Haskins Laboratoria Statlu !Upon on SpeechRu«Jrch1992, SR·I09/110, 59-72

Reiterant Speech as a Test of Nonnative Speakers' Masteryof the Timing of French*

Andrea Levittt

The reiterant speech of ten native speakers of French was analyzed to develop baselinemeasures for syllable and consonant/vowel timing for a series of two-, three-, four-, and five­syllable French words spoken in isolation. Ten native speakers of English, who learned Frenchas a second language, produced reiterant versions of both the French words and a comparableset of English words. The native speakers of English were divided into two groups on the basisof their second language experience. The first group consisted of four university-level teachers,who were relatively experienced learners of French, and the second group of six lessexperienced learners of French. The French reiterant imitations of the two groups of nativespeakers of English were compared to the native French speakers' productions. The timingpatterns of the experienced group of non-native speakers did not differ significantly fromthose of the native French speakers, whereas there was a significant difference between thesetwo groups and the group of six less experienced second-language learners. Deviations fromthe French baseline measures produced by the less experienced group are discussed in termsof the influence of the timing patterns of English and the literature on a sensitive period forsecond language acquisition.

INTRODUCTION

Although considerable research shows that na­tive language phonetic habits influence secondlanguage productions, even for experienced sec­ond-language speakers (see Flege, 1986, for an ex­tensive review), little work has been done on theinfluence of first language timing patterns on sec­ond language rhythmic patterns. One such study

This research was supported by NIH grant DC..()().403 andNICHD Grant HD01994 and BRS Grant 05596 to HaskinsLaboratories. I am grateful to Wellesley College for providingadditional funding and to LeBlie LeBarron and Katy IIIlUIC8 forconducting preliminary statistics on lome of the earlydurational measurements. Mindy Levitt provided valuableassistance with data entry, and Tony Hatoun supplied lpecialstatistical programa. The French subjects were ron while theauthor was supported by the North Atlantic TreatyOrganization under a grant awarded in 1983 and held at theCentre National de la Recherche Scientifique in Paria. I thankJacques Mehler for making available the facilitiel of hislaboratory. Catherine Best, Jane Baier, Jim Flege, Bob Frye,Alice Healy, Patti Price, Bruno Repp, Michael Studdert­Kennedy, Doug Whalen, and an anonymous reviewer providedhelpful comments on previous versions ofthiB manuscript.

59

(Wenk, 1985) found an influence of native Frenchrhythmic patterns on the timing of English as asecond language. However, the effect of Englishtiming patterns on the acquisition of French hasnot been directly tested.

The use of reiterant speech to test for suchinfluence presents several advantages. Inreiterant speech studies, subjects are asked tosubstitute a single syllable, often /ma/, for each ofthe original syllables in a word or sentence.Acoustic and perceptual analyses of reiterantspeech have shown that it preserves the prosodiccharacteristics of the original utterance (Larkey,1983; Liberman & Streeter, 1978; Nakatani,O'Connor, & Aston, 1981; Oller, 1973).Furthermore, because measurements of segmentand syllable durations are easy with reiterantspeech and are generally unconfounded bysegmental variation, many studies have used suchduration measurements in English for analyzingrhythm (e.g., Nakatani et a1., 1981), for studyingthe perceptual effects of timing variations(Larkey, 1983; Nakatani & Schaffer, 1978), andespecially for determining how durations vary as afunction of utterance position and stress (e.g.,

60 Lantt

Oller, 1973). Reiterant speech durationmeasurements have also been made on Swedish(e.g., Lindblom & Rapp, 1973), and comparisons ofthe rhythmic features of a group of languageshave been made on the basis of reiterant speech(Hoequist, 1983; Vatikiotis-Bateson, 1986).However, very little work has been done withreiterant speech on the rhythmic features ofFrench, aside from that done by Vatikiotis­Bateson (1986), where reiterant speech was usedto determine universal and language-specificeffects on articulator timing in native speakersfrom a group of languages. The use of reiterantspeech as a means of testing a non-nativespeaker's mastery of the timing patterns of aforeign language has not been previouslyattempted. In learning a second language,speakers need to learn new timing patterns forindividual segments, often as a function of context(Mack, 1982), as well as new rhythmic patterns.Reiterant speech is particularly well suited totesting the acquisition of new rhythmic patternsindependently from the effects of timing for non­native segments.

The speech rhythm of French and that ofEnglish are quite distinct. French has been tra­ditionally classified as a "syllable-timed- language(e.g., Pike, 1945), with syllables essentially equalin length. This characterization of French rhythmhas been criticized (e.g., Dauer, 1983; Fletcher,1991; Wenk & Wioland, 1982) for failing torecognize the important final-syllable lengtheningthat is characteristic of French rhythmic groups,which may be either the individual "sense groups­of a French sentence or individual French wordsspoken in isolation. Thus, nonfinal syllableswithin unemphatic French rhythmic groups are,except for effects of phonetic variation, essentiallyequal in length, whereas final syllables showconsiderable lengthening. English, on the otherhand, has been traditionally classified as a "stress­timed- language (e.g., Pike, 1945). Because ofvariable word stress, any English sentencepresents a series of stressed syllables whichalternate with unstressed syllables. A stress­timed language is supposed to maintain equalintervals between stressed syllables. Thus, if aninterval between two stressed syllables containsmore unstressed syllables than another, thoseunstressed syllables should show relativelygreater compression. English also exhibitscharacteristic patterns of final-syllablelengthening, including word-final, phrase-final,and utterance-final lengthening (Oller, 1973).

Altho~h the characterization of English as a"streas-timed- language has also been criticized(e.g., Dauer, 1983; Wenk & Wioland, 1982), itsrhythmic pattern is nonetheless quite differentfrom that of French, especially in two salient re­spects. First, in English, nonfinal syllables willvary in length as a function of stress, whereas inunemphatic French, nonfinal syllables within arhythmic group are essentially equal in length.Second, although both languages exhibit final­syllable lengthenini, the magnitude of the final­syllable lengthening effect and its location bothvary. Thus, the magnitude of utterance-finallengthening is greater in French than in English(Delattre, 1966). In addition, in English, utter­ance-finallengthening appears to be greater than.phrase- or word-final lengthening (e.g., Oller,1973). A similar difference in the magnitude offinal-syllable lengthening has been observed forutterance-final compared to phrase-final lengthen­ing in French (Benguerel, 1971; Fletcher, 1991;but cf. Allen, 1973), but not for words. Frenchwords exhibit final lengthening only at the ends ofrhythm groups or when uttered in isolation.

Which of these rhythmic differences are second­language learners of French likely to master first?On the one hand, since both languages exhibitfinal-syllable lengthening, English-speakinglearners of French might find it easier to adjustthe magnitude of such lengthening as they acquirethe rhythm of French. On the other hand, Flege(e.g., Flege, 1981; Flege, 1987; Flege &Hillenbrand, 1984) has proposed that second­language learners are more likely to master thetotally new phonetic features of a second languagethan those that can be assimilated te· their nativerepertoire. In that case, English-speakIng learnersof French might find it easier to acquire therelatively equal timing of nonfinal syllables inFrench, which is not found in English.

In order to conduct a test of the acquisition ofFrench rhythmic patterns by native speakers ofEnglish, it is first necessary to establish baselinemeasures for timing patterns in French using thereiterant productions of native speakers ofFrench. Not all speakers are equally good atproducing reiterant speech that preserves thetiming of the original utterance (Larkey, 1983).Thus, it is important that the baseline measuresbe based on the fluent productions of the bestreiterant speakers. Once these measures havebeen established, they can be compared topublished findings about the durations ofconsonants, vowels, and syllables in French.

Reiterant Spe«h lIS il Test ofNo1l71Jltive 5pt4km' MllStery ofthe Timing ofFrmch 61

Experiment I reports the results of an experiJnentdesigned to produce such data.

We may then ask how well non-native speakersof French match the timing patterns of the nativeFrench productions. In Experiment II, reiterantversions of both French and English words madeby native speakers of English were analyzed inorder to establish a similar set of baselinemeasures for reiterant English, to determine howwell the non-native speakers of French differing indegree of experience with the language match thetiming of the productions of the French speakers,and to see whether any deviations from theFrench baseline measure stem from the influenceof English timing patterns.

I. EXPERIMENT 1A. Subjects. Ten subjects, five male and five

female, participated in the study. All were nativespeakers of French from the Paris region. All ofthe subjects have advanced graduate degrees.Although the majority of their daily verbalexchanges took place in French, all the subjectshad some experience with other languages, as istypical ofhighly educated Europeans.

B. Test materials. The materials for theexperiment consisted of a set of 30 French words,6 two-syllable, 12 three-syllable and 6 each offour- and five-syllable words. (See the Appendixfor a complete list of the stimuli.) As stress inFrench is on final syllables, and all of the wordswere produced in isolation, all of the two-, three-,four-, and five-syllable words in French werestressed on their final syllable. Each word wastyped on the center of a 3 x 5 card. The cards werepresented in the same random order to allsubjects.

C. Procedure. Recordings were made in asoundproof booth using a Sony tape recorder(model TC-510-Z) and a Sennheiser microphone(model MD 441-V). The subjects read the wordtyped on the card out loud and then reproducedwhat they had just said by substituting thesyllable Imal for every syllable of the original,while preserving both its timing and the melodiccontour. They were asked to be careful to use thesyllable Imal in all cases and to repeat a stimulusitem and its reiterant version, if they felt they hadmade an error.

D. Equipment and measurement methods. The30 French words and their reiterant versions werelow-pass filtered at 4.9 kHz, digitized at 10 kHz,and stored on disk, using Haskins Laboratories'Vax 11-780 computer. All durational measure

ments were made by the author on the reiterantspeech using large-scale waveform displays, witha resolution of 0.1 ms. Differences in amplitudebetween the consonant and the vowel, as well asdifferences in the appearance of the waveformsassociated with 1m! (the nasal murmur) and laI,made segmentation relatively easy. This was par­ticularly true for reiterant productions by Frenchspeakers. It was very easy in almost all cases tosegment the 1m! and the Ial because French 1m!and Ial are kept quite distinct, whereas Englishoral vowels in a nasal environment often showsome nasalization <Clumeck, 1975). When therewas a question about the location of a particularboundary, it was resolved through listening to thesegments in question. The most common segmen­tation difficulty arose in determining the locationof the end of the word. A consistently conservativecriterion was applied, such that the termination ofperiodicity was used to mark the end point. Thisexcluded breathy releases, but seemed best forconsistent comparisons across speakers.

In order to test the reliability of the durationmeasurements, a random sample of 12 Frenchreiterant utterances containing 82 separatemeasurements were measured a second time bythe author. Absolute duration measurementdifferences were within 4 ms of the original on theaverage overall and within 9 ms on the average onthe 12 final vowel measurements.

Not all individuals are equally adept atproducing reiterant speech that faithfully mimicsthe prosodic characteristics of the originalutterances. To construct accurate timing models,we must require that the reiterant utteranceschosen for analysis come from subjects who havedemonstrated that they are capable ofneutralizing inherent segmental lengthdifferences. That is, the subject must producereiterant syllables of the same length, all otherthings being equal, for both original syllables thatare inherently long and for ones that areinherently short. Reiterant speech studiestypically use specially constructed sentences thatare rhythmically matched, based on their stresspatterns, although one sentence of each paircontains words with inherently long syllables andone sentence contains words with inherently shortsyllables. Thus, the sentences in each pair arerhythmically the same, with the same number ofsyllables and the same locations for stressedsyllables, but the individual syllables vary inlength. Subjects should produce essentiallyidentical reiterant productions for both sentences

62 Lermt

in a set, if, in fact, they are neutralizing intrinsicdifferences in the durations of individualsegments.

In the present study, each of the two-, three-,four- and five-syllable word-length types had syl­lables composed of segments of inherently differ­ent lengths. Thus, instead of using a sentence­length test, measures of subjects' duration mea­surement variability in producing word types wereused as an indication of their ability to neutralizeinherent segmental length differences. Eachreduplicative version of a particular word of agiven length was considered a token of that word­length type. The standard deviations for compa­rable measurements, e.g., first syllable length,were calculated across tokens for each subject foreach word-length type and averaged. Separatevalues were calculated for each of the four word­length types because it is generally more difficultto produce good reiterant versions for longer ut­terances. Finally, an overall mean (measure A)and a standard deviation (measure B) of each sub­ject's mean standard deviations for the fourFrench word-length types were calculated. Theoverall group mean was 25 ms for measure A and20 ms for measure B. Subjects were rank orderedon both measures, and three subjects, one femaleand two males, showed means and standard devi-

400

300

E 200

100

o

ations that were consistently longer than theother subjects (35 ms for measure A and 33 ms formeasure B for the group of three). They had alsoproduced more errors between them (16) than theother seven subjects combined. Their data wereexcluded from the construction of the Frenchbaseline measures for timing. For the remainingseven subjects, the mean for measure A was 19 mswith a mean 14 ms for measure B.

E. Re,ult,. There were only seven errors madeacross the seven subjects (3%), most of whichinvolved the addition or deletion of a syllable,usually on words of four or five syllables. Allerrors were excluded from the construction of thebaseline measures for timing. There were also twoinstances of missing data (1%).

Figure 1 shows the mean durationalmeasurements for the syllables of the reiterantversions of each of the four word-types in terms ofthe mean durational measurements of theconsonants (1m/) and vowels (la/) of each syllable.The mean duration of ImI in nonfinal syllables was83 ms, of ImI in final syllables was 103 ms, of /aJ innonfinal syllables was 93 ms and of final /aJ was171 ms. Nonfinal syllables averaged 175 ms inlength, whereas final syllables measured 274 mson the average, an increase of almost 100 ms or afinal/nonfinal ratio of 1.6,1

• V• C

1w2 s 1w2w3 s 1w2w3w4 s 1w2w3w4w5 s

Two- to five-syllable words

Figure 1. Consonant and vowel durations, as a function of wordlensth, syllable position, and stress, for reiterantproductions of French words spoken in isolation by native speake.. of French. (Numben indicate syllable position, Sindicates stressed syllables, and W indicates unatrelHd syllables).

Reiterrmt Speech lIS II Test ofN01J1JIltifJe Spetl1cers' Mutery ofthe Timing ofFrench 63

Ratios

Ratios

Word Length in Syllables

Two 1bree FolD' FiveMean Word Length 448.2 611.6 776.2 1027.5

Table 1. Mean word lengths (in ms) and CIV andCV/length ratios in reiterant speech productions ofFrench words by native speakers ofFrench.

F. Discussion

The results of this experiment showed fairlygood agreement with the published data onFrench, especially with respect to French syllableduration ratios. The segment measurements willbe considered first and then the syllablemeasurements.

The duration measurements for French nonfinal1m! and Ial and for final 1m! tended to be roughly20 me longer than the durations found for thesame segments by other researchers (Di Cristo,1980; O'Shaughnessy, 1984; Smith, 1977). Thisdiscrepancy is most likely due to the fact that thesubjects in the present experiment spoke at aslower rate in producing reiterant speech than thesubjects in the other studies, who read Frenchtexts. The measurement for utterance-final/al wasroughly 10 ms longer than that of O'Shaughnessy(1984). The smaller discrepancy in final position isprobably due to the conservative segmentationcriterion adopted in the present study. Thus,given the segment values of the present study, thenasal consonant 1m! accounted for 47% of theduration of nonfinal syllables, whereas for finalsyllables, it accounted for 38%.

In general, nonfinal syllables were remarkablyclose in duration (see Figure 1). The present datadid not show an initial syllable shortening ascompared to medial syllables, which disagreeswith Crompton's (1980) finding of decreasedlength for initial syllables. In fact, anotherresearcher <Vaissiere, 1983) has found growingevidence in French of a tendency to stress wordinitial syllables, and presumably to lengthenthem. Indeed, one of the subjects showed a regularlengthening of initial syllables. Crompton (1980)also found evidence for prenuclear lengthening, orlengthening of a syllable just prior to a nuclearstress. An analogous penultimate syllablelengthening has been described by Smith (1977)as characteristic of Parisian French (althoughonly one of Crompton's four subjects was fromParis, while the other three came from Brittany).The present pooled data show no overall effect ofpenultimate syllable lengthening, although datafrom two of the speakers do show such an effect.

The ratio of final syllable to non-final syllablelength in the present data was 1.6, which agreesexactly with Parmenter and Blanc's measure of1.6 (1933), with Benguerel's (1971) measure of 1.6,and with Allen's (1983) finding of an overall ratioof 1.6 when he compared the median lengths offinal to penultimate vowels in French children'sproductions of French words. It does not matchDelattre's (1966) measure of 1.8, perhaps because

.9 .9 .8 .9

.6 .8 .9 1.0.6 .9 1.0

.7 .9.7

.4 .3 .2 .2

.6 .3 .2 .2.4 .2 .2

.4 .2.3

CINIC2N2C3N3C4IV4CSNS

CVIILCV2ILCV31LCV4ILCVSIL

This final-syllable lengthening was found to besignificant in the results of a two-way analysis ofvariance comparing the subjects' mean nonfinaland final syllable lengths for the four word-lengthtypes [F(1,6)=130.19, p < 0.0000]. There were noword-length type and no word-lenrth type bysyllable position interactions. Analyses comparinlsubjects' mean nonfinal syllable lengths for each ofthe four word-length types were also notsignificant. 2 A separate two-way analysis ofvariance to explore segment length in final andnonfinal syllables again showed a highlysignificant effect of syllable position JF(1,6)::105.8,p < 0.0000]. There was also a significant effect ofsegment type [F(1,6)=46.01, p <. 0005], and asyllable by segment type interaction JF(1,6)::60.26,p < 0.0002]. Post hoc tests (Newman-Keuls)revealed that final Ial was significantly differentfrom nonfinal Ial and from final and nonfinal 1m!and that final /m/ was significantly different fromnonfinal 1m!, all at the p < 0.05 level or better.Nonfinal 1m! and Ial were not significantlydifferent from one another.

Table 1 shows the mean length of each of theword types and the ratio of the mean length of theconsonant to that of the vowel in each syllable.The overall mean CN ratio was .9 for nonfinalsyllables and the CN ratio was .6 for finalsyllables. In addition, Table 1 presents the ratiosof the mean syllable length to the word as a whole.

64

of differences in the criteria used for measuringfinal syllable lengths.

In summary, our French ti~ data based onreiterant speech productions of French wordsspoken in isolation showed generally consistentsyllable durations for nonfinal syllables and aratio of finallnonfinal syllables of 1.6. Individualsubjects showed some slight l~eningof initialor penultimate syllables, but no consistentevidence for any shortening effects. Insofar asintrasyllabic timing is concerned, in nonfinalsyllables, the nasal accounted for 47% of theduration, and in final syllables, it accounted for38%. How well then do non-native speakers ofFrench match these characteristic durationpatterns when they produce reiterant speechversions of French words?

II. EXPERIMENT 2A. Subjects. Ten subjects, five male and five

female, participated in the study. All of thesubjects except for one have advanced graduatedegrees. All are native speakers of English,currently living in the Boston area, who havestudied standard French. Four of the subjects (twomen and two women, including the author) teachFrench at the university level. One subjectlearned French from his French wife, whom hemet after graduate school. The other subjects allhad some formal training in French; sevensubjects began the study of French in high schooland the remaining two in junior high school. Thefour teachers of French and the other subjects,with the exception of the subject who learnedFrench at home, averaged over two years of highschool French. The four French teachers, however,studied French for four years in college, ascompared to an average of slightly over 1 m yearsin college for the others. The four French teachersalso completed postgraduate training in Frenchand had traveled more extensively in French­speaking countries than had the other subjects.

B. Test materials. The same French deck of 3 x 5cards used in the previous experiment was used inthis second study. An additional deck consisting ofthe English cognates of the French words was alsoused. The 30 English words consisted of two,three, four or five syllables. There were tenpossible stress patterns represented. For words oftwo syllables, both initial and final primary stresspatterns occurred (sacred and degree.) For wordsof three syllables, initial, medial and final primarystress patterns occurred (compliment, instructive,and engineer). For words offour syllables, three ofthe four possible primary stress patterns occurred

(commentary, economy, and expoaition). For wordsof five syllables, two possible patterns occurred(electricity and communication). There were threedifferent words representing each of the syllableand stress types.3 Although in general most of thecognates had the same number of syllables in thetwo languages, there were three items for whichthe syllable count differed. (See the Appendix for acomplete list of the stimuli used).

C. Procedure. Subjects first filled out a shortquestionnaire about their years of experience withFrench and were then recorded in a quiet room,onto a Teac tape recorder (model X-7MKII) usinga Realistic dynamic microphone (model 33-984A).The rest of the procedure was the same as in theprevious experiment, except that subjects readand produced reiterant versions the words of theEnglish deck first.

D. Equipment and measurement methods. All 30French and 30 English words and their reiterantversions were low-pass filtered at 4.9 kHz,digitized at 10 kHz, and stored on disk on HaskinsLaboratories'Vax 11/780. The same criteria usedin the previous experiment were used here todetermine the consonant and vowel boundariesand the end of the reiterant speech utterance.

A random sample of fourteen reiterantproductions of English words containing 102separate measurements were measured a secondtime. The absolute duration measurements werewithin 4 ms of the original measures on theaverage overall, and within 9 ms on the averagefor the fourteen final vowel measurements.

The errors from both sets of reiterant produc­tions will be discussed first. The data fromExperiment 2 will then be presented as a set ofbaseline measures for consonant, vowel, and syl­lable timing for English words of various lengthsand stress patterns based on the productions ofthe most consistent reiterant speakers. Third, theEnglish speakers' reiterant versions of the Frenchwords will be examined for patterns of intra- andintersyllabic timing. Finally, the durations of theproductions of the French native speakers will bestatistically compared to those of the non-nativespeakers, broken into two groups, the relativelyexperienced teachers of French and the other, lessexperienced group of French leamers.

As with the French subjects, measures of theAmerican subjects' duration measurement vari­ability in producing word types were used as anindication of their ability to neutralize inherentsegmental length differences. Each reduplicativeversion of a particular word of a given length andstress pattern was considered a token of that

Reiterrmt Spe«h lIS" Test ofNomrJItiTJe 5pe41rm' Mutny 01the Timing ofFrench 65

word-lengtblstress-pattem type. The standard de­viations for comparable measurements, e.g., firstsyllable length, were calculated across tokens foreach subject for each of the ten word­length/stress-pattern types and averaged.Separate values were calculated for each of theten word-lengtb/stress-pattem types because it isgenerally more difficult to produce good reiterantproductions for longer utterances and becausevariable word stress in English affects the dura­tion of syllables in comparable positions. Finally,an overall mean (measure A) and a standard devi­ation (measure B) of each subject's mean standarddeviations for the ten word-lengthlstress-pattemtypes were calculated. For the English words, thegroup mean on measure A was 18 ms with a groupmean on measure B of 17 ms. When the subjectswere rank ordered on these two measures, twosubjects, one male and one female, showed thehighest scores on both measures (for measure A,their mean was 26 ms, with a mean of 24 ms formeasure B). The remaining eight subjects showeda group mean of 17 ms on measure A and 15 mson measure B. In constructing the baseline mea­sures for timing for the English words, only thedata from the eight most consistent subjects wereincluded.

E. ResultsThe American subjects made relatively few

errors in their reiterant versions of the Englishwords. The twelve errors across the eight mostconsistent subjects gave an error rate of 5%, withmost errors due to a subject's producing anincorrect number of syllables for one of the longerwords or to a subject's clearly stressing the wrongsyllable in the reiterant production. There wereonly two missing tokens (.8%). The Americansubjects made many more errors in their reiterantversions of the French words. There were twenty­nine such errors (12%) across the eight subjects.Twenty-four of those errors (83% of the tota}),were words ending in "ion" or containing the vowelsequence "ie" as in "societe," which the Frenchcount as a single syllable, but which many of theAmericans counted as two. There was only onemissing token (.4%).

Figure 2 presents the averaged durationalmeasurements of the eight American speakers foreach of the ten word types as a function of theconsonants (1m/) and vowels (la/). For initialstressed syllables;' /m/ averaged 56 ms and /al 92ms, for medial stressed syllables, Im/ averaged 79ms and Ial 108 ms, for final stressed syllables, /m/averaged 82 ms and Ial 255 ms. For unstressed

syllables, Im/ averaged 45 ms and Ial 70 ms ininitial syllables, /m/ was 65 ms and Ial was 76 msin medial syllables, and /m/ was 79 ms and Ial was155 ms in final syllables. The mean duration ofsyllables bearing primary stress5 were 160 ms ininitial position, 190 ms medially, and 336 msfinally. Syllables with secondary stress averaged137 ms initially and 168 medially. Syllables thatwere not stressed averaged 113 ms initially, 138ms medially and 233 ms finally.

Table 2 shows the overall mean length for each.word type, the consonant/vowel ratios for eachsyllable and the ratios of each of the individualsyllables to the length of the word.

Figure 3 shows the mean durational measure­ments for the reiterant versions of the syllables ofeach of the four French word-length types, as pro­duced by the native speakers of English, in termsof consonants (1m/) and vowels (/a/). The mean du­ration of Im/ in nonfinal syllables was 73 ms, of1m!in final syllables was 95 ms, of Ial in nonfinal syl­lables was 85 ms, and of Ial in final syllables was235 ms. Nonfinal syllables thus averaged 157 ms,whereas final syllables averaged 330 ms. The dif­ference in syllable length averaged over 170 msand produced a final/nonfinal ratio of2.1.

The results of a two-way analysis of variancecomparing the subjects' mean nonfinal and finalsyllable lengths for the four word-length typesshowed a highly significant effect of syllable posi­tion [F(1,9)=182.22, p < 0.0000], but no word­length type and no word-length type by syllableposition interaction. Separate analyses comparingsubjects' mean nonfinal syllable lengths for each ofthe four word-length types were also notsignificant.6

Table 3 shows the mean length of each of theword-length types and the ratio of the meanlength of the consonant to that of the vowel ineach syllable. The overall mean CN ratio was .9for nonfinal syllables, which was comparable tothat of the French subjects, but the overall meanCN was .45 for final syllables, which was differentfrom that of the French subjects.

In order to test how well the American subjectsconformed to the French baseline measures fortiming for nonfinal and final syllables in theirreiterant productions of French words, theirtiming measures were subjected to an analysis ofvariance with one between group factor with threelevels (native French versus teachers of Frenchversus English speakers) and two within groupfactors (syllable position [nonfinal versus final]and segment duration [consonant versus vowellength]).

66 Ler1itt

- • V•E .c.E-•Ei=

1s2W 1w2s 112W3w 1w213w 1w~s

Two- Mel thr....yllab.. words400

300-CIlE-.E 200 -

~i=

100 -

o..,. ~.... .1s2w3w+lw lw2s3w4w lw2w3s'lw lw2w3Mw5w lw2w3w+.s5w

Four- and fty.-.yllable words

Figure 2. Consonant and vowel durations, .. a function of word Imgth, syllable poeition, and stress, for niterantimitations of English words spokm in isolation by native speaken of English. (Numbers indicate syllable poeition, Sindicates stressed syllables, and W indicat. unatreMed syllabI. or th.e bearing secondary stress).

Table 2. Mean word lengths (in ms) and CIV and CVlLength ratios in reiterant speech productions ofEnglish wordsby native speakers ofEnglish.

Word Length in Syllables

Two Three Four Five

Stress Type 1 2 3 4 5 6 7 8 9 10Mean Word Length 408.0 457.1 552.4 542.5 624.4 663.0 651.3 7033 825.7 861.5

RatiosCINI .5 .6 .6 .6 .7 .7 .6 .6 .6 .7C7JV2 .5 3 .9 .7 .9 .8 .8 .9 .8 .9C3N3 .5 .6 .4 .9 .8 .7 .8 .8C4N4 .8 .5 .5 .8 .7C5N5 .6 .5

RatiosCVIIL .4 .2 3 .2 .2 .2 .2 .2 .1 .1CV2IL .6 .8 3 .4 .2 .2 .3 .2 .2 .2CV31L .5 .4 .5 .3 .2 3 .2 .2CV4IL .3 .4 3 .2 .2CV51L .3 3

Tokens of types: l=eounter; 2=eontrol; 3=eompliment; 4=concluaion; 5=engineer; 6=c:ommentary; 7=economy; 8=exposition;9=elasticity; l0=c0mmunication.

Reiterant Speech lIS a Test ofNonnative Spetl1cm' Mastery of the Timing ofFrench 67

tnE

400

300

200

100

o

• V• C

1YI2. s 1Yl2.Yt3 s 1Yl2.Yt3w4 s 1Yl2.Yt3w4Y6 sTwo- to five-syllable words

Figure 3. Consonant and vowel duratiON, as a function of word length, syllable p06ition, and stress, for reiterantFrench words spoken in isolation by nOlHlative .peak.... (NumbelS indicate .yllable position, S indicates stressedsyllables, and W indicates unstressed syllables).

Although there was no significant main effect ofgroup, there was a significant effect of syllable po-

Table 3. Mean word lengths (in ms) and CIV andCV/Length ratios in reiterant speech productions ofFrench words by native speakers ofEnglish.

Mean Word Length

Ratios

CINIC'2JV2C3N3C4N4C5N5

Ratios

CVIILCV2ILCV31LCV4ILCV51L

Word Length in Syllables

Two Three Four Five500.5 656.1 786.1 943.6

.7 .8 1.0 1.0

.4 .9 1.0 .9.4 .8 1.0

.5 .9.5

.3 .2 .2 .2

.7 .3 .2 .2.5 .2 .2

.4 .2.2

sition [F(1,17}::417.87, p < 0.0000] and of segmentduration [F(1,17)=121.42, p < 0.0000], and both ofthese effects interacted significantly with thegroup factor [F(2,17)=15.41, p < 0.0003], in thecase of syllable position, and [F(2,17)=8.28, p <.0032], in the case of consonant versus vowellength. There was also a significant two-wayinteraction of syllable position and segmentduration [F(1,17)=145.20, p < 0.0000] that alsointeracted significantly with the group factor[F(2,17)=10.88, p < 0.001]. Figure 4 shows thepattern of results for the three groups.

An exploration of the group interactions withsyllable position and consonant versus vowelrevealed that the source of the interactions wasthe differences in final syllable length among thethree groups, in particular due to differences inthe vowel length, as can be seen in Figure 4. Aseparate analysis of variance conducted on finalsyllable vowel length was significant [F(2,17)=7.65, p < .0044]. Post hoc (Newman-Keuls) testsrevealed that in terms of final vowel length, theproductions of the native speakers of French andthe French teachers did not differ from oneanother but the productions of both groupsdiffered from those of the other native Englishspeakers (p<.05).

68

300 • CJFrench H.Iv..

* VJFrench NatIY..-0- CJEngli.h SubJect-

tv VJEngliM SubJect-

200 .'-0-' 1 CJFrench Tuchers

rn ···ts··1 VJFrench TeachersE

100

0+----""1"""---.......---......---....Nonflnal

SyllableFinal

Figure 4. Mean consonant and vowel length for fiMl and nonfiMlsyllables for French native speakers, experiencedlearners of French (French Teachers), and relatively inexperienced learnen of French (English Subjects).

F. Discussion

The American subjects' productions of theEnglish segment and syllable durations will firstbe discussed, followed by an examination of theways in which their reiterant productions of theFrench words deviate from the French baselinemeasures. Finally the possible effects of Englishtiming patterns on the French productions will beconsidered.

In the English reiterant speech, the nasal mur­mur accounted for 38% of the syllable in stressedinitial syllables, 42% in stressed medial syllablesand 24% in stressed final syllables. For unstressedsyllables the percentages were 39% initially, 45%medially and 34% finally. These percentagesclearly differ from those found in French inExperiment 1, which suggests that the intrasyl­labic timing is not the same in the two languages.

There was also clearly an effect of utterance­final lengthening carried largely by the vowel inthe English data. For stressed syllables,lengthening for final vowels was roughly 150 msand for unstressed syllables it was roughly 75 IDS.

These durationallengthenings are comparable tothose found by Oller (1973).7

Insofar as the syllable measurements areconcerned, the present data showed clenr effectsboth of stress and of utterance-finallel thening.There also appeared to be increments due tosecondary stress, although Nakatani et at foundonly marginal increases in length for suchsyllables and only for some speakers. The ratio of

final/nonfinal syllables was 1.7, which is greaterthan the 1.5 found by Delattre (1966), but whichmay be due to the unusually short initial syllablesfound in this study. Indeed, if initial syllables areeliminated from consideration, the ratio becomes1.6, which is closer to Delattre's measure. Theratio of accented to unaccented syllables was 1.43in initial syllables, 1.38 in medial syllables and1.44 in final syllables. These ratios, which do notinclude the somewhat problematic syllables thatbear secondary stress, correspond fairly well toHoequist's measure of 1.45, although they arelower than the measure given by Delattre (1966)of 1.7. Hoequist's (1983) suggestion that Delattre'shigher ratio is due to the inclusion in theunstressed group of very short I:l! syllables, whichare generally not found in reiterant speech, seemsquite reasonable.

As can be seen in Figure 4, for the reiterantversions of the French words, there was littledifference in the consonant and vowel lengths innonfinal syllables for the three groups. Thus, thepercentage represented by the nasal in nonfinalsyllables was 47% for the native speakers ofFrench, 49% for the American teachers of French,and 44% for the less experienced French speakers.There was also little difference in the mean lengthof 1m! in final syllables for the three groups ofsubjects. The striking difference in the reiterantproductions of the three groups occurs in thelength of utterance-final Ia! which was 171 ms forthe French natives, 199 ms for the Frenchteachers, and 260 ms for the less experienced

Reitenmt Speech lIS fl Test ojN01UIfltitJe SpetUcm' Mtlstery ofthe Timing ofFrmch 69

group. Thus, the nasal consonant accounts for 38%of the final syllable for French natives, 33% forFrench teachers, and only 26% for the les8experienced group. Intrasyllabic timing appears tobe more native-like in nonfinal than in finalsyllables. The ratio of final to nonfinal syllableswas 1.6 for the French natives, 1.9 for the Frenchteachers, and 2.2 for the others. Although thereiterant productions of the American teachers ofFrench were not significantly different from thoseof the French natives, in almost all cases, theteachers' productions, while close to those of theFrench natives, fall between that group and theother group ofnative speakers of English.

Surprisingly, the Americans had a durationalpattern in their reiterant versions of Englishwords that turned out to be very close to theFrench timing pattern. Thus, the averageduration of the first syllable in two syllable wordswith stress on the first syllable (see Figure 2) was173 ms while the final syllable was 236 ms on theaverage, which is comparable to the Frenchnatives' 176 ms average length for nonfinalsyllables and 274 ms average length for finalsyllables. Yet many of the Americans who wereless experienced in French seemed to match thedurational pattern of the final syllable of Frenchwords uttered in isolation (353 ms) by patterningit after the duration of their own stressed syllablesin final position (336 ms) whereas the teachers ofFrench achieved a closer match to the Frenchbaseline measure (296 ms).

Insofar as the nonfinal syllables are concerned,all the Americans showed that they can generallyproduce syllables of quite equal length (see Figure3), and there was no indication in their reiterantversions of French of the systematic initialsyllable shortening that was found with the samesubjects in the English reiterant productions,although some individual subjects continued toshow such a pattern.

Thus, the American teachers ofFrench producedreiterant timing patterns that, while not identicalto those of the native French speakers, did notdiffer significantly from them. On the other hand,the American teachers of French and the Frenchnatives both produced final vowel timing patternsthat were significantly different from those of theother Americans.

G. General DiscussionThere is a growing body of acoustic-phonetic

literature that suggests that the non-nativeproductions of late second language learners areinfluenced, sometimes in subtle ways, by their

native language speech patterns (see Flege, 1986,for a review). Most of the research has focused onthe analysis of the phonetic characteristics ofbilingual speech. Thus the influence of nativelanguage phonetic habits has been demonstratedfor voice onset time (vOT) in stop consonants forEnglishlFrench bilinguals (Flege & Hillenbrand,1984) and for ArabirJEnglish bilinguals (Flege &Port, 1981), because bilinguals show a range ofVOT values whens~ their second languagethat are intermediate between the valuesproduced by monolingual native speakers of thetwo languages. Native language influences havealso been shown for English vowel durations thatdepend on the voicing of the final consonant,because FrencblEnglish bilinguals showed voweldurations, when speaking English, that werecloser to those of French monolinguals (whichvary less with respect to the voicing of a syllable­final consonant) than to those of English-speakingmonolinguals (Mack, 1982).

A similar effect of the rhythmic pattern of thenative language on the acquisition of the rhythmicpatterns of English by native speakers of Frenchhas been found by Wenk (1985) who has describedhis subjects as passing through a transitional"interlanguage- phase, characterized by featuresof both language systems. Intermediate-levelspeakers of French who were learning Englishapparently mastered post-tonic reduced vowels (asin matter) before pre-tonic reduced vowels (as inJapan), when their productions of such words wasjudged by native speakers of English. In thepresent study, native speakers of English whohave studied French appear to master therelatively equal durations of nonfinal syllables inFrench before they master the appropriate Frenchfinal syllable length, because both groups ofAmerican subjects produced essentially equalnonfinal reiterant syllables in French, but only themore experienced group of American subjects, theteachers of French, also produced French-like finalsyllables. Flege (e.g., Flege, 1981; Flege 1987;Flege & Hillenbrand, 1984) has hypothesized thatsecond language learners may acquire inore rapid,accurate pronunciation of a sound that is totallyforeign to their native repertoire, because they areunable to assimilate it to one of their nativephonemes. Equally-timed nonfinal syllables arenot typical. of English words, whereas final­syllable stress does occur. Perhaps nativespeakers of English who learn French are moresuccessful in producing essentially equal nonfinalsyllables in their reiterant versions of French thanin producing the correct final-syllable lengthening,

70 Leuitt

because the former pattern is more foreign to theirnative repertoire.

Many have argued that language learners whobegin their study of a second language relativelylate fail to master fully the phonetic details of thatsecond language because of biological limitationsimposed by a critical or sensitive period for speechacquisition (Lenneberg, 1967; Long, 1990; Oyama,1979; Scovel, 1988). The notion of a critical periodfor language acquisition is a strong one anddescribes a period that is genetically determined,clearly delimited, and not susceptible to theinfluence of the environment. The notion of asensitive period for language acquisition, on theother hand, while still a maturational effect, issubject to greater variability, including a lessclearly delimited time-frame. Although for someresearchers in the field, the onset of adolescence(roughly twelve years of age) was seen as the pointafter which second language learners were likelyto speak their non-native language with a notableforeign accent, others have pushed for acquisitionof a foreign accent to six, at least for someindividuals (see Long, 1990, for a review). Indeed,Long (1990) has written:

Thus, while somewhat weaker than the claim for acritical period for first language learning, the claimfor a sensitive period for second languageacquisition is still a strong and interesting one. Thematurational processes underlying it are held to beuniversal. Hence, learners who begin a secondlanguage after its supposed closure (which will herebe claimed to be as early as age 6 for phonology inmany individuals and around 15 for morphology andsyntax), and who nevertheless attain native-likeability in those areas, will falsify the hypothesis(p.253).

However, all of the native speakers of English inthe present study were late learners of French(beginning in junior high school at the earliest),yet the more experienced group of learners(American teachers of French) produced timingpatterns that were not significantly different fromthose of the native French speakers.

Two possible explanations for this pattern ofresults can be suggested. Either the acquisition ofsecond.language rhythm patterns is exempt fromthe sensitive period constraint or factors such aslength of exposure, training, language aptitude, ormotivation may play an important role. Whereasthere has been little empirical investigation of thefirst hypothesis, the role of experience andtraining has been supported by a number ofstudies. For example, Wenk (1985) found that hisadvanced French students of English, unlike those

at the intermediate level, had mastered the vowelreduction patterns associated with English wordstress. Similarly, Flege and Eefting (1987) foundthat Dutch speakers of English who majored inthe subject were judged to have significantlybetter pronunciation scores than Dutch studentsof English who studied to become engineers,although both groups' productions were judged tobe significantly different from those of nativeEnglish speakers. As in the present study,however, experience may have been confoundedwith aptitude. The English majors, like theuniversity-level teachers of French in the presentstudy, were more experienced second-languagelearners, but they also probably had greateraptitude for second-language learning. In fact,aptitude rather than experience may be the sourceof the performance of the group of Frenchteachers. However, in either case, if good reasonsfor exempting the acquisition of second-languagerhythm patterns from the sensitive periodconstraint are not found, then these results callinto question the notion of a sensitive period ascurrently formulated.

Future research needs to compare directlysecond-language segmental and rhythmiclearning, to see if rhythmic patterns are easier toacquire, and to determine the relativecontribution of rhythmic and phonetic factors tothe detection of non-native pronunciation.

REFERENCESAllen, G. D. (1983). Some suprasegmental contours in French two­

year-<>ld children's speech. Phoneticll, 40, 269-292.Benguerel, A. (1971). Duration of French vowels in unemphatic

stress. LAnguAge iii Speech, 14, 383-391.Qumeck, H. A. (1975). Cross-linguistic investigation of vowel

nasalization: An instrumental study. In Nasalfest: Pllpers frum II

Symposium on Nasals lind Nasaliution (pp. 133-151). Stanford:Stanford University Press.

Crompton, A. (1980). Timing patterns in French. Phoneticll, 37,205-234.

Dauer, R M. (1983). Stress-timing and syllable-timing reanalyzed.Jounuzl ofPhonetics, 11, 51-62.

Delattre, P. (1966). A comparison of syllable length conditioningamong languages. IRAL,4, 183-198.

Di Cristo, A. (1980). La Dur'e intrinHque des voyelles du&an~ais. TT'IlVllll% de l'Institut Phonitique d'Aix, 7,211-35.

Duez, D., &: Nisronuma, Y. (1985). Le rythme en fran~ais:

altemance des dur~ syllabiques. TT'Ilwux de l'Instit"t dePhonitiqued'Aix, 10,151-169.

Plege, J. E. (1981). The phonological basis of foreign accent: Ahypothesis. mOL QuIlrtely, 15,443-455.

Plege, J. E. (1986). The production and perception of foreignlanguage speech sounds. In H. Winitz (Ed.), HlDflllnc:ormrllmiclltion IDUl its disorders Vol. II (pp. 224-4(1). Norwood,N}: Ablex.

Flege, J. E. (1987). The production of 'new' and 'similar' phones ina foreign language: evidence for the effect of equivalenceclassification. JOU17fIIl ofPhonetics, 15,47-65.

Reiterant Speech as a Test o[Nonnative Speakers' MDstery o[the Timing ofFrmch 71

Flege, J. E. (1988). Factors affecting degree of perceived accent inEnglish sentences. JOUTTUIl of the Acoustit'Jll Society ofAmerit'Jl, 84,70-79.

Flege, J. E., &: Eefting, W. (1987). Cross-language switching in stopconwnant perception and production by Dutch speakers ofEnglish. Sp«ch Communication, 6, 185-202.

Flege, J. E., &: Hillenbrand, J. (1984). Umits on phonetic accuracyin foreign language speech production. JOUTTUIl of the Acoustit'JllSociety ofAmerit'Jl, 76, 7C8-21.

Flege, J. E., &: Port, R. (1981) Cross-language phonetic interfer­ence: Arabic to English. LAnguage and Speech, 24, 125-146.

Fletcher, J. (1991). Rhythm and final lengthening in French. JOUTTUIlofPhonetics, 19, 193-212.

Hoequist, C. (1983). Syllable duration in stress-, syllable-, andMora-timed languages. Phonetit'Jl, 40, 203-37.

Klatt, D. (1976). Linguistic uses of segmental duration in English:Acoustic and perceptual evidence. Journal of the Acoustit'JllSociety ofAmerit'Jl, 59, 12C8-1221.

Krashen, S. D., Long, M. A., &: Scarcella, R. C. (1979). Age, rateand eventual attainment in second language acquisition.TESOL QuIIrterly, 13,263-167.

Larkey, L. (1983). Reiterant speech: An acoustic and perceptualvalidation JOUTTUIl of the Acoustit'Jll Society ofAmerit'Jl, 73, 1337­45.

Lenneberg, E. (1967). Biologit'Jllfoundi3tions of/Qnguage. New York:Wiley.

Liberman, M. Y., &: Streeter, 1. A. (1978). Use of nomense-syllablemimicry in the study of prosodic phenomena. Journal of theAroustit'Jll Society ofAmerit'Jl, 63, 231-33.

Lindblom, B. E .F., &: Rapp, K. (1973) Some temporal regularitiesof spoken Swedish. Quarterly Progress Report of the SpeechTmnsmission LAborIltory, (Stockholm, Sweden: Royal Institute ofTechnology) 2, 1-12.

Long, M. (1990). Maturational constraints on language develop­ment. Studies in Second unguage Acquisition, 12, 251-285.

Mack, M. (1982). Voicing-dependent vowel duration in Englishand French: Monolingual and bilingual production. Jounuzl ofthe Aroustit'Jll Society ofAmerit'Jl, 71, 173-178.

Nakatani, 1. H., &: Schaffer, J. A. (1978). Hearing "Words"without words: Prosodic cues for word perception. Jounuzl ofthe Aroustit'Jll Society ofAmerit'Jl, 63, 234-45.

Nakatani, 1. H., O'Connor, K. D., &: Aston, C. H. (1981). Prosodicaspects of American English speech rhythm. Phonetit'Jl, 38, 84­106.

Oller, D. K. (1973). The effect of position in utterance on speechsegment duration in English. Jounuzl of the Aroustit'Jll Society ofAmerit'Jl, 54, 1235-47.

O'Shaughnessy, D. A (1984). Multispealcer analysis of durationsin read French paragraphs. Jounuzl of the Acoustit'Jll Society ofAmerit'Jl, 76, 1664-1672.

Oyama, S. C. (1976). A sensitive period for the acquisition of aphonological system. JOUTTUIl of Psycholinguistic Resetzrch, 5, 261­283.

Parmenter, C. E., &: Blanc, A. V. (1933). An Experimental Study ofAccent in French and English. MLA Publit'Jltions, 48, 598-607.

Pike, K. (1945). Intonation of American English. Ann Arbor:University of Michigan Press.

Scovel, T. (1988). A time to SpetIk: A psycholinguistic inquiry into thecritiad pmodfor IrumiIn speech. Rowley, MA: Newbury House.

Smith, A. (1977). The timing of French, with reflections on syllabletiming. Edinborough Work in Progress, 10,97-107.

Vaissiere, J. (1983). Language-independent prosodic features. InA. CuUer &: D. R. Ladd (Eds.), Prosody: Models and measurements(pp. 53-66). New York: Springer-Verlag.

Vatikiotis-Bateson, E. (1986). DynIlmic anlllysis of reiterant speechproduction. Unpublished doctoral dissertation, IndianaUniversity.

Wenk, B. J. (1985). Speech rhythms and second languageacquisition.LAnguagund Speech, 28,157-175.

Wenk, B. J., &: Wioland, F. (1982). Is French really syllable-timed?JotmVIlofPlwmetics, 10, 193-216.

FOOTNOTES·JOUTTUIl ofthe Awustical SocietyofAmerit'Jl, 90(6),3008-3018 (1991).t Also Wellesley College.1All ratios reported in the paper are to 1.2Results of these analyses of variance were essentially the same,

even when all ten original subjects were included. The onlysignificant effect was that of syllable position [F(1,9)=121.16, p<.0000). None of the other effects were significant.

3In the case of five-syllable words, there were actually four wordsrepresenting one of the five-syllable word types and two wordsrepresenting the other.)

'For comparability with Oller (1973) secondary stress syllableswere grouped with WISU essed syllables.

5The syllables were here divided into those with primary,secondary and no stress for comparability with Nakatani et al.(1981). The two initial syllables of the second set of five syllablewords had complementary stress patterns (one of the words hada secondary stress where the other had no stress and vice versa),so the averaged durations of those syllables were excluded fromthese calculations.

6The results of this analysis and all subsequent analyses includeall of the original subjects from both groups. Similar analysesincluding only the subjects who produced the most consistentreiterant speech produced essentially the same results.

7However, the present data exhibit a consistent effect of initialsyllable shortening (see Figure 2), which disagrees with findingsby Oller (1973), Klatt (1976) and Nakatani et al. (1981). The mostlikely explanation for this discrepancy is that the reiterantproductions in this study were produced as citation forms,rather than in a sentence frame. The present study used citationforms in order to reduce the number of syllables that subjectsneeded to remember for the reiterant production of individualwords (but d. Nakatani et al., 1981 for a different method). Itmay be the case that the sentence frame gives extra prominenceto the word to be imitated and that such prominence results inthe pattern of word-initial syllable length found in the otherstudies.

72 Leuitt

APPENDIX

French Words

Two syllables comptoirsamprogrescontrolesurprisedegre

Three syllables complimentinstrumentsolitudeingemeurindiscretjaponaisconclusioninstructUsolutioncommentairelegendairesociete

Four syllables

television

economiepubliciteexpositionpopulationsatisfaction

Five syllables automatiquemenelasticiteelectricitepossibilit.ecommunicationcivilisation

English Words (Stress Pattern)

counter (SW)sacred (SW)progress (SW)control (WS)surprise (WS)degree (WS)

compliment (SWW)instrument (SWW)solitude (SWW)engineer (WWS)indiscrete (WWS)japanese (WWS)conclusion (WSW)instructive (WSW)solution (WSW)

commentary (SWWW)legendary (SWWW)television (SWWW)society (WSWW)economy (WSWW)publicity (WSWW)exposition (WWSW)population (WWSW)satisfaction (WWSW)

automatically (WWSWW)elasticity (WWSWW)electricity (WWSWW)possibility (WWSWW)communication (WWWSW)civilization (WWWSW)

(S=primary stress, W=secondary stress or no stress)