Neural Indices of Vowel Discrimination in Monolingual and Bilingual Infants and Children

Introduction

Considerable research has tracked the time course of first-language speech perception development and has revealed that experience with a language influences how an infant categorizes speech from birth (and even in the womb) through early childhood, typically measured via behavioral preference and novelty tasks (Best, et al., 1982; DeCasper et al., 1994; Kuhl et al., 1992, Kuhl 2004). By one year of age, children have learned which phonological patterns are contrastive (phonemes) in their native language and they have already shown sensitivity to language-specific phonological information (e.g., Werker & Tees, 1983; Jusczyk, et al., 1994). However, phonological development is incomplete well into the grade-school years (Nittrouer & Miller, 1997). For example, one-year olds will fail to discriminate difficult phoneme contrasts (e.g., /ba/ and /da/ in more challenging tasks such as lexical learning) until 18 months of age, even though they demonstrate discrimination in a simpler task (e.g., Werker, et al., 2002).

Both experience and auditory cortical maturation are likely to be necessary to attain an adult-like phonological system. Studies of speech perception development have largely focused on the former variable. More recent models have acknowledged the importance of considering the developmental level of a child when interpreting speech perception results (Werker & Curtin, 2005; Curtin, et al., 2011). The goal of the current study is to examine both experience and maturation in the development of speech discrimination.

Studies of neural maturation of the auditory system have revealed considerable cortical immaturity before six months of age (e.g., Kostović & Jovanov-Milosević, 2006; Moore & Linthicum, 2007). Considerable structural maturation (e.g., number of axon collaterals, amount of myelination) and functional maturation (e.g., latency and topography of obligatory, auditory components) occurs during the first year of life, with the rate of maturation slowing down through the preschool and into the grade-school years (Moore & Lithicum, 2007). Thus internal, biological factors constrain the rate of auditory cortical maturation in neonates and young infants (Leppänen et al., 2004; Kostović & Jovanov-Milosević, 2006; Moore & Lithicum, 2007). Studies of children with congenital deafness followed by cochlear implant indicate that early input also plays a role in maturation of the auditory system (Sharma, et al., 2002). What is less clear is the degree to which the characteristics of the input (e.g., monolingual versus bilingual speech input) modulate auditory development.

An additional source of evidence for biological factors constraining maturation comes from studies of sex differences. Sex differences during prenatal and early infancy periods have been observed (Basic sensory difference: see Alexander & Wilcox, 2012 for a review; Oral-motor function: Miller et al., 2006; Nagy et al., 2007). Male and female infants show different rates of cortical maturation driven by differing levels of sex hormones (e.g., Shucard et al.,1981; Shucard & Shucard, 1990; Friederici et al., 2008). In particular, female infants initially showed more rapid maturation of electrophysiological responses to tones over left compared to right hemisphere sites and male infants showed the opposite pattern (Shucard & Shucard, 1990); furthermore, one study found that female infants generally showed more linguistic rule-based learning than male infants (Mueller et al., 2012). Irrespective of the exact underlying sources of these sex differences related to cortical maturation, variations in cortical maturation can be inferred from these findings of sex differences.

Cross linguistic studies of speech perception have shown that listeners often have difficulty discriminating and categorizing non-native phonological contrasts not found in their first language (e.g., Werker & Tees, 1984; Best, 1985). In the case of second-language (L2) learning, late learners of an L2 (after approximately 11 years of age) may show poor L2 speech perception (e.g., Strange, 2011). In contrast, early learning of an L2 results in better behavioral perception of L2 phonological contrasts (e.g., Flege, 1995; Sundara & Polka, 2008; Gonzales & Lotto, 2013). Some studies, however, suggest that even early learning of an L2 (before five years of age) may result in less robust brain discrimination of L2 speech contrasts, as compared to monolingual listeners for some (e.g., Hisagi, et al., 2014), but not all early bilinguals (Datta, et al., 2019). Understanding the factors that influence L2 speech perception in adult bilinguals who had learned the L2 early in life is a challenging task. Previous studies have shown that amount of input in the first language (L1) versus the second language (L2) is likely to be a major factor (Flege, 1995; Flege & Mackay, 2004). Thus, examining L2 speech perception in the time period when amount of input can be better evaluated is desirable (Datta, et al., 2019).

Behavioral studies of bilingual language development in infants have shown that speech processing in one or both of a bilingual’s languages can be delayed if the two languages are difficult to separate (e.g., prosodically-similar languages such as Spanish and Catalan) (Bosch & Sebastián-Gallés, 2003; Sebastián-Gallés & Bosch, 2009). In cases that the languages (e.g., English and French) are easier to separate, the time course of speech perception development for both languages appears to be somewhat similar for bilinguals and monolinguals (Sundara, et al., 2008). However, in a previous event-related potential (ERP) study we found a small but significant effect of bilingual exposure on vowel speech processing in children between 3 months and 36 months for two languages (Spanish and English) that should be relatively easy to separate (Shafer, et al., 2011; 2012). Therefore, the relationship between language similarity and bilingual development may be less straightforward.

In addition to language similarity, the amount of language exposure in each language also affects speech and language development in young children (Hoff, et al., 2012; Conboy & Mills, 2006). Hoff and colleagues reported that the rate of development of each language in bilingual children is a function of their relative amount of language exposure (Hoff et al., 2012). Conboy and Mills also found a large effect of language experience on electrophysiological measures to words in the dominant versus non-dominant languages of Spanish-English toddlers (e.g., Conboy & Mills, 2006). Recently, Cattani and colleagues reported that 60% or more English exposure will lead to monolingual-like performance on English language tests in bilingual 2;6-year-olds (Cattani et al., 2014).

There has been an increased interest in examining the neurodevelopment of speech processing in bilingual children in the past 15 years (e.g., Rivera-Gaxiola et al., 2005; Shafer et al., 2010, 2012, 2015; Ferjan Ramirez et al., 2016; Garcia-Sierra et al., 2016; Rinker et al., 2017). Most studies have used an auditory oddball paradigm, in which a frontocentral neural response called Mismatch Negativity (MMN) is used to measure auditory processing (Näätänen, et al., 1997). The mature MMN is a negative peak between 100–250 ms post stimulus onset. Children older than approximately four years of age show somewhat similar neural measures of sound discrimination to adults, but sometimes at later latencies (Gomes, et al. 1999; Shafer, et al., 2000; Shafer et al., 2010). For example, Shafer et al. (2010) observed an MMN in four- to seven-year-old children to a speech sound contrast; this MMN peaked between 200 and 400 ms in the children, rather than between 100 and 250 ms, as expected for adults. The MMN shifts earlier in latency with increasing age, but does not reach adult latencies for more difficult sound contrasts until past puberty for some auditory and speech contrasts (Ponton, et al., 2000; Shafer, et al., 2000, Shafer et al., 2010).

Infants typically show a slow positive mismatch response (pMMR) in a passive oddball paradigm (Dehaene-Lambertz & Dehaene, 1994; Dehaene-Lambertz & Baillet, 1998; Trainor et al., 2001; Kushnerenko et al., 2002; Morr, et al., 2002; Leppänen et al., 2004; Friederici et al, 2007; Lee et al., 2012; Shafer et al., 2012; Partanen, et al., 2013a; Partanen, et al., 2013b). However, a few studies have observed a negative MMR (nMMR) rather than a pMMR (e.g., Cheour, et al., 1998; Trainor et al., 2001; Kushnerenko et al., 2002; He et al., 2007, 2009). With increasing age, the pMMR declines in amplitude and the nMMR emerges, peaking between 100 and 400 ms. The pMMR and nMMR may co-occur and partially overlap in infants and young children (e.g., Leppänen et al., 1997; Morr et al., 2002). It is likely that the timeframe for the emergence of the nMMR is dependent on the degree of acoustic or phonetic difference for the stimulus contrasts. For example, the nMMR emerges after three years of age to the subtle vowel contrast /I/ versus /ε/ (Shafer, et al., 2010), but to a large acoustic difference, such as 1000 versus 2000 Hz, the nMMR is observed in infants under one year of age (Morr, et al., 2002; see Kushnerenko, et al., 2013, for review). The pMMR continues to be observed (preceding the nMMR, if it co-occurs) to somewhat more difficult contrasts into the grade-school ages (Shafer, et al., 2010; Lee et al. 2012; Liu et al., 2014).

The nMMR may be equivalent to the MMN observed in adults and older children. The MMN indicates a mismatch between a stimulus or stimulus pattern (deviant) and the predictions generated by the prior context (standards) (Kushnerenko, et al., 2013). The nature of pMMR is less clear. Some studies have suggested that the pMMR is an analogue of adult P3a, indexing involuntary attention switch to stimulus change (Alho et al, 1990; Trainor et al., 2001; Kushnerenko et al., 2002). Others suggest that it reflects “an automatic categorization of the stimulus” (Friedrich et al., 2004), or an index of a general process as a result of neural adaptation (He, et al., 2009). Recent studies have found larger pMMR to more difficult than to easier stimulus contrasts (e.g., Lee, et al., 2012); these findings are more consistent with one of the latter two explanations for the pMMR. There is also evidence that a much later positive component (PC) peaking later than 300 ms and observed to novel sounds is the precursor of the adult P3a response, rather than the earlier occurring pMMR (Kushnerenko, et al., 2013).

Infants, children and adults may also show a second negative discriminative peak at superior anterior scalp sites. In an early study with adults, this response was called the re-orienting negativity (RON) (Schroger & Wolff, 1998); in studies with children, a similar response has been called the late discriminative negativity (LDN)(e.g., Cheour et al., 2001), the late MMN, the MMN2 (Korpilati, et al., 2001), or the late negativity (LN) (Datta et al., 2019; Shafer, et al., 2005; Yu et al., 2017; Yu et al., 2018). The Nc in infants may be comparable to this late negativity in children and adults. The Nc is found in a fairly late time interval (500–600 ms; Kurtzberg et al., 1984; Courschesne, 1990; Kushnerenko, et al., 2013).

Differences in the LN timing and topography, as well as different findings across studies, currently limit our understanding of what the LN reflects. Some studies suggest that the LN is modulated by phonological or lexical aspects of processing. For example, Korpilati and colleagues (2001) reported that a late MMN was significantly larger for deviant words than deviant tones, and suggested that it was related to automatic detection of lexical difference. Yu et al. (2017) found that the LN amplitude to lexical tone contrast was larger in native Mandarin listeners than monolingual English listeners when presented at a relatively long interstimulus interval (at the rate of approximately 3 second). In other studies, the LN was related to language status (language-impaired versus typically-developing) (Bishop et al. 2010; Datta et al., 2010). Alternatively, the LN may reflect enhanced auditory attention, considering that it is also observed to non-speech auditory stimuli and is modulated by novelty (Kushnerenko, et al., 2013). We observed positive responses in infants and younger children in the LN time window, and this positive response gradually became more negative. We therefore use the term “late MMR” to refer to the responses in the later time window for the remainder of the article. We refer to the MMR in the time window of 160–360 ms as “early MMR.” (eMMR). Depending on age and language background, this eMMR may be positive or negative. Other studies have used the term eMMN to refer to early Mismatch Negativity in this time window (e.g., Choudhury, Parascando & Benasisch, 2015).

The present study examines ERP correlates of processing the American English vowel contrast [ɛ]versus [ɪ] in monolingually-exposed and bilingually-exposed infants and children between three and 47 months of age. The goals of the current study were: 1) to examine how level of maturity modulates the neurodevelopment of speech processing throughout the period in which the child is mastering language (infancy to four years of age); and 2) how age, sex, and language experience modulate event-related potentials (ERPs) associated with speech-sound discrimination. We hypothesized that even a minimal amount of experience with an L2 would influence speech processing in a young child. We also expected to find age-related modulation of the ERPs. In particular, we expected to see increased negativity in both early and late MMR amplitude with increasing age (Shafer, et al., 2010). Finally, we predicted that sex would influence the early and late MMRs but more so during the first year of life than at later ages.

Materials and Methods

The experiment reported here was approved by The Graduate Center, CUNY Institutional Review Board, and was conducted in compliance with the Declaration of Helsinki. Parental consent was obtained from the parent/legal guardian(s) for each participant.

Participants

Fifty-nine bilingual Spanish-English children and 59 monolingual English children were recruited. Ten children from the bilingual group were not included in this study for the following reasons: five refused to be tested (no data), two had too few trials, two had corrupt data, and one had missing data files. Of the remaining 49 bilingually exposed children, 30 were male and 19 were female. Seventy-eight data sessions were collected with 16 (nine males, seven females) of these bilingually-exposed children being tested two to five times at different ages in a longitudinal design. Ten children from the monolingual English group were not included for the following reasons: three were tested in a different paradigm, three refused to be tested, two showed indications of language impairment, and two had missing or corrupt data files. Of the remaining 49 monolingual children, 26 were male and 23 were female. Ninety data sessions were collected with 20 (10 males, 10 females) of the monolingually-exposed children being tested two to six times at different ages in a longitudinal design.

A total of 168 data sets were collected from the 98 children (Bilingual: 47 male and 31 female sessions; Monolingual: 48 male and 42 female sessions; Age: 3–5, 6–12, 13–24, and 25–47 months old). Table 1 presents a summary of the number of ERP sessions in terms of participant age, sex, and language for the cross-sectional and longitudinal data combined. There was no session number difference in terms of the language or sex factor (Chi Square test: X² = 0.558, p = 0.46). An additional eleven bilingual sessions and seventeen monolingual sessions are included here compared to our previous report (Shafer, et al., 2011). In addition, the previous report did not include the standardized language measures, or the subtypes of bilingual exposure (more English versus less English), and it examined maturation in a cross-sectional design. Shafer et al. (2012) focused exclusively on a comparison of the bilingual and monolingual 6–8 month-old infants.

Language input was estimated from a language background questionnaire (LBQ) filled out by the principal caretaker, using a seven-point scale for rating input across multiple persons (e.g., father, mother, babysitter, siblings) and situations (e.g., home, playground), with 3 indicating all Spanish (labeled on the questionnaire as “Spanish all the time”), −3 indicating all English (labeled on the questionnaire as “English all the time”), and 0 indicating equal Spanish and English input (labeled on the questionnaire as “equal English and Spanish”). Note that on the questionnaire the values 1 to 7 were used and were then transformed to a scale of −3 to 3 for ease of interpretation. None of the bilingual children had a language background score of 3 (indicating all Spanish exposure), or −3 (indicating all English exposure). Twenty-eight out of 49 children had relatively-balanced English-Spanish exposure with an LBQ score between the values of −1 and 1. In one descriptive analysis and one mixed model regression analysis, we divided the bilingual children into two subcategories based upon LBQ scores (see Table 2). A total of 20 (8 females) children were categorized as Spanish-dominant with an LBQ score of 0 to 2.9 and a mean of 0.62 ((SD= 0.80), and the remainder (N=29, 11 females) were categorized as English-dominant with an LBQ score between −0.1 and −2.9, mean −1.47 (SD = 0.76). The mean LBQ score for all the bilingual children was −0.62 (SD = 1.3).

Language measures were obtained using the Preschool Language Scale-4 (PLS-4) in English and, for bilinguals, in both English and Spanish. Three-year-old children also received the English Peabody Picture Vocabulary Test-3 (PPVT-3) (Dunn & Dunn, 1997) and, for bilinguals, the Spanish Test de Vocabulario en Imagenes Peabody (TVIP) (Dunn and Dunn, 1986). Table 3 shows standard language scores on the PLS-4 (Zimmerman et al., 2002) by language group (monolingual and bilingual, Spanish dominant versus English dominant bilinguals). All children had language measures on the English tests; Spanish measures were available for 16 of the 20 Spanish-dominant children and for 19 of the 29 English-dominant bilingual children. Note that for children under 12 months of age, the test items on the PLS-4 reflected general communication development (until first words are reported), and thus we did not collect measures for both the English and Spanish versions of the PLS at this age. The only difference in these two tests on early development is whether the question to the caretaker is asked in English or Spanish. Of the bilingual children lacking Spanish scores, only four of these children were older than nine months of age. In these cases, absence of Spanish scores was due to failure to complete the Spanish testing due to fatigue.

In the main statistical analyses, we used language as a categorical measure (monolingual versus bilingual; monolingual, balanced bilingual versus Spanish-dominant bilingual) because 1) the amount of English exposure does not follow a linear distribution since all monolingual children have 100% of English; 2) the monolingual and two bilingual groups of children did not significantly differ on the English language tests for any pairwise comparison between monolinguals and bilinguals in the longitudinal and cross-sectional groups, or between English- versus Spanish-dominant groups (t-tests, p-values > 0.20). The only differences were between English-dominant versus Spanish-dominant groups on the Spanish language tests. Unsurprisingly, bilingually-exposed children with relatively greater exposure to Spanish (0 to 2.9 on the LBQ) showed significantly higher receptive Spanish language standard scores than those with less Spanish exposure (−0.1 to – 2.9; p = 0.05 for receptive language, and p = 0.11 for expressive language).

We used age as a continuous variable for the main statistical analyses. We also used age as a categorical variable in the t-tests, figures and some tables for easy summary of participant characteristics. In some analyses, we divided children into the following age groups: 3–5 months, 6–12 months, 13–24 months, and 25–47 months. Our rationale for these four groups was derived from literature on early brain development and behavioral speech discrimination development. Moore and Linthicum (2007) divide early brain development into five distinct stages: transitioning to perinatal (27 to 29 fetal weeks), perinatal (third trimester to sixth postnatal month), transition to childhood (six months to 12 months), early childhood (2 to 5 years of age) and later childhood (6 to 12 years of age). Up to 6 months of age, the auditory cortex is still quite immature. Behavioral studies indicate that by six months of age, monolingual infants already show language-specific modulation of vowel discrimination (Eilers et al, 1979; Kuhl et al., 1992). Between six month and 1 years of age, there is a marked reduction of axons in the marginal layer, and filament-filled axons become visible in the auditory radiations in the core of the temporal lobe and the deeper layers of the cortex. Behavioral studies indicate language specific discrimination effects of more fine-grained speech information (Werker & Tees, 1984; Polka & Werker, 1994) and differences between monolinguals and bilinguals in vowel perception emerge during this period (Bosch & Sebastián-Gallés, 2003). Thirteen-24 months is a time of rapid lexical growth, whereas from 24–48 months, children are expected to acquire the grammatical patterns of language (Werker & Yeung, 2005).

The Bayley Scales of Infant and Toddler Development (Bayley, 1993) was used as a measure of general cognitive development. A t-test showed no group differences (monolingual mean = 103, SD = 15; bilingual mean = 98, SD = 17, p = 0.14). All infants were full-term, and had a normal birth history with no history of cognitive, neurological, speech, language, or hearing deficits in immediate family members. All had passed a newborn hearing screening according to parental report, and most passed a Transient Evoked Otoacoustic Emissions (TOAE) hearing screening in the laboratory. We did not have records of TOAE for several children due to incomplete behavioral testing or participant lack of compliance. All infants in the analyses showed a clearly-defined P1 obligatory auditory evoked potential (AEP) at frontocentral sites, indicating encoding of the auditory sound (Sharma, et al., 2009). Socio-economic status (SES) information was collected for all children using the Hollingshead Four-Factor Index of Socioeconomic Status (Hollingshead, 1975). Most of the children came from families with a middle-class or higher SES, but the monolingual group had a higher average SES than the bilingual group (Monolingual: mean = 50.7, SD = 12.1; Bilingual: mean = 39.5, SD = 16.9, p < 0.001).

Stimuli

Two, 250-ms re-synthesized vowels [ε] as in “bet” and [I] as in “bit” that differed in F1 and F2 formant frequencies (F1, 650 Hz and 500 Hz, F2: 1980 Hz and 2160 Hz, respectively) were used (See Figure 1). These stimuli were from a continuum of nine stimuli (deviant [I]: step 3, standard [ε]: step 9 of the continuum) used in previous studies with four- to ten-year old children with and without language impairment (Datta, et al., 2010; Shafer et al., 2010; Rinker, et al., 2017). Previous studies in our laboratory using all nine stimuli (step 1 to 9) showed that the categorical perception crossover point was at step 6 for both adults and children; thus, step 3 and step 9 were equal step distances from the crossover point. Moreover, step 3 was consistently identified as [I], and step 9 as [ε] by both adults and children as reported in our previous studies (Shafer, et al., 2007; Datta et al., 2010). In addition, other studies indicated poor categorization and neural responses MMN to these vowels by late adult learners of English with Spanish as a first language (Hisagi et al., 2014), and poor categorization of these vowels in children with specific language impairment (e.g., Datta, et al., 2010). The stimuli were presented at 86.5 dB SPL in sound field over two speakers, located 1 m in front of and 1 m slightly above and behind the child’s head.

Electrophysiological Design and Procedures

The electroencephalogram (EEG) was recorded from a 63-site Geodesic net using Net Station amplifiers and Net Station 4.1 software at a 250 Hz sampling rate and 0.1 to 30 Hz bandpass. The EEG was timelocked to the onset of the vowel stimuli. The stimuli were delivered via Eprime software (Psychology Software Tools, Pittsburgh, PA, United States) in sequences of 10 stimuli with an ISI of 400 ms between stimuli in the sequence, and 1500 ms between each sequence of 10. Three train types were randomly presented (see Table 4). The deviant stimulus occurred in the 4^th and 8^th position for 100 trains, in the 5^th and 10^th position for 50 trains, and in the 6^th and 10^th position for 50 trains, for a total of 1600 standards [ε] (80%) and 400 deviants [I] (20%). This is the same paradigm that was used in Shafer, et al. (2010; 2011 & 2012). We did not reverse the standard and deviant because the study time would be too long and our goal was to compare group effects.

The procedures were explained to the caretakers, who then read and signed informed consent. The electrode net was first soaked in a saline solution and then excess solution was removed. The nets were placed on the participant’s scalp while the participant was entertained by a laboratory assistant. The impedances of electrodes were maintained at or below 50 kOhms, which is sufficient for high input impedance amplifiers (200 Megaohms). During the experiment, older children sat in a chair, whereas infants sat on the caretaker’s lap while watching a video (e.g., Elmo, Baby Einstein) with the sound muted. Caretakers wore headphones playing music so that they would not orient to the vowel changes. Up to 200 sequences of ten stimuli were delivered (1600 standards, 400 deviants) in an approximately 40-minute session (28 minutes of stimulus delivery plus preparation time). Sessions were ended if a child became fussy and could not be calmed by changing the distracting task (e.g., by a researcher blowing bubbles or showing silent toys).

The continuous EEG was re-filtered offline using a low pass filter of 20 Hz, and was segmented with a pre-stimulus duration of 200 ms and 800 ms post-stimulus onset and baseline corrected using the pre-stimulus 100 ms amplitude. Any epochs with electrical activity exceeding +/− 140 μV at any electrode site were rejected, and bad channels (on 20% of segmented trials) were replaced by spline interpolation algorithm from surrounding channels. Over 90% of the children had more than 100 trials for the deviant stimulus after artifact rejection. The mean number of trials for the standard and deviant ERPs after artifact reject did not differ significantly between the monolingual and bilingual groups although there was a trend towards fewer trials obtained for the bilingual groups (p-values > .07, Standards: Monolingual Mean = 601, SD = 217, Bilingual Mean = 540, SD = 224; Deviants: Monolingual Mean = 241, SD = 88, Bilingual Mean = 216, SD = 91). Segments (−200 to 800 ms) were averaged for type (standard, deviant).

Analysis

Results

Discussion

This study examined the modulation of MMRs to English vowels in three-month to 47-month-old children by the factors of age, sex, hemisphere, and language experience. The current analysis showed that age, sex, and language experience influenced the amplitude of MMR in the early time window of 160–360 ms. In contrast, age was the only factor that influenced the amplitude of the late MMR in the time window of 400–600 ms. The new findings that we contribute to the current literature include: 1) Relative amount of bilingual exposure in English versus Spanish modulated the amplitude of the early MMR, but not the late MMR. Specifically, bilingual children with relatively-balanced English and Spanish exposure or more English than Spanish had similar MMR amplitude to their monolingual peers in the early time window (i.e., eMMR). However, bilingual children with Spanish-dominant exposure showed a difference in MMR amplitude from English monolinguals in the early interval. 2) Both the early and late MMRs to the vowel contrast shifted gradually from robust positivity to negativity from infancy as age increased. 3) No language exposure or sex effects were observed for the amplitude of the MMR in the late time window between 400 and 600 ms (i.e., late MMR). Below, we discuss our findings in greater detail in relation to the general literature.

Conclusion

Our findings demonstrate that neural indices of speech processing during the first few years of life undergo considerable development. Both maturational factors (age, sex) and experiential factors (language experience) modulate the infant/child MMRs. These developmental changes cannot be inferred from behavior. The considerable changes in neural discrimination (e.g., from a positive to a negative MMR dominating the response) provide clues indicating that the mechanisms for encoding and discriminating auditory information differ greatly from those used by older children and adults. Specifically, we found that immaturity in neural discrimination was indexed by a positive MMR prior to 350 ms to the English [ε] versus [I] speech contrast. Increased maturity, in terms of age, was indexed by increased negativity of the MMR in this early time frame. In addition, female infants showed increased negativity of the MMR, consistent with more rapid maturation than male infants. The finding that language experience modulated the MMR in the early time frame, provides further evidence for the importance of early speech experience in auditory-speech development. We hypothesize that this increasing negativity reflects increasing automaticity of discriminating the target contrast. We also suggested that the different pattern found for bilinguals could be the result of increased attention to the speech signal. Future studies need to further explore language-specific experience using other language pairs to fully understand the development of neural measures of speech processing.