Chinese subjects'perception of the word-final English/t/-/d/ contrast: Performance before and after training James Emil Flege Departmentof Biocommunication, Unioersity of A!abamaat Birmingham,Uniuersity Station. Birmingham, .4 labarea 35294
(Received8 August 1988;accepted for publication20 July 1989) Chinesewords may beginwith/t/and/d/, but a/t/-/d/contrast doesnot existin word-final position.The questionaddressed by experiment1 waswhether Chinese speakers of English couldidentify the final stopin wordslike beatand bead.The Chinesesubjects examined approachedthe near-perfect identification rates of nativeEnglish adults and children for words that wereunedited, but performedpoorly for wordsfrom whichfinal release bursts had been removed.Removing closure voicing had a smalleffect on the Chinesebut not the English listeners'sensitivity. A regressionanalysis indicated that the Chinesesubjects' native language (Mandarin, Taiwanese,Shanghainese) and their scoreson an Englishcomprehension test accountedfor a significantamount of variancein sensitivityto the (burstless)/t/-/d/ contrast.In experiment2, a small amountof feedbacktraining administeredto Chinese subjectsled to a small, nonsignificantincrease in sensitivityto the English/t/-/d/contrast. In experiment3, more training trials were presentedfor a smallernumber of words.A slightly larger and significanteffect of training wasobtained. The Chinesesubjects who were native speakersof a languagethat permitsobstruents in word-finalposition seemed to benefitmore from the trainingthan those'whosenative language (L1) hasno word-finalobstruents. This wasinterpreted to meanthat syllable-processingstrategies established during L 1 acquisition may influencelater L2 learning. PACS numbers: 43.71.Es, 43.71.Hw
INTRODUCTION aL, 1981;Werker and Tees, 1983, 1984a,b).An apparent consequenceof this lost ability is the difficultyadults may Much secondlanguage (L2) acquisitionresearch has havein comprehendingan L2 later in life (Johansson,1978; beenmotivated by the desireto learnwhether a biologically van Balen, 1980;Oyama, 1982b;Florentine, 1985;Koster, (rather than cognitively)based "critical period" exists for 1987;Mack and Tierney,1987). Suchcomprehension diffi- humanspeech learning. Lenneberg ( 1967,p. 176) claimed culty may derive,at leastin part, from difficultyperceiving that "foreignaccents cannot be overcomeeasily after pu- phonesin the L2. berty."Rating-scale experiments have amply comfirmed the EchoingTrubetzkoy's (1939) observationthat the pho- presenceof foreignaccents in the speechof L2 learners,even nologyof L1 actslike a "sieve"through which L2 •unds are thosewho began learning L2 longbefore the onset of puberty processed,Borden et al. (1983} suggestedthat adult L2 (Asher and Garcia, 1969;Fathman, 1975; Flege and Eeft- learnersare "constrained"to categorizethe phonesof an L2 ing, 1987;Flege, 1988a,b; see also Seliger etal., 1982).1 The "accordingto the phonemecontrasts" of.their native lan- mispronunciationof particularphones, which contributes to guage(p. 500}. This mightexplain why, for example,Japa- foreign accent (see Johansson,1978; F!ege, 1984, 1988a; nesespeakers have difficulty in discriminatingEnglish/r/'s Schneidermanet al., 1988), may derivein part from insuffi- and/1/'s (Miyawaki et aL, 1975). It doesnot explainother cientmotor learning.This, in turn, may be impededby an experimentalevidence so well, however.Flege and Hillen- inability to perceiveL2 phonesor phoneticcontrasts in a brand (1985) found that Swedesand Finns, whoseL 1 hasno nativelikemanner. Difficulty distinguishing phonemes in an /z/, were able to differentiallyidentify the membersof a L2 mayalso impede lexical development (see Leonard et al., continuumranging from peace (/pis/} to peas(piz). They 1982). The purposeof the presentstudy was to provide usedonly vowel duration, however, whereas native speakers further insight into how phonesin an L2 are perceivedby of English used a combination of vowel and fricative dura- examiningthe perception of thevoicing feature in word-final tion cues. Englishobstruents. The presentstudy examinedthe identificationof final Prelinguistieinfants are able to discriminatemany, if /t/ and /d/ by Chinesesubjects. The Ll's of the subjects notall, of the.phonesused contrastively in human languages examinedpossess a contrast between phonologieally voiced (see,e.g., Jusezyk, 1979; Kuhl, 1987), but thisability maybe andvoiceless stops, but only in word-initialposition (Cheng, attenuated near the end of the first year of life for certain 1973;Howie, 1976;Li and Thompson,1981; Shinn, 1985; phoneticcontrasts. Werker and her colleaguesfound that Heyer, 1986). The word-initial phonemiecontrast in Chin- young children lost the ability to discriminatephones not eseis implementedas a phoneticcontrast between voiceless usedcontrastively in the languagespoken around them, even unaspiratedstops with short-lagvoice onsettime (VOT) though they were able to do so earlier as infants (Werker et valuesand voicelessaspirated stops with long-lagVOT val-
1684 J. Acoust.Soc. Am. 86 (5), November1989 0001-4968/89/111684-14500.80 ¸ 1989 AcousticalSociety of America 1684 ues(Lisker and Abramson,1964; Clumeck eta!., 1981). theycontained some of the informationthat cues the voiced- Littleis known about how Chinese subjects perceive fi- voicelesscontrast in word-initialposition. Such a strategy nal stopsin English,but much is knownabout how they could be expectedto work lesswell for stopsfrom which produce word-final stops.As predicted by the contrastlye releasebursts had been removed,however. Alternatively, analysishypotheses (Lado, 1957), Chineselearners of Eng- the Chinesesubjects might regardword-final tokens of/t/ lishare oftenunable to producea perceptuallyeffective con- and/d/as representinga "novel" phoneticcontrast, and trast betweenvoiced and voicelessstops in word-finalposi- begin seekingcues other than those used for word-initial tion. Transcriptionaldata from severalstudies have shown stops,such as precedingvowel duration. If so, the Chinese that Chinesespeakers devoice English/b,d,g/, deletevoiced subjectsmight perform more poorly than the native English and voicelessstops, and/or add vowelsfollowing word-final subjectsfor the tokenswith releasebursts, but the effectof stops(Tarone, 1980;Eckman, 1981; Anderson, 1983, 1987; removingrelease burst (and closure voicing) cues should be F!egeand Davidian, 1985;Heyer, 1986;Weinberger, 1987; no greaterfor them than for native Englishsubjects. Flegeet al., 1987). Detailed analysesof the productionof The resultsof experiment1 supportedthe first alterna- /p/ and/b/ in word-finalposition have shown that Chinese tive, so two feedbacktraining experiments were carried out subjectsproduce a much smaller(albeit significant)dura- to betterenable the Chinese subjects to perceivethose cues to tion differencebetween vowels preceding/p/vs/b/, are the/t/-/d/contrast that remainedafter the closurevoicing significantlyless likely than nativespeakers to activelyen- and releaseburst cues had beenremoved. The first training largethe oral cavityto sustainvoicing in/b/, and do not experimentinvolved subjects who had participated in exper- produce/p/with a greaterforce ofbilabial constriction than iment1. A greateramount of feedbacktraining was present- /b/(Flege eta!., 1987;Flege, 1988e). edin thesecond training experiment to a newgroup of Chin- Thesefindings show that the ability to producevoiced esesubjects. A "neural atrophy" hypothesis(see Werker andvoiceless stops in word-initialposition does not guaran- and Tees, 1984b) impliesthat it would be difficultto in- teethe ability to producean effectivevoiced-voiceless con- creaseChinese subjects' sensitivity to the/t/-/d/contrast trast in the word-final position.This suggeststhat speech throughshort-term laboratory training. The notionthat au- productionis not basedon freely cornmutablephoneroes ditory sensitivityis neverreally lost,but is simply"difficult { seeFlege and Port, 1981) and that speechproduction skills to apply"to novelphonetic contrasts as the resultof a "lan- must be learnedon an allophone-by-allophonebasis (see guagebased reorganization" of contrastlyesound units also Bri•re, 1966). This inferenceis consistentwith the ob- (Werker, 1989) suggestsa differentoutcome, however. servationthat soundsmay be morevulnerable to disruption Trainingmight be effective if it succeedsin enablingsubjects in initial than final positionfollowing injury to the brain topay attention to relevantacoustic cues ( Strange,1986), or (Kent and Rosenbek1983; Kelso and Tuller, 1981). It is if it helpssubjects engage a phonetic,as opposed to a phone- alsoconsistent with the conclusion,based on performance mic,mode of processing(Werker and Tees, 1984b}. errors in normal individuals, that word-initial consonants areprocessed separately from subsequent consonants (Shat- I. EXPERIMENT 1 tuck-Hufnagel,1987). A. Introduction The questionaddressed by experimentI was whether speechperception skills also need to be learnedon an allo- The closureintervals of Englishfo,d,g/and/p,t,k/are phone-by-allophonebasis. To determinethis, nativeand distinguishedby thepresence versus absence of voicingdur- Chinesesubjects were asked to identifythe finalstop in une- ingoral constrictionin isolated,clearly spoken words, but as ditedmultiple natural tokens of Englishwords such as beat a rule voicingdoes not extendthrough the entireconstric- and bead,and in copiesfrom whichthe final releasebursts tion of/b,d/g/. As discussedlater (seealso the Appendix), and/or closurevoicing had been removed. In conversational the removalof closurevoicing may or may not lead to an speech,English/t/and/d/are frequentlyproduced with- increasein voicelessstop responses by nativeEnglish-speak- out audible release bursts, and/d/is "devoiced" (that is, inglisteners. English word-final stops are oftennot released producedwithout closure voicing through some part or all of audibly(W. ardrip-Fruin, 1982), so it is hardlysurprising the closureinterval; see, e.g., Smith, 1979). Perhapsbecause that removingrelease bursts may havelittle perceptualef- of the unreliabilityof thesecues, native speakers of English fect. 3 can identifystops without closure voicing and/or final re- Previousresearch has shownthat native speakersof leasebursts nearly as well as stopswith all acousticcues Englishcan identify the voicingfeature of word-finaltokens available (see below}. of/b,d,g/and/p,t,k/fairly well evenwhen both closure Two alternativepossibilities existed concerning how the voicingand release bursts have been removed. They do so on Chinesesubjects would perform in the identificationtask. the basisof temporaland spectralcues residing in the pre- Even thoughthe acousticcues to the/t/-/d/contrast differ cedingvowel and in thefinal consonant transitions (Halle et in the initial versusfinal positionof Englishwords (Strange, al., 1957;Ohde and Shaft, 1977;Revoile et al., 1982;Hillen- 1986), the Chinese subjectsmight attempt to identify the brand et al., 1984). For example, Wolf (1978) found that final/t/'s and/d/'s usingthe perceptualcues that distin- /t/'s and/d/'s were identifiedcorrectly nearly as often in guish/t/from/d/in theinitial position of Chinesewords. 2 thefinal position of wordsfrom which release burst and clo- Sucha strategywould probablywork well for the unedited surevoicing cues had beenremoved as in uneditedwords words in which the final release bursts were intact, because (89% vs100% correct).4 The aim of experiment I was to de-
1685 J. Acoust.Soc. Am., Vol. 86, No.5, November1989 JamesEmil Flege: Chinese perception of word-final/t/-/d/ 1685 terminehow well nativeand Chinesesubjects could identify native speakersof Mandarin from mainland China, others /t/'s and/d/'s in the final positionof Englishwords that (N = 5) werenative speakers of Taiwanese,and the remain- wereunedited and in copiesfrom which closure voicing and/ ing subjectswere individualsfrom southernChina who or release burst cues had been removed. spokeeither Shanghainese(N= 8) or Hakka (N= 1). A secondaryaim wasto examinethe effectof removing None of the ANOVAs testingfor differencesbetween the acousticcues for native English-speakingchildren. Speech subgroupsin chronologicalage, age of arrival in the U.S., productionresearch has shown that Chineseadults are less lengthof residencein the U.S., percentagedaily useof Eng- skillful in producinga contrastbetween word-final/p/and lish, or comprehensionability reachedsignificance at the /b/than nativeEnglish children who, in turn, are lessskill- 0.10 level. ful in doingso than native English adults (Flege et al., 1987; Flege,1988c). Parnell and Amerman(1978) foundthat 4 2. Stimuli yearolds identified the place of articulationof consonantsin An adult native speakerof American English (the CV syllablesless often than 11 yearolds or adultswhen the author) read real wordsformed by insertingseven vowels steady-stateportion of thevowel was removed (76% vs98% (/i,•,e t ,e,a•,a,u/)into/b-t/and/b-d/frames in thecarrier for the two older groups). It appearsthat both children phrase"Now I will say --." The first four tokensof each (Cole and Perfetti, 1980) and nonnativeadults may require longerstretches of speechthan mature native speakersto recognizewords (Truin, 1981, reportedby Kester, 1987). If speechperception skills undergoa gradual attune- TABLE [. Characteristicsof the Chinesesubjects in three native-language subgroupswho participated in experimentI. POBis placeof birth.AOA is ment to the multiple coarticulatedcues used to signalpho- ageof arrivalin the U.S., MOS indicatesmonlhs lived in the U.S., COMP is netic contrasts(Parnell and Amerman, 1978; Eguchi and the scoreon an Englishcomprehension test, and USE is self-estirrmtedper- Hirsch, 1969;Flege and Eefting, 1986), then removingclo- centagedaily useof English. surevoicing and/or releaseburst cues may affectnative Eng- lish childrenand the Chineseadults similarly. Alternatively, POB Sex AOA MOS COMP USE
the childrenmight resemble native English adults in show- Mandarin ing little effectof editing.This couldbe expectedif the chil- Beijing M 36 19 74 80 dren basedtheir voicingjudgments on the vowel duration Beijing M 35 8 56 20 cuesthat distinguishedthe edited words endingin/t/and Beijing M 32 12 85 50 Beijing F ! 7 84 91 --- /d/(see Krause, 1982a,b). Xian M 30 7 72 50 Xian F 28 28 76 60 B. Methods Tientsin M 29 5 72 60 Lanzou M 26 23 70 20 1. Subjects Harbin F 27 16 87 80 Three groupsof listenersparticipated as paid subjects, Changchun M 35 4 83 80 Shenyang M 32 8 56 35 all of whom wereable to detectfour pure tonesin the 500- Chengdu M 27 I 83 60 4000-kHz rangeat 20 dB HL. The nativeEnglish adult sub- Nanchang M 26 I 76 20 jects (N= 8) and the nine Englishchildren (N= 9) were Taipei, ROC F 25 132 91 85 Taoyuan, ROC M 28 5 78 20 monolingualspeakers of English from Birmingham, Ala- Kaohsiung, ROC F 22 36 83 80 bama, who were recruited through an advertisementin the Universitynewspaper. The adultswere in their late twenties M 28 24 77 49 and early thirties;the meanage of the childrenwas 9.2 years. (5) (35) (11) (25) Most of the Chinesesubjects (19 males, 11 females) Taiwanese weregraduate students or facultymembers at the University Taipei,ROC F 29 32 74 60 of Alabamaat Birminghamwho spoke English with foreign Taipei, ROC F 24 3 79 40 Taichung,ROC M 26 4 70 20 accents(see Flege, 1988b). As summarizedin Table I, the Taichung,ROC F 31 27 46 20 Chinesesubjects had a meanage of about 30 yearsand had Kaoshung,ROC M 27 10 81 20 livedin theU.S. for a little more than2 yearson average.The M 27 15 70 32 Chinesesubjects reported that they spokeEnglish about half (3) (13) (14) (18) the time,but someof themseemed to havedifficulty compre- hending English. This suggestedthat their estimated daily $hanghainese Shanghai M 27 18 80 40 useof Englishmight have been too high (seeGras, 1983). To Shanghai F 25 2 53 20 test this, each subjectwas administeredthe listeningcom- Shanghai M 24 72 87 95 prehensionsubtest of the Michigan English LanguageAs- Shanghai M 35 47 79 20 sessmentBattery (English Language Institute, University of Shanghai M 27 52 70 60 Shanghai M 29 4 72 50 Michigan, 1986). The Chinesesubjects received a 29th per- Shanghai M 42 27 78 20 centilescore, which confirmed the inferenceconderning rel- Shanghai F 25 36 91 50 ativelylimited native-speaker inputfi Pingtong,ROC F 25 25 78 50
For the purposeof regressionanalysis (see below), the M 29 31 77 45 Chinesesubjects were subdivided into threesubgroups based (6) (23) (12) (24) on nativelanguage. Of the 30 subjects,some (N = 16) were
1686 J. Acoust.Sec. Am.,Vol. 86, No. 5, November1989 JamesEmil Flege: Chinese perception of word-final/t/-/d/ 1686 word werelow-pass filtered at 4 kHz, digitizedat 10 kHz identifiedstops in the burst- and V q- B-removedwords cor- with 12-bit resolution, then normalized for overall rms am- rectly at nearly the samerates (99%, 98%) as the unedited plitude.The initial/b/'s of the 56 testwords were all pre- and voicing-removedwords (100%, 99%). Similarly, the voiced.Each final stophad an audiblerelease burst, and the native English children identifiedthe burst- and V q- B-re- final/d/'s all had at leastsome voicing. As shownin Table movedwords correctly at aboutthe samerates (96%, 96% ) II, the vowelsaveraged 133 ms (59%) longerbefore/d/ as the unedited and voicing-removedwords (97% and than/t/. 98% ). The Chineseadults, on the other hand, identifiedthe Copiesof the "unedited"words were modifiedas fol- burst- and V + B-removed words far lesswell (64% correct lows.A set of "voicing-removed"words was madeby at- in both instances)than the uneditedand voicing-removed tenuatingto zeroany energypresent in the closureinterval words (94%, 93%). of the finalstops. Stop closure duration information was pre- An unbiasedestimate of perceptualsensitivity to the servedin thesestimuli. 6 "Burst-removed" words were made /t/-/d/contrast wasdetermined by calculatingA ' scoresfor by removingall energyin the releasebursts. Both closure eachminimal pair (Grier, 1971). The A ' scoreswere submit- voicingand releasebursts were removedfrom the "V q- B- ted to a group (English adults, Englishchildren, Chinese removed"words by zeroingany energypresent after the con- adults) X vowel ( sevenlevels ) X editingcondition ( four lev- strictionof the final stops. els) ANOVA with repeatedmeasures on the last two fac- tors.• As shownin Fig. 1, theA ' scoreswere only slightly 3. Procedures lower in the uneditedand voicing-removedconditions for The wordswere presented binaurally via TDH-49 head- the Chineseadults than for the Englishadults or children. phonesat a comfortablelevel (74-dB peaksyllable intensi- However, in the burst- and V + B-removed conditions, the ty). The subjectswere told to push a button marked "t" or Chinesesubjects' scores were substantiallylower than the "d," and to guessif uncertain.Each new word waspresented nativespeakers', which resultedin a significantgroup X con- 1.0 s after a responsewas receivedfor the precedingword. dition interaction[F(6,132) = 19.4,p <0.05]. Three randomizations of the 56 words (seven vowels X two The simplemain effectof editingcondition was nonsig- final stopsXfour replicate tokens) were presentedin the nificant for the Englishadults and alsofor the Englishchil- four editing conditions(unedited, voicing-removed,burst dren [F(3,21) = 0.96, F(3,24) = 2.77] but highly signifi- removed, V d- B-removed). The order of the four conditions cantfor the Chineseadults [F(3,87) = 75.5,p < 0.05]. Post was only partiallycounterbalanced. Subjects always heard hoc tests (Newman-Keuls, a =0.05) revealed that the the unedited words first and the V q- B-removed words last. Chinesesubjects' A' scoreswere lower in the burst- and This orderingwas used because pilot experimentssuggested V + B-removedconditions than in theunedited and voicing- that the uneditedwords were relativelyeasy for the Chinese removedconditions. The simplemain effectof group was subjectsto identify,whereas the V q- B-removedwords were significantfor stopsin the voicing-,burst-, and V + B-re- difficult. The voicing-removed and burst-removed words movedconditions [F(2,44) = 4.73, 28.0, 21.9,p < 0.05] and were presentedin counterbalancedorder as the secondand narrowly missedreaching significance in the uneditedcondi- third conditions,which affordedthe opportunityto deter- tion [F(2,44) = 3.16, p = 0.052]. Post hoc tests revealed mine the relativeimportance of closurevoicing and release that, for stopsin the voicing-removedcondition, the English burst cues. adults'but not the Englishchildren's scores were significant- ly greaterthan the Chineseadults' scores. In the burst-and C. Results V + B-removedconditions, both the English adults' and L Effects of editing children's,4 ' scoreswere significantly higher than the Chin- eseadults' scores.The Englishadults and childrendid not The mean percentageof correct identificationsof the differsignificantly in anycondition? final stopin the 14 testwords was calculated for eachlisten- er.7 As shownin Table III, the overall ratesfor/d/and/t/ œ.Predictors of sensitivity differed little (85% vs 86% correct). The English adults SomeChinese subjects succeeded better than othersin identifyingstops without release bursts. For example,the TABLE II. The meanduration of vowelsin the English/bVt/and/bVd/ correct identificationrate for stopsin the burst-removed wordsused in experimentl, in ms. Each meanis basedon four tokens; conditionranged from 35%-96%. Two forwardstepwise standarddeviations are in parentheses. multipleregression analyses were carriedout to help ac-
Word final/t/ Word final/d/ countfor intersubjectvariability among the 30 Chinesesub- jects. beat 198 (12) bead 335 (28) a. Methods.The criterionvariable in the regressionanal- bit 178 (19) bid 304 (10) ysiswas sensitivity to the/t/-/d/contrast in the burst-and bait 236 (8) bade 364 (13) bet 201 (8) bed 327 (15) V + B-removedconditions, as measuredby d' (rather than bat 264 (21) bad 410 (9) A ' ) scores.t• The resultsof previousresearch suggested five boot 220 (13) booed 353 (15) of the sixpredictor variables examined. It isknown that abi- bought 261 (15) baud 396 (9) lity to comprehendL2 sentencesmay be correlatedpositive- M 223 M 356 ly with the ageof learning(Oyama, 1982b),length of resi- dencein an L2-speakingcountry (Gras, 1983;Underbakke
1687 J. Acoust.Soc. Am., Vol. 86, No. 5, November1989 JamesEmil Flege: Chinese perception of word-final/t/-/d/ 1687 TABLE IlL Meanpercent correct identifications of word-final stops by native English adults (EA), nativeEnglish children (EC), andChinese adults (CA) in Englishwords that were unedited (UNED), hadclosure voicing removed (RVOI), hadfinal release bursts removed (RBUR), or hadboth voicing and burstsremoved (RV + B ). The meansfor groupEA arebased on eight subjects X sevenminimal pairs = 56 scores.Means for groupEC arebased on data for 9 subjects,those for CA on 30 subjects.Standard deviations are in parentheses.
/d/ /t/ UNED RVOI RBUR RV + B UNED RVOI RBUR RV + B M
EA 99.6 98.4 99.1 97.2 99.6 99.6 99.6 99.4 99.0 (1.9) (5.1) (3.0) (8.0) (1.9) (1.9) (1.9) (2.2) (3.9) EC 9g.0 97.8 96.5 95.6 96.8 97.9 95.4 96.8 96.9 (3.9) (4.8) (5.6) (7.8) (6.8) (4.7) (7.4) (4.8) (5.9) CA 91.5 88. ! 70.0 69.0 96.5 97.9 58.5 58.7 78.5 (15.6) (17.0) (26.6) (28.7) (7.9) (5.8) (27.8) (29.8) (26.7)
M 94.1 91.7 80.3 78.9 97.1 98.2 72.5 72.9 (13.0) (14.7) (25.2) (26.8) (7.1) (5.2) (29.3) (30.5)
et aL, 1988), overallexperience with and proficiencyin the In the analysesof stopsin the V + B-removedcondi- L2 (van Balen, 1980; Mack and Tierney, 1987), and the tion, LI backgroundand the Englishcomprehension scores extentto which Englishis usedboth activelyand passively wereagain identified as significant predictors of thed' scores on a daily basis(Gras, 1983;Underbakke et al., 1988;see [F(2,26) = 3.66, p<0.05], but the two-factormodel ac- alsoMacKain et al., 1981). Sincesentence comprehension countedfor only22% of thevariance in d' scores.The simple may be relatedto the ability to perceivespecific phonetic correlationbetween the d' scoresand the Englishcompre- contrasts,the followingpredictor variables were examined: hensionscores was significant[r(28)= 0.372, p<0.05], chronologicalage, age of arrivalin the U.S., lengthof resi- but not the correlation between the d' scores and L1 back- dencein the U.S., percentageof daily useof English,and ground [r(28) = 0.287]. The correlationbetween the d' scoreson the Englishcomprehension test. L1 background scoresand the comprehensionscores remained significant servedas the sixth predictor variablebecause preliminary whenvariation due to L1 waspartialed out [r(28) = 0.393, analysesof the correctidentification scores revealed signifi- p < 0.05]. All subjectssubjects participated in the burst-re- cant differencesbetween the subjectswho spokeShanghain- moved condition before the V + B-removed condition, ese,Taiwanese, and Mandarin (seeTable I)." which may haveleveled differences between subjects in the b. Result• Analysisof data from the burst-removedcon- threeChinese subgroups in the V + B-removedcondition. dition indicatedthat L1 backgroundand ability to compre- hendEnglish were independent predictors of sensitivityto the English/t/-/d/contrast. A modelwtih thesetwo fac- D. Discussion tors accounted for 31% of variance [F(2,26)=5.79, p < 0.05]. The d' scorescorrelated significantly with the L Perceptual sensitivity comprehensiontest scores, even when the L1 factorwas par- As expected,native Englishadults identifiedtokens of tialedout [ r(28) = 0.441,p < 0.05]. The d' scorescorrelat- final/t/and/d/(from which closurevoicing and/or final ed significantlywith the L 1 factor, evenwhen the compre- releasedburst cueswere edited out) at high rates.Taken hensionscores were partialed out [ r(28) = 0.420,p < 0.05]. togetherwith the resultof previousreseaxeh, this indicated that acoustic information in the vowel and consonant transi- tionsof EnglishCVC words,including the spectralquality • EA I EC • CA and durationof the precedingvowel and F• offsetfrequency, 1.0 1.0 are sufficientto cueaccurate voicing judgments. The nativeEnglish children who were examinedper- 0.8 0.8 formed much like the English adults,but this doesnot mean
0.6 0.6 necessarilythat their perceptionwas identical to that of adults (seeFlege and Eefting, 1986,for word-initialstops). 0.4 0.4- Bernstein Rather and Lubereft (1984) reported that mothersspeaking to their childrenwere perceivedto have 0.2 0.2 deleted40% of/t/'s and/d/'s (seealso Shoekey and Bond,
0.0 0.0 1980). These samemothers exaggerated vowel duration UNED RVOI RBUR RVB cues,so it is hardly surprisingthat other research(e.g., Krause, 1982a,b)has shown that vowelduration may pro- FIG. 1.Mean sensitivity(A') to theword-final English/t/-/d/contrast by vide a sufficientperceptual cue to the distinctionbetween nativeEnglish adults (EA), nativeEnglish children (EC), and Chinese voiced and voicelessstops for young children. Thus any adults(CA) in experiment1. The stimuliwere unedited (UNED), had closure voicing removed (RVOI), the final release bursts removed adult-child differencethat did existmay havebeen obscured (RBUR}, or had both closure energy and release bursts removed by a ceilingeffect brought about by the useof vowelduration (RV + B). The bracketsenclose + 1 standarderror. cues.The childrenmay also have used F• offsetfrequency as
1688 J. Acoust.Sec. Am., Vol. 86, No. 5, November1989 James Emil Flege: Chineseperception of word-final/t/-/d/ 1688 a perceptualcue (seeSimon and Fourcin, 1978,for word- word-finalEnglish stops because their L I permitsan ob- initialstops). struent(a glottalstop) in word-finalposition. One might Onemight infer that the Chinese subjects were unable to speculatethat this caused them to focusattention on phonet- makethe same use of vowelduration as the native English ic informationfound at theend of syllablesto a greaterex- childrenbecause they performed more poorly than the chil- tentthan the Mandarin subjects, whose L 1 permitsno final dren when releaseburst cueswere removed. This inference is obstruents.However, there is indirect evidencethat Manda- consistentwith the observationthat Chinesespeakers of rin speakersof Englishmay paygreater attention to release Englishproduce only a relativelysmall (albeit significant) burstsin word-finalEnglish stops than Shanghainesesub- vowelduration difference between English words ending in jects.In examiningthe Englishspeech production of Chin- /p/ and /b/ (Flege, 1988c). ese subjects,Heyer (1986) found that Mandarin subjects The removalof closurevoicing had little effecton the "hyperaspirated"English/p,t,k/more frequentlythan na- nativeEnglish subjects, whereas it significantlylessened the tive speakersof southernChinese dialects (46% vs 13% of Chinesesubjects' sensitivity to theEnglish/t/-/d/contrast errorsobserved). •2 (it lowered the correct identificationscores from 95% in the The relationshipbetween the comprehension test scores uneditedcondition to 93% in the voicing-removedcondi- andsensitivity to the/t/-/d/contrast suggests,not surpris- tion). Thisdoes not meannecessarily that theChinese sub- ingly,that the ability to identifyphonetic segments isimpor- jectswere more sensitive to closurevoicing than the English tant to comprehension.Variations in the comprehension adults.The lack of a decreasein sensitivityfor theEnglish scoresmay havebeen related to other factorssuch as amount adultswas probably the resultof a ceilingeffect resulting of English-languageinput, but neitherof the questionnaire from the use of other cues such as vowel duration. A short variablesrelating directly to amountof English-languageex- experimentpresented in the Appendixshowed that, in a perience(length of residencein theU.S., percentdaily use of paired-comparisontask, native English adults significantly English)was a significantpredictor of perceptualsensitivity preferredfinal/d/tokens with closurevoicing over those to Englishstops. withoutclosure voicing. It mustbe recognized, of course,that comprehending an Removingrelease burst had a muchgreater effect on the L2 dependson more than just bottom-up segmental process- Chinesesubjects than removing closure voicing. Their rate ing (van Balen, 1980;Koster, 1987). Moreover,some of the of correctidentifications plummeted to 64% correctin the intersubjectvariability not accountedfor by the regression two conditions in which the release bursts were removed. analysismay have derived from differencesin overall listen- Thissuggests that theChinese subjects were identifying the ing skill, whichwas not assessed.A recentstudy examining word-final/t/'s and/d/'s on the basis of acoustic cues that word recognitionrevealed a greatdeal of variabilityin sub- distinguishstops in word-initialposition, where a contrast jects' ability to perceivespeech in noiseand to exploit con- betweena voicelessunaspirated /d/ anda voicelessaspirat- textual constraints(Boothroyd and Nittrouer, 1988). In a ed/t/exists in the subject'sLl's (Lisker and Abramson, study examiningDutch high schoolstudents, van Balen 1964;Clumeck et al., 1981;Shinn, 1985). (1980) notedthat studentswith high scoreson an English An alternativeexplanation is that theeffect of altering listeningproficiency test also tendedto receiverelatively cuesmight simply be greater for listenerswhose speech per- highscores on a comparableDutch listeningtest. More re- ceptionskills are not maturethan for maturenative speak- search will be needed to establish the basis for individual ers.Previous research has shown that nonnativespeakers perceptualdifferences in both L1 and L2 speechprocessing. recognizewords more poorly than native speakers, especial- ly in non-ideallistening conditions (Johansson, 1978; van II. EXPERIMENT 2 Balen,1980; Oyama, 1982b;Florentine, 1985; Koster, 1987; Mack and Tierney, 1987). Like children,nonnatives may The near-perfectperformance of the nativeEnglish sub- requirelonger stretches of speechthan mature native speak- jects in experiment 1 showedthat sufficientacoustic cues ers to recognizewords (Truin, 1981, reportedby Koster, remainedin the V + B-removedstimuli to supportaccurate 1987). This explanationis probablynot correct,however, voicingjudgments. The aim of this trainingexperiment was becauseremoving the closure voicing and release burst cues to enablethe Chinesesubjects to makebetter use of acoustic did notsignificantly lower the nativeEnglish children's sen- cuesin the V + B-removedstimuli. The reasonfor focusing sitivity to the/t/-/d/contrast. on the V + B-removedstimuli is that phonologicallyvoiced stopssuch as/d/are oftenproduced partially or completely without closurevoicing in conversationalspeech, and both 2. Factors predicting sensitivity /d/ and /t/ often do not havean audiblerelease burst (e.g., The multiple regressionanalyses indicated that L1 Smith, 1979; BernsteinRatner and Luberoff, 1984). backgroundand L2 comprehensionability weresignificant It was uncertainwhether the feedbacktraining would predictorsof sensitivityto theEnglish/t/-/d/contrast. The improvesensitivity to the English/t/-/d/contrast. If pho- nativespeakers of Shanghainesewere somewhat more sensi- netic contrastsare not learned (or maintained) in early tive to the/t/-/d/contrast than the native Mandarin sub- childhood, they may be difficult to learn (or train) later in jects;and the betterthe Chinesecomprehended English, the life. There is evidence,however, that the perceptionof at more sensitivethey wereto the/t/-/d/contrast. least some normatire phoneticcontrasts can be improved The basisfor the L 1 effectis uncertain.The Shanghain- through appropriatestimulation. MacKain et al. (1981) ese subjectsmay have been relatively good at identifying found that relativelyinexperienced Japanese subjects per-
1689 J. Acoust.Sec. Am., Vol. 86, No. 5, November 1989 James Emil Flege: Chinese perceptionof word-final/t/-/d/ 1689 formedat near-chancelevels in identifyingand discriminat- The presentexperiment used methods similar to those ingsynthetic/r/'s and/l/'s, whereasmore experienced Jap- of Jamiesonand Morosan ( 1986;see also Logan eta!.,1989), anesesubjects (28 monthsresidence in the U.S., 55% daily who employedidentification with feedbackto train French useof English)identified and discriminated/r/'sand/1/'s subjectson the word-initialEnglish/0 / vs/•J/contrast. The in a nativelike fashion. 56 V + B-removedwords (sevenvowels X two final stops- The successof a trainingexperiment, or the rapidityof X four repetitions)from experiment1 were presentedto learningin naturalisticacquisition, may dependto someex- Chinesesubjects who had participatedpreviously in that ex- tent on the acousticnature of the L2 contrastbeing learned/ periment.Seven subjects each were randomlyassigned to trained. Certain normatire contrasts,such as thosebetween experimentaland control groups.All 14 participatedin Zulu clicks,may require little or no trainingbecause they are three blocks,in which they were told to identify the final salientfor auditoryreasons alone (Best et aL, 1988;see also stopsas "t" or "d." Three randomizationsof the 56 stimuli Burnham,1986; Polka, 1987). Othersmay prove to be more werepresented without feedback to both groupsin the first difficult. Werker and her colleagues(Werker and Tees, and third blocks (designated"pre-training" and "post- 1984a,b;Werker and Logan, 1985) found that English training"blocks). Four randomizationsof the stimuliwere adults could not discriminatea relatively rare place of ar- presentedin the secondblock, in which feedbackwas pre- ticulationcontrast between uvular and velar stopsin a cate- sentedto the experimentalbut not the control subjects. gory-changetask, although they could easily discriminate the releaseburst portions of suchstops, and coulddiscrimi- B. Results and discussion natefull stops,when a moresensitive testing procedure was Table IV presentsthe rate at which the/t/and/d/ used.In otherresearch, Werker showedthat Englishadults tokens were identified correctly. The mean identification learned to discriminate a Hindi contrast between voiceless ratesin the pre-trainingblock resembledthose obtained in aspiratedand breathy voiceddental stopsmore readily than experiment1, which showsthat the Chinesesubjects' diffi- a Hindi place of articulation contrastbetween dental and culty with the V + B-removedstimuli persisted.The experi- retroflexstops (Werker etaL, 1981;Werker and Tees, 1983; mental subjects'overall identification rate was somewhat Tees and Werker, 1984). higher after than before training (74% vs 65% correct), This last finding suggeststhe possibilitythat voicing whereasthere was little differencefor the control subjects contrastsare relatively "robust" becauseof a "strong psy- (59% vs 60%). choacousticbasis" (Burnham, 1986,p. 222) and thuswill be The 9% increaseobserved here for the experimental easyto train. However,the Englishsubjects' success on the subjectsis comparableto the 11% increasein correctidenti- Hindi voicingcontrast may havebeen due to previouspho- ficati6nsobtained by Jamiesonand Morosan(1986) for neticexperience. Even though the primary contrast between Frenchlearners of English.As in that study,subjects in the English/b,d,g/and/p,t,k/is not betweenprevoiced and presentstudy continuedto identify the voicing feature in long-lagstops, native speakersof Englishare nevertheless obstruents at rates that were far lower than those of native exposedto prevoicedrealizations of/b,d,g/, whichmay have Englishspeakers. (Recall that, in experiment1, nativeEng- helpedthe Werker et al. ( 1981)subjects to perceivethe Hin- lish adults identified the V + B-removed stimuli in 99% of di voicingcontrast. instances.)Feedback training resultedin fairly large in- In recentyears, the technologyused in speechpercep- creasesin percentcorrect identifications for two experimen- tion researchhas beenadapted to train nonnativephonetic tal subjects,but the increasesnoted for most of the other contrasts.A numberof trainingstudies have focused on the subjectswere quite small. English/r/-/1/contrast (Cochrane,1977; Gillette, 1980; Table IV alsopresents the A' scorescomputed to esti- Bordeneta!., 1983;Strange and Dittmann, 1984;Logan et matelisteners' sensitivity to the/t/-/d/contrast. The •t' ' al., 1989). The trainingtechniques and the amountof train- valueswere calculatedbased on the averageproportion of ingadministered have varied widely? Eachstudy has re- sultedin someimprovement in discriminationand/or iden- tification, although performancehas generallyremained TABLE IV. Mean percentageof correctidentifications of/t/and/d/be- belownative-speaker levels. Other studieshave focused on fore,during, and after training. The Chinesesubjects in the experimental word-initialstop consonants.Most but not all of the stop groupreceived feedback training in thesecond ofthrec blocks (FB), where- studieshave met with success,and several provided evidence asthose in thecontrol group did not.The `4' scoresrepresent an unbiased measureof sensitivityto the/t/-/d/contrast in sevenminimal pairs. Stan- for transferof training(Lisker, 1970;Kalikow and Swets, dard deviationsare in parentheses. 1972; Strange and Jenkins, 1978; Pisoni et al., 1982, McClasky et al., 1983). No previousstudy has focusedon Before FB After the voiced-voicelesscontrast in word-finalstops, however. Experimentalgroup /d/ 70.4% (24) 74.8% (25) 77.4% (23) A. Methods /t/ 60.0% (26) 62.8% (30) 71.6% (32) .4' 0.716 (0.126) 0.751 (0.151) 0.802 (0.161) Some previousstudies have used discriminationtrain- ing. One drawbackof this techniqueis that it may highlight Controlgroup /d/ 67.3% (29) 58.4% (33) 63.2% (32) differencesat the boundariesbetween the phoneticcatego- /t/ 52.8% (28) 59.7% (31) 54.5% (34) ries being trained rather than helping to define category `4' 0.658 (0.125) 0.636 (0.150) 0.628 (0.163) centers(Repp, 1984).
1690 d. Acoust.Soc. Am., VoL86, No. 5, November1989 James EmilFlege: Chinese perception of word-finalItl-ldl 1690 /d/ responsesgiven by eachsubject to theseven words end- in experiments1and 2. Onequestion addressed was whether ing in/d/and the sevenwords ending in/t/in the three improvementsin identifyingthe final stopsof thesewords blocks.The.4' scoreswere submitted to a group(experimen- wouldgeneralize to tokensof otherwords (bit, bid,bought, tal, control)X block(pre-training, feedback, post-training) baud)that had not been trained. If feedbacktraining affects ANOVA, which yieldeda significanttwo-way interaction only the phonological/phoneticspecification of individual [F(2,24) = 3.53, p(0.05]. The two-way interactionoc- words,one would not expectto seegeneralization. However, curredbecause the experimental but not thecontrol subjects' if feedbacktraining leads to an increasein subjects'tacit sensitivityto the/t/-/d/contrast was somewhatgreater in knowledgeabout the propertiesof/t/and/d/ (or differ- the third than the first block. The block main effect was encesbetween them), onemight expect generalization from nonsignificant,however, for boththe controland the experi- trained to untrained words. mentalsubjects [F(2,12') = 1.06,2.56, 1.06,p•0.10]. The Themultiple natural token approach to speechtraining experimentaland control subjects did not differsignificantly assumesthat exposureto the acousticvariation between to- in the first (pre-training) or the second block kensof a singlecategory will inducesubjects to derivea more IF(1,12) = 0.76, 2.03, p)0.10]. In the third (post-train- generalrepresentation than they would derive had they been ing) block,the experimental subjects' sensitivity was greater trainedon just a singletoken. Another issue addressed was than the control subjects'at a marginallysignificant level whetherchanges in vowelcontext would increase the effec- [F(1,12) = 4.04,p = 0.067]. tivenessof training.Native English-speakinglisteners use A larger overall effectof training might have been ob- variationsin vowelquality as a perceptualcue to the voiced- tainedhad moresubjects from southernChina beenincluded voicelesscontrast (e.g., van Summers,1987). Subjectswho in the experimentalgroup, for the subjectsshowing the lar- receivetraining on beat-beadand bet-bedin a singleblock gesteffects were from southernChina. Of the three Manda- mighttherefore show a largereffect of trainingand/or more rin subjects,two showedsmall effectsof training, and one generalizationthan subjects who receive training on the two showedno effect.The training effectsmight alsohave been minimal pairsseparately. largerhad more trainingtrials beenpresented. These possi- In experiment1, nativespeakers of Shanghainesetend- bilitieswere testedfurther in experiment3. ed to identifyEnglish stops better than nativespeakers of Mandarin.In experiment2, therewas some evidence that III. EXPERIMENT 3 Shanghainesesubjects benefited more from feedbacktrain- A. Introduction ingthan native Mandarin subjects, perhaps because Manda- rin has no obstruentsin word-final position,whereas The trainingadministered in experiment2 might be re- Shanghainesehas words ending in a glottalstop. The pres- gardedas effective, but just barelyso. A secondtraining ex- enceof a word-finalobstruent in their LI may causenative perimentwas therefore carried out in an attemptto find a speakersof Shanghaineseto attendto theend of syllablesto a bettermethod for trainingthe contrastbetween/t/and/d/ greaterextent than Mandarin subjects.Children's prior in the final positionof Englishwords. Identification with knowledgemay affect to some extent the effectivenessof feedbackwas again used as the trainingmodality, but a larg- languagetraining (Johnston,1988). If the sameholds true er numberof trainingtrials were given for a fewernumber of for speech,differences in howsyllables are processed might words. The experimentalso exploredissues pertaining to therefore influence the results obtained from feedback train- generalizationof training,acoustic variability of trainingto- ing. kens,and L1 background. Little empiricalevidence exists concerning the general- B. Methods ization of new speechperception abilities for adult learners of an L2. Neitherapplied behavioHsts nor Piagetian struc- The stimuli usedin this experimentwere the "V + B- turalistshave viewed generalization of trainingas an impor- removed"versions of minimal pairs usedin experiments1 tant researchissue because they see it as an inherentpart of and 2. The experimentwas divided into three blocks.In thelearning process. While it maybe true that generalization each,the subject's task was to identifythe final stop as "t" or is a "naturalproperty" of humanintelligence that can be "d." The subjectsreceived feedback training on just two "takenfor granted"for certainaspects of languagelearning minimalpairs (beat-bead,bit-bM) in the secondblock, but (Johnston,1988, p. 325), it is neverthelessknown (see the were testedon all four minimal pairs (beat-bead,bit-bM, Introduction)that Chinesesubjects have difficulty produc- bet-bed,bought-baud) in the firstand third blocks(desig- ing word-finalEnglish stops. Evidently, their prior experi- natedthe "before-"and "after-training"blocks). In the be- enceproducing/t/and/d/in the initial positionof Chinese fore- and after-trainingblocks, three separaterandomiza- wordsdoes not automaticallyenable them to producea per- tions of the 32 stimuli (eight wordsxfour tokens) were ceptuallyeffective contrast between these stops in thefinal presented.In the secondblock, 15 randomizationsof the 16 positionof Englishwords (Flege and Davidian,1985; Flege stimuli (four tokenseach of beat,bead, bit, bM) were pre- etal., 1987;F!ege, 1988c). Moreover, Strange and Dittmann sented for identification with immediate feedback. Thus (1984) assessedtransfer of training following speechper- more feedbacktrials were provided for each word here than ceptiontraining. They found incomplete transfer to newsyn- in experiment2 (60 vs 16). theticmaterials, and no transferto natural speech. A total of 16 Chinesesubjects participated, none of The subjectsin the presentexperiment were trained us- whom had taken part in experiment 1 or 2. Eight subjects ing the four tokensof beat,bead, bet, and bedused previously eachwere randomly assigned to "blocked"and "unblocked"
1691 J. Acoust.Soc. Am., Vol. 86, No. 5, November1989 James Emil Flege:Chinese perception of word-final/t/-/d/ 1691 conditions.Those in the blocked group were trained on who were assignedto the blockedcondition showed about beat-bit and bit-bid separatelyin counterbalancedorder, the sameeffects of training, and about the sameamount of whereasthose in the unblockedcondition receivedtraining generalization,as subjectswho receivedtraining on both on both minimal pairs at the sametime. Six subjectswere minimal pairs at the same time. The overall rate of correct native speakersof Mandarin, six were native speakersof identificationfor subjectsin the blockedand unblockedcon- Taiwanese,and four were native speakersof a southern ditions was 70%. Chinesedialect. As summarizedin Table V, the subjectsin Table VI alsopresents the A' valuesrepresenting the experiment3 resembledthose from experiment2 in termsof subjects'sensitivity to the/t/-/d/contrast beforeand after English-languageexperience. All had learnedEnglish as training. The /1' scoreswere submittedto a condition adultsand spoke English with a foreignaccent. Most of them (blocked versus unblocked)xtime (before versus after had arrived in the U.S. in their late twenties and had lived training))
1692 d. Acoust.Soc. Am., Vol. 86, No. 5, November1989 James Emil Flege: Chineseperception of word-final/t/-/d/ 1692 TABLEVI. Mean percent correct identifications of/t/and/d/by subjects whoreceived training separately ortogether onmultiple tokens oftwo English minimalpairs (blocked versus unblocked conditions). Performance wasassessed before and after feedback training both for words that were trained (bit- bid,bet-bed) ornot (bet-bed, bought-baud). The.4' scores represent anunbiased estimate ofsensitivity tothe/t/-/d/contrast. Standard deviations arein parentheses.
Trained words Untrained words Before After Before After
Blocked condition Percentcorrect 66%(26) 75%(23) 63%(30) 75%(26) .4' 0.736(0.117) 0.832(0.086) 0.667(0.210) 0.817(0.119)
Unblocked condition Percentcorrect 66%(32) 80%(27) 62%(29) 72%(30) .4' 0.702(0.184) 0.825(0.227) 0.654(0.289) 0.755(0.227)
identifythe word-final English stops using cues they applied beenremoved. The firstresulted in a nonsignificant9% in- to theword-initial/t/-/d/contrast in theirnative language crease in the rate of correct identifications.The second, (L1). The voiced and voicelessbursts that occurred at re- which includeda larger numberof trainingtrials on a leaseof lingualocclusion for finalstops in theunedited con- smallernumber of words,resulted in a slightlylarger and ditionof experiment1 arelikely to be similarto the release significantincrease (11%) that generalizedto untrained bursts for word-initial/t/and/d/in Chinese words. This words.The trainingeffects obtained here were similar in mayexplain the relatively good performance of Chinesesub- magnitudeto the one obtainedby Jamiesonand Morosan jectsfor the uneditedstimuli. If the Chinesesubjects relied (1986), who usedsimilar techniquesto train French sub- onword-initial cues, it wouldalso explain their poor perfor- jectson the word-initialEnglish/0/-/•3/contrast (seealso manceon the stimuli without releasebursts, since cues that Loganet al., 1989). Althoughthe Chinesesubjects' perfor- remainedafter editing(e.g., precedingvowel duration and manceimproved significantly, they continuedto identify quality) areabsent in word-initialstop voicing distinctions. stopsmuch lesswell than nativespeakers of English,who This interpretationis consistentwith the fir•dingthat performedat near-perfectlevels. Thus an importantques- the Chinesesubjects showed only a very small decreasein tion to ask is: Why did the Chinesesubjects benefit so little sensitivityto the/t/-/d/contrast whenclosure voicing was from the training? removed.One wouldnot expectthe Chinesesubjects to rely There are two reasons to think that the small effect of on closurevoicing as a cuefor word-finalstops in English training probablycannot be attributedto the nature of the because word-initial tokens of/t/and/d/are not distin- phoneticcontrast being trained. First, the Englishsubjects guishedby closurevoicing in Chinese.The very smalleffect identifiedstops in the V -t- B-removedstimuli quite accu- of removingclosure voicing was not due to the fact that the rately. Second,previous research suggests that nonnative voicingwas inaudible. A follow-upstudy (see the Appen- phoneticcontrasts which involve the voicingfeature are rel- dix) revealedthat nativeEnglish speakers showed a signifi- ativelyeasy to trainbecause they are relatively robust in psy- cant preferencefor final /d/'s with closurevoicing over choacousticterms (Tees and Werker, 1984;Burnham, 1986; thosewithout closure voicing. The Chinesesubjects' sharp Bestet al., 1988;Werker et al., 1981;Werker and Tees, 1983, drop in correctidentifications when release burst cues were 1984a,b;Werker and Logan,1985). removedwas not an artifactof the waveformediting, since The difficultyof the word-finalEnglish/t/-/d/con- stopsin conversationalspeech often have little or no closure trast may have derivedinstead from a syllable-processing voicingand no audiblerelease burst. Moreover, the English strategyestablished during L1 acquisition.This inference childrenwho were tested closely resemble the nativeEnglish was drawn from the observationthat certain subgroupsof adultsin showinglittle effectof editing. Chinesesubjects tended to performbetter than othersin ex- It is not known if the Chinesesubjects' di•culty in iden- periment1, and to benefitmore from the feedback training in tifying/t/'s and/d/'s without releasebursts would affect experiments2 and 3. In experiment3, threeof four subjects their comprehensionof English,nor is it knownif theyestab- from southern China whose L1 includes word-final ob- lishedcentral phonetic representations for word-finalEng- struents(but not/d/) showedgains that weremuch greater lishstops. If theydid, their centralphonetic representations thanaverage ( 18%-30% correct).These subjects may have wereprobably less elaborated than nativeEnglish speakers' learnedto focusgreater attention on the syllabletermination (seeMassaro and Oden, 1980;Samuel, 1982; Flege, 1974, propertiesof wordsin their L1 than the native Mandarin 1988a), which presumablyspecify parameter values for a subjects(whose L 1 hasno word-finalobstruents) in orderto largenumber of dimensionssuch as precedingvowel dura- distinguishbetween open syllables and syllablesending in a tion, vowel quality, and F• offsetfrequency in addition to glottal stop.This hypothesisreceived support in a recent information associated with final release bursts, such as study by Flege and Wang (1989). In that study, it was overall intensity. shownthat subjectswhose L1 (Cantonese)permits unre- Two trainingexperiments were undertaken to improve leased/p,t,k/in finalposition identified edited tokens of/t/ Chinesesubjects's ability to identify/t/and/d/in English and/d/significantlybetter than subjects whose L 1 (Manda- wordsfrom which closurevoicing and releasebursts had rin) permitsno final obstruents.
1693 J. Acoust.$oc. Am., Vol. 86, No. 5, November1989 James Emil Fiego:Chinese perception of word-final/t/-/d/ 1693 Locke(1983) speculatedthat syllableoffset properties identificationsusing other cues such as vowel duration. are importantfor childrenlearning their L1. Adultsmay To testthis, a shortpaired-comparison experiment was deletefinal/t/'s and/d/'s, or replacethem with glottal stop carriedout to determinewhether native speakers of English (BernsteinRatner and Lubereft,1984), which might ex- do or do not useclosure voicing in a phoneticallyrelevant plainEnglish-speaking children's propensity for omitting fi- manner.The stimuli consistedof oneof the four tokenseach nalstops or replacingthem with glottal stops. The hypothe- of the/d/-final word from experiment1, and a copyfrom sisoffered here is that, if the LI doesnot distinguishwords whichclosure voicing had been removed. In agreementwith accordingto informationspecified in syllableoffsets, then previousreports (e.g., Revoile etal., 1982), the closure voic- L2 learnerswill focusrelatively little attentionlater in life to ingin thefinal/d/'s wasless intense (by 18riB) thanthe the offsetproperties of syllables(or words)in an L2. This vowel and died out before the end of the closureintervals in assumesthat syllable-processingstrategies are indeed sixof the sevenwords selected. Closure voicing filled 75% of learned.Consistent with this assumption,Walley (1987) the closureintervals on average.Pairs of/d/-final words foundthat in contextswhere meaning was not semantically with andwithout closure voicing were presented, separated constrained,4 and 5 yearolds did not detectmore mispro- by0.5 s, to 11normal-hearing native English subjects with a nunciationsin initialthan final position. They did soin con- meanage of 29 years.The subjectswere to decidewhich strainingcontexts, however, thereby resembling adults. memberof eachpair sounded"better." Each pair waspre- Walley(1988) found that adults but not children gave lower sentedrandomly 20 timeseach (10 timeswith the voiced ratingsto word-initialthan word-final consonants that had /d/'s in thefirst position, and 10 timesin thesecond posi- beendegraded by noise. tion). In conclusion,Chinese subjects identified fully specified The resultssuggested that removingclosure voicing in English/t/'sand/d/'s as well as native English speakers but experiment1 did not affectthe Englishsubjects' perfor- performedmore poorly when release bursts were removed. mancebecause of a ceilingeffect. Subjects chose/d/'s with Feedbacktraining increased significantly Chinese subjects' closurevoicing more often than thosewithout (64% vs sensitivityto stopswithout release bursts, but theirgains 36%). A significantvowel X voicinginteraction was ob- weresmall in absoluteterms. This wasespecially true for tained [F(6,60) ----7.34, p < 0.05] becausethe simplemain nativespeakers of Mandarin,a languagethat does not per- effectof voicingwas significantin all sevenwords except mit obstruentsin word-finalposition. Additional research booed(p • 0.01). As expectedfrom previous research, inter- will be neededto furthertest the hypothesisthat the effec- subjectvariability was great. The rate at which/d/'s with tivenessof feedbacktraining for word-finalstops will belim- voicingwere chosen over/d/'s withoutvoicing ranged from itedif subjectslearn to focus relatively little attention on the 55% to 73%. endof syllablesduring L1 acquisition. •Oyama(1982a) noted a significantcorrelation between degree of foreign accentand age of learningfor Italian menwho began learning English as ACKNOWLEDGMENTS an L2 betweenthe agesof 6 to 20 years.Tahta et ai. ( 1981) reportedno accentfor nonnativespeakers who began learning English by the ageof 6 The researchwas supported by NIH grantNS20963. years,moderage accents for thosewho began learning English between the The authorthanks E. Jamesfor helpwith all phasesof the agesof7-11 years,and relatively strong accents for individualswho began learningEnglish after the ageof 12 years.Amount of L2 experiencehas research.Thanks are due also to O.-S. Bohn, W. Strange, generallybeen found to predictdegree of foreignaccent less well thanage andtwo anonymousreviewers for commentson a previous of learning.Tahta et al. ( 1981) foundthat ageof learningbut notlength of draft of the article. residenceaccounted for a significantamount of variancein foreignaccent ratings,perhaps because subjects with lessthan 2 years'experience were not included.Oyama (1982a) found that a correlationbetween degree of APPENDIX: PERCEPTION OF CLOSURE VOICING accentand ageof learningremained significant when differences in length Massareand Oden (1980) hypothesizedthat listeners of residencewere partialedout (seealso Fathman, 1975; Seliger et al., 1982, but cf. Asher and Garcia, 1969). evaluateall potentialacoustic cues to a phoneticcontrast 2Theauthor thanks W. Strangefor pointingthis out. independentlyand continuously, even though no one prop- 3Thedecrease in correct identificationsthat occurswhen releasebursts are ertyis necessary to establishcategory identity. However, re- removedis generallygreater for/p,t,k/than/b,d,g/(Malcot, 1958;Win- movingclosure voicing did not significantlylower English itz et al., 1972; Raphael, 1981; Revoileet al., 1982), perhapsbecause voicelessstops are releasedmore consistentlythan voicedstops (Wang, subjects'sensitivity to the/t/-/d/contrast in experiment1. 1959), or becausethey are more intense. Thisraises the possibility that the subjects may not use clo- 4Wolf's(1978) subjectshad three response alternatives: voiced, voiceless, surevoicing as a cueto the/t/-/d/contrast. or no final stop.The percentagesgiven here for the editedwords were The nonuseof closurevoicing may occurbecause/d! is basedonly on responsesfor wordsidentified as havinga final stop. SComprehensonwas examined using a 45-item,tape-recorded test that as- not reliablydistinguished from/t/in Englishby closure sessessyntactic knowledge as well as word recognition. The test, a portion voicing(Flege and Hillenbrand,1987), especiallyin the of the EnglishLanguage Assessment Battery (English Language Insti- speechof children(Smith, 1979; Flege et al., 1987).Pre- tute,University of Michigan,1986) does not gauge directly the ability to identifyphonemic categories. The Chinesesubjects' average score was viousresearch has shownthat the perceptualeffect of re- 34.2 correctout of 45. The percentilescore was based on the performance movingclosure voicing from/b,d,g/varies considerably for of 1486students who had takenthe testwhile enrolled in Englishlanguage individualnative speakers of English(Wardrip-Fruin, 1982; classesin Ann Arbor prior to enrollingas full-timedegree seeking stu- Revoileet al., 1982;Raphael, 1981 ). Alternatively,the na- dents. ørhe closureinterval extended from constrictionof the finalstop, identified tiveEnglish subjects may not have shown a decreasein sensi- by a decreasein signalamplitude and a concomitantsimplification in wa- tivity to the/t/-/d/contrast whenclosure voicing was re- veshape,to the beginningof the final releaseburst. moved because they were able to make accurate 7Preliminaryanalyses for the uneditedand the burst-removedwords
1694 d. Acoust.Sec. Am., Vol. 86. No.5. November1989 JamesEmil Flege: Chinese perception of word-final/t/-/d/ 1694 showedthat therewas not a significantcorrelation between the text fre- Boothroyd,A., andNittrouer, S. (1988). "Mathematicaltreatment of con- quenciesof the 14 wordsexamined and the rate of correctidentifications text effectsin phonemeand word recognition,"J. Acoust.Soc. Am. 84, It(12) = O.172, 0.3751, nor wasthere a differencein the rateof correct 101-114. identificationsfor male and femaleChinese subjects for /t/ (57% vs Borden,G., Gerber,A., andMilsark, G. (1983). "Productionand percep- 62%) or/d/(68% vs71%) [F(1,28) = 1.01,0.18,p>0.10]. tion of the/r/-/l/contrast in Koren adultslearning English," Lang. •Thesum of squares of thebetween-subjects factors were adjusted ina hier- Learn. 33, 499-526. archical fashionbecause between-subjects factors tend to be correlated Burnham,D. (1986). "Developmentalloss of speechperception: Exposure whensample sizes are unequal. to and experiencewith a firstlanguage," Appl. Psycholing.3, 207-241. " significantvowelxediting condition interaction was obtained Bri•re, E. (1966). "An investigationof phonologicalinterference," Lan- [ F( 18,792) = 2.66,p < 0.05] becauseof a differingeffect of thevowel fac- guage42, 769-796. tor for the four editingconditions. This seemsto haveoccurred because Cheng,C.-C. (1973). `4 SynchronicPhonology of Mandarin Chinese(Mou- the differencein sensitivitybetween the uneditedand editedconditions ton, The Hague). wasgreater for some minimal pairs than others. For example, the decrease Clumeck,H., Barton,D., Macken,M., andHuntington, D. (1981). "The in ,4' scores from the unedited to the V + B-removed condition was aspirationcontrast in Cantoneseword-initial stops: Data from children greaterfor bit-bid (0.257) than for bought-baudor bet-bed (0.137, and adults,"J. Chin. Ling. 9, 210-224. 0.166). Vowel-related effects will not be discussedfurther becausethe Cochrane,R. (1977). "The acquisitionof/r/and/1/by Japanesechildren groupX vowelX editing condition interaction was nonsignificant and adultslearning English as a secondlanguage," unpubl. Ph.D. thesis, [F(36,792) = 0.781,p> 0.10] and becausethe exactacoustic basis for Universityof Connecticut,Storrs, CT. thesedifferences is uncertain(see Hillenbrand et al., 1984;van Summers, Cole, R., and Perfetti, C. (1980). "Listeningfor mispronunciationsin a 1987). children'sstory: The useof contextby childrenand adults,"J. Acoust. •øThed' valueswere calculated according tothe formula d' = z(proportion Soc. Am. 67, 1323-1332. of hits) - z(proportionof falsealarms). The ,4' scoreswere examined in Eckman, F. (1981). "On the naturalnessof interlanguagephonological theANOVA reportedearlier because of thesmall number of trialsand the rules,"Lang. Learn.31, 195-216. requirementof normal distributionsfor d', which could not be met for Eguchi,S., andHirsch, I. (1969). "Developmentof speechsounds in chil- manyof the nativeEnglish subjects owing to perfectscores. Even though dren," Acta Oto-Laryngol.Suppl. 257, 1-55. ,4' and d' scoresmay be correlatedsignificantly (see Snodgras et al., EnglishLanguage Institute (1986). Manualfor theEnglish Language Insti- 1972), the regressionanalysis examined d"scores because they seemed to tuteListening Comprehension Test (Ann Arbor, MI), 16 pp. be more revealingof between-groupdifferences than the//' scores.An Fathman, A. (1975). "The relationshipbetween age and secondlanguage ANOVA examiningd' scoresshowed the samesignificant main effects productiveability," Lang. Learn. 25, 245-253. and interactionsas the ,4 ' analysis,but posthoc testsrevealed between- Flege,J. (1984). "The detectionof French accentby Americanlisteners," groupdifferences not seenin the ,4 ' analysis(significantly less sensitivity J. Acoust. Soc. Am. 76, 692-707. for the Chinesethan Englishadults in the uneditedcondition, and less Flege, J. (1988a). "The productionand perceptionof foreign language sensitivityto the/t/-/d/contrast by the Englishchildren than English speechsounds," in Human Communicationand Its Disorders,,4 Review, adultsin the burst-removedcondition; p <0.05). When the six-factor 1988,edited by H. Winitz (Ablex, Norwood,NJ), pp. 224-401. regressionmodel was applied to ,4 ' scores,it did notaccount for a signifi- Flege,J. (1988b). "Factorsaffecting degree of foreignaccent in English cant amount of variancefor either the burst-removed[F(1,27) = 3.67, sentences,"J. Acoust. Soc. Am. 84, 70-79. p = 0.063] or the V + B-removed[F(1,27) = 2.96,p = 0.069] condi- Flege,J. (1988c)."The developmentof skillsin producingword-final/p/ tion. and/b/: Kinematicparameters," J. Acoust.Soc. Am. 84, 1639-1652. •The effectof LI backgroundon thecorrect identification rate for/d/'s Flege,J., andDavidian, R. (1985). "Transferand development processes in wassignificant for stopsin the voicing-removedand burst-removed condi- adult foreignlanguage speech production," J. Appl. Psycholing.Res. 5, tions [F(2,207) = 15.4,4.43, p<0.05]. Posthoc tests revealed that the 323-347. Shanghainesesubjects identified/d/'s correctlysignificantly more often Flege,J., andEefting, W. (1986). "Linguisticand developmental effects on thanthe native Mandarin or Taiwanesespeakers. Nonsignificant trends in the productionand perceptionof stopconsonants," Phonetica 43, 155- the same direction were evident for final/t/'s. 171. •2Hyperaspirationwas not notedin fo,d,g/. Thissuggests, in agreement Flege,J., and Eefting,W. (1987). "Cross-Languageswitching in stopcon- with two recent speechproduction studies (Flege et al., 1987; Flege, sonantproduction and perceptionby Dutch speakersof English,"Speech 1988c),that the Chinesesubjects produced voiceless but not voicedword- Commun. 6, 185-202. final stopsin Englishwith a laryngealdevoicing gesture. Flege,J., and Hillenbrand,J. (1985). "Differentialuse of temporalcues to •3Thesestudies employed a wide variety of methods:discrimination train- the Is-z] contrastby native and non-nativespeakers of English,"J. ingwith immediatefeedback (Strange and Dittmann, 1984); perceptual Acoust. Soc. Am. 79, 508-517. feedbackconcerning consonant identity together with feedbackconcern- Flege,J., andHillenbrand, J. (1987). "A differentialeffect of releasebursts ing the differencebetween correct and incorrectproductions, and addi- on thestop voicing judgments of nativeFrench and Englishlisteners," J. tional information relating to "visual" and tactile-kinesthetic"differ- Phon. 15, 203-208. encesbetween/r/and/1/(Borden et al., 1983);instructions concerning Flege,$., McCutcheon,M., and Smith,S. (1987). "The developmentof tongueplacement and imitation of a nativespeaker's correct productions skill in producingword-final English stops," J. Acoust.Soc. Am. 82, (Cochrane,1977); and articulatory descriptions followed by individual 433-447. andgroup practice in producing/r/and/1/that wereevaluated by fellow Flege,J., andPort, R. (1981). "Cross-languagephonetic interference: Ara- studentsand a nativeEnglish-speaking instructor (Gillette, 1980). The bic to English,"Lang. Speech 24, 125-146. lengthof trainingvaried widely, from two45-min clinical sessions (Bor- Flege,J., andWang, C. (1989). "Native-Languagephonotactic constrains denet al., 1983)to 5 of clinicaltraining (Cochrane, 1977) to 15laborato- affecthow well Chinesesubjects perceive the word-finalEnglish/t/-/d/ ry sessions(Strange and Dittmann, 1984) to 5 weeksof classroomtrain- contrast,"J. Phon. (in press). ing (Gillette, 1980). Florentine,M. (1985). "Speechperception in noiseby fluent,non-native listeners,"J. Acoust.Soc. Am. Suppl.1 77, S106. Anderson,J. (1983). "The difficultiesof Englishsyllable structure for Gillette,S. (1980). "Contextualvariation in the perceptionof L andR by ChineseESL learners,"Lang. Learn. Commun. 2, 53-61. Japaneseand Koreanspeakers," Minn. PapersLing. Philos.Lang. 6, 59- Anderson,J. (1987). "The markedhesSdifferential hypothesis and syllable 72 ( Universityof Minnesota,Minneapolis). structuredifficulty," in InterlanguagePhonology, edited by G. Ioup and Gras, S. (1983). "Integrafiveattitude and difficultiessecond language $. Weinberger(Newbury House,Rowley, MA), pp. 279-291. learnersencounter in comprehendinginformal speech: A studyof Kore- Asher, J., and Garcia. R. (1969). "The optimal age to learn a foreign lan- an graduatestudents attending the University of Texas." unpubl. Ph.D. guage,"Mod. Lang. J. 53, 334-341. thesis,University of Texas at Austin. BernsteinRatner, N., andLuberoff, A. (1984). "Cuesto post-vocalicvoic- Grief, J. (1971). "Nonparametricindexes for sensitivityand bias:Comput- ing in mother-childspeech," 1. Phon. 12, 285-289. ing formulae,"Psychol. Bull. 75, 424-429. Best,C., McRoberts,G., and Sithole,N. (1988). "Examinationof percep- Halle, M., Hughes,G., and Radley,J.-P. (1957). "Acousticproperties of tual reorganizationfor nonnativespeech contrasts. Zulu click discrimi- stopconsonants," J. Acoust.Soc. Am. 29, 107-116. nationby Englishspeaking adults and infants,"JEP:HPP 14, 345-360. Heyer,S. (1986). "Englishfinal consonantsand the Chineselearner," u n-
1695 J. Acoust.Soc. Am.,Vol. 86, No. 5, November1989 JamesEmil Fiege: Chinese perception of word-final/t/-/d/ 1695 publ.MA thesis,Dept. of Linguistics,Southern Illinois University,Car- Oyama,S. (1982b). "The sensitiveperiod and comprehension of speech," bondale, IL. in Child-AdultDifferences in SecondLanguage Acquisition, edited by S. Hillenbrand,J., Ingrisano,D., Smith,B., andHege, J. (1984). "Perception Krashenand R. Scarcella(Newbury House, RowIcy, MA), pp.39-52. of the voiced-voicelesscontrast in syllable-finalstops," 3. Acoust.Sec. Parnell,M., andAmerman, 3. ( 1978). "Maturationalinfluences on percep- Am. 76, 18-28. tion of coarticulatoryeffects," J. SpeechHear. Res.21, 682-701. Howie,I. (1976). AcousticalStudies of Mandarin Vowelsand Tones( Cam- Pisoni,D., Aslin,R., Percy,A., andHennessy, B. (1982). "Someeffects of bridgeU.P., London). laboratorytraining on identificationand discrimination of voicingcon- /amieson,D., and Morosan,E. (1986). "Training non-nativespeech con- trasts," JEP:HPP S, 297-314. trastsin adults:Acquisition of the English/ /-//contrast by franco- Polka,L. (1987). "Perceptionof Persianuvular and velar stops by speakers phones,"Percept. Psychephys. 40, 205-215. of AmericanEnglish," J. Acoust.Sec. Am. Suppl.I, 81, S66. Johannson,S. (1978). Studiesin Error Gravity,Gothenburg Studies in Eng- Raphael,L. (1981). "Durationsand contexts as cues to word-finalcognate lish 44 (Dept. of English,Gothenburg University of Sweden}. oppositionsin English,"Phonefica 3•1, 126-147. Johnston,J. (1988). "Generalization:The nature of change," Lang. Repp, B. (19181)."Categorical perception: Issues, methods, fmdingg" in Speech,Hear. Serv.Schools 19, 314-329. Speechand Language: Advances in BasicResearch and Practice, VoL I0, Jusczyk,P. (1979). "Infant speechperc•tion: A criticalappraisal," in Per- editedby N. Lass(Academic, New York), pp. 243--335. spectioeson theStudy of Speech,edited by P. Eimasand J. Miller (Erl- Revoile,S., Picket, J., Holden,L., andTalkin, D. (1982). "Acousticcues to bantu, Hillside, NJ). finalstop voicing for impaired- and normal-hearing listeners," J. Acoust. Kalikow, D., and Swets,J. (1972). "Experimentswith computer-con- Sec. Am. 72, 1145-1154. trolleddisplays in second-languagelearning," IEEE Trans.Audio Eke- Samuel,A. (1982). "Phoneticprototypes," Percept. Psychephysics 31, troacoust. AE-20, 23-28. 307-314. Kelso,S., andTuller, B. ( 1981). "Toward a theoryof apraeriesyndromes," Schneiderman,E., Bourdages,L, and Champagne,C. (1988). "Second- Brain Lang. 12, 224-245. languageaccent: The relationshipbetween discrimination and perception Kent, R., and Rosenbek,J. (1983). "Acoustic patterns of apraxia of in acquisition,"Lang. Learn.38, 1-21. speech,"L SpeechHear. Res. 26, 231-249. Scliger,H., Krashen,S., and Ladefoged,P. (1982). "Maturationalcon- Kester, C. (1987). Word Recognitionin Foreign and Native Language straintsin the acquisitionof secondlanguages," in Child-Adult Differ- (Foris, Dordrecht, The Netherlands). encesin SecondLanguage Acquisition, edited by S. Krashen,R. ScarceHa, Krause,S. {1982a). "Vowel durationas a perceptualcue to postvocalic andM. Long(Newbury House, Rowlcy, MA), pp. 13-19. consonantvoicing in youngchildren and adults," J. Acoust.Sec. Am. 71, Shattuck-Hufnagel,S. (1987). "The role of word-onsetconsonants in 990-995. speechproduction planning: New evidence from speech error patterns," Krause,S. (1982b)."Developmental use of vowelduration as a cueto post- in Motorand Sensory Processes of Language, edited by E. Keller and M. vocalicstop consonant voicing," L SpeechHear. Res.25, 388-393. Gopnik(Eribaum, Hillside, N]), pp. 17-52. Kuhl, P. (1987). "Perceptionof speechand soundin early infancy," in Shinn,P. (1985). "A ernss-languageinvestigation of thestop, affricate, and Handbookoflnfant Perception,Vol. 2, editedby P. Salapatekand L. Co- fricativemanners of articulation," unpubl. Ph.D. thesis, Brown Universi- hen (Academic,New York), pp. 275-382. ty, Providence,RI (Univ. Microfilms,No. 8519906). Lade,R. ( 1957}.Linguistics•4cross Cultures (University of Michigan,Ann Shoekey,L., andBond, Z. (1980). "Phonelogicalprocesses in speech ad- Arbor). dressedto children," Phonetica37, 267-274. Leonard,L., Schwartz,R., Chapman,K., Rowan,L., Prelocke,P., Weiss, Simon,C., andFourtin, C. (1978). "Cross-languagestudy of speech-pat- A., and Messick,C. (1982). "Early lexicalacquisition in childrenwith tern learning,"L Acoust.Sec. Am. 63, 925--935. specificlanguage impairment," J. SpeechHear. Res. 25, 554-564. Smith,B. (1979). "A phoneticanalysis of consonantaldevoicing in chil- Lenneberg,E. ( 1967}. BioiogicolFoundations of Language(Wiley, New dren'sspeech," L Child Lang. 6, 19-28. York). Snodgras,J., Volvoritz,R., andWaifish, E. (1972). "Recognitionmemory Li, C., andThompson, S. (1981). MandarinChinese: A FunctionalRefer- for words,pictures, and words+ pictures,"Psychenora. Sci. 27, 345- enceGrammar (University of California, Berkeley,CA ). 347. Lisker,L. (1970}. "On learninga new phoneticcontrast," Haskins Lab. Strange,W. (1986). "Speechinput and the development of speechpercep- Stat. Rep. SpeechRes. 24, 1-15. tion,"in OtitisMedia and ChildDevelopment, edited by J. Kavanagh Lisker,L., andAbramson, A. ( 1964}."A cross-languagestudy of voicingin (York, Parkton,MD), pp. 12-26. initial stops:Acoustical measurements," Word 20, 384-421. Strange,W., andDittmann, S. (1984). "Effectsof discriminationtraining Locke, J. (1983). Phonelogical,4CCluisition and Change(Academic, New onthe perception of/r-l/by Japaneseadults," Percept. Psychephys. 36, York ). 131-145. Logan,J., Pisoni,D., andLively, S. (1989). ''TrainingJapanese listeners to StrangeW., andJenkins, J. (1978). "The roleof linguisticexperience in the identify/r/and/i/," Res.Speech Percept. 14, 1-22 (Dept. of Psycho- perceptionof speech,"in Perceptionand Experience, edited by R. Walk logy, IndianaUniversity). andH. Pick,Jr. (Plenum,New York), pp. 125-169. Mack, M., andTierney, J. ( 1987}. "The inte!ligibilityof naturaland reced- Tahta,S., Wood, M., and Lowenthal, K. (1981). "Forelgp_ ed semanticallyanomalous sentences: A comparativeanalysis of English relatingto transferof accentfrom the firstlanguage to the secondlan- monolingualsand German-Englishbilinguals," Tech. Rep. 792 (Lin- guage,"Lang. Speech 24, 265-272. coin Lab.,MIT, Cambridge,MA). Tarone,E. (1980). "Someinfluences on the syllablestructure of interlan- MacKain,K., Best,C., andStrange, W. ( 1981}. "Categoricalperception of guagephonology," lnt. Rev.Appl. Ling. 18, 139-152. English/r/and/1/by Japanesebilinguals," Appl. Ling. 2, 369-390. Tees,R., and Werker,L (1984). "Perceptualflexibility: Maintenance or Ma16cot,A. {1958). "The role of releasesin the identificationof released recoveryof theability to discriminate non-native speech sounds," Canmi. final stops,"Language 34, 370-380. J. Psych. 38, 579-590. Massare,D., andeden, G. (1980). "Evaluationand integration ofaconstic Trubetzkoy,N. (1939). Grundzgeder phenoleRie (Travaux du cerclelin- features in sl•ech oerceotion." J. Acoust_$oc. Am_ 67. 996--101•. gui•tiqu• de Profitre,7)_ MeClasky,C., Pisoni,D., andCarrell, T. (1983). "Transferof trainingof a Truin, P. (1981). "Herkenningvan fragmentenvan gesprokenNeder- new linguisticcontrast in voicing,"Percept. Psychephys. 34, 323-330. landsewoofden door Nederlanders en niet-Nederlanders,"Toegepaste Miyawaki, K., Strange,W., Verbrugge,R. Liborman,A., Jenkins,J., and Taalwetenschapin Artikelen 9, 170-182. Fujimura,O. (1975). "An effectof linguisticexperience, The discrimina- Underbakke,M., Polka,L., Gottfried,T., andStrange, W. (1988). "Trad- tionof [r] and [I] by nativespeakers of Japaneseand English," Percept. ingrelations in theperception of/r/-/l/by Japaneselearners of Eng- Psychephys.lg, 331-340. lish," J. Acoust. See. Am. 84, 90-100. Ohde,R., and Shaft, D. (1977). "Order effectof aeonsticsegments of VC vanBalen, C. (1980). lnteiligibilityof SpeechFragments (Drukkerijelink- andCV syllableson stopand vowel identification," J. SpeechHear. Res. wijk, Utrecht). 20, 543-554. vanSummers, W. (1987). "Effectsof stressand final-consonant voicing on Oyama,S. (1982a). "A sensitiveperiod for the acquisitionof a nonnative vowelproduction: Articulatory and acousticanalyses," J. Aeonst.Sec. phonelogical'system," in Child-AdultDifferences in SecondLanguage Am. 82, 847-863. Acquisition,edited by S. Krashen and R. Scareella(Newbury House, Walley,A. (1987}. "Youngchildren's detections of word-initialand -final RowIcy, MA}, pp. 20-38. mispronunciationsin constrained and uneonstrainedcontext," Cogn.
1696 J. Acoust.Sec. Am., Vol. 86, No. 5, November1989 James Emil Fiego:Chinese perception of word-final/t/-/d/ 1696 Dev. 2, 145-167. Werker, J. (1989). "Becominga nativelistener," Am. Sci. 77, 54-59. Walley, A. (1988). "Spokenword recognitionby youngchildren and Werker, J., and Logan,J. (1985). "Cross-languageevidence for threefac- adults,"Cogn. Der. 3, 137-165. torsin speechperception," Percept. Psychophys. 37, 35-44. Wang,W. (1957). "Transitionand release as perceptual cues for finalplo- Werker,J., andTees, R. (1983). "Developmentalchanges across childhood sives,"J. SpeechHear. Res.2, 66-73. in the perceptionof non-nativesounds," Can. J. Psychol.37, 278-286. Wardrip-Fruin,C. (1982). "On the statusof temporalcues to phonetic Werker,J., andTees, R. (1984a). "Cross-languagespeech perception: Evi- categories:Preceding vowel duration as a cueto voicingin finalstop cate- dencefor perceptualreorganization during the first year of life," Infant gories,"J. Acoust.Soc. Am. 71, 187-195. Behar. Der. 7, 49-63. Weinberger,S. (1987). "The influenceof linguisticcontext of syllable Werker,i., andTees, R. (1984b). "Phonemicand phonetic factors in adult structuresimplification," in InterlanguagePhonology, edited by G. Ioup cross-languagespeech perception," J. Acoust.Soc. Am. 75, 1866-1878. and S. Weinberger(Newbury House, Rowley, MA), pp. 401-418. Winitz, H., Scheib,M., and Reeds,J. (1972). "Identificationof stopsand Werker,J., Gilbert,J., Humphrey,K., andTees, R. (1981). "Developmen- vowelsfrom the burstportion of/p,t,k/isolated from conversational tal aspectsof cross-languagespeech perception," Child Dev. 52, 349- speech,"J. Acoust.Soc. Am. 51, 1309-1317. 355. Wolf, C., (1978). "Voicingcues in Englishfinal stops," J. Phon.6, 299-309.
1697 J. Acoust. Soc. Am., Vol. 86, No. 5, November 1989 James Emil Flege: Chinese perceptionof word-final/t/-/d/ 1697