Mimicry of Non-distinctive Phonetic Differences Between Language Varieties *

James Emil Flege Ro bert M. Ham mond Unil'l![sifyof Alabama. Birmingham Miami·Dade Community College

1\ delayed mimicry paradigm w.as used to assess speakers' ,awareness of non-dlSlLnCf!Vc phonetic differences which in part dislinguisn I:mguagcs, . The notion of " rho no logIcal fillu­ Lng" IffiroheS ttur second Lmsuagc learners may not be lIblc to perceIve: phonetic differences between thCIT native language and :01 fOrf:.gn langua~ unless the phonetic differences arc linguistically relevant in the nallve bngu2gc. If cross-bogulIgc phoncuc differences arc In fact perceIved poorly, It is unlikely [hal phon.::!;\; mod,ficallon , ..,II OCcur .n Ihc course of nalUu.hsllc second language acqu15iuon. In th.s study natlvC' English speakers familiar Wllh Spamsh·accentcd English ulempted to lead sentences With a Spamsh accent. Acoustic mC::l5Uremc:nts showed thai twO phonellc characlens!!cs of Engllsh-Ihe long VOT values aSSOC iated '''llh 'p, t, kI and hnal-syll:Jble lengthening- were altered In the direction of Spanuh anu Spanish.accented Enghsh. These results prOVide tentative eYldence that nOn -dlStlnCIIVe rhonetle differences between languages arc detecuble by I:lnguage learners and thus do nO! present an msupeuble bamer to phonetIC learning in second 1:lOgU .. gc "cqulsHlun

For several decades phonologists have believed that "phonological fillering" con· tnbutes Importantly to the presence of foreign accent In the speech of second languagc learners ITrubct;:koy 1969:5 1 ffI. Having acquired the phonology of one language, the language learner will tend to mterpret sounds heard in a foreign language in terms of phonological umts found In the native language. For phonolog­ ic:.a l filtenng to occur phonological unllS III two different languages must ~ iudged to be the "s:.ame" or "Slmil:.ar" IWetnreich 1953:71. A maJor problem for COnt3Ct analyses In general and contrastive analyses, in particular, IS 10 determme how such equlv:.alencies are est:.ablished by bi linguals, and at what levellsl of an.. lysls the process of tnterlingual identification IS best descnbed.

Most tnvestigations of second-language phonology h:.ave used either the allophone or the distinctive feature as Ihe prllne uni t of analysiS. Work done Wlthlll 1\ structural phonemic framework le.g., Weinreich 1953; Shen 1959: Moulton 1962) has focused on between-language differences in phonemic inventory and their effect on the system of linguistically relevant contrasts, on the distribution of allophones, and on constraints affecting permissible sequences of surface phones. In the words of Robert Lado 11957: J II. we tend to "transfer ... our phonemes and their variants" when leamlOg a foreign language. Between-language differences Involvmg non-segmental characteristics of speech such as stress, intonation, and Juncture have also been handled wlthm a framework of phonemIC contrast. 2 Swrues in Second um.ruoge Acqulsllion. 5. No. I, 19H2

Orner approaches have made use of tbe distinctive feature as the prime unll of analysIs. Ritchie (1968) proposed tJut.a perceptual hierarchy of distinctive fC2turCs could be used to predict which native-language phone would be used In plat'! of a foreign-l.anguage phone absent from the leamer's native-language Inventory !e.g., l tJ or(sl for English /BII. Eckman (I977)proposed that markedness, as establi shed by cross-language implicational n:Jations (typological markedness), might be used to predict how and where phonological contrasts Will be neutralized by language learners. Stili other investigations have focused on rules or processes relating surface phones to underlying phonological categories and their possible effect on pronunciation of fortlgn-Ianguage words (Lovins 1975 1 Eckman 1981 1.

An assumption that often setms to be madt is that inter-lingual contact can be adequately describtd in ttrms of :I small number of discrete linguistic categorics: i.e., allophones, phonemes, or segmemal changes describable in tenns of distinc­ tive fe.lltures. Relatively little importance has been attached to the phonetic sub­ stance of surface phones found in the tutive and target languages In many th,"-'Oreu­ ul accounts, except for its possible role in influencing which phonemes will be considered the "same:" in the tWO languages. Once assigned a structural phone:mlc description, a target-language: phone is exptcted to be replaced by a phonetically similar allophont of the: corre:sponding native language: phone:me: o r, if no such allophone: e:XIStS, to be phone:rically approximated. For e:xample:, learners rn:lY substitute: the: alveolar StOp of the: native language for a dental Stop which occurs in a foreign language, or replace the uvular III of a language: they arc Icarning with the approximate Irf they learned as a child IWeinreich 19531.

It may be Important, howe:ver, to pay somewh.u closer ancorion [0 the rhonetlc chaucteristics of sounds in the native and foreign language (B riere 19661. Cross­ language interference at the phonetic level probably contnbutes .:I great deal (Q what is pc=rceivtd as foreign accent. Phonetic Implementation of comparable phonological unIts in two languages may differ in Important ways IFromkin 1979; Laddoged 1980). In such cases it appears that articulatory habits pre:vlously estab­ lished torealite phonological units in tbe native languagt: will be used to producing a fort:ign language: IFlegt: 1981; Flege and Pon 1981).

Phonological filtering may be at least partially responsible for cross-language phonetic tnte:rference, for knowledge of the native-language phonology may lead a second-Ianguagt: learner to ignore purely phonetiC differences between languages.' Bloomfitld fI933:79l, for t:xample, proposed that spc=ake:rs learn to attend to JUSt those features of speech that are linguistically distinctive, ignoring other purely phonetic differences which are " redundant" or non-distinctivt: (see also GIbson and Levin 1975, for a more recent statement of this position). If this is true, language lea.rners may fa tl to modify pre-existing phonetic habits when faced With cross­ language differences that are nOt linguistically disonctlve: in their native language because they do not puce:ive such differences. For example, an Ameriun may make the phonemically long vowels of Swedish much shorter than do most Swedes- producmg them with relatively shorter durations which approXimate the durations of spectr.dly Similar [ng.ll~h vowels- because vowel duration is not "phonemiC" (i.e ., hngulstically relevant) in English fJon.asson and McAllister Mimicry of lIon·diSllnctn'e phonetic differences I Flcge aJ Hummond 3

1972). On th~ oth~r hand, I~arn~rs may b~com~ aware of purely phonetic differ­ enc~s ~tween their nativ~ langu:lge and a foreign language in the process of leamjng rhe foreign languag~ . U this were the C:.llse, one might conclude th:u non· authentic pronunciation of foreign languag~s h:.lls primacily an articulatory rather than perceplUal origin, resulting perh2ps from general neurophysiological constuints on the acquISition or modification of fine motOr skills rath~r than from perceptual limiuuons Imposed by phonologtcal filtering or previous linguistic e~rience Is~e Strange :.lind Jenkins, 19781.

The present study was conducted in the hope of shedding some light on the ability of spe:lkers to detect phonetic differences between languages. If it can be demon­ strated that speakers in a language contact situation are aware of purely phonetic Ii.e., "redundant" or "non·distinctive"l differences between language varieties It would suggest th.u phonological filtering does not impose an insuperable barrier [Q phonetic le:lming in second·language acquisition. Two phonetic differences be· tween English andSpamsh provided a means by which to assess the hypothesis that speakers are able to detect non·distinctive phonetic differences between language varieties.

Spanish-English Phonetic D ifferences One way the voiceless StOps of Spanish and English differ is according to VaT [voice onset time). English /p,t,kI are produced with "Iong·lag" VOT values, whereas voicing begins soon after stop release for the unaspirated "short-lag" /p, t,kI of Spanish ILisktr and Abramson 1964; Williams 1977). As a result, native English speakers often ttnd to produce voiceless SpllOish StopS wuh VaT values that are tOO long Ii.e., with too much aspiration; Stockwell and Bowen 1965) willie Spanish speakers predictably under·aspmne English ip,t,klIWiIliams 1979; cf. lones 1948; Flege 1980).

There are several reasons to thmk that the kInd of cross·language differencts which lead to inappropriate VaT values during pronunciation of a foreign language will not be linguistlcal1y relevant (i.e., lead to the perception of segmental substitu· tions) for native·speakers expt:riencing a foreign accent. The distinctive difference between English Ip,t,kI vs. Ib,d,gI is ordinanly considered to be one of "voicing" or "tensity" rather than one of degree of aspirationorVOT. Although VaT may in fact be used to distinguish between natural classes of SlOpS in English and Other languages le.g., Ip,t,kI vs Ib,d,gI), VaT IS not equivalent to the distinctive linguistic feature "voicing" or "tensity". In fact, degree of aspirauon is ordinarii y considered by phonologJsts to be a redundant property of English stopS because It is predictable from phonetic context Isee POrt and Romnno 1979; Weismer 1979a).

It IS well known that mampularion of the VaT value of synthetic CV syllables (such as 1p:aJ vs /bal) will have an important effect on whether the syllable is ,udged to be initiated by a voiced or voiceless Stop Isee, e.g. Abramson and Lisker 1970, 1973). However, other acoustic dimensions which co· vary with VaT in natural speech may also importantly affect StOP voicing judgments made by English speak· ing listeners in the abSence of clear VaT differences between StOpS (Lisker 1978a). 4 Studlu in ~GOtld ulll1!//Qg/! ACI/IIISIIU)IJ, 5, No " /9H2

Thus il is not surpnsmg that an English s~akcr mlglll heaT :l short-I.;Ig li.c. vOiceless unaspiratedl SlOp produced by a deaf speaker of English (Monsen 19761 or by an En&lish·learning chdd IEilers and Oller 1976) R. Eders, personal commUni' cation):IS phonologicallyvOJuJes.s fi.c" /P,t,kli even though Its VOTvaluc might be less than the approXimately 35 msec needed to cue voiceless SlOp ludgments In pe:rcepuon experiments usmg synthetic speech stimuli. Similarly, phoncuc:ally naive English ~kers are strongly inclined to labd unasplu.tcd Dutch or Korean stopS as phonologically vOlcclcssli.c., as Ip,t,kll. even though they are produced with VaT values which are much shoner th.:m those typic.. ) of corresponding English voicelc55 stops (Liska 1978b) d. Brihe 19661.

Even if VOT or degree of aspiration were a distlnctlvt! rather than redundant propeny of English Stops, inappropriate VOT values in foreign·accented English might not lead to the perception oi linguistIcally relevant differences. In most natural communicative lettLngs context will serve to disambigu~te an ambiguous phonenc contt~s[ between minimal p&irs hke "to" VI!; "do". At the same time, Il:luve English speakers might be expected to ad.1pt their perceptu:.I1 ex-pe:ct~uons when listening to Spanish speakers of Enghsh. It tus been shown th:.lt Spanish· English bilinguals modify their judgments las " VOiced" or" voiceless" l of synthetic CV syll.1bles dilfenng an VOT as a function of whether they are listening in a Spanish orin an Enghsh mode for "set") to the stimuli (Elman, Diehl and Buchwald 1977; d . Monsen 1978).

Another phonetic difference between Spanish and English pertams to the duration of syllables occunng in utterance-hnal position. In English the vocahc nucleus of utterance·finallpre-pausall syllables may be 60-80% (about 100 msecllonger than the same vowel occumng an utterance-medial positions. Allhough thiS prosodic charactensuc of English IS acquired at an early age by chddren (Smith 1978) and has been found to be important for English speech to hi! ludged as " ruuural" sounding ICarlson, Granstrom and Klatt 19791, it does not affect propoSItional meaning.

The magnitude of final-syllable lengthening is known to be considerably smaller In SJnnish than English IDeiattre 19661 Oller 1973, 19771. As a result of cross· language phonetic interference (Flege and Port 19811 native Spanish s~akers seem to produce somewhat less final·syHable lengthenmg In English than do :1auve speakers of English (Pinkenon 19731.

For the present study it is Important to note that both the VOT and final -syllable lengthening differences between English and Spanish: III fall withm the range of duration differmces which arc detectable by human listeners; 121 are considered " redundant" or "non·distlnctive" phonetic ddfen:nces In English ILe., for monohngual Enghsh s~akersl l and (31 are more aptly descnbc:d by scalar phonetic features th.n with bmary dlStmctlve features. U phonologJcal flltenng does nOt serve to obscure phonetic differences like these, both phoneuc djfferences should be detectable by English speakers learning Spamsh and by English speakers who are experiencing Spamsh·accented English. In the present study a delayed·mlmlcry paradicm was used to assess awareness of these two phonetic differences by Amer­ icans familiar With Spanish·accented English. Mimiuy IIf Ilpn-disl ill(l JI'C phOIll: l h: diffcre /H':cs I FJ.:gL' u J I-h!l]lJllOlld 5

Methods The tas k used in t he present study c.:J!led (or the del.:Jyed mimcry li.e., from m emory l of a variety of English rather than the direct and immediate imitation of so me externally present model. College-aged su ject , alt English native speakers, were instructed to rcad a ty ped list of21 Engl ish sentences with what subjectively seemed lO them to represent a "typica l Spanish accent". That is, they were to produce English sentences as best they could like " someone whose native language is Spanish" without introducing unnatural pauses to convey rhe impression of accentedness. N o demonstration of Spanish-acc nred En gli sh was given, nor were any explicit instructions provided concerning how one might produce the eHect of accentednes .

We can be reasonably certain that t he speakers in t h is study were familiar with Spanish-accented English si nce al l were enrolled at the ti m e of the study in a beginning-level Spanish language class taught by a native Spanish speaker who spoke English with a "Spanish accent " . In addition, speakers' r sponses to a language- background questionnai re we re used to further insure that subjects had been exposed to Spanish-accen ted Engl ish . From rhe original population of 125, fi fty speakers (25 males, 25 females) were random ly chosen who indicated that they: (II were ac tually acquainted wi th nati ve speakers or Spanish; and (21 had lived in Fl orida (home to m any Sranish speakers) (or at least five years.

The mim ic ked sen tences were all of rhe rL'rm "T ht: __ is on tilt: __ ." The sentence-m edial and -fina l blanks in the carrier phrase eac h contained one of the ClC)V C English words shown in Tab!..: I (e .g. "The boo);. is on the wig. "J. The lexical items were chosen to include thn_'e tokens c.lch o f English sounds thought susceptible to m ispronunciation by Sp'lI1ish speakers of English . ~ T he apparent sllbstitution of one Enelish Ullno by ;.I n()th.;-r English sound may be rhe aspect of Span ish·accented Engli sh mos t evident to monolingual Engli s h speakers. It is known that estimates concerning the degree of Sp<.lnish ac ccntedncss m ade by native English speakers are correlated I\" i t h the nt! m ber of sound substi.tutions present in the speech sample (B rennan, Ry;.ln, aod Dawson (975).

Lexical Items Substitute Frequency nose, cheese, hose s: 14 1 (47%1 vice, veil, ~' ase b \. 129 (4 3% 1 fi g, pig, wig i I 12 7 (42% 1 book, hook, crook uU 61 (20%1 shell, sheet, sh eep c S 49 (16%) bean. phone, bone 1")0 1 (0 .3%) l .J pe,wbe, (oad dt 0 (0.0%1

Table 1. Lexical items inserted into the cJrricr ph rase " The _ _ is on the __ " At right, the number of sound substituti ons produced by SO native English speakers reading the test sentence with a Spanish accent. 6 Sludies in Second Lang uage Acqlll. ilion. S. No. 1. 19H2

rn addition to Engl ish sounds having predictable sound substitutes in Spanish­ accented English, three words with pre-vocalic ItJ (c ape. lOad. and tube) were also included in the test sentences. T hese items provide d a m eans of determining whether a non-distinctive phonetIc diff erence between u naccented Engli sh and Spanish-accented English would lead to a segmental substi tu tion ([ dJ for JtJ), a phonetic modification lit] for [thl )' or no modification a t all by native.: English speakers m imicking Spanish-accented English. At the sam e ti m e, Ihe presence of each word in both sentence-medial and sentence-fi nal posi (ion permi tt ~d the quantification of final-syllable lengthening in the speech of Americans mimicking a Spanish accent.

Speakers recorded the test sentences on Wollensak cassette tape recorde.: rs through fixed head- set m i cr o phon e~ while seated in alternate booths in a language labora­ lOry following a mid- term oral examination in Spanis h. The record level of each machine was set before the expe riment so that recordings for each speaker were of approximately eq ual intensity. Speakers paced themselve , and were permitted to repeat any sentence with which they were unsatisfied .

Two phonelically-trained listeners independenrly transcri be d one sound in each of the key words found in the tes t sen tences. One judge OEF) is a native speaker of English who does nm speak Spanish; the other (RMH) is bi lingual in Spanish and English . Transcriptions by the twO judges agreed in the great majority of cases (91. 2%). The few cases (2.2% ) where agreement could not b reached aft er further listening were subm icted to a thi rd ju dge fo r a final decision.

The frequency with which a substitute was heard for the Engl ish target sounds Izl, lvi, I I JU I, Isf, Inl and Itl are presented in Table 1. Speakers produced a tota I of 508 segmental sound substitutions. Not included in Table J are the el even cases where some other segm ental substi tution was heard.

The two judges heard no segmental substitution for ltl. Th is stop was transcribed as [(hI 70% of rhe time. and as a voiceless unaspirated stop Ut il 30% of the time.

The frequency with which speakers produced segmental sound substitutions while mimickinga Spanish accem was used as the basis fO Tselecting t\\'o sub-groups from tbe larger popul ation fo r an instrum ental phonetic anal ys is. ft was hypothesized that speakers who de monstrate relatively grea t' awareness of segmental propert ies of accented spe ech might also be most aware of non ·segmen ta I phonetic differences between language varieties. Accordingly, sentences of the ten speakers (Group Al l who produced the largest number of segmen tal substitutions (ranging from 17-27 per speaker) were considered separately from [he sentences produced by the ten speakers (Group A21 producing the fewest segmental ubsti t u lions (0-4 per speaker). A control group (Gro up U) of ten native English speakers (5 males, S females) producing unaccented English was also established. The e speakers, si m i­ lar in background and age to speakers in G roups A I and A2 , read the same lis t of 21 entences produced by speakers m imicking Span ish accent. The control group was recorded (Ampex. Model 602) in a sound-m:arcd bQoth j a constant mouth-to­ microphone distance was insured by m eans of a cephalostat. MlmicTY o{ non - (Ii.~lmclive pllonelic di{{ere1ll: c..( I FJc8c ttl Hammond 7

The original recordings of speakers in all three groups were dubbed onto a Crown !Model 700) tape recorder for spectrographic analysis {Voiceprint, Model 700)_ Any variatiun in the strength of the origin:l! signal was largely neutf:l.lized during the dubbing process. Only the six sentences with tape, lube, and toad occuning in the sentence-medial or sentence-final blank were analyzed.

Three acoustic intervals were measured by hand to the nearest five msec from the spectrograms: 11) Voice-onsellime.IVOT) of Itl was measured from the beginning of the transient noise burst signalling StOp release, to the first vertical striation signalling onset of phonation; 121'vowel' dUTUlion of lowl, luwl and leyl was mea­ sured from the onset of phonation following release of Itl to the offset of energy in the region of F2 and F3 signalling implosion of a following StOp; and /3j'senrence' du ration was measured from the beginning of the vowel in the sentence-medial test word to the end of the vowel in the sentence-final test word. This provided six measures each of 'sentence' duration, VOT, and 'vowel' duration for each spe.aker.

Results The tWO groups of speakers mimicking Spanish accent (AI, A21 differed from speakers producing unaccented English [Group U) on all three phonetic dimensions that were examined. The mean duration of phonetic intervals [VaT, 'vowel' and 'sentence') produced by the three speaker groups are presented in Table 2.

Speaker Group

Position AI A2 U Medial 44(15) 54120) 81116) VOT Final 46/21) 63)251 89114) Medial 194(44) 202155) 141(24) VOWEL Final 184(42) 2091 4 8) 217135) SENTENCE 1208)1801 1182/270) 9991104)

Table 2. Mean duration in msec of phonetic IIltervals in the speech of three speaker groups. Each mean is based on the averaged values for ten speakers. Standard deviations are in parentheses.

First, the effect of speaker group on VaT was siglllficant IFI2, 27)= 14.09, p = .(X}(}6). Figure 1 displays the VOT values measured foreachof the stops produced by speakers in the three groups. Nott: that VOT values in the StOPS produced by speakers in Group Al and Group A2 ranged from 10 msec to over 110 msee, and that no Stop produced by speakers of unaccented English IGroup UI had a VOT value of , Swdies In StCo1II1 LtlllgWIKC AttlurSI/IUII, 5, No. 1, 19H2

less th:m 50 msec. In fact, the most frequent VOT value for Stops produced by speakers in both Group Al and Group A'l(30-J5 msecl was less than that for tiny stOp produced by speakers in Croup U.

Group At Group A2 _ Group U "

". " "c "C- "~ - 5

0- 10- 20- )0- 100 - 50- 60- 70- 30· '0- 10Q- 1'::10* ) 15 25 JS ~5 5i '5 15 ~5 ~5 105 VOT in msec

Figure 1. Frequency of VOT values measurc:d In StOpS produced by thn:c: groups of speakers.

The mean VOT of speakers producing unaccented English (85 msecl agrees closely with values reported previOusly for English StOpS m a similar phonetic context le.g., Summerfidd 1975; Pon and Rotunno 19791. However, as can be seen in Figmc 2, speakers OlimlcklOg a Spanish accent (Groups AI, All produced It} with VOT values avenging about 30 msec shorter than those produc~d by spnkers in Group U.

Post-hoc tests revealed that the VOT values of both sentence-medial and sentence-final StOpS produced by speakers m Croup U were greater than those produced by speakers In Group AljTukey's HSD, alpha - .011 . The speakers who produced fewer segmenul subslHutions when mimlckmg Spanish accent !Group A1Jdid not differ from Group U, nor was there a difference between the tWO Spanish accent groups (AI, A21.

The three speaker groups were alike in producmg StOpS with slightly longer VOT values in sentence-linal than in sentence-medial words [Ff 1,17) = 4.72, P ~ .008]. ""filiI/cry of IItJ1Hh

u w ~ E .s of­ >

Al A2 u speaker group Figure 2, Mean VOT In n1SCC 10 StOps produced In sentence-medial and se ntence-flOal words by tlu ct: groups of sfl'!'a kers. Each bar is based on 30 measure, ments_

The size of thiS effcct-dboul 6 m:5CC or II %-1:5 sl mli.ar to that noted 10 a prcvlOus study by Summeriield p97S1.

Second, Figure 3 rC\'t'als tlUt the three spt'akcr groups diffcrL'd in temlS of final­ syllable kngtheOing. Group U made vowelS:1O aVl'ragc 76 mscc !54% l longer in semcncc-final than semence-ll1l.:dml words. Spcakcrs in Group A2, o n the other hand, madc fill31 \'1I\\'ds only 7 mscc (-t%l lungl'r in sentence-final words; and vowels pruduccd III scm cncc-final words by the mhcr group u f speakers minllcking Spanish accent (A I) actuall}' an'raged 10 mSt'C (fi% I slwlte! than sentence-medial vowels_

The lack oi final-syllabic lengthening m St'ntenecs pruduced by the speakers mimicking SpaOlsh aecent appears 10 be the basis fm the signiiicant interaction of the facto r S~akc r Group IA I, Al, or UI with tht' facwr Position-in-Se ntence imcdiai vs fma ll on the mean duration of \'owds [FIl, 271 = 14_04, p =' -000071. Post-hoc te5151Tukey's HSD, U -01) rcvl.';Jlcd thai only the speakers producing unaccented English IGruup UI madt' sentence-fmal vow!.'ls longer than sentence' medial vuwels, ThiS seems to bI.' due pnm.lril)' to tht' fact that the sentence-medial vowels produced by spl.'akcrs mllnlckmg SPdlllSh :lccl.'nt IGroups AI, and A21 ""ere !>ignlficantly 1 0n~l.'r than lim!>1.' uf Gnlup U Ip < .0 11_ Tht:rt: were no between-group differences fur sentcnCl'·flO31 vtJwcl~ S",dle.~ In Second iAnJ:IIQ"c ACl/iII~HIOII. 5, Nu. J. 19H2

ulleren<; e - medl.. , 0

uu • • .,nce- U IInlll G ~ E c • c o ~, '0 0; ~ >

Al A2 u speaker group

Figurr: 3. Mean duration in mscc of vowels found In sentencc-ml'dla! and sentence-final words produced by th ree speaker groups. Each bar IS based on olD m easurt:ments.

Finally, insIH=ctlon of Figure 4 reveals a difference between groups In the mcan duration of sen tt=nces{FI2, 27) == 3.34, P "" .05021. Sentences prod uced by Croup Al and G roup A1 averaged ahom 200 msec (20% I longer than scntences produced by speakers in Group U. Post-hoc tests revealed that Croup A I but nm Croup Al produced sign ifica ntly longer sentences than thl! unaccented grnup lUI. Th,S'd lffer­ ence may ~ due in part to the longer durallon of uttcrancc·mcdi.. l vowels pH1duced by speakers m imicking Spanish acceOi. It may :.I lso result from the presence of pauses. Twelve sentences produced by speakers In Group A I had III be excluded from analysis btC:.luse the presence of pauses c:.lusc:d [henllO exceed the length of a spectTogram I2.S secl. However, a finer differeOliation of sentences according to the presence or absence of pauses was nO[ attempted because of the difhculty of distinguishing fluen t from non·f1uent pauses.

Discussion The present study demonstrates the ability of speakers to modify nun·segmental phonetic dimenSions of theu speech when m imicking a vanety of English different from their own native dialect. Spanish speakers of English may produce English MimJcry of JJO/Hlis/JIICllllf: /l/lOn!!IIC ,li"eICtlcl:s I Flege.v Hammond II

StopS with VOT valul!s which are shorter than those typically seen in stops pro­ duc~d by nativ~ English speak~ r s (Wi lliams 1979). Spanish speakers of English are also known to produc~ a lesser magnitude of fmal-sy llabl ~ h:ngtheoing than do nallv~ sp~a k crs of English (Pinkerton 1973)_ 80th charact~ ri s tics of Spanish­ acc~ntcd English appear to result from the carry-over from Spanish into English of Spamsh-specific phonetic patterns (Flege 1980; Flegc and Port 198 1; d. Jones 1948).

1200

um ~ E c c .2 " . -~ " m "u c ~ m" -" '"

A1 A2 u speaker group Figure 4. Mean du ration in msee of sent~nc~s produced by three speaker groups. Numb~r of s~nt~nces analyzed was 60 for Groups U and A2, 4R for Group A I.

The modihcanon of both VOT and final -srllab1e lengthening by English native speak~ r s asked to mlmick Spanish-accented English suggests that speakers may be awar~ of non-distinctive phonetic differences d IStingUishing their own native dialect from otherdial~cts or language vaTlt.'t]es. It Seems unlikely that Ihe speakers of this study could have produced the kmd of phonctlc modifications seen here without some form of perceptual "awareness".

However, several objections might be made againSt Interpreting the data of [his study to mean that speakers are able to nllmick purely phonetic differences be ­ tween language varieties. FIrst, spcaking rare]s knoh'n to innuence VOT and vowel duration, so that the dfects observed Ln thIS study may be an artifact of between­ group differences in speaking rate. HO\".c\,er, r:ne dIfferences probably cannot account for the observed tendency of speakers mlm]cking Spa nish-accented Eng- " Sludie.s in SeCOlll1 L"nglw,!(l! Al;qlll.~H/lm. 5, No. I. lo,lR2 lish to produce /11 with VOT values that are shorter Ihan usual for English. As speaking rate decreases and phonetic intervals increase, VOT values tcnd also to increase (Summerfield 1975; POrt 19761. Speakers mimicking Spanish accent made sentences about 20% longer than speakers producing unaccented English. If f3tC were responsible for between-group differences in veT we should hav\.' observed longer VOT values for Croup Al and Group A1 tnan for Croup U. But what \"'3S actually observed was a substantial decrease in VOT by speakers mimicking Spanish-accented English. Nor is it likely that a more localized change in r:HC--­ that is, a difference in the duration of vowels immedilncly following the VOT intervals measured-can explain the between-group VOT differences. VOT valut.'s tend to be relatively long before vowels lengthened by phonctic f:lctors !pan and Rotunno 1979; Weismer 1979al_ The utterance-medial vowels produced by speakers mimicking Spanish-accented English in this study were significantly longer than those of speakers producing unaccented English, yet the VOT values of StOpS directly preceding these vowels were signiflcandy shorter.

Second, one might argue that the absence of final-sylLable lengthenmg in tht.' English of native English speakers mimicking a Spanish accent may not truly be due to a suppression of this prosodic characteristic of English. The absence of final lengthening in sentences produced by speakers in Group Al and Group A2 derives primarily hom the relatively great h:ngtb of their sentence-medial vowels ramer than to the shamming of their utterance-final vowels Isc!! Figure 31. Perhaps speakers prolonged sentence-medial vowels while planning the segmental sound substitution they would produce later in the sentence as pan of thcir cffon to effect a Spanish accent. This, rather than an intent to alter the Tatio of semencc-final and sentence-medial vowels, might explain the absence of final lengthening by speakers mimicking Spanish accent.

Anomer possibility is that speakers in Group Al and Group A2 sometimes iOlro­ duced shan pauses after sentence-medial test words, thereby placing medial \'ow­ els in pre-pausal position. Pauses following sentence-medial vowels were, in fact, perceptually evident in a number of cases. Accordingly, spectrograms were re­ examined to determine if there was temporal acoustic evidence for pauses follow­ ing sentence-medial test words. In about one-third 122J601 of the sentences there was a delay of more than 200 msec between the end of the vowel of the medial test word and onset of the following vowel lin "is"l_ This included cases where sp~akers added a vowel to the test word (e.g., [tep" iyl "tape", (tad... ] "tOad"l. Using 200 msec as a conservative criterion lor the presence of pauses, data for the eight speakers who provided no perceptual or acoustic evidence of IOtra-sententl3l pauses (4 from AI, 4 from A21 were re-analyzed. Their sentence-linal vowels averaged 190 msec (SO = 59 msecl; and their sentence-medial vowels averaged 175 msec (SO = 64 msecl_ Although this represents somewhat more fina l lengthening than seen for ei ther Group A I or Group A1 taken as a whole, it is considerably less than the 76 msec 154%1 of final lengthening produced by speakers of unaccented English (Group VI. Moreover, it does approximate the magnitude of finallengthemngfound in languages other than Eoglish such as Spanish /Delattre 1966; Pinkerton 1973; Mlmla)' of /l(1II ' (/'SIII1I: III'C 1,/" ,m'lIl' tldrcrel1l': I: .~ I Fk...:./! iiI Hammond JJ

Oller 19771 . Thus t he presence of in tea·sententlal pauses probably cannot explain the h ck of final lengthening 111 sentences produced by speakers mimicking Spanish accent.

Fmally, It IS possible that spcakl-rs In thiS study adupreu a general "foreign accent" mode Instead of actually mimicking SpIJlli!;h-accen tcd En~li sh. Producing speech in a general "forclgn .:accen t" mode might cnnsist in part of reducing the physiolog­ ical complexity of sIX-ech sound production. Fur example, it has been proposed that long-lag ILe., voiceless aspuatcdl SlOpS an' nl'urologlcally more complex than arc shan -lag {i .e., unasplratcdl StopS such as the Ip,t,kJ of Spanish {Kewley-Port and Preston 19741. It also :appears that the magnitude of fmal-syllable lengthening is considerably greater 111 English than in mher languages IDclattre 1966; O ller 19771 so th:u thIS phonetic dimcnsion might also hi.: considered relatively " marked" o r "difficult " for speakcrs. Thus \I would be of great intercst to determine if, for example, C ubans asked til mimic English-accented Spanish would produce m ore final-syllabic lengthening than is cha racterisllc of Spal1l sh or increase VaT in pre·vocalic StOps. Either finding would provide important additional support for the hypOthesis that speakcrs do detect purcl}' phonetic differences between lan­ guage varieties such 3S those which may In fact distinguish foreign-accented speech from that produced by native speakers.

Data from this study arc alsu relevant to a questllln ra ised by phuneticlans concern­ ing the nature of a speaker's control nvcr VaT In speech production. Stops arc generally regarded as falling into nne uf sen-ral modal ranges of VOT (the "lead", "short -lag," and " long-lag" categories I. In running speech, however, the StopS produced by individual spe3kcrs arc often internlediatc in value to Idealized modal catcgory valucs. It may be the case that thc iaT)'ngeal devoicing gcsture which rftsults III the aspiral/tlil associated with pre-vocalic StopS is "pre·programmed" {Weismer 1979bl, and thus not under the conscious or voluntary control of s~ake r s . Any overlap between VOT categories for stops observed in running specch could be conSidered to be the mdirect and perhaps ul1lntcntional conse­ qucnce of, for example v3natiU Il In phonetic context, stress, (l r speaking rate. But If, on the uther hand, speakers :Ire able to exert control ovcr the time course of thc laryngeal dcvoicing gesture and thus the yaT value which is ultimatdy measured lLlsker and Abramsun 1\J71 :77Hl speakers should also bc able to vary VOT In the absence of contextual v:malloos.

The present data show Ihal some sp..:akers. at kast, can produce StopS Wllh VaT values spanning the shon-Iag and long-lag ranges. ThiS supportS the observam)l1 that thcre is no physLcallllnLt

Conclusions This slUdy provides support for the hypOIhcsls that spcakcfs arc tacitly aware of purely phonetic differences distinguishing thClf native dIalect from Illner language variClics. Native English spc:l.kc:rs who mimicked Sp.mish-acccnu:d English in Ihc absence of an explicit external model either reduced or climin.lti.:d final -syllabic lengthening. a prosodic characteristic WillCh is much less prominent m Spanish and Spanish-accented English than It is In English. The speakers mimlckmg Span­ ish accent also produced Itl with VOT ... alues that were considerably sho n~' r Ihan is typical of English. They thereby approximated the shon-Iag StopS of Spanish and the frequently unde:raspirated voiceless stops that are typical of Spanish-accented English.

Further research using delayed mimicry is needed to determine If an alternative interpreuTion, namely that speakers in this srudy were producmg speech in a general " foreign accent" mode, might not better explain the present fesuhs. How­ ever there is reason to believe that the spt:akers in the present study actu::l. lly did mimick phonetic characteristics of Spanish-accented English. First, both phonetic dimensions examined here were modified in the expected direction, (that is, they approximated Spanish phonetic patterns). Second, in the segmental analysIS pre­ ceding this study, speakers produced many segmental substitutions associated with Spanish·accented English but very few that were not. Those who demon­ strated the greatest awareness of Spamsh accent (as determined by the number of segmental substitutions producedl also modified their speech most at the phonetic level.

Thus the present results suggest a tacit knowledge of phonetic Jiifcrences between language varieties is part of the phonological competence of speakers fam ilIar \.,.ith multiple varieties of language. Spcakers in thIS study seem to have been able to detect twO of the non·distinctive phonetic differences which dIstinguish Spanish from English. From this we tent3tively conc:lude that phonological filtering docs not make it impossible for second language learners to achieve 3 phonetically accurate pronuociauon of foreign languages.

Footnotes . Tho: au thors would like to thank personnel of rh(' Language Laboratory at the University of Flonda for their [ooperarion and Alan Burns, Caa!'1 C~nto:r. ~nd I"hn Lucke ftlr c::mnmems on an e~r1y draft of thLS ~rtic::lc . ThIS rcscarch was.fundo:d in p~n by NIH grant NS 07107 to thc Instllutc for the Adnnced Study of theCommunic::adon Processes anJ br NIH grant NS 07100 10 the Department of Cnmmunicauvl' DI5(Lrdcrs. Northwcstern Universuy.

Notes

I. Phonologlcal lihenng has much in common with thc nOtlon "cau:goTlcal percCptLon" discussed In sreech perception rC5ellrch. Speech snmull ue saId to be "cato:goncally" perceIved when a h,tenu's ~bdLty to dlscnmmate between any paIr of speech sounds is predicuble on the blisls of how each has been labelled (iden nficd l b)' tho: hSlo:ncr. Truly "categoncal" perceptIon predicts, for example, that two StimUlI whleh dIffer m VOT but are: none:thtIM$ both eonm!e:n!ly l.1obtlled .liS ItJ .... rli not bt reliably discnmmatcd, Mimicry 01 nnJl·c1is l i llr lil't' p/)(mellr d iffcrcnLl!.s I Flcgc 0) HamTllond 15

whereas another pair of slim u li di£(e ring by the same amount of VaT will be reliably discriminated from one another if o ne is olten I:Jbelled ldi and the othe( Ir). Speech perception research has shown, however, t ha I wi t hi n.category discrimination is gen­ erally bener than rro!dlcled by ca tegory labelli ng Isce Liberman, Harris, Hoffman, and Griffith 195 71 and can cven be im proved with training ICarne y, Widen, and Vierneis!er 1977 ). This is intuiti\'ely plausi ble. If such w ere not t he case it wouLd be impossible, for example, co detect a "distorted" production of lsi Or to detect some cross·dialect differ­ ences. 2. The possible substitution of one soun d by anmher in Spanish·accented English is to some extent understandable lh rough a comparison of Spanish and English. In Spanish there are no lax vowels such as English ru o r lUI; orthographiC ''\-'' represerHs either [b] or (b]. depending o n phonological environ ment; and [z i does not occur ei ther intervocalically or in utternace-(jnal posi tion. Further, in Caribbean dialects of Spa nish (~) and f!) may vary as surface realizations of ItJ, and finallnJ m ay be phonetically re3lized ilS (l)).

References Abramson, A. a.nd L. Li sker. 1970. Dlsc riminabi ltty ol ong the voicing continuum: C ross­ language tes ts. Proceedings of the Si x th International Congress of Phone tic Sciences, 569- 573. Prague, 1967. Prague: Academ ia. ____ . 1973. Voice liming perception in Spanish word·ini!lal StOps. Journal of Phonetics \.J·8. Bloomfield, L. 1933. Language. New York: Holt Brennan, E. , E. Ryan, and W, Dawso n. 1975. Scali ng of apparenl accentedncss by magnirude cs(imallon and sensory modality matching. Journal of Psycholinguistic Research 4.27-36. Briere, E. 1966. Art investig.ation of pho n ological interference. Language 42.768-798. Carlson, R., B. Granstrom, and D. Kla tt. 1979. So me notes on the perception of (cmporal pattems in sper:ch. Fr ontiers in speech communication research. ed . by B. Lindblom wd S. Ohman, 233·243. New York: Acade m ic Press. Carney, A., G. Widen, and N . Vi emeister. 1977. Noncatcgorical perceprion of srop consonants differing io VaT Journal Acou stical Society 01 America 62.961 ·970. Delanre, P. 1966. A comparison of syl lable lenglh cond,tloning among languagc:s. interna­ tional Review of Appli ed Li nguis ti cs 4 . 183· \98. Eckman, F. 1977. Markedness and the contrastjve analysi s h ypochesls. Language Learning 27.3l5·330. ____ . 19H 1. On the naturalness of in terl anguage pho nologic al rules. Language Learning 31.1 95·216. Ei lers, R. and O .K . Oller. 1976. T he role of speech discri minat Ion in developmental sound su bstitutions. Journal 01 Ch il d Language 3.319·329. El man, I., R. Diehl, and S. Buchwald. 1977. Perceptual switching in bilinguals. Journal Acoustical Society of America 62.971 ·974 . Fl ege, J. 1980. Phoneti c approxima tion in second language acquisition. Language Learning 30.117 ·134. ____ . 198 1. The phonological basis of foreign accent. TESOL Quarterly 15 .443·455. ____ and R. POrt. 19Rt. Cross-la nguage phonetic interference: Arabic to English. Lan· guage :l nd Speech 24. 125- 146. 16

Ffomkin, V. 1919. Pcrslsunt qUCS!Lons concc:rn;ng dlstlnctlve fcalUru. FroOlu:rs of speech communication research, ed. by B. Lindblom and S. Ohman. New York: AcademIc Press. C,b$On, E. and H Ltvln. 1975. The psychology of reading Dmhndgc' The MIT Press Jones, D. 194811972. An oUlllnt: of En&ll,h "honeue, Cambmlsc. Cambrldge Umverslty p,.,u Kewley·Port, O. and M. Preston. 1974. Enly lIpLcal SlOp ptoducllon. loum:!.1 of Phonetics 2.195·210.

udcf~d , P. 1980. What He lingulslic sounds made aU Language 56.485·502. Lado, R. 1957. Linguistics across cultures: applu:d hngulSucs for language 'c:lehtrs. Ann Arbor: Univcrslly of Mien;",n Press. unistc, I. 1970. Suprasegmcntals. Cambnd,c: MIT Press. L,bennan, A., K. Hams, H. Holfman and B. Griffith 1957. The dlscnmmauon of speech sounds within and across phoneme boundaries. Journal of Expcnmenul Psychology 54 .358· 368. Usker, L. 1978",. RabId vs. rapId: A catalogue of ",caustic fealUres th;1[ may cue ,he distlnc· tion. Haskins Labor"'toncs SUtus Report on Speech Resc:arch SR-54 . I27·131 ..,-_-,-_ . 1978b. Speech acrOal • !lngulslic boundary C""eeory namlne ",nd phonetic description. Hukms Laborllotones SUIU' Report on Speech ResraJl;h SR·54.105-111. ____ , and A. Abu.mson. 1964 . A cross·langu"'ge study of vOlcmg m mlllal StOrs: AcOUStical mc:.a5urements. Word 20.384-422. ____ . 1971. Distinctive fealUres and laryngeal control. Language 47.7fi7.7M5. Lovins, ,. 1975. Lo",nwords and the phonological SITUcture of Japanesr. Unrubhshed Ph.D. dissenallOn, Umveuuy of Chicago.

McAllister. R. and J. )onasson. 1972. Foreign ~ccent and umlng· An InStrumental phnncllc slUdy. Papc:rs from the fnstltute of Linguisllcs. UnlversllY of Stockholm 14 11 -40. Monsen, R. 1976. The productiOn of English StOp consonants In the speech of deaf children. Journal of Phonetics 4.29-41. Monsc:n. R. 1978. Toward measunne how well hearmg Impaned children srcak. Journal Speech and Hearing Research 21.197·219. Moulton, W. 1962. Towards a claSSIfication of pronunciallon errors. Modcm Language Jour· naI46.101 · 109. Oller, O.K. 1973 The effect of position 10 uneunce on spc:ech sc:gment durallon 10 EngJ.sh Journal Acousllul SOCiety of Amenca 54 .1235-1.247 . ____ .1977. Syll tlmlOg in Spanish. English, and FinnIsh. Current tssues 10 the phonetiC SCIences, cd. by H. Hollein and P. Hollein, .1)1 -345. Amsterdam: John Benl",mlnS Pmkenon, 5. 1973. The learning of English supra sc:gmental rules for suess and fln:!.J sylbblc.s by Sp:!.OIsh speakers. Paper presented at the Mld·Amenca Conference on Llngulsllcs. UOI' VCTSUy of Iowa, October, 1973. ED 098786. POri. R. 1976. The influence of spt:JklOg tempo on the dura lion of slressed vo ..... el and medIal stops 10 English trochee words. Unpublished Ph.D. d,ssen:!.tlOn. University of ConnectlcUl :-_--:--;- ' and It. ROlunno. 1979. Relation between VOlCe' On set' llme and vowel duration. Journal Acou31iCllI SoclelY of America. 66.654 ·662. Ritchie, W. 1968 On the explanatIon of phoniC mterference. language Lumlng IH . IR3-197

Shen, Y. 1959. Some allophones can be imrortant. Langu:age Learning 9 . 7 · 1~ . Mimll"ry of nOlHlis,illl";"t' pllUm'lll" cJtfh'fe/lCl's./ Fit'.":'c d' H,ItIUI)/IIl1l "

Smidl, B. 1978. Temporal as[>CClS of Engbsh spctth proJucllon "developmenul perspec· tive. Journal of PhonetiCS 6 ~17-l>7.

Stockwell, R. and J. Bow~n. 1'165. Th~ sounds of Enl:,lish :Jnt.1 Sr:lOlsl1, Ch,cago: University of Chicago Press. Strange, W, and J. Jenkins, 1978. The role of linguistic experience In the perctplion of spctch. Percep"lon and experience, cd. by. R. Walk and H. Pick, Jr., 12S· 169. New York: Plenum Publishing Co. Summerfield, A.O. 1975 . How a full account of scgmental pcrceptlon dcpend, on prowdy and vice·ver,a. Struclure and process In spccch perctptlon, cd. by A. Cohen and S. Nooteboom, 51·66. New York: Srnngcr Verlag. Trubeu;koy, N. 1969. Principles of phonology. Berkeley: Unl\'ersity (If California Press.

Weinreich, U. 1953. Languages In eOntilCt, findings and pWblems. The Hague: Mouton. Weismer, G. 1979a. 5ensi{Jvity nf vOlee·onset time (VOT\ measures to eenaJn segmental features in speech production. (ou rn31 of Phonetics 7.1\17·20-1. Wdsmcr, C . 1979b. Control of Ihe vOIcmg distinction fur Interyoe.. lic slaps and fric:nives: Some data and ,heorl:tlcal conSiderations. Journal of ('h"n~'l1cs 1'l...J27·-I3S. Williams, L. 1977. The VOicing eonnast In Spanosh. Inurnal of Phonelics S.169·IH4.

~ ___ . 1979. The modific.ation of speech pen:eptlon :lod production In secood·langu.age leaming. Perception &. PsychophYSIC5 26.95·104.