You are on page 1of 12

Brain & Language 137 (2014) 50–61

Contents lists available at ScienceDirect

Brain & Language


journal homepage: www.elsevier.com/locate/b&l

Bilingualism trains specific brain circuits involved in flexible rule


selection and application
Andrea Stocco ⇑, Chantel S. Prat
Institute for Learning and Brain Sciences, University of Washington, United States
Department of Psychology, University of Washington, United States

a r t i c l e i n f o a b s t r a c t

Article history: Bilingual individuals have been shown to outperform monolinguals on a variety of tasks that measure
Accepted 18 July 2014 non-linguistic executive functioning, suggesting that some facets of the bilingual experience give rise
Available online 24 August 2014 to generalized improvements in cognitive performance. The current study investigated the hypothesis
that such advantage in executive functioning arises from the need to flexibly select and apply rules when
Keywords: speaking multiple languages. Such flexible behavior may strengthen the functioning of the fronto-striatal
Bilingualism loops that direct signals to the prefrontal cortex. To test this hypothesis, we compared behavioral and
Executive function
brain data from proficient bilinguals and monolinguals who performed a Rapid Instructed Task Learning
Cognitive control
Basal ganglia
paradigm, which requires behaving according to ever-changing rules. Consistent with our hypothesis, bil-
fMRI inguals were faster than monolinguals when executing novel rules, and this improvement was associated
Rule-based behaviors with greater modulation of activity in the basal ganglia. The implications of these findings for language
Cognitive flexibility and executive function research are discussed herein.
Rapid Instructed Task Learning Published by Elsevier Inc.

1. Introduction sorting paradigms), the biological nature of the bilingual advantage


still remains largely unexplained.
One of the most puzzling findings in recent cognitive neurosci- In this paper, we test a proposed neurocognitive mechanism
ence research is that individuals who develop bilingually outper- underlying the bilingual advantage in executive functioning,
form monolinguals on tests that measure non-linguistic namely, that the advantage stems from the strengthening of a gen-
executive functions. After decades of research failing to find gener- eral circuit for routing signals to the prefrontal cortex (PFC)
alized ‘‘training’’ effects (Melby-Lervåg & Hulme, 2012; Shipstead, through the basal ganglia (Stocco et al., 2014). Specifically, because
Redick, & Engle, 2012), the fact that children born into bilingual this circuit is known to be involved in rule-based behaviors
families exhibit superior performance on non-linguistic tasks sug- (Muhammad, Wallis, & Miller, 2006; Pasupathy & Miller, 2005),
gests that some aspect of the experience of managing multiple lan- we test the hypothesis that the need for flexible linguistic rule
guages must ‘‘train the brain’’ in a way that gives rise to application in bilingual language use gives rise to greater flexibility
generalized improvements in cognitive functioning (Stocco, in adapting behaviors according to novel or changing rules. Using a
Yamasaki, Natalenko, & Prat, 2014). It is particularly interesting neuroimaging instantiation of a Rapid Instructed Task Learning
that these improvements are found in executive functions, which (RITL) paradigm (Cole, Laurent, & Stocco, 2013), we compared the
in many ways represent the apex of human cognitive abilities. behavioral and neural responses of monolinguals and proficient
Although this finding has been demonstrated throughout the life- bilinguals while executing novel and practiced arithmetic opera-
span (e.g., Bialystok, Craik, Klein, & Viswanathan, 2004; Carlson & tions. Below we summarize the research on the bilingual advan-
Meltzoff, 2008), using paradigms that test different facets of exec- tage in executive functioning that provided the rationale for this
utive functioning (e.g., susceptibility to interference using the paradigm choice, and a more complete description of the theory
Simon or Stroop tasks; flexibility using task switching or card that motivated the research herein.

⇑ Corresponding author. Address: Department of Psychology and Institute for 1.1. The bilingual advantage in executive functioning
Learning and Brain Sciences, University of Washington, Campus Box 357988,
Seattle, WA 98915, United States. The term ‘‘executive functions’’ refers to the set of higher-level
E-mail address: stocco@uw.edu (A. Stocco). cognitive processes that oversee and coordinate the control of

http://dx.doi.org/10.1016/j.bandl.2014.07.005
0093-934X/Published by Elsevier Inc.
A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61 51

lower-level behavior (Elliott, 2003; Miyake et al., 2000). In other one to bridge the link between the bilingual advantage as mea-
words, executive functions are the mechanisms that enable human sured in the laboratory and its implications for real-world process-
cognition to move away from automatic responses, towards more ing advantages in bilingual individuals.
complex, goal-driven behaviors. Thus, we argue that our understanding of the bilingual advan-
Experimental psychologists have largely tried to characterize tage can be advanced by carefully considering how bilingual par-
executive functions by adopting a ‘‘divide and conquer’’ approach. ticipants are defined and selected, and by choosing an
For instance, in an influential study, Miyake et al. (2000) used experimental paradigm that captures the robustness of flexible
latent variable analysis to identify distinct mechanisms that under- information processing in bilingual individuals. In the current
lie performance on a number of executive function paradigms, study, the critical feature considered for defining bilingualism
such as Stroop, N-back, and task switching paradigms. The authors was L2 proficiency, with only participants who were highly profi-
were able to break down the monolithic notion of ‘‘executive func- cient in reading, speaking, and understanding both of their lan-
tions’’ into three loosely independent faculties, namely: inhibition, guages being considered as part of the bilingual group. We chose
updating, and set shifting. to focus on proficiency as a defining characteristic of bilingualism
Research on the nature of the bilingual advantage in executive because several neuroimaging investigations have shown that pro-
functions has largely mimicked this ‘‘divide and conquer’’ ficiency is a more powerful determinant of patterns of neural acti-
approach, attempting to identify the specific components of exec- vation observed in bilinguals than is age of L2 acquisition (e.g.,
utive functioning that are strengthened through bilingual language Briellmann et al., 2004; Chee, Hon, Lee, & Soon, 2001; Perani
use. Unfortunately, results from these experiments have not con- et al., 1998). Additionally, in the population of bilingual partici-
verged on a straightforward answer. For instance, a number of pants tested at the University of Washington, great variability
experiments carried out with the Simon task (a task that requires can be observed in the frequency and proficiency with which early
inhibition of interference) seem to suggest that bilinguals have bet- bilinguals use their non-English language. Thus, in the current
ter inhibitory processes than do monolinguals (Bialystok, 1999; study, we chose to use degree of proficiency as a defining charac-
Bialystok et al., 2004). However, Colzato et al. (2008) found little teristic for inclusion in the bilingual group.
evidence for a bilingual advantage in tasks that target different fac- Furthermore, rather than adopting a divide-and-conquer
ets of inhibitory control, such as the Stop-Signal or the Inhibition of approach, we employed a complex task that involves many
Return tasks. Similarly, there is evidence that bilinguals have better components of executive functioning, and more closely emulates
capacity to shift between mental sets than do monolinguals, as real-world scenarios. Specifically, one of the goals of executive
indicated by their reduced switch cost in task-switching paradigms functions is to flexibly adapt behavior to accomplish a task that
(Garbin et al., 2010; Prior & MacWhinney, 2009). However, these is non-automatic. In fact, some researchers have recently argued
results do not generalize to other set-shifting paradigms, such as that the capacity to change and modify behavior according to
the Dimensional Change Card Sort Task (Bialystok, 1999; internal plans and rules might be the core property of executive
Bialystok & Martin, 2004). Finally, some authors have failed to rep- functioning, which can be studied using RITL paradigms designed
licate the bilingual advantage altogether (Paap & Greenberg, 2013). to measure rearrangements of behavior in response to rules (Cole
There may be two reasons behind the heterogeneity of these et al., 2013). In this experiment, we used a RITL paradigm to test
findings. One possibility is that the bilingual groups recruited in the hypothesis that bilingualism trains the general capacity to
these studies were not comparable with one another. This is behave adaptively, with language serving as an ever-present condi-
because bilingualism is not a dichotomous, or nominal category, tion that provides a dynamic ‘‘context’’ for selecting appropriate
but a multi-dimensional variable in which any of a number of fac- behaviors (e.g., Buchweitz & Prat, 2013; Green & Abutalebi, 2013;
tors may drive the bilingual advantage in executive functioning. Stocco et al., 2014).
For instance, bilingual individuals differ from one another in terms
of relative language proficiencies, age and method of L2 acquisi- 1.2. Rapid Instructed Task Learning (RITL)
tion, and patterns of language use (e.g., speaking two languages
every day vs. speaking one language most of the time). All of the RITL paradigms differ significantly from ‘‘traditional’’ experi-
above factors have been shown to modulate the effects of bilin- ments because of the nature of the question they were designed
gualism. For instance, Luk, De Sa, and Bialystok (2011) found that to investigate. In traditional research paradigms, participants are
the age of acquisition modulates the performance of bilinguals in typically initially instructed on how to perform an experimental
the Flanker task, with early bilinguals showing the least interfer- task, and subsequently given a series of homogeneous experimen-
ence to incongruent stimuli. Studies conducted in different loca- tal stimuli to perform the task upon. In RITL paradigms, on the
tions typically have access to populations that vary significantly other hand, participants are given new instructions at the begin-
along these dimensions, thus making some findings harder to ning of each trial; thus each trial in a RITL paradigm represents a
reproduce. ‘‘novel task’’.
Another possibility is that the ‘‘divide and conquer’’ method Different novel tasks are typically generated by combining a set
may not be the approach best suited for understanding the mech- of basic operations in unique ways, so that each resulting task is
anism behind the bilingual advantage in executive functions. First, comparable in terms of complexity and structure to all the others.
bilinguals do not consistently outperform monolinguals across dif- For instance, in the experiment by Hartstra, Kühn, Verguts, and
ferent tasks that rely on the same sub-component process (e.g., Brass (2011), each task consisted of responding with finger presses
inhibition) of executive functioning. Additionally, although inhibi- to specific classes of visual stimuli, and different tasks were gener-
tory processes have received a considerable amount of attention in ated by varying the stimuli and the fingers used to respond. Thus,
the literature, the bilingual advantage has been demonstrated in one task could be ‘‘respond with your index finger to cars and with
tasks that measure updating (Carlson & Meltzoff, 2008) and set your middle finger to animals’’, and a different task could be
shifting (Prior & MacWhinney, 2009) as well. Furthermore, in nat- ‘‘respond with your middle finger to plants and with your index
ure, many of the component processes of executive functioning finger to houses’’. Similarly, Stocco, Lebiere, O’Reilly, and
happen in tandem. For instance, when you switch to a new task, Anderson (2012) generated different tasks by combining basic
the previously appropriate stimulus–response rules must be inhib- arithmetic operations (such as add, subtract, and multiply) into
ited before the new ones can be executed (Mayr & Keele, 2000). sets of three. For instance, one task could be ‘‘add one to x, divide
Finally, using a divide-and-conquer method does not easily allow y by two, and sum the results’’ and a subsequent task could be
52 A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61

‘‘multiply x by two, subtract one from y, and multiply the results’’. higher level linguistic goals and rules, as evidenced by its activity in
Thus, RITL paradigms stress the human capacity to generate new paradigms that involve language switching (Hernandez, 2009;
behaviors by combining simpler operations according to given Hernandez, Dapretto, Mazziotta, & Bookheimer, 2001). This role is
rules. reminiscent of the corresponding role of the PFC in holding internal
Because they stress the top-down, goal-directed control of representations of the task in RITL. As for the basal ganglia, both
behavior, RITL paradigms provide a new window into the mecha- neuropsychological and experimental research suggest that they
nisms that underlie executive functions. In addition, by separating are involved in controlling which target language to use. For
the encoding of instructions from their execution, RITL paradigms instance, bilingual individuals with basal ganglia lesions show a
offer a convenient means for isolating how task-specific rules and pathological tendency to switch back and forth between languages
goals are encoded from how they are actually used to guide behav- (Abutalebi, Miozzo, & Cappa, 2000; Fabbro, 2001), and it has been
ior during rule execution. show that direct electrical stimulation in the left basal ganglia
impairs the ability to voluntarily switch between languages
1.3. Relating bilingualism to flexible rule application as measured in (Wang, Wang, Jiang, Wang, & Wu, 2012).
RITL paradigms One important neuroimaging study conducted by Crinion et al.
(2006) examined the nature of multiple language representations
In addition to providing a new strategy for investigating the by using a cross-language priming paradigm in bilinguals. In the
nature of executive functions in more complex forms, RITL para- experiment, individuals responded to words that were preceded
digms also share important features with the bilingual language by either semantically related or unrelated primes. The prime word
experience at both the behavioral and neural levels. An under- was presented in either the same language as the target word, or in
standing of these similarities is essential for describing the mech- a different language. Semantic priming crosses the language
anism by which skill transfer from bilingual language management boundary, so that seeing the prime word ‘‘Salmon’’ in English
to executive functions can occur. results in a decreased reaction time to respond to the target word
At the behavioral level, the execution of tasks that involve novel ‘‘Forelle,’’ (the German word for ‘‘Trout’’). At a neural level, this
combinations of simpler rules, as implemented in RITL paradigms, priming effect manifested itself in reduced activation for a seman-
mirrors the bilinguals’ need to quickly select and combine learned tically related target across most of the brain regions that were
linguistic rules according to the language being used. The two lan- recruited by the task (such as prefrontal cortex and the anterior
guages that a bilingual speaks serve the same communication temporal lobes), independent of whether the prime and the target
goals, but each consists of unique mechanisms for expressing the were presented in the same or different languages. The only excep-
message content. Because each language comes with its own set tion to this rule was in the left caudate nucleus (part of the basal
of pragmatic, syntactic, and phonetic rules, speaking two languages ganglia circuit), where the priming effect (i.e., reduction of activa-
requires extensive training in quick and flexible rule selection and tion for the semantically related target) was present only when
application. For example, to form the plural form of the English prime and target were in the same language. In other words, the
word ‘‘cat’’, an Italian/English bilingual needs to select the specific left part of the basal ganglia seemed to be the only region of the
pluralization rule (e.g., ‘‘add s’’ in English vs. ‘‘change the final semantic circuit where lexical representations were separated
vowel into i’’ in Italian), select the target ‘‘variable’’ of interest according to ‘‘target language’’. This is consistent with the hypoth-
(‘‘cat’’ in English vs. ‘‘gatto’’ in Italian), and apply the rule to the tar- esis that the basal ganglia are involved in controlling the selection
get variable, keeping the higher-order goal of communicating in of the proper language-specific variables during the application of
the target language (in this case, English) consistent. During natu- linguistic rules, with target language serving as an ever-present
ral language use, this isolated morphological example would occur constraint in the bilingual mind.
in combination with unique grammatical, pragmatic, and cultural
rules associated with each language, such as the plausibility that 1.4. Experimental predictions
the word ‘‘cats’’ is used in a metaphoric context (as in the English
expression ‘‘herding cats’’). In general, the need to flexibly select, Based on the overlap between bilingual language control and
combine, and apply rules consistent with the target language is the neural correlates of RITL paradigms, we made the following
similar to the demands that RITL tasks impose, where simpler predictions: First, we predicted that bilingual individuals should
operations need to be selected, combined, and applied according perform better than monolinguals when executing novels tasks
to the given instructions. in RITL paradigms, as evidenced by faster response times to novel
The similarity between bilingual language use and RITL para- trials. The reason behind this prediction is that the bilingual expe-
digms is also apparent at the neural level, where they recruit over- rience stresses the need to quickly combine diverse linguistic rules
lapping brain networks. The network of regions underpinning RITL in a flexible way, therefore training the very same skills that also
has been well characterized (Cole, Bagic, Kass, & Schneider, 2010; underlie RITL paradigms.
Cole et al., 2013; Stocco, Lebiere, O’Reilly, & Anderson, 2012). Second, at a neural level, we expect that the improved perfor-
Within this network, two regions are particularly relevant. The first mance of bilinguals in RITL paradigms should be reflected in differ-
is the anterior part of the prefrontal cortex (PFC), which seems to ent patterns of activity in those brain regions that are commonly
be important for forming and maintaining an internal representa- recruited by the bilingual experience and the RITL tasks, which
tion of rules (Wallis, Anderson, & Miller, 2001), relationships are the PFC and the basal ganglia. Specifically, the research out-
(Wendelken, Nakhabenko, Donohue, Carter, & Bunge, 2008), and lined by Prat and colleagues (Prat & Just, 2011; Prat, Keller, &
goals (Miller & Cohen, 2001). The second key circuit is the basal Just, 2007; Prat, Mason, & Just, 2012), suggests that improved per-
ganglia, a set of subcortical nuclei necessary for learning and formance may be manifest by group differences in either neural
applying rules (Muhammad et al., 2006), and controlling the selec- efficiency (i.e., reduced amounts of brain activation to perform
tion of the appropriate operations and target variables for each the same task), or in neural adaptability (i.e., greater modulation
operation (Stocco & Anderson, 2008; Stocco, Lebiere, O’Reilly, & of neural activity in the face of changing task demands). In partic-
Anderson, 2012). ular, neural efficiency is measured by comparing the brain activa-
Interestingly, there is significant evidence that both the basal tion of two groups during the execution of an identical tasks; the
ganglia and the PFC play important and similar roles in bilingual better group should show decreased activation, indicating the
language use. For instance, PFC seems to be involved in representing capacity to achieve the same task with less neural resources
A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61 53

(Prat, 2011; Prat & Just, 2011; Prat et al., 2007). Neural adaptability, classes offered after the age of 13, when many children begin foreign
on the other hand, is measured by comparing the difference language classes in school. Finally, all of the participants were
between an easy and a difficult task across the two groups; with healthy, right-handed adults, with no previous history of neurologi-
higher-performance being associated with greater differences in cal disease at the time of the experiment.
activation between the two conditions, indicating a greater capac-
ity to recruit neural resources to cope with the increased difficulty 2.2. Materials
(Prat & Just, 2011; Prat et al., 2007).
When applied to our experiment, this framework predicts that, The paradigm employed was a modified version of the RITL par-
if bilinguals are better than monolinguals in RITL tasks, they should adigm originally used by Stocco, Lebiere, O’Reilly, and Anderson
also show a smaller amount of active voxels, or less activity within (2012) and described above. In this paradigm, participants receive
the same voxels (the signature of neural efficiency), or alternatively different instructions at each trial on how to mentally manipulate
greater changes in either neural activity or number of active voxels two numbers. Because the instructions change, participants are
(the signature of neural adaptability) in the PFC or in the basal effectively performing a new task at every trial. Tasks of compara-
ganglia. ble difficulty were generated by systematically permuting ten basic
arithmetic operations [add, subtract, multiply, divide, half, double,
1.5. Summary third, triple, increment (i.e., +1), and decrement (i.e., 1)] into sets
of three, which were then applied on two input numbers, denoted
Decades of research suggest that bilinguals outperform monol- as x and y. Thus, each arithmetic operation is a learned rule; the
inguals on tasks that measure executive functions, but the cogni- instructions for each trial constitute a novel combination of rules;
tive and neural nature of this advantage remain unclear. We and the two numbers represent the rule variables. All of the instruc-
hypothesized that the bilingual advantage originates from the tions contained one binary (e. g. ‘‘multiply’’), and two unary (e.g.,
increased pressure to flexibly select and combine linguistic rules ‘‘double’’ or ‘‘increment’’) operations. The three operations were
according to different target languages. Previous research has dem- applied in a fixed order, whereby the first operation was applied
onstrated that the basal ganglia circuit is important for selecting to x, the second operation to y, and the third operation to the previ-
appropriate rules and applying them to specific variables. There- ous two results. The third (and last) operation was always a binary
fore, we hypothesized that this enhanced capacity depends, at one. For instance, the instructions ‘‘HALF/DOUBLE/ADD’’ were
the neural level, on an increased capacity for controlling the trans- interpreted as specifying the task ‘‘divide x by 2, multiply y by 2,
mission of signals to the prefrontal cortex through the basal gan- and then add the two results to each other’’, or ‘‘x/2 + 2*y’’. Note
glia. Based on these assumptions, we predicted that bilingual that, although the instructions were presented using words, and
individuals would be faster than monolinguals when executing the performance of mental arithmetic does require the use of inter-
novel, instructed tasks, and that this ability would be associated nal verbal codes (Dehaene & Cohen, 1995, 1997; Venkatraman,
with increased efficiency or increased adaptability in the basal Siong, Chee, & Ansari, 2006), the task itself stresses a non-linguistic
ganglia and the PFC. ability, i.e. to dynamically re-combine and coordinate mental activ-
ity according to novel templates (Cole et al., 2013).
As in other RITL paradigms (Cole et al., 2010; Stocco, Lebiere,
2. Methods
O’Reilly, & Anderson, 2012), each trial consisted of three consecu-
tive phases: (a) An Encoding phase, where the instructions were
2.1. Participants
presented; (b) An Execution phase, where the two input numbers
x and y (variables) were unveiled and participants had to mentally
Seventeen bilingual individuals (9 female, mean age
execute the task; and (c) A Response phase, where participants
21.1 ± 4.7 years) and 14 monolingual participants (7 female, mean
indicated whether a given number was the result of executing
age 22.4 ± 4.2 years) were recruited from the local student popula-
the instructions. Both Encoding and Execution were self-paced
tion at the University of Washington in exchange for monetary
and preceded by a 1-s fixation. Participants were instructed to
compensation. The demographic and linguistic characteristics of
press a key as soon as they had encoded the instructions or calcu-
the two groups are detailed in Table 1. The linguistic profile of
lated the result. The three phases were separated by blank screens,
bilingual subjects was characterized using the language experience
the durations of which were randomly jittered, and varied between
and proficiency questionnaire (LEAP-Q), a structured questionnaire
0 and 8 s according to an exponential distribution (Dale, 1999). The
designed to collect measures of second language usage and acqui-
fact that the intermediate blank screen could potentially be very
sition (Marian, Blumenfeld, & Kaushanskaya, 2007). Only bilingual
short and that only 1.5 s was given to respond to the probe dis-
participants who rated themselves at least 7 on a 1–10 scale in
couraged participants from extending either the encoding of the
speaking, understanding, and reading in their L2 were selected
instructions or the mental calculations into the delay period. The
for the experiment (mean proficiency score: 8.5 ± 1.3). The great
time course of a sample trial is illustrated in Fig. 1A.
majority of our bilingual subjects had acquired their L2 in child-
hood (mean age of acquisition: 3.1 ± 3.9 years), with only one
2.3. Procedure
highly proficient bilingual acquiring a second language later in life
(see Table 1).1 At the time of testing, the English proficiency of the
Data were collected individually for each participant over the
bilingual participants was comparable to that of the English monol-
course of three sessions, occurring on three different days. During
inguals (as indicated by self-report and confirmed by their scores on
the first session, a number of psychometric measures were taken to
the Nelson–Denny test: Brown, Fishco, & Hanna, 1993; See Table 1).
ensure that the bilingual and monolingual groups did not differ on
All of the monolinguals were proficient in English only, and had not
some important non-linguistic indices of cognitive capacity. In par-
been exposed to any other languages outside of foreign language
ticular, we collected measures of English proficiency (the Reading
Comprehension and Vocabulary portions of the Nelson–Denny
1
To make sure that the heterogeneity of our bilinguals in terms of age of test: Brown et al., 1993), working memory capacity (the
acquisition did not affect our results, we conducted a number of brain-wise
correlation analyses analysis to see if any of the contrast of interests was significantly
Automated Operation Span test: Unsworth, Heitz, Schrock, &
affected by the age of acquisition. As it turns out, no significant voxels were found at a Engle, 2005), and language experience and proficiency (LEAP-Q:
threshold of p < 0.001, uncorrected. Marian et al., 2007). See Table 1 for further details.
54 A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61

Table 1
Composition and psychometric scores of the bilingual and monolingual groups.

Bilinguals Monolinguals Comparison


Group composition Size and gender N = 17 (9 female) N = 14 (7 female) v = 0, n.s.
Mean age ± SD 21.5 ± 4.9 years 22.4 ± 4.3 years t(29) = 0.45, n.s.
Linguistic background L1 Cantonese (N = 1) English (N = 14)
Danish (N = 1)
English (N = 5)
Friulian (N = 1)
Japanese (N = 1)
Korean (N = 2)
Mandarin (N = 3)
Portuguese (N = 1)
Russian (N = 1)
Spanish (N = 1)
L2 English (N = 12) N/A
Japanese (N = 1)
Korean (N = 2)
Mandarin (N = 1)
Spanish (N = 1)
Age of acquisition for L2 Age 0 (N = 1) N/A
Age 2 (N = 6)
Age 3 (N = 2)
Age 4 (N = 3)
Age 5 (N = 1)
Age 6 (N = 3)
Age 18 (N = 1)

English proficiency (Nelson–Denny reading test) Reading comprehension (Mean Percentile ± SD) 57.1 ± 23.2 61.3 ± 21.6 t(28) = 0.51, n.s.
Vocabulary (Mean Percentile ± SD) 55.7 ± 27.1 62.3 ± 21.9 t(28) = 0.76, n.s.

Working memory Operation span (Mean Score ± SD) 50.5 ± 16.4 52.2 ± 14.3 t(28) = 0.30, n.s.

A total of 93 participants were screened in this session. From new variables. The remaining half of the trials consisted of novel
this pool, we selected a group of 18 bilinguals and 18 monolinguals instructions (i.e., previously unseen combinations of operations).
that were matched as a group in terms of English proficiency and All of the fMRI scanning sessions took place within 24 h of the
working memory capabilities.2 These 36 participants were invited practice session.
to return for the second and third sessions. Four monolinguals did
not return to complete the fMRI session; data from one bilingual
individual with the lowest working memory score was consequently 2.4. Image acquisition and analysis
expunged to re-balance the groups, thus yielding the final number of
17 bilinguals and 14 monolinguals. Neuroimaging data were collected from each participant using a
During the second session, participants practiced several trials 3.0T Philips Achieva scanner at the Bio-Molecular Imaging Center
with only two instructions, i.e., two different combinations of the (BMIC) at the University of Washington. Functional images were
basic arithmetic operations. This practice session consisted of 4 collected using a gradient echo planar pulse sequence with a
blocks of 10 trials each. Notice that, although each set of task 2000 ms TR and a 21 ms TE. Each volume acquisition consisted of
instructions was practiced for 20 trials, it was applied on a different 44, 3.2 mm thick oblique axial slices with no gap in-between; each
pair of input numbers at every trial, thus preventing participants slice contained 80  80 voxels with an in-plane resolution of
from forming simple instruction-response associations. Therefore, 2.75  2.75-mm. To improve the normalization process, an
even if the specific combination of rules (i.e., the instructions) to MP-RAGE structural image was also acquired for each participant,
be used in a given trial had been practiced before, participants still consisting of 176 1-mm think sagittal slices containing 256  256
had to perform the three component operations to find the correct voxels with an in-plane resolution of 1  1 mm.
result. This practice session served the dual purpose of familiarizing The data were analyzed using SPM8 (Wellcome Department of
participants with the nature of RITL paradigms, and creating a pool Imaging Neuroscience, www.fil.ion.ucl.ac.uk/spm). Images were
of instructions that, because they had been practiced in advance, are corrected for differences in slice acquisition time, spatially rea-
easier to perform and can be used to assess the hypothesis of neural ligned to the first image in the series, normalized to the Montreal
adaptability (by comparing them against the more difficult cases in Neurological Institute (MNI) ICBM 152 template, resampled to
which the instructions to be performed are completely new). 2  2  2 mm voxels, and finally smoothed with an 8  8  8-
During the third session, participants performed the RITL exper- mm Full-Width Half-Maximum Gaussian kernel to decrease spatial
iment in a fMRI scanner. Like the practice session, the experiment noise and accommodate for individual differences in anatomy. Sta-
consisted of 4 blocks of 10 trials each. Half of the trials consisted of tistical analyses were performed on individual and group data
the previously practiced instructions, which were again applied to using the general linear model as implemented in SPM8 (Friston,
Ashburner, Kiebel, Nichols, & Penny, 2011). For individual partici-
pants, fixed-effects models that incorporated a high-pass filter
2
Note that making the groups comparable in terms of working memory span with a cut-off of 128 s and an auto-regressive AR(1) correction
works against our experimental hypothesis, because measures of working memory for serial autocorrelation were used to estimate parameters. The
and executive functions are correlated. Thus, if bilingualism trains executive
functions, it may also result in improved working memory performance. In fact, in
individual models included one regressor for the response and
the larger sample of participants that were screened in our lab, bilingual individuals one regressor for each of the four experimental conditions
did score higher than monolinguals in the Automated Operation Span task. (Novel Instructions, Practiced Instructions, Execution of Novel
A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61 55

Fig. 1. (A) A sample trial of the RITL paradigm used in our experiment. During each trial, participants were first shown the instructions describing the mathematical task to
perform, such as ‘‘DOUBLE/HALF/ADD’’, and given up to 8 s to memorize them. After a variable delay, participants were then shown two input numbers, X and Y, on which the
task specified by the instructions need to be applied. Participants performed the required calculation in their head, and indicated that they had reached a result by pressing a
button within 8 s. Finally, a numeric probe, such as ‘‘3 ?’’ was presented on the screen, and participants had to indicate within 1.5 s whether the number was the result of their
mental calculation. (B–D) Average reaction times of bilinguals and monolinguals during the Encoding (B), Execution (C), and Response (D) phases of the RITL paradigm. Bars
represent means ± SEM. Line segments highlight within-group statistically significant differences: *p < 0.01, **p < 0.001.

Instructions, Execution of Practiced Instruction). Behavioral error pallidus, and thalamus) and one encompassing the left anterior
trials were modeled separately and not included in the analysis. PFC. These two volumes were selected because the corresponding
Group analyses were performed on the beta values obtained from brain regions have been commonly associated to language control
the individual subject analysis using a mixed-effects model in in bilinguals (Crinion et al., 2006; Fabbro, 2001; Hernandez, 2009;
which participants were treated as a random factor. Hernandez et al., 2001), RITL paradigms (Cole et al., 2010; Stocco,
Lebiere, O’Reilly, & Anderson, 2012), and rule execution (Wallis
2.5. Predefined volumes of interest et al., 2001). The two VOIs also encompass the majority of cen-
troids reported in a recent review of the overlapping roles of basal
Because we had specific predictions about the location and ganglia and PFC in rule learning, syntactic parsing, and language
direction of the neural differences between bilinguals and monol- switching (Buchweitz & Prat, 2013). The use of two small VOIs
inguals, our primary analysis focused on anatomically predefined maximizes the sensitivity of our analysis (which tests a specific
volumes of interest (VOIs). Specifically, our analysis was restricted anatomical hypothesis) while at the same time allowing for some
to two search volumes in the left hemisphere, one encompassing degrees of freedom in the exact anatomical localization of the
the left basal ganglia circuit (caudate nucleus, putamen, globus regions of interest.
56 A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61

Practice [F(1, 29) = 11.87, p = 0.002], no main effect of Group


[F(1, 29) = 1.05, p = 0.31], and a significant interaction among the
two [F(1, 29) = 5.60, p = 0.02]. This effect was due to the fact that
bilinguals were faster than the monolinguals in the execution of
novel instructions [3483 vs. 3964 ms; see Fig. 1C]. In fact, the dif-
ference between the execution of novel and practiced tasks was
significant for monolinguals [t(13) = 4.41, p < 0.001], but not for
bilinguals [t(16) = 0.91, p = 0.38], who were essentially as fast at
executing novel instructions as practiced ones (3368 vs.
3427 ms). Finally, there was no significant effect of either Group
or Practice in the Response phases [F(1, 29) < 0.8, p > 0.38; Fig. 1D].
In summary, as predicted, bilinguals were faster than monoling-
uals at executing novel tasks, with bilinguals exhibiting no signif-
icant differences in execution times for novel vs. practiced tasks.
This finding is consistent with the hypothesis that the bilingual
advantage in executive functions is related to a better capacity to
dynamically reconfigure behavior.

3.2. Neuroimaging results

Consistent with previous experiments, completion of the RITL


Fig. 2. 3D rendering of the two volumes of interests (VOIs) within the MNI paradigm elicited a distributed network of activation, which cru-
template. Blue = Prefrontal search volume; Red = Basal ganglia search volume. See cially includes both the prefrontal cortex and the basal ganglia. A
Table 2.
3D rendition of the overlapping networks of activation in monol-
inguals and bilinguals is depicted in Fig. 3 and detailed in the Sup-
In the case of the PFC, the VOI consisted of a 20-mm radius plementary Materials, Tables S1–S4. The figure only depicts data
sphere encompassing the prefrontal foci reported in the RITL stud- from novel instructions because they are most representative of
ies by Stocco, Lebiere, O’Reilly, and Anderson (2012) and Cole et al. the types of processes that we are interested in investigating. Note
(2010). The position of the sphere was adjusted so that its surface that Fig. 3 is only provided to give an idea of the distribution of
followed the profile of the dorsolateral part of the prefrontal cor- activation across groups and conditions, and does not depict the
tex, including prefrontal grey matter, but excluded regions outside result of a direct comparison between the groups. The next section
of the brain. In the case of the basal ganglia, the VOI consisted of an reports the results of the comparisons between the two groups.
18-mm radius sphere that enclosed the anterior part of caudate
nucleus, the putamen, the globus pallidus, and part of the thala- 3.3. VOI analysis with predefined volumes of interest
mus. The position of the sphere was adjusted so that its surface fol-
lowed the profile of the left lateral ventricle without including it, We first looked for the neural correlates of the bilinguals’
thus maximizing the inclusion of relevant grey matter. The result- improved performance in the two VOIs that are commonly
ing sphere includes most of the foci reported in a previous meta- recruited in bilingual language control, rule application, and RITL
analysis of striatal involvement in bilingual studies (Stocco et al., (see methods, above). As previously discussed, the two possible
2014), as well as the striatal foci reported in previous RITL studies, signatures of superior performance (such as we expect from biling-
such as the left striatal focus in Ruge and Wolfensteller (2010) and uals) are neural efficiency (reduced neural activity for the same task
the left counterpart of the striatal region in Stocco et al. (2012). The condition), and neural adaptability (larger differences in neural
two VOIs are shown in Fig. 2 and described in Table 2. activity between easier and harder conditions; Prat et al., 2007).
To assess neural efficiency, we examined differences in brain activ-
ity between the two groups during the Execution phase of Novel
3. Results and Practiced tasks. No voxels were found in either VOI that were
significantly less active in bilinguals than in monolinguals.
3.1. Behavioral results To assess neural adaptability, we examined the difference in
brain activity between the Execution of Novel vs. Practiced tasks
Participants were highly accurate overall (mean accuracy 91%). between the two groups in the two predefined volumes. No signif-
Because accuracy is a binomially distributed variable, accuracy icant differences were found in the prefrontal VOI. In the case of
data were analyzed with a mixed-effect logistic regression model, the basal ganglia, however, the analysis identified three clusters
which uncovered no difference between groups or between novel of 36, 24, and 16 voxels whose activity was differentially modu-
and practiced tasks [z < 0.58, p > 0.56]. Data for reaction times were lated in the two groups. Two clusters were located in the left puta-
collected from correct trials only, and analyzed separately for the men, and one in the left dorsal nucleus of the thalamus, which is
Encoding, Execution, and Response phases. In the case of the the target of basal ganglia outputs. These clusters are shown in
Encoding phase, a Group (Bilingual vs. Monolingual) by Practice Fig. 4A, and characterized in Table 3. The set-level corrected prob-
(Novel vs. Practiced) mixed-effects ANOVA model uncovered a sig- ability (Friston, Holmes, Poline, Price, & Frith, 1996) of identifying
nificant main effect of Practice [F(1, 29) = 58.25, p < 0.001], but not three clusters within the given volume of interest was calculated as
of Group [F(1, 29) = 0.12, p = 0.73], and no significant interaction P < 0.005. Thus, bilinguals showed greater modulation of activity to
(Fig. 1B). The main effect of Practice was due to reaction times novel instructions in the basal ganglia region than did monoling-
being predictably slower for encoding novel instructions uals, a finding that is consistent with previous investigations link-
(M = 2695 ± 785 ms) than practiced ones (M = 1982 ± 806 ms). ing individual differences in executive abilities to neural
As predicted, the case was quite different for the Execution adaptability (Prat & Just, 2011; Prat et al., 2007).
response times. A Group (Bilingual vs. Monolingual) by Practice To further characterize the nature of the differential modulation
(Novel vs. Practiced) ANOVA uncovered a significant effect of of this region between the two groups, we extracted the mean beta
A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61 57

Table 2
Characteristics of the two search volumes used in the predefined volumes analysis. Anatomical locations were identified using the Automated Anatomical Labeling software
package (Tzourio-Mazoyer et al., 2002).

Location Center (MNI) Radius Volume (Voxels) Grey matter (Voxels)


1 Prefrontal cortex 36, 40, 4 20 mm 4169 Middle frontal (1450)
Inferior Frontal (1416)
Inferior pars triangularis (845)
2 Basal ganglia 20, 4, 0 18 mm 3071 Left putamen (906)
Left caudate (272)
Left claustrum (118)
Left globus Pallidus (288)
Left thalamus (71)

Fig. 3. Network of regions involved during the encoding (top) and during the execution (bottom) of novel tasks in bilingual and monolingual participants. Significant clusters
were identified by combining a spatial extent threshold with an uncorrected voxel-wise threshold of p < 0.0005, yielding a cluster-wise FWE-corrected value of p < 0.01. The
spatial extent threshold was calculated separately for each image via Monte Carlo simulations (as performed by AFNI program 3dClustSim). Regions in red were significantly
active in bilinguals; regions in blue were significantly active in monolinguals; purple regions were significantly active in both groups.

value for each condition in each participant across all three clus- recruiting a larger number of voxels within the circuit. Consistent
ters’ voxels using the MarsBaR software (Brett, Anton, with previous research on adaptability, such basal ganglia modula-
Valabregue, & Poline, 2002). The results of this analysis showed tion was associated with improved performance when correlated
that basal ganglia activity moved from negligible (during the exe- with ‘‘cognitive costs’’ (i.e., the timed needed to execute novel tri-
cution of practiced instructions) to significantly positive (during als) in all participants.
the execution of novel instructions) in bilinguals, but remained
negligible in both conditions for monolinguals (see Fig. 4B). 3.4. Follow-up exploratory analysis
This finding suggests that bilinguals may more flexibly recruit
neural resources to execute novel rules, thus offsetting the addi- In addition to our predefined volumes analysis, four whole-
tional cognitive demands and reducing the cost of executing novel brain exploratory neuroimaging analyses were conducted to rule
rules in terms of reaction time. To investigate whether these two out alternative interpretations of our data. The first two analyses
effects (recruitment of additional neural resources and reduced looked for brain regions where activation was either stronger for
response times) were related, we performed a correlational analy- bilinguals than monolinguals (first analysis), or vice versa (second
sis in which each individual subject’s mean difference in basal gan- analysis) during the execution of novel instructions. These two
glia activity for novel and practiced trial execution was correlated analyses were performed to rule out the possibility that bilinguals
with his or her mean execution time for novel trials. The analysis were simply executing novel instructions in a fundamentally dif-
uncovered a significant negative correlation between increases in ferent way than monolinguals. For example, significantly less acti-
basal ganglia activity and corresponding reaction times vation in the bilinguals’ dorso-lateral prefrontal cortex could
[t(29) = 2.89, r = 0.47, p = 0.007; Fig. 4C]. In other words, across indicate a reduced need to rehearse the instructions in working
language groups, larger increases in basal ganglia activity were memory. The third and fourth exploratory analyses compared the
reliably related to correspondingly faster reaction times in all indi- two groups’ brain activity during the encoding (instead of
viduals when executing novel instructions, and accounted for the execution) of novel instructions between the two groups. These
22.1% of the variance in the distribution of reaction times. analyses were performed to rule out the possibility that the
In summary, the hypothesis that bilinguals are more flexible in between-group difference in execution was an effect of encoding
selecting and combining rules was supported by the fact that they differences – for instance, the possibility that bilinguals were
were significantly faster in executing novel instructed mathemati- simply encoding the instructions in a deeper way.
cal tasks than were monolinguals. During such tasks, bilinguals All of these analyses were conducted using the same threshold
also exhibited greater modulation of activity in the basal ganglia, and multiple-error correction procedure (FWE) used in our
58 A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61

Fig. 4. (A) 3D rendition of the cluster of voxels in the left basal banglia, whose activity was more modulated in bilinguals than in monolinguals. Modulatory activity was
assessed by comparing the difference between the execution of novel and practiced instructions across the two groups. (B) Mean beta values (±SEM) during the Execution
phase across all the voxels in the three clusters, divided by group (Bilinguals vs. Monolinguals) and practice (Novel vs. Practiced instructions). Beta values were extracted
using the MarsBaR software (Brett et al., 2002) (C) Scatterplot illustrating the correlation between execution times and activation difference (Novel – Practiced trials) in the
left putamen cluster.

Table 3
The three clusters of regions that were found to be significantly more modulated in bilinguals than monolinguals within the basal ganglia VOI.

Peak MNI coordinates Anatomical locations (% of total cluster volume) Cluster size in voxels Peak intensity (t value)
1 10, 6, 4 Left thalamus (65.71%); 35 3.70
2 22, 14, 2 Left lenticular nucleus, putamen (37.50%); Left lenticular nucleus, pallidum (37.50%); 16 3.08
3 24, 0, 10 Left lenticular nucleus, putamen (85.19%); 27 3.14

predefined volumes analysis. None of them uncovered any brain individual differences investigations (Prat & Just, 2011; Prat
regions whose activity was significantly different between biling- et al., 2007), this dynamic recruitment of resources in the face of
uals and monolinguals. changing task demands was associated with improved cognitive
performance (in this case through reduced cognitive costs, as mea-
4. Conclusions sured by a shorter time needed to apply novel rules).
The centrality of the basal ganglia to this neural adaptability is
This study investigated the neural underpinnings of the bilin- consistent with previous findings that show how this circuit is
gual advantage in executive functions, and the larger implications involved in bilingual language control (Crinion et al., 2006), execu-
of such ‘‘bilingual brain training’’ (Stocco et al., 2014). Consistent tive functions (Avila et al., 2012; O’Reilly and Frank, 2006; Stocco,
with our hypothesis, the behavioral results demonstrated that bil- Lebiere, & Anderson, 2010), and rule learning and application
inguals had a general advantage when flexibly reconfiguring their (Muhammad et al., 2006). The fact that the bilinguals’ improved
behavior, as shown by their ability to execute tasks based on novel performance can be traced back to the basal ganglia is also consis-
rules more quickly than monolinguals, and as quickly and accu- tent with two previous neuroimaging investigations that have
rately as practiced ones. At a neural level, this advantage was asso- shown that greater modulation of basal ganglia activation to
ciated with a correspondingly greater adaptability in the basal changing task demands were related to individual differences in
ganglia in the face of novel task execution. In other words, biling- higher-level cognitive abilities (Prat & Just, 2011; Prat et al.,
uals are more effectively modulating basal ganglia activity in 2007). In particular, Prat and Just (2011) showed that greater mod-
response to changing task demands. Consistent with previous ulation of the basal ganglia in linguistic tasks is also strongly
A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61 59

associated (r = .78) with higher working memory capacity, thus and their mutual dependencies (e.g., Stocco & Anderson, 2008). It
suggesting that adaptability in the basal ganglia strongly relates is therefore possible that extensive bilingual practice results in a
to general cognitive abilities. larger quantity of basal ganglia neurons devoted to binding tempo-
rary variables, which ultimately translates into an advantage when
4.1. Comparing bilinguals to monolinguals novel behaviors need to be implemented. This mechanism is con-
sistent with Crinion et al.’s (2006) conclusions that bilinguals must
It is worth noting, as mentioned by Bialystok, Craik, and Luk ultimately set aside a specific population of basal ganglia neurons
(2012), that ‘‘bilingualism’’ is not a dichotomous variable. In our to represent language-specific information. It is also consistent
study, we carefully balanced our monolingual and bilingual partic- with Prat and Just’s (2011) finding that individuals who show
ipants on non-verbal indices of working memory. This increased higher modulation of basal ganglia activity exhibit higher perfor-
the specificity of our findings, while reducing the probability of mance on working memory tasks, since the capacity of simulta-
false positives due to confounds. The price for this specificity, how- neously maintaining multiple bindings is key for manipulating a
ever, may be seen in terms of power to detect group differences, larger amount of information in working memory (e.g., O’Reilly
because the behavioral difference between the two groups may and Frank, 2006).
have been reduced by equating them on working memory capacity, Previous research has established that, in RITL experiments,
and the number of participants that could be scanned was limited variable binding is crucially needed when applying novel rules
(as only a subset of all the participants whose psychometric mea- (Stocco et al., 2012). This is because executing novel rules requires
sures formed balanced groups were selected). Therefore, while we keeping track of the operations that must be performed, in which
are confident in our findings concerning the role of the basal gan- order, and onto which variables. Our results suggest that, when
glia, we cannot definitely rule out the possibility that bilinguals facing increased demands for variable binding, monolinguals and
and monolinguals selected from groups more representative of bilinguals respond differentially. Specifically, monolinguals seem
the general population might show differential recruitment of to employ an almost constant pool of basal ganglia resources,
other brain regions. For instance, it is possible that bilingual indi- and must compensate for the additional amount of variable bind-
viduals also use less anterior prefrontal resources, but this effect ings by going through additional control steps, thus taking longer
may have gone undetected in the current analysis due to limited to execute novel rules than practiced ones. Bilinguals, on the other
power. This possibility definitely warrants more research in the hand, seem to be able to recruit a larger set of basal ganglia
future. It should be noted, however, that our findings are consistent resources (see Fig. 4A and B), and thus can manage a larger set of
with previous studies that have identified the basal ganglia as a bindings in parallel, resulting in shorter times to apply novel rules.
privileged locus of language control for bilinguals (Crinion et al., This interpretation may be viewed as differences in ‘‘bandwidth’’,
2006), as a region contributing to executive functions (McNab & which determines the number of variables that can be routed dur-
Klingberg, 2008), and whose activation in bilinguals correlates ing a single cognitive step in monolinguals and bilinguals.
with improved behavioral performance (Garbin et al., 2010).
4.3. Real-world applications and future research
4.2. Variable binding and the conditional routing model of bilingual
brain training A better understanding of the cognitive and neural mechanisms
underlying the benefits of bilingual development may ultimately
One important remaining question is ‘‘What is the nature of the lead to the design of training paradigms that produce sustained,
neural computations that underpin the increased basal ganglia and long-lasting improvements of cognitive functioning. The pres-
adaptability for bilinguals?’’ Although our data alone cannot ent study suggests two important features that such paradigms
answer this question, it does provide important preliminary con- should have. First, they should engage participants in the flexible
straints. One possible theory is that the bilingual experience results reconfiguration of behaviors, stressing quick adaptability to novel
in enhanced signal routing functions of the basal ganglia. This the- environments (or tasks). This is a crucial feature of RITL paradigms
ory was originally proposed by Stocco et al. (2014), and is based on that has not been previously investigated in either bilingual
the Conditional Routing model of the basal ganglia proposed by research or in research on cognitive training. Second, at the neural
Stocco et al. (2010). level, an ideal paradigm should involve the signal routing mecha-
According to this model, the main role of the basal ganglia is to nisms of the basal ganglia. Because the advantages of bilingualism,
rapidly select and route neural signals to the prefrontal cortex. This as well as other indices of cognitive functioning (such as working
mechanism is essential to the process of assigning temporary val- memory capacity), are correlated with modulation of basal ganglia
ues to variables during rule-based behaviors (Stocco et al., 2010). activity, this region is an ideal target for training. It may be feasible,
Natural languages often employ very complex rules, thus resulting then, to identify potential training procedures by analyzing their
in increased demands to keep the variable-value bindings consis- relative impact on the adaptability of the basal ganglia circuit.
tent. The Italian pluralization rule, for example, can be expressed The results reported herein also have a direct connection to the
as ‘‘change the final vowel of X to ‘i’ if (1) the gender of X is mas- field of education, as the use of an instructed rule-application par-
culine and (2) if the ending vowel of X is ‘‘e’’. Thus, to pluralize adigm mimics the most common method of learning in educa-
‘‘cane’’ (dog) into ‘‘cani’’ (dogs), one must maintain X bound to tional settings (Cole et al., 2013). Because of the similarity
‘‘cane’’ across two different tests (the gender test and the ending between RITL and traditional educational settings in which indi-
vowel test). Bilingualism adds an additional level of complexity viduals learn to perform novel tasks based on instructions, our
to this rule application process, since bilinguals need to first con- finding that bilinguals are faster at executing new instructions
sider a higher-level variable corresponding to the desired target (and can recruit more neural resources to do so), suggests that they
language. For instance, an English/Italian bilingual needs to con- may also be quicker at learning certain types of rule-based materi-
sider the gender of X (i.e., ‘‘cane’’, masculine) only if the desired als at school. In fact, the 350-ms difference between bilinguals and
output language Y is Italian, and then to access the phonological monolinguals corresponds to 10% faster processing times (see
representation of X (i.e., /kane/, ending vowel ‘‘e’’) in language Y. Fig. 2, center), which translates to significant savings of time for
According to the Conditional Routing model, this increased more complex instructions and longer tasks. For example, bilingual
complexity results in the recruitment of more basal ganglia neu- children might have a significant advantage over monolingual chil-
rons, which are needed to represent a larger number of variables dren when first learning complex rule-based procedures at school,
60 A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61

such as those that are part of the algebra and math curricula. Crinion, J., Turner, R., Grogan, A., Hanakawa, T., Noppeney, U., Devlin, J. T., et al.
(2006). Language control in the bilingual brain. Science, 312(5779), 1537–1540.
Although we are currently not aware of any study that has explic-
http://dx.doi.org/10.1126/science.112776Pf1.
itly investigated this possibility, further investigation seems war- Dale, A. M. (1999). Optimal experimental design for event-related fMRI. Human
ranted. Such findings may have important implications on Brain Mapping, 8(2–3), 109–114.
education policies, as they imply that devoting additional time to Dehaene, S., & Cohen, L. (1995). Towards an anatomical and functional model of
number processing. Mathematical Cognition, 1, 83–120.
second-language curricula could indirectly improve the students’ Dehaene, S., & Cohen, L. (1997). Cerebral pathways for calculation: Double
performance in subjects (such as mathematics, algebra, and phys- dissociation between rote verbal and quantitative knowledge of arithmetic.
ics) that heavily rely on instructed rule application. Cortex, 33(2), 219–250.
Elliott, R. (2003). Executive functions and their disorders. British Medical Bulletin,
In summary, results from the current experiment provide evi- 65(1), 49–59.
dence that links improved executive functions in bilinguals to a Fabbro, F. (2001). The bilingual brain: Cerebral representation of languages. Brain
corresponding enhancement of basal ganglia function. Because this and Language, 79(2), 211–222. http://dx.doi.org/10.1006/brln.2001.2481.
Friston, K. J., Ashburner, J. T., Kiebel, S. J., Nichols, T. E., & Penny, W. D. (Eds.). (2011).
improved performance acquired through bilingual language expe- Statistical parametric mapping: The analysis of functional brain images. London,
rience is connected to basal ganglia functioning, it also provides a UK: Academic Press.
possible biological target to use in developing training to enhance Friston, K. J., Holmes, A., Poline, J. B., Price, C. J., & Frith, C. D. (1996). Detecting
activations in PET and fMRI: Levels of inference and power. Neuroimage, 4(3),
or rehabilitate executive functioning. We see these as exciting 223–235.
opportunities for future research. Garbin, G., Sanjuan, A., Forn, C., Bustamante, J. C., Rodríguez-Pujadas, A., Belloch, V.,
et al. (2010). Bridging language and attention: Brain basis of the impact of
bilingualism on cognitive control. NeuroImage, 53(4), 1272–1278.
Author note Green, D. W., & Abutalebi, J. (2013). Language control in bilinguals: The adaptive
control hypothesis. Journal of Cognitive Psychology, 25(5), 515–530.
Hartstra, E., Kühn, S., Verguts, T., & Brass, M. (2011). The implementation of verbal
This research was supported by a start-up award from the Insti-
instructions: An fMRI study. Human Brain Mapping, 32(11), 1811–1824.
tute for Learning and Brain Sciences to Chantel Prat. The authors Hernandez, A. E. (2009). Language switching in the bilingual brain: What’s next?
would like to thank Justin A. Abernethy for his help during data Brain and Language, 109(2), 133–140.
Hernandez, A. E., Dapretto, M., Mazziotta, J., & Bookheimer, S. (2001). Language
collection and preprocessing, and Theresa Becker for providing
switching and language representation in Spanish–English bilinguals: An fMRI
comments on an earlier draft of this manuscript. study. NeuroImage, 14(2), 510–520.
Luk, G., De Sa, E., & Bialystok, E. (2011). Is there a relation between onset age of
bilingualism and enhancement of cognitive control. Bilingualism: Language and
Appendix A. Supplementary material Cognition, 14(4), 588–595.
Marian, V., Blumenfeld, H. K., & Kaushanskaya, M. (2007). The language experience
Supplementary data associated with this article can be found, in and proficiency questionnaire (LEAP-Q): Assessing language profiles in
bilinguals and multilinguals. Journal of Speech, Language and Hearing Research,
the online version, at http://dx.doi.org/10.1016/j.bandl.201 50, 940.
4.07.005. Mayr, U., & Keele, S. W. (2000). Changing internal constraints on action: The role of
backward inhibition. Journal of Experimental Psychology: General, 129(1), 4.
McNab, F., & Klingberg, T. (2008). Prefrontal cortex and basal ganglia control access
References to working memory. Nature Neuroscience, 11(1), 103–107.
Melby-Lervåg, M., & Hulme, C. (2012). Is working memory training effective? A
Abutalebi, J., Miozzo, A., & Cappa, S. (2000). Do subcortical structures control meta-analytic review. Developmental Psychology, 49(2), 270–291.
‘‘language selection’’ in polyglots? Evidence from pathological language mixing. Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex
Neurocase, 6, 51–56. function. Annual Review of Neuroscience, 24(1), 167–202.
Avila, C., Garbin, G., Sanjuán, A., Forn, C., Barrós-Loscertales, A., Bustamante, J. C., Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D.
et al. (2012). Frontostriatal response to set switching is moderated by reward (2000). The unity and diversity of executive functions and their contributions to
sensitivity. Social, Cognitive, and Affective Neuroscience, 7(4), 423–430. complex ‘‘frontal lobe’’ tasks: A latent variable analysis. Cognitive Psychology,
Bialystok, E. (1999). Cognitive complexity and attentional control in the bilingual 41(1), 49–100.
mind. Child Development, 70(3), 636–644. Muhammad, R., Wallis, J. D., & Miller, E. K. (2006). A comparison of abstract rules in
Bialystok, E., Craik, F. I., Klein, R., & Viswanathan, M. (2004). Bilingualism, aging, and the prefrontal cortex, premotor cortex, inferior temporal cortex, and striatum.
cognitive control: Evidence from the Simon task. Psychology and Aging, 19, Journal of Cognitive Neuroscience, 18(6), 974–989.
290–303. O’Reilly, R. C., & Frank, M. J. (2006). Making working memory work: A
Bialystok, E., Craik, F. I. M., & Luk, G. (2012). Bilingualism: Consequences for mind computational model of learning in the prefrontal cortex and basal ganglia.
and brain. Trends in Cognitive Sciences, 16(4), 240–250. Neural Computation, 18(2), 283–328.
Bialystok, E., & Martin, M. M. (2004). Attention and inhibition in bilingual children: Paap, K. R., & Greenberg, Z. I. (2013). There is no coherent evidence for a bilingual
Evidence from the dimensional change card sort task. Developmental Science, advantage in executive processing. Cognitive Psychology, 66(2), 232–258.
7(3), 325–339. Pasupathy, A., & Miller, E. K. (2005). Different time courses of learning-
Brett, M., Anton, J. L., Valabregue, R., & Poline, J. B. (2002). Region of interest analysis related activity in the prefrontal cortex and striatum. Nature, 433(7028),
using an SPM toolbox. In 8th international conference on functional mapping of 873–876.
the human brain. Sendai, Japan. Perani, D., Paulesu, E., Galles, N. S., Dupoux, E., Dehaene, S., Bettinardi, V., et al.
Briellmann, R. S., Saling, M. M., Connell, A. B., Waites, A. B., Abbott, D. F., & Jackson, (1998). The bilingual brain. Proficiency and age of acquisition of the second
G. D. (2004). A high-field functional MRI study of quadri-lingual subjects. Brain language. Brain, 121(10), 1841–1852.
and Language, 89(3), 531–542. Prat, C. S. (2011). The brain basis of individual differences in language
Brown, J. A., Fishco, V. V., & Hanna, G. (1993). Nelson–Denny reading test: Manual for comprehension abilities. Language and Linguistics Compass, 5(9), 635–649.
scoring and interpretation, forms G & H. Rolling Meadors, IL: Riverside Publishing. Prat, C. S., & Just, M. A. (2011). Exploring the neural dynamics underpinning
Buchweitz, A., & Prat, C. (2013). The bilingual brain: Flexibility and control in the individual differences in sentence comprehension. Cerebral Cortex, 21(8),
human cortex. Physics of Life Reviews, 10(4), 428–443. 1747–1760.
Carlson, S. M., & Meltzoff, A. N. (2008). Bilingual experience and executive Prat, C. S., Keller, T. A., & Just, M. A. (2007). Individual differences in sentence
functioning in young children. Developmental Science, 11(2), 282–298. comprehension: A functional magnetic resonance imaging investigation of
Chee, M. W., Hon, N., Lee, H. L., & Soon, C. S. (2001). Relative language proficiency syntactic and lexical processing demands. Journal of Cognitive Neuroscience,
modulates BOLD signal change when bilinguals perform semantic judgments. 19(12), 1950–1963.
Neuroimage, 13(6), 1155–1163. Prat, C. S., Mason, R. A., & Just, M. A. (2012). An fMRI investigation of analogical
Cole, M. W., Bagic, A., Kass, R., & Schneider, W. (2010). Prefrontal dynamics mapping in metaphor comprehension: The influence of context and individual
underlying rapid instructed task learning reverse with practice. The Journal of cognitive capacities on processing demands. Journal of Experimental Psychology:
Neuroscience, 30(42), 14245–14254. Learning, Memory, and Cognition, 38(2), 282–294.
Cole, M. W., Laurent, P., & Stocco, A. (2013). Rapid instructed task learning: A new Prior, A., & MacWhinney, B. (2009). A bilingual advantage in task switching.
window into the unique cognitive flexibility of the human brain. Cognitive, Bilingualism: Language and Cognition, 13(02), 253.
Affective & Behavioral Neuroscience, 13(1), 1–22. Ruge, H., & Wolfensteller, U. (2010). Rapid formation of pragmatic rule
Colzato, L. S., Bajo, M. T., Van den Wildenberg, W., Paolieri, D., Nieuwenhuis, S., La representations in the human brain during instruction-based learning.
Heij, W., et al. (2008). How does bilingualism improve executive control? A Cerebral Cortex, 20(7), 1656–1667.
comparison of active and reactive inhibition mechanisms. Journal of Shipstead, Z., Redick, T. S., & Engle, R. W. (2012). Is working memory training
Experimental Psychology: Learning, Memory, and Cognition, 34(2), 302–312. effective? Psychological Bulletin, 138(4), 628.
A. Stocco, C.S. Prat / Brain & Language 137 (2014) 50–61 61

Stocco, A., & Anderson, J. R. (2008). Endogenous control and task representation: An Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version
fMRI study in algebraic problem-solving. Journal of Cognitive Neuroscience, of the operation span task. Behavior Research Methods, 37(3), 498–505.
20(7), 1300–1314. Venkatraman, V., Siong, S. C., Chee, M. W., & Ansari, D. (2006). Effect of language
Stocco, A., Lebiere, C., & Anderson, J. R. (2010). Conditional routing of information to switching on arithmetic: A bilingual fMRI study. Journal of Cognitive
the cortex: A model of the basal ganglia’s role in cognitive coordination. Neuroscience, 18(1), 64–74.
Psychological Review, 117(2), 541–574. Wallis, J. D., Anderson, K. C., & Miller, E. K. (2001). Single neurons in prefrontal
Stocco, A., Lebiere, C., O’Reilly, R. C., & Anderson, J. R. (2012). Distinct contributions cortex encode abstract rules. Nature, 411(6840), 953–956.
of the caudate nucleus, rostral prefrontal cortex, and parietal cortex to the Wang, X., Wang, Y. -Y., Jiang, T., Wang, Y. -Z., & Wu, C. -X. (2012). Direct evidence of
execution of instructed tasks. Cognitive, Affective, & Behavioral Neuroscience, the left caudate’s role in bilingual control: An intra-operative electrical
12(4), 611–628. stimulation study. Neurocase, 37–41 (September 2012).
Stocco, A., Yamasaki, B., Natalenko, R., & Prat, C. S. (2014). Bilingual brain training: A Wendelken, C., Nakhabenko, D., Donohue, S. E., Carter, C. S., & Bunge, S. A. (2008).
neurobiological framework of how bilingual experience improves executive ‘‘Brain is to thought as stomach is to?’’: Investigating the role of rostrolateral
function. International Journal of Bilingualism, 18(1), 67–92. prefrontal cortex in relational reasoning. Journal of Cognitive Neuroscience, 20(4),
Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, 682–693.
N., et al. (2002). Automated anatomical labeling of activations in SPM using a
macroscopic anatomical parcellation of the MNI MRI single-subject brain.
Neuroimage, 15(1), 273–289.

You might also like