Considering that an increasing percentage of the population worldwide masters even structurally distant languages, this study aimed to investigate how L1-L2 linguistic distance may have an effect on L2 brain representation. We carried out a quantitative meta-analysis to investigate the role of this factor, although the linguistic distance was not directly assessed or quantified in the studies we selected.

In particular, we focused on studies with bilinguals speaking English (i.e., an Indo-European language) as L2 and either a European language with Latin script or Chinese (i.e., a Sino-Tibetan language) as L1. These specific language groups were chosen for two reasons as follows: the overall marked linguistic distance between English and Chinese (Chiswick and Miller, 2005; Wichmann et al., 2012) and the fact that there is an adequate number of neuroimaging studies involving these language pairs. As detailed below, we limited our choice to European languages with Latin script, which are phylogenetically close to English and have a close common progenitor, as shown by several language trees (Serva and Petroni, 2008; Chang et al., 2015). Furthermore, as reported in Chiswick and Miller (2005), English appeared to be structurally much closer to the selected Indo-European languages (with a score of 2.25–2.50) than to Chinese (with a score of 1.00). In other words, we aimed to inspect whether linguistic distance as derived from this index can actually translate into differential English-L2 functional activations.

Second, we controlled for the effect of L2 AoA and proficiency, given that both these factors were identified by prior meta-analyses as fundamental influencers of the L2 language network in healthy bilinguals (Sebastian et al., 2012; Liu and Cao, 2016; Cargnelutti et al., 2019; Sulpizio et al., 2020) and bilinguals with aphasia (Kuzmina et al., 2019). Specifically, given that late AoA and low proficiency are likely to pose greater L2 demands and therefore translate into additional functional activations, we investigated whether these factors could further modulate the effect of linguistic distance on L2 brain representation.

We provided an almost exploratory description because only a few studies to date have explored these specific bilingual patterns. However, we believed that investigating how L2-related network changes in response to L1-L2 linguistic distance are a relevant topic of growing interest. With our current preliminary findings, we also aimed to prompt further research in the field using different language pairs and a better conceptualization of the linguistic distance concept.

We performed a keyword search (i.e., “fMRI” or “functional MRI,” “PET,” “bilingual*,” “multilingual*”) in databases such as PubMed/MedLine, Scopus, and Scholar to extract papers published in the 1995–2019 period and focused on language representation in the bilingual brain. This sample was complemented by papers identified from inspection of the reference lists of the identified papers. We selected peer-reviewed papers written in English, a process that could lead to a publication bias. However, coordinate-based meta-analyses look for spatial convergence between reported coordinates and thus they differ from effect-size meta-analyses, which quantify the effect size, and are instead prone to bias. Current meta-analyses are less prone to region- and task-dependent biases and less affected by the exclusion of unpublished data (Fox et al., 1998; Rottschy et al., 2012). To limit potential sources of bias, we excluded findings related to a priori selected ROIs and chose to include only those from whole-brain analyses.

The paper selection procedure is shown in Figure 1 (PRISMA flow chart, Moher et al., 2009). We included only studies performed on healthy adult individuals (aged 18–60 years) addressing the main language structural domains (i.e., lexical semantics, phonology/articulation, and morphosyntax, see Sebastian et al., 2012; Liu and Cao, 2016). To reduce data variability, we excluded studies investigating non-purely linguistic tasks (e.g., affective/emotional components of language, numbers, and mathematics) and tasks that were specific to bilinguals (e.g., translation/interpretation, switching, and language control). The latter tap additional brain processes as both languages are simultaneously activated during these tasks and, therefore, deserve a specific meta-analysis (Luk et al., 2011). We also excluded studies assessing language performance after either a learning/training period or some kind of manipulation in the degree of exposure to a given language.

The paper selection consisted of several steps, which were guided by the literature and the language pairs spoken by bilinguals in the eligible papers. Paper inclusion was ultimately determined by the languages addressed in the candidate papers and by how much they were represented in the literature. In the first step, we scrutinized all studies with bilinguals having an Indo-European language as L2 and either another Indo-European language or one of the languages spoken in Eastern Asia as L1 (following language classification in Wichmann et al., 2012 and measures of language distance in Chiswick and Miller, 2005).

This first selection resulted in 108 potentially candidate papers, which were then accurately read to possibly exclude papers due to: (i) absent or incomplete (not full 3D) coordinates, only coordinate ranges (i.e., no single coordinate corresponding to the peak activation in a given functional cluster), coordinates reported only for a single subject (n = 2); (ii) coordinates from a priori selected ROIs (n = 3); (iii) analyses with contrasts not being informative (e.g., they did not differentiate between different languages or between bilinguals and monolinguals) or that were too specific or related to a very low-level of linguistic processing, such as passive viewing of single letters (n = 3); (iv) assessment after language training (n = 1).

A sample of 99 papers survived this scrutiny as they all met the criteria for our meta-analysis. Then, we focused on the available language pairs. We observed that East-Asian languages considered as L1 included Chinese, Korean, and Japanese, which belong to three different language families (i.e., Sino-Tibetan, Koreanic, and Japonic, respectively). In the large majority of these studies, L2 was English. This led us to choose this language as a common denominator between the two groups. Concerning L1, given that only the studies with Chinese-native speakers had the number of experiments (n > 15) sufficient to perform robust meta-analyses, we restricted the analyses to this language. In the group with an Indo-European language as L1, we chose languages with less linguistic distance from English (Chiswick and Miller, 2005; Serva and Petroni, 2008; Chang et al., 2015), namely, alphabetic European languages with Latin script (as previously described). Then, Dutch, French, German, Italian, Portuguese, and Spanish were included (and Bengali, Hindi, Macedonian, and Fārsí excluded). The resulting sample consisted of 31 studies: 13 studies with a European language as L1 and 18 studies with Chinese as L1. Hereafter, we named the two groups as European group and Chinese group, respectively. The full list of papers is reported in Table 1.

The analyses were carried out using the GingerALE software (brainmap.org), relying on a coordinate-based activation likelihood estimation (ALE) algorithm, looking for consistency in functional coordinates across reported contrasts (Turkeltaub et al., 2012; Eickhoff et al., 2009; Laird et al., 2009a,b). This algorithm is based on a random-effect approach that accounts for spatial uncertainty by treating the reported foci as centers for 3D Gaussian probability distributions. The provided probability distribution maps, which were weighted on the number of subjects, described the probability for a given focus to be within a given voxel.

(i) Investigation of the functional brain networks associated with the two languages, and specifically with L2, in both groups and subsequent comparison between the two groups. We were confident in including different language tasks in the same analysis, given that the algorithm, as just described, looks for areas showing a convergence of activation across different experiments and therefore provides only consistently recurring activations. This choice was also motivated by the restricted number of selected studies, which prevented robust analyses for specific linguistic domains, tasks, or stimulus presentation modalities. However, in Supplementary Results, we reported an exploratory analysis on lexical semantics, which was the most frequently represented domain.

(ii) Investigation of the role of AoA. After carrying out the analysis on the whole groups of bilinguals—which we expected to provide the most general and robust functional activations irrespective of influencers –, we identified two groups of early and late bilinguals following the most commonly adopted AoA cutoff (i.e., L2 appropriation either before or after the age of 6 years). We found that studies including early bilinguals (AoA < 6) were limited in number, especially for the Chinese group. For this reason, we restricted the analyses to late bilinguals. We expected the analysis to focus on late bilinguals to better highlight potential differences between the two groups.

(iii) Investigation of the role of L2 proficiency. Assessment of proficiency is sometimes inappropriate for the purpose of systematic analyses. Accordingly, many studies reported either qualitative or quantitative—yet non-objective—proficiency rates (e.g., self-ratings in rating scales). Only a small percentage of studies tested the proficiency level by a comprehensive screening and standardized batteries, even though neither proved an optimal measure of proficiency (Grosjean, 2010). Although conscious of possible biases, we retained only studies declaring a high proficiency or with high self-rating or screening test scores. As long as proficiency was observed to reduce differences in brain activations between L1 and L2 (Sebastian et al., 2012), it was interesting to investigate to what extent these differences could be modulated by linguistic distance.

We performed (i) main effect analyses, which provide results consisting in the functional activations associated with a specific condition (i.e., L2 in the European group and L2 in the Chinese group) and (ii) contrast analysis, which provide conjunction results, namely, the areas activated in both conditions (i.e., in both the European and the Chinese groups), and subtraction results, which reveal specific activations, namely, areas emerging from the direct comparison between the two conditions and being activated in one condition but not in the other (i.e., the functional activations of the European group survived after subtracting activations of the Chinese group and vice versa).

To set thresholds for both main effect and contrast analyses, we followed the software user instructions and reference articles (Eickhoff et al., 2016; Xu et al., 2020), in order to use approaches already observed to be highly reliable. For main effect analyses, we used a p < 0.05 permutation-based cluster-level family-wise error (FWE) corrected threshold (1,000 permutations; voxel-wise threshold p < 0.001, uncorrected) and a minimum cluster size of 200 mm³ (25 voxels). This is a conservative thresholding approach. For contrast analyses, we based on the conservative minimum statistics (Nichols et al., 2005); therefore, only areas that resulted to be significant in the individual analyses were included. We thresholded probability values at p < 0.001, with 10,000 p-value permutations and minimum cluster size of 80 mm³ (10 voxels); however, for the conjunction results, we only retained clusters of at least 120 mm³ (15 voxels), in order to avoid incidental overlap between individual ALE maps (Rottschy et al., 2012; Krall et al., 2015).

We transferred the coordinates to Montreal Neurological Institute (MNI) standard space. Coordinates in Talairach and Tournoux (1988) space were converted to the MNI space by icbm_spm2tal transform before running the analyses. Anatomical localization and labeling of resultant clusters of activation were performed using the SPM Anatomy toolbox (Eickhoff et al., 2005), which assigns activations to the most probable cytoarchitectonic area.

Results of both main effect and contrast analyses are reported in Table 2. Figure 2 shows the results from main effect analyses and Figure 3 from contrast analyses.

The main effect and contrast analysis results are detailed in Table 3. Figure 4 shows the results from main effect analyses and Figure 5 from contrast analyses.

The main effect and contrast analysis results are detailed in Table 3. Figure 6 shows the results from main effect analyses and Figure 7 from contrast analyses.

The first analysis included all the selected studies independently of language domain, task, and presentation modality or factors relevant to bilingualism, such as AoA or proficiency. Interestingly, contrast analysis revealed a specific activation of the left insula for the European group and of the inferior frontal gyrus in the territory of the DLPFC for the Chinese group. Furthermore, the Chinese group selectively activated a cluster peaking in the intraparietal sulcus and extending to both IPL and SPL.

Activation of the insula and of the DLPFC was not surprising as these areas subserve more general higher cognitive functions including cognitive control, monitoring, and attention. Although not language-specific, these functions were observed to play a relevant role in language, particularly in cognitively demanding situations such as those posed by bilingualism. Concerning the specific activation in the European group, the insula has a relevant role in diverse cognitive functions including language (Oh et al., 2014) and is particularly involved in supporting the execution of tasks posing conflict and interference (Wager et al., 2005). In light of these findings, it could appear counterintuitive for this region to be specifically recruited in bilinguals knowing closer languages. In point of fact, one may hypothesize a greater need for cognitive support when dealing with two languages significantly differing in many respects (e.g., phonology and writing system) due to the need to switch to a different processing modality (e.g., the way graphemes are converted into phonemes).

Nevertheless, there is strong evidence that handling two languages is more challenging when they share linguistic features rather than when they are more structurally distant (Paradis, 1994, 2004, 2009; Fabbro, 1999); this is also shown by interference occurring, in specific conditions, between cognate words (Colomé and Miozzo, 2010; Muscalu and Smiley, 2019). This phenomenon was also described in some clinical studies in which languages sharing cognate words were observed to interfere with each other, thus hampering recovery in some bilingual patients with aphasia (Kurland and Falcon, 2011).

Regarding the Chinese group, selective activation of another crucial area involved in general cognitive functions was observed, namely, the DLPFC (MacDonald et al., 2000; Mansouri et al., 2007). In bilinguals, the DLPFC located in the inferior frontal gyrus included the control network, which regulates selection and control, in particular when a weaker language is in use (Abutalebi and Green, 2007, 2016). A cluster in the inferior/middle frontal gyrus (MNI coordinates: x = -44, y = 18, and z = 20) was also found in the meta-analysis by Liu and Cao (2016) with an L2 which was orthographically deeper (i.e., opaquer) than L1, as in the case of Chinese vs. English. The authors hypothesized that this area was recruited to handle the greater difficulty to coordinate sounds and meaning when L2 is an opaque language.

Taken together, these findings seem to suggest that, irrespective of similarity between known languages, bilinguals do tend to strongly rely on cognitive resources when dealing with their L2; however, the cognitive demands posed by closer vs. more distant languages could be different and translate into the specific recruitment of different areas. This hypothesis is more deeply discussed in the next paragraphs, where we illustrated the results obtained when controlling for AoA and proficiency.

The other specific activation for the Chinese group involved the IPL and SPL.² The IPL has been mainly (although not uniquely) associated with phonological processing in both Chinese (Wu et al., 2012) and European languages (Hartwigsen et al., 2010) as well as with lexical knowledge in both (Mechelli et al., 2004; Abutalebi et al., 2015). Inspection of the contrasts contributing to this cluster showed that they pertained to different language domains and suggested that this activation was not task-specific.

The cluster we found also covered the SPL, which is generally recruited during spelling-to-sound mapping (Taylor et al., 2013) and acts as a station for audio-visual multisensory integration (Molholm et al., 2006). This area was observed to be particularly crucial for Chinese language processing, for instance when processing characters across diverse tasks (e.g., orthographic and semantic; see meta-analysis by Wu et al., 2012), but it is also involved in processing Japanese Kanji characters (Matsuo et al., 2000). For this reason, the SPL is considered a typical logographic language area, with a hypothesized involvement in holistic visuospatial processing (Bolger et al., 2005).

Activation of this area can be tentatively explained by drawing on the assimilation-accommodation hypothesis (Perfetti et al., 2007), which postulates that bilinguals may either rely on assimilation (i.e., applying the same procedures developed for L1 to L2 as well) or on accommodation (i.e., “abandoning” the procedures associated with L1 processing to activate those specific for L2). Our results cautiously suggest that Chinese-English bilinguals tend to rely on assimilation when using English.

Overall, these findings (Liu and Cao, 2016) suggest that the application of long-established mechanisms, although not perfectly fitting with the new language demands, could be the default strategy used to deal with a new distant language. This can more likely occur with late language learning, as findings the related paragraph seem to suggest. Bilinguals would use these mechanisms as long as these are effective and develop new-language-specific processing mechanisms only when those associated with L1 are inefficient.

Another interesting activation cluster in the Chinese—but not the European—group was represented by the posterior-medial frontal cortex, although it did not survive the statistical thresholds in the subtraction analysis. The cluster was located in close proximity to the pre-SMA/ACC region described by Abutalebi and Green, another fundamental station in language control, as it is involved in appropriate language selection and switching between languages (Abutalebi and Green, 2007, 2016). This activation, too, showed the high cognitive demand posed by L2 in this group of bilinguals.

These preliminary results probably reflect the activation clusters that are more tightly associated with the two conditions (closer vs. more distant language pairs). These activations are also likely to bypass language domain-specificity and represent the brain areas that are generally recruited when performing any task in L2, as they can reflect the cognitive load of managing an additional language (see also the exploratory lexical-semantic analysis in Supplementary Results).

Several previous meta-analyses addressed the role of AoA and showed that, overall, late language learning (typically after the age of 6 years) is associated with the recruitment of additional and/or wider brain areas (Liu and Cao, 2016; Cargnelutti et al., 2019). This reflected a higher cognitive effort with respect to early L2 acquisition. The impact of AoA can be even more interesting in relation to linguistic distance: one may hypothesize that learning a new structurally distant language might be more demanding when it takes place late, therefore, when the cognitive mechanisms for L1 are already established.

For this reason, we replicated the analysis while controlling for AoA. Unfortunately, data for early bilinguals were limited, and therefore, comparison between early and late bilinguals was not possible. However, we can make indirect inferences on the results of late bilinguals. Our analyses showed that the European group activated specifically the left insula and hippocampus, whereas the Chinese group activated the left DLPFC and SPL. The main effect analysis showed that this group also activated the posterior-medial frontal gyrus. Also, in this case, the two groups did not share any suprathreshold activation clusters.

Findings for the Chinese group replicate those achieved for the whole group (i.e., without distinction based on AoA), suggesting that the specific activations we observed could be more relevant for the Chinese bilinguals having learned English late. This is reasonable as greater cognitive control—reflected in the specific activation of the DLPFC—can be necessary to learn L2 when the L1 network is already developed. Furthermore, as previously commented, late learners may tend to approach a new language by first applying—and, if successful, keeping—the same cognitive mechanisms developed to process L1 (assimilation). Activation of the SPL in late bilinguals suggested that this could be the case.

Concerning the European group, it was interesting to observe that the left insula was activated for the whole group. By inspecting the contrasts contributing to the cluster, we observed that these were represented by tasks involving potential conflict, for instance, lexical decision (i.e., decide if letter strings were real words or not). This finding contributes to supporting the previously mentioned hypothesis of insula recruitment to solve potential conflicts, which are amplified when dealing with L2.

With regard to the other European-group-specific activation, this did not match any macroanatomic probability map but it predominantly included the hippocampus. Contrasts contributing to this cluster mostly emerged from the comparison of morphological tasks in late vs. early bilinguals, suggesting that the hippocampus might be more likely recruited in the case of late AoA. This specific activation was observed to be driven by tasks tapping grammar. This was in agreement with the memory models describing language learning vs. acquisition (Paradis, 1994, 2004, 2009; Ullman, 2001, 2005, 2006). In fact, late learning involves the explicit memory system, whereas early acquisition relies on implicit processes, allowing the automatic performance of these tasks. In light of this theoretical background, the European group could rely on hippocampal activation, which is fundamental for memory retrieval, when performing morphological tasks in English. The different European languages included in our study are significantly more similar to English than Chinese (Chiswick and Miller, 2005). Accordingly, it is plausible that European bilinguals mainly rely on direct recall to perform these tasks in order to bypass potential interference between grammar rules of close languages. A retrieval could be a more effective strategy, whereas recruitment of regions supporting cognitive processing might be more successful for Chinese bilinguals. Another hypothesis comes from the recently observed role of the hippocampus in the flexible use and processing of language (Covington and Duff, 2016) and, remarkably, in associative learning (Brasted et al., 2003). Accordingly, its activation could reflect a more efficient learning process, where the potential obstacle represented by late AoA is overcome by the development of flexible and associative strategies for new rule application. These are only speculations, which need to be confirmed by a sufficient number of studies comparing activations associated with the same linguistic tasks in different groups of bilinguals; in this way, it would be possible to understand whether hippocampal involvement depends on specific tasks, languages, or profile of bilinguals (e.g., AoA and proficiency).

Another crucial aspect concerns the role of proficiency as discussed in previous meta-analyses (Sebastian et al., 2012; Cargnelutti et al., 2019): low proficiency is associated with greater cognitive effort and then with greater activation of areas involved in cognitive control. The role of proficiency in relation to linguistic distance could be particularly interesting to explore.

As previously detailed, we could not make a direct comparison between low- and high-proficiency bilinguals owing to the paucity of available data. The exploratory analysis carried out on bilinguals who were proficient in their L2 shows—partially unexpectedly—that both the European and Chinese groups activated regions supporting general cognitive functions. European bilinguals activated the insula bilaterally, although this activation emerged from the main effect analysis only and not from the subtraction analysis (again, probably due to the paucity of studies, which is a limitation especially in subtraction analyses). In contrast, the Chinese group activated specifically the posterior-medial frontal gyrus. Taken together, these findings indicate the role of these regions in supporting even proficient performance.

The insula was previously observed to activate in balanced bilinguals (i.e., bilinguals with native-like performance in both languages) compared with less proficient bilinguals and also in response to increased task difficulty (Chee et al., 2004). This seems to confirm that this region is recruited to cope with increased cognitive demands (Vigneau et al., 2011). In light of these findings, we can also hypothesize that the insula is activated when L2 is mastered very well: a very good mastery of more than one language is a demanding process but the insula could promote an anyway proficient performance.

By looking at peak location, it was similar between proficient bilinguals and late bilinguals (note that analysis in the group of late bilinguals included six contrasts referring to high-proficiency bilinguals and 10 to low-proficiency bilinguals, suggesting that a similar result could have been affected independently by both AoA and proficiency). For both groups, we can hypothesize a similar role of the insula in supporting the cognitive effort, although for different reasons: (i) the need for proficient bilinguals to perfectly master the two languages in conflict situations (i.e., bilingual environment) and (ii) the need for late bilinguals to overcome obstacles posed by late learning. Therefore, the other activation clusters are likely to work in synergy with the insula to meet the specific demands of the two conditions.

Regarding the Chinese group, the lack of SPL activation in this analysis is noteworthy. Although a direct comparison between high- and low-proficiency bilinguals was not possible, we could speculate that a high level of proficiency can be more easily reached when accommodation—rather than assimilation—processes take place and may also be prompted by accommodation. High proficiency is more likely gained throughout proper recruitment of the areas supporting general cognitive functions than by areas specifically associated with a given language.