ebook img

The variability of early accent peaks in Standard German PDF

24 Pages·2010·0.77 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview The variability of early accent peaks in Standard German

The variability of early accent peaks in Standard German Tamara Rathcke and Jonathan Harrington Thispaperis concernedwiththerelationships between‘early’pitch accentsin German and with whether downstep in German is phonological or phonetic. Atthe coreof our analysis isan investigation into thedifferences betweentwo kinds of pitch accents in which the pitch peak precedes the accented vowel: theseareH+!H*andH+L*whichareclaimedtobephonologicallycontrastive in German. We made use of two experimental procedures: (1) a production experiment in which speakerswere askedto imitate synthetically manipulated sentences and (2) a semantic differential experiment in which listeners rated the perceived meaning of those sentences on eight semantic scales.Although boththeimitationandperceptionexperimentprovidedevidenceforadistinction between an early and a later (H*) peak accent, the results pointed neither to a three-way distinction, nor to a categorical distinction between H+!H* and H+L*. Finally, we present some results from an analysis of both English and Germancorporawhichsuggestthatthedifferencebetweenthesetwopeaksmay be phonetic and attributable to the number of syllables following the nuclear accent in the tail. Based on these results and from theoretical considerations, wearguethatH+!H*isaninappropriatepitchaccentcategoryintheinventory ofparadigmatic phonological intonational contrasts ofstandard German. 1. Introduction Theautosegmental-metrical(AM)theoryofintonationrestsonthepremisethat anf0-contour isderived fromasequenceofabstractphonological tone-targets thataresystematicallyalignedwithsegmentalunits(Pierrehumbert1980,Beck- manandPierrehumbert1986, Ladd1996).Thetonal inventory is reducedtoa basicdistinctionbetweenH(igh)andL(ow)tonesandastarrednotationisused in pitch accents to indicate the phonological association between a tone and a rhythmically strong syllable, which in turns has a major influence on their temporal alignment. In one sense, temporal alignment is clearly phonological because it is the result of the way that the tune is associated with the text, but 534 TamaraRathckeandJonathanHarrington inanotheralignmentisalsocontext-dependentandphoneticasvariousexperi- mentshaveshown(SilvermanandPierrehumbert1990).Thefrequencyscaling ofthe High andLow tones is assumedto be predictablefromphonetic factors and controlled by a system of rules. As Liberman and Pierrehumbert (1984) have argued,downstep is oneof themechanismsusedto justify theadherence toatwo-tonemodelinthefaceofwhatareevidentlymanydifferentf0-levelson the surface.Thus, in the famous stepping contour, the progressive lowering of f0-levelsonthestrong syllablesofsuccessiveaccentedwordscanbemodelled by downstep that is iteratively (and in theory infinitely) applied to a H* tone withinasinglephrase.Thisspecificationofdownstep isclearlyphonetic. However, it has often been difficult to place ‘downstep’ along the divide betweenphonological categoriesandphonetic implementation. Downstep was originallyinvokedintonelanguagestoexpresstheideathatthereisahightone thatisautomaticallylowered afteralowtoneinanH-L-Hsequenceandinthe AMmodel,downstepreferstotheloweringofsuccessivehightonetargetswithin a prosodic phrase. Unlike Liberman and Pierrehumbert (1984), Ladd (1996: 97) argues that there is also evidence that downstep is largely an ‘orthogonal phonologicalvariable’withthemeaningof‘finality’or‘completeness’(cf.Ladd 1996:90–92; thiswasalsoarguedinLadd1983). ThedownsteppedpitchaccentsinAmericanEnglishincludemonotonal!H* andbitonalL+!H*(BeckmanandAyers1994).Sincetheynecessarilyfollowa non-downstepped and often bitonal accent, these downstepped accents cannot occurinphrase-initialposition. Ourinterestliesinathirdtypeofdownstepped pitchaccentH+!H*whicharoseintheevolutionoftheToBIsystem,basedonthe AMmodel,fortranscribingstandardAmericanEnglishintonation. IntheToBI system,H+L*cametobereplacedbyH+!H*(Ladd1996:96–97;Gussenhoven 2004: 132) both for phonetic reasonsand becausethe levelof thestarred tone was not usually at the bottom of the speaker’s range, as would be expected if the second component of this bitonal accentwere actually L*. Given that one pitchaccentessentiallyreplacesanotherintheToBIsystem,thereisobviously nosenseinwhichtheycanbephonologically contrastiveinAmericanEnglish. However, the researchon intonational phonology by Grice(1995)in Southern British English and by Grice and Baumann (2002) and Grice, Baumann and Benzmüller (2005) in standard German does provide evidence that there may indeedbeacontrastbetweenH+L*andH+!H*anditisthisaspectthatwewish toexploreforGermanusingacousticandperceptualtechniques. SomeexamplesofthiscontrastarepresentedintheGermanGToBItraining materials(Benzmüller,GriceandBaumannURL2).InbothH+L*andH+!H*, thereisapitchpeakduetotheH+thatprecedestheaccentedvowel:thediffer- encebetweenthesepitchaccentsisthatinH+L*alowpitchtargetisreachedin ThevariabilityofearlyaccentpeaksinStandardGerman 535 theaccentedvowel,whereasinH+!H*thereisadownsteppedpeak thatoccurs duringtheaccentedvowel,sothatthetargetstepisfromhightomidinH+!H* butnotinH+L*.ThisisfurtherillustratedinFig.1whichshowsanH+L*pitch accentonschräg (‘diagonally’)intheleftpanelandanH+!H*pitchaccenton schön (‘pretty’) in the right panel.There is a markedstep down in pitch from theprecedingsyllabletoapitch-troughthatoccursduringthe[E]ofschräg;on the other hand, there is no evidence of a trough in the [2] vowel of schön and the pitch does not reachthe bottom ofthe speaker’srangeasit does forH+L* inschräg. If we allow the system to contrast H+L* and H+!H* pitch accents, then this meansthat the frequencyscaling of the(starred) tones that areassociated with the accented syllable is no longer phonetic as argued in e.g., Pierrehum- bert(1980),butisimplicitly phonologicalbecausethereisnowaparadigmatic contrastbetweenthree tonallevels: high-starin(L+)H*,low-starinH+L*and effectively a mid-star in H+!H*. Furthermore, H+!H* represents a departure from thephonotactics ofdownstep discussed earlier,since – in contrast to the other downstepped accents – there is no obligation for the H+!H* to follow a pitch accent (as a result of which, H+!H*, unlike !H* or L+!H*, could be the firstandindeedonlypitchaccentinaphrase). Figure1. Synchronized time-waveform and f0-trajectory with word labels (top) and GToBIlabelsofnuclearaccents(bottom)takenfromBenzmüller,Grice,and Baumann(URL2)oftwoGermanutterances:... immernochweiterschräg. (‘still further along diagonally’) and ... wie schön sie war. (‘how pretty it was’). Thereissomedifficultyinestablishingthemeaningdifferencesassociatedwith thesepitchaccents.Baumann(pers.comm.)suggests thatH+L*mightconvey areassuring,soothingtoneofvoiceincontrasttothegreatermatter-of-factness or neutrality of H+!H*. In a previous study (Rathcke and Harrington 2006), we assumed the general semantic differenceto be that between general/polite 536 TamaraRathckeandJonathanHarrington (H+!H*)andresolute/self-evident (H+L*)statements, but wewerenot ableto verifyanyofthesehypothesisedmeaning-tuneassociationsempirically.Further- more,ourprevious studyshowed thatitwasverydifficult tofindanysemantic contextdifferringbetweentheusageofH+!H*andH+L*. InothermodelsoftheintonationofstandardGerman(e.g.,Féry1993;Kohler 1997),thereisclearevidenceforacontrastbetweenH*andH+L*(orH*Lvs. HH*Lin Fe´ry’sanalysis), i.e. foracontrast oftiming in whichthe pitch peak is synchronised near the temporal midpoint of the vowel in H* as opposed to its muchearliersynchronisation with the immediatelyprecedingsyllable in H+L*,buttherehavebeennosuggestionstoourknowledgethatthereisathree- way phonological contrast between H* and what would be two types of early peaks,H+L*andH+!H*.Wethereforehavetoconsiderthepossibility thatthe distinction between H+L* and H+!H* may not be phonological but phonetic, i.e. predictably relatedto factors such asphrasal position, number ofsyllables andspeechrate. Our basis for assessing whether the distinction between two pitch accents is categoricalis firstly that native listeners canhearand reproducethedistinc- tion andsecondly thatit isassociatedwith semanticdifferences.Since,forthe reasons outlined above, it is difficult to be certain about the type of meaning differencethat is likely to be associated with the pitch accent distinctions be- ing investigated, the present paper draws upon a semantic differential test for thispurpose (Osgood,Suci,andTannenbaum1957;Uldall 1964;Dombrowski 2003; Ambrazaitis 2005), in which listeners were asked to make judgements on various semantic scales. We also carried out an imitation test of the kind usedinPierrehumbertandSteele(1989)inwhichsubjectsimitatetokensfrom acontinuumspanningthetonalcategoriesunderinvestigation. WeusedtheH*pitch accentasacontrolstimulus andcreatedacontinuum includingf0-trajectoriesofthethreepitchaccents:H*,H+!H*andH+L*.Ifthe distinction between H+!H* and H+L* is categorical, then listeners should be morelikely to perceive, and to produce, between- rather than within-category differences. On the other hand, subjects’ responses will be bimodal due to a contrastbetweenH*vs.acollapsedH+!H*/H+L*category,ifthelatterarejust phonetic variants ofasingle ‘earlypeak’category. Similarly,we would expect threedifferenttypesofresponsestothecontinuum inthesemanticdifferential testifallthreepitchaccentsarephonologically distinct. ThevariabilityofearlyaccentpeaksinStandardGerman 537 2. Experiment 2.1. Stimuli Wecreatedasyntheticcontinuuminthef0-scalingdomainspanningthreepitch accentcategories, H*,H+!H*and H+L*with the H*endof thecontinuum as controlstimuli.Wechosetomakeourcontinuumspanthreeratherthanjusttwo (H+!H*, H+L*) categories, because wewanted to makesure that therewas at leastoneunequivocalcontrastrepresentedinthestimuli (i.e.,H*vs.H+L*). The test sentence Sie mag Bananen (‘she likes bananas’) was read several times by a femalespeaker as a simple statement with a nuclear accent on the medialrhythmically strong syllable -na-inBananen.Theproduction inwhich she producedthe finalweaksyllable -nenwith aschwaandin which thepitch accentwasclearlyH+L*waschosenasthebasicitemforresynthesispurposes. Thiswasdoneprimarilybecauseourpilotstudy(RathckeandHarrington2006) had shown that tokens of Bananen with no final schwa vowel were often imi- tatedwith acreakyvoiceextending fromsomepartoftheaccentedsyllable of Bananentoitsoffset,therebymakingf0difficulttomeasure. Theconstructionofthecontinuumwasbasedontheidea,proposedbyGrice et al. (Griceand Baumann2002; Grice,BaumannandBenzmüller2005), that themaindifferencesbetweenH*,H+!H*,andH+L*areinthef0-heightofthe accentedvowel.Thecontinuumwasthereforedesignedtoextendoverthreecat- egoriesfromH*throughH+!H*toH+L*byincrementingf0-heightvariations in the accented vowel. More specifically, we created a stylised H+L* contour fromthesubject’sactualproductionofH+L*bymakingthef0-contourfallover theaccentedsyllable/na:/inastraightlinefromtheonsetoftheinitial[ν]at210 Hzto140Hzatapoint1/3ofthewayintothevowel.ForthisstylisedH+L*(st 16 inFigure2),thef0waskeptlevelat140Hztotheendof[a:].Twopoints(atthe onset ofthe nuclearsyllable andattheoffsetofthe following nasal)remained fixedintimeandfrequencyforallsynthesisedtokens.Thecontinuumwassyn- thesisedbyraisingthef0-levelinfifteen0.5semitonestepsoverthemiddle1/3 sectionofthevowelandthenbyconnectingthebeginningandendofthisraised sectionwithstraightlinesasshowninFigures2and3.Thisprocedureresulted in 16 stimuli. Weassumedthat the firststimuli (st -st ) corresponded to H* 01 06 giventhatthef0dropfromtheinitial/ν/ofBananen(timepointref1inFigure2) to themidpoint of /α:/ was less than 3 st, i.e. generally less than that required forantonalfalltobeperceived(‘tHart1981).Thestimulist -st represented 07 11 clearfallsofbetween3and5stfromtheprecedinghighpeakatref1inFigure2 toamidrangedf0-areaintheaccentedvowel(i.e.,H+!H*).Thestimulist -st 12 16 corresponded totheacousticformoftheH+L*pitch accentdefinedasasteep 538 TamaraRathckeandJonathanHarrington Figure2. Time-waveformwithwordandsegmentlabelsofthetestsentence(top)syn- chronized with f0-trajectories of the first (st ) and the last (st ) stimulus 01 16 ofthecontinuum.Foralltokensfromthecontinuum,thetimeandfrequency pointsmarkedbyref1andref2onthetimeaxiswerefixed.Theverticaldashed linesmarktheintervaloverthemiddle1/3sectionofthevowelthatwasraised insixteen0.5semitonesteps. Figure3. Schematic outlineof the componentsthat makeup thestimuluscontinuum andtheirpresumedassociationwiththepitchaccentsshownontheright. fall of between 5.5 and7.5 st to the verylow f0-valuesnearthe bottom ofthe speaker’srange(seealsoFigure3). 2.2. Semanticscales The semantic differential measures subjects’ reactions to stimuli in terms of ratingsonbipolarscalestypicallydefinedwithcontrastingadjectivesateachend. Thelistofsemanticscalesresultinginasemanticdifferentialisprimarilychosen based on hypotheses about the meaning of the categories under investigation. As has been shown in various studies (e.g., Heise 1970), ratings on bipolar scalestendtobecorrelated,andthethreebasicdimensionsofresponse(labelled ‘evaluation’, ‘potency’and ‘activity’) account for most of the co-variation in ThevariabilityofearlyaccentpeaksinStandardGerman 539 ratings. By using a few pure scales like ‘good-bad’for evaluation, ‘powerful- powerless’ for potency, and ‘fast-slow’ for activity, reliable measures can be obtained ofsubjects’overallattitudinal responsestodifferentstimuli. In contrast to the earlier semantic differential studies (e.g., Uldall 1964) whichwerecharacterisedbyanaposteriori factoranalysisforthepurposesof extractingtheunderlyingsemanticdimensions,acompletesemanticdifferential including the basic semantic dimensions was constructed a priori (see e.g., Dombrowski 2003 and Ambrazaitis 2005 for more details). For the present investigation, eight scaleswerecreated.Thechoiceofscaleswasbased onthe following considerations: 1. Scales 1–3:These correspond to meaning contrasts that have already been demonstratedfordistinctions analogoustoH*vs.H+L*(e.g.,Kohler1987; 2005).So,thesescaleswereselectedasthecontrolcondition. 2. Scale4:Thechoiceofthisscalewasinspiredbythefrequencycode(Ohala 1984):perhapsH+!H*isusedparalinguisticallytoavoidverylowf0-values (whichareassumedtobetypicalforH+L*pitchaccents).Asiswellknown (e.g.,Gussenhoven2002),lowtonestendtoconveyspeakers’dominanceor tosignalunalterablefactsinagreatnumberoflanguages. 3. Scale 5: As outlined in the introduction, this scale’s semantic distinction has alreadybeenhypothesised as oneoftheways that meaningdifferences betweenH+!H*andH+L*maybeconveyed. 4. Scales6–8:Thesewereaddedtocompletethesemanticdifferentialwiththe threebasisdimensions(evaluation,potency,activity,respectively).Arelevant factorforthechoiceofcontrasting adjectivesforscale6isthatGriceetal. (GriceandBaumann2002;Grice,BaumannandBenzmüller2005)consider ‘politeness’tobeoneofthemeaningsassociatedwiththeH+!H*vs.H+L* distinction.Thesemanticcontrastschosenforthescales7and8werethought tobegoodexponents of‘potency’and‘activity’indialogues. The scalesareshown inTable 1.Taking into account the meaningdistinctions associated with the frequency code (Ohala 1984) and what is known about howmeaningdifferencesareconveyedbythedistinctions betweenH*,H+!H* and H+L*(GriceandBaumann2002; Grice,Baumannand Benzmüller2005; Kohler 1987, 2005), the negative pole of each scale was assumed to be more appropriateforH+L*pitchaccents,thepositivepoleinscales1–3forH*,and thepositivepoleofscales4–8forH+!H*pitchaccents.Scales1–3areassociated primarilywithdifferencesoflinguistic meaning,whiletheotherscalesexpress predominantly differencesofparalinguistic meaning(cf.Ladd1996:33–36). Thescaleswereembeddedinacarriersentence‘Thespeakersoundslike...’. Theinstructions to thesubjects forjudging the scalarvalues madeitclearthat 540 TamaraRathckeandJonathanHarrington Table1. Eightpragmaticscaleschosenfortheperceptiontestwithsemanticdifferential. Germantermsgivetheexactmeaningofthescalepolesusedinthetest. Scale Negativepole(−) Positivepole(+) no. Germanterm Englishtranslation Germanterm Englishtranslation 1 wissend knowing erkennend realising 2 abschließend concluding diskussions-eröffnend openingadiscussion 3 akzeptierend accepting kontrastierend refusing 4 resolut outofthequestion kompromissbereit readytocompromise 5 beruhigend calming ermunternd encouraging 6 unhöflich impolite höflich polite 7 sicher certain unsicher uncertain 8 gelangweilt bored interessiert interested we were interested in potential meaning differences evoked by the speaker’s melody.Althoughsemanticjudgementscanbeinfluencedbythemorpho-syntax andlexisofanutterance,wedecidedtouseonlyonesentenceforthesemantic differentialtestinordertolimitthelengthofthetestto30min. 2.3. Procedures Atapecontaining128stimuliwascreatedfromeightrepetitions ofeachofthe 16 stimuli.Thus eachstimulus waspaired oncewith eachscale andpresented inrandomisedorder.Thetapewasplayedtothesubjectswhohadtomarktheir answers on a sheet of paper containing the scales. They also had to rank the utterance between ±3, i.e. between the extreme negative and positive ends of the scale.They were told to interpret these values as follows: 0 was ‘neutral’, ±1was‘slightly’(e.g.,assigning-1onscale8correspondsto‘slightlybored’), ±2was‘quite’,and±3‘extremely’. Thesubjects wereinstructed to waitforthe auditory prompt precededby a beepandtojudgespontaneouslyonthegivenscalehowtheythoughtthespeaker sounded.Theydidthisforeachstimulusseparatedbya4spause.Theperception task wasprecededby atrial session including eight stimuli presentations (one for each scale) to familiarise the subjects with scales and with the procedure. After the trial session the subjects had the opportunity to ask questions. The stimuli were presented fromaCD-player in asound treated roomat the IPDS Kiel. Fortheimitationtest,thesixteenstimulidetailedabovewereeachcopiedten times resulting in 160 items.These 160 items wererandomised and then each presented twice with a preceding beep.After the second presentation of each stimulus,therewasapauseduringwhichthesubjectwasinstructedtoimitateit, paying particularattention tocopying themelodyascloselyaspossible. Inthe ThevariabilityofearlyaccentpeaksinStandardGerman 541 event ofahesitation orspeecherror,theitemwasrepeated.Notimelimit was imposedforresponses.Thestimuli werepresentedtosubjectsviaheadphones. TheimitationtestwascarriedoutinasoundtreatedrecordingstudioattheIPDS Kiel. Twelve speakers of standard North German, five M and seven F between 20and22yearsofagewithoutknownspeechorhearingdisordersparticipated intheexperiment.AllsubjectswerebeginnerstudentsofphoneticsattheIPDS Kiel.Noneofthesubjectshadanyexperienceinprosodynorwastold thepur- poseoftheexperiment.Allsubjectswerepaidforparticipation.Theexperiment (containing bothsemanticandimitation test)tookaboutonehour. 2.4. Results 2.4.1. SemanticDifferential If there is a separate range of meanings conveyed by H+!H* as opposed to H* and H+L*, then we would expect to find a marked judgement profile for stimuli 7–11 on any of the scales 4–8: that is, these stimuli should get high positiveornegativescalevalueswhereasstimuli01–06aswellas12–16should belabelled‘neutral’orwithscalevaluesoppositetothosefromstimuli 7–11. The evaluation of the 16-point continuum is shown in Figure 4: the x-axis representsthestimulusnumber(01–16)andthey-axisisthescalevalue(seven points).Thedataforeachscalewereanalysedseparatelywith arepeatedmea- sures analysis of variance using SPSS (Brosius 2002). The f0-level in the ac- centedvowel(i.e.,stimulusnumber)wastheindependentvariableandthejudge- ment score the dependent variable.We used a repeated measuresANOVA be- cause the subjects judged more than one stimulus within one analysis. The Greenhouse-Geisser correction for non-homogenity of variances was applied (Leonhart 2004).These statistical results are summarised inTable 2.The sig- nificancelevel(alpha)wassetat1%sincethesamplewasonlyn=12. There weresignificant differencesin subject responses on five out of eight scales, i.e. on scales 1–3 aswell as on scales 5 and 8 (Figure 4).As expected, scales1–3showedchangesofjudgementthatarecompatiblewiththesemantic differencesforthedistinction between‘medial’and‘early’peaks(Kohler1987 and 2005), corresponding roughly to the H* vs. H+L* difference in an AM analysis.Scale5showssimilarresultsforscales1–3,i.e.themeaning‘soothing’ shows differencesprimarily between the beginning and end ofthe continuum, andhencebetweenH*andH+L*andnot–ashypothesised –betweenthepart of the continuum corresponding to the H+!H*vs. H+L* distinction (in which case,wewouldhaveexpectedamajordifferencebetweenthemiddleandtheend ofthecontinuum).Thesameistrueforscale8althoughtheprofileofjudgments 542 TamaraRathckeandJonathanHarrington e. u al v e al c s e h t s i s xi a - y r, e b m u n s u ul m sti s i s xi a - x s:s. ee calalu sv hte gal eisc onthesitive o 16p –st1with 0 std e muliciat stisso ea hs ti re foal )2sc 1h =ac ne (e ntsbov ea m m e judgftter Meanhele T 4. e r u g Fi

Description:
German and with whether downstep in German is phonological or phonetic. At the core . phonetic variants of a single 'early peak' category. Similarly
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.