JournalofPhonetics45(2014)12–25 ContentslistsavailableatScienceDirect Journal of Phonetics journal homepage: www.elsevier.com/locate/phonetics Synchronic and diachronic factors in the change from pre-aspiration to post-aspiration in Andalusian Spanish ⁎ Hanna Ruch , Jonathan Harrington InstituteofPhoneticsandSpeechProcessing,Ludwig-MaximiliansUniversityofMunich,Schellingstr.3,D80799München,Germany A R T I C L E I N F O A B S T R A C T Articlehistory: Thestudyisconcernedwiththemechanismsbywhichpre-andpost-aspirationarerelatedsynchronicallyand Received4April2012 diachronically in Andalusian Spanish. An analysis of isolated word productions by 24 speakers of a Western Receivedinrevisedform Andalusianandby24ofanEasternAndalusianvarietyeachdividedintotwoagegroupsprovidedevidencefora 16February2014 soundchangeinprogress:youngerspeakersandthoseoftheWesternAndalusianvarietyweremorelikelyto Accepted28February2014 produce/st/withashorterpre-aspirationandwithalongerpost-aspirationphasethantheirolderandEastern Andalusiancounterparts.Theresultsoffurtherbetween-groupdurationanalysesshowthatthesoundchange Keywords: cannotbeentirelyexplainedbyarealignmentfromananti-phasetoanin-phasetimingrelationshipbetweenthe Pre-aspiration closureandglottalopening;theyalsoprovideevidenceforthedevelopmentofatradingrelationshipbetweenthe /s/-Aspiration durationoftheclosureandthedurationofpost-aspiration.Inordertotestwhetherperceptualfactorsmighthave Soundchange contributed to this sound change, a perception experiment was carried out. Listeners of Argentinian Spanish AndalusianSpanish judgedstimuliofaminimalpairpasta-patawithvariablepre-aspirationdurationinaforced-choiceperceptiontest. VOT Whenthecontinuumwassynthesizedwithaslightlypost-aspiratedstoprelease,listenersweremoreinclinedto perceivepasta,whichsuggeststhat post-aspirationisparsedwithpre-aspiration andservesasacueforthe underlying/st/eveninanon-post-aspiratingvarietyofSpanish. &2014ElsevierLtd.Allrightsreserved. 1. Introduction Aspirationisproducedwhentheabductionofthevocalfoldsforavoicelessobstruent,typicallyastop,istimedasynchronouslywithrespecttothe oralconstrictionandisanticipated/prolongedduringthepreceding/followingsonorant.Inthevastmajorityoflanguagesinwhichaspirationoccurs,it typicallyfollowstheobstruentreleaseandthestopissaidtobepost-aspirated;butoccasionallyandmuchmorerarely,languageshavebeenshownto havepre-aspiratedstopsinwhichtheaperiodicenergythatcharacterizesaspirationacousticallyprecedesthestopclosure.Insomelanguages,such astheNortherndialectofIcelandic,pre-aspiratedstopsarephonologicallycontrastivetopost-aspiratedstopsword-medially(Helgason,2002,186), but in Southern Icelandic and in other cases there is phonetic variation between pre- and post-aspirated stops: thus Silverman (2003) presents examples of a Mexican language Tarascan (based on Foster, 1969) and of Gaelic (based on Borgstrøm, 1940) in which pre-aspirated and post- aspirated stops have a complementary distribution between word-medial and word-initial position respectively (see also Helgason, 2002, for an extensive review of the occurrence and distribution of pre-aspiration in the world's languages). The phonetic characteristics of pre-aspiration, comparedtothoseofpost-aspiration,arereportedtobehighlyvariable,bothbetweenthefewlanguagesinwhichitoccursandevenwithinthesame languagegroup(Silverman,2003).Itsvariabilitymaycomeaboutbecausetheconditionsthathavegivenrisetopre-aspirationhistoricallyarealso varied:thuswhereaspre-aspirationinIcelandic/hphthk/hasdevelopedhistoricallyfromOldNorsegeminates/p:t:k:/(Hansson,2001;seealso Stevens,2010foranalogousdevelopmentsinsomeItalianvarieties),insomevarietiesofEnglish,inwhichpre-aspirationhasbeenshowntohavea sociolinguisticfunction,ithasbeenbroughtaboutpredominantlyasaresultofpre-obstruentvoweldevoicing(Docherty&Foulkes,1999).Thetypeof pre-aspiration inAndalusianSpanishthatis themainfocus ofthepresentstudy hasyet another originwhichis dueto thedebuccalization ofthe fricative in /s/+voiceless stop sequences, a phenomenon known as /s/-aspiration, either word-medially (casco [ˈkahko] ‘helmet’, caspa [ˈkahpa] ‘dandruff’,pasta[ˈpahta]‘paste,pasta’)oracrosswordboundaries(lastapas[lahˈtapah]‘thelids’,losperros[lohˈperoh]‘thedogs’,máscaro[mah ˈkaɾo]‘moreexpensive’)./s/-aspirationisnotlimitedto/s/+voicelessstopssequencesinAndalusianSpanishbutmayaffectanysyllablefinal/s/or/θ/ (e.g.isla[ˈihla]‘island’;capaz[kaˈpah]‘able’).Weakeningofsyllable-finalfricativesisacommonphenomenoninmanylanguages,andaspirationas nCorrespondingauthor.Nowat:URPPLanguageandSpace,UniversityofZurich,Plattenstr.54,CH8032Zürich,Switzerland.Tel.:+41446340222. E-mailaddress:[email protected](H.Ruch). 0095-4470/$-seefrontmatter&2014ElsevierLtd.Allrightsreserved. http://dx.doi.org/10.1016/j.wocn.2014.02.009 H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 13 aresultofvoicelessfricativelenitionis reportedfor anumberofotherlanguagesandRomancedialects(seeSolé, 2010,for anoverviewandan explanationbasedonanaerodynamicaccount). /s/-aspirationoccursinnumerousvarietiesofSpanish:ForEuropeanSpanishithasbeendocumentedmainlyinthesouthofSpain,butithasbeen shownrecentlythatthephenomenonisextendingtocentralvarietiesofSpain(CaleroFernández,1993;Momcilovic,2009).ForAmericanSpanish, /s/-weakeningoccursinCaribbeanandSouthAmericanvarietiessuchasChileanandArgentinianSpanishandinsomeCentralAmericandialects (see e.g. Canfield, 1981; Lipski, 1994, for an overview). Sociolinguistic and linguistic factors have also beenshown to influence the realization of coda-/s/.Theweakeningof/s/typicallyoccursinaword-medialpre-consonantalpositionbutitmayalsooccursyllable-initially(e.g.cincocentavos [ˈsiŋkohenˈtaβo]inDominicanSpanish,JiménezSabater,1975,77).Thereisconsiderableagreementthatweakeningcomesaboutpre-consonantally andword-mediallybeforeitspreadstoothercontextsbeforevowelsorpauses(Alba,1990;CaleroFernández,1993;Cedergren,1978;Momcilovic, 2009;SamperPadilla,2011;Terrell,1979,1981).Thereisalsoconsiderablephoneticvariationinsyllable-final/s/-weakeninginSpanish:itcanbe producedasafullalveolarfricative,or[h],ordeleted,dependingonphonologicalcontext(e.g.Alther,1935;Marrero,1990;Sánchez-Muñoz,2004), socialfactors(e.g.Cedergren,1978;Lafford,1986),anddialect. ForAndalusianSpanish,/s/-lenitionisverycommon(CarboneroCano,1982;MoyaCorral,1979;VillenaPonsoda,2008)andispresumed–as inferred from transcriber errors – to have existed since at least the 16th century (Lapesa, 1986; for a review see Romero, 1995b). In Western AndalusianSpanish(WAS),numerousvariantsof/sp,st,sk/arepossibleincludingwithpre-aspirationorbreathyvoicing,withclosurelengthening, andwithpost-aspiration(O'Neill,2010;Parrell,2012;Torreira,2007a,2007b,2012).Althoughdegreeofpost-aspirationwasfoundtobeaffectedby placeofarticulation(Torreira,2012),resultsofTorreira(2007a,2012)suggestpost-aspirationoccursinallthreestoptypesinWesternAndalusian Spanish. Torreira(2007a)acousticallyanalysed/sp,st,sk/inthreevarietiesofSpanish:PuertoRican,ArgentinianandWesternAndalusianSpanish.He observedlongervoiceonsettime(VOT)andshorterpre-aspirationfortheAndalusianspeakers,comparedtotheothertwovarieties.VOTwasalso longerin/sp,st,sk/thanintherespectivesingletonsor/lC/sequencesintheAndalusianvariety.Torreira(2007b)compared/sp,st,sk/sequencesof threeWesternAndalusianandthreeCastilianspeakers(avarietyinwhich/s/isnotaspirated)andfound,asinTorreira(2007a),longpost-aspiration fortheAndalusian,butnotfortheCastilianspeakers,andconsiderablevariationinhowpre-aspirationwasrealized.Heconcludedthatpost-aspiration inAndalusianSpanishwastheresultofcoarticulationandextensivearticulatoryoverlap,andproposedanon-goingsoundchangefrompre-topost- aspirationfortheAndalusianvariety. ForEasternAndalusianSpanish(EAS),thephonologicalcontrast/st,sk/vs./t,k/seemsnottodependonpost-aspirationdifferences,butinstead on either how the stopclosure and preceding vowelduration are related or on the presence of pre-aspiration or breathy voicein /sC/sequences (Gerfen,2002).AstudybyO'Neill(2010)alsofoundthat,whereasinthetwocitiesofWesternAndalusia(Seville,Cádiz)/s/+stopwasrealizedwith considerablepost-aspiration,forthoseinEasternAndalusia(Granada,Almería)alongerstopclosuredurationwasfound.Between60%(Granada) and70%(Seville)ofintervocalic/p,t,k/stopswerefoundbycontrasttohaveavoicedorpartlyvoicedclosure(O'Neill,2010,33).Similarfindingsare reported in Torreira & Ernestus (2011) for /p, t, k/ in spontaneous Madrid Spanish. A further phenomenon that has been documented for Eastern AndalusianSpanishistheuseofvowellowering(andfrontinginthecaseof/a/)inthecontextof/s/-lenition(e.g.tienes‘youhave’[ˈtjɛnɛ])whichmay beaccompaniedbyvowellengtheningand/orvowelharmony(Narbona,Cano,&Morillo,2003,170;seeMartínezMelgar,1994foraphoneticstudyof vowelharmonyinEasternAndalusianSpanish). Inarecentproductionstudy,Torreira(2012)testedifVOTincreasedwithstressanddecreasedatslowerspeechratesinWesternAndalusian Spanishjustasitdoesinlanguagesinwhichpost-aspiratedstopsarephonemic(Kessinger&Blumstein,1997).Neitherspeechratenorlexicalstress hadaconsistenteffectonpost-aspiration.Instead,Torreira(2012)observedanegativecorrelationbetweenthecombineddurationofpre-aspiration andclosuredurationandthedurationofpost-aspiration;hisconclusionfromtheseresultswasthatpost-aspirationwasmorelikelytobephoneticand theresultofextensivearticulatoryoverlapandnotaprincipalcuetothe/p,t,k/vs./sp,st,sk/distinction(Torreira,2012,61). SeveralstudieshavesuggestedasoundchangeinprogressintheWesternvarietyofAndalusianSpanishbywhichpre-aspirationmaybegiving waytopost-aspirationoraffrication(MoyaCorral,2007;O'Neill,2010;Ruch,2012;Torreira,2007a,b).However,thisquestionhasnotsofarbeen addressedsystematically.OneofthemainaimsinthisstudyistoassesswhetherinAndalusianSpanishthereisasoundchangeinprogressfrom pre-topost-aspiration.Parrell(2012)carriedoutasynchronicstudyofchangeinspeechratetotestwhethersuchashiftfrompre-topost-aspiration canbeexpressedintermsofamodelofgesturalre-phasingbetweentheopenglottisandclosureoftheoralstopduring/s/+voicelessstopclusters. According to this model which is inspired by articulatory phonology (Browman & Goldstein, 1992; Goldstein, Byrd, & Saltzman, 2006), the synchronisationbetweenglottalopeningandtheoralclosureinproducingpre-aspiratedstopsisanti-phase(becausetheoralclosureislateinrelation totheonsetoftheglottalopeningin/st/clusters),butitisin-phaseinproducingpost-aspiratedstops(inwhichtheonsetofglottalopeningandtheoral closureonsetaresynchronised).1Anti-phasetimingrelationshipsareingeneralmoreunstablethanthosethatarein-phasenotonlyinspeech(Marin &Pouplier,2010;Nam,Goldstein,&Saltzman,2010)butalsoinothertypesofskilledaction,suchasfingertapping(Kelso,1984;Turvey,1990)but alsoinothertypesofskilledaction,suchasfingertapping(Kelso,1984;Turvey,1990):consequently,timingchangesbywhichpre-aspirationchanges topost-aspirationsynchronicallyanddiachronicallyshouldbeprobable,ormoresothanintheotherdirectioninwhichtimingrelationshipswouldbe changedfromin-phasetoanti-phase.OneofthemainpredictionsofthismodelthatistestedinbothParrell(2012)andTorreira(2012)isthatthe durationofpre-andpost-aspirationshouldbeinverselycorrelatedasthecoordinationbetweentheoralclosureandglottalspreadingswitchesfrom anti-phasetoin-phase.Someevidenceinsupportofthisideaisprovidedinbothofthesestudies:InParrell's(2012)analysisofthe/st/clusterin pastándola(‘grazingit’)producedatdifferentratesby20subjects,durationalmeasurementsofpre-andpost-aspirationwerefoundtobenegatively correlated.Torreira(2012)found,asmentionedabove,thedurationofthecombinedpre-aspirationandstopclosuretobenegativelycorrelatedwith thedurationofpost-aspirationinproductionsofvarious/sp,st,sk/clustersproducedbythreenativespeakersofWesternAndalusianSpanish. Our second aim in the present study is to assess whether this synchronic evidence for a change from an anti-phase to an in-phase timing relationshipprovidedby Parrell(2012)canalsobeobserveddiachronicallyas inferredfromcomparisonsbetweenyoungerandolderspeakersof Andalusian Spanish on some of these measures. Another is to seek to resolve a controversy regarding closure duration in the course of this realignment between oral and glottal gestures. This issue is both controversial and unresolved because on the one hand, the closure duration is predicted to remain constant during changesto glottal-supraglottal coordination, as noted by Parrell (2012): “since the phase shift is independent of the duration of the gestures involved, there should be no changes in the oral constriction gesture itself”. On the other hand, the closure of 1However,asareviewernotes,itshouldbekeptinmindthatananti-phaserelationshipcorrespondstotheplannedvoicelessfricative+stopsequence. 14 H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 /s/+voicelessstopclustersinWesternAndalusianSpanishhasbeenshowntobelongerthanbothintervocalicstopclosuresofthesamevarietyand longerthantheclosuresthatoccurin/s/+voicelessstopclustersinothervarietiesofSpanish(Torreira,2007a).ForTorreira(2012,53),thisfindingofa lengthened closure provides a challenge for an explanation of pre- to post-aspiration change based entirely on phase relationships because it is “difficulttoimaginehowadelayed[h]glottalgestureinan[hp]clusterorareducedandoverlappedtongue-tipgestureinan[sp]clustermightleadto anonlinelengtheningoftheupcoming[p]closureproducedatthelips”.Torreira(2012,61)suggeststhatthislengtheningofthesupraglottalgesture mayhaveariseninorderto“compensatefortherealignmentof[h]withrespecttoupcomingconsonants”,andpointstovariousstudiesthatreport lengtheningoftheconsonantfollowinganaspiratedordeleted/s/forWesternAndalusianSpanish(e.g.Mondéjar,2001;Romero,1995a).However, longer closures in /st/-clusters than in intervocalic /t/ are also reported for the Eastern Andalusian variety by Gerfen (2002), and are described in dialectologicalstudiesnotonlyforAndalusian/sp,st,sk/(Alther,1935;Alvar,1955)butalsoforCubanSpanish/s/+voicelessstopsequences(Ruiz Hernández&MiyaresBermúdez,1984).ItseemsthusunlikelythatthelongstopclosureinWASistheresultofcompensationfortherealignment with[h]. While Parrell (2012) notes that gestural re-phasing “does not predict any changes in the closure duration for the stop”, he nevertheless also suggests,referringtophoneticuniversals(Maddieson,1997),thattheclosuredurationmightvaryinverselywithpost-aspirationontheassumption thatthedurationoftheglottal spreadinggestureis invariant. Parrell's(2012)resultsof thechangesdueto ratedemonstratejustsuch anegative association between closure and post-aspiration duration. His findings are in line with Torreira (2007b) who also observed a negative (but not significant)correlationbetweenVOTandclosuredurationfor/sk/and/st/,andattributedthisrelationshiptouniversalarticulatoryconstraints(suchas aerodynamicfactorsortheconstantdurationofaglottalcycle:seeCho&Ladefoged,1999).Thenegativecorrelationbetweenclosuredurationand VOTisalsonotinlinewithTorreira's(2012)assumptionofcompensatorylengtheningmentionedabove. Thus,thereisevidentlyconfusionaboutthedurationalchangestotheclosureaspre-aspirationwanesandpost-aspirationincreasesgivenfirstly thepredictionthatitshouldremainunchanged,secondlythedemonstrationbyTorreira(2007a,b)ofalongclosuredurationinWesternAndalusian Spanish,andthirdlyParrell's(2012)otherfindingthattheclosuredurationshortensaspost-aspirationdurationlengthens.Inthepresentstudy,oneof theaimswillbetoinferfromourapparent-timeinvestigationwhetherthereisanyevidencethatthehypotheticaldiachronicchangefrompre-topost- aspirationinAndalusianSpanish/st/clustersisassociatedwithachangeinthedurationoftheoralclosure. Theattemptssofartoexplaintheemergenceofpost-aspiratedstopsinAndalusianSpanishwereallbasedonarticulatoryaccounts.Butthereis alsocompellingevidencethatthesourceofthesynchronicinstabilityleadingtomanytypesofdiachronicchangedependsonauditoryfactors(Beddor, 2009; Ohala,1993).Accordingto perceptually-basedmodelsof soundchange(Ohala,1993),a post-aspirated releaseintheproductionof a pre- aspiratedstopmayleadtosomeconfusionforthelistenerintheserialorderoftheaspirationandclosure(seealsoGarrett&Blevins,2004,foran interpretationofcertainkindsofmetathesisinvolvingfricativesasduetoaperceptualconfusionintheserialorderatwhichthefricativeisproducedin relationtotheothersegments).Accordingtosuchamodel,theemergenceofpost-aspirationmightbeperceptuallyconfusedwithpre-aspirationand thereforeinterpretedasa(post-vocalic)underlying/s/. Insummary,therearethreemainaimsofthisstudy.ThefirstistocompareyoungerandolderspeakersofAndalusianSpanishonthedurationof pre-andpost-aspirationin/st/clustersinorderassesswhetherthereisasoundchangeinprogressinthisvarietyofSpanish(Section2).Thesecond aimistoassesstheplausibilityofgesturalre-phasingasthebasisforthediachronicchangefrompre-topost-aspiration,andtotestwhetherthe synchronic change found by Parrell (2012) under fast speech by which pre-aspiration changes into post-aspiration (assuming minimal change to closureduration)mightbethebasisforthediachronicchange(Section3).Thethirdaimistoassesswhetherthelengtheningofpost-aspirationmight haveaperceptualbasis(Section4). Thefirst two aims will additionally be tested in two separate but closelyrelated varieties: Western Andalusian Spanish whichis spoken in the provincesofCádiz,Huelva,Málaga,andSeville;andEasternAndalusianwhichisspokeninAlmería,Córdoba,Granada,andJaén(Mondéjar,2001). Thepurpose of doingso wastoprovideanindicationof thesizeof thediachronicchangeintheWesternAndalusianvariety, giventhat,withthe possibleexceptionsofMoyaCorral(1979)andO'Neill(2010),therearenoreportseitherthat/sp,st,sk/arepost-aspirated,northatachangefrom pre-topost-aspirationistakingplaceinEasternAndalusianSpanish. 2. Theproductionofpre-aspirationandpost-aspiration 2.1. Method Werecruited48subjectsfromAndalusia:24fromthecityofSeville,partofWesternAndalusia,and24fromthecityofGranada,partofEastern Andalusia.Foreachofthesetwovarieties,therewasayoungergroup(agerangebetween20and36years)andanoldergroup(agerange55–79 years).Thefourspeakergroups–olderandyoungerspeakersofWesternandEasternAndalusia–wereequallydistributedingender(sixwomenand sixmenineachgroup),and,asfaraspossible,ineducationallevel.Allsubjectsofourstudyhadlivedforatleastthelast20yearsintheirplaceof residence,SevilleorGranada,withtheexceptionofsixspeakerswholivedinthenearbysurroundingareaoftherespectivecity. Thematerialsincludedfourisolatedwordsthatcontained/st/sequencesandtwowordswith/t/stopsthatalloccurredintervocalicallyin/e_a/in trisyllabicwordswithlexicalstressonthesecondsyllable.Thefour/st/sequencesoccurredinthewordsestaba(/esˈtaba/,‘tobe’,1st/3rdperson singular,imperfectpast),estado(/esˈtado/,‘state’),estanco(/esˈtanko/,‘kiosk’),andpestaña(/pesˈtaɲa/,‘eyelash’);thetwointervocalic/t/stopsin amatchedvocaliccontextoccurredinthewords(henceforth/t/-words)etapa(/eˈtapa/,‘period’, ‘stage’)andretara(/reˈtaɾa/,‘tochallenge’, 1st/3rd person singular, past subjunctive). Each of these six words was repeated three times resulting in 6 (words)×3 (repetitions)×48 (speakers) 864 ¼ /V (s)tV /targetwordtokens.Inadditiontothe6targetwords,therewere130itemscontainingintervocalicsingletonstops,/sC/-sequencesorstop 1 2 clusterswhichwererecordedforarelatedstudy,and45itemsforanindependentproject,whichatthesametimeservedasfillersforourstudy.Each ofthese181wordswasrepeatedthreetimesinarandomisedorder,resultinginatotalof543wordsforeachspeaker. Subjectsreadeachitemdisplayedindividuallyonacomputermonitorataconstantratewhichwasfixedatjustover40itemsperminute.The recordingswerecarriedoutwiththeSpeechRecordersoftware(Draxler&Jänsch,2004),usingaportablecomputerandaCakewalkUA-25EXCV2 orM-AudioMobilePreUSBdevice,andaheadsetmicrophoneBeyerdynamicOpus54.16-3.Wheneverpossible,therecordingswerecarriedoutin thephoneticslaboratoryoftheUniversityofSevilleorintheradiostudioofGranadaUniversity;ifnot,welookedforaquietroominthesubjects' residenceorworkplace.Atthebeginningoftherecordings,speakerswereaskedtospeakintheirdialectinanaturalwayandasiftheyweretalking toafriend. H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 15 V1.i V1.f ci cf V2.i Hz) 6 k y ( nc 4 ue q Fre 2 50 ms Fig.1. Waveformandspectrogramoftheproductionofestaba.V1.iandV1.faretheacousticonsetandoffsetoftheword-initialvowel,V2.iistheacousticonsetof/a/.Theacousticonset andoffsetoftheclosureareciandcfrespectively.VTTisdefinedasci−V1.fandVOTasV2.i−cf. Fromthese864/V (s)tV /targetitems,73werediscarded,eitherbecausethewordhadbeenincorrectlyread(adifferentwordwasproducedfrom 1 2 theonethatwasdisplayedonthescreen)orbecauseofhesitationsorfalsestarts.Afurther33werediscardedbecauseV wascompletelyaperiodic 1 sothatnoframesofpitchdatacouldbecalculated.Fortheremaining758/V (s)tV /tokens,thespeechdataweredigitizedat44,100Hzandinitially 1 2 segmentedautomaticallywiththeMunichAutomaticSegmentationSystem(MAuS;Schiel,2004).TheMAuS-segmentationwasbasedonabroad phonetictranscriptionofthetargetwordstakingintoaccountthetypicalAndalusianpronunciationforms(ofe.g.[ehˈtaðo]forestado).Theboundaries werethen readjustedmanuallyto providesegmentboundariesfor theonset of V (V ),the offset of V (V ) and theonset (c) and offset (c) of 1 1.i 2 2.f i f the/t/-closure.V andV weremarkedatthebeginningandendofperiodicityrespectivelybyinspectingthewaveform;c wasmarked,asjudged 1.i 2.f i fromwaveformandspectrographicdisplays,wheretheenergydecreasedclearly,andc wasmarkedatthereleaseofthetargetstop(Fig.1).Theaim f ofthisprocedurewastofindasemi-automaticandgradientwaytodeterminevoiceterminationtime(VTT)asameasureforpre-aspirationduration, andvoiceterminationtime(VOT)asameasureforpost-aspirationduration. Oncetheseboundarieshadbeenmarked,twopitchtrackers,onebasedonESPS/wavesandtheotheronScheffers(1983),bothofwhicharepart oftheEmuspeechdatabasesystem(Harrington,2010),wereusedtocalculateautomaticallytheacousticoffsetofV (V )andtheacousticonsetof 1 1.f V (V ).TheacousticoffsetofV wasmarkedatthecalculatedcessationofperiodicityinframesofpitchdata(interval5ms)workingforwardsintime 2 2.i 1 betweenV andc.Asubsequentinspectionofwaveform/spectrographicdisplaysshowedthattheESPS/wavespitch-trackerwasmorereliablethan 1.i f thealgorithmbasedonScheffers(1983)formarkingthecessationofperiodicitywhenthevoicingoffsetprecededc,theacousticonsetofthestop i closure(i.e.incasesinwhichthestopclosurewasfullyvoiceless).Butifperiodicityextendedintotheclosure,thenthealgorithmbasedonScheffers (1983)morereliablydetectedwhenperiodicityceased(whereastheESPS/wavespitch-trackerwouldoftendeclaretheentireclosuretobevoiced, eventhoughtherewasnoevidenceofthisfromvisualdisplays).2Accordingly,V wasdefinedasthepointatwhichperiodicityceasedascalculated 1.f bytheESPS/wavespitchtrackerifthistimepointprecededc,butotherwisefromthepitchtrackerbasedonScheffers(1983).TheonsetofV was i 2 calculatedinanexactlyanalogouswaybutworkingbackwardsintimebetweenV andc. 2.f f Finally,the368tokensof/V stV /-wordsforwhichthe/s/-durationwasgreaterthanzero(i.e.forwhichc>V )wereauditorilyclassifiedbythefirst 1 2 i 1.f authorofthispaperaccordingtowhetherthemedialclusterwasjudgedtohavebeenproducedinoneofthreeways:eitheras[st](thefirstconsonant wasproducedwithanalveolarfricative);oras[ht](thefirstconsonantwaslenitedto[h]);oras[0t](thefirstconsonantwasdeleted).Ouraimwasto removeasmanyofthefirstcategoryproducedwithanalveolarfricativeaspossiblefromfurtherconsiderationusinganautomatic,acoustictechnique. Inordertodoso,k-meansclustering(Hartigan&Wong,1979)wasappliedtothemeanzero-crossingdensitycalculatedovertheintervalbetweenV 1.f andc:themotivationforthisparameterwasthatanalveolarfricativetendstohaveenergyconcentratedinahigherfrequencyrange(asaresultof i whichthezero-crossingdensityishigher)thanpre-aspiratedfricationi.e.thanfricationfromalenited/s/.Thedesiredoutputofthek-meansclustering algorithmwasthatonegroupwouldcontain/st/sequencesthathadbeenjudgedauditorilytohavebeenproducedwithanalveolarfricative;andthat theothergroupderivedfromk-meansclusteringwouldcontainpredominantlythetokensauditorilylabelledas[ht]or[0t].AsTable1shows,oneofthe twogroupsoutputfromk-meansclusteringcontained54/75tokensthatwehadauditorilyclassifiedas[s]andtwotokensthatwehadnot.These56 tokensinthisgroupwereremovedfromfurtheranalysis. Thefinalbreakdownoftheremaining702tokensthatwereanalysedinthispaperisshowninTable2.Weusehenceforththeshorthandnotation ht-wordstorefertotheremainingtokensproducedforthelargepartwithalenited/aspirated/s/inestaba,estado,estanco,pestaña(row1ofTable2); andt-wordstorefertoetapaandretara(row2ofTable2). ForalltokensinTable2,wedefinedvoiceterminationtime(VTT)asthedurationbetweenthecessationofperiodicityinV andtheonsetofthe 1 followingclosurei.e.VTT¼ci−V1.f(Fig.1).VTTwastypically(butnotexclusively)positiveinht-words(i.e.,periodicityceasedbeforetheonsetofthe 2TheESPS/Wavesalgorithmisbasedoncomputingthenormalizedcross-correlationfunction(NCCF)atlowandhighsamplesratesofthespeechsignalandthenusingdynamic programmingtoselectthesetofNCCFpeaksorunvoicedhypotheses(Talkin,1995).Thepurposeofusingdynamicprogrammingistoincorporatecontinuityconstraintsacrossadjacent framestoreducepitchdoublingandhalvingerrors.Thealgorithmisalsobasedonatwo-statevoiced/unvoicedclassification.Bycontrast,thealgorithminScheffers(1983)isbasedon findingauditorilyweightedharmonicsinFFTspectraandusesaso-calledharmonicsievetoidentifycomponentsthatbearaharmonicrelationtooneanother.Thereasonwhythetwopitch algorithmsperformdifferentlyhasnotyetbeensystematicallyinvestigated.ItispossibleforthepresentdatathattheESPS/Wavesalgorithmperformedlessaccuratelyindetectingthe offsetofperiodicityintheclosurebecause“inthesecases…highcorrelationsmaybeobservedthatcanleadtoanincorrectvoicingdecision”(Talkin,1995).Otherfactorsmaybedueto theusedynamicprogrammingintheESPS/Wavesalgorithmwhichmightover-estimatethedurationforwhichsuccessiveframesintheclosurearevoicedorbecausethebinaryvoicedvs. unvoiceddetectionislessaccurateindetectingthecessationofvoicingforverylowamplitude,quasi-periodicsignalsthattypicallyoccurduringtheclosure. 16 H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 Table1 Comparisonbetweenauditorylabels(columns;st,ht,0t)andautomaticclassification(rows;groups1,2,3).Thefirsttworowsincludethenumbersoftokens(inestaba,estado,estanco, pestaña)forwhichci>V1.f(i.e.voicingendedbeforeclosurebeginning)separatedintotwogroupsbyk-meansclustering.Thethirdrowshowsthenumbersoftokensinthesamewordsnot analysedbyk-meansclusteringforwhichVTT<5ms.Thenumbersarebrokendownbyvariety,age-group,andauditorylabel.Numbersinboldarethoseinwhichtheautomaticand auditoryanalysesdisagreed(inwhich/st/,producedaccordingtotheauditoryanalysiswithafullalveolarfricative,wasassignedtogroup2;orinwhich/st/,producedaccordingtothe auditoryanalysiswithpre-aspirationoranelidedsyllablefinal/s/,wasincludedingroup1).All56tokensingroup1werediscardedfromfurtheranalysis. Group EastAndalusian WestAndalusian Σ Younger Older Younger Older Label: st ht 0t st ht 0t st ht 0t st ht 0t 1(Higherzcr) 14 0 0 14 1 1 9 0 0 17 0 0 56 2(Lowerzcr) 2 50 24 7 64 17 1 24 37 11 51 24 312 3(VTT<5ms) 0 1 35 0 0 17 0 2 17 0 6 55 133 Σ 16 51 59 21 65 35 10 26 54 28 57 79 501 Table2 Thefirstrow(ht-words)showsthenumberofremainingtokensinestaba,estado,estanco,pestañathatwereincludedintheanalysis(thesearethesumsofrows 2and3inTable1)byagegroupandvariety.Thesecondrow(t-words)showsthedistributionoftokensinetapaandretarathatwerealsoanalysed. Category Variety EastAndalusian WestAndalusian Σ Agegroup: Younger Older Younger Older ht-Words 112 105 81 147 445 t-Words 69 61 65 62 257 Σ 181 166 146 209 702 closure)whereasint-words,VTTwasclosetozeroornegative(voicingfromV extendedintotheclosure).Wedefinedvoiceonsettime(VOT)also 1 foralltokensinTable2asthedurationbetweentheoffsetoftheclosureandtheonsetofperiodicityi.e.VOT V –c.Itturnedoutthatforalltokens, ¼ 2.i f VOTwaspositive.3 The results below are based on tests of whether VTT was smaller in younger speakers and VOT larger as a reflection of a sound change in progressbywhichpre-aspiration(asinferredfromVVT)iswaningandpost-aspiration(asinferredfromVOT)isincreasinginht-wordsinthesetwo varietiesofAndalusianSpanish.Thehypothesiswasalsotestedthatthedistinctionbetweenht-wordsandt-wordswouldbebasedforolderspeakers predominantlyonVTT(ontheassumptionthattheyproducedht-wordswithlongpre-aspiration)butforyoungerspeakersonVOT(assumingthatfor them,ht-wordswereproducedwithlongpost-aspiration). 2.2. Results WepresenttheresultseparatelyforVTT,closureduration,andVOTbelow.Themeanvaluesandstandarddeviationsontheseparametersforthe categoriesthatwereanalysedareshowninTable3. Fig.2showsbothalargerVTTforolderthanyoungerspeakersandthattheVTT-differencebetweenht-wordsandt-wordswaslargerforolder speakers.AmixedmodelinR(RDevelopmentCoreTeam,2012)wasappliedtoall702tokens(Table2)withVTTasthedependentvariable,with fixedfactorsage(oldvs.young),variety(WesternvsEasternAndalusianSpanish),andword-type(twolevels:ht-wordsvs.t-words);andwiththe speaker (48 levels) and word (6 levels) as random factors. Since the three-way interaction between the fixed factors was not significant, it was removedfromthemodel.Thisupdatedmodelwithoutthethree-wayinteractionthenshowedthatVTTwassignificantlyinfluencedbyage(χ2 20.7, 3¼ p<0.001)andbyword-type(χ2 22.8,p<0.001);thissecondresultcomesaboutbecause,asisevidentfromFig.2,VTTwassignificantlylargerin 3¼ ht-wordsthanint-words.TheinfluenceofvarietyonVTTwasnotsignificant.Theonlytwo-wayinteractionthatwassignificantwasbetweenageand word-type(χ2 6.3,p<0.05):thisresultcomesaboutbecause,asFig.2shows,thedifferenceinVTTbetweenht-wordsandt-wordswaslargerfor 1¼ olderthan for youngerspeakers.This age-dependent influenceonword-type was confirmed by multiplepairedcomparisonsusing alpha-adjusted Tukey-testswhichshowedthatVTTwassignificantlylargerforolderthanforyoungerspeakersinht-words(z 4.7,p<0.001),butnotint-words. ¼ AsFig.3shows,closuredurationwasoveralllargerinht-wordsthanint-words;anditwaslargerforolderthanforyoungerWASspeakers.Sincea mixedmodelwiththeclosuredurationasthedependentvariableandwiththesamefixedandrandomfactorsasfortheVTT-analysisaboveshowed thatthethree-wayinteractionwasnotsignificant,itwasdroppedfromthemodel.Theupdatedmodelwithoutthisthree-waytermshowedasignificant influenceonclosuredurationofage(χ2 13.1,p<0.01),variety(χ2 22.4,p<0.001),andword-type(χ2 38.1,p<0.001).Therewasalsoasignificant 3¼ 3¼ 3¼ interactionbetweenageandword-type(χ2 5.6,p<0.05)aswellasbetweenvarietyandword-type(χ2 14.0,p<0.001)butnotbetweenageand 1¼ 1¼ 3InordertoassessthereliabilityoftheautomaticmeasurementsforVTTandVOTinht-words,wecomparedthesevalueswiththemanuallymeasureddataforpre-andpost- aspirationthathadbeenobtainedforearlierversionsofthispaper.Inthemanualmeasurementstheonsetofpre-aspirationwasmarkedatthepointatwhichtheenergyofthesecondand thirdformantsdecreasedclearly,togetherwiththeappearanceofaperiodicnoiseinthewaveform.Thepost-aspirationphaseextendedfromthereleaseofthestoptotheonsetofthefirst periodic waveform. The mean difference between manually measured and automatically measured durations was 1.1ms (sd 13.9ms) for VTT (i.e. pre-aspiration), and 8.0ms ¼ (sd 5.5ms)forVOT(i.e.post-aspiration).ThedifferenceinVOTcameaboutbecauseinthemanualtechniquewetooktheonsetofobservableperiodicvoicingtobetheoffsetofpost- ¼ aspiration,whiletheautomaticprocedurewasbasedontheonsetofvoicing(therebypotentiallyincludingsomeaperiodic,low-amplitudeperiodswhichwerenotincludedmanually).Mixed modelswithageandvarietyasfixedfactorsandwordandspeakerastherandomfactorswererunonthetoken-wisedifferencebetweentheautomaticandthemanualmeasuresforVOT andVTT.Themodelsshowednoeffectofthefixedfactorsonthedependentvariable,whichsuggeststhatthemanualandtheautomaticmeasurementsperformednotdifferentlyneither withinnorbetweenthegroups. H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 17 Table3 Meanvalues(greyshadowedcolumns)andstandarddeviationinmsofvoiceterminationtime(VTT),closuredurationandvoiceonsettime(VOT)foreachspeakergroup. Category Variety EastAndalusian WestAndalusian Agegroup: Younger Older Younger Older VVT(ms) ht-Words 10.3 13.2 18.3 19.0 5.6 10.7 17.8 20.6 t-Words −10.0 22.2 −5.0 19.4 −11.6 17.5 −10.2 20.1 Closureduration(ms) ht-Words 113.4 26.2 115.2 23.8 88.7 20.3 110.0 22.3 t-Words 80.4 17.1 81.0 18.5 68.9 16.9 81.1 17.7 VOT(ms) ht-Words 35.8 14.0 20.4 8.1 55.2 17.5 26.8 10.3 t-Words 18.7 6.4 16.3 7.5 18.7 7.4 19.2 11.5 West Andalusia West Andalusia ht-words t-words 40 20 0 -20 s) m -40 n ( East Andalusia East Andalusia o ati ht-words t-words ur D 40 20 0 -20 -40 Older Younger Older Younger Fig.2. Distributionofvoiceterminationtime(VVT)inht-wordsandint-wordsinolderandyoungerEastAndalusianspeakersandinolderandyoungerWestAndalusianspeakers.Eachboxplot consistsofone(averaged)valueperspeakeronthismeasure.Therectanglesspantheinterquartilerangeandthedotinthecentreoftherectangleisthemedianofthedistribution. West Andalusia West Andalusia ht-words t-words 140 120 100 80 s) 60 m n ( East Andalusia East Andalusia o ati ht-words t-words ur D 140 120 100 80 60 Older Younger Older Younger Fig.3. DistributionoftheclosuredurationsinolderandyoungerEastandWestAndalusianspeakersinht-wordsandint-words.Eachboxplotconsistsofoneaveragedvalueperspeaker onthismeasure. 18 H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 West Andalusia West Andalusia ht-words t-words 80 60 40 20 s) m n ( East Andalusia East Andalusia atio ht-words t-words ur 80 D 60 40 20 Older Younger Older Younger Fig.4. Distributionofvoiceonsettime(VOT)inht-wordsandt-wordsinthefourspeakergroups.Eachboxplotconsistsofoneaveragedvalueperspeakeronthismeasure. variety.PairedcomparisonsusingTukey-testsshowedthattheclosuredurationwassignificantlylargerinht-wordsthanint-wordsforallfourgroups (olderEAS:z 10.0,p<0.001;youngerEAS:z 9.6,p<0.001;olderWAS:z 8.5,p<0.001;youngerWAS:z 6.0,p<0.001).Theyalsoshowedthat ¼ ¼ ¼ ¼ closuredurationinht-words(butnotint-words)wassmallerforyoungerWASspeakersthanfortheotherspeakergroups(olderWAS:z 3.2,p<0.05; ¼ youngerEAS:z 3.6,p<0.01;olderEAS:z 6.5,p<0.01).Thusthissecondsetofresultsshowsthat,whilealargerclosuredurationdistinguished ¼ ¼ ht-wordsfromt-wordsinallfourgroups,thedifferenceinclosuredurationbetweentheseword-typeswassmallerforyoungerWASspeakersthanit wasfortheothergroups.ForyoungerWASspeakerstherewasatrend(althoughnotsignificant)toproduceashorterclosuredurationthanolder speakersalsoint-words. FinallyFig.4suggeststhatVOTwaslargerinyoungerthaninolderspeakersandthattheVOT-differencebetweenht-wordsandt-wordswas largerforyoungerthanforolderspeakers.SincethemixedmodelwithVOTasthedependentvariableandwiththesamefixedandrandomfactorsas beforeshowedsignificantthree-wayandtwo-wayinteractionsbetweenallcombinationsofthefixedfactors,multiplepairedcomparisonsusingalpha- adjustedTukey-testswerecarriedout.TheseshowedthatVOTwassignificantlylargerinht-wordsthanint-wordsinallspeakergroupsexceptfor olderEASspeakers(youngerEAS:z 8.9,p<0.001;olderWAS:z 4.3,p<0.001;youngerWAS:z 19.3,p<0.001).Theresultsofthemixedmodels ¼ ¼ ¼ alsoshowedthatVOTinht-words(butnotint-words)wassignificantlylargerforyoungerthanforolderspeakers(youngervs.olderEAS:z 4.8, ¼ p<0.001;youngervs.olderWAS:z 9.1,p<0.001)andlargerforyoungerWASthanforyoungerEASspeakers(z 6.4,p<0.001).Thusoverall,and ¼ ¼ compatiblywiththetrendsinFig.4,theresultsshowthatVOTwaslargerinht-wordsthanitwasint-wordsforallspeaker-groupsexceptforolderEAS speakers; and that the VOT-difference between ht-words and t-words was (a) larger for younger than it was for older speakers and (b) larger for youngerWASspeakersthanfortheotherthreegroups. 2.3. Discussion The findings show that Andalusian speakers made use of VTT, closure duration, and VOT for distinguishing ht-words from t-words. t-words showedmostlynegativeVVT,i.e.voicingextendedintotheclosure,whileht-wordsshowedpredominantlypositiveVVT,i.e.voicingceasedpriorto theclosureonset. This finding of partiallyvoiced closuresin Spanish intervocalic stops is consistent with Torreira and Ernestus(2011) for Madrid spontaneous speech, and with O'Neill (2010) for Andalusian Spanish. The idea that the closure may be long in Andalusian /s/+voiceless stop sequencesisconsistentbothwiththeresultsfromrecentphoneticstudies(Gerfen,2002;Torreira,2007a,b)aswellaswithearlierstudiessuggesting ageminatestopproductioninAndalusianSpanish/sp,st,sk/(Alther,1935;Alvar,1955;seeTorreira,2012forfurtherexamplesanddiscussionof consonantlengtheninginWesternAndalusianSpanish);anditisalsoconsistentwithotherevidenceshowingthatpre-aspiratedstopsoftenseemto beaccompaniedbysomeformofstrengtheningofthefollowingoralgestureinmanylanguages(Silverman,2003). TherewasaneffectofageontheanalysedparameterssuchthatVTTwassmallerandVOTwaslargerforyoungerthanforolderspeakers.Tothe extentthatdiachroniceffectscanbeinferredfromapparent-timedata(Bailey,Wikle,Tillery,&Sand,1991;Weinreich,Labov,&Herzog,1968),then thisresultsuggeststhatthereisasoundchangeinprogressinAndalusianSpanishsuchthatpre-aspirationiswaningandpost-aspirationisbecoming moredominantasacuefordistinguishingintervocalics-lenited/aspirated/st/fromintervocalic,singleton/t/. Thereweretwomaininfluencesofvarietyonthispatternofresults:firstly,olderEASspeakersincontrasttotheotherthreespeakergroupsdidnot useVOTtodistinguishht-wordsfromt-words;secondly,youngerWASspeakershadthesmallestclosuredurationandlargestVOTofallfourgroups. Thefirstoftheseresults,combinedwiththefindingofalargerVOTinyoungerWASthaninyoungerEASspeakers,isconsistentwiththeideathat thesoundchangehasprogressedtoagreaterextentintheWesternthanintheEasternvariety.Thesecondresultmaybeindicativeofafurther innovationtothesoundchangeproducedbyyoungerWASspeakers:withtheprogressofthesoundchangebywhichpre-aspirationgiveswayto post-aspiration,thedurationoftheclosurealsobeginstoshorten.4 4Asareviewernotes,thisfindinghastobeinterpretedwithcautionbecauseyoungerWASspeakerscomparedtoolderWASspeakersalsoshowedslightlyshorterclosuredurations int-words. H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 19 If post-aspiration lengthening is accompanied by closure-shortening, then there should be evidence of a negative correlation between closure durationandVOTinht-words,especiallyfortheyoungerWASgroup(whichhadthelargestVOT).Wetestedwhethertherewasanyevidenceforthis in the next section. Another issue is whether there is compensatory lengthening, i.e. whether pre-aspiration shortening is correlated with closure lengthening.WealsotestedtheideafromParrell(2012)thatpost-aspirationarisesasanincreasingtendencytoproducetheglottalopeningandthe closureonsetinphase.Accordingtohismodel,post-aspirationarisesbecausetheoralclosureofaconstantdurationistimedtooccurearlierduring theintervalofglottalopening.Ifthisisthecase,thenthereshouldbeanegativecorrelationbetweenVTTandVOT,thatis,alongerVOTshouldbe accompaniedbyashorterVTT. 3. Therelativetimingofpre-aspiration,closure,andpost-aspiration 3.1. Method Theanalysiswascarriedoutforalltokensofht-words(Table2).AmixedmodelwasusedtoestimatetheextenttowhichVOTcouldbepredicted fromalinearcombinationofeitherclosuredurationorVTTwithvarietyandageasfixedfactorsandthespeakerandwordasrandomfactors.The speakerwasenteredintothemodelinsuchawaytoadjustbothfortheinterceptandfortheslope.Thepurposeofdoingsowastobeabletotestnot onlywhethertherewasanegativeassociationbetweenthenumericpredictor(VTTorclosureduration)andVOT,butalsotoexplorewhetherthesize ofthiseffectdifferedacrossthefourspeakergroups. 3.2. Results TheresultsinFig.5provideverylittleevidenceofaninverserelationshipbetweenVOTandVTTwithinanyofthespeaker-groups;noristhereany suggestionthattherewasalargerinverserelationshipbetweentheseparametersforyoungervs.olderspeakers.IndeedtheR2valuesforthesedata whichvarybetween0.0(youngerEAS)and0.05(olderEAS)suggestthattherewasnolinearrelationshipbetweentheseparameters.Amixedmodel inwhichthedependentvariableVOTwaspredictedfromVTTcombinedwithageandvarietyasfixedfactorsandthespeakerandwordasrandom factorsshowednoevidenceofanylinearrelationshipbetweenVTTandVOT(χ2 3.0,p 0.39). 3¼ ¼ WetheninvestigatedwhetheranyVTTorVOTchangeswereaccompaniedbychangesinclosureduration.Fig.6showsevidenceforanegative relationshipbetweenVTTandclosuredurationamongtheolderEASspeakers,butmuchlesssofortheotherthreespeakergroups.Giventhata mixedmodelwithVTTasthedependentvariable,closureduration,varietyandageasfixedfactorsandwordandspeakerasrandomfactorsshowed a significant interaction between the fixed factors, we re-ran the mixed models separately per speaker group. These results confirmed the trend observedinFig.6:therewasasignificantnegativecorrelationbetweenclosuredurationandVTTforolderEASspeakers(χ2 13.1,p<0.001),avery 1¼ weaknegativecorrelationforyoungerEASspeakers(χ2 4.5,p<0.05)andnosuchnegativecorrelationforWASspeakers. 1¼ TheresultsinFig.7showevidenceforanegativerelationshipbetweenVOTandclosureduration:R2variedbetween0.02(olderWAS)and0.32 (youngerWAS).TheresultsofamixedmodelinwhichVOTwaspredictedfromclosuredurationcombinedwiththesamefixedandrandomfactors as before showed a significant association between the two acoustic parameters (χ2 35.0, p<0.001). We explored the group differences in this 3¼ associationbyderivingfromthemixedmodel48slopes(intheplaneofclosureduration×VOT),oneforeachspeaker:thatis,thereisoneslopeper speakerderivedfromthelinearassociationbetweenclosuredurationandVOTcalculatedbythemixedmodel. TheresultsinFig.8show,compatiblywiththedatainFig.7,thattheslopeswerenegativeforallfourgroupsi.e.thattherewasatendencyfor largerVOTtobeassociatedwithsmallerclosuredurations.Fig.8alsoshowsthatthistrendwasingreaterevidencefortheyoungerthanfortheolder speakersandtoalesserextentforEAScomparedwithWASspeakers.TheresultsofanANOVAwithslopeasthedependentvariable(the48values, oneperspeaker,inFig.8)andwithindependentfactorsageandvarietyshowedasignificanteffectforbothfactors(age:F[1,44] 51.8,p<0.001; ¼ -50 0 50 West Andalusia West Andalusia Older Younger R2=0.03 R2=0.01 100 80 60 40 20 s) m East Andalusia East Andalusia T ( Older Younger O V 100 R2=0.05 R2=0.00 80 60 40 20 -50 0 50 VTT (ms) Fig.5. Scatterplotsofthedurationofpost-aspiration(VOT;y-axis)inht-wordsasafunctionofpre-aspirationduration(VVT;x-axis)forWest(row1)andEast(row2)Andalusianspeakers andforolder(left)andyounger(right)speakers. 20 H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 50 100 150 West Andalusia West Andalusia Older Younger R2=0.00 R2=0.01 50 0 -50 s) m East Andalusia East Andalusia T ( Older Younger T V R2=0.19 R2=0.01 50 0 -50 50 100 150 Closure duration (ms) Fig.6. Scatterplotsofthedurationofpre-aspiration(VTT;y-axis)inht-wordsasafunctionofclosuredurationforWest(row1)andEast(row2)Andalusianspeakersandforolder(left) andyounger(right)speakers. 50 100 150 West Andalusia West Andalusia Older Younger R2=0.02 R2=0.32 100 80 60 40 20 s) m East Andalusia East Andalusia T ( Older Younger O V 100 R2=0.15 R2=0.30 80 60 40 20 50 100 150 Closure duration (ms) Fig.7. Scatterplotsofthedurationofpost-aspiration(VOT;y-axis)inht-wordsasafunctionofclosureduration(x-axis)forWest(row1)andEast(row2)Andalusianspeakersandfor older(left)andyounger(right)speakers. East Andalusia West Andalusia 0.0 -0.1 e p o -0.2 Sl -0.3 -0.4 Older Younger Older Younger Fig.8. Speaker-specificslopesderivedafterfittingamixedmodeltothedatainFig.7forthefourspeakergroups.Eachboxplotconsistsofoneslopevalueperspeaker. H.Ruch,J.Harrington/JournalofPhonetics45(2014)12–25 21 variety:F[1,44] 18.0,p<0.001)andnosignificantinteractionbetweenthesefactors.Thusthissecondsetofresultsshowsthattherewasanegative ¼ associationbetweenVOTandclosureduration;andthatthisnegativeassociationwasmoremarkedinyoungerthaninolderspeakers,andmore markedforWesternthanforEasternAndalusianspeakers. 3.3. Discussion The results have shown that there was a greater negative correlation between closure duration and the duration of post-aspiration in the Western comparedwiththeEasternAndalusianvarietyandinyoungerthanolderspeakers:thisfindingisconsistentwithTorreira(2007b)andParrell(2012)who alsofoundsynchronicevidenceforanassociationbetweenclosuredurationshorteningandpost-aspirationlengthening.Ourresultssuggestthatthesound changeisaccompaniedbythedevelopmentofatradingrelationshipinwhichthegroupforwhichthesoundchangeismostadvanced,i.e.theyounger speakers,producedwordswithanunderlying/st/sequenceeitherwithalongclosureandshortpost-aspirationorwithashortclosureandalongpost- aspiration.OlderEasternAndalusianspeakersincontrastshowedanegativerelationshipbetweenVTTandclosureduration,i.e.compensatorylengthening ofthesubsequentstopclosure.Takentogether,theresultssuggestthatclosuredurationmayco-varyeitherwithpre-orwithpost-aspirationduration. Ontheotherhand,contrarytoParrell's(2012)modelbasedonsynchronicanalysesofratechangesinpastándola,thereisnoevidencefromthepresent apparent-time study of a correlation between the shortening of pre-aspiration and the lengthening of post-aspiration. Our findings suggest that post- aspirationlengtheningistoacertaindegreeindependentofpre-aspirationshortening,andthattheformertakesplacegraduallyratherthancategorically. Asanalternativetoanexplanationforpost-aspirationlengtheningandpre-aspirationshorteningbasedonphysiologicaltimingrelationships,we carriedoutaperceptionexperiment.Wetestedhowaslightlypost-aspiratedstopisperceivedinvarietiesofSpanishthataspirate/s/bycarryingout aforced-choiceperceptionexperimentwithlistenersofArgentinianSpanish.Theaimwastoinvestigatehowpost-aspirationisparsedbylistenersof aSpanishvarietywithpre-,butwithoutpost-aspiration.Thegeneralhypothesistobetestedwasthatinacontinuumwithaslightlypost-aspiratedstop (i.e.alongerVOT),listenersaremoreinclinedtoperceive/pasta/thaninacontinuumwithashortVOTbecausepost-aspirationisparsedperceptually withpre-aspirationandthereforeassociatedwiththesameunderlyingphonological/st/. 4. Theinfluenceofincreasingpost-aspirationontheperceptionofpre-aspiration Wesynthesizedtwocontinuabetween/pasta/and/pata/inwhichpre-aspirationwasshortenedinninesteps;onecontinuumwassynthesizedwith aslightlypost-aspiratedstop,theotheronewithatypicalVOTforSpanishof12ms.ThereasonforcarryingoutthetestwithlistenersofArgentinian Spanishisthatwordswithmedial/st/areproducedwith/s/-lenitioninmostregionsofArgentina(AlezaIzquierdo&EnguitaUtrilla,2002;i.e.pastais realizedas[pahta]);however,incontrasttoAndalusianSpanishthereisnoevidenceofasoundchangeinprogress(suchthatpre-aspirationisgiving waytopost-aspiration)whichcoulditselfbiastheperceptualresponsestowards/pasta/(seeTorreira,2007aforacomparisonbetweenWAS,Buenos AiresSpanish,andPuertoRicanSpanish). TheaimofthisexperimentwastosimulateasituationsimilartotheoneinAndalusianSpanishattheinitiationofthesoundchange.Theideawas totestwhetherthereisaperceptualbasistothesoundchangefrompre-aspirationtopost-aspiration. 4.1. Method 4.1.1. Syntheticstimuli AtokenofpastaproducedbyayoungerWASmalespeakerwasselectedfromthecorpusdescribedinSection2.1.Theselectedtokenhadalong VTT(57ms),aVOTofcomparableduration(45ms),andaclosuredurationof139ms.Sincethesecondvowelofthistokenwasweakanddevoiced itwasreplacedbyastronger,fullyvoiced/a/takenfromthesecondvowelofa/pata/tokenproducedbythesamespeaker(thecutforthereplacement wasmadeatV inFig.1).Subsequently,andinordertomakepre-andpost-aspirationdurationsmoresimilartothedurationsfoundintheproduction 2.i dataoftheEASspeakers(Section2.2)andthedurationsfoundbyTorreira(2007a)forArgentinianSpanish,VOTwasshortenedto29ms,closure durationto100ms,andVTTto34ms.ThiswasdonebyshorteningproportionallythewholeVOT(closure,orVTT)intervalusingPSOLAtechniques andtheAkustykplugin(Plichta,2012)inPraat(Boersma&Weenink,2012).TheresultingtokenwithaVTTof34ms,aclosuredurationof100ms, andaVOTof29msservedasthebase-wordfromwhichallotherstimuliwerederived. Inordertocreatethelong-continuum,wereducedthisbase-word'sVTTinnineequalstepsfrom34msto0msusingAkustyk,creating10stimuliwhich differedonlyinVTT(i.e.pre-aspirationduration).Theshort-continuumwasgeneratedbycopyingthe10stimuliofthelong-continuumandbysubstitutingthe VOTintervalineverytokenbyashortenedVOTof12msattheV cutpointinFig.1.Throughthisprocedure,weobtainedtwoVTT-continuawhichdiffered 2.i onlyintheVOTofthestoprelease(longvs.short).AsummaryofallstimuliusedintheperceptionexperimentisshowninTable4. Table4 VTT,VOTandclosuredurationvalues(ms)ofthestimuliusedinthepasta/pata-perceptionexperiment. Step Short-continuum Long-continuum VTT Closureduration VOT VTT Closureduration VOT 0 0 0 1 5 5 2 12 12 3 16 16 4 19 100ms 12ms 19 100ms 29ms 5 22 22 6 25 25 7 29 29 8 32 32 9 34 34
Description: