When One Personʼs Mistake Is Anotherʼs Standard Usage:
The Effect of Foreign Accent on Syntactic Processing
Adriana Hanulíková1,2, Petra M. van Alphen1, Merel M. van Goch3,
and Andrea Weber1,4
Abstract
■ How do native listeners process grammatical errors that are
frequent in non-native speech? We investigated whether the neural correlates of syntactic processing are modulated by speaker
identity. ERPs to gender agreement errors in sentences spoken
by a native speaker were compared with the same errors spoken
by a non-native speaker. In line with previous research, gender
violations in native speech resulted in a P600 effect (larger P600
for violations in comparison with correct sentences), but when
INTRODUCTION
Modern trends of global mobility have increased the number of people who need to communicate in languages
other than their mother tongue. As a result, people are
encountering non-native (L2) speakers of their language
(Cheng, 1999). These L2 speakers are often recognizable
by an accent in pronunciation or by grammatical errors,
as it is difficult to achieve native-like proficiency (Birdsong
& Molis, 2001; Flege, 1995). Despite the differences encountered in language usage, communication normally
succeeds, as native (L1) listeners can rapidly adjust to
foreign-accented speech and improve their comprehension of it (Bradlow & Bent, 2008; Clarke & Garrett, 2004).
They can even become desensitized to L2 grammatical errors. Analyses of natural conversations between L1 and L2
speakers show that L1 speakers rarely correct L2 syntactic
errors (Chun, Day, Chenoweth, & Luppescu, 1982) and
possibly do not detect otherwise salient mistakes. Given
L1 listenersʼ experience with frequent or fossilized L2
errors, the impact of these mistakes on language comprehension could be modulated by language usage. However,
little is known about how and when the brain processes
such grammatical mistakes. Using EEG, this study examines the neural consequences of foreign-accented speech
1
Max Planck Institute for Psycholinguistics, Nijmegen, the Netherlands, 2Basque Center on Cognition, Brain, and Language,
Donostia, Spain, 3Radboud University, Nijmegen, the Netherlands,
4
Donders Institute for Brain, Cognition, and Behaviour, Nijmegen,
the Netherlands
© 2012 Massachusetts Institute of Technology
the same violations were produced by the non-native speaker
with a foreign accent, no P600 effect was observed. Control sentences with semantic violations elicited comparable N400 effects
for both the native and the non-native speaker, confirming no
general integration problem in foreign-accented speech. The
results demonstrate that the P600 is modulated by speaker identity, extending our knowledge about the role of speakerʼs characteristics on neural correlates of speech processing. ■
on processing grammatical errors frequently produced by
L2 speakers.
Listening to spoken language is usually effortless, despite the fact that listeners have only a very short time to
process and integrate various sources of information, including phonology, semantics, syntax, and pragmatics.
This proficiency stems from a lifetime of experience with
their native language that gives L1 listeners certain expectations about these different sources of information. Such
expectations can result in (rather unconscious) anticipations, helping listeners interpret and predict upcoming
events (for an overview on mechanisms of expectations or
predictions across various domains, see, e.g., Bar, 2007).
For example, world knowledge about Dutch trains helps
comprehenders interpret the phrase Dutch trains are white
as false, because Dutch trains are yellow (e.g., Van Berkum,
Van den Brink, Tesink, Kos, & Hagoort, 2008; Hagoort, Hald,
Bastiaansen, & Petersson, 2004). Similarly, listeners know
that a man is unlikely to say Iʼm pregnant, because it is
biologically implausible (Van Berkum et al., 2008). Listeners
anticipate what might be said and use their world knowledge or stereotype-driven inferences about the speaker,
contributing to effective language processing. Studies on
changes in the electrical activity of the brain show that conflicts with inferences about what a given speaker might say
lead to qualitatively distinct brain ERPs in L1 comprehenders
(Van Berkum et al., 2008). Van Berkum et al. presented
participants with utterances that could be interpreted as
either consistent or inconsistent with the speakersʼ age,
the gender, or the social economic status, as conveyed by
the speakersʼ voice. Van Berkum et al. found that speaker
inconsistency (e.g., hearing a man producing the word
Journal of Cognitive Neuroscience 24:4, pp. 878–887
pregnant in the sentence I might be pregnant because I
feel sick) elicited a larger N400 compared with speaker
consistency (e.g., hearing a woman producing the word
pregnant in the same sentence). This modulation of the
N400 effect suggests that listeners used what they inferred
about the speakers based on their voice characteristics in
the earliest stages of meaning construction. It is less clear,
however, whether inferences about the speaker could
affect neural correlates of syntactic processing.
Listeners can no doubt anticipate likely syntactic structures of their language. For example, listeners expect a
noun phrase after the verb selected in the English utterance the scientist selected *to win the prize. If their expectations are not met, distinctive ERPs to such phrases can be
observed (Osterhout & Holcomb, 1992, 1993). Osterhout
and Holcomb observed a larger P600 effect to the word
to in the above low-probability syntactic construction as
compared with the same word in a more probable construction such as the scientist hoped to win the prize.
The modulation of the P600 effect in their study suggests
that listeners use the knowledge about probable structures
in their L1 during syntactic processing. However, none
of the previous studies on syntactic processing varied
speakerʼs characteristics. It is, therefore, less clear whether
neural correlates of syntactic processing are also affected
by listenerʼs knowledge about frequent or infrequent syntactic structures as a function of speakerʼs identity.
The majority of studies on syntactic processing employing grammatical conflicts were conducted in the visual
modality. In the studies using spoken language, grammatical conflicts were usually produced by L1 speakers (for a
review, see Kutas & Federmeier, 2007). However, it can
be assumed that L1 comprehenders know that such grammatical errors are uncommon among L1 speakers, whereas
they are frequent in L2 speech. This raises the following
questions: How do L1 listeners solve syntactic conflicts in
L2 speech, which is a more natural environment for error
occurrences? And does speaker identity influence the neural dynamics of processing grammatical errors? To answer
these questions, we take advantage of the presence of
grammatical errors in L2 speakers of Dutch.
Among the most troublesome features of acquiring an
L2 is grammatical gender, yet it plays an important role
in language processing in many languages. Dutch nouns,
for example, belong to either neuter gender, associated
with the determiner het, or common gender, associated
with the determiner de. A nounʼs gender is relevant in
adjectival agreement (Booij, 2002). Prenominal adjectives
in singular indefinite noun phrases have the inflection
-e (creating an additional syllable) for common nouns
(groene winkel, “green store”) but not for neuter nouns
(groen huis, “green house”). Because the gender assignment to nouns differs cross-linguistically (Corbett, 1991),
it is often difficult to predict the gender class of a noun
in a second language. Unsurprisingly, L2 learners across
all proficiency levels and across different languages experience difficulties with gender agreement (Sabourin,
2006; Franceschina, 2005; Guillelmon & Grosjean, 2001),
and they have even more difficulties when their L1 lacks
grammatical gender (Sabourin, 2006; Franceschina, 2005).
Listeners are sensitive to the correct use of grammatical gender in their L1, as is reflected in studies showing
that matching gender is easier to process than mismatching gender (Boelte & Connine, 2004; for a review, see
Friederici & Jacobsen, 1999). ERP studies employing contrasts between grammatical violations and correct sentences have shown several components responsive to
morphosyntactic agreement involving gender (e.g., Barber
& Carreiras, 2005; Gunter, Friederici, & Schriefers, 2000;
Münte & Heinze, 1994), such as a LAN and a late posterior
positivity (P600). ERP studies in Dutch employing gender
violations such as noun phrases preceded by an incorrect
article (*het winkel, “store”) or by an incorrectly inflected
adjective (*groen winkel, “green store”) have revealed
a clear P600 effect (but no LAN) in L1 comprehenders
(Van Berkum, Brown, Zwitserlood, Kooijman, & Hagoort,
2005; Hagoort & Brown, 1999). This late positive deflection is assumed to index, among other, processes of syntactic reanalysis or recovery from well-formedness conflicts
(Bornkessel-Schlesewsky & Schlesewsky, 2009; Friederici,
1995; Hagoort, Brown, & Groothusen, 1993; for a review, see Kutas & Federmeier, 2007; Osterhout & Holcomb,
1992) and may reflect controlled and strategic processes
(Hahne & Friederici, 1999; Coulson, King, & Kutas, 1998;
Gunter, Stowe, & Mulder, 1997).
If the underlying processes associated with syntactic
analysis differ as a function of speaker identity, this should
be reflected in the late positive deflection. One of the
largest immigrant communities in the Netherlands are
Turkish speakers (Statistics Netherlands, 2010), thus making many Dutch listeners familiar with Turkish-accented
Dutch. Turkish lacks grammatical gender, and corpus
studies of L2 Dutch show that Turkish learners of Dutch
frequently omit gender marking or use incorrect gender
agreement up to 67% of the time (e.g., Orgassa, 2009;
Blom, Polišenská, & Weerman, 2006; Weerman, Bisschop,
& Punt, 2006). Gender agreement errors are common
even in later generations of Turkish immigrants (Cornips,
2008) and are also used as a stereotype about L2 speakers
or immigrants (e.g., on Dutch entertainment television programs). In a communicative situation involving L1 speakers,
L1 listenersʼ default expectation is for gender agreement
to be well formed (as determined by their linguistic experience). In an interaction with an L2 speaker, in contrast, listenersʼ default expectations might be governed by higher
probabilities of gender errors in foreign-accented speech.
This could result in a different strategy of dealing with these
mistakes and possibly relaxing their sensitivity to grammatical errors. Previous research indeed suggests that L1
listeners can develop a keen ear for meaning in foreignaccented speech (Cheng, 1999; Galloway, 1980), while
focusing less on errors and foreign accents. Listeners could
then allocate processing resources away from incorrect
syntax in L2 speech. This would not necessarily apply in L1
Hanulíková et al.
879
speech, where the probability of gender errors is considerably lower than in L2 speech. As a consequence, frequent
gender agreement errors produced with a Turkish accent
might not elicit the same brain responses as would such
errors spoken with an L1 Dutch accent.
We used a 2 × 2 factorial design and compared Dutch
listenersʼ ERP responses to Dutch nouns agreeing or
disagreeing in gender with previous context in sentences spoken either by a native speaker of Dutch or
by a Turkish learner of Dutch. On the basis of previous
research, we predicted that gender violations spoken by
the L1 speaker should result in a P600 compared with
correct sentences. If listeners are sensitive to higher error
probabilities in L2 speech, the same grammatical errors
produced by the L2 speaker, however, should result in
a modulation of the P600 amplitude. On the basis of
previous research on the role of speakerʼs identity, we
predicted either a smaller P600 or no P600 compared
with correct sentences. We thus expected to observe an
interaction of the two P600 effects as a function of the
accent of the speaker.
To control for possible effects of shallow processing or
integration difficulties in foreign-accented speech compared with native speech, a control set of sentences including semantic world knowledge violations was also
created. An ERP response frequently observed for these
violations is the N400 effect (e.g., Kutas & Federmeier,
2007; Kutas & Hillyard, 1980). A consistent finding in the
literature on the N400 effect is that its amplitude is negatively correlated with the fit of a word in the (semantic)
context. We hypothesized that semantic violations such
as I put a thick evening on my bed create a conflict with
both L1 and L2 speakersʼ typical productions, because
neither of the two speaker groups is likely to produce such
sentences. We therefore predicted that these violations
should elicit comparable N400 effects in both native and
foreign-accented speech. This would confirm that there
are no general comprehension difficulties because of a
foreign accent.
Materials and Pretests of Materials
The critical stimuli consisted of 240 sentences, selected
from an initial set of 320 sentences after auditory recordings
and pretesting (see below). Each sentence was recorded
in four versions; a correct and an incorrect version of each
sentence were spoken by a native speaker and by a nonnative speaker (resulting in 960 sentences). Half of the incorrect sentences contained gender disagreement between
the definite determiner and the noun, and the other half
contained incorrectly inflected adjectives (see Table 1 for
example sentences). A further set of 104 control sentences
was selected from 120 sentences after auditory recordings.
Each sentence was recorded in four versions in the same
way as the critical stimuli (resulting in 416 sentences). The
incorrect sentences contained semantic violations (see
Table 1). The critical nouns at which the grammatical or
semantic violation became apparent were embedded in a
subordinate clause following a main clause, at least five
syllables before the end of the subordinate clause. All critical
nouns were balanced for gender (half neuter) and frequency
(using the Spoken Dutch Corpus; Oostdijk, 2000).
The sentences were spoken by a female Dutch native
speaker and by a female Turkish speaker of Dutch, respectively. The Turkish speaker spoke Dutch fluently, but
with a clear foreign accent, as determined with accent ratings stemming from a questionnaire collected in a pretest
(see below). On a scale from 1 (strong foreign accent) to 5
(no foreign accent), the Turkish speaker had an average
of 2.4. Both speakers received a list of all sentences in both
versions (correct and incorrect) in a randomized order.
They were asked to read the sentences at a natural speech
rate. To minimize possible differences in the speech rate
and the intonation across speakers and conditions, each
sentence was first produced by the native Dutch speaker
Table 1. Sentences with English Translation
Sentences with Gender Agreement
Ik wil een reis naar China maken, omdat de/*het cultuur
daar zo anders is dan hier.
METHODS
Participants
Thirty-four native speakers of Dutch with no hearing, neurological, or psychiatric disorders volunteered to participate.
They were students (17 women, all right handed except
for two ambidextrous, mean age = 22 years, range = 18–
30 years) at Radboud University Nijmegen. None of the
students reported a Turkish background or knowledge of
the Turkish language. After the experiment, 28 participants
correctly recognized that the speaker was from Turkey.
One of the participants reported not to be familiar with
the foreign accent of the speaker, two participants reported the speaker to be Moroccan, and the remaining
three participants reported hearing a German, an Iraqi,
and a Japanese accent, respectively.
880
Journal of Cognitive Neuroscience
“I want to make a trip to China, because the culture there
differs from the one over here.”
Mijn moeder belde in paniek op, omdat een duur/*dure
juweel uit haar tas was gestolen.
“My mother panicked because expensive jewelry had been
stolen from her bag.”
Control Sentences with Semantic Manipulation
Het was vannacht best koud, dus ik had een dikke deken/
*avond op mijn bed gelegd.
“It was very cold last night, so I put a thick blanket/evening
on my bed.”
Critical words are in italics. Asterisk indicates an incorrect word in a
given context.
Volume 24, Number 4
and immediately repeated by the Turkish speaker of Dutch.1
In native speech, the mean duration of the critical words
was 411 msec (SD = 98 msec) and the mean duration
of the whole sentence was 4993 msec (SD = 582 msec).
In non-native speech, the mean duration of the critical
word was 387 msec (SD = 101 msec) and 5124 msec (SD =
614 msec) of the whole sentence.2 All sentences were adjusted for amplitude. The mean logarithmic critical word
form frequency per million was 2.23 (SD = 0.61) in sentences with the gender agreement manipulation and 1.93
(SD = 0.63) in sentences with the semantic manipulation.
To select the best material, all recorded sentences containing gender agreement manipulations were pretested
in an off-line error detection task. Four randomized lists
were created with only one of the four different versions
of each sentence occurring in a given list. Participants (20
L1 Dutch university students in Nijmegen, mean age =
21 years, range = 18–25 years, 18 women, none of whom
participated in the EEG experiment) were asked to press
a response button when they heard a grammatical error.
No information about the nature of the errors was provided. Sentences in which errors were either missed or
false-reported by more than five participants were excluded from the EEG experiment. Participants showed
93% correct error detections in non-native speech and
97% correct error detections in native speech (one participant was excluded because of a misdetection rate of
3 SD below the participantsʼ average).3
Four experimental lists were then created with one
version of each of the 240 critical sentences and the 104
control sentences occurring only in one experimental list
(each list thus contained altogether 344 sentences). The
frequency of the critical words was matched across conditions. In total, 35% (n = 120) of all sentences within
an experimental list contained a gender violation (half of
which were spoken by the native speaker, and the other
half were spoken by the Turkish speaker), and 15% (n =
52) of all sentences contained a semantic violation (half
of which were spoken by the native speaker, and the other
half were spoken by the Turkish speaker).
Procedure
Participants were seated in a comfortable chair and were
told that they would hear two speakers talking about their
lives. No information about the background or the accent
of the speakers was provided. The 344 utterances were
presented over loudspeakers. Participants were asked to
listen for comprehension and were told that they would
have to answer comprehension questions following some
sentences. The questions (12 yes or no questions, half of
which required a “yes” response) were added to ensure
that participants paid attention during the experiment. No
grammaticality judgment or acceptability task was used.4
After the presentation of each utterance, a cross appeared
in the middle of the screen to indicate that participants
could blink or move. Participants were given button-press
control over the initiation of the next trial, which started
with a silence of 1000 msec followed by the utterance.
The experiment consisted of eight blocks (each block
lasted on average 8 min) and seven breaks. Participants
completed a language-background questionnaire after the
EEG study.
EEG Recording and Analysis
EEG was recorded from 34 Ag–AgCl electrodes (impedance was kept below 10 kΩ) at standard 10–20 locations
(Fz, Cz, Pz, Oz, Fp1/2, F3/4, F7/8, F9/10, FC1/2, FC5/
6, FT9/10, C3/4, T7/8, CP1/2, CP5/6, P3/4, P7/8, O1/2,
PO9/10). All recordings were referenced to the left mastoid during recording (eye movement and blink artifacts
were recorded from F9 to F10 and from Fp1 to an additional EOG electrode below the left eye), amplified with
BrainAmp DC amplifiers (0.016–100 Hz band pass, digitized at 500 Hz), and re-referenced off-line to the mastoid
average. EEG segments ranging from 200 msec before to
1500 after critical word onset were extracted and baseline
corrected to a 200-msec pre-onset baseline. Two participants were excluded because of recording problems,
a further participant was excluded because of extensive
alpha (exceeding 3 SD from the mean alpha across all participants and all trials), and one participant was excluded
because his error rate on the comprehension questions
during the EEG session exceeded 3SD from participantsʼ
average (4%). Across the remaining 30 participants, the
segments with potentials above ±75 μV were marked as
artifacts and deleted from further analyses (average segment loss = 13%, range = 12–15%, no difference between
conditions). The segments were averaged per participant
and condition, and mean amplitudes in specific time windows were analyzed with repeated measures ANOVAs. First,
the variation of effect size over all electrodes was evaluated,
after which a topography-oriented analysis was conducted
involving anterior (Fp1/2, F3/4, F7/8, FC1/2, FC5/6, F9/10,
FT9/10, Fz) and posterior distribution (CP5/6, CP1/2, P7/8,
P3/4, PO9/10, O1/2, Pz, Oz).
For the statistical analyses, the time window after the
critical word onset was chosen in line with previous research within the auditory modality and on the basis of
visual inspection of the averaged data. For the P600 effect, the time window was 800–1200 msec (for a similar
time window, see, e.g., Friederici & Oberecker, 2008;
Rossi, Gugler, Hahne, & Friederici, 2005; Koester, Gunter,
Wagner, & Friederici, 2004), and for the N400 effect, it
was 300–500 msec (see, e.g., Kutas & Federmeier, 2007).
Note that the detection of the agreement error is only
possible once the critical noun has been heard and recognized and the gender of the noun has become available. The resulting P600 effect can therefore be delayed
in speech. Although the N400 condition also requires the
recognition of the critical noun, the retrieval of the nounʼs
Hanulíková et al.
881
gender information is not necessary for the detection of
the semantic incongruity.
RESULTS
As can be seen in Figure 1A, gender violations in native
speech resulted in a larger P600 compared with correct
sentences, with a clear posterior distribution. The variation of the effect size over all electrodes was confirmed in
an ANOVA with the factors Correctness (violation, control) and Electrodes (all 34), revealing a significant interaction in the 800–1200 msec window (F(1, 33) = 6.61,
p < .001, ηp2 = .186). To examine whether the effect indeed had a posterior distribution, a topography-oriented
analysis was conducted, with the factors Distribution and
Correctness. This analysis confirmed that the effect was
larger over the posterior than the anterior area (Distribution × Correctness: F(1, 29) = 12.21, p = .002, ηp2 =
.296). Follow-up analyses revealed a significant P600 effect to violations compared with correct sentences across
all posterior electrodes (F(1, 29) = 8.66, p = .006, ηp2 =
.230) but not across all anterior electrodes (F < 1). No
other significant differences were found in an earlier time
window of 300–500 msec.
Interestingly, when the same listeners heard the same
type of gender violations in non-native speech, no such
P600 effect was observed (see Figure 1B). The lack of
variation in effect size across all electrodes was confirmed
by a nonsignificant interaction of the factors Correctness
and Electrodes (F(1, 33) = 1.95, p = .10, ηp2 = .063).
None of the topography-oriented and follow-up analyses showed significant effects, and no other significant
differences were found in an earlier time window of
300–500 msec.
To examine whether the size of the P600 effect depended on the accent of the speaker, we compared the
ERP effect elicited by the native speaker to that elicited
by the non-native speaker, considering posterior electrodes only. A 2 (Accent of the Speaker) × 2 (Correctness)
ANOVA confirmed that the two effects differed in size
across posterior electrodes (F(1, 29) = 7.18, p = .01, ηp2 =
.198), suggesting that the P600 effect was modulated by
the accent of the speaker.
Although the results so far suggest that the P600 effect
was only present in L1 speech, we examined the possibility that the P600 effect was present in L2 speech too but
diminished during the course of the experiment because
of an adjustment to grammatical errors. For this purpose,
the data was split into two blocks corresponding to the
first and second halves of the experiment. As can be seen
in Figure 2, there was a clear posteriorly distributed P600
to violations compared with correct sentences spoken by
the L1 speaker in the first block (as confirmed by a main
effect of condition over posterior electrodes, F(1, 29) =
11.94, p = .002, ηp2 = .292). This was not found in the
second block (F < 1). For the L2 speaker, neither of the
two blocks showed a P600 effect over posterior electrodes (both Fs < 1). Follow-up analyses over all posterior electrodes and both accents confirmed that the size
of the P600 effect was modulated by the accent of the
Figure 1. (A, B) Grand average ERPs from nine scalp sites elicited by syntactically incorrect nouns (dashed lines) and syntactically correct nouns
(solid lines) in native speech (A) and non-native speech (B). Waveforms are filtered (5 Hz high cutoff, 12 dB/oct) for presentation purpose only.
882
Journal of Cognitive Neuroscience
Volume 24, Number 4
Figure 2. Grand average ERPs
from the parietal site PZ elicited
by syntactically incorrect nouns
(dashed lines) and syntactically
correct nouns (solid lines) in
native and non-native speech
in the first and second block.
speaker in the first block (Accent of the Speaker × Correctness: F(1, 29) = 4.99, p = .03, ηp2 = .147), but not in
the second block (F < 1). We return to this result in the
discussion below.
To exclude the possibility of shallow processing or
integration problems in foreign-accented speech, control
sentences with semantic manipulation were analyzed. As
can be seen in Figure 3A and B, semantic violations elicited
a larger posteriorly distributed N400 than correct sentences
in the 300–500 msec window for both the native speaker
and the non-native speaker. An ANOVA with the factors
Correctness and Electrodes (34) revealed a significant interaction for both the native speech (F(1, 33) = 6.54, p < .001,
ηp2 = .184) and the non-native speech (F(1, 33) = 5.87,
Figure 3. (A, B) Grand average ERPs from nine scalp sites elicited by semantically correct nouns (dashed lines) and semantically incorrect
nouns (solid lines) in native speech (A) and non-native speech (B).
Hanulíková et al.
883
p < .001, ηp2 = .168). These results confirm that the N400
effects varied in size across all electrodes. A topographyoriented analysis showed that the effect was indeed larger
over the posterior than the anterior area for the native
speech (Distribution × Correctness: F(1, 29) = 8.32, p =
.007, ηp2 = .223) but not for the non-native speech (F(1,
29) = 1.62, p = .21, ηp2 = .053). Follow-up analyses revealed a significant N400 effect across all posterior electrodes for both the native speech (F(1, 29) = 22.71, p <
.001, ηp2 = .439) and the non-native speech (F(1, 29) =
13.54, p = .001, ηp2 = .318). Across all anterior electrodes,
the N400 was marginally significant in the native speech
(F(1, 29) = 3.73, p = .06, ηp2 = .114) and reliable in the
non-native speech (F(1, 29) = 7.61, p = .01, ηp2 = .208).
The size of the posterior N400 effect did not depend
on the accent of the speaker (Accent of the Speaker × Correctness: F < 1), suggesting that the N400 effect was not
modulated by the accent of the speaker. Although the
results indicated that the N400 had a broader distribution
for the non-native speaker (anterior and posterior distribution) than for the native speaker (posterior distribution
only), the size of the anterior N400 effect did not depend
on the accent of the speaker (Accent of the Speaker × Correctness: F < 1). This result confirms that listeners were
attending equally well to sentences spoken by both the
native speaker and the non-native speaker.
DISCUSSION
Unlike native speech, the speech of second language learners typically contains grammatical errors. However, there
is a dearth of studies on this more intuitive scenario in
which errors are likely to occur in actual speech, despite
the large literature on ERP effects observed to grammatical
violations. The present study showed that gender agreement errors led to a change in the electrophysiological response to the subsequent noun when L1 listeners heard
these mistakes made by an L1 speaker, but not when the
same mistakes were made by an L2 speaker with a clear
non-native accent. Importantly, L1 listeners did not appear
to experience comprehension or shallow processing problems in L2 speech, as indicated by almost equivalent electrophysiological responses to world knowledge violations
produced by either the L1 speaker or the L2 speaker.
Although processing world knowledge in L2 speech thus
did not differ from L1 speech, syntactic processing as reflected by the P600 clearly was affected.
What is the source of this difference in the electrophysiological response to grammatical errors? It seems that listenersʼ previous experiences with the correlation between
L2 accent and error likelihood have modified their expectations about the grammatical well-formedness of foreignaccented speech. Consequently, this could have led to a
differential impact of grammatical errors on processing. It
can be assumed that L1 comprehenders have experienced
that the most problematic aspects of L2 acquisition are
morphosyntax and pronunciation, both of which do not
884
Journal of Cognitive Neuroscience
necessarily preclude successful communication in the
absence of native-like proficiency. Persevering gender
agreement errors, for example, could be relatively well
overlooked during the process of sense making, because
they might not affect the communicative intention of the
speaker. On the basis of the accumulated evidence of such
L2 errors in real life, listeners might have developed sensitivity to distributional differences of grammatical errors
between L1 and L2 speech and could thus reduce attempts
to repair these grammatical errors. The absence of a P600
effect in the present study might reflect such a reduction
of grammatical repairs.5 In native speech, in contrast, gender errors are less probable. L1 listeners are therefore more
likely to attempt grammatical repair processes, as indicated
by a P600 in the present study. This would suggest that
L1 listeners have adjusted their probability model about
error occurrences in L2 speech but not in L1 speech.
An adjustment to errors seems in line with our block
analysis showing no P600 effect to grammatical violations
spoken by the L2 speaker in either of the two blocks. Interestingly, grammatical violations spoken by the L1 speaker
elicited a P600 only in the first but not the second block.
This suggests that adjustments to errors occur even in L1
speech. However, unlike in L2 speech, the adjustments in
L1 speech seem to be rather gradual. It thus appears that
listeners adjust their probability model about error occurrences resulting in fewer attempts of grammatical repairs
even when errors are repeatedly produced by an L1 speaker.
This is in line with previous studies that showed flexibility
of brain responses to L1 grammatical errors as a function
of error probability within an experiment (e.g., Hahne &
Friederici, 1999; Coulson et al., 1998; Gunter et al., 1997).
Hahne and Friederici, for example, compared ERP responses to phrase structure violations with responses to
correct sentences when the proportion of violations within
an experiment was either 20% or 80% of all trials. L1 comprehenders showed a P600 to violations in the 20% condition, but the P600 was largely diminished in the 80%
condition. Although no split analysis was provided to directly evaluate the adjustment to errors across the course
of the experiment, the overall results suggest that listeners
refrained from repair attempts because of the extremely
high proportion of errors even in L1 speech. In other
words, if an error becomes the “standard” (as in the high
error proportion manipulation), listeners are able to adjust
to these errors and can direct their attention to what they
hear in a way needed for a successful comprehension. Note
that the present study contained a constant number of
grammatical violations (35%) in both the L1 and the L2
speech. The modulation of the P600 effect was, therefore,
not driven by proportion manipulation within the study
but rather by inferences about what a certain speaker is
likely to say. How quickly listeners adjust can be, thus, also
modulated by speakerʼs identity. As the present study
demonstrates, adjustments in brain responses observed
to grammatical violation depend on whether the speaker
has a native or a non-native accent.
Volume 24, Number 4
Why would adjustments differ as a function of the accent of the speaker? One possibility is that the outcome
of the block analysis reflects a short-term adjustment to
errors during the experiment for the L1 speaker and a
long-term adjustment to errors for the L2 speaker. Longterm adjustment results from the amount of accumulated experience with L2 speakers, leading to changes in
estimates of the probability of correct formal features.
Non-native speech often contains errors. Moreover, it is
associated with a salient cue to the different probability
distribution—namely, the foreign accent (indeed, over
80% of the participants recognized the Turkish accent).
For these reasons, the error patterns in non-native speech
might be easier to overlook, whereas the error patterns in
native speech are less typical and require a certain amount
of evidence to achieve adjustment for a specific native
speaker. A second possible interpretation of the adjustment results could be based on a stronger decline in attention for L2 speech during the experiment. In this case,
however, a similar decline in attention would be expected
for the N400 effect. We examined this possibility with
an additional block analysis of the semantic condition,
which revealed a constant N400 over blocks in both the
L1 and the L2 speech. This suggests that adjustment to
syntax but not to world knowledge was modulated by
prior experience.
Overall, this result extends our knowledge about the role
of speakerʼs characteristics on neural correlates of speech
processing. Previous research has already shown that listeners take into account speakerʼs identity during meaning
construction (Van Berkum et al., 2008). Speaker inconsistency, such as the improbable sentence I am pregnant
uttered by a male voice elicited a larger N400 than the
same sentence uttered in a more probable context of a
female voice. The present study has shown that syntactic
processing as reflected by the P600 too can be modulated
in a similar manner. Inconsistencies, on the basis of infrequent events, such as a native speaker producing a
gender error, elicited a larger P600 compared with the
same gender errors produced in a more likely context of
foreign-accented speech. The linguistic brain thus takes
into account all information available to achieve an effortless and successful comprehension of spoken language.
This study not only suggests a certain degree of tolerance with respect to a suboptimal syntactic fit in L2 speech
but also shows tolerance with respect to the foreign accent.
Many studies on effects of foreign-accented speech on language comprehension (for a review, see Munro & Derwing,
1995) propose that phonology interferes with comprehension to a greater extent than syntax (Van Heuven, 1986).
Other studies claim that grammar is more detrimental than
phonology (Ensz, 1982), whereas yet others believe that the
probability of a given error is a more decisive factor in the
comprehension of L2 speech (Albrechtsen, Henriksen, &
Færch, 1980). However, a foreign accent does not necessarily impinge upon communication. Ultimately, the listenerʼs
concern is to rapidly extract meaning from the speech sig-
nal and, as the present study shows, listeners succeed
in doing so relatively well in both native and non-native
speech, even if it sometimes requires overlooking the syntactic rules of their language. Future research needs to examine whether other types of mistakes by L2 speakers
show similar results. It is possible that not all errors are
equal in gravity and that grammatical errors altering meaning affect parsing more than errors which are less crucial
for meaning.
The present results also speak to a larger debate about
the functional interpretation of the P600 effect. The P600
effect has been often associated with the processing of
incorrect syntactic constructions (Hagoort et al., 1993)
as well as infrequent syntactic constructions (Osterhout
& Holcomb, 1992, 1993). Several studies have shown a
dependence of the P600 response on manipulations of
participantsʼ expectations concerning the frequency of
error occurrences within an experiment (e.g., Hahne &
Friederici, 1999; Coulson et al., 1998; Gunter et al., 1997;
but see Osterhout, McKinnon, Bersick, & Corey, 1996,
who do not confirm such dependency). On the basis of
these studies, the P600 has been described as a controlled
rather than an automatic response. Some researchers have
therefore argued that the P600 is not language specific
but that it is part of the P300 family (e.g., Coulson et al.,
1998). Specifically, it resembles the P3b component, which
is sensitive to probability manipulations of an event. The
present result indeed confirms that the late positivity is
modulated by error probability, and shows that the error
probability can be inferred on the basis of the accent of
the speaker as well as on manipulations of the amount of
errors within an experiment. Thus, the late positivity can
be modulated by participantsʼ inferences about error occurrences as a function of speaker. Although the present
results do not allow resolving the debate on whether this
late negativity is a P600 or rather part of the P300 family,
they do demonstrate that the late component is a suitable tool for investigation of experiential factors in spokenlanguage processing. Future research could further explore
this issue and examine the effect of accent of the speaker
on earlier, automatic syntactic processes.
In conclusion, achieving mutual understanding despite
the large amount of variation in spoken language is an
amazing human achievement. People are rapid in adapting
to new situations and new speakers, and are able to reduce
interference effects following conflicts. Given that cognitive control is dynamic, monitoring of conflicts leads to
behavioral adjustments. Listeners can effectively use a
foreign accent as a cue for non-nativeness and adjust their
probability model to make the communication a successful enterprise. Although L1 listeners can certainly detect
grammatical errors in L2 speech if explicitly required to
do so, the impact that these errors have on them in a
naturalistic listening situation is not detrimental. Moreover,
L1 comprehenders can adjust to real-life probabilities of
error occurrence. This is good news for L2 speakers who
are often embarrassed for producing grammatical errors.
Hanulíková et al.
885
Overlooking grammatical errors in a conversation could
then be seen as a natural and automatic attempt to ensure
successful communication between native and non-native
speakers.
Acknowledgments
We thank Laurence Bruggeman and Karina Visser for their help
with sentence construction and Laurence Bruggeman and Anne
Blankenhorn for assistance in testing participants. We also thank
three anonymous reviewers for their constructive feedback on
the manuscript. This research was funded by the Max-PlanckGesellschaft, Germany.
Reprint requests should be sent to Adriana Hanulíková, Basque
Center on Cognition, Brain and Language, Mikeletegi Pasealekua 69,
Solairua 2, 20009 Donostia, Spain, or via e-mail: Adriana.Hanulikova@
mpi.nl.
Notes
1. No editing or cross-splicing of the critical nouns was administered to avoid coarticulation inconsistencies across different
conditions (e.g., the phonetic properties of the onset of the
critical noun are affected by the preceding speech sound, which
differed across conditions because of differing determiners or
adjectival inflections).
2. The differences in duration between the native and the
accented speech were significant for both sentence duration
(t(687) = 15.30, p < .001) as well as word duration (t(687) =
9.24, p < .001).
3. The high error detection rates suggest that the morphological
manipulations were phonologically salient. Note that the correct
and incorrect adjectives preceding half of the critical words differed
in the number of syllables (e.g., groen vs. groene).
4. One advantage of using EEG is that it allows investigating
brain activity as the speech unfolds over time, without an additional task that could interfere with the natural spoken language
processing. It is often argued that the P600 is more likely to
be elicited with a judgment task. However, numerous task-less
studies have elicited ERP effects, some of which are closely related to the present study (e.g., Van Berkum et al., 2008; Hagoort
& Brown, 2000). To keep the task as natural as possible and to
keep the study comparable to previous studies, we asked participants to listen for comprehension only.
5. One might wonder whether a P600 in L2 speech could be
observed with a sentence judgment task. Judging a sentence may
strengthen control processes, possibly resulting in a modulation of
the ERP effects. However, the aim of the present study was to
better understand the natural situation of processing grammatical
errors produced by L1 and L2 speakers rather than to investigate under which task-related conditions a P600 effect might be
elicited.
REFERENCES
Albrechtsen, D., Henriksen, B., & Færch, C. (1980). Native
speaker reactions to learnersʼ spoken interlanguage.
Language Learning, 30, 365–396.
Bar, M. (2007). The proactive brain: Using analogies and
associations to generate predictions. Trends in Cognitive
Sciences, 11, 280–289.
Barber, H., & Carreiras, M. (2005). Grammatical gender
and number agreement in Spanish: An ERP comparison.
Journal of Cognitive Neuroscience, 17, 137–153.
886
Journal of Cognitive Neuroscience
Birdsong, D., & Molis, M. (2001). On the evidence for
maturational constraints in second-language acquisition.
Journal of Memory and Language, 44, 235–249.
Blom, E., Polišenská, D., & Weerman, F. (2006). Effects of age
on the acquisition of agreement inflection. Morphology,
16, 313–336.
Boelte, J., & Connine, C. M. (2004). Grammatical gender
in spoken word recognition in German. Perception
and Psychophysics, 66, 1018–1032.
Booij, G. E. (2002). The morphology of Dutch. Oxford, UK:
Oxford University Press.
Bornkessel-Schlesewsky, I., & Schlesewsky, M. (2009).
Processing syntax and morphology: A neurocognitive
perspective. Oxford, UK: Oxford University Press.
Bradlow, A. R., & Bent, T. (2008). Perceptual adaptation to
non-native speech. Cognition, 106, 707–729.
Cheng, L. R. L. (1999). Moving beyond accent: Social and
cultural realities of living with many tongues. Topics in
Language Disorders, 19, 1–10.
Chun, A. E., Day, R. R., Chenoweth, N. A., & Luppescu, S.
(1982). Errors, interaction, and correction: A study of
native-nonnative conversations. TESOL Quarterly, 16,
537–547.
Clarke, C. M., & Garrett, M. F. (2004). Rapid adaptation to
foreign-accented English. Journal of Acoustical Society
of America, 116, 3647–3658.
Corbett, G. (1991). Gender. Cambridge, MA: Cambridge
University Press.
Cornips, L. (2008). Loosing grammatical gender in Dutch:
The result of bilingual acquisition and/or an act of identity?
International Journal of Bilingualism, 12, 105–124.
Coulson, S., King, J. W., & Kutas, M. (1998). Expect the
unexpected: Event-related brain response to
morphosyntactic violations. Language and Cognitive
Processes, 13, 21–58.
Ensz, K. Y. (1982). French attitudes toward typical speech
errors of American speakers of French. The Modern
Language Journal, 64, 210–215.
Flege, J. E. (1995). Second language speech learning:
Theory, findings and problems. In W. Strange (Ed.),
Speech perception and linguistic experience
(pp. 233–272). Baltimore: York Press.
Franceschina, F. (2005). Fossilized second language
grammars: The acquisition of grammatical gender.
Amsterdam: Benjamins.
Friederici, A. D. (1995). As time goes by: The time-course
of syntactic activation during language processing. Brain
and Cognition, 28, 259–281.
Friederici, A. D., & Jacobsen, T. (1999). Processing grammatical
gender during language comprehension. Journal of
Psycholinguistic Research, 28, 467–484.
Friederici, A. D., & Oberecker, R. (2008). The development
of syntactic brain correlates during the first years of life.
In A. D. Friederici & G. Thierry (Eds.), Early language
development: Bridging brain and behaviour
(pp. 215–231). Amsterdam: John Benjamins.
Galloway, V. B. (1980). Perception of the communicative
efforts of American students of Spanish. The Modern
Language Journal, 64, 428–433.
Guillelmon, D., & Grosjean, F. (2001). The gender marking
effect in spoken word recognition: The case of bilinguals.
Memory and Cognition, 29, 503–511.
Gunter, T. C., Friederici, A. D., & Schriefers, H. (2000).
Syntactic gender and semantic expectancy: ERPs reveal
early autonomy and late interaction. Journal of Cognitive
Neuroscience, 12, 556–568.
Gunter, T. C., Stowe, L. A., & Mulder, G. (1997). When
syntax meets semantics. Psychophysiology, 34, 660–676.
Volume 24, Number 4
Hagoort, P., & Brown, C. M. (1999). Gender electrified: ERP
evidence on the syntactic nature of gender processing.
Journal of Psycholinguistic Research, 28, 715–728.
Hagoort, P., & Brown, C. M. (2000). ERP effects of listening
to speech compared to reading: The P600/SPS to syntactic
violations in spoken sentences and rapid serial visual
presentation. Neuropsychologia, 38, 1531–1549.
Hagoort, P., Brown, C. M., & Groothusen, J. (1993). The
syntactic positive shift (SPS) as an ERP measure of syntactic
processing. Language and Cognitive Processes, 8, 439–483.
Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M.
(2004). Integration of word meaning and world knowledge
in language comprehension. Science, 304, 438–441.
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.
Koester, D., Gunter, T. C., Wagner, S., & Friederici, A. D.
(2004). Morphosyntax, prosody, and linking elements:
The auditory processing of German nominal compounds.
Journal of Cognitive Neuroscience, 16, 1647–1668.
Kutas, M., & Federmeier, K. D. (2007). Event-related brain
potential (ERP) studies of sentence processing. In G. Gaskell
(Ed.), The Oxford handbook of psycholinguistics (pp. 385–406).
Oxford, UK: Oxford University Press.
Kutas, M., & Hillyard, S. A. (1980). Reading senseless
sentences: Brain potentials reflect semantic incongruity.
Science, 207, 203–205.
Munro, M. J., & Derwing, T. M. (1995). Foreign accent,
comprehensibility, and intelligibility in the speech of second
language learners. Language Learning, 45, 73–97.
Münte, T. F., & Heinze, H. J. (1994). ERP negativities during
syntactic processing of written words. In H. J. Heinze, T. F.
Münte, & G. R. Magnun (Eds.), Cognitive electrophysiology.
Boston: Birkhäuser.
Oostdijk, N. (2000). The Spoken Dutch Corpus project.
The ELRA Newsletter, 5, 4–8.
Orgassa, A. (2009). Specific language impairment in a
bilingual context: The acquisition of Dutch inflection
by Turkish-Dutch learners. Utrecht: LOT.
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, 413–437.
Osterhout, L., McKinnon, R., Bersick, M., & Corey, V.
(1996). On the language specificity of the brain response
to syntactic anomalies: Is the syntactic positive shift a
member of the P300 family? Journal of Cognitive
Neuroscience, 8, 507–526.
Rossi, S., Gugler, M. F., Hahne, A., & Friederici, A. D. (2005).
When word category information encounters morphosyntax:
An ERP study. Neuroscience Letters, 384, 228–233.
Sabourin, L. (2006). Transfer effects in learning a second
language grammatical gender system. Second Language
Research, 22, 1–29.
Statistics Netherlands. (2010). http://statline.cbs.nl/StatWeb/
publication/ ?DM=SLEN&PA=37325eng&D1=0&D2=0&
D3=0&D4=0&D5=0-1,84,102,139,145,210,225&D6=a&
LA=EN&HDR=G2,G3,G4,T&STB=G1,G5&VW=T.
Retrieved on June 1, 2010.
Van Berkum, J. J., Brown, C. M., Zwitserlood, P., Kooijman, V.,
& Hagoort, P. (2005). Anticipating upcoming words in
discourse: Evidence from ERPs and reading times. Journal
of Experimental Psychology: Learning, Memory, and
Cognition, 31, 443–467.
Van Berkum, J. J., Van den Brink, D., Tesink, C. M., Kos, M., &
Hagoort, P. (2008). The neural integration of speaker and
message. Journal of Cognitive Neuroscience, 20, 580–591.
Van Heuven, V. J. (1986). Some acoustic characteristics
and perceptual consequences of foreign accent in Dutch
spoken by Turkish immigrant workers. In J. Van Oosten
& J. F. Snapper (Eds.), Dutch Linguistics at Berkeley,
papers presented at the Dutch Linguistics Colloquium
held at the University of California, Berkeley on
November 9th, 1985 (pp. 67–84). U.C. Berkeley, CA:
Berkeleyʼs Dutch Studies Program.
Weerman, F., Bisschop, J., & Punt, L. (2006). L1 and L2
acquisition of Dutch adjectival inflection. ACLC Working
Papers, 1, 5–36.
Hanulíková et al.
887