Devine y Stephens - Dionysius of Halicarnassus, De Compositione Verborum XI, Reconstructing the...

8/4/2019 Devine y Stephens - Dionysius of Halicarnassus, De Compositione Verborum XI, Reconstructing the Phonetics of the

1/59

American Philological Association

Dionysius of Halicarnassus, De Compositione Verborum XI: Reconstructing the Phonetics ofthe Greek AccentAuthor(s): A. M. Devine and Laurence D. StephensSource: Transactions of the American Philological Association (1974-), Vol. 121 (1991), pp. 229-286Published by: The Johns Hopkins University PressStable URL: http://www.jstor.org/stable/284454 .

Accessed: 27/09/2011 17:31

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

American Philological Association and The Johns Hopkins University Press are collaborating with JSTOR to

digitize, preserve and extend access to Transactions of the American Philological Association (1974-).

http://www.jstor.org
http://www.jstor.org/action/showPublisher?publisherCode=jhuphttp://www.jstor.org/stable/284454?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/284454?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=jhup


2/59

Transactionsof the AmericanPhilologicalAssociation121 (1991) 229-286

DIONYSIUS OF HALICARNASSUS, DE COMPOSITIONEVERBORUM XI: RECONSTRUCTING THE PHONETICS OFTHE GREEK ACCENT.

A. M. DEVINE LAURENCE D. STEPHENSStanfordUniversity TheUniversityof NorthCarolina at ChapelHill

P.O'UOtKIIyp TXt;S rV KOatiTI V oX3ttiTi&vX6oyov int5XaITLi ZTnoa< 6tlakkXaTToxza Tfi; Ev 6sfi Ka' 6pydvotP6y i oxi ( oto.)Kai yap iv xratx lcaai ?Xo; ixoVutv a X??Ct; Kaic pVbpO.v Kail?vcapokivV Kai tpeov, oE ictai t xaTlTS j d&icolxT?pI7?ativ Toi0; p??(Eatv, ayErat 6? Tol; p'6p;Ooi;, a7dna?lTat 6?? Tax;p?a3apoX6d;, xtoOi &' iti a6v-Tcov oiKciov, i ? ttaXayticKaTaTo p.a&kov Ka'i fIxov.6taX?KxroV aEV OVV ?Log; ?vi g?cp?ivat 6tao(aritarti %T

??yOpEV(p6ta 7VT?VtE ; ?yytoTa, Kaiot eOtLEivEtx?tat n?pa TwVTpt&vt6vcv icat iqLtovioU kt. To 5 o) OT' d&viEvat OvX(opiouTolTxovt,?ov ?Cxixb p,apv. o0 plnavltaoa 4tiS lnKca0' vO6plOVXO6yoV TaTTOJ.t?VTnEti Ti; av fi; XyerTat TadEo);, a&XX't?v E7i TTfi;60?eac;, h 65'EnitfiT; pape?iac, h S' E7' dlapotv.T&v6? &apoT?opa; xTa; Ta6oEt; iXOoox&vai p?lv ica'a pLiavo0uXXapnvOuvEpOapp?vovi?ouoat To 64?i TO apu&,a; 6il t?eptlncog?va;KcaXo)pv?V ai ? EV ?iT?p E ia? Kai??p a Xwpi; KcXal?pOV ?p'

E?aUiroV inv oiK?iav p-uX&TLOv p61tv. Kai rais; aEv 6to-oXXapoStoO6?v 'r 6tt ?0oou Xopiov 13apnlor6l; Te Kaic6TotI'To;. Ta;i6i? oX;aouXXadpot;, xhial tCo'' av dotv, hi 6rv 6bi)v r6VOVEX0oora ia ?v 7okkalS; ai;S aXaat; pap?ialS;EV?ToV. h S'6pyavt TEKcal i6ti !oioa 6tiaoTx'Aaoi re Xpijrat tXeiootv,o'i) 8t TCTvrE lo6vov, d&X' &doOV 8ta aocov d&p4agevnKai o 6tla n?VT? ?cp?l Kai To St?a ' p'r'pv Kait ('r 6tlxptiv Kai x v) -Ovov cKaic i7iuo6vtov...(De Compositione Verborum76.13 Usher)

This paperhas two aims: first, to interpretand evaluate Dionysius' reportof the pitch intervalassociatedwith the Greek wordaccentand to flesh out theskeletal informationhe offers with experimentallybased typological data;andsecond, to supplement Dionysius' brief account of the tones of unaccentedsyllables with an analysis of the evidence of the Greekmusicalrecordsfor thephonetic implementation of both accented and unaccented syllables. Thiscombinationof ancientgrammatical vidence with typologicalandphilologicaldataleads to a reconstructionof the tonalpatternsof the Greeklexical word inits varioussententialcontexts.


3/59

A. M. Devine and LaurenceD. Stephens1. TheHigh-Lowinterval1.1 The textl

Dionysius' argument is as follows: the distinction between speech,includingoratoricaldelivery,andmusic is one of degree,not of kind:speechhasmelody and rhythm,just like music. The melody of speech uses one interval,namelythe fifth-the pitchobtrusionassociatedwith the accentdoes not exceedthreetones anda semitone-whereas music additionallyuses a varietyof otherintervals, such as the octave, the fourth, the tone, and thesemitone. On the face of it, one could hardlywish for a morevaluablepiece ofevidence aboutthephoneticsof the Greekaccent.Apparently,one can concludethat if we could take a male Greekinto a phonetics laboratoryand ask him tosay a sentenceassigninga fundamental requencyof 123.5 Hz (B on the secondoctave of thetempered cale) to the lowest unaccented yllables,his fundamentalfrequency would rise on accented syllables to 185 Hz (F sharpon the thirdoctave). We shall seek to establish to what extent such a conclusion can bedrawn from Dionysius' report, and if it can, whether it is a believableconclusion. Afterall, ancientgrammarians an give us whatthe Germanscall agoldcorn or they can give us what the Americanscall garbage.Nevertheless,Dionysius should be treated with the respect normally accorded a nativeinformant.We cannot be sure thatall characteristicsof the word accent in thespeechof educatedGreeksat thetime of Dionysiuswereidenticalto those of theclassical period, but any qualificationson this score will be mitigatedby thelikelihoodthatDionysius' reportrests on anearliergrammaticalradition.As a first step in the analysis, we need to be as clear as possible exactlywhat it is thatDionysius is tryingto tell us; the text has its fair shareof uncer-tainties of interpretation.Pitch excursions are measured(pcxErpo) in intervals(6iatoct'xtoac):ogtKcp6oxaTov... i6t&aLa' plETpirTov Plato, Republic531a7, e'TE te?pexTcai ttvt xTOv oXarro6v0ovatoTorlplaTovAristoxenus,Harmonics 1.24.There is no basis for the claim thatpETpEtxat SaotaiLoaCxtnDionysius cannotreferto an intervaldefininga maximumrangeof pitchexcur-sion: hi&vOpcontvrlpoviRpE?LptpTlxt (p)OIlKOi; X Tp.;Sp &nOCC&vtaOTi-gOCxtnonymusBellermanIII.94.An additionalrestriction s introducedby theterm ivi: music uses various intervals,but speech only one. This analysis ismotivatedby thedoctrineof Kivctat ocveXi;. Ancienttraditions quiteexplicitthat in speech, as opposed to music, pitch is continuously changing: conse-quentlythe rangeof mid tones (Tx6it&Etoo Xcopiov)nterveningbetween thetargetpeakandthe targetvalley of a pitchexcursionis not definablein termsof

1 S. Usher,Dionysius of Halicarnassus:The Critical Essays in Two Volumes(Cambridge,Mass. 1985); G. Aujac and M. Lebel, Denys d'Halicarnasse:OpusculesRhetoriques Paris1981);W. R. Roberts,Dionysiusof Halicarnassus:On LiteraryComposition London1910). W. Corssen,Aussprache, Vokalismusund Betonung der lateinischen Sprache ed. 2, II (Leipzig 1870) 797; J. P.Postgate,A ShortGuideto the Accentuation f AncientGreek(London1924);E.Sturtevant,The Pronunciationof Greekand Latined. 2 (Philadelphia 940);J.Carson,"GreekAccent and the Rational," HS 89 (1969) 24; W. SidneyAllen,VoxGraecaed. 3 (Cambridge987) 120.

230


4/59

ThePhoneticsof theGreek Accentsmaller intervals (nor, for most ancient treatments,are the targetsthemselves,since they too have insufficientduration).AristidesQuintilianushas a ternarydistinctionamong KiVrlOat6taS[ttlcatucit (music), KivTiot;aovexi; (speech),KtivloT;)oal (apparently typeof chantusedforreadingpoetry):

ovxv?X;eVg Oovv ?t (povi TX ; Te avo??t; KicaxaS; EittiaaErtXO?knl6ooX Sta TITlxo; xotoIl?Vrl. 5aOtalXpaXTiKtci TI Xa;a?V a?etg (pave?paS; exooa, ra ?6? Toov gu:pa~ [codd.p?Tpa] XekXri06xa,?oq 6&?h ? pp&poiv UoYK?1tpiVTl(De Musica1.4).

Pitch excursion in speech, Dionysius continues,does not have a rangeofmore than three tones and a semitone from valley to peak nor from peak tovalley. The implication of the terms ipopa and iXFeov s not clear. They maysimply imply that,given the largerintervalsoccurring n music, speech woulda priorialso be expected to allow such largerintervals. Or they may indicatethataccentualexcursions frompeak to valley or vice versa of more thanthreetones and a semitone do not occur, but those of less than three tones and asemitone do occur. Accentualexcursions would thenvaryfromone instance toanotherof the word accent up to the maximumof a fifth. This interpretationshardto reconcile with the whole point of Dionysius' argument,which is thatpitchexcursionsin speech arerestricted o a single interval,whereasmusic usesmany differentintervals. One possibility is that the O6' and papv refer to theaccentual peak and valley of the largestaccentualexcursion in the utterance.Smaller accentual excursions would be discounted or treated as part of theKivrloit o-vexnig rather than as measurable points for establishing thei6taoxilClaa of speech. It is also conceivable that,at this point, Dionysius ishaving difficulty reconciling different and partiallycontradictorygrammaticaltraditions.It has also been suggested thatthe interval of a fifth is measurednot fromthe lowest unaccented yllable to theaccentedsyllable,andvice versa,but froma mean pitch or "midline"from which pitch is supposed to rise a fifth to thetopline for accented syllables and fall a fifth to the bottomline for the lowestunaccentedsyllable. This midlinehypothesiscomplicatesthe analysisby intro-ducing a thirdmeasuring point into the KIVTrlOYtTveXri of speech; since as6taoTrTiais, according to one of Bryennius's definitions, 656b;not&a&a6p3apxTllTToS;i otTX-qra a&vdarcaXtv,he midline would representa p3apv orthe exixraot and an $6S for the av?E(a;.It has also been argued that on theusual interpretation,he phraseoVI' &avtCat..iXEov is tautologouswith whatprecedes, but, quiteapartfromthe impliedjudgmentaboutthe rarityof tautol-ogy in ancient grammaticalwritings, this is not so unless pitch inevitablyfallsby just the same amountthatit has previouslyrisen, which is not the case. Onthe other hand, it has been suggested that&viexratTOuXcopiouTOx-oouxnXovmeans "falls from this pitch level to a greaterextent" ratherthan"falls morethan this'pitch range,"a suggestion that is incompatiblewith the midline hy-pothesis. However, in geometrical terminologyXwpiovrefers to an area,par-ticularly a rectangle (EoTtv ov Texp&CyCvovcopovPlato, Meno 82b), and

231


5/59

A. M. Devine andLaurenceD. Stephenswith the phraseTx6t&cpgeaoo%copiov apTrxTos;TeKica 06xTxTo; few linesdown Dionysius seems to be denotingthepitchrangebetween the J3ap6andthe6o6 rather hana single intermediate one.This is aboutas far as the discussion can be taken on the basis of textualanalysis alone. Dionysius' evidence for the phoneticsof the Greek accent nowneeds to be interpreted n the overall frameworkof what is known about howtone is producedandperceived in speech;otherwise,there is a dangerthatanyinterpretationwill be limitedto an abstractandsuperficialexercise in deductionunconnectedwith thephysicalrealitiesit seeks to clarify.1.2. Theproductionof tone2

This section briefly surveys the main factors that can influence thefundamentalrequencyexcursionassociatedwitha word accent.Thesefactorsareimportantfor an appreciationof the acoustic events that underlie Dionysius'account of the Greekaccent andconsequently,when takenin conjunctionwiththe dataon perception n the next section, for the interpretation f thataccount.Intervals reconstructed for Greek are provisionally viewed as involvingmeasurementof fundamentalfrequencyfrom peak to valley ratherthansomeothertheoreticallyconceivable measuresuch as mean or midpointfundamentalfrequency. The terms High, Mid and Low, as they refer to tones, are usedthroughout his paperas a first level of phonologicalabstraction o indicatethegeneralphoneticlevel of a tone;no ultimatephonologicalanalysis in termsof aminimal specification of target tones is implied. The actual peak to valleyexcursions found in speech do not necessarily correspond to pure musicalintervals, that is they often cannot accurately be expressed by a simplenumericalratio; however, it may be that the frequencyof pure intervals washigher in Greek than it is in English, since one study found about twice asmanypureintervals n Norwegian,whichhas a tonalaccent,as in English.3High vowels have greater ntrinsic undamentalrequency han ow vowels.This difference has been establishedfor a varietyof languages:for instance, astudy on English founda differenceof 12-16 Hz;4in Danisha rangeof 10-35Hz was found.5Intrinsicfundamental requencydifferences are more markedwhere pitch is boosted by otherfactors,thatis at the beginningof an utteranceand in emphasizedsyllables: in such contexts differencesin English of about2

2 D. R. Ladd, "Declination:a Review and Some Hypotheses,"PhonologyYearbook 1 (1984) 53. M. Liberman and J. Pierrehumbert,"IntonationalInvariance under Changes in Pitch Range and Length,"Language Sound Structure,ed. M. Aronoff and R. T. Oehrle (Cambridge,Mass. 1984) 157. W. J. Poser, ThePhonetics and Phonology of Tone in Japanese (Ph.D. diss. MIT 1985). J.Pierrehumbert and M. Beckman, Japanese Tone Structure (Cambridge, Mass.1988).3 I. Lehiste, Suprasegmentals (Cambridge, Mass. 1970) 81. For somespeculation on musical intervals in speech see D. Hirst, "PhonologicalImplications of a Production Model of Intonation,"Phonologica 1980 ed. W. U.Dresler, O. E. Pfeiffer, and J. R. Rennison, 195 (Innsbruck 1981).4 G. E. Peterson and H. L. Barney, "Control Methods Used in the Study ofVowels," JASA 24 (1952) 175.5 N. R. Petersen, "Intrinsic Fundamental Frequency of Danish Vowels," JPh 6(1978) 177.

232


6/59

The Phoneticsof theGreek Accentto 3 semitones were found as compared with about 0.5 to 1.5 semitones inothercontexts.6The reason for the difference n intrinsic undamentalrequencyof vowels has been the subjectof intense investigation:one possibility is thatwhen the tongue is raised for the articulationof high vowels, it exerts extratension thatis, in one way or another, ransmitted o the vocal cords andconse-quentlyinduces higherfundamental requency7. t is easy to see that the differ-ences just cited are large enough to make a considerable mpacton Dionysius'reported interval of a fifth, particularlywhen cases in which the interval isincreased by intrinsic fundamentalfrequency boosting the peak and furtherlowering a contiguous valley are compared with cases showing the inverserelation that constricts the interval. However, it is likely that differences ofintrinsicpitch were somewhat less extensive in Greekthanthey are in English,because they seem to be inverselycorrelatedwith the functional oad of pitch ina languageand/orpositively correlatedwith the physiological effort associatedwith stress: the differences have been found to be smallerin the tone languageItsekirithan n thepitch-differentiatedtresslanguageSerbocroat, ndsmaller nSerbocroat than in the simple stress language English. The suggestion8thatapassage of the Poetics-xr oaotoeIca 6ia(pEpet....6o6-lTtl Kicdp3aprlxtt1456b-refers to intrinsicpitchrather han accentualpitch is very unlikely, cp.PriscianIII.519.10 Keil: "habetquippe litteraaltitudinem n pronuntiatione."The potentialrelevance of intrinsicfundamental requencyof vowels in Greekmusic is tested in section 2.3.1 below.

The fundamental requencyof a vowel is also liable to be influencedby thepresenceor absenceof voice in a precedingstopconsonant.9Exceptimmediatelyfollowing the stop release, fundamental requencytends to be falling from anabsolutehighervalue aftervoiceless stops and level or rising from an absolutelower value after voiced stops, although some reportsof rising fundamentalfrequencyafter voiced stops may be owing to a failure to factor out the intona-tionalcontext. In Swedish, peakfundamental requencywas found to be 15 Hzlower and25 msec later into the vowel aftervoiced than after voiceless conso-nants.10 The physiological basis for this is again poorly understood; oneinterpretationupposesa difference n tension of thevocal cordsassociated withconsonant voicing." The effect weakens as the vowel progresses;it also lasts6 S. A. Steele, Vowel Intrinsic FundamentalFrequencyin Prosodic Context(Ph.D. Diss. Texas 1985).7 The most recent discussion we have seen is E. Vilkman,O. Aaltonen,I.Raimo,P. Arajarviand H. Oksanen,"ArticulatoryHyoid-Laryngeal hangesvs.CricothyroidMuscleActivity in the Controlof IntrinsicFo of Vowels,"JPh 17(1989) 193.8 Carson above,note 1).9 For recentdiscussions see J. C. Kingston,The Phonetics and Phonology ofthe Timingof Oral and Glottal Events(Ph.D. diss. Berkeley 1984); R. Ohde,

"Fundamental requencyas an AcousticCorrelateof Stop ConsonantVoicing,"JASA 75 (1984) 224; K. Silverman,"FoCuesDependon Intonation," honetica43 (1986) 76.10 A. Lofqvist, "Intrinsic and Extrinsic Fo Variations in Swedish TonalAccents,"Phonetica 31 (1975) 228.11A. L6qvist,T. Baer,N. S. McGarr nd R. S. Story,"TheCricothyroidMusclein Voicing Control," ASA 85 (1989) 1314-21.

233


7/59

A. M. Devine andLaurenceD. Stephenslonger into the vowel in English than in tone languages such as Thai,12Yoruba,13 ndHausa,14where it is a comparatively ransient ffect. Ina numberof African tone languages, voiced obstruents have the phonological effect oflowering the tones of contiguous vowels, and in such cases they are called'depressorconsonants'.15nJapanese, hepitch peakin accentedsyllablesoccurslaterafter voiced stops than aftervoiceless stops.16We areprobably ustified inconcluding that the cumulative impact of these segmental factors on thefundamental requencyof anaccentualexcursion n Greek,for instancein .0iOo;versus pdaOoS,or sequentially in Enfpaotg, would certainly be fairlysubstantial: n English, a difference of 30 Hz was found between the peak fun-damentalfrequenciesof [pi]and [ba].17The potentialrelevance of the effect ofpreceding stop voicing on fundamental requencyin Greek music is tested insection 2.3.1 below. A small butconsistentdifferencein fundamentalrequencyis also occasioned by the voice of a following obstruent;18 lthough this dif-ference can be used as a perceptualcue, it is minorcomparedwith the effect ofthe voice of a precedingobstruentand is discounted n the following analysis.Anotherfactor thatpresumablyaffects themagnitudeof an accentualexcur-sion is the numberof pre- and postaccentualmorae or syllables in the word.Theoretically, hree situationscan be envisaged:either the targetpeakand/orthetargetvalley are fixed and the slope from one to the other variesaccordingto thenumber of unaccentedsyllables, or the slope is fixed and the targetsvary, orboth the targetsand the slope vary. In Japanese, the valley, and, to a smallerdegree, the following and preceding peaks are sensitive to the numberof post-accentual morae: the greaterthe numberof postaccentualmorae,the lower thevalley and the higherthe preceding peak;moreover,the peak of the followingminorphraseis somewhat lower after a deep valley than aftera shallow valley;none of these effects generallyexceeded 15Hz for a malespeaker,and theremaybe an absolutelimitbeyondwhich a valley cannotbe depressed o accommodateadditional postaccentual morae.19 n Danish, the degree of pitch rise onto theposttonic syllable is sensitive to the number of unstressedsyllables following

12 J. Gandour,"ConsonantTypes and Tone in Siamese," JPh 2 (1974) 337.13 J. M. Hombert, "Consonant Types, Vowel Quality, and Tone," Tone: ALinguistic Survey, ed. V.A. Fromkin (New York 1978) 77.14 L. Myers, Aspects of Hausa Tone. Working Papers in Phonetics, UCLA no.32 (1976).5 J. A. Goldsmith,Autosegmental and Metrical Phonology (Oxford 1990).16 H. Kawasaki, "Fundamental Frequency PerturbationCaused by Voiced andVoiceless Stops in Japanese,"JASA 73 (1983) S88.17 I. Lehiste and G. E. Peterson, "Some Basic Considerations in the Analysis ofIntonation," JASA 33 (1961) 419. Intrinsic and contextual fundamentalfrequencyhave been found to combine or neturalize each other in French too: A. Di Cristoand D. J. Hirst, "Modelling French Micromelody: Analysis and Synthesis,"Phonetica 43 (1986) 11.18 K. E. A. Silverman, "The Separation of Prosodies," Papers in LaboratoryPhonology I, ed J. Kingston and M. E. Beckman (Cambridge 1990) 139.19 H. Kubozono, The Organization of Japanese Prosody. Ph.D. diss., Universityof Edinburgh. 1987.

234


8/59

ThePhoneticsof the Greek Accentthe accent.The evidence for such variation n Greekis analyzedin section 2.3.3below.

When rapid speech is comparedwith slow speech, and when connectedspeech having tonal coarticulation is compared with citation form pro-nunciations,significantdifferencesare foundin the executionof the movementbetween target tones: pitch range can be compressed (that is, peaks can belowered and valleys raised), complex movements can be replacedby simplermovements,and the association of targettones with segmentscan be changed.Such phenomena have been investigated in Swedish,20 Japanese,21 andChinese.22Pitch sandhieffects in Greek areanalysed n section 2.3.4 below.Whenotherpitch modifyingfeaturesare factoredout, almost all languageshave an overall decrease n fundamentalrequency rom thebeginningto the endof a declarativeutterance.Languages n which some form of downtrend s absentareextremelyrare,althoughthe phenomenonseems to be more marked n sen-tences read underexperimentalconditions thanin ordinarydiscourse.23Down-trendcanbe analyzedas arisingfrom threeseparate ontributingactors:declina-tion, which is a tendency for fundamentalfrequency to fall graduallyas theutteranceprogresses;catathesis,as it is generallytermednowadays,which is aresettingof fundamental requency n alternatingHigh-Lowstructures; nd ter-minal fall, which is a significantadditional oweringof fundamental requencyat the end of a domain. Declination has been estimated to contribute o down-trendat a very approximate ateof 10 Hz persecond. Terminal all at the end ofan utterancecan producea steep fall in fundamental requencyto the point oflaryngealization, nd the loweringeffect can spreadbackwards nto nonfinalsyl-lables. Catathesis is the major contributorto downtrend: one of the conse-quences of catathesis is thatin a tone languagea High tone late in the utterancecan have lowerfundamentalrequency hana Low tone earlier n thesame utter-ance, and in an accent language,an accentedsyllable late in the utterancecanhave lower fundamentalfrequency than an unaccentedsyllable earlier in theutterance.The physiologicalbasis of downtrendapparentlynvolves both actionof laryngealmuscles anddecliningsubglottalairpressure.24

20 E. Ghrdingand J. Zhang, 'Temporaland Prosodic Pattern n Chinese andSwedish,"WPGLPLundUniversity 1 (1987) 11.21 Y. Nagano-Madsen,"Effect of TemporalTonal Context on FundamentalFrequencyContoursn Japanese,"WPGLPLundUniversity 1 (1987) 103.22 C.-L. Shih, "Tone and Intonation n Mandarin,"WorkingPapers of theCornellPhoneticsLaboratory (1988) 83-109.23 W. E. Cooper, S. J. Eady and P. R. Mueller, "Acoustical Aspects ofContrastiveStress in Question-Answer ontexts," ASA77 (1985) 2142.24 For recentdiscussionssee R. Collier, "FODeclination:The Control of itsSetting,Resetting,andSlope,"LaryngealFunction n Phonationand Respiration,ed. T. Baer, C. Sasaki and K. S. Harris(Boston 1987) 27; C. E. Gelfer, ASimultaneousPhysiological and Acoustic Study of FundamentalFrequencyDeclination. PhD.Dissertion.CUNY 1987.

235


9/59

A. M. Devine andLaurenceD. Stephens

200190Ie&C 70140150140

5 10Naalau Nubu / yas hb&r Lvwae / b*rva vurit reaaoa

'Kalam Nuhu/prevented Lawan/'r:Lirg to plac. of the n'u

Figure 1:Tonal Downtrendin a HausaDeclaration2

jrrl

(U~~~~s .

V*"

Figure 2:Pitch accent catathesis in a Japanesedeclaration.Sore wa nursiumni nomfmno26

25 Reproduced in modified form by permission of W. R. Leben. See S. Inkelasand W. R. Leben. "Where Phonology and Phonetics Intersect: The Case of HausaIntonation," Papers in Laboratory Phonology I, ed. J. Kingston and M. E.Beckman (Cambridge 1990) 17.26 Reproduced with permission from Poser (above, note 2).

236


10/59

The Phonetics of the Greek Accent

260240220200180160140120100

.-00* \O

0

W- od with short tolling ocCents-- - Wbrd%with shot ,isinq occents

*-- -0\

TimeFigure 3:Downtrend with pitch-differentiated stress accents in sentencesconsisting of three trisyllabic words in Serbocroat,Jagpe se druiebibama with short falling accents and Snasice raznose travicu.2

15-

05t.

5

0 100 [cs] 200 300Figure 4:Schema for downtrend with stress accents in Danish terminaldeclarative utterances. Large dots represent stressed syllables,small dots unstressed syllables.28

27 Reproduced with permission from I. Lehiste and P. Ivic, Word and SentenceProsody in Serbocroatian (Cambridge, Mass. 1986).28 Reproduced in modified form with permission of N. Thorsen. See, e.g., N.Thorsen, 'Two Issues in the Prosody of StandardDanish," Prosody: Models andMeasurements, Ed. A. Cutler and D. R. Ladd (Berlin 1983) 27.

237

v

0 "

I..


11/59

A. M. Devine and LaurenceD. StephensDowntrend s illustrated or a tone language(Hausa) n Figure 1, for a pitchaccent language (Japanese) n Figure2, for a pitchdifferentiated tresslanguage(Serbocroat)in Figure 3, and for a stress language (Danish) in Figure4. Theeffects of catathesisare dramaticwhether heHigh-Lowsequenceis tonal or ac-centual.In Danish, thepitchpeak is actuallydislocated one syllable to the rightof the stress onto the first unaccentedsyllable. Since catathesis does not onlylower frequencybut also compresses the frequencyrange of accentual excur-sions, the peak to valley and valley to peak measurementswill be smallertowardthe end of a catathesisdomainthan at the beginning.Some of this differ-ence is absorbedwith thechangefrom a linearfrequencyscale in Hertz to a log-arithmicpitch scale in semitones,but a significantdifference can remain,as is

overtly shown in Figure 4 in which the scale is in semitones.29The evidencefrom Greekvocal music for downtrend n Greekspeech is analyzed in section2.3.5 below.The phenomenonof downtrendcomplicatesourexploitationof Dionysius'evidence for the purposesof reconstructinghe accentualexcursionsin a Greeksentence. One possible assumption would be that Dionysius is reportingthecentralvalues likely to occur aroundthe midpointof a catathesisdomain,andthat we should increasehis reported ntervalof a fifth to reconstruct he compar-atively largefalls at thebeginningof the domainandcorrespondingly ecrease tafter the midpointin the domain.However, to the extent thatthe data cited sofar containrepresentativentervals, hey suggestthatan excursionof as much asa fifth would be morelikely to occur at or close to thebeginningof an utterancewith catathesis. For instance, the fundamentalfrequency contour illustratingdowntrend n Japanese(Figure2) has peak to valley ratios startingat 150%(afifth) and falling about 10%per pitch accent to approximately125%(a majorthird),factoringout the additional erminal all. In Dutch,which is a stress lan-guage, accentual prominencerarely involves excursions of more than half anoctave in normal,nonemphaticspeech.30Anotherpossibility is thatDionysius'report s based on words uttered n isolation.An early studyof Kyoto Japanesedisyllablesreporteda differenceof just undera fifth in paroxytonewords.31 t isalso interesting that in Thai words uttered in isolation the pitch excursionsinvolved in the falling and rising lexical tones were typically just under afifth.32 This view involves two difficulties. It is not clear to what extentDionysius has discountedthe additionalterminalpitch fall which characterizesthe end of words utteredin isolation, and which, if included, would, even by

29 It is interesting that in long sentences in Hausa, although frequencyexcursions are in a higher register in the first phrase than in the last, theintervalsbetweenHighandLow aregreatern the last phrase han n the first:R.0. Silverstein,"A Strategy or UtteranceProductionn Hausa,"Studiesin AfricanLinguistics Suppl.6 (1976) 233.30 J. 't Hart,"Differential ensitivity o PitchDistance,Particularlyn Speech,"JASA 69 (1981) 811.31Y. Homma,"An AcousticStudyof JapaneseVowels,"TheStudyof Sounds16(1973) 347.32A. S. Abramson,"StaticandDynamicCues in DistinctiveTones,"Languageand Speech 21 (1978) 391. J. S. Gandourand S. H. Petty, "PerceptionandProduction f Tonein Aphasia,"BrainandLanguage35 (1988)201.

238


12/59

ThePhoneticsof theGreekAccentitself, resultin a sizeable discrepancybetween iriotato andoiveor. StudiesofJapanesewords uttered n isolationhave found rather teep postaccentual alls.33Furthermore, s far as we can tell, utterances, ncludingwords in citationform,tend not to start with a fully low tone but with some form of mid tone. Forinstance, in Kinyarwanda a tone languagein which the tones have become sorestrictedthat they patternas in a pitch accent language), the tonal sequenceHigh High High when uttered n isolationis modifiedto Mid High Low, ratherthan Low High Low or Low High Mid.34It seems to be Dionysius' implicitassumptionthat rise and fall involved identicalintervals,althoughthe text doesnot explicitly say so. But the view thatDionysius' report s based on words intheir citation form does not fit too well with this assumption. The evidencefromGreekvocal music for a Mid High Low patternof words in Greekspeechis analyzedin section 2.3.3 below.Downtrend s conditionedby theoverallpitchrangechosen by the speaker.Whensubjects n anexperimentareinstructed o "speakup,"reflectinga greaterdegreeof overallemphasisorexcitement,not only do the absolutevalues of ac-centual peaks significantly increase, but the intervals between consecutivepeaks, expressed logarithmically,also increase.If this is also true for the peakto valley intervalsthatDionysius is concernedwith, thenit is a legitimateques-tion whether the pitch range he is referring to is the normal conversationalregisteror the higher registerthatwould characterizeordinary,unexaggerated(not moreAsiatico)rhetoricaldeclamation,given Dionysius' professionand thestatedobjectivesof his book.In additionto the overall variation n the pitch rangeattributable o down-trend,there is also the factor of local variation, akingthe shapeof a localizedincreasein pitch level andpitch intervalsto convey focus or emphasison a par-ticularword or phrase;this is typically done to separatenew informationcon-tributedby a sentenceto thediscourse fromold andalreadyestablished nforma-tion or to pinpointsemanticmaterial nvolved in a contrastor an erroneousas-sumption.The emphasizedmaterial s given greaterduration, ntensityandpitchobtrusion,and neighbouringpitch excursions tend to be constricted. Pitch ob-trusionas one implementationof emphasis is not confined to stress languagessuch as English, but is well established also in tone languages such as Zulu,Hausa, Thai and Chinese, in pitch differentiated stress languages such asSerbocroatandin the purepitchaccent languageJapanese.Japaneserelies moreon wordorderandparticlesthanEnglishdoes, butexpansionof thepitchrangeis also found. In Hausa, focused materialthat is not frontedregularlyreceivesspecial intonation.35 n Japanese,a boost of approximately20 Hz from 165 to185 Hz was found foremphasizedas comparedwithunemphasized djectives.36In Serbocroat, the interval between the peak frequency of the accented first

33 H. Fujisaki and M. Sugito, "Acoustic and Perceptual Analysis of Two-MoraWord Accent Types in the Osaka Dialect," Annual Bulletin, Research Institute ofLogopedics and Phoniatrics 10 (1976) 157.3 R. Furere and A. Rialland, "Tons et accents en Kinyarwanda," AfricanLinguistics: Essays in Memory of M. W. K. Semikenke, ed. D. L. Goyvaerts(Amsterdam 1985) 99.35 S. Inkelas, Prosodic Constituency in the Lexicon. Ph.D. diss. Stanford, 1989.36 Poser (above, note 2).

239


13/59

A. M. Devine and LaurenceD. Stephenssyllable and the terminalfrequencyof the postaccentualsyllable in the secondword in a sentence increased from 40 Hz without emphasis to 130 Hz withmaximum emphasis.37 The evidence from Greek vocal music for pitchobtrusion as an implementationof emphasis in Greek speech is analyzed insection 2.3.7.1 below.A few otherfactors deservebrief mention.In longerutterances,downtrendcan be reset atphraseboundaries,normallyrestartingat a lower level thanat thebeginningof the utterance.The evidencefrom Greekvocal music for the resetofdowntrend at phraseboundariesin Greek speech is analyzed in section 2.3.6below. Sequencesof shortsentencescan also be arrangedntohierarchical own-trend structures;when a single sentence was comparedwith a three-sentencestructure n Swedish, it was found thatthe peaks and valleys of the single sen-tence were lowerthan thoseof the initial sentence of thethree-sentence tructureand higherthanthose of the final sentence.38The internalsyntacticor prosodicstructureof a catathesischain can also affect the patternsof downtrend.39 heevidence from Greekvocal music for the effect of syntacticstructureon down-trend n Greekspeech is discussedin sections2.3.6 and2.3.7.1 below. Terminalpitch fall in declarativeutteranceshas alreadybeen alludedto. Downtrendsarelikely to be constrainedin at least some types of interrogativesentences, andthere may be a terminalrise: in Japanesethere is a steep rise in pitch on thefinal syllable of a yes-no questionof 110 Hz or more.40Questionscan also havea higheroverallpitchrange.41 none studyon English,for male subjectssayingmonosyllabicwords with rising (interrogative)ntonation,e.g. "me?","John?",fundamental requencyrose from 118 to 265 Hz, which is well over an octave;for the same monosyllables with falling (declarative)intonation,fundamentalfrequencyfell from 159 to 102 Hz, which is just over a fifth.42Ourevaluationof Dionysius' evidence and our use of it to reconstruct hephoneticsof the Greekaccentwill dependin parton the extentto which we canassume he is abstractingaway fromrather hanaveraging n theskewing effectsof any or all of the factors ust discussed.1.3. Theperceptionof tone43

Up to now, we have been concerned with the acoustic properties of theGreekaccent in its variouscontexts. We must now give some considerationtothe questionof how and to whatextentDionysius, as a Greeklistener,wouldbeaware of those varying acoustic propertiesand thereforein a position to takeaccountof themin his descriptionof the Greekaccent.37 Lehisteet al. (above,note 27).38G. Bruce,"TextualAspectsof Prosody n Swedish,"Phonetica39 (1982)274.39 H. Kubozono,"Syntacticand RhythmicEffects on Downstep n Japanese,"Phonology 6 (1989) 39.40 Poser(above,note 2).41 Inkelaset al. (above,note 25).42 R. B. MonsenandA. M. Engebretson, Studyof Variationsn the Male andFemaleGlottalWave,"JASA62 (1977) 981.43 't Hart(above,note 30). Hombert above,note 13). M. E. Beckman,Stressand Nonstress Accent (Dordrecht 1986) 107. A. C. M. Rietveld and C.Gussenhoven,"Onthe RelationbetweenPitch ExcursionSize and Prominence,"JPh 13 (1985) 299-308.

240


14/59

The Phoneticsof the GreekAccentPeople who can tell the pitchof a single puretone to withina semitonearecomparativelyrare. But many people who do not have absolute pitch can tell

whether two sequentially presentedtones have identicalfrequenciesor not: fortones in the region of 1000 Hz, just noticeabledifferencescan be below 1 Hz.With syntheticvowels havingconstant fundamental requencythroughout heirduration, t was found that istenerswere likewise able to perceivedifferencesofless than 1 Hz,44but, when frequency is changing, as it normally is in realspeech situations, ust noticeable differences were in the orderof 2 to 2.5 Hz.45In anotherstudy, adjacentsyllables were found to differperceptually n pitch ifthey were 5%or a little less than a semitone apart.46 he accuracywith whichpeople can differentiate wo consecutively presentedpitchintervals s obviouslyrelevant to any descriptionof word accents. Subjectsshowed notablevariationin theircapabilitiesandin theirstrategies n this task,with the most successfulperformersable to discriminate differences as low as 1.5 semitones. A peakheight difference of 1.5 semitones was also sufficient to create a perceptibledifference in the prominenceof a Dutch focus accent. These resultspertaintothe questionwhetherthe factorsmentioned n theprevioussection as potentiallyconditioningfundamental requencyhave effects largeenoughto impacton theintervalsreportedby Dionysius.Identical ntervals n semitones havebeen foundto be subjectively argerathigher frequencies than at lower frequencies.47The direction of fundamentalfrequencychangealso seems to affect theperceptionof the magnitudeof the in-terval.The changein pitch perceivedfor a risingfundamental requencyexcur-sion is greaterthanthat for a falling excursion of the same magnitude.48Thisfact mayprovidesomejustification orassumingthatfundamentalrequency ellmore fromthe accentualpeakthan it rose to it, despitetheprobable mplicationof Dionysius' text thatthe rise and the fall wereof equalmagnitude.Provided hemagnitudeof a frequency hangeand theduration f the transi-tion are not too large, a single unchanged pitch will be attributed o a vowelhaving continuouslyrising or falling fundamental requency.Accordingto onestudy, if the fundamentalfrequencyexcursion was less than about 2.5 semi-tones, a single pitch was assigned to the vowel corresponding o the frequencyapproximately wo-thirdsof the way into theexcursion.49Thisperceptualeffectpotentiallycomplicatesthesimpleprocedureadopted n theprecedingsectionofreconstructingntervals n termsof peakandvalley measurements.At a less physical and more linguistic level of perception,we need to askwhether Dionysius is including in his reported intervals differences in

44 J. L. Flanaganand M. G. Saslow, "Pitch Discrimination or SyntheticVowels,"JASA 30 (1958) 435.45 D. H. Klatt,"Discriminationf FundamentalrequencyContoursn SyntheticSpeech,"JASA53 (1973) 8.6 A. V. IsacenkoandH. J. Schadlich,A Modelof StandardGerman ntonation(TheHague1970).47 Rietveld and Gussenhoven (above, note 43).48 See Beckman (above, note 43) 116.49 M. Rossi, "Le seuil de glissando ou seuil de perception des variations tonalespour les sons de la parole," Phonetica 23 (1971) 1. Id. "La perception desglissandos descendants dans les contours prosodiques," Phonetica 35 (1978) 11.

241


15/59

A. M. Devine and LaurenceD. Stephensfundamental frequency that listeners can hear but partially or completelydiscount as linguistically insignificant.

Thereseems to be an intrinsicpitcheffect inverselycorrelatedwith the in-trinsic fundamental requencyeffect noted above for vowels andpartiallycom-pensating for it. When speakersof English were presentedwith low and highvowels (a versusi, u) havingthe same fundamental requencyand instructed ojudge which one had the higher pitch, they chose the low vowel two and a halftimes as often as the high vowel.50However, the intrinsic fundamental fre-quency of vowels can be exploited in whistle languages, as at Kuskoy inTurkey,where a is whistled at 1750 Hz, e at 2100 Hz andi at 2600 Hz.51In thedrumlanguage of San Cristoval in the Solomon Islands,a low-toned drumisused to representa and a high-toneddrumfor the mid and high vowels e, o, i,U.52 Theperceptualeffects of consonantvoicing on vowel pitchcan be expectedto be less significantbecauseof the comparatively ransientnatureof theresult-ing fundamental requencyperturbations.Nevertheless,they can serve as a cueto consonant voicing,53and comparabledifferences in syntheticvowel stimulican be heard as small differences in pitch. In the Zulupraisepoems, which arechanted n theupperpitchrangeof theperformer'svoice, consonantvoicing canaffect the note assignedto a syllable.54There is also evidence that listenerspartiallycorrect for downtrendeffectswhen estimating pitch. When subjects were asked to judge which of twostressedsyllables in a nonsense sentencewas higherin pitch, the configurationfor which theiranswers wererandomactuallyhad 10 Hz lower fundamentalre-quency on the second stressedsyllable;there was also a correctionfor the pitchrange factor.55Nevertheless, in SubsaharanAfrica,where most tone languageshavedowndrift,musicalmelodiesare oftenreported o be characterized y a pre-dominantly downwardslope. For instance, the highest note in a typical Igbo

50 Hombert above,note 13). C. K. ChuangandW. S. Wang,"PsychophysicalPitch Biases Related to Vowel Quality, Intensity Difference, and SequentialOrder,"JASA 64 (1978) 1004; G. Stoll, "Pitchof Vowels: Experimental ndTheoretical Investigations of its Dependence on Vowel Quality," SpeechCommunication3 (1984) 137-50.51 C. Leroy, "Etudede phonetiquecomparativede la langueturquesifflee etparlee,"Revuede phonetiqueappliquee14-15 (1970) 119.52 J. Snyders, "Le langage par tamboursa San Cristoval,British SolomonIslands,"Journalde la societedes oceanistes24 (1969) 133.53 M. Haggard,S. Amblerand M. Callow,"Pitchas a VoicingCue,"JASA47(1970) 613. The perceptual system is apparentlyelastic enough to "hear"intrinsicand contextual undamentalrequency ffects for segmentaldentificationand to discount them for prosodic purposes: R. Petersen, "PerceptualCompensationor SegmentallyConditionedFundamentalrequencyPerturbation,"Phonetica 43 (1986) 31.54 D. Rycroft, "Melodic Features in Zulu Eulogistic Recitation,"AfricanLanguage Studies 1 (1970) 60. Id., "Zuluand Xhose Praise-Poetry nd Song,"African Music 3.1 (1972) 79.5 J. Pierrehumbert,ThePerceptionof Fundamental requencyDeclination,"JASA 66 (1979) 363. C. Gussenhovenand A. C. M. Rietveld, "FundamentalFrequencyDeclinationn Dutch:TestingThreeHypotheses,"Ph 16 (1988)355.

242


16/59

ThePhoneticsof the GreekAccentmelody is the first High tone, afterwhich the tune works its way down to theend.56

In the above discussion, we have tried to providethe readerwith the infor-mation needed for making a decision about the meaning and the value ofDionysius' evidence. It is clearthatDionysius has abstractedaway from a vari-ety of effects that condition accentual excursions in the acoustic records ofliving languages. The reportedinterval of a fifth conforms quite well to theupperend of the rangeof intervalsassociatedwith ordinary,unemphaticaccen-tualprominence,but it does not conformat all with the overall frequencyrangeof an utterancemeasuredfrom the peak of the highest prominenceto the low-point of the terminalfall, a measurementwhich can easily exceed an octave.This suggests thatDionysius' reportpertains o the domain of the word,andnotto the domainof the utterance.The 6Otandthe 13apvieparatedby a fifth are inthe same word;they are not the highest o6t and the lowest papp6 n the utter-ance. For Dionysius to assign the single intervalof a fifth to accentualexcur-sions that in fact varied considerably is a naturalgeneralization that couldequally well be madeby nonexperimentalield investigatorsnowadays.The in-terval of a fifth seems to fit best with the first accentualfall in the utterance,orwith the fall in a citation formallowing thepossibilityof some factoringout ofterminal fall. On the other hand, the midline hypothesis, which we do notfavour,is unrealistic or the domain of the word andcanonly be adoptedfor thedomainof the utterance.2. Reconstructionof thepitch movementsof Greekspeechon the basis of theDelphic Hymns2.1. Thephoneticimplementation f accentedandunaccented yllables

Dionysius tells us that in Greek thereis only one high toned syllable peraccentedword:i xTovtDv TxvoveXoixca tax.The othersyllable in thecase ofa disyllabic word,and the othersyllables in the case of polysyllabic words,arelow toned: ev noXkad; rtC; aaokkt; pocapeiaot;he same traditionappearsinPseudo-SergiusIV.532.10K: gravesnumerosuntplures.He also tells us thatindisyllabic words there is no Mid tone: Txoi; evtoxV )a3ot; o68ev 1rO6&aeioou)Xcopiovp3aptpxrlxog;KcatidUTTl ;. Thatpolysyllables do have Midtones follows logically and is clearly suggested by the antithesis xtaC;p.iv5itofxkpot;.. .atc; 6e rtoX)oaXkapotg;;hether hese Mid tones are the noto-riousprosodia media is a separatequestion.57 apv; means unaccentedphono-logically and low toned (as opposed to mid toned) phonetically. Dionysius'reportprovidesclearconfirmation hat n Greeknone of the unaccented yllablesis high toned; Greek is not like Tokyo Japanese,in which all morae betweenthe first and the accentual fall are phonologically high toned. But Dionysius56 L. E. N. Ekwueme, "Linguistic Determinantsof Some Igbo MusicalProperties," ournalof AfricanStudies1 (1974) 355. Id., "AnalysisandAnalyticTechniquesn AfricanMusic,"AfricanMusic6 (1980)89.57 E. Pohlmann,"Der PeripatetikerAthenodor ber Wortakzent nd Melodie-bildungim Hellenismus,"WSt79 (1966) 201.

243


17/59

A. M. Devine andLaurenceD. Stephensdoes not tell us how the Mid and Low tones are distributedn unaccentedsylla-bles. Does fundamental requencyrise graduallyto the accentualpeak and fallgraduallyfrom the peak?or is the accentual fall concentratedn the postaccen-tual moraor syllable?Is the lowest postaccentual one normallythe final sylla-ble of the word?Is the lowest tone a fixed targetor does it vary accordingto thenumberof postaccentualmorae?Can the lowpointbe anticipatedor postponedin sandhi environments? Is it possible to draw a straightline connecting themeasurementpoints of the fundamental requencyof unaccentedsyllables? orwould the line have kinks in it, as would happenif any syllable carried morethan its proportional hare of the rise or fall? Is the lowest preaccentual one thesame as the lowest postaccentual one of theprecedingword or higheror lower?One might thinkthat the answers to such questionsare beyond our grasp,seeing that it is not possible to conduct in vivo experimentalstudies on a deadlanguage. However, the surviving musical records, particularlythe Delphichymns, providea valuablein vitro source of evidence thatcompensatesto someconsiderabledegreeforthe lack of experimentalmeasurements.We shall see thatcareful and detailedanalysisof the musicalrecords, ncludingstatisticalevalua-tion where possible, can provide answers to the questions just posed. Thismethodwill also be used to resolve problemsassociatedwith accentedsyllables,specifically to estimate the tonalheightof the graveaccent andto test for directevidence of downtrendn Greek.2.2. The correlationof song melodyand speechtoneIn manytone languages,notablyin AfricaandAsia, thecorrelationof songmelody with speech tone is so high that it can generallybe expressedin termsof rules. This characteristicof the vocal music of tone languages is obviouslyrelated to the important unctionalrole playedby pitch in those languages.Forexample, in the Nigeriantone languageIgbo, a song might contain the follow-ing line: Akwaadigh n'elu igwe. This line has a numberof differentmeanings,depending on the tones of both syllables of the first and last word. If a songcomposer arbitrarily omposed a melody for this line, he would be arbitrarilyassigning one of the following meanings to the line: 'There is no sorrow inheaven,' 'There are no tears on the bicycle,' 'There are no eggs in the sky,''There is no clothing among the crowd,' 'Thereis no egg on the iron,' etc. Inorderfor the composerto convey the correctmeaningof the text, he mustcom-pose a melody that reflects the speech tones of the two words in question.58Since nearly all Igbo music is based on song, linguistic tone (and intonation)have a formativeinfluenceon the melody of Igbo music. However, it is not thecase that the melody of song in tone languagesis faithfulto the linguistic tonesonly when ambiguitywouldotherwisearise.Thecorrelationof speechandsongmelody is an overallruleand so moreprobablyrelatedto thegeneralhigh func-tional yield of pitch differences in these languages.In fact, because of the in-built redundancyof language, Chinese without tones, as often spoken by for-eigners, is fairly intelligible provided that the pronunciationand syntax are

58 Ekwueme (above, note 56).

244


18/59

ThePhoneticsof theGreek Accentotherwise free of distortingerrors.59Greeklikewise has a significantnumberofminimalpairsof wordsdistinguishedonly by an accentualdifference,involvingboth lexical and inflectional morphemes;and, apartfrom its functionalyield,the general importanceof pitch in a pitch accent languageis greaterthanin astress language, since in stress languages the word accent can be realized byfeaturesotherthanpitch.For the avoidance of speech-songtonal mismatches na pitch accent language,a passage may be cited from Zeami (1363-1443), theforemost writerandtheoreticianof theJapanesesungnoh theatre,whichproba-bly indicatesthat suchmismatchesareundesirablen major exical categories.60In general, it is much more importantfor song melody to preserve thedirection of pitch movement than it is for it to maintain he truepitch of an ut-teranceor to replicatethepitch intervalsprecisely.Althoughwe shall cite somedataquantifyingaverageintervals n semitones,theobjectof most of ouranaly-sis will be to establish relational or directional properties, that is "same,""higher,""lower,"and"level,""rising,""falling."In the Hausa tonalsequenceHigh Low High, the Low tone can sometimesbe realizedby the same musicalnote as the preceding High; this is not a mismatch but reflects a feature ofphraseintonation n fluent speech.61This observationabout Hausais importantin thatit indicatesthat the tonal relationsreflectedin the music are those of thesurfacephoneticoutput,after the lexical tones have beenadjustedby factors ikesandhiand intonation. It follows that,conversely, we are in principlejustifiedto exploit the evidence of Greekmusic to reconstruct he surface tonalrelationsof Greekspeech.2.3. Theevidence of theDelphic hymnsThe Delphic hymns show the strictest rules of correspondencebetweenlinguistic pitch accent and musical setting of all the remains of Greek vocalmusic, andare, therefore,thebest potentialsource of dataagainstwhich to testfor the existence in Greek of the typological regularities governing theproductionof tone discussed in section 1.2 and from which to seek to answerthe questions raised in section 2.1. Previous understanding f those rules maybe summarizedas follows:62

1. The accented syllable (whether acute, circumflex or grave) is not setlower than any other syllable of the word.Specifically by accent type:2. The acute accent in trisyllables and longer words is preceded by arise and followed by a fall in pitch.59 Y. R. Chao, "Tone, Intonation, Singsong, Chanting, Recitative, TonalComposition, and Atonal Composition in Chinese," For Roman Jakobson, ed.M. Halle, H. G. Lunt, H. McLean and C. H. van Schooneveld (The Hague 1956)52.60 A. Omote and S. Kato, Zeami, Zenchiku, Nihon shiso taikei (Tokyo 1978)

24.61 W. R. Leben, "On the Correspondencebetween Linguistic Tone and MusicalMelody," African Linguistics, ed. D. L. Goyvaerts (Amsterdam1985) 335.62 S. Eitrem, L. Amundsen, and R. P. Winnington-Ingram, "Fragments ofUnknown Greek Tragic Texts with Musical Notation, II: The Music," by R. P.Winnington-Ingram, SO 31 (1955) 28-87. E. P6hlmann, Denkmiler alt-griechischer Musik (Niumberg1970).

245


19/59

A. M. Devine andLaurenceD. Stephens3. The circumflexaccentis most frequentlyet to a melism,the secondnote of which is generally ower,and not higher,than the first.4. A graveaccent is not set lowerthanany othersyllableof the word;moreover, t cannotbe higherthanthe next followingacute or circum-flex accent and often is not higherthanfollowingunaccentedyllables.The strictness with which these rules are observed establishes for theDelphic hymns a level of respect for pitch patternsand movements in Greekspeech that is likely to hold for the pitch patternsof otherlinguistic structuresand units in addition to those involved in the rules above. The questions ofsection 2.1 are empirical ones, designed on the basis of the typologicalinvestigation of sections 1.2 and 1.3 to lead to answers in terms of regularpatternsandprocessesof the tones of Greekspeech.Therefore f we can observea regularitygoverningthe musical settingsof the linguisticunitsand structuresinvolved in one of our questions, and if we can show that this regularity isstatistically significant,and not merelya coincidentaloccurrence,thenwe maysafely assumethat the regularityof the musicalsettingreflects the regularityofspeech. Conversely, any statisticallysignificantcorrelationbetween music andspeech revealedby more detailedanalysis of the hymns may be tested againstthe typological generalizationsof sections 1.2 and 1.3 to judge whether it islinguisticallynatural,andthereforeanotherprimafacie patter in Greekspeech,or a musical artificiality. Thus the triple controls of typological naturalness,musical respect for the surfacephoneticpatternsof pitch in Greekspeech, andstatistical testing will permit us to reconstruct the tone relations andmovements of words andutterancesof ancient Greekin far moredetailthanhashithertobeen possible.The Delphic hymnsarecomposedin thediatonicand chromaticgenera,63othat intervallic measurementsmay be given in tones and semitones. We haveignoredtheoretical,smalldifferencesbetweenintervals n the tetrachorduchastones of 9/8 and 8/7 for Archytas' diatonic genus. Examples are cited inPohlmann's transcriptionof the vocalic and instrumentalnotations into staffrepresentation. t shouldbe remembered hatmeasurementn semitonesconsti-tutes a logarithmic transformation with base of 12j2 in a temperedscale) oflinear measurementin Hertz; i.e. the interval, D, in semitones between twopitchesof f1andf2frequency s D=12 log2(fi/f2).2.3.1. Intrinsicand segmentallyconditioneddifferencesof pitchThere are two methods by which we can test for any tendency in theDelphic hymns for the musicalsettingto reflect the effects of vowel heightandstop voicing on pitch discussed in section 1.2 above. The first is to test for anyeffect on the intervalbetween thepitchpeakof a nonfinalaccentedsyllableandthepitchvalley of the final (unaccented) yllableof the same word.This methodis chosen for the test of the effect of vowel height. If therewere any tendency

for thesettingto reflect thedifferences n fundamentalrequencyassociatedwithdifferences n vowel height,then,on theaverage,the intervalbetween theaccen-tualpeakon a high vowel and the valley on a final low vowel shouldbe greater63 Pohlmann (above, note 62, hereafter P.) 67, 76. Jacques Chailley, La musiquegrecque antique (Paris 1979) 154-166.

246


20/59

ThePhoneticsof theGreekAccentthan the correspondingintervalwhen the final vowel was mid, and the lattershould be greater han when the final vowel was high, andsimilarlythroughallcombinationsof vowel height to a minimumwhen the accentualpeakwas on alow vowel and the final valley on a high vowel. An initial searchfor examplesto concretizethe foregoingdeductioncertainlydoes not appear o support t; fora high vowel - low vowel interval one may cite KuvOifavP.20.13 set to U C(f1 el), for a high vowel - mid vowel intervaloapcov P.20.25 set to : U (bbb), for a high vowel - high vowel intervalAcMXpictvP.19.6 set to ei1 ( ebldiebl), andfor a low vowel - high vowel intervaloup6vto; P.20.8 C< (e1 d1).For a statisticaltest it is sufficient to classify vowels anddiphthongssimply ashigh, mid,andlow, ignoringthedifferencebetweenhigh-midandlow-mid.Themeanintervalsas just definedarecalculatedforeach of theresultingnine com-binations of vowel height in the hymn by Limenius (P. 20). Only wordsbearingthe acute accent are considered,but the circumflexbehaves similarly.The results aregiven in Table 1.

LL LM LH ML MM MH HL HM HH0.93 1.19 1.00 1.39 1.64 1.56 1.57 1.86 2.67Table 1: Mean intervals between accentual peak and valley of theunaccented final syllable in words bearing the acute in P.20classified according to vowel height combinations.

Theresultsdo not conformto thedeductions.An analysisof variance,presentedin Table 2, confirms that the differencesamong the mean intervalsare merelyrandom.Source of Sums of df MeanSquares F = MSb/MSwVariation SquaresSS 203.24 76 -SSb 9.33 8 1.17SSw 193.92 68 2.85 0.4p >>> .05

Table 2: Analysis of variance of data underlying Table 1.Consequencesfor reconstruction,1: the test resultsof Tables 1 and 2indicate that the intrinsic effect of vowel height on pitch in ancientGreekwas not of sufficientperceptualsalience to be reflectedin musi-cal settings.The second methodis to calculatetheaveragepitchesof syllablesof differ-ent types. Since the Greek notationdoes not provide absolutepitches (as im-plied by the staff transcription),he pitch measurementmust be made in termsof intervals (above or below) an arbitrary eference. This methodwas used to

test for any tendencyfor the musical settingsto reflect contextualdifferenceinfundamental requencyassociatedwith stopvoicing. If therewere any such ten-dency, we should expect that, on the average, the pitch of a syllable with avoiced consonant onset such as the -65- of Ca86po); P.19.15, set to F (f'),would standlower in relationto thereferencepitchthan a syllable witha voice-less stop onset, such as the -09- of 0UyaTxpe;P.19.2, set to I (dl). Since

247


21/59

A. M. Devine andLaurenceD. Stephensaccented syllables have a higher mean pitch than unaccentedones, the twoclasses must be treatedseparately.Accentedsyllableswere limited to the acute,and theirmean calculated for P.19 and P.20, unaccentedsyllables were takenfrom P.19 only. The arbitrary eference was chosen as M (cl). The results arepresented n Table 3, whereTVrepresentsa syllablewith a voiceless (aspiratedor unaspirated) toponset and DV one with a voiced stoponset.64

Acute No AccentTV m = 1.86 m= 1.03s= 1.58 s= 1.47DV m = 2.31 m = 1.45s= 1.69 s= 1.80Table 3: Mean pitch relative to an arbitraryreference of accentedand unaccented syllables classified according to stop voicing ofthe onset.

For both accentedandunaccentedsyllables the mean intervalsabove the refer-ence pitch are actually greaterfor syllables with voiced onsets, although thedifferences are not statistically significant. To control for the possibility thatcollapsing over differencesof vowel heightmighthave dampenedthe observeddifferences,that s that theremightbe an interaction ffect betweenstopvoicingand vowel height of the following vowel in determining pitch, the data forunaccentedsyllables were crossclassified accordingto vowel height and stopvoicing. (Crossclassification of the data for accented syllables reduces thesample sizes so severely that the test would lack adequatepower to detectsignificantdifferences.)The greatestdifferencebetweenthemeanswas foundforsyllables with low vowels (L): the mean for TLas 1.25 and that for DL 1.88.Again the difference was not statistically significant: t = 0.92, df = 20, p >.05.65Consequencesfor reconstruction,2: the test results of Table 3 andfurther controlled testing indicate that the contextual effect of stopvoicing on pitch in ancient Greek was not of sufficient perceptualsalience to be reflected n musicalsettings.The effects of intrinsic fundamentalfrequencyare reduced in singing ascomparedto speech; Swedish choir singerscompensated airly successfully forintrinsicfundamental requencydifferencesby auditorymonitoringof theirper-formance.66However,the intrinsicfundamentalrequencyeffect thatremainsinsinging is still well above the thresholdof detection; in unaccompaniedsolosinging, listeners-including the singers themselves-seem to compensate

64 m indicates the sample mean, and s the sample standarddeviation.65 df means degrees of freedom and is 1 less than the sum of the number ofobservations in each sample. p indicates the level of significance of the observeddifference measured by t and means the probability of a t as large or larger thanthat observed occurring if in fact there were no difference between the samplemeans.66 S. Ternstrom,J. Sundberg and A. Collden, "ArticulatoryFo Perturbations andAuditory Feedback,"Journal of Speech and Hearing Research 31 (1988) 187.

248


22/59

ThePhonetics of theGreek Accentperceptually orintrinsic undamentalrequencydifferences.67t is not clearfroma purely theoreticalpoint of view if, or how, the setting of the Delphic hymnswould be sensitive to any such performance endencies. So the results of ourtests for musical reflexes of inherentand contextualsegmentaleffects on pitch,while negative, are nevertheless of considerable value. First of all, given theanatomical/physiological asisandempiricaluniversalityof the effects of vowelheight and stop voicing discussed in section 1.2, those effects must also havebeen operativein ancientGreek.Consequently henegativetest resultsindicatethat the strength of those effects was not so great that the musical settingswould be constrained to reflect them. Such a conclusion fits well with thefurther ypological evidence thatthese intrinsicand contextual effects on pitchare inversely proportional o the functionalvalue of pitch in a language;sincethis was relatively high in Greek,the intrinsicandcontextualeffects would havebeen proportionally ower. Secondly, it is methodologically important hatthemusical settings do not reflect intrinsicand contextualsegmentaleffects; theymay be ignoredin all the following tests, therebygreatlysimplifyingthedesignof those tests and also increasing their power to detect real effects on tonemovementby permittingus to operatewith largersamplesizes.2.3.2. Thegrave accentIt has long been recognized that the musical documents treat the graveaccent as a loweredHightone,not as an unaccented yllable.Statisticalanalysisdefinitely confirms this effect in the Delphic hymns. Since, as demonstratednsection 2.3.6, thepitchexcursionto thegrave is stronglyinfluencedby its loca-tion in catathesisdomains,and these domainscorrelatevery stronglywith syn-tactic units,not all graveaccentsmaylegitimatelybe treated ogetherforanaly-sis. Since catathesisdoes not applybefore the accentualfall of the initial wordof a clause or majorphrase,adequatecontrolmay be obtainedby excludingthegrave accents in these positions and treatingonly clause or phrase internalgraves. (The grave, of course, does not occur clause finally.) Now if the gravewere a loweredHigh tone, it would follow thatthereshould be a positive meanrise to it, but that this mean rise shouldbe significantlyless thanthe meanriseto the acute. In orderto achieve comparability n the measurementsof the re-spective meanrises andto control for the fact thatsyllablesfollowing theaccen-tual peak within a word are lower than the syllables precedingthatpeak (seesection 2.3.3), it is necessary to limit the analysis to polysyllables and to ex-clude the rise to accentsacrosswordboundaries.Thus the rise to monosyllablesbearingthe grave, such as 5e, and therise to word initial acutes are excluded.Consequentlyourprocedure s to calculate themeanrise to thegravein polysyl-lables such as Xoxc6;P.19.14, set to OOM (b bcl), eoscg; P.19.17, set to F F(fl fl), KXrxa&v.20.14, set to < [ (dl el), and EniP.20.14, set to CU (el fl ),and the meanrise to the acute in words such as 'ArO6' P.20.14, set to EC (e1a') and eKcivac;P.20.19, set to K< M_b d' cl) .Propernames andnonlexicalwords,as well as lexical words,bearingthegraveor acuteaccentsare included.The grave cases are taken from P.19 and P.20; for the acute P.20 provides an

67 N. R. Petersen, "The role of Intrinsic FundamentalFrequency in thePerception f Singing,"WPGLPLundUniversity 4 (1988) 99.

249


23/59

A. M. Devine andLaurenceD. Stephensample data set. Of course, to maintaincomparability,only clause and majorphraseinternalacutes are admitted.The results arepresentedin Table4, alongwith Student'st-testof the differencesbetween the mean rise to the grave,mg,and the meanrise to the acute, ma.

To Grave To Acutemg = 0.31 ma= 0.92g = 0.53 Sa = 1.43t = 2.01, df= 68 mg


24/59

The Phoneticsof the GreekAccentP.20.39

6dip Xav cv - ErrThe statusof the graveas a loweredHightone may also be establishedbycomparing ts pitch to that of the nongraveaccentof an immediately followingword.Inordernot to introduce heinteractingactorsdiscussed in sections 2.3.6and2.3.7 below, propernames andnonlexicalwords are excluded. Accents onthe first lexical word of a clause or participialphraseare also excluded. Themean rise from the grave to the following nongraveaccent, gA, is given inTable 5 along with the t-test of thehypothesis hatthis rise is greater hanzero.-S#(...)SS(...)#gA = 1.22s = 0.57

gA > 0 : t = 6.12, df = 8p


25/59

A. M. Devine andLaurenceD. Stephensvowel grave syllables such as -av in coSav above. Accordingly, the peak ofthe grave has been taken to be located on the final morain all cases. There areno data for pyrrhic-shapedexical wordsin clause andphrasemedialpositions.Table 6 presentsthe mean total rise from the beginningof grave words for thethree classes distinguished. Below the table are given the results of theJonckheere-Terpstraest68of the hypothesisthat the rise to graveincreases fromproclitics to lexicals with two pregrave morae, and to lexicals with three ormore pregravemorae:MG(p,x3), with at least onestrictinequality.

Graveclass:pregravemorae Mean total riseMG(p,x3) 0.63MG(p,x3):J = 116, z = 1.61p =.054

Table 6: The rise of the grave ncreaseswiththe number fpregravemoraeand statusas lexical word.The rise to the grave triplesas one moves from proclitics (0.07) to disyl-labic lexicals (0.21) to trisyllabic and longer lexical words (0.63 tones). Thevalue of the z-approximationo the distributionof the test statistic J shows thatthere is just slightly more than a one in 20 chance thata J this large or largerwould ariseat random f the meansactuallydid not differ.Consequences or reconstruction,5: the test resultsof Table 6 indicatethat the magnitudeof the excursionto the grave accent is afunction oftwovariables:1) the rise to thegraveis greaterin lexicalwordsthaninproclitics, and 2) the rise is greater the larger the numberof morae itcovers. This reconstruction may be represented schematically asfollows:

S S s s s (S) SProclitics ShorterLexicals LongerLexicalsAlthough in the medial positions studiedproclitics have almost no rise tothe grave, in sentenceinitialpositionthereis a rise of an octave on the proclitic'AXXaP.20.26, as large a rise as any to the acuteor circumflexin initialposi-tion. This fact suggests thatdisyllabic proclitics are not phonologically unac-centedlike enclitics, butratherhave an underlyingaccent which is substantiallyreduced in most circumstances,but which can emerge as an excursion under

68 See W. W. Daniel, Applied Nonparametric Statistics (Boston 1990) 234-38.

252


26/59

The Phoneticsof the GreekAccentspecial circumstances. The correlation of the magnitudeof the excursion toacute and circumflex accents with lexical status and numberof syllables is dis-cussed in section 2.3.3.2.3.3. Pre- and postaccentual syllables and the Mid-High-Lowcontour ofmediallyaccentedpolysyllablesThere is ample evidence from the Delphic hymns to confirm Dionysius' im-plicit claim that medially accented Greek polysyllables had Mid tones and toshow that in fact they followed a Mid-High-Low pitch contour,not a Low-High-Low contour. There are a number of settings of polysyllables whichclearlyexemplify this Mid-High-Lowpattern,e.g.

P.19.15

P.19. a - 8-poP.19.18

E-y? --ov

P.20.23

i - pa -The hypothesisof such a phonetic Mid-High-Lowcontourcertainlywouldexplain Winnington-Ingram'sobservation69 hat "thetendencyto fall from theaccented syllable is distinctlystronger n these hymnsthanthe tendencyto riseto it." More generally, it is true that preaccentualsyllables had a higher pitchrelative to the accentualpeak of the word thanpostaccentualsyllables, so thatpitch fall to the first syllable afteran initial accent was greater,on the average,than pitch rise to the circumflex on final syllables, as comparison of thesettings of ueXlWnTeith thatof &dcpovpwifn the following example makesclear.

P.19.4

69 Winnington-Ingram(above, note 62) 66.

n - 4! I

253

j0 a -_ r\ - TE


27/59

A. M. Devine andLaurenceD. StephensP.19.19

6-KPO - VI-We shall test the hypothesis of a Mid-High-Lowcontourfirst at the nominallevel of its implications, i.e. in terms of the proportionsof unaccentedplusaccented and accentedplus unaccented yllables withinthe wordwhich areset,respectively, to risingandfalling intervals,regardlessof the size of those inter-vals. If preaccentual yllables werelinguisticallycloser in pitch to the accentualpeak thanpostaccentual yllables, it would follow that there would be less of amismatch between the phonetic pitch contour and the musical setting whenpreaccentualsyllables were set to the same note as the accentualpeak, such aspaO9)5ev5pov P.19.1 with pa- and -OU-both at I (dl) (but -8ev- at M [c']),or xpuoea P.19.15 with xpv- and -oe- at F (fl) (but -a at A [db]), oroupavito; P.20.8 with o)- and -pd- at C(el) (but -vi- at < [d']), than whenthe postaccentualsyllable was set to the same note as the accentualpeak, suchas ai6tov P.19.23 with -6- and -kov at + (abl), n6zI P.20.21 at U (fl), orvacEta P.20.36 with -E- and -cra at V (ebl) (but with va- at < [dl]).Consequently, t wouldbe expectedthattheproportion f preaccentual yllablesset to the same note as the following accentualpeak would be greaterthantheproportion of postaccentual syllables set to the same note as the precedingaccentualpeak. Similarly,reversalof the phonetic pitchrelations,such that thepreaccentualsyllable is actually set higher than the accentualpeak should bemore common than reversal with postaccentual syllables. It is immediatelystriking that there is only one case in the two hymns of a reversal with thepreaccentual syllable, (pepo6ioto P.19.10 with wpE- t A (dbl) and -6- at M(cl), but no case of thecorresponding eversalwith thepostaccentual yllable.70To test this nominal level implicationof the Mid-High-Low hypothesis, wecollected all the cases in which the settings of the relevantsyllables were pre-served for trisyllabicandlongerwordsbearing heacuteaccent(includingpropernames)and calculatedthe proportions f the immediatelypreacutesyllables sethigherthan or equal to thepeakof theacuteandcompared he resultto thecor-respondingcalculationfor the immediatelypostacutesyllables. The resultsaregiven in Table7 separately or eachhymn.

P.19 P.20S>S S s S< s...)# 62.75% 37.25% 63.51% 36.49%#(..)SS(...)# 87.04% 12.96% 87.65% 12.35%odds ratio= 0.25 odds ratio = 0.24

70 There is one case of a second postaccentual syllable set higher than theaccentual peak SiK6po9ov P.20.2 with -x6- at C (el) and -pov- at U (fl), but theimmediately postpeak syllable shows a fall: -p-- at < (dl). It may not becoincidental that pcp6epOoto and 8txK6p-uovare compound words with recessiveaccentuation.

254


28/59

The Phoneticsof the GreekAccentchi-square= 8.31 chi-square= 12.40

Table 7: Postacute fall is more regular than preacuterise.The resultsshow that there s hardlyanydifferencebetweenthe two hymns.Thevalues of the odds ratios,0.25 and 0.24 respectively, mean that the odds for apreacute syllable to be set lower than the acute (approximately5:3) are onlyabout a quarter f thecorresponding dds forpostacutesyllables (approximately7:1). The values of thechi-squares,8.31 and 12.40respectively,mean thatthesedifferences in the treatmentof pre-andpostacutesyllables vis-a-vis the peak oftheacute arestatisticallysignificantand cannotbe attributed o chance.The Mid-High-Low hypothesis obviously implies that the mean absoluteinterval of rise from an immediatelypreaccentual yllable to the accentualpeakis significantly less than the mean absolutefall from the peak to the immedi-ately following syllable. Now the simple fact just establishedthat the propor-tion of immediatelypostaccentualsyllables set lower thanthe accentualpeak isgreater han theproportion f immediatelypreaccentual yllableset lowerwouldby itself cause the mean rise to be less than the meanfall, even if the cases in-volving only rises and fall were of identical intervals.A test, therefore,mustbeconducted only on those cases which, in fact, show rises and falls, with theinstances of equal settings(andthe one case of a reversal)excluded.The wordsstudied continue to be trisyllabicandlongerones bearing he acute on a nonfinalsyllable. The resultsare given in Table 8, wheremrindicates the meanrise andmfthe meanfall (andthese are bothreportedas positive).

#(...)SS...) # (...)s(...)#mr= 0.80 mf= 1.33Sr= 0.21 sf= 1.12mr < mf. t = 2.03, df = 65p < .025Table 8: Mean rise to the acute less than mean fall from the acute.

The averagepitchfall fromthe acute(1.33 tones)is greater hantheaveragerise to the acute (0.80 tones) by a little more than a semitone. The value t =2.03 with 65 degrees of freedom(i.e. a total samplesize of 67), meansthat thisdifference of approximatelya semitone is statistically significant.As the valuep < .025 indicates,there is less thantwo and a half chances in a hundred hat adifferencegreater hanorequalto thatobservedcouldhave arisenat random.Consequences or reconstruction,6: the test resultsof Tables 7 and 8indicate thatthemagnitudesof theexcursions o andfromtheaccentualpeak are not equal: thefall from thepeak is considerablygreater thanthe rise to it. Prepeak,peak, andpostpeak syllables thus ormed a Mid-High-Low contour. This reconstruction may be representedschematicallyasfollows:

255


29/59

A. M. Devine andLaurenceD. Stephens

s s sMid Low HighWe may extend ourexplorationof thepitchcontourof polysyllables furtherfrom the accent in both directions. Pitch tends to rise from the second to thefirstsyllablebeforetheaccent,as exemplified by a'cpovtqpP.19.19 above andP.20.20-21

i 18 T.X- Vi - TrO-VP.19.13-14

8 _ a- Kis - va,-TarConversely, pitch tendsto fall fromthe first to the second postaccentualsylla-ble, as exemplifiedbyP.19.16

TrCI Vol cr vO - -o-aivP.19.2

0i- ya-'rpesAs would be expectedfromtheparallelscited in section 12, the magnitudeof accentualexcursionsto andfromthe acute andcircumflexvariesaccordingtothe lexical status and the numberof syllables covered. We demonstrate thiseffect first for the peak to valley excursion. To control for other factors whichcan influence the magnitudeof excursions, such as emphasis, focus, terminalfall and the like, we excludepropernames,wordswhich standin initialandfinalpositions in clauses and majorphrases,words which seem, prima facie, to beespecially focused, and all wordswhich have a secondaryrise in pitch afterthefirstpostpeakfall, since the last aregenerallyassociatedwith special contextualfactors involving emphasis (see sections 2.3.4 and 2.3.7). To control for wordlength we crossclassify paroxytonesandproparoxytonesaccordingto the pres-ence or absence of an unaccentedsyllable before the acute. There are so fewwords with two or morepreacutesyllables thatthey must be ignored.Similarly,there aretoo few circumflexwords to permitthis crossclassification.The resultsarepresentedin Table9, wherethemean totalpeakto valley interval s denoted

256


30/59

ThePhoneticsof theGreekAccentas a function of lexical status(n = nonlexical, I = lexical) andthe number(1 or2) of the postpeaksyllables. The means for words with a preacutesyllable aredistinguishedby prefixingan M (for Mid tone) to the symbolsPV (:: ).S. . . SS ...PV(n,1) = 0.70 MPV(n,l): ?KEiV(x (lx) = 1s = 0.84PV(I, 1) = 0.92 MPV(I, 1) = 1.18s =0.69 s =0.82PV(I, 2) = 2.20 MPV(I, 2) = 1.58s = 0.98 s = 1.28

PV(n,l) < PV(I,1) < PV(1,2):J = 377.5, z= 3.50, p< .001Table9: Correlationf peakto valley intervalwith lexical statusandnumberof syllables.

In each column the peak to valley intervalincreasesfrom nonlexical to lexicalwords (although there is only the single datapoint, ?Keiva;, for nonlexicalswith a preacutesyllable)andfromparoxytone o proparoxytoneexicals. Belowthe #? . . . column for initial acute accents are given the results of theJonckheere-Terpstra7'est forthe hypothesisthat thepeakto valley intervalsareordered as PV(n, 1) < PV(I, 1) < PV(I,2), with at least one strictinequality.Asthe value 3.50 of the z-approximationindicates, the increase over the threeclasses is statistically significant with only one chance in a thousand that theobserved trendcould arise at random.The fact that there is a fairly substantialdecrease across the columns for PV(I, 2) to MPV(I, 2), but a small increasefrom PV(I, 1) to MPV(I, 1) indicates that there is no meaningful correlationbetweenpeakto valley intervaland thepresenceof a rise to the peak.Maintain-ing the same controls as above, but averaging over paroxytones,proparoxy-tones, perispomena,and properispomena,we find that the meanrise over onesyllable to the peak is MP(I, 1)= 0.70 and the mean rise over two or more isMP(I,x > 2) = 1.00. Althoughthe differencecannot be demonstrated o be sta-tistically significantin the small dataset available,it is in the same directionasthe peak to valley increase,andwe mayconclude thatrise to thepeak is also anincreasingfunctionof the numberof syllableswhich it covers.

Consequencesfor reconstruction, 7: the test results of Table 9 andadditional data indicate that the magnitudeof the excursions to andfrom the peak of a full accent is a function of two variables: 1) boththerise to and the allfrom thepeak aregreaterin lexicalwords than innonlexicals, and 2) both the rise and thefall increase the more moraethey cover. A schematic representationof this reconstruction s givenbelow in Figure 5, following the discussion of variation of peak andvalley pitch levels.For unemphatic exical words, variation n the magnitudeof accentual ex-cursions resultsalmost entirely from variation n the pitch of word initial Mid71 Daniel (above,note 68) 234-238.

257


31/59

A. M. Devine andLaurenceD. Stephenstones and word final Low tones, ratherthanfrom variationin the pitch of theaccentualpeak. In fact, in the sampleof unemphatic, exical proparoxytonesnmedial position, a weak, negative correlation(r = - 0.33) is observedbetweenthe total intervalof fall, HL2, and the heightof the peak,H, above a referencepitch. This correlation,however, is not statistically significant,so that we mayassume that there is no relation between the pitch of accentualpeaks and thetotal peak to valley fall in the absence of factors which might boost pitch.There is a statistically significant, although weak, positive correlation (r =+0.33) between the pitch of H and the total interval of rise from the beginningof the word to H among unemphatic exicals in medialposition which have atleast one prepeak syllable. Since this correlation s weak, however, the majorportionof the variation n the interval of rise to the peak is producedby varia-tion in thepitchof the initial Mid tone, with less dependenceon the pitchof H.The fact that the peak to valley excursionin nonlexicalparoxytonesis smallerthan in lexical paroxytonesof the same length is in substantialparta result ofthe process that subordinates he pitch of nonlexical acutes and circumflexestothe peaks of following lexical acutes and circumflexes (see section 2.3.7.2):nonlexical peak to valley excursions are smaller than lexical ones to someextent because nonlexical peaks are never higherthan lexical peaks. It seems,therefore, that when variables that influence the level of all the pitches of aword, Mid, High, andLow, such as lexicality, catathesis,emphasis, focus, etc.(see sections 2.3.5-2.3.7) are factoredout, the accentual peaks of words arerelatively independentof wordlength. It is the final Low and initial Mid tonesthat fall as additional yllablesare addedon theirrespectivesides of the accentualpeak, although the peak shows some tendency to be higher with increasingintervals of rise to it. The peaks represent a fairly stable phonetic target,whereas the valleys are variable, being contextually determined.This recon-struction,along with the small correlationof peak level withprepeakmorae,isrepresentedschematically in Figure 5, where the length of the vertical linesindicates the magnitude of the rise to the peak, P, from one preaccentualsyllable, M(1), over two, M(21,or three,M(3), and the fall from the peakto theword's pitch valley over one, V()., two, V(2), or three, V(3) postaccentualsyllables: PP

PI I M(2) M(3)

|(1) V(2V) V(2)V(3) V(3)

Figure 5: Intervals of rise and fall increase with morae covered;peaks are relatively stable.

258


32/59

The Phoneticsof theGreekAccentWhenthepeakto valley excursioncovers two syllables, thepitch trajectoryof the fall does not lie along a single straight ine connectingthe H of the peak,

the low of the first postpeaksyllable, LI, and the low of the second postpeaksyllable, L2, as in Figure6:H

Figure6: Peakto valley fall with constant lope over twosyllables.Neither does the trajectory onsist of a steep fall fromH toLI and a flat line toL2, as in Figure 7:

H

L1 L2Figure7: Peak to valley with a flat trajectoryromL1 toL2.Rather,the fall from H to L1 is greaterthan the fall fromLj to L2, so that theline HL1 has a muchsteeper slope than the line LjL2, as represented n Figure8:

H

L1/L2Figure8: Schematic epresentationf the actualpeakto valleytrajectory ver two syllablesin Greek.Table 10 gives the means HLj andLjL2 for lexical proparoxytonesof alllengths. The same controls on position and emphasis are maintained;note inparticularthat secondaryrises fromL1 to L2 are excluded, so that the meansreflect the normalpattern,H > L1 > L2, andLjL2 is not artificiallydecreased.

#... sss#HL = 1.29 L1L2= 0.50s = 1.16 s = 0.50HLI > LjL2: t = 2.33, df = 32p < .025Table10: The interval rom thepeakto the low of the firstfollowing syllable is significantlygreater han the interval romthat low to the low of the final syllable.

259


33/59

A. M. Devine and LaurenceD. StephensThe interval of fall from H to LI is over two and a half times as great as theinterval fromL1 to L2, and the difference is statistically significant.The samepatter holds for the few perispomenaandproperispomenawhose pitchrelationsare fully preserved. The same pattern HLI > LIL2 also holds for lexicalproparoxytones not propernamesand not havingsecondaryrises) which standat the beginnings and ends of clauses or majorphrases.Table 11 presentsthedata for the means in initialposition, [HLj, [LjL2and final position HLj] andL1L2]. #... SSS# in Initialand FinalPositions[HLI=1.30 HLj] = 2.00s = 1.44 s = 1.98

[LIL2= 0.60 LjL2] = 0.13s = 0.89 s= 0.23Table 11: The interval from the peak to the Low of the firstsyllable remains greater than the interval from that Low to theLow of the final syllable in initial and final positions as well asin medial positions in clauses and major phrases.The fact that, despite the compressive effects of catathesis,HLI is greaterinfinal position thanin initial or medial position reflects terminalfall. The factthat L1L2 is smaller in final position than in either initial or medial positionindicates that terminal fall was concentratedin HL1, leaving little range forfurther all over L1L2.The intervalof fall fromH to LI in all positions, initial, medial, and final,is over twice as greatas the interval fromLj toL2. Pitchdoes not fall at a con-stantrate from peak to valley, but there is a discretechange in the rate of fallafter L1. Not only is the immediate postpeak fall in proparoxytones (andproperispomena) he majorcomponentof the total peak to valley excursion, itis statisticallyindistinguishable rom the meanpeakto valley interval n lexicalparoxytonesof all lengths:PV (1,1)= 1.10, s = 0.73. This relativeconstancyofthe immediatepostpeakfall along with its greatermagnitude hantheLjL2 fallsuggests that the HLI segment is the most salient cue of a full accent. Themuch smallerintervalof theLjL2 fall segmentandits nearlyflat averageslopein final position (LIL2]= 0.13) indicatethattheLJL2segment may in fact notbe a necessary part of signalling the accent at all, but may be subject tocontextuallymotivatedreduction.This inference s important or the analysisofthephenomenonof secondaryrise discussed in section2.3.4.Data for the trajectoryof the rise to the accentualpeak covering two ormore syllables are quite restricted.Table 12 gives the meanrise from the thirdto the second preaccentualsyllable, M3M2,the second to the first,M2MJ, andfrom the first to the peak, MjH. The same controls as in Table 10 are main-tained,butwordswiththe circumflexarenow included.

M3M2 = 0.13 s = 0.25M2MJ = 0.59 s = 0.49MlH = 0.36 s = 0.32Table 12: Data on the mean rise, syllable by syllable, to theaccentual peak.

260


34/59

Devine y Stephens - Dionysius of Halicarnassus, De Compositione Verborum XI, Reconstructing the...

Documents

Transcript of Devine y Stephens - Dionysius of Halicarnassus, De Compositione Verborum XI, Reconstructing the...