The UCM and Cross-Language Similarity

The idea that similarity between languages can influence L2 processing is not new and has been examined extensively in the past (e.g., Bates & MacWhinney, Reference Bates, MacWhinney, MacWhinney and Bates1989). This issue has been brought back into focus in current bilingualism research (e.g., Chen, Shu, Liu, Zhao, & Li, Reference Chen, Shu, Liu, Zhao and Li2007; Jing, Reference Jing2008; Sabourin & Stowe, Reference Sabourin and Stowe2008; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005), and the UCM posits general neurocognitive mechanisms to account for effects of similarity between a speaker’s two languages. This review focuses only on the specific predictions derived from the (morpho)syntactic domain of the UCM and evaluates these predictions in the context of neurocognitive data.

According to the UCM, adult L2 learners are highly subject to the influence of their L1 linguistic system (e.g., [morpho]syntax, phonology, etc.), which has been learned from an early age and used extensively throughout their lives and has therefore become highly entrenched in the bilingual’s brain. Cross-language similarity—along with L2 age of acquisition (AoA), learning environment, and so on—determines the outcome of this crosslinguistic interaction.

Therefore, the L1 (morpho)syntactic system (e.g., word order, agreement) will likely influence L2 comprehension and production through the transfer of processing strategies. Although general findings on crosslinguistic transfer have also been reported (e.g., Kellerman, Reference Kellerman1995), under the UCM, transfer processes can be investigated in terms of cross-language similarity. Specifically, when an adult learns a L2, two particular outcomes are likely to occur: transfer and competition. Positive transfer takes place when grammatical features are similarly instantiated in the L1 and L2, which thus allows the bilingual to easily and effectively apply the L1 system to the processing of the weaker L2. Negative transfer, in contrast, occurs when linguistic features are present in both the L1 and L2 but are instantiated differently in the L2, thus affecting L2 processing. Such a situation may give rise to online competition—which is likely to be experienced even by highly proficient bilinguals—and to subsequent ungrammatical utterances if the bilingual unsuccessfully uses the entrenched L1 system in the processing of the L2.

The outcome of this competition is ultimately determined not only by cross-language similarity but also by the strength of linguistic cues, which provide surface information about the underlying function of particular linguistic items (e.g., the German determiner der signals the function of masculine gender). Cue strength, in turn, depends on both cue availability and reliability: If the correspondence between form and function is highly available in the L2 input and if the cue is reliable in its prediction of function, cue strength will increase and linguistic items (e.g., words, sentences) associated with the strongest cues will have a processing advantage. Therefore, the UCM’s overarching assumption is that “whatever can transfer will” (MacWhinney, Reference MacWhinney, Kroll and de Groot2005, p. 55). Cross-language similarity between types of cues as well as their relative strengths in the input are seen as crucial factors in L2 acquisition that will determine, in part, how potentially conflicting cues in the two languages are resolved and thus influence L2 comprehension and production.

In this review, cross-language similarity is operationalized in the same manner as in Tokowicz and MacWhinney (Reference Tokowicz and MacWhinney2005), with a focus on similarity between specific (morpho)syntactic features on the basis of word-by-word translation. Within this framework, a given construction can be conceptualized as grammatically similar, dissimilar, or unique to a L2. These categories are exemplified through sentences in English and Spanish, as in (1)–(3).Footnote ¹

1. (1) Similar construction

   a. Este chicoes alto.

   “This boy is tall.”

   b. Estos chicosson altos.

   “These boys are tall.”

2. (2) Dissimilar construction

   a. Elchicoes alto.

   The-sing boy-sing is tall

   “The boy is tall.”

   b. Los chicosson altos.

   The-pl boys-pl are tall

   “The boys are tall.”

   c. *El chicosson altos.

   The-sing boys-pl are tall

3. (3) Unique construction

   a. El chicoes alto.

   The-masc boy-masc is tall

   “The boy is tall.”

   b. *La chicoes alto.

   The-fem boy-masc is tall

A construction is considered similar in the L1 and L2 if the (morpho)syntactic feature exists in both the L1 and L2 and is instantiated similarly in both languages. In this case, translation results in a legal sentence in the corresponding language while retaining its overall meaning. For example, as shown in (1), demonstrative determiner-noun number agreement is instantiated similarly in English and Spanish (which generally follows the same word order) because number is marked on both the demonstrative determiner and the noun (gender effects can be ignored because gender is held constant in the contrasting sentences). In contrast, a construction is considered dissimilar if the (morpho)syntactic feature is present in both the L1 and L2 but is instantiated differently in the L2, as illustrated by the example of definite determiner-noun number agreement in (2). Spanish marks number on the definite determiner and the noun, and so these must always agree in number. In contrast, in English, number is not marked on the definite determiner, and so number agreement is not required between the definite determiner and the noun. Therefore, in this case, word-by-word translation can result in an ungrammatical L2 sentence, as in (2c). Finally, a construction is considered unique to the L2 if it includes a (morpho)syntactic feature that is absent in the L1—for example, determiner-noun gender agreement, as shown by the examples in (3). The outcome in this case is less clear because the bilingual neither benefits from positive transfer nor suffers from competition and negative transfer. The UCM holds that the success of processing this type of construction depends on the strength of the L2 cue such that the bilingual will eventually succeed when the new feature has been acquired (e.g., Hernandez, Hofmann, & Kotz, Reference Hernandez, Hofmann and Kotz2007; Kotz, Holcomb, & Osterhout, Reference Kotz, Holcomb and Osterhout2008).

The SSH

One account that examines the specific mechanisms of L2 syntactic comprehension is the SSH (Clahsen & Felser, Reference Clahsen and Felser2006a, Reference Clahsen and Felser2006b). According to the SSH, adult L2 syntactic comprehension differs fundamentally from L1 comprehension in that L2 speakers build shallower and less detailed grammatical representations than adult native speakers. Furthermore, to compensate for this structure-building deficit, nonnative speakers overly rely on nonstructural, lexicosemantic, and pragmatic information to comprehend sentences in the L2. The SSH specifically claims that adults who learn a L2 after childhood are restricted to shallow online processing because they lack adequate L2 grammatical representations; that is, L2 speakers fail to construct the complex hierarchical structures that constitute the basis for the segmentation and organization of incoming discourse and instead rely on other sources of information to extract meaning. This is not the result of a lack of knowledge of L2 syntactic structures per se but is rather due to incomplete or just fundamentally different internal L2 grammatical representations. Thus, based on evidence from ambiguity resolution (e.g., relative clause attachment) and syntactic dependency processing (e.g., filler-gap dependencies), the SSH predicts that L2 speakers are likely to exhibit differences relative to their L1 (or to native speakers of the L2) during online L2 sentence processing. Clahsen and Felser make an important distinction between the processing of syntax and morphology such that shallow processing applies to syntactic but not necessarily to morphological processing, which rests primarily on local agreement rather than on complex phrase-structure building processes. Moreover, differences between L2 and native speakers of a language are more likely to be observed during online than offline processing because online tasks reduce the availability of explicit L2 grammatical knowledge.

It is relevant to note that the SSH predicts little transfer of L1 syntactic processes to the L2. Because bilinguals are restricted to shallow, underspecified syntactic processing in the L2, they cannot easily apply their sophisticated L1 parsing skills to the processing of L2 structures. Therefore, according to the SSH, transfer from the L1 is only possible indirectly: “Given the processing model […], we would expect L1 transfer to influence L2 processing only as a consequence of one or more of the knowledge sources that feed the processing system being affected by properties of the L1” (Clahsen & Felser, Reference Clahsen and Felser2006a, p. 118). How such transfer effects manifest themselves at the sentence processing level is not entirely clear.

Therefore, the two models discussed thus far make very different predictions with regard to L1 transfer effects. Unlike the UCM, the SSH predicts no (or few) transfer effects at the sentence level (e.g., phrase structure, word order); additionally, L2 speakers are predicted to always differ from native speakers during online sentence-level processing but not necessarily during morphological processing. Therefore, the SSH does not make specific predictions about cross-language similarity effects except that this should not be a relevant modulating factor in L2 processing. In contrast, the UCM predicts that if a grammatical feature is similarly instantiated in the L1 and L2, its neural signature will probably be more nativelike because it will benefit from positive transfer. For example, the Spanish sentence in (4a) would be correctly translated into English following the same subject-verb-object (SVO) order. However, the UCM predicts that the processing of a L2 sentence that follows non-SVO (i.e., dissimilar) word order, as in (4b), will often result in comprehension errors (e.g., “she kisses the boy”).Footnote ²

1. (4) a. El chico besa a la chica.

   “The boy kisses the girl.”

   b. La besa el chico.

   Her-obj kisses the boy-subj

   “The boy kisses her.”

The SSH, in contrast, predicts that the online processing of both the similar and dissimilar sentence types (assuming comparable structural complexity) would be equally underspecified by the bilingual and therefore nonnativelike, regardless of his or her L1.

Furthermore, for unique L2 features, the UCM predicts that if cues are widely available and highly reliable—as they are, for example, in Spanish noun-adjective gender agreement in sentences like those in (5)—it is likely that the bilingual will eventually acquire the unique feature and show nativelike neural signatures during processing.

1. (5) a. El chico es muy alto.

   The boy-masc is very tall-masc

   “The boy is very tall.”

   b. *El chico es muy alta.

   The boy-masc is very tall-fem

In contrast, the SSH predicts underspecified processing of nonlocal (i.e., across phrase boundaries) L2 syntactic features because these cannot be mapped onto fully parsed L1 structures, regardless of how crosslinguistically similar (or unique) the structures are.

Physiology-Based Methods: fMRI

Many of the findings from behavioral studies on L2 syntactic processing have been complemented by more recent research with neurocognitive methods. Brain imaging techniques allow the investigation of L2 processing based on physiological markers such as vascular and electrical activity in the brain. Specifically, fMRI involves the acquisition of the blood-oxygen-level–dependent signal, which reflects changes in magnetism caused by the amount of oxygen in the blood in a given area of the brain. fMRI is used as an index of the level of activity of particular brain areas as a result of carrying out specific cognitive tasks. Thus, for example, increased blood flow to Broca’s area (Brodmann areas 44 and 45 of the left inferior frontal gyrus) during a language task as compared to blood flow during rest (or a control task) provides support for the involvement of this particular brain area in language processing. fMRI typically has a spatial resolution of approximately 3 mm³, and experimental results are usually reported in terms of the number of activated voxels (three-dimensional volume elements) in a brain region above a specified statistical threshold as well as the percent blood-oxygen-level–dependent signal change from a baseline or control level.

Despite its relatively high precision in the localization of brain activity related to specific functions, the fMRI technique has some shortcomings—namely, it is a slow, indirect measure of brain activity because it relies on changes in blood flow and oxygenation, which themselves are associated with changes in neuronal signaling. Furthermore, there are issues of variability in the obtained signal. For example, considerable intersubject variability in the activation of brain areas during language tasks exists even among monolinguals (or during L1 processing), irrespective of effects related to gender or handedness differences. Such variability is greater in a L2 than a L1 (Dehaene et al., Reference Dehaene, Dupoux, Mehler, Cohen, Paulesu and Perani1997) and can affect the localization of language-related activation independently of manipulated variables (e.g., Springer et al., Reference Springer, Binder, Hammeke, Swanson, Frost and Bellgowan1999).

It is even more important to note that a large amount of within-subject variability also exists between separate fMRI scans that employ identical tasks and stimuli. As an illustration, Mahendra, Plante, Magloire, Milman, and Trouard (Reference Mahendra, Plante, Magloire, Milman and Trouard2003) had bilingual participants perform word fluency and sentence generation tasks in both their L1 and L2 in two identical but separate fMRI scans. Differences in the extent of activated brain areas while processing a L1 versus a L2 were comparable to those observed due to between-scan (within-subject) variability in each language and for either task. Thus, independently of manipulated variables, some degree of inconsistency in the localization of L1- and L2-related activation within and between experiments examined in this review is to be expected, especially because stimulus materials were necessarily different in the L2 as compared to the L1 and because different tasks were used in different experiments. Therefore, a lack of differences in activation between the L1 and L2 suggests processing that is highly similar crosslinguistically; however, the large amount of variability may limit the ability to detect differences.

Physiology-Based Methods: ERPs

In contrast, the ERP method involves measuring the electrical activity of large populations of neurons in response to internally or externally generated events through electrodes placed on the surface of the scalp. ERPs constitute segments of the continuous electroencephalographic record that are time-locked to the occurrence of such cognitive or motor events. Therefore, ERPs measure internal responses associated with processes like language comprehension, within tens of milliseconds. ERPs consist of specific components associated with a variety of motor and cognitive functions. Components are identified by characteristics such as polarity, latency, and scalp distribution. The components typically observed in (morpho)syntactic processing include the early left anterior negativity (ELAN), the left anterior negativity (LAN), and the P600, each considered an index of a specific aspect of language processing. The ELAN is a negative-going wave measured at left-hemisphere frontal areas of the scalp with a latency of approximately 150–250 ms and is considered to reflect highly automatized syntactic parsing (e.g., Hahne & Friederici, Reference Hahne and Friederici1999). The LAN is a slightly later left-hemisphere anterior negative-going wave with a latency of approximately 300–500 ms; its amplitude increases in response to (morpho)syntactic violations such as errors in subject-verb agreement (e.g., Ojima, Nakata, & Kakigi, Reference Ojima, Nakata and Kakigi2005; Rossi, Gugler, Friederici, & Hahne, Reference Rossi, Gugler, Friederici and Hahne2006). Although both the ELAN and the LAN have been observed during language processing in native speakers (e.g., Hahne, Reference Hahne2001; Hahne & Friederici; Ojima et al.; Rossi et al.), it is often the case that the ELAN is not found (e.g., Allen, Badecker, & Osterhout, Reference Allen, Badecker and Osterhout2003; Frenck-Mestre, Osterhout, & McLaughlin, Reference Frenck-Mestre, Osterhout and McLaughlin2008; Kim & Osterhout, Reference Kim and Osterhout2005; Tolentino & Tokowicz, Reference Tolentino and Tokowicz2010). These two components are observed even less often in L2 speakers (Ojima et al.; Rossi et al.).

The final component relevant to this review, the P600, is a positive-going wave that extends from approximately 500 to 800 ms poststimulus and is centroparietally distributed (Osterhout & Holcomb, Reference Osterhout and Holcomb1992). The P600 has been widely used in L1 and L2 syntactic processing paradigms as an index of brain responses to (morpho)syntactic violations (e.g., Frenck-Mestre et al., Reference Frenck-Mestre, Osterhout and McLaughlin2008; Osterhout & Holcomb; Sabourin & Stowe, Reference Sabourin and Stowe2008; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005).

The degree of temporal precision of ERPs stands in contrast to that of other brain imaging methods, such as fMRI, that rely on slower blood-flow differences. Nevertheless, ERPs, like fMRI, have certain limitations: Current evidence suggests that there is not a one-to-one correspondence between ERP components and underlying cognitive functions, an issue that also applies to fMRI data. For example, modulations of the amplitude of the P600 component can result not only from the processing of syntactic violations but also syntactic ambiguity (e.g., Kaan & Swaab, Reference Kaan and Swaab2003; Kotz et al., Reference Kotz, Holcomb and Osterhout2008), syntactic complexity (e.g., Kaan & Swaab), and music parsing (e.g., Patel, Gibson, Ratner, Besson, & Holcomb, Reference Patel, Gibson, Ratner, Besson and Holcomb1998). Thus, certain ERP components may reflect more general cognitive functions, such as sensitivity to hierarchically structured sequences in the case of the P600 (Patel, Reference Patel2003). Another limitation is that ERP data are typically obtained through multiple levels of averaging over several trials, electrodes, participants, and experimental conditions. Such averaging, if not carefully handled, can obscure relevant data or misshape component waveforms. Finally, ERPs suffer from spatial limitations; it is difficult to map the effects observed at the scalp to any precise neural source.

Such limitations highlight the importance of integrating findings from both physiology-based and behavioral methods, thus capitalizing on the strengths of each. Therefore, although the focus of the present review is on studies that employed physiology-based methods, relevant behavioral findings are also briefly discussed. It is interesting to note that results from behavioral tasks have often been at odds with those obtained with neuroimaging methods. This divergence between behavioral and physiology-based findings has assumed one of two forms: Some studies have reported neural activation indicative of brain sensitivity to grammatical violations or lexical status despite behavioral performance at chance levels (McLaughlin, Osterhout, & Kim, Reference McLaughlin, Osterhout and Kim2004; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005). In contrast, other studies have shown no neural activation indicative of brain sensitivity despite largely above-chance behavioral performance (Chen et al., Reference Chen, Shu, Liu, Zhao and Li2007; Hahne & Friederici, Reference Hahne and Friederici2001; Sabourin & Stowe, Reference Sabourin and Stowe2008).

A combined approach may shed some light on the controversy that surrounds theories of L2 acquisition (e.g., the critical period hypothesis vs. age-related decline), which have relied almost exclusively on behavioral measurements to support their conclusions, despite evidence to suggest that physiology-based methods sometimes demonstrate higher sensitivity in L2 processing (e.g., McLaughlin et al., Reference McLaughlin, Osterhout and Kim2004; Thierry & Wu, Reference Thierry and Wu2007; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005).

Neural Commonalities in L1 and L2 Processing

Findings from an increasing body of literature generally converge on the notion that the neural mechanisms and structures that subserve L2 processing are largely the same as those that underlie L1 processing (see Abutalebi, Reference Abutalebi2008; Stowe & Sabourin, Reference Stowe and Sabourin2005, for recent reviews). Imaging studies have revealed that these structures are primarily located in the left hemisphere of the brain and comprise a network of mainly frontal and temporal areas. The left inferior frontal gyrus (which comprises Broca’s area) and the left temporal lobe (superior and middle gyri that comprise Wernicke’s area) are consistently activated during both (morpho)syntactic comprehension and production (e.g., Hasegawa, Carpenter, & Just, Reference Hasegawa, Carpenter and Just2002; Suh et al., Reference Suh, Hyo, Lee, Chung, Cho and Park2007; Tatsuno & Sakai, Reference Tatsuno and Sakai2005; Vingerhoets et al., Reference Vingerhoets, Borsel, Tesink, Noort, Deblaere and Seurinck2003; Wattendorf et al., Reference Wattendorf, Westermann, Zappatore, Franceschini, Lüdi and Radü2001). Additional areas such as the visual word form area (in the occipito-temporal cortex) have been implicated in specific functions such as reading (e.g., McCandliss, Cohen, & Dehaene, Reference McCandliss, Cohen and Dehaene2003).

Results from ERP studies that examined L1 and L2 processing have yielded a comparable picture to the results from fMRI studies in that several commonalities exist between L1 and L2 processing. In particular, ERP markers of specific linguistic processes are often present in both native and nonnative speakers of a variety of languages (see Mueller, Reference Mueller2005, for a recent review). For example, nonnative speakers often exhibit P600 effects consistent with sensitivity to (morpho)syntactic violations in the L2 (e.g., Frenck-Mestre et al., Reference Frenck-Mestre, Osterhout and McLaughlin2008; Hahne, Reference Hahne2001; Kotz et al., Reference Kotz, Holcomb and Osterhout2008; Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006; Sabourin & Stowe, Reference Sabourin and Stowe2008; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005; Weber-Fox & Neville, Reference Weber-Fox and Neville1996). Furthermore, L2 speakers sometimes also show particular ERP effects that allegedly reflect highly automatic (morpho)syntactic processing such as the ELAN (e.g., Hahne & Friederici, Reference Hahne and Friederici1999) and the LAN (e.g., Rossi et al.; Weber-Fox & Neville). The presence of specific ERP component effects—in particular, the ELAN and LAN—appears to depend on a variety of factors, including the type of syntactic structure, the parser’s expectations, proficiency level, and cross-language similarity (see van Hell & Tokowicz, Reference van Hell and Tokowicz2010, for a review).

Neural Differences in L1 and L2 Processing

Despite the many commonalities between L1 and L2 processing—and between bilinguals and native speakers—differences are nevertheless often observed. These differences have arisen mainly as a function of proficiency, L2 AoA, and, to a lesser-known extent, cross-language similarity (e.g., Chen et al., Reference Chen, Shu, Liu, Zhao and Li2007; Jeong, Sugiura, Sassa, Haji, et al., Reference Jeong, Sugiura, Sassa, Haji, Usui and Taira2007; Sabourin & Stowe, Reference Sabourin and Stowe2008; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005). Therefore, L2 processing becomes more nativelike and increasingly similar to L1 processing when bilinguals have attained a high level of L2 proficiency (e.g., Ojima et al., Reference Ojima, Nakata and Kakigi2005; Perani et al., Reference Perani, Paulesu, Sebastián-Gallés, Dupoux, Dehaene and Bettinardi1998; Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006). Additionally, L2 (morpho)syntactic (and phonological) processing is strongly affected by L2 AoA, with poorer performance (Birdsong, Reference Birdsong, Kroll and de Groot2005; Johnson & Newport, Reference Johnson and Newport1989) and nonnativelike brain patterns (e.g., Hernandez et al., Reference Hernandez, Hofmann and Kotz2007; Wartenburger et al., Reference Wartenburger, Heekeren, Abutalebi, Cappa, Villringer and Perani2003; Weber-Fox & Neville, Reference Weber-Fox and Neville1996) being characteristic of later AoAs.

In fact, brain activation differences between L1 and L2 processing are often found using fMRI (as measured by blood-oxygen-level–dependent signal changes) and ERPs (as measured by component changes in latency or amplitude, or both). Of 11 imaging studies that examined L2 (morpho)syntactic comprehension, 9 reported increased L2 activation (mainly in left inferior frontal gyrus and left temporal areas) compared with the L1 or native speakers (Hasegawa et al., Reference Hasegawa, Carpenter and Just2002; Hernandez et al., Reference Hernandez, Hofmann and Kotz2007; Jeong, Sugiura, Sassa, Haji, et al., Reference Jeong, Sugiura, Sassa, Haji, Usui and Taira2007; Jeong, Sugiura, Sassa, Yokoyama, et al., Reference Jeong, Sugiura, Sassa, Yokoyama, Horie and Sato2007; Rüschemeyer, Fiebach, Kempe, & Friederici, Reference Rüschemeyer, Fiebach, Kempe and Friederici2005; Rüschemeyer, Zysset, & Friederici, Reference Rüschemeyer, Zysset and Friederici2006; Vingerhoets et al., Reference Vingerhoets, Borsel, Tesink, Noort, Deblaere and Seurinck2003; Wartenburger et al., Reference Wartenburger, Heekeren, Abutalebi, Cappa, Villringer and Perani2003; Yokoyama et al., Reference Yokoyama, Okamoto, Miyamoto, Yoshimoto, Kim and Iwata2006). The most common explanation for the increased activation associated with processing a L2 is that it reflects greater effort or difficulty—and therefore increased processing demands (e.g., Hasegawa et al.; Hernandez et al.; Indefrey, Reference Indefrey2006; Leikin, Reference Leikin2008; Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006; Rüschemeyer et al., Reference Rüschemeyer, Fiebach, Kempe and Friederici2005; Stowe & Sabourin, Reference Stowe and Sabourin2005; Vingerhoets et al.; Yokoyama et al.). Indeed, results from fMRI studies by Hasegawa et al. (Reference Hasegawa, Carpenter and Just2002) and Yokoyama et al. (Reference Yokoyama, Okamoto, Miyamoto, Yoshimoto, Kim and Iwata2006) suggested that increased L2 activation reflects greater L2 processing demands because, in these studies, increasing stimulus complexity resulted in increased L2 activation, mainly in left temporal and fusiform areas (see also Just, Carpenter, Keller, Eddy, & Thulborn, Reference Just, Carpenter, Keller, Eddy and Thulborn1996, for similar L1 results).

It is interesting to note that all of the studies that reported increased L2 activation tested late AoA (a range of 6–20 years of age) participants who had achieved moderate to high levels of proficiency (however, measures of proficiency varied widely in these studies). Nevertheless, the only study that directly contrasted different AoAs and proficiency levels was that of Wartenburger et al. (Reference Wartenburger, Heekeren, Abutalebi, Cappa, Villringer and Perani2003), which reported greater L2 than L1 activation in bilateral inferior frontal gyrus (and subcortical structures) in response to agreement violations only in late AoA participants, regardless of proficiency level. Therefore, it appears that increased L2 activation during syntactic processing as compared to L1 is mainly associated with task or stimulus difficulty, or late AoA, or both, and may reflect increased L2 processing effort (but see Perani et al., Reference Perani, Paulesu, Sebastián-Gallés, Dupoux, Dehaene and Bettinardi1998). It may also be that task difficulty and AoA interact with other factors that also increase L2 activation compared to L1, such as the degree of cross-language similarity. Thus, more carefully controlled studies as well as comprehensive reviews of the experimental literature are needed to determine the specific ways in which cross-language similarity may interact with AoA and proficiency level to influence L2 processing.

fMRI Cross-Language Similarity Findings

Results from the restricted set of fMRI studies that directly investigated cross-language similarity suggest an important role of this variable on the neural activation patterns of moderate and higher proficiency bilinguals during L2 (morpho)syntactic processing (see Table 1).

Functional magnetic resonance imaging findings from late AoA, moderate to higher proficiency bilinguals showed increased brain activation relative to the L1 (or to native speakers) during the processing of L2 syntactic features that were dissimilar to the L1 or unique to the L2. Such results point to differences in L1 and L2 brain activation when aspects of (morpho)syntax in the L1 differ markedly from the L2, even in the absence of behavioral differences.

Jeong, Sugiura, Sassa, Haji, et al. (Reference Jeong, Sugiura, Sassa, Haji, Usui and Taira2007) and Jeong, Sugiura, Sassa, Yokoyama, et al. (Reference Jeong, Sugiura, Sassa, Yokoyama, Horie and Sato2007) examined the effect of dissimilar word order employed in English and Mandarin Chinese (canonical SVO), as opposed to Korean and Japanese (SOV). In these two studies, Jeong and colleagues tested L1 Mandarin Chinese and Korean speakers with similar (moderate) levels of L2 English and Japanese proficiency. Despite comparable behavioral performance on an auditory sentence comprehension task in all languages, differences in brain activation as a function of cross-language similarity were observed. Example sentences used in the studies are provided in (6).

1. (6) a. English

   John met the teacher at school.

   b. Korean

   Chelswu-nun hakkyo-eyse sensayngnimkwa manna-ss-ta.

   Chelswu-topic school-at teacher-with meet-past

   c. Japanese

   Tarou-wa gakkou-de sensei-to a-tta.

   Tarou-topic school-at teacher-with meet-past

   d. Mandarin

   Zhãngsãn zài xuéxiào jiàn-le laoshi.

   Zhãngsãn at-school meet-perfect teacher

In Jeong, Sugiura, Sassa, Haji, et al.’s (Reference Jeong, Sugiura, Sassa, Haji, Usui and Taira2007) study, Korean L1 speakers displayed increased activation during L2 English sentence processing relative to the L1, mainly in left inferior frontal gyrus and temporal areas. In contrast, their patterns of activation during the processing of L2 Japanese, which relies on similar word order parameters as the L1, were similar to those displayed in L1 processing; differences in script between Korean and English can be safely ignored because the sentences were presented auditorily. Analogous results were obtained with L1 Mandarin trilinguals who spoke English and Japanese as L2s (Jeong, Sugiura, Sassa, Yokoyama, et al., Reference Jeong, Sugiura, Sassa, Yokoyama, Horie and Sato2007). Nevertheless, it is unclear (a) to what extent such findings would extend to a different task, such as a grammaticality judgment task (GJT), and (b) what the influence of possible confounding factors, such as systematic differences in prosody between the various languages, would be.

In terms of unique L2 features, Hernandez et al. (Reference Hernandez, Hofmann and Kotz2007) reported on a gender decision task in Spanish that consisted of irregular items (ending with the letters e, l, n, r, s, t, or z; e.g., papel “paper”) as well as regular items (ending in o for masculine and a for feminine; e.g., pueblo “town” vs. casa “house”). L1 English speakers proficient in L2 Spanish displayed a greater increase in left inferior frontal gyrus activation in response to the gender-irregular items compared to gender-regular items, relative to native speakers. Once again, these results were obtained despite comparable behavioral performance between L2 and native speakers.

Thus, findings from fMRI studies that examined the role of cross-language similarity in higher proficiency bilinguals using both sentence comprehension (Jeong, Sugiura, Sassa, Haji, et al., Reference Jeong, Sugiura, Sassa, Haji, Usui and Taira2007; Jeong, Sugiura, Sassa, Yokoyama, et al., Reference Jeong, Sugiura, Sassa, Yokoyama, Horie and Sato2007) and gender decision tasks (Hernandez et al., Reference Hernandez, Hofmann and Kotz2007) suggest an influence of cross-language similarity because differences in processing were found mainly when linguistic structures were crosslinguistically dissimilar or unique to the L2, in line with UCM—but not necessarily SSH—predictions. Moreover, the results of Hernandez et al. (Reference Hernandez, Hofmann and Kotz2007) suggest that the effects of L2 stimulus or task difficulty may interact with cross-language similarity. The increased activation observed for gender-irregular as compared to gender-regular items suggests that although L2 gender agreement cues associated with gender-regular items were sufficiently reliable as to allow for the learning of a morphological feature unique to the L2, the processing of gender-irregular items was nevertheless associated with increased processing demands (to a greater extent than in native speakers). The SSH, on the other hand, would predict no cross-language similarity effects during online L2 processing.

Because only a few studies have directly examined cross-language similarity, it can be informative to also examine results obtained in other studies that employed different languages, even if cross-language similarity was not directly controlled or investigated as it was in the main set of studies; cross-language similarity of the linguistic structures was approximated according to Tokowicz and MacWhinney’s (Reference Tokowicz and MacWhinney2005) classification scheme. Findings from this extended set of fMRI studies generally agree with the main set of results: Differences in brain activation were found during the processing of unique (Vingerhoets et al., Reference Vingerhoets, Borsel, Tesink, Noort, Deblaere and Seurinck2003) and dissimilar features in late AoA participants (Hasegawa et al., Reference Hasegawa, Carpenter and Just2002; Wartenburger et al., Reference Wartenburger, Heekeren, Abutalebi, Cappa, Villringer and Perani2003; Yokoyama et al., Reference Yokoyama, Okamoto, Miyamoto, Yoshimoto, Kim and Iwata2006; it should be noted that no similar features were tested in these studies) and therefore are in line with predictions of the UCM for (morpho)syntactic processing within a neurocognitive framework. Early bilinguals, on the other hand, showed no differences or effects of similarity (Wartenburger et al.), in line with SSH predictions of no transfer effects in general.

ERP Cross-Language Similarity Findings

The ERP studies that directly examined the effect of cross-language similarity (see Table 2) covered a wider range of AoAs and proficiency levels than the fMRI studies. The findings indicate that the ERPs of participants with early and late AoAs as well as of participants of various levels of proficiency show differences relative to L1 or native speakers, often as a function of cross-language similarity. The results for moderate and higher proficiency bilinguals are described first, followed by the results for lower proficiency bilinguals.

Moderate and higher proficiency bilinguals. De Diego Balaguer, Sebastián-Gallés, Díaz, and Rodríguez-Fornells (Reference De Diego Balaguer, Sebastián-Gallés, Díaz and Rodríguez-Fornells2005) and Kotz et al. (Reference Kotz, Holcomb and Osterhout2008) examined early, higher proficiency bilinguals and found some differences in L2 processing compared to that of native speakers. De Diego Balaguer et al. reported comparable priming effects during a primed lexical decision task that examined Spanish verb inflection in L1 Spanish-L2 Catalan and L1 Catalan-L2 Spanish bilinguals for regular verbs (i.e., with predictable inflectional suffixes; e.g., ganar “to win,” gano “I win”) but found differential priming effects for irregular verbs (i.e., with deviant inflectional patterns due to stem or root changes; e.g., sentir “to feel,” siento “I feel”). De Diego Balaguer et al. examined the N400 component because although typically associated with semantic processing, the N400 can also be elicited by violations of irregular verb inflection. Results suggested an influence of the differences in inflectional morphology between Catalan and Spanish for irregular verbs only: Irregular verb processing in L2 speakers compared to native speakers was associated with distributional differences in the topography of the N400 priming effect. Regular verb processing, in contrast, was associated with a comparable centroparietal N400 priming effect in both L2 and native speakers.

In contrast, results from Kotz et al. (Reference Kotz, Holcomb and Osterhout2008) showed a comparable P600 effect in native and L2 speakers of English (with Spanish as the L1) during the processing of temporary syntactic ambiguities (based on transitive-intransitive verb use; e.g., The broker persuaded to sell the stock was sent to jail), which is a syntactic feature unique to English. This effect was not found during the processing of phrase structure violations (e.g., *The broker hoped to sell the stock was sent to jail), which follows similar rules in both languages. Whereas both groups exhibited a P600 effect in response to (crosslinguistically similar) phrase structure violations in a GJT, the P600 was right-lateralized in native speakers but left-lateralized in L2 speakers. Nevertheless, Kotz et al. interpreted these differences with caution and suggested that a right-hemisphere P600 effect in L2 speakers may have been attenuated by a preceding negativity observed over posterior sites, thus eliciting the observed distributional differences between the two groups.

Event-related potential differences in the processing of dissimilar and unique features were also observed in studies that examined cross-language similarity in late AoA participants. Chen et al. (Reference Chen, Shu, Liu, Zhao and Li2007) compared the performance of L1 Mandarin Chinese speakers of moderate to higher L2 proficiency to that of native-speaker controls. In particular, they processed sentences with English number agreement violations (e.g., *The price of the cars were too high; *The price of the car were too high), a feature that does not exist in Mandarin. The L2 participants exhibited different ERP patterns in response to these violations during a GJT as compared to native-speaker controls: Whereas native English speakers showed a LAN and a subsequent P600 effect to ungrammatical sentences, Mandarin speakers did not. Similarly, Sabourin and Stowe (Reference Sabourin and Stowe2008) found that whereas L1 Dutch and German speakers exhibited a P600 effect in response to Dutch gender agreement violations, as in (7), L1 Romance language speakers failed to show this effect, despite high L2 Dutch proficiency. This result stands in contrast to the finding in the same study of a P600 effect in all groups in response to Dutch verbal domain dependency violations (i.e., main verb inflection) as in (8), a feature of all three groups of languages.Footnote ³

1. (7) Dutch gender agreement violation

   *De kleine kind probeerde voor het eerst te lopen.

   The-com small child-neut tried to walk for the first time

2. (8) Verbal domain dependency

   a. Dutch

   Ik heb in Groningen gewoond.

   I have in Groningen lived-past.part

   b. German

   Ich habe in Groningen gewohnt.

   c. French

   J’ai habitéen Groningen.

   “I have lived in Groningen.”

The findings are attributed to the influence of cross-language similarity such that although both verbal domain dependencies and gender agreement are characteristics of all languages examined, gender agreement is implemented differently in Romance languages. Thus, the L1 German speakers could apply a direct conversion algorithm for equating masculine and feminine German words with the common gender in Dutch (with neuter remaining the same in the two languages); however, the Romance language speakers could not rely on any direct mapping and thus had to map a new gender to each specific lexeme. It therefore appears that the presence of a given morphosyntactic feature or system in a L1 does not constitute a sufficient basis for positive L1-L2 transfer; rather, the system needs to be implemented in a comparable—or at least easily predictable—manner for transfer to be beneficial (see Costa, Kovacic, Frank, and Caramazza, Reference Costa, Kovacic, Frank and Caramazza2003, for behavioral results on the functional independence of grammatical gender systems).

Overall, like results from fMRI studies, ERP findings for higher proficiency bilinguals suggest an influence of cross-language similarity in neural activation during L2 (morpho)syntactic processing. Such results were obtained in the context of priming (De Diego Balaguer et al., Reference De Diego Balaguer, Sebastián-Gallés, Díaz and Rodríguez-Fornells2005) and the GJT (Chen et al., Reference Chen, Shu, Liu, Zhao and Li2007; Kotz et al., Reference Kotz, Holcomb and Osterhout2008; Sabourin & Stowe, Reference Sabourin and Stowe2008). Although the distributional differences in the P600 observed for the phrase structure violations in the Kotz et al. (Reference Kotz, Holcomb and Osterhout2008) study seem to contradict the UCM prediction of shared processing of crosslinguistically similar features (and to be more in line with the SSH prediction of no L1-L2 transfer), this particular result should be interpreted with caution.

It is important to note that the N400 effect observed in the De Diego Balaguer et al. (Reference De Diego Balaguer, Sebastián-Gallés, Díaz and Rodríguez-Fornells2005) study has not traditionally been used to assess sensitivity to (morpho)syntactic violations, thus making this particular result difficult to compare across studies. Nevertheless, just as specific ERP components may index more than one particular cognitive function, a given cognitive function may be indexed by more than one ERP component, which may vary with the task employed. For example, at least one study has indicated a longitudinal shift from an N400 to a P600 for L2 morphosyntactic violations (Osterhout, McLaughlin, Kim, Greenwald, & Inoue, Reference Osterhout, McLaughlin, Kim, Greenwald, Inoue, Carreiras and Clifton2004).

Lower proficiency bilinguals. Tokowicz and MacWhinney (Reference Tokowicz and MacWhinney2005) also examined the effect of cross-language similarity in late learners; however, they investigated L1 English speakers who were lower proficiency bilinguals of Spanish (in the first four semesters of university study). Participants exhibited a P600 effect in response to grammatical violations similar in the L1 and L2, such as errors in aspect marking as in (9a), and those unique to the L2, such as grammatical gender agreement as in (9b). However, this effect was not found in response to grammatical violations of constructions that exist in both the L1 and L2 but are instantiated somewhat differently in the L1 and L2, as with determiner-number agreement in (9c).

1. (9) a. *Su abuela cocinando muy bien.

   “*His grandmother cooking very well.”

   b. *Ellos fueron a un fiesta.

   They went to a-masc party-fem

   “They went to a party.”

   c. *El niños están jugando.

   The-sing boys-pl are playing.

   “The boys are playing.”

In a similar study, Tolentino and Tokowicz (Reference Tolentino and Tokowicz2010) also reported a P600 effect in response to violations of crosslinguistically similar constructions (demonstrative determiner-number agreement; e.g., *Este niños están jugando “*This boys are playing”) in a comparable participant population, but, in contrast to Tokowicz and MacWhinney (Reference Tokowicz and MacWhinney2005), they did not observe a P600 effect during the processing of unique constructions (grammatical gender agreement). Instead, a P600 effect was obtained for dissimilar constructions (definite determiner number agreement). It is likely that differences in stimulus set and experimental design (number or type of grammatical constructions, or both; presence of accuracy feedback; etc.) were responsible for the discrepancies between these two studies. Although these two studies directly contrasted cross-language similarity, they did not test native Spanish speaker controls with no knowledge of English, which thus prevented a direct comparison between native and L2 (morpho)syntactic processing; native Spanish speakers with no knowledge of English would be the appropriate comparison group because transfer from the L2 to the L1 is also possible (e.g., Dussias & Sagarra, Reference Dussias and Sagarra2007). (Nevertheless, both studies measured ERPs in response to violations during L1 processing.)

Both Kotz et al. (Reference Kotz, Holcomb and Osterhout2008) and Tokowicz and MacWhinney (Reference Tokowicz and MacWhinney2005) found that participants were sensitive to violations of morphosyntactic features that are unique to the L2, according to ERP measures. This finding is interesting because although the pair of languages examined in the two studies was identical (English and Spanish), the participant population differed in terms of level of proficiency, AoA, and L1. This would seem to indicate that both lower and higher proficiency as well as early and late AoA participants are sensitive to morphosyntactic features unique to the L2. However, the findings from Tolentino and Tokowicz (Reference Tolentino and Tokowicz2010) suggest that this may not always be the case, as indicated by the lack of a P600 effect in response to unique L2 features in that study. Furthermore, additional counterevidence comes from Chen et al. (Reference Chen, Shu, Liu, Zhao and Li2007), a study in which—unlike native English speakers—L1 Mandarin speakers failed to show a P600 in response to number agreement violations in L2 English (e.g., *The price of the cars were too high; *The price of the car were too high). It is unlikely that this result is due to either AoA or proficiency level because participants in this study performed at approximately 88% accuracy on a GJT despite being late AoA participants. One possibility is that although prevalent in English, specific cues associated with number agreement may not be as salient in English as gender agreement is in Spanish, thus making it more difficult for bilinguals to exhibit online sensitivity to this particular feature. Support for this idea comes from additional findings in Chen et al.’s (Reference Chen, Shu, Liu, Zhao and Li2007) experiment, which suggest that Mandarin speakers did not consistently apply number agreement cues in English, at least during online processing: In contrast to the LAN-P600 pattern exhibited by native English speakers in response to ungrammatical sentences, L1 Mandarin speakers showed an N400 followed by a P600 to sentences that were grammatical but number-incongruent based on surface features (e.g., The price of the cars wastoo high). This result suggests that bilinguals were sensitive to number cues in the L2 but perhaps needed stronger cues (or more targeted practice) to exhibit nativelike online sensitivity to this nonnative linguistic system.

In sum, for (morpho)syntactic features that are similar in the L1 and L2, results from these ERP studies are largely consistent and in accordance with predictions of positive transfer made by the UCM. Thus, all of the studies that examined the processing of shared morphosyntactic features between the L1 and L2 found that participants exhibited physiological markers of sensitivity to features as varied as regular verb inflection (De Diego Balaguer et al., Reference De Diego Balaguer, Sebastián-Gallés, Díaz and Rodríguez-Fornells2005), violations of phrase structure (Kotz et al., Reference Kotz, Holcomb and Osterhout2008), verbal domain dependencies (Sabourin & Stowe, Reference Sabourin and Stowe2008), aspect marking (Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005), and determiner number agreement (Tolentino & Tokowicz, Reference Tolentino and Tokowicz2010). In contrast, morphosyntactic features that differed in their implementation across languages were generally associated with differences in neural signatures during L2 and L1 processing (or compared with native speakers) or sometimes failed to elicit ERP markers of sensitivity in bilinguals. This was the case for irregular verb inflection (De Diego Balaguer et al.), determiner number agreement (Tokowicz & MacWhinney), and gender agreement (Sabourin & Stowe). These effects of cross-language similarity were obtained despite varying AoAs and L2 proficiency levels. One exception is the presence of a P600 effect in response to dissimilar constructions—although smaller than that for similar constructions—found by Tolentino and Tokowicz (Reference Tolentino and Tokowicz2010). It is possible that the presence of accuracy feedback and the use of two types of number violations, which were distinct features of this study, resulted in a differential increase in cue strength associated with determiner number agreement.

Such comparable processing in a L2 and L1 (or in native speakers) in response to crosslinguistically similar constructions is also largely in accordance with SSH predictions, given that the majority of the studies examined local violations of morphosyntax. However, the results of Kotz et al. (Reference Kotz, Holcomb and Osterhout2008) constitute an exception because they do not support the SSH prediction of differences between L2 and native speakers during the processing of phrase structure violations.

In the case of unique constructions, results were mixed in that whereas two studies (Kotz et al., Reference Kotz, Holcomb and Osterhout2008; Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005) found bilinguals to be sensitive to violations of this type of construction, others did not (Chen at al., Reference Chen, Shu, Liu, Zhao and Li2007; Tolentino & Tokowicz, Reference Tolentino and Tokowicz2010). The lack of physiological markers of sensitivity observed in the Chen et al. study in response to number violations and local incongruency could be attributed either to a lack of positive L1 transfer or to a restriction on agreement processing to local domains (e.g., *The price of the cars weretoo high), as predicted by the SSH. However, because participants also did not show a nativelike pattern (or even sensitivity) in response to what would be perceived as local violations (*The price of the car weretoo high), it is not clear that such results support the SSH.

Results from an extended set of ERP studies that did not directly examine the influence of cross-language similarity generally agree with the main set of findings in that the processing of crosslinguistically similar constructions by late, moderate to higher proficiency bilinguals were generally associated with comparable neural patterns in the L2 and L1 (and in native speakers; Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006). In contrast, the processing of dissimilar constructions was sometimes associated with comparable neural patterns in the L1 and L2 (Isel, Reference Isel2007; see also Isel, Hahne, Maess, & Friederici, Reference Isel, Hahne, Maess and Friederici2007)—and sometimes not (Frenck-Mestre et al., Reference Frenck-Mestre, Osterhout and McLaughlin2008). Furthermore, the processing of unique constructions was generally associated with distinct neural patterns (Hahne & Friederici, Reference Hahne and Friederici2001; Ojima et al., Reference Ojima, Nakata and Kakigi2005).

These results are largely consistent with UCM predictions; effective resolution of L1-L2 competition and the acquisition of a unique L2 feature in part depended on the strength of the cues associated with it (among other factors). For example, L1 German-L2 French bilinguals tested by Frenck-Mestre et al. (Reference Frenck-Mestre, Osterhout and McLaughlin2008) showed a P600 effect similar to that of native speakers during silent reading of phonologically realized inflectional morphology (a dissimilar feature). In contrast, the bilinguals did not show this sensitivity to agreement violations in which the errors were silent, thus constituting weaker cues. Cross-language similarity did not influence lower proficiency bilinguals (Osterhout, McLaughlin, Pitkänen, Frenck-Mestre, & Molinaro, Reference Osterhout, McLaughlin, Pitkänen, Frenck-Mestre and Molinaro2006; Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006; Weber-Fox & Neville, Reference Weber-Fox and Neville1996), although caution is required in the interpretation of these results because similarity was not explicitly contrasted in these studies.

Results from a few studies in this extended set are worth highlighting because they provide experimental evidence that can be used to evaluate theories of L2 syntactic processing and cross-language similarity effects. In particular, Rossi et al. (Reference Rossi, Gugler, Friederici and Hahne2006) reported that moderately proficient late L1 Italian-L2 German and L1 German-L2 Italian bilinguals exhibited an ELAN effect in response to word category violations that were crosslinguistically similar, as in (10a) and (10b).

1. (10) a. German:

   *Der Junge im singtein Lied

   The boy in-the sings a song

   b. Italian:

   *Il signore nel beveun caffè

   The man in-the drinks a coffee

Therefore, like native speakers, L2 speakers exhibited early, automatic sensitivity to phrase structure violations (see also Isel, Reference Isel2007). Relevantly, the purported absence of ELAN or LAN effects in response to syntactic violations is often offered as evidence in support of the SSH, according to which the absence of early syntactic component effects in L2 speakers reflects the fact that they do not use structural information at the same level of detail as native speakers (Clahsen & Felser, Reference Clahsen and Felser2006a, Reference Clahsen and Felser2006b). These results suggest that this may not always be the case, even for moderately proficient bilinguals.

Finally, it is interesting to note that in addition to the main qualitative differences (i.e., presence and distribution of an ERP component, or both) found between groups by Sabourin and Stowe (Reference Sabourin and Stowe2008), quantitative differences (i.e., changes in amplitude or latency of an ERP component, or both) were also observed. Thus, whereas all groups exhibited a P600 effect in response to verbal domain dependency violations, the effect was smaller in amplitude and showed delayed peak latencies in the L2. Furthermore, the P600 effect exhibited by L1 German speakers in response to gender agreement violations was also smaller and delayed relative to that of native Dutch speakers. This grading of responses within and between groups is suggestive of a possible continuum of L2 grammatical sensitivity (see also Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006; Tolentino & Tokowicz, Reference Tolentino and Tokowicz2010). It could be that the display of an identical P600 effect to native speakers indicates nativelike sensitivity; the presence of a delayed or smaller effect is indicative of near-native sensitivity; and the absence of an effect indicates nonnative sensitivity. However, this is not to be equated with degree of nativeness in behavioral performance. As some of the studies reviewed here have shown, accurate performance can occur in the absence of nativelike neural activation, and neural activation can exist even when overt performance is at chance.

It could be that higher proficiency bilinguals who perform at behavioral levels comparable to native speakers—but who do not display the same indicators of neural sensitivity—may rely on different L2 processing mechanisms that yield the same behavioral output. This concept is referred to as degeneracy, which occurs when the same function or outcome can be achieved through different structural constituents (Green, Crinion, & Price, Reference Green, Crinion and Price2006). This is also in accordance with the idea that functional anatomy prevails over structural anatomy. Hagoort (Reference Hagoort2006) argued that the pattern of input to a specific brain area and its connectivity to other areas, rather than the cytoarchitecture of the tissue, is what determines its cognitive function. It is reasonable to assume that, in most cases, the input pattern of L2 speakers varies considerably from that of native speakers (or from L1 input patterns), mainly due to specific learning settings (e.g., classroom vs. home) and, often, to AoA. This difference in the input could lead to alternate neural patterns in language representation and processing and possibly result in degenerate systems, at least in higher proficiency bilinguals. Cross-language similarity (or the absence of competition) may nevertheless encourage reliance on similar neural generators during processing.

In contrast, the case in which ERP sensitivity is observed despite lower levels of overt performance (Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005) could reflect reliance on the same neural mechanisms but at a level below behavioral threshold. In other words, lower proficiency bilinguals may be implicitly sensitive to (morpho)syntactic violations in the L2 but may not be able to access or report their knowledge during a grammaticality judgment or similar task. As noted by Ellis (Reference Ellis2005), bilinguals often have difficulties with L2 functional morphology, for example, because they continue to implicitly rely on L1 processes instead of adopting a more conscious approach necessary for noticing and acquiring nonsalient aspects of the L2. Consistent with this idea, Tolentino and Tokowicz (Reference Tolentino and Tokowicz2010) aimed to improve bilinguals’ ability to rely on their implicit knowledge in making end-of-sentence grammaticality judgments and observed that isolating grammatical violations and providing accuracy feedback were associated with increased brain sensitivity, as measured by ERPs, even for the processing of constructions unique to the L2.

Nevertheless, the possibility remains that results obtained with techniques that are based on physiological markers of sensitivity differ from those obtained with behavioral methods simply because the two types of methods measure L2 processing at different stages. Therefore, whereas fMRI and ERPs continuously track physiological responses (albeit with different temporal resolutions), the most common behavioral methods (e.g., GJT, comprehension questions) rely on global, post hoc judgments that likely involve metalinguistic knowledge.