglottalizedplosivesarepronouncedas [ʔ]. (Thevariationbetween preglottalized plosives and glottal stops is stylistic; i.e. both pronunciations maybeheardfromthesamespeaker.)

(6) pɪʔktʃə or pɪʔtʃə picturemɪiʔtwɪljəm or mɪiʔwɪljəm meet Williamnɪʔknæʔks or nɪʔnæʔks knick-knacksnɔʔtnæʊ or nɔʔnæʊ not nowkɪiʔpsmɑɪlɪŋ or kɪiʔsmɑɪlɪŋ keep smiling

InNewWorldSpanish,[s]hasbeenlostincodaposition.Thatis,thepluralof[klase]‘class’,whichinPeninsularSpanishis[klases],is[klasε].(The[ε]arisesfromaseparateprocesswhichisnotrelevanthere.)However,the[s]hasnotleftwithoutatrace:an[h]-likeoffglidecanbeheardafterthepluralincarefulspeech.Similarly,voicelesssonorantshavearisenword-internallyafterthelossoftheoralarticulationof[s],asshownin(7).

(7) Stage I Stage IImismo miimoor mimmo ‘same’fosforo fooforo ‘match’

Bothprocessescanbedescribediflaryngealandsupralaryngealfeaturesaresep­arateconstituents.Underthatassumption,boththeEnglishandtheSpanishdatawould receive the same description: deletion of the supralaryngeal constituent. InthecaseofEnglish,thiswouldleave[+constr,–voice,–spread]behind,i.e.[ʔ],andinthecaseofSpanishitwouldleave[–constr,–voice,–spread]behind,i.e.avoicelessintervalatthatpointintheword.Suchorphanedlaryngealfeaturesthusgenerallyendupas[ʔ]and[h],respectively.ToexpressLass’sproposalinourfeaturetree, we need to assume that segments have a laryngeal and a supralaryngeal

Feature geometry236

constituent.Toensurethatsegmentswithonlyoneoftheseconstituentsalsohavespecificationsforthemajor-classfeatures([±consonantal]and[±sonorant]),thesefeaturesmustbepresentinbothconstituents.AccordingtoMcCarthy(1988),theymake up the root node. This decision can safely be made, since no processes have beenreportedthattransfer[±consonantal]or[±sonorant]fromonesegmenttothenext.Thismeansthatourtreenowlookslike(8).Wewilltreatthefeature[±approx­imant]asamannerfeature.

(8) ROOT(8)consson

spread voice constr

distround ant back high low

LAR

LABIAL DORSAL

PLACE/SUPRALAR

RAD

tense

COR

14.3.3 The suPralaryngeal node

In the preceding sections, we have been able to establish a constituent structure for themajor-classfeatures,thelaryngealfeaturesandtheplacefeatures.Whatremainstobedeterminedisthelocationofthemannerfeatures[±nasal],[±continuant]and[±lateral],inadditionto[±strident]and[±approximant].Dotheyformathirdcon­stituent, or are they located under either the laryngealor the supralar/place node?Wearguethat[continuant]isnotlocatedinthelaryngeal constituent in (8)onthebasisofanassimilationprocesschanging[h]to[ç]inFrisian(Tiersma1985).Recallfromchapter5that[hʔ]arenotspecifiedfor[+cont],andthatthiswas the explanation for the fact that they do not trigger nasalization. This rule nasalizesavowelbefore[n],provideda[+cont]consonantfollows.In(9a)theeffectoftheruleisillustrated,while(9b)showsthat[h]doesnottriggerit.

(9) Underlying Surfacea in-jɑːn ĩjɑːn ‘togivein’

oən-flɪənə oəflɪənə ‘toattack’b in-hεljə inhεljə ‘toholdin’

oən-hɪərə oənhɪərə ‘tolistento’

Knowingthat[±continuant]isnotintheleft-handbranchof(8)isnotenoughto decide whether it is dominated by the supralar/place node or directly by the rootnode.ThesouthernvarietyofFrisianhasaruleofdiphthongshift,which –amongothervocalicchanges–turns[iə]into[jε].Whenthis[j]appearsafter[h],itassimilates[h]to[ç].Thatis,theplacefeaturesof[j]areaddedtotherepresenta­tionof[h].Thequestionnowiswhether[±continuant]istransferredalongwiththeplace features. If it is dominated by the supralar/placenode,[+cont]of[j]willbetransferredto[h],alongwithitsplacefeatures.Ifitisincludedasathirdbranchunder the root node, [h] would only receive a place specification, but would

Feature geometry 237

continuetolackthespecification[+cont].Thatis,transferof[+cont]predictsthataftertheassimilation,[ç]willtriggernasalization, while no transfer predicts that [ç]willnottriggernasalization.Thedatain(10)showthatthenew[ç]triggersnasalization andmustthereforebe[+cont].Thissegmentisnow[–son,–cons,+spread,–voice](whichiswhat[h]contributes),aswellas[+cont,–lat,–nas,cor, –ant](whichiswhat[j]contributes).

(10) Underlying Northern Southernin-hɪəkjə inhɪəkjə ɪçjεkjə ‘tohookin’oən-hiərə oənhɪərə oəçjεrə ‘tolistento’

Thetreein(8)predictsthatsegmentsotherthan[hɦ ʔ]canloseoracquireman­ner features independently of the laryngeal features. This can be shown for the lat­eralconsonantsofKlamath.Inadditiontoavoiced[l],Klamathhasaglottalized[lˀ]andvoiceless[l].Whenappearingafter[n],alateralconsonanttransfersitslat­eral articulation to the preceding segment, but without its laryngeal features, which remainbehindas [ʔ]and[h],asshownin(11).Thisagainshows that there isaconstituent in the feature tree that contains manner features but not the laryngeal features.Sincethenewsegmentisnot[+nas],thisKlamathprocesssuggeststhatthe supralar/placenodeof(8)contains[±lateral](becauseitistransferredfromthesecondtothefirstsegment)aswellas[±nasal](becauseitisdeletedfromthefirstsegment),onthecorrectassumptionthatthesefeaturesaremutuallyexclusivein the language.

(11) [nlˀ] → [lʔ][nl] → [lh]

Q133

1 Whichconstituentistransferredfromaconsonanttoapreceding[tdnl]byEnglishcoronal assimilation?

2 WhichconstituentistransferredfromaconsonanttoaprecedingnasalbyHindi nasal assimilation?

While it is therefore clear that themanner features are inside the supralar/place constituent, it is also clear that they cannot be included under the place node.InHindi,asinmanyotherlanguages,itisjusttheconsonant’splace node thatistransferredtoaprecedingnasal,notits[–nas]specification.Fromthenodelabelled supralar/placein(8),therefore,wemustsplitofftheplacenode.Sincethere have been no reports of processes that transfer [±nasal], [±continuant],[±lateral],[±strident],[±approximant],en bloc, from one segment to the next, these

Feature geometry238

features are not grouped in a manner constituent, but form separate terminal nodes dominated by supralar,asin(12).

(12) ROOT(12)consson

LAR SUPRALAR

LABIAL

PLACE

spread voice constr

distround ant back high low tense

strid approxnas cont lat

RADDORSALCOR

Singlemannerfeaturescanbeshowntotransferindependentlytoneighbour­ingsegments.Processes that transfer [+nas]werediscussed forFrisian in thissectionandEnglishinsection8.6.2.Transferof[±continuant]ismuchrarer.InAmericanEnglish,formslike[gɪbm]forgiven, [hiːdn]forheathen, [bɪdnəs]forbusiness and [wʌdnt] forwasn’t occur (Bailey1985:63).Transferof [±lateral]occurs inKlamath, as seen above.No case has beenmade for the transfer of[±strident] or [±approximant]. There have been a number of proposals thatmodify the structure of (12); however, (12)may be seen as a ‘consensus’ tree(Broe1992).

14.4 SPREADING AND DELINKING

In this section, we will deal with the way features or subconstituents of the seg­menttreespreadordelink.Inourrepresentation,segmentsarearrangedfromlefttoright,whichistheusualmetaphorfortheorderfrom‘early’to‘late’.Alongthistime axis, the segment trees are arranged much in the way the records in an old­fashionedjukeboxarearranged:arowofparalleldisks,whereadiskrepresentsasegment.Eachdiskdefinesaplaneinwhichsomeinstantiationofthefeaturetreeistobefound.Ifyouweretotakeoutadisktolookatitsrightorleftface,youwouldseesomeversionof(12).Thisisknownasanendview.

Howcananodeofsomesegmentbedominatedbytheappropriatenodeofanadjacentsegment?ImaginetwoadjacentsegmentsS1andS2.Nowmentallydrawalinefromthe[±nasal]nodeofS1tothe[±nasal]nodeofS2,andalinefromthesupralarnodeofS1tothesupralarnodeofS2.Thelinesyouhavejustdrawnare,ofcourse,tiers.Allcorrespondingnodesformtiersinthisway,sothatwecantalkaboutthe‘nasaltier’,the‘supralaryngealtier’,the‘roottier’,etc.Evidently,tiersareadjacentifnotierintervenesbetweenthem.Thisisthecasewiththeplacetierandthesupralaryngealtier,forinstance:nonodeintervenesbetweenthem.Twoadjacenttiersdefineaplane.

Feature geometry 239

Q134 Draw two feature trees in pencil, next to each other. Draw all tiers. Now erase all pencil lines that are ‘behind’ the planes nearest to you. As you will realize, your picture shows three dimensions.

Wecannowdraw,insomeplane,anassociationlinefromanynodetotheimme­diatelydominatingnodeinanadjacentsegment.Thatis,fromthe[labial]nodeofS1wecandrawanassociationlinetotheplacenodeofS2,andfromtheplace nodeofS1tothesupralaryngealnodeofS2,andsoon.Thisspreadingofnodesin the feature tree is entirely comparable with the spreading of tones, or with the association of one vowel to two moras.

Q135 Can you draw an association line from the place node of S1 to the root node of S2? If not, why not?

InHindiplace assimilation, the place node of the right­hand segment asso­ciates with the supralaryngealnodeoftheleft-handsegment.Thisisshownin(13)bytheinterruptedassociationline.Noticethatthereissomethingwrongnow:whatwasan[n]isnowspecifiedfortwosetsofplacefeatures.Whatweneedtodoisdelinktheoriginalspecification.Asusual,thisdelinkingoperationissymbolizedby ‘=’.

(13) ROOT(13)cons ROOT

son cons

SUPRALAR SUPRALAR

PLACE PLACE

[ nas]

[...] [...]

14.4.1 Writing rules

It would be a very daunting prospect indeed if all rules were to be formulated by drawing these elaborate three-dimensional pictures.Whatwewouldneed todoisdraw just thoseelements in theconfiguration thatare involved in the spread­ing and delinking, and make sure that the appropriate contextual conditions are

Feature geometry240

Q136 Distinctive features are phonetically non-abstract in the sense that they can be defined in terms of phonetic concepts. Would you say that natural feature classes are phonetically abstract?

expressed.Wewilldo this inanadaptationofanotationproposedbyClements(1985).Theadaptationconsistsinseparatingouttwoaspectsintherule.First,thestructuralchange(SC)isidentified:whichnodespreadstotheappropriatedomi­natingnodeintheadjacentsegment(withorwithoutsubsequentdelinkingoftheoriginalassociationlineinthatsegment)?Thisisshownbyspecifyingtwotiers,oneof which immediately dominates the other, together with the corresponding nodes oftwoadjacentsegments.Toexpresstheassimilation,itshowseitheranassocia­tiontowardstheleft(regressiveassimilation)oronetowardstheright(progressiveassimilation).In(14)thesearethesupralaryngeal tier and the place tier, while the right­hand placenodeisspecifiedasthespreadingone.Second,thestructuraldescription (SD) specifies what phonological properties the two segments con­cernedmusthaveforthechangetogothrough.Theseproperties(features,usually)aresimplylistedbelowthetwosegmentsconcerned.InnotspecifyingtheSDinthe tree itself, we express the fact that the tree was motivated on the basis of natural featureclassesthatareactuallymanipulatedbyphonologicalprocesses.Bycontrast,featuresintheSDmerelyconditionthoseprocesses.

Therulein(14)thusreadsasfollows:theplace node of any consonant spreads to (the supralaryngealnodeof) anasal on the left,while theplace node of thelatterisdelinked.ObservethatinourinterpretationofClements’sdisplay,theupperandlowertierarealwaysadjacent.Thehighernodelabel(here:supralar)is therefore redundant.

(14) SUPRALARNASAL ASSIMILATION

SD:

SC:

PLACE

nas cons

(14)

Similarly,Englishcoronal assimilation isgivenin(15): ‘Spreadthecoronalnode(withallthefeaturesitdominates)toanoncontinuantcoronalontheleft.’

(15)

SD:

SC:

PLACECORONAL ASSIMILATION

cont

CORONAL

(15)

Feature geometry 241

Q137 English has an optional rule assimilating [t d n] to the place of articulation of following velar and labial plosives and nasals, as shown in the data below. Formulate this English place assimila-tion in the display format.

θɪn θɪm bʊk thin bookðæt ðækkʌp that cupgʊd gʊb bɔɪ good boytεn tεmmaɪlz ten milesðæt ðætnaɪt that night

14.5 IMPLICATIONS OF UNDERSPECIFICATION

Featuretreeshaveledtomoreinsightfulformulationsofphonologicalrules.Under­specificationhashelpedtoreachthisresult.Wehavejustseenthatnotallsegmentshave all nodes. Thus, laryngeal segments have no supralaryngeal node, and the place node has only those articulator nodes that positively specify the place of articulationof thesegment,which implies, for instance, that [t]doesnothavealabial, dorsal or radicalnode(Sagey1986;McCarthy1988).Inaddition,pre­dictable information can be considered absent in the underlying representation, being suppliedbydefault rulesat theendof thederivation.Below,anumberofadvantages of such underspecification of predictable surface features are briefly outlined.

14.5.1 Default rules

Leavingpredictablefeaturesunspecifiedmayhaveasimplifyingeffectonthefor­mulationofrulesthatneutralizecontraststosomeneutral,‘default’realization.Letustake(Low)Germanfinal devoicing, which neutralizes the underlying forms /bʊnt/ ‘variegated’ and /bʊnd/ ‘union’ to [bʊnt], asanexample (cf. the inflectedforms[buntəs],[bundəs]).Theexpectedlaryngealfeaturespecificationforobstru­entsis[–voice,–spread,–constr],whichcanbesuppliedasadefaultlaryngeal node before the phonetic implementation rules start their work. final devoicing can now consist of a rule delinking any underlying laryngeal node in coda posi­tion,asin(16).

(16)

SD: Coda

SC:

ROOT

LARYNGEAL

(16) FINAL DEVOICING

Feature geometry242

A furtherquestion iswhether voicelessobstruents in a language likeGermancouldbeleftunspecifiedforthefeature[–voice]intheentirelexicon.Eventhough[±voice] iscontrastiveforobstruents inGerman,whichhascontrasts like[pain]‘pain’–[bain]‘leg’,itwouldbepossibletoleavevoicelessobstruentsunspecified,whilespecifyingvoicedobstruentsas[+voice].Thevoicelessobstruentsthenonlyacquire their [–voice]specificationat theendof thederivation,or, if the featureneeds to be referred to by some rule, as early as it is needed. This is the assump­tionofRadicalUnderspecification(Kiparsky1982a,1993;Archangeli1988),whichgoesonestepfurtherthanthetheoryofContrastiveUnderspecificationdiscussedinsection6.3.AsurveyoftheissuesisprovidedbyRoca(1994:ch.2)andSteriade(1995).

Q138 Two northern dialects of British English each have a postlexi-cal assimilation rule affecting consonant sequences across word boundaries (Wells 1982; Kerswill 1987). Neither rule occurs in the standard accent (Received Pronunciation (RP)).

1 GiveSPE formulations of the two rules.2 Explainwhyitwouldnotbepossibletoassumethatadefaultrulesuppliesthe

feature[–voice]toEnglishvoicelessobstruentsattheendofthederivation.

RP Yorkshire Durhamwhite blouse [tb] [tb] [db]wide shot [dʃ] [tʃ] [dʃ]ripe beans [pb] [pb] [bb]drab conditions [bk] [pk] [bk]black velvet [kv] [kv] [gv]five votes [vv] [vv] [vv]rough boys [fb] [fb] [vb]this village [sv] [sv] [zv]bad joke [ddʒ] [ddʒ] [ddʒ]live performance [vp] [fp] [vp]Bradford [df] [tf] [df]that night [tn] [tn] [dn]at last [tl] [tl] [dl]all true [lt] [lt] [lt]in Spain [ns] [ns] [ns]

Q139 In Klamath, the underlying consonant pairs listed in the first column are pronounced as in the second column (cf. the data in (11); after Halle and Clements 1983). Two ordered rules are sufficient to derive

Feature geometry 243

the outputs correctly. Assume that voiced sonorant consonants are unspecified for [+voice], and that a default rule supplies this feature to unspecified sonorants at the end of the derivation. Also assume that if a geminate arises as a result of a spreading operation, the two original feature trees merge under a single root node.

Underlying Derivednl llnl lhnlˀ lʔll lhllˀ lʔ

1 Stateinprosewhatthesetworulesshouldaccomplish.2 Formulatetherulesinautosegmentaldisplaynotation.(Hint:Drawpar­

tialtreesofatypicalsequencelike[nlˀ]beforeandafterthechange.Alsodrawafeaturetreeof[ll]afterthemergerunderasinglerootnode.)

Q140 In Brao-Krung, voiced, voiceless unaspirated, voiceless aspirated and preglottalized voiced plosives are contrastive in the onset, as illustrated by [dɑk] ‘walk (verb)’, [trɯː] ‘fish’, [thun] ‘season’, [ʔduːr] ‘type of flute’ (the last may vary with an implosive [ɗ]). In the coda, the only plosives that can occur are voiceless unaspirated, as illus-trated by [neːt] ‘drink (verb)’ (Keller 2001).

1 Providethefulllaryngealfeaturespecificationsforthefourtypesofini­tial plosive.

2 Which features could be supplied by default rules, assuming RadicalUnderspecification?

3 Whatunderlying laryngeal feature specificationwouldyouassume forthecodaplosives?

14.5.2 Long-distance assimilation

Underspecification equally aids the expression of long-distance assimilations.Articulatornodesfrequentlyspreadtothedominatingnodeofanapparentlynon-adjacent segment. If theNoCrossingConstraint introduced inchapter10 isvalid in the segment tree, which of course it is, then this means that the intervening segmentorsegmentsmustlackthedominatingnodeinquestion.Asanexample,considertheRussiandatain(17)(Kiparsky1985).SequencesofobstruentsinRus­sian, whether or not separated by a word boundary, agree in voicing with the right­mostone.Thisisshownin(17a).Asshownin(17b),however,itdoesnotappear

Feature geometry244

tobethecasethattheobstruentsmustbestrictlyadjacent:asonorantconsonantmay intervene.

(17) a gorod-k-a gorotka ‘littletown(gen)’mtsenskbɨl mtsenzgbɨl ‘itwasMtsensk’

b izmtsensk-a ismtsenska ‘fromMtsensk’otmzd-ɨ odmzdɨ ‘fromthebribe’

Russian devoices final obstruents, a rule that is assumed to apply before theassimilation of voice illustrated above. The spreading feature can accordingly be assumedtobe[+voice].Russianvoicingassimilationisgivenin(18).Thefea­ture[+voice]spreadsleftandanyoriginalspecificationfor[voice]islostthroughdelinking. This spreading may take place more than once, to deal with sequences of obstruents.

(18) LARYNGEAL

SD:

SC:

voice

sonson

voice

(18) RUSSIAN VOICING ASSIMILATION

However, the spreading of [+voice] from the rightmost obstruent to a non-adjacentobstruentcanonlyproceed if sonorantsdonothavea specification for[±voice]intheunderlyingrepresentation,justas[±constrictedglottis]and[±spreadglottis]areabsent.Russiandoesnotinfactcontrastvoicedandvoicelesssonorants.In(19),weassumethatthelaryngealnode is absent, though given as a bullet to identify the segment. The assimilation goes right through an intervening sonorant consonant,andthesubsequentspecificationofthefeature[+voice]insonorantsisachieved by a default rule.

(19) LARYNGEAL

SD:

SC:

voice

son sonvoiceson

(19)

Again,giventheabsenceofsupralaryngealfeaturesinglottalconsonants,wecansafely spread supralaryngealnodesacross[hʔ].Thefactthatnasalizationpro­cessesfrequentlyspreadthroughlaryngealsegmentstovowelsinadjacentsyllablesthusfinds a ready explanationunder this assumption.Another oft-cited case oflong­distance spreading relies on the absence of a coronal node in noncoronal (labialanddorsal)segments.Sanskritretroflexionturnsalveolar[n]intoretroflex[ɳ]after[ʈ ʂ ɳ ɻ].Retroflexconsonantsarecontrastively[–ant,–distr],asshownin(20)(Steriade1987;Kenstowicz1994a).

Feature geometry 245

(20) Alveolar Retroflex Prepalatalt s n ʈ ʂ ɳ ɻ c ʃ ɲ

ant + + + – – – – – – –distr – – – – – – – + + +

Wewould at first sight expect the rule to be blockedby any consonant otherthan[n].However,noncoronalconsonantsaswellasvowelsmayfreelyintervenebetweentargetandtrigger,asshowninthefirstcolumnof(21).Bycontrast,theremaynotbeacoronalconsonantotherthan[n]betweentargetandtrigger,asshownin the second column of (21). (Underlyingly voiceless obstruents are shown asvoicedbetweensonorantsegments.)

(21) Applies Blocked-na ‘present’ ɪʂ-ɳa ‘seek’ mɻd-na ‘begracious’-na ‘pass.partic.’ puːɻ-ɳa ‘fill’ bhug-na ‘bend’-aːna ‘middlepartic.’ puɻ-aːɳa ‘fill’ maɻɟ-aːna ‘wipe’

kʂubh-aːɳa ‘quake’ kʂved-aːna ‘hum’-mːana ‘middlepartic.’ kɻp-a-maːɳa ‘lament’ kɻt-a-ma:na ‘cut’

Theruletoaccomplishthisisgivenin(22),whichreliescruciallyontheabsenceof a coronal node between the two segments. If there is a coronal node, the nasal cannot be reached, as the spreading association line would cross the associa­tion line between the coronal node and the place node,asshownin(23).

(22) PLACE

SD:

SC:

CORONAL

distrnas

antdistr

(22) RETROFLEXION

(23) PLACE

SD:

SC:

CORONAL

distrnas

antdistr

(23)

Q141 In Sundanese, nasalization of vowels is predictable as a result of a phonological rule, nasalization. Consider the following data (Robins 1957; Cohn 1990).

/siar/ [siar] ‘seek’/mawur/ [mãwur] ‘spread(Active)’

Feature geometry246

/paŋliŋ/ [paŋliŋ] ‘unrecognizable’/ŋajak/ [ŋãjak] ‘sift(Active)’/ŋatur/ [ŋãtur] ‘arrange(Active)’/kana/ [kanã] ‘for the purpose’/tiʔis/ [tiʔis] ‘relax in a cool place’/niʔis/ [niʔis] ‘relaxinacoolplace(Active)’/mahal/ [mãhãl] ‘expensive’/caŋra/ [caŋra] ‘very nice weather’/ŋarahitan/ [ŋãrahitan] ‘wound(Active)’/ɲaho/ [ɲãhõ] ‘know(Active)’/ɲiar/ [ɲiãr] ‘seek(Active)’

1 Giveaprosestatementofnasalization.2 Argueonthebasisofthesedatathatthefeaturetreeforlaryngealsdoes

not include a supralaryngeal node.

Q142 Stem-initial obstruents in Bakairí, a language whose syllable struc-ture is (C)V, are realized as voiceless plosives, unless preceded by a vowel, when they are voiced (Wetzels 1997b). This is shown by [ɔkɔ] ‘bow’, [tɔ-dɔka­ge] ‘have a bow’. A stem-internal obstruent, however, can be voiced or voiceless, as shown by [itubi] ‘skin’, [ədəpigɨ] ‘heat’.

1 Assume that stem-initial obstruents are unspecified for [±voice] inunderlyingrepresentations.Whichtwodefaultrulesmustbeassumed?

The distribution of stem­internal obstruents is not free. If stem­initial conso­nants are not counted, there can be only one voiceless obstruent in the stem, whichmust be either the first or the second obstruent (again, not count­inganystem-initialconsonant).Thus,while[tɔkɔ]‘bow’,[ədəpigɨ]‘heat’and[ʊdɔdɔ]‘jaguar’arefine,formslike*[itupi],*[titupi],*[idebiko]areimpos­sible.ThisstateofaffairscanbecapturedbyassumingthatBakairístemshavea limitednumberof [±voice]melodies,much in theway thatEtunghasalimited number of tone melodies.

2 Assuming that [–son] segments are the [±voice]-bearing segments,which[±voice]melodieswouldyouassume?Wouldthefeaturespread?

3 Howare[ʊdɔdɔ]and[ədəpigɨ]accountedfor?4 Givetheunderlyingformsof[tɔkɔ]‘bow’and[tɔdɔkage]‘haveabow’,indi­

catingobstruentsunspecifiedforvoicewithcapitallettersP,T,K.Illustratetheassociationof[±voice]melodiesandtheeffectsofthedefaultrules.

5 In [kɔnɔpiɔ] ‘littlebird’, the voiceless [p] is the third consonantof thestem.Whyisn’tthisformungrammatical?

Feature geometry 247

14.5.3 Vowel harmony

Vowelharmonyisasubclassoflong-distanceassimilation.Manylanguagesexcludecertaincombinationsofvowels in theword(fora survey, seevanderHulstandvan deWeijer 1995). For instance, a Finnish word containing one of the backvowels[uoa]mustnotcontainanyofthefrontcounterparts[yœæ].Thus,pos­siblewordsare[mitæ] ‘what’, [suomi] ‘Finland’, [talo] ‘house’,while*[tymo]and*[tumæ]areill-formed.Sincethevowelsmustagreeforthefeature[±back],FinnishrepresentsacaseofBackHarmony.LanguagesmayalsooradditionallydisplayRoundHarmony,HeightHarmonyandATR-Harmony.InAkan,thefour[+ATR]vowels[ieou]mustnotco-occurwithanyoftheir[–ATR]counterparts[iεɔ ʊ].Thissitu­ationisagainreminiscentoftherestrictionontonalpatternsinEtungandMendediscussedinchapter10.Specifically,itwouldbereasonabletoassumethat,under­lyingly,backnessinFinnishisapropertyoftheword,ratherthanoftheindividualvowels.Thus,underlyingly,thevowelsofaFinnishwordlike[pysæhtyæ]‘tostop’wouldbeunspecifiedfor[±back].Instead,thewordappearsinthelexiconwitha‘floating’ feature [–back] (24a),whichassociateswith thedorsalnodesofall thevowels,asshownin(24b),whichisequivalentto(24c).(In(24a,b)theunderspecifi­cationfor[±back]isindicatedbythecapitalletters.Thus,[A]couldbe[æ]or[a],etc.)Noticethatthereislittlepointinassumingaparticulardirectionofassocia­tion, as there is only one feature to be associated. The important advantage of the lexicalrepresentationin(24a)isthatthemutualexclusionof[+back]and[–back]vowels in the same word is naturally accounted for.

(24) (24) pUsAhtUAbpUsAhtUAa c [pysæhtyæ][ back]

[ back]

The vowel systems of languages with vowel harmony frequently have one or more vowels which do not contrast for the harmony feature of the language. Thus, in additiontothetwosetsoffourvowelsmentionedabove,Akanhasa[–ATR][ɑ],forwhichthereisno[+ATR]counterpart*[a].Similarly,Finnishhasthetwo[–back]vowels[ie],forwhichthereareno[+back]counterparts*[ɯ ɤ].Dependingonthelanguage, such noncontrasting vowels behave in either of two ways. They may be opaque, in which case they stop the spreading of the feature with the opposite value and, moreover,imposetheirvalueonanyfollowingvowels.Alternatively,theyaretrans­parent, in which case they allow the spreading of the harmony feature to go right throughthem.Opaquevowelsthusblockandimposeharmony,whiletransparentvowelsneitherundergonorimposeit(GafosandDye2011).Akan[ɑ]isanopaquevowel,whileFinnish[ie]aretransparentvowels.

AswasshownbyClements(1981),theautosegmentalmodelhasnoproblemincharacterizing the behaviour of opaque vowels. If we assume that the floating fea­ture only spreads to vowels that are unspecified for the harmony feature, we predict thatthespreadingisstoppedbyanyvowelthatalreadyhasaspecification.Akanwordsaremadeupofaroot,whichhaseitherfloating[+ATR]orfloating[–ATR],plusanumberofprefixesandsuffixes.Ifthewordcontainsonlynonlowvowels,all

Feature geometry248

thevowelswillagreewiththefeaturevalueoftheroot,asillustratedby[e-bu-o]‘nest’,whichcontainsthe[+ATR]root[bu],and[ε-bʊ­ɔ]‘stone’,whichcontainsthe[–ATR]root[bʊ].However,[ɑ]mayoccurincombinationwitheither[+ATR]or[–ATR]vowels,asshownbythedisyllabicroots[pɪrɑ]‘tosweep’,[jɑrɪ]‘tobesick’,[bisɑ]‘toask’and[kɑri]‘toweigh’,wherethelasttwoare‘disharmonic’(Clements1981).Theirlexicalrepresentationsaregivenin(25).(Thevowel[ɑ]hasanumberofallophones,whichwehavenotindicated.)

(25) (25) pIrA

[ [ATR] [ATR] ATR][ [ATR] ATR][ ATR]

jArI kArIbIsAba dc

Disharmonicrootswillhaveprefixesandsuffixesthatsurfacewithoppositeval­ues for [ATR]: thefloating [+ATR] featureand theassociated featureof [ɑ]willeachspreadtotheir‘half ’oftheword,asshownin(26),where(26a,b)areequiva­lentto(26c,d),respectively.

(26)

Q143 Akan does not have words with [+ATR] vowels on both sides of [ɑ]. Does the above account explain this?

Bycontrast,thebehaviouroftransparentvowelshasbeenproblematicfortheautosegmental model. It would be reasonable to want to represent transparent vowels as underspecified for the harmony feature of the language, so that they can let thespreading feature through.However, sinceharmonizingvowelsarecru­ciallyunspecifiedfortheharmonizingfeature,aBackHarmonyrule,forinstance,cannot be expected to distinguish between harmonizing and transparent vowels during its search for empty dorsal nodes in vowels. There is no hope for a general solution based on leaving the entire dorsal node out in the case of transparent vowels. This will work in the case of languages that have only one transparent vowel,butnotinthecaseofFinnish,whichhastwotransparentvowels,[i]and[e],whichcontrastforafeaturewhichintheconsensusmodelispresentonthedorsalnode:[±high].Thesolutionheremustbeanappealtoaconstraintban­ning ill-formedvowels.ForFinnish, theconstraintwouldban [ɯ]and [ɤ], i.e.*[–round,+back,–low].Theexistenceoftransparentvowelsmaythusserveasafurther argument for abandoning rule­based descriptions in favour of constraint­based ones.

Feature geometry 249

Q144 The Finnish word game Siansaksa (‘Pig German’) produces the forms in the second column for the Finnish forms in the first (Vago 1988).

a saksalaisiahætyːtetːiːn hæksælæisiæsatuːtetːiːn‘theGermanswereattacked’

b tykːæːn urheilusta ukːaːntyrheilystæ‘I like sports’

c otsansa hiesːæ hitsansa oesːa‘in the sweat of his brow’

d pitæː kalasta kataː pilasta‘likes fish’

1 GiveananalysisofSiansaksa.2 Doesyouranalysisaccountforthefactthatthefirstwordinthegame

formin(a)isnot*[hæsælæisiæ]?3 Doesyouranalysisaccountforthefactthatthesecondforminthegame

formin(a)isnot*[satyːtetːiːn]?4 Doesyouranalysisaccountforthefactthatin(c)thefirstwordinthe

game form is not *[hitsænsæ]?

14.6 COMPLEX SEGMENTS

Unlike simplex segments, complex segments have more than one specificationeither for place of articulation or a manner feature. Three types of complex seg­ments can be distinguished.

1 Complex-place segments.A complex-place segment hasmore thanonearticulator node, and more than one articulator therefore participates in realizing the constriction specified by the features in the root. That is, a doublearticulationisproduced.Examplesare[gbkpŋm],allofwhicharelabial­dorsal. English [w] is a labial­dorsal approximant. Labial-coronal[pt]occursinBuraandMargi.In(27),alabial­coronal seg­mentisdepicted.Ofcourse,abovetheplacenodethesegmentlooksjustlike a simplex segment. This is an end view: the two articulator nodes lie in the segment plane, and are thus phonologically simultaneous.

(27) (27) PLACE

LABIAL CORONAL

Feature geometry250

2 Manner-contour segments. Manner-contour segments have a sequenceofdifferentlyvaluedoccurrencesof the samemanner feature.For exam­ple,prenasalizedstopslike[mbndŋg]are[+nas]aswellas[–nas].Since[±nasal]definesafeatureplane,thiscanonlyberepresentedbyarrangingthe two specifications in sequence, and a side view of such a segment would looklike(28).

(28)

[ nas] [ nas]

SUPRALAR

ROOT(28)

A second type of manner-contour segment has been claimed by Sagey(1986), among others, for affricates, like German [ts] as in [tsuː] ‘closed’or English [tʃ] as in [tʃuː] chew. In this view, affricates have a sequence of [–cont][+cont], insteadofbeingsimplexsegmentsthatarecharacterizedby[+strid],todistinguishthemfrom[–strid]plosives.Themostimportantargu­ment for the view that affricates are complex is that they are pronounced as complex segments, a closure followed by a narrowing that causes friction. However,thereisevidencethataffricatesfunctionasstridentplosivesandaresimple[–cont]segments.Iftheywere(also)specifiedas[+cont],theywouldbe expected to pattern with fricatives, but this appears not to be the case. Forinstance,Polish[s]isassimilatedto[ɕ]byafollowing[tɕ],asin[ɔɕtɕe]‘thistle+sg, locative’, from[ɔstɕe],buttheaffricate[ts] isunaffected,as in[ɔtstɕe]‘vinegar+sg;locative, *[ɔtɕtɕe](Rubach1994;Kim2001).Ofcourse,under either analysis, an affricate is a single segment, but the evidence appears tobeinfavourofasimplexsinglesegment(seealsoQ47andBerns2013).1

3 Secondary articulations. Consonantswitha secondaryarticulationhavetwo place specifications: one to indicate the location of the manner of articulation, and one to indicate a simultaneous vocalic articulation. That is,whilethesegmentis[+cons],andcanhaveanyconfigurationofmannerfeatures, it has an additional component specifying a vowel­like gesture of eitherthelipsorthetonguebody.Asecondaryarticulationlikelabializa­tion can combine with a labial, a coronal or a dorsal segment. The labialized segments [pw mw tw nw kw ŋw] are only some of the contrastively labial­ized segments ofNambakaengo. Similar freedomof combination can beshown to exist for palatalization, velarization and pharyngealization. There is, however, a strong tendency not to combine identical place specifications. Whilesegmentslike[xw],[χw]arerelativelycommon,labializedlabialsarerare.(Velarizedvelarshavenotbeenreported; indeed, it isnotclearhowsuchsegmentscouldbedifferentfromplainvelars.)

Feature geometry 251

The representation of secondary­articulation segments involves the addi­tion of a place node.However,unlike complex-place segments, the twoplace specifications of secondary­articulation segments differ in manner of articulation. In order to indicate which of the place features specifies the secondary articulation and which the consonantal place, one or the other mustbemarkedwithsomediacritic,asin(29),whichshowsapartialrep­resentationof[pj].2

(29)

[–cont] PLACE

SUPRALAR(29)

LABIAL DORSALsecondary

[–back]

The two place nodes in a segment with secondary articulation are not sequenced intime.AlthoughinIPAthesuperscriptsindicatinglabialization,velarization,etc.conventionally appear to the right of the consonant symbol, the two components of a secondary­articulation segment are phonologically simultaneous. That is, a side view would show a straight line.

Q145 There have been no reports of complex segments like [px] or [dŋ]. How do the representations assumed above account for this?

14.6.1 Evidence for complex segments

Likesingle segments,complexsegmentshavea singleroot node, and as such distinguish themselves from sequences of segments. This structural difference is reflected in their behaviour in a number ofways.An obvious place to lookfor such single­segment behaviour is the way they are incorporated into syllable structure.Atfirstsight,wemightbetemptedalwaystoregardphoneticsequenceslike[mb]or[nd]assequencesofphonologicalsegments.However,thefactthatinBakwiri, for example, these are the only phonetic sequences ever to appearprevocalicallyisastrongindicationthat[mb],[nd],[ŋg]aresinglesegmentsinthislanguage,asthisanalysiswillallowustorestrictitssyllablestructuretoCV.Itexplainswhynoothercombinationsofconsonantseveroccur.Underthealter­nativeanalysiswithCC-onsets,anexplanationwouldhave tobe foundfor thepeculiar restriction that the only complex onsets consist of nasals followed by plosives with the same place of articulation, even though complex onsets like

Feature geometry252

[st-]or[pj-]arecommoninother languages.Anotheralternative thatmustberejectedisanassignmentofthenasalandtheplosivetodifferentsyllables,assum­ingCV(C)asthesyllablestructure.Inthatanalysis,[limba]wouldbesyllabifiedas[lim.ba]ratherthan[li.mba],whileawordlike[mbila]wouldhavetobe[m.bi.la],withasyllabic[m].ThisanalysiscanbeempiricallyexcludedonthebasisoftheBakwiriwordgamethatappearedinQ96,inwhichthesyllablesofdisyllabicwordsarereversedand[kóndì]‘rice’istransformedto[ndíkò](Hombert1973,1986).Clearly,ifthelanguageweretoallowcodas,wewouldexpectthesyllablestructure of the word to be *[kón.dì],andtheinvertedformtobe*[díkòn].This,however, is not the correct game form.

Q146 In Ewe, verb stems reduplicate to form a present participle, noun or adjective. Determine what segments of the verb stem are redupli-cated. Why do these data (from Sagey 1986: 86) demonstrate that Ewe has compound-place segments and an affricate?

Stem Derivationfo ‘to beat’ fofo ‘beating’zo ‘to walk’ zozo ‘walking’aha+no ‘liquor+drink’ ahanono ‘liquor­drinking’fja ‘to burn’ fafja ‘burning’bja ‘to ask’ babja ‘asking’fle ‘to buy’ feflee ‘buying’ɲra ‘to rave’ ɲaɲrala ‘a raver’kplo ‘to lead’ kpokplo ‘leading’avo+sja ‘cloth+todry’ avosasja ‘cloth­drying’tsi ‘to grow’ tsitsi ‘grown up’gbla ‘to exert oneself ’ gbagblam ‘exerting oneself ’

14.6.2 The representation of palatals and palatoalveolars

Intheconsensusmodel(Broe1992),roundnessrequiresalabialarticulatornode,and tongue body specification requires a dorsal articulator node; in this account, roundvowelshavetwoarticulatornodes.Segmentslike[yøu]arethereforecom­plex, i.e. labial­dorsal.RecallfromQ145thattheplacespecificationoftheseg­ment[u]isidenticaltothatofthelabial­dorsalapproximant[w],aswellastothat of the labial­dorsal plosive [gb].Butwhile the assumptionsmade so farcorrectly characterize [u] and [w] as a natural class, they fail to characterize [i]and[j]ashavingthesameplaceofarticulation.Theassumptionthat[i j]and[uw]formnaturalclassesissupportedbyprocessesinanumberoflanguageswhichhavealternationsbetween[i]and[j]aswellasbetween[u]and[w].Insuchcases,

Feature geometry 253

[iu]appear inthesyllablepeak,while[jw]appearelsewhere.Ourtreatmentsofarhasassignedthefeature[coronal]topalatoalveolarsand(pre)palatalslike[j],but[dorsal]tovowels,includingfrontvowelslike[i].Notonlydoesthismakeitimpossibletodescribealternationsbetween[i]and[j]asinvolvingadifferenceinpositioninsyllablestructure,italsofailstoaccountforanumberofeffects[i]and[j]haveincommon(Bhat1978;LahiriandEvers1991).

1 Thesegment[j]triggersfrontingofvelars,justlike[i](andpossiblyotherfront vowels). For instance, French [kjɔsk] ‘kiosk’ has the same [k] asdoes[ki]‘who’.Thefrontingiseasilyexplainedfor[i],whichis[dorsal,–back].Butwhy should [j],which is [coronal, –ant, +dist], have thesameeffect?

2 Inmanylanguages,[k]before[i]haschangedintoprepalatal(i.e.[coro­nal])[c],andfromthereintopalatoalveolar[tʃ].Takingthisroute,Latin[kiːwiˈtaːtem]and[ˈkirkaː]endedupas[tʃitˈta]‘city’and[ˈtʃirka]‘approxi­mately’inItalian.If[i]is[dorsal],how can it affect a dorsal consonant so astobecomecoronal?

3 Many languages, of which Japanese is an example, have processes thatchange[ts]into[tɕ ɕ]before[i],asin[sitimi],pronounced[ɕitɕimi],‘pep­per’. [tɕ ɕ] are coronal. If [i] is [dorsal], thenwhy don’twe get dorsal[çikimi]instead?

BeginningwithKeating(1988),manyresearchershaveanalysedsegmentslike[j]as compound­place segments with both a coronal and a dorsal articulator. In this view,[ji]bothhavethestructureasin(30).WithJacobs(1989)andJacobsandvandeWeijer(1992),weassumethatpalatoalveolarsarealsocoronal­dorsal. This view is supported by articulatory data, in the sense that all these segments involve articulations between the forward part of the tongue and a section of the palate stretchingsome2cmbackfromthealveolarridge.ThesedatahavebeenobtainedwiththehelpofX-raypictureswithsideviewsofthetongue,aswellaswiththehelpof electropalatography, a technique for recording contact areas of the roof of the mouth(Recasensetal.1995).

(30)

naslat

voicecont

SUPRALAR

(30)

back lowhigh

DORSAL

ant dist

CORONAL

ROOTconsson

LAR

PLACE

Feature geometry254

This feature analysis accounts for all three earlier observations. First, Frenchvelar frontingof[k]to[k]intheenvironmentof[ji]canbeaccountedforasspreadingofthe[±back]nodeof[j]andfrontvowelstothedorsal node of the velar.This isshownin(31),wherethe[–back] featureof[j i]willassociatewiththe dorsalnodeof[k].(Dependingonthelanguage,therulemayhavetobecon­strainedtopreventotherfrontvowelsfromtriggeringit.)

(31) DORSAL

SD:

SC:

± back

–back

(31) VELAR FRONTING

Thisanalysisalsoaccountsfortheoccurrenceof[ç] insteadofvelar[x] intheenvironmentof[j]andfrontvowels,asinGreek[ˈçilia]‘thousand’versus[xoˈros]‘dance’.Thatis,thefrontingof[x]to[ç]amountstothereplacementof[+back]with[–back]inthedorsalfricative.3

The assimilationof [k] to [c] or [tʃ] canbedescribed as the spreadingof thecoronal­dorsalnodeof[ji]tothesupralarnodeofthevelar,asshownin(32),wheretheleft-handsegment,[kg],receivestheplacenodeofthefollowing[ij].

(32) SUPRALAR

SD:

SC:

PLACE

DORSALnascont

(32)

conshighback

And, to complete the story, (33)describes the assimilationof [t s] to [tɕ ɕ] or[tʃ ʃ]asthespreadingoftheplacenodefrom[ji]totheconsonant.Thisanalysispredictsthatthecoronalspecificationof[tdsz]becomes[–ant],whichiscorrect.Onthedistinctionbetweenprepalatal[ɕ]andpalatoalveolar[ʃ],seethenextsec­tion.Note that thesespreadingoperationsonlyaccountfortheplaceassimilation.Themannerfeatureswillinmanycasesbesuppliedseparately.Forinstance,iftheproductof(33)is[c],nothingneedstobedone,butifitis[tʃ],[+strid]needstobeadded(cf.Kim2001).

(33)

SD:

SC:

–cons+high–back

COR–son

(33) SUPRALAR

PLACE

Feature geometry 255

Q147 Standard Chinese contrasts three coronal fricatives [s ʂ ɕ] in the onset, as shown in (a,b,c). Before [i], only [ɕ] occurs; after [s ʂ], syllabic approximants occur which are homorganic with the fric-atives, and which historically go back to [i] (Lee-Kim 2014).

a [sa55] ‘spill’ d [sɹ55] ‘tear(Verb)’b [ʂa55] ‘murder’ e [ʂɻ55] ‘wet’c [ɕa55] ‘foolishly’ f [ɕi55] ‘West’

1 Whatisthecontrastivespecificationunder[coronal]forthethreefrica­tives(cf.section6.3)?

2 IftheV-positionhasasegmentalnodewith[−cons,+son],cantheapprox­imantconsonantbedescribedbythespreadingofasinglenode?

3 Whileitisclearthattheapproximantconsonantisinthenucleusofthesyllable, the question may arise whether this segment is a vowel or a syl­labicconsonant.Assumingthatthisdistinctionismeaningful,canitbecharacterizedasaconsonantinourfeaturemodel?

4 Whatdoesthisassumptionimplyforthecontrastbetween[i]and[j]?5 Whatdoesthisassumptionimplyforthecontrastbetween[u]and[w]?

14.6.3 Distinguishing among palatoalveolars

Thecoronal-dorsal representationof [ʃ ӡ tʃ dӡ]wouldappear tooffer a satisfactorydescription of these consonants, which are typically pronounced with the tongue tip raised towards, or articulating with, the alveolar ridge and a simultaneous raising of the tongue blade and forward part of the front towards the postalveolar area of the hard pal­ate. The coronal node is responsible for specifying the first type of articulation, while the dorsalnodeisresponsibleforthesecondaspect.Ofcourse,thecoronal node candominateanumberofspecificationsforthefeatures[±anterior]and[±distributed].Eventhoughdifferentpalatoalveolarconsonantsrarelycontrastinthesamelanguage,the implicit prediction that there are different types of palatoalveolar consonants is cer­tainlycorrect.First,a[–ant,+distr]articulationwillhavethebladearticulatingwiththepostalveolar region, while allowing the tongue tip to be behind the lower teeth. This is typicallythetonguepositionfor[j]and[i].Inthecaseof[+cons]segments,thistypeissometimesreferredtoas‘prepalatal’or‘alveopalatal’,symbolized[ɕ ʑ ʨʥ].Thefrica­tivesandaffricatesmaycontrastwithotherpalatoalveolars.InSerbo-Croat,forinstance,[tɕ] contrastswith[tʂ],forwhichthecoronalcontactispostalveolarorretroflex,i.e. [–ant,–distr],asshownby[tʂâːr]‘enchantment’vs[tɕâːr]‘profit’.Bothcoronal­dor­salaffricatesaredistinctfromthecoronalaffricate[ts],asshownby[ˈtsuːriti]‘toleak’vs[ˈtɕuːriti]‘toblow(ofwind)’and[tsêh]‘guild,union’vs[tʂêh]‘Czechman’.Similarly,Polishcontrastsretroflex[ʂ tʂ]withalveopalatal[ɕ tɕ], as in prosze [ˈprɔʂε]‘please’,czas [tʂas]‘time’vsprosie [ˈprɔɕε]‘piglet’andciasny [ˈtɕasnɨ]‘tight’.

Feature geometry256

English[ʃ ӡ tʃ dӡ]are[+ant,–distr]:inadditiontotheraisedfrontand/orblade,there is an apical articulation with the alveolar ridge. This representation accounts for an apparent asymmetry in the application of coronal assimilationinEnglish.Recallthatthisprocesscausescoronal[tdnl]toadopttheplaceofarticulationofafollowingcoronalconsonant,sothatthe[t]ofthat is dental in that thing, alveolar in that tin and postalveolar in that trip (cf.section14.2).Theapparentasymmetryisthatbeforepalatoalveolarstheplaceofarticulationof[tdnl]isunaffected.Eventhough the forward part of the front may show some anticipatory raising towards thehardpalate,[tdnl]retaintheircontactbetweenthetipandthealveolarridge,justastheydowhenavowelfollows,sothatthelocationofthecrownduring[t]inthat egg isthesameasthatduring[t]inthat chore, andthe[n]inkeen officer is no differentfromthe[n]inkeen judge.4Ouranalysisreadilyaccountsforthesedata: [tʃ dӡ] have a coronal nodewith [+ant, –distr], which explains the articulatoryactionofthetonguetip,inadditiontoadorsalnodewith[–back],whichexplainstheraisingofthefrontofthetongue.Sinceitisonlythecoronal node that spreads toapreceding [tdn l], theeffecton theirplaceofarticulation is entirelyvacu­ous,becausetheseconsonantsarealready[+ant,–distr].Bycontrast,ifweweretoanalysepalatoalveolarsassimplexcoronalsegments,aswasdoneinchapter6,theywouldhave tobe [–ant,+distr],a featuralcharacterization thatwouldbeneces­sary to explain the largely postalveolar place of articulation. This would incorrectly predict that coronal assimilation causes a considerable retraction of the place of articulationof[tdnl]beforepalatoalveolars.Wesummarizeourfeatureanalysisofcoronal, coronal­dorsal and dorsal consonantsinTable14.1.

Table 14.1Featureanalysisofcoronal, coronal­dorsal and dorsal consonants. stands for the presence of a unary feature.

Coronal Coronal-dorsal dorsal

Apical [ts] Prepalatal [ij ɲ ɕ] Eng. [ ʃ ӡ tʃ dӡ] [k ç ] [kx]

Coronal Anterior + – + Distributed – + –Dorsal Back – – – +

Q148 Tahltan has the following system of oral coronal [+cons] segments (Shaw 1991):

t tɬ tθ ts tʃd dl dð dz dӡt’ tɫ’ tθ’ ts’ tʃ’

ɬ θ s ʃl ð z ӡ

Feature geometry 257

The dental, alveolar and palatoalveolar consonants in the last three col­umns are involved in coronal harmony. Withintheword,therightmostconsonant in this group determines the place of articulation of all preceding consonants in thesamegroup.This is illustratedwith[s] ‘firstpersonsin­gularsubject’,andwith[θi]‘firstpersondualsubject’inthefollowingdata.The(a)examplesshowthesesuffixesintheirunassimilatedform,whilethe(b) examplesshowtheeffectofcoronal harmony. The consonants in the first two columns above are transparent to coronal harmony: they nei­thertriggernorblocktheprocess.Thisisillustratedbythe(c)examples.Tar­get consonants are underlined.

Underlying [s] Underlying [θi]a εsk’aː ‘I’m gutting fish’ dεθigɪtɬ ‘we threw it’

εsdan ‘I’m drinking’ θiːtθædi ‘we ate it’nεstεɬ ‘I’m sleepy’ naθibaːtɬ ‘we hung it’

b hudiʃtʃa ‘I love them’ iʃitʃotɬ ‘we blew it up’ɬεnεʃtʃuːʃ ‘I’m folding it’ uʃidӡε ‘we are called’εθduːθ ‘I whipped him’ xasiːdεts ‘we plucked it’mεθεθεq ‘I’m wearing’ dεsidzεl ‘we shouted’

c εdεdεθduːθ ‘I whipped myself ’ dεsit’ʌs ‘we are walking’jaʃtɬ’εtʃ ‘I splashed it’ mεʔεʃit’otʃ ‘we are breast­

feeding’

1 How can we tell from these data that the two suffixes have differentunderlyingforms?

2 What are the three place specifications of the consonants involved incoronalharmony?

3 Whyisitimpossibletoviewcoronalharmonyastheleftwardspreadingofa[coronal]nodetoconsonantsthatareunspecifiedfor[coronal]?

4 Isitpossibletoviewcoronalharmonyasarulethatspreadsa[coro­nal]node to consonants thathave a [coronal]node,while skippingconsonantsthatdonot?Motivateyouranswer.

14.7 CONCLUSION

Many languages place restrictions on the combination of different values of thesamephonological features in theword. Suchword-baseddistributional restric­tions include vowel harmony and consonant harmony, as well as word­based distributional patterns of nasalization and obstruent voicing. The autosegmentali­zation of features, together with the underspecification of certain features in cer­tain segments, allows us to describe these patterns with the help of many­to­one associations, a more insightful solution than is possible if segments are represented

Feature geometry258

as unstructured lists of features, as was usual in the linear theory of SPE.Tofullyaccount for transparent segments, however, a rule­based description will need to fall back on constraints. This can be seen as an argument for abandoning the con­cept of rules that literally change one representation into another, and for adopting atheorylikeOptimalityTheory,whichreliesonconstraintsthatsimplytelluswhatsurfacerepresentationsmustlooklike.Wealsosawthatthefeaturetreeallowsforanatural representation of complex segments. It was shown that, in addition to obvi­ouslycomplexsegments like[mb], [kp]and[tw],palatalandpalatoalveolarcon­sonantscanfruitfullybe interpretedas(complex)coronal­dorsal segments, a type of representation that makes it possible to describe various place assimilations involving palatoalveolar articulations.

In this book, we have seen an impressive array of the ways in which languages organizethesoundresourcesthatareavailabletothem.Wehaveseenhowphono­logical theory has responded to this variation by constructing a representational framework in which phonological content is slotted into a hierarchically organized array of structural positions. These structures and the nature of the phonological contentinthemintrigueusbecauseoftheshiftingperspectivetheyofferbetweenhow similar they are across languages and how extraordinarily different. Phono­logicaltheoryhasbeenquitesuccessfulinlayingouttheserepresentations.Yetwehave also seen how theoretical accounts of the relations between different levels ofrepresentationarenot fullysatisfactory.Such issuesare likely tobeaddressedinbehavioural andneurologicalpsycholinguistic research.Equally,wemay lookforward to insights that a continued expansion of the data base will bring through perceptive analyses of undescribed and underdescribed languages.

NOTES

1 Tocaterforthetwokindsofevidence,Lombardi(1990)proposedanunderlyingrep­resentationwithanunordered[–cont]/[+cont]specification,whichbecomeorderedonthe surface.

2 InClementsandHume(1995),thefeaturetreeincludesavocalic place node in addi­tion to a consonantal place node, each of which dominates the usual articulator nodes.Afeaturelike coronal is interpreted as a front articulation in the case of vowels and as a coronal pronunciation in the case of consonants, while dorsal is used to char­acterizebackvowelsordorsalconsonants.Aseparatenodeisusedtospecifythedegreeofopeningofvowels.SeealsoKenstowicz(1994a).

3 The[ç]isclassedasacomplexcoronal-dorsalbyKeating(1988).Inadditiontomakingitimpossiblefor[ʃ]tobecoronal-dorsal,thischaracterizationdoesnotdojusticetothemore retracted articulatory contact for [ç]than for the palatoalveolars and[j].Also,theimpliedpredictionthatthefrontingof[k]isadifferentprocessfromthefrontingof[x]has not been confirmed.

4 Clements(1985)notedthatthearticulationof[tdn],butnotof[l],changesbeforepalato-alveolars.AtleastforstandardBritishEnglish,webelievethereisnosuchdistinction.

References

ThislistincludesreferencestoworksintheKeytotheQuestions,availableintheeResourceatwww.routledge.com/9781138961425.

Abercrombie,David(1967).Elements of general phonetics.Edinburgh:EdinburghUniversityPress.

Abolhasanizadeh,Vahideh;Bijankhan,Mahmood;Gussenhoven,Carlos(2012).ThePersianpitchaccentanditsretentionafterthefocus.Lingua122:1380–1394.

Agostini,PàuluM.(1995).L’usu di a nostra lingua.Biguglia,Corsica:PrivatelyprintedbySammarcelli.

Aitken, Adam J. (1981).The Scottish vowel-length rule. InM. Benskin&M.L. Samuels(eds.),So meny people longages and tonges: Philological essays in Scots and Mediaeval English presented to Angus McIntosh.Edinburgh:Authors.131–157.

———(1984).Scottishaccentsanddialects.InP.Trudgill(ed.),Language in the British Isles. Cambridge:CambridgeUniversityPress.94–114.

Andersen,Torben(1987).ThephonemicsystemofAgarDinka.Journal of African Languages and Linguistics9:1–27.

Anderson,StephenR.(1974).The organization of phonology.NewYork:AcademicPress.Archangeli,Diana(1988).Underspecification in Yawelmani phonology and morphology.New

York:GarlandPress.Árnason,Kristján(1985).Icelandicwordstressandmetricalphonology.Studia Linguistica

39:93–129.Arvaniti,Amalia;Ladd,D.Robert;Mennen,Ineke(2000).Whatisastarredtone?Evidence

fromGreek.InM.Broe&J.B.Pierrehumbert(eds.),Papers in laboratory phonology V: Acquisition and the lexicon.Cambridge:CambridgeUniversityPress.119–131.

Ashby,Patricia(2011).Understanding phonetics.Abingdon:Routledge.Baerman,Matthew(1999).The evolution of fixed stress in Slavic.Munich:Lincom.Bailey,Charles-JamesN.(1985).English phonetic transcription.Dallas:SummerInstituteof

Linguistics&Arlington:UniversityofTexas.Baković,Eric(2004).Unboundedstressandfactorialtypology.InJ.McCarthy(ed.),Opti-

mality Theory in phonology: A reader.London:Blackwell.202–214.Bao,Zhiming(1999).The structure of tone.NewYork&Oxford:OxfordUniversityPress.Barak,Barış;Vogel,Irene(2001).ThephonologicalwordandstressinTurkish.Phonology

18:315–360.Bartels,Christine(1999).The intonation of English questions and statements: A compositional

interpretation.NewYork:GarlandPress.Bateman,Janet(1990).Iausegmentalandtonephonology.NUSA: Linguistic Studies of Indo-

nesian and Other Languages in Indonesia32:29–42.Beckman,JillN.(1988).Positional faithfulness.PhD,UniversityofMassachusetts,Amherst.Beckman,MaryE.(1986).Stress and non-stress accent.Dordrecht:Foris.

References260

Bell,ArthurJ.(2003).Gemination,degeminationandmoraicstructureinWolof.Working Papers of the Cornell Phonetics Laboratory15:1–68.

Berg,Thomas(1998).Linguistic structure and change: A perspective from language processing. Oxford:ClarendonPress.

Bermúdez-Otero,Ricardo (forthcoming).Stratal Optimality Theory.Oxford:OxfordUni­versity Press.

Berns,Janine(2013).Friction between phonetics and phonology.Utrecht:LOT.Bhat,D.N.S.(1978).Ageneralstudyofpalatalization.InJ.H.Greenberg,C.A.Ferguson&

E.Moravcsik(eds.),Universals of human language II: Phonology.Stanford,CA:StanfordUniversityPress.47–92.

Bickmore,Lee(2007).Cilungu phonology.Stanford:CSLI.Blevins, Juliette (1995).The syllable in phonological theory. In J.A.Goldsmith (ed.),The

handbook of phonological theory.Cambridge,MA:Blackwell.206–244.Boersma,Paul(1998).Functional phonology.Utrecht:LOT.Bolinger,DwightL.(1986).Intonation and its parts: Melody in spoken English.Stanford,CA:

StanfordUniversityPress.Bolognesi,Roberto(1998).The phonology of Campidanian Sardinian. TheHague:Holland

AcademicGraphics.Booij,Geert(1977).Generatieve fonologie van het Nederlands. Utrecht:Spectrum.———(1985).Coordinationreductionincomplexwords:Acaseforprosodicphonology.In

H.vanderHulst&N.Smith(eds.),Advances in non-linear phonology.Dordrecht:Foris.143–160.

———(1988).Ontherelationbetweenlexicalandprosodicphonology.InP.M.Bertinetto&M.Lopocaro(eds.),Certamen phonologium: Papers from the 1987 Cortona phonology meeting.Turin:Rosenberg&Sellier.389–418.

———(1995).The phonology of Dutch.Oxford:OxfordUniversityPress.———(1996).Cliticizationasprosodicintegration:ThecaseofDutch.Linguistic Review13:

219–242.———(1997).Non-derivationalphonologymeetslexicalphonology.InI.M.Roca(ed.),Der-

ivations and constraints in the lexicon.Oxford:ClarendonPress.261–288.Bowden, John;Hajek, John(1996).Taba. Journal of the International Phonetic Association

26:55–57.Boxwell,Helen;Boxwell,Maurice(1966).Weriphonemes.Papers in New Guinea Linguistics

5:77–93.Broe,Michael(1992).Anintroductiontofeaturegeometry.InG.J.Docherty&D.R.Ladd

(eds.),Papers in laboratory phonology II: Gesture, segment, prosody.Cambridge:Cam­bridgeUniversityPress.149–165.

Browman,CatherineP.;Goldstein,Louis(1989).Articulatorygesturesasphonologicalunits.Phonology6:201–251.

Brunelle,Marc(2009).ToneperceptioninNorthernandSouthernVietnamese.Journal of Phonetics37:79–96.

Busenitz,RobertL.;Busenitz,Marilyn J. (1991).Balantakphonologyandmorphophone­mics.InJ.N.Sneddon(ed.),Studies in Sulawesi linguistics II.Jakarta:NUSA(UniversitasKatolikIndonesiaAtmaJaya).29–47.

Butcher,AndrewR.(2006).AustralianAboriginal languages:Consonant-salientphonolo­gies and the ‘place-of-articulation imperative’. In J. Harrington & M. Tabain (eds.),Speech production: Models, phonetic processes, and techniques. New York: PsychologyPress.187–210.

Catford,JohnC.(1988).A practical introduction to phonetics.Oxford:ClarendonPress.

References 261

Chao,Yuen-Ren(1930).A system of tone­letters. Le Maître Phonétique45:24–27.Charette,Monik(1989).Theminimalityconditioninphonology.Journal of Linguistics25:

159–187.Chatterji,SunitiK.(1921).Bengali phonetics. Bulletin of the School of Oriental and African

Studies2:1–25.Chen,Matthew(1987).ThesyntaxofXiamentonesandhi.Phonology Yearbook4:109–149.Chen,Yiya;Gussenhoven,Carlos(2008).EmphasisandtonalimplementationinStandard

Chinese.Journal of Phonetics36:724–746.——— (2015). Shanghai Chinese. Journal of the International Phonetic Association 45:

321–338.Chesswas,JohnD.(1963).The essentials of Luganda.London:OxfordUniversityPress.Cho,Taehong(2004).Prosodicallyconditionedstrengtheningandvowel-to-vowelcoarticu­

lationinEnglish.Phonetica32:141–176.Chomsky,Noam;Halle,Morris(1968).The sound pattern of English.NewYork:Harper&

Row.Clements,GeorgeN.(1981).Akanvowelharmony:Anonlinearanalysis.InG.N.Clements

(ed.),Harvard studies in phonology.Vol.2.Bloomington:IndianaUniversityLinguisticsClub.108–177.

———(1985).Thegeometryofphonologicalfeatures.Phonology Yearbook2:223–252.——— (1986). Compensatory lengthening and consonant gemination in LuGanda. In

W.L.Wetzels&E.Sezer(eds.),Studies in compensatory lengthening.Dordrecht:Foris.36–77.

———(1990).Theroleof thesonoritycycle incoresyllabification.InJ.Kingston&M.E.Beckman(eds.),Papers in laboratory phonology I: Between the grammar and physics of speech.Cambridge:CambridgeUniversityPress.283–333.

——— (1993). Lieu d’articulation des consonnes et des voyelles: une théorie unifiée.InB. Laks&A.Rialland (eds.),L’architecture et la géométrie des représentations pho-nologiques.Paris:CNRS.101–145.

———(1999).Affricatesasnoncontouredstops.InO.Fujimura,B.Joseph&B.Palek(eds.),Item order in language and speech.Prague:CharlesUniversityPress.271–299.

———(2004).Featureeconomyinsoundsystems.Phonology20:287–333.———;Ford,K.(1979).Kikuyutoneshiftanditssynchronicconsequences.Linguistic Inquiry

10:179–210.———;Hume,E.V.(1995).Theinternalorganisationofspeechsounds.InJ.A.Goldsmith

(ed.),The handbook of phonological theory.Cambridge,MA:Blackwell.245–306.———;Keyser,SamuelJ.(1983).CV-phonology.Cambridge,MA:MITPress.Cohn,Abigail (1990).Phonetic and phonological rules of nasalization. PhD,University of

California,LosAngeles.Connell,Bruce;RobertLadd,D.(1990).AspectsofpitchrealisationinYoruba.Phonology

7:1–29.Coupez,André(1969).Uneleçondelinguistique.Africa-Tervuren5:1–5.Cutler, Anne (2011). Native listening: Language experience and the recognition of spoken

words.Cambridge,MA:MITPress.Davis,Stuart(2011).Geminates.InM.vanOostendorp,M.Ewen,C.Colin,E.V.Hume&

K. Rice(eds.),The Blackwell companion to phonology. 5volumes.Malden,MA&Oxford:WileyBlackwell.837–859.

deGraaf,Tjeerd(1990).Grasuglas.FonetischecontrastenJapans-Nederlands.Tabu20:47–57.deLacy,Paul(ed.)(2007).The Cambridge handbook of phonology.Cambridge:Cambridge

UniversityPress.

References262

DiCanio, Christian T. (2012). Coarticulation between tone and glottal consonants in Itu-nyoso Trique. Journal of Phonetics 40: 162–176.

Dogil, Grzegorz; Williams, Briony (1999). The phonetic manifestation of word stress. In H. van der Hulst (ed.), Word prosodic systems in the languages of Europe. Berlin: Mouton de Gruyter. 273–334.

Duanmu, San (2000). The phonology of Standard Chinese. Oxford: Oxford University Press.Dupoux, Emmanuel; Peperkamp, Sharon; Sebastián-Gallés, Nuria (2001). A robust method

to study stress ‘deafness’. Journal of the Acoustical Society of America 110: 1606–1618.Edmondson, Tom; Bendor-Samuel, John T. (1966). Tone patterns of Etung. Journal of Afri-

can Languages 5: 1–6.Elbert, Samuel H.; Pukui, Mary Kawena (1979). Hawaiian grammar. Honolulu: University

Press of Hawaii.Elimelech, Baruch (1976). A tonal grammar of Etsako UCLA Working Papers in Phonetics

35: 1–170.Elordieta, Gorka (1997). Accent, tone and intonation in Lekeitio Basque. In F. Martínez-Gil

& A. Morales-Front (eds.), Issues in the phonologies of the major Iberian languages. Wash-ington, DC: Georgetown University Press. 3–78.

Evans, Nick (1995). Current issues in the phonology of Australian languages. In J.A. Gold-smith (ed.), The handbook of phonological theory. Cambridge, MA: Blackwell. 723–761.

Everett, Daniel (1986). Pirahã. In D.C. Derbyshire & Geoffrey K. Pullum (eds.), Handbook of Amazonian languages. Vol. 1. Berlin: Mouton de Gruyter. 200–326.

Ewen, Colin J. (1995). Dependency relations in phonology. In J.A. Goldsmith (ed.), The handbook of phonological theory. Cambridge, MA: Blackwell. 570–585.

Firth, John R. (1957). Phonological features of some Indian languages. In J.R. Firth (ed.), Papers in linguistics 1934–1951. Oxford: Oxford University Press. 121–138.

Flemming, Edward (1995). Auditory representation in phonology. PhD, University of Califor-nia, Los Angeles.

Friberg, Timothy; Friberg, Barbara (1991). Notes on Konjo phonology. In J.N. Sneddon (ed.), Studies in Sulawesi linguistics II. Jakarta: NUSA (Universitas Katolik Indonesia Atma Jaya). 71–115.

Fry, Dennis (1955). Duration and intensity as physical correlates of linguistic stress. Journal of the Acoustical Society of America 35: 765–769.

——— (1958). Experiments in the perception of stress. Language and Speech 1: 120–152.Furby, Christine E. (1974). Garawa phonology. Papers in Australian Linguistics 7: 1–11.Gafos, Adamantios I.; Dye, Amanda (2011). Vowel harmony: Opaque and transparent

vowels. In M. van Oostendorp, M. Ewen, C. Colin, E.V. Hume & K. Rice (eds.), The Blackwell companion to phonology. 5 volumes. Malden, MA & Oxford: Wiley Blackwell. 2164–2189.

Giegerich, Heinz J. (1992). Onset maximisation in German: The case against resyllabification rules. In P. Eisenberg, K.H. Ramers & H. Vater (eds.), Silbenphonologie des Deutschen. Tübingen: Gunter Narr. 134–171.

Gimson, Alfred C. (1989). An introduction to the pronunciation of English (2008; 4th edn) revised by A. Cruttenden. London: Arnold.

Goldsmith, John A. (1976). Autosegmental phonology. Bloomington: Indiana University Lin-guistics Club.

Gomez-Imbert, Elsa; Kenstowicz, Michael (2000). Barasana tone and accent. International Journal of American Linguistics 66: 419–463.

Gordon, Matthew J. (2000). Re-examining default-to-opposite stress. Proceedings of the Berkeley Linguistic Society 26: 101–112.

References 263

Grønnum,Nina(1992).The groundworks of Danish intonation: An introduction.Copenha­gen:MuseumTusculanum.

Gurevich,Naomi(2004).Lenition and contrast: The functional consequences of certain pho-netically motivated sound changes.NewYork&London:Routledge.

Gussenhoven,Carlos (1986).Englishplosiveallophonesandambisyllabicity.Gramma10:119–141.

———(1993).TheDutchfootandthechantedcall.Journal of Linguistics29:37–63.———(1994).Englishstressinlexicalphonology.InW.U.Dressler,M.Prinzhorn&J.J.Ren­

nison(eds.),Phonologica 1992: Proceedings of the 7th International Phonology Meeting. Turin:Rosenberg&Sellier.87–96.

———(2000).Theboundarytonesarecoming:Onthenonperipheralrealizationofbound­ary tones. InM.Broe& J.B. Pierrehumbert (eds.),Papers in laboratory phonology V: Acquisition and the lexicon.Cambridge:CambridgeUniversityPress.132–151.

——— (2009). Vowel duration, syllable quantity, and stress in Dutch. In K. Hanson & S.Inkelas(eds.),The nature of the word: Essays in honor of Paul Kiparsky.CambridgeMA&London,UK:MITPress.181–198.

———(2011).SententialprominenceinEnglish.InM.vanOostendorp,M.Ewen,C.Colin,E.V.Hume&K.Rice(eds.),The Blackwell companion to phonology.5volumes.Malden,MA&Oxford:WileyBlackwell.2778–2806.

———(2017).Zwara(Zuwārah)Berber.Journal of the International Phonetic Association. ———;Peters,Jörg(2004).AtonalanalysisofCologneSchärfung.Phonology21:252–285.———;Wang,Lu;Hendrix,Louis(2016).YuhuanWutonesandhiandtonecontrastmain­

tenance. Speech Prosody 2016:1052–1055.http://www.isca-speech.org/archive/sp2016/Haas,MaryR.(1977).TonalaccentinCreek.InL.Hyman(ed.),Studies in stress and accent.

Southern California Occasional Papers in Linguistics 4. Los Angeles: University ofSouthernCalifornia.195–208.

Hale,Mark;Reiss,Charles(2008).The phonological enterprise.Oxford:OxfordUniversityPress.Halle,Morris(1959).The sound pattern of Russian.TheHague:Mouton.———(1985).Speculationsabouttherepresentationsofwordsinmemory.InV.Fromkin

(ed.),Phonetic linguistics: Essays in honor of Peter Ladefoged.Orlando, FL:AcademicPress.101–104.

———;Clements,GeorgeN.(1983).Problem book in phonology.Cambridge,MA:MITPress.———;Kenstowicz,Michael (1991).The free element condition and cyclic vs. noncyclic

stress. Linguistic Inquiry22:457–501.———;Mohanan,KaruvannurP.(1985).SegmentalphonologyofmodernEnglish.Linguistic

Inquiry16:57–116.Hanssen,Judith(2017).Non-standardmelodiesandmelodypreferencesindialect-accented

Dutch.InJ.Hanssen(ed.), Regional variation in the realization of intonation contours in the Netherlands. Utrecht:LOTdissertations.125–155.

Harris,JamesW.(1983).Syllable structure and stress in Spanish.Cambridge,MA:MITPress.Harris,John(1994).English sound structure.Oxford:Blackwell.———(2004).Releaseofthecaptivecoda:Thefootasadomainofphoneticinterpretation.

InJ.Local,R.Ogden&R.Temple(eds.),Phonetic interpretation: Papers in laboratory phonology VI.Cambridge:CambridgeUniversityPress.103–129.

———(2013).Wide-domainr-effectsinEnglish.Journal of Linguistics49:329–365.Harry,Otelemate;Hyman,LarryM.(2014).Phrasalconstructiontonology:ThecaseofKala­

bari. Studies in Language38:649–689.Hayes,Bruce(1981).A metrical theory of stress rules.PhD,MIT,Cambridge,MA.Distrib­

utedbyIndianaUniversityLinguisticsClub,Bloomington,IN.

References264

Hayes, Bruce (1981). A metrical theory of stress rules. PhD, MIT, Cambridge, MA. Distrib-uted by Indiana University Linguistics Club, Bloomington, IN.

——— (1985). Iambic and trochaic rhythm in stress rules. Berkeley Linguistics Society 13: 429–446.

——— (1987). A revised parametric metrical theory. Northeastern Linguistics Society 17: 274–289.

——— (1989a). Compensatory lengthening and moraic phonology. Linguistic Inquiry 20: 253–306.

——— (1989b). The prosodic hierarchy in meter. In P. Kiparsky & G. Youmans (eds.), Rhythm and meter. Orlando, FL: Academic Press. 201–260.

——— (1990). Precompiled phonology. In S. Inkelas & D. Zec (eds.), The phonology-syntax connection. Chicago: University of Chicago Press. 85–108.

——— (1994). ‘Gesture’ in prosody: Comments on the paper by Ladd. In P. Keating (ed.), Papers in laboratory phonology III: Phonological structure and phonetic form. Cambridge: Cambridge University Press. 64–75.

——— (1995). Metrical stress theory: Principles and case studies. Chicago: Chicago University Press.

———; Lahiri, Aditi (1991). Bengali intonational phonology. Natural Language and Linguis-tic Theory 9: 46–99.

Hayward, Katrina; Hayward, Richard J. (1992). Amharic. Journal of the International Pho-netic Association 22: 48–52.

Hellmuth, Sam (2007). The relationship between prosodic structure and pitch accent distri-bution: Evidence from Egyptian Arabic. The Linguistic Review 24: 291–316.

Himmelmann, Nikolaus P. (1991). Tomini-Tolitoli sound structures. In J.N. Sneddon (ed.), Studies in Sulawesi linguistics II. Jakarta: NUSA (Universitas Katolik Indonesia Atma Jaya). 49–70.

Holst, Tara; Nolan, Francis (1995). The influence of syntactic structure on [s] to [ʃ] assimila-tion. In B. Connell & A. Arvaniti (eds.), Papers in laboratory phonology IV: Phonology and phonetic evidence. Cambridge: Cambridge University Press. 315–333.

Hombert, Jean-Marie (1973). Speaking backwards in Bakwiri. Studies in African Linguistics 4: 227–235.

——— (1986). Word games: Some implications for analysis of tone and other phonological constructs. In J.J. Ohala & J. Jaeger (eds.), Experimental phonology. Orlando, FL: Aca-demic Press. 175–186.

Hooper, Joan B. (1976). An introduction to natural generative phonology. New York: Aca-demic Press.

Hualde, José I. (1992). Catalan. London: Routledge.Huisman, Ronald; Huisman, Roberta; Lloyd, Joy (1981). Angaatɨha syllable patterns. In P.M.

Healey (ed.), Angan languages are different: Four phonologies. Huntington Beach, CA: Summer Institute of Linguistics. 51–62.

Hyman, Larry M. (1985). A theory of phonological weight. Dordrecht: Foris.——— (2001). Tone systems. In M. Haspelmath, E. König, W. Oesterreicher & W. Raible

(eds.), Language typology and language universals: An international handbook. Vol. 2. Berlin & New York: De Gruyter. 1367–1380.

——— (2006). Word-prosodic typology. Phonology 23: 225–257.——— (2007). Kuki-Thaadow: An African tone system in Southeast Asia. UC Berkeley Pho-

nology Lab Annual Report: 1–19.——— (2009). How (not) to do phonological typology: The case of pitch-accent. Language

Sciences 31: 213–223.

References 265

———(2016).Lexicalvsgrammaticaltone:Sortingoutthedifferences.Proceedings of the 5th International Symposium on Tonal Aspects of Languages (TAL).Buffalo,NY.http://www.isca-speech.org/archive/TAL_2016/

———;Ngunga,ArmindaS.A.(1994).Onthenon-universalityoftonalassociation‘conven­tions’:EvidencefromCiyao.Phonology11:25–68.

Inkelas,Sharon(1989).Prosodic constituency in the lexicon.NewYork:GarlandPress.Itô,Junko;Mester,Armin(1986).ThephonologyofvoicinginJapanese:Theoreticalconse­

quences for morphological accessibility. Linguistic Inquiry17:49–73.Jacobs,Haike(1989).Nonlinear studies in the historical phonology of French.PhD,Nijmegen.———;Gussenhoven,Carlos(2000).Loanphonology:Perception,salience,thelexiconand

OT.InJ.Dekkers,F.vanderLeeuw&J.vandeWeijer(eds.),Optimality Theory: Phono-logy, syntax and acquisition. Oxford:OxfordUniversityPress.193–210.

———;vandeWeijer,Jeroen(1992).Ontheformaldescriptionofpalatalisation.InR.Bok-Bennema&R.vanHout(eds.),Linguistics in the Netherlands 1992.Amsterdam:Benja­mins.123–135.

Jakobson,Roman(1968).Child language, aphasia, and phonological universals.TheHague:Mouton.———(1990).On language.EditedbyL.R.Waugh&M.Monville-Burston.Cambridge,MA:

HarvardUniversityPress.———;Fant,Gunnar;Halle,Morris(1952).Preliminaries to speech analysis.Cambridge,MA:

MITPress.Jeanne,LaVerneM.(1982).SomephonologicalrulesofHopi.International Journal of Ameri-

can Linguistics48:245–270.Jones,Daniel(1967).The phoneme: Its nature and use.London:CambridgeUniversityPress.———(1977).English pronouncing dictionary (14thedn).Extensivelyrevisedandeditedby

A.C.Gimson.London:Dent&Sons.Jones,RobertB.(1961).Karen linguistic studies.Berkeley&LosAngeles:UniversityofCali­

fornia Press.Joos,Martin(1942).AphonologicaldilemmainCanadianEnglish.Language18:141–144.Kager,René(1999).Optimality Theory.Cambridge:CambridgeUniversityPress.Kahn,Daniel(1976).Syllable-based generalizations in English.PhD,MIT,Cambridge,MA.

DistributedbyIndianaUniversityLinguisticsClub,Bloomington,IN.Kahn,SameerudDowla(2010).Bengali(BangladeshiStandard).Journal of the International

Phonetic Association40:221–225.Kaisse,EllenM.(1985).Connected speech.London:AcademicPress.Kaye,Jonathan(1975).Afunctionalexplanationofruleorderinginphonology.Parasession

on Functionalism.CLS2:244–252.———(1989).Phonology: A cognitive view.Hillsdale,NJ:LaurenceErlbaum.———(1990).WhateverhappenedtoDialectB?InJ.Mascaró&M.Nespor(eds.),Progress

in grammar: GLOW essays for Henk van Riemsdijk.Dordrecht,NL:Foris.259–263.Keating,PatriciaA.(1988).Palatalsascomplexsegments:X-rayevidence.UCLA Working

Papers in Phonetics69:77–91.———(1990).Phoneticrepresentationsinagenerativegrammar.Journal of Phonetics18:321–334.———;Cho,Taehong;Fougeron,Cécile;Hsu,Chai-Shune(2004).Domain-initialarticula­

torystrengtheninginfourlanguages.InJ.Local,R.Ogden&R.Temple(eds.),Papers in laboratory phonology VI.Cambridge:CambridgeUniversityPress.145–163.

Keller,CharlesE.(2001).Brao-Krungphonology.The Mon-Khmer Studies Journal31:1–13.Kennedy,Robert(2008).BugotuandChekeHoloreduplication:Indefenceoftheemergence

of the unmarked. Phonology25:61–82.

References266

Kenstowicz,Michael(1994a).Phonology in generative grammar.Oxford:Blackwell.———(1994b).Cyclicvs.noncyclicconstraintevaluation.MIT Working Papers in Linguistics

21:11–42.———;Kisseberth,C.(1977).Topics in phonological theory.NewYork:AcademicPress.Kerswill,PaulE.(1987).LinguisticvariationinDurham.Journal of Linguistics23:25–49.Key,MaryR.(1968).Comparative Tacanan phonology.TheHague:Mouton.Kim,Hyunsoon(1997).The Phonological representation of affricates: Evidence from Korean

and other languages.PhD,CornellUniversity,Ithaca,NY.———(2001).Aphoneticallybasedaccountofphonologicalstopassibilation.Phonology8:

81–108.Kingston, John(2005).ThephoneticsofAthabaskan tonogenesis. InS.Hargus&K.Rice

(eds.),Athabaskan phonology.Amsterdam:Benjamins.137–184.———;Diehl,RandyL.(1994).Phoneticknowledge.Language70:419–454.Kiparsky,Paul(1968).Linguisticuniversalsandlinguisticchange.InE.Bach&R.Harms

(eds.),Universals in linguistic theory.NewYork:Holt.170–202.———(1973a).‘Elsewhere’inphonology.InS.Anderson&P.Kiparsky(eds.),A festschrift for

Morris Halle.NewYork:Holt.93–106.———(1979).Metricalstructureassignmentiscyclic.Linguistic Inquiry10:421–442.———(1982a).Fromcyclicphonologytolexicalphonology.InH.vanderHulst&N.Smith

(eds.),The structure of phonological representations.PartI.Dordrecht:Foris.131–175.———(1982b).Explanation in phonology.Dordrecht:Foris.———(1985).Someconsequencesoflexicalphonology.Phonology Yearbook2:82–138.———(1991).Catalexis.Unpublishedmanuscript.StanfordUniversityandWissenschaftkol­

legzuBerlin.———(1993).Blockinginnonderivedenvironments.InS.Hargus&E.Kaisse(eds.),Studies

in lexical phonology.SanDiego,CA:AcademicPress.277–313.———(2015).StratalOT:AsynopsisandFAQs.InY.E.H.Hsiao&L.-H.Wee(eds.),Captur-

ing phonological shades within and across languages.Newcastle-upon-Tyne,UK:Cam­bridgeScholarsPublishing.2–42.

Kluender,KeithR.;Diehl,RandyL.;Wright,BeverlyA. (1988).Vowel-lengthdifferencesbeforevoicedandvoicelessconsonants:Anauditoryexplanation.Journal of Phonetics 16:153–170.

Koefoed, Geert A.T. (1979). Paradigmatische invloeden op fonetische processen. InT. Hoekstra&H.vanderHulst(eds.),Glot special.VakgroepNederlands:UniversityofLeiden,AkademischePers.51–72.

Koster,Jan(1975).DutchasanSOVlanguage.Linguistic Analysis1:111–136.Kraehenmann,Astrid(2001).SwissGermanstops:Geminatesallovertheword.Phonology

18:109–145.Kubozono,Haruo(1993).The organization of Japanese prosody.Tokyo:Kurosio.———(1999).Moraandsyllable.InN.Tsujimura(ed.),The handbook of Japanese linguistics.

Oxford:Blackwell.31–61.———;Itô, Junko;Mester,Armin(1997).Non-finality inJapanesephonology.Proceedings

of the16th International Congress of Linguists.Amsterdam&NewYork:Elsevier.89–93(cd-rom1998).

Ladd,D.Robert(2008).Intonational phonology(2ndedn).Cambridge:CambridgeUniver­sity Press.

Ladefoged,Peter(1971).Preliminaries to linguistic phonetics.Chicago:UniversityofChicagoPress.

———;Johnston,Keith(2010).A course in phonetics(6thedn).Boston,MA:Wadsworth.

References 267

———;Maddieson,Ian(1996).The sounds of the world’s languages.Oxford:Blackwell.Lahiri,Aditi;Evers,Vincent(1991).Palatalizationandcoronality.InC.Paradis&J.-F.Prunet

(eds.),The special status of coronals: Internal and external evidence.Cambridge:Cam­bridgeUniversityPress.79–100.

———;Marslen-Wilson,WilliamD. (1991).Themental representationof lexical form:Aphonological approach to the recognition lexicon. Cognition38:245–294.

Laniran,YetundeO.;Clements,GeorgeN.(2003).Downstepandhighraising:InteractingfactorsinYorubatoneproduction.Journal of Phonetics31:203–250.

Lass,Roger(1976).English phonology and phonological theory.Cambridge:CambridgeUni­versity Press.

Laughren, Mary (1984). Tone in Zulu nouns. In G.N. Clements & J. Goldsmith (eds.),Autosegmental studies in Bantu tone.Dordrecht:Foris.183–234.

Laver,John(1994).Principles of phonetics.Cambridge:CambridgeUniversityPress.Leben,WiliamR.(1973).Suprasegmental phonology.PhD,MIT,Cambridge,MA.———(1976).ThetonesinEnglishintonation.Linguistic Analysis2:69–107.———(1978).Therepresentationoftone.InV.Fromkin(ed.),Tone: A linguistic survey.New

York:AcademicPress.177–219.Lee,HyunBok(1993).Korean.Journal of the International Phonetic Association23:28–31.Lee-Kim,Sang-Im(2014).RevisitingMandarin‘apicalvowels’:Anarticulatoryandacoustic

study. Journal of the International Phonetic Association44:261–282.Levelt,WillemJ.M.(1989).Speaking from intention to articulation.Cambridge,MA:MITPress.Lewis,M.Paul;Simons,GaryF.;Fennig,CharlesD.(eds.)(2016).Ethnologue: Languages of

the World(19thedn).Dallas,TX:SILInternational.Onlineversion:http://www.ethno­logue.com

Li,Bing(1996).Tungusic vowel harmony: Description and analysis.TheHague:HollandAca­demicGraphics.

Liberman,Mark;Pierrehumbert,JanetB.(1984).Intonationalinvarianceunderchangesinpitch range and length. InM.Aronoff&R.Oehrle (eds.),Language sound structure. Cambridge,MA:MITPress.157–233.

———;Prince,Alan(1977).Onstressandlinguisticrhythm.Linguistic Inquiry8:249–336.Lindau,Mona(1978).Vowelfeatures.Language54:541–563.Lloret,Maria-Rosa(1995).TherepresentationofglottalsinOromo.Phonology12:257–280.Lombardi,Linda(1990).Thenonlinearorganizationoftheaffricate.Natural Language and

Linguistic Theory8:375–425.Maddieson,Ian(1984).Patterns of sounds.Cambridge:CambridgeUniversityPress.Mascaró,Joan(1976).Catalan phonology.PhD,MIT,Cambridge,MA.Maskikit-Essed,Raechel;Gussenhoven,Carlos(2016).Nostress,nopitchaccent,nopro­

sodicfocus:ThecaseofAmboneseMalay.Phonology33:353–389.McCarthy,JohnJ.(1985a).Formal problems in Semitic phonology and morphology.NewYork:

Garland.———(1985b).SpeechdisguiseandphonologicalrepresentationinAmharic.InH.vander

Hulst&N.Smith(eds.),Advances in non-linear phonology.Dordrecht:Foris.305–312.———(1988).Featuregeometryanddependency:Areview.Phonetica45:84–108.———(1998).Prosodicmorphology.InA.Spencer&A.M.Zwicky(eds.),The handbook of

morphology.Oxford:Blackwell.283–305.———(2007).Hidden generalizations: Phonological opacity in Optimality Theory.London:

Equinox.———;Prince,Alan(1993).Generalizedalignment.InG.Booij&J.vanMarle(eds.),Year-

book of morphology.Amsterdam:Springer.79–154.

References268

———(1998).Prosodicmorphology. InA. Spencer&A.Zwicky (eds.),The handbook of morphology. Oxford:OxfordUniversityPress.283–305.

McCawley,JamesD.(1986).Todaytheworld,tomorrowphonology.Phonology Yearbook3:27–43.

McQueen,JamesM.;Cutler,Anne(1997).Cognitiveprocessesinspoken-wordrecognition.InW.J.Hardcastle&J.Laver(eds.),The handbook of phonetic sciences.Oxford:Blackwell.566–585.

Meyer,AntjeS.;Belke,Eva (2007).Word formretrieval in languageproduction. InM.G.Gaskell(ed.),The Oxford handbook of psycholinguistics.Oxford:OxfordUniversityPress.471–487.

Michaud,Alexis(2004).Finalconsonantsandglottalization:NewperspectivesfromHanoiVietnamese.Phonetica61:119–146.

Mielke,Jeff(2008).The emergence of distinctive features.Oxford:OxfordUniversityPress.Mohanan,KaruvannurP.(1986).The theory of lexical phonology.Dordrecht:Reidel.———(1992).Describingthephonologyofnon-nativevarietiesoflanguage.World Englishes

11:111–128.Morais,José;Cary,Luz;Alegria,Jésus;Bertelson,Paul(1979).Doesawarenessofspeechasa

sequenceofphonesarisespontaneously?Cognition7:323–331.Moulton,WilliamG.(1962).ThevowelsofDutch:Phoneticanddistributionalclasses.Lin-

gua6:294–312.Nadeu,Marianna(2016).Phoneticandphonologicalvowel reduction inCentralCatalan.

Journal of the International Phonetic Association46:33–60.Nagano-Madsen, Yasuko (1993). Phrase-final intonation in West Greenlandic Eskimo.

Working Papers in General Linguistics and Phonetics40:145–155.LundUniversity.Nespor,Marina;Vogel,Irene(1986).Prosodic phonology.Dordrecht:Foris.———(1998).Stressdomains.InH.vanderHulst(ed.),Word prosodic systems in the lan-

guages of Europe.Berlin:MoutondeGruyter.117–159.Nettle,Daniel(1995).Segmentalinventorysize,wordlength,andcommunicativeefficiency.

Linguistics33:359–367.Newman,Paul(1970).A grammar of Tera.Berkeley&LosAngeles:UniversityofCalifornia

Press.Nieuweboer,Rogier(1998).The Altai dialect of Plautdiitsch (West-Siberian Mennonite Low

German).PhD,UniversityofGroningen.Odden,David(1986).OntheObligatoryContourPrinciple.Language62:353–383.Oftedal,Magne(1985).Lenition in Celtic and in Insular Spanish: The secondary voicing of

stops in Gran Canaria.Oslo:Universitetsforlaget.Ohala,JohnJ.(1986).Aconsumer’sguidetophonologicalevidence.Phonological Yearbook

3:3–26.———(1989).Soundchangeisdrawnfromapoolofsynchronicvariation.InL.E.Breivik&

E.H.Jahr(eds.),Language change: Contributions to the study of its causes.Berlin:MoutondeGruyter.173–198.

Ohala,Manjari(1983).Aspects of Hindi phonology.NewDelhi:MotilalBanarsidass.Ortega-Llebaria,Marta;Prieto,Pilar(2011).AcousticcorrelatesofstressinCentralCatalan

andCastilianSpanish.Language and Speech54:73–97.Özçelik,Öner(2014).Prosodicfaithfulnesstofootedges:ThecaseofTurkishstress.Phonol-

ogy31:229–269.Padgett,Jaye(2008).Glides,vowels,andfeatures.Lingua 118:1937–1955.Pasch,Helma(1986).Die Mba-Sprachen. Die Nominalklassensysteme und die genetische Glie-

derung einer Gruppe von Ubangi-Sprachen. Hamburg:Buske.

References 269

Pet,WilliamJ.A.(1979).AnotherlookatSurinamArawakphonology.International Journal of American Linguistics45:323–331.

Peters,Jörg(2008).ToneandintonationinthedialectofHasselt.Linguistics46:983–1018.Phelps,ElaineF.(1975).Iterationanddisjunctivedomainsinphonology.Linguistic Analysis

1:137–172.Pierrehumbert,JanetB.(1980).The phonetics and phonology of English intonation.PhD,MIT,

Cambridge,MA.———(1990).Phonologicalandphoneticrepresentation.Journal of Phonetics18:375–394.———(2002).Word-specificphonetics.InC.Gussenhoven&N.Warner(eds.),Laboratory

phonology7.Berlin:MoutondeGruyter.101–139.———;Beckman,MaryE.(1988).Japanese tone structure.Cambridge,MA:MITPress.———;Hirschberg,Julia(1990).Themeaningofintonationalcontoursintheinterpretation

ofdiscourse.InP.Cohen,J.Morgan&M.Pollack(eds.),Intentions in communication. Cambridge,MA:MITPress.271–311.

Pike,EuniceV.;Wistrand,Kent(1974).Step-upterracetoneinAcatlánMixtex.InR.Brend(ed.),Advances in tagmemics.Amsterdam:North-HollandPublishing.83–104.

Poeppel,David(2003).Theanalysisofspeechindifferenttemporalintegrationwindows:Cer­ebral lateralization as asymmetric sampling in time. Speech Communication41:245–255.

Port, Robert;O’Dell,Michael (1986).Neutralization of syllable-final voicing inGerman.Journal of Phonetics13:455–471.

Post,Brechtje(2000).Pitchaccents,LiaisonandthephonologicalphraseinFrench.Probus 12:127–164.

Postal,Paul(1968).Aspects of phonological theory.NewYork:Harper&Row.Prince,Alan;Smolensky,Paul(1993).Optimality Theory: Constraint interaction in genera-

tive grammar.TechnicalReport2.NewBrunswick,NJ:RutgersUniversityCenter forCognitiveScience.

Pulleyblank,Douglas(1986).Tone in lexical phonology. Dordrecht:Reidel.Rahmani, Hamed; Rietveld, Toni; Gussenhoven, Carlos (2015). Stress ‘deafness’ reveals

absence of lexical marking of stress or tone in the adult grammar. PLOS ONE. doi:10.1371/journal.pone.0143968

———(forthcoming).Post-focalandfactivedeaccentuationinPersian.Glossa.Recasens,Daniel;Fontdevila, Jordi;Pallarès,MariaD.(1995).Aproductionandperceptual

accountofpalatalization.InB.Connell&A.Arvaniti(eds.),Papers in laboratory phonology IV: Phonology and phonetic evidence.Cambridge:CambridgeUniversityPress.265–281.

Reetz,Henning;Jongman,Allard(2011).Phonetics: Transcription, production, acoustics, and perception.Chichester:Wiley-Blackwell.

Remijsen,Bert(2007).LexicallycontrastivestressandlexicaltoneinMa’ya.InC.Gussen­hoven&N.Warner(eds.),Laboratory phonology7.Berlin:Mouton.585–614.

Riad,Tomas(2014).The phonology of Swedish.Oxford:OxfordUniversityPress.Rietveld,A.C.M. (Toni); Loman,HenkG. (1985). /h/, Stemloos of stemhebbend.Glot 8:

275–285.Rischel,Jørgen(1974).Topics in West Greenlandish phonology.Copenhagen:Akademisk.Roberts,John(1987).Amele.London:CroomHelm.Robins,RobertH. (1957).Vowelnasality inSundanese:Aphonologicalandgrammatical

study.InJ.R.Firth(ed.),Studies in linguistic analysis.Oxford:Blackwell.87–103.Roca,IgnacioM.(1994).Phonological theory.London:Routledge.Ross,MalcolmD.(1993).TonogenesisintheNorthHuonGulfchain.InJ.A.Edmondson

& K.J. Gregerson (eds.), Tonality in Austronesian languages. Honolulu: University ofHawaiiPress.133–153.

References270

Rubach,Jerzy(1990).FinaldevoicingandcyclicsyllabificationinGerman.Linguistic Inquiry 21:79–94.

———(1994).AffricatesasstridentstopsinPolish.Linguistic Inquiry25:119–143.———(1996).ShorteningandambisyllabicityinEnglish.Phonology13:197–237.———(2000).GlideandglottalstopinsertioninSlaviclanguages:ADOTanalysis.Linguistic

Inquiry31:271–317.Sacks,Oliver(1989).Seeing voices.Berkeley:UniversityofCaliforniaPress.Sagey,Elizabeth.C.(1986).The representation of features and relations in non-linear phonol-

ogy.PhD,MIT,Cambridge,MA.Schachter, Paul;Otanes, FeT. (1972).Tagalog reference grammar. Berkeley:University of

CaliforniaPress.Schane,SanfordA.(1971).Thephonemerevisited.Language47:503–521.Seiler,Walter(1985).Imonda, a Papuan language.Canberra:AustralianNationalUniversity.Selkirk,Elisabeth (1978).Onprosodic structureand its relation to syntactic structure. In

T. Fretheim(ed.),Nordic prosody II.Trondheim:TAPIR.111–140.———(1980).Prosodicdomainsinphonology:Sanskritrevisited.InM.Aronoff(ed.),Junc-

ture.Saratoga,CA:AnmaLibri.107–129.———(1982).Syllables.InH.vanderHulst&N.Smith(eds.),The structure of phonological

representations.PartII.Dordrecht:Foris.337–383.———(1984).Phonology and syntax: The relation between sound and structure.Cambridge,

MA:MITPress.———(1986).Onderiveddomainsinsentencephonology.Phonology Yearbook3:371–405.———(2000).Theinteractionofconstraintsonprosodicphrasing.InM.Horne(ed.),Prosody:

Theory and experiment: Studies presented to Gösta Bruce.Amsterdam:Kluwer.231–261.———(2011).Thesyntax-phonologyinterface.InJ.Goldsmith,J.Riggle&A.C.L.Yu(eds.),

The handbook of phonological theory(2ndedn).Oxford:Wiley-Blackwell.435–484.———;Lee,SeunghunJ.(2015).Constituencyinsentencephonology:Anintroduction.Pho-

nology32:1–18.Senft,Gunter (1986).Kilivila: The language of the Trobriand islanders.Berlin:Moutonde

Gruyter.Sezer,Engin(1983).Onnon-finalstressinTurkish.Journal of Turkish Studies5:61–69.Shattuck-Hufnagel,Stefanie(1992).Theroleofwordstructureinsegmentalserialordering.

Cognition42:213–259.Shaw,Patricia(1991).Consonantharmonysystems:Thespecialstatusofcoronalharmony.

InC.Paradis&J.-F.Prunet(eds.),The special status of coronals: Internal and external evidence.Cambridge:CambridgeUniversityPress.125–157.

Silverman,Daniel(1992).Multiplescansionsinloanwordphonology.Phonology9:289–328.Sluijter,AgaathM.C.;vanHeuven,VincentJ.(1996).Spectralbalanceasanacousticcorre­

late of stress. Journal of the Acoustical Society of America100:2471–2485.Sluyters,Willibrord(1990).Lengthandstressrevisited:Ametricalaccountofdiphthongi­

zation, vowel lengthening, consonant gemination and word­final vowel epenthesis in modern Italian. Probus2:65–102.

Smolensky,Paul(1996).Onthecomprehension/productiondilemmainchildlanguage.Lin-guistic Inquiry27:720–731.

Spajić,Siniša;Ladefoged,P.;Bhaskarararo,P.(1996).ThetrillsofToda.Journal of the Inter-national Phonetic Association26:1–21.

Stampe,David(1980).A dissertation on natural phonology.NewYork:GarlandPress.Stankiewicz,Edward(1993).The accentual patterns of the Slavic languages.Stanford:Stan­

fordUniversityPress.

References 271

Smolensky, Paul (1996). On the comprehension/production dilemma in child language. Lin-guistic Inquiry 27: 720–731.

Spajić, Siniša; Ladefoged, P.; Bhaskarararo, P. (1996). The trills of Toda. Journal of the Inter-national Phonetic Association 26: 1–21.

Stampe, David (1980). A dissertation on natural phonology. New York: Garland Press.Stankiewicz, Edward (1993). The accentual patterns of the Slavic languages. Stanford: Stan-

ford University Press.Steriade, Donca (1987). Redundant values. In A. Bosch, B. Need & E. Schiller (eds.), Par-

asession on autosegmental and metrical phonology. Chicago: Chicago Linguistic Society. 339–362.

——— (1995). Underspecification and markedness. In J.A. Goldsmith (ed.), The handbook of phonological theory. Cambridge, MA: Blackwell. 114–174.

Stevens, Kenneth N. (1997). Articulatory–acoustic–auditory relationships. In W.J. Hardcas-tle & J. Laver (eds.), The handbook of phonetic sciences. Oxford: Blackwell. 462–506.

Ternes, Elmar (1973). The phonemic analysis of Scottish Gaelic. Hamburg: Helmut Buske.Thelwall, Robin; Sa’adeddin, M. Akram (1990). Arabic. Journal of the International Phonetic

Association 20: 37–39.Tiersma, Pieter M. (1985). Frisian reference grammar. Dordrecht: Foris.Todd, Terry L. (2002). A grammar of Dimili, also known as Zaza. Stockholm: Iremet.Topintzi, Nina (2010). Onsets: Suprasegmental and prosodic behaviour. Cambridge: Cam-

bridge University Press.Tranel, Bernard (1981). Concreteness in generative phonology. Berkeley: University of Cali-

fornia Press.Trommelen, Mieke; Zonneveld, Wim (1983). Generatieve fonologie van het Nederlands. Muiden:

Coutinho.Tyler, Stephen A. (1969). Koya: An outline grammar. Berkeley: University of California Press.Vago, Robert M. (1988). Vowel harmony in Finnish word games. In H. van der Hulst & N. Smith

(eds.), Features, segmental structure and harmony processes II. Dordrecht: Foris. 185–205.Valenzuela, Pilar; Gussenhoven, Carlos (2013). Shiwilu (Jebero). Journal of the International

Phonetic Association 43: 97–106.van den Berg, Berend (1975). De morfonologische regels voor de vorming van het verklein-

woord in het dialect van de stad Utrecht. Taal en Tongval 27: 95–102.van den Broecke, Marcel P.R. (1976). Hierarchies and rank orders in distinctive features.

Assen: Gorcum.van der Hulst, Harry (1984). Syllable structure and stress in Dutch. Dordrecht: Foris.——— (1985). Ambisyllabicity in Dutch. In H. Bennis & F. Beukema (eds.), Linguistics in the

Netherlands. Dordrecht: Foris. 57–66.——— (2008). The Dutch diminutive. Lingua 118: 1288–1306.———; Goedemans, Rob; van Zanten, Ellen (2010). A survey of word accentual patterns in the

languages of the world. Berlin: Walter de Gruyter. Mouton.———; van de Weijer, Jeroen M. (1995). Vowel harmony. In J.A. Goldsmith (ed.), The hand-

book of phonological theory. Cambridge, MA: Blackwell. 495–534.van de Weijer, Jeroen M. (2002). An optimality theoretical analysis of the Dutch diminutive.

In H. Broekhuis & P. Fikkert (eds.), Linguistics in the Netherlands 2002. Amsterdam: Benjamins. 199–209.

Venneman, Theo (1972). On the theory of syllabic phonology. Linguistische Berichte 18: 1–18.——— (1988). Preference laws for syllable structure and the explanation of sound change. Ber-

lin: Mouton de Gruyter.

References272

Vergnaud, Roger J.; Halle, Morris (1978). Metrical structures in phonology. Unpublished manuscript, MIT, Cambridge, MA.

Volk, Erez (2011). Mijikenda phonology. PhD, Tel Aviv University, Tel Aviv.Vos, Frits (1963). Dutch influences on the Japanese language. Lingua 12: 341–388.Wang, H. Samuel; Derwing, Bruce C. (1986). More on English vowel shift: The back vowel

question. Phonology Yearbook 3: 99–116.Warner, Natasha; Jongman, Allard; Sereno, Joan; Kemps, Rachèl (2004). Incomplete neutral-

ization and other sub-phonemic durational differences in production and perception: Evidence from Dutch. Journal of Phonetics 32: 251–276.

Wells, John C. (1982). Accents of English. London: Arnold.——— (2008). Longman pronunciation dictionary (3rd edn). Harlow: Pearson-Longman.Wetzels, W. Leo (1997a). Bakairí and the feature ‘voice’. Número Especial do Boletim da Asso-

ciação Brasileira de Lingüística (ABRALIN) am Homenagem a Aryon Rodrigues. Maceió, Brazil: Universidade Federal de Alagoas (UFAL).

——— (1997b). The lexical representation of nasality in Brazilian Portuguese. Probus 9: 203–232.

Xu, Yi (2006). Tonal alignment, syllable structure and coarticulation: Toward an integrated model. Journal of Italian Linguistics 18: 125–157.

Yip, Moira (1993). Cantonese loanword phonology and Optimality Theory. Journal of East Asian Linguistics 2: 261–291.

Zec, Draga; Inkelas, Sharon (1990). Prosodically constrained syntax. In S. Inkelas & D. Zec (eds.), The phonology-syntax connection. Chicago: University of Chicago Press. 365–378.

Zhang, Jie (2002). The effects of duration and sonority on contour tone distribution: A typologi-cal survey and formal analysis. New York: Routledge.

Zsiga, Elizabeth C. (1997). Features, gestures, and Igbo vowel assimilation: An approach to the phonology/phonetics mapping. Language 73: 227–274.

——— (2012). The sounds of language: An introduction to phonetics and phonology. New York: Wiley & Sons.

Language index

ThelanguageindexbelowwasmodelledontheoneinLaver(1994)andcomposedwiththehelpoftheEthnologue(Lewisatal.2016)andtheGlottolog(www.glottolog.org).

Language Family Geographical area Page numbers!Xũ Khoisan Angola,Namibia 4,37,41,45Akan Niger-Congo,Kwa Ghana(South) 78,247Amele Trans-NewGuinea,

GumPapuaNewGuinea

(Madangprovince)182–3

Amharic Semitic Egypt,Ethiopia 85,132Angaatɨha Trans-NewGuinea,

AnganPapuaNewGuinea

(Morobeprovince)39–40

Arabic Semitic Algeria,Bahrain,Chad,Egypt,Ethiopia,Iraq,Jordan,Kuwait,Lebanon,Libya,Morocco,Oman,Qatar,SaudiArabia,Somalia,Sudan,Syria,Tunisia,UnitedArabEmirates,Yemen

29,33,38,46,79,130,132,149,178

CaireneArabic

Egypt(Cairo) 184

GulfArabic Bahrain,Kuwait,Oman,Qatar,UnitedArabEmirates

29

Arawak Arawakan Surinam(CommewijneandParadistricts)

31

Bakairí Carib Brazil(MatoGrossostate) 45,246Bakwiri Niger-Congo,Bantu Cameroon(South) 38,157,252Balantak Austronesian,

Malayo-Polynesian

Indonesia(EastCentralSulawesi)

105

Barasana Tucanoan Colombia(Amazonasdepartment)

160–1,163

Basque Isolate Spain(Basquecountry) 204Bengali(akaBangla)

Indo-European,Indo-Aryan

Bangladesh,India 77,178–9,204,207

Brao-Krung Austro-Asiatic,Brao-Kravet

Cambodia(Ratanakiriprovince),Laos

243

Bugotu Austronesian,Oceanic

SolomonIslands(Isabelprovince)

140

Bura Afro-Asiatic,Chadic Nigeria(Bornostate) 249

Language index274

Language Family Geographical area Page numbersBurmese Sino-Tibetan,

Tibeto-BurmanMyanmar 45

Cantonese Sino-Tibetan,Chinese

China(South),HongKong

51

Catalan Indo-European,Romance

Spain(Catalonia,Valenciana,BalearicIslands)

31,39,136,175

Cavineña Tacanan Bolivia(ElBenidepartment)

194,197

Chama(akaEseEja)

Tacanan Bolivia(ElBenidepartment)

197

Chipewyan(akaDene)

Athabaskan Canada(SouthCentral) 32

Chuave Trans-NewGuinea,Chimbu

PapuaNewGuinea(Chimbuprovince)

41,46

Cilungu Niger-Congo,Bantu

Zambia(Northeast) 164

Ciyao Niger-Congo,Bantu Malawi,Mozambique,Tanzania

154

Corsican Indo-European,Romance

France(Corsica) 71,81,224

Creek Muskogean USA(Alabama;Oklahoma)

195

Czech Indo-European,Slavic

CzechRepublic,Slovakia 31,39,46,255

Danish Indo-European,Germanic

Denmark 81

Dinka Nilo-Saharan SouthSudan 22–3,39,137,165Dizi Afro-Asiatic Ethiopia(Southwest) 46Dutch Indo-European,

GermanicBelgium,France,

Netherlands,NetherlandsAntilles,Surinam

12–15,22,29,31,39,43,46,51–2,62,67–8,72,74–5,79,81,87–9,91,96,101–2,104,106–8,110–12,114–15,118–19,127–8,139–40,166,169,179,186–7,198,201,206,209–12,219,221,232

KaatsheuvelDutch

Netherlands(Kaatsheuvel)

219–20,222

UtrechtDutch

Netherlands(Utrecht) 107–8

ZeelandDutch

Netherlands(Zeelandprovince)

166

Language index 275

Language Family Geographical area Page numbersEnglish Indo-European,

GermanicAustralia,Canada,Hong

Kong,India,Ireland,NewZealand,Pakistan,Singapore,SouthAfrica,UK,USA,Zimbamwe;linguafranca

2,4,6–7,9,11–15,17,19–23,26–7,29–31,33–4,37,39–40,42,45–6,48–51,53–7,59,61,65,69–71,73,75,77–8,88,90,92,94–104,108–9,111,113–15,118,120–4,127–8,135–6,139,142–3,145,147–9,161,166–7,175–8,180,185–6,196,198,200,204,208,213,217,222,233–5,240–2,249,256,258

AmericanEnglish

USA 22,26,34,39,70,94,114,121,123,142–3,147,208,238

BritishEnglish

UK 20–2,26,34,39,77,95–6,101,122–3,139,145,206,218,242

CanadianEnglish

Canada 217,222

DurhamEnglish

UK(Durham) 242

LondonEnglish

UK(London) 235

NorthernIrish English

UK(NorthernIreland) 118

NYCEnglish USA(NewYorkCity) 120OldEnglish Extinct 37,90,97,142–3,234Scottish

EnglishUK(Scotland) 29,71,73,114

SingaporeEnglish

Singapore 99–100

YorkshireEnglish

UK(Yorkshire) 242

Estonian Uralic,Finnic Estonia,Latvia 39Etsako Niger-Congo,

Benue-CongoNigeria(Edostate) 155–6,159

Etung Niger-Congo,Bantu Cameroon,Nigeria 151–4,158–60,163,233,246–7

Ewe Niger-Congo,Kwa Ghana,Togo 71,252

Fijian Austronesian,Oceanic

Fiji 38

Finnish Uralic,Finnic Finland 7,31–2,39,216,247–9French Indo-European,

RomanceAlgeria,Belgium,

Canada,France,Luxemburg,Monaco,Morocco,Switzerland;linga franca

4,12,23,26–7,29–30,37–8,48,58,88,90,97–8,116–17,119,124,127,133,146,178–9,213,219,222,253–4

Frisian Indo-European,Germanic

Germany(Northeast);Netherlands(Friesland)

73–4,135,236–8

Language index276

Language Family Geographical area Page numbersGaelic Indo-European,

CelticUK(OuterHebrides) 133

Garawa Australian,Garawan Australia(NorthernTerritory;Queensland)

181–2,193–4

German Indo-European,Germanic

Austria,Belgium,Brazil,Germany,Italy,Liechtenstein,Luxembourg,Namibia,Poland,Romania,Ukraine,Russia,Slovakia,Switzerland

2–3,7,13,23,26–7,29–32,38–40,42,50–1,71,78–9,81,87,101,115–16,127,135,149,166,169,176,185,207,209,241–2,249,250

CentralFranconiandialects (German)

Rhineland-Westfalia,Rhine-Palatinate;alsoBelgium,Luxemboug,theNetherlands(Limburgprovinces)

169

CologneGerman

Cologne 176

Giryama Niger-Congo,Bantu Kenya 172Gothic Indo-European,

GermanicExtinct 135

Greek Indo-European,Hellenic

Cyprus,Greece 71,78,89,209,254

Haida Isolate Canada(BritishColumbia)

40–1

Hausa Afro-Asiatic,Chadic

BurkinaFaso,Niger,Nigeria

46,213

Hawaiian Austronesian,Oceanic

USA(Hawai’i) 23,31–2,38–40,46,48–50,53,59,136,191

Hindi Indo-European,Indo­Iranian

India,Nepal 22,29,233–5,237,239

Hixkaryana Cariban Brazil(AmazonasandParástates)

46

Hopi Uto-Aztecan USA(Arizona) 136,182,190,193,196

Icelandic Indo-European,Germanic

Iceland,USA 181–2,189–90

Igbo Niger-Congo,Benue-Congo

Nigeria(Southeast) 40,95–6

Imonda Border,Waris Indonesia(EasternWestPapua),PapuaNewGuinea(Sandaunprovince)

81

Indonesian (akaBahasaIndonesia)

Austronesian,Malayic

Indonesia 48,51,59–60,179

Inuit Eskimo-Aleut Canada,Greenland 2Italian Indo-European,

RomanceItaly,USA 26–7,31–2,38,79,83,137,

178–9,185,195,201,209,211–12,253

Language index 277

Language Family Geographical area Page numbersItunyosoTrique Otomanguean,

MixtecanMexico(Oaxacastate) 170

Japanese Isolate Brazil,Japan,Taiwan,USA

2–3,5–6,31–2,38,45,52–3,61,101,127,160,176–7,179,195,253

Kalabari Niger-Congo,Ijoid Nigeria(Riversstate) 158Karen Sino-Tibetan,

Tibeto-BurmanMyanmar 23

Kayardild Australian,Non-Pama-Nyungan

Australia(SouthWellesleyIsland)

77

Kikuyu Niger-Congo,Bantu Kenya 153–4Kilivila Austronesian,

OceanicPapuaNewGuinea

(TrobriandIslands)76

Kipare(akaAsu)

Niger-Congo,Bantu Tanzania(Kilimanjaroregion)

168

Kisanga Niger-Congo,Bantu DRCongo(South) 157Klamath Penutian USA(Oregon) 38–9,237–8,242Konjo Austronesian,

Malayo-Polynesian

Indonesia(EastSulawesi) 48,51,59–60,62

Korean Isolate NorthKorea,SouthKorea 26,95,97,116Koya Dravidian India(Telangana,Andhra

Pradesh)137

Kuki-Thaadow Sino-Tibetan,Kuki-Chin

India(Northeast),Myanmar

164

Latin Indo-European,Romance

Extinct 83,253

Lenakel Austronesian Vanuatu 191Limburgish Indoeuropean,

Germanic,CentralFranconian

Belgium,Netherlands(Limburgprovinces)

74,169

GeleenLimburgish

Netherlands(Geleen) 74

HasseltLimburgish

Belgium(Hasselt) 169

Luganda Niger-Congo,Bantu Uganda 84,129–30,135

Ma’ya Austronesian,Malayo-Polynesian

Indonesia(WestPapua,RajaAmpatarchipelago)

176,198

Macedonian Indo-European,Slavic

Macedonia 197

Malay Austronesian,Malayic

Indonesia,Malaysia 179

AmboneseMalay

Indonesia(CentralMoluccas)

179

Malayalam Dravidian India(Kerala;TamilNadu)

129

Language index278

Language Family Geographical area Page numbersMandarin Sino-Tibetan,

ChineseChina(North);many

other locations38,148,162,170

HenanMandarin

China(Henanprovince) 38

Margi Afro-Asiatic Nigeria(East) 76,249Mauritian

CreoleFrench

French-basedcreole Mauritius 58

Mba Niger-Congo,Ubangi

DRCongo(Orientaleprovince)

38

Mende Niger-Congo,Mande

SierraLeone 153,233,247

Mixtec Otomanguean Mexico(Southern) 163Axcatlan

MixtecMexico(Pueblastate;

Oaxacastate)163

Mwera Niger-Congo,Bantu Tanzania(Southeast) 220

Nama Khoisan Namibia 34Nambakaengo EastPapuan SolomonIslands(Santa

Cruzisland,Temotuprovince)

76,250

Ndunga Niger-Congo,Ubangi

DRCongo(Equateurprovince)

38

Norwegian Indo-European,Germanic

Norway 41,43

Nubi Arabiclexifiedcreole

Kenya(Nairobi);Uganda(Bombo)

177,198

Orochen Altaic,Tungusic China(Heilongjiang) 79Oromo Afro-Asiatic,

CushiticEthiopia 70

Pawaia Trans-NewGuinea,Teberen-Pawaian

PapuaNewGuinea(Chimbuprovince)

41

Persian(akaFarsi)

Indoeuropean,Indo­Iranian

Iran 176–7,198

Pirahã Muran Brazil(Amazonasstate) 38,40Plautdiitsch Indo-European,

GermanicRussia(WesternSiberia) 97

Polish Indo-European,Slavic

Poland 39,188–9,250,255

Portuguese Indo-European,Romance

Brazil,Portugal,USA 27,145

BrazilianPortuguese

Brazil,USA 145

EuropeanPortuguese

Portugal 27

Proto­Indo­European

reconstructed 148,178

Provençal Indo-European,Romance

France,Italy,Monaco 96

Language index 279

Language Family Geographical area Page numbersRotokas NorthBougainville PapuaNewGuinea

(Bougainvilleprovince)37–8,40

Russian Indo-European,Slavic

Russia,USA,formerUSSR

111,243–4

Sanskrit Indo-European,Indo-Aryan

India(UttarPradesh) 244

Sardinian Indoeuropean, Romance

Italy(Sardinia) 224

Serbocroat Indo-European,Slavic

Bosnia,Croatia,Serbia 255

Shiwilu(akaJebero)

Kawapanan Peru(AltoAmazonasprovince)

190,196

Shona Niger-Congo,Bantu Mozambique,Zambia,Zimbabwe

158

Sinhalese Indo-European,Indo­Iranian

SriLanka 29

Slovak Indoeuropean, Slavic

Slovakia 223,225–8

Spanish Indo-European,Romance

Argentina,Belize,Bolivia,Chile,Colombia,Costa Rica,Cuba,DominicanRepublic,Ecuador,ElSalvador,Guatemala,Honduras,Mexico,Nicaragua,Panama, Peru, Puerto Rico,Spain,Uruguay,USA,Venezuala

23,29,31,38–9,67,97–8,119,127,144,175–6,189–90,222–4,226,228,230,235–6

ColombianSpanish

Colombia 38

GranCanarianSpanish

Spain(CanaryIslands) 222–4,226,228,230

NewWorldSpanish

CentralandSouthAmerica

235

Sundanese Austronesian Indonesia(Java;Bali) 245Swedish Indo-European,

GermanicFinland,Sweden 96,128,175–6,198

Tacana Tacanan Bolivia(LaPazprovince) 297Tagalog Austronesian,

Malayo-Polynesian

Philippines 138,201

Tahltan Na-Dene Canada(BritishColumbia)

256

Tamil Dravidian India(TamilNadu;AndhraPradesh),SriLanka(North)

129,137

Telugu Dravidian India(AndhraPradesh) 84Tera Afro-Asiatic,Chadic Nigeria(Gombestate) 138

Language index280

Language Family Geographical area Page numbersThai Daic, Tai Thailand 42, 46Tharaka Niger-Congo, Bantu Kenya (Central) 153–4Tiv Niger-Congo,

BantoidCameroon, Nigeria 154

Tlingit Na-Dene Canada (British Columbia), USA (Alaska)

29

Toda Dravidian India (Tamil Nadu) 31Tolitoli Austronesian,

Malayo-Polynesian

Indonesia (Sulawesi) 96, 104

Tonkawa Isolate USA (Oklahoma) 55Tsou Austronesian Taiwan 23Turkish Altaic Cyprus, Germany, Turkey 71, 82, 89, 93, 184–5, 209

Venda Niger-Congo, Bantu South Africa, Zimbabwe 71Vietnamese Austro-Asiatic,

Mon-KhmerVietnam 34, 46, 170

Welsh Indo-European, Celtic

UK (Wales) 31

Weri Trans-New Guinea, Kunimaipa

Papua New Guinea (Central province; Morobe province)

190

West Greenlandic

Eskimo-Aleut Greenland 155

Wolof Niger-Congo, Atlantic

Gambia, Senegal 39, 195–6

Wu Sino-Tibetan, Chinese

China (East Central) 38, 170–1, 195

Shanghai Wu Shanghai municipality, China

170–1

Yuhuan Wu Zhejiang province, China 171

Xhosa Niger-Congo, Bantu Lesotho, South Africa, Swaziland

34

Yabem Austronesian, North Huon Gulf

Papua New Guinea (Morobe province)

187

Yanyuwa Pama-Nyungan Australia, Northern Territory

82

Yoruba Niger-Congo, Benua-Congo

Benin, Nigeria, Togo 46, 165–6

Zenzontepec Chatino

Otomanguean, Zatopecan

Mexico (Oaxaca province)

148

Zulu Niger-Congo, Bantu Malawi, South Africa, Swaziland

154, 160

Zwara (aka Zuwārah)

Afro-Asiatic, Berber Libya (Northwest) 39

Subject index

ι[aɪˈəʊtə]see intonational phraseμ[mjuː]see morasφ[faɪ]see phonological phrase σ[ˈsɪgmə];see syllableωBrE[ˈəʊməgə];AmE[əʊˈmeɪgə];see

phonological words

absoluteslicinghypothesis150accents159–61;non-stressvsstress176–7;

pitch163,175accentualanalysis160adjustment,phonological48,63;adjustment

processes49–54;adjustmentsinthenativevocabularyand54;choosingbetweenrulesandconstraintsin58–62;constraintsapproach56–8;processofnativizationand49–53;rulesapproach54–5;startingphonologyand48–9

affixes8–9,54,192affricates30airstreammechanisms18Aitken’sLaw71alignment,tone166–8allomorph98;morphemealternantandallophones94allophonicandstylisticvariation95–8allophonicrules101,109–10alveolar29alveolarridge25alveolopalatal95ambiguity and non­specification, distinctive

features72–5ambisyllabicity142–4,147analysis,segment43Andersen,Torben166apical77approximant30,31;lightornear-contactand

30aspiration20–2assimilations45,89;coronal233,240,256;

long-distance243–6;nasal240;place239,241;vowelharmony247–9

associationconvention151,153–5,172;accentand159–60

associationlines130autosegments130–2

back(tongue)25backvowels26Beckman,MaryE.176bilabial29bimoraicsyllables189binarydistinctivefeatures64,66,180bladezone(tongue)25bleedingorder219Booij,Geert200,209,211boundary:strength200–1;symbols85;tones

163,201boundedness and unboundedness in word

stress182–3boundmorphemes8braceinSPErules88bracketing,labeled9breathyvoice22,23;toneand170

catalexis184categorymembership8centralvowels26Chatterji,SunitiK.178Chaucer,Geoffrey103Chen,Matthew211children:earlylanguagelearningby1–2;

learning sounds of their language in a particularorder42;vowelproductionin28

Chomsky,Noam54,76,92,111,119clashes196–8clauses9Clements,GeorgeN.25,66,128,129,133,247clicks33–4cliticization203–4clitics212–13close-midvowels26closevowels26coarticulation201

Subject index282

coda38,126coda­µ189–90cognitivefactors5Cohn,Abigail121compensatorylengthening134–6complementarydistribution96complexconsonants32–3complex-placesegments249complexsegments249–57;evidencefor

251–2;representationofpalatalsandpalatoalveolars252–5

complexwords8consonants37;apical77;cluster38;complex

32–3;deletion,final219,222;doublearticulations33;liaison133,143–4;manner-contour33;nonpulmonic33–4;numbercorrelatedtonumberofvowels42;secondaryarticulations32–3;segmentaldurationof31–2;syllabic39,135

constraints:-basedapproachtoadjustment56–8;choosingbetweenrulesand58–62;gradient violation and unranked constraints in61–2;hierarchy56;unranked61–2

constrictedglottis70constriction28–31;major-classfeatures66–8;

placesofarticulationand29;typesof30–1contextinSPErule87contourtones151contrastiveunderspecification84,242contrastivevsredundantfeatures82–5coronal29,76–7coronalassimilation233,240,256counterbleeding:opacity225;order219–20counterfeeding:opacity223–4;order219creakyvoice22,23;toneand170crown29

declination,pitch165degeneratefeet188–92delinkingandspreading238–41dental29derivationalaffixes9derivationalapproach54devoicing20–2;bleedingorderand219;

simultaneous-orderingtheoryand92diacritics17diphthongs27,44dissimilation83distinctivefeatures11,64,75;ambiguityandnon-

specification72–5;binary64,66;contrastiveandredundant82–3;featurevalues66;laryngeal66,69–70;major-class66–9;manner66,70–2;motivating64–6;place76–82;system66

diversity,language2–3dominanceinwordstress180–1,188–92Do-re-mi3–4dorsal29,78–82dorsum see tonguedoublearticulations33downstep,tonal164dualstructure14–16;nonisomorphic

structuresand15duplicationproblem59duration,segmental31–2,146

edgeinwordstress180–1,192–5;extrametricalityand184

egressivespeechsounds18ejectives34Eliot,T.S.128ElsewhereCondition216–17English Pronouncing Dictionary176ergonomics,speech44–5exemplarmodels122extrametricalelements184extrinsicandintrinsicordering216–18

faithfulnessconstraints56–7featuregeometry232–3,257–8;complex

segments249–57;implicationsofunderspecification241–9;long-distanceassimilation243–6;naturalfeatureclasses233–4;spreadinganddelinking238–41;treediagram234–8;vowelharmony 247–9

featurematrix85feedingorder218finallengthening201finalloweringoftones165–6flapping143–4,147,217floating164,166,247focusofchangeinSPErule87foot34;degenerate188–92;dominance180–1,

188–92;edgesanditeration192–5;andfooting179–83;independentevidencefor186–7;non-finalityandclashes196–8;shapes188–92

formants21f0:perturbations149;tonedistributionand

170–2;see also fundamental frequencyfreemorphemes8fricatives30,44friction22,30front(tongue)25frontvowels26fundamentalfrequency21–2

Subject index 283

front vowels 26fundamental frequency 21–2

geminates 31geminates and weight 195–6generalizations: involving prosodic

constituents 200–2; syllable-based 139–41gestures 122; and gestural model 121–3Gimson, A. C. 176glottalic airstream mechanism 34glottal stop 23; tone distribution and 171glottis 18; closing 23; laryngeal features 69–70;

open and vibrating 19Goldsmith, John A. 149, 150, 151, 158Gomez-Imbert, Elsa 160Gordon, Matthew 183Grabe, Esther 166gradient violation 61–2Gurevich, Naomi 224Gussenhoven, Carlos 185

Halle, Morris 54, 66, 224; on levels of representation 111, 119; on places of articulation 76, 92; on word stress 179, 185

Hammerstein, Oscar 3hard palate 25harmony, vowel 247–9Harris, Zellig S. 111Hayes, Bruce 179–80, 182–3, 188, 191,

199–200, 207, 213–14hierarchical phrasal structure 9hierarchy, constraint 56high vowels 26Hockett, Charles F. 111Hooper, Joan B. 88H-raising 165Hyman, Larry M. 154, 160, 174, 178

iambic structure 180–3, 188; edges and iteration 193–5; quantity-sensitivity bias 183; uneven 188

implementation, phonetic 109, 119–20; models of 120–2; of tone 162–8

implosives 34inflectional affixes 9information structure of sentences 204initial strengthening 201Inkelas, Sharon 208, 211innate features of language 37interpolations 120–1intonational phrase 206–7intonational tones 149, 166inventory, lexical 113

Jacobs, Haike 253Jakobson, Roman 41, 42Jones, Daniel 176

Kahn, Daniel 88, 143Kaisse, Ellen M. 213Kaye, Jonathan 66, 224Keating, Patricia A. 253Kenstowicz, Michael 160, 185, 193Keyser, Samuel J. 129, 133Kiparsky, Paul 216, 222

labelled bracketing 9labial-coronal segments 249labial-dorsal segments 249, 252labiodental 29Lang, Walter 161language(s): assimilation in 45; awareness

of structure of 2; diversity 2–3; -external evidence 6–8; -internal evidence 5; learning by children by age 5 1–2; system gaps 45–6; tone 23, 148–9, 178; varying complexity of 37–40

laryngeal features 66, 69–70laryngeals and no manner or place feature

specifications 73–4; laryngeals and no supralaryngeal specification 235–6

laryngealized voice 22laryngeal node 235–6, 241, 244larynx 17, 18–19Lass, Roger 235lateral plosion 135laterals 31Leben, William R. 151, 158Lehrer, Tom 4level tones 151lexical inventory 113lexical and postlexical phonological rules

111–12, 124–5; application across word boundaries 114; controversial properties of 118–24; exceptions 112–13; native-speaker intuitions 114; non-derived environment blocking 118; reference to morphological tables 112; structure preservation 113–14

lexical and postlexical prosody 178–9lexical representation 109, 248;

surface representation and 118–24; underlying representation and 118–24

liaison consonants 133; ambisyllabicity and 143–4

Liberman, Mark 164, 179linear order 91–3

Subject index284

linearrepresentations128;tonesandinadequacyof149–50

linguisticstructures1;awarenessof2;cognitive,phonetic,andsocialfactorsin5;diversityof2–3;morphosyntacticstructurein1,8–10;observing1–4;phonologicalstructurein1,10–14;roleofspellingin4

linguists4–8;language-externalevidenceusedby6–8;language-internalevidenceusedby5

Lloyd,Joy39–40long-distanceassimilation243–6;vowel

harmony247–9lowvowels26lungs17,18–19;plosivesand30

Maddieson,Ian42major-classfeatures66–9manner-contourconsonants33,250mannerfeatures66,70–2marginalsegments43matchtheory212maximalruleapplication222maximalsurfacetransparency222maximumonsetprinciple(MOP)126–8,209;

syllabificationrules,post-141–6McCarthy,John57,74,130,192melodies,word149,150–5midvowels26minimalpairs95,111monophtongs26–7moras3,6,136–9,146–7;bimoraicsyllables

189;moraicconsonants137;tonalanalysisand160

morphemes8–9,14–15;boundary85;choosingtheunderlyingformfor101–4;recurringelementsin11;segmentsin11;tonal157;variationinpronunciation of94

morphologicallabels112morphology8;effectsandexceptionsinword

stress185–6morphosyntacticstructure1,8–10;inevery

linguisticexpression14–15;morphemesandwords8–9;syntacticstructure9–10

motivatingdistinctivefeatures64–6mouth24,25;nasalsand31;teeth25,29;

uvula25,31;see also tonguemultivaluedfeatures66

narrowing,vocalfolds22narrowtranscription17nasalcavity24,25,236–7

nasals30,31native-speakerintuitions114nativizationprocess49–53naturalfeatureclasses233–4naturalnesscondition65,188–92naturalsegmentclasses64,65Nespor,Marina200,205,211,212neutralization95,101,109–10Ngunga,ArmindaS.A.154nocrossingconstraint154–5,172,243nocrowdingconstraint152–3,172non-derivedenvironmentblocking118non-finalityandclashesinwordstress196–8nonpulmonicconsonants33–4

obligatorycontourprinciple(OCP)132,149,158–9,172

obstruents30,67,70–1Odden,David158onset126opacity223;counterbleeding225;

counterfeeding223–4;optimalitytheoryand223–30;rulesandconstraints222–3

open-midvowels26openvowels26operativelevel50–1optimalform57optimalitytheory(OT)56–7,159,211,

215–16,230–1;counterbleedingopacityasproblematicfor225;counterfeedingopacityasproblematicfor223–4;extrinsicandintrinsicordering216–18;feeding,counterfeeding, bleeding, counterbleeding 218–22;opacityand223–30;stratal225–30;stressand187–98

organsofspeech17;lungsandlarynx17,18–19

over-application223

palatal33;featuregeometryrepresentationof252–5

palatoalveolar29;distinguishingamong255–6;featuregeometryrepresentationof252–5

parenthesisnotationinSPE89peak38,39,126perceptuallevel49–51perturbations,f0149;tonedistributionand

170–2pharyngeal33pharynx24,25phonation19,22phonemes12–13,94,111phonemictranscription109

Subject index 285

phoneticfactors5,65–6phoneticimplementation109,119–20;

decidingbetweenphonologyand122–4;oftone162–8

phoneticstress176;absenceof176–7;asobligatoryfeature178

phoneticsymbols17phonologicalcontext94phonologicalphrase207–8phonologicalstructure1,10–14;as“builtup”

41;ineverylinguisticexpression14–15;extendingbeyondthesyllableandfoot200;similaritiesbetweendifferentlanguages36,47;writingrules85–91;see alsoadjustment,phonological

phonologicalutterance205–6phonologicalword15,209–10phonology36–7;decidingbetweenphonetic

implementationand122–4;informationinthelexicon115–17;lexical111–15;postlexical111–15;-sensitivesyntax201–2;systemgaps45–6;universal12–14,37;universalsandimplicationalrelations40–6;andvaryingcomplexityoflanguages37–40

phrases,syntactic9Pierrehumbert,JanetB.162,164Pike,EuniceV.163pitch23placenode234–5,241,245;secondary

articulations250–1placesofarticulation76;coronal29,76–7;

dorsal29,78–82;features76–82;labial29,32–3;linearorder91–3;pharynx29,33;radical79

plosives30,44,46Postal,Paul111,188postalveolar29postlexical phonological rules see lexical and

postlexical phonological rulespostlexical prosody see lexical and postlexical

prosodypost-MOPsyllabificationrules141–6preboundarylengthening201precompiledforms213prefixes8prenasalizedstops33prepalatal29primarystress34Prince,Alan57,179,192privativefeatures66privativetone149,168–9pronunciation,variationin94,122;received

pronunciation(RP)and122–4

prosodichierarchy125,200,214;constituents205–10;derivingprosodicconstituents211–13;factorsdeterminingprosodicphrasing203–5;generalizationinvolving200–2;intonationalphrasein206–7;phonologicalphrasein207–8;phonologicalutterancein205–6;phonologicalwordin209–10;strictlayerhypothesis(SLH)202–3

prosodicphrasingfactors203–5prosody,lexicalvspostlexical178–9Pulleyblank,Douglas154pulmonicairstreammechanism18

quantity(syllables)136quantity-sensitivitybiasinwordstress183,

188–92

radical79radicalunderspecification242rankingofconstraints56receivedpronunciation(RP)122–4recurringelementsinmorphemes11recursiveness9–10redundantvscontrastivefeatures82–5Reiss,Charles224representation,levelsof109;beyondsurface

118–24;defininganintermediate109–11;hierarchiesandautosegments128–33;lexicalphonology109,111–15,124–5;lexicalrepresentation119;phonologicalinformationinthelexiconand115–17

representations,underlyingandsurface94–5,104–8;allophonicandstylisticvariation95–8;argumentsfortwolevelsof98–101;choosingtheunderlyingformfrom101–4;neutralization95,101

restructuring204resyllabification141–6retroflex29rhyme38,126rhythmrule207–8Roca,IgnacioM.242Rodgers.Richard3rootnode236root,tongue25Rubach,Jerzy116,226ruleapproachinadaptation54–5;choosing

betweenconstraintsand58–62ruleinteractions218

Sagey,ElizabethC.66scaling,tonal163–6secondaryarticulations32–3,250–1

Subject index286

secondarystress34segmentalduration31–2,146segments11–12,15;associationlines130;

auto-130–2;avoidanceofcomplexityin43;commonversusunusual40–1;complex249–57;complex-place249;distinctivefeatures(seedistinctivefeatures);duration31–2,146;informationofsyllables,complexityof38;manner-contour250;marginal43;naturalsegmentclasses64,65;phoneticsymbolsrepresenting17;skeletalslots129–30;writingrules85

Selkirk,Elisabeth147,179,200,206,211,212sentences9;informationstructureof204Sezer,Engin185signlanguages12Silverman,Daniel49,50simplexwords8simultaneousruleordering91–3skeletalslots129–30;unfilledand

unassociated133softpalate25;nasalsand31sonorants30,67sonorityprofile127–8Sound Pattern of English, The (SPE)54speech:complexconsonants32–3;

constrictionsin28–31;devoicingandaspiration20–2;ergonomics44–5;glottalstopin23;lungsandlarynxin17,18–19;nonpulmonicconsonants33–4;openandvibratingglottisin19;organsof17;phoneticsymbolsand17;pitch23;segmentalduration31–2;specialtypesofphonation22;stress34;vocaltractin24–5;vowels25–8;waveform20

spelling4SPEmodel85–91,215,230;absoluteslicing

hypothesisand150;additionalnotations88–91;bracenotation88;generalformat86–8;parenthesisnotation89;transformationalruleformat90;variablefeaturevalues89

spreadglottis69–70spreadinganddelinking238–41stability,tone149Stampe,David188stem(s)9Steriade,Donca83,242stops30;glottal23,171;prenasalized33;

voiced222–3;voiceless123strataloptimalitytheory225–30stress see word stressstrictlayerhypothesis(SLH)202–3

structuralchange(SC)87structuraldescription(SD)87structure preservation, lexical phonology

113–14style95stylisticvariants95subcategories8–9suffixes8–9supralaryngealnode236–8,241,244surfacecontrast111syllabicconsonants135syllabification126–8;autosegments130–2;

compensatorylengthening134–6;domain127;hierarchiesandautosegments128–36;liaisonconsonants133–4;maximumonsetprinciple126–8;post-MOPrules141–6;skeletalslots129–30;sonorityprofile127–8;unfilledandunassociatedslots133–4

syllable(s)11,14–15,126;ambisyllabicity142–4,147;-basedgeneralizations139–41;bimoraic189;coda38;contactlaw128;degreeofcomplexityin38–9;footing179–83;moras3,6,136–9,146–7;onset38,39;peak38,39;segmentsbuilding15;stressed34;tone-bearingunit(TBU)151–5,159

syntacticresidue213syntacticstructure9–10syntax8;phonology-sensitive201–2systemgaps45–6

tableauxinoptimalitytheory57target-basedphoneticimplementation122,162–8teeth:coronalarticulationand29;lowerfront

25;upperfront25timingtier129tonalanalysis160tonalmorphemes1157tone(s):accentand159–61;alignment166–8;

associationconvention151,153–5,172;-bearingunit(TBU)151–5,159,164,166;boundary163,201;contour151;declination165;distributionandf0perturbations170–2;downstep164;finallowering165–6;H-raising165;inadequacyoflinearmodelfor149–50;intonational149,166;languages23,148–9,178;level151;obligatorycontourprinciple(OCP)132,149,158–9,172;phoneticimplementationof162–8;privativerepresentationof149,168–9;scaling163–6;stability149,155–8;twinsisterconvention156,172;undershoot166;upstep163;voicequalityand170;wordmelodiesand149,150–5

Subject index 287

tongue25;bodyfeatures78;see also dorsal articulation; mouth

transcription,narrow17transformationalruleformat90transparency and opacity: rules and

constraints222–3treediagramandfeaturegeometry234–8trills31trisyllabiclaxing113,118trochaicstructure180–3,188;edgesand

iteration193–5Twaddell,W.Freeman111twinsisterconvention156,172

unaryfeatures66under-application223undershoot,tonal166underspecification:implications241–9;theory

83undominatedconstraints59uneveniamb188unfilledandunassociatedslots133univalentfeatures66,76universalphonology12–14,37universals40–6unmarkedpronunciation56unrankedconstraints61–2upstep,tonal163utterance,phonological205–6uvula25;trillsand31

vacuousruleapplication87values,feature66variablefeaturevalues89velaricairstreammechanism33velarization33velum25Venneman,Theo88,127,128Vergnaud,RogerJ.179vocal cords see vocal foldsvocalfolds18;devoicingandaspiration20–2;

openandvibratingglottis19;inproducingvowels25;specialtypesofphonation22;tonedistributionand170–2

vocaltract24–5

Vogel,Irene200,205,211,212voice19;breathy22,23,170;creaky22,

23,170;devoicingandaspiration20–2;laryngealfeatures69;laryngealized22;onsettime20;pitch23;qualityandtone170

voicedstops222–3voicedvowelsandconsonants19voicelesssounds19,23;fricatives44,71;

plosives44,46voicing:specialtypesofphonation22;whisper

22,23vowel(s)25–8,37;commonversusless

common45;diphthong27,44;harmony247–9;monophthong26–7;nasalization27–8;numbercorrelatedtonumberofconsonants42;productioninchildrenversusadults28

weight-by-position(Coda-µ)189–92,193weightinwordstress180–1;geminatesand

195–6weight-to-stressprinciple(WSP)189–92,193Weijer,Jeroenvande253whisper22,23Wistrand,Kent163word(s):boundary85,114;feet34;melodies

149,150–5;morphemesand8–9;phonological15,209–10

wordstress34,174–5,198–9;absenceofphoneticstress176–7;degeneratefeetand188–92;describingpatternsof179–87;dominancein180–1,188–92;edgein180–1,184,192–5;flexiblerightedges184;footing179–83;footshapesand188–92;independent evidence for the foot in 186–7;lexicalvspostlexicalprosody178–9;manifestationsof175–9;morphologicaleffectsandexceptions185–6;non-finalityandclashesin196–8;optimalitytheoryand187–98;phoneticstressasobligatoryfeature178;phonologicalfeaturesof174;weightand180–1,195–6

writingrules85–91;boundarysymbols85;featurematrix85;spreadinganddelinking239–41