The Influence of Language Proficiency on Lexical Semantic - IA de Investigación especializada en el MIT

La influencia del dominio del lenguaje en la semántica léxica
Processing in Native and Late Learners of English

Aaron J. Newman1, Antoine Tremblay2*, Emily S. Nichols1,
Helen J. Neville3, and Michael T. Ullman2

Abstracto

■ We investigated the influence of English proficiency on ERPs
elicited by lexical semantic violations in English sentences, en
both native English speakers and native Spanish speakers who
learned English in adulthood. All participants were administered
a standardized test of English proficiency, and data were analyzed
using linear mixed effects (LME) modelado. Relative to native
learners, late learners showed reduced amplitude and delayed
onset of the N400 component associated with reading semantic
violations. As well, after the N400 late learners showed reduced
anterior negative scalp potentials and increased posterior poten-
tials. In both native and late learners, N400 amplitudes to seman-
tically appropriate words were larger for people with lower
English proficiency. N400 amplitudes to semantic violations,

sin embargo, were not influenced by proficiency. Although both
N400 onset latency and the late ERP effects differed between
L1 and L2 learners, neither correlated with proficiency. Different
approaches to dealing with the high degree of correlation be-
tween proficiency and native/late learner group status are dis-
cussed in the context of LME modeling. The results thus
indicate that proficiency can modulate ERP effects in both L1
and L2 learners, and for some measures (en este caso, N400
amplitude), L1–L2 differences may be entirely accounted for by
proficiency. Por otro lado, not all effects of L2 learning can
be attributed to proficiency. Bastante, the differences in N400
onset and the post-N400 violation effects appear to reflect fun-
damental differences in L1–L2 processing. ■

INTRODUCCIÓN

With increasing age of acquisition (AoA) of a second lan-
guage (L2), evidence suggests decreasing ultimate achieve-
ment and increasing variance of proficiency among
individuals (Birdsong & Molis, 2001; Flege, Yeni-Komshian,
& Liu, 1999; Johnson & Newport, 1989). The reasons for
these changes in language learning ability are not well un-
derstood and likely involve a combination of maturational
cambios, differences in social and learning environments,
and the cumulative effects of learning and use of an L1 over
tiempo (Morgan-Short, Sanz, & Ullman, 2010; Weber-Fox &
Neville, 1996; Johnson & Newport, 1989; Lenneberg,
1967). One approach to better understanding the effects
of age on L2 abilities is the use of neuroimaging techniques
such as ERPs and fMRI. Studies have typically compared
highly proficient L1 speakers with L2 speakers of varying
(and generally lower) proficiency. Because L1/L2 status
and proficiency are confounded, it is often unclear whether
any observed L1–L2 differences are attributable to L2s being
processed in qualitatively different ways from L1s or simply
because of differences in the amount of effort required for
language processing. The answer to this question funda-

1Universidad de Dalhousie, 2Georgetown University, 3Universidad de
Oregón
*Antoine Tremblay is now at the Issak Walton Killam Health
Centre.

mentally impacts how we interpret neuroimaging studies
of L2 learners. One approach to this question is to compare
L1 and L2 learners of comparable proficiency, but differing
AoA. Sin embargo, because most L2 learners do not achieve
native levels of proficiency, such an approach is restricted
either to a narrow sample of L2 learners, or compares
higher-than-average proficiency L2 learners with lower-
than-average L1 learners. An additional concern is that most
studies seem to treat L1 learners as having universally
maximal proficiency, when in fact their scores on vocabu-
lary and grammar tests vary (Hammill, Marrón, Larsen, &
Wiederholt, 1994). en este estudio, we addressed these issues
directly by measuring both brain activation using ERPs
and English proficiency in both L1 and L2 learners, treating
proficiency as a continuous variable.

Several studies have previously demonstrated effects of
proficiency in L2 learners. Some areas show increased
activation among more proficient learners. Using fMRI,
Wartenburger et al. (2003) found that high-proficiency late
learners showed greater activation than lower-proficiency
late learners in temporo-parietal regions when performing
a grammatical judgment task in their L2, thus suggesting
that activation in this area is sensitive to proficiency when
AoA is controlled. Newman-Norlund, Frey, Petitto, y
Grafton (2006) found an effect of proficiency in the left
inferior frontal gyrus (LIFG) activation in a longitudinal
study of artificial grammar learning, with activation during

Revista de neurociencia cognitiva 24:5, páginas. 1205–1223

D
oh
w
norte
yo
oh
a
d
mi
d

F
r
oh
metro

/
j

t
t

F
/

i
t
.

:
/
/

h
t
t
pag
:
/
D
/
oh
metro
w
i
norte
t
oh
pag
a
r
d
C
mi
.
d
s
F
i
r
oh
yo
metro
v
mi
h
r
C
pag
h
a
d
i
i
r
r
mi
.
C
C
t
.
oh
metro
metro
/
j
mi
d
oh
tu
C
norte
oh
/
C
a
norte
r
a
t
r
i
t
i
C
C
yo
mi
mi
–
pag
–
d
pag
d
2
F
4
/
5
2
4
1
/
2
5
0
/
5
1
1
2
9
0
4
5
3
/
7
1
7
7
7
7
oh
7
C
6
norte
8
_
8
a
/
_
j
0
oh
0
C
1
norte
4
3
_
a
pag
_
d
0
0
b
1
y
4
gramo
3
tu
.
mi
pag
s
t
d
oh
F
norte
b
0
y
7
S
METRO
mi
I
pag
t
mi
metro
l
i
b
b
mi
r
r
a
2
r
0
2
i
3
mi
s

/
j

tu
s
mi
r

oh
norte

1
7

METRO
a
y

2
0
2
1

sentence processing increasing as mastery of the language
increased. En cambio, Meschyan and Hernandez (2006)
found greater activation of components of the articula-
tory motor system when Spanish–English bilinguals
read words in their less proficient L1 than in their more
proficient L2.

Not all differences in brain activation between L1 and L2
learners can be attributed to proficiency, sin embargo. Para
ejemplo, when late L2 learners were compared with early
L2 learners of comparable high proficiency, greater activa-
tion was found in the LIFG—an effect of later acquisition
where the possible effect of proficiency was controlled for
( Wartenburger et al., 2003). Similarmente, Perani et al. (2003)
found increased LIFG activity for a phonological fluency
task performed in L2 versus L1, in early L2 learners who
were assumed to have relatively high proficiency (a pesar de
proficiency was not explicitly measured). These studies
provide evidence that separable effects of proficiency and
AoA may be detected using neuroimaging.

Several ERP studies have also investigated the relation-
ship between proficiency and brain activation in L2 learn-
ers. These studies generally report more native-like
patterns of scalp activity in more proficient L2 learners.
The ERP components that have been most studied are
the following: the N400, a negativity peaking around
400 msec postword onset, which is larger for semantically
anomalous than congruent words; the P600, a positivity
that is typically larger in response to syntactic violations;
and a LAN around 150–500 msec, which is also sensitive
to syntactic congruity. In studies of both lexical semantic
and syntactic processing, less proficient learners show
delayed onsets and/or peaks of components, reduced
amplitudes, and in some cases qualitatively different or
even absent components (Midgley, Holcomb, & Grainger,
2009; Hahne, Mueller, & Clahsen, 2006; Rossi, Gugler,
Friederici, & Hahne, 2006; Elston-Güttler, Paulmann, &
Kotz, 2005; Moreno & Kutas, 2005; Ojima, Nakata, & Kakigi,
2005; Kotz & Elston-Güttler, 2004; Phillips, Segalowitz,
Brien, & Yamasaki, 2004). Similar results have been found
in studies of artificial or miniature languages where the
language exposure is known and controlled (Morgan-Short
et al., 2010; Mueller, Oberecker, & Friederici, 2009; Mueller,
Hahne, Fujii, & Friederici, 2005; Friederici, Steinhauer, &
Pfeifer, 2002).

Tomados juntos, these data suggest that it is critical to
account for proficiency in neuroimaging studies of L2
learners. One methodological issue, aunque, is that “high”
and “low” proficiency are often defined fairly arbitrarily,
such as a median split of a proficiency measure or on
self-reported amounts of usage. Such an approach makes
it hard to compare one group of “high-proficiency” learners
a otro. Además, the practice of dichotomizing con-
tinuous variables such as proficiency leads to a loss of power
and reduced effect sizes as well as increasing the likelihood
of finding spurious significant effects (cohen, 1983). A
preferable approach is to use a standardized test of profi-
ciency and treat scores as falling along a continuum.

A second question that has not been well addressed in
the literature is how proficiency affects neurocognitive
measures of language processing in L1 learners. L1 profi-
ciency is rarely measured in lieu of the implicit, but erro-
neous, assumption that all L1 speakers perform at ceiling.
It is crucial to determine whether the variation in brain
activation associated with proficiency is similar in L1 or
L2 learners or qualitatively different. If proficiency modu-
lates neural activation in the same way in L1 and L2 learn-
ers, then we gain insight into the associated processes,
but not into the question of why L2 learners generally
have lower proficiency. En cambio, if differences remain
after controlling for the effects of proficiency, then we
can be assured that we are looking at the effects of L1/L2
learner status.

Two studies have investigated the effects of L1 profi-
ciency on ERPs. Pakulak and Neville (2010) found earlier-
latency anterior negativities and higher-amplitude P600
components in higher- than lower-proficiency native speak-
ers of English, in response to syntactic phrase structure vio-
laciones. Weber-Fox, davis, and Cuadrado (2003) comparado
“high” and “normal” proficiency L1 English learnersʼ re-
sponses to felicitous and semantically anomalous words
during sentence processing. Late (400–600 msec) negative
responses to semantically congruous open class (es decir., estafa-
tent) words were reduced over posterior electrodes but
enhanced over anterior ones, in the high proficiency group.
As well, responses to semantically anomalous words were
reduced in high-proficiency learners, and the amplitude of
this effect correlated with the standardized measure of pro-
ficiency used (the TOAL-3; Hammill et al., 1994; also used
by Pakulak & Neville, 2010).

One study of L2 speakers did treat proficiency as a con-
tinuous variable. Moreno and Kutas (2005) found that a
measure of vocabulary knowledge correlated with the tim-
En g, but not the amplitude, of the N400 elicited by se-
mantic violations in both L1 and L2 learners. N400 peak
latencies over a left posterior electrode were earlier in peo-
ple with higher vocabulary scores and in participantsʼ
dominant language (regardless of whether this was their
L1 or L2). N400 peak latency also increased with AoA in this
grupo, but using stepwise linear regression Moreno and
Kutas showed that whichever variable (proficiency or
AoA) was entered first, the second explained additional
variance. These results suggest separable effects of AoA
and proficiency.

The evidence thus indicates that proficiency is an
important factor affecting brain activation in L2 and
even in L1. In L2, effects of proficiency and AoA may
be separable, and so it is important to include both as
predictors. Además, it is important to determine
whether proficiency affects patterns of brain activation
similarly in L1 and L2. By taking this approach, tenemos
the power to determine whether any observed differ-
ences between L1 and L2 learners are simply because
of the groupsʼ falling, on average, at different points
along the continuum of proficiency, or if the differences

1206

Revista de neurociencia cognitiva

Volumen 24, Número 5

D
oh
w
norte
yo
oh
a
d
mi
d

F
r
oh
metro

/
j

t
t

F
/

i
t
.

:
/
/

/
j

tu
s
mi
r

oh
norte

1
7

METRO
a
y

2
0
2
1

can be attributed to differences in how L1 and L2 are
procesado.

meanings into episodic memory (Lau, Phillips, & Poeppel,
2008; Kutas & Federmeier, 2000).

The goals of this study were to (1) characterize effects of
proficiency on brain activation during the processing of se-
mantically congruous and incongruous sentences, in both
L1 and L2, treating proficiency as a continuous measure,
y (2) separate the effects of proficiency on lexical se-
mantic processing from those of L1 versus late L2 learner
estado. Previous studies have taken important first steps
into exploring the relationship between proficiency and
brain activation. en este estudio, we aimed to make several
further advances. For one, we used a standardized test
of proficiency (the TOAL-3) that does not show ceiling
effects in native speakers and tested both L1 and late L2
learners. As well, we used linear mixed effects (LME) modificación-
eling so that proficiency could be treated as a continuous
variable included alongside numerous factorial variables
(p.ej., condición; electrode position). This improves on
previous approaches that have either dichotomized profi-
ciency to incorporate it into an ANOVA framework (en el
cost of statistical power/sensitivity), or used simple linear
regression (at the expense of considering numerous pre-
dictive factors in a single analysis, p.ej., limiting the regres-
sion to a single electrode). Another issue that has not
typically been addressed is the fact that proficiency and
L1/L2 learner status are typically highly collinear. Moreno
and Kutas (2005) took one approach in using stepwise lin-
ear regression to test whether, after one of these variables
had been entered, the addition of the other explained ad-
ditional variance. We built upon this foundation by using
both the stepwise approach, and an alternative in which
the variance in proficiency because of group was first re-
moved and then the residual variance included in the
LME models alongside group, condición, and electrode
factores. In doing so we were further able to assess the
robustness of any effects that were found.

L2 learners in this study were native Spanish speakers
with a mean age of first exposure to English of 10 años,
mean age of first arrival in an English-speaking country
de 24 años, and an average of 8 years living in an English-
speaking country. Most participants reported that, Alabama-
though they had childhood exposure to English, típicamente
through school and/or television, they did not feel they
had achieved any significant level of fluency until moving
to an English-speaking country in adulthood. We mea-
sured vocabulary and grammatical proficiency in both na-
tive and late learners of English, using the TOAL-3. This is a
standardized test battery that assesses both grammar and
vocabulary skills, has norms up to age of 24, does not typ-
ically show ceiling effects even among English L1 speakers,
and has been used in previous ERP studies of language
(Pakulak & Neville, 2010; Weber-Fox et al., 2003). Nosotros
used a paradigm involving lexical semantic violations
(p.ej., The Irishman sipped Toddʼs thunder at the party),
which typically elicit an N400 relative to well-formed con-
trol sentences. The N400 is thought to reflect aspects of
lexical access and the postlexical integration of word

We predicted that learner status (native or late) and pro-
ficiency would have separable effects on ERPs. Específicamente,
following previous studies, we predicted that the latency
and amplitude of the N400 elicited by semantic anomalies
would be later and smaller, respectivamente, in late learners.
Sin embargo, we further predicted, on the basis of Moreno
and Kutasʼ (2005) datos, that the differences in latency
and possibly in amplitude would largely be accounted
for by proficiency, controlling for L1/L2 learner status.
De este modo, we predicted that a similar relationship between
proficiency and N400 latency and amplitude would be
found for L1 and L2 learners of English. To the extent that
this prediction did not hold, and group differences were
observed once proficiency had been accounted for in
el modelo, we would attribute such group differences as
likely stemming from an effect of late L2 acquisition inde-
pendent of proficiency. An additional possibility was that
while proficiency might modulate one or more properties
of the N400 in both groups, it might do so in different ways
between groups. Such an interaction could be interpreted
as evidence that, although proficiency modulates lexical
Procesando, it does so in different ways, depending on
whether the language is acquired from birth or in early
adulthood as an L2.

MÉTODOS

Participantes

Nineteen native English speakers (edad media = 23.3 años,
DE = 7.1 años, range = 18–51 years; mean years of
education = 14.4 años, DE = 1.7 años) y 19 native
Spanish speakers (edad media = 34 años, DE = 6.6 años,
range = 21–46 years; mean years of education = 17.9 años,
DE = 3.0 años) took part in this study. The L2 learners were
thus older, R = 0.33, F(1, 33) = 17.5, pag = .0002, and had
more years of education, R = 0.31, F(1, 33) = 16.6, pag =
.0003. These differences in age and education level were
taken into account in the analyses as described below.
All participants were men, right-handed, and without any
reported neurological or psychiatric pathology. Native
Spanish speakersʼ mean age of first exposure to English
era 9.6 años (DE = 9.6 años; de 0 a 24 años). Primero
exposure was typically in school, taught by a nonnative
Inglés hablante, with 1–6 hr/week of formal instruction.
The mean age at which native Spanish speakers moved
to an English-speaking country where they were immersed
in English was 24.2 años (DE = 6.9 años, range = 18–
40 años) and the average length of time they had lived in
an English-speaking country was 8.1 años (DE = 5.2 años,
range = 1–19 years). All but 3 of the native English speak-
ers reported some knowledge of at least one other lan-
guage. A los participantes se les pagó $20 por su participación.
Study procedures were reviewed by the Georgetown
University Institutional Review Board.

Newman et al.

1207

D
oh
w
norte
yo
oh
a
d
mi
d

F
r
oh
metro

/
j

F
/

t
t

i
t
.

:
/
/

/
j

tu
s
mi
r

oh
norte

1
7

METRO
a
y

2
0
2
1

Materials

Procedimiento

The target stimuli for this experiment consisted 64 simple
declarative English sentences. Two versions of each sen-
tence were created, one that was semantically acceptable
(p.ej., The Irishman sipped Toddʼs whiskey at the party)
and the other in which the direct object of the verb was
replaced with a noun, matched in lexical frequency, eso
did not make contextual sense (p.ej., The Irishman sipped
Toddʼs thunder at the party). Stimuli were counter-
balanced across participants, such that each participant
saw only the control or the anomalous version of a given
oración. An additional 192 sentences were used, incluir-
En g 32 sentences each with violations of regular past tense
morfología, irregular past tense morphology, and syntac-
tic phrase structure. The remaining sentences were gram-
matically and semantically acceptable. The complete set
of stimuli are available in Newman, Ullman, Pancheva,
Waligura, and Neville (2007). Because the present article
is focused on the relationship between language profi-
ciency and lexical semantic processing, the results of the
grammatical violations will not be discussed here.

Participants were administered a general health screen-
ing and a language history questionnaire that included
self-ratings of proficiency in each language known (en un
5-point Likert scale). The following subtests from the Test
of Adult and Adolescent Language, third edition (Hammill
et al., 1994), were also administered: Reading and Listen-
ing Vocabulary, and Listening, Reading, and Speaking
Grammar.

After giving informed consent, los participantes completaron el
questionnaires and were administered the TOAL-3 sub-
pruebas. The EEG cap was then applied, and impedances
were lowered to <5 kΩ. Participants were then seated in a dimly lit, sound-attenuating booth 135 cm from a CRT monitor; stimulus words subtended 0.5° vertically and 1°–3° deg horizontally. given response button box to hold in both hands. Each sentence was ini- tiated by button press the participant and began with outline of (7° × 3° visual angle) appearing on computer monitor for random period 300– 1100 msec. The pres- ented one at time, with each word displayed 300 msec and 200-msec delay between words. the box remained 1500 msec after last the sentence replaced response prompt, “Good or bad?”, on screen. This remained visible until responded (response buttons were counterbalanced across participants the four sets stimuli), which point fixation cross dis- played pressed initiate next trial. Participants short breaks every 50 sen- tences could any other time. Order of presentation randomized partic- ipant. Before experimental stimuli being presented, participants performed practice session consisting of 16 sentences, receiving feedback their performance. Feedback not provided stimuli. ERP Recording Preprocessing Statistical Analyses Continuous EEG data recorded participant via 64 tin electrodes sewn into tight-fitting cap (Electro- Cap, Eaton, OH), referenced on-line an electrode on the right mastoid bone (later rereferenced average of left locations). Electrode positions were specified International 10–20 system (FP1>
The Influence of Language Proficiency on Lexical Semantic image

Descargar PDF