1. Introduction
1.1 Phonological organization of simultaneous bilinguals
The phonological organization of bilinguals has been of great interest to speech perception and production research in recent decades. Most studies have focused on the abilities of sequential bilinguals, who acquired their first language (L1) from birth, and started to learn a second language (L2) at later stages of their lives (Best, Reference Best1995; Flege, Reference Flege1995). One of the general conclusions has been that sequential bilinguals are able to achieve native monolingual-like phonological performance in the L2, if the bilinguals’ onset of L2 acquisition is close to the onset of L1 acquisition (birth) and/or their L2 proficiency is high (e.g., Bongaerts, Mennen & van der Slik, Reference Bongaerts, Mennen and Slik2000; Flege, Schirru & MacKay, Reference Flege, Schirru and MacKay2003; Flege, Yeni-Komshian & Liu, Reference Flege, Yeni-Komshian and Liu1999; Guion, Reference Guion2003; MacKay, Meador & Flege, Reference MacKay, Meador and Flege2001; Perani, Paulesu, Sebastián-Gallés, Dupoux, Dehaene, Bettinardi, Cappa, Fazio & Mehler, Reference Perani, Paulesu, Sebastián-Gallés, Dupoux, Dehaene, Bettinardi, Cappa, Fazio and Mehler1998; Piske, Flege, MacKay & Mador, Reference Piske, Flege, MacKay and Meador2002; Yamada, Reference Yamada1995; Yamada, Yamada & Strange, Reference Yamada, Yamada and Strange1995; but see Pallier, Bosch & Sebastián-Gallés, Reference Pallier, Bosch and Sebastián-Gallés1997). By contrasting sequential bilinguals’ L2 abilities with the performance of native monolinguals, it is inferred, however, that fully successful second language acquisition and an intact bilingual phonological organization require native monolingual-like performance. This phenomenon has been defined as the monolingual or fractional view on bilingualism (Grosjean, Reference Grosjean1998).
Based on this fractional view, one might expect that simultaneous bilingual (SB) individuals – those who developed two languages within the same timeframe from birth and are therefore native speakers of two languages – would exhibit monolingual-like abilities within both of their native languages. However, findings of behavioral studies (e.g., Caramazza, Yeni-Komshian, Zurif & Carbone, Reference Caramazza, Yeni-Komshian, Zurif and Carbone1973; Guion, Reference Guion2003; Mack, Reference Mack1989; Sebastián-Gallés, Echeverría & Bosch, Reference Sebastián-Gallés, Echeverría and Bosch2005; Sundara & Polka, Reference Sundara and Polka2008; Sundara, Polka & Baum, Reference Sundara, Polka and Baum2006a; Sundara, Polka & Genesee, Reference Sundara, Polka and Genesee2006b) seem to converge toward the conclusion that speech perception and production patterns of SB individuals do not completely match the patterns observed in their monolingual peers.
For instance, in terms of consonants, SB adults accommodate additional phonetic detail that is not consistently produced and perceived by either of their monolingual peers (e.g., Sundara, Reference Sundara2005; Sundara & Polka, Reference Sundara and Polka2008). As well, there is evidence that SB adults’ production patterns are more streamlined than their monolingual peers; SB adults appear to selectively omit free variant forms that fail to enhance phonetic category differences in either of their languages (MacLeod & Stoel-Gammon, Reference MacLeod and Stoel-Gammon2005; Sundara et al., Reference Sundara, Polka and Baum2006a). Similar findings have emerged for vowels. For instance, even though simultaneous bilingual speakers of Quechua and Spanish produce vowel categories of the two languages similar to those of the monolingual speakers, the bilingual productions (in terms of phonetic detail) differ from the monolingual ones. The bilingual productions of the Quechua vowels are raised upward along the F1 axis in the acoustic/phonetic vowel space, as compared to monolingual productions of the same vowels (Guion, Reference Guion2003). This systemic manipulation of the bilingual vowel categories probably serves to sustain separate vowels across and within the two languages.
Considering SB adults’ processing of cross-language vowels and consonants, SB adults do not appear to have a dual set of entirely monolingual-like speech processing abilities, even though they are native speakers of both languages under investigation. Rather, it seems that SBs represent a group of language users distinct from monolingual users of both of their languages – a conclusion that is in line with the so-called holistic view of bilingualism (Grosjean, Reference Grosjean1998).
Surprisingly, the neural correlates of SB language processing that contribute to such unique behavioral patterns have only rarely been investigated (Kim, Relkin, Lee & Hirsch, Reference Kim, Relkin, Lee and Hirsch1997; Klein, Zatorre, Milner, Meyer & Evans, Reference Klein, Zatorre, Milner, Meyer, Evans and Paradis1995; Peltola, Tamminen, Toivonen, Kujala & Näätänen; Reference Peltola, Tamminen, Toivonen, Kujala and Näätänen2012; Perani et al., Reference Perani, Paulesu, Sebastián-Gallés, Dupoux, Dehaene, Bettinardi, Cappa, Fazio and Mehler1998). With respect to phonological processing, the most common tool to measure vowel or consonant processing in monolinguals and sequential bilinguals has been a specific type of neurophysiological measure, the mismatch negativity or MMN (e.g., Näätänen, Lehtokoski, Lennest, Luuki, Alliki, Sinkkonen & Alho, Reference Näätänen, Lehtokoski, Lennest, Luuki, Alliki, Sinkkonen and Alho1997). However, this measure has been only recently applied in a SB population (Peltola et al., Reference Peltola, Tamminen, Toivonen, Kujala and Näätänen2012).
The MMN is an early event-related brain potential (ERP) component sensitive to change detection and discrimination in auditory stimulation (Näätänen & Picton, Reference Näätänen and Picton1987). The classic method for eliciting MMN is via the oddball paradigm (for review: Näätänen, Paavilainen, Rinne & Alho, Reference Näätänen, Paavilainen, Rinne and Alho2007), which involves presenting a sequence of identical stimuli (referred to as standards) in which rare irregular stimuli (referred to as deviants) appear 5 to 30 percent of the time. Participants normally listen to such stimuli streams passively without paying direct attention, while watching silent videos or reading books. In the corresponding ERPs, both standards and deviants elicit an early negativity (around 100 ms after stimulus onset), called N100 or N1, followed by a positive peak (around 200 ms after stimulus onset) called P200 or P2. Importantly, only the deviant stimulus evokes an additional MMN component (e.g., a negative wave with peak latencies between 150 and 300 ms). This MMN effect is most evident in the ERP difference wave, when the ERP response to the standard is subtracted from the response to the deviant stimuli.
It has been demonstrated that the phonemic status of the speech sounds might enhance the amplitude or decrease the latency of the MMN as compared to components elicited by non-phonemic contrasts, indicating that language experience shapes the early pre-attentive processing of speech sounds (e.g., Dehaene-Lambertz, Reference Dehaene-Lambertz1997; Näätänen et al., Reference Näätänen, Lehtokoski, Lennest, Luuki, Alliki, Sinkkonen and Alho1997; Sharma & Dorman, Reference Sharma and Dorman1999, Reference Sharma and Dorman2000; Winkler, Kujala, Tiitinen, Sivonen, Alku, Lehtokoski, Czigler, Csépe, Ilmoniemi & Näätänen, Reference Winkler, Kujala, Tiitinen, Sivonen, Alku, Lehtokoski, Czigler, Csépe, Ilmoniemi and Näätänen1999). For example, Winkler et al. (Reference Winkler, Kujala, Tiitinen, Sivonen, Alku, Lehtokoski, Czigler, Csépe, Ilmoniemi and Näätänen1999) investigated the vowel discrimination abilities of native Finnish and native Hungarian participants in response to Finnish and Hungarian vowel contrasts that either represent an across-phonemic category change in both languages, or contrasts that represent a phonemic change in Finnish, but a within-category vowel contrast in Hungarian (and vice versa). Both the within-category and the across-phonemic category changes elicited an MMN in both language groups; however, the participants exhibited an earlier latency for the across- than the within-category vowel contrasts.
To explain the enhancement of the MMN (in terms of larger amplitude or shorter latency) in the case of phonemic (versus non-phonemic) contrasts, Näätänen et al. (Reference Näätänen, Lehtokoski, Lennest, Luuki, Alliki, Sinkkonen and Alho1997) suggested that traces of categorical prototypes in cortical long-term memory (auditory cortex) serve as “recognition patterns” for native phonemes, facilitating the correct perception of speech sounds. Thus, if a given vowel happens to be a prototype of a phonemic category, its percept is reinforced by the corresponding recognition pattern stored in long-term memory. During the presentation of deviant sounds that are prototypical (or phonemic), this effect elicits a larger MMN when contrasted with non-phonemic deviants. Furthermore, it has been generally concluded that phoneme representations are based on language-specific cortical long-term memory traces, which begin to develop in the first year of life (e.g., Peltola, Kujala, Tuomainen, Ek, Aaltonen & Näätänen, Reference Peltola, Kujala, Tuomainen, Ek, Aaltonen and Näätänen2003).
A more recent study by Peltola et al. (Reference Peltola, Tamminen, Toivonen, Kujala and Näätänen2012) investigated the neural correlates of speech perception in Finnish–Swedish simultaneous (who began to learn both languages during infancy) and sequential bilinguals (who began to learn Finnish during infancy and Swedish later in life), and reported differences in the neural organization of vowel discrimination across the two bilingual groups. MMN responses elicited by two closed rounded vowels (/y/, and /ʉ/) were investigated. Importantly, the two vowels cross a vowel category boundary in Finnish (phonemic difference), but the same vowel pair represents a within-category change in Swedish (allophonic difference). Each bilingual participant was tested twice on different days using the same oddball paradigm: once in a Finnish linguistic context, and once in a Swedish linguistic context.
Peltola et al. (Reference Peltola, Tamminen, Toivonen, Kujala and Näätänen2012) reported two types of effects in the MMN responses. First, only the MMN amplitudes elicited by sequential but not by simultaneous bilinguals (SBs) were affected by the linguistic context. Sequential bilinguals exhibited a larger MMN response in the Finnish (L2) context (in which the vowels represented two different phonemes), and showed no significant MMN response in the Swedish (L1) context (in which the vowels represented an allophonic change); yet, the MMN amplitudes exhibited by SBs were not affected by linguistic context. Second, the latency of the MMN was slightly influenced by the linguistic context in the SB group, as the ERP response was somewhat faster in the Finnish context (phonemic change) as compared to the Swedish context (allophonic change). According to the authors’ interpretation, the results related to amplitude changes clearly illustrate that SBs, but not sequential bilinguals, rely on a shared phonological space across their languages (“one-store model”), as the MMNs of SBs were not affected by linguistic context.
Similar results have been reported for Spanish–English sequential bilinguals’ discrimination of consonants (Garcia-Sierra, Ramirez-Esparza, Silva-Pereyra, Siard & Champlin, Reference Garcia-Sierra, Ramirez-Esparza, Silva-Pereyra, Siard and Champlin2012; but see Winkler, Kujala, Paavo & Näätänen, Reference Winkler, Kujala, Paavo and Näätänen2003). When sequential bilingual participants passively attended to phonemic and non-phonemic VOT changes in English and Spanish language contexts, their MMN patterns suggested that pre-attentive auditory change detection is shaped by linguistic context.
1.2 The present study
The present study was designed to explore the MMN responses to both acoustically similar and distinct cross-language vowel category representations in Canadian English and Canadian French SB adults and in monolingual speakers of these languages. As opposed to previous investigations with bilinguals (Garcia-Sierra et al., Reference Garcia-Sierra, Ramirez-Esparza, Silva-Pereyra, Siard and Champlin2012; Peltola et al., Reference Peltola, Tamminen, Toivonen, Kujala and Näätänen2012), we were particularly interested in the processing of cross-language speech sounds when no specific external linguistic context is established by using a language-specific task during the passive oddball paradigm (e.g., to establish language context, participants could read magazines in the context-appropriate language).
The advantage of obtaining MMNs without establishing an external language context is that identical testing conditions across monolinguals and bilinguals are provided. The presence of language context effects in monolingual settings tends to minimize actual differences between monolinguals and bilinguals (e.g., Garcia-Sierra et al., Reference Garcia-Sierra, Ramirez-Esparza, Silva-Pereyra, Siard and Champlin2012) making it difficult to establish the degree to which learning two languages from birth shapes the perceptual processing of speech. On the other hand, a potentially more revealing “bilingual mode” (when both languages are activated) can only be applied, naturally, with bilinguals but not with monolinguals. Thus, using a measure such as the MMN (without language context manipulations) that is sensitive to subtle differences in phoneme perception but not necessarily susceptible to external context effects may fill an important methodological gap for understanding bilingual processing, because it taps into the earliest processing level that precedes conscious perception of a sound that is unavailable for investigation by traditional behavioral methods (Sams, Paavilainen, Alho & Näätänen, Reference Sams, Paavilainen, Alho and Näätänen1985; Näätänen, Reference Näätänen1987).
Based on a previous behavioral investigation of SB and monolingual populations in our laboratory (Molnar, Polka, Baum & Steinhauer, Reference Molnar, Polka, Baum and Steinhauer2010), we selected three vowels to be used in the oddball paradigm, as illustrated in Figure 1. The first vowel (V1) was an /u/ sound corresponding to the best exemplar (or prototype) of the native Canadian French vowel (as in the French word douze “twelve”); V2 was an /u/ sound representing the prototypical Canadian English vowel (as in the English word you); and V4 corresponded to the best /y/ vowel (as in the French word sucre “sugar”) as identified by native Canadian French speakers. Importantly, this vowel (/y/) is non-phonemic in English. In a second step, we also selected a fourth control vowel that falls between the English /u/ and the French /y/, along an F1/F2 continuum. With the purpose of achieving a fully symmetrical design in terms of psychoacoustic differences, we chose this control vowel (V3) such that its perceptual distance (on a Bark scale) from the French /y/ (V4) was identical to the distance between the English /u/ (V2) and the French /u/ (V1).
Vowel stimuli as described by the F1 and F2 formant values in the acoustic vowel space. The relative acoustic distance as calculated by Bark values is identical between V1 vs. V2, and V3 vs. V4. The ovals are schematic representations of the vowel categories. The exact F1, F2, and F3 values of the stimuli are listed in the Method section.
In four multiple-deviant oddball paradigms (see details in Section 2 below), the discrimination of all four vowels was measured resulting in six contrasts (V1 vs. V2, V3 vs. V4, V1 vs. V4, V2 vs. V3, V1 vs. V3, and V2 vs. V4). Two of these contrasts represent a within-phonemic category change: within the French [y] (V4 vs. V3), and within the English or French [u] (V1 vs. V2). The V1–V2 contrast represents allophonic variations, and it includes vowels that are prototypical in both languages, thus similar MMN responses across all groups are expected. The V3–V4 contrast (French [y] vs. control vowel), on the other hand, represents a within-category change only for the French and SB groups. Thus different (e.g., reduced) MMN responses are expected to be elicited from the English listeners as compared to the two other groups in response to the V3–V4 vowel pair.
The rest of the four contrasts involve across-phonemic category changes: between (good or poor) exemplars of the French [y] and (good or poor) exemplars of the French/English [u]. Based on the language background of the participants, if a good/prototypical exemplar of a native vowel is present in the contrast, a MMN response should be present as well. In the case of the V2–V3 vowel pair (control vowel vs. English [u]), only poor exemplars of French vowels are present, therefore an MMN might be present in the English, but not in the French group. However, the relatively small acoustic difference between V2 and V3 might not be sufficient to elicit a MMN in either of the groups. Based on the fractional view of bilingualism, SBs are expected to show a pattern similar to the monolingual English group. In contrast, in keeping with the holistic view of bilingualism, SBs would be expected to show a different pattern from the monolingual groups.
The V1–V2 and V2–V4 contrasts represent two relevant pairs for the bilinguals, because both vowels within each pair belong to different languages. However, the V1–V2 also reflects within-category variations for both the English and French vowels, whereas the V2–V4 pair represents clear categorical changes across the languages. If SBs are sensitive to differences that are relevant for distinguishing phonemes across languages, SBs are expected to show different MMN patterns (e.g., reflected in amplitude or latency) from the monolingual listeners, at least in response to the V2–V4 contrasts.
2. Method
2.1 Participants
Forty-nine right-handed participants with no history of speech, language, or hearing impairment signed an informed written consent to take part in the experiment. EEG recordings of four participants were excluded from the analysis due to technical problems (two) and poor data quality caused by artifacts (another two). Participants’ language background was assessed using three measures: (i) a detailed language questionnaire filled out via interview with a proficient bilingual research assistant; (ii) the Language Experience and Proficiency Questionnaire (LEAP-Q) that was developed specifically to evaluate bilingual and multilingual individuals’ language background (Marian, Blumenfeld & Kaushanskaya, Reference Marian, Blumenfeld and Kaushanskaya2007); and (iii) a five-minute speech sample evaluated by monolingual speakers of Canadian English (n = 3) and Canadian French (n = 3) using a scale from 1 to 5, where 5 represents native-like ability and 1 indicates no ability in the given language. Based on this assessment (as detailed below), 15 monolingual English speakers from English-speaking Canada (average age = 25 years, seven females), 15 monolingual French speakers from French-speaking Canada (average age = 26 years, seven females), and 15 simultaneous bilingual speakers of Canadian English and Canadian French (average age = 23 years, nine females) were included in the final analysis.
Only participants who fit the following criteria were included in the monolingual English and monolingual French groups: (i) parents of the participants were monolingual speakers of the appropriate language; (ii) participants were educated in monolingual school settings in the appropriate language; (iii) they began formal learning of a second language in school settings not earlier than age 10; (iv) they did not use a second language on a regular basis, having rated their ability in a second language with a maximum of 4 out of 10; and (v) their native speech sample was rated 5 (native-like) on average.
Inclusion criteria for the SB group were the following: (i) participants had been exposed to both languages since birth, both languages are spoken in their homes, and the participants still use both languages on an everyday basis; (ii) their schooling was completed, at different points, in both English and French; (iii) participants rated their language abilities for both languages with a minimum of 9 out of 10 and the bilingual interviewer confirmed their self-ratings; (iv) they had been living in bilingual areas of Canada, such as Montreal or the Ottawa region; and (v) the monolingual raters evaluated their speech samples in both languages with a minimum average of 4.5 out of 5.
2.2 Stimuli and procedure
Based on our previous behavioral experiment (Molnar et al., Reference Molnar, Polka, Baum and Steinhauer2010), we selected four synthesized vowels to represent a good exemplar of Canadian French /u/ (F1 = 275 Hz, F2 = 745 Hz), a good exemplar of Canadian English /u/ (F1 = 300 Hz, F2 = 979 Hz), a good exemplar of Canadian French /y/ (F1 = 275 Hz, F2 = 2011 Hz), and a control vowel representing a non-prototypical Canadian French /y/ (F1 = 300 Hz, F2 = 1597 Hz), as illustrated in Figure 1 above. Each token had fixed F3 (2522 Hz) values and was 400 ms in duration including a 35 ms rise and fall time.
The vowels were presented at 65 dB HL via insert earphones (Etymotic Research) in four different experimental blocks, as illustrated in Figure 2. Each block contained a total of 1000 stimuli: 790 standard sounds and 210 deviants (70 of each deviant) presented with a 1600 ms ISI. Within each block, standard and deviant sounds were in a pseudo-random order ensuring that at least two standard sounds preceded one deviant token. All four blocks were played within one testing session in a pseudo-randomized order, counterbalanced across participants (Latin square) and across groups. During the EEG recording, participants were sitting in a comfortable armchair in an electrically shielded sound-attenuated booth watching a silent movie. The experimental sessions lasted approximately 3.5 hours including preparation time (approximately 40 minutes), and breaks (approximately 30 minutes).
(Colour online) Four experimental blocks presented to each participant in a randomized order. The two arrows illustrate the ERP averaging technique applied in the current study. For instance, when calculating the averages for standards and deviants in the case of the V1 vs. V2 contrast, first the averages of V1 and V2 as standards, then the averages of V2 and V1 as deviants were calculated (but only for those cases where the respective other relevant vowel served as the standard; see arrows). In this way, we obtained an average ERP for the standards and for the deviants that was unaffected by the physical characteristics of the stimuli, and only the oddball effect was present when comparing the standard and deviant waveforms. The same averaging technique was applied with the rest of the contrasts.
In the oddball paradigm, the vowels were presented using a relatively long inter-stimulus–interval (ISI) of 1600 ms. The ISI is the duration of the silent gap between two sounds as measured from the offset of one stimulus sound to the onset of the next stimulus. EEG oddball paradigms normally employ relatively short ISIs (approximately 500 ms) in order to sufficiently strengthen the short-term memory trace for the repeated standard sounds that develops online during the experiment. The MMN is elicited when the deviant stimulus is compared against this trace and results in a mismatch. That is, the shorter the ISI, (i) the less likely it is that this trace in short-term memory will decay between two stimuli and (ii) the more standards can contribute to the trace within the relevant time window (a few seconds). In contrast, with a longer ISI, the short-term memory trace decay is more complete, and thus the sound discrimination process underlying the MMN may rely to a larger extent on the categorization of the first sound based on the recognition pattern in long-term memory representations. In fact, behavioral speech discrimination studies (Carney, Widin & Viemeister, Reference Carney, Widin and Viemeister1977; Pisoni, Reference Pisoni1973; Werker & Logan, Reference Werker and Logan1985) have illustrated that language-specific (“phonemic”) effects, which rely on neural traces developed for long-term memory representations of speech sounds, are more obvious when the ISI is relatively long (at least 1000 ms), while a short ISI tends to primarily activate acoustic short-term memory, thus tapping into acoustic processing that may be influenced by both short-term acoustic memory and long-term memory of phonemic traces.
The primary interest of the current study is to examine memory traces developed in long-term memory (since birth), which plays a major role in everyday speech processing, and to minimize the role of online developing memory traces. To achieve this, a relatively long ISI (1600 ms) was implemented in the experiment presented here.Footnote 1
2.3 EEG recording
EEG was continuously recorded (500 Hz/32 bit sampling rate; Neuroscan Synamps2 amplifier) from the scalp with cap-mounted Ag-Ag/Cl electrodes (Electro-cap International, Inc, Eaton, OH) from 20 sites on the scalp, based on the international 10–20 system of electrode placement: Fp1, Fp2, F7, F8, F3, F4, T3, T4, C3, C4, T5, T6, P3, P4, O1, O2, Fpz, Fz, Cz, Pz, and Oz. Vertical (EOGV) and horizontal (EOGH) eye-movements were monitored by bipolar electrodes placed above and below the left eye, and at the outer corner of each eye, respectively. All electrodes were referenced against the right mastoid, and an electrode located between Fz and Fpz provided the ground. Electrode impedances were kept under 3 kOhm.
2.4 Data analysis
EEG data were analyzed using the Brain Vision Analyzer software (Brain Products GmbH, Germany), including offline band-pass filtering (0.5 to 30 Hz)Footnote 2 and artifact rejection with a ±50 μV deviation criterion at all channels except for Fp1 and Fp2, which were clearly more affected by eye movements than the rest of the channels. Consequently, Fp1 and Fp2 were excluded from any further ERP analysis and data processing. Artifact rejection resulted in data loss within a range of 3.45% and 11.09% across participants.
ERP averages were time-locked to vowel onset and were computed separately for the standard and deviant conditions of each vowel. Only averages for standards immediately preceding a deviant stimulus were included in the analyses; therefore, approximately the same number of standards and deviants contributed to the final grand averages. Epochs (900 ms) included a 100 ms baseline starting 50 ms before and ending 50 ms after stimulus onset, as each stimulus was fully audible after 50 ms of stimulus onset only.
As illustrated in Figure 2 above, all the vowels were presented both as standards and as deviants across the four different presentation blocks, unlike in most MMN studies which assign the role of standard to one specific sound alone and the role of deviant to the other (remaining) sound(s) of interest. However, different sounds, due to differences in their pure physical properties, might elicit different N1 and P2 components (overlapping with the MMN) even when they are not presented in an oddball task. Therefore, if the role of standard and deviant is confounded with the different sounds used in an oddball paradigm, the MMN components might be partially influenced by the physical differences between the sounds, in addition to the oddball effect under investigation (for related discussion of ERP artifacts, see also Steinhauer & Drury, Reference Steinhauer and Drury2012).
Another reason why it is crucial to employ all the sounds as both standards and deviants is to avoid any confounds with the order of stimulus presentation. It has been demonstrated in behavioral studies (Natural Referent Vowel Hypothesis; Polka & Bohn, Reference Polka and Bohn2003; Polka & Bohn, Reference Polka and Bohn2011), and even in ERP investigations (Polka, Molnar, Baum, Ménard & Steinhauer, Reference Polka, Molnar, Baum, Ménard and Steinhauer2009), that the discriminability of sounds is affected by the order in which they are presented relative to each other. In addition, whether the sound presented first is more prototypical in the given language compared to the following sounds (or the other way around) has also been shown to influence discrimination performance (Perceptual Magnet Effect; Kuhl, Reference Kuhl1991).
In order to minimize the ERP effects due to the pure physical properties of the vowels, and to diminish potential effects due to the order of presentation, averages for each contrast of interest were constructed to include both vowels of the pair in both positions (as standard and as deviant). For instance, in case of the V1–V4 contrast, the standard condition included ERPs collected during the presentation of V1 (while V4 was the deviant) and V4 (while V1 was the deviant) as standards; naturally, the deviant condition included ERPs collected for V1 (when V4 served as standard) and V4 (when V1 was presented as standard). In this way, all participants contributed a minimum of 51 trials (out of 70) for each vowel (e.g., a minimum of 102 trials for each of the 12 main conditions reported below) contributed to the final averages and entered the statistical analyses.
ERP components were quantified by means of amplitude averages in three representative consecutive time windows (140–180 ms, 210–260 ms, and 440–580 ms), selected according to visual inspection of the grand average waveforms, and corresponding to time points associated with the N1, MMN, and the late negativity that emerged (see Figure 3). Using amplitude average as the dependent variable, separate Analyses of Variance (ANOVAs) were carried out for each time window (140–180 ms, 210–260 ms, 440–450 ms) and vowel pair (V1–V2, V2–V3, V3–V4, V1–V4, V2–V4, and V1–V3) across six electrodes (F3, F4, Fz, C3, C4, and Cz) out of the 20, as the effects were most prominent at these sites and no additional components of interest emerged at the other sites. Each ANOVA included the within-subject factors Condition (standard, deviant) and Electrode (F3, F4, Fz, C3, C4, Cz) as well as the between-subject factor Group, with three levels based on the participants’ language background (monolingual French, monolingual English, and simultaneous bilingual).
Average ERPs across all three language groups and all six vowel contrasts on the Fz electrode reflecting the overall ERP pattern elicited by the standard (blue line) and deviant (red line) sounds. The black line illustrates the difference wave (deviant condition minus standard condition). The first grey area corresponds to the N1 effect (140–180 ms), the second area corresponds to the MMN effect (210–260 ms), and the final grey are corresponds to the late negativity effect (440–580 ms).
3. Results
As expected (based on Figure 3), ERP plots collapsed across all contrasts and all three groups were generally characterized by an N100-P200 complex in both the standard and the deviant conditions, and the deviants elicited enhanced negativities in three distinct time intervals (140–180 ms, 210–260 ms, and 440–450 ms). Difference waves computed by conditions revealed the expected MMN component in most of the conditions, which began at around 210 ms and reached its peak around 230 ms. In addition, an early negativity with an onset around 140 ms and peak around 160 ms emerged in certain conditions, corresponding to the typical N1 auditory component. Finally, a late negativity peaking between 440 ms and 580 ms after stimulus onset is also visible in the difference waves of most conditions. We focus on the MMN results first, followed by the N1, and finally the results related to the late negativity.
3.1 MMN (210–260 ms)
In all figures, the thin solid line represents the ERPs in response to the standard sounds, and the thick line shows the average ERPs in response to the deviant vowels. The dashed line represents the difference wave obtained by subtracting the standard condition from the deviant one revealing the MMN components that begin around 210 ms with an offset at about 260 ms.
Figure 4a illustrates the ERPs averaged across groups recorded in response to the contrast of V1 (French /u/) vs. V2 (English /u/), separately for each group. The difference wave (dashed line) reveals the emergence of an MMN between 210 ms and 260 ms, confirmed in an ANOVA, which yielded a significant main effect of condition (F (1,42) = 12.534; p = .001) with no significant group effect (F < 1) or interaction (F < 1). Figure 4b displays ERPs computed for each language group for the V3 (control /y/) vs. V4 (French /y/) vowel contrast. The ANOVA revealed a main effect of condition (F (1,42) = 17.6; p = .001), as well as a group x condition interaction (F (2,42) = 3.651; p = .037). Follow-up analyses within groups showed that only the French group (F (1,14) = 6.42; p = .026), and the SB group (F (1,14) = 9.15; p = .01) exhibited a significant MMN response, whereas the English group did not (F < 1).
Average ERPs recorded on the Fz electrode presented for the three language groups reflecting discrimination between within-phonemic categories acoustically characterized by both F1 and F2 changes: (a) V1 and V2, (b) V3 and V4. Information on the phonemic status of the vowel pairs is provided for each contrast by language group. For significant MMN responses the peak latency values (in ms) and the standard errors in brackets are also listed. The grey shaded areas represent the interest of analysis and are only used if significant differences between standard and deviant sounds reached significance within the time windows.
Figure 5a illustrates the V1–V4 contrast. The ANOVA revealed a main effect of condition (F (1,42) = 6.118; p = .017), with no group effect (F < 1), nor any interaction (F < 1), confirming that all three language groups showed a similar MMN between 210 ms and 260 ms. ERPs computed for the V2 (English /u/) vs. V3 (control /y/) contrast are illustrated in Figure 5b. The ANOVA revealed a main effect of condition (F (1,42) = 7.1; p = .01), as well as an interaction between group and condition (F (2,42) = 2.02; p = .03). A clear MMN in the 210–260 ms time window was only recorded for the SB group. Follow-up analyses within the groups revealed a significant effect of condition only in the SB group (F (1,14) = 6.5; p = .025), with no significant effects in either monolingual group (mE group, F < 1, mF group, F < 1).
Average ERPs recorded on the Fz electrode presented for the three language groups reflecting discrimination between across-phonemic categories acoustically characterized by F1 changes only: (a) V1 and V4, (b) V2 and V3. Information on the phonemic status of the vowel pairs is provided for each contrast by language group. For significant MMN responses the peak latency values (in ms) and the standard errors in brackets are also listed. The grey shaded areas represent the interest of analysis and are only used if significant differences between standard and deviant sounds reached significance within the time windows.
Results obtained in response to the other two contrasts (V1–V3, and V2–V4) that represent across-phonemic changes (with acoustic differences in both F1 and F2) are illustrated in Figure 6. For the V1–V3 vowel contrast (Figure 6a), the ANOVA revealed a main effect of condition (F (1,42) = 5.307; p = .027), as well as an interaction between group and condition (F (2,42) = 3.8; p = .031). Follow-up analyses within the groups revealed a significant effect of condition in the SB group (F (1,14) = 6.5; p = .021, and in the French group (F (1,14) = 5.5; p = .031), with no significant differences in the English group (F < 1). ERPs computed for the V2–V4 (control /y/) contrast are illustrated in Figure 6b. The ANOVA revealed a main effect of condition (F (1,42) = 4.082; p = .05), as well as an interaction between group and condition (F (2,42) = 3.062; p = .036). Consistent with the ERP plots, follow-up analyses within the groups revealed a significant effect of condition only in the English group (F (1,14) = 3.5; p = .049), and in the French group (F (1,14) = 3; p = .05) with no significant differences in the SB group (F < 1).
Average ERPs recorded on the Fz electrode presented for the three language groups reflecting discrimination between across-phonemic categories acoustically characterized by both F1 and F2 changes: (a) V1 and V3, (b) V2 and V4. Information on the phonemic status of the vowel pairs is provided for each contrast by language group. For significant MMN responses the peak latency values (in ms) and the standard errors in brackets are also listed. The grey shaded areas represent the interest of analysis and are only used if significant differences between standard and deviant sounds reached significance within the time windows.
In addition to the amplitude analyses, we examined peak latencies for all significant MMNs within each language group for each vowel pair. These values, which reflect the group average peak latencies found within the time window of the MMN, are displayed in each figure. A one way ANOVA conducted on the MMN peak latencies showed no significant differences across language groups, with the exception of the V1 vs. V2 vowel pair, where a marginally significant main effect of language group was observed (F (1,42) = 3.309; p = .052). Post-hoc analyses contrasting the three groups showed that the MMN peak for the SB group (213 ms) occurred significantly earlien than those in the other two groups (SB vs. English p = .05; SB vs. French p = .05), whereas the two monolingual groups did not differ from each another (peak latencies in both groups: 247 ms).
3.2 N1 (140–180 ms)
Prior to the MMN effects, the ERP averages revealed that the deviant condition elicited a larger N1 component than the standard condition in response to the V1 vs. V4 and the V1 vs. V2 contrasts (clearly reflected in the difference waves). This occurred within the time window of 140 ms and 180 ms after stimulus onset, as highlighted by the first grey shaded area in the figures. In the case of the V1 (French /u/) vs. V4 (French /y/) contrast, a main effect of condition was observed (F (1,42) = 8.78; p = .005) with no interaction with group, demonstrating that all three groups exhibited this N1 condition effect. Similarly, the main effect of condition approached significance (F (1,42) = 3.7; p = .0611) in the ANOVA for the V1 (French /u/) vs. V2 (English /u/) contrast, with no group interaction. No main effects or interactions were observed for the remaining four vowel pairs.
3.3 Late negativity (440–580 ms)
Interestingly, discrimination effects were also signaled by a late negativity following the MMN components in five (out of six) vowel contrasts. This late negativity effect is visible in the grand average ERPs highlighted by the third grey-shaded area, as well as in the corresponding difference waves, emerging at about 440 ms and lasting almost until 600 ms after stimulus onset.
The ANOVA for the V1 vs. V4 contrast (Figure 4) revealed that this late negativity was elicited across all groups, as reflected in a main effect of condition with no group interaction (F (1,42) = 10.122; p = .003). A similar shared pattern emerged for the V1 vs. V2 contrast, as shown in Figure 4a (main effect of condition with no group interaction (F (1,42) = 13.412; p = .0001), and for the V3 vs. V4 comparisons, as shown in Figure 4b (main effect of condition without group interaction: F (1,42) = 10.1; p = .003). In the case of the V1 vs. V3 contrast (Figure 6), the main effect of condition was also significant (F (1,42) = 11.03; p = .002) with no group interaction, as well as in the V2 vs. V4 contrast (F (1,42) = 12.03; p = .001).
4. Discussion
The overall goal of the current study was to investigate the long-term neural memory traces (as reflected by the MMN component) developed for native vowels when the overall language organization of the participants is based on acquiring one or two languages from birth. The experiment was designed to test: (i) whether the pre-attentive speech processing in SBs is tuned to the linguistic characteristics of the speech input from two languages, such that both within-language and cross-language phonetic differences are present in a single phonetic space in which access to sub-phonemic differences across the two languages is similar to that of monolingual peers; and (ii) whether SB listeners are sensitive to additional sub-phonemic detail in each language which might be irrelevant for monolinguals. The latter possibility is based on the assumption that the SBs’ ability to adapt their perception to function optimally in different language contexts may require early, automatic access to a higher level of phonetic detail.
Given the overall ERP results, it is evident that SBs and monolinguals (who are all native users of English and/or French) exhibit different neurophysiological patterns in response to native and cross-language vowels, in support of the holistic view of bilingualism (Grosjean, Reference Grosjean1998). Discrimination of the vowels was signaled by different patterns of the MMN triggered in the time window of 210 ms to 260 ms across the three language groups. Vowel discrimination was also apparent in the N1 effect (140–180 ms), and in a late negativity occurring between 440 ms and 580 ms after stimulus onset. The latter two ERPs showed no evidence of variation as a function of linguistic background.
4.1 Mismatch negativity (MMN)
Focusing on the MMN findings first, the within category phonemic contrasts (V1–V2 and V3–V4) produced results similar to previous investigations (e.g., Aaltonen, Niemi, Nyrke & Tuhkanen, Reference Aaltonen, Niemi, Nyrke and Tuhkanen1987; Näätänen et al., Reference Näätänen, Lehtokoski, Lennest, Luuki, Alliki, Sinkkonen and Alho1997; Peltola et al., Reference Peltola, Kujala, Tuomainen, Ek, Aaltonen and Näätänen2003). As expected, the English monolingual speakers did not show an MMN in response to the V3–V4 contrast, as this part of the vowel space carries no relevant phonemic information in English (as a reminder, /y/ is not part of the English vowel inventory). The relatively small acoustic difference between these vowels (in terms of F1) appeared to be insufficient to elicit an MMN component in this group, (unlike the larger acoustic change between V1 and V4). However, the V3–V4 contrast (good vs. poor French /y/) did elicit an MMN response in the French and the SB groups, thus illustrating language-specific effects, as V3 and V4 are within-phoneme-category representatives for both the French and SB listeners. Since the MMN showed identical features in terms of latency and amplitude across the two groups, having acquired French as a monolingual or bilingual does not seem to lead to a different perceptual organization for this specific phonemic category, which has no close English “neighbors”.
The V1–V2 stimulus pair represents an interesting phonetic contrast, as it contains the best exemplar of the English [u] (V2) and the best exemplar of the French [u] (V1), differentiated by the exact same acoustic–perceptual distance as the V3–V4 contrast. In response to this contrast, each group exhibited a reliable MMN. Similar MMNs across the groups are expected because at least one of the vowels in this particular contrast is prototypical to the listeners. Interestingly, the SB listeners exhibited a faster peak latency compared to the other groups, pointing to an MMN response that is distinct from those of the monolingual listeners. This finding cannot be attributed to the SB group being generally faster than monolinguals in pre-attentive, automatic processing of speech sounds because this group difference was only present for the V1–V2 contrast, and not for the V3–V4 contrast, which represents a comparable acoustic–perceptual difference; the two sets of vowel contrasts only differ in terms of phonemic status.
Typically, the latency of MMN responses decreases when the stimuli pair crosses a phonemic boundary (Näätänen, Jacobsen & Winkler, Reference Näätänen, Jacobsen and Winkler2005). Therefore, this distinct neurophysiological response in terms of latency suggests that SB listeners distinguish these vowels as different phonemic categories (English /u/ and French /u/) rather than as within-category variants (allophones). Several behavioral experiments have already illustrated that sequential bilinguals are able to perceptually separate acoustically overlapping cross-language categories when the ongoing linguistic context is highly controlled (e.g., Bohn & Flege, Reference Bohn and Flege1993; Elman, Diehl & Buchwald, Reference Elman, Diehl and Buchwald1977; Garcia-Sierra et al., 2009). Even though such experiments have demonstrated that proficient sequential bilinguals are able to adjust their perception to the ongoing language context (L1 or L2), it has still remained unclear whether a bilingual's ability of perceptual adjustment was supported by separate phonemic categories across the languages, or whether bilinguals rely on allophonic information during the process.
The across-phonemic category contrasts that are defined by changes in F1 only (V1–V4, V2–V3) elicited different MMN patterns as a function of SB vs. monolingual status only in the case of the V2–V3 vowel pair; the V1–V4 contrast elicited comparable MMNs across the groups. The largest acoustic difference extends between stimuli V1 and V4 (French /y/ vs French /u/) and corresponds to a clear cross-category (phonemic) difference in French and to a perceptible acoustic change (and/or a possible phoneme category change in English.) As shown by our previous behavioral investigations, the prototypical French [y] is categorized as a very poor exemplar of the English [i] by most English listeners. Therefore, the MMN observed in the English group might be due to categorical overlap between the French [y] and the English [i] vowels. Additionally, language differences are likely minimized by the large acoustic differences, given that ERPs recorded during the processing of contrasts with smaller acoustic differences (V1–V2, V3–V4, V2–V3, V1–3, and V2–V4) did reflect language-specific effects on the MMN, similar to previous investigations (e.g., Aaltonen et al., Reference Aaltonen, Niemi, Nyrke and Tuhkanen1987; Näätänen et al., Reference Näätänen, Lehtokoski, Lennest, Luuki, Alliki, Sinkkonen and Alho1997; Peltola et al., Reference Peltola, Kujala, Tuomainen, Ek, Aaltonen and Näätänen2003).
The finding that the monolingual English group exhibited an MMN only in response to either a large acoustic difference (V1 vs. V4) or to a native within-category change (V1 vs. V2), and the monolingual French and the SB group showed MMN responses to across- and within-category variations, implies that MMNs can be elicited by either a large acoustic difference or by contrasts where at least one of the vowels belongs to (or is a good category exemplar of) the native vowel system. Similar MMN effects have emerged in previous investigations in regard to consonant perception (e.g., Rivera-Gaxiola, Csibra, Johnson & Karmiloff-Smith, Reference Rivera-Gaxiola, Csibra, Johnson and Karmiloff-Smith2000; Shafer, Schwartz & Kurtzberg, Reference Shafer, Schwartz and Kurtzberg2004).
The processing of the V2–V3 contrast revealed an important difference between SB and monolingual processing. The V2 and V3 vowels only differ in their F2 values, sharing a common F1 value (300 Hz); in comparison, the other two within-category contrasts (e.g., V1–V2 and V3–V4) differ in both F1 and F2 values. The SB listeners’ MMN response to the V2–V3 pair suggests that the subtle difference between these two vowels is perceptually prominent for SB listeners only; thus SB listeners show sensitivity to a wider range of sub-phonemic differences (specifically, changes along the F2 axis of the vowel space) than their monolingual peers.
Interestingly, SB listeners in the current study appear to resolve and encode more detail in a part of the vowel space that is “crowded” when all native cross-language vowels are considered. Moreover, it is noteworthy that the SB listeners are resolving more detail in F2, given that the highest vowels in French (/i/, /u/ and /y/) and English (/i/, /u/) are differentiated across these languages primarily by their F2 values. Therefore, it is highly adaptive for SBs to be more sensitive to changes in F2 than their monolingual peers who encounter only a subset of these high vowels in a less crowded space. This finding is consistent with several previous studies in showing that SB listeners differentiate, rather than assimilate, highly similar phones across their languages (e.g., Guion, Reference Guion2003; Sundara & Polka, Reference Sundara and Polka2008; Sundara et al., Reference Sundara, Polka and Baum2006a). The present findings, similar to prior reports, also show that this perceptual adaptation in SBs may involve resolving differences that are not phonemic in either language and appears to give rise to some unique, and highly efficient, perception (and production) patterns in comparison to either of their monolingual peers (e.g., MacLeod & Stoel-Gammon, Reference MacLeod and Stoel-Gammon2005; Sundara & Polka, Reference Sundara and Polka2008; Sundara et al., Reference Sundara, Polka and Baum2006a, Reference Sundara, Polka and Geneseeb; Sundara, Polka & Molnar, Reference Sundara, Polka and Molnar2008).
The final pair of vowel contrasts (V1–V3, and V2–V4) that also represent across-vowel category changes (but are defined by changes in both F1 and F2) demonstrated clear differences between the phonological organization of SBs and the monolingual listeners. The V1–V3 contrast represents the prototypical French [u] (V1) and the control vowel (V3; a very poor exemplar of the French vowel [y]). The English group showed no reliable MMN response, as expected, given that none of the vowels is prototypical in Canadian English; as opposed to the V1–V4 contrast, this acoustic difference was not sufficient to elicit an MMN irrespective of the non-phonemic status of the stimuli. The French and SB groups, however, showed comparable MMN patterns, also as expected, given the French phonemic status of V1.
The ERPs in response to the V2 and V4 contrast provided a unique insight into the phonological organization of monolinguals and SBs. This vowel pair (prototypical French [u] and prototypical English [u]) signals an across-phonemic category change, but also a switch between languages or linguistic contexts for the bilinguals. Similarly to the previous vowel pairs, when there was a prototypical vowel present in the contrast, the monolingual listeners exhibited an MMN pattern; the MMN responses of SBs were attenuated, however. Similar patterns in monolinguals have been demonstrated before; monolingual German listeners’ processing of a non-native (Polish) deviant consonant was weaker in the presence of a linguistically relevant native contrast, and their responses to a native deviant consonant was increased by the presence of non-native sounds (Lipski & Mathiak, Reference Lipski and Mathiak2008). In the present study, it appears that the presence of either a prototypical French or English vowel as standard (played 70% of the time) defined a linguistic context in the oddball paradigm. Listeners exhibited MMNs when the deviants represented within-language and within-phonemic category changes (e.g., V1–V4, V1–V2, and V3–V4), although this particular contrast represented a clear change between two linguistic modalities of SBs. Accordingly, it seems that when V2 or V4 served as standards, SBs showed MMN responses to within-linguistic context variation but showed an attenuated MMN for a vowel that clearly belonged to another linguistic modality. In this case, the late negativity signaled auditory discrimination only.
Accordingly, it is possible that the phonemic status of vowels in a multiple oddball paradigm defines an internal linguistic context (independently of the presence or absence of an external linguistic context), and affects the pre-attentive auditory processing of bilingual listeners. This implies that SBs do not rely on a completely shared phonological space across their languages (contra Peltola et al., Reference Peltola, Tamminen, Toivonen, Kujala and Näätänen2012), and that language context affects bilinguals’ vowel processing, as has been demonstrated in several behavioral paradigms (e.g., Bohn & Flege, Reference Bohn and Flege1993; Elman et al., Reference Elman, Diehl and Buchwald1977; Garcia-Sierra et al., 2009).
4.2 Late negativity in the ERPs
In addition to the MMN findings, interestingly, discrimination abilities across all the vowels and language groups were reflected in a late negativity as well (except for V2–V3). The late negativity was reliably elicited in the monolingual speakers even when no MMN was present. The overall pattern of the late negativity suggests that it might be related to, but not dependent upon, the MMN component, as it can appear in the absence of the MMN as well; and it appears to signal a process related to discrimination, as it is elicited by sounds with deviant status in the oddball paradigm.
Similar late negativities have been reported in the EEG literature (e.g., Alho, Woods, Algazi & Nätäänen, Reference Alho, Woods, Algazi and Näätänen1992; Čeponienė, Cheour & Nätäänen, Reference Čeponienė, Cheour and Näätänen1998; Korpilahti, Salmela, Lang, Porn & Crause, Reference Korpilahti, Salmela, Lang, Porn and Crause1997; Trejo, Ryan-Jones & Kramer, Reference Trejo, Ryan-Jones and Kramer1995); however, the exact role of this component remains unclear. Studies investigating infants’ and children's auditory perception often record a late negativity (termed late discriminative negativity) peaking between 400 ms and 500 ms (for review, see Cheour, Korpilahti & Martynova, Reference Cheour, Korpilahti and Martynova2001), that has been shown to be greater in response to speech sounds than to tones of equal complexity (e.g., Korpilahti, Salmela, Lang, Porn & Crause Reference Korpilahti, Salmela, Lang, Porn and Crause1997). As the late discriminative negativity has been linked to developmental changes – given that it is obtained most reliably in children and its amplitude shows a decrease with age and seems to completely disappear by adulthood (Cheour et al., Reference Cheour, Korpilahti and Martynova2001) – it may not be directly related to the late negativity recorded in the current study. However, it is important to note that Čeponienė et al. (Reference Čeponienė, Cheour and Näätänen1998) demonstrated that the second negativity in children is more prominent when the ISI is longer; consequently, the duration of the ISI in the present investigation might have influenced the emergence of a second negativity in adults as well.
Only a few adult studies have reported a late negativity in a non-attended paradigm (Alho et al., Reference Alho, Woods, Algazi and Näätänen1992; Trejo et al., Reference Trejo, Ryan-Jones and Kramer1995), and it is somewhat unclear whether these negativities were related to the standard stimuli preceding the deviants (a “spill-over” processing of the previous sound) or whether they reflected a discriminatory process associated with the deviant sound, since these studies relied on a short ISI (< 400 ms). In the current study, due to the elongated ISI between standards and deviants, the late negativity cannot be related to the processing of the preceding sounds (as the late negativity occurred before the presentation of the following stimuli), and it is more likely associated with change detection in the auditory input. In addition, given that the SB group exhibited a somewhat different pattern for the late negativity as compared to monolinguals, this late component might reflect the effects of language experience as well.
4.3 N1 component
In one contrast (V1–V4), a significant N1 difference was recorded. The auditory N1 component has been shown to be primarily sensitive to the physical and temporal characteristics of the auditory stimuli; however, certain subcomponents of the N1 have been shown to be related to condition-dependent effects (for review, see Näätänen & Picton, Reference Näätänen and Picton1987). Recall that we relied on an averaging method that ruled out any negativities in the final grand averages due to the physical characteristics of the stimuli; therefore the N1 effect in our ERPs more likely reflects effects related to the characteristics of the given conditions.
Long ISIs have also been shown to modify the N1 component; for instance, Czigler, Csibra and Csontos (Reference Czigler, Csibra and Csontos1992) reported enhanced N1 amplitudes to changes in tones when the relatively long ISIs of 800 ms, 2400 ms, and 7200 ms were employed in different experimental blocks. Additionally, Rivera-Gaxiola et al. (Reference Rivera-Gaxiola, Csibra, Johnson and Karmiloff-Smith2000) recorded N1 effects in response to certain contrasts when using a long ISI of 1500 ms to illustrate cross-linguistic effects on consonant perception during a passive auditory oddball paradigm. In the current study, the N1 was most prominent in response to the largest acoustic change in the stimulus set, thus it is a possibility that the N1 effect is related to acoustic changes and not to phonemic information.
5. Conclusion
The overall MMN findings demonstrate three unique characteristics of the English–French SB perceptual system. First, acoustically similar cross-language vowel categories present in the SB speech input appear to be stored as separate categories (rather than within-category variants) in the form of long-term phonemic memory traces that are retrieved automatically during the neural path of speech processing at the pre-attentive level. Similar findings emerged from behavioral studies investigating both the production and perception of acoustically similar speech sounds (e.g., Guion, Reference Guion2003; Sundara et al., Reference Sundara, Polka and Baum2006a, Reference Sundara, Polka and Geneseeb; Sundara et al., Reference Sundara, Polka and Molnar2008), supporting the idea that bilingual speakers, instead of assimilating overlapping cross-language categories, rather tend to maintain separate categories (dissimilation). This has been primarily shown in behavioral studies when the ongoing language context was optimized to display such effects (e.g., Bohn & Flege, Reference Bohn and Flege1993; Elman et al., Reference Elman, Diehl and Buchwald1977; Garcia-Sierra et al., 2009), unlike in the current study where external language context and other task effects were minimized.
Second, SB listeners appear to be more attuned to sub-phonemic changes as compared to monolinguals, most obvious in the V2–V3 contrast. Because this contrast contains a prototypical English vowel ([u]), it was expected to elicit a significant MMN response at least in the English and SB group. However, only the SB group revealed a MMN. It appears that relatively small acoustic changes in one dimension only (F2) were not perceptually salient enough for the monolingual English listeners to elicit a statistically reliable MMN when the stimuli were presented with a long ISI. Changes in the F2 value were more relevant to the SBs as compared to the monolinguals. This can be explained by the specific organization of the English–French vowel space: F2 is the most relevant cue for differentiating the English and French vowels in this particular part of the vowel space. Therefore, SB listeners utilize the most relevant phonetic information across the two languages at the very early stages of speech perception.
Third, when the standard vowels provided a clear cue for linguistic contexts, the SB listeners’ pre-attentive processing (at the level of the MMN) appeared to be less sensitive in response to vowels that were not part of the specific context. Therefore, it appears that linguistic context (without the presence of an external context) specified by the standard sounds of the oddball paradigm affects SBs’ vowel discrimination, and the passive auditory oddball-paradigm itself can create an internal linguistic context. Also, it appears that SBs are sensitive to context effects at an early stage of processing as reflected by the MMN, but only when non-ambiguous linguistic cues are provided (e.g., V2 vs. V4), but not when the linguistic cues are ambiguous (e.g., V1 vs. V2).
In sum, results indicate that the pre-attentive process involved in SB vowel perception is shaped to detect the phonetically most salient information that is specific to each of the languages, as demonstrated by the MMN responses to the V1–V2 and V2–V3 contrasts. SB listeners of English and French exhibit a speech perception pattern that is distinct from both monolingual listener groups even during the earliest levels of speech processing, as the SB pre-attentive system is tuned to access sub-phonemic detail with respect to both of their input languages, including detail that is not readily accessed by either of their monolingual peers. This automatic access to fine phonetic detail may be essential in supporting the SBs’ ability to make rapid, effortless shifts in perception across different communication/linguistic contexts. Moreover, bilinguals appear to function in a hybrid phonological space: when a clear linguistic context is present, the context-appropriate information is the most salient; when no clear context is established, bilinguals show heightened sensitivity to cues that are salient for discriminating acoustically similar cross-language categories.
