Effects of sex and proficiency in second language processing as revealed by a large‐scale fNIRS study of school‐aged children

Abstract

Previous neuroimaging studies in adults have revealed that first and second languages (L1/L2) share similar neural substrates, and that proficiency is a major determinant of the neural organization of L2 in the lexical‐semantic and syntactic domains. However, little is known about neural substrates of children in the phonological domain, or about sex differences. Here, we conducted a large‐scale study (n = 484) of school‐aged children using functional near‐infrared spectroscopy and a word repetition task, which requires a great extent of phonological processing. We investigated cortical activation during word processing, emphasizing sex differences, to clarify similarities and differences between L1 and L2, and proficiency‐related differences during early L2 learning. L1 and L2 shared similar neural substrates with decreased activation in L2 compared to L1 in the posterior superior/middle temporal and angular/supramarginal gyri for both sexes. Significant sex differences were found in cortical activation within language areas during high‐frequency word but not during low‐frequency word processing. During high‐frequency word processing, widely distributed areas including the angular/supramarginal gyri were activated in boys, while more restricted areas, excluding the angular/supramarginal gyri were activated in girls. Significant sex differences were also found in L2 proficiency‐related activation: activation significantly increased with proficiency in boys, whereas no proficiency‐related differences were found in girls. Importantly, cortical sex differences emerged with proficiency. Based on previous research, the present results indicate that sex differences are acquired or enlarged during language development through different cognitive strategies between sexes, possibly reflecting their different memory functions. Hum Brain Mapp 36:3890–3911, 2015. © 2015 Wiley Periodicals, Inc.

INTRODUCTION

Learning a second language (L2) or a foreign language is of growing importance in globalizing economies and societies. Research on L2 and the brain has been of considerable interest to scientists as well as to L2 educators and learners. Over the past two decades, a large body of neuroimaging studies has opened new perspectives to understanding the brain‐language relationship and has shed new light on the neural basis of L2 processing [Abutalebi, 2008; Kotz, 2009; Perani and Abutalebi, 2005; Rüschemeyer et al., 2006; Suh et al., 2007]. A basic issue in the neuroscience of L2 is whether an L2 is processed through the same neural mechanism underlying native or first language (L1) acquisition or processing, and a more practical and crucial issue is to clarify the factors that affect the neural organization of an L2 and its relationship to behavioral performance. The available neuroimaging data support that, in principal, similar brain areas are recruited in L2 and L1 [e.g., Rüschemeyer et al., 2006; Suh et al., 2007; for reviews, see Abutalebi, 2008 and Kotz, 2009]. They also reveal that the extent of activation, peak activation latency, and/or detailed regions included vary as a function of proficiency [Golestani et al., 2006; Tatsuno and Sakai, 2005; for review, see Kotz, 2009]. These differences in activation are prominent in the initial stages of L2 acquisition and/or when L2 is processed with a non‐native‐like proficiency [for reviews, see Abutalebi, 2008 and Kotz, 2009].

A previous functional magnetic resonance imaging (fMRI) study showed that both proficiency and age of acquisition affect the neural substrates of L2 processing with a differential effect on syntax and semantics [Wartenburger et al., 2003]. It has been widely believed that the age of L2 acquisition plays no major role in the lexico‐semantic domain but rather that L2 proficiency is the main determinant [Indefrey, 2006; Perani and Abutalebi, 2005]. An L2 acquired late in life can be semantically processed through the same brain areas that process L1. Late learners with native‐like L2 proficiency activate the same areas for both languages [for review, see Perani and Abutalebi, 2005]. Conversely, age of acquisition has long been considered to have a major role in the neural organization of L2 in the syntactic domain [Golestani et al., 2006; Rüschemeyer et al., 2005; Wartenburger et al., 2003]. A recent review of L2 syntactic processing, however, suggested that proficiency is a major factor influencing the peak and extent of brain activation as a function of learning, implying that age is not as critical as it was thought to be [Kotz, 2009]. Given these findings, in both syntactic and lexico‐semantic domains, data to date reveal that L1 and L2 share similar neural structures, and proficiency is the major determinant of the neural organization of L2 processing.

With respect to the phonological domain, the neural organization of L2 relative to L1 has not been thoroughly investigated. A study of monolingual English‐speaking adults learning Hindi suggested that the successful learning of a nonnative phonetic contrast results in the recruitment of the same areas that are involved during the processing of native contrasts, based on the data of only 10 adults [Golestani and Zatorre, 2004]. In the initial stage of language acquisition, the phonological domain is particularly important, as one must have exquisite phonological abilities to allow segmentation of the speech signal into words, and to extract the meaning of words to understand sentences. Recent models of early language acquisition suggest that phonological awareness and word learning bidirectionally influence one another [Kuhl et al., 2008; Werker and Curtin, 2005]. Kuhl et al. have stated that infants with better phonetic learning skills advance more quickly toward language because phonetic skills assist phonotactic pattern and word learning [Kuhl et al., 2005, 2008]. Furthermore, a number of studies have shown that phonological awareness (or sensitivity) is strongly predictive of reading and spelling acquisition across languages [for review, see Ziegler and Goswami, 2005]. In short, despite its importance, data for neural representation of phonological processing in L2 learner children and its relation to learning is sparse [for studies of infants' responses to native and nonnative phonemes, see Kuhl, 2004, 2010 and Kuhl and Rivera‐Gaziola, 2008].

Another intriguing aspect of L2 research is whether sex differences exist in the behavioral performance and the functional organization of the brain in early L2 acquisition. Previous behavioral studies on L1 acquisition have reported that girls begin talking earlier [Murray et al., 1990], and acquire vocabulary faster than boys [Roulstone et al., 2002]. Faster language development in girls is consistently found [Bauer et al., 2002; Doran, 1907; Huttenlocher et al., 1991; Nelson, 1973]. These results led us to consider that sex differences might appear in the initial stages of L2 acquisition when the learners have yet to achieve high‐level proficiency with native‐like neural responses.

Moreover, a large number of neuroimaging investigations on sex differences have indicated greater bilateral cortical activation during language tasks in women compared to men [Baxter et al., 2003; Clements et al., 2006; Kansaku et al., 2000; Phillips et al., 2001; Pugh et al., 1996; Shaywitz et al., 1995], although controversy remains [Frost et al., 1999; Sommer et al., 2004, 2008; Weiss et al., 2003]. The existence or nonexistence of sex differences is likely to be task‐dependent. Previous studies observed sex differences in passive listening tasks (listening to stories) [Kansaku et al., 2000; Phillips et al., 2001] and phonological tasks [Clements et al., 2006; Pugh et al., 1996; Shaywitz et al., 1995], whereas others reported no sex differences in language comprehension tasks and verbal fluency tasks [Frost et al., 1999; Wallentin, 2009; Weiss et al., 2003]. Be that as it may, observation of sex differences both in behavioral and neuroimaging studies may reflect different cognitive strategies between the sexes in language processing.

Although sex differences in children have been found in L1 in a number of behavioral studies [Bauer et al., 2002; Doran, 1907; Huttenlocher et al., 1991; Nelson, 1973; Roulstone et al., 2002], and neuroimaging studies [Burman et al., 2008, 2013; Plante et al., 2006], such differences have not been examined in young children just beginning to learn a second language. Given that a number of neuroimaging studies in adults have shown sex differences in phonological tasks, it is possible that sex differences in L2 phonological processing exist in young children in the initial stages of L2 acquisition.

Hence, a large‐scale functional near‐infrared spectroscopy (fNIRS) study was conducted to examine the relationship between cortical activation and L2 proficiency during word processing in school‐aged children (6–10 years, n = 484). Prior to the current study, using fNIRS we examined the children's cortical activation in the language network for a word repetition task in L1 and L2 of linguistically distant languages: Japanese (L1) and English (L2) [Sugiura et al., 2011; for linguistically distant languages, see Chiswick and Miller, 2005]. We investigated the similarities and differences in cortical activation between L1 and L2 in children, and reported that children used largely overlapping neural substrates when processing words in both L1 and L2. We also found that the extent of activation differed between the languages and hemispheres. L1 words elicited significantly greater brain activation than L2 words, regardless of semantic knowledge, particularly in the superior/middle temporal and inferior parietal regions (angular/supramarginal gyri). The greater L1‐elicited activation in these regions suggests that they are phonological loci, reflecting processes tuned to the phonology of the native language. Real‐word repetition is thought to reflect a lexical level of representation (both phonological and semantic), but our prior study together with a previous study [Petersen et al., 1988] suggest that word repetition strongly reflects phonological aspects of word processing, although that does not mean that there is no semantic effect. Thus, we expect that word repetition tasks enable us to investigate cortical response, particularly in the phonological domain.

Here, we focused on L2 to examine the relationship between cortical activation and L2 proficiency. First, we present similarities and differences in cortical activation between L1 and L2 considering the sex effect, which has been implicated in L1 as aforementioned. Then, we move on to our main examination of the effect of proficiency in early L2 learning by using separate analyses for boys and girls, as the prior analyses indicated significant sex differences in L2 representation. Finally, we report sex differences in cortical activation associated with L2 proficiency during early L2 learning in children.

MATERIALS AND METHODS

Participants

Participants of this study initially included 484 normally developing Japanese elementary school children (236 boys and 248 girls, aged 6–10 years, M = 8.93, SD = 0.89). All participants completed a questionnaire before commencing this study. Each participant's parent gave written informed consent before his or her child's participation in this study. All of the procedures in this study were approved by the Human Subject Ethics Committee of Tokyo Metropolitan University. The Edinburgh Handedness Inventory [Oldfield, 1971] was used to determine hand dominance. Participants who failed to meet the criteria of the present study were excluded: The left‐handed (8) and ambidextrous (38) were excluded from the analyses, as those participants are more likely to have atypical language lateralization. Participants whose L1 was not Japanese (11) and participants with psychiatric disorders (12) were excluded from the analyses. A further 23 participants whose task performance or behavior did not meet our criteria (as described below) were also excluded from the current analyses. Consequently, 392 participants were used for the current analyses.

While most L2 studies have focused on the natural course or environment of L2 acquisition, the present study dealt with L2 learning in an environment where the target language is not typically spoken in daily life and is generally acquired through formal instruction in a classroom setting. We had participants of the same age with different levels of English proficiency as they had had different levels of exposure to L2. Some public schools provided English lessons (11–35 school hours/year) while others did not. Some children had been exposed to English through commercial language schools and/or home study. A few children who had at least one parent who was a native English speaker were excluded from the analyses. Our study also included some children who went to a private school with an immersion program, which are often associated with bilingual societies. As this Japanese private school is located in a monolingual city and is not an international school, these children were included.

In the present study, age of acquisition was not systematically manipulated and investigated. This is because Japan is a monolingual country in terms of language policy, and our participants had not been in an L2 environment with many opportunities for exposure to L2 input and interaction inside or outside of the home in their early childhood (this was also confirmed by our questionnaire survey).

Experimental Task and Data Acquisition

We used a noninvasive and unrestrictive fNIRS system (ETG‐4000, Hitachi Medical Co., Tokyo, Japan) to monitor cortical hemodynamic changes, as it has a major advantage in developmental studies with children, especially for large‐scale studies. fNIRS was installed in our original neuroimaging vehicle and was transported to seven different elementary schools in Japan. A 3 × 5 array of optodes consisting of eight laser diodes and seven light detectors, alternately placed at an inter‐optode distance of 3 cm to yield 22 channels, was applied on each side of the participant's head (Fig. 1A). The middle column of the 3 × 5 array was placed along the coronal reference curve (T3‐C3‐Cz‐C4‐T4) of the international 10/20 system [Jurcak et al., 2005, 2007], so that the lower edge of the array was placed directly above the ear. The highest sensitivity of hemodynamic changes in the lateral cortical region encompassing a pair of optodes is expected to be localized at the midpoint between the optodes [Okada et al., 1997]. This point served as the location of a channel. Optical data from individual channels were collected at two different wavelengths (695 and 830 nm) and analyzed using the modified Beer‐Lambert Law for a highly scattering medium [Cope et al., 1988]. While fNIRS has potential as a noninvasive brain monitor in a wide range of research fields, note that it has limited penetration depth (about 1–2 cm) due to the high level of light scattering within the tissue, thus allowing for sensitivity only to superficial cortical areas. Changes in oxygenated (oxy‐Hb), deoxygenated (deoxy‐Hb), and total hemoglobin (total‐Hb) signals were calculated in units of millimolar‐millimeter [Maki et al., 1995]. Optical signals were sampled at a rate of 10 Hz.

Figure 1.

Functional near‐infrared spectroscopy (fNIRS) measurements. (A) Close‐up view of the fNIRS equipment. fNIRS data were obtained using a 44‐channel spectrometer (Hitachi ETG‐4000). A 3 × 5 array of eight laser diodes and seven light detectors was applied, resulting in 22 channels on each side of the participant's head. (B) Cortical projection points of fNIRS measurements (location of 22 channels) and the six defined ROIs for language processing are mapped onto the MNI standard brain coordinate system by spatial registration. The locations of the 22 channels and six ROIs on the left and right hemispheres are symmetrical. The six defined ROIs: (a) the primary and auditory association cortices (PAAC) consisting of BAs (BA 41, 42) with channel 12, (b) the vicinity of Wernicke's area, the posterior part of the superior/middle temporal gyri (SMTG) (BA 21, 22) with channels 16, 17, and 21, (c) the angular gyrus (AG) (BA 39) with channels 4, 9, and 13, (d) the supramarginal gyrus (SMG) (BA 40) with channels 3 and 8, (e) the pars opercularis (POP), part of Broca's area, (BA 44) with channels 1 and 6, and (f) the pars triangularis (PTR), part of Broca's area, (BA 45) with channels 5, 10, and 14.

We measured the children's cortical hemodynamic changes while performing an oral repetition task in participants' L1 (Japanese) and L2 (English). As statistically significant differences in activation between factors appeared to be more robust in the high‐frequency word condition than in the low‐frequency word condition, we used high‐frequency words, which are defined as words that have more than 50 occurrences per million, for our main analysis. We used 60 single words: 30 words each for L1 and L2. For the supplementary analysis, we also used low‐frequency words, which are defined as words that have less than five occurrences per million. Here, we used another 60 single words: 30 words each for L1 and L2. Different word sets were used for L1 and L2 for both high‐ and low‐frequency word conditions. All words used in this experiment were emotionally neutral, and taken from two corpora: one by Amano and Kondo [2000] for Japanese and the other by Kučera and Francis [1967] for English. A list of all the words used in this study is reported in our previously published study [Sugiura et al., 2011]. All Japanese words contained four morae (Japanese syllabic unit) and English words consisted of two syllables. The length of Japanese and English words was kept approximately equal (within +/−10% difference). Children were seated in a chair and given instructions to repeat the words presented from a loud speaker. They were asked to overtly repeat each word once shortly after they heard it during the task period. The children heard the stimuli through the loud speaker at a comfortable volume (around 65 dB SPL). Each condition included only one language, and consisted solely of low‐ or high‐frequency words. The order of L1 and L2 conditions was counterbalanced, and the stimuli within each condition were presented in blocks of 5 different words. A total of six blocks was presented for each condition. The blocks were presented in random order while stimuli in each block were kept in the same sequence. One block was 35 s: a 5‐s prestimulus period, 15‐s stimulus period, and 10‐s recovery period, followed by a 5‐s poststimulus period. Their oral repetitions were recorded while children performed the task.

As English had not been introduced as a mandatory academic subject at the elementary school level in Japan at the time the data were collected, and therefore, the children had different levels of exposure to L2 regardless of age, a meaning comprehension test of spoken English tailored for Japanese children (score range: 0–100 points) was prepared and conducted to determine their levels of English proficiency.1

Note that our previous study with the same group of participants revealed that L2 proficiency is highly correlated with L2 exposure, which was expressed as the total hours of exposure to English that each child had received up to the time of the experiment [Ojima et al., 2011]. Thus, L2 proficiency is assumed to be mainly governed by learning at the early stage of L2 acquisition in the current study.

Data Analyses

Functional NIRS data were preprocessed using the Platform for Optical Topography Analysis Tools (Adv. Res. Lab., Hitachi, Saitama, Japan), a plug‐in‐based analysis platform that runs on Matlab (The MathWorks, South Natick, MA). To remove components originating from slow fluctuations of cerebral blood flow and heartbeat noise, the Hb signals were bandpass‐filtered between 0.02 and 1 Hz. A participant's motion during fNIRS measurement is tolerated to a higher degree than in fMRI and PET [Hull et al., 2009; Ikegami and Taga, 2008; Watanabe et al., 1998], in which the head position must be strictly fixed and vocalization may induce severe motion artifacts. Given that elicited imitation is necessarily accompanied by articulation and small motions of the participant's head, this advantage makes fNIRS a primary candidate for the language task used in the current study. Nevertheless, repetition causes some motion artifacts locally around the temple. Thus, an assessment of motion artifacts was performed by detecting rapid changes in total‐Hb signals (signal variations > 0.1 mmol·mm over two consecutive samples); all blocks that had been affected by movement artifacts were subsequently identified and removed. Following this elimination process, participant data that contained a minimum of three out of six data blocks for each condition were used. In addition, by visual inspection, we discarded an entire condition when there was insufficient optical signal (i.e., when the peak signal of oxy‐Hb during the task period was lower than approximately 0.01 mmol·mm as determined with reference to the standard deviation of the rest period) due to obstruction by hair or for other reasons. We utilized the channels that had more than a 60% survival rate of data across all subjects after the motion check. As two channels (15 and 20) did not reach the criterion due to movement in the temporal muscles, they were not used for further analyses.

For each individual set of hemoglobin data, we extracted data blocks from time course data. As described above, each data block consisted of 5 s prior to stimulus onset, 15 s of stimulus, 10 s of recovery, and a 5 s poststimulus period. As the hemodynamic response gradually fades back toward the baseline after the stimulus period, and returns to near baseline levels after several minutes, a recovery period of 10 s was set for the full recovery of hemodynamic response prior to the poststimulus period. For each channel in nonrejected blocks, a first‐degree baseline fit to the mean of the 5 s prestimulus period and 5 s of the poststimulus period was performed. The results of only the oxy‐Hb analyses are included here because the oxy‐Hb signal is reported to be more sensitive to changes in cerebral blood flow than are deoxy‐Hb and total‐Hb signals [Hoshi, 2003; Hoshi et al., 2001], to have a higher signal‐to‐noise ratio [Strangman et al., 2002], and also to have a higher retest reliability [Plichta et al., 2006]. For each child, the mean change in concentration of oxy‐Hb over 25 s after the onset of stimulus was calculated for each condition and for each channel. We used a virtual registration method [Tsuzuki et al., 2007] to register NIRS data to the Montreal Neurological Institute (MNI) standard brain space. A detailed description of the application of this method was reported previously [Sugiura et al., 2011].

For recorded oral data, whether the words were correctly repeated or not during the NIRS measurement was evaluated phoneme by phoneme by a native Japanese and bilingual (Japanese and English) speaker. The participants with a repetition success rate of less than 70% were excluded from the analyses. A total of 392 children (boys/girls: 190/202) who satisfied all entry criteria were used for analyses.

Statistical Analyses

Statistical analyses were carried out using the SPSS statistical package (SPSS, Chicago, IL) unless otherwise indicated. We defined six regions of interest (ROIs) bilaterally (Fig. 1B) according to the results of spatial registration and referring to a standard macroanatomical atlas (Automatic Anatomical Label) [Tzourio‐Mazoyer et al., 2002] as follows: (a) the primary and auditory association cortices (PAAC) consisting of Brodmann areas (BA 41, 42) with channel 12; (b) the vicinity of Wernicke's area, the posterior part of the superior/middle temporal gyri (SMTG) (BA 21, 22) with channels 16, 17, and 21; (c) the angular gyrus (AG) (BA 39) with channels 4, 9, and 13; (d) the supramarginal gyrus (SMG) (BA 40) with channels 3 and 8; (e) the pars opercularis (POP), part of Broca's area (BA 44), with channels 1 and 6; and (f) the pars triangularis (PTR), part of Broca's area (BA 45), with channels 5, 10, and 14. The overall oxy‐Hb signal level in a single ROI was obtained by calculating the mean oxy‐Hb signal level of all the channels within the ROI [Sugiura et al., 2011]. The ROI‐wise statistical results were corrected for multiple comparisons using false discovery rate (FDR) control [Singh and Dan, 2006].

We first examined cortical activation, emphasizing sex differences, to clarify similarities and differences in cortical activation between L1 and L2. Statistical analyses using a 3‐way repeated‐measures ANOVA were conducted for 6 ROIs to evaluate the effects of sex (boys and girls), language (Japanese: L1 and English: L2), and hemisphere (left hemisphere: LH and right hemisphere: RH).

Then, to investigate the effect of L2 proficiency on cortical activation, we conducted separate analyses for boys and girls as we had found significant sex differences in cortical activation in L2 in the prior analysis. As stated above, we used an English test to estimate children's L2 proficiency. Figure 2 shows a histogram of the English test scores. It appeared that the distribution of the test scores was bimodal rather than unimodal, suggesting two populations segregated with an obvious discontinuity around the score 50. The bimodality coefficient (b) was computed using the MODECLUS procedure of SAS (SAS Institute, Cary, NC) using the following formula,

$$b=\frac{g^2+1}{k+\frac{3{\left(n-1\right)}^2}{\left(n-2\right)\left(n-3\right)}}$$

where g is skewness and k is kurtosis. As the bimodality coefficient was 0.713, above the criterion value of 0.555 for multimodal distribution, the current distribution was considered highly bimodal. Thus, we conducted group analyses by categorizing participants who scored below 50 points in the English test as low proficiency (Low) and those who scored above 50 points as high proficiency (High). The means +/− SD of the English test scores were 41.0 +/− 5.3 for Low and 73.0 +/− 15.4 for High. As the distribution of scores for both sexes with low‐ and high‐English proficiency was similar, the total population was consequently divided into similar numbers of both sexes in each proficiency group (Low: boys = 49.4%, girls = 50.6%; High: boys = 46.6%, girls = 53.4%). There were no statistically significant differences in school grade and age between both sexes in both low and high English proficiency groups.

Figure 2.

Histogram of English test scores. As the children had different levels of exposure to L2 regardless of age, an English test was administered to estimate their L2 proficiency. The scores of the test (range: 0–100 points) were used as the criterion for selecting two proficiency groups. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

To further examine whether cortical activation during L2 processing could be predicted based on age while simultaneously taking into account whether collinearity between L2 proficiency (English test scores) and age exists or not, we used stepwise multiple regressions. We analyzed to what extent cortical activation within six ROIs could be predicted as a function of age, while assessing collinearity among the predictor variables (age and English test scores), only when the age effect was significant. Collinearity among the predictor variables was assessed utilizing tolerances, as well as variance inflation factors (VIF), and we checked whether the tolerances of the predictors were higher than the evaluation criteria of 0.01, and VIF values were not higher than the evaluation criteria of 10 [Stevens, 2002]. Participants' age was calculated by dividing the number of days between birth and fNIRS data collection by 365 days, because there can be a difference of almost a full year between the oldest and youngest participants within the same age group. FDR control was adopted for simultaneous comparisons.

RESULTS

Similarities and Differences in Cortical Activation Between L1 and L2

The effects of sex (boys and girls), language (L1 and L2), and hemisphere (LH and RH) were evaluated using 3‐way repeated‐measures ANOVAs for six ROIs. The statistical results for the high‐frequency word condition are shown in Table 1, and a summary of the results is depicted in Figure 3 with bar graphs showing the relative changes in cortical activation for the six ROIs during L1 and L2 processing. The ANOVA revealed main effects of sex, language and/or hemisphere in some ROIs, which will be described below in detail, but no significant interactions between the factors tested.

Table 1.

Results of 3‐way repeated‐measures ANOVA (high‐frequency words)

Brain area	Source of variation	d.f.	F	P corrected	Remarks
PAAC	Sex	1,361	0.383	n.s.
	Language	1,361	3.034	n.s.
	Hemisphere	1,361	0.123	n.s.
	Language × sex	1,361	1.763	n.s.
	Hemisphere × sex	1,361	1.818	n.s.
	Language × hemisphere	1,361	0.011	n.s.
	Language × hemisphere × sex	1,361	0.012	n.s.
SMTG	Sex	1,376	4.519	< 0.1+	girls > boys
	Language	1,376	7.629	< 0.05*	L1 > L2
	Hemisphere	1,376	1.985	n.s.
	Language × sex	1,376	0.578	n.s.
	Hemisphere × sex	1,376	0.595	n.s.
	Language × hemisphere	1,376	0.163	n.s.
	Language × hemisphere × sex	1,376	0.01	n.s.
AG	Sex	1,383	9.394	< 0.05*	boys > girls
	Language	1,383	19.346	< 0.001***	L1 > L2
	Hemisphere	1,383	15.172	< 0.001***	LH > RH
	Language × sex	1,383	1.321	n.s.
	Hemisphere × sex	1,383	0.717	n.s.
	Language × hemisphere	1,383	0.081	n.s.
	Language × hemisphere × sex	1,383	0.884	n.s.
SMG	Sex	1,379	9.048	< 0.01**	boys > girls
	Language	1,379	28.688	< 0.001***	L1 > L2
	Hemisphere	1,379	0.513	n.s.
	Language × sex	1,379	2.676	n.s.
	Hemisphere × sex	1,379	0.247	n.s.
	Language × hemisphere	1,379	0.006	n.s.
	Language × hemisphere × sex	1,379	0.893	n.s.
POP	Sex	1,381	1.413	n.s.
	Language	1,381	0.079	n.s.
	Hemisphere	1,381	19.863	< 0.001***	RH > LH
	Language × sex	1,381	2.834	n.s.
	Hemisphere × sex	1,381	0.261	n.s.
	Language × hemisphere	1,381	2.951	n.s.
	Language × hemisphere x sex	1,381	0.172	n.s.
PTR	Sex	1,379	0.088	n.s.
	Language	1,379	0.496	n.s.
	Hemisphere	1,379	43.044	<0.001***	RH > LH
	Language × sex	1,379	2.586	n.s.
	Hemisphere × sex	1,379	0.765	n.s.
	Language × hemisphere	1,379	4.28	n.s.
	Language × hemisphere × sex	1,379	0.118	n.s.

Note: Statistical analyses using 3‐way repeated‐measures ANOVA were conducted for 6 ROIs to evaluate the effects of two within‐subject factors: language (Japanese: L1 and English: L2) and hemisphere (left: LH and right: RH), and an inter‐subject factor: sex (boys and girls). P values were FDR corrected for 6 tests with a significance level of P < 0.05 after correction for multiple testing. Asterisks and plus sign indicate statistically significant results (⁺ P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001). d.f.: degree of freedom, PAAC: primary and auditory association cortices, SMTG: superior/middle temporal gyri, AG: angular gyrus, SMG: supramarginal gyrus, POP: pars opercularis, part of Broca's area, PTR: pars triangularis, part of Broca's area, and n.s.: not significant.

Figure 3.

Summary of 3‐way repeated‐measures ANOVA results. Cortical activation during high‐frequency L1 and L2 word processing were examined. The effects of two within‐subject factors, language (Japanese: L1 and English: L2) and hemisphere (left: LH and right: RH), and an intersubject factor, sex (boys and girls), were evaluated for six ROIs. The bar graphs show the relative changes in [oxy‐Hb] in units of millimolar·millimeter (mmol·mm), and error bars indicate standard error. Blue bars show boys and red bars show girls. The ANOVA results are shown in the colored boxes. P values were FDR corrected for six tests with a significance level of P < 0.05 after correction for multiple testing. Asterisks and plus sign indicate statistically significant results (⁺ P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001, corrected). PAAC: primary and auditory association cortices, SMTG: superior/middle temporal gyri, AG: angular gyrus, SMG: supramarginal gyrus, POP: pars opercularis, part of Broca's area, PTR: pars triangularis, part of Broca's area, and n.s.: not significant.

Sex effects (differences between sexes)

A significant main effect of sex was found in the SMG (P < 0.01, boys > girls) and AG (P < 0.05, boys > girls), and a trend toward a main effect of sex was found in the posterior part of the SMTG (Wernicke's area) (P < 0.1, girls > boys), as shown in Table 1. Specifically, girls exhibited a trend toward greater activation in Wernicke's area than did boys, while, in contrast, boys had significantly greater activation in the AG and SMG than did girls, irrespective of language. Interestingly, the AG and SMG were moderately activated in boys for L2 compared to L1, whereas these regions were not activated at all in girls for L2. Although there were no interactions between sex and other factors, direct comparisons of cortical activation between sexes for each ROI using unpaired, two‐tailed t‐tests revealed that the significant sex differences were most clearly demonstrable for L2 (Table 2).

Table 2.

Direct statistical comparisons of cortical activation between sexes for each ROI

Condition	Hemisphere	Brain area	t	d.f.	Mean diff	Std. err. diff	P corrected	Remarks
L1	LH	PAAC	0.580	373	0.003	0.006	n.s.
		SMTG	−2.085	383	−0.010	0.005	n.s.
		AG	2.388	388	0.012	0.005	n.s.
		SMG	1.584	384	0.009	0.005	n.s.
		POP	0.364	386	0.002	0.005	n.s.
		PTR	−0.404	388	−0.002	0.005	n.s.
	RH	PAAC	−0.581	384	−0.003	0.006	n.s.
		SMTG	−1.567	383	−0.007	0.004	n.s.
		AG	1.238	386	0.006	0.005	n.s.
		SMG	1.173	385	0.007	0.006	n.s.
		POP	0.180	387	0.001	0.005	n.s.
		PTR	−1.345	384	−0.006	0.004	n.s.
L2	LH	PAAC	2.019	379	0.012	0.006	n.s.
		SMTG	‐.764	386	−0.004	0.005	n.s.
		AG	3.164	390	0.017	0.005	<0.05*	Boys > girls
		SMG	3.573	388	0.019	0.005	<0.01**	Boys > girls
		POP	1.898	387	0.010	0.005	n.s.
		PTR	1.085	386	0.006	0.005	n.s.
	RH	PAAC	0.628	385	0.004	0.006	n.s.
		SMTG	−0.869	386	−0.004	0.004	n.s.
		AG	2.650	387	0.016	0.006	<0.05*	Boys > girls
		SMG	3.243	388	0.020	0.006	<0.01**	Boys > girls
		POP	1.621	389	0.009	0.005	n.s.
		PTR	0.326	387	0.001	0.004	n.s.

Note: Unpaired, two‐tailed t‐tests were used to compare means of cortical activation between sexes in each ROI. P values were FDR corrected for 12 tests with a significance level of P < 0.05 after correction for multiple testing. Asterisks indicate statistically significant results (*P < 0.05, **P < 0.01). d.f.: degree of freedom, PAAC: primary and auditory association cortices, SMTG: superior/middle temporal gyri, AG: angular gyrus, SMG: supramarginal gyrus, POP: pars opercularis, part of Broca's area, PTR: pars triangularis, part of Broca's area, LH: left hemisphere, RH: right hemisphere, and n.s.: not significant.

Cortical activation independent of sex (similarities between sexes)

Language effects

While no difference in cortical activation was observed between languages in the primary auditory area, the differences between L1 and L2 (L1 > L2) increased from SMTG to AG, and further to SMG in the posterior language areas: PAAC (F(1,361) = 3.034, n.s.) < SMTG (F(1,376) = 7.629, P < 0.006) < AG (F(1,383) = 19.346, P < 1.42E‐5) < SMG (F(1,379) = 28.688, P < 1.45E‐7; for comparison, the effect sizes of the differences, as measured by Cohen's d statistic are 0.087 (PAAC) < 0.148 (SMTG) < 0.259 (AG) < 0.314 (SMG)).

Hemisphere effects

While bilateral activation was observed in the auditory and the temporal regions, significant hemispheric asymmetry was observed in the AG (P < 0.001, LH > RH) in the posterior language area and in the POP and PTR (P < 0.001, RH > LH) in the inferior frontal regions.

Taken together, for the high‐frequency word condition, there were little or no differences in cortical activation between languages and hemispheres near the auditory area, whereas there were significant differences in cortical activation between languages (SMTG, AG, and SMG) and hemispheres (AG, POP, and PTR) in the posterior and anterior language areas, and these patterns were revealed as common to both sexes.

Supplementary analysis for low‐frequency word condition

The effects of sex (boys and girls), language (L1 and L2), and hemisphere (LH and RH) for the low‐frequency word condition were also analyzed using 3‐way repeated‐measures ANOVAs for six ROIs. The statistical results are shown in Table 3. Overall tendencies of differences in activation between the factors observed in the high‐frequency word condition were replicated except for the differences between some factors, which will be described below. However, overall statistical significance between the factors tended to be smaller in low‐frequency word than high‐frequency word conditions. As for the sex effect, uncorrected results showed statistically significant (AG: P = 0.05: boys > girls) or trending toward significant (SMG: P = 0.09: boys > girls) differences in activation between sexes in the low‐frequency word condition. However, the results turned out to be statistically insignificant after correction for multiple testing.

Table 3.

Results of 3‐way repeated‐measures ANOVA (low‐frequency words)

Brain area	Source of variation	d.f.	F	P corrected	Remarks
PAAC	Sex	1,356	0.770	n.s.
	Language	1,356	1.376	n.s.
	Hemisphere	1,356	0.023	n.s.
	Language × sex	1,356	0.189	n.s.
	Hemisphere × sex	1,356	1.193	n.s.
	Language × hemisphere	1,356	2.197	n.s.
	Language × hemisphere × sex	1,356	3.561	n.s.
SMTG	Sex	1,378	1.346	n.s.
	Language	1,378	4.018	< 0.1+	L1 > L2
	Hemisphere	1,378	0.211	n.s.
	Language × sex	1,378	0.285	n.s.
	Hemisphere × sex	1,378	2.522	n.s.
	Language × hemisphere	1,378	0.442	n.s.
	Language × hemisphere × sex	1,378	0.070	n.s.
AG	Sex	1,388	3.762	n.s.
	Language	1,388	7.400	< 0.05*	L1 > L2
	Hemisphere	1,388	1.348	n.s.
	Language × sex	1,388	0.016	n.s.
	Hemisphere × sex	1,388	0.151	n.s.
	Language × hemisphere	1,388	0.218	n.s.
	Language × hemisphere × sex	1,388	0.045	n.s.
SMG	Sex	1,383	2.872	n.s.
	Language	1,383	12.053	< 0.01**	L1 > L2
	Hemisphere	1,383	15.462	< 0.001***	RH > LH
	Language × sex	1,383	0.834	n.s.
	Hemisphere × sex	1,383	3.331	n.s.
	Language × hemisphere	1,383	5.680	n.s.
	Language × hemisphere × sex	1,383	3.520	n.s.
POP	Sex	1,385	2.166	n.s.
	Language	1,385	7.221	< 0.05*	L2 > L1
	Hemisphere	1,385	27.707	< 0.001***	RH > LH
	Language × sex	1,385	0.001	n.s.
	Hemisphere × sex	1,385	0.953	n.s.
	Language × hemisphere	1,385	0.004	n.s.
	Language × hemisphere × sex	1,385	0.131	n.s.
PTR	Sex	1,384	0.050	n.s.
	Language	1,384	1.491	n.s.
	Hemisphere	1,384	50.502	< 0.001***	RH > LH
	Language × sex	1,384	0.028	n.s.
	Hemisphere × sex	1,384	1.299	n.s.
	Language × hemisphere	1,384	2.082	n.s.
	Language × hemisphere × sex	1,384	0.003	n.s.

Note: Statistical analyses using 3‐way repeated‐measures ANOVA were conducted for six ROIs to evaluate the effects of two within‐subject factors: language (Japanese: L1 and English: L2) and hemisphere (left: LH and right: RH), and an intersubject factor: sex (boys and girls). P values were FDR corrected for 6 tests with a significance level of P < 0.05 after correction for multiple testing. Asterisks and plus sign indicate statistically significant results (⁺ P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001). d.f.: degree of freedom, PAAC: primary and auditory association cortices, SMTG: superior/middle temporal gyri, AG: angular gyrus, SMG: supramarginal gyrus, POP: pars opercularis, part of Broca's area, PTR: pars triangularis, part of Broca's area, and n.s.: not significant.

Different results between low‐ and high‐frequency word conditions were observed for some factors in some ROIs. (1) A significant hemisphere effect was observed in the AG (P < 0.001, LH > RH) but not in the SMG for the high‐frequency word condition; on the contrary, a significant hemisphere effect was observed in the SMG (P < 0.001, RH > LH) but not in the AG for the low‐frequency word condition. (2) While language effect was not observed in the high‐frequency word condition in the POP, its effect was revealed as significant (P < 0.05, L2 > L1) in the low‐frequency word condition.

Second Language Processing

Next, we focused on L2 processing to investigate whether, and if so how, cortical activation is associated with proficiency during early L2 learning in children. Here, we emphasized sex differences, as the prior analysis revealed sex differences in L2 processing.

Behavioral results

Behavioral data were analyzed to examine task performance (repetition success rates) in L2. L2 proficiency (Low and High) and sex (boys and girls) effects were investigated. A 2 × 2 ANOVA (L2 proficiency × sex) showed significant differences in the repetition success rates between low‐ and high‐proficiency groups (F (1, 388) = 195.088, P < 0.001, Low < High), but there was no significant difference in the rates between sexes, and no interaction between L2 proficiency and sex.

Proficiency effects on cortical activation during L2 processing

Differences in activation between the L2 low‐ and high‐proficiency groups (Low and High) for six ROIs in each hemisphere are shown in Figure 4A (boys) and 4B (girls). Statistical significances between Low and High in the bar graphs are indicated with asterisks. For reference, L1 data for the whole group (Low and High) are also shown. We used the whole group for L1 data as there were no statistical differences in activation between the L2 low and high‐proficiency groups in the L1 condition.

Figure 4.

Relationships between L2 proficiency and cortical activation during high‐frequency L2 word repetition for (A) boys and (B) girls. The results of ROI‐wise group analyses are shown in the bar graphs to delineate the differences in cortical activation between L2 low‐ and high‐proficiency groups (low and high; blue bars denote boys; red bars denote girls), and activation during L1 word repetition for the whole group (L2 low‐ & high‐proficient boys and girls; shown as gray bars for reference). The results of whole‐group analyses are shown for the L1 condition because there were no significant differences in activation during the L1 condition between L2 low‐ and high‐proficient groups. The y‐axes show the relative changes in [oxy‐Hb] in units of millimolar·millimeter (mmol·mm), and error bars indicate standard error. Asterisks and plus sign in the bar graphs indicate statistically significant results (⁺ P < 0.1, *P < 0.05, **P < 0.01; corrected). (Although there were no statistically significant results for girls, bar graphs are included for comparison with the boys' data and/or L1 data.) PAAC: primary and auditory association cortices, SMTG: superior/middle temporal gyri, AG: angular gyrus, SMG: supramarginal gyrus, POP: pars opercularis, part of Broca's area, and PTR: pars triangularis, part of Broca's area.

As shown in Figure 4, both boys and girls in the low‐proficiency group showed less activation during L2 compared to L1 processing across all regions examined; however, boys and girls in the high‐proficiency group exhibited completely different patterns from one another. In boys, significant differences in cortical activation between the low and high‐proficiency groups were revealed broadly and bilaterally in the language‐related cortical regions (Fig. 4A). In other words, cortical activation in the language‐related regions in boys significantly increased with L2 proficiency. Conversely, no such differences were observed between the low and high‐proficiency groups in girls (Fig. 4B). Significant differences in cortical activation between the low‐ and high‐proficiency groups in boys were not attributable to age differences, as no significant age differences were found between the two groups.

For statistical confirmation, language ROI‐wise sex differences in cortical activation in the low‐ and high‐proficiency groups were examined respectively. The results are shown in Figure 5. There were no significant differences in activation between sexes in any of the ROIs in the low‐proficiency group. In contrast, significant differences in activation were detected between sexes in the high‐proficiency group in the multiple language areas, markedly in the bilateral AG and SMG.

Figure 5.

ROI‐wise statistical sex differences in cortical activation during high‐frequency L2 word processing in the L2 low‐ and high‐proficiency groups. The y‐axes show the relative changes in [oxy‐Hb] in units of millimolar·millimeter (mmol·mm), and error bars indicate standard error. PAAC: primary and auditory association cortices, SMTG: superior/middle temporal gyri, AG: angular gyrus, SMG: supramarginal gyrus, POP: pars opercularis PTR: pars triangularis, and n.s.: not significant. Asterisks indicate statistically significant results (*P < 0.05, **P < 0.01; corrected).

Confirming age effect and its noncollinearity with L2 proficiency

As noted above, English had not been introduced as a mandatory academic subject at the elementary school level in Japan at the time the data were collected; therefore, our participants had different levels of English proficiency regardless of age. Nevertheless, to further examine possible effects of age while simultaneously considering collinearity between L2 proficiency (English test scores) and age, stepwise multiple regression analyses were used. The results revealed a significant negative relationship between age and cortical activation only in the left temporal region for both sexes: PAAC for boys (β = −0.016, P < 0.05) and SMTG for girls (β = −0.009, P < 0.05). As for the relationship between proficiency and cortical activation, significant (left AG, P < 0.01; right AG and SMG, bilateral SMTG, POP, and PTR, P < 0.05) or marginally significant (left SMG, bilateral PAAC, P < 0.10) positive relationships were found in all six ROIs for boys, but no significant relationships between them were found in any of the ROIs in girls. Although the discontinuity in the test score distribution was not taken into consideration in the regression analysis, the results of positive relationships between English test scores and cortical activation in boys and no relationships between them in girls were roughly consistent with the results of the prior group analyses regarding L2 proficiency. Multiple regression analyses demonstrated that age remained a significant predictor of cortical activation in the temporal region. At the same time, the analyses indicated that while English test scores and cortical activation were positively correlated in boys, age showed an inverse correlation with cortical activation in both sexes. Furthermore, as both age and L2 proficiency effects were significant in the left PAAC for boys, collinearity among the predictor variables was assessed. The tests gave satisfactory results with a tolerance of 0.97 (>0.01) and a VIF of 1.03 (<10), indicating that there was no collinearity between the test scores (L2 proficiency) and age [Stevens, 2002].

DISCUSSION

In the present study of school‐aged children, we first investigated, by taking particular note of sex differences, similarities and differences in cortical activation between L1 and L2. Second, we examined whether, and if so how, cortical activation varies with proficiency during early L2 learning in children, and whether sex differences exist in its variation. The results are summarized and the details discussed below.

Similarities and Differences in Cortical Activation Between L1 and L2

Consistent with previous findings in the lexico‐semantic and syntactic domains, the present study focusing on phonological domain in children clarified that, in principle, L1 and L2 share similar neural structures for both sexes even in the case of linguistically distant languages. Nevertheless, significant differences were observed in the levels of activation between languages and hemispheres, and more importantly, differential effects were observed between boys and girls during high‐frequency word processing, which suggest that the neural organization of language is complex and dependent on various factors.

Sex effects (differences between sexes)

Significant differences between sexes were found in cortical activation, and post hoc analyses suggested that the results were most clearly demonstrable for L2. Both boys and girls showed significantly less activation during L2 than L1 processing in the posterior language areas (SMTG, AG, and SMG), especially the parietal region (AG and SMG) (Fig. 3). However, significant sex differences were found in resource allocation. While boys moderately recruited the AG and SMG in L2, girls did not, although they definitely did during L1 processing (Fig. 3, AG and SMG). This comparison of L1 and L2 results in girls suggests that the absence of parietal activation in L2 for girls is not due to an anatomical difference of this system between sexes but due to functional differences between sexes during L2 development.

Another noteworthy sex difference was found in the posterior part of the SMTG, which is the vicinity of Wernicke's area, a region associated with auditory speech perception. More specifically, two meta‐analyses have pointed to a specific role for the posterior superior temporal gyrus in lexical phonological access [Indefrey and Levelt, 2000, 2004]. Contrary to the activation in the AG and SMG, girls exhibited a trend toward greater activation in this area than did boys, irrespective of language. These results suggest that while boys allocate processing resources in a widely distributed language network that includes the AG and SMG, girls allocate resources in more restricted regions, with little or no cognitive load in the AG and SMG, but with a greater cognitive load in the SMTG than boys during speech perception in the early stage of L2 learning. Vadlamudi et al. [2006] reported in their MRI study that the absolute and proportional planum temporale volumes (often identified with Wernicke's area) were not significantly associated with age or sex in children ages 4.2–15.7 years, which includes the age range of our participants. This finding suggests that the sex differences observed in the SMTG in the current study can be attributed to the functional differences of this region between sexes.

Our results revealed sex differences in the functional usage of language‐related cortical regions. The interpretation of the results will be discussed later.

Cortical activation independent of sex (similarities between sexes)

In both sexes, the differences in cortical activation between L1 and L2 were enhanced during word processing: PAAC < SMTG < AG < SMG (L1 > L2). The present results suggest that there is no difference in acoustic processing between L1 and L2 initially at the level of non‐domain‐specific processing in the auditory‐associated region (PAAC), but that differences emerge in the SMTG, and are likewise enhanced at subsequent stages of language processing in the inferior parietal cortex (AG and SMG), which exhibits higher‐level functional specializations. At the same time, while both sexes showed bilateral symmetric activation in the auditory and temporal regions, significant hemispheric asymmetry was observed in the posterior (AG: LH > RH) and anterior (POP and PTR: RH > LH) language areas.

Furthermore, as shown in Figure 3, a decrease in activation for L2 was observed in all areas that differed by language, which were limited to the posterior language areas (SMTG, AG, and SMG). These results are in agreement with those of previous studies in adults with an age range of about 18–36 [Dehaene et al., 1997; Perani et al., 1996, 1998; Rüschemeyer et al., 2005, 2006]. Although these comparable studies all involved semantic comprehension, they provided evidence of reduced activation in L2 relative to L1 in various tasks, especially when learners do not have an excellent command of L2 (low proficiency and/or late acquisition learners). Given these findings, the context of our participants (i.e., they were at the early stage of L2 learning), and the evidence of significantly less activation in boys in the low‐proficiency group compared to boys in the high‐proficiency group as shown in Figure 4A, the observed differences in L1 and L2 in the present study are presumed to arise from differences in proficiency, and reduced activation in L2 relative to L1 would be common to both children and adults in the initial stage of L2 learning.

Word repetition as a reflection of phonological processing

Repetition of words may involve access to phonological processing including phonological store, as well as to lexical‐semantic knowledge. Although it is unknown to what extent access to each of these occurs during word repetition, and although the design of the present study was not aimed at distinguishing between them, a previous study [Petersen et al., 1988] and our overall results suggest that semantic analysis of words is unlikely to be dominant compared to phonological analysis during word repetition: (1) L1 words elicited significantly greater brain activation than L2 words during word repetition, regardless of semantic knowledge, particularly in the superior/middle temporal and inferior parietal regions (angular/supramarginal gyri) [Sugiura et al., 2011]. Also, the magnitude of cortical activation in these ROIs was not associated with semantic knowledge, but with word repetition success rates, which reflects phonological processing [Sugiura et al., 2011]. (2) Our participants did not exhibit significant activation in the left prefrontal cortex (left PTR), an area that reflects semantic processing [Friederici et al., 2000], which is consistent with the results of a previous study that utilized a word repetition task [Petersen et al., 1988]. (3) Rather, activation in the right Broca homolog, which is associated with phonological processing [Poldrack et al., 1999], was more significant than that in the left Broca's area.

In the present study, cortical activation in the low‐frequency word condition, which is assumed to reflect only the phonology of words but little or no semantics (as participants rarely know the meaning of the words), was also examined. Overall trends of the results were similar to those of the high‐frequency word condition, but statistical significance was lower in the low‐frequency word than in the high‐frequency word condition regardless of the factors examined. Note that the lower statistical significance in the low‐frequency word condition than in the high‐frequency word condition does not necessarily reflect only the difference in semantic knowledge, but also reflects the difference in lexical familiarity (phonological word familiarity regardless of whether the meaning is known or not). The phonological processing of words consists of multiple levels, such as levels of phonemes, diphones, syllables, and/or words (i.e., word or lexical phonology). Thus, the differences in statistical significance between the low‐ and high‐frequency word conditions are presumed to be attributable to the presence or absence of lexical phonology (whole‐word phonology) in the mental lexicon. In other words, even if children do not access the lexical‐semantic knowledge in their mental lexicon during word repetition, they would most likely access the lexical phonology in their mental lexicon to retrieve the heard word and to check whether the phonology of the heard word matches with that of a word in their lexicon. As a consequence of this retrieval, differences in cortical response would appear depending on phonological word familiarity.

Right‐dominant activation was observed in the inferior frontal regions (Broca's area) during word processing among right‐handers irrespective of word frequency. Previous literature showing greater activation of the right Broca homolog for phonological than for semantic processing [Poldrack et al., 1999] suggests that the right‐dominant activation in the frontal region is due to phonological processing rather than semantic processing.

Broca's area in the left inferior frontal gyrus, particularly the POP [Dapretto and Bookheimer, 1999; Friederici, 2002], is known to be involved in cortical response related to working memory [Gabrieli et al., 1998]. Our repetition task undoubtedly incorporated aspects of working memory. However, inconsistencies arise if we attempt to explain the present results for the POP primarily with working memory function. We observed significant right‐hemispheric activation in the inferior frontal gyrus, but activation in this region during memory tasks generally appears to be lateralized to the left. Thus, although repetition of words is presumed to incorporate working memory to some extent, the observed cortical activation in the present study cannot be accounted for solely by working memory. Rather, the activation is more likely to reflect phonological processing considering its association with the right Broca homolog reported previously [Poldrack et al., 1999].

Another important aspect to be considered is prosody, that is, the suprasegmental features of natural speech including rhythm, intonation, and stress. The role of the right POP in prosodic processing has been demonstrated in pitch assessment [Celsis et al., 1999; Pugh et al., 1996; Zatorre et al., 1999] and sentence melody processing [Meyer et al., 2002]. On the basis of previous studies, our results of right‐hemispheric activation in Broca's area are presumed to be due to suprasegmental (prosodic) phonological information processing. Therefore, significantly greater activation in the POP in L2 than in L1 for the low‐frequency word condition is probably as a consequence of greater cognitive load of suprasegmental (prosodic) phonological information processing during both phonologically and semantically unfamiliar low‐frequency L2 word repetition, compared to phonologically familiar (but semantically unfamiliar) low‐frequency L1 word repetition.

Functional differences between anterior and posterior language‐related areas

The present study revealed significant differences in cortical activation between languages in the posterior language‐related areas (SMTG, AG, and SMG), but not in the anterior language‐related areas (POP and PTR). In other words, while the posterior language‐related areas were sensitive to language difference, the anterior language‐related areas were not, suggesting that phonological processes, such as perceiving and discriminating phonemes, are handled by areas for phonological analysis in the posterior language‐related areas prior to transmitting information to anterior language‐related areas.

Second Language Processing

Proficiency effects on cortical activation during L2 processing

The present results reveal that the high‐proficiency group outperformed the low‐proficiency group in behavioral performance in L2, suggesting that proficiency plays a measurable role in behavioral performance (ability to repeat L2 words correctly and precisely) in the phonological domain, just as in the syntactic and lexico‐semantic domains. We confirmed that L2 proficiency was not associated with age. Proficiency was clarified as a major factor influencing cortical activation in boys, although the same observations were not made in girls, which will be discussed in further detail later.

Age effects on cortical activation during L2 processing

The age effect appeared to have less of a significant effect on cortical activation than one might have thought. The most likely reason for the smaller effect of age on cortical activation with L2 is that participants within the same age group had different levels of English proficiency/exposure (which is entirely different from the case of L1 in terms of language education). Other possibilities include the task used and the narrow age range of participants; an age effect may be found in a task with a higher degree of difficulty. Nevertheless, an age effect was found in the temporal region in both boys (PAAC) and girls (SMTG). More specifically, cortical activation in this region decreased with age for both sexes. As discussed earlier, the vicinity of Wernicke's area is associated with auditory speech perception, especially lexical phonological access [Indefrey and Levelt, 2000, 2004], and the absolute and proportional volumes of this area were not significantly associated with age or sex in our participants' age range [Vadlamudi et al., 2006]. This suggests that the observed age effects in this region are due to functional changes with age.

The decrease in activation in the posterior temporal region (encompassing Wernicke's area) during development is in line with previous studies. Bitan et al. [2007] examined developmental changes in activation and effective connectivity among brain regions during a phonological processing task, using fMRI. Their participants were aged between 9 and 15, which partially overlaps with the age of the children in the present study. Thus, the present study extends their observations, suggesting that the developmental decrease in activation in the posterior temporal region would have already started from the age of 6–10 years. In addition, Bitan et al. [2007] revealed a developmental increase in activation in the dorsal inferior frontal gyrus, accompanied by a decrease in the dorsal superior temporal gyrus. The coupling of the dorsal inferior frontal gyrus with other specific brain regions involved in phonological decision‐making increased with age, while the coupling of the superior temporal gyrus decreased with age. They suggested that during development there is a shift from reliance on sensory auditory representations to reliance on phonological segmentation and covert articulation during word processing. They mentioned that a decrease in activation may reflect less engagement of a specific cognitive process in a given task, and consequently reduced activity in the neural substrates associated with that process. Alternatively, a developmental decrease in activation may reflect increasing neural efficiency in a specific brain region while it continues to contribute to the task. Similarly, a developmental decrease in extrastriate activation during word processing has been observed in participants aged 7 to 32 years [Brown et al., 2005]. This developmental decrease in activation has been suggested to result from improved “tuning” of lower level mechanisms (i.e., brain regions become more selectively activated over age for the same tasks) [Brown et al., 2005].

In summary, previous studies suggest that there is reduced reliance on primary sensory processes as task‐relevant processes mature and become more efficient during development, and the present results are consistent with this interpretation.

Sex differences in L2 learning and development

There were significant sex differences in proficiency‐related cortical activation in L2 processing. More specifically, as shown in Figure 4, activation was significantly increased in boys with L2 proficiency in a wide range of bilateral language areas, while girls did not show significant differences in activation between the two proficiency groups. In other words, while both boys and girls exhibited similar cortical activation at the initial stage of L2 learning (low‐proficiency group in Fig. 5), sex differences emerged when L2 proficiency increased (high‐proficiency group in Fig. 5). Given that both sexes showed significant improvement in behavioral performance (repetition success rates) as L2 proficiency increased, and that both sexes in the high‐proficiency group had comparable behavioral performance, the sex differences detected in brain activation could be attributed to the different cognitive functions or control strategies used.

Based on previous findings, there are several possible interpretations of these results: (1) there are sex differences in memory functions, (2) sex differences are acquired during language development through different cognitive strategies between sexes, or (3) a combination of both (1) and (2): sex differences are acquired or enlarged during language development through different cognitive strategies between sexes, which is resultant of different memory functions between sexes. These possibilities will be discussed here.

Sex differences in memory functions

A recent study investigated sex differences and their underlying mechanisms for a linguistic task (i.e., novel word learning) in adults [Kaushanskaya et al., 2011] and in children aged 5 to 7 [Kaushanskaya et al., 2013]. Both studies examining subjects' abilities to remember novel words revealed that females outperformed males on phonologically familiar novel words, but not on phonologically unfamiliar novel words. These findings suggest that females are more likely than males to recruit native‐language phonological knowledge during novel word learning. Namely, novel words that are more familiar in form and/or meaning are retained better than novel words that are less familiar [e.g., Ellis and Beaton, 1993; Service and Craik, 1993; Storkel, 2001], and females may be more sensitive to such linguistic familiarity effects than males.

Kaushanskaya et al. [2011, 2013], focusing on word learning within the declarative/procedural framework [Ullman, 2001, 2004; Ullman et al., 2008], attributed sex differences in language acquisition and processing to the superior declarative memory system of females [Herlitz et al., 1999; Kimura and Harshman, 1984; Larsson et al., 2003; Loonstra et al., 2001; Maitland et al., 2004]. The authors demonstrated that females outperform males on lexical learning tasks only when the novel words are constructed using native‐language phonological categories, supposedly stored in long‐term linguistic knowledge in a declarative memory system.

If this is the case, our results for word frequency effects and L2 proficiency effects could be interpreted as follows. With regard to low‐frequency words, sex differences in cortical activation might not be found as boys and girls rarely recognize the low‐frequency L2 words, lacking lexical knowledge of the words in their mental lexicon. Conversely, sex differences could be found in the case of high‐frequency words if girls have a better declarative memory system: girls would impose a smaller cognitive load on the phonological store than boys. Given that the SMG and AG are considered to be parietal regions crucial to the function of the phonological store [Awh et al., 1996; Becker et al., 1999; Paulesu et al., 1993; Vallar and Papagno, 1995], this interpretation seems reasonable.

As for L2 proficiency effects, different memory functions could also account for the existence or lack of sex differences in each group. In the low‐proficiency group, no sex differences would be observed in cortical activation regarding phonological store, as there is little or no lexical knowledge to draw from. Conversely, in the case of the high‐proficiency group, there should exist sex differences if girls have a better declarative memory system than boys: girls would place a smaller cognitive load on the memory system, including phonological store (SMG and AG), than would boys.

Sex differences in learning strategies

Given that significant sex differences were observed in the high‐proficiency group, while no sex differences were detected in the low‐proficiency group, our findings may indicate that sex differences in language function are acquired during language development, emerging from different cognitive strategies between the sexes. This interpretation is consistent with previous reports on L1 sex differences in children, which suggest that sex differences could actually increase during development as the differing cognitive strategies become entrenched [Burman et al., 2013]. Thus, our results revealing a sex effect in high‐proficiency groups compared to no effect in low‐proficiency groups might solve the controversy over sex differences in brain activation studies among adults. Sommer et al. [2004, 2008] mentioned that studies with a larger number of subjects are less likely to find sex differences, while they are well observed in small subject pools. It may well be that larger studies are likely to include more subjects with low language proficiency, increasing group variability and decreasing the incidence of subjects who do exhibit sex differences. Our results may demonstrate the importance of individual differences in language proficiency in analyzing sex differences.

With regard to learning strategies, ample empirical research has shown that women use a wider range of strategies than men do [Božinović and Sindik, 2011; Dreyer and Oxford, 1996; Lee and Oxford, 2008; Oxford and Nyikos, 1989]. For instance, a large‐scale study conducted by Catalán [2003] on 581 Spanish‐speaking students learning Basque and English as L2 show that male and female L2 learners differed significantly in both the number and range of strategies used. Females frequently use rehearsal strategies and input elicitation strategies, while males often use image vocabulary learning strategies, suggesting distinct learning styles or preferences by sex: females often say a new word when studying, and study the sound of a word, while males use the loci method, the peg method, form an image of the word's meaning, use physical action, and put English labels on objects. Other researchers also point out an auditory learning style for females and a visual and tactile one for males [Hodgetts, 2008; Oxford, 1994; Reid, 1987]. These previously observed sex differences in behavioral performances lead to the possibility that the cognitive/learning strategies utilized by boys might well differ from those used by girls, as in our study.

The present results suggest that while boys allocate processing resources in a widely distributed language‐related area including the inferior parietal cortex (AG and SMG) with more resources imposed to achieve higher performance, girls allocate resources in a more restricted region with little or no cognitive load in inferior parietal regions, but with a greater cognitive load in the SMTG (Wernicke's area) during speech perception in the early stage of L2 learning.

It is reported that Wernicke's area, especially around the posterior superior temporal gyrus, is the substrate for acoustic‐phonetic processing, and that the posterior superior temporal gyrus, inferior parietal cortex and inferior frontal gyrus support phonological decoding [Boatman, 2004]. In particular, the inferior parietal cortex has been associated with phonological processing in a number of neuroimaging studies [Celsis et al., 1999; Demonet et al., 1994; Paulesu et al., 1993]. Also, the phonological storage component of the phonological loop, postulated by a model of working memory [Baddeley, 2003], has been associated with inferior parietal cortex activation [Awh et al., 1996; Becker et al., 1999; Paulesu et al., 1993; Vallar and Papagno, 1995].

In a review, Boatman [2004] described that acoustic–phonetic analysis is one of the earliest stages of speech perception and refers to a listener's ability to discriminate speech sounds based on their phonetic features. At the same time, phonological processing refers to the decoding and mapping of acoustic–phonetic information onto the listener's internal representations, which are subsequently used to access other language systems including word information (lexical and semantic). Given these findings, girls may be considerably engaged in performing acoustic–phonetic analyses but may not apply higher‐level control during word processing in the same way as boys, while boys may utilize a different mode of word analysis in the inferior parietal regions: decoding the phonological identity of a sound and comparing each sound with an internal representation of the sound (and maybe also with an internal representation of lexical‐semantic knowledge) in addition to performing acoustic–phonetic analyses, especially as proficiency increases.

A cooperative interpretation: development of sex differences in language learning derived from a sex difference in memory functions

While we have seen numerous previous studies of sex differences in memory functions and learning strategies, there is no decisive evidence at present that would allow us to identify which interpretation is most conceivable. Therefore, based on the present and previous evidence of the development of sex differences in language processing, together with the possible explanations being discussed here, our results are most plausibly interpreted as memory function and learning strategies combining to produce observed sex differences. Namely, sex differences in cortical activation may be acquired or enlarged during language development through different cognitive strategies between sexes, which originally results from different memory functions between sexes.

Finally, the female characteristic of nonsignificant differences in cortical activation in the context of significant improvement in performance (repetition success rates) with L2 proficiency could be interpreted in the following way. Unlike the unconscious, effortless acquisition of a first language, the early or initial stages of second‐ or foreign‐language learning require volitional and conscious effort with a large cognitive load, such as in working memory [Newell, 1990]. However, the mental process changes from controlled or conscious processing to automatic or unconscious processing as proficiency increases [Dekeyser, 2001; Segalowitz, 2003]. It is also well known that, while brain activation increases with cognitive load, it decreases with fluency and proficiency when the execution of language tasks becomes easy, effortless and automatic [Chee et al., 2001; Musso et al., 2003]. During language development, cortical activation increases at the onset of acquisition by controlled operations and is followed by the maintenance of the activation by more automatic operations (a plateau phase with either no or only a slight decrease of activation) [Sakai, 2005]. Based on these previous findings together with the interpretation of differing strategies between sexes, it is possible to postulate that boys develop lexical competence, including phonological awareness skills, gradually with deeper analysis of the words using a wide range of language areas including inferior parietal regions, and their cognitive load increases with L2 proficiency to improve performance accuracy. In contrast, girls may acquire phonological awareness skills efficiently utilizing both a function of Wernicke's area, which allows them to make the best use of accoustic‐phonetic analyses of words, and a superior declarative memory system. Thus, as proficiency increases, girls gain automatic operations in L2 phonological processing, leading to significantly smaller activation compared to boys with efficient use of resources. This does not necessarily mean that girls acquire or develop lexical competence, including semantics of words, efficiently. Other possibilities remain,2 and it should be noted that our study revealed sex differences in L2 phonological processing, but does not suggest sex differences in all language functions in this age range.

CONCLUSIONS

The present study, focusing on the phonological domain in children, clarified that L1 and L2 share similar neural structures even with linguistically distant languages. We found differential effects for brain activation in boys and girls in early L2 learning, while decreased activation in L2 compared to L1 in the posterior language‐related areas (SMTG, AG, and SMG) were common to both sexes. The observed cortical sex differences were found during high‐frequency word processing, but not during low‐frequency word processing. An L2 proficiency effect was also observed in that boys had significantly increased activation with proficiency in L2, whereas girls did not. Taken together with the previous findings, the present results suggest that sex differences are acquired or enlarged during language development through different cognitive strategies between sexes, enhancing originally existing differing memory functions between sexes. Sex differences in linguistic functions are enthusiastically debated in both behavioral and neuroscience studies, which often provide evidence for female superiority; but, it is also controversial. In the present study, sex differences were not found in the behavioral performance, but significant sex differences were found in cortical activation patterns especially during phonologically familiar word processing in L2, and when L2 proficiency was high. The present large‐scale neuroimaging study of school‐aged children demonstrates that sex differences are not always observable in linguistic tasks, even within the phonological domain, but can be observed under certain conditions. The results suggest female sensitivity to phonological familiarity effects, an interpretation that is corroborated with evidence of a lower cognitive load in the inferior parietal regions involved in the phonological store, and an interaction between sex and language proficiency that has been overlooked in most studies.