EmmanuelFerragne&Franc - PDF document

EmmanuelFerragne&Franc¸oisPellegrinoTable1AccentsoftheABIcorpusandnumberofsubjectsandvowels. AbbreviationAccentPlaceSubjectsinmonophthongSubjectsindiphthongNumberofvoweltokens BirminghamBirmingham76517CornwallTruro22161EastAngliaLowestoft75473EastYorkshireHull86588GlasgowGlasgow77523iloInnerLondonLondonnotincluded(seesection2.2)LancashireBurnley109763LiverpoolLiverpool85561NewcastleNewcastle22147NorthWalesDenbigh77544RepublicofIrelandDublin76528ScottishHighlandsElgin1111865StandardSouthernEnglishLondon66472UlsterBelfast65482Total88776624 overviewofgeographically-inducedphoneticandphonologicalvariationintheBritishIsles.Thearticlecanalsoserveasabasisformoresophisticatedpronunciationmodellingforspeechtechnologypurposes(speakerindexing,automaticaccentclassiÞcationÐseeFerragne&Pellegrino2007).Materialandmethod2.1CorpusTheAccentsoftheBritishIsles(ABI;DÕArcyetal.2004)corpusincludesrecordingsfrom14areascoveringtheBritishIsles.Onaverage,20subjects(equallydividedintomenandwomen)tookpartforeachregion.Thespeakershadlivedalloftheirlivesintheregioninquestion,andtheirparentshadlivedthere,too.Thetargetagerangewasfrom18to50yearsoldbut,insomeareas,theactualrangeis16Ð79.Therecordingstookplaceatthebeginningof2003.Table1showstheabbreviationsthatwillbeusedthroughouttodesignatetheaccents,themeaningoftheseabbreviations,wheretherecordingstookplace,thenumberofsubjectswhosevowelswillbeactuallygraphed,andthenumberofvoweltokensperaccent.Thenumberofsubjectsissplitintotwocolumns:theleftmostcolumncontains,foreachaccent,thenumberofsubjectswhosedatacontributetomonophthongplots,andthenumberofsubjectsusedinthediphthongplotsappearsintheothercolumn.TheÞguresinthecolumnsdifferasaresultofthescreeningprocedure(seesection2.2).Itisworthmentioningthatnoinformationontheparticipantsisavailable,whichdefactoprecludesthestudyofphoneticvariationcausedbysociological(inabroadsense)factors.Inparticular,itwouldhavebeendesirabletoknowtheageoftheparticipantsbecausemanystudies(e.g.Labov1994,Foulkes&Docherty1999,Hawkins&Midgley2005)haveshownthataspeakerÕsageisacentralfactorinexplainingphoneticandphonologicalvariation.Nevertheless,althoughtheseimportantfactorscannotbecontrolledforaposteriori,standardspeakernormalizationtechniqueswillbeappliedtothedata(seesection2.2). VowelsinBritishIslesaccentsTable2Test-words,(assumed)correspondinglexicalsets,andconventionalvowelcategory. Test-wordLexicalsetConventionallabel FLEECEmonophthongKITmonophthongDRESSmonophthongTRAPmonophthongSTARTmonophthongLOTmonophthonghoardFORCEmonophthongFOOTmonophthongwho’dmonophthongSTRUTmonophthongNURSEmonophthongFACEdiphthongPRICEdiphthonghoiddiphthonghoedGOATdiphthonghowdMOUTHdiphthongSQUAREdiphthongNEARdiphthongCUREdiphthong Thespeechmaterialweareconcernedwithhereisalistof19//wordsthatthesubjectswereaskedtoreadÞvetimes:hardhoardwho’dheardhowdharedheeredhured.Theuseofsuchstimuliismeanttofactoroutboththeeffectsofcoarticulationandtheacousticvariationdeterminedbyinformation-relatedphenomenasuchaslinguisticredundancy(Wright2003,Aylett&Turk2006).Oneobviousdrawbackofthisapproachliesinthequestionablesuitabilityofnon-wordsorrarewords(wereturntothispoint,andotherdisadvantages,below,intheDiscussionsection).However,DÕArcyandcolleagues,whorecordedthedata,providedthesubjectswithcommonwordsthattheÔdifÞcultÕ//targetsweresupposedtorhymewith.Forinstance,thesubjectswereinstructedthatwastorhymewithvoid,withshowedhowd,with,etc.(DÕArcyetal.2004:116).Intheremainderofthepresentarticle,bothtest-wordsandstandardlexicalsets(Wells1982)willbeused,ofteninterchangeably.Forthecorrespondencetobeexplicitfromthestart,test-wordsandstandardlexicalsetshavebeenmatchedintable2;thismakescomparisonswithotheraccentstudieseasier.ThetableshowswhichlexicalsetofStandardSouthernBritishEnglishisÐsotheauthorsthinkÐexempliÞedbyeachtest-word;butitdoesnotgosofarastoimplythatallspeakersphonologicallyinterpretedeachtest-wordasamemberofthelexicalsetitisassociatedwithinthetable(seetheDiscussionforamorethoroughtreatmentofthebiasinducedbyreasoningintermsofthelexicalsetsofStandardSouthernBritishEnglish).Forthesakeoflegibility,thevowelsweresplitbetweenmonophthongs,closingdiphthongs,andcentringdiphthongs,followingthephonologicallabelstheyaregiveninclassicdescriptionsofReceivedPronunciation(RP;Jones2003,Wells2008).Ofcourse,thisconventionaldivisiondoesnottellusanythingabouttheactualphoneticrealizationofavowelinagivenaccent.Theconventionallabelsforvowelsarelistedintable2.Inthispaper,weonlyreportmaledataonthegroundsthatgenderisknowntobeanimportantfactorofphoneticvariation;thus,includingdatafromtwogenderswouldhavenecessitatedseparategraphsandcomments,whichwouldhaveconsiderablyincreasedthe EmmanuelFerragne&Franc¸oisPellegrinolengthofthispaper.Thechoiceofmaleratherthanfemaledatawasdeterminedbythefactthattypicalityscores(seebelow)wereonlyavailableforthemaleparticipants.2.2MethodGiventhelackofinformationonindividualsubjects,aphonetician,nativespeakerofEnglish,wasaskedtolistentoallspeechsamplesrecordedbythe145malespeakersandtocomeupwithscores(onaÞve-pointscale)reßectingthetypicalityofthespeakerÕsaccentaswellasintra-accenthomogeneity.TheexpertphoneticianwastoldtolistentotheÞrstpartofthereadpassage(theABIcorpuscontainsareadpassageofabout300words)andcouldstopplayingasoundassoonashehadformedanopinionastothetypicalityofthesampleunderscrutiny.Thisassessmentledustoleaveoutthesamplelabelledonaccountofsubstantialwithin-samplephoneticvariability(mainlybroughtaboutbydifferencesinethnicbackgrounds).AmorethoroughassessmentwithmorelistenersandsomemeasureoflistenersÕagreementconstitutesapotentialfollow-upstudy.Formantshavelongbeenheldtobewell-suitedparametersfordescribingvowelquality,mainlybecausetheF1/F2planecorrelateswithtraditionalarticulatoryÐauditorytrapezoidalrepresentationsofvowels.However,DeWetetal.(2004)emphasizethat,todate,noautomatictoolcanreliablyextractformants.Asaconsequence,thereproducibilityofmeasurementsisnotguaranteedandtheprocessingofverylargespeechcorporamayproveaverytime-consumingtask.WeoptedforacompromisethatwouldminimizehumaninterventionwhilepreservingthebeneÞtsoftheF1/F2representation.Theremainderofthissectionexplainshowformantmeasurementswerecalculated.Themethodologyincludestherejection,aftervisualassessment,ofvoweltokenswithanobviousmismatchbetweenspectrogramandestimatedformanttracks,theautomaticdeterminationofvowelboundariesbasedonF0detection,theconversionofHertzvaluestotheBarkscale,thesmoothingofformantcontourswithregressiontechniquesinordertoruleoutremainingformanttrackingerrors,andthecomputingofLobanov-transformedvalues(seebelow).Abroad-bandspectrogramwithsuperimposedformanttrackingwasobtainedwithPraat(Boersma&Weenink2008)ÐandtheBurgalgorithmsettodefaultparametersÐforeachofthe26,408vowels.Eachsuchrepresentationwasvisuallyinspected:whenacompletemismatchwasobservedbetweenformanttrackingandspectrogram,thevowelinquestionwasrejected;otherwiseÐandeveniftherewasaslightmismatchÐthevowelwaskeptforfurtheranalysis.Thescreeningprocedureleftuswith22,331vowels.TheABIcorpuscomescompletewithword-levelsegmentation,sothattheboundariesofeach//wordareclearlyidentiÞed.Inordertoextractthevocalicnuclei,automaticF0detectionwasrunusingtheSnackSoundToolkit(Sjolander2004)andthevoicedpartwaskept.Itshouldbeborneinmindherethatinsuchtest-wordsashardheard,etc.,a/followsthevowelsinrhoticaccents(roi).Inthoseaccentswhere//isrealizedasanapproximant,F0detectiondidnotallowforseparationbetweenthevowelandthe//(besides,somevowelswerer-colouredthroughout),whichledusÐforwantofabettermethodÐtokeepthewholespeechportion.Asaconsequence,wehaveincludedcommentsonrhoticityandthephoneticrealizationof//whererelevant.Priortoacousticmeasurements,formantvalueswerere-sampledusinglinearinterpolationinorderforeachvoweltohavethesamenumberofvalues,namely13,whateveritsduration.Thiswasdoneforthesakeofconvenience:forinstance,fornumericaltechniquessuchasregression,numbersweremoreeasilyhandledwhenstoredinvectorsofconstantsize.However,itmustbenotedthatthedurationratiobetweenformantsteadystatesandformanttransitionsvariesacrossvowelsandspeechrates(Gay1968,1978),soourmethodcausesaslightlossofinformation.Beyondourconventionalsplitofthevowelsintomonophthongsanddiphthongs,westillhadtodecideifavowelwouldberepresented,intermsofphoneticrealization,asoneortheothertype.Afterlisteningtoeachvoweltokenandvisualizingthecorresponding VowelsinBritishIslesaccents Figure1F2fromthevowelofbyspeakerPDKfrom:originalformanttrack(circles),classicregressionline(solidline),androbustregressionline(dashedline).spectrogram,itwasdecidedthatallthevowelsthathadbeenlabelledasmonophthongswouldbetreatedassuch.Althoughsomeofthemareslightlydiphthongizedinsomeaccents(seee.g.below),therulewasappliedsystematically.Similarly,allthevoweltokensconventionallycalleddiphthongswerevisuallyandauditorilyinspected,and,ineachaccent,thepatternfoundmostoften(which,inpractice,alwaysprovedfarmorefrequentthanalternativevariants)determinedhowavoweltypewouldbeanalyzedinthisparticularaccent.Giventhesmallnumberofalternativevariantsineachaccent,wereckonedthatcomputingmediansandinterquartilerangesforallvoweltypeswouldguaranteeinsensitivitytothosevariantsandthereforespareusanadditionalsortingofspeakers.F1andF2valueswerethenBark-transformed(Traunmuller1990),androbustandpolynomialregressiontechniqueswereappliedtosmoothformanttrajectoriesandruleoutoutlyingvalues.ThespeciÞckindofregressiontobeemployedwasdecidedforeachtypeofvowelanddependedontheaccent.Forinstance,thevowelofFACEinmostaccentsofthenorthofEnglandisamonophthong,sorobustlinearregressionwasdeemedsufÞcienttocharacterizetheformant.Figure1showsthesecondvocalicformantofafterlinearinterpolationÔre-samplingÕ(circles)producedbyspeakerPDKfromwithregressionlinesobtainedwithclassicÐordinaryleastsquaresÐregression(solidline)androbustÐweightedleastsquaresÐregression(dashedline).TheÞguremakesitclearthatthelattertechnique(whichweadoptinthepaperformonophthongs)provessuperiorinthat,intuitively,itmatchesthemonophthongpatternmoreclosely,whereastheclassicregressionline,mostlybecauseoftheÞnalformantvalues,leadstoanoverestimationofstartingvaluesandanunderestimationofmostvaluesinthesecondhalfofthevowel.Asfarasthevowelisconcerned,acubicpolynomialmodelwasadoptedsothattheS-shapedtrajectoriesofF1andF2inmanyaccentswouldbeaccuratelyrendered.InÞgure2,thecirclesrepresentthesecondformantofthevowelspokenbyspeakerCTSfromthesolidlinewasobtainedthankstocubicpolynomialregression,whichallowsacurvilineartrajectorywithatmostoneinßectionpoint.Theso-calledLobanov(z-score)transformwasappliedtoeachformantseparatelyandforeachspeakerindependentlyinordertoerasephysiologically-inducedacousticvariation(Lobanov1971,Adanketal.2004).Inpractice,forexample,allF1valuesmeasuredattemporalmid-pointontheregressionlineinthemonophthongsofonesinglespeakerwerez-scored,andthesamewasappliedtoF2.Suchanormalizationprocedureimpliesthataspeakerproducesatleastonetokenofeachvoweltype,whichwasnotalwaysthecase(asa EmmanuelFerragne&Franc¸oisPellegrino Figure2F2fromthevowelofbyspeakerCTSfrom:originalformanttrack(circles)andcubicregressioncurve(solidline).consequenceofthescreeningproceduredescribedabove).Speakerswithincompletevowelsetswerethereforeremovedfortheremainderoftheacousticanalysis,hencethevariablenumberofremainingspeakers(table1)acrossaccentsandvowelcategory(i.e.monophthongsvs.diphthongs)andthereductionofvoweltokensfrom22,331to6,624.Incidentally,thelownumberofnclspeakerswarnsusagainstbeingtooconÞdentabouttheacousticresultsforthesetwoaccents.However,alongwiththeacousticdata,anauditoryanalysisofallspeakersÐincludingthosewhowereremovedduringthescreeningÐwascarriedoutbytheauthors.InadditiontoF1/F2plots,probabilitydensityestimatesbasedonkernelsmoothing(Everitt,Landau&Leese2001:16Ð20)willbepresentedforsomepairsofvowelsnecessitatingÞner-grainedcomparison(seee.g.Þgure16below).TheresultingÞgurescanbeinterpretedassmoothedhistograms,andtheyareparticularlyhelpfulwhenitcomestodecidingtowhatextenttwovowelscanbesaidtobelongtoasinglephoneme.Durationvalueswillonlyoccasionallybereported(e.g.inÞgure16),whenspectralqualityalonewasnotenoughtodetermineiftwovowelswereidentical.But,althoughdurationplaysacrucialroleinthephonologicalsystemofmanyaccents,wechosenottoincludethisparametersystematically.Thereasonisthat,toamuchlargerextentthanformants,durationisinßuencedbythespeciÞctypeofspeechmaterialanalyzedhere.Inotherwords,thefactthatmonosyllabicwordsappearedinalistresultedinhugeinter-individualvariationand,onaverage,durationvaluesweremuchhigherthanonewouldexpectinnaturalspeech.TheresultingF1/F2plotsarepresentedbelowforeachaccent.appearsÞrst,andthenallremainingaccentsarepresentedinalphabeticalorder.Giventhatonlytwospeakerscouldbeanalyzedacousticallyinncl,itwasdecidedthatoneplotforeachspeakerwouldbeprovidedintheseaccents.Thez-scoredBark-transformedF1/F2valuesplottedforthemonophthongsweremeasuredatthetemporalmidpoint.Onaccountoflargewithin-dialectdifferencesinphoneticrealization,themedianandtheinterquartilerangeswereusedasestimatorsofcentraltendencyandspread,respectively,ratherthanthearithmeticmeanandthestandarddeviation.Fortheclosingdiphthongs,thearrowsrepresenttheinitialandÞnal(arrowhead)medianF1andF2valueswitherrorbarsspanningtheinterquartilerange.Theactualmeasurementsweremadeat2/13and11/13ofthedurationofthevowelinanattempttominimizetheimpactofcoarticulation.CentringdiphthongsarenotplottedintheF1/F2plane,mainlybecauseoftheirwithin-dialectvariabilityandespeciallyduetotheinconsistencieselicitedbytheill-suitedtest-wordhured(seeDiscussion).Nevertheless,someofthemwillbeillustratedwithspectrograms. VowelsinBritishIslesaccents Figure3Medianandinterquartilerangeforthe11monophthongsproducedbysixmalespeakersofAcompletedescriptionofeachandeveryvowelforall13accentswouldindeedbeextremelytedious.Instead,wewillfocusonarestrictedsetÐvaryingfromonedialecttothenextÐofthemostnotablephenomena.Results3.1StandardSouthernEnglish(Theaccentservesasareferenceagainstwhichtheremainingaccentsarecompared.Ithasbeendescribedindetailinmanypublications(Nolan1998,Jones2003,Upton2004,Hawkins&Midgley2005,Wells2008,etc.).AsÞgure3shows,thevowelsofwho’d(whicharesupposedtoinstantiatetheFOOTsets,respectively)aremorefrontthanrecentpronunciationdictionariessuggest(Jones2003,Wells2008).ThisÞndingishoweverinaccordancewithup-to-dateacousticÐphoneticdescriptions(Hawkins&Midgley2005,McDougall&Nolan2007).Asfarasclosingdiphthongsareconcerned,Þgure4showsthatthevowelofhasaratherbackstartingelementwhilethatofisratherfront,thedifferencebeingaudiblyperceptible.Inearlierdescriptions,OÕConnor(1973)andGimson(1980)usedthesymbols/and//forthestartingelementof,respectively.Nowadays,althoughJones(2003)andWells(2008)useasinglesymbolfortheÞrstvowelofthevowelplotsprovidedinthesedictionaries(Jones2003:viii;Wells2008:xxiii)explicitlyshowthattheauthorskeepconsideringthattheÞrstelementinisslightlymorebackthanthatof.Incontrasttothelatter,otherrecentpublicationshavesymbolsforthataremuchmoreinlinewithourÞndings.Upton(2004)recordstwovariantsforfora]intraditionalRPand[]incontemporaryRP.He,however,notesjustonepossibilityforfora].Olausson&Sangster(2006)alsouse[)andda](MOUTH).Inareviewofvowelsymbolsindictionaries,WindsorLewis(2003:147)commentsontheuseof[]insteadof[]:accordingtotheauthor,Ô[t]heapparentUptonsuggestionthat//and//havenowreversedtheirrelativestartingpositionsinmainstreamusageisnotsupportedbymyobservationsandIknowofno-oneelseofsuchanopinionÕ.WindsorLewisgoessofarastoclaimthatthesymbols[]and[](for EmmanuelFerragne&Franc¸oisPellegrino Figure4Medianstartingandendpointforthe5closingdiphthongsproducedbysixmalespeakersofwouldstillbetotallydefensible.Inotherswords,heseemstorejectanypossiblephoneticdevelopmentinthepronunciationofthesetwovowels.HealsoremindsthereaderthatwhenGimson(inthefourteentheditionoftheEnglishpronouncingdictionary,in1977)departedfromearlierversionsofJonesÕsdictionarybydropping[]infavourof[],itwasforthesakeofsimplicity,anditcertainlydidnotreßectanyphoneticchange.Now,focusingonthequestionofapotentialreversalbetweenthestartingpointsof(andnotontheprecisephoneticqualityoftheformer),ouranalyses,bothacousticandauditory,suggestthatthespeakersinoursampledohavesuchareversal.Whetheritistheresultofanevolutionaryprocessisstilldebatableandwouldneedmorethoroughresearch(e.g.withagreatersample),butotherfactorscomeintoplayinthiscontroversy.Firstly,theanswercanbeinßuencedbyhowconservativethedeÞnitionoftheaccentis.Forinstance,Upton(2004:219)aimsatdescribingaÔkindofmodern,ÒdilutedÓReceivedPronunciationÕ,Jones(2003:v)describesthepronunciationofBBCnewsreaders,andWells(2008:xix)providestranscriptionsthatÔcoververymuchmorethananarrowlydeÞnedRPÕ.Asfarassampleisconcerned,theassessmentbyaBritishphonetician(seesection2.2)yieldedtypicalityandhomogeneityscoresof3/5;whichsuggeststhatourdeÞnitionof(whichweequatewithwhatotherslabelRP)isindeedquitelax.Secondly,mostofthereferences(especiallydictionaries)quotedheretargetlearnersofEnglishasasecondlanguageandnon-specialistnativespeakers.Asaconsequence,thetranscriptionsreßectacompromisebetweenphoneticaccuracyandsimplicity.Insuchbooks,distinguishingonlybetween[]and[]makesperfectsenseintermsofparsimony;besides,symbolsaremereconventions.Sothisisanotherreasonwhywewillguardagainstinferringpotentialphoneticchangesfromsymbolsusedinbroadtranscriptions.So,fromoursample,wehavenomeansofknowingwhethertheratherbackpositionoftheÞrstelementof(relativetothatof)istheresultofahistoricalprocess,andwecannottelltowhatextentoursamplematchesthepopulationwhosepronunciationisdescribedindictionaries.Wells(1982:310)recordedthisphenomenon(whichhecallsCrossover)inthePopularLondonaccent;giventhatoursubjectshaveallbeenrecordedinLondon,itmaywellbethattheirtypeofStandardEnglishisinßuencedbylocalfeaturessuchastheCrossover.Inoursample,thevowelofharedismostoftenamonophthong(Þgure5),andmorerarelyamonophthongfollowedbyashortoffglidetowardsthecentreoftheF1/F2space.ThisagainsupportsthetranscriptionusedbyUpton(2004)andOlausson&Sangster(2006)(i.e.[ VowelsinBritishIslesaccents Figure5SpectrogramofproducedbythespeakerDMEfrom Figure6SpectrogramofproducedbythespeakerHAKfromandsomewhatcontradictsJones2003andWells2008,whosuggestthattheSQUAREvowelisacentringdiphthong(Upton2004:226claimsthatthediphthongcharacterizestraditionalUnliketheSQUAREvowel,themostfrequentvariantfortheheeredvowelisacentringdiphthong(Þgure6);butmonophthongsarenotuncommon.ThediphthongalnotationcanbefoundunanimouslyinJones(2003),Upton(2004),Olausson&Sangster(2006)andWellsThetest-wordhuredwasquiteproblematic:itveryfrequentlyyielded,withinthesameaccent,between-andalsowithin-speakervariation.Giventheapparentlackofagreementastohowthespeakersphonologicallyinterpretedthisword,wedecidedtoexcludeitfromtherestofouranalysis.3.2Birmingham(TheaccentofBirmingham()isgenerallythoughttopatternwiththenorthernEnglishaccents,partlybecauseneithertheFOOTSTRUTsplitnortheBATHsplitissaidtohaveoccurredintypical(Clark2004).But,accordingtoWells(1982:354),Ômanyeducatednortherners...wouldnotbecaughtdeaddoingsomethingsovulgarastopronounceSTRUTwordswith[]Õ;thus,sociolinguisticvariationmustbeexpected.Thissuggeststhat,sincesociolinguisticfactorshavenotbeencontrolledinthecorpus,theaverageformantvaluesforFOOTSTRUTmaybemisleadinginthattheypotentiallyreßectthepronunciationofbothspeakerswithandspeakerswithouttheFOOTSTRUT EmmanuelFerragne&Franc¸oisPellegrino Figure7Medianandinterquartilerangeforthe11monophthongsproducedbysixmalespeakersofFigure7displaysthemonophthongsofthemalespeakersof.Giventhatthedistributionsofonlyslightlyoverlap,itcanbethoughtthatthemajorityofspeakerslikelyhavetwophonemeshere.Anauditoryanalysisactuallyrevealsthatonlyhalfofthe20(maleandfemale)speakersfromthesamplehaveaperfecthomophonybetween,whichexempliÞeswhycomputingaveragesonformantvaluesmayprovemisleading.InsomespeakerswhosesystemhasseparatevowelsforFOOTSTRUTSTRUTvowelisveryclosetothatofheard),totheextentthatdurationseemstobetheonlyreliablecuetodistinguishthem.Relativeto,whereSTRUThasaratheropenquality,thevowelisclosertothecentreofthevowelspace.ThiscentralqualityforSTRUTistypicalofeducatedandmobilenortherners,asevidencedinaperceptualstudybyEvans&Iverson(2004):whenaskedtoÞndbestexemplarsofSTRUTwordsinacarriersentencespokeninsouthernEnglish,northernsubjectslivinginLondonadjusttheirrepresentationwhilenorthernersstilllivinginthenorthdonotnormalizeforaccent.However,theperceptualadjustmentperformedbymobilenorthernersdoesnotmatchvowelsactuallyproducedbysoutherners;itonlycomesclosetoaschwa-likequality.Now,turningtothediphthongsof(Þgure8),thestartingqualityofslightlylessopenthanin.Actually,anauditoryinspectionincludingallspeakersrevealsthatitisnoticeablybackandclose.Wehavefoundnoevidencesupportingapotentialmergerinvolving(thispossibilityismentionedbyWells1982:363Ð364),althoughinsomespeakersthetwovowelsareperceptuallyveryclosetoeachother.Thetest-wordharedhasalongmonophthongforallspeakers;heered,acentringdiphthong.3.3Cornwall(AccordingtoWells(Wells1982),theaccentofCornwall()clusterswiththeaccentsofthesouth-westofEngland,althoughithasadifferentlinguistictraditionfromsurroundingisrhotic,whichmeansthatthegraphicinthetest-wordshasaphoneticequivalent.Theinßuenceofarealized//ontheprecedingvowelmanifestsitselfindifferentways:forinstance,insomerealizationsofheard,F3collapsesabruptlytowardsthemiddleoftheautomaticallyextractedsignalportion,whichindicatesthatthevowelandthe//occurinsuccession,whileothervoweltokensdisplayastablelowF3throughout,suggestingthatthevowelisr-colouredthroughout.Figure9andÞgure10showthemonophthongsproducedbyeachofthetwospeakers.Auditorily,thevowelisparticularlyvariable;backvowels(Þgure11),comparativelyfrontqualities(Þgure12)anddiphthongizedvariantsareattested. VowelsinBritishIslesaccents Figure8Medianstartingandendpointfortheveclosingdiphthongsproducedbysevenmalespeakersof Figure9MedianF1/F2forthe11monophthongsproducedbyspeakerMJMof Figure10MedianF1/F2forthe11monophthongsproducedbyspeakerRPGof EmmanuelFerragne&Franc¸oisPellegrino Figure11Spectrogramofabackvariantofwho’dproducedbyspeakerRPGfrom Figure12Spectrogramofafrontvariantofwho’dproducedbyspeakerMJMfrom Figure13MedianstartingandendpointfortheveclosingdiphthongsproducedbyspeakerMJMofAsÞgure14suggests,thestartingqualityofhowd)forspeakerRPGisnotasopenasin,whilespeakerMJM(Þgure13)hasacomparativelybackstartingpoint.howd)andGOAT)diphthongstendtopointtothesamedirection,whichwasnotthecasein(Þgure4)wherehowdpointedtohoard,whilepointedto VowelsinBritishIslesaccents Figure14MedianstartingandendpointfortheveclosingdiphthongsproducedbyspeakerRPGofThevowelofharediseitheralongmonophthong(sometimesr-coloured)oramonophthongfollowedbyanapproximant.Fromtheauditoryanalysisofallspeakers,heeredisalmostunanimouslyrealizedasamonophthongplusapproximant.3.4EastAnglia(ThevowelsystemoftheEastAnglia()samplefromtheABIcorpusisverymuchlikethat.Asapreliminaryremark,notethat,inÞgure15,thevowelsofhavebeentreatedasmonophthongs,althoughTrudgill(1999:129)referstothemasdiphthongs.Itistruethatmanyrealizationsdeviatefromperfectlysteadyformantcontours(asisalsothecaseinotheraccents,e.g.).However,forthesakeofconsistencyinthepresentation,andfollowingourauditoryimpressionthatthedegreeofdiphthongizationinwho’dwasfarfrommatchingthatof(etc.),wedecidednottoregardthevowelsofwho’dasdiphthongs.AsÞgure15shows,thevowelsofheard)andSTRUTareconspicuouslyclosetooneanother.AnalternativerepresentationofthesetwovowelscanbefoundinÞgure16,wheretheestimatedprobabilitydensitiesofF1,F2,anddurationareplotted.ThedashedlinestandsforthevowelandthesolidlinerepresentsSTRUT.Theestimateddensitiesfordurationshowsomeoverlap,but,onaverage,isalmosttwiceaslongasSTRUT.TheasymmetryoftheF1curveandthebimodalityoftheF2curveforSTRUT(solidline)tendtosuggestthatthevariationfoundhereisnotamereconsequenceofrandomßuctuationaroundthemean.OurhypothesisÐafterauditoryassessmentÐisthatsomespeakershavearobustspectraldistinctionbetweenthetwovowels(STRUTslightlymoreopenandback),whileothersproducevowelqualitiesthatareclosertooneanother(althoughperceptiblynothomophonous),andtheymaythereforerelymoreheavilyondurationtoimplementthedistinction.InÞgure17wecanseethatthestartingqualitiesof)and)arecomparativelyclosevowels,whileGOAT)hasaratherbackstartingpoint.Thevowelofharedshowsanexclusivelymonophthongalrealizationasdomostrealizationsofheered3.5EastYorkshire(TheaccentofHullhasbeendescribedinWilliams&Kerswill(1999).Asatypicalnorthernaccent,neithertheFOOTSTRUTsplitnorBATH-broadeninghasoccurred.TheSQUAREmergeriscommon(Williams&Kerswill1999:146).EastYorkshireEnglishisnotrhotic.ThemonophthongsoftheEastYorkshire(eyk)sampleareshowninÞgure18.Comparedtomostofthesystemswedescribeinthispaper,eykretainscomparativelybackqualities EmmanuelFerragne&Franc¸oisPellegrino Figure15Medianandinterquartilerangeforthe11monophthongsproducedbysevenmalespeakersof Figure16Probabilitydensityestimatesfor(solidline)and(dashedline)bymalespeakers(F1,F2,andduration).forthevowelsofFOOTSTRUT(thelattertwobeingclearlyonesinglephoneme).TheproximitybetweenSTARTisconÞrmedbyauditoryanalysis:itevenseemsthatsomespeakersrealizethecontrastusingdurationonly.AÞnalremarkablepointregardingÞgure18liesintheoverlappingdistributionsofheardontheF1andF2dimensions.Themediandurationvaluesare166msand281ms,respectively:thetwovowelscannotthereforebesaidtobephonetically(andprobablyphonologically)identical,butitseemsthatsomespeakersonlyrelyondurationtodistinguishthetwo.FACE)andGOAT)vowelsaremonophthongs(Þgure19)althoughsomeclosingdiphthongsareattested.The)vowelshowsnarrowerformantmovements(fromabout[]to[])withadeÞnitefrontvowelasitsstartingpoint.Somemonophthongalrealizationscanalsobeheard;accordingtoWilliams&Kerswill(1999:146)thisvariantisrestrictedtocaseswherethevowelisnotfollowedbyavoicelessconsonant,anditisfoundinworking-classspeechonly.Haredhasamonophthongformostspeakers,inwhichcaseitrhymeswithheardHeeredhasacentringdiphthong(sometimesnearlymonophthongal). VowelsinBritishIslesaccents Figure17Medianstartingandendpointfortheveclosingdiphthongsproducedbyvemalespeakersof Figure18Medianandinterquartilerangeforthe11monophthongsproducedbyeightmalespeakersof Figure19Medianstartingandendpointfortheveclosingdiphthongsproducedbysixmalespeakersof EmmanuelFerragne&Franc¸oisPellegrino Figure20Non-rhoticversionofproducedbythespeakerWNHfrom Figure21Medianandinterquartilerangeforthe11monophthongsproducedbysevenmalespeakersof3.6Glasgow(TheaccentofGlasgow()isrhotic;but,asrecentpublicationshaveshown(Stuart-Smith2007,Scobbie2007),thereisevidencesupportinganongoingchangepossiblyleadingtothelossof//incodaposition.Fromourdata,therealizationsof//varyquitealot(alveolarßaporapproximant,schwa-likevowel,ornoacousticequivalent).Bywayofexample,Þgure20isaspectrogramofanon-rhoticversionofhardproducedbythespeakerWNHof.Asfaraswecantell,theredoesnotseemtobeanyspectralcuesupportingthepresenceofaphoneticeventcorrespondingto//,whichconÞrmsourauditoryimpression.TypicalvowelsystemsofScottishEnglishhavegenerallyundergonetheFOOTmergerandtheLOTmerger(Stuart-Smith1999,2004).Inaddition,theoppositionbetweenNORTHisgenerallypreserved(Wells1982:408).Unfortunately,exceptfortheÞrstonejustmentioned,thesephenomenacouldnotbetestedduetothelackofappropriatetest-words.AlsotypicalofScottishEnglishisthelengtheningofsomemorpheme-ÞnalvowelswhenfollowedbyasufÞxal//aspartoftheScottishVowelLengthRule(Wells1982:400Ð401;Scobbie,Hewlett&Turk1999).Thevowelinwho’dcanthereforebeexpectedtobelongerthanitscounterpartin.AsÞgure21shows,theFOOTmergerseemstobesupportedbyacoustic(spectral)evidence.However,thedifferenceinmediandurationbetweenwho’d(163msand244ms,respectively)suggeststhat,asweanticipated,theScottishVowelLengthRuleappliestowho’dhere.Thisraisesthepossibilitythatthevowelswho’dareactuallyseparatephonemes(wereturntothisintheDiscussion). VowelsinBritishIslesaccents Figure22MedianstartingandendpointfortheveclosingdiphthongsproducedbysevenmalespeakersofOneofthemoststrikingrealizationalfeaturesofisexempliÞedbythetest-wordvowelisindeedquiteretractedtotheextentthatitcanbesaid,fordescriptivepurposes,toclusterwithcentralÐratherthanfrontÐvowels.ThisfeatureagreeswellwithpreviousÞndingsbyEremeeva&Stuart-Smith(2003).Asfarasdiphthongsareconcerned,Þgure22showsthetypicallyScottish[]realizationofthe)vowel.FACE)andGOAT)aremonophthongs.Haredhasamonophthongofthe[]or[]type,followedbya//.Thestartingelementofheeredisthesamevowelasin,anditisalsofollowedby/3.7Lancashire(TheaccentofLancashire()belongstothelinguisticnorthofEngland.MorespeciÞcally,accordingtothetypologyofEnglishdialectsbyHughes,Trudgill&Watt(2005),thetownofBurnleyispartoftheCentralLancashirearea,whichisrhotic,althoughthispeculiarityisreceding.Infact,almostallspeakersinoursamplearenon-rhotic.Wells(1982:367Ð368)alreadynotedthat,althoughLancashireaccentswerepopularlythoughttoberhotic,thisfeaturewasrestrictedtoanevershrinkingnumberofareas.AscanbeseenonÞgure23,theabsenceofFOOTSTRUTsplitiswellattested.Contrarytowhathasbeenobservedforeyk(Þgure18),thevowelhasarelativelyfrontquality.Thevowelsofhardareonaveragefurtherapartthanineyk,althoughpartialoverlapisvisible.ThediphthongsinÞgure24followapproximatelythesamepatternasineykFACEGOAT)arerealizedasmonophthongs,thequalityofthelatterbeingperceptiblymorebackthantheaverageeykrealization.Somespeakershaveaperfecthomophonybetweenhoard.Asineyk)showsrathernarrowformantmovements,buthere,themonophthongismorefrequent.Haredisalongmonophthongandmostoccurrencesofheeredarecentringdiphthongs.Figure25andÞgure26showspectrogramsofamonophthongalandadiphthongalrealizationofproducedbythesamespeaker.TheseÞguresillustratethatwithin-accentvariationoccursbothbetweenspeakersandwithinonesinglespeaker.3.8Liverpool(TheLiverpoolaccent()isalsotypicallynorthern(intermsofvowelsystem)althoughithassalientdistinctivepronunciationfeaturesofitsown.MuchofitsphoneticspeciÞcityseemstostemfromthemassiveintakeofIrishimmigrantsduringthe19thcentury(Wells EmmanuelFerragne&Franc¸oisPellegrino Figure23Medianandinterquartilerangeforthe11monophthongsproducedby10malespeakersof Figure24Medianstartingandendpointfortheveclosingdiphthongsproducedbyninemalespeakersof1982:371).AsKnowles(1978:80)remarks,Ô[itis]aninterestinghybrid:onthephonologicallevel,itremainssimilartothedialectsofneighbouringnortherntowns,butphoneticallyithasbeenheavilyinßuencedbyAnglo-IrishÕ.Themosttypicalsystemicfeature,especiallyinworking-classspeech,istheSQUAREmerger(Beal2004:125).TheFOOTSTRUTsplithasnottakenplaceandBATHbroadeningisattestedfromthemiddleclassesupwards(Watson2006:57).Figure27shows,amongotherthings,theabsenceofaFOOTSTRUTsplit,theproximityhard,andthatofheard.TheauditoryanalysisconÞrmsthesmallspectraldistancebetweenhard;butthedistinctionisneverthelessrenderedbydifferencesinduration.Notealsothat,whilethevowelofremainsback,who’disclearlyfront.HeardhasalongmonophthongwhosequalityissimilartothatofAsÞgure28shows,,contrarytomostnorthernaccents,hasphoneticclosingdiphthongsinFACE)andGOAT).Incidentally,hereagain,theÞrstelementofismorebackinmenthaninwomen.Haredisalongmonophthong,anditishomophonousheard:theSQUAREmergercanbeseenmoreclearlyinÞgure29. VowelsinBritishIslesaccents Figure25MonophthongversionofproducedbythespeakerPDKfrom Figure26DiphthongversionofproducedbythespeakerPDKfrom Figure27Medianandinterquartilerangeforthe11monophthongsproducedbyeightmalespeakersof3.9Newcastle(Newcastle(ncl)clusterswithnorthernEnglishaccents.Inaccordancewithpreviousdescriptions(Watt&Allen2003),thevowelisratherfrontandclose;anauditoryassessmentconÞrmsthat,formostspeakers,thisvowelliesintheregionbetween[¿]and[Ï].Figure30andÞgure31showacomplexpatternfor,andwho’d.Inmost EmmanuelFerragne&Franc¸oisPellegrino Figure28Medianstartingandendpointfortheveclosingdiphthongsproducedbyvemalespeakersof Figure29Probabilitydensityestimatesfor(solidline)and(dashedline)bymalespeakers(F1,F2,andduration).speakers,thelackofFOOTSTRUTsplitisevidentbutsomespeakersÐincludingspeakerTXRinÞgure31Ðrhymewho’d:wethereforelistenedtopotentialFOOTwordsinthereadpassageoftheABIcorpusinordertomakeoutwhetherthehomophonyofwho’dwastheconsequenceofahypotheticalFOOTmergerortheoutcomeofßuctuatinglexicalincidence.Ifthelatterweretrue,itwouldmeanthatbelongstothesetformanyspeakersinournclsample.Wells(1982:362)indicatesthatthemayincludesomewordswiththespellinginthenorth.Asitturnedout,theauditoryanalysisshowedthatthevowelsinfootwereperceptiblydifferentfromthatofwho’d.Itthereforeseemsthatthehomophonybetweenwho’disaresultofpatterningwithinmanynclspeakers,whichemphasizeshowcrucialthechoiceoftest-wordsthataremeanttoelicitthestructureofvowelsystemscanbe.InÞgure32,speakerGGCproducesaclosingdiphthonginFACE)andamonophthonginGOAT).SpeakerTXR(Þgure33)hasacentringdiphthonginandamonophthongin.Thevastmajorityofmalespeakersinournclsample(i.e. VowelsinBritishIslesaccents Figure30MedianF1/F2forthe11monophthongsproducedbyspeakerGGCof Figure31MedianF1/F2forthe11monophthongsproducedbyspeakerTXRofaccordingtoourauditoryjudgementbasedonthewholenclsample,priortothescreeningthatleftuswithjusttwospeakers)haveacentringdiphthonginFACEHaredhasalongmonophthongandheered,acentringdiphthong.3.10NorthWales(AccordingtoPenhallurick(2004:98Ð100),accentgeographyinWaleshasbeentoagreatextentfashionedbythetwodialectsoftheWelshsubstratum(north-westvs.mid-south),thevaryingdegreesofresistancetoanglicization,andtheinßuenceofpronunciationfeaturestypicalofneighbouringEnglishcounties.WelshEnglishisusuallyconsiderednon-rhotic(Wells1982:378);but,intraditionalWelsh-speakingareas(i.e.thewest)orregionsclosetorhoticEnglishcounties,//canbesystematicallyrealizedinallpositions(Penhallurick2004:110Ð111).Asfarasvowelsareconcerned,bothWells(1982:380)andPenhallurick(2004:103)mentionthepossibilityofSTRUTbeingmergedwithschwa.Inthenorth-eastÐrecallfromtable1thatoursamplewasrecordedinDenbighÐsomeSTRUTwordshavethetypicalFOOTvowelasinthenearbyaccentsofthenorthofEngland.Figure34showsthat,incontrastwithmanysystemsdealtwithinthispaper,thevowelisclearlyback.Allthepairsconsistingoftwovowelsseparatedbyasmallspectral EmmanuelFerragne&Franc¸oisPellegrino Figure32MedianstartingandendpointfortheveclosingdiphthongsproducedbyspeakerGGCof Figure33MedianstartingandendpointfortheveclosingdiphthongsproducedbyspeakerTXRofdistanceintheÞgureareeasilydistinguishablebyear:outofthe20speakersinthenwa13apparentlyrelysolelyondurationtoseparatehard,whereastheremaining7usebothdurationandthefront/backaxis.Figure35showsthediphthongsofnwa.ItshouldbenotedthatsomespeakershaveamonophthonginHaredhasalongmonophthongandheered,acentring3.11RepublicofIreland(roiFromWellsÕsaccountofIrishEnglishitseemsthattheaccentofEnglishspokeninSouthernIrelandhasremainedparticularlyimpermeabletoAnglo-EnglishandAmericaninnovations(Wells1982:418).ThepoliticalindependenceoftheRepublicofIrelandisparalleledlinguisticallybythefactthatReceivedPronunciationisconsideredanÔextra-nationalnormnotaspiredtoÕ(Hickey1999:265).IrishEnglishisusuallythoughttoberhotic,andinsomecases,thisrhoticitysurfacesasanr-colouringspanningthewholeoftheprecedingvowel.However,accordingtoHickey(1999:272),rhoticityvariesevenwithinDublin,themorerhoticspeakersbeingassociatedwithhigherprestige.Turningtovowels,thebest-knownsystemicphenomenonisprobablytheSQUAREmerger,whichis,accordingtoWells VowelsinBritishIslesaccents Figure34Medianandinterquartilerangeforthe11monophthongsproducedbysevenmalespeakersof Figure35Medianstartingandendpointfortheveclosingdiphthongsproducedbysevenmalespeakersof(1982:421),notastigmatizedfeature(unlikeinLiverpool).MoredetailsonphonologicalandphoneticphenomenacanbefoundinHickey(1999,2004).Figure36illustratesthemonophthongsofroi.TheoverlapoftosupportthelackofFOOTSTRUTsplit.AccordingtoHickey(2004:91),thisfeaturecharacterizesÔPopularDublinÕasopposedtoÔFashionableDublinÕ.ThediphthongsareplottedinÞgure37.Somerealizationsofaremonophthongal.Dependingonthespeaker,thestartingqualityofiseitherfrontorback.Thevowelinthetest-wordrangesfromaquasi-monophthongtoadiphthong.Inmanyspeakers,theÞnalelementofhowdisclearly[3.12ScottishHighlands(ThetownofElginissituatedattheeasternendoftheScottishHighlands.ScottishGaelicisstillsporadicallyspokeninthisarea,hencetheinßuenceofGaelicphonologyÐespeciallyconsonantalphenomenaÐonthevarietyofEnglishspokenthere(Wells1982:412Ð414;Stuart-Smith2004:50).Contraryto,rhoticityismaintainedinoursamplebyallspeakers;someofthemhaveatrillratherthananapproximant. EmmanuelFerragne&Franc¸oisPellegrino Figure36Medianandinterquartilerangeforthe11monophthongsproducedbysevenmalespeakersof Figure37MedianstartingandendpointfortheveclosingdiphthongsproducedbysixmalespeakersofThevowelinisnotascentralizedasitscounterpart;inaddition,thecentralizedvariantisnotasfrequentasin(seeÞgure38foraclosevariantandÞgure39foracentralvariant).ContrarytowhatweexpectedfromaScottishaccent,Þgure40doesnotseemtosupportapotentialFOOTmerger.Acloserlookatthispair(Þgure42)showsthatwho’dhavesimilarF1anddurationvalues,butthedistributionsofF2valuesdonotoverlap.So,atleastsomespeakershaveanacousticdifferenceintheF2dimension.Fromthedurationdistributions,itcanbeconcludedthatcontrarytowhatwehaveobservedforwho’dnotacandidatefortheScottishVowelLengthRule.Actually,anauditoryassessmentshowsthattherealizationsofwho’darenothomogeneousacrossspeakers.Whilemostofthemproduceidenticalvowels,othershaveaspectraldifference.hoardareveryclosetoeachotherinspectralquality,the//inhoardstillmarksthedifference.ButtheproximitybetweenthetwovowelsleadsustothinkthatatleastsomespeakershaveaLOTmerger(which,ofcourse,couldnotbeelicitedforwantofanadequatetest-word).FACEvowel(Þgure41)isunanimouslymonophthongal,andsoisGOAT;bothareparticularlyclose,totheextentthat)probablyhasthevowelthatcomesclosesttocardinal[]inthewholecorpus.Thevowelinisofthe[]type. VowelsinBritishIslesaccents Figure38SpectrogramofaclosevariantofthevowelinbyspeakerCNBfrom Figure39SpectrogramofacentralvariantofthevowelinbyspeakerGDWfrom Figure40Medianandinterquartilerangeforthe11monophthongsproducedby11malespeakersof3.13Ulster(EventhoughtheprovinceofUlsterbelongstothesamegeographicalentityastheRepublicofIreland,UlsterEnglishsharesmanypronunciationfeatureswithScottishEnglish,mainlyonaccountofanimportantinßuxofScotssettlersinthe17thcentury(Hickey2004:68).Theaccentisrhotic;//isgenerallyrealizedasaretroßexapproximant(Wells1982:446). EmmanuelFerragne&Franc¸oisPellegrino Figure41Medianstartingandendpointfortheveclosingdiphthongsproducedby11malespeakersof Figure42Probabilitydensityestimatesfor(solidline)andwho’d(dashedline)bymalespeakers(F1,F2,andduration).Thevowelsystemandthephoneticqualityofvowels(asdescribedinWells1982:438Ð440)shareanumberoffeatureswithScottishEnglish.Forinstance,vowellengthplaysarestrictedsystemicrole.FOOTareonesinglephoneme,andsoare(potentially)LOTThemonophthongsofareplottedinÞgure43.AmongthemostnotablefeaturesistheoverlapbetweenFOOT)andwho’d)andtheretractedqualityof).TheFOOTvowelsseemtobemorefrontthantheircounterparts;thelengtheningofwho’disnotunanimous.InÞgure44,thearrowcorrespondingtopointstowardsthetopoftheplot:thisisprobablyanartefactduetothemultiplevariantsproducedbythespeakersinthesample.Thegeneral(auditory)picturesuggeststhatthemainvariantisacentringdiphthong;monophthongsandclosingdiphthongscanalsobefound.TheÞrstelementofthevowelisalsoquitevariable.AsÞgure44suggests,itisquiteopenformostspeakers.Haredheardarehomophonousin14(outof20,menandwomeninclusive)speakers(hencethe VowelsinBritishIslesaccents Figure43Medianandinterquartilerangeforthe11monophthongsproducedbysixmalespeakersof Figure44MedianstartingandendpointfortheveclosingdiphthongsproducedbyvemalespeakersofSQUAREmergerisattestedforhalfofthespeakers).Thevoweliseitherrealizedas[or[3.14SummaryTheF1/F2plotsaboveillustratethevowelsystems(derivedfromacousticdata)of13accentsoftheBritishIsles.Butthesegraphsshouldbegeneralizedwithcaution.Thespectrogramsandprobabilitydensityplotsthathavebeenincludedhereclearlyillustratethatwithin-dialectvariationoccursbothattheinter-andintra-individuallevel.Now,hadwecarriedoutaqualitativeanalysis,itwouldhavebeeneasiertoshowhowmanyspeakersofanaccentproducesuchorsuchvariantofagivenphoneme.ButinourcaseÐi.e.withquantitativeformantdataÐmeasuresofcentraltendencywithoutpriorqualitativeassessmentcanleadtocomputingaveragesoverformantdataemanatingfromverydifferentphoneticevents.Forinstance,computingaverageformantvaluesoverthethreevariantsoftheFACEvowelinncl(monophthong,centringandclosingdiphthong)withoutdistinguishingeachvariantaprioriisnotreallyinformative.Butfromwhatdegreeofformantinstabilityshouldweconsiderthatthevowelunderstudyisadiphthongratherthanamonophthong?Shouldwetrustour EmmanuelFerragne&Franc¸oisPellegrinoTable3Medianformantvaluesforthe11monophthongs(Hz). AccentFormantheedhidheadhadhardhodhoardhoodwho’dHuddheard F1289350502679639576454414318482491F22219205818111479110312938821194162012461573F1285402500641629556461400332569508F22346202218181560120310629051400129713651568F1335407499692702580428416325569562F22277214220471717124111438351509166615771570F1281394560700725578551399278426588F22266205818731463131611149881210103512161717F1301446473636693530463327345480543F22164178020461490117812159731723175115451561F1310423576697689615571483354485542F222762024181114541112113810371144174611301575F1299465607730638599536491328496488F222111854163113931200118210331126169011301796F1279430514694633591489382309423478F22263179316681333102010968371079108411291492F1276444596762767612541493283545471F2228019251733141513001056954117299915401567F1276420561709639674495484317509529F222471912183115101491121411001212155512091552F1248364410558618439434258258427532F222171849194913351183120710731587144115141646F1273386527751655552452397291623527F2228920381801155810449867931550167213701528F1279413573681642614480334376466537F222001813182514951347118310971747175413291553 judgementastrainedphoneticians,orasknaõvenativespeakers;inwhichcase,shouldtheyjustbenativespeakersofEnglishorofTynesideEnglish?Despitethisweakness,themedianformantvaluesforeachmonophthongineachdialectareshownintable3.ThevaluesareinHertzsinceitseemstousthatitisthemostwidelyusedfrequencyscale,anditthereforemakescomparisonswithotherpublicationseasier.Followingourtwofoldgoal,thissectionÞrstprovidesadiscussionoftheÞndingsandthentacklessomemethodologicalissues.Overall,onthesystemiclevel,ourÞndingsagreewellwithpreviousdescriptions.However,insomeaccents,considerablevariationbetweenspeakerscanbeobserved.Forinstance,aswehaveseen,thesampleisequallydividedbetweenspeakerswith,andspeakerswithout,theFOOTSTRUTsplit.AccordingtoWells(1982:352),intheWestMidlandsarea,allspeakershavetwophonemeshere,buttheoppositionmaybeneutralizedinsomecases,andlexicalincidenceisuncertain.Thelistof//wordsdidnotallowustotestpotentialneutralizationsorproblemsoflexicalincidence,butitisclearfromourdatathat,ifSTRUTvowelisthekeycriterionfortellingtheEnglishlinguisticnorthfromitssoutherncounterpart,then,Birminghamcanprobablybeseenasatransitionalarea.Theotheraccents VowelsinBritishIslesaccentsofthenorthofEnglandrepresentedintheABIcorpusare,inthisrespect,undoubtedlynorthern.SQUAREmergeristhenorminthefollowingsamples:eykroi,andFromourdata,theFOOTmergerseemstohold,atleastpartially,for,and,thevowelsofwho’darespectrallyidenticalbutthelatterissystematicallylonger,asaconsequenceoftheScottishVowelLengthRule.In,thelengtheningofwho’dnotsystematic,andthereseemstobeadifferenceinF2.In,thereisnospectraldifferencebetweenthetwovowels,andlengtheningoccursinmanyspeakers.Thedifferencebetweenthemediandurationofwho’dandthatof,andare81ms,32ms,and32ms,respectively.ThisclearlyshowsthatthemanifestationoftheScottishVowelLengthRulethroughvowellengtheningcausedbysufÞxal//isprobablymoretypicalofGlasgowthantheothertwo.Throughoutthearticle,who’dhavebeenusedastypicalmembersoftheFOOTsets.Wesawinournclsamplethepossibilitythatmayactuallybeamemberforsomespeakers,whichissimplyaconsequenceofßuctuatinglexicalincidence.Amorechallengingsituationariseswhenitcomestousingwho’dasatypicalmemberofinthoseaccentsweretheScottishVowelLengthRuleapplies.Thevowelinwho’dbeinterpretedaspartofasubsetofwordswhosevowelislongduetomorphologicalconditioning.Inordertofacilitatedialectologicalanalysis,itmaybeagoodideatosignalthedistinctionbyusingadifferentkeyword(e.g.)asiscommonpracticewhenthephonologyofanaccentrequiresit(Foulkes&Docherty1999:7;foranexampleoflexicalsetsadaptedtolocalneeds,seee.g.Stuart-Smith2004:53Ð56).Asfortheinclusionofthisvowelinthephonemicinventory,who’dcanbeseenasaminimalpairinandtheirdifferenceinmedianduration(81ms)isprobablysufÞcienttomaintainaviablephonologicalopposition(Bywayofcomparison,thedifferenceinmediandurationfortheduration-onlybasedcontrastbetweenhardeykis106ms).So,surely,thevowelinwordsisagoodcandidateforphonemicstatusbut,quitecircularly,italldependsonhowgradientthephonologicalanalysisusedtointerpretthisphenomenonis(seeChitoran&Cohn,toappear,foranoverviewofgradientvs.categoricalphonology).Turningtonotablephenomenaontherealizationallevel,thequalityofthevowelisparticularlyvariableacrossaccents.But,aswehaveshownelsewhere(Ferragne2008:295Ð296),itdoesnotconstituteareliableaccentindicatorbecausebetween-accentvariationiscompensatedbyhugewithin-accentvariability.Thegeneraltrendshowsanincreasinglyfrequentfrontedqualityin.Insomeaccents,thisistheresultofarecentchange(e.g.:Hawkins&Midgley2005,McDougall&Nolan2007),whileinothers(e.g.:Wells1982:402;Stuart-Smith2004:58Ð59),itisalong-establisheddiagnostictrait.Someoftheaccentsinourdatasetareratherconservativewithrespecttoeyknclnwaroi(notfullyback,though),and(asopposedto,butnotsystematically).Wherefrontingoccurs,itishardtoseeorforeseetheconsequencesofthisphenomenon,anditisdifÞculttotellwhetherthiscouldleadtogenuinechainshifting.LabovÕsthirdprincipleofvowelshiftingstatesthatbackvowelsmovetothefront(Labov1994:116ff.);asithappens,theprincipleagreeswellwithFOOTinourdataformanyaccents(especially).Note,however,thatinfrontingisnotaccompaniedbyFOOTfronting.Itmaywellbethatourdatashowthreedifferentstagesofapotentialpartialchainshift:(i)neitherFOOTseemstohavemovedfromitsbackposition(typicallyineyk)Ðincidentallythereisnothinginourdatathattellsusthatthechangewilleverapplyhere;(ii)only,butnotFOOT,hasmovedtoafronterposition();and(iii)FOOTareratherfront(typicallyinChainshiftingnormallyimpliesthatthemovingentityinthesystemeitherÔdragsÕorÔpushesÕadjacentphonemes.Concerningapotentialpush-chainmechanism,thisseemsveryunlikely,fortworeasons.Firstly,fromourdata,thevoweldoesnotseemtodeparttoanynoticeabledegreefromitspositioninpreviousstudies.Secondly,vowelsystemsintheworldÕslanguagescaneasilyÔhandleÕtwofrontclosevowelsÐonerounded,theother,notÐas EmmanuelFerragne&Franc¸oisPellegrinoinFrench,German,Norwegian,etc.(24outof451languageshavethetwophonemesintheUCLAPhonologicalSegmentInventoryDatabase(UPSID);Maddieson1984,Maddieson&Precoda1990).Thus,notonlyistherenoevidenceshowingapossiblemovementofbutalso,thereisnointrinsicreasonwhythecurrentpositionofshouldbethreatenedbythefrontingof.Now,adrag-chainmechanismÐwhichwouldcausehoard,ormonophthongal,tomoveonestephigherÐremainspossible,allthemoresoas,accordingtoLabovÕssecondprinciple,longvowelsriseinchainshifts(Labov1994:116ff.).IfweÐtwoaccentswhereFOOTarecomparativelyfrontÐwitheyknwaÐtwoaccentswhereFOOTarecomparativelyback,itappearsthathoardisclearlymorecloseintheformer,whichisconsistentwithadrag-chainmechanism.However,onlyalongitudinalstudywouldshedlightonthis.Forthetimebeing,thedataonRPinHawkins&Midgley(2005)donotsupportourtentativehypothesis:theirresultsshowthat,whileF2inwho’dmovedfrom994Hzonaverage(maleover65yearsold)to1616Hz(malebetween20and25yearsold),F1inhoardremainedconstant(391Ð392Hz).Rhoticityconstitutesanotherimportantaspectofaccentvariation,whichmayhavedramaticconsequencesonvowelsystems.OurresultstendtoconÞrmongoingchanges.Contrarytoourexpectations,oursampleisvirtuallynotrhotic.Thelossof//hereisprobablyjustacontinuationofthewell-knownchangeinitiatedinRPintheeighteenthcentury.ThischangehasbeenassociatedwithhighprestigeinEngland,notablybecause,accordingtoTrudgill(1990:53),BBCnewsreadersadoptednon-rhoticityveryearly.Butclearly,thefactthatnon-rhoticitycorrelateswithhighsocialstatusinEnglanddoesnotnecessarilyapplytoneighbouringcountries.Forexample,inherreviewofrhoticityinScottishEnglish,Stuart-Smith(2004:62Ð63)showsthattheongoinglossof//ismainlyaworking-classphenomenon.Note,incidentally,thatderhoticizationinGlasgow(Stuart-Smith2007)isagradientphenomenoninthatthetotalabsenceofaphoneticevent(asshowninÞgure20)isnottheonlypossibility.Besides,amereperceptualanalysismaynotbesufÞcient:Scobbie(2007)usedultrasoundtongueimagingtoshowthathisacousticallyandauditorilyÔderhoticizedÕsubjectretainedacovertrhotic-liketonguegesture.TheconsequencesofderhoticizationonvowelsystemshavebeenespeciallywelldocumentedinRP(see,amongothers,Wells1982:213Ð222).OurdatasuggestthatthemonophthongingofcentringdiphthongsÐwhichWellssawasapotentialsubsequentdevelopment(centringdiphthongsinRPbeingtheresultof//loss)Ðseemstobethenormforthetest-wordhared,butnotforheered.TowhatextenttheformerresultmaybeinterpolatedtoothermembersoftheSQUAREset,wedonotknowÐbearinmind,though,thatUpton(2004:226)encouragestheuseofamonophthongsymbolforthewholelexicalsetÐbutifmonophthonginggoesontoaffectallcentringdiphthongs(asitalreadydoesthevowelÐWells1982:361),phonologicallengthwillbecomeanincreasinglyimportantparameterintheRPsystem.Post-//-lossmonophthongingisattestedinourdatainsomeaccentstovariousdegrees:monophthongalharedcanbeheardinallnon-rhoticaccentsand,veryoften,themonophthongismorefrequentthanthediphthong.SoiftheseaccentsfollowthesamepathasRP,thenincreasingimportanceofdurationasaphonologicalfeaturemaywellaffectthemtoo.Nowweturntomethodologicalissues.Exceptforthesemi-automaticprocedurewherebyformantvalueswerecomputed,themethodwehaveusedinthisdescriptionisfairlyclassicandconsensual.However,itinvolvesanumberofshortcomingswhichwewouldliketocommentupon,namely,thebiascausedbyz-scoringthedata,andtheweaknessesofthewordThez-scoretransformisexpectedtoreducewithin-dialectvariabilitybyplayingdowntheimpactofvariablevocaltractsizesbetweenspeakers.Infact,thetransformaltersallindividualF1/F2plotssothattheyhavesimilarmeanandspread.Whilethisisconvenientforwithin-accentnormalization(insomeway,wewantspeakersofthesameaccenttobeassimilaraspossible),italsoerasespotentialbetween-accentdifferences.Itisgenerally VowelsinBritishIslesaccents Figure45Meanand95%condenceintervalsformalespeakers.acknowledgedthatpermanentfeaturesÐsubsumedundergenericheadingssuchasÔvoicequalityÕandÔarticulatorysettingsÕÐplayanimportantroleinaccentdistinction(see,,Stuart-Smith1999:211ff.),andsomeofthesefeaturesmaywellbereßectedinvowel-spacecentroidandspread,i.e.exactlytheinformationthathasbeenlostafterthez-scoretransform.Asapreliminarycheck,Þgure45showsmeansand95%conÞdenceintervalsofvowelsystemcentroidsperaccentintheF1andF2dimensionsforthemalespeakers.MeanF1andF2valuesinBarkwereÞrstcomputedforeachtypeofmonophthong,yielding11F1and11F2valuesperspeaker.Thenindividualcentroidswerecalculatedbyaveragingthese11values.Finally,themeanandstandarddeviationforeachaccentwereobtainedbycomputingmeanindividualcentroids.TheresultsintheÞguretendtosuggestthattherearedifferencesincentreofgravitybetweendialects.Now,whetherthisisamereconsequenceofasamplingbiasoragenuinedifference,wecannottell.Butifitcouldbeshownthatthedifferenceisreliable,thenitdeÞnitelyshouldbetakenintoaccount.Thiscouldbeachieved,forinstance,bycomputingcentroidsforallaccents,andaddingtheappropriateaccentcentroidvaluestoaspeakerÕsF1andF2z-scoredvalues.Now,letusconsiderthewordlist,and,inparticular,itscomprehensiveness.ThewordlistusedfortheABIcorpusismeanttoelicitthephonemesofwhatwehavecalledStandardSouthernEnglish().Relevantquasi-phonemiccontrasts(e.g.BATH)cannotbeelicited.Whatismore,whenrecordedbyspeakersfromotherregions,thelistallowsfortestingifthespeakershavefewervowelphonemes(than),butitisunfortunatelyinadequateifonewantstoaddressthequestionofapotentialgreaternumberofphonemes(orstandardlexicalsets).Forinstance,itseemsthatthereisadistinctioninsomeaccentsinthenorthofEnglandbetweenthevowelsofwaitweightighte]and[],respectively;Beal2004:123).Thephoneticvalueofthevowelinnclhasbeenshowntobeconditionedbythefollowingconsonant:[]beforevoicelessstopsandfricatives,[]elsewhere(Watt&Milroy1999:28Ð29).Inaddition,inworking-classspeech,thevowelinHull(eyk)isadiphthongbeforevoicelessconsonantsandamonophthongelsewhere(Williams&Kerswill1999:146).TheScottishVowelLengthRule(Scobbieetal.1999)alsogeneratesdialect-speciÞcpronunciations,bothintermsofdurationandspectralquality.AÞnalexample,takenfromWells(1982:421),showsthatin(southern)IrishEnglish,thewords EmmanuelFerragne&Franc¸oisPellegrinopurrcanhavethreeortwoseparatephonemesorjustonesinglephonemedependingonthespeciÞcaccent.Itfollowsfromthesechosenexamplesthatsuchvariation,whateveritslinguisticstatusÐi.e.whetheritisphonemic,quasi-phonemic,lexical-incidental,allophonic,etc.Ðdoesplayanindexicalroleandshouldthereforebeelicitedthroughappropriatetest-words.Anotherkeyelementrelatedtothewordlistliesinthatsometest-wordswerelikelynotwell-suitedtoelicittheintendedlexicalset.Asmentionedatthebeginningofthearticle,thetest-wordhuredprovedproblematiconaccountofgreatwithin-accentvariation,bothacrossspeakersandwithinasinglespeaker.ThisturnedouttobequitefrustratingsincesetisknownÐasaresultofanÔall-EnglandtrendÕ(Wells1982:361)Ðtohaveundergoneaphonologicalchangeinrecentyears.Bywayofexample,theresultsofanopinionpollonpronunciationpreferencesinWells(2008:628)showthat74%ofpeoplesurveyedfavoured//ratherthan//(26%)in;andthedifferenceappearstobecorrelatedwithage.Nowwhyhuredtriggeredsomuchvariationandhesitationremainsunclear.Rarityorobsolescencecannotcountastheonlyfactorssinceotherequallyrare(ornon-existent)test-words(heered,etc.)elicitedtheintendedlexicalsetperfectlywell.Atentativeanswerwouldbethatitseemsherethatthespelling-to-pronunciationcorrespondenceisnotasstraightforwardasinotherwords;but,surely,othertest-words(cure,etc.)wouldbeneededtocheckifsomebodyuses//inwordsandwhethertheirhasbeentotallyoronlypartiallyaffected.Ourgoalhasbeentoprovideanup-to-dateacousticdescriptionofthevowelsof13accentsoftheBritishIsles.Foreachaccent,F1/F2graphsformonophthongsanddiphthongswithz-scoredBarkfrequencieshavebeendiscussedwithparticularfocusonpairsofvowelswhosetwomemberswereconspicuouslyclosetoeachother.Somesuchpairswereinspectedmorecloselywiththehelpofprobabilitydensityestimateplots.Spectrogramswerealsoincludedinordertoexemplifytheextentofwithin-dialectindividualÐandsometimeswithin-speakerÐvariation.Atablecontainingformantfrequenciesattemporalmidpointforthe11monophthongsinthe13accentshasalsobeenprovidedsoastofacilitatecomparisonwithotherstudies.Overall,thestudywascomplicatedbyindividualvariation,whichwouldhaveprobablybeenreduced(thoughsupposedlynoteliminated)byastratiÞedsamplingstrategyinvolvingÞne-grainedcriteria(age,socialfactors)leadingtoclusteringspeakersintomorehomogeneoussub-accents(seee.g.Labov2001).Thesecondmainobstaclearisesfromtheuseofautomaticformantextraction.Asstatedpreviously,formantmeasurementsgenerallynecessitatehumanintervention,whichraisesreproducibilityissues.Onthecontrary,ourmethodisentirelyreproduciblebutitledtosigniÞcantlyreducingthenumberofvowelsandspeakersand,despitetheuseofrobustestimatorsandsmoothingtechniques,somefrequencyvaluesweredeÞnitelyerroneous.WhiletheÞrstdrawbackmentioned(lackofserioussamplingforspeakerrecruitment)canhardlybeovercomewithoutrecordinganothercorpus,ourfutureeffortswillconcentrateonotherfrequencyparameterswhoseextractionistotallyAcknowledgementsTheauthorswouldliketothankJohnEsling,FrancisNolan,JaneStuart-Smithandtwoanonymousreviewersfortheirinvaluablehelp. VowelsinBritishIslesaccentsReferencesAdank,Patti,RoelSmits&RoelandvanHout.2004.Acomparisonofvowelnormalizationproceduresforlanguagevariationresearch.JournaloftheAcousticalSocietyofAmerica116(5),3099Ð3107.Aylett,Matthew&AliceTurk.2006.Languageredundancypredictssyllabicdurationandthespectralcharacteristicsofvocalicsyllablenuclei.JournaloftheAcousticalSocietyofAmerica119(5),3048ÐBeal,Joan.2004.EnglishdialectsinthenorthofEngland:Phonology.InSchneideretal.(eds.),113Ð133.Boersma,Paul&DavidWeenink.2008.Praat:Doingphoneticsbycomputer(version5.0.30).http://www.praat.org/(7August2008).Chitoran,Ioana&AbigailC.Cohn.Toappear.Complexityinphoneticsandphonology:Gradience,categoriality,andnaturalness.InFrancüoisPellegrino,EgidioMarsico,IoanaChitoran&Christophee(eds.),Approachestophonologicalcomplexity.Berlin:MoutondeGruyter.Clark,Urszula.2004.TheEnglishWestMidlands:Phonology.InSchneideretal.(eds.),134Ð162.DÕArcy,Shona,MartinJ.Russell,SueR.Browning&MikeJ.Tomlinson.2004.TheaccentsoftheBritishIsles(ABI)corpus.Inelisationpourl’identicationdeslanguesetdesvariesdialectalesParis,115Ð119.DeWet,Febe,KatrinWeber,LouisBoves,BertCranen,SamyBengio&HerveBourlard.2004.Evaluationofformant-likefeaturesonanautomaticvowelclassiÞcationtask.JournaloftheAcousticalSocietyof116(3),1781Ð1792.Eremeeva,Viktoria&JaneStuart-Smith.2003.AsociophoneticinvestigationofthevowelsOUTandBITinGlaswegian.15thICPhS,Barcelona,1205Ð1208.Evans,BronwenG.&PaulIverson.2004.Vowelnormalizationforaccent:AninvestigationofbestexemplarlocationsinnorthernandsouthernBritishEnglishsentences.JournaloftheAcousticalSocietyofAmerica115(1),352Ð361.Everitt,BrianS.,SabineLandau&MorvenLeese.2001.Clusteranalysis.London:Arnold.Ferragne,Emmanuel.2008.Etudephonetiquedesdialectesmodernesdel’anglaisdesIlesBritanniques:versl’identicationautomatiquedudialecte.Ph.D.dissertation,UniversityofLyon.Ferragne,Emmanuel&FrancüoisPellegrino.2007.AutomaticdialectidentiÞcation:AstudyofBritishEnglish.InChristianMuller(ed.),SpeakerclassicationII,243Ð257.NewYork:Springer.Foulkes,Paul&GerardJ.Docherty(eds.).1999.Urbanvoices:AccentstudiesintheBritishIsles.London:Arnold.Gay,Thomas.1968.Effectofspeakingrateondiphthongformantmovements.JournaloftheAcousticalSocietyofAmerica44(6),1570Ð1573.Gay,Thomas.1978.Effectofspeakingrateonvowelformantmovements.JournaloftheAcousticalSocietyofAmerica63(1),223Ð230.Gimson,A.C.1980.AnintroductiontothepronunciationofEnglish.London:Arnold.Hawkins,Sarah&JonathanMidgley.2005.FormantfrequenciesofRPmonophthongsinfourage-groupsofspeakers.JournaloftheInternationalPhoneticAssociation35(2),183Ð199.Hickey,Raymond.1999.DublinEnglish:Currentchangesandtheirmotivation.InFoulkes&Docherty(eds.),265Ð281.Hickey,Raymond.2004.IrishEnglish:Phonology.InSchneideretal.(eds.),68Ð97.Hughes,Arthur,PeterTrudgill&DominicWatt.2005.Englishaccentsanddialects:AnintroductiontosocialandregionalvarietiesofEnglishintheBritishIsles.London:HodderArnold.Jones,Daniel.2003.Englishpronouncingdictionary,editedbyPeterRoach,JamesHartman&JaneSetter.Cambridge:CambridgeUniversityPress.Knowles,Gerry.1978.ThenatureofphonologicalvariablesinScouse.InPeterTrudgill(ed.),SociolinguisticpatternsinBritishEnglish,80Ð90.London:Arnold.Labov,William.1994.Principlesoflinguisticchange:Internalfactors.Malden,MA:Blackwell.Labov,William.2001.Principlesoflinguisticchange:Socialfactors.Malden,MA:Blackwell.Lobanov,B.M.1971.ClassiÞcationofRussianvowelsspokenbydifferentspeakers.JournaloftheAcousticalSocietyofAmerica49(2),606Ð608.Maddieson,Ian.1984.Patternsofsounds.Cambridge:CambridgeUniversityPress. EmmanuelFerragne&Franc¸oisPellegrinoMaddieson,Ian&KristinPrecoda.1990.UpdatingUPSID.UCLAWorkingPapersinPhoneticsMcDougall,Kirsty&FrancisJ.Nolan.2007.Discriminationofspeakersusingtheformantdynamicsof/inBritishEnglish.16thICPhS,Saarbrucken,1825Ð1828.Nolan,FrancisJ.1998.TheshiftingsandsofEnglishpronunciation.2ndInternationalTartuConferenceonBritishStudies,Tartu,UniversityofTartu,76Ð93.OÕConnor,JosephD.1973..London:Penguin.Olausson,Lena&CatherineSangster.2006.OxfordBBCguidetopronunciation.Oxford:OxfordUniversityPress.Penhallurick,Robert.2004.WelshEnglish:Phonology.InSchneideretal.(eds.),98Ð112.Schneider,EdgarW.,KateBurridge,BerndKortmann,RajendMesthrie&CliveUpton(eds.).2004.AhandbookofvarietiesofEnglish:Phonology.Berlin:MoutondeGruyter.Scobbie,JamesM.2007.Biologicalandsocialgroundingofphonology:Variationasaresearchtool.,Saarbrucken,225Ð228.Scobbie,JamesM.,NigelHewlett&AliceTurk.1999.StandardEnglishinEdinburghandGlasgow:TheScottishVowelLengthRulerevealed.InFoulkes&Docherty(eds.),230Ð245.olander,Kare.2004.SnackSoundToolkit(version2.2).http://www.speech.kth.se/snack/(7AugustStuart-Smith,Jane.1999.Glasgow:Accentandvoicequality.InFoulkes&Docherty(eds.),203Ð222.Stuart-Smith,Jane.2004.ScottishEnglish:Phonology.InSchneideretal.(eds.),47Ð67.Stuart-Smith,Jane.2007.Asociophoneticinvestigationofpostvocalic//inGlaswegianadolescents.,Saarbrucken,1449Ð1452.Traunmuller,Hartmut.1990.Analyticalexpressionsforthetonotopicsensoryscale.JournaloftheAcousticalSocietyofAmerica88(1),97Ð100.Trudgill,Peter.1990.ThedialectsofEngland.Malden,MA:BasilBlackwell.Trudgill,Peter.1999.Norwich:Endogenousandexogenouslinguisticchange.InFoulkes&Docherty(eds.),124Ð140.Upton,Clive.2004.ReceivedPronunciation.InSchneideretal.(eds.),217Ð230.Watson,Kevin.2006.Phonologicalresistanceandinnovationinthenorth-westofEngland.EnglishToday22(2),55Ð61.Watt,Dominic&WilliamAllen.2003.TynesideEnglish.JournaloftheInternationalPhoneticAssociation33(2),267Ð271.Watt,Dominic&LesleyMilroy.1999.PatternsofvariationandchangeinthreeNewcastlevowels:Isthisdialectlevelling?InFoulkes&Docherty(eds.),25Ð46.Wells,JohnC.1982.AccentsofEnglish.Cambridge:CambridgeUniversityPress.Wells,JohnC.2008.Longmanpronunciationdictionary.Harlow:Pearson.Williams,Ann&PaulKerswill.1999.Dialectlevelling:ChangeandcontinuityinMiltonKeynes,ReadingandHull.InFoulkes&Docherty(eds.),141Ð162.WindsorLewis,Jack.2003.IPAvowelsymbolsforBritishEnglishindictionaries.JournaloftheInternationalPhoneticAssociation33(2),143Ð152.Wright,Richard.2003.Factorsoflexicalcompetitioninvowelarticulation.InJohnLocal,RichardOgden&RosalindTemple(eds.),PapersinLaboratoryPhonologyVI:Phoneticinterpretation75Ð87.Cambridge:CambridgeUniversityPress. Formantfrequenciesofvowelsin13accentsoftheBritishIslesEmmanuelFerragne&Franc¸oisPellegrinoLaboratoireDynamiqueduLangage,UMR5596CNRS,UniversiteLyon2Emmanuel.Ferragne@univ-lyon2.frFrancois.Pellegrino@univ-lyon2.fr