Neurocognitive determinants of performance variability among world-language users

2.2.1 The age of appropriation factor

A world language may be appropriated as an L2 at varying ages, ranging from early childhood to adulthood. The ability to incidentally acquire linguistic information is bound to optimal periods, which renders AoA another key factor underlying inter-user variability. By age five, approximately, procedural memory circuits begin to lose plasticity (Paradis 2009), which progressively undermines the brain’s ability to acquire new implicit, automatic routines. Studies with humans and rodents confirm that the acquisition of procedural information, via the basal ganglia, is subject to short-lived critical periods (Fredriksson et al. 2000; Schlaug 2001). At the same time, declarative memory performance steadily improves throughout childhood (DiGiulio et al. 1994), and then decays during adulthood (Kirasic et al. 1996). One reason behind these changes would be the increase of estrogen levels during childhood and adolescence, as this substance seems to inhibit the procedural system while promoting activation in the declarative system (Ullman 2004).

Birdsong (1999) showed that, in languages appropriated after late childhood (i.e., around age seven), grammatical skills are poorer than lexical abilities. Möhring (2001) tested early bilinguals and found that gender agreement rules were easily acquired before age three, but became progressively more difficult at later ages. Moreover, the review offered by Birdsong (2006) shows that most of the tasks in which bilinguals’ L2 performance is comparable to their performance in L1 are offline in nature – meaning that they allow for conscious control via declarative and executive processes.

These data warrant the postulation that the critical AoA to distinguish between early and late L2 learners lies between ages five and seven. This critical period is determined by maturational brain processes rendering individuals above age seven better prepared to learn – rather than acquire – linguistic information. Thus, whereas early learners may promptly achieve native-like levels of performance in processing implicit abstract routines, late learners only rarely manage to do so. Indeed, the neural representation of an early learned L2 is much more similar to that of a native language than is a lately learned L2. In reference to the declarative/procedural model of bilingualism, Fabbro (2001b, 219) states that:

[…] the acquisition or learning modality seems to determine a different participation of procedural memory systems vs. declarative memory systems. If L1 and L2 are acquired in informal contexts and both are at a high level of proficiency, their phonologic and morphosyntactic aspects are stored in procedural memory systems. On the other hand, traditional learning of L2 after the age of 7, along with limited proficiency in production, seems to involve the declarative memory systems to a greater extent.

Indeed, early bilinguals tend to represent the grammar and the lexico-semantics of their L2s in the same broad neuroanatomical regions as their L1s – i.e., grammar is represented in procedural memory (frontobasal regions), and lexical information is processed by declarative memory (temporal/temporo-parietal areas). On the contrary, late bilinguals tend to represent both the grammatical and the lexical information of their L2 in declarative memory. Aphasiological evidence supports this claim. Ku et al. (1996) report the case of a late Mandarin-English bilingual with a lesion focused on the left temporal lobe, which compromised his syntactic abilities more markedly in L2 than in L1. Aladdin et al. (2008) discuss two Ukranian-English bilingual epileptics whose seizures originated in left temporal regions. In both cases, for roughly 20 minutes post-seizure, the patients would find themselves unable to speak in L2 while their L1 remained virtually unaffected (comprehension abilities, however, remained fully functional in both languages).

However, when late bilinguals sustain lesions to frontobasal areas, their grammatical competence in L1 is selectively compromised. Evidence for this dissociation is offered by Fabbro and Paradis (1995), who review four cases of bilingual aphasia subsequent to basal ganglia damage. All four patients (late learners of their respective L2s) were more significantly impaired in their native than in their non-native languages (e.g., they omitted more functors in obligatory contexts in L1 than in L2). A similar pattern was observed by Garcia-Caballero et al. (2007) in their case study of a late bilingual suffering from crossed aphasia.

Converging evidence has been found through neuroimaging studies with non-pathological subjects. Single-word tasks consistently show that lexico-semantic processing in L1 and L2, regardless of AoA, engages the same macroanatomical regions, especially within the temporal lobe (Illes et al. 1999; Paradis 2004; Ullman 2005; Mondt et al. 2009). However, tasks involving syntactic processing (parsing or stripping) indicate a marked AoA effect.

For example, in the Chee Michael et al. (1999) experiment, early Chinese-English bilinguals were asked to decide whether sentences presented in one or the other language were true or false. Stronger activations were detected in the middle and inferior prefrontal cortices, the left temporal region, the left angular gyrus, the supplementary motor cortex, and bilateral occipital and parietal areas. What is remarkable is that all these areas were equally activated in both languages, which supports the view that syntactic processes in early bilinguals engage the same neurocognitive mechanisms for both languages. For their own part, in a positron emission tomography (PET) experiment, Perani et al. (1996) asked late Italian-English bilinguals to listen to stories in L1, L2, and a third language unknown to them. In L1, differential activations were observed in the inferior frontal gyrus, the medial and superior temporal gyri, the temporal pole, the angular gyrus, and the right cerebellum. However, L2 yielded significant activations only in bilateral middle and superior temporal areas and parahippocampal regions. These data are consistent with the claim that, in late bilinguals, L2 syntactic processing relies more heavily on posterior regions implicated in declarative memory.

Other studies provide direct comparisons between early and late bilinguals. Hirsch et al. (1997), in an fMRI experiment, compared the activation patterns of early and late bilinguals during a silent sentence-generation task. Both L1 and L2 yielded overlapping activations in Broca’s area in the “early” group, but they engaged separate frontal regions in the “late” group. No differences between groups were observed in Wernicke’s area. In a more recent fMRI study, Waldron and Hernandez (2013) compared brain activation patterns in early and late Spanish-English bilinguals with similar proficiency in both languages. Participants were asked to covertly generate the past tense form of visually presented L2 verbs. The “early” group demonstrated greater activations throughout frontal-temporal regions associated with automatic processes, whereas the “late” group presented differential activations in posterior and prefrontal areas associated to controlled processes.

There is also behavioral evidence demonstrating the role of AoA in L2 processing. For example, Marinis and Chondrogianni (2011) showed that the development of comprehension of pronouns and reflexives is similar between early bilingual children’s L2 and monolinguals’ native language, but not between the former group and late L2 learners. To the extent that such word classes involve grammatical processing, these results are also consistent with the claim that grammar is subserved by different cognitive mechanisms for early and late bilinguals.

In sum, the evidence indicates that AoA has an impact on how non-native languages are neurocognitively represented and processed (but see Frenck-Mestre et al. 2005). If these are appropriated at an early age, they will tend to resemble native languages both in the anatomical distribution of their subsystems and in the type of cognitive mechanism responsible for the latter’s processing. Hence, inter-user performance variability in typically procedural (e.g., grammatical) functions will not be as marked as variability in declarative (e.g., lexical) functions. On the contrary, if a language is learned at a late age, all of its subsystems will tend to be processed declaratively, via conscious, controlled mechanisms – including executive processes (Waldron and Hernandez 2013), leading to greater interindividual variability in performance.

2.2.2 The level of proficiency factor

Another factor underlying performance variability in world languages is LoP. LoP and AoA effects are often confounded in experimental designs, since early bilinguals tend to have higher LoPs than late bilinguals. However, each variable contributes independently to performance variability among users.

There is abundant behavioral evidence demonstrating that LoP modulates processing in a non-native language. For instance, translation asymmetries, asymmetrical switching costs, and differences in conceptual involvement between L1 and L2 tasks are greater in low-proficiency than in high-proficiency bilinguals (for reviews, see French and Jacquet [2004], Brysbaert and Duyck [2010], and Kroll et al. [2010]).¹ Moreover, LoP has an impact on the neurological representation of languages. In this sense, a leading proposal is the so-called convergence hypothesis:

According to this hypothesis, as proficiency in L2 increases, the representation of L2 and its processing profile (i.e., ERP and neuroimaging data) converge with those of native speakers of that language. That is, any qualitative differences between native speakers of a language, and L2 speakers of that language, disappear as proficiency increases […] Notice the convergence hypothesis is a claim about neural representation and processing profiles and not a claim about whether or not an L2 speaker of a language can simulate or pass off as a native speaker of that language (Green 2004, 5).

Support for this view has been found in varied studies. For example, Ibrahim (2009) reports the case of M.H., an aphasic bilingual patient who suffered an intracraneal hemorrhage compromising his left temporal lobe (declarative memory) but sparing frontobasal regions (procedural memory). A native speaker of Arabic, M.H. learned Hebrew metalinguistically, through formal instruction, since the fourth grade. Although he was a late bilingual, he had achieved a high LoP in Hebrew (not only did he use it daily in academic, professional, and private contexts, but he graduated from a Hebrew university). Subsequent to his lesion, M.H. presented with similar deficits in both languages, in fluency, comprehension, and repetition. Moreover, after three months of therapy, grammatical deficits were not observed in either language, although vocabulary deficits remained. This case suggests that even the grammar of a lately learned L2 may eventually be processed through the more automatic mechanisms of procedural memory, provided the user has developed a sufficiently high LoP.

More evidence has been offered in neuroimaging experiments. Using the PET technique, Perani et al. (1998) formed two groups which differed in AoA but possessed similar LoPs, and asked them to listen to stories in L1, L2, and a third language unknown to them. The “early” group comprised native speakers of Italian who spoke either Spanish or Catalan as L2. The “late” group was composed of Italian-English bilinguals. The latter group showed left temporal and hippocampal activations for both languages, without significant differences between them. For its own part, the “early” group presented bilateral hippocampal and left-lateralized temporal and parietal activations. Left temporal activations, however, were less pronounced in this group. This study shows that differences in L2 neural representation between age groups may be attenuated with increased LoPs by late learners – i.e., extensive practice and exposure lead to greater macroanatomical convergence between corresponding L1 and L2 subsystems.

In an fMRI study, Videsott et al. (2010) administered a picture naming task to a sample of native Ladin² speakers with a high LoP in Italian and an intermediate LoP in English. While word production in all three languages engaged common brain areas, the two fluently spoken languages involved enhanced right prefrontal activity relative to English. Given the well-documented role of the right prefrontal cortex in cognitive control, the authors suggest that such a structure may support language proficiency by supervising effective word retrieval.

LoP also modulates brain potentials associated with lexical processing. Moreno and Kutas (2005) examined N400 deflections in two groups of Spanish-English bilinguals during a semantic decision task. One group was composed of early bilinguals with high vocabulary proficiency in English, whereas the other comprised late learners with lower English vocabulary proficiency. In both subgroups, the N400 effect peaked with a significant delay in the non-dominant than in the dominant language. Among other results, it was found that both AoA and LoP independently accounted for a statistically significant amount of the variance in N400 latency. The authors further suggested that early AoA does not necessarily guarantee a fast response to semantic incongruity. Moreover, a language dominance effect was found which resembled abstractness/concreteness effects observed at a similar time range within a single language (i.e., relative to abstract words, concrete words yielded a larger anterior negativity). This led the authors to conjecture that LoP may influence a language’s overall processing style, making it more literal or metaphorical. For further evidence that L2 LoP is systematically associated with ERP responses, see Hahne (2001), Elston-Güttler et al. (2005), and Kotz and Elston-Güttler (2007).

All in all, LoP-related evidence seems largely consistent with the convergence hypothesis. A non-native world language spoken at a low LoP is processed via conscious, declarative mechanisms, and its grammar is represented in posterior brain areas that are separate from those subserving native-language grammar. However, as LoP increases, both its processing and gross neural representation tend to converge with those of native languages (i.e., processing depends on frontobasal regions, becoming more automatic and implicit). Moreover, LoP differences account for performance variability even in strictly declarative functions, such as vocabulary processing: a greater LoP reduces word activation latencies and executive involvement.

2.2.3 The interlinguistic influence factor

Unlike monolingual users of a world language, bilinguals represent information about (at least) two languages, each of which may influence processing in the other. Thus, performance in a world language that is represented as either a native or a non-native language may also vary depending on specific properties of the other language.

As a rule, native speakers of European languages tend to develop greater proficiency in English as an L2 than do Chinese learners. When matched for AoA, years of L2 instruction, and years spent in the United States, European bilinguals outperform Asian bilinguals in listening and reading tasks. Such differences in L2 attainment are likely related to the fact that, unlike Chinese, most European languages use alphabetic scripts, which would facilitate processing in other alphabetic-script languages, such as English (Bialystok and Miller 1999; Birdsong and Molis 2001; Jia et al. 2002; Jia 2006).

Moreover, the formal properties of a native language may influence the neurocognitive representation of a lately learned L2. Yang et al. (2011) examined whether the neural representation of lexical categories in a non-native language is influenced by that of a given native language. Strategically, the authors relied on fMRI to explore the neural distribution of nouns and verbs in late Chinese-English bilinguals. This language pair is well-suited to answer such a question. In monolingual speakers of English, nouns and verbs are differentially represented in left frontal/prefrontal regions and left middle/posterior temporal areas, respectively (Damasio and Tranel 1993; Martin et al. 2000; Shapiro et al. 2006). The neural dissociation between nouns and verbs also holds in non-native speakers of English whose native language also presents a morphological distinction between both classes (Hernández et al. 2007; Willms et al. 2011). However, no such dissociation occurs in Chinese monolinguals (Li et al. 2004), arguably because Chinese features no inflections signaling a formal distinction between such lexical classes (Kao 1990). A previous fMRI study with early Chinese-English bilinguals raised in Hong Kong found that each language presented the same profile observed in monolingual speakers (Chan et al. 2008). However, in their study with late bilinguals, Yang et al. (2011) found that nouns and verbs, in both Chinese (L1) and English (L2), were subserved by largely overlapping regions, although the network for English was more widely distributed than that involved in Chinese – which likely reflects more effortful processing in the former. The key finding was that, unlike what occurs in native English speakers and western learners of English as an L2, these subjects showed no neural dissociation between nouns and verbs. The authors suggest that “the lack of grammatical morphology and the high degree of noun-verb ambiguity in Chinese might lead to the speaker’s insensitivity to noun-verb differences”, so that late bilinguals “may be applying the neural mechanism for Chinese (L1) to the processing of English (L2) and hence also show no neural sensitivity to nouns versus verbs in English” (Yang et al. 2011, 680).

Basnight-Brown et al. (2007) offer further evidence that the level of formal similarity between a non-native world language and a native language influences processing in the former. Using a cross-modal priming paradigm, the authors compared magnitudes of facilitation between Serbian-English and Chinese-English bilinguals. The study focused on different types of English verbs: irregular nested-stem verbs (drawn-draw), irregular change-stem verbs (ran-run), and regular past tense/present tense verbs (guided-guide). While both groups showed significant facilitation for regular verbs and no facilitation for change-stem irregulars, only the Serbian group exhibited facilitation effects for nested irregulars. In light of these results, Basnight-Brown et al. (2007, 76–77) posit that “due to familiarity with a highly inflected language and with a written language that captures inflectional variation, L1 speakers of Serbian appear to be analytic with respect to form in their L2,” thus being “better able to transfer their sensitivity to word structure from the L1 to the L2 so as to exploit orthographic and phonological similarities between various inflected forms.” On the contrary, the authors claim, “the Chinese speakers tend to be less attuned to morphological relatedness overall and more rigid in their criterion for similarity so that it includes only those pairs that entail affixation of -ed.” More generally, these findings suggest that speakers of logographic languages are less consistent than speakers of alphabetic languages in their reliance on formal information during English (L2) processing.

Formal similarities between a native language and a non-native world language also influence lexical processing in the latter. This is particularly evident in the so-called cognate effect, that is, the empirical finding that cognates (i.e., words which are phonologically/orthographically similar to their translation equivalents) are processed faster than noncognates (de Groot and Nas 1991; Sáchez-Casas et al. 1992; Gollan et al. 1997; Kiran and Lebel 2007; Schwartz et al. 2007; Dimitropoulou et al. 2011).

Notably, this effect is sensitive to the degree of formal similarity between equivalents. In an English lexical decision task with Dutch-English bilinguals, Dijkstra et al. (2010) observed that cognate facilitation effects were proportional to the degree of orthographic overlap between equivalents. Reaction times decreased as formal similarity increased across counterparts (e.g., lamp-lamp < flood-vloed < song-lied). Even word pairs with only partial formal overlap (e.g., guide-gids, rhythm-ritme) elicited facilitation effects. However, the cognate effect was modulated by task demands, as results were completely opposite in Dutch-English language decision. In this task, Dijkstra et al. (2010) observed a cognate inhibition effect, whose size depended on the degree of orthographic overlap between equivalents (identical cognates were processed more slowly than non-identical cognates). It follows from these findings that the greater the degree of formal overlap between the English and the native-language lexicons, the greater facilitatory and inhibitory effects will tend to be in the former during specific language processes. Typologically related languages, then, are particularly susceptible to these crosslinguistic effects – as well as others, such as the effect of first language polysemy on L2 semantic processing (Elston-Güttler et al. 2005; Elston-Guttler and Williams 2008).

Although the cognate facilitation effect is generally larger in L2s than in L1s (Kroll et al. 1999), it also affects L1 processing. Direct evidence for this was obtained by García et al. (in preparation), who administered identical single-word translation tasks to three groups of Spanish-English bilinguals, and two groups of English-Spanish bilinguals. The cognate effect emerged in both translation conditions (backward translation, forward translation), irrespective of the participants’ LoP and language pair. This shows that lexical processing in English is modulated by the degree of formal similarity with another language, regardless of whether the former is a non-native or a native language.

Interlinguistic effects may even modulate brain activity in the absence of significant behavioral correlates. Thierry and Wu (2007) tested Chinese-English bilinguals via an implicit priming paradigm. Participants were asked to read and listen to pairs of English words so as to decide whether they were semantically related. Critically, the Chinese equivalents of some of the English pairs had a repeated character (e.g., the words train and ham are not semantically related, but their Chinese counterparts, Huo Che and Huo Tui, have a character in common). Since the tasks were performed entirely in English, the participants were unaware of this manipulation. Although this hidden factor did not affect behavioral performance, it did significantly modulate brain potentials by reducing the amplitude of the N400 component (an effect that was also observed in monolingual Chinese controls). Since this component is associated with unconscious semantic and repetition priming, the results demonstrate that the native language is unconsciously activated during second-language comprehension. They further indicate that implicit neurocognitive processing of a non-native world language is sensitive to covert, L1-specific structural features.

In conclusion, cognitive processing of, and behavioral performance in, a non-native world language is modulated by its level of formal similarity with a specific native language. Similarly, the appropriation of a second or foreign language may modulate both processing and performance in a native language. Hence, part of the performance variability among users of English as a world language depends on the idiosyncratic properties of the other language(s) they have appropriated, either as L1 or as L2. Remarkably, some of these effects may operate below the threshold of conscious awareness.