Magnetoencephalography detection of early syntactic processing in

humans: comparison between L1 speakers and L2 learners of English

Mikio Kubotaa,b,*, Paul Ferrarib, Timothy P.L. Robertsb

aDepartment of English, Seijo University, 6-1-20, Seijo, Setagaya-ku, Tokyo, 157-8511, JapanbDepartment of Radiology, University of California San Francisco, San Francisco, CA, USA

Received 12 June 2003; received in revised form 9 September 2003; accepted 9 September 2003

Abstract

In previous brain imaging studies of human syntax processing, only phrase structure (grammatical category) violations have been shown to

elicit a very early (,140 ms) neural response. This has led to interpretations about the nature of phrase structure encoding in the brain,

particularly its relationship to early automatic brain processes. Utilizing different sentence structures that contrasted within- vs. across-phrase

violations, the current study examined whether an early response could be elicited by non-phrase-structure violations. Magnetoencephalo-

graphy fields were recorded, while both first-language speakers (L1) and second-language learners (L2) were tested. A prominent syntactic

magnetic field component, peaking at around 150 ms post-onset (labeled ‘SF-M150’), was observed in both hemispheres of only the L1

speakers in response to within-phrase violations but not across-phrase violations. The results provide evidence that L1 speakers possess the

ability for automated detection of non-phrase-structure violations, particularly within-phrase violations, and that L2 learners may not have

sufficient neural representation available for an early automated response to the target violations.

q 2003 Elsevier Ireland Ltd. All rights reserved.

Keywords: Language; Syntax; Case; Syntactic violation; Magnetoencephalography; Early left anterior negativity; SF-M150; Within-phrase violation; Across-

phrase violation

Previous event-related brain potential (ERP) studies of

sentence comprehension have reported that three com-

ponents of the response waveform reflect syntactic pro-

cesses at a neural level: the early left anterior negativity

(ELAN) [2], the left anterior negativity (LAN) [10], and the

P600, a positive deflection peaking at about 600 ms [6]. It

has been found that the ELAN component peaks between

125 and 180 ms after the critical word and has a

topographical distribution over the anterior region of the

left hemisphere [3,13]. It has been proposed that the ELAN

component is elicited exclusively by phrase structure (PS)

violations, induced by grammatical category errors –

realized by substitution of a word of a correct grammatical

category (e.g. noun) with an incorrect one (e.g. verb) – and

is related to the operations of building up the initial syntactic

structure [2–5,7,9,11,13,16]. This early neural response has

also been observed using magnetoencephalography (MEG)

during PS violations aurally presented in German, having its

peak around 140 ms in temporal and frontal regions of both

hemispheres [4,11]. Further, in previous ERP studies,

second-language learners failed to exhibit the ELAN

response [7,9].

However, not all grammatical violations disrupt the

syntactic building process. Within-category violations do

qualify, structurally speaking, as phrases and are considered

non-phrase-structure (non-PS) violations. These errors refer

to inflection errors (e.g. tense/case errors) of the obligatory

grammatical category and do not theoretically disrupt the

PS building process. Furthermore, within-category viola-

tions can occur within the phrase boundary in which they

belong (within-phrase violations) or can be otherwise

dependent on a relation outside of the phrase in which

they belong (across-phrase violations). Up to now, the

within- vs. across-phrase (within-category) violation types

have not been tested and it is not clear whether a very early

syntactic response may be evoked by syntactic violations

not strictly related to PS rules, and whether it may be

elicited by ‘advanced’ second-language (L2) learners.

0304-3940/03/$ - see front matter q 2003 Elsevier Ireland Ltd. All rights reserved.

doi:10.1016/j.neulet.2003.09.019

Neuroscience Letters 353 (2003) 107–110

www.elsevier.com/locate/neulet

* Corresponding author. Department of English, Seijo University, 6-1-20,

Seijo, Setagaya-ku, Tokyo, 157-8511, Japan. Tel.: þ81-3-3482-1181; fax:

þ81-3-3482-7740.

E-mail address: mkubota@seijo.ac.jp (M. Kubota).

The current study tested whether such an early syntactic

response would be elicited by English sentences possessing

non-PS (within-category) violations regarding nominative

and accusative case-features, whether the MEG component

would be generated by both within- and across-phrase

violations, and whether advanced L2 learners would display

the same neuromagnetic responses as first-language (L1)

speakers. The subjects were five L1 speakers of American

English, aged 28–49 (average ¼ 38; two females) and five

Japanese L2 advanced learners of English, aged 23–29

(average ¼ 26; four females). All subjects were healthy and

right-handed with normal hearing, the same educational

background, and no known neurological disorders. L2

subjects started studying English as a foreign language at

age 13, and their TOEFL scores ranged from 570 to 590

(average ¼ 575:4). The approval of the Committee for

Human Research and subjects’ informed written consent

were obtained.

Grammatically correct and incorrect versions of two

syntactic structure types of English case features were

compared, tensed complement (TC) and infinitival com-

plement (IC). The number of syntactically correct and

incorrect sentences was equal, and there was no difference

in meaning across conditions. Each sentence followed the

template: ‘I [verb] [pronoun] [is/to be] a spy.’ Three

personal pronouns and 13 matrix verbs were used. For

example,

TC condition: TC(a) I believe he is a spy.

*TC(b) *I believe him is a spy.

IC condition: IC(a) I believe him to be a spy.

*IC(b) *I believe he to be a spy.

(*: ungrammatical)

The sentences were recorded with a natural speech rate and

natural intonation by an American male on a Windows

computer at a sampling rate of 44,100 Hz. The stimuli were

edited using SoundEdit 16 software (version 2; Macro-

media, Inc.). To ensure intonation homogeneity across all

stimuli and avoid prosodic target clues, incorrect stimuli

were created by cross-splicing correct sentences. The time

between the onset of the pronoun (e.g. ‘he’) and the onset of

the critical word (‘is’ or ‘to’) remained the same across all

stimuli. Four L1 speakers listened to the edited sentences

and evaluated the stimuli as natural.

Neuromagnetic fields were recorded, time-locked to the

onset of the mid-sentence critical word, using a dual 37-

channel gradiometer system (MAGNES II, BTi). The

subjects listened to English sentences and made a covert

grammaticality judgment on each stimulus. One hundred

and four different sentences were repeated four times, and

all 416 sentences were presented randomly by PsyScope

software (version 1.2 [1]) on a Macintosh computer. The

onset-to-onset inter-trial interval varied randomly between

3900 and 4100 ms in steps of 100 ms. Subjects lay on a bed

with the head positioned between the two MEG sensors.

Prior to the experiment, positioning of each sensor over the

auditory cortex, overlying the temporal regions, was

confirmed by neuromagnetic responses to 1 kHz pure tones.

Recording epochs of 1100 ms duration, including 100 ms

pre-trigger baseline, were acquired at a sampling rate of

1041.7 Hz and subject to an online 1–100 Hz band-pass

filter. The stimuli were binaurally delivered at an intensity

of at least 55 dB SL, using insert earphones and plastic air

tubes (ER-3A; Etymotic Research). After the 30 min MEG

experiment, the subjects took a computer-based behavioral

grammaticality judgment test.

All epochs were inspected for artifacts, and epochs were

rejected if the min-max value of any sensor exceeded a

threshold of 3000 fT. The waveforms averaged by condition

for each hemisphere were filtered off-line with a 1–40 Hz

band-pass filter and adjusted to the 100 ms pre-trigger

baseline to correct for the drift associated with the DC

offset. The root mean square (RMS) of the magnetic field

strength across sensor channels was calculated. The latency

of the early syntactic processes was determined as the time

point corresponding to the peak RMS field value in the time

window of 80–250 ms. The alpha level with the Bonferroni

adjustment was a ¼ 0:025 for multiple tests. TomTom2000

histogram software was used to plot the RMS distribution

across subjects to ensure that the data displayed a normal

distribution allowing further interpretable statistical

analysis.

RMS evoked field amplitudes for two groups (L1/L2

speakers) and four conditions (correct and incorrect

conditions in the left/right hemisphere (LH/RH)) were

compared by a two-way repeated-measures ANOVA (Fig.

1). For the TC condition, the group by condition interaction

was statistically significant (Fð3;24Þ ¼ 5:98, P ¼ 0:003). The

analysis of the simple main effect revealed that the magnetic

strength of L1 subjects (84.15 ^ 14.59 fT) was larger than

that of L2 learners (45.16 ^ 16.39 fT) in *TC(b) in the LH

(Fð1;8Þ ¼ 12:63, P , 0:01). No other between-group com-

parisons were not statistically significant. Fisher’s Least-

Significant-Difference multiple comparisons of conditions

showed that for L1 speakers, the magnetic strength of the

early response to *TC(b) (LH: 84.15 ^ 14.59 fT, RH:

94.18 ^ 15.01 fT) was larger than that to TC(a) (LH:

57.66 ^ 17.68 fT, RH: 65.68 ^ 10.58 fT) in each hemi-

Fig. 1. Peak RMS amplitudes (magnetic field strength) of an early syntactic

response averaged across subjects for the TC and IC conditions within a

subject group for each hemisphere [error bar, 1 SD]. The RMS values in

both hemispheres of L1 subjects are only statistically larger in the incorrect

*TC(b) than in the correct TC(a).

M. Kubota et al. / Neuroscience Letters 353 (2003) 107–110108

sphere (P , 0:01), with no hemispheric difference being

observed for *TC(b). For the IC condition, the group by

condition interaction (Fð3;24Þ ¼ 0:25, P ¼ 0:86), and the

main effects for groups (Fð1;8Þ ¼ 0:14, P ¼ 0:71) and for

conditions (Fð3;24Þ ¼ 0:99, P ¼ 0:42) were not statistically

significant (see Fig. 2). There was no significant difference

in the latency of early syntactic responses between groups or

conditions for either the TC or IC, e.g. TC: Fð1;8Þ ¼ 3:52,

P ¼ 0:10; IC: Fð1;8Þ ¼ 5:17, P ¼ 0:05 for the main effect of

groups. The mean latency of an early syntactic component

elicited by *TC(b) for L1 subjects was 145.73 ^ 13.27 ms

in the LH and 139.20 ^ 17.10 ms in the RH (no statistical

difference between the LH and the RH).

Hence, a prominent component, peaking at around 150

ms after the onset (labeled ‘SF-M150’) in *TC(b), was

elicited from both hemispheres of L1 speakers. This

syntactic magnetic field component was robustly evoked

by all five L1 subjects. No such component was observed for

any other condition. Additionally, L2 learners failed to

demonstrate a prominent magnetic field in any condition.

For one representative L1 subject, the generators of the

SF-M150 component obtained using BESA software

(MEGIS [15]) were overlaid onto the anatomical MR

axial image within an estimated 4 mm error of measurement

[12], as shown in Fig. 3. For *TC(b), two sources were

identified in the LH: the first source was located in the

lateral fissure and the second one in the superior temporal

gyrus. In the RH, a single source was in the superior

temporal sulcus. All sources were estimated by an in-house

neuro-radiologist. Localizations found in this study are in

line with the previous MEG results [4,11,16]. Detailed

investigation of source localizations may be motivated in

another subsequent study. However, in light of the current

and previous MEG studies finding early (,150 ms) syntax-

related component localizations not only in left anterior

regions but also in temporal regions bilaterally, we chose to

label this response the ‘SF-M150’, instead of the magnetic

counterpart to the ELAN, a term that implies only anterior

generators involved in early syntactic processing.

The major finding was that L1 subjects exhibited a

prominent magnetic field component, evoked by *TC(b),

but not by *IC(b). The difference of evoked responses may

be due to the syntactic difference between *TC(b) and

*IC(b). According to current syntactic theory, the case

feature for TC (tensed complement) is checked internally

within a phrase (e.g. he is a spy), but that of IC (untensed,

infinitival complement) is checked externally across a

phrase by an immediately preceding transitive matrix verb

(e.g. believe) [14]. In TC(a) below, the verb in TC (is)

assigns the TC subject to the nominative case (he). On the

contrary, as in IC(a) the matrix verb which is located outside

IC assigns the IC subject to the accusative case (him).

TC(a): I believe [CP [IP he is a spy]].

IC(a): I believe [IP him to be a spy].

(CP ¼ complementizer phrase)

(IP ¼ inflectional phrase)

Therefore, while both *TC(b) and *IC(b) contain within-

category violations and satisfy the requirement of a

syntactic hierarchy, *TC(b) constitutes within-phrase viola-

tions, whereas *IC(b) constitutes across-phrase violations.

Hence, the SF-M150 component was generated in response

to within-phrase violations as in *TC(b) in each hemisphere

of L1 subjects, but not across-phrase violations as in *IC(b).

The latency of the early response was observed to be

142.46 ^ 15.19 ms averaged over both hemispheres, and

this is in accord with a previous study of MEG syntactic

processing in German [16]. Previous studies have tested and

characterized the automaticity of such early syntax-related

processes and found that the early stage (ELAN) is an

automated process compared to later stage processes (e.g.

P600). These studies supported the theoretical view that

first-pass parsing of syntactic structures is highly automated

[8]. It is speculated then that L1 speakers in the current

study processed the within-phrase violations automatically,

possibly without accessing explicit grammatical knowledge.

The behavioral data showed that no significant difference

was observed between L1 and L2 subjects on a grammati-

cality judgment test (e.g. Fð1;8Þ ¼ 1:96, P ¼ 0:20 for the

main effect of groups). The rate of correct responses was

98.46% for L1 and 92.81% for L2 in *TC(b), and 97.41%

Fig. 2. Seventy-four MEG channels are overlaid from the left and right

hemispheres on the ordinate axis. Responses to correct (TC(a), IC(a)) and

incorrect sentences (*TC(b), *IC(b)) for a representative subject of each

group are shown for each stimulus condition. Only the response to *TC(b)

for the L1 subject has a prominent deflection peaking at around 150 ms.

Fig. 3. Estimated source localizations of SF-M150 elicited by the incorrect

condition *TC(b) for a representative L1 subject. Two sources are identified

in the left superior temporal region and a single source in the right superior

temporal region.

M. Kubota et al. / Neuroscience Letters 353 (2003) 107–110 109

for L1 and 96.15% for L2 subjects in *IC(b). The results

revealed that L2 learners were able to detect the violations

as correctly as L1 speakers at a behavioral level, but they

failed to show the neuronal responses exhibited by L1

speakers. ‘Low-advanced’ L2 learners in this study (TOEFL

575.4) may not possess or have yet to develop such neuronal

mechanisms subserving rapid automated syntactic proces-

sing of the target structures, as apparently the L1 speakers

did. This is in accord with previous ERP studies in which

Russian and Japanese learners of German as L2 failed to

exhibit the ELAN component [7,9]. It remains to be

determined whether ‘high-advanced’ learners (e.g. TOEFL

650) or truly balanced bilinguals will elicit the SF-M150

component as L1 speakers.

Other researchers have interpreted the ELAN and its

neuromagnetic equivalent as reflecting PS violations [3,4,

11,16]. The current study extends their findings, to the

extent that both this study and previous studies [3,4,11,16]

utilized structures containing ‘within-phrase’ violations

(grammatical category violations are inherently within a

phrase), and provides neuromagnetic evidence of an early

syntactic component reflective of non-PS violations,

particularly implicating local (phrase-internal) processes

as a trigger of the SF-M150. Thus, the SF-M150 component

is interpreted as reflecting the automated detection mech-

anism for within-phrase syntactic violations that disrupt the

listener’s structural expectations, whether a PS rule is

violated or not. In other words, this component may

represent the process of failing to integrate the current

word into the initial syntactic structure (cf. Ref. [7]). L1

speakers may possess more neuronal sensitivity to the

detection of within-phrase violations than across-phrase

violations, because the case-feature checking that goes

beyond a phrase may demand other resources, subsequently

diminishing the robustness of encoding. Further, to the

extent that advanced L2 learners may not possess an early

automated mechanism available for detecting non-PS

violations, they most likely rely on different neuronal

mechanisms to process this syntactic information. The

observation that neuronal activity modulation depends on

the particular syntactic structures in the present study

suggests the need to further investigate the possibility of a

neuronal error gravity depending on structure types (e.g.

strong violations as in *TC(b) vs. weak violations as in

*IC(b)).

References

[1] J.D. Cohen, B. MacWhinney, M. Flatt, J. Provost, PsyScope: a new

graphic interactive environment for designing psychology exper-

iments, Behav. Res. Methods 25 (1993) 257–271.

[2] A.D. Friederici, A. Hahne, D.Y. von Cramon, First-pass versus

second-pass parsing processes in a Wernicke’s and a Broca’s aphasic:

electrophysiological evidence for a double dissociation, Brain Lang.

62 (1998) 311–341.

[3] A.D. Friederici, E. Pfeifer, A. Hahne, Event-related brain potentials

during natural speech processing: effects of semantic, morphological

and syntactic violations, Cogn. Brain Res. 1 (1993) 183–192.

[4] A.D. Friederici, Y. Wang, C.S. Herrmann, B. Maess, U. Oertel,

Localization of early syntactic processes in frontal and temporal

cortical areas: magnetoencephalographic study, Hum. Brain Mapp. 11

(2000) 1–11.

[5] T.C. Gunter, A.D. Friederici, Concerning the automaticity of syntactic

processing, Psychophysiology 36 (1999) 126–137.

[6] P. Hagoort, C.M. Brown, L. Osterhout, The neurocognition of

syntactic processing, in: C.M. Brown, P. Hagoort (Eds.), The

Neurocognition of Language, Oxford University Press, Oxford,

1999, pp. 273–316.

[7] A. Hahne, What’s different in second-language processing? Evidence

from event-related brain potentials, J. Psycholinguist. Res. 30 (2001)

251–266.

[8] A. Hahne, A.D. Friederici, Electrophysiological evidence for two

steps in syntactic analysis: early automatic and late controlled

processes, J. Cogn. Neurosci. 11 (1999) 194–205.

[9] A. Hahne, A.D. Friederici, Processing a second language: late

learners’ comprehension mechanisms as revealed by event-related

brain potentials, Bilingualism 4 (2001) 123–141.

[10] R. Kluender, M. Kutas, Bridging the gap: evidence from ERPs on the

processing of unbounded dependencies, J. Cogn. Neurosci. 5 (1993)

196–214.

[11] T.R. Knosche, B. Maeß, A.D. Friederici, Processing of syntactic

information monitored by brain surface current density mapping

based on MEG, Brain Topogr. 12 (1999) 75–87.

[12] R.C. Knowlton, K.D. Laxer, M.J. Aminoff, T.P. Roberts, S.T. Wong,

H.A. Rowley, Magnetoencephalography in partial epilepsy: clinical

yield and localization accuracy, Ann. Neurol. 42 (1997) 622–631.

[13] H.J. Neville, J.L. Nicol, A. Barss, K.I. Forster, M.F. Garrett,

Syntactically based sentence processing classes: evidence from

event-related brain potentials, J. Cogn. Neurosci. 3 (1991) 151–165.

[14] A. Radford, Syntactic Theory and the Structure of English: A

Minimalist Approach, Cambridge University Press, Cambridge, 1997.

[15] M. Scherg, P. Berg, Use of prior knowledge in brain electromagnetic

source analysis, Brain Topogr. 4 (1991) 143–150.

[16] Y. Wang, U. Oertel, M. Meyer, C.S. Herrmann, B. Maess, A.D.

Friederici, MEG source localization of early syntactic processes.

Poster presented at the 6th annual meeting of the Organization for

Human Brain Mapping, San Antonio, TX, 2000.

M. Kubota et al. / Neuroscience Letters 353 (2003) 107–110110