Abstract

EEG studies employing time–frequency analysis have revealed changes in theta and alpha power in a variety of language and memory tasks. Semantic and syntactic violations embedded in sentences evoke well-known ERPs, but little is known about the oscillatory responses to these violations. We investigated oscillatory responses to both kinds of violations, while monolingual and bilingual participants performed an acceptability judgment task. Both violations elicited power decreases (event-related desynchronization, ERD) in the 8–30 Hz frequency range, but with different scalp topographies. In addition, semantic anomalies elicited power increases (event-related synchronization, ERS) in the 1–5 Hz frequency band. The 1–5 Hz ERS was strongly phase-locked to stimulus onset and highly correlated with time domain averages, whereas the 8–30 Hz ERD response varied independently of these. In addition, the results showed that language expertise modulated 8–30 Hz ERD for syntactic violations as a function of the executive demands of the task. When the executive function demands were increased using a grammaticality judgment task, bilinguals but not monolinguals demonstrated reduced 8–30 Hz ERD for syntactic violations. These findings suggest a putative role of the 8–30 Hz ERD response as a marker of linguistic processing that likely represents a separate neural process from those underlying ERPs.

INTRODUCTION

Electroencephalography (EEG) is a popular technique for investigating the neural correlates of language processing. Most commonly, EEG data are analyzed using the ERP technique, in which electrical signals are averaged across trials time-locked to an event such as stimulus presentation or response onset. For instance, semantically unexpected words (e.g., He likes his coffee with cream and dog) elicit the N400 response, a negative-going waveform peaking about 400 msec after word onset, maximal over midline central-parietal electrodes (Kutas & Federmeier, 2000; Osterhout & Nicol, 1999; Kutas & Hillyard, 1980). Syntactic processing has been associated with the early left anterior negativity (about 150–200 msec poststimulus) and the left anterior negativity (about 400 msec poststimulus). The early left anterior negativity has been observed to phrase structure violations or word category errors (Hahne & Friederici, 1999; Friederici, Pfeifer, & Hahne, 1993), whereas left anterior negativity has been observed to a variety of syntactic violations, including morphological agreement violations (Coulson, King, & Kutas, 1998; Münte, Heinze, & Mangun, 1993). In addition, syntactic anomalies such as inappropriately inflected verbs (e.g., The cat will eating the food) elicit the P600, a positive-going waveform with a longer latency than the N400 (∼600 msec) and a slightly more posterior central-parietal distribution (Osterhout & Mobley, 1995; Osterhout & Holcomb, 1992, 1993). The P600 has also been observed for syntactically complex (Kaan, Harris, Gibson, & Holcomb, 2000) or ambiguous structures (Osterhout, Holcomb, & Swinney, 1994) and may index reanalysis and repair of the syntactic structure after a violation has been detected (Friederici, Hahne, & Saddy, 2002; Osterhout & Holcomb, 1992). In addition, the amplitude and latency of these ERPs have been shown to be modulated by subject-specific attributes such as language proficiency (Newman, Tremblay, Nichols, Neville, & Ullman, 2012; Pakulak & Neville, 2010) and age of acquisition (Sanders, Weber-Fox, & Neville, 2008; Weber-Fox & Neville, 1999).

Although the ERP method has been extensively utilized in neurolinguistic research, ERPs are only sensitive to neural activity that is phase-locked to the event onset and ignore non-phase-locked activity that is cancelled out by the averaging procedure. As ERP signals depend on precise time-locking to a stimulus, neural events that are delayed or simply more variable in latency generate a smaller signal in the time domain average (Mouraux & Iannetti, 2008) because the peaks and troughs of voltage perturbations do not line up precisely across trials. Therefore, in cases where one group (such as a clinical population) exhibits a decreased ERP response (Kielar, Meltzer-Assher, & Thompson, 2012; Kawohl et al., 2010; Swaab, Brown, & Hagoort, 1997), it is difficult to distinguish between reduced neural activity and greater temporal variability using this technique.

An attractive alternative approach is time–frequency analysis of the EEG signal. This technique reveals changes in the amplitude of ongoing oscillations, induced by behavioral events (Mouraux & Iannetti, 2008; Pfurtscheller & Lopes da Silva, 1999). Each individual trial is subjected to a moving window spectral analysis technique, such as the short-time Fourier transform or the wavelet transform, resulting in a two-dimensional data matrix on each trial, consisting of spectral power across time points and frequencies. These power values are then averaged across trials and usually normalized as deviations from power in a prestimulus baseline period. One version of this approach, adopted in this study, is known as the event-related spectral perturbation (ERSP), a term that encompasses both increases and decreases in power, termed event-related synchronization (ERS) and desynchronization (ERD), respectively. Although ERSP responses are induced by a stimulus in a particular time period (and frequency range), they do not depend on precise phase-locking of the induced amplitude change across trials and may therefore be more sensitive to signal changes that exhibit more temporal variability across trials and participants. Thus, the basic characterization of the ERSP responses to semantic and syntactic violations is an important first step in applying ERSP analysis to the study of the differences in linguistic processing between groups.

Thus far, only a handful of ERP studies have examined time–frequency modulations in linguistic violation paradigms. A rather heterogeneous pattern of results has emerged, presumably owing to considerable differences in experimental design across studies. For semantic anomalies that evoke an N400 response, studies have reported both ERS and ERD in the 4–8 Hz theta band (ERS: Davidson & Indefrey, 2007; Hald, Bastiaansen, & Hagoort, 2006; ERD: Allefeld, Frisch, & Schlesewsky, 2005) as well as ERD in the 8–12 Hz alpha band (Willems, Oostenveld, & Hagoort, 2008). For syntactic anomalies, studies have reported theta power ERS (Roehm, Schlesewsky, Bornkessel, Frisch, & Haider, 2004; Bastiaansen, van Berkum, & Hagoort, 2002a) and alpha and beta ERD (Bastiaansen, Magyari, & Hagoort, 2009; Davidson & Indefrey, 2007). Few studies have examined both semantic and syntactic violations within the same paradigm (cf., Davidson & Indefrey, 2007), and still fewer have used the ERSP technique to study differences in linguistic processing across tasks and groups, despite its potential advantages.

To investigate the similarities and differences in oscillatory responses related to processing semantic and syntactic information, we applied time–frequency analysis to a multifactorial EEG data set from a previous ERP study (Moreno, Bialystok, Wodniecka, & Alain, 2010), which examined responses to sentence-embedded semantic and syntactic anomalies. In addition, the effects of these linguistic anomalies were examined in two different tasks and two groups of participants, namely bilinguals and monolinguals. Differences in linguistic processing between these two groups are the subject of a rich behavioral literature and therefore present a good test case to evaluate the potential of the ERSP technique to reveal differences in linguistic processing across groups.

For example, bilinguals often show an advantage on tasks requiring control of attention and cognitive flexibility (Moreno, Rodrigues-Fornells, & Laine, 2008; Bialystok, Craik, & Ryan, 2006; Bialystok, Klein, Craik, & Viswanathan, 2004; Bialystok & Martin, 2004; Jackson, Swainson, Cunnington, & Jackson, 2001). Furthermore, electrophysiological studies have reported differences related to bilingualism in ERP patterns for processing of semantic and syntactic information (e.g., Moreno & Kutas, 2005; Neville & Weber-Fox, 1996). In some studies, the N400 effect has been found to be delayed in bilinguals for their second, less dominant language (Moreno & Kutas, 2005; Hahne, 2001; Neville & Weber-Fox, 1996; Ardal, Donald, Meuter, Muldrew, & Luce, 1990). Other studies reported either a delayed or significantly reduced P600 effect in bilinguals (Hahne, 2001; Hahne & Friederici, 2001; Neville & Weber-Fox, 1996). However, Osterhout and colleagues (Osterhout et al., 2008; Kotz, Holcomb, & Osterhout, 2007) found similar P600 responses elicited by syntactic violations in monolinguals and bilinguals.

In this study, we reanalyzed the EEG data from the Moreno et al. (2010) study using a time–frequency analysis. There were two main goals of this study. The first was to characterize the nature of the ERSP responses in this paradigm. This experiment featured both semantic and syntactic anomalies occurring in the same task, the same part of speech, and the same position in the sentence. Given the heterogeneous nature of the findings in previous studies of ERSP in violation paradigms, this data set offers a good opportunity to compare the brain's response to both kinds of anomalies under similar conditions. We sought to establish the frequencies that are modulated by linguistic violations, the timing of the responses, the directionality (power increase or decrease), and the scalp topography. In addition, we directly investigated the relationship between oscillatory activity and time-locked responses present in the ERP signal evoked by the same semantic and syntactic anomalies.

The second goal was to compare the strength of the ERSP response across tasks and groups to see if the previous findings with the ERP technique are replicated using ERSP analysis. Moreno et al. (2010) found equivalent responses to semantic and syntactic anomalies for bilinguals and monolinguals on the acceptability judgment task, but different responses on the grammaticality judgment task, consistent with the behavioral advantage for bilinguals on the latter task. We tested whether a similar dissociation was present in the ERSP data and whether the directionality of the effects matched that seen in the ERP data. By comparing ERSP responses across groups and tasks on an established paradigm that distinguishes these groups, we hope to validate the use of ERSP as a routine tool for assessing differences in linguistic processing across other groups of participants, particularly in clinical populations.

METHODS

Participants

EEG data were obtained from 14 healthy English monolingual (mean age = 23.06 ± 3.1 years, seven men) and 14 bilingual (mean age = 23.5 ± 4.5 years, two men) university students. Monolingual participants were all born and raised in either Canada or the United States. Bilingual participants were born in Canada (4), Russia (1), Romania (1), or Israel (8). Twelve of the 14 bilingual participants learned their second language (L2) before the age of 12 (mean = 6), and two participants learned English later in life, one at 15 years old and the second at 14 years old. On average, bilinguals spoke English at home 26% of the time and at work 85% of the time. Similarly, bilinguals heard English at home 33% of the time and at work 89% of the time. One bilingual reported to have English as first language, and six bilinguals considered English as their dominant language. Their average proficiency self-ratings on a 0–100% scale (100% corresponding to native proficiency) was 97.1% for dominant and 81.9% for nondominant language. All participants reported normal vision and provided informed consent before the start of the study, in line with the ethical guidelines outlined by the University of Toronto and Baycrest Centre for Geriatric Care. More detailed demographic and behavioral profiles for each group are presented in Moreno et al. (2010).

Materials

The materials are described in detail in Moreno et al. (2010), and examples are presented in Table 1. One hundred twenty sentence frames, originally used in Osterhout and Nicol (1999), were used to construct sentences that were correct, grammatically incorrect but semantically acceptable, and semantically anomalous but grammatically correct. A fourth condition containing words that were both semantically and syntactically anomalous was included, but treated as fillers and not analyzed, because the semantic and syntactic effects could not be unambiguously isolated. The responses to semantically or syntactically anomalous words were compared with nonanomalous words in the same sentence position. The semantically anomalous sentences were created by introducing a mismatch in animacy between the verb and the agent of the sentence (e.g., computer–paint), creating a selectional restriction violation. The syntactically incorrect structures were formed by introducing verb tense violations (e.g., will lasting). The correct and semantically inconsistent critical words were matched on frequency (semantically acceptable: mean = 96, semantically anomalous: mean = 70, p > .2; Kučera & Francis, 1967) and length (semantically acceptable: mean letters = 4.94, semantically anomalous: mean letters = 4.52, p > .03). These experimental materials were used to create three stimulus lists. Each list contained 30 exemplars of each of the four experimental sentence types. Items were counterbalanced between lists such that only one version of each sentence was presented on a given list. Each participant was presented with one list of 120 sentences for the grammaticality task and a separate list of 120 sentences for the acceptability task. Sentence frames were repeated between the two tasks, but appeared in different conditions, and lists were counterbalanced across participants.

Table 1. 

Examples of Sentences Used in the Acceptability and Grammaticality Tasks

Semantically CorrectSemantically Incorrect
Syntax correct A new computer will last for many years A new computer will paint for many years 
Syntax incorrect A new computer will lasting for many years A new computer will painting for many years 
Semantically CorrectSemantically Incorrect
Syntax correct A new computer will last for many years A new computer will paint for many years 
Syntax incorrect A new computer will lasting for many years A new computer will painting for many years 

Procedure

Cognitive Assessment

Before EEG recording, participants completed a battery of psychological tests including the Language and Social Background Questionnaire, Peabody Picture Vocabulary Test (PPVT)-III (Dunn & Dunn, 1997), Cattell Culture Fair Intelligence (Cattell, 1957), and Corsi Block tests (used after Fischer, 2001). The purpose of these tests was to assess the similarity of the two participant groups on vocabulary (PPVT), intelligence (Cattell), and working memory (Corsi Block test). The cognitive assessment session lasted approximately 45 min.

EEG Recording during Sentence Judgment Tasks

The EEG recording was conducted in a soundproof room. On each trial, participants were presented with a 500-msec fixation cross, followed by a sentence. Sentences were presented word by word, with each word appearing in the center of the screen for 300 msec. Words were separated by a blank screen interval of 350 msec. A 1450-msec blank screen interval was inserted at the end of each sentence, after which a response prompt appeared. In this study, the processing demands were manipulated by varying task instructions. Participants completed both a standard acceptability judgment task and the grammaticality judgment paradigm previously used by Bialystok and colleagues (Bialystok & Majumder, 1998; Bialystok, 1986). In the acceptability task, the participants were instructed to answer “yes” if they judged the sentence to be acceptable and “no” if the sentence was either semantically or grammatically incorrect. In the grammaticality task, participants responded “yes” if the sentence was grammatically correct and “no” if it was grammatically incorrect, irrespective of meaning. Thus, in the grammaticality task, sentences such as “A new computer will paint for many years” would be judged as correct, whereas the same sentence will be judged as incorrect in the acceptability task. These manipulations created a higher level of attentional control in the grammaticality task, because it was necessary to attend to only one dimension in spite of conflicting information from the other dimension. On the basis of prior behavioral studies (Bialystok & Majumder, 1998; Bialystok, 1986), in the grammaticality task, a higher degree of executive control is required to attend to syntax when the meaning is inconsistent. It is more difficult to approve the grammatically correct sentences and ignore the conflicting semantic information. Participants used two different hands to respond. They indicated their responses by pressing one button for “yes” and another one for “no,” with the left- and right-hand assignments counterbalanced across participants. For all participants, the grammaticality task was performed first.

Participants had a brief break (<5 min) and then performed the acceptability task. For the original ERP study of Moreno et al. (2010), this order was chosen to maximize sensitivity to group differences in the grammaticality task, which had previously been shown to elicit behavioral differences between monolinguals and bilinguals (Bialystok & Majumder, 1998; Bialystok, 1986). In this study, for our initial characterization of oscillatory responses, we focus on the acceptability task, because it has been the most commonly used task in linguistic violation paradigms. In addition, in the original Moreno et al. (2010) paper, the two language groups showed comparable levels of neural activity on the acceptability task, as assessed with ERPs.

EEG Recording and Analysis

Electrophysiological signals were recorded continuously from an array of 64 electrodes with a bandpass of 0.05–100 Hz and a sampling rate of 500 Hz using a NeuroScan Synamps2 system (Compumedics, El Paso, TX). Electrode Cz served as a reference during recording, but all electrodes were re-referenced to an average reference before spectral analysis (Picton et al., 2000). Vertical and horizontal eye movements were monitored using electrodes placed at the superior and inferior orbit and in the outer canthi.

The data preprocessing and spectral analyses were done using EEGlab software (version 9.0.4.5) running in the Matlab 2010 (version 7.6) environment. The continuous data were first segmented into individual whole sentences. Trials were reviewed visually, and those with obvious artifacts were removed, along with electrodes that were consistently bad throughout a recording. Next, the data were subjected to independent component analysis (ICA; Jung et al., 2000) to remove artifacts that occurred throughout the data set. Scalp projections and time courses of all components were reviewed by two authors (AK and JM), who came to a consensus on which components to remove. Components that were judged to comprise eyeblinks, eye movements, and facial muscle activity were selected for removal. The cleaned continuous EEG data were then average-referenced and segmented into individual epochs around the critical word in each condition (semantic violations, syntactic violations, and control), from −1 sec to +2 sec. Epochs were baseline-corrected in the interval −200 to 0 msec. Data sets from the acceptability and grammaticality tasks were analyzed separately, with each task data set undergoing its own ICA decomposition. Because the scalp impedances were checked and electrodes were “touched up” in between the two tasks, it would be suboptimal to combine the two data sets for ICA because of possible changes in the scalp distribution of components.

Time–frequency analysis was performed on single-trial epochs, using a moving window short-time Fourier transform with 200 overlapping time windows per trial. The length of the time window in the spectrogram analysis was 0.5120 sec (256 samples at a sampling rate of 500 Hz). All values were averaged over trials of each specific condition. The average log-power in the baseline period for all three conditions was used as a common baseline, subtracted from log-power at each time–frequency point, yielding the measure conventionally known as “event-related spectral perturbation” or ERSP (Makeig, 1993). This procedure ensured that the same baseline power values were used across all conditions; thus, any differences between conditions could not be because of the differences in the baseline. We would like to note that, because a relatively long time window was used to capture lower-frequency activity, the power estimates from the baseline interval may include some influence from early evoked activity occurring after word onset. However, that would play only a minimal role in the baseline calculation given the use of Hanning windows for the short-time Fourier transform and would not affect relative differences between conditions, only the absolute values of the power estimates. Because the critical responses that distinguish the conditions begin around 400–500 msec, it is unlikely that the window parameters made a major difference in the estimated magnitude of the power changes.

Cluster Analysis Procedure

Because we did not have a priori information on the frequency range and time window in which linguistic anomalies elicit ERSP responses in this paradigm, we first needed to characterize the responses in an unbiased manner before testing for differences related to region, task, or group. To accomplish this, we used cluster analysis routines (Maris & Oostenveld, 2007; Maris, 2004) implemented in the Fieldtrip toolbox (fieldtrip.fcdonders.nl/; accessed Dec 18, 2008). Applied to time–frequency data, this form of cluster analysis identifies responses that differ significantly between two conditions in clusters that span adjacent time points, frequency bins, and electrodes. Permutation tests are used to establish the statistical significance of clusters as a whole, correcting for multiple comparisons.

For initial characterization of the ERSP response using cluster analysis, we elected to analyze only the acceptability task, as it is the most commonly used task in N400/P600 paradigms, and previous behavioral studies have shown that monolinguals and bilinguals complete it in a similar manner (Bialystok, 1986, 1988, 2001). We combined data from the bilingual and monolingual participants, thus yielding a sample size of 28 to increase power. The goal of this was to first identify the peak time and frequency range of the ERSP responses to semantic and syntactic violations before submitting ERSP data to further analysis of effects of task conditions and bilingualism. In each violation condition, we contrasted the response to the anomalous words with the response to the correct version in the control condition, using a paired t test across participants. We used a statistical threshold of p < .025 for individual time–frequency points at single electrodes and p < .05 for corrected cluster-level significance. Comparisons were conducted in the frequency range of 0.05–50 Hz and in the time range of −2 to +3 sec relative to the onset of the critical word.

On the basis of the results of the cluster analysis (see Results section below), we selected frequency ranges of 1–5 Hz and 8–30 Hz within a time window of 500–1500 msec after critical word onset, for both semantic and syntactic violations. ERSP values were averaged in these windows for each electrode, violation condition, and participant, separately for each task. The resulting averages were exported to SPSS for statistical analysis. To assess differences in scalp topography between conditions, we grouped together electrodes into 11 a priori regions, as shown in Figure 1. For the acceptability task, we conducted repeated-measures ANOVAs on difference scores (subtracting the response to the control condition from violations) with within-subject factors of Violation condition (semantic vs. syntactic) and Electrode region. For the within-subject factor of Region having more than two conditions, the Huyn–Feldt correction for nonsphericity was applied. To investigate the effects of bilingualism on violation type in each task, we conducted ANOVAs on an ROI consisting of five frontal electrodes (i.e., Fz, F1, F2, F3, F4) that were highly responsive to syntactic violations in the initial cluster analysis of the acceptability task across all 28 participants. For each task, ANOVAs were computed using Group as a between-subject factor and Condition (syntactic violation and semantic violation, with the control condition subtracted from both) as within-subject factors.

Figure 1. 

The 64-channel montage, with grouping of the electrodes into 11 regions. The data from the electrodes indicated by the solid lines were used in the statistical analyses: CC (CZ, CPZ), PP (PZ, POZ), FC (FZ, FCZ), LF (F1, F3, F5, F7), RF (F2, F4, F6, F8), LFC (FC1, FC5); RFC (FC2, FC6), LCP (C1, C3, C5, CP1, CP5), RCP (C2, C4, C6, CP2, CP6), LP (P1, P3, P5, P7); RP (P2, P4, P6, P8).

Figure 1. 

The 64-channel montage, with grouping of the electrodes into 11 regions. The data from the electrodes indicated by the solid lines were used in the statistical analyses: CC (CZ, CPZ), PP (PZ, POZ), FC (FZ, FCZ), LF (F1, F3, F5, F7), RF (F2, F4, F6, F8), LFC (FC1, FC5); RFC (FC2, FC6), LCP (C1, C3, C5, CP1, CP5), RCP (C2, C4, C6, CP2, CP6), LP (P1, P3, P5, P7); RP (P2, P4, P6, P8).

RESULTS

Behavioral Data

The group scores for the background language and cognitive measures are presented in Table 2. There were no significant group differences in age, education, or in performance on the Corsi and Cattell tests (ps > .05). The two language groups differed on the PPVT scores, F(1, 27) = 5.36, p < .005, with monolinguals obtaining higher scores than bilinguals.

Table 2. 

Means (SD) for Demographic and Background Cognitive Measures in Two Language Groups: PPVT-III (Dunn & Dunn, 1997), Cattell Culture Fair Intelligence (Cattell, 1957), and Corsi Block Tests (Working Memory Measure, Used after Fischer, 2001)

Age (years)Education (years)PPVT-IIICorsi ScoreCattell
Monolinguals 23.6 (3.1) 17.3 (1.8) 109.2 (5.7) 17.4 (2.4) 117.1 (9.8) 
Bilinguals 23.5 (4.5) 16.8 (3.4) 103 (8.8) 16.6 (2.4) 115.6 (12.4) 
Age (years)Education (years)PPVT-IIICorsi ScoreCattell
Monolinguals 23.6 (3.1) 17.3 (1.8) 109.2 (5.7) 17.4 (2.4) 117.1 (9.8) 
Bilinguals 23.5 (4.5) 16.8 (3.4) 103 (8.8) 16.6 (2.4) 115.6 (12.4) 

Accuracy rates on the grammaticality and acceptability judgment tasks (performed during EEG acquisition) were analyzed with repeated-measures ANOVA with Condition as a within-subject factor and Group as a between-subject factor. Before the ANOVA, the proportion correct for each participant and condition was subjected to the arcsine-square-root transformation (Sokal & Rohlf, 1995), with values of 1 replaced by [1 − (1/(4n)], with n equal to the number of trials in each condition (30). The untransformed accuracy scores and standard deviations for the two groups are presented in Table 3. In the acceptability task, there was a significant main effect of group, indicating that monolinguals were significantly more accurate than bilinguals, F(1, 26) = 4.66, p < .05. The main effect of Condition was also significant, F(2, 52) = 14.98, p < 0 .001, reflecting overall higher accuracy for correct and syntactically incorrect sentences than for semantically anomalous structures (both ps < .001). There were no significant differences between correct and syntactically anomalous sentences (p > .05), and the Group × Condition interaction was not significant, F < 1. In the grammaticality task, there were no significant main effects of Group, F(1, 26) = 1.755, p > .05, or condition, F < 1. The Group × Condition interaction was not significant (F < 1), indicating that in the grammaticality task monolinguals and bilinguals did not differ in their ability to judge the syntax of the sentences.

Table 3. 

The Mean Accuracy (Standard Deviations) Judgment Scores in Percent Correct for Each Language Group in the Acceptability and Grammaticality Tasks

Language GroupConditionAcceptability Mean (SD)Grammaticality Mean (SD)
Monolinguals Correct 95 (2) 96 (1) 
Semantic errors 87 (3) 95 (2) 
Syntactic errors 95 (2) 96 (2) 
Bilinguals Correct 90 (2) 95 (1) 
Semantic errors 81 (3) 93 (2) 
Syntactic errors 93 (2) 93 (2) 
Language GroupConditionAcceptability Mean (SD)Grammaticality Mean (SD)
Monolinguals Correct 95 (2) 96 (1) 
Semantic errors 87 (3) 95 (2) 
Syntactic errors 95 (2) 96 (2) 
Bilinguals Correct 90 (2) 95 (1) 
Semantic errors 81 (3) 93 (2) 
Syntactic errors 93 (2) 93 (2) 

Oscillatory Responses: ERSP

Illustration of ERSP Characteristics

The primary statistical analysis of ERSP responses in this experiment concerns the difference in the responses between anomalous and correct words in the same sentence position. However, as time–frequency analysis is somewhat less familiar than the ERP technique to most neuroscientists, we chose to illustrate additional characteristics of the responses measured in this study, although they are not the direct subject of statistical testing. We wish to demonstrate to the reader in a qualitative fashion how ongoing oscillations are modulated in a rapid serial visual presentation sentence comprehension task, both in correct sentences and in the presence of violations. Averaged time–frequency decompositions for one representative electrode, P5, are presented in Figure 2A. These are averaged across all 28 participants (both monolinguals and bilinguals) on data from the acceptability task. Power in the alpha and beta bands (approximately 6–25 Hz) is modulated by the periodic presentation of words in the sentence. Additionally, there is a general trend for power to decrease from the beginning to the end of the sentence. Superimposed on these two effects, linguistic violations induce a sharp drop in power that is maximal approximately 500–1500 msec after the onset of the critical word. The drop is larger for semantic violations than syntactic ones. For semantic violations, we also observe an increase in power in the delta and theta bands (approximately 1–5 Hz), which is consistent with some prior reports (Davidson & Indefrey, 2007; Hald et al., 2006). To avoid potential overlap with power increase in the theta range, we chose a lower limit of 8 Hz for our ANOVAs on differences in the alpha-beta power decrease across tasks and groups. Similarly, for delta-theta power increases, a 1–5 Hz frequency range was used for computing ANOVAs. Note that the 1–5 Hz ERS response has a slightly earlier onset and peak than the 8–30 Hz ERD on most electrodes. However, the statistical analyses (see below) indicated that significant differences in ERS between the conditions are predominantly observed in the same range as the ERD, 500–1500 msec after the onset of the critical word.

Figure 2. 

Example representations of oscillatory reactivity (ERD and ERS) in the acceptability task, combined across monolinguals and bilinguals. (A) Time–frequency maps of ERSP signal time-locked to control, semantic violations, and syntactic violations shown at one representative electrode, P5. (B) Time courses of 8–30 Hz power decreases at site P4, time-locked to the critical word onset. Peaks and troughs of power before and after the critical word are because of visual presentation of other words in the sentence. (C) Time courses of 8–30 Hz power decreases at site P5, time-locked to critical word onset. (D) Time courses of 1–5 Hz power increases at site Pz, time-locked to critical word onset.

Figure 2. 

Example representations of oscillatory reactivity (ERD and ERS) in the acceptability task, combined across monolinguals and bilinguals. (A) Time–frequency maps of ERSP signal time-locked to control, semantic violations, and syntactic violations shown at one representative electrode, P5. (B) Time courses of 8–30 Hz power decreases at site P4, time-locked to the critical word onset. Peaks and troughs of power before and after the critical word are because of visual presentation of other words in the sentence. (C) Time courses of 8–30 Hz power decreases at site P5, time-locked to critical word onset. (D) Time courses of 1–5 Hz power increases at site Pz, time-locked to critical word onset.

Figures 2B and 2C displays the time course of power fluctuations in each condition, averaged over the 8–30 Hz range, at two representative electrodes (P4 and P5). These time course plots allow the reader to easily discern the three patterns of modulation present in the 8–30 Hz range: periodic rise-and-fall driven by the stimulus presentation, an overall decreasing trend across the sentence, and a specific decrease induced by linguistic violations. In this example, P4 (right parietal) responds mainly to semantic violations, whereas P5 (left parietal) responds to both semantic and syntactic violations. Similarly, Figure 2D displays the time course of power fluctuations in the 1–5 Hz frequency band for semantic and syntactic violations at one of the central electrodes (Pz) that showed the strongest responses in this frequency range. The 1–5 Hz response is specifically modulated by semantic violations but is not periodically driven by the visual word presentation.

Cluster Analysis Results

For semantic violations versus control words, we observed two significant clusters of ERSP differences. Semantically anomalous words induced a power decrease (ERD), occurring between 500 and 1500 msec, in the frequency range of 6–30 Hz (Figure 3A). In addition, semantic violations elicited a power increase (ERS) in the time range of 500–1500 msec, in the frequency range of 1–5 Hz (Figure 3B). To simplify the display of this cluster across the dimensions of time, frequency, and scalp topography, we computed a measure that we refer to as T-sum. This is the sum of t values for all data points (time, frequency, and electrode) that are part of the cluster (Meltzer & Braun, 2011). By summing across different dimensions of the cluster, different aspects of the cluster can be highlighted. Note that this procedure of summing t values across individual dimensions is done solely for purposes of summarizing the cluster's spatiotemporal characteristics in a format convenient for display and does not play a role in the assessment of cluster significance as a whole. Figure 3A and B shows the T-sum for the clusters of ERD and ERS in response to semantic violations versus control words. At each time–frequency point, significant t values are summed across electrodes. A given point can achieve a higher t score either by having more significant electrodes or by having higher t values at the individual electrodes (generally both). Thus, the region of maximal response can be easily discerned.

Figure 3. 

Significant clusters of EEG oscillatory reactivity in the acceptability task, combined across monolinguals and bilinguals. (A) t values of 8–30 Hz ERD for semantic violations versus control words, summed across all channels at each time and frequency point and the scalp topography of the one significant cluster of ERD for semantic violations versus control words, represented as sums of t values across time and frequency points at each electrode. (B) t values of 1–5 Hz ERS for semantic violations versus control words and the scalp topography of the one significant cluster of ERS for semantic violations versus control words. (C) t values of 8–30 Hz ERD for syntactic violations versus control words and the scalp topography of the one significant cluster of ERD for syntactic violations versus control words.

Figure 3. 

Significant clusters of EEG oscillatory reactivity in the acceptability task, combined across monolinguals and bilinguals. (A) t values of 8–30 Hz ERD for semantic violations versus control words, summed across all channels at each time and frequency point and the scalp topography of the one significant cluster of ERD for semantic violations versus control words, represented as sums of t values across time and frequency points at each electrode. (B) t values of 1–5 Hz ERS for semantic violations versus control words and the scalp topography of the one significant cluster of ERS for semantic violations versus control words. (C) t values of 8–30 Hz ERD for syntactic violations versus control words and the scalp topography of the one significant cluster of ERD for syntactic violations versus control words.

For the comparison of syntactic violations versus control words, we obtained one significant cluster, in the same time range of 500–1500 msec, in the frequency range of 6–22 Hz (see Figure 3C). Thus, both kinds of violations induced ERD in largely overlapping time and frequency ranges, although the response to semantic violations was larger and extended to higher frequencies. The observed frequency range includes the commonly defined alpha band (8–12 Hz) as well as the beta band (15–30 Hz). In addition, semantic violations induced ERS in the delta and theta bands (1–5 Hz).

To illustrate the topography of these clusters, we computed T-sum values at each electrode, summing over both time and frequency dimensions. The resulting topographies of ERD and ERS for semantic violations are shown in Figure 3A and B. ERD induced by semantic violations was maximal over bilateral parietal electrodes, whereas ERS was maximal at the frontal and central sites, with additional increases observed at the temporal sites. The topography of ERD for syntactic violations is shown in Figure 3C. The distribution of the syntactic response appeared to be somewhat more left-lateralized and also stronger at bilateral fronto-central electrodes. However, the statistical significance of differences in topography cannot be tested directly from the cluster analysis, which operates on each electrode separately. Because the cluster analysis identified consistent time–frequency windows in which ERSP responses to linguistic anomalies occurred, we then conducted repeated-measures ANOVAs on ERSP values averaged within the relevant windows to formally assess differences in scalp topography and effects of bilingualism within each task.

Relationship between ERP and ERSP Measures

Because the cluster analysis revealed significant time–frequency responses in the alpha-beta (8–30 Hz) and delta-theta (1–5 Hz) frequency bands, we investigated the extent to which these ERSP responses may be driven by time-locked signals. To do that, we performed a time–frequency analysis on the ERP signal using the same procedure as was applied to compute ERD and ERS responses but applied to the signal after averaging across trials within each participant. This “ERSP of ERPs” analysis was computed for the acceptability task across all 28 participants. The analysis revealed power increases for both semantic and syntactic anomalies from 1 to 5 Hz spanning 500–1500 msec time window (see Figure 4A). These effects were the strongest over the central electrode sites and were greater for semantic than syntactic violations (see Figure 4A and B).

Figure 4. 

Representation of oscillatory activity obtained from the ERP signal using time–frequency analysis in the acceptability task, combined across monolinguals and bilinguals, compared with 1–5 Hz ERS. (A) Time–frequency maps of ERSP of ERP signals time-locked to control, semantic violations, and syntactic violations shown at one representative electrode, CPz. (B) Time courses of 1–5 Hz ERSP of ERPs at central site CPz, time-locked to the critical word onset, and the scalp topography for semantic violations versus control in 1–5 Hz band at 500–1500 msec time window. (C) Time courses of 1–5 Hz ERS at central site CPz, time-locked to the critical word onset, and the scalp topographies for semantic violations versus control at 500–1500 msec time window. (D) The topographical maps of correlations computed for semantic violation versus controls between ERSP measures (1–5 Hz ERS and 8–30 Hz ERD) and the ERSP of ERPs effects, in the 1–5 Hz frequency range. The electrodes showing significant correlations are marked with symbols (black diamonds: p < .05, uncorrected; blue circles: p < .0008, Bonferroni-corrected).

Figure 4. 

Representation of oscillatory activity obtained from the ERP signal using time–frequency analysis in the acceptability task, combined across monolinguals and bilinguals, compared with 1–5 Hz ERS. (A) Time–frequency maps of ERSP of ERP signals time-locked to control, semantic violations, and syntactic violations shown at one representative electrode, CPz. (B) Time courses of 1–5 Hz ERSP of ERPs at central site CPz, time-locked to the critical word onset, and the scalp topography for semantic violations versus control in 1–5 Hz band at 500–1500 msec time window. (C) Time courses of 1–5 Hz ERS at central site CPz, time-locked to the critical word onset, and the scalp topographies for semantic violations versus control at 500–1500 msec time window. (D) The topographical maps of correlations computed for semantic violation versus controls between ERSP measures (1–5 Hz ERS and 8–30 Hz ERD) and the ERSP of ERPs effects, in the 1–5 Hz frequency range. The electrodes showing significant correlations are marked with symbols (black diamonds: p < .05, uncorrected; blue circles: p < .0008, Bonferroni-corrected).

Correlations were computed between the ERSP measures (8–30 Hz ERD and 1–5 Hz ERS) and the ERSP of ERPs effects in the 1–5 Hz frequency range and within the 500–1500 msec time window. These analyses were computed for the acceptability task, separately for semantic and syntactic violations on the difference scores (semantic- control; syntactic-control) across the 28 participants. The correlations were computed on the average response combined across all electrodes and also separately within each of the previously defined 11 electrode clusters. The results are presented in Table 4. The analysis revealed significant correlations between the 1–5 Hz ERS and the “ERSP of ERPs” signals. For semantic violations, significant positive correlations were present when computed across all electrodes, as well as separately at the central, parietal, and frontal regions. These positive correlations reflect correspondence between power increases in 1–5 Hz frequency range for the total time–frequency response (ERS) and the portion of it that is attributable to phase-locked signals (ERSP of ERPs). For syntactic violations, significant correlations were also found at the left central parietal, left frontal, and parietal regions, but these correlations were somewhat weaker than those seen for semantic violations and did not meet the Bonferroni-corrected significance threshold for multiple tests done in 11 electrode regions. Also, correlation in the average signal across all electrodes was not significant for syntactic violations.

Table 4. 

Correlations between ERSP of ERPs and 1–5 Hz ERS, 8–30 Hz ERD and between 8–30 Hz ERD and 1–5 Hz ERS Computed for Semantic and Syntactic Anomalies in Acceptability Task, Combined across Monolinguals and Bilinguals

ConditionERSP of ERPs vs. 1–5 Hz ERSERSP of ERPs vs. 8–30 Hz ERD8–30 Hz ERD vs. 1–5 Hz ERS
RegionCorrelation (r)Significance (p)RegionCorrelation (r)Significance (p)RegionCorrelationSignificance (p)
Semantic Average 0.434 .021a Average 0.001 ns Average 0.226 ns 
CP 0.113 ns CP −0.120 ns CP 0.344 ns 
PP 0.454 .015a PP 0.169 ns PP 0.117 ns 
RCP 0.379 .047a RCP 0.126 ns RCP 0.170 ns 
LCP 0.190 ns LCP 0.090 ns LCP 0.454 0.015a 
FC 0.550 .002bc FC −0.005 ns FC −0.238 ns 
LF 0.568 .002bc LF −0.104 ns LF 0.138 ns 
RF 0.353 ns RF 0.214 ns RF 0.005 ns 
RFC 0.384 .044a RFC −0.287 ns RFC −0.230 ns 
LFC 0.345 ns LFC −0.231 ns LFC 0.252 ns 
RP 0.467 .012a RP −0.097 ns RP 0.060 ns 
LP 0.469 .012a LP 0.169 ns LP 0.340 ns 
Syntactic Average 0.116 ns Average 0.010 ns Average 0.179 ns 
CP 0.074 ns CP 0.124 ns CP 0.254 ns 
PP 0.291 ns PP 0.081 ns PP 0.058 ns 
RCP 0.144 ns RCP 0.071 ns RCP 0.169 ns 
LCP 0.451 .016a LCP 0.128 ns LCP 0.269 ns 
FC −0.062 ns FC −0.268 ns FC 0.223 ns 
LF 0.432 .022a LF −0.221 ns LF −0.066 ns 
RF 0.061 ns RF −0.006 ns RF −0.020 ns 
RFC 0.162 ns RFC −0.123 ns RFC 0.108 ns 
LFC 0.427 .023a LFC 0.086 ns LFC 0.227 ns 
RP 0.175 ns RP −0.107 ns RP 0.036 ns 
LP 0.413 .029a LP 0.048 ns LP 0.000 ns 
ConditionERSP of ERPs vs. 1–5 Hz ERSERSP of ERPs vs. 8–30 Hz ERD8–30 Hz ERD vs. 1–5 Hz ERS
RegionCorrelation (r)Significance (p)RegionCorrelation (r)Significance (p)RegionCorrelationSignificance (p)
Semantic Average 0.434 .021a Average 0.001 ns Average 0.226 ns 
CP 0.113 ns CP −0.120 ns CP 0.344 ns 
PP 0.454 .015a PP 0.169 ns PP 0.117 ns 
RCP 0.379 .047a RCP 0.126 ns RCP 0.170 ns 
LCP 0.190 ns LCP 0.090 ns LCP 0.454 0.015a 
FC 0.550 .002bc FC −0.005 ns FC −0.238 ns 
LF 0.568 .002bc LF −0.104 ns LF 0.138 ns 
RF 0.353 ns RF 0.214 ns RF 0.005 ns 
RFC 0.384 .044a RFC −0.287 ns RFC −0.230 ns 
LFC 0.345 ns LFC −0.231 ns LFC 0.252 ns 
RP 0.467 .012a RP −0.097 ns RP 0.060 ns 
LP 0.469 .012a LP 0.169 ns LP 0.340 ns 
Syntactic Average 0.116 ns Average 0.010 ns Average 0.179 ns 
CP 0.074 ns CP 0.124 ns CP 0.254 ns 
PP 0.291 ns PP 0.081 ns PP 0.058 ns 
RCP 0.144 ns RCP 0.071 ns RCP 0.169 ns 
LCP 0.451 .016a LCP 0.128 ns LCP 0.269 ns 
FC −0.062 ns FC −0.268 ns FC 0.223 ns 
LF 0.432 .022a LF −0.221 ns LF −0.066 ns 
RF 0.061 ns RF −0.006 ns RF −0.020 ns 
RFC 0.162 ns RFC −0.123 ns RFC 0.108 ns 
LFC 0.427 .023a LFC 0.086 ns LFC 0.227 ns 
RP 0.175 ns RP −0.107 ns RP 0.036 ns 
LP 0.413 .029a LP 0.048 ns LP 0.000 ns 

The analyses were computed on the mean amplitude of the difference scores as averaged across all electrode regions and separately at each of the 11 electrode regions. The significant correlations are marked in bold font. The electrodes showing significant correlations are marked (a: p < .05, uncorrected; bc: p < .004, Bonferroni corrected).

The same analysis was performed for 8–30 Hz ERD. However, the analysis revealed that there were no significant correlations between the 8–30 Hz ERD and the ERSP of ERPs measures. These results indicate that 1–5 Hz ERS, but not 8–30 Hz ERD, correlated with power increases in the 1–5 Hz frequency range present in the ERP signals. Comparison of Figure 4B and C illustrates the similarity between 1–5 Hz ERS power computed over single trials (Figure 4C) and the time–frequency characteristics of the ERP signal averaged across trials (Figure 4B). Topographical maps of the correlations at each electrode are presented in Figure 4D for the two comparisons. Electrodes showing significant correlations are highlighted at both an uncorrected level (p < .05) and a level Bonferroni-corrected for multiple comparisons across individual electrodes (p < .0008).

We also computed correlations between 8–30 Hz ERD and 1–5 Hz ERS. This analysis revealed one significant positive correlation for semantic violation limited to the left central-parietal region, but this correlation did not survive Bonferroni correction for multiple comparisons over the 11 electrode regions. No other correlations were significant, and there were no significant correlations for the syntactic anomalies.

These results indicate that participants who exhibit a strong 1–5 Hz ERS response in single trials also tend to exhibit a strong response in the time domain average or ERP, whereas no such relationship was apparent with the 8–30 Hz ERD. This pattern is consistent with a common neural process driving the expression of 1–5 Hz ERS and ERPs, distinct from that underlying the 8–30 Hz ERD.

These 1–5 Hz power increases represent phase-locked signals evoked by the violations but collapse across polarity, as both negative and positive components contain power. Thus, it is possible to observe significant ERS in the delta and theta ranges, even if ERP responses are highly variable across participants. The relationship between ERP responses to semantic violations, 1–5 Hz ERSP of ERPs, and 1–5 Hz ERS are presented in Figure 5. Figure 5A illustrates the topographical distribution of ERP responses to semantic violations versus control words. In this experiment, semantic violations evoked both negative (N400) and positive (P600) waveforms at different electrode sites, with a biphasic waveform at central sites. Figure 5B shows that 1–5 Hz ERSP of ERPs picks up signals of either polarity, both positive and negative. The strong correlations observed across participants (Figure 4) suggest that ERP components of either polarity contribute to the 1–5 Hz ERS seen in the time–frequency analyses. For a representative electrode C3, we display the time courses of ERP, 1–5 Hz ERSP of ERPs, and 1–5 Hz ERS responses in Figure 5C, D, and E, respectively. These graphs illustrate that the violation induced both negative and positive ERP components, both of which are reflected in 1–5 Hz ERS.

Figure 5. 

The relationship between ERP responses to semantic violations, 1–5 ERSP of ERPs and 1–5 Hz ERS. (A) The topographical distribution of ERP responses to semantic violations versus control words. Semantic violations evoked both negative (N400) and positive (P600) waveforms at different electrode sites, with a biphasic waveform at central sites. (B) The topographical distribution of 1–5 Hz ERSP of ERPs responses to semantic violations versus control words. 1–5 Hz ERSP of ERPs picks up signals of either polarity, both positive and negative. (C) Time courses of ERPs at C3. (D) Time courses of 1–5 Hz ERSP of ERPs at C3. (E) Time courses of 1–5 Hz ERS at C3.

Figure 5. 

The relationship between ERP responses to semantic violations, 1–5 ERSP of ERPs and 1–5 Hz ERS. (A) The topographical distribution of ERP responses to semantic violations versus control words. Semantic violations evoked both negative (N400) and positive (P600) waveforms at different electrode sites, with a biphasic waveform at central sites. (B) The topographical distribution of 1–5 Hz ERSP of ERPs responses to semantic violations versus control words. 1–5 Hz ERSP of ERPs picks up signals of either polarity, both positive and negative. (C) Time courses of ERPs at C3. (D) Time courses of 1–5 Hz ERSP of ERPs at C3. (E) Time courses of 1–5 Hz ERS at C3.

Statistical Analysis of 8–30 Hz ERD and 1–5 Hz ERS

To investigate the scalp distributions of ERD and ERS effects and to test for significant interactions between violation conditions and language groups in the two tasks, we chose to analyze ERD in the time window of 500–1500 msec and frequency range of 8–30 Hz. In addition, ERS was analyzed in the 1–5 Hz frequency range and 500–1500 msec time window. This range comprises the bulk of the significant clusters (Figure 3) and is consistent with previous studies reporting modulations of brain oscillations in this range by language tasks (Willems et al., 2008; Davidson & Indefrey, 2007). Although significant ERD does extend below 8 Hz at some electrodes, we also observed ERS up to about 6 Hz in semantic condition at certain electrodes (e.g., Figure 2A, center), so a lower bound of 8 Hz helps to avoid overlap with a potential theta-band effect of opposite directionality.

Scalp Distribution of ERD and ERS Violation Responses

8–30 Hz ERD

To characterize the scalp distribution of the ERD and ERS effects for semantic and syntactic violations, the analysis was conducted collapsing across all 28 participants (both monolinguals and bilinguals) in the acceptability task, using 11 a priori ROIs to test for differences in topography. These results are summarized in Table 5A. The repeated-measures ANOVA on the difference scores (violations − control) with Violation condition (semantic vs. syntactic) and Region as within-subject factors showed a significant main effect of Violation condition, reflecting overall larger power decreases for semantic violations compared with syntactic violations (see Figure 3A and C). The main effect of Region was also significant, as well as the Violation condition × Region interaction, indicating that effects for semantic and syntactic violations had significantly different scalp distributions. Semantic effects were distributed bilaterally over the central and parietal sites, whereas syntactic effects were weaker overall but were maximal at fronto-central and left parietal electrodes (see Figure 3A and C).

Table 5. 

Statistical Results in the 500–1500 msec Time Window

A. Effects of Violation Condition and Region in Acceptability Task
Frequency RangedfFp
8–30 Hz ERD Violation condition (semantic vs. syntactic) (1,27) 9.73 .004 
Region (10,270) 2.48 .019 
Condition × Region (10,270) 6.65 <.001 
1–5 Hz ERS Violation condition (semantic vs. syntactic) (1,27) 20.09 <.001 
Region (10,270) 2.6 .014 
Condition × Region (10,270) 2.67 .01 
 
B. Effect of Bilingualism in Frontal ROI in Grammaticality Task 
Violation Type  df F p 
Syntactic violation Sentence condition (syntactic vs. control) (1,26) 4.31 .048 
Condition × Group (1,26) 4.79 .038 
Group (monolinguals vs. bilinguals) (1,26) 1.373 ns 
Semantic violation Sentence condition (semantic vs. control) (1,26) 11.31 .002 
Condition × Group (1,26) 0.53 ns 
Group (monolinguals vs. bilinguals) (1,26) 0.262 ns 
A. Effects of Violation Condition and Region in Acceptability Task
Frequency RangedfFp
8–30 Hz ERD Violation condition (semantic vs. syntactic) (1,27) 9.73 .004 
Region (10,270) 2.48 .019 
Condition × Region (10,270) 6.65 <.001 
1–5 Hz ERS Violation condition (semantic vs. syntactic) (1,27) 20.09 <.001 
Region (10,270) 2.6 .014 
Condition × Region (10,270) 2.67 .01 
 
B. Effect of Bilingualism in Frontal ROI in Grammaticality Task 
Violation Type  df F p 
Syntactic violation Sentence condition (syntactic vs. control) (1,26) 4.31 .048 
Condition × Group (1,26) 4.79 .038 
Group (monolinguals vs. bilinguals) (1,26) 1.373 ns 
Semantic violation Sentence condition (semantic vs. control) (1,26) 11.31 .002 
Condition × Group (1,26) 0.53 ns 
Group (monolinguals vs. bilinguals) (1,26) 0.262 ns 

(A) The results of statistical analyses of 8–30 Hz ERD and 1–5 Hz ERS in the 500–1500 msec time window. Effects of violation condition and scalp region in the acceptability task. (B) Effects of bilingualism on 8–30 Hz ERD in the frontal ROI, in the 500–1500 msec time window. Group comparisons are between monolinguals and bilinguals, in the grammaticality task.

1–5 Hz ERS

In the 1–5 Hz frequency band, the repeated-measures ANOVA on the difference scores (violations − control) with Violation condition (semantic vs. syntactic) and Region as within-subject factors showed a significant main effect of Violation condition, reflecting larger power increases for semantic violations compared with syntactic violations (Figure 2A and D), which was to be expected as only the semantic response had achieved significance in the cluster analysis. There was also a significant main effect of Region and a Violation condition × Region interaction, reflecting a stronger fronto-central distribution of 1–5 Hz ERS for semantic anomalies (Figure 3B).

Effects of Bilingualism

The EEG data in this study were originally collected for an ERP study examining differences in linguistic processing between bilinguals and monolinguals in a grammaticality judgment task, in which bilinguals have been shown to have a behavioral advantage attributable to superior executive function and cognitive control (Bialystok & Majumder, 1998; Bialystok, 1986). For that reason, the grammaticality task was always performed first, but the acceptability task was included as a secondary control. In this study, we used the data from the acceptability task to characterize the oscillatory responses to semantic and syntactic violations, in terms of their temporal extent, their frequency range, and their scalp topography. This left the grammaticality task as an independent sample in which to test for group differences in the most relevant time–frequency windows and regions. Cluster analysis of the acceptability task revealed that violations were consistently associated with 8–30 Hz ERD in the 500–1500 msec time window, widespread and bilateral for semantic violations, and maximal at left parietal and frontal electrodes for syntactic violations.

Previous ERP studies suggest that bilingualism has a stronger influence on syntactic than semantic processing (e.g., Moreno et al., 2010; Hahne, 2001; Hahne & Friederici, 2001; Neville & Weber-Fox, 1996). The bilingual advantage in grammaticality judgment has been linked to superior cognitive control in bilinguals, and various neuroimaging modalities have suggested a key role for frontal cortex in that function (MEG: Bialystok et al., 2005; ERP: Jackson, Jackson, & Roberts, 1999; fMRI: Konishi et al., 1998, 1999). Therefore, to test for group differences in the grammaticality task, we focused our analysis of group differences on an ROI consisting of five frontal electrodes that were highly responsive to syntactic violations in the acceptability task.

In brief, we found that the responses to semantic violations at frontal electrodes were similar in both groups, whereas the 8–30 Hz power responses to syntactic violations were reduced in the bilingual group, in the grammaticality task (Figure 6A). Scalp topographies of 8–30 Hz ERD in each group and task are presented in Figure 6B. To test for significant interactions between violation conditions and language groups, we conducted ANOVAs with Sentence condition (semantic and syntactic violations vs. control sentences) as a within-subject factor and Group (monolinguals and bilinguals) as a between-subject factor, for the grammaticality task in the frontal region. The summary of results is presented in Table 5B.

Figure 6. 

8–30 Hz ERD for semantic and syntactic violations, by task and language group. (A) The bar graph shows 8–30 Hz ERD values (violation–control) at the frontal region, separately for monolinguals and bilinguals in each task. The error bars represent SEM. Oval symbols on the scalp map indicate location of electrodes averaged together to form the frontal ROI. (B) The scalp topographies of 8–30 Hz ERD effects for each language group modulated by task and violation condition in the 8–30 Hz frequency range and 500–1500 msec time window.

Figure 6. 

8–30 Hz ERD for semantic and syntactic violations, by task and language group. (A) The bar graph shows 8–30 Hz ERD values (violation–control) at the frontal region, separately for monolinguals and bilinguals in each task. The error bars represent SEM. Oval symbols on the scalp map indicate location of electrodes averaged together to form the frontal ROI. (B) The scalp topographies of 8–30 Hz ERD effects for each language group modulated by task and violation condition in the 8–30 Hz frequency range and 500–1500 msec time window.

The ANOVAs revealed a significant main effect of Sentence condition and Group × Sentence condition interaction for syntactic violations. Post hoc tests of this interaction revealed that for bilinguals, power decreases for syntactic anomalies did not significantly differ from control sentences, F < 1, whereas monolinguals displayed significantly greater power reduction to syntactic violations as compared with control sentences, F(1, 13) = 6.87, p < .05 (see Figure 6A). For semantic violations (semantic vs. control), there was a significant main effect of Sentence condition, but no main effect of Group or Group × Sentence condition interaction.

These results indicate that frontal responses to semantic anomalies were equivalent across both groups, whereas 8–30 Hz ERD responses to syntactic violations were reduced for bilinguals in the grammaticality task.

Split-half Analysis

In this study, the experimental design was constrained by the desire to have both anomalies occur at the same position in the sentence. This resulted in a somewhat restricted set of sentence materials (see Methods). We were concerned that the observed effects may be because of participants adopting experiment-specific processing strategies, rather than more general mechanisms of language comprehension. To rule out this possibility, we performed a split-half analysis on the data separately for 8–30 Hz ERD and 1–5 Hz ERS, collapsing across all 28 participants (both monolinguals and bilinguals) in the acceptability and grammaticality tasks. If the observed ERD and ERS responses were attributable to participants adopting experiment-specific strategies, we would expect the magnitude of the responses to increase as the experiment proceeds. In contrast, if the responses are attributable to more general mechanisms, we would expect no difference between the first and second halves of each task or even a decrease in reactivity as the task becomes more practiced.

8–30 Hz ERD

For the acceptability task, we separated the trials for each participant into the first 15 and the second 15 in each condition. We ran a repeated-measures ANOVA on the difference scores (violations − control) with Data portion (first half, second half), Violation condition (semantic vs. syntactic), and Region as within-subject factors. The summary of results is presented in Table 6. There was a significant main effect of Condition, reflecting greater 8–30 Hz ERD for semantic than syntactic violations. In addition, there was a significant main effect of Region and Violation condition × Region interaction, indicating bilateral central-parietal distribution for semantic violations, whereas syntactic effects were maximal at fronto-central and left parietal electrodes. This analysis did not show a significant main effect of Data portion or interactions with Condition or Region (Violation condition × Data portion, Region × Data portion, Condition × Region × Data portion), indicating that effects were equally strong in the second and first halves of the experiment.

Table 6. 

The Results of Split-half Analysis Performed Separately for 8–30 Hz ERD and 1–5 Hz ERS, Collapsing across All 28 Participants (Both Monolinguals and Bilinguals) in the Acceptability and Grammaticality Tasks

TaskFactordfFp
8–3 Hz ERD 
Acceptability Condition (semantic vs. syntactic) (1,27) 8.94 .006 
Data portion (first half vs. second half) (1,27) 0.57 ns 
Region (10,270) 2.78 .009 
Condition × Data portion (1,27) 0.74 ns 
Condition × Region (10,270) 6.77 <.001 
Region × Data portion (10,270) 0.75 ns 
Condition × Data portion × Region (10,270) 0.45 ns 
Grammaticality Condition (semantic vs. syntactic) (1,27) 13.30 .001 
Data portion (first half vs. second half) (1,27) 1.94 ns 
Region (10,270) 3.22 .005 
Condition × Data portion (1,27) 1.32 ns 
Condition × Region (10,270) 5.76 <.001 
Region × Data portion (10,270) 1.26 ns 
Condition × Data portion × Region (10,270) 1.04 ns 
 
1–5 Hz ERS 
Acceptability Condition (semantic vs. syntactic) (1,27) 21.31 <.001 
Data portion (first half vs. second half) (1,27) 1.46 ns 
Region (10,270) 2.49 .019 
Condition × Data portion (1,27) 3.70 ns 
Condition × Region (10,270) 2.71 .009 
Region × Data portion (10,270) 1.17 ns 
Condition × Data portion × Region (10,270) 1.58 ns 
Grammaticality Condition (semantic vs. syntactic) (1,27) 15.38 .001 
Data portion (first half vs. second half) (1,27) 0.82 ns 
Region (10,270) 1.14 ns 
Condition × Data portion (1,27) 2.49 ns 
Condition × Region (10,270) 2.51 .02 
Region × Data portion (10,270) 1.24 ns 
Condition × Data portion × Region (10,270) 0.94 ns 
TaskFactordfFp
8–3 Hz ERD 
Acceptability Condition (semantic vs. syntactic) (1,27) 8.94 .006 
Data portion (first half vs. second half) (1,27) 0.57 ns 
Region (10,270) 2.78 .009 
Condition × Data portion (1,27) 0.74 ns 
Condition × Region (10,270) 6.77 <.001 
Region × Data portion (10,270) 0.75 ns 
Condition × Data portion × Region (10,270) 0.45 ns 
Grammaticality Condition (semantic vs. syntactic) (1,27) 13.30 .001 
Data portion (first half vs. second half) (1,27) 1.94 ns 
Region (10,270) 3.22 .005 
Condition × Data portion (1,27) 1.32 ns 
Condition × Region (10,270) 5.76 <.001 
Region × Data portion (10,270) 1.26 ns 
Condition × Data portion × Region (10,270) 1.04 ns 
 
1–5 Hz ERS 
Acceptability Condition (semantic vs. syntactic) (1,27) 21.31 <.001 
Data portion (first half vs. second half) (1,27) 1.46 ns 
Region (10,270) 2.49 .019 
Condition × Data portion (1,27) 3.70 ns 
Condition × Region (10,270) 2.71 .009 
Region × Data portion (10,270) 1.17 ns 
Condition × Data portion × Region (10,270) 1.58 ns 
Grammaticality Condition (semantic vs. syntactic) (1,27) 15.38 .001 
Data portion (first half vs. second half) (1,27) 0.82 ns 
Region (10,270) 1.14 ns 
Condition × Data portion (1,27) 2.49 ns 
Condition × Region (10,270) 2.51 .02 
Region × Data portion (10,270) 1.24 ns 
Condition × Data portion × Region (10,270) 0.94 ns 

The identical analysis for the grammaticality task revealed a significant main effect of violation condition, region, and Violation condition × Region interaction, but no main effect of data portion or interactions.

1–5 Hz ERS

The same analyses in the 1–5 Hz frequency window for the acceptability task revealed a significant main effect of condition, indicating larger power increases for semantic violations compared with syntactic violations. There was a significant main effect of region, and a Violation condition × Region interaction, reflecting a stronger fronto-central distribution of 1–5 Hz ERS for semantic anomalies. However, there was no significant main effect of Data portion or interactions with Condition or Region, indicating that effects were equally strong in the second and first halves of the experiment.

Similarly, for the grammaticality task there was a significant main effect of Violation condition and Violation condition × Region interaction, but no significant main effect of Data portion or interactions.

These analyses confirm that effects are equally strong in the earlier and later halves of the experiment for both tasks. Therefore, it is unlikely that participants developed special strategies to complete the task instead of comprehending the sentences in a natural manner.

DISCUSSION

Modulation of Oscillatory Activity to Semantic and Syntactic Anomalies

In this study, we investigated modulations of ongoing EEG oscillatory activity while participants made judgments of sentences containing semantic and syntactic errors. The results showed that both semantic and syntactic violations elicited power decreases in the alpha and beta bands (approximately 8–30 Hz), with distinct but overlapping scalp distributions. For semantic violations, oscillatory power decreases were observed bilaterally over parietal sites, whereas power decreases for syntactic violations were weaker overall and were maximal at left parietal and frontal sites. In addition, semantic violations elicited a power increase in the frequency range of approximately 1–5 Hz, which was distributed over fronto-central sensors with additional peaks in bilateral temporal areas. We also investigated the relationship between responses present in the time-locked ERP and induced oscillatory activity. The results indicated that the 1–5 Hz ERS was strongly phase-locked to stimulus onset and highly correlated across participants with time domain averages, but the 8–30 Hz ERD response varied independently of these.

The 1–5 Hz ERS to semantic violations in this study is consistent with previous reports demonstrating power increases in the theta band for sentence-embedded semantic anomalies (Willems et al., 2008; Davidson & Indefrey, 2007; Hald et al., 2006). Modulations in theta power have been widely reported during studies of word processing (Bastiaansen, Oostenveld, Jensen, & Hagoort, 2008; Hald et al., 2006; Bastiaansen, van der Linden, ter Keurs, Dijkstra, & Hagoort, 2005) and have also been reported over longer time periods spanning the length of a sentence (Bastiaansen et al., 2009; Bastiaansen, van Berkum, & Hagoort, 2002b; Röhm, Klimesch, & Doppelmayr, 2001). Theta increases have also been observed in response to verbal working memory demands in general (Klimesch, Schack, & Sauseng, 2005; Klimesch, 1999), although EEG-fMRI studies have suggested that such increases reflect deactivation of the default-mode network rather than positive activation of task-specific cortical areas (Scheeringa et al., 2009; Meltzer, Negishi, Mayes, & Constable, 2007; Mizuhara, Wang, Kobayashi, & Yamaguchi, 2004).

Despite these similarities, several findings in this study cast doubt on a close association between the observed 1–5 Hz ERS and the increases in theta oscillations observed in working memory paradigms. First, the observed frequency band, 1–5 Hz, is lower than that typically described as “theta,” which tends to be about 4–7 Hz, and overlaps more heavily with the delta band (0–4 Hz). Second, we observed a close correlation across participants between the 1–5 Hz ERS measured across single trials and the amount of 1–5 Hz power increase present in the time domain average ERP signal. The close relationship between ERPs and 1–5 Hz ERS may have been expected, as most ERP signals found in language studies (e.g., N400, P600) have the bulk of their power in this lower frequency range. The duration of the 1–5 Hz ERS response constitutes less than a full wave cycle at the lowest frequencies, and so these signals should not necessarily be interpreted as oscillations, but rather as waveforms of limited temporal extent. In the present analysis, we used a time window of 0.512 sec for the short-time Fourier transform. To characterize very low frequency oscillations in detail, a much longer analysis window would be necessary, which would have compromised the time resolution necessary to capture the relatively short event-related responses occurring in this paradigm. Thus, we consider the 1–5 Hz ERS response to be a reflection of low-frequency waveforms elicited by semantic violations, but not necessarily true oscillations.

Our results indicate a close relationship between ERPs and 1–5 Hz ERS but do not suggest that these signals are identical or interchangeable. For example, the visual presentation of each word (including the words preceding the critical control or violation word) evokes a visual ERP, which is reflected in the ERSP of ERP signal (Figure 4B), but not in the ERS signal (Figure 4C). This distinction indicates that the enhanced 1–5 Hz ERS in response to semantic violations primarily reflects the later components of the evoked potential, such as N400 and P600, but not the early visual components. These early components also contain power in the 1–5 Hz range (and higher frequencies), but this power is not high enough to stand out above the background activity that is present in all single trials, unless the background activity is attenuated by averaging the signal in the time domain.

Another distinction between 1–5 Hz ERS and ERPs is that ERS in this range picks up signals of either polarity, both positive and negative (as illustrated in Figure 5). Therefore, one may detect significant ERS in the delta and theta ranges, even if differences in polarity across trials or participants result in relatively weak ERP signals. Our findings suggest that ERS in such a low range likely reflects a strong contribution from the same neural generators that underlie ERP responses, and caution should be taken in its interpretation. On the other hand, differences in polarity across participants may have more to do with variability in neural geometry than with differences in cognitive mechanisms engaged by a task. Therefore, low-frequency ERS may in some cases be more sensitive than traditional ERPs for quantifying neural reactivity. This possibility awaits further study. Furthermore, additional information on the relationship between ERPs and time–frequency reactivity may be obtained from alternative methods of studying their correlations. Whereas we examined correlations across participants in this study, other studies have examined correlations across individual trials within participants (Davidson & Indefrey, 2007) or across different electrode sites (Engell & McCarthy, 2011).

In contrast to the strong relationship between 1–5 Hz ERS and evoked signals, there was very little correlation between the 1–5 Hz signals and the 8–30 Hz ERD. In addition, scalp maps of 1–5 Hz ERS and time domain averages showed similar topographic distributions, and those were distinct from the 8–30 Hz ERD topographies. The dissimilarity between the time-locked signals and 8–30 Hz ERD for either semantic or grammatical violations suggests that at least a portion of the activity that gives rise to these effects originates from different cortical populations.

Power decreases in the alpha and beta (8–30 Hz) frequency bands have also been reported in numerous cognitive paradigms. Research in the memory domain indicates that alpha and beta power decreases play an important role in semantic encoding and retrieval from long-term memory (Hanslmayr, Staudigl, & Fellner, 2012; Klimesch et al., 2005). Although power decreases in the alpha and beta range have been less consistently observed in response to sentence-embedded semantic anomalies, evidence suggests that oscillations in this frequency range may be more directly related to processing of linguistic information than theta responses (Bastiaansen et al., 2009; Willems et al., 2008).

Power changes in the alpha and beta bands have been associated with a variety of processes related to sentence comprehension (Bastiaansen et al., 2005, 2008; Davidson & Indefrey, 2007; Weiss et al., 2005; Röhm et al., 2001). For example, alpha desynchronization has been found for a semantic retrieval task in a sentence context (Röhm et al., 2001) and for semantic judgments on pairs of words (Klimesch, Doppelmayr, Pachinger, & Russegger, 1997). In another study, Willems et al. (2008) observed decreases in alpha band power in response to sentence-embedded semantic anomalies. This effect was suggested to be specifically related to processing of linguistic information, as the decrease in alpha power was greater to semantic mismatches within the sentence than to mismatching pictures. Similarly, in a more recent MEG study, Wang et al. (2012) observed left temporal decreases in alpha and beta power in response to semantic anomalies in sentence-final position.

Power decreases in the alpha and beta frequency bands were also observed in response to grammatical violations in this study. Previous studies have linked various aspects of syntactic processing to reactivity in these frequency ranges (Bastiaansen et al., 2009; Davidson & Indefrey, 2007). For example, larger alpha and beta power suppression has been reported for open-class words versus closed-class words occurring in the sentence context (Bastiaansen et al., 2005). In a different study, Weiss et al. (2005) observed a change in EEG coherence in the beta band (11–18 Hz) during processing of sentences with relative clauses. More directly relevant to this study, syntactic violations occurring in a sentence context have been found to elicit power suppression in the alpha and beta bands. For instance, Bastiaansen et al. (2009) reported power decreases in these frequency ranges in response to sentences containing word category violations. Similarly, Davidson and Indefrey (2007) found power decreases in the alpha and beta bands for phrase structure errors, whereas number agreement violations elicited responses in the alpha band only. In these studies the functional role of alpha-beta power suppression has been associated with attentional control demands (Davidson & Indefrey, 2007), lexical retrieval from semantic memory (Röhm et al., 2001), or increased processing effort associated with detection of a linguistic violation (Bastiaansen et al., 2009; Willems et al., 2008).

In this study, we observed 8–30 Hz ERD for both semantic and syntactic anomalies. The latency of this effect (between 500 and 1500 msec) is rather long, and given the 650-msec interval between word presentations, the bulk of it occurs as the following one to two words in the sentence are presented. Nonetheless, the power subsequently returns to baseline before the end of the sentence, indicating that the effect is unlikely to be related to motor preparation for the overt response. These findings suggest that alpha and beta power reduction most likely reflects reprocessing or reanalysis of the sentence after a violation is detected. Alternatively, it may signal increased attentional control associated with the monitoring of violations. In general, alpha and beta power reductions are likely to reflect increases in neural activation in cortical areas that are involved in reprocessing linguistic input after a semantic or syntactic violation is encountered, as it has recently been proposed that alpha power regulates the information flow in the brain by deactivation of task-irrelevant regions (Hanslmayr et al., 2012). In light of the previous studies, there appears to be a complex relationship between alpha/beta oscillations and sentence comprehension. The present findings are most consistent with the idea that alpha and beta oscillations reflect increased engagement of the task-relevant brain regions for both semantic and syntactic anomalies (also discussed by Wang et al., 2012; Davidson & Indefrey, 2007).

Our data are broadly consistent with previously observed dissociations between semantic and syntactic processing, although it is uncertain whether ERD effects are coming from the same regions that generate ERP signals and to what extent either of these effects colocalize with fMRI signal changes. Previous studies have found that ERD in the alpha and beta bands corresponds closely to fMRI activation and is associated with increased neural activity (Brookes et al., 2005; Pfurtscheller & Klimesch, 1992), not only in sensorimotor paradigms but also in higher-level language tasks (Meltzer & Braun, 2011; Singh, Barnes, Hillebrand, Forde, & Williams, 2002).

We note that the present results were obtained from a highly constrained set of sentence materials, and further study will be necessary to determine to what extent the observed differences between semantic and syntactic violations extend to other forms of violations, occurring in different contexts. For example, all syntactic anomalies in this study consisted of an inappropriate “-ing” ending attached to a verb. The split-half analysis showed that effects were similar in the first and second halves of each task, suggesting that they were not attributable to participants' adoption of experiment-specific strategies, but it would still be interesting to explore the extent to which these findings generalize to more varied forms of semantic and syntactic violations.

The constrained nature of the present materials arose largely from the desire to have both kinds of anomalies occur in a comparable sentence-medial position. Whereas most studies of semantic anomalies in English use sentence-final nouns as the critical words, this study used sentence-medial verbs and achieved semantic anomalies by creating a mismatch in animacy between the sentence subject and the main verb. Although the original ERP study employing these sentences, Osterhout and Nicol (1999), observed a fairly clean dissociation between semantic and syntactic anomalies, similar animacy mismatches have frequently been found to elicited ERP responses more closely linked to syntactic than semantic violations, despite their grammatical correctness (e.g., P600 responses; for review, see Bornkessel-Schlesewsky & Schlesewsky, 2008; Kuperberg, 2007). Thus, we suggest that further studies employing different violation types or reanalyses of existing ERP studies using time–frequency methods will shed additional light on the distinction between semantic and syntactic processing mechanisms in the brain.

Effect of Task and Bilingualism

We found that bilingualism modulated 8–30 Hz ERD for syntactic violations as a function of the executive demands of the task. When the executive function demands were increased in the grammaticality judgment task, bilinguals demonstrated reduced alpha/beta ERD for syntactic violations. As discussed above, increased and more widespread ERD has been associated with increased task complexity, more effort, or increased attentional demands. It is hypothesized that tasks with greater cognitive demands require involvement of a larger neural network or more neuronal assemblies in information processing, contributing to stronger ERD (Klimesch et al., 1996, 1997; Sterman, Kaiser, & Veigel, 1996; Dujardin et al., 1993), whereas more efficient task performance is associated with decreased ERD (Zhuang et al., 1997). Thus, reduced ERD in bilingual participants found in this study may indicate that they require fewer neural resources to judge grammaticality when the meaning needs to be ignored. Because of their superior executive control abilities (Bialystok et al., 2004, 2006), bilinguals are more efficient at managing conflict between competing alternatives, and this is reflected in less effortful processing of conflict in the grammaticality task.

One limitation of this study is that the grammaticality task was always conducted before the acceptability task, rather than counterbalancing the order across participants. This was a deliberate choice by the authors of the original ERP study (Moreno et al., 2010) made to maximize the sensitivity to group differences in the grammaticality task. Thus, we did not analyze task as a factor in the statistical analysis. Rather, we used the acceptability task to characterize the spatiotemporal nature of the oscillatory response to violations to identify the relevant time windows, frequency bands, and scalp regions. This reserved the grammaticality task data as an independent sample in which to assess group differences. However, the finding of group differences in the grammaticality task but not in the acceptability task raises the question of whether such a group difference could have been eliminated by the temporal order confound, especially because the same 120 sentence frames (but not individual sentences in the same conditions) were repeated. We consider such an explanation unlikely, though. Increased amounts of task practice and stimulus repetition are typically associated with a reduced neural response, both in fMRI (Meltzer, Postman-Caucheteux, McArdle, & Braun, 2009) and EEG (Romero, McFarland, Faust, Farrell, & Cacace, 2008). In this study, bilinguals had a reduced response to syntactic violations compared with the monolinguals in the grammaticality task, which was presented first. Therefore, reduction of the overall response because of repetition does not appear to account for the specificity of the group difference for the grammaticality judgment task.

Our results suggest that bilinguals required a lesser amount of neural activity at frontal sites to judge the grammaticality of sentences. These results support previous imaging studies implicating regions of the frontal cortex in cognitive control functions (MEG: Bialystok et al., 2005; ERP: Jackson et al., 1999; fMRI: Konishi et al., 1998, 1999). Overall, behavioral and imaging studies support the idea that constant management of two languages in bilinguals leads to changes in the engagement of frontal executive control processes.

The behavioral results are also consistent with bilinguals' enhanced executive control abilities through greater selective attention to conflicting values. Even through the monolingual and bilinguals were equivalent in all background measures, except vocabulary (measured by the PPVT), on which monolinguals scored higher (consistent with previous findings, e.g., Portocarrero, Burright, & Donovick, 2007), the type of instruction produced different behavioral performance in the two groups. In the acceptability task, bilinguals performed less accurately. This is attributable to lower language proficiency overall, related to their smaller vocabulary. However, in the grammaticality task that requires greater involvement of executive control to monitor and resolve conflict, bilinguals performed as well as monolinguals. Thus, bilinguals were able to reach the behavioral performance level of monolingual participants when task demands required greater executive control.

The behavioral results in this study contrast slightly with previous findings of superior performance in bilinguals on grammaticality judgments in the presence of semantic anomalies (Bialystok & Majumder, 1998; Bialystok, 1986), whereas in this study we observed equivalently high performance in both groups. This difference can be explained by a difference in task design. In prior studies, participants were required to detect grammatically incorrect words immediately on their presentation. Monolinguals were more likely to inappropriately reject grammatically correct but semantically anomalous words, that is, producing a go response instead of a no-go. In this study, the decision on sentence acceptability/grammaticality was postponed to the end of the sentence. This was done to avoid confounding the correct/violation distinction with a motor response and to keep accuracy levels high. This made the task easier, resulting in fewer errors overall, and resulted in equivalent levels of high performance in both groups. This allowed us to observe differences in the neural response to anomalous words without the presence of confounding differences in accuracy.

The present results complement the ERP findings of Moreno et al. (2010) described in the Introduction. In that study, bilinguals showed smaller P600 amplitude than monolinguals, but only in the more difficult grammaticality task. The findings reported in the literature indicate that executive control mechanisms modulate late positivity components, suggesting a link between control processes and the P600 (Swainson, Jackson, & Jackson, 2006; Slagter, Kok, Mol, & Kenemans, 2005; Jackson et al., 2001). The decreased P600 amplitude observed in Moreno et al.'s (2010) study is consistent with the proposal that a more efficient control mechanism influences the processing of linguistic information in bilinguals. The results of this study support earlier findings from bilingual children and adults that report a bilingual advantage on verbal and nonverbal tasks demanding selective attention and conflict resolution (Moreno et al., 2008; Bialystok et al., 2004, 2005, 2006; Bialystok & Martin, 2004; Bialystok & Majumder, 1998).

One aspect of the present results diverges from the ERP analyses in the study of Moreno et al. (2010). In the grammaticality task, Moreno et al. observed a significant N400 response to semantic violations only in bilinguals, not in monolinguals. In contrast, we observed robust ERSP responses to semantic anomalies in both groups of approximately equal magnitude. This difference most likely reflects the relative sensitivity of the time domain averaging and time–frequency analysis to different types of information present in the EEG signal. Whereas the ERP procedure attenuates any responses in the EEG signal that are not strictly time- and phase-locked to stimulus onset, time–frequency analysis can effectively detect both phase-locked and non-phase-locked activity. Because ERP responses reflecting higher level cognitive processes have greater across-trial variability in response latency, their amplitude is more likely to be reduced by time domain averaging of the brain signal. In the present case, the response to semantic violations for monolinguals seems to have been greatly attenuated in the grammaticality task as assessed by ERPs, but not by time–frequency analysis. This is an example of the complementary nature of the two techniques. In cases where greater temporal variability may reduce the amplitude of observed ERPs, time–frequency analysis may be especially useful in assessing the degree of neural reactivity to a given experimental manipulation. Another possibility is that differences in time windows being analyzed may have weakened group differences. In the Moreno et al. (2010) study, the N400 was quantified during the 350–480 msec time interval, whereas in this study, we used cluster analysis to identify significant patterns of time–frequency modulation over the entire trial and found that ERSP responses occurred predominantly at longer latencies (500–1500 msec).

Conclusions

The present results contribute to a growing literature on changes in oscillatory neural activity induced by linguistic stimuli. We have shown that semantic and syntactic violations both elicit 8–30 Hz ERD responses that are maximal approximately 500–1000 msec after the onset of the critical word. The scalp topography differed between violation types, with syntactic violations producing effects that are more left lateralized and frontal. Semantic anomalies produced an additional 1–5 Hz power increase, in the delta and lower theta bands, with peaks over fronto-central and temporal electrodes. The 1–5 Hz ERS was strongly phase-locked to stimulus onset and highly correlated with time domain averages, but the 8–30 Hz ERD response varied independently of these. These results indicate that the 1–5 Hz ERS is a reflection of similar neuronal activity that underlies the ERP response. However, more interestingly, the 8–30 Hz ERD likely reflects a separate neural process.

We also compared the strength of the ERSP response between monolinguals and bilinguals, in keeping with the original motivation for the EEG study of Moreno et al. (2010). Similar to the ERP findings of that study, we found that bilingualism influences how the brain processes sentence-level semantic and syntactic aspects of language. The two language groups showed comparable levels of neural activity on the acceptability task, which required primarily linguistic knowledge, whereas responses to the linguistic stimuli were modified for bilinguals when the task demanded greater executive control. In the grammaticality task, bilingual participants showed significantly decreased ERD in response to syntactic errors. These results are consistent with the body of work showing a bilingual advantage on tasks which require greater executive processing and inhibitory control. In this study, this advantage was reflected in the modulation of 8–30 Hz oscillatory power during processing of linguistic stimuli under differential task demands.

In conclusion, this study serves to characterize the nature of the oscillatory responses in sentence judgment tasks featuring both semantic and syntactic errors and serves as a starting point for future studies that may investigate how these processes are altered in neurological disorders. By comparing ERSP responses across groups and tasks on an established paradigm that distinguishes these groups, we show that oscillatory analyses can be a valuable tool for assessing differences in linguistic processing across groups of participants. This is particularly useful in assessing language processing in clinical populations, which may show significant delays or variability in the latency of neural responses.

Reprint requests should be sent to Jed A. Meltzer, Rotman Research Institute, Baycrest Centre, 3560 Bathurst Street, Toronto, Ontario, M6A 2E1, Canada, or via e-mail: jmeltzer@research.baycrest.org.

REFERENCES

REFERENCES
Allefeld
,
C.
,
Frisch
,
S.
, &
Schlesewsky
,
M.
(
2005
).
Detection of early cognitive processing by event-related phase synchronization analysis.
NeuroReport
,
16
,
13
16
.
Ardal
,
S.
,
Donald
,
M. W.
,
Meuter
,
R.
,
Muldrew
,
S.
, &
Luce
,
M.
(
1990
).
Brain responses to semantic incongruity in bilinguals.
Brain and Language
,
39
,
187
205
.
Bastiaansen
,
C. M.
,
Oostenveld
,
R.
,
Jensen
,
O.
, &
Hagoort
,
P.
(
2008
).
I see what you mean: Theta power increases are involved in the retrieval of lexical semantic information.
Brain and Language
,
106
,
15
28
.
Bastiaansen
,
M.
,
Magyari
,
L.
, &
Hagoort
,
P.
(
2009
).
Syntactic unification operations are reflected in oscillatory dynamics during on-line sentence comprehension.
Journal of Cognitive Neuroscience
,
22
,
1333
1347
.
Bastiaansen
,
M. C.
,
van Berkum
,
J. J.
, &
Hagoort
,
P.
(
2002a
).
Syntactic processing modulates the theta rhythm of the human EEG.
Neuroimage
,
17
,
1479
1492
.
Bastiaansen
,
M. C.
,
van Berkum
,
J. J.
, &
Hagoort
,
P.
(
2002b
).
Event-related theta power increases in the human EEG during online sentence processing.
Neuroscience Letters
,
323
,
13
16
.
Bastiaansen
,
M. C.
,
van der Linden
,
M.
,
ter Keurs
,
M.
,
Dijkstra
,
T.
, &
Hagoort
,
P.
(
2005
).
Theta responses are involved in lexical-semantic retrieval during language processing.
Journal of Cognitive Neuroscience
,
17
,
530
541
.
Bialystok
,
E.
(
1986
).
Factors in the growth of linguistic awareness.
Child Development
,
57
,
498
510
.
Bialystok
,
E.
(
1988
).
Levels of bilingualism and levels of linguistic awareness.
Developmental Psychology
,
24
,
560
567
.
Bialystok
,
E.
(
2001
).
Bilingualism in development: Language, literacy, and cognition
.
New York
:
Cambridge University Press
.
Bialystok
,
E.
,
Craik
,
F. I. M.
,
Grady
,
C.
,
Chau
,
W.
,
Ishii
,
R.
,
Gunji
,
A.
,
et al
(
2005
).
Effect of bilingualism on cognitive control in the Simon task: Evidence from MEG.
Neuroimage
,
24
,
40
49
.
Bialystok
,
E.
,
Craik
,
F. I. M.
, &
Ryan
,
J.
(
2006
).
Executive control in a modified antisaccade task: Effects of aging and bilingualism.
Journal of Experimental Psychology: Learning Memory and Cognition
,
32
,
1341
1354
.
Bialystok
,
E.
,
Klein
,
R.
,
Craik
,
F. I. M.
, &
Viswanathan
,
M.
(
2004
).
Bilingualism, aging and cognitive control: Evidence from the Simon task.
Psychology and Aging
,
19
,
290
303
.
Bialystok
,
E.
, &
Majumder
,
S.
(
1998
).
The relationship between bilingualism and the development of cognitive processes in problem solving.
Applied Psycholinguistics
,
19
,
69
85
.
Bialystok
,
E.
, &
Martin
,
M. M.
(
2004
).
Attention and inhibition in bilingual children: Evidence from the dimensional change card sort task.
Developmental Science
,
7
,
325
339
.
Bornkessel-Schlesewsky
,
I.
, &
Schlesewsky
,
M.
(
2008
).
An alternative perspective on “semantic P600” effects in language comprehension.
Brain Research
, Review,
59
,
55
73
.
Brookes
,
M. J.
,
Gibson
,
A. M.
,
Hall
,
S. D.
,
Furlong
,
P. L.
,
Barnes
,
G. R.
,
Hillebrand
,
A.
,
et al
(
2005
).
GLM-beamformer method demonstrates stationary field, alpha ERD and gamma ERS co-localisation with fMRI BOLD response in visual cortex.
Neuroimage
,
26
,
302
308
.
Cattell
,
R. B.
(
1957
).
Culture Fair Intelligence Test, a measure of “g”: Scale 3, forms A and B
.
Savoy, IL
:
Institute for Personality and Ability Testing
.
Coulson
,
S.
,
King
,
J.
, &
Kutas
,
M.
(
1998
).
Expect the unexpected: Event-related brain response to morphosyntactic violations.
Language and Cognitive Processes
,
13
,
21
58
.
Davidson
,
D. J.
, &
Indefrey
,
P.
(
2007
).
An inverse relation between event-related and time–frequency violation responses in sentence processing.
Brain Research
,
1158
,
81
92
.
Dujardin
,
K.
,
Derambure
,
P.
,
Defebvre
,
L.
,
Bourriez
,
J. L.
,
Jacquesson
,
J. M.
, &
Guieu
,
J. D.
(
1993
).
Evaluation of event-related desynchronization (ERD) during a recognition task: Effect of attention.
Electroencephalography and Clinical Neurophysiology
,
86
,
353
356
.
Dunn
,
L. M.
, &
Dunn
,
L. M.
(
1997
).
Peabody Picture Vocabulary Test
(3rd ed.).
Circle Pines, MN
:
American Guidance Service
.
Engell
,
A. D.
, &
McCarthy
,
G.
(
2011
).
The relationship of gamma oscillations and face-specific ERPs recorded subdurally from occipitotemporal cortex.
Cerebral Cortex
,
21
,
1213
1221
.
Fischer
,
M. H.
(
2001
).
Probing spatial working memory with the Corsi blocks task.
Brain and Cognition
,
45
,
143
154
.
Friederici
,
A. D.
,
Hahne
,
A.
, &
Saddy
,
D.
(
2002
).
Distinct neurophysiological patterns reflecting aspects of syntactic complexity and syntactic repair.
Journal of Psycholinguistic Research
,
31
,
45
63
.
Friederici
,
A. D.
,
Pfeifer
,
E.
, &
Hahne
,
A.
(
1993
).
Event-related brain potentials during natural speech processing: Effects of semantic, morphological and syntactic violations.
Cognitive Brain Research
,
1
,
183
192
.
Hahne
,
A.
(
2001
).
What's different in second-language processing? Evidence from event-related brain potentials.
Journal of Psycholinguistic Research
,
30
,
251
266
.
Hahne
,
A.
, &
Friederici
,
A. D.
(
1999
).
Electrophysiological evidence for two steps in syntactic analysis: Early automatic and late controlled processes.
Journal of Cognitive Neuroscience
,
11
,
194
205
.
Hahne
,
A.
, &
Friederici
,
A. D.
(
2001
).
Processing a second language: Late learners' comprehension mechanisms as revealed by event-related brain potential.
Bilingualism: Language and Cognition. Special Issue: The Cognitive Neuroscience of Bilingualism
,
4
,
123
141
.
Hald
,
L. A.
,
Bastiaansen
,
M. C.
, &
Hagoort
,
P.
(
2006
).
EEG theta and gamma responses to semantic violations in online sentence processing.
Brain and Language
,
96
,
90
105
.
Hanslmayr
,
S.
,
Staudigl
,
T.
, &
Fellner
,
M. C.
(
2012
).
Oscillatory power decreases and long-term memory: The information via desynchronization hypothesis.
Frontiers in Human Neuroscience
,
6
,
1
12
.
Jackson
,
G. M.
,
Swainson
,
R.
,
Cunnington
,
R.
, &
Jackson
,
S.
(
2001
).
ERP correlates of executive control during repeated language switching.
Bilingualism: Language and Cognition
,
4
,
169
178
.
Jackson
,
S. R.
,
Jackson
,
G. M.
, &
Roberts
,
M.
(
1999
).
The selection and suppression of action: ERP correlates of executive control in humans.
NeuroReport
,
17
,
861
865
.
Jung
,
T. P.
,
Makeig
,
S.
,
Westerfield
,
M.
,
Townsend
,
J.
,
Courchesne
,
E.
, &
Sejnowski
,
T. J.
(
2000
).
Removal of eye activity artifacts from visual event-related potentials in normal and clinical subjects.
Clinical Neurophysiology
,
111
,
1745
1758
.
Kaan
,
E.
,
Harris
,
A.
,
Gibson
,
E.
, &
Holcomb
,
P.
(
2000
).
The P600 as an index of syntactic integration difficulty.
Language and Cognitive Processes
,
15
,
159
201
.
Kawohl
,
W.
,
Bunse
,
S.
,
Willmes
,
K.
,
Hoffrogge
,
A.
,
Buchner
,
H.
, &
Huber
,
W.
(
2010
).
Semantic event-related potential components reflect severity of comprehension deficits in aphasia.
Neurorehabilitation and Neural Repair
,
24
,
282
289
.
Kielar
,
A.
,
Meltzer-Assher
,
A.
, &
Thompson
,
C. K.
(
2012
).
Electrophysiological responses to argument structure violations in healthy adults and individuals with agrammatic aphasia.
Neuropsychologia
,
50
,
3320
3337
.
Klimesch
,
W.
(
1999
).
EEG alpha and theta oscillations reflect cognitive and memory performance: A review and analysis.
Brain Research Review
,
29
,
169
195
.
Klimesch
,
W.
,
Doppelmayr
,
M.
,
Pachinger
,
T.
, &
Russegger
,
H.
(
1997
).
Event-related desynchronization in the alpha band and the processing of semantic information.
Cognitive Brain Research
,
6
,
83
94
.
Klimesch
,
W.
,
Schack
,
B.
, &
Sauseng
,
P.
(
2005
).
The functional significance of theta and upper alpha oscillations.
Experimental Psychology
,
52
,
99
108
.
Klimesch
,
W.
,
Schimke
,
H.
,
Doppelmayr
,
M.
,
Riper
,
B.
,
Schwaiger
,
J.
, &
Pfurtscheller
,
G.
(
1996
).
Event-related desynchronization (ERD) and the Dm-effect: Does alpha desynchronization during encoding predict later recall performance?
International Journal of Psychophysiology
,
24
,
47
60
.
Konishi
,
S.
,
Nakajima
,
K.
,
Uchida
,
I.
,
Kameyama
,
M.
,
Nakahara
,
K.
,
Sekihara
,
K.
,
et al
(
1998
).
Transient activation of inferior prefrontal cortex during cognitive set shifting.
Nature Neuroscience
,
1
,
80
84
.
Konishi
,
S.
,
Nakajima
,
K.
,
Uchida
,
I.
,
Kikyo
,
H.
,
Kameyama
,
M.
, &
Miyashita
,
Y.
(
1999
).
Common inhibitory mechanism in human inferior prefrontal cortex revealed by event-related functional MRI.
Brain
,
122
,
981
991
.
Kotz
,
S. A.
,
Holcomb
,
P. J.
, &
Osterhout
,
L.
(
2007
).
ERPs reveal comparable syntactic sentence processing in native and non-native readers of English.
Acta Psychologica
,
128
,
514
527
.
Kučera
,
H.
, &
Francis
,
W. N.
(
1967
).
Computational analysis of present-day American English
.
Providence, RI
:
Brown University Press
.
Kuperberg
,
G. R.
(
2007
).
Neural mechanisms of language comprehension: Challenges to syntax.
Brain Research
,
1146
,
23
49
.
Kutas
,
M.
, &
Federmeier
,
K. D.
(
2000
).
Electrophysiology reveals semantic memory use in language comprehension.
Trends in Cognitive Sciences
,
4
,
463
470
.
Kutas
,
M.
, &
Hillyard
,
S. A.
(
1980
).
Reading senseless sentences: Brain potentials reflect semantic incongruity.
Science
,
207
,
203
205
.
Makeig
,
S.
(
1993
).
Effects of exposure of pure tones on event-related dynamics of the EEG spectrum.
Electroencephalography and Clinical Neurophysiology
,
86
,
283
293
.
Maris
,
E.
(
2004
).
Randomization tests for ERP topographies and whole spatiotemporal data matrices.
Psychophysiology
,
41
,
142
151
.
Maris
,
E.
, &
Oostenveld
,
R.
(
2007
).
Nonparametric statistical testing of EEG- and MEG-data.
Journal of Neuroscience Methods
,
164
,
177
190
.
Meltzer
,
J. A.
, &
Braun
,
A. R.
(
2011
).
An EEG-MEG dissociation between online syntactic comprehension and post hoc reanalysis.
Frontiers in Human Neuroscience
,
5
,
1
15
.
Meltzer
,
J. A.
,
Negishi
,
M.
,
Mayes
,
L. C.
, &
Constable
,
R. T.
(
2007
).
Individual differences in EEG theta and alpha dynamics during working memory correlate with fMRI responses across subjects.
Clinical Neurophysiology
,
118
,
2419
2436
.
Meltzer
,
J. A.
,
Postman-Caucheteux
,
W. A.
,
McArdle
,
J. J.
, &
Braun
,
A. R.
(
2009
).
Strategies for longitudinal neuroimaging studies of overt language production.
Neuroimage
,
47
,
745
755
.
Mizuhara
,
H.
,
Wang
,
L. Q.
,
Kobayashi
,
K.
, &
Yamaguchi
,
Y.
(
2004
).
A long-range cortical network emerging with theta oscillation in a mental task.
NeuroReport
,
15
,
1233
1238
.
Moreno
,
E. M.
, &
Kutas
,
M.
(
2005
).
Processing semantic anomalies in two languages: An electrophysiological exploration in both languages of Spanish-English bilinguals.
Cognitive Brain Research
,
22
,
205
220
.
Moreno
,
E. M.
,
Rodrigues-Fornells
,
A.
, &
Laine
,
M.
(
2008
).
Event-related potentials (ERPs) in the study of bilingual language processing.
Journal of Neurolinguistics
,
21
,
477
508
.
Moreno
,
S.
,
Bialystok
,
E.
,
Wodniecka
,
Z.
, &
Alain
,
C.
(
2010
).
Conflict resolution in sentence processing by bilinguals.
Journal of Neurolinguistics
,
23
,
564
579
.
Mouraux
,
A.
, &
Iannetti
,
G. D.
(
2008
).
Across-trial averaging of event-related EEG responses and beyond.
Magnetic Resonance Imaging
,
26
,
1041
1054
.
Münte
,
T. F.
,
Heinze
,
H. J.
, &
Mangun
,
G. R.
(
1993
).
Dissociation of brain activity related to syntactic and semantic aspects of language.
Journal of Cognitive Neuroscience
,
5
,
335
344
.
Neville
,
H. J.
, &
Weber-Fox
,
C. M.
(
1996
).
Maturational constraints on functional specializations for language processing: ERP and behavioural evidence in bilingual speakers.
Journal of Cognitive Neuroscience
,
8
,
231
256
.
Newman
,
A. J.
,
Tremblay
,
A.
,
Nichols
,
E. S.
,
Neville
,
H. J.
, &
Ullman
,
M. T.
(
2012
).
The influence of language proficiency on lexical semantic processing in native and late learners of English.
Journal of Cognitive Neuroscience
,
24
,
1205
1223
.
Osterhout
,
L.
, &
Holcomb
,
P. J.
(
1992
).
Event-related brain potentials elicited by syntactic anomaly.
Journal of Memory and Language
,
31
,
785
806
.
Osterhout
,
L.
, &
Holcomb
,
P. J.
(
1993
).
Event-related potentials and syntactic anomaly: Evidence of anomaly detection during the perception of continuous speech.
Language and Cognitive Processes
,
8
,
413
438
.
Osterhout
,
L.
,
Holcomb
,
P. J.
, &
Swinney
,
D. A.
(
1994
).
Brain potentials elicited by garden-path sentences: Evidence of the application of verb information during parsing.
Journal of Experimental Psychology: Learning, Memory, and Cognition
,
20
,
786
803
.
Osterhout
,
L.
, &
Mobley
,
L. A.
(
1995
).
Event-related brain potentials elicited by failure to agree.
Journal of Memory and Language
,
34
,
739
773
.
Osterhout
,
L.
, &
Nicol
,
J.
(
1999
).
On the distinctiveness, independence, and time course of the brain responses to syntactic and semantic anomalies.
Language and Cognitive Processes
,
14
,
283
317
.
Osterhout
,
L.
,
Poliakov
,
A.
,
Inoue
,
K.
,
McLaughlin
,
J.
,
Valentine
,
G.
,
Pitkanen
,
I.
,
et al
(
2008
).
Second-language learning and changes in the brain.
Journal of Neurolinguistics
,
21
,
509
521
.
Pakulak
,
E.
, &
Neville
,
H. J.
(
2010
).
Proficiency differences in syntactic processing of monolingual native speakers indexed by event-related potentials.
Journal of Cognitive Neuroscience
,
22
,
2728
2744
.
Pfurtscheller
,
G.
, &
Klimesch
,
W.
(
1992
).
Functional topography during a visuoverbal judgment task studied with event-related desynchronization mapping.
Journal of Clinical Neurophysiology
,
9
,
120
131
.
Pfurtscheller
,
G.
, &
Lopes da Silva
,
F. H.
(
1999
).
Event-related EEG/MEG synchronization and desynchronization: Basic principles.
Clinical Neurophysiology
,
110
,
1842
1857
.
Picton
,
T. W.
,
Bentin
,
S.
,
Berg
,
P.
,
Donchin
,
E.
,
Hillyard
,
S. A.
,
Johnson
,
R.
,
et al
(
2000
).
Guidelines for using human event-related potentials to study cognition: Recording standards and publication criteria.
Psychophysiology
,
37
,
127
152
.
Portocarrero
,
J. S.
,
Burright
,
R. G.
, &
Donovick
,
P. J.
(
2007
).
Vocabulary and verbal fluency of bilingual and monolingual college students.
Archives of Clinical Neuropsychology
,
22
,
415
422
.
Roehm
,
D.
,
Schlesewsky
,
M.
,
Bornkessel
,
I.
,
Frisch
,
S.
, &
Haider
,
H.
(
2004
).
Fractionating language comprehension via frequency characteristics of the human EEG.
NeuroReport
,
15
,
409
412
.
Röhm
,
D.
,
Klimesch
,
W.
, &
Doppelmayr
,
M.
(
2001
).
The role of theta and alpha oscillations for language comprehensions in the human electroencephalogram.
Neuroscience Letters
,
310
,
137
140
.
Romero
,
S. G.
,
McFarland
,
D. J.
,
Faust
,
R.
,
Farrell
,
L.
, &
Cacace
,
A. T.
(
2008
).
Electrophysiological markers of skill-related neuroplasticity.
Biological Psychology
,
78
,
221
230
.
Sanders
,
L. D.
,
Weber-Fox
,
C. M.
, &
Neville
,
H. J.
(
2008
).
Varying degrees of plasticity in different subsystems within language.
In
J. R.
Pomerantz
(Ed.),
Topics in integrative neuroscience: From cells to cognition
(pp.
125
153
).
Cambridge, UK
:
Cambridge University Press
.
Scheeringa
,
R.
,
Petersson
,
K. M.
,
Oostenveld
,
R.
,
Norris
,
D. G.
,
Hagoort
,
P.
, &
Bastiaansen
,
M. C.
(
2009
).
Trial-by-trial coupling between EEG and BOLD identifies networks related to alpha and theta EEG power increases during working memory maintenance.
Neuroimage
,
44
,
1224
1238
.
Singh
,
K. D.
,
Barnes
,
G. R.
,
Hillebrand
,
A.
,
Forde
,
E. M. E.
, &
Williams
,
A. L.
(
2002
).
Task-related changes in cortical synchronization are spatially coincident with the hemodynamic response.
Neuroimage
,
16
,
103
114
.
Slagter
,
H. A.
,
Kok
,
A.
,
Mol
,
N.
, &
Kenemans
,
J. L.
(
2005
).
Spatio-temporal dynamics of top–down control: Directing attention to location and/or color as revealed by ERPs and source modeling.
Cognitive Brain Research
,
22
,
333
348
.
Sokal
,
R. R.
, &
Rohlf
,
F. J.
(
1995
).
Biometry: The principles and practice of statistics in biological research
(3rd ed.).
New York
:
W.H. Freeman
.
Sterman
,
M. B.
,
Kaiser
,
D. A.
, &
Veigel
,
B.
(
1996
).
Spectral analysis of event-related EEG responses during short-term memory performance.
Brain Topography
,
9
,
21
30
.
Swaab
,
T.
,
Brown
,
C.
, &
Hagoort
,
P.
(
1997
).
Spoken sentence comprehension in aphasia: Event-related potential evidence for lexical integration deficits.
Journal of Cognitive Neuroscience
,
9
,
39
66
.
Swainson
,
R.
,
Jackson
,
S. R.
, &
Jackson
,
G. M.
(
2006
).
Using advance information in dynamic cognitive control: An ERP study of task-switching.
Brain Research
,
1105
,
61
72
.
Wang
,
L.
,
Jensen
,
O.
,
van den Brink
,
D.
,
Weder
,
N.
,
Schoffelen
,
J. M.
,
Magyari
,
L.
,
et al
(
2012
).
Beta oscillations relate to the N400m during language comprehension.
Human Brain Mapping
,
33
,
2898
2912
.
Weber-Fox
,
C.
, &
Neville
,
H. J.
(
1999
).
Functional neural subsystems are differentially affected by delays in second-language immersion: ERP and behavioral evidence in bilingual speakers.
In
D.
Birdsong
(Ed.),
New perspectives on the critical period for second language acquisition
(pp.
23
38
).
Hillsdale, NJ
:
Lawrence Erlbaum
.
Weiss
,
S.
,
Mueller
,
H. M.
,
Schack
,
B.
,
King
,
J. W.
,
Kutas
,
M.
, &
Rappelsberger
,
P.
(
2005
).
Increased neuronal communication accompanying sentence comprehension.
International Journal of Psychology
,
57
,
129
141
.
Willems
,
R. M.
,
Oostenveld
,
R.
, &
Hagoort
,
P.
(
2008
).
Early decrease in alpha and gamma band power distinguish linguistic from visual information during spoken sentence comprehension.
Brain Research
,
1219
,
78
90
.
Zhuang
,
P.
,
Toro
,
C.
,
Grafman
,
J.
,
Manganotti
,
P.
,
Leocani
,
L.
, &
Hallett
,
M.
(
1997
).
Event-related desynchronization (ERD) in the alpha frequency during development of implicit and explicit learning.
Electroencephalography and Clinical Neurophysiology
,
102
,
374
381
.