JI"

LANGUAGE AND SPEECH, 1994,37(2), 103 123

103

EVIDENCE FOR TONE-SPECIFICACrIVITY OF THE STERNOHYOID MUSCLE lN MODERN STANDARD CHINESE *

PIERREA. HALLÉ Laboratoire de PsychologieExpérimentale CNRSand Paris V (Paris) The foie of the cricothyroid muscle (CT) in raising Fois weil understood,but the activily of Fo.lowering strap musclessuch as the stemohyoid (SH) has been less thoroughly investigated,especially in speech.This study focusedon the active participation of the SH in the production of IODes2 (mid-rising) and 4 (high-falling) in Modem StandardChinese.The other tones,however,togetherwith the foie of the CT and vocalis muscles, were also investigated ilJ.order to replicate earlier findings and to provide a more comprehensivepicture of the production of Chinese tones. EMG data recorded from two male speakersshow that the SH is consistently utilized to resetF0 to a mid-low value at the onsetof tone 2. Basedon a comparisonwith earlier resultsfor Thai speakers, we argue that this is a mandatory manoeuvrefor producing rising F0 contours in most contexts.The SH muscle also participates in the F0 fall of tone 4, but less consistently. We argue that the latter manoeuvremay DOtbe obligatory, especially in the case of speakerswith a high-pitched voice. Key Words: Mandarin, toneproduction, e/ectromyography,/aryngea/ musc/es

INTRODUCI'ION Although Modem Standard Chinese (Mandarin, for short) is one of the most extensively studied tone languages,little attention bas been paid to the articulatory processesinvolved in the production of ilS tones. The author knows of only one published study that bas addressedthis issue directly by collecting electromyographic

This work was supported in part by a grant to the author from the JapaneseSociety for the Promotion of Science(1989-1990). 1 thank the ResearchInstitute of Logopedics and Phoniatrics of Tokyo University for gracious hospitality; 1 am especially indebted to Prof. Hajime Hirose and Dr. Seiji Niimi for helpful advice and discussion, and to Dr. Satoshi Imaizumi for assistance.For their invaluable contribution, 1 give my thanks to Drs. Kohichi Tsunodaand Kiyoshi Ohshirna who inserted the electrodes.Special thanks are also due to the subjects who bravely performed while stoically enduring the discomfort of the experiment.

Correspondenceconceming this manuscript should be sent to P. A. Hailé, Laboratoirede PsychologieExpérimentale,28 rue Serpente,'75006 Paris (France), E-mail:[email protected]: (33) (1) 40 517085

104

Tone-SpecificSH Act;v;ty

(EMG) data (Sagart,Hallé, Boysson-Bardies,and Arabia-Guidet, 1986).The study was limited to the cricothyroid (Cf) and sternohyoid (SH) muscles; it used two female speakerswho reada small corpusof syllablesembeddedin a carrier sentence.ln contrast, the tones of Central Thai have been more thoroughly investigatedby Erickson (1976). Her study bore on Cf, vocalis, and strap muscles,whoseactivities were recordedfrom four subjects.The four tones of Mandarin have counterpartsin terms of pitch contour among the five tones of Thai. Not surprisingly, then, the EMG activity patterns in Mandarin tones observedby Sagart et al. generally bear a strong resemblanceto those observedin Thai by Erickson.There were,however,noticeabledifferencesin the activity patternsof the SH muscle betweenMandarin tones2 and 4 and their Thai counterparts. Mandarin tone 2 and the Thai rising tone are bath characterizedby a mid-to-high rising Fa contour, with a slight trough after tonal onset. Both Erickson's and Sagart et al. 's studiesfound a burst of CT activity precedingthe rising part of the tonal contour. ln Erickson's data, the initial trough of the tonal contour clearly resulted from a burst of activity of ail three strap muscles, especialjY the thyrohyoid but also the SH and the sternothyroid.Evidencefor a similar pattern in Mandarin tone 2 is very dim in Sagartet al.'s data, mainly becausesegment-relatedactivity of SH almost completely blurred Fa related activity, while, in Erickson's data, segment-relatedactivity of strap muscleswas negligible. Mandarin tone 4 and the Thai falling tone are both characterizedby a high-to-iow falling Facontour.Both studiesfound the high initial Falevel to result from a peakof Cf activity. ln Thai, the Fa fall was clearly assisted,in the secondhalf of the tonal contour, by a substantialburst of activity of aIl three strap muscles.ln Mandarin, no increaseof SH activity during the secondhalf of tone 4 was round, except for the utterance-final syllable /zi4/ ([tSl] in tone 4)', an activity which was probably related to utterance-final downdrift. Sagartet al. speculatedthat the difference betweenMandarin tone 4 and the Thai falling tone may reflect "different acousticcharacteristics".No obvious differencecan be seen,however, in the shapeor in the height of their Fa contours that could result from such a radical difference in strap muscle activity. Since acoustically similar outputs can be producedby different means,the discrepancymay simply reflect different individual strategies.But the differencebetweenthe Thai and the Mandarin datamay also be related to duration or stress2:Although Sagartet al. indicate that targetsyllables in their material alwaysreceived"strong stress",the duration of the tone-carryingpart in tone 4 was only

~~

The 'pinyin' transcription of Mandarin is used here as a phonological transcription; digits 1 to 4 are appended to tonal syllables to denote tone. Phonetic transcriptions include symbols that are traditionally used in Chinese studies (e.g., [1], a high-front apical vowel round after [s], [ts], and [tsb]). ln Mandarin, local stress and global tempo are the primary factors determining tone duration (Coster and Kratochvil, 1984). Tone duration also depends on the segmentaI structure, and on the tone itself for syllables in citation form (Howie, 1976), where tone 3 il; round to be longer; in running speech,however, tone duration dependslittle on the tone (Coster and Kratochvil, 1984).

P.A. Hailé

lOS

about 200 msec; in Erickson's study, where long vowels were used3,tonal contour duration rangedfrom 300 msec to 425 msecfor the falling tone - a sizeabledifference. The active participation of strap musclesin lowering Fa may emergeonly at such long durations. Long durations may also be necessaryfor a burst of strap muscle activity to emergeat the onset of Mandarin tone 2. This view is supportedby the "variable norms" proposedby Kratochvil which expresslawful relationshipsbetweenduration and tonal contour (Kratochvil, 1985).Kratochvil usedschematicrepresentationsof tone shapesFa values at six equally spacedpoints of a syllable's tone-carrying part) measuredin a corpusof spontaneousspeech.He found that the tone shapeswere largely determinedby tone duration for tonal syllables (though Dot for non-tonal syllables), and were best modelled by a seriesof linear relationshipsbetweenFa (at each point of the schematic tone shape)and tone duration. Figure 1 showsthe idealized tone shapesfor each of the four tonesat different durations,as modelled by theselinear relationships.Kratochvil's findings strongly suggest that, for tone durations longer than 150-200 msec, some active Fa-loweringdevice is at wor. before the onsetof tone 2 and in the secondhalf of tone 4. The Fa contour of non-tonal syllables (not shown in Figure 1) was largely flat, around the 210 Hz level. This FI) lever may be consideredas a baselineor 'neutral' Fa lever in Kratochvil's data: Fa levels departing from the neutrallevel conceivably result from an active manoeuvreof raising or lowering Fa-Such is the casefor the initial Fadip in tone 2 and for the tonal offset of tone 4: They both becomeproportionally lower with increasedduration and, from about 150-200 msec on, are both below the neutral Fa level. Thesepatternssuggestthat for sufficiently long durations,an Fa-loweringactivity (e.g., SH contraction) is at work.in toRes2 and 4. Kratochvil's findings for Mandarin tones2 and4 fit weil with the EMG datafor the Thai rising and falling tones.They suggestthat EMG patternssimilar to thoseobserved for the Thai rising and faUing tones should be found for Mandarin tones 2 and 4, provided that they are of long duration. ln addition, they suggestthat increasedduration or stresscoincides with increasedintensity of laryngeal activity. The reasonwhy Sagart el al. did DOtfind Fa-relatedactivity of SH in tone 4 may be an insufficient degreeof stress of the target syllables. ln the case of toRe 2, an additional problem was the participation of the SH in segmentaiarticulation, especiallyaroundsyllable onset,which obscuredits role in Facontrol. There was someindication of Fa-relatedactivity of the SH muscle in tone 2 only in the syllable /bi/ ([pi]), where segment-relatedSH activity was the weakest.ln contrast, the syllable /buu/ ([bu:]), which was usedin Erickson's study, inducedalmost no segment-relatedstrap muscleactivity. This study was primarily designedto re-examinethe role of strap musclesin the production of Mandarin tones2 and 4. ln addition, tones 1 and 3 and the cole of the CT and VOC muscles were also investigated, in order to replicate, and if possible to complement,earlier findings on Mandarin tone production. The SH muscle participates in segmentaIarticulation by lowering or fixing the hyoid bone during jaw opening and

Thai bas a phonemic vowellength distinction. Mandarin does not. Erickson chose CV syllables with a long vowel becausethey may carry any of the Civetones. CV syllables with a short vowel cao only carry 'static' tones: high tone or low tone.

Tone-SpecificSH Activity

106

iODe 1 350

250

325

225

300

200

175

175

250

It:::::::'

150

--~~

225

~ --oc..::'-' -. -"-.~~72.~'~'-

100

200 'N == -

~

125

0

100

200

300

400

0

100

200

300

400

0

100

200

300

400

tone 2 325 300 275 250

225 200

175

Tone duration (ms)

Fig.

Tone contour as a function of tone duration, in the four tones (source: Kratochvil, 1985)The baselineFa level of210 Hz (seetext) is indicated by the horizontal dashed-dottedlines. The dashedcontoursin tones2 and 4 correspond to 200 msecduration.

during longue lowering and backing gestures(Collier, 1975).Therefore, we usedtarget syllables whosesegmentaIstructure required thesegesturesonly to a limited extent. We usedmore refined techniquesfor lime alignment and lime normalization than in the two aforementionedstudiesso as to reducetemporal distortions due to variability in segment duration. Finally, attemptswere made to quantitatively assessdifferences in Fo-related

P.A. Hailé

107

EMG activities. We relied on the most widely acceptedaccounts of Fo-relatedlaryngeal muscle activities: The cr is the main detenninant of Fo rises in ail Fo registers (Hirose, Simada,and Fujimura, 1970;Ohala and Hirose, 1970; Gârding, Fujimura, and Hirose, 1970; Hirose and Gay, 1972; Collier, 1975; Erickson, 1976; Atkinson, 1978; Harris, 1981).The vocalis (VOC) muscle's activity bas also been observed to correlate with Fo(Erickson, 1976;Atkinson, 1978), although Jessconsistently than cr activity (e.g., Sawashima,Gay, and Harris, 1969). The role of VOC is possibly limited to counterbalancing cr tension in static configurations of the folds, as is suggested by Erickson's (1976) data: VOC activity correlates with Fo in Thai 'static' tones, not in 'dynamic' tones. The strap muscles are often regarded as synergic (Erickson, Libennan, and Niimi, 1977; Atkinson, 1978); they show a negative correlation with Fo and seemto actively contribute to lowering Fo only below an Fo 'threshold' level close to the Fo midrange (Erickson, 1976; Ericks~ and Atkinson, 1976), similar to the Fo level that we described above as 'neutral'. We assume,then, that strap muscles may cause or assistFo falls in medium and low Fo ranges(seealso Ohala and Hirose, 1970; Simada and Hirose, 1970, 1971; Atkinson, 1973, 1978; Collier, 1975; Sagan et al., 1986), whatever the exact nature of the mechanisms involved (for a discussion, see Ohala, 1972; Erickson, Baer, and Harris, 1983). Somestrap muscles also seemto be active in the high Fo range, especially in singing (Sonninen, 1956, Faaborg-Andersen and Sonninen, 1960, Niimi, Horiguchi, and Kobayashi, 1991). Niimi et al. (1991) reasonedthat the stemothyroid should play the sameraie as the cr, since it also helps tilting the thyroid cartilage downward, and found supporting evidence in the high Fo rangefor trained singers producing high-pitched 'covered voice'. The SH may also be active in the high Fo range (Roubaut, 1993); a plausible explanation is that SH cocontracts with the geniohyoid (GH): This co-contraction pulls the hyoid bone forward and downward, and conceivably helps tilt the thyroid cartilage forward and raise Fo (Honda, personal communication; see also Yoshida, Honda, and Kakita, 1993); however, it is usually observed only in extreme gestures for Fo raising, that is, DOtin Donnai speech. We did DOtinvestigate other muscles that may also contribute to Fo contrai: for example, the lateral cricoarytenoid, found to parallel the cr (Atkinson, 1978); the GH, found to act as "an extra boost" to raise Fo(Erickson et al., 1977; seealso Honda, 1983), and the cricopharyngeal muscle, active in lowering Fo (Honda, 1988; Honda and Fujimura, 1991). Finally, subglottal pressure(Ps) is generally regardedas playing a secondary role in Fo contrai (Ohala, 1978; but see Atkinson, 1973, 1978). Rose (1984) bas shawn how bath Ps and vocal fold tension contribute to the production of iODesin the Chinese dialect of Zhenhai, a nonhem Wu dialect. It seems,however, that the domain of tone in (northem) Wu dialects is, unlike Mandarin, wider than the syllable: Tone-spreading olten occurs as, for example, in Shanghai dialect (Zee and Maddieson, 1980) where the domain of tone may be a whole breath group. Since Ps panicipation in Fo contrai is more likely to occur at the breath group level than within syllable-sized domains (Atkinson, 1978; Collier, 1975), we may assumethat the role of Ps is secondarv.not primarv, in the production of Mandarin svllabic tones.

108

Tone-SpeciftcSH Activiry

We thug limited ourselves to studying the rote of three laryngeal muscles in the production of Mandarin tones: CT, VQC, and SH.

METHOD Speechmaterial Like Sagartet al. (1986) and Erickson (1976), we used syllables embeddedin a carrier sentenceto avoid contamination by non-speechmuscular activity. The carrier sentencewas/yi2ge S zi4/ ("a characterS"). The targetsyllableS underscrutiny wasoneof four syllablesproducedwith eachof the four tones.ln orderto minimize SH contribution to segmentaiarticulation,the syllablesusedwere/bi/, fmi/, /yi/, and/hu/ ([pi], [mi], [ji], and [xu»: High vowels [i] or [u] following either a bilabial closure or an homorganic approximantminimize jaw opening and ton&uelowering. Tonguebacking is expectedto occur for [xu], but early in the utterance.(Recall that /buu/ in Erickson'sdatainducedvery little strap muscle activity.) The target syllable S was not in prepausalposition, was stressed,and was precededand followed by unstressedsyllables.This was to avoid strong tonal contexteffectsas well as utterance-finalintonationdowndrift on the targetsyllable. Subjects Two male subjects, L and Z, were successfully tested. Two additional subjects participated in the experiment. but their data could not be useddue to displacementof EMG electrodes or to contamination by unwanted muscular activity. Both retained subjectswere native speakersof Mandarin, bom and raised in Beijing, aged 26 and 39 respectively,with no known speechpathology. Experimentalprocedure Hooked wire EMG electrodeswere insertedin the Cf, VOC (only for L), and SH muscles,using the long-establishedtechnique of the Researchlnstitute of Logopedics and Phoniatricsat the University of Tokyo. Correct insertion was controlled with various non-speechmanoeuvresbefore and after the experiment, and periodically during its course.The subjectswere askedto pronouncethe 16sentences(4 syllables x 4 tones)at a comfoftable speechrate; there were ten separateblocks, so that each sentencewas repeatedten times. Electrodechecking was performedevery three blocks. The audio and EMG signaIs were recorded by means of a U-matic video recorder for subject L, by means of a multi-channel DAT recorder for subject Z. The raw signaIs were then replayed,digitized, and storedin computerfiles. Data analysis Signal processing. Each of the original three- or four-channel interleaved signal files was first split into single-channel files so as to take advantage of the many available single-channelsignal processingprograms.Fa,amplitude, and rate of spectral change4 were computedfrom the audio signal every 10 msecusing time framesof 31.2 msecor 35.7 msecfor Fa (adaptingto the speaker'slowest pitch), 20 msecfor amplitude,

109

P.A. Halli

Wavefonn

:-ic-;:c:;:'-:= Amplitude

1\ ~

~

~ Rate of spectral change,\ 2 dB/DIS 4

1\

"'" ~ FO

~

1 dB/ms

~

~ ISO Hz

~' """"0

100Hz

T..Fig.2

Processingof the speechsignal: one utteranceof /mi4/ (subject L

and 32 msec for rate of spectralchange.(An example is shown in Figure 2.) The latter function is useful to localethe major speechevents.We usedit for lime alignmentandrime normalization,as discussedbelow. The raw EMG signaiswereflrst low-passedfiltered (1 kHz cut-off frequency).Their amplitudewasthencomputedevery 10msecwith a lime frame of 20 msec;this processing is equivalentto 'rectification and integration'. As a rule, the SH and cr signais were rather intenseand clean, but the VOC signal !rom subject L was somewhatweaker:Although the VOC electrode had beencorrectly inserted, the electric signal had been less amplified than for CT or SM. However, subsequentanalysesproved the recordedactivity of the VOC to be meaningful.

The spectral distance between two adjacent time frames divided by the frame duration was taken as the rate of spectral change.The spectral distance between two frames was defined as the RMS difference in dB between the two con'esponding shol1-term Barkscaled energy spectra (dB per msec is the unit used in Figure 2).

110

Tone-S~cific SH Activity

Averaging method.The usual method of averaginga physiological function, such as integratedEMG activity, acrossrepetitions of a sentenceconsists in ftrst locating a specific acousticevent in eachutterance,the 'line -up' point, then aligning aIl utterances on that point, and finally averagingthe function acrossthe lined-up utterances,within a domain of interest. This 'ensemble averaging' procedureis quite valid as long as the domain of interest lies close to the line-up point and the fluctuations in articulation rate are small acrossrepetitions.A novel method of non-linear lime alignment has recently been proposedto cope with articulation rate differences(Strik and Boves, 1991): Basically, dynamic programmingis usedto optimally time-warp eachrealization of the same sentencesoas to minimize ils acousticdistancefrom a referencerealization.This method can be understoodas using a seriesof many line-up points insteadof just one: It may be neededfor long, complex sentences.We usedan alternativemethodusing only two lineup points between which lies the domain of interest, that is the target syllable. The releaseburst of [k] in the syllable Igel and the vocalic onset in the syllable Izi4/, which precedeand follow the target syllable and ca'.J.always be easily located(at a clear peakof the rate of spectralchangecurve for Izi4/) were chosenas the two line-up points. For eachsentence,the referenceutterancewas the one whoselime distancebetweenline-up points was the median of aIl such distances. ln order to gel them aligned with the referenceutterance,the other utteranceswere linearly compressedor expandedso that line-up points coincided.As a result, both lime alignment and lime normalization (to the lime scaleof the referenceutterance)were performed.As expected,this methodgreatly reducedthe timing differences betweenrepetitions of each sentence.The vowel of the target syllable was of special interest since it bore the tonal contour underexamination5: The variability of its onset and offset (relative to the first line-up point) and of ils duration was computedfor each sentencebefore and after lime normalization. Standard deviationsdroppedconsiderably,down to 10-15 msec.(10 msec was the sampling rate of ail functions derived from the audio or EMG signaIs.)The meantOGeduration of the target syllable rangedfrom 170 to 245 msec according to syllable type. Averageduralions ofboth tOGes2 and 4 were about 215 msec.

RESULTS EMG patterns in thefour tones The timing of cr and SH activity related to the production of the target syllable tone was stableand consistentacrosssyllable-type.This is illustrated in Figure 3 where /bi4/ and/mi4/ are superimposed.Note that the timing of EMG activities is stablerelative

Traditionally, phonological descriptions of Mandarin (e.g., Chao, 1969) allegro that the domain of tone is the entire voiced part of the syllable. Howie's (1974) phonetic smilles showed,however,that the domain of tone in Mandarin is confmedto the myrne: the syllabic vowel and any voiced segmentthat may follow it. ln particular,Fo movementsin syllableinitial sonorantsare irrelevant to the tonal contourper se. ln the material we used,all target syllableswere CV syllabJes.Hence,their tone-carryingpart was V, the syllabic vowel.

111

P.A. Hailé

Line-up#l

Fig.3.

Line-up#2

Superimposedcurves for averaged/bi4/ and /mi4/ utterances.Vertical bars for Fo and thin lines for amplitude are for /bi4/. horizontal bars and thick lines for /mi4/ (subjectL).

to the vowel of the target syllable. It may be thought of aslargely invariant relative to the whole syllable only if, for example,the occlusion is consideredpart of the /bi/ syllable. ln any case,the timing is suchthat similar contoursare obtainedon the vowel, whatever the initial portion of the syllable. The Fo movementobservedin the initial sonorantlm/, for example,reflects the continuousregulationof Foand is usually not takento characterize the tone of the Irnil syllable (seeNote 5). The patterns of cr and SH activity were found to be largely invariant across segmentaIvariations and acrossthe two subjects.The EMG activity patternsshown in Figures4 (L's data) and 5 (Z's data) are averagedacrossaIl four syllables. The results regardingtarget syllablesobtainedby Sagartet al. (1986) were largely replicated.Fo rises (precedingtonal onset in tones 1 or 4, in the secondhalf of the tonal

~

Tone-SpecificSH Activity

112 (Me

1)

~

~

(toae3)

LiDe-upn

~

-"

~

~

Ampütude

1 AmpUtude

A '"

.,./\

~

=:.i-.f

SR

--'"

p;r~-,

/'-

~

ill

j=~~~b:~ cr

cr

~

1

~

/'--

~

1\

zoo ~v

100~V

zoo~v .~\

"i'O

~v ~v

voc 1 1

~v

!

l'~V

-

(IODe 2)

FO

1JDe-apn

Liae-apli ~

1..,.

'---

-."'"

...'

LiDe-llp#l

,

,

n

'

'.""

Une-upn , ,

~,

";1

"

"1 100Hz

:LJ

Amptitude

j\

\'

SB

SR

~

Itone4) -~ ~

,

1\ ~ f'.ll\ /'"\

Y1\

~

;

1""-

~

,

cr

/""

1".

~

/

(

\

+W

100"V

~

Fig.4.

Fa.amplitude. and EMG patternsfor the four tonesaveragedacrossrepetitions and syllables (subject L).

~

P.A. Hailé

113

u..upIZ

--

'-

-

1zoo Hz

! 100Hz

lOO,.V

,z u 1

~

Fig.5.

1

,

200&

1""-

lCM!Hz

Fo. amplitude. and EMG patternsfor the four tones averagedacrossrepetition: and syllables (subjectZ).

114

Tone-SpecificSH Activity

contour in tone 2) were precededby a burst of cr activity. A moderatelyhigh level of cr activity was maintainedthroughout tone 1 whose tonal contour was high and level. cr activity otherwise remained at a minimal, 'rest' level. ln particular, cr activity was at this rest level throughout tone 3. Extremely high SH activity was found in tone 3; it is presumablyrelated to the 10wFolevel in this tone. Finally, we also found an increaseof SH activity in the initial part of aIl target syllables; this increasewas only moderatein tone 1 where it was presumablymainly related to segmentaiarticulation: It was lowest for /yi/ (L) or /hu/ (Z) and somewhatlarger for /mi/ and /bit (both subjects). Somenew patterns,however, emergedin our data. First, a moderateburst of SH activity occurred in the second part of tone 4, whatever the segmentai structure for subject L, only for syllables /bi4/ and /hu4/ in the caseof subject Z. Second,a similar burst of SH activity was very clear precedingthe initial part of tone 2 for subjectL in aIl syllables,for subjectZ in /hu2/ and, to a lesserextent, in /bi2/. Finally, VOC activity (L) was much weaker than that of other muscles.ln consequence,VOC activity patterns,as shown in Figure 4, remain unclear. Consi.*nt patterns were revealed, however, by a comparisonmethod exposedin the following section: VOC activity roughly paralleled cr activity; it was the weakestduring tone 3, and reacheda higher level during tone l, in the beginning of rODe4, and in the secondhalf of tone 2. The reliability of theseSH and VOC patternsof activity is examinedin the following sections. Aside from EMG patterns related to target syllables, we observed a consistent pattern of EMG activities at the ends of aIl utterancesfor both subjects: a very intense burst of SH activity, centredon the offset of the utterance-finalsyllable Izi4/, followed by a rather large increaseof activity of the cr (and VOC for L). This intense burst of SH activity may be responsiblefor the intonation downdrift that terrninatesbreath groups. The subsequentincreasein cr and VOC activity presumablyincreasedfold stiffnessand thickness,which, combined with vocal fold abductionresultedin voice terrnination. SH activity in tones2,3, and 4 Since the SR is generally involved in both Fo control and segmentaIarticulation (Collier, 1975),the questionarisesof whetherthe SR activity found in tones2, 3, and4 is at aIl relatedto Focontrol. As can be seenin Figures4 and 5, the SR activity relatedto the targetsyllable is the weakestin tone 1, where it precedessyllable onset.This activity, if it was related to Fo contrai, couId only contribute to the initial Fo-raising since tone 1 is essentially high-ievel. Such SR contribution can be observedin extreme gesturesfor raising Fo but is unlikely in the presentcase.Rence,we may assumethat SR activity in tone 1 is mainly segment-related.The SR activity profile in tone 1 may thus serveas a 'baseline' for estimating Fo-relatedSR activity in other tones. For example, in order to estimateFo-relatedSR activity in tone 2, two setsof utteranceswere compared:the same targetssyllables (to minimize differencesin segment-relatedSR activity), producedwith tone l and with tone 2. The two sets of utterances were first time-aligned and time-normalizedtogether; the difference in SR activity levels betweenthe two setswas then assessedby meansof Student's t values computedat each point in time. The plot of t valuesalongtime indicateswheredifferencesaresignificant, hence,whereSR activity in the tone comparedwith tone l is Fo-related.Figures6 and 7 are an illustration of this

~

P. A. Haiié

Ils

-../~.J'"r~.t:~-~/J

~(::;:~t~~~~=~~:t~~J

-

~

~\

1Studat's

1.

t (df=11) i

J\

11-. 11\

~V

1\

6..wz

4

2.947

~ Fig.6.

W~~I

.,~!f1-

Comparisonof SH activity between(a) tone 2 and tone l, and (b) tone 4 and tone 1 (subjectL, syllable /yi/). The two labelled levels of t valuesshown in this figure and in Figures 7 and 8 correspond to p

= 0.01 and p = 0.0000 1. Vertical

bars(Fo) and thin lines (EMG) are for tone l, horizontal bars and thick lines are for tones2 or 4. procedure for lyil (L) and /hu/ (Z) respectively. They show that SH activitY in the initial part of tone 2, and in the second part of tone 4 was significantly larger than in the corresponding portions of tone 1. This was especially clear in L's data. ln Z's data, Forelated SH activity in tone 4 was more spread out; for this subject, an increase of SH activitY also occurred at the onset of utterance-final lzi41 much more markedly after tone 4 than after tone 1,2, or 3, as can be seen in Figure 5 (all syllables), or in Figure 7 (/hu/ syllable). The intense SH activitY observed in tone 3 for both subjects (see Figures 4 and 5) leaves tittle doubt as to its Fo-related nature: Indeed, this was confirrned by t-test comparisons.

cr and VOC activity in fanesl, 2, and 4 Just like the SH in iODel, the cr remains at a minimal activity level in iODe3: Indeed,iODe3, the low-falling ioDe,doesnot require Fo-raisingactivity. The cr activity profile in iODe3 may thus serve as a baselinefor estimating Fo-relatedCT activity in other tones. It is already fairly weIl known, however, that CT activity is mainly Forelated.Not surprisingly, then, ibis was confirmed by t-test comparisonssimilar to those conductedfor the SH. As to the VOC muscle,since its activity profile in iODe3 was the lowest and unlikely to be Fo-related,it was used as a baseline in t-test comparisons

116

1

(al

Tone-SpecificSB Act;v;ty

"_(L ~~

.

~

. ~.-.;

~

l.8I-..,z

'-

1n..c~ ~

t~

~

~

1-. ..

1... ..Ha

,-_V

L-.s

- , '. t C«~-..

~

I.~V

~

,.. 2.H7

~ Fig.7

~

-t'8"

Comparisonof SH activity between(a) tone 2 and tone l, and (b) tone 4 and tone 1 (subject Z. syllable /hu/). Vertical bars (Fo) and thin lines (EMG) are for tone l, horizontal bars and thick lines are for tones2 or 4.

betweentone 3 and the other tones.VOC activity in tones 1,2, and 4 strikinglY paraileled cr activity: It was significantly higher than the baselinealmost in the saIneregionsas for the Cf, as illustratedin Figure 8 showingL's datafor /yi 11and/yi41 comparedto /yi3/. Overa// differencesin EMG activity ln the previous sections, tone-specific patternsof EMG activity were isolated by visual inspectionof EMG and Focurvesas in Figures4 and S, and by local comparisons with baseline EMG profiles that were assumedto reflect minimally Fo-relatedEMG activity. Another means for assessingtone-relateddifferences and for 'factoring out' segment-relateddifferences consists in comparing the variability induced by tone variation alone (keeping the segmentsconstant)to that induced by segmentaivariation alone (keeping the tonesconstant).ln other words, we May compareinter-tone variability to inter-segmentaivariability. A simple measureof the overall differencebetweenIwo activity profiles of a given muscle is given by the following distance: 1

"

d(x.y)= Llx(t) I-~

If

- Y(t)1

L