Brain & Language 121 (2012) 194–205

Contents lists available at SciVerse ScienceDirect

Brain & Language

journal homepage: www.elsevier .com/locate /b&l

Assessing the double phonemic representation in bilingual speakers of Spanishand English: An electrophysiological study

Adrián García-Sierra a,⇑, Nairán Ramírez-Esparza a, Juan Silva-Pereyra b, Jennifer Siard c,Craig A. Champlin c

a Institute for Learning & Brain Sciences, University of Washington, United Statesb Universidad Nacional Autónoma de México – Iztacala, Mexicoc University of Texas at Austin, United States

a r t i c l e i n f o

Article history:Accepted 29 March 2012Available online 23 April 2012

Keywords:Double phonemic representationLanguage contextsBilingual speech perceptionMismatch negativitySpanish–English bilingualsSpeech discriminationERPs

0093-934X/$ - see front matter � 2012 Elsevier Inc. Ahttp://dx.doi.org/10.1016/j.bandl.2012.03.008

⇑ Corresponding author. Address: Institute forUniversity of Washington, Box 357988, Seattle, WA 9

E-mail address: gasa@u.washington.edu (A. García1 We use the term phonetic to indicate sounds with d

articulatory patterns. The term phonemic indicates sounin the language.

a b s t r a c t

Event Related Potentials (ERPs) were recorded from Spanish–English bilinguals (N = 10) to test pre-atten-tive speech discrimination in two language contexts. ERPs were recorded while participants silently readmagazines in English or Spanish. Two speech contrast conditions were recorded in each language context.In the phonemic in English condition, the speech sounds represented two different phonemic categories inEnglish, but represented the same phonemic category in Spanish. In the phonemic in Spanish condition,the speech sounds represented two different phonemic categories in Spanish, but represented the samephonemic categories in English. Results showed pre-attentive discrimination when the acoustics/phonet-ics of the speech sounds match the language context (e.g., phonemic in English condition during the Eng-lish language context). The results suggest that language contexts can affect pre-attentive auditorychange detection. Specifically, bilinguals’ mental processing of stop consonants relies on contextual lin-guistic information.

� 2012 Elsevier Inc. All rights reserved.

1. Introduction perceive an acoustic event as two different speech sounds depending

It is well known that speech perception differs in bilinguals andmonolinguals (Flege & Eefting, 1986; Flege, MacKay, & Meador,1999; Mackain, Best, & Strange, 1981; Sundara & Polka, 2008).However, little is known about the way speech perception in bil-inguals is affected by the language they are using at the moment.The present investigation explores bilinguals’ ability to discrimi-nate speech sounds that have similar acoustic characteristics inEnglish and Spanish, but have different phonemic-meaning in eachof these languages. For example, the English /ga/ is acousticallysimilar to the Spanish /ka/. Therefore, Spanish–English bilingualsmay perceive an acoustic–phonetic speech sound based on the lan-guage they are using at the moment – bilinguals may develop twophonemic representations associated with a single acoustic–pho-netic speech sound.1 This phenomenon has been called bilinguals’double phonemic boundary or bilinguals’ double phonemic repre-sentation (Elman, Diehl, & Buchwald, 1977; Garcia-Sierra, Diehl, &Champlin, 2009). The term ‘‘double’’ indicates that bilinguals may

ll rights reserved.

Learning & Brain Sciences,8195, United States.-Sierra).istinct acoustic properties andds that are used contrastively

on the language context.The double phonemic boundary in bilinguals has been assessed

primarily in the perception of stop consonants. Voicing is one ofthe primary features that differentiates the stop consonants of alanguage. Stop consonants can take one of two voicing values;voiced or voiceless. Voiced and voiceless stops vary in the timingof the release of lip closure, or other constriction of the vocal tractassociated with the consonant, and onset of the voicing of the vo-wel. The release of the consonant is accompanied by a strong burstof air, or aspiration, which persists until the onset of voicing. Thetime period between the release of the consonant and the onsetof the voicing of the vowel is defined as voice onset time or VOT(Abramson & Lisker, 1970, 1972; Lisker & Abramson, 1970). There-fore, VOT is a temporal property that robustly specifies voicing instop consonants. Accordingly, by varying the amount of VOT, it ispossible to generate a continuum ranging from voiced/short-aspi-rated sounds (e.g., a clear ‘ga’) to voiceless/long-aspirated sounds(e.g., clear ‘ka’).

In the English language, voiceless stop consonants have largetime periods of aspiration (large VOT duration) and voiced stop con-sonants have short aspiration (short VOT duration). In contrast,Spanish language voiceless stop consonants have short aspiratedperiods and voiced stops are not aspirated, with voice onset beforethe release of the consonant. Voiced stops in the Spanish languageare also described as pre-voiced or negative VOT because voicing

A. García-Sierra et al. / Brain & Language 121 (2012) 194–205 195

occurs before the release of the consonant articulation. It is impor-tant to note that voiceless sounds in Spanish (/p, t, k/) have VOT/aspiration durations that are similar to voiced sounds in English(/b, d, g/). In fact, when monolingual speakers are asked to identifythe speech sounds from a voiced-to-voiceless continuum, the place-ment of the boundary dividing voiced from voiceless sounds willdepend on their native language. For example; a velar stop with+15 ms of VOT denotes a clear /ga/ for English monolingual speak-ers but, it denotes a clear /ka/ for Spanish monolingual speakers.

The fact that the placement of the phonemic boundary dependson the native language raises important questions regarding secondlanguage acquisition. For example, would phonemic categories inbilinguals resemble monolinguals’ phonemic categories? Or will bil-inguals develop independent phonemic categories for speechsounds that share acoustic–phonetic information but have differentphonemic – meaning in each of their languages? If the latter is true,then it would be indicative that bilinguals’ mental representations ofstop consonants depend on previous linguistic information (i.e., lex-ical, syntactic, semantic information of the language in use).Researchers have hypothesized that bilinguals can develop doublephonemic representations; that is, perception can change dependingon the language used at the moment. However, studies on doublephonemic representation in bilinguals have yielded contradictoryfindings. Although most studies have tested double phonemic repre-sentation using a speech continuum, there have been methodologi-cal differences, especially in the way that language contexts areestablished and in the use of monolingual control groups.

Caramazza, Yeni-Komshian, Zurif, and Carbone (1973) askedFrench–English bilinguals to categorize 3 VOT continua differingin place of articulation (/b-p/, /d-t/, and /g-k/). Language contextswere established by brief conversations in English or French beforethe perceptual task. The results showed no shift in bilinguals’ pho-nemic boundaries as a function of language contexts. However, bil-inguals’ phonemic boundaries were not aligned with those ofmonolingual speakers of English or French. These results suggestedthat the phonemic boundaries of L1 (first language learned) and L2(second language learned) are merged into a single category thatincorporates the phonetic inventories of both languages. Theauthors concluded that bilinguals do not shift between phoneticrules; rather they adopt a single rule that fits the needs of both lan-guages. The same finding was reported by Williams (1977), whoreplicated Caramazza et al. (1973) with bilingual speakers of Span-ish and English. Although some bilingual participants did showphonemic-boundary-shifts in accordance with language contexts,the language effect was not present in the group average.

In a follow-up study, Elman and colleagues (1977) proposedthat language contexts should be established by exposing biling-uals to the language of interest throughout the perceptual task. In-deed, the investigators presented precursor sentences in thelanguage of interest throughout a VOT identification task. Theresults showed support for the double phonemic representation:Bilinguals perceived identical acoustic–phonetic information dif-ferently depending on the language context. These results indi-cated that language contexts can affect speech perception if theyare properly established. This method was used also to test thedouble phonemic boundary in bilingual speakers of English andDutch (Flege & Eefting, 1987), and bilingual speakers of Englishand French (Hazan & Boulakia, 1993). Both investigations reporteda significant voicing-boundary-shift in accordance with the lan-guage contexts.

In a more recent investigation, Bohn and Flege (1993) suggestedbilinguals’ double phonemic boundary could be explained by theacoustic history of the language used in the precursor sentencesrather than by the linguistic information per se. For example, stud-ies have shown that speech perception is affected by the acoustichistory of neighboring sounds, even if the acoustic characteristics

of the target sound are remained constant (Diehl, Elman, & McCus-ker, 1978; Holt, 2005; Holt & Lotto, 2002). Consequently, theacoustic history of the precursor sentences could have affectedthe phonemic boundary placement in bilinguals. Bohn and Flegetested this hypothesis by asking Spanish–English bilinguals andEnglish monolinguals to identify members of a VOT continuumin two language contexts. A voicing-boundary-shift as a functionof language contexts seen only in bilinguals would indicate a dou-ble phonemic boundary in bilinguals. However, a voicing-bound-ary-shift as a function of language contexts seen in both groupswould indicate that the acoustic properties of the precursor sen-tences influence phonetic judgments. The results showed a voic-ing-boundary-shift in accordance with the language context forboth groups, suggesting that the acoustics of the sentences, andnot the linguistic information in the sentence, was responsible ofthe phonemic boundary shift.

Garcia-Sierra et al. (2009) tested the double phonemic bound-ary in Spanish–English bilinguals using a variation of Elman’set al. (1977) precursor sentence method. Precursor sentences weredelivered only 13% of the time to immerse the participants in thelanguage of interest, without excessive exposure to the acousticsof the precursor sentences. The results showed that both groupswere affected by the acoustic history of the sentences, but biling-uals showed stronger boundary-shifts than monolinguals. Further-more, the magnitude of the phonemic-boundary-shift correlatedwith L1/L2 proficiency (confidence in using Spanish and English)in bilinguals, but not in monolinguals. In other words, large bound-ary shifts were correlated with high confidence in using Englishand Spanish. However, since both groups showed language effects,it is clear that the acoustics of the precursor sentences influencedboundary shift.

In the present investigation we assessed bilinguals’ double pho-nemic representation by means of Event Related Potentials (ERPs).In contrast to behavioral paradigms, ERPs provide a continuousmeasure of the processes occurring between a stimulus and re-sponse, making it possible to determine which stage or stages ofprocessing is affected by a specific experimental manipulation(Luck, 2005; see Shtyrov & Pulvermüller, 2007 for a review). We as-sessed speech discrimination in the form of the Mismatch Negativ-ity or MMN (Näätänen, 1992). The MMN reflects electric brainactivity associated with early auditory processing in change detec-tion (Näätänen, 1992; Näätänen, Gaillard, & Mantysalo, 1978;Näätänen & Michie, 1979). The MMN is elicited by presenting arepetitive sound that establishes an auditory memory trace for agiven sound. Then, a new sound that differs from the memory tracein frequency (or localization, or duration, or intensity, etc.) is pre-sented. The degree of deviance between the memory trace and thenew sound is reflected by the ERP amplitude, so that the MMN re-sponse increases as the acoustic differences between standard(memory trace) and deviant increase (Tiitinen, May, Reinikainen,& Näätänen, 1994). It has been shown that this ERP componentcan also be recorded in response to speech sounds (syllables orwords) that differ in phonetic properties (Aaltonen, Niemi, Nyrke,& Tuhkanen, 1987; Diesch & Luce, 1997a, 1997b; Näätänen,2001; Winkler et al., 1999a). Therefore, it can be used to testspeech discrimination of consonants and vowels. The MMN hasbeen used in cross-linguistic studies to assess discrimination ofspeech continua varying in frequency and time domains (Rivera-Gaxiola, Csibra, Johnson, & Karmiloff-Smith, 2000; Rivera-Gaxiola,Johnson, Csibra, & Karmiloff-Smith, 2000; Shafer, Schwartz, &Kurtzberg, 2004; Sharma & Dorman, 2000).

The MMN is an excellent tool to explore bilingual’s double phone-mic boundary because it is an early auditory brain response and isnot affected by participants’ attention (Näätänen, 1986; Näätänen,Simpson, & Loveless, 1982). For example, participants can read abook or watch a silent movie while the brain data is being collected.

196 A. García-Sierra et al. / Brain & Language 121 (2012) 194–205

As a result, the MMN offers a unique way to investigate bilinguals’double phonemic boundary because it allows establishment of lan-guage context (reading a book in the language of interest) withoutrelying on precursor sentences that can influence speech perception(Bohn & Flege, 1993; Diehl et al., 1978; Holt, 2005; Holt & Lotto,2002). Since the MMN precedes conscious perception (see Sussman,Winkler, Huotilainen, Ritter, & Näätänen, 2002), it can be used to ex-plore the effects of language context early in the perceptual pro-cesses of mapping acoustic/phonetic cues to categorical mentalrepresentations. Most importantly, recent research has shown thatthe MMN can be modulated by lexical, semantic and syntactic infor-mation (Shtyrov & Pulvermüller, 2002). Consequently the MMN canbe used to explore how higher-levels of information processinginteract with early levels of sensory processing to jointly determinewhat we perceive (see: McClelland, Mirman, & Holt, 2006; McClel-land & Rumelhart, 1981). Therefore, the MMN is uniquely suitedfor investigating bilinguals’ double phonemic representation be-cause does not require participants’ attention, it provides a measureof phoneme categorical allocation and it is sensitive to higher order-linguistic information.

To our knowledge there is only one MMN study exploring bil-inguals’ double phonemic representation. Winkler, Kujala, Alku,and Näätänen (2003) tested the effects of language contexts onthe MMN. Bilingual Hungarian–Finnish speakers were assessed intheir ability to discriminate two Finnish words /pæti/ (was quali-fied) and /peti/ (bed). These are perceived as two distinct wordsby monolingual Finnish speakers and allophones of one word,/Peti/ (Peter) by monolingual Hungarian speakers. Bilinguals’ worddiscrimination was tested in a Hungarian language context and in aFinnish language context. Language context was established bypresenting a word in Finnish or Hungarian, which was differentfrom the standard and deviant, on 1.3% of the trials. The research-ers expected bilinguals to discriminate the standard and deviantwords in the Finish language context, but not in the Hungarian lan-guage context. However, the results showed no MMN amplitudechange as a function of language contexts, suggesting no doublephonemic representation in bilinguals. The authors concluded thatlearning a second language establishes new phonemic categoriesthat are used irrespective of language context.

In a relevant study, Paulmann, Elston-Güttler, Gunter, and Kotz(2006) tested the effects of language context on the N400 – a laterERP component sensitive to semantic integration. Bilingual Ger-mans speakers of English were tested in their ability to suppressL1, while doing a semantic task in L2. The task consisted in makinglexical decisions of English words that are spelled the same in Ger-man but have different meaning in each of these languages (inter-lingual homographs, for example, ‘‘chef’’ means ‘cook’ in Englishand ‘boss’ in German). Contrary to the authors’ expectations, theresults showed no differences in the N400 with language context.However, in a similar study with bilingual German speakers ofEnglish, Elston-Güttler, Gunter, and Kotz (2005) provided more ro-bust language context by presenting the interlingual homographsin full grammatical sentences. For example, the prime word gift(denoting ‘poison’ in German) was presented in the sentence Thewoman gave her a pretty gift followed by the target word poison.Immediately after the target word, participants were asked to tellif gift and poison were related. As expected, these results showedan interaction between the amplitude of the N400 and languagecontexts, suggesting that it is possible to produce more robustlanguage context effects by visually presenting sentences in thelanguage of interest throughout the lexical task.

In summary, both ERP and behavioral studies demonstrate thatresults vary depending on methods used to establish language con-text. Three findings are of particular interest. First, short exposureto the language of interest, before the perceptual task or duringthe perceptual task, may not be sufficient to establish language

contexts (Caramazza et al., 1973; Paulmann et al., 2006; Williams,1977; Winkler et al., 2003). Second, if language contexts are estab-lished by presenting auditory precursor sentences, the acoustic his-tory of the sentences can influence the perception of speech soundsregardless of language status (Bohn & Flege, 1993; Garcia-Sierraet al., 2009). Third, it is possible to establish strong language con-texts by visually presenting sentences in the language of interest(Elston-Güttler et al., 2005). The success or failure in establishinglanguage contexts can be explained by what Grosjean (2001) con-ceptualizes as language mode. Grosjean proposes that bilingualsfunction along a continuum that reflects the state of activation ofa given language at a given point in time in their everyday activities.At one end of the continuum, bilinguals are in monolingual modeand at the other end of the continuum, bilinguals are in bilingualmode. In the monolingual mode, bilinguals use one language whiledeactivating the other language to the greatest extent possible. Inthe bilingual mode, bilinguals choose a base language and bringthe other language as needed. However and importantly, Grosjeanhypothesizes that bilinguals are governed by a ‘‘base language’’which controls language processing at any time. Therefore, in Gros-jean’s view, in order for language context to set a ‘‘mode’’, bilingualsmust be engaged in the language of interest throughout the percep-tual task or the base language will predominate. The present studywas designed to immerse bilinguals in the language of interestthroughout the perceptual task to keep them in monolingual mode(i.e., Spanish mode in the Spanish language context and Englishmode in the English language context), but the immersion occursvisually and does not produce biases due to acoustic input.

2. Study overview

The present investigation explores bilinguals’ double phonemicrepresentation by means of MMN. The MMN is an ideal measure totest speech discrimination across language contexts since it doesnot requires participants’ attention. Participants were immersedin the language of interest by silently reading magazines duringthe ERP sessions. Two speech contrast conditions were recordedin each of the language contexts. In the phonemic in English condi-tion, the speech sounds tested represented two different phonemes(across category) for the English language, but represented thesame phoneme (within category) for the Spanish language. In thephonemic in Spanish condition, the speech sounds tested repre-sented two different phonemes for the Spanish language but, rep-resented the same phoneme for the English language. We expectedthat bilinguals in the phonemic in English condition would showbetter discrimination (larger MMN amplitudes) during the Englishlanguage context recordings than during the Spanish languagecontext recordings. The opposite pattern was expected for the pho-nemic in Spanish condition. That is, bilinguals would better discrim-inate the sounds, as indicated by the MMN, during the Spanishlanguage context recordings than during the English language con-text recordings.

A control group of English monolinguals was recruited to verifythat the speech sounds tested would produce the expected MMNresponses. For example, we expected that monolinguals would eas-ily discriminate the sounds used in the phonemic in English condi-tion, but not the sounds in the phonemic in Spanish condition. Themonolingual group was tested only in the English language context.

3. Method

3.1. Participants’ exposure to English and Spanish

All participants responded a background language question-naire that was designed to assess English and Spanish language

A. García-Sierra et al. / Brain & Language 121 (2012) 194–205 197

exposure from childhood to adulthood (see Garcia-Sierra et al.,2009). Nine bilingual participants were born in the US and twoin Mexico. All 12 monolingual participants were Americans bornin the US

3.1.1. Amount of English and Spanish exposureOur Spanish–English background questionnaire revealed a lan-

guage pattern in bilinguals that changed from Spanish dominant(ages 3–6) to English dominant (ages 18–21), transitioning througha period of relatively balanced exposure to both languages (ages 9–12).

From ages 3 to 6, 73% of bilinguals were exposed to Spanish100% of the time. From ages 9 to 12, 64% of the bilinguals were ex-posed equally to Spanish and English. From ages 18 to 21, 82% ofthe bilinguals were exposed more to English than to Spanish. Mon-olinguals reported being exposed only to English from ages 3 to 21.

3.1.2. Amount of English and Spanish useFrom ages 3 to 6, 64% of bilinguals reported using (speaking)

Spanish 100% of the time. From ages 9 to 12, 64% of bilinguals re-ported using equal amount of English and Spanish. From ages 18 to21, 64% of the bilinguals reported using more English than Spanish.Monolinguals reported speaking only English from ages 3 to 21.

3.2. Participants’ English and Spanish level of bilingualism

Proficiency in English and Spanish was assessed by self-reports.Participants were asked to rate themselves on four factors: profi-ciency in speaking, reading, writing, and listening comprehension.We report only speaking and listening comprehension. For speak-ing proficiency a Likert scale from 1 to 5 was used (1 = ‘‘I cannotspeak the language, I have a few words or phrases and, I cannotproduce sentences,’’ and 5 = ‘‘I have a native-like proficiency withfew grammatical errors and I have good vocabulary’’). The meansfor bilinguals were 4.73 (SD = .47) for English and 4.36 (SD = .67)for Spanish. For monolinguals, the overall means showed that Eng-lish was indeed their dominant language (mean = 5.00, SD = 0.0),while Spanish proficiency was poor (mean = 2.22, SD = 1.10).

For listening comprehension a Likert scale from 1 to 5 was used(1 = ‘‘I only understand few words of what is being said,’’ and5 = ‘‘I understand all of what is being said’’). Bilinguals showedequivalent mean scores for English and Spanish (M = 4.73, SD =.47; M = 4.37, SD = .47; respectively). Monolinguals showed a meanscore of 5.00 (SD = 0.00) for listening comprehension in English anda mean score of 3.00 (SD = 1.125) for listening comprehension inSpanish.

Only those bilinguals who reported in the Likert scale to be atleast 75% confidence in speaking and comprehending English andSpanish were invited to participate in the experiment. Monoling-uals, on the other hand, were invited only if they reported to be25% (or less) confidence in speaking and listening in Spanish.

3.3. Participants

The sample was intended to be as homogeneous as possible. Forthis reason only right handed females were recruited to participatein the investigation. Participants with auditory thresholds that ex-ceeded 20 dB at any frequency .25, .5, 1, 2, 4, 6, and 8 kHz received$5.00 and were excused from the experiment. Eleven female bilin-gual speakers of Spanish and English, and 12 female monolingualspeakers of English participated in the study. Nineteen participantswere retained for subsequent analysis because they showed clearevoked potentials, passed the hearing screen, and attended to bothERP sessions (i.e., four participants were excluded). The final sam-ple was 10 bilinguals (mean age = 19.73, SD 1.10), and nine monol-inguals (mean age = 20.33, = SD 1.97). All participants were

undergraduate students from the University of Texas at Austin.Participants were recruited by means of flyers and word of mouthand were paid $ 70.00 for two experimental sessions of approxi-mately 120 min each. Subjects gave written informed consent afterthe procedures of the experiment were explained to them.

3.4. General procedure

3.4.1. Language contextsA single language context was established in each experimental

session and context sessions were counterbalanced. Language con-texts were established by exposing participants to the language ofinterest before and during the experimental task. Experimental ses-sions consisted of 30–45 min of Quick-cap placement and 1 h ofdata collection. The interactions that occurred between the experi-menter and the participants during electrode-cap placement werein the language of interest. In the Spanish language context, theprincipal investigator (Spanish native speaker) spoke to participantsin Spanish. In the English language context, the research assistant JS(native English speaker) spoke to them in English. During data col-lection, participants were instructed to read a magazine in the lan-guage of interest. Each language context session was divided in twoERP recordings (two speech contrast conditions; see below). EachERP recording consisted of five data collection blocks (3 min long)and four relaxation blocks (5 min long). Data collection blocks andrelaxation blocks were alternated throughout the experiment. Dur-ing the relaxation blocks, the researcher asked the participants todescribe in the language of interest, the amount of Spanish and Eng-lish they used during everyday activities in a given weekday andduring the weekend. At the end of the 5 min conversation, theexperimenter resumed ERP data collection. The time between theend of a relaxation block and the start of data collection was1 min. According to previous studies a 1 min silence is sufficientto minimize the influence of acoustic history (conversation with re-searcher) on speech perception (Holt, 2005; Holt & Lotto, 2002).

3.4.2. StimuliThe stimuli selected were taken from a VOT speech continuum

synthesized by Garcia-Sierra et al. (2009). The speech continuumranging from /ga/ to /ka/ was synthesized in 27 VOT steps, and var-ied from �100 ms to +100 ms of VOT in 10 ms steps except from 0to 60 ms were it varied in 5 ms VOT steps. The continuum was gen-erated using the cascade method described by Klatt (1980). Eachstimulus was 500 ms long including 10 ms burst. Vowel durationvaried from 490 ms (0 VOT) to 390 ms (+100 VOT) depending onVOT duration. Only three stimuli were used to collect ERP data(see below). For the purpose of the present investigation, the vowelduration of the stimuli selected was shortened 123 ms. The F0 con-tour was kept constant across the stimuli. The total duration forthe stimuli used was 376 ms.

3.4.3. Speech contrast conditionsGarcia-Sierra et al.’s (2009) continuum was tested in a pilot

study of monolingual speakers of English living in the US (N = 15)and monolingual speakers of Spanish living in Mexico (N = 18).Fig. 1 shows that both monolingual groups perceived stimulus�20 ms and stimulus +50 ms as ‘ga’ and ‘ka’; respectively. In con-trast, stimulus +15 ms was perceived as ‘ka’ by Spanish monolin-gual speakers and it was perceived as ‘ga’ by Englishmonolingual speakers. Accordingly, these three stimuli were se-lected to establish the speech contrast conditions. Two speech con-trast conditions were tested: the phonemic in English condition andthe phonemic in Spanish condition. The phonemic in English condi-tion used stimulus +50 as standard and stimulus +15 as deviant.We expected that during the English language context, bilingualswould perceive stimuli +50 and +15 as sounds belonging to

Fig. 1. Identification of a VOT continuum by monolingual English speakers andmonolingual Spanish speakers. The continuum ranges from Spanish /ga/ (�100 ms)to English /ka/ (+100 ms) passing through Spanish /ka/ (+10 ms) and English /ka/(approx +25 ms). The arrows indicate the stimuli used to collect ERPs.

198 A. García-Sierra et al. / Brain & Language 121 (2012) 194–205

different speech categories (‘ka’ and ‘ga’; respectively). However,we expected that during the Spanish language context, bilingualswould perceive stimuli +50 and +15 as sounds belonging to thesame speech category (‘ka’ and ‘ka’). In the phonemic in Spanishcondition we used stimulus �20 as standard and stimulus +15 asdeviant. We expected that during the Spanish language context,bilinguals would perceive stimuli �20 and +15 as sounds belong-ing to different speech categories (‘ga’ and ‘ka’; respectively). How-ever, we expected that during the English language context,bilinguals would perceive stimuli �20 and +15 as sounds belong-ing to the same speech category (‘ga’ and ‘ga’).

3.5. Electrophysiological procedure

3.5.1. Stimuli presentationA classic odd-ball paradigm was used to collect the evoked

potentials. This paradigm consists in delivering infrequent stimuli(physically deviant) within a repetitive homogeneous stimulus se-quence (standard stimulus). The standard sounds occurred with aprobability of 0.85 (850 stimulus repetitions) and the deviantsounds occurred with a probability of 0.15 (150 stimulus repeti-tions). Two rules governed stimuli presentation: (1) The deviantsound could not occur two times consecutively and, (2) there wereat least three standard sounds between deviant sounds. The timebetween the offset of a stimulus and the onset of the next stimulus(inter stimulus interval) was 500 ms. An insert earphone (EARTone, model 3A 10 kO) was used to present the speech sounds.The peak sound intensity (dB SPL) of each stimulus was measuredwith a sound-level meter that was connected to a 2-cc coupler. Allstimuli were delivered at 85 dB peak-equivalent SPL, which is con-sidered a comfortable listening level.

3.5.2. Electrophysiological recordingThe electroencephalogram was recorded with Ag/AgCl sintered

surface electrodes using NeuroScan SynAmp amplifier (16 bit A/Dconverter), and Scan4.3 software from 32 inverting electrodes(Quik-Cap). The resolution of the SynAmp amplifier was set to0.168 lV (A/D accuracy). Electrodes were referenced to linked ear-lobes, and the ground electrode was placed 1.5 cm anterior to thecentral frontal electrode (FZ). All leads were placed according tothe 10–20 International System. Eye blinks were monitored withelectrodes placed on the orbis ocularis muscle above and belowthe left eye. Lateral and horizontal eye movements were monitoredwith electrode placed on the left and right outer canti, approxi-mately 2 cm later to either eye. An electro-oculorgram artifact cor-rection method developed by Semlitsch, Anderer, Schuster, andPresslich (1986) was applied to the data off-line. The recorded

electroencephalogram was digitized at 500-Hz sampling rate andfiltered using a band-pass filter with low and high cut-off frequen-cies at 0.05 Hz and 40 Hz, respectively. Electrode impedances weremaintained below 5 KO across language contexts and conditions.Epochs of 700 ms with a 100 ms pre-stimulus interval were de-rived from the continuous electroencephalographic recording afteroff-line filtering the data with a band-pass filter from 1 to 30 Hz.The ERP responses to the first ten stimuli in each block as well asepochs showing voltage changes exceeding +100 mV were omittedfrom the final average. The ERP averaging process was done by pre-senting a 1 ms trigger that was time-locked to the presentation ofeach stimulus (Stim2 Neuroscan Compumedics).

3.6. Number of ERP trials accepted

3.6.1. Phonemic in English conditionThe number of trials accepted for the standard sound – exclud-

ing the standards following deviants – during the English languagecontext was 324.0 (SD = 123.0) and 316.4 (SD = 125.5) during theSpanish language context in bilinguals. No significant differencewas found across language contexts (t(8) = .104, p = .92). The num-ber of trials accepted for the deviant sound during the English lan-guage context was 105.4 (SD = 22.1) and 98.1 (SD = 22.1) duringthe Spanish language context in bilinguals. No significant differ-ence was found between language contexts (t(8) = .82, p = .44).ERPs were collected only in the English language context for mon-olinguals. Therefore, no statistical comparisons were done acrosslanguage context. The number of standard trials – excluding thestandards following deviants – accepted for monolinguals was652 (SD = 101) and 102 (SD = 3) for the deviant sound.

3.6.2. Phonemic in Spanish conditionThe number of trials accepted for the standard sound – exclud-

ing the standards following deviants – during the English languagecontext was 337.1 (SD = 151.3) and 332 (SD = 131.5) during theSpanish language context in bilinguals. No significant differencewas found across language contexts (t(8) = �.063, p = .95). Thenumber of trials accepted for the deviant sound during the Englishlanguage context was 95.0 (SD = 25.0) and 93.1 (SD = 14.1) duringthe Spanish language context in bilinguals. No significant differ-ence was found across language contexts (t(8) = .18, p = .86). ERPswere collected only in the English language context for monoling-uals. Therefore, no statistical comparisons were done across lan-guage context. The number of standard trials – excluding thestandards following deviants – accepted for monolinguals was618.5 (SD = 136) and 100 (SD = 11) for the deviant sound.

3.7. Data analysis

3.7.1. Standard and deviant ERP amplitude analysisThe onset of the MMN was expected at 220–235 ms after stim-

ulus onset because the acoustic differences of the speech contraststested occurred in the first 20–35 ms (Näätänen, 1992). For bothmonolingual and bilinguals, the electric brain activity associatedwith phonemic discrimination was analyzed from 220 to 280 msafter stimulus onset. The amplitude for the standard and deviantresponse was calculated by averaging the voltage values from theERP time window (i.e., mean-amplitude). Mean-amplitude wasused because the measured amplitude is not biased by the numberof trials accepted in each of ERP-responses (Luck, 2005). The mean-amplitude of the deviant ERP response was compared with themean-amplitude of the standard ERP response. Baseline correctionwas computed by subtracting the average pre-stimulus voltage(100 ms) from the average voltage occurring in the time windowof interest.

A. García-Sierra et al. / Brain & Language 121 (2012) 194–205 199

Eight regions of interest were computed from the 32 electrodes,each containing the average of a group of three or four electrodes.The regions were Left-Frontal (F7, F3, FT7 and, FC3), Left-Central(T7, C3, TP7 and, CP3), Left-Parietal (P7, P3 and, O1), Mid-Fronto-central (FZ, FCZ and, CZ), Mid-Centroparietal (CPZ, PZ, and OZ),Right-Frontal (F8, F4, FT8 and, FC4), Right-Central (T8, C4, TP8and, CP4) and, Right-Parietal (P8, P4 and, FT8). In order to testthe effects of language context on speech perception in bilinguals,we observed changes between standard and deviant (ERP type)across the eight electrode regions of interest for the phonemic toEnglish condition (i.e., standard +50 ms of VOT; deviant +15 ms ofVOT) and the phonemic to Spanish condition (standard �20 ms ofVOT; deviant +15 ms of VOT). A 2 (language context; English andSpanish) � 2 (phonemic condition: phonemic in English andphonemic in Spanish) � 2 (ERP type; standard and deviant) � 8(Electrode regions: Left-Frontal, Left-Central, Left-Parietal, Mid-Frontocentral, Mid-Centroparietal, Right-Frontal, Right-Centraland, Right-Parietal) repeated measures ANOVA was performed.Greenhouse-Geisser epsilon (e) was used for non-sphericity correc-tion when necessary.

3.7.2. MMN Amplitude analysisResearch has shown that the scalp distribution of the MMN has

its maximal amplitude in fronto-central electrodes (Schröger,1998). Also, studies show that there is a MMN polarity inversionat electrodes positioned over the opposite side of the Sylvian fis-sure, such as the mastoid leads (Alho, 1995; see Deouell, 2007 fora review). Accordingly, in this study the MMN polarity inversionwas investigated to verify that the ERP-effects were the conse-quence of MMN generators. In particular, the fronto-central MMNresponse (FCZ) was compared with the average of MMN responseat leads T7 and T8 (nearest electrodes to the mastoids). The MMNwas calculated by subtracting the standard response from the devi-ant response (deviant-minus-standard). Dependent T-tests weredone independently for each condition and each language context.

3.7.3. Additional analysesTwo ERP time windows were analyzed. An early ERP time win-

dow (160–220 ms from stimulus onset) investigated the earlystages in stimulus-specific sensory integration. This ERP-responseis known as the N1 ERP-component and it precedes conscious rep-resentation of the stimulus (Näätänen & Picton, 1987; Näätänen &Winkler, 1999). The later ERP time window (280–480 ms fromstimulus onset) assessed an ERP positivity that in some instancesis followed by the MMN. The positive component is known asthe P3a and occurs about 280 ms after stimulus onset. Typically,individuals will show a P3a response when they detect a differencebetween standard and deviant while attending to another task(Escera, Alho, Schröger, & Winkler, 2000; Friedman, Cycowicz, &Gaeta, 2001; Squires, Squires, & Hillyard, 1975).

4. Results

Our first goal was to investigate the effects of language contextsand speech contrast condition on the amplitude of standard anddeviant responses in bilinguals. The repeated measure ANOVAshowed a main effect for ERP type (F(1,8) = 7.5, p = .026g2

p ¼ 0:50) with no significant main effect for language context orphonemic condition. The main effect for ERP type showed thatthe deviant sound produced a significant more negative ERP re-sponse than the standard sound (Standard Mean = �.20 lVSD = .76; Deviant Mean = �.51 lV SD = .75). Also, and more rele-vant to our research questions, the results showed a significantinteraction between language context, speech contrast conditionand, ERP type (F(1,8) = 10.0, p = .014 g2

p ¼ 0:55). This interaction

suggested that bilinguals’ deviant response differed from the stan-dard response as a function of language context and speech con-trast condition. This finding was further investigated by a 2 (ERPtype: standard and deviant) � 8 (Electrode regions: Left-Frontal,Left-Central, Left-Parietal, Mid-Frontocentral, Mid-Centroparietal,Right-Frontal, Right-Central and, Right-Parietal) repeated mea-sures ANOVAs. The ANOVAs were done independently for group,condition and language context.

4.1. Do bilinguals show an amplitude change depending on thelanguage context?

A double phonemic representation would result in differencesin the standard and deviant responses in the phonemic in Englishcondition and in the phonemic in Spanish condition across languagecontexts.

4.1.1. Phonemic in English conditionOur expectation was that bilinguals in the English language

context would perceive standard (+50 ms VOT) and deviant(+15 ms VOT) as belonging to different phonemic categories (/ka/and /ga/; respectively). On the other hand, bilinguals in the Spanishlanguage context would perceive both speech sounds as belongingto the same phonemic category (/ka/). Specifically, a significantamplitude difference between standard and deviant ERP-responsesin the English language context was expected, but not in the Span-ish language context.

4.1.1.1. English language context. Fig. 2A shows standard and devi-ant ERP-responses from the eight electrode regions. The overalldeviant-ERP response (Mean = �.94 SD = 1.20) was more negativethan the overall standard-ERP response (Mean = .10 SD = .80). Thisdifference was statistically significant (F(1,8) = 18.1, p = .003,g2

p ¼ 0:70) with no significant interaction between standard anddeviant responses as a function of electrode regions (F(1.8,15.0) = 2.0, p = .20, g2

p ¼ :20). There was a significant MMN ampli-tude difference between electrode FCZ and the averaged mastoids(FCZ Mean = �1.5 SD = 1.0; Mastoids’ Mean = �.44 SD = .60; t(8) =�6.0, p = .0003).

4.1.1.2. Spanish language context. Fig. 2B shows standard and devi-ant ERP responses for the eight electrode regions. The overall devi-ant-ERP response (Mean = �.37 SD = 1.23) was similar to theoverall standard-ERP response (Mean = �.23 SD = .71). Differencesbetween standards and deviants were not significant (F(1,8) =.30, p = .60, g2

p ¼ :036) and there were no significant interactionsbetween standard and deviant responses and electrode regions(F(3.0,24.1) = 1.5, p = .23, g2

p ¼ :16). The MMN amplitude inversionbetween electrodes FCZ and averaged mastoids was not significant(FCZ Mean = �.20 SD = 1.24; Mastoids’ Mean = �.22 SD = .74; t(8) =.11, p = .92).

4.1.2. Phonemic in Spanish conditionOur expectation was that bilinguals in the Spanish language

context would perceive standard (�20 ms VOT) and deviant(+15 ms VOT) as belonging to different categories (/ga/ and /ka/;respectively). On the other hand, during the English language con-text, both speech sounds would be perceived as belonging to thesame phonemic category (/ga/). Specifically, a significant ampli-tude difference between standard and deviant ERP-responses inthe Spanish language context was expected, but not in the Englishlanguage context.

4.1.2.1. Spanish language context. Fig. 3A shows the standard anddeviant ERP-responses from the eight electrode regions. Frontal re-gions produced a more negative response for the deviant than for

Panel A Panel B

FCZ

MID-CENTROPARIETAL

LATEIRAPTFELLATEIRAPTFEL

Phonemic in English ConditionEnglish Language Context

LATNORFTHGIRLATNORFTFEL

LARTNECTHGIRLARTNECTFEL

MID-FRONTOCENTRAL

StandardDeviant

MastoidMMN

FCZ

MID-CENTROPARIETAL

LATEIRAPTFELLATEIRAPTFEL

Phonemic in Spanish ConditionEnglish Language Context

LATNORFTHGIRLATNORFTFEL

RLARTNECTFEL IGHT CENTRAL

MID-FRONTOCENTRAL

StandardDeviant

MastoidMMN

Fig. 2. Bilinguals’ (N = 10) scalp distributions of the standard and deviant ERP responses in the phonemic in English condition during the English language context (Panel A)and during the Spanish language context (Panel B). The standard response is represented by the black line and the deviant response by the gray line. Negative is plotted down.At the bottom of each panel, MMN amplitudes from electrode FCZ (gray line) and mastoid substitute electrodes (black line; T6/7) are presented. The short black lines in eachof the boxes depict the time windows of interest (220–280 ms and 280–480 ms).

200 A. García-Sierra et al. / Brain & Language 121 (2012) 194–205

the standard, and the overall deviant-ERP response (Mean = �.26SD = .90) was more negative than the overall standard-ERP re-sponse (Mean = .21 SD = 1.12) in a significant way (F(1,8) = 10.0,p = .01, g2

p ¼ 0:55). There was no significant interaction betweenstandard and deviant responses as a function of electrode regions(F(2.3,18.3) = 1.0, p = .41, g2

p ¼ :11). There was a significant MMNamplitude difference between electrode FCZ and the averagedmastoids (FCZ Mean = �.86 SD = .72; Mastoids’ Mean = �.33SD = .36; t(8) = �3.0, p = .02).

4.1.2.2. English language context. Fig. 3B shows standard and devi-ant ERP-responses for the eight electrode regions. The overall devi-ant-ERP response (Mean = �.46 SD = .90) was similar to overallstandard-ERP responses (Mean = �.90 SD = .94). Differences be-tween standards and deviants were not significant (F(1,8) = 2.6,p = .14, g2

p ¼ :25) and there was no significant interaction betweenstandard and deviant responses and electrode regions(F(2.6,21.0) = .73, p = .53, g2

p ¼ :08). The MMN amplitude inversionbetween electrodes FCZ and averaged mastoids showed to be nosignificant (FCZ Mean = .47 SD = 1.3; Mastoids’ Mean = .15 SD =.40; t(8) = .90, p = .40).

4.2. Do monolinguals perceived the stimuli as belonging to differentphonemic categories?

We compared standard and deviant responses in the phonemicin English condition and in the phonemic in Spanish condition in

the English language context in monolinguals to verify that stimuliwere discriminated in the expected way.

4.2.1. Phonemic in English conditionOur expectation was that monolinguals in the English language

context would perceive the standard (+50 ms of VOT) and the devi-ant (+15 ms of VOT) as belonging to different phonemic categories(/ka/ and /ga/; respectively). That is, a significant amplitude differ-ence between standard and deviant ERP-responses was expectedfor the phonemic in English condition. Please note that ERPs wereobtained only during the English language context.

Fig. 4A shows the standard and deviant ERP responses from theeight electrode regions. The overall deviant-ERP response(Mean = �.70 SD = .67) was more negative than overall standard-ERP response (Mean = �.10 SD = .74). The results showed a signifi-cant main effect for ERP type (F(1,8) = 14.2, p = .006, g2

p ¼ 0:64)with no significant interaction between standard and deviant re-sponses and electrode regions (F(2.8,22.44) = 2.65, p = .08, g2

p ¼:25). There was a significant MMN amplitude difference betweenelectrode FCZ and the averaged mastoids (FCZ Mean = �.93SD = .70; Mastoids’ Mean = �.33 SD = .60; t(8) = �2.7, p = .02).

4.2.2. Phonemic in Spanish conditionOur expectation was that monolinguals in the English language

context would perceive the standard (�20 ms of VOT) and thedeviant (+15 ms of VOT) as belonging to the same phonemic cate-gory (/ga/ and /ga/; respectively). Namely, no significant amplitude

Panel A Panel B

FCZ

MID-CENTROPARIETAL

LATEIRAPTFELLATEIRAPTFEL

Phonemic in Spanish ConditionSpanish Language Context

LATNORFTHGIRLATNORFTFEL

LARTNECTHGIRLARTNECTFEL

MID-FRONTOCENTRAL

StandardDeviant

MastoidMMN

FCZ

MID-CENTROPARIETAL

LATEIRAPTFELLATEIRAPTFEL

Phonemic in Spanish ConditionEnglish Language Context

LATNORFTHGIRLATNORFTFEL

LARTNECTHGIRLARTNECTFEL

MID-FRONTOCENTRAL

StandardDeviant

MastoidMMN

Fig. 3. Bilinguals’ (N = 10) scalp distributions of the standard and deviant ERP responses in the phonemic in Spanish condition during the Spanish language context (Panel A)and during the English language context (Panel B). The standard response is represented by the black line and the deviant response by the gray line. Negative is plotted down.At the bottom of each panel, MMN amplitudes from electrode FCZ (gray line) and mastoid substitute electrodes (black line; T6/7) are presented. The short black lines in eachof the boxes depict the time windows of interest (220–280 ms and 280–480 ms).

A. García-Sierra et al. / Brain & Language 121 (2012) 194–205 201

difference between standard and deviant ERP-responses were ex-pected for the phonemic in Spanish condition during the Englishlanguage context.

Fig. 4B shows the standard and deviant ERP responses from the8 electrode regions. The overall deviant-ERP responses (Mean =�.36 SD = 1.13) was similar when to the overall standard-ERP re-sponses (Mean = �.10 SD = .70). The results showed no main effectfor ERP type (F(1,8) = 1.6, p = .23, g2

p ¼ :17) and, no interaction be-tween standard and deviant responses across electrode regions(F(1.6,12.8) = .92, p = .4, g2

p ¼ :10). The MMN amplitude inversionbetween electrodes FCZ and averaged mastoids were not signifi-cantly different (FCZ Mean = �.57 SD = .85; Mastoids’ Mean = �.24SD = .60; t(8) = �1.8, p = .11).

The stimuli tested represented clear phonemic categories, asindicated by the MMN, for the monolingual participants. Resultsobtained from the bilinguals indicate that language contexts caninfluence the early stages of speech representation as indicatedby the changes observed in the MMN amplitude. Specifically, iden-tical acoustic–phonetic information was perceived as belonging todifferent phonemic categories (presence of the MMN), or asbelonging to the same phonemic category (absence of the MMN),depending on language context.

4.3. Additional analyses

For the bilingual group a 2 (language context; English and Span-ish) � 2 (phonemic condition: phonemic in English and phonemicin Spanish) � 2 (ERP type; standard and deviant) � 8 (Electrode

regions: Left-Frontal, Left-Central, Left-Parietal, Mid-Frontocentral,Mid-Centroparietal, Right-Frontal, Right-Central and, Right-Parie-tal) repeated measures ANOVA was done to test the N1 ERP re-sponse and P3a ERP response. For the monolingual group, a 2(phonemic condition: phonemic in English and phonemic in Span-ish) � 2 (ERP type; standard and deviant) � 8 (Electrode regions:Left-Frontal, Left-Central, Left-Parietal, Mid-Frontocentral, Mid-Centroparietal, Right-Frontal, Right-Central and, Right-Parietal) re-peated measure ANOVA was done to test the N1 ERP response andP3a ERP response. In both groups’ analyses (bilingual and monolin-gual) a Greenhouse-Geisser epsilons (e) was used for non-spheric-ity correction when necessary.

4.3.1. Bilinguals’ N1 responseNo significant main effects or interactions were found for ERP,

language context or speech contrast conditions or electroderegions.

4.3.2. Bilinguals’ P3a responseNo main effects for ERP type, language context, speech contrast

condition or electrode regions were found. A significant interactionbetween ERP type, language context, speech contrast condition andelectrode regions was found (F(7,19.2) = 11.4, p = .0003 g2

p ¼ 0:58).Further analyses revealed that in the phonemic in English conditionduring the English language context the deviant response wasmore positive than the standard response at frontal electrode re-gions (i.e., Left-Frontal, Mid-Frontocentral and, Right-Frontalshowed a significant P3a response). No significant P3a effects were

Panel A Panel B

FCZ

MID-CENTROPARIETAL

LATEIRAPTFELLATEIRAPTFEL

Phonemic in English ConditionEnglish Language Context

LATNORFTHGIRLATNORFTFEL

LARTNECTHGIRLARTNECTFEL

MID-FRONTOCENTRAL

StandardDeviant

MastoidMMN

FCZ

MID-CENTROPARIETAL

LATEIRAPTFELLATEIRAPTFEL

Phonemic in Spanish ConditionEnglish Language Context

LATNORFTHGIRLATNORFTFEL

LARTNECTHGIRLARTNECTFEL

MID-FRONTOCENTRAL

MastoidMMN

StandardDeviant

Fig. 4. English monolinguals’ (N = 9) scalp distributions of the standard and deviant ERP responses in the English language context in the phonemic in English condition (PanelA) and in the phonemic in Spanish condition (Panel B). The standard response is represented by the black line and the deviant response by the gray line. Negative is plotteddown. At the bottom of each panel, MMN amplitudes from electrode FCZ (gray line) and mastoids’ substitute electrodes (black line; T6/7) are presented. The short black linesin each of the boxes depict the time windows of interest (220–280 ms and 280–480 ms).

202 A. García-Sierra et al. / Brain & Language 121 (2012) 194–205

found in the phonemic in English condition during the Spanish lan-guage context.

In the phonemic in Spanish condition during the Spanish lan-guage context the deviant response was more positive than thestandard response at frontal electrode regions (i.e., Left-Frontalshowed a significant P3a response). No significant P3a effects werefound in the phonemic in Spanish condition during the English lan-guage context.

4.3.3. Monolinguals’ N1 responseNo significant main effects or interactions were found for ERP,

language context or speech contrast conditions or electroderegions.

4.3.4. Monolinguals’ P3a responseNo main effects for ERP type, language context, speech contrast

condition or electrode regions were found. A significant interactionbetween ERP type and speech contrast condition was found(F(1,8) = 7.7, p = .024 g2

p ¼ 0:50). Further analyses revealed that inthe phonemic in English condition the deviant response was morepositive than the standard response at frontal electrode regions(i.e., Mid-Frontocentral, Right-Frontal and Right-Central electroderegions showed a significant P3a response). No significant P3a ef-fects were found in the phonemic in Spanish condition.

The results showed a predominant frontal distribution for theP3a ERP component. Bilinguals’ and monolinguals’ P3a responsewas significant only for the speech contrast conditions that

matched the language contexts (e.g., Phonemic in English conditionduring the English language context). These results suggested thatlanguage context can enhance involuntarily detection of speechcues among standard and deviant sounds.

5. Discussion

The present investigation explored the double phonemicboundary in Spanish–English bilinguals. Specifically, our goal wasto test bilinguals’ ability to discriminate speech sounds that shareacoustic–phonetic information in English and Spanish, but havedifferent categorical mental representations in each of these lan-guages. Behavioral research of the double phonemic boundaryhas yielded contradictory findings. Some studies show that mono-lingual and bilingual speakers develop fixed but distinct phonemicboundaries (i.e., bilinguals merge the phonetic rules of their twolanguages into one boundary, Caramazza et al., 1973; Williams,1977). Other investigations have shown that phonemic boundariesin bilinguals shift depending on the language in use (Elman et al.,1977; Flege & Eefting, 1987; Garcia-Sierra et al., 2009; Hazan &Boulakia, 1993). The previously reported shifts in the phonemicboundary suggest that bilinguals can develop independent pho-netic rules for speech sounds that share acoustic information intheir two languages, but have different meaning in each of theselanguages. However, it has been suggested that bilinguals’ doublephonemic representation is not linguistic in nature but rather is

A. García-Sierra et al. / Brain & Language 121 (2012) 194–205 203

a consequence of the methods used to establish language contexts(Bohn & Flege, 1993).

In the present investigation we assessed the effects of languagecontext on brain activity associated with speech discrimination inbilingual and monolingual participants. Specifically, we expectedthat bilinguals’ categorical mental representation would changefrom /g/ to /k/ as a function of language context. Unlike previousresearch, we established language context by asking participantsto read silently in the language of interest during ERP recordings.This procedure allowed us to set bilinguals in a specific languagemode (Grosjean, 2001) throughout the perceptual task.

Our results showed significant amplitude changes betweenstandard and deviant responses as a function of language contextsin bilinguals. Differences between standard and deviant ERP re-sponses were significant only for the expected stimulus conditionand language context combinations: in the phonemic in Englishcondition during English language context and in the phonemic inSpanish condition during Spanish language context. Our resultssuggest that our data collection and language context methodswere successful and that bilingual speakers of English and Spanishpossess a double phonemic representation. These results are con-sistent with behavioral studies supporting the double phonemicboundary in bilinguals (Elman et al., 1977; Flege & Eefting, 1987;Garcia-Sierra et al., 2009; Hazan & Boulakia, 1993) but conflictwith previous behavioral and ERP investigations of double phone-mic representation in bilinguals (Bohn & Flege, 1993; Caramazzaet al., 1973; Williams, 1977; Winkler et al., 2003).

Winkler and colleagues (2003) explored double phonemic rep-resentation in Hungarian–Finnish bilinguals using the MMN. Theirresults showed no significant MMN amplitude changes as a func-tion of language contexts. The authors concluded that bilingualsapply L2-phonetic-distinctions irrespectively of the language inuse. In this study we argue that in order to create strong languagecontext effects, participants should be immersed in the language ofinterest throughout the perceptual task. Winkler et al. establishedlanguage context by presenting a word in Finnish or Hungarian,which was different from the standard and deviant, on 1.3% ofthe trials, which may not have produced a strong language contexteffect. In fact, in a set of studies using language context to primebilinguals, the primer-effect was shown only when language con-texts were presented consistently throughout the entire task(Elston-Güttler et al., 2005).

The inclusion of a monolingual group served to verify that ourspeech sounds represented clear phonemic categories. Indeed,monolingual English speakers categorized the speech sounds, asindicated by the MMN, in the expected way. Standard and deviantresponses differed significantly for the stimuli pair used in the pho-nemic in English condition during the English language context. Incontrast, standard and deviant did not differ significantly for thestimuli pair used in the phonemic in Spanish condition, during theEnglish language context. Although our findings suggested thatlanguage context influenced speech perception, it is essential totest our methodology in other bilingual and monolingual popula-tions that show an overlap in their voicing boundaries (e.g., Frenchvs. English, German vs. Spanish).

Bilinguals’ and monolinguals’ difference waveform show a posi-tive peak around 300 ms after stimulus onset. We interpreted thispositivity as the P3a ERP component. The P3a response has beendescribed to be elicited when attention is being switched fromthe background sound (standard) to the target sound (Esceraet al., 2000; Friedman et al., 2001; Squires et al., 1975). The statis-tical analyses for the P3a response showed that it was frontally dis-tributed and significant for the conditions where the phonetics ofthe speech contrasts matched those of the language contexts (pho-nemic in English condition during English language context andphonemic in Spanish condition during Spanish language context)

in bilinguals and monolinguals. The presence of P3a componentsuggests that the phonetic differences between standard and devi-ant captured participants’ attention in an involuntary way. In fact,both groups showed the strongest P3a response in the speech con-trast condition were the duration of aspiration (+VOT) was themain cue (i.e., phonemic in English condition). This suggests thatour participants perceived aspiration more saliently than pre-voic-ing. The fact that bilinguals’ attention to phonetic differences (P3a)changed as a function of language context, supports our interpreta-tion that bilinguals possess a double phonemic representation.

We are aware that our findings regarding the MMN responsemay not necessarily reflect a language effect (i.e., be based onknowledge of a second language) but rather a familiarity effect.For example, it has been shown that target words and non-speechsounds are better discriminated in familiar contexts (familiar stan-dard words/non-speech sounds) than in unfamiliar contexts (unfa-miliar standard words/non-speech sounds) (Jacobsen, Schröger,Winkler, & Horváth, 2005; Jacobsen, Horváth, Schröger, Lattner,Widmann, & Winkler, 2004). The fact that familiarity-effects werefound in non-speech sounds suggests that enhanced discrimina-tion in familiar context may reflect a more general feature of audi-tory processing that is not linguistic in nature. Our languagecontext methods could have produced a familiarity-effect insteadof a language-effect per se. Bilinguals showed significant MMN re-sponses when the phonetics of the speech sounds match the lan-guage context (i.e., familiar context) but, no MMN was recordedwhen the phonetics of the speech sounds did not match the lan-guage context (i.e., unfamiliar context). In our opinion, testingspeech discrimination of monolinguals in two language contextscould disambiguate language effects and familiarity effects. Iffamiliarity effects were influencing our results, then we would ex-pect that monolinguals would show the same speech discrimina-tion pattern as bilinguals. Unfortunately, in this investigation ourmethod did not allow us to test monolinguals’ brain responses intwo language contexts since monolinguals reported no confidencein reading Spanish.

However, there is some evidence suggesting that MMN ampli-tude changes as a function of language contexts are not the conse-quence of familiarity effects. Specifically, Garcia-Sierra (2007)established two language contexts by exposing participants to vi-deo-clips and pre-recorded audio samples in the language of inter-est between ERP recoding blocks. This method allowed datacollection from monolinguals and bilinguals in two language con-texts since previous knowledge of a second language was not re-quired. Results from this study showed evidence of strongerMMN responses for bilinguals, but not monolinguals, as a functionof language context, indicating double phonemic representationrather than familiarity effects.

In the present investigation, we observed robust MMNs forspeech sounds belonging to different phonetic categories (be-tween-category) and reduced MMN amplitudes for speech soundsbelonging to the same phonetic category (within-category).Namely, our results indicate that the MMN to speech is sensitiveto specific linguistic processes, such as categorical perception ofspeech (Cheour et al., 1998; Dehaene-Lambertz, 1997; Diaz, Baus,Escera, Costa, & Sebastian-Galles, 2008; Näätänen et al., 1997;Peltola, Kujala, Tuomainen, Ek, Aaltonen, & Näätänen, 2003;Rivera-Gaxiola, Csibra et al., 2000; Sharma & Dorman, 2000; Win-kler et al., 1999a, 1999b). However, it has also been postulated thatthe MMN represents a more general acoustic process rather than alinguistic one. For example, equivalent MMN amplitudes have beenreported for speech sounds belonging to the same phonetic cate-gory and speech sounds belonging to different phonetic categories(Maiste, Wiens, Hunt, Scherg, & Picton, 1995; Pettigrew et al., 2004;Sams, Aulanko, Aaltonen, & Näätänen, 1990; Sharma, Kraus, McGee,Carrell, & Nicol, 1993; Tampas, Harkrider, et al., 2005).

204 A. García-Sierra et al. / Brain & Language 121 (2012) 194–205

The contradictory findings regarding MMN sensitivity to acous-tic vs. phonetic representations of speech can be explained bymethodological differences rather than neural speech processingsensitivity (Dehaene-Lambertz, 1997). Research showing within-category sensitivity tested monolingual speakers’ abilities to dis-criminate sounds occurring in their native language (e.g., two stopconsonants with different VOT durations but representing thesame sound; Maiste et al., 1995; Pettigrew et al., 2004; Samset al., 1990; Sharma et al., 1993; Tampas et al., 2005); whereas re-search showing no within-category sensitivity tested monoling-uals’ abilities in discriminating non-native speech contrasts (e.g.,two vowels sounds, one present in their native language and theother not present in their language: Cheour et al., 1998;Dehaene-Lambertz, 1997; Diaz et al., 2008; Näätänen et al.,1997; Peltola et al., 2003; Sharma & Dorman, 2000; Winkleret al., 1999a, 1999b). These findings can be explained by Best’s per-ceptual assimilation theory (Best, 1993, see also Best, McRoberts, &Goodell, 2001) that proposes that monolingual speakers will showpoor within-category discrimination for sounds not present intheir native language due to ‘‘phonetic perceptual assimilation’’.In other words, non-native speech sounds are perceptuallygrouped to the ‘‘nearest’’ native-phonetic category. Our resultsare consistent with Best’s assimilation theory (1993) in the sensethat monolingual participants showed reduced within-categorysensitivity for the speech contrast condition consisting of a nativeand a non-native speech sound for the English language (phonemicin Spanish condition). That is, monolinguals grouped the standard(�20 ms VOT/non-native) and the deviant (+15 ms VOT/native)into the same phonemic category. Negative VOT has been shownto be a temporal cue that English monolingual speakers learn todetect only after rigorous auditory training (Tremblay, Kraus,Carrell, & McGee, 1997; Tremblay, Kraus, & McGee, 1998). Overall,it seems that the MMN is especially sensitive to phonetic differ-ences between native and non-native languages (Näätänen, 2001).

Bilinguals showed a change in within-category sensitivity as afunction of language contexts. To be precise, bilinguals’ showedless within-category sensitivity in the phonemic in English condi-tion during the Spanish language context than during the Englishlanguage context. The same pattern was found in the phonemic inSpanish condition. These results cannot be compared with themonolingual results because the target speech sound (+15 ms ofVOT) is ambiguous for the bilingual speaker but not for the mono-lingual speaker. More research is necessary in order to comparebilinguals’ and monolinguals’ sensitivities to sounds belonging tothe same phonetic categories (within-category sensitivity). Forexample, bilinguals could be tested with speech sounds that arenot ambiguous for either of their languages but belong to the samephonetic category (e.g., �50 vs. �30 ms of VOT). Also, bilingualscould be tested in their ability to discriminate speech contraststhat are not present in either of their native languages.

Our findings support the assumption of bilinguals’ doublephonemic representation; this finding is in accordance with theidea that perception is a parallel, rather than a serial process.McClelland and Rumelhart (1981) proposed that perception isthe result of an interaction between previous information andsensory input. Our results showed that in the case of bilinguals’speech perception, the early stages of speech representation canbe affected by the language being used in a given moment(language mode). More research is needed in this field, especiallyto incorporate this idea with current and new models onbilingualism.

There are two limitations associated with this study. One limi-tation is that data was not collected from mastoid electrode sites.Ideally, the polarity inversion associated with the MMN is quanti-fied by comparing the MMN response form fronto-central elec-trodes and the mastoid electrodes (Alho, 1995). In the present

investigation we used electrodes T6 and T7 as substitutes for mas-toid electrodes. Although the T6 and T7 electrodes are positionedmore frontally and higher than the mastoid electrodes, they are lo-cated below the Sylvian fissure as indicated by the 10–20 Interna-tional System and can be used to explore MMN brain generators.Our comparisons between the fronto-central electrode (FCZ) andthe mastoid substitutes (average of T6 and T7) showed a significantamplitude change in the expected conditions. For example, theMMN response from FCZ and the averaged of T6/T7 differed in asignificant way in the phonemic in English condition during Englishlanguage context and in the phonemic in Spanish condition duringSpanish language context. Therefore, these results suggest thatthe enhanced negativity associated with the deviant response doesreflect the MMN generators.

Another limitation in this study is that the deviant response wasnot compared against a deviant control. For example, in Jacobsenand colleagues (2004, 2005), the amplitude of the deviant responseis compared against the amplitude of the same deviant soundwhen presented in isolation. This method controls for the acousticdifferences between standard and deviant sounds. Further researchassociated with language contexts should consider introducing adeviant control condition.

In summary, our results indicate that language context caninfluence the formation of phonemic memory traces involved inspeech discrimination. These findings suggest that bilingual speak-ers of English and Spanish can assign identical acoustic–phoneticinformation to different phonemic categories depending on thelanguage they are using at the moment. Nevertheless, our resultsshould be interpreted with caution since monolingual speakers ofEnglish were tested only in the English language context. Newstudies exploring double phonemic representation should testbilingual and monolingual speakers in identical conditions andshould control for the acoustic differences between deviant andstandard sounds.

Acknowledgments

This study was supported by the US-Mexico/Borderlands Re-search Award, College of Liberal Arts, The University of Texas atAustin, by the Jesse H. Jones Fellowship, College of Communication,The University of Texas at Austin, and by The National ScienceFoundation Science of Learning Program (SBE-0354453). We thankDenise Padden for her valuable contribution in reviewing this man-uscript. We also thank Dr. Javier Alvarez Bermudez at the Autono-mous University of Nuevo Leon, Mexico for making the behavioraltesting possible.

References

Aaltonen, O., Niemi, P., Nyrke, T., & Tuhkanen, M. (1987). Event-related brainpotentials and the perception of a phonetic continuum. Biological Psychology, 24,197–207.

Abramson, A. S., & Lisker, L. (1970). Discriminability along the voicing continuum:Cross-language tests. Proceedings of the Sixth International Congress of PhoneticSciences, 569–573.

Abramson, A. S., & Lisker, L. (1972). Voice-timing perception in Spanish word-initialstops. Journal of Phonetics, 1, 1–8.

Alho, K. (1995). Cerebral generators of mismatch negativity (Mmn) and its magneticcounterpart (Mmnm) elicited by sound changes. Ear and Hearing, 16, 38–51.

Best, C. T. (1993). Emergence of language-specific constraints in perception of non-native speech: A window on early phonological development. In B. de Boysson-Bardies & S. de Schonen, et al. (Eds.), Developmental neurocognition: Speech andface processing in the first year of life (pp. 289–304). Norwell, MA, US: KluwerAcademic Publishers.

Best, C. T., McRoberts, G. W., & Goodell, E. (2001). Discrimination of non-nativeconsonant contrasts varying in perceptual assimilation to the listener’s nativephonological system. Journal of the Acoustical Society of America, 109, 775–794.

Bohn, O. S., & Flege, J. E. (1993). Perceptual Switching in Spanish-English bilinguals.Journal of Phonetics, 21, 267–290.

Caramazza, A., Yeni-Komshian, G. H., Zurif, E. B., & Carbone, E. (1973). Theacquisition of a new phonological contrast: The cast of stop consonants in

A. García-Sierra et al. / Brain & Language 121 (2012) 194–205 205

French–English bilinguals. Journal of the Acoustical Society of America, 54,421–428.

Cheour, M., Ceponiene, R., Lehtokoski, A., Luuk, A., Allik, J., Alho, K., et al. (1998).Development of language-specific phoneme representations in the infant brain.Nature Neuroscience, 1, 351–353.

Dehaene-Lambertz, G. (1997). Electrophysiological correlates of categoricalphoneme perception in adults. Neuroreport, 8, 919–924.

Deouell, L. Y. (2007). The frontal generator of the mismatch negativity revisited.Journal of Psychophysiology, 21(3), 188–203.

Diaz, B., Baus, C., Escera, C., Costa, A., & Sebastian-Galles, N. (2008). Brain potentialsto native phoneme discrimination reveal the origin of individual differences inlearning the sounds of a second language. Proceedings of the National Academy ofSciences of the United States of America, 105, 16083–16088.

Diehl, R. L., Elman, J. L., & McCusker, S. B. (1978). Contrast effects on stop consonantidentification. Human Perception and Performance, 4, 599–609.

Diesch, E., & Luce, T. (1997a). Magnetic fields elicited by tones and vowel formantsreveal tonotopy and nonlinear summation of cortical activation.Psychophysiology, 34, 501–510.

Diesch, E., & Luce, T. (1997b). Magnetic mismatch fields elicited by vowels andconsonants. Experimental Brain Research, 116, 139–152.

Elman, J. L., Diehl, R. L., & Buchwald, S. E. (1977). Perceptual switching in bilinguals.Journal of Acoustical Society of America, 62, 971–974.

Elston-Güttler, K. E., Gunter, T. C., & Kotz, S. A. (2005). Zooming into L2: Globallanguage context and adjustment affect processing of interlingual homographsin sentences. Cognitive Brain Research, 25, 57–70.

Escera, C., Alho, K., Schröger, E., & Winkler, I. (2000). Involuntary attention anddistractibility as evaluated with event-related brain potentials. Audiology andNeurotology, 5, 151–166.

Flege, J. E., & Eefting, W. (1986). Linguistic and developmental effects on theproduction and perception of stop consonants. Phonetica, 43, 155–171.

Flege, J. E., & Eefting, W. (1987). Cross-language switching in stop consonantperception and production by Dutch speakers of English. Speech Communication,6, 185–202.

Flege, J. E., MacKay, I. R. A., & Meador, D. (1999). Native Italian speakers’ perceptionand production of English vowels. Journal of the Acoustical Society of America,106, 2973–2987.

Friedman, D., Cycowicz, Y. M., & Gaeta, H. (2001). The novelty P3: An event-relatedbrain potential (ERP) sign of the brain’s evaluation of novelty. Neuroscience andBiobehavioral Reviews, 25, 355–373.

Garcia-Sierra, A. (2007). Language contexts in speech perception: Testing the doublephonetic standard in bilinguals. Doctoral Disseration: University of Texas atAusitn.

Garcia-Sierra, A., Diehl, R. L., & Champlin, C. (2009). Testing the double phonemicboundary in bilinguals. Speech Communication, 51, 369–378.

Grosjean, F. (2001). The bilingual’s language modes. In J. L. Nicol (Ed.), One mind, twolanguages: Bilinguals language processing (pp. 1–22). Blackwell.

Hazan, V. L., & Boulakia, G. (1993). Perception and production of a voicing contrastby French–English bilinguals. Language and Speech, 36, 17–38.

Holt, L. L. (2005). Temporally nonadjacent nonlinguistic sounds affect speechcategorization. Psychological Science, 16, 305–312.

Holt, L. L., & Lotto, A. J. (2002). Behavioral examinations of the level of auditoryprocessing of speech context effects. Hearing Research, 167, 156–169.

Jacobsen, T., Horváth, J., Schröger, E., Lattner, S., Widmann, A., & Winkler, I. (2004).Pre-attentive auditory processing of lexicality. Brain and Language, 88, 54–67.

Jacobsen, T., Schröger, E., Winkler, I. N., & Horváth, J. (2005). Familiarity affects theprocessing of task-irrelevant auditory deviance. Journal of CognitiveNeuroscience, 17(11), 1704–1713.

Klatt, D. H. (1980). Software for a cascade/parallel formant synthesizer. Journal ofthe Acoustical Society of America, 67, 971–990.

Lisker, L., & Abramson, A. S. (1970). The voicing dimension: Some experiments incomparative phonetics. Proceedings of the 6th international congress of phoneticsciences, Prague, 1967. 1970. Prague: Academia, pp. 563–567.

Luck, J. S. (2005). An introduction to the event-related potential technique. CambridgeMassachusetts: The MIT Press.

Mackain, K. S., Best, C. T., & Strange, W. (1981). Categorical perception of English Rand I by Japanese bilinguals. Applied Psycholinguistics, 2, 369–390.

Maiste, A. C., Wiens, A. S., Hunt, M. J., Scherg, M., & Picton, T. W. (1995). Event-related potentials and the categorical perception of speech sounds. Ear &Hearing, 16, 68–89.

McClelland, J. L., Mirman, D., & Holt, L. L. (2006). Are there interactive processes inspeech perception? Trends in Cognitive Sciences, 10, 363–369.

McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model ofcontext effects in letter perception. Part 1. An account of basic findings.Psychological Review, 88, 375–407.

Näätänen, R. (1986). A classification of N2 kinds of ERP-components. In W.C.McCallum, R. Zappoli, F. Denoth (Eds.). Cerebral psychophysiology: Studies inevent related potentials, Electroencephalography & Clinical Neurophysiology(Suppl. 38, pp. 169–172).

Näätänen, R. (1992). Attention and brain function. Hillsdale, New Jersey: LawrenceErlbaum Associates, Publishers.

Näätänen, R. (2001). The perception of speech sounds by the human brain asreflected by the mismatch negativity (MMN) and its magnetic equivalent(MMNm). Psychophysiology, 38, 1–21.

Näätänen, R., Gaillard, A. W. K., & Mantysalo, S. (1978). Early selective attentioneffect on evoked potential reinterpreted. Acta Poycholoica, 42, 313–329.

Näätänen, R., Lehtokoski, A., Lennes, M., Cheour, M., Huotilainen, M., Iivonen, A.,et al. (1997). Language-specific phoneme representations revealed by electricand magnetic brain responses. Nature, 385, 432–434.

Näätänen, R., & Michie, P. T. (1979). Early selective-attention effects on the evoked-potential – A critical-review and reinterpretation. Biological Psychology, 8,81–136.

Näätänen, R., & Picton, T. (1987). The N1 wave of the human electric and magneticresponse to sound – A review and an analysis of the component structure.Psychophysiology, 24, 375–425.

Näätänen, R., Simpson, M., & Loveless, N. E. (1982). Stimulus deviance and evokedpotentials. Biological Psychology, 14, 53–98.

Näätänen, R., & Winkler, I. (1999). The concept of auditory stimulus representationin cognitive neuroscience. Psychological Bulletin, 125, 826–859.

Paulmann, S., Elston-Güttler, K. E., Gunter, T. C., & Kotz, S. A. (2006). Is bilinguallexical access influenced by language context? Neuroreport, 17, 727–731.

Peltola, M. S., Kujala, T., Tuomainen, J., Ek, M., Aaltonen, O., & Näätänen, R. (2003).Native and foreign vowel discrimination as indexed by the mismatch negativity(MMN) response. Neuroscience Letters, 352, 25–28.

Pettigrew, C. M., Murdoch, B. M., Kei, J., Chenery, H. J., Sockalingram, R., Ponton, C.W., et al. (2004). Processing of English words with fine acoustic contrasts andsimple tones: A mismatch negativity study. Journal of the American Academy ofAudiology, 14, 47–66.

Rivera-Gaxiola, M., Csibra, G., Johnson, M. H., & Karmiloff-Smith, A. (2000a).Electrophysiological correlates of cross-linguistic speech perception in nativeEnglish speakers. Behavioural Brain Research, 111, 13–23.

Rivera-Gaxiola, M., Johnson, M. H., Csibra, G., & Karmiloff-Smith, A. (2000b).Electrophysiological correlates of category goodness. Behavioural Brain Research,112, 1–11.

Sams, M., Aulanko, R., Aaltonen, O., & Näätänen, R. (1990). Event-related potentialsto infrequent changes in synthesized phonetic stimuli. Journal of CognitiveNeuroscience, 2, 334–357.

Schröger, E. (1998). Measurement and interpretation of the mismatch negativity.Behavior Research Methods Instruments & Computers, 30, 131–145.

Semlitsch, H. V., Anderer, P., Schuster, P., & Presslich, O. (1986). A solution forreliable and valid reduction of ocular artifacts, applied to P300 ERP.Psychophysiology, 23, 695–703.

Shafer, V. L., Schwartz, R. G., & Kurtzberg, D. (2004). Language-specific memorytraces of consonants in the brain. Cognitive Brain Research, 18, 242–254.

Sharma, A., & Dorman, M. F. (2000). Neurophysiologic correlates of cross-languagephonetic perception. Journal of the Acoustical Society of America, 107, 2697–2703.

Sharma, A., Kraus, N., McGee, T., Carrell, T., & Nicol, T. (1993). Acoustic versusphonetic representation of speech as reflected by the mismatch negativityevent-related potential. Electroencephalography and Clinical Neurophysiology/Evoked Potentials Section, 88, 64–71.

Shtyrov, Y. C. A., & Pulvermüller, F. (2002). Neurophysiological evidence of memorytraces for words in the human brain. Neuroreport, 13, 521–525.

Shtyrov, Y., & Pulvermüller, F. (2007). Language in the mismatch negativity design:Motivations, benefits, and prospects. Journal of Psychophysiology, 21, 176–187.

Squires, N. K., Squires, K. C., & Hillyard, S. A. (1975). Tow varieties of long-latencypositive waves evoked by unpredictable auditory-stimuli in man.Electroencephalography and Clinical Neurophysiology, 38, 387–401.

Sundara, M., & Polka, L. (2008). Discrimination of coronal stops by bilingual adults:The timing and nature of language interaction. Cognition, 106, 234–258.

Sussman, E., Winkler, I., Huotilainen, M., Ritter, W., & Näätänen, R. (2002). Top-down effects can modify the initially stimulus-driven auditory organization.Cognitive Brain Research, 13, 393–405.

Tampas, J. W., Harkrider, A. W., et al. (2005). Neurophysiological indices of speechand nonspeech stimulus processing. Journal of Speech Language and HearingResearch, 48, 1147–1164.

Tiitinen, H., May, P., Reinikainen, K., & Näätänen, R. (1994). Attentive Noveltydetection in humans is governed by pre-attentive sensory memory. Nature, 372,90–92.

Tremblay, K., Kraus, N., Carrell, T. D., & McGee, T. (1997). Central auditory systemplasticity: Generalization to novel stimuli following listening training. Journal ofthe Acoustical Society of America, 102, 3762–3773.

Tremblay, K., Kraus, N., & McGee, T. (1998). The time course of auditory perceptuallearning: Neurophysiological changes during speech-sound training.Neuroreport, 9, 3557–3560.

Williams, L. (1977). The perception of stop consonant voicing by Spanish-Englishbilinguals. Perception & Psychophysics, 21, 289–297.

Winkler, I., Kujala, T., Alku, P., & Näätänen, R. (2003). Language context and phoneticchange detection. Cognitive Brain Research, 17, 833–844.

Winkler, I., Kujala, T., Tiitinen, H., Sivonen, P., Alku, P., Lehtokoski, A., et al. (1999a).Brain responses reveal the learning of foreign language phonemes.Psychophysiology, 36, 638–642.

Winkler, I., Lehtokoski, A., Alku, P., Vainio, M., Czigler, I., Csepe, V., et al. (1999b).Pre-attentive detection of vowel contrasts utilizes both phonetic and auditorymemory representations. Cognitive Brain Research, 7, 357–369.