ERPs

Applied Linguistics 2014: 35/4: 393–417 ß Oxford University Press 2014 doi:10.1093/applin/amu028 Advance Access published on 26 June 2014

1,2

KARSTEN STEINHAUER

1

School of Communication Sciences and Disorders; McGill University, Montreal, Canada and 2Centre for Research on Brain, Language and Music, Montreal, Canada E-mail: [email protected] This article provides a selective overview of recent event-related brain potential (ERP) studies in L2 morpho-syntax, demonstrating that the ERP evidence supporting the critical period hypothesis (CPH) may be less compelling than previously thought. The article starts with a general introduction to ERP methodology and language-related ERP profiles in native speakers. The second section presents early ERP studies supporting the CPH, discusses some of their methodological problems, and follows up with data from more recent studies avoiding these problems. It is concluded that well-controlled ERP studies support the convergence hypothesis, according to which L2 learners initially differ from native speakers and then converge on native-like neurocognitive processing mechanisms. The fact that ERPs in late L2 learners at high levels of proficiency are often indistinguishable from those of native speakers suggests that ageof-acquisition effects in SLA are not primarily driven by maturational constraints.

1. INTRODUCTION Why should applied linguists care about neuroscience? The temporal dynamics of integrating different types of linguistic information while we acquire and process a first or second language may be relevant to determining which are the best ways of learning—and teaching—languages. For example, recent electrophysiological research involving applied linguists has started to investigate how SLA in classroom and immersion settings may rely on different neurocognitive processing mechanisms—even though these differences do not necessarily show up behaviorally (Morgan-Short et al. 2010, 2012). A related topic of interest, and a main focus of this article, concerns the degree to which age-of-acquisition (AoA) effects on second language (L2) proficiency are biologically constrained by ‘critical periods’. If it turns out that factors other than an ‘inevitable’ loss of brain plasticity in childhood are responsible for AoA effects, language learners (and teachers) may have considerably more

Downloaded from http://applij.oxfordjournals.org/ at McGill University Libraries on September 9, 2014

Event-related Potentials (ERPs) in Second Language Research: A Brief Introduction to the Technique, a Selected Review, and an Invitation to Reconsider Critical Periods in L2

394 EVENT-RELATED POTENTIALS IN SECOND LANGUAGE RESEARCH

2. A BRIEF INTRODUCTION TO ERPS Readers already familiar with ERPs can either skip this section or focus on any of its three subsections: 2.1 General approach, advantages and limitations; 2.2 Recording and analyzing data; and 2.3 Psycholinguistic ERP components in L1.

2.1 General approach, advantages and limitations In psycholinguistic research, we are interested in how the human mind/brain processes and integrates different types of linguistic information in order to understand words or utterances. If native speakers or language learners have difficulties understanding utterances, we want to know exactly what causes these, and at which point in the sentence they occur. With processing rates of two to five words per second in listeners and readers, identifying the critical processes involved is a nontrivial task. Behavioral measures are often too slow, influenced by metalinguistic and response-related (e.g. decision and motor) processes, limited to one data point per utterance, or applicable only to sufficiently cooperative populations. In contrast, ERPs provide brain waves that reflect the neural activity without delay, with a temporal resolution of less than a millisecond, across the whole utterance (e.g. for each word, syllable, phoneme), in almost any population (including newborns and coma patients) without requiring metalinguistic tasks, and can distinguish between different cognitive processes. As described below, ERPs are computed by first recording EEG (electroencephalography) data from the scalp (Figure 1) and then analyzing the brain waves time-locked to the relevant event (e.g. the onset of a critical word on the screen or in the speech signal). The vast majority of psycholinguistic processes take place within 1,000 ms after word onset and elicit characteristic positive or negative going brain waves (ERP components) with specific latencies. For instance, automatic discrimination between two phonemes elicits frontally distributed mismatch negativities (MMNs) around 200 ms (Figure 2), while syntactic processing difficulties are often reflected


influence on final attainment in L2 acquisition than suggested by the critical period hypothesis (CPH). The present review has two goals. First, I will give a brief introduction to event-related brain potentials (ERPs) to enable readers not familiar with this technique to understand this approach, its findings, as well as advantages and limitations. This section should allow readers to easily access the other parts of this article without having to consult other introductory readings. Secondly, I will provide a selective review of ERP findings in second language research, paying particular attention to methodological shortcomings of early studies supporting the CPH—and the surprisingly different results of more recent work that tried to avoid these shortcomings. Through this, the reader should develop a critical stance regarding the CPH and the work that seems to overwhelmingly support it.

K. STEINHAUER

395

by left-anterior negativities between 300 and 500 ms, and subsequent parietal positivities around 600 ms (P600s), see Figure 3. ERP profiles for different linguistic processes differ in terms of polarity (positive vs negative), latency, duration, and scalp distribution. A given scalp distribution does not indicate the location of underlying neural generators, but rather the area of the scalp toward which activated nerve cells are oriented (similar to a flash light beam). This absence of direct relationship between the scalp distribution of an ERP component and the location of its neural generators in the brain is a limitation of ERPs compared with functional brain imaging (fMRI), but their temporal resolution is 500 times better and can help us understand the temporal dynamics of language processes in real time. Although ERP measures do not depend on task performance, most ERP studies employ tasks to ensure participants’ attention and stimuli processing. Research has shown that certain ERP patterns are influenced by task requirements, either because they direct attention to specific linguistic aspects of the experiment or because task-related ERP responses superimpose ERPs related to the linguistic processes (e.g. Royle et al. 2013). However, many ERP patterns associated with psycholinguistic processes have been very consistent across labs and languages (see Section 2.3). Therefore, it is possible to determine if second language learners show similar


Figure 1: EEG set-up, N400 effect, and voltage map. EEG data collected from an electrode-cap during sentence processing are averaged time-locked to the target words (broccoli and democracy). The (schematic) ERP plot shows similar N100s and P200s in both conditions and an enhanced N400 for the semantic mismatch condition (negative polarity is plotted upwards). The voltage map illustrates the centro-parietal scalp distribution of this N400 effect between 300 and 500 ms (mismatch minus control). See main text for details


ERP patterns as first language speakers, and whether these change as a function of AoA, L2 proficiency, etc.

2.2 Recording and analyzing data In order to obtain ERP data, three steps are necessary: (i) collection of EEG raw data while participants read or listen to linguistic input; (ii) preprocessing these raw data (e.g. filtering and artifact rejection); (iii) averaging across multiple events of the same type and comparing the resulting ERPs between conditions. To record the ‘raw’ EEG signals, EEG electrodes (sensors) are placed at specific positions on the participant’s scalp. Almost all labs now use elastic caps that contain between 20 and 256 sensors at predefined standard positions (Figures 1 and 2). Once the cap has been adjusted, the amplified EEG signal can be viewed on a monitor and stored on a computer hard drive. Special eye electrodes monitor eye movements (blinks and saccades) that elicit large artifacts in the EEG signal and have to be removed before data analyses. To minimize artifacts, ERP reading studies typically present sentences one word at a time (for 500 ms) in the center of a screen, a technique called rapid serial visual presentation (RSVP). The participant reads, or listens to, a considerable number of linguistic stimuli (e.g. syllables, sentences), typically thirty to several hundred stimuli per condition, depending on their complexity and the research question. A stimulus-specific ‘trigger code’ is sent to the EEG system each time a target stimulus is presented, marking the stimulus onset in the continuous stream of EEG data. These codes are used to identify those sections of the EEG signal that reflect the processing of target stimuli (‘time-locking’).


Figure 2: MMN with difference wave and voltage map. This early ERP component reflects the brain’s response to perceived sound differences in an oddball paradigm, used to measure abilities in phoneme discrimination. (a) ERPs for standards and deviants at frontal midline electrode FZ. (b) MMN difference wave (deviant minus standard) at FZ. (c) Voltage map of the difference wave between 140 and 190 ms after stimulus onset. (Adapted from Steinhauer and Connolly 2008, with permission from Elsevier)

K. STEINHAUER

397

Once the raw data are recorded, data ‘preprocessing’ is used to remove artifacts and ‘clean up’ the data. Common steps include filtering the data, and rejecting (or correcting) eye-blinks and other artifacts. Next, the EEG signals of all ‘clean’ trials are averaged, time-locked to the onset of the target stimulus (i.e. to the trigger codes described above). Averaging across many trials is necessary, because the brain waves initially contain both the ‘event-related’ signal of interest (reflecting psycholinguistic processes triggered by the target stimulus) and random noise (other irrelevant spontaneous brain activity). During the averaging procedure, only the relevant signal that is shared across all trials ‘survives’, whereas EEG waves reflecting random noise cancel each other out. As most ERP studies are group studies, the ERPs of individual participants are then averaged in a final step, thereby creating a ‘grand average’. The length of an ERP epoch is typically in the range of 1,000 ms, beginning at the onset of the target stimulus. ERPs for isolated stimuli (e.g. single words) are characterized by a sequence of positive and negative deflections (ERP components). During the first 200 ms, early (exogenous) ERP components primarily reflect perception processes and are strongly influenced by physical


Figure 3: Biphasic LAN and P600 effects elicited by a syntactic word category violation in native English readers. (a) ERPs at seven electrodes show a LAN between 300–500 ms and a subsequent posterior P600 (600–1000 ms) for the ungrammatical target word (dotted line). (b) Voltage maps illustrating the scalp distribution of both ERP effects. (Reproduced from Steinhauer and Connolly 2008, with permission from Elsevier)


2.3 Psycholinguistic ERP components in L1 2.3.1 Lexical-semantic processing: the N400 A good example for a ‘violation’ paradigm is found in the seminal N400 study by Kutas and Hillyard (1980), who presented sentences such as ‘He spread his warm bread with . . .’ and completed them either with ‘butter’ or with ‘socks’ (target words for ERP analyses are underlined). During the first 250 ms of the ERP, implausible completions such as ‘socks’ elicited the same N100 and P200 components as plausible words, but the subsequent N400 component between 300 and 500 ms was much larger for implausible words (Figure 1). Hundreds of follow-up studies have shown that the N400 is a reliable brain signature of lexical-semantic processing difficulties and that its amplitude decreases when a word’s cloze probability increases or processing is facilitated by a preceding semantically related prime word (e.g. doctor–nurse; see Steinhauer and Connolly 2008). The N400 is larger for content compared with function words as well as for unfamiliar (e.g. low-frequency) compared with familiar (e.g. high-frequency) words (e.g. ‘sextant’ vs ‘bottle’). Thus, the N400 allows us to study both the acquisition of vocabulary knowledge and its semantic integration in second language learners.

2.3.2 Morpho-syntactic processing: LANs and P600s Similar to behavioral studies using grammaticality judgments (GJs), ERP studies on morpho-syntactic processing typically compare the ERPs elicited by various kinds of grammatical errors to matched control sentences. Common paradigms include word order (phrase structure) violations (‘He criticized Max’s *of proof . . .’) and agreement violations of number, case, etc. (e.g. ‘The children *plays in the garden’). Other paradigms have employed anomalies such as


characteristics of the stimulus (e.g. loudness, font size, and physical contrast to the background). A Negativity around 100 ms (N100) signals processing in the primary visual or auditory cortex and is followed by a positivity around 200 ms (P200), already reflecting more complex pattern recognition (Figures 1 and 2). These early ERP components are followed by later (endogenous) ERP components reflecting higher-level cognitive processing. For example, the N400 (a centro-parietal negativity around 400 ms) gradually increases in amplitude the more difficult lexical-semantic processing is (Figure 1). Importantly, only an ERP comparison between two conditions (e.g. a linguistic violation and its correct control, or a deviant vs standard phoneme) can tell us something about the relationship between psycholinguistic processes and their ERP profile. To better illustrate the scalp distribution of ERP effects (i.e. the difference between two conditions), ERP papers often also provide voltage maps displaying this difference (usually the violation condition minus its control) in a representative time window (e.g. 300–500 ms for N400 effects, see Figure 1).

K. STEINHAUER

399

2.3.3 Phonological ERP components: MMN and Closure Positive Shift (CPS) Another domain that has received much attention in ERP research is phonological processing, especially phoneme discrimination (e.g. /ba/vs/pa/). Most studies have used the so-called ‘oddball’ paradigm, which presents a sequence of sounds consisting of a frequent standard (e.g. 80%/ba/) intermixed with one or more rare deviants (e.g. 20%/pa/). Both standards and deviants elicit N100s and P200s. But only if the listener’s brain is able to distinguish the two sounds, the deviant additionally elicits an MMN between 120 and 250 ms (superimposing the P200; see Figure 2 for an example). As the MMN was also found to be sensitive to phonemic categories (Na¨a¨ta¨nen et al. 2007), it can be used to investigate cross-linguistic differences in phoneme boundaries and respective adjustments in bilinguals (Molnar et al. 2014). Another ERP component related to phonological processing is the CPS, a large centro-parietal positivity elicited at prosodic boundaries. It reflects the prosodic phrasing of spoken utterances in listeners (Steinhauer et al. 1999). Prosodic boundaries often guide syntactic parsing and sentence comprehension; for example, the sentence ‘Mary said Peter’s brother is a nice girl’ makes sense only with boundaries before and after ‘said Peter’s brother’. Without such boundaries, the word ‘girl’ would be implausible and elicit an N400. Thus, combining measures of the CPS and other ERP components


garden-path sentences involving temporal ambiguities between active past tense vs passive reduced-relative-clause readings (‘The broker persuaded the investor to sell . . .’ vs ‘The broker persuaded to sell the stock . . .’; Osterhout and Holcomb 1992) or syntactic complexity more generally (Kaan et al. 2000). The most reliable ERP finding across morpho-syntactic studies has been a parietal positivity between 600 and 900 ms after word onset (P600, see Figure 3). However, P600 amplitudes are larger in combination with GJ tasks and, therefore, likely reflect both genuine morpho-syntactic processing demands as well as task-related processes (e.g. Royle et al. 2013). The second, less consistent ERP component for morpho-syntactic violations is a negativity preceding the P600 in a time window similar to that of the N400 (300–500 ms). Due to its left-anterior distribution in a number of early reading studies, this component is often referred to as ‘left-anterior negativity’, or LAN (e.g. Friederici 2002). However, both its scalp distribution and its latency have shown some variation across studies. It is controversial whether sustained bilateral anterior negativities (ANs), left-temporal negativities (LTNs) and ‘early left-anterior negativities’ (ELANs) reflect similar or distinct cognitive processes, whether some differences between them are modality-dependent (spoken vs written language), and whether some of these ‘components’ must even be viewed as artifacts (e.g. Steinhauer and Drury 2012; see below). The core message here is that morpho-syntactic processing difficulties generally elicit an ERP profile that is qualitatively distinct from the N400 found for lexical-semantic processing (see Section 3.2 for a functional interpretation of LANs and P600s).


(N400s, P600s) can shed light on the online integration of prosodic and morpho-syntactic information (Pauker et al. 2011) in L1 and L2. The CPS is also elicited by subvocal prosodic phrasing, allowing us to investigate implicit prosody during silent reading (Steinhauer 2003; Hwang and Steinhauer 2011).

2.3.4 Language specificity of ERP components

3. ERPS IN L2 AND THE CPH Having reviewed some important ERP components, we can ask what ERP evidence exists for similarities and differences between L1 speakers and L2 learners. I will focus on a subset of selected studies in morpho-syntax bearing on this issue. After clarifying what the claims of the CPH are and which linguistic domains seem to be most affected by it (Section 3.1), we will look at early ERP studies supporting these claims (3.2). Next, I will discuss a number of severe methodological shortcomings in these studies. The last section (3.3) will present data from more recent ERP studies attempting to avoid these problems.

3.1 The CPH in L2 acquisition: not all linguistic domains are equally affected ‘Age-of-acquisition’ effects are typically attributed to biological factors, most prominently to loss of brain plasticity after a ‘critical’ or ‘sensitive’ period for language learning in childhood. As a consequence, adult language learners are often viewed as having to rely on fundamentally different neurocognitive mechanisms than child L1 learners. This ‘critical period hypothesis’ (CPH; Lenneberg 1967), as well as the related ‘fundamental difference hypothesis’ (FDH; Bley-Vroman 1989), make intuitive sense, especially in the context of Chomsky’s claim for a biologically based ‘language acquisition device’ representing core principles and parameter options of a ‘Universal Grammar’ (UG)


Comparisons with other cognitive domains suggest that none of the ERP components described above are exclusively elicited by linguistic materials. For example, N400s can also be found for semantically mismatching pictures (e.g. Royle et al. 2013), and LAN-like negativities, P600s, MMNs, and CPS have been reported for music (see Patel et al. 1998, for negativities and P600s to stimuli violating the melodic ‘syntactic’ structure). However, even if these ERP components are not specific to language, they may still reflect cognitive processes that are both characteristic and tightly linked to natural language processing. Moreover, there is ongoing debate whether relevant neurocognitive mechanisms underlying language processing are ‘language specific’ in the Chomskian sense (e.g. Ullman 2001; Goodluck and Zweig 2013).

K. STEINHAUER

401

3.1.1 What are the assumed differences between L1 and L2? Simply put, an early acquired L1 is assumed to be (i) acquired implicitly and effortlessly (possibly guided by UG) and (ii) to be quasi-automatic and rapidly processed (potentially relying on language-specific brain modules), whereas late-acquired L2 would be acquired more explicitly and effortfully (possibly not guided by UG), and rely on ‘domain-general’ cognitive mechanisms. As a consequence, childhood L1 learners are said to follow a universal learning trajectory with common milestones, and always converge on the target grammar of the input language. Late L2 learners, in contrast, vary in their trajectories and would not attain full target grammar competence (Bley-Vroman 1989). A few initial clarifications about the CPH controversy seem to be in order. First, almost everyone agrees that there are critical periods in first language acquisition. Seminal case studies and more systematic recent work provide overwhelming support for irreversible problems after delayed L1 acquisition for both L1 and L2 (Mayberry and Lock 2003). The crucial controversy in SLA is about whether early exposure to L2 is as critical as for L1 acquisition, or whether one’s L1 provides a brain network that can then also be used to acquire an L2. Secondly, hardly anyone disputes AoA effects in SLA. The question is whether these point to loss of brain plasticity during a specific period early in life, or rather to other factors. Thirdly, AoA effects in SLA do not seem to affect all linguistic domains in the same way. Consistent findings of problems in L2 phonology (foreign accents) and morpho-syntax suggest that these domains are more sensitive to AoA effects than, for example, the acquisition of new vocabulary (which is also necessary in one’s L1 throughout the life span) (Hahne 2001; Newport et al. 2001; Birdsong 2006). Accordingly, these two domains have been a major focus in SLA research by proponents and opponents of the CPH. In the following, I will focus primarily on morpho-syntax (in a broad sense, i.e. both syntax proper and morpho-syntax).


underlying all natural languages. Studies such as those by Johnson and Newport (1989) and more recent work by DeKeyser (2012), Abrahamsson (2012) and Abrahamsson and Hyltenstam (2009), demonstrating that AoA is a reliable predictor for ultimate attainment in L2, seem to provide indisputable support for the CPH (but see Birdsong 2006; Herschensohn 2009; and Hopp 2010). In addition, neuroscientific data in the 1990s and early 2000s contributed to establishing the CPH as current textbook wisdom backed-up by ‘hard brain data’. Some of these studies seemed to provide direct evidence that language-specific structures in the brain could not be accessed by late learners (e.g. Hahne 2001; Newport et al. 2001; Sakai 2005). Below I will explain why this evidence may not be as convincing as it seems.


3.1.2 Alternatives to the CPH Three possible scenarios have been suggested for the relationship between L1 and L2 processing: A. B. C.

Difference Hypothesis: Similarity Hypothesis: Convergence Hypothesis:

L2 6¼ L1 L2 = L1 L2 ! L1 (L2 6¼ L1 ! L2 = L1) Downloaded from http://applij.oxfordjournals.org/ at McGill University Libraries on September 9, 2014

The view in (A) is basically the position of the CPH and the FDH: Once maturational constraints during childhood or puberty have resulted in loss of plasticity in brain systems needed for native-like language acquisition, late learners have to allocate other brain systems for their L2. Such differences between L1 and L2 are fundamental and irreversible and predict distinct brain activation patterns as a function of AoA (especially, before vs after the CP). According to the hypothesis in (B), L2 acquisition and processing largely rely on the same mechanisms as the L1 (Hernandez et al. 2005). There is, in principle, no reason to assume systematic differences between L1 and L2. However, the assumption that both L1 and L2 use the same neurocognitive systems and compete for the same resources can account for different learning trajectories in L1 vs L2, as well as L1 attrition: an already established L1 likely influences the acquisition of a new language, and L2 dominance would have effects on one’s L1. In extreme cases, an L2 can virtually replace the L1 (Pallier et al. 2003). AoA effects are linked to the degree of L1 entrenchment and other factors such as motivation, executive control skills, etc. Thirdly, the Convergence Hypothesis in (C) is somewhat similar to (B). However, neurocognitive mechanisms are assumed to be dynamic and to change over time with increasing practice and language proficiency (thereby converging on L1 patterns). That is, as long as L2 learners are not very proficient, this hypothesis expects differences between L1 and L2 (partly because L2s are acquired differently from L1s). However, there are no biological constraints that would prevent L2 learners from using the same brain mechanisms found in native speakers. Therefore, even adult L2 learners are expected to reach native-like proficiency for some or all aspects of L2, in which case the brain activation patterns should be indistinguishable from those of native speakers. Some versions of this hypothesis include specific predictions of how brain activation should change with increasing L2 proficiency (Green 2003; Steinhauer et al. 2009). The similarities between (B) and (C) are these: If one assumes that children acquiring their L1 also undergo dynamic neurocognitive changes, then their trajectory toward the ‘native’ patterns of adult L1 speakers may largely resemble the trajectory found in L2 learners. The appropriate comparisons between L1 and L2 should take the respective proficiency levels of language learners into account. Another related issue is that having two languages in one brain may even affect one’s first language, such that differences between monolingual native speakers and second language learners may partly be due to differences between monolingual and bilingual brains, irrespective of critical periods.1

K. STEINHAUER

403

3.1.3 An ERP model of L1-L2 convergence

3.2 ERP studies supporting the CPH and their methodological problems The most influential ERP study on morpho-syntactic processing in L2 learners is that by Weber-Fox and Neville (1996). The authors replicated parts of


Our model (Steinhauer et al. 2009) predicts that increasing levels of proficiency in L2 morpho-syntax should be reflected by systematic changes in ERP profiles. At low levels of proficiency, one expects no effects or only N400s reflecting a fundamentally nonnative-like approach to morpho-syntactic structures and violations (often relying on explicit, classroom-instructed rule knowledge). Once the brain starts to ‘proceduralize’ or ‘grammaticalize’ L2 morpho-syntactic rules, small and typically delayed P600s (e.g. Osterhout et al. 2006; Rossi et al. 2006) will be elicited (see Stowe and Sabourin 2005; Osterhout et al. 2006; Kotz 2009; and McLaughlin et al. 2010, for similar ideas regarding proficiency-dependent changes). With increasing L2 proficiency, these P600s increase in amplitude, while their latency decreases (becoming more native-like). Finally, native-like L2 proficiency (for the target structure) should reliably elicit the same ERP components found in native speakers (or simultaneous bilinguals).2 However, since similarities between the L1 and L2 for certain structures result in distinct learning trajectories (L1 transfer effects), L2 learners are expected to differ as to whether they reach native-likeness for a given structure X prior to structure Y, or vice versa. Therefore, measures of overall proficiency in L2 may not predict the ERP profile for a specific structure, whereas specific proficiency measures for that structure should. Additionally, even though the time-course and order of achievements may differ among L2 learners, for a given structure the sequence of ERP profiles (starting with no effects or N400s and progressing toward the native-like ERP profile) is expected to be largely universal (important exceptions are linked to factors like the type of L2 exposure, see below). In contrast, although the CPH would allow for ERP changes with increasing L2 proficiency, the ERP profiles should never converge on those of native speakers. Before we turn to ERPs in L2 learners, a few clarifications regarding LANlike ERP effects and ‘native ERP profiles’ are necessary to avoid misleading oversimplifications. First, not all morpho-syntactic violations reliably elicit LAN-like negativities in native speakers; sometimes native speakers show N400s followed by P600s (e.g. Steinhauer et al. 1999; Tanner and van Hell 2014; see also Figures 5 and 6c). In those cases, the absence of a LAN or presence of an N400 in L2 speakers can obviously not be viewed as an indication of nonnativeness (Steinhauer et al. 2009). Secondly, ERP components can overlap in time, such that a (negative-going) central N400 and an overlapping (positive-going) right-posterior P600 may cancel each other out at midline electrodes and look like a left lateralized LAN followed by a P600.3


3.2.1 Methodological problems However, all three studies mentioned above have two methodological problems that raise serious questions about the validity of their findings. First, they entirely confounded AoA with L2 proficiency: the higher the AoA, the lower the L2 proficiency level. This is a problem because it is unclear whether the ERP differences should be attributed to AoA (as the authors did) or to their proficiency level. In other words, the studies claim to have found evidence for a causal relationship between biological constraints and L2 attainment, but what they actually showed was only that, unsurprisingly, less proficient L2 learners do not process L2 morpho-syntax like native speakers. In order to support their much stronger claim, one must show that either there really are no late L2 learners that can achieve native-likeness, or that, if nativelike L2 proficiency levels exists, even successful learners rely on other brain mechanisms than native speakers.


Johnson and Newport’s (1989) study with ERPs and investigated various types of morpho-syntactic violations in adult Chinese and Korean learners of English (e.g. He criticized Max’s *of proof the theorem). Data from their reading study were analyzed depending on the participants’ AoA and compared with native speakers. The main ERP findings were that, (i) except for very early L2 learners (AoA 2 years), none of the subjects showed the early left-anterior and temporal (LAN-like) negativities found in native speakers, and that (ii) those with late AoA (>16 years) did not even elicit a P600. Moreover, their behavioral GJ data revealed decreasing accuracy levels with increasing AoA. The authors took this as strong evidence that early syntax-related brain processes in the left hemisphere reflected by LAN-like components are not available to late learners, as predicted by the CPH. As semantic anomalies (John sailed Mary’s *cloud to Boston) elicited an N400 in all groups, they also concluded that syntactic, but not semantic, L2 processing is subject to critical periods. Findings of this reading study were then extended by various auditory studies in late L2 learners of German. Hahne and Friederici (2001) and Hahne (2001) employed a German phrase structure violation paradigm and reported that, whereas the late P600 component was elicited at least in some participants, the earlier LAN-like negativities between 100 and 500 ms were not found in any L2 group. Within the framework of Friederici’s (2002) neurocognitive model of sentence processing, early LANs are linked to highly ‘automatic’ processes of phrase structure building (involving Broca’s area and the left-anterior temporal lobe), whereas late P600s reflect more controlled processes. Their studies suggested that late L2 learners cannot recruit Broca’s area for native-like early automatic processes and have to rely on different mechanisms, again supporting the CPH and the FDH. In sum, by the year 2001, behavioral and ERP evidence seemed to strongly back up these hypotheses, and these findings are unfailingly cited in articles and chapters on critical periods in L2 acquisition.

K. STEINHAUER

405


The second problem of the studies cited above is that their unbalanced experimental paradigms are prone to certain ERP artifacts that may only look like syntax-related LAN components. Specifically, all these studies, as well as other follow-up studies (Mueller et al. 2005; Rossi et al. 2006; Pakulak and Neville 2011), kept the target word (e.g. of in our example) constant and created violations by changing the pre-target context (e.g. He criticized Max’s proof of the theorem vs He criticized Max’s *of proof the theorem). As discussed in Section 2.2, ERP effects are calculated by comparing the relevant conditions. If the words preceding the target word differ between the violation condition (Max’s) and its control (proof), it is possible that ERP differences that we see on the target word (of) are actually due to ERP differences elicited by the pre-target context. In fact, Steinhauer and Drury (2012) demonstrate that this is the case in many ERP studies, in both L1 and L2. In other words, LAN-like components, which have been taken as a hallmark brain signature of ‘native-like’ automatic syntax processing, may often not even be related to syntactic processing at all. Instead, they may reflect ‘spill-over’ and ‘DC-offset’ (baseline) artifacts4 due to context-dependent ERPs reflecting either word category differences in the context, prosodic differences, or processing strategies acquired over the course of an experiment (see Steinhauer and Drury 2012 for details). The most striking cases illustrating these problems come from studies modeled on the Hahne and Friederici (2001) paradigms, but Weber-Fox and Neville’s paradigm has similar problems. For example, Rossi et al. (2006), who adopted Hahne and Friederici’s paradigm, reported early LAN-like negativities for phrase structure violations in both German and Italian spoken sentences. However, as shown by Steinhauer and Drury (2012), these effects occurred some 400 ms before the violation was even presented—clearly pointing to context-driven artifacts. Similarly, a reading study by Newman et al. (2007), modeled after WeberFox and Neville (1996), as well as Mueller et al.’s (2005) auditory study on ‘mini-Japanese’ also found very early LAN-like components that either preceded the onset of the target word carrying the violation (Newman et al. 2007) or were elicited in novice L2 learners who had no morpho-syntactic knowledge (Mueller et al. 2005), such that these authors refrained from interpreting the negativities as a real violation effects. Recent work has shown that some very early negativities (e.g. Friederici 2002; Dikker et al. 2009, 2010; Kim and Gilley 2013) may be due to the very constrained selection of sentence materials that encourage artifactual processing strategies and tell us little about normal everyday sentence processing outside of the laboratory, that is, their ecological validity may be questionable (e.g. Bourguignon and Steinhauer, under revision). In conclusion, since early LAN-like components are often either obvious artifacts or ambiguous, their absence in L2 learners may simply point to differences in developing certain strategies during laboratory experiments, rather than to differences in morpho-syntactic processing in the real world. Moreover, as long as ERP studies in L2 confound AoA and language proficiency, group differences can be attributed to either of these two factors.


3.3 Recent ERP evidence challenging the CPH This section shows that ERP data typically do not support the CPH when experimental designs and analyses are better controlled for. Some studies also elucidate the role of other factors contributing to AoA effects and variability among late L2 learners.

3.3.1 AoA or L2 proficiency? Downloaded from http://applij.oxfordjournals.org/ at McGill University Libraries on September 9, 2014

To my knowledge, the first two ERP studies systematically teasing apart AoA and proficiency effects in L2 morpho-syntax were those by Friederici et al. (2002) and Rossi et al. (2006). In Friederici et al. (2002), we employed an artificial miniature language (Brocanto) to test language learning while controlling for previous exposure to it. Participants learned to speak and understand Brocanto by playing a computer-implemented (chess-like) board game and communicating their moves in this language. After reaching a high level of proficiency (>90% accuracy), their ERPs were measured while they listened to both grammatical and ungrammatical sentences. Grammar violations (including phrase structure violations) elicited early LAN-like negativities around 100 ms, a second negativity around 300 ms, and a P600; the ERP profile viewed as ‘native-like’ for this kind of violation. Importantly, a control group that only acquired the vocabulary but not the grammar of Brocanto did not elicit any ERP effects to these violations, strongly suggesting that the LANP600 pattern was indeed related to grammatical processing. Although our study successfully avoided trivial context-driven artifacts of the sort discussed above, in retrospect, the early LAN effects may need to be reinterpreted as reflecting phonological mismatches between expectations developed during the study and the target words presented in violation conditions. This reinterpretation is motivated by the fact that Brocanto’s vocabulary was very limited (16 words) and word initial phonemes were largely confounded with word categories. In other words, although better designed than most other studies, the Brocanto study still shared a shortcoming characterized above as ‘ecologically invalid’. In contrast, the subsequent negativity and the P600 are likely real violation effects found in natural languages as well. The interesting finding was that, for the first time, adult L2 learners were shown to elicit the same biphasic ERP profile (negativities followed by P600s) that had previously been thought to be limited to native speakers. Another critique of this study was that it used an artificial language. However, its findings have been confirmed by studies on natural language acquisition (see below). The second ERP study teasing apart AoA and L2 proficiency by Rossi and colleagues (2006) tested late Italian learners of German and late German learners of Italian, both at low and high levels of L2 proficiency. This study is remarkable in two ways. On the one hand, it provides the first clear evidence using natural languages that late L2 learners at high proficiency do indeed elicit the LAN-P600 pattern previously found only in native speakers, whereas late

K. STEINHAUER

407

3.3.2 L2 proficiency and L1 background (L1 transfer) Since 2005, ERP studies have tested a variety of morpho-syntactic anomalies in late post-CP L2 learners (with AoAs of more than 18 years) at both low and high levels of L2 proficiency. The general pattern suggests, again, that proficiency rather than AoA predicts observed brain signatures, with native-like L2 proficiency resulting in native-like ERP profiles. In addition, typological similarity between L1 and L2 modulates results, pointing to transfer effects and coactivation of both languages. For example, we developed a novel balanced design of phrase structure violations in English that consists of two violation and two correct control conditions, which together avoid any kind of contextdriven artifacts (White et al. 2006; Steinhauer et al. 2013). In a reading study, we tested this paradigm in both French and Chinese late learners of English at low and high proficiency. As predicted, at low proficiency, French and Chinese groups differed both from native speakers (no native-like LANs) and from each other (with French learners displaying more native-like P600s). However, at high L2 proficiency, there was no difference between French and Chinese participants: they all showed a (N400-)LAN-P600 pattern that was indistinguishable from that of the English control group. Figure 4 illustrates the main findings collapsed across French and Chinese learners. Our results demonstrate that L1 background plays a significant role during the first stages of L2 acquisition, but much less so when L2 proficiency approaches native-likeness. A replication of this study in the auditory domain with German and Chinese learners of English (currently underway) points to a similar pattern. Together with Rossi et al.’s (2006) agreement data, these results confirm the Brocanto findings in natural language, and strongly support neurocognitive convergence of L2 learners on L1 processing. Moreover, Bowden et al.’s (2014) balanced reading study replicated these findings with late English learners of Spanish, thus providing cross-linguistic evidence for the convergence hypothesis. Similarly, Gillon Dowens et al. (2010) found that proficient late EnglishSpanish bilinguals who acquired Spanish after an age of 20 years displayed ERP profiles for gender and number agreement violations ‘qualitatively similar to those of native speakers’ (p. 1870), consisting of a sequence of LANs, P600s, and late negativities. However, while all native-like components were present


L2 learners at lower proficiency (matched on AoA with the first group) only show a P600. This replication of Friederici et al.’s (2002) findings, for morphosyntactic agreement violations, suggests that the previous reports of absent LANs in late L2 learners (Weber-Fox and Neville 1996; Hahne and Friederici 2001) must be attributed to low proficiency levels, not to late AoA. On the other hand, the phrase structure violation condition in this study (discussed above in Section 3.2) is the one that elicited context-driven negativities 400 ms before the actual violation occurred. Therefore, the ERPs in this condition, which—to the authors’ surprise—were replicated even in low-proficiency L2 learners, must be viewed as artifacts.


at determiner–noun violations within the noun phrase, violations for long distance noun–adjective agreement did not elicit the LAN found in the Spanish native speakers. In the gender disagreement condition, these group differences may be related to the much lower accuracy/proficiency for this structure in the L2 group (likely due to L1–L2 transfer), but group differences


Figure 4: Voltage maps of LAN and P600 effects in L1 and high vs low proficiency L2. Native speakers and late L2 learners at high proficiency both show the same bi-phasic LAN-P600 pattern of left-lateralized negativities and subsequent posterior positivities. In contrast, L2 learners at lower proficiency only show a broadly distributed (and relatively small) P600, but no LAN. Differences between the L2 groups show that the absence of LANs is due to lower proficiency levels, not delayed AoA. (Reproduced from Steinhauer et al. 2009, with permission from SAGE publishers)

K. STEINHAUER

409

for number disagreement suggest that language proficiency alone does not always fully predict the ERP profile (see also below). A follow-up study by Gillon Dowens et al. (2011) with proficient Chinese learners of Spanish-L2 found P600s and late frontal negativities, but no LANs, for either determiner–noun or noun–adjective violations. The authors attributed the differences between their two studies to transfer effects as a function of L1 background (where Spanish-L2 is more similar to English-L1 than Chinese-L1).

Another paradigm tested in our lab included adjective placement (John bought the vase *white) and bare singulars that lacked a determiner (article) (Yesterday he bought *book) (Steinhauer et al. 2010). As expected, Chinese learners, whose L1 does not have determiners, missed the ungrammaticality of bare singulars more often than the French learners. In contrast, for adjective placement, which is prenominal in both English and Chinese (the white vase) but often postnominal in French (le vase blanc), French learners showed a pattern that differed from both Chinese learners and native speakers. These data illustrate that L2 proficiency is not equal for all structures, and that native-like proficiency in L2 is reached earlier for those structures that are more similar between the L2 and one’s mother tongue. As predicted by the convergence hypothesis, at high proficiency, both Chinese and French learners showed the native-like ERP patterns of English L1 speakers (e.g. an N400-P600 on the ungrammatical postnominal adjective). Interestingly, only high-proficiency French learners additionally showed an early N400-P600 effect on the grammatical prenominal adjective of the control condition, however, only if the French translation would have been ungrammatical (e.g. for the white vase [le *blanc vase], but not for the small vase [le petit vase]; Figure 5). This finding shows that the L1 grammar is partly co-activated even when high-proficiency L2 learners process sentences that are exclusively presented in L2, extending previous findings by Thierry and Wu (2007), who demonstrated phonological L1 activation during L2 reading. Intriguingly, in contrast to most previous reports on the impact of L1 on L2, co-activation does not necessarily lead to interference between competing L1 and L2 structures, at least not at high-proficiency levels. To summarize, there is strong evidence that L1 transfer plays a role in L2 acquisition (especially at early stages). At very high levels of L2 proficiency, the L1 is still partly co-activated, but does not interfere with the appropriate processing of the L2.

3.3.4 Training environments As already mentioned, a major difference between L1 acquisition in infancy and L2 acquisition at older ages is that infants learn implicitly, whereas late learners are typically exposed to classroom instruction that explains grammar


3.3.3 L1 co-activation even at high L2 proficiency


rules before we start using them. In fact, it is often assumed that adults have lost the ability to learn languages implicitly. This is a highly controversial issue. Interestingly, the common layman advice for efficient L2 acquisition (i.e. ‘Find yourself a native boyfriend/girlfriend’) also suggests that immersion, not classroom instruction, is viewed as the most efficient way of learning a language. However, research in applied linguistics appears to paint a different picture: studies have quite consistently demonstrated the superiority of explicit instruction, thus supporting the CPH’s assumption that adults are better off learning an L2 differently from children. In the late 2000s, Morgan-Short and colleagues decided to investigate this question with ERPs using a modified (more Romance-like) version of our artificial language Brocanto. This study introduced implicit (immersion-like) and explicit (classroom-like) training conditions. Unlike the first Brocanto study, where the dynamics of the computer game decided which moves had to be made, and thus what kinds of sentences subjects had to produce and comprehend, the new paradigm controlled a priori both groups’ exposure to the exact same sentences. The only


Figure 5: ERP effects for adjective placement in high-proficient French learners of English. (a) ERPs elicited by ungrammatical postnominal adjectives (the vase *white) show the same large N400-P600 violation effect found in native speakers (see arrows). (b) Unlike native English speakers and Chinese learners, only French learners show an additional early N400-P600 effect for grammatical prenominal adjectives (the white vase) that would have been ungrammatical in French (le *blanc vase), compared with those that would have been grammatical in French (the small vase – le petit vase). These ERPs illustrate co-activation of L1 grammar during L2 processing. (Reproduced from Steinhauer et al. 2010)

K. STEINHAUER

411

3.3.5 Near-native ERPs without immersion? Whereas the Morgan-Short et al. (2012) study, as well as many studies in natural language learners, found ‘native-like’ ERP profiles only after extended periods of immersion-like exposure to the L2, it remains unclear whether immersion is mandatory. A few studies using relatively simple sentence paradigms and typologically close L1-L2 pairings found near-native ERP patterns after long years of classroom instruction, followed by less than 1 year of immersion (Bowden et al. 2013; Nickels et al. 2013). The study by Nickels and colleagues (2013) tested the processing of prosody–syntax mismatches (e.g. When a bear is approaching #1 the people #2 the dogs come running, where the first boundary #1 is superfluous) in advanced German learners of English. They found that the late L2 learners displayed virtually the same ERP components both at the boundaries (eliciting CPS components, Figure 6a, b) and at positions disambiguating the mismatches (N400s and P600s, Figure 6c). However, since a follow-up study with Chinese learners of English showed clear differences from this pattern (unpublished data), Nickel et al.’s data also seem to illustrate L1-L2 transfer effects. A recent ERP reading study by Reichle and Birdsong (2014) further extended the range of linguistic phenomena studied in advanced classroom-instructed L2 learners. They examined focus


difference between groups was that the ‘explicit’ group spent a certain amount of time learning the explicit grammar rules of Brocanto_2, whereas the ‘implicit’ group spent the same amount of time listening to and producing sentences without formal instruction. Replicating previous studies in the applied linguistics literature, explicit learners initially outperformed implicit learners in terms of accuracy (although their judgments were somewhat slower). However, in follow-up sessions, the implicit group caught up, such that at the end of the last training period both groups’ performances were indistinguishable. In contrast, the ERP patterns for ungrammatical sentences (containing phrase structure violations) in Brocanto_2 differed significantly between groups: whereas implicit learners displayed the full set of ERP components found in native speakers of natural languages (including LANs and P600s), explicit learners basically elicited only P600s. These data suggest a number of things. First, even adults can implicitly learn language grammars to a certain degree (see also, Reber 1967 on implicit learning). Secondly, previous studies comparing implicit and explicit learning may have captured only early phases where explicit learners still outperform implicit learners. Thirdly, implicit (immersion-like) training environments may result in native-like processing earlier than explicit learning environments, even in adulthood. Before we can draw any firm conclusions, these findings require replication, ideally in natural language learners. Nevertheless, they are exciting and promising; a study of comparable quality with natural language learners would likely have taken years.


structure in French and observed a native-like negativity for contrastive focus in the more proficient L2 group.

3.3.6 Individual differences Even though recent evidence seems to support Hypothesis (C) above, according to which L2 learners converge on native-like processing as a function of L2 proficiency, their specific trajectories may differ not only depending on L1, but


Figure 6: CPS components and prosody–syntax mismatch effects in L1 and L2 (a) ERP plots at electrodes CZ and PZ (collapsed across L1 and L2 groups) show similar closure positive shifts at an early prosodic boundary (#1) in condition B (gray line) and at a later boundary (#2) in condition A (black line). (b) Voltage maps illustrate that the scalp distribution of these CPS components was very similar in native speakers (L1) and German learners of English (L2). (c) ERP difference waves (mismatch minus control) at electrodes CZ and PZ for the disambiguating region of a syntax–prosody mismatch in native English speakers (black line) and late German learners of English (gray line). Both groups show very similar N400 and P600 ERP components in the same latency range. The slightly less posterior P600 topography for the L2 group is in line with previous findings of advanced (but nonnative-like) L2 learners (see also Figure 4). (Adapted from Nickels et al. 2013, with permission from Elsevier)

K. STEINHAUER

413

4. CONCLUSION In sum, early ERP studies that are reliably cited as providing ‘hard’ evidence supporting the CPH in L2 morpho-syntax have a number of methodological issues including confounds of AoA with proficiency levels, and ambiguous ERP effects. More recent (and better controlled) ERP studies that come from various labs and use a large variety of paradigms across different stages of L2 proficiency largely support the convergence hypothesis: they demonstrate quite systematic neurocognitive changes as a function of (individual) L2 proficiency and show that ERPs of late L2 learners at very high levels of L2 proficiency are usually indistinguishable from those of native speakers. These findings cast serious doubt on the claim that genetically based maturational constraints in brain development are responsible for difficulties in late L2 acquisition.


also across subjects with similar L1 backgrounds. One exciting finding is that ERPs may be able to identify such differences even before they manifest in behavioral performance. In line with McLaughlin et al.’s (2004) observation of early ERP differences related to vocabulary knowledge, a recent study by Tanner et al. (2013) demonstrated that individuals at relatively early stages of L2 acquisition may differ in whether their brains respond to grammar violations with N400s or with P600s. N400s (linked to lexical-semantic processing and declarative memory) reflected less successful trajectories than P600s. These findings concur with our model linking ERP components to proficiency levels, according to which P600s reflect a higher level of L2 proficiency than N400s (Steinhauer et al. 2009). Tanner et al. found that these ERP differences correlate with individual proficiency, not with the hours of L2 exposure (see White et al 2012 for similar ERP data on individual differences). If it turns out that lack of native-like achievement in L2 acquisition is not primarily due to brain maturational constraints that prevent the learner from using the optimal neurocognitive mechanisms, individual motivation levels to overcome L1 influences might play a powerful role in accounting for variability among L2 learners. In fact, Tanner and colleagues (2014) reported that highly motivated L2 learners were more likely to elicit P600s than N400s in response to morphosyntactic violations. However, since even native speakers showed either N400s or P600s (Tanner and van Hell 2014), Tanner et al.’s measure of ‘L2-motivation’ may point to other interindividual differences. Other work by Wong et al. (2012) goes a step further and suggests that individual genetic differences affecting the neurotransmitter dopamine (DA) may play a crucial role in successful language learning. If these authors are right, this hypothesis may point to individually tailored language learning. Importantly, the dopamine hypothesis is related to relative reliance on different memory systems (e.g. Ullman 2001), not on genetic factors modulating critical periods. In other words, this hypothesis does not seem to provide a theoretical basis to identify successful L2 learners as ‘genetic exceptions’ from the critical period hypothesis, but rather points to distinct types of learning processes.


However, whether such L1–L2 differences might be found for other, more subtle morpho-syntactic structures or in other linguistic domains is an empirical question that awaits clarification by future studies.

ACKNOWLEDGEMENTS

NOTES 1 Such effects of L2 on L1 can be found in bilingual immigrants who are primarily immersed (and may even be dominant) in their L2 (e.g. Kasparian et al. 2013). 2 In other words, the main point of our model is that, with increasing L2 proficiency for a target structure, L2 learners increasingly converge on the ERP pattern found in native speakers, not that they always show N400s at low proficiency or LAN components at high proficiency. 3 Tanner and van Hell (2014) demonstrated that an ERP group average may show a LAN-P600 pattern, while most of the individual data sets showed either only an N400 or only a P600. This non-representative group average also illustrates how important

it is to inspect individual data sets in addition to ERP group averages. 4 ‘Spill-over’ artifacts are due to pretarget differences that spill over into the target word (and may be mistaken for an ERP effect triggered by the target word). ‘DC offset effects’ have to do with baseline corrections of ERPs that force the ERPs of two conditions together in the pre-target ‘baseline interval’. If the pre-target word of a control condition shows an enhanced N400 in this baseline interval, baseline correction would create an early sustained negativity at the target word of the violation condition, which can be misinterpreted as a real violation effect.

REFERENCES Abrahamsson, N. 2012. ‘Age of onset and nativelike L2 ultimate attainment of morphosyntactic and phonetic intuition,’ Studies in Second Language Acquisition 34: 187–214. Abrahamsson, N. and K. Hyltenstam. 2009. ‘Age of onset and nativelikeness in a second language: Listener perception versus linguistic scrutiny,’ Language Learning 59: 249–306. Birdsong, D. 2006. ‘Age and second language acquisition and processing: A

selective overview,’ Language Learning 56 (Suppl. 1), 1–49. Bley-Vroman, R. 1989. ‘What is the logical problem of foreign language learning?’ in S. M. Gass and J. Schacter (eds): Linguistic Perspectives on Second Language Acquisition. Cambridge University Press, pp. 41–68. Bourguignon, N. and K. Steinhauer. Under revision. Levels of mental imagery


The author would like to thank Darren Tanner and two anonymous reviewers for their constructive comments. This work was supported by grants awarded to K.S. by the Canada Research Chair program/Canada Foundation for Innovation (CRC/CFI; project # 201876), the Natural Sciences and Engineering Research Council of Canada (NSERC; #RGPGP 312835, #RGPIN 402678-11), as well as by a team grant from the Fonds de Recherche Socie´te´ et Culture (FQRSC; #2010SE-103727).

K. STEINHAUER

brain potentials,’ Journal of Psycholinguistic Research 30: 251–66. Hahne, A. and A. D. Friederici. 2001. ‘Processing a second language:Late learners’ comprehension mechanisms as revealed by event-related brain potentials,’ Bilingualism: Language and Cognition 4: 123–41. Hernandez, A., P. Li, and B. MacWhinney. 2005. ‘The emergence of competing modules in bilingualism,’ Trends in Cognitive Sciences 9/ 5: 220–5. Herschensohn, J 2009. ‘Fundamental and gradient differences in language development,’ Studies in Second Language Acquisition 31: 259–89. Hopp, H 2010. ‘Ultimate attainment in L2 inflection: Performance similarities between nonnative and native speakers,’ Lingua 120: 901–31. Hwang, H. and K. Steinhauer. 2011. ‘Phrase length matters: The interplay between implicit prosody and syntax in Korean ‘garden path’ sentences,’ Journal of Cognitive Neuroscience 23/ 11: 3555–75. Johnson, J. S. and E. L. Newport. 1989. ‘Critical period effects in second language learning: The influence of maturational state on the acquisition of English as a second language,’ Cognitive Psychology 21: 60–99. Kaan, E., A. Harris, E. Gibson, and P. Holcomb. 2000. ‘The P600 as an index of syntactic integration difficulty,’ Language and Cognitive Processes 15: 159–201. Kasparian, K., F. Vespignani, and K. Steinhauer. 2013. ‘My Italian is not what it used to be: Investigating the neural correlates of L1 attrition and late L2 acquisition,’ Eighth International Workshop on Neurobilingualism, NBL, Groningen, The Netherlands, August. Kim, A. E. and P. M. Gilley. 2013. ‘Neural mechanisms of rapid sensitivity to syntactic anomaly,’ Frontiers in Psychology 4: 45. Kotz, S. A. 2009. ‘A critical review of ERP and fMRI evidence on L2 syntactic processing,’ Brain and Language 109/2: 68–74. Kutas, M. and S. A. Hillyard. 1980. ‘Reading senseless sentences: Brain potentials reflect semantic incongruity,’ Science 207: 203–5. Lenneberg, E. H. 1967. Biological Foundations of Language. Wiley. Mayberry, R. I. and E. Lock. 2003. ‘Age constraints on first versus second language


predict brain responses during written sentence processing: The contribution of cognitive styles in approaching the eLAN debate. Bowden, H. W., K. Steinhauer, C. Sanz, and M. T. Ullman. 2013. ‘Native-like brain processing of syntax can be attained by university foreign language learners,’ Neuropsychologia 51/ 13: 2492–511. DeKeyser, R. M. 2012. ‘Age effects in second language learning’ in S. M. Gass and A. Mackey (eds): The Routledge Handbook of Second Language Acquisition. Routledge, pp. 442–60. Dikker, S., H. Rabagliati, T. A. Farmer, and L. Pylkka¨nen. 2010. ‘Early Occipital Sensitivity to syntactic category is based on form typicality,’ Psychological Science 21/5: 629–34. Dikker, S., H. Rabagliati, and L. Pylkkanen. 2009. ‘Sensitivity to syntax in the visual cortex,’ Cognition 110: 293–321. Friederici, A. D. 2002. ‘Towards a neural basis of auditory sentence processing,’ Trends in Cognitive Sciences 6: 78–84. Friederici, A. D., K. Steinhauer, and E. Pfeifer. 2002. ‘Brain signatures of artificial language processing: evidence challenging the critical period hypothesis,’ Proceedings of the National Academy of Sciences 99: 529–34. Gillon Dowens, M., T. Guo, J. Guo, H. Barber, and M. Carreiras. 2011. ‘Gender and number processing in Chinese learners of Spanish—Evidence from event related potentials,’ Neuropsychologia 49: 1651–9. Gillon Dowens, M., M. Vergara, H. Barber, and M. Carreiras. 2010. ‘Morphosyntactic processing in late second language learners,’ Journal of Cognitive Neuroscience 22: 1870–87. Goodluck, H. and E. Zweig. 2013. ‘Introduction: Formal vs. processing explanations of linguistic phenomena,’ Language and Cognitive Processes 28/1–2: 1–8. Green, D. W. 2003. ‘The neural basis of the lexicon and the grammar in L2 acquisition: The convergence hypothesis’ in R. van Hout, A. Hulk, F. Kuiken, and R. Towell (eds): The Interface Between Syntax and the Lexicon in Second Language Acquisition. John Benjamins, pp. 197–218. Hahne, A 2001. ‘What’s different in second-language processing? Evidence from event-related

415


second language learners rely on the same prosodic cues in syntactic parsing as native speakers,’ Neuroscience Letters 557: 107–11. Osterhout, L. and P. J. Holcomb. 1992. ‘Event-related brain potentials elicited by syntactic anomaly,’ Journal of Memory and Language 31: 785–806. Osterhout, L., J. McLaughlin, I. Pitka¨nen, C. Frenck-Mestre, and N. Molinaro. 2006. ‘Novice learners, longitudinal designs, and event-related potentials: A means for exploring the neurocognition of second language processing,’ Language Learning 56(Suppl. 1), 199–230. Pakulak, E. and H. J. Neville. 2011. ‘Maturational constraints on the recruitment of early processes for syntactic processing,’ Journal of Cognitive Neuroscience 23/10: 2752–65. Pallier, C., S. Dehaene, J. B. Poline, D. Lebihan, A. M. Argenti, E. Dupoux, J. Mehler, U. Inserm, S. Hospitalier, F. Joliot, and P. D. G. Leclerc. 2003. ‘Brain imaging of language plasticity in adopted adults: Can a second language replace the first?,’ Cerebral Cortex 13/2: 155–61 Patel, A. D., E. Gibson, J. Ratner, M. Besson, and P. J. Holcomb. 1998. ‘Processing syntactic relations in language and music: an eventrelated potential study,’ Journal of Cognitive Neuroscience 10/6: 717–33. Pauker, E., I. Itzhak, S. R. Baum, and K. Steinhauer. 2011. ‘Effects of cooperating and conflicting prosody in spoken English garden path sentences: ERP evidence for the boundary deletion hypothesis,’ Journal of Cognitive Neuroscience 23/10: 2731–51. Reber, A. S. 1967. ‘Implicit learning of artificial grammars,’ Journal of Verbal Learning and Verbal Behavior 6: 855–63. Reichle, R. V. and D. Birdsong. 2014. ‘Processing focus structure in L1 and L2 French. L2 proficiency effects in ERPs,’ Studies in Second Language Acquisition. doi:101017/ S0272263113000594. Rossi, S., M. F. Gugler, A. D. Friederici, and A. Hahne. 2006. ‘The impact of proficiency on syntactic second-language processing of German and Italian: evidence from eventrelated potentials,’ Journal of Cognitive Neuroscience 18: 2030–48. Royle, P., J. E. Drury, and K. Steinhauer. 2013. ‘ERPs and task effects in the auditory processing of gender agreement and semantics in French,’ The Mental Lexicon 8/2: 216–44.


acquisition: Evidence for linguistic plasticity and epigenesis,’ Brain and Language 87/3: 369–84. McLaughlin, J., L. Osterhout, and A. Kim. 2004. ‘Neural correlates of second-language word learning: minimal instruction produces rapid change,’ Nature Neuroscience 7: 703–4. McLaughlin, J., D. Tanner, I. Pitka¨nen, C. Frenck-Mestre, K. Inoue, G. Valentine, and L. Osterhout. 2010. ‘Brain potentials reveal discrete stages of L2 grammatical learning,’ Language Learning 60: 123–50. Molnar, M., L. Polka, S. Baum, and K. Steinhauer. 2014. ‘Learning two languages from birth shapes pre-attentive processing of vowel categories: Electrophysiological correlates of vowel discrimination in monolinguals and simultaneous bilinguals,’ Bilingualism: Language and Cognition 17: 526–41. Morgan-Short, K., C. Sanz, K. Steinhauer, and M. T. Ullman. 2010. ‘Second language acquisition of gender agreement in explicit and implicit training conditions: An eventrelated potential study,’ Language Learning 60: 154–93. Morgan-Short, K., K. Steinhauer, C. Sanz, and M. T. Ullman. 2012. ‘Explicit and implicit second language training differentially affect the achievement of native-language brain patterns,’ Journal of Cognitive Neuroscience 24/4: 933–47. Mueller, J. L., A. Hahne, Y. Fujii, and A. D. Friederici. 2005. ‘Native and nonnative speakers’ processing of a miniature version of Japanese as revealed by ERPs,’ Journal of Cognitive Neuroscience 17: 1229–44. Na¨a¨ta¨nen, R., P. Paavilainen, T. Rinne, and K. Alho. 2007. ‘The mismatch negativity (MMN) in basic research of central auditory processing: A review,’ Clinical Neuropsychology 118: 2544–90. Newman, A. J., M. T. Ullman, R. Pancheva, D. Waligura, and H. Neville. 2007. ‘An ERP study of regular and irregular English past tense inflection,’ NeuroImage 34/1: 435–45. Newport, E., D. Bavelier, and H. Neville. 2001. ‘Critical thinking about critical period: Perspectives on a critical period for language acquisition’ in E. Dupoux (ed.): Language, Brain and Cognitive Development: Essays in Honor of Jacques Mehler. MIT Press, pp. 481–502. Nickels, S., B. Opitz, and K. Steinhauer. 2013. ‘ERPs show that classroom-instructed late

K. STEINHAUER

Tanner, D., K. Inoue, and L. Osterhout. 2014. ‘Brain-based individual differences in on-line L2 grammatical comprehension,’ Bilingualism: Language and Cognition 17: 277–93. Tanner, D., J. McLaughlin, J. Herschensohn, and L. Osterhout. 2013. ‘Individual differences reveal stages of L2 grammatical acquisition: ERP evidence,’ Bilingualism: Language and Cognition 16/2: 367–82. Thierry, G. and Y. J. Wu. 2007. ‘ChineseEnglish bilinguals reading English hear Chinese,’ Journal of Neuroscience 30: 7646–51. Ullman, M. T. 2001. ‘The neural basis of lexicon and grammar in first and second language: The declarative/procedural model,’ Bilingualism: Language and Cognition 4: 105–22. Weber-Fox, C. M. and H. J. Neville. 1996. ‘Maturational constraints on functional specializations for language processing: ERP and behavioral evidence in bilingual speakers,’ Journal of Cognitive Neuroscience 8: 231–56. White, E. J., F. Genesee, and K. Steinhauer. 2012. ‘Brain responses before and after intensive second language learning: proficiency based changes and first language background effects in adult learners,’ PLoS One 7/12: e52318. White, E., F. Genesee, L. White, E. King, and K. Steinhauer. 2006. ‘Factors contributing to native-like processing of a late-acquired second language: Evidence from event-related brain potentials,’ Boston University Conference on Language Development, Boston, November. Wong, P., K. Morgan-Short, and M. Ettlinger. 2012. ‘Linking neurogenetics and individual differences in language learning: The dopamine hypothesis,’ Cortex 48/9: 1091–02.


Sakai, K. L. 2005. ‘Language acquisition and brain development,’ Science 310/5749: 815–9. Steinhauer, K 2003. ‘Electrophysiological correlates of prosody and punctuation,’ Brain and Language 86/1: 142–64. Steinhauer, K. and J. F. Connolly. 2008. ‘Event-related potentials in the study of language’ in B. Stemmer and H. A. Whitaker (eds): Handbook of the Neuroscience of Language. Elsevier, pp. 91–104. Steinhauer, K. and J. E. Drury. 2012. ‘On the early left-anterior negativity (ELAN) in syntax studies,’ Brain and Language 120/2: 135–62. Steinhauer, K., J. E. Drury, N. Bourguignon, and K. Kasparian. 2010. ‘L1 grammar influences L2 processing: ERP evidence of transfer effects,’ Journal of Cognitive Neuroscience (Suppl. 2010). Steinhauer, K., A. D. Friederici, and K. Alter. 1999. ‘Brain potentials indicate immediate use of prosodic cues in natural speech processing,’ Nature Neuroscience 2: 191–6. Steinhauer, K., E. White, and J. E. Drury. 2009. ‘Temporal dynamics of late second language acquisition: Evidence from event-related brain potentials,’ Second Language Research 25/1: 13–41. Steinhauer, K., E. J. White, and F. Genesee. 2013. Proficiency and first language transfer effects in a late acquired second language: ERP evidence from a balanced word-category violation paradigm (in preparation). Stowe, L. A. and L. Sabourin. 2005. ‘Imaging the processing of a second language: Effects of maturation and proficiency on the neural processes involved,’ International Review of Applied Linguistics 43: 329–54. Tanner, D. and J. G. Van Hell. 2014. ‘ERPs reveal individual differences in morphosyntactic processing,’ Neuropsychologia 56: 289–301.

417