ArticlePDF AvailableLiterature Review

Bilingual Minds


Abstract and Figures

The regular use of two languages by bilingual individuals has been shown to have a broad impact on language and cognitive functioning. In this monograph, we consider four aspects of this influence. In the first section, we examine differences between mono-linguals and bilinguals in children's acquisition of language and adults' linguistic processing, particularly in terms of lexical retrieval. Children learning two languages from birth follow the same milestones for language acquisition as mono-linguals do (first words, first use of grammar) but may use different strategies for language acquisition, and they generally have a smaller vocabulary in each language than do monolin-gual children learning only a single language. Adult bilinguals typically take longer to retrieve individual words than monolin-guals do, and they generate fewer words when asked to satisfy a constraint such as category membership or initial letter. In the second section, we consider the impact of bilingualism on nonverbal cognitive processing in both children and adults. The primary effect in this case is the enhancement of executive control functions in bilinguals. On tasks that require inhibition of distract-ing information, switching between tasks, or holding information in mind while performing a task, bilinguals of all ages outperform comparable monolinguals. A plausible reason is that bilinguals recruit control processes to manage their ongoing linguistic per-formance and that these control processes become enhanced for other unrelated aspects of cognitive processing. Preliminary evi-dence also suggests that the executive control advantage may even mitigate cognitive decline in older age and contribute to cognitive reserve, which in turn may postpone Alzheimer's disease. In the third section, we describe the brain networks that are responsible for language processing in bilinguals and demon-strate their involvement in nonverbal executive control for bilinguals. We begin by reviewing neuroimaging research that identifies the networks used for various nonverbal executive control tasks in the literature. These networks are used as a ref-erence point to interpret the way in which bilinguals perform both verbal and nonverbal control tasks. The results show that bilinguals manage attention to their two language systems using the same networks that are used by monolinguals performing nonverbal tasks. In the fourth section, we discuss the special circumstances that surround the referral of bilingual children (e.g., language delays) and adults (e.g., stroke) for clinical intervention. These referrals are typically based on standardized assessments that use normative data from monolingual populations, such as vocabulary size and lexical retrieval. As we have seen, however, these measures are often different for bilinguals, both for children and adults. We discuss the implications of these linguistic differences for standardized test performance and clinical approaches. We conclude by considering some questions that have important public policy implications. What are the pros and cons of French or Spanish immersion educational programs, for example? Also, if bilingualism confers advantages in certain respects, how about three languages—do the benefits increase? In the healthcare field, how can current knowledge help in the treatment of bilingual aphasia patients following stroke? Given the recent increase in bilingualism as a research topic, answers to these and other related questions should be available in the near future.
Content may be subject to copyright.
Bilingual Minds
Ellen Bialystok
, Fergus I.M. Craik
, David W. Green
, and
Tamar H. Gollan
Department of Psychology, York University,
Rotman Research Institute, Baycrest Centre,
Department of Cognitive,
Perceptual, and Brain Sciences, University College London,
University of California, San Diego
The regular use of two languages by bilingual individuals has
been shown to have a broad impact on language and cognitive
functioning. In this monograph, we consider four aspects of this
In the first section, we examine differences between mono-
linguals and bilinguals in children’s acquisition of language
and adults’ linguistic processing, particularly in terms of
lexical retrieval. Children learning two languages from birth
follow the same milestones for language acquisition as mono-
linguals do (first words, first use of grammar) but may use
different strategies for language acquisition, and they generally
have a smaller vocabulary in each language than do monolin-
gual children learning only a single language. Adult bilinguals
typically take longer to retrieve individual words than monolin-
guals do, and they generate fewer words when asked to satisfy a
constraint such as category membership or initial letter.
In the second section, we consider the impact of bilingualism on
nonverbal cognitive processing in both children and adults. The
primary effect in this case is the enhancement of executive control
functions in bilinguals. On tasks that require inhibition of distract-
ing information, switching between tasks, or holding information
in mind while performing a task, bilinguals of all ages outperform
comparable monolinguals. A plausible reason is that bilinguals
recruit control processes to manage their ongoing linguistic per-
formance and that these control processes become enhanced for
other unrelated aspects of cognitive processing. Preliminary evi-
dence also suggests that the executive control advantage may even
mitigate cognitive decline in older age and contribute to cognitive
reserve, which in turn may postpone Alzheimer’s disease.
In the third section, we describe the brain networks that are
responsible for language processing in bilinguals and demon-
strate their involvement in nonverbal executive control for
bilinguals. We begin by reviewing neuroimaging research that
identifies the networks used for various nonverbal executive
control tasks in the literature. These networks are used as a ref-
erence point to interpret the way in which bilinguals perform
both verbal and nonverbal control tasks. The results show that
bilinguals manage attention to their two language systems
using the same networks that are used by monolinguals
performing nonverbal tasks.
In the fourth section, we discuss the special circumstances
that surround the referral of bilingual children (e.g., language
delays) and adults (e.g., stroke) for clinical intervention. These
referrals are typically based on standardized assessments that
use normative data from monolingual populations, such as
vocabulary size and lexical retrieval. As we have seen,
however, these measures are often different for bilinguals, both
for children and adults. We discuss the implications of these
linguistic differences for standardized test performance and
clinical approaches.
We conclude by considering some questions that have
important public policy implications. What are the pros and
cons of French or Spanish immersion educational programs,
for example? Also, if bilingualism confers advantages in
certain respects, how about three languages—do the benefits
increase? In the healthcare field, how can current knowledge
help in the treatment of bilingual aphasia patients following
stroke? Given the recent increase in bilingualism as a research
topic, answers to these and other related questions should be
available in the near future.
As the world becomes more interconnected, it is increasingly
apparent that bilingualism is the rule and not the exception. Not
only do some countries support bilingual populations because
of cultural and linguistic diversity within its citizenry, but also
increased global mobility has enlarged the number of people
who have become bilingual at all levels of society. For exam-
ple, a recent survey of language use in the United States
obtained from the American Community Survey in 2007
reported that approximately 20%of the population spoke a
non-English language at home, a proportion that has increased
by 140%since 1980 (Shin & Kominski, 2010). These numbers
are higher when considering world figures: Crystal (1997) esti-
mates bilingualism that includes English and another language
Corresponding Author:
Ellen Bialystok, Department of Psychology, York University, 4700 Keele Street,
Toronto, Ontario M3J 1P3, Canada
Psychological Science in the
Public Interest
10(3) 89–129
ªThe Author(s) 2009
Reprints and permission:
DOI: 10.1177/1529100610387084
represents about 235 million people worldwide and that two
thirds of the children in the world are raised in bilingual
Recently, evidence indicating that this common experience
has a systematic and significant impact on cognitive functioning
has accumulated. In this review, we examine the nature of that
impact across the lifespan and consider what these effects con-
tribute to our understanding of cognition in general. We begin
by examining the linguistic dimensions of bilingualism in terms
of children’s language acquisition and adult language process-
ing. In the second section, we investigate the consequences of
bilingualism on nonverbal cognitive functioning. The third sec-
tion describes research documenting how the brain supports
bilingual functioning and how it changes in response to it. In the
fourth section, we review the clinical implications of bilingual-
ism for diagnosis and intervention. We conclude by identifying
and discussing some specific issues for bilinguals in society. By
adopting this cognitive perspective, there are a number of topics
we do not cover, such as reading, lexical and syntactic process-
ing, and linguistic consequences of brain damage, all of which
are beyond the scope of the present review.
There are many ways to be bilingual: Some people are born
bilingual, some aspire to bilingualism, and others have bilingu-
alism thrust upon them later in life. Underlying these differ-
ences, a myriad of factors make the bilingual experience
deeply heterogeneous and potentially alter its consequences.
Some of the reasons for bilingualism include immigration, a
family that speaks a heritage language, formal education in
another language, temporary residence in another country, or
a national situation in which the official language is different
from the community language. Each of these circumstances
is associated with a different set of social, cognitive, and per-
sonal factors, and these factors undoubtedly intervene in as
well as determine any potential effect of bilingualism. Each
of the situations associated with multiple language use also car-
ries different assumptions about expectations for education,
values around literacy, standards for language proficiency,
the purposes for which one or both of the languages are used,
the level of community support for the home language, and the
identity of the individual as a member of a majority or minority
culture. Therefore, there can be no single outcome and no defi-
nitive consequence that follows from incorporating more than
one language into daily life. And yet the consequences of
bilingualism affect educational policy, social organization, and
conceptions of mind.
1. Language Learning and Language Use in
Language acquisition in bilingual children
The most striking feature of a young child’s acquisition of
language is the extraordinary ease with which the process appears
to progress. Perhaps more remarkable than this achievement,
therefore,is that this facility for learning a complex symbolic sys-
tem is not diminished when the child faces the task oflearning two
of them. Bilingual language acquisition is as effortless, efficient,
and successful as monolingual acquisition. It is now clear that lan-
guage acquisition is not a simple matter of biological unfolding,
as some had previously believed, but rather a process that is finely
tuned to features of the environmental input, the child’s atten-
tional and perceptual abilities, and the development of cogni-
tive and conceptual competencies. All of these factors conspire
as well to shape the process of acquiring two languages. More-
over, as we describe later, the major milestones concerning
competence in sounds, words, and sentences that are the
foundation of acquiring language are passed at equivalent
times for children growing up with one language in the home
and those growing up in a multilingual home.
The acquisition of the phonological system by infants has
been well documented for the case of monolingual acquisition:
Infants can detect the contrasts that define the phonological
system for all human languages almost from birth (e.g., /pa/
vs. /ba/; Eimas, Siqueland, Jusczyk, & Vigorito, 1971), but
their ability to perceive these contrasts in languages that are not
heard in the environment (e.g., /r/ vs. /l/ for children being
raised in Japanese homes) begins to decline at about 6 months
of age (Werker & Tees, 1984; see also Kuhl et al., 2006). Thus,
until about 6 months old, there is no detectable difference in the
perception of phonetic contrasts by infants in monolingual and
bilingual environments but diverging patterns appear as bilin-
gual babies maintain and develop the categorical distinctions
for the phonetic system in both languages and monolingual
infants lose the ability to detect contrasts that are not part of the
language they are about to learn (Burns, Yoshida, Hill, &
Werker, 2007; Sebastian-Galles & Bosch, 2005). By about
14 months old, infants being raised in bilingual environments
have established a clearly demarcated phonological representa-
tion for both languages. Therefore, bilingual infants develop
the phonological basis for both languages on roughly the same
schedule as monolingual children do for their only language.
It may be that it is this very early experience that leaves its life-
long trace as a foreign accent when childhood monolinguals
attempt to learn new languages later in life.
Beyond the phonetic constituents, infants also need to learn
the more general phonological structure of language. Recently,
Kovacs and Mehler (2009a) presented auditory stimuli to
12-month-old infants who were being raised in a monolingual
or bilingual environment. The stimuli were three-syllable
combinations that had the syllabic structure of either ABA or
AAB. These stimuli were artificial creations and were not
words in any language. The crucial manipulation was that each
structure was associated with a different response—namely,
look either to the right or to the left to see an interesting toy.
The experimental results showed that the monolingual babies
could learn only one of the responses but that the bilingual
babies learned both, a difference the researchers interpreted
as demonstrating more flexible learning in bilinguals. They
offer their results as part of the explanation for how bilingual
children can learn twice as much about language as monolin-
gual children in the same amount of time (although it is not
clear that they do, as will be discussed below), but the task was
90 Bialystok et al.
only marginally linguistic. If anything, it is more similar to
word learning than to speech perception, a process that rests
on different perceptual and cognitive processes than phonologi-
cal development (Burns et al., 2007). In fact, bilingual babies
apply their developing phonological system to the learning of
new words later than monolingual children do (Fennell,
Byers-Heinlein, & Werker, 2007), although a recent study
testing 17-month-old infants raised with French and English
did not replicate this finding and attributed the difference
between studies to details of the phonetic input (Mattock,
Polka, Rvachew, & Krehm, 2010). Nonetheless, the results
reported by Kovacs and Mehler provide compelling evidence
for different levels of performance in a phonological task in the
first year of life that can be traced to the experience of building
up two linguistic systems.
Undoubtedly the most salient evidence for children’s
progress in language acquisition is word learning, particularly
the appearance of the child’s first word. As with the developing
phonological system, the basic milestones associated with this
achievement are similar for children learning one or more
languages. The child’s first word appears on average at about
1 year old, regardless of how many languages are in the
environment (Pearson, Fernandez, & Oller, 1993) and, more
dramatically, regardless of whether the languages are both
spoken or one is spoken and one is signed (Petitto et al.,
2001). However, two factors may be different for monolingual
and bilingual children: the strategies for word learning and the
rate and extent of vocabulary acquisition.
One strategy that allows children to rapidly learn new words
is to assume that novel words signify unfamiliar objects,
presenting a simple pairing of word and concept. This strategy
of word–meaning assignment follows from what Markman and
Wachtel (1988) posit as the mutual exclusivity constraint—the
assumption that a thing can only have one name—although this
assumption need not be innately determined. The evidence for
mutual exclusivity is that children appear to create mappings
between new words and new objects—for example, if a child
hears the word ‘‘bik’’ while looking at a cup and an unknown
object, the child will assume that the novel item is called a bik.
But bilingual children already know that things can have more
than one name—the known object could be ‘‘a cup’’ or ‘‘une
tasse.’’ Do bilingual children follow the strategy of mapping
unknown words to unknown objects? The evidence is mixed,
with some studies reporting less reliance on this strategy for
bilingual children (Bialystok, Barac, Blaye, & Poulin-Dubois,
2010; Davidson & Tell, 2005) but others reporting
no difference between monolingual and bilingual children
(Au & Glusman, 1990; Merriman & Kutlesic, 1993). More con-
vincing, however, is evidence from a study by Byers-Heinlein
and Werker (2009) in which they compared the adherence to
this strategy by children learning one, two, or three languages.
Their results showed a systematic decline in the reliance on this
heuristic with the number of languages being learned. These
results, in conjunction with those reported by Kovacs and
Mehler (2009a) suggesting that phonological word structures
are perceived differently by monolingual and bilingual
children, are consistent with a view in which the actual
mechanisms of word learning used by monolingual children
differ from those used by bilingual children. Importantly,
however, the essential cognitive landmark that guides these
mechanisms, namely, the time at which the child is able to pro-
duce the first meaningful word, is comparable for all children.
The second difference in word learning between monolin-
gual and bilingual children is in the size of their developing
vocabularies. As in phonological discrimination and first word
production, the timetable for the critical milestone is similar for
children with both types of experience. In this case, the crucial
landmark is the establishment of a vocabulary of 50 words,
which is achieved by both monolingual and bilingual children
at about 1½ years old (Pearson et al., 1993; Petitto, 1987;
Petitto et al., 2001), at least for total vocabulary across the two
languages. Beyond that, however, the evidence is compelling
that, on average, bilingual children know significantly fewer
words in each language than comparable monolingual children.
A careful investigation examining how many words children
between 8 and 30 months old knew in each language confirmed
that, on average, this number was smaller in each language for
bilingual children than for monolingual learners of that
language (Pearson et al., 1993). The number of words in the
total vocabulary of a bilingual child, however, is difficult to
estimate: Do proper names count for one language or two?
Do cognates count once or twice, especially if the pronuncia-
tion is unclear? Do childish sounds that are not quite words
count as words if they have a consistent meaning?
A clearer illustration of the relative vocabulary size of
monolinguals and bilinguals comes from a study of children
who were older than those in the Pearson et al. (1993) analysis.
Bialystok, Luk, Peets, and Yang (2010) measured the receptive
vocabulary of over 1,700 children between the ages of 3 and
10 years old. All the bilingual children spoke English and
another language, with English being the language of the com-
munity and school for all children. Across the sample and at
every age studied, the mean standard score on the English Pea-
body Picture Vocabulary Test (PPVT) of receptive vocabulary
(Dunn & Dunn, 1997) was reliably higher for monolinguals
than for bilinguals. These results are shown in Figure 1. At least
in one of the two languages and, importantly, the language of
schooling, monolingual children had an average receptive
vocabulary score that was consistently higher than that of their
bilingual peers. It is important to note, however, that the dispa-
rities were not equivalent for all words. In a subset of 6-year-
olds in the sample, all children achieved comparable scores
on words associated with schooling (e.g., astronaut,rectangle,
writing) but bilinguals obtained significantly lower scores for
words associated with home (e.g., squash,canoe,pitcher).
Therefore, the nature of the smaller vocabulary of bilingual
speakers of each language than that of monolingual speakers
is in fact somewhat complex (Bialystok, Luk, et al., 2010).
The hallmark of human language, however, is not sounds or
words, but the grammatically constrained combinations of
units to form utterances or sentences. Again, the transition into
this stage of language acquisition occurs on the same timetable
Bilingual Minds 91
for children learning one or more languages: The first word
combinations for all children appear at about 1½ years old
(Pearson et al., 1993; Petitto et al., 2001), with utterances becom-
ing incrementally more complex on a similar trajectory (de
Houwer, 1995). The details of children’s increasing grammatical
sophistication appear to be tied to the specific language, with
examples for this point coming from children learning English
and Spanish (Gathercole, 1997), English and French (Paradis &
Genesee, 1996), and French and German (Meisel, 1990).
Current theories of language acquisition are based on the
idea that there is a deep connection between words and
structure: Grammar is part of the linguistic system and emerges
seamlessly when the lexicon has reached a critical mass. The
first evidence for structure occurs when the child knows about
50 words, a relationship demonstrated for both monolingual
(Bates & Goodman, 1997) and bilingual (Conboy & Thal,
2006) children. In this sense, discussion of children’s early
grammar is not different in kind from the discussion of their
early lexicon, but the issues in their development present them-
selves in different ways. And if language acquisition is not
guided by dedicated modules equipped to detect and record
grammatical structure, then what directs this process? From the
cognitive perspective, the linguistic and cognitive systems are
intimately interconnected, each guiding the other and profiting
from the symbiotic relationship. What happens when a child is
learning two languages?
Across the major linguistic features—sounds, words,
grammar—the acquisition of language by monolingual and
bilingual children follows a similar timetable for milestones that
largely reflect cognitive ability, but the linguistic competence
that is developing is different. Partly because linguistic
knowledge for bilingual children is divided across two
languages, the organization and richness of the representational
system in each language is different from that acquired by a
monolingual speaker of one of the languages. Similarities in
developing cognitive abilities keep the process of language
acquisition on a common time course, but variation in input and
use make the developing linguistic systems quite different
both qualitatively and quantitatively. Understanding bilingual
language ability and the bilingual mind more broadly requires
understanding these interfaces between the linguistic and
cognitive systems.
Two languages in the mind
The bilingual mind presents an intriguing set of puzzles. Are
the two languages represented in separate or in overlapping
systems? Are concepts duplicated or shared across languages?
Do interactions between languages facilitate or impede
language production? How are the selection of the target
language and avoidance of the nontarget language achieved?
How does the bilingual switch between languages, both inten-
tionally and unintentionally? None of these questions applies to
monolingual language use, so from the outset, the presence of
two languages in mind changes fundamental aspects of lan-
guage processing. Moreover, these questions are all inherently
about cognitive systems at least as much as they are about
linguistic ones; switching between representational systems
and avoiding interference are processes routinely handled by
the general executive control system. Therefore, bilingual
language use must be intimately tied to a cognitive system in
a way that is less essential for monolingual speech. It is those
3 years
4 years
5 years
6 years
7 years
8 years
9 years
10 years
Age Group
Mean PPVT std. score
Monolinguals Bilinguals
Fig. 1. Mean Peabody Picture Vocabulary Test (PPVT) standard score and standard error by age and language group
(monolinguals, M, vs. bilinguals, B). From Bialystok, Luk, Peets, and Yang (2010).
92 Bialystok et al.
relations between language and cognition that will be examined
in this section: How is language processing different when there
are two fully elaborated linguistic systems available? How does
that situation change the cognitive processes whose responsibil-
ity it is to manage those language systems? There is an active
body of research examining these questions, comparing how
bilinguals can carry out these tasks in their two languages (for
excellent reviews of this literature, see Kroll & de Groot,
2005). However, the present question is not to compare process-
ing of the two languages of bilingual speakers but to compare
monolinguals and bilinguals as they perform similar tasks.
To understand how the simple act of speaking may be
different for monolinguals and bilinguals, it is necessary to
acknowledge two crucial differences between these groups.
First, the knowledge base from which all language processing
proceeds is less rich or less interconnected for a bilingual in
each language than it is for a monolingual speaker of one of
those languages. The most salient difference in the language
competence of monolingual and bilingual children is in the
vocabulary scores obtained in a given language, as described
earlier (Bialystok, Luk, et al., 2010)—a pattern that may persist
into adulthood. Although it is more difficult to attribute reliable
differences in adults’ vocabulary size to bilingualism versus
monolingualism than it is for that of children because of the
enormous variation in adults’ knowledge of words, there is none-
theless evidence that such systematic differences exist (e.g., Bia-
lystok, Craik, & Luk, 2008a; Portocarrero, Burright, &
Donovick, 2007). Gollan and colleagues argue that the essential
feature of bilingual representations is the ‘‘weaker links’’ that are
established within the network because of less frequent use of
each language (Gollan, Montoya, Cera, & Sandoval, 2008); sim-
ply using each language less often produces weaker connections
in the network than would emerge from greater use. In this view,
the knowledge resources underlying language performance for
monolinguals and bilinguals who are comparable on many other
cognitive abilities are not equivalent.
Second, it is now well documented that both languages of a
bilingual are jointly activated even in contexts that strongly
bias towards only one of them. Evidence for this claim comes
from both behavioral (Beauvillain & Grainger, 1987; Colome´,
2001; Grainger, 1993; Hernandez, Bates, & Avila, 1996;
Francis, 1999; Kroll & de Groot, 1997) and imaging studies
(Marian, Spivey, & Hirsch, 2003; Martin, Dering, Thomas, &
Thierry, 2009; Rodriguez-Fornells, Rotte, Heinze, Nosselt, &
Munte, 2002). One of the first pieces of evidence for this
conclusion comes from an ingenious experiment by Guttentag,
Haith, Goodman, and Hauch (1984, Experiment 2). On each
trial, bilingual participants viewed a word drawn from one of
four semantic categories (e.g., metals, clothing, furniture, and
trees); two categories were assigned to one response key and
the other two categories to a second key. The participant’s task
was to press the designated key to indicate the category mem-
bership of the target word as rapidly as possible. Each stimulus
word also had copies of a further word above and below it as
flanker items. These flankers were always in the participant’s
other language and belonged to one of four categories:
translations of the target word, a different word drawn from the
same semantic category as the target, a word from a different
category but requiring the same response, or a word from a
category requiring a different response. The crucial result is
that response times were significantly longer in the second two
conditions, showing that participants were unable to ignore the
flankers and that some analysis of the flankers’ categories
(and possibly responses) took place despite the fact that the
flankers were in the nonused language.
This joint activation of the two languages creates a unique
need for selection in bilinguals in which language processing
must resolve competition not only from within-language
alternatives as monolinguals do to select among close semantic
neighbors (words that share semantic features, e.g., cup vs. mug;
Luce & Large, 2001; Mirman & Magnuson, 2008; Vitevitch,
2002) but also from between-language alternatives for the same
concepts (e.g., cup vs. tasse). The predominant view is that lan-
guage selection does not normally occur prior to speech, making
this selection part of bilingual speech production (Kroll, Bobb,
& Wodniecka, 2006). For this reason, a somewhat different set
of attention and control procedures is necessary for speech pro-
duction in bilinguals than is necessary for monolinguals (Green,
1998). However, there is less agreement on what those special
processes might be. Some studies have shown that the nontarget
language is actually inhibited while using the other language
(e.g., Levy, McVeigh, Marful, & Anderson, 2007; Philipp &
Koch, 2009), but others indicate that correct selection can be
achieved by increasing the activation of the preferred response
(Costa, Santesteban, & Ivanova, 2006; see Costa, 2005, for a dis-
cussion of these views). As we describe later, these alternatives
need not be mutually exclusive: Selection depends on the activa-
tion level of both the target item to be selected and that of the
competing items, incorporating as well views that reject the role
of competition and instead focus on selection (Caramazza, 1997;
La Heij, 1988; Mahon, Costa, Peterson, Vargas, & Caramazza,
2007; Roelofs, 2003). Therefore, selection is facilitated by either
preferentially enhanced activation of the target, inhibition of the
competitor, or both. Whatever the mechanism, selection of
appropriate lexical items for bilinguals involves either
different or additional processes than does the same activity for
monolinguals. Taken together, the differences in the linguistic
representations and differences in the selection mechanisms lead
to sustained differences between monolinguals and bilinguals in
fluent speech production.
Although ordinary conversation does not generally signal
observable deficits in bilingual language processing, controlled
experimental procedures can reveal more subtle differences
between these two groups. Two such features are the speed
with which target words can be retrieved in response to a cue
and the number of words that can be generated to satisfy a
criterion. Evidence for the first comes primarily from studies
of picture naming or semantic classification, and evidence for
the second comes from studies of verbal fluency.
Lexical retrieval in bilinguals. Much of the research in lexical
retrieval compares the relative ability of multilingual speakers
Bilingual Minds 93
to perform such tasks as naming the pictures in their two (or
more) languages (Costa & Santesteban, 2004; Hernandez,
Martinez, & Kohnert, 2000), making semantic classifications
for words in the two languages (Dufour & Kroll, 1995), or
translating between languages (Kroll & Stewart, 1994). The
purpose is to compare lexical access to the two languages and,
in some cases, as in the study by Dufour and Kroll (1995), to
compare bilinguals who are more or less fluent in the two
languages. The issue we are discussing here is different: to
compare monolingual and bilingual speakers naming pictures
in the same language. The comparison is inherently fraught
with difficulty: If we assume that bilinguals never have identi-
cal proficiency in their two languages and, moreover, that even
their ability in their stronger language may not fully resemble
the language competence of a monolingual speaker of that
language, then any comparison of monolinguals and bilinguals
seems unfair. And yet, proficient bilinguals manage to function
perfectly well, belying the notion of an underlying handicap.
Thus it may be that the task of rapidly accessing target lexical
items is carried out differently by monolinguals and bilinguals,
an outcome that would be important in understanding
the relation between language and cognitive systems in the
bilingual mind.
Research shows that bilingual participants take longer and
make more errors than monolinguals on naming tasks. Using
the Boston Naming Task (Kaplan, Goodglass, & Weintraub,
1983), bilinguals produced fewer correct responses (Roberts,
Garcia, Desrochers, & Hernandez, 2002; Gollan, Fennema-
Notestine, Montoya, & Jernigan, 2007) and made more errors
on a speeded version of the task (Bialystok et al., 2008a) than
did monolinguals. On timed picture naming, bilinguals
performed more slowly than did monolinguals (Gollan,
Montoya, Fennema-Notestine, & Morris, 2005). Similar results
(slower responses in bilinguals) are found in both comprehend-
ing (Ransdell & Fischler, 1987) and producing words (Ivanova
& Costa, 2008), even when bilinguals respond in their first and
dominant language. The simple act of retrieving a common
word seems to be more effortful for bilinguals.
Healthy aging is frequently accompanied by a reduction in
productive language abilities—searching for words and names
becomes a more salient part of every conversation. Consistent
with this trend, picture naming is carried out more slowly by
older adults than by younger adults, even for monolinguals
(e.g., Albert, Heller, & Milberg, 1988). Therefore, older
bilinguals should find lexical access particularly difficult, since
both age and language status are associated with poorer
performance. The situation is even more problematic for older
bilinguals who may have spent the majority of their adult lives
using one of their two languages, usually the second language
(L2), and have been removed from a daily context that supports
the first language (L1). The outcome of this situation can be
attrition of the L1. Therefore, difficulties in performance on
tests of lexical access such as picture naming can be attributa-
ble to normal aging, L1 attrition, or both. These possibilities
were evaluated in a study by Goral, Libben, Obler, Jarema, and
Ohayon (2008) comparing younger and older Hebrew-English
bilinguals who lived in an English-speaking or Hebrew-
speaking society. Their conclusion was that the slower retrieval
time for older bilingual adults in their L1 was caused primarily
by attrition of that language and not by aging. These results
point to the importance of gauging proficiency level, such as
vocabulary knowledge, in linguistic processing and in perfor-
mance on psycholinguistic tasks.
Linguistic differences between monolinguals and bilinguals
go beyond vocabulary size. The consistent result showing
longer picture-naming times for bilinguals suggests that word
retrieval is carried out differently for bilinguals than for
monolinguals. To explore a possible explanation for this effect,
Hernandez and Meschyan (2006) conducted a functional
magnetic resonance imaging (fMRI) study in which Spanish-
English bilinguals who learned the L2 in adolescence named
pictures in both languages. The results showed that naming the
pictures in the weaker second language produced greater
activity in the executive control network, a system that will
be described in more detail in Sections 2 and 3. Extrapolating
to monolingual performance, where naming is always carried
out in a strong language, it appears that this executive control
network is involved in word retrieval for bilinguals in a way not
required by monolingual language production. We will return
to this idea later.
Studies of verbal fluency. The second experimental paradigm
in which reliable differences between monolinguals and
bilinguals have been reported is the verbal fluency task. The
basic procedure is to ask participants to generate as many words
as possible in 60 seconds that satisfy a criterion determined
either by the category (semantic fluency) or the initial letter of
the word (phonological fluency). There are standardized ver-
sions of the task, such as in the Delis-Kaplan Executive Function
Battery (DKEFS; Delis, Kaplan, & Kramer, 2001) and the
Controlled Oral Word Association Test (COWAT; Strauss,
Sherman, & Spreen, 2006), that allow performance to be inter-
preted in terms of normalized tables and used as an instrument
for neuropsychological assessment. The clinical applications of
this test are explained in Section 4, but in the present discussion
we consider the task as an experimental tool. The semantic and
letter versions assess different aspects of competence and engage
different processes. The demands of category fluency are
congruent with normal procedures for word retrieval in that the
meaning is cued and words associated with that meaning are
primed and available. Thus, when asked to generate names of
fruits, the inherent associations among various fruits in semantic
memory facilitate recall. In contrast, the letter fluency condition
imposes an arbitrary criterion on word generation: Conversation
does not normally require the generation of words by virtue of
their initial letter. Moreover, the letter fluency task additionally
imposes a set of restrictions that exclude repetitions of words in
different forms and therefore requires more intensive monitoring
and working memory. Thus, category fluency is strongly indica-
tive of vocabulary size (how many types of fruit can you name?)
and letter fluency requires additional and effortful procedures for
monitoring and controlling attention (how well can you keep
94 Bialystok et al.
track of the words already produced and initiate a new search to
satisfy a different criterion?). Supporting this interpretation of
distinct processes involved in each condition, Grogan, Green,
Ali, Crinion, and Price (2009) related the results of structural
MRI scans of high-proficiency bilinguals to their performance
on category and letter fluency tasks. They found that grey matter
density in a medial frontal region (the presupplementary motor
area) and one subcortical region (the left caudate; see Section
3 for the neural bases of language control) was related to letter
fluency performance whereas higher grey matter density in left
inferior temporal cortex was related to semantic fluency
The typical outcome of studies comparing monolingual and
bilingual adults performing verbal fluency tasks is for bilin-
guals to generate fewer words than monolinguals, with greater
disparity between groups in the category fluency task (Bialystok
et al., 2008a; Gollan, Montoya, & Werner, 2002; Portocarrero
et al., 2007; Rosselli et al., 2000). In a dramatic demonstration,
Linck, Kroll, and Sunderman (2009) reported that English-
speaking college students living in a Spanish-speaking environ-
ment for 1 year produced fewer words on a verbal fluency test in
English than did monolinguals who did not travel abroad!The
scores of the students who had been abroad were restored shortly
after returning home. Moreover, as with picture naming
(Connor, Spiro, Obler, & Albert, 2004), performance in verbal
fluency declines with healthy aging, so this task may be espe-
cially difficult for older bilingual adults (Brickman et al., 2005).
Several possible reasons for the difference in verbal fluency
between monolinguals and bilinguals have been suggested.
First, bilinguals may simply have a smaller overall vocabulary
than monolinguals in each language, a deficit that would partic-
ularly affect the category fluency test. Indeed, it is primarily on
category fluency that lower scores for bilinguals have been
most often observed, with some researchers reporting no differ-
ence between groups in letter fluency (e.g., Rosselli et al.,
2000). Second, as demonstrated in the research on picture
naming, bilinguals take longer to retrieve each item, so the
60-second limit in a verbal fluency trial may curtail bilingual
performance. One possible reason for slower word retrieval
in bilinguals is the need to deal with the competition from the
other language, as stated earlier. Managing this competition
takes time, and this can delay word production for bilinguals and
result in fewer words being generated. Note that both of these
reasons—vocabulary limitations and competition resolution—
apply primarily to category fluency where multiple exemplars
for the given category are activated, including exemplars from
the nontarget language, and much less to letter fluency. In con-
trast, letter fluency relies less on the richness of vocabulary in a
semantic domain and the automatic activation of exemplars in
the other language. Therefore, there is no reason to expect mono-
linguals and bilinguals to perform differently on letter fluency
tasks. In fact, the additional requirements for working memory
and monitoring in the letter fluency condition should actually
favor bilinguals who, as will be explained later in Section 2, are
generally better than monolinguals in tasks requiring working
memory and monitoring.
A more detailed understanding of performance on the verbal
fluency task comes from examining the function showing the
production of words in real time across the 1 minute allotted
to each trial. Following the logic explained by Rohrer, Wixted,
Salmon, and Butters (1995), a deficit in vocabulary size should
manifest itself in a function that shows very few words being
produced toward the end of the time period because the poten-
tial set of items has been exhausted. In this case, monolinguals
would continue producing words later into the time course than
would bilinguals. In contrast, slower time to produce each item,
possibly because of the need to resolve competition from the
nontarget language, would produce a function that continues
longer into the time period than one representing faster retrie-
val of the same total number of words. In this case, bilinguals
would produce words later in the time course than
These predictions were tested in two studies using time-
course analysis to compare monolinguals and bilinguals per-
forming a verbal fluency task. A study by Sandoval, Gollan,
Ferreira, and Salmon (2010) compared monolinguals and
Spanish-English bilinguals who reported high proficiency in
both languages for their performance on several category and
letter fluency conditions in English, and in a second experiment
also compared the time course of retrieval from bilinguals’ two
languages (English vs. Spanish). In another study by Luo, Luk,
and Bialystok (2010), a standardized version of the category
and letter fluency tasks in English was administered to mono-
linguals and bilinguals who were either matched on English
vocabulary or had a lower English receptive vocabulary. In
both studies, the bilinguals produced words later into the
allotted time, indicating slower and more effortful retrieval for
each word produced, likely due to interference from the nontar-
get language (Sandoval et al., 2010). In addition, the compari-
son between the two English proficiency groups in the study by
Luo et al. indicated a second effect attributable to vocabulary
size. Once vocabulary was matched, the bilinguals with Eng-
lish proficiency comparable to that of monolinguals performed
as well as the monolinguals on the category fluency task and
better than monolinguals on letter fluency. Having equated for
differences in vocabulary resources, the bilinguals were able to
display better control than the monolinguals in the condition
that required monitoring and working memory. Figure 2a dis-
plays the results for category fluency in which monolinguals and
high-vocabulary bilinguals show identical retrieval patterns
because performance is driven primarily by vocabulary size,
which in this case is matched. Figure 2b displays the results for
letter fluency; in this case, the high-vocabulary bilinguals main-
tain a higher production rate throughout the time course than do
the other two groups because the task additionally requires high
levels of executive control.
These results point to the need to guarantee that participants
who are performing a language task have linguistic resources
adequate to carry out the task. Without explicitly controlling
for language proficiency, it is impossible to localize the effects
of bilingualism as opposed to the effects of weaker proficiency
in the language of testing. Moreover, when proficiency in the
Bilingual Minds 95
two languages had been controlled by using receptive vocabu-
lary as a matching variable, a bilingual advantage emerged in
the letter fluency task. This pattern was replicated in a
comparison between monolinguals, bilinguals with matched
vocabulary, and bilinguals with lower vocabulary on a simple
behavioral comparison of the number of words produced in
each of these fluency tasks (Bialystok, Craik, & Luk, 2008b).
Clearly, not all tasks requiring processing of linguistic material
are performed more poorly by bilinguals.
Control over linguistic resources. To this point, the studies
described have generally found more effortful (longer response
Fig. 2. Number of items produced as a function of time in (A) category task and (B) letter task
for monolinguals, high-vocabulary (HV) bilinguals, and low-vocabulary (LV) bilinguals. Best fit
lines are logarithmic functions. From Luo, Luk, and Bialystok (2010).
96 Bialystok et al.
time, RT) or poorer (more errors) performance by bilinguals
than by monolinguals when rapid retrieval of specific lexical
items is required. When language proficiency is matched,
however, bilinguals perform as well as monolinguals in
category fluency (which depends on vocabulary) and better
than monolinguals on letter fluency (which depends more
extensively on cognitive control). Therefore, at least some of
the differences observed between monolinguals and bilinguals
on language production tasks reflect a simple difference
in linguistic resources and may mask a potential advantage in
control over those resources once proficiency has been equated.
If bilinguals do have better control over linguistic resources
than do monolinguals, then it should be possible to demonstrate
this processing difference in tasks that require monitoring or
manipulation of verbal stimuli. Two tasks meet these criteria.
The first is a paradigm developed by Jacoby (1991), called the
process dissociation procedure (PDP), that is designed to
distinguish between automatic (familiarity) and controlled
(recollection) aspects of memory. The second is a paradigm
called release from proactive inhibition (PI) that assesses the
ability to monitor items for their source (e.g., Kane & Engle,
2000). Both paradigms have been widely used in studies of cog-
nitive processes involved in memory performance. Although
substantially different from each other, they share the feature
that participants are asked to remember words for later recall
when an intervening event has made it difficult to keep track
of the source of the target words. In the case of PDP, words are
presented in two lists, or two formats (for example, visually or
orally), and the crucial recall test requires responding only to the
words presented in one of them (for example, visually) and
ignoring the others. In the case of PI, lists of different words
from the same semantic category are presented successively and
participants are asked to report the words on the list just heard
without reporting words from the previous lists. ‘‘Release’ from
PI is observed when words from a different category are pre-
sented. Both tasks, therefore, require monitoring and control to
attend to the target words and inhibition to avoid making errors
on the distractor words. As predicted, bilinguals obtained lower
scores than monolinguals on tests of receptive vocabulary but
performed better than monolinguals on both PDP (Wodniecka,
Craik, Luo, & Bialystok, 2010) and release from PI tasks
(Bialystok & Feng, 2009). Again, separating verbal ability
from control over verbal processing produces a more complex
picture in which bilinguals demonstrate better processing in
the context of poorer verbal performance.
Cognitive control and bilingual language
All the illustrations of language acquisition and use described
in this section have demonstrated the importance of considering
the interaction between language and cognitive systems in
explaining outcomes for bilinguals. Bilingual children acquire
language on the same timetable as monolingual children, largely
because this timetable is determined by the process of cognitive
development. As acquisition proceeds, however, bilingual
children develop different types of competence (e.g., smaller
vocabulary in each language) and probably use different
strategies (e.g., phonemic cues and mutual exclusivity for word
learning). In adulthood, the ability of bilinguals to effectively
use language in such tasks as word retrieval and word generation
depends on both linguistic competence and cognitive proce-
dures for access and monitoring. Thus, levels of vocabulary
determine how many words can be associated with a meaningful
category but levels of control determine how many words can be
selected to fit an arbitrary restrictive criterion.
What is the source of these interactions? One possibility is
that the interacting systems are set in motion because the joint
activation of the two languages for a bilingual creates a problem
not experienced by monolinguals—namely, the need to select
from the target system in the context of compelling and active
alternatives. There is substantial evidence, described in Sections
2 and 3, that the response to this conflict is to recruit the execu-
tive control system that has evolved to resolve conflict across all
domains of perceptual and cognitive processing. The constant
use of this executive control system for bilingual language man-
agement opens the possibility that the system itself is modified,
changing its valence or efficiency for all tasks. That is, the use of
a set of executive control procedures to manage attention to lan-
guage, to avoid interference from the nontarget language, and to
monitor two simultaneously active languages may alter the
nature or efficiency of those executive control processes more
generally. This possibility is examined in the next section. To
anticipate, the evidence suggests that whereas bilingual children
and adults have somewhat lower vocabulary levels than their
monolingual counterparts, the bilinguals possess an advantage
in cognitive control that generalizes beyond language processing
to other aspects of cognitive functioning.
2. How Bilingualism Affects Cognitive
For many years it was assumed that while bilingualism might
be an asset for adults—in terms of culture, travel, and trade,
for example—it was a handicap for children in the educational
system. The idea was that learning in two languages imposed
an additional burden on schoolchildren who must learn two
vocabularies, two sets of grammar, and probably two sets of
cultural habits and expectations. This negative view of bilin-
gualism was at least questioned by the results of a study by
Peal and Lambert (1962). They gave a battery of intelligence
fluent English speakers. They expected to find that monolin-
gual and bilingual children would be equivalent on measures
of nonverbal intelligence but that bilinguals would obtain
lower scores on verbal measures. To their surprise, however,
bilingual children outperformed their monolingual peers on
virtually all of the tests, including tests of nonverbal intelli-
gence. Further analysis revealed that there was little differ-
ence between the groups on spatial-perceptual tests but that
the bilingual children showed an advantage on tests requiring
symbol manipulation and reorganization. This latter finding
Bilingual Minds 97
has the interesting implication that extra effort and more
extensive learning in the area of language apparently confers
benefits to nonverbal mental abilities, refuting the idea that
language is a separate module of mind and brain that relies
on dedicated processes (e.g., Fodor, 1983); instead, language
must be viewed as recruiting processes from the general cogni-
tive system. On the basis of their unexpected findings, Peal and
Lambert suggested that bilingual children may show enhanced
mental flexibility, perhaps as a consequence of having to
switch between their two languages.
The study by Peal and Lambert (1962) may be criticized on
the grounds that francophone children in Montreal in 1960 who
spoke English were likely of higher than average social class,
or at least were the children of intelligent and ambitious
parents, and were therefore less representative than their mono-
lingual counterparts (Bialystok, 2001). Nevertheless, the study
was important in showing both that bilingualism in children
might help rather than hinder the development of other abilities
and also that language learning may influence nonverbal
cognitive processes supporting the view that language is not
a separate and independent module of mind.
Some decades following the Peal and Lambert study, support-
ing evidence for a bilingual advantage in general cognitive func-
tioning for children was found in studies using a variety of
experimental paradigms. For example, Bialystok (1992) reported
that bilingual children performed better than their monolingual
counterparts on the Embedded Figures Test. In this test, partici-
pants must find a simple visual pattern concealed in a larger com-
plex figure. Bialystok suggested that the better performance of
bilingual children might reflect their superior ability to focus
on wanted information and ignore misleading information. That
is, the advantage might be one of enhanced selective attention,
involving the ability to inhibit irrelevant or unwanted information
and the complementary ability to concentrate on relevant aspects.
This interpretation was in line with another demonstration in
which children were asked to judge whether phrases were gram-
matically correct, regardless of meaning. Bilingual children were
better than their monolingual age-mates at ignoring the mislead-
ing meaning in sentences such as ‘‘Apples grow on noses’’ or
‘Why is the cat barking so loudly?’’ and stating that the grammar
was correct (Bialystok, 1988). More generally, research
demonstrated enhanced metalinguistic awareness in bilingual
children compared to their monolingual peers (Ben-Zeev, 1977;
Cummins, 1978; Galambos & Hakuta, 1988; Ricciardelli, 1992)
Why might bilingual children show an advantage in the
ability to inhibit attending to unwanted information and select
relevant aspects? The answer may follow from the surprising
finding described earlier: that when bilingual speakers use one
language, the other language is still active. However, this does
not mean that a full analysis of incoming stimuli in the nonused
language inevitably takes place, nor that formulating speech in
one language fully activates the relevant words and grammar of
the other language. It seems rather that the second language is
potentially active, that some analysis is typically carried out,
and that more analysis takes place when combinations of
context and meaning increase the likelihood that words and
phrases from the nonused language are in fact relevant to the
speaker’s or listener’s concerns.
The idea that the nonrelevant language is always potentially
active accounts for another observation on bilingual speakers:
that they occasionally intrude words from the alternate
language during speech. Though such intrusions are rare
(Poulisse, 1997; Poulisse & Bongaerts, 1994; Sandoval et al.,
2010), these instances reflect occasions in which the appropriate
word in the language being used is difficult to locate or the word
or phrase in the nonused language is made particularly likely
because of the context or its salience. Bialystok (2001) commen-
ted that such intrusions are more common in bilingual children
than in adults and are also more common (anecdotally at least)
in older than in younger adults (Sandoval, 2010). In turn, this
age-related pattern suggests that the brain mechanisms responsi-
ble for maintaining attentional set (in this case maintaining atten-
tion on the selected language) are less effective in childhood and
in older adulthood. One candidate for such mechanisms is integ-
rity of frontal lobe functioning, since it is well established that
the frontal lobes develop slowly in childhood and are among the
first parts of the brain to decline in efficiency in older adulthood
(Craik & Grady, 2002; Diamond, 2002; Raz, 2000).
Our suggestion is that bilingual speakers must develop an
unusually strong ability to temporarily inhibit access to the
nonrelevant language while maintaining attentional set
(‘‘maintaining concentration’’) on the language in current use.
This ability may be mediated by the frontal lobes and may
therefore exhibit a lifespan developmental trend that peaks in
young adulthood. The further suggestion is that the constant
necessity to exercise this inhibitory control leads to the
development of particularly effective attentional functions that
are then drawn on to mediate good performance on a variety of
nonverbal tasks requiring inhibition of unwanted or misleading
material and concurrent selection of relevant aspects.
Inhibition or selection?
What would it mean to have enhanced control over attentional
functions? When Bialystok (2001) surveyed studies of the
effects of bilingualism on children’s cognitive processes,
she concluded that ‘‘the most consistent empirical finding
about the cognition of bilingual children is their advantage in
selective attention and inhibition’’ (Bialystok, 2001, p. 246).
This conclusion was based on some of her own work (e.g.,
Bialystok, 1988, 1992) as well as on a growing number of
studies from other laboratories. An example that illustrates how
these processes are used by children is the dimensional change
card sort task (DCCS) developed by Zelazo, Frye, and Rapus
(1996). This is a game in which images that vary on two dimen-
sions, usually shape and color, are sorted according to one of
them. For example, cards containing either red or blue circles
or squares are sorted into containers marked by an image of
either a red square or a blue circle. Children are asked to first
sort the cards by one dimension—blues in this box and reds
in this box—and then to switch to the other—circles in this box
and squares in this box. The dramatic finding is that young
98 Bialystok et al.
children can easily state the new rule but continue to sort by the
first rule; they have great difficulty overriding the habit set up
in the first phase. When this experiment was repeated with
bilingual and monolingual children aged between 4 and 5 years,
the bilingual children were markedly better at switching to the
new rule (Bialystok, 1999; Bialystok & Martin, 2004). This
result was obtained despite there being no difference in
pre-switch performance. The researchers thus concluded that
the constant need to inhibit the nonused language generalized
to more effective inhibition of nonverbal information.
These demonstrations were followed by studies that
extended the investigation to adults and used other paradigms
in which a prepotent response tendency must be inhibited. One
such situation is embodied in the Simon task. The participant
views a screen on which either a red or green square appears;
there are two response keys, one for red squares and the other
for green squares. The keys are positioned below the sides of
the screen, and the squares can appear either immediately
above their relevant response key (congruent condition) or
above the other key (incongruent condition). Response laten-
cies are longer in the incongruent case, and the difference
between incongruent and congruent latencies is termed the
Simon effect. If participants are able to resist the misleading
information carried by spatial position in the incongruent situ-
ation, the Simon effect will be smaller, and we may conclude
that they have well-developed inhibitory control mechanisms.
Using this logic, Bialystok, Craik, Klein, and Viswanathan
(2004) tested groups of younger and older adults who were
either monolingual or bilingual on a version of the Simon task.
When the colored squares are presented centrally, there is no
conflict between the position of the stimulus and side of the
appropriate response, and in this case there were no differences
in reaction time between monolinguals and bilinguals, although
older participants took longer to respond (Fig. 3a). When the
colored squares appeared laterally, however, Simon effects
were found, and these were larger for monolinguals—especially
older monolinguals (Fig. 3b). This evidence for a bilingual advan-
tage in inhibitory control in adults extended the results of previous
studies on children. Moreover, the bilingual advantage was
especially strong in older adults, suggesting that bilingualism
may afford some protection against at least some forms of
cognitive aging.
Two other unexpected results emerged from this study. The
first is that the bilingual advantage in response time was found
for congruent as well as incongruent stimuli. This result was
obtained in all three experiments and has been consistently
observed in subsequent studies (e.g., Costa, Herna´ndez, &
Sebastian-Galles, 2008). Why should there be a bilingual
advantage for congruent stimuli when there is no misleading
information to inhibit? Most experiments of this sort are run
under mixed conditions in that experimental runs contain both
congruent and incongruent stimuli, so participants must keep
the rule in mind throughout the experimental run and monitor
each trial for the type of processing needed (conflict or no
conflict). It may be that bilinguals are also better at these
aspects of executive control. The test of this conjecture is to
check what happens in experiments containing pure runs of all
congruent or all incongruent stimuli, and the finding there is
that the bilingual advantage disappears (Bialystok, Craik, &
Ryan, 2006).
The second unexpected result found by Bialystok et al.
(2004) was that prolonged practice on the Simon task reduced
the difference between monolinguals and bilinguals.
In Experiment 3, participants performed the Simon task for
10 consecutive blocks of 24 trials; by the end of the session the
monolingual disadvantage had disappeared and both groups
showed minimal differences between congruent and incongru-
ent stimuli. It is interesting to speculate that everyone may be
able to inhibit the effects of misleading information in specific
situations with sufficient practice but that bilinguals can learn
this type of inhibition more rapidly.
The Stroop effect may be considered the ‘‘gold standard’’ of
tests of inhibition. In this paradigm, participants name colors as
rapidly as possible, both when the colors are presented as
colored squares on a screen and when the stimuli are color
names (e.g., ‘‘red,’’ ‘‘green,’’ ‘‘blue’’) but presented in a differ-
ent colored font (e.g., the word ‘‘red’’ printed in green ink). The
difference in speed between naming colored squares and the
color of words is the Stroop effect; again, a smaller Stroop
effect indicates a strong ability to inhibit the misleading
tendency to name the word rather than its color. Bialystok
et al. (2008a) tested groups of 24 younger and older adults who
were monolingual or bilingual on this paradigm. In four differ-
ent conditions, participants named the color of displays of Xs,
named a color word presented in black font, named the font
color of words printed in their own color (congruent condition),
and named the font color of words printed in a different color
(the incongruent Stroop condition). For the control conditions
(naming words and colored Xs), naming times were faster for
words and for younger participants but there were no
language-group differences. Response times for the congruent
and incongruent colored-word conditions are shown in Figure 4
as differences (positive or negative) from the time taken to
name colored Xs. The figure shows that congruent stimuli are
associated with relatively faster response times (a facilitation
effect) and are indicated by positive RT differences in the
figure, whereas incongruent stimuli show the classic Stroop
pattern in which slower response times are indicated by nega-
tive RT differences. Statistical analysis revealed a significant
three-way interaction of age, language, and congruence; both
younger and older bilinguals sustained smaller costs than their
monolingual peers, but only the older bilinguals showed greater
facilitation. We may thus conclude that the older bilinguals
exhibited greater degrees of cognitive control than their mono-
lingual counterparts, in that they both took greater advantage of
congruent conditions and at the same time were less impaired
by incongruent conditions. Younger bilinguals showed the
latter effect but not the former.
Other results from the Bialystok et al. (2008a) study
included a bilingual advantage for the older participants in a
version of the Simon task using directional arrows, but no bilin-
gual advantage for either age group in a condition in which
Bilingual Minds 99
participants were instructed to respond in the direction opposite
to that indicated by a single arrow. There was also no bilingual
advantage on the Sustained Attention to Response Task
(SART; Robertson, Manly, Andrade, Baddeley, & Yiend,
1997), which involves withholding a response to the number 3
while responding rapidly to all other digits. In both these latter
tasks, the participant can encode a simple rule (e.g., ‘press in the
opposite direction’’) and then follow that rule; there is essentially
no need to select one aspect of the stimulus and suppress other
aspects, as with the Simon, Stroop or flanker tasks. This account
claiming no need for control in these tasks is reinforced by other
results showing no bilingual advantage in children who were
instructed to respond ‘‘day’’ when shown a picture of a dark
night, and ‘‘night’’ when shown a sunny day (Martin-Rhee &
Bialystok, 2008). These investigators also replicated the finding
of no bilingual advantage in children given the reverse arrow
task, even though the same children demonstrated a bilingual
advantage when the arrows were placed in side positions on the
display that created conflict.
This pattern of presence and absence of advantages is in line
with the distinction between interference suppression and
1500 (a) Control Condition
(b) Simon Effect
30-39 40-49 50-59 60-69 70-79
Mean RT (ms)
30-39 40-49 50-59 60-69 70-79
Monolingual Bilingual
RT Difference (ms)
Fig. 3. Mean reaction time (RT) on Simon task by decade for monolinguals and bilinguals.
Graph a shows mean RT for the control condition; Graph b shows mean RT cost as the
difference between congruent and incongruent trials (Simon effect). From Bialystok, Craik,
Klein, and Viswanathan (2004).
Type of Difference Score
Mean RT (ms)
Facilitation Cost
You n g Mo no
Young Biling
Old Mono
Old Biling
Fig. 4. Mean reaction time (RT) and standard error for
facilitation and cost for young monolinguals and bilinguals
and older monolinguals and bilinguals in the Stroop task.
The values are mean differences from baseline (0 millise-
conds) calculated as the average difference in the time
taken to name colors from the time taken to name neutral
stimuli (Xs). From Bialystok, Craik, and Luk (2008a).
100 Bialystok et al.
response inhibition proposed by Bunge, Hazeltine, Scanlon,
Rosen, and Gabrieli (2002). Interference suppression refers to
situations in which misleading information evokes a faulty
response and must therefore be ignored or suppressed; this
appears to be the type of situation that bilinguals can deal with
particularly well. Response inhibition is the ability to avoid
responding in error to a habitual or highly salient cue, and bilin-
guals show no advantage under these circumstances. In other
words, the bilingual advantage appears when there is conflict
between two potential responses, but not when there is a need
to withhold a single primed response.
As a final converging point, Kimberg, D’Esposito, and
Farah (1997) have commented that patients with lesions in the
prefrontal cortex are impaired on tasks in which the most
salient cue evokes the wrong response and must therefore be
suppressed to select the cue associated with the correct
response. If one effect of bilingualism is to boost frontal lobe
functions, it follows that bilingual children and adults should
be adept at tasks involving interference suppression.
Converging evidence from other studies has supported the
conclusion that bilinguals show strong abilities to inhibit irre-
levant or interfering information. Zied and colleagues (2004)
found that balanced bilingual adults of various ages responded
more rapidly than unbalanced bilinguals on the Stroop task. In
an ingenious series of studies, Philipp and colleagues (Philipp,
Gade, & Koch, 2007; Philipp & Koch, 2009) asked participants
who were fluent in three languages (English, French, &
German) to switch among their languages in a number naming
task; thus ‘‘2’’ was named either ‘‘two,’’ ‘‘deux,’’ or ‘‘zwei’
depending on a concurrent instruction. The main finding was
that naming in language A was slower on the third trial of a
sequence ABA than in a sequence CBA. That is, A (e.g.,
French naming) was slower on the third trial of a sequence
French, German, French than it was in the third trial of a
sequence English, German, French, suggesting that in the first
sequence French was subjected to a temporary global inhibi-
tory effect to permit access to German. When French was
needed immediately after that, negative priming slowed access
to the target name. Although there is no bilingual advantage in
this study—monolinguals were not tested, and the study was
not designed to test for bilingual advantages—the results
demonstrate the role of a general inhibitory process applied
to the nonused language in order to avoid interference effects
in the selected language.
Negative priming was also used in an experiment by
Treccani, Argyri, Sorace, and Della Sala (2009). Targets could
appear at one of four positions on a screen and participants
responded by pressing one of four keys. When a target was
accompanied by a distractor stimulus in another location,
bilingual adults were better able to ignore it (interference
suppression) and so made fewer errors than did their monolingual
counterparts. However, the bilinguals were more negatively
affected (making more errors than monolinguals) when a target
appeared in the position previously occupied by a distractor item.
In this situation, the better inhibition of the distractor carried over
to the next trial, providing more negative priming to the bilingual
participants. The authors concluded that whether bilinguals show
an advantage or a disadvantage relative to monolinguals depends
on task characteristics.
The studies reviewed so far have endorsed the notion that the
bilingual advantage found in these studies is due to an advantage
in inhibition or suppression of interfering material, but there
remains the possibility that bilinguals show an advantage in the
positive selection of wanted information. The latter interpreta-
tion is favored by a number of investigators. Costa, Miozzo, and
Caramazza (1999) argue that although lexical candidates in both
languages are active during the planning of an utterance, the
intention to speak in one language rather than another effectively
restricts selection to words in the target language. Colzato and
colleagues (Colzato et al., 2008) set out to compare what they
termed ‘‘active inhibition’’ with ‘‘reactive inhibition.’’ By active
inhibition they mean general global suppression of the nonrele-
vant language (cf. inhibition in the study by Philipp & Koch,
2009) and by reactive inhibition they mean lack of suppression
of specific interfering stimuli. Evidence for the latter was found
in the attentional blink paradigm in which detection of a target
stimulus is impaired if the same stimulus was presented earlier
in a rapid sequence of events. The authors predicted that if bilin-
guals show more reactive inhibition, then they will process the
first presentation of the target to a greater extent and therefore
show less suppression of intervening items. Without suppres-
sion, these items would then interfere more with the second
presentation of the target, creating a larger attentional blink
effect. This is what they found, and so they suggested that the
bilingual advantage is not due to constant exercise of inhibition
of the nonused language but rather to prolonged practice at
maintaining the relevant attentional set, though they grant that
such selection may involve strong inhibition of competing items.
The debate over inhibition versus selection may rest on a
false dichotomy: Inhibition may not be an all-or-none phenom-
enon but may rather be found to different degrees under some
circumstances. One such factor that might influence the degree
of inhibition required to perform a task is the effect of context.
Kroll, Bobb, Misra, and Guo (2008) describe work showing
that cross-language cognates were activated (that is, naming
a word in one language activated its cognate in bilinguals’
second language) when a word was named out of context, but
this cognate facilitation was eliminated in contexts that were
semantically constrained in that the required word was more
clearly determined from the context (see also Schwartz &
Kroll, 2006; van Hell & de Groot, 2008). One possibility, then,
is that the degree to which both languages are active may not be
constant but may vary probabilistically with the contextual
constraints provided by language, topic, and the external
Another possible effect of context was suggested by Costa,
Herna´ndez, Costa-Faidella, and Sebastian-Galles (2009). They
tested monolinguals and bilinguals on versions of a flanker task
in which different conditions contained varying proportions of
incongruent trials: 8%,25%,50%,or92%(therefore mixed with
92%,75%,50%,or8%congruent trials, respectively). The bilin-
gual advantage was strongly present in the 50%/50%version,
Bilingual Minds 101
reduced in the 75%/25%version, and entirely absent in the 92%/
8%version. The authors conclude that the bilingual advantage is
related to their greater ability to monitor the environment when
the probability of change is high, as in the 50%/50%condition.
Under low-monitoring conditions, when most of the trials are of
one type, there is little need to monitor and thus no bilingual
advantage is found. The notion of monitoring is similar to the
idea of set maintenance described previously by Colzato et al.
(2008). Costa and his colleagues also make the interesting pre-
diction that bilinguals who live in situations in which their two
languages are used in different contexts (e.g., Italian at home,
English at work) rarely need to monitor language changes and
so may not develop strong monitoring abilities and thus show
no bilingual advantage.
Finally, the distinction between selection and inhibition was
examined in a study by Herna´ndez, Costa, Fuentes, Vivas, and
Sebastian-Galle´s (2010), in which participants rapidly judged
how many items (letters or numerals) appeared on a screen. The
items appeared either in a congruent form (1, 22, 333), an incon-
gruent form in which the displayed numerals did not match the
required response (e.g. 3, 11, 222), or a neutral form (Z, GGG,
MM). Relative to the neutral baseline, congruent stimuli were
associated with faster response times (facilitation) and incongru-
ent stimuli with slower response times (interference). Bilingual
participants showed smaller interference effects but larger facil-
itation effects than their monolingual counterparts (cf. Bialystok,
Craik, & Luk, 2008a), so their advantage may be described as
one of better executive control of perception/action processing.
The conclusion of Costa and colleagues is that the bilingual
advantage is reasonably high level, involving top-down working
memory processes, and is manifested as enhanced set mainte-
nance or monitoring. This description suggests that the advan-
tage may stem from enhanced frontal lobe effectiveness, as
suggested by Bialystok (2001).
Selective attention and executive control
We have seen in the previous section that research aimed at
assessing inhibitory abilities in bilinguals evolved to consider
such concepts as selection, set maintenance, and monitoring.
However, the distinction between these concepts and notions
of attention and executive control is difficult to discern.
In many ways, all these concepts are simply aspects of attention
and executive control. Therefore, in this section we consider
work that assesses group differences in attention and control
more directly.
Costa et al., (2008) examined the performance of monolin-
gual and bilingual participants on the attentional network task
(ANT) developed by Fan, McCandliss, Sommer, Raz, and
Posner (2002). The bilinguals were young adults who spoke
Catalan and Spanish; the monolinguals were young adults who
spoke Spanish only. The ANT task assesses abilities on three
different attentional networks: alerting, orienting, and execu-
tive control. The test is a flanker task in which the participant
responds to the direction of a central arrow that is flanked by
two arrows on each side pointing in the same (congruent) or
different (incongruent) direction as the central target arrow.
Alerting is studied by presenting a cue before the target stimu-
lus, and orienting is assessed by the presence or absence of a
cue signaling the future spatial position of the target. The
results supported the hypothesis of greater attentional control
by bilinguals in the alerting and executive control networks.
The bilingual participants responded faster than the monolinguals
on all conditions and showed a smaller cost for the incongruent
trials, indicating better conflict resolution. Two final results from
this study were that this bilingual advantage disappeared by the
third block of trials (cf. Bialystok et al., 2004, Study 3) and that
bilinguals had smaller switching costs between congruent and
incongruent trials, a point to which we will return.
Similar results were obtained by Carlson and Meltzoff
(2008) with much younger participants. They administered a
battery of executive function tests to 50 kindergarten children
who were English-speaking monolinguals, English-Spanish
bilinguals, or children who were in a language immersion ele-
mentary school. The major finding was that the native bilingual
children performed better on the executive function battery
than did both other groups, once differences in age, vocabulary,
and parents’ education and income levels were statistically
controlled (recent work extends this finding that bilingualism
can offset the negative effects of lower socioeconomic status
on task switching to young adults; Prior & Gollan, 2010).
The effects were specific to only some aspects of control: There
were no bilingual advantages in suppressing a motor response
on delay-of-gratification tasks (response inhibition) but
significant advantages on conditions requiring memory and
inhibition of attention to a prepotent response (interference
suppression; cf. Martin-Rhee & Bialystok, 2008). The authors
conclude by endorsing the notion that ‘‘language experiences
can influence further development of frontal lobe functions such
as inhibition and the control of attention’’ (p. 293).
Task switching
The features of executive control discussed to this point
are somewhat invisible in ordinary cognitive performance.
The interference suppression that allows us to perform a Stroop
task or ignore misleading flankers in the ANT seems to have
little role in everyday cognition. A more noticeable aspect of
executive control might be task switching—the ability to move
easily between two tasks, keeping two protocols simultane-
ously active. Task switching might come closest to the special
processes bilinguals engage in as they switch between
In one of the first studies to find positive things to say about
bilingualism, Peal and Lambert (1962) suggested, as we noted
earlier, that bilingual children may show an advantage in men-
tal flexibility—an idea presumably stemming from the fact that
bilinguals must switch easily from one language to another. A
large body of research investigates task switching, typically by
asking participants to classify a long series of two-dimensional
stimuli by one criterion or the other as rapidly as possible. Such
sorting times are relatively short when successive trials
102 Bialystok et al.
continue with the same criterion (e.g., continue sorting by
shape), but local switching costs are incurred when instructions
change to sort by the other dimension (e.g., switch and sort by
color). Some runs of trials involve only one dimension (e.g., all
trials require sorting by color), so it is also possible to measure
mixing costs, defined as the difference in time taken to classify
a set of trials under single- and dual-criterion conditions
(Meiran & Gotler, 2001; Pashler, 2000). Typically, sorting
times are longer when it is necessary to bear in mind the
requirement to switch when the instruction changes.
Several studies have now explored monolingual—bilingual
differences in such paradigms, with the prediction that
bilinguals should show reduced costs, owing perhaps to their
prolonged practice in switching languages and monitoring
which language may be spoken in which context. The
prediction with regard to which type of cost might be affected
by bilingualism is less clear. To take an analogous
difference between individuals—development and aging over
the lifespan—the typical finding is that younger adults have
smaller mixing costs than children or older adults do, whereas
the age groups do not differ markedly on local switch cost
(Reimers & Maylor, 2005; for review, see Mayr & Liebscher,
2001). The relatively large value for mixing costs in young
children and older adults was speculatively attributed to their
greater difficulty in simultaneously maintaining two task sets.
Given bilinguals’ apparent advantage in maintaining task set
(Colzato et al., 2008), it should follow that they should also
show reduced mixing costs. This result was indeed reported
by Bialystok et al. (2006) in an experiment in which partici-
pants needed to respond on the same or opposite side as a target
depending on a cue. Participants performed single-task runs in
which only one cue was used and mixed runs in which either
cue might appear. Response times to the target were slower
under mixed conditions, and mixing costs were greater for
monolingual participants.
Three other task-switching studies investigating monolin-
gual and bilingual college students have yielded mixed results.
First, Prior and MacWhinney (2010) asked participants to clas-
sify stimuli by color (red/green) or shape (circle/triangle). They
found no mixing-cost advantage to bilinguals and no speed
differences between the two groups on non-switch trials, but
the bilinguals were faster than monolinguals on switch trials
when instructions changed to sort on the alternate dimension.
Thus, their study found a local switch-cost advantage to bilin-
guals with no mixing-cost advantage. Subsequent experiments
replicated the switching advantage in bilinguals who reported
that they frequently switched languages and no switching advan-
tage in a less balanced group, although this less-balanced group
exhibited significant associations between fluency in a nondomi-
nant language and switching and mixing costs (Prior & Gollan,
2010). These results suggest dissociations of switching and
mixing costs with respect to group differences and imply that
multiple aspects of bilingualism may influence task shifting.
Frequent language switching may lead to task-switching advan-
tages, whereas close monitoring of which language may be
spoken when (and avoiding switching) may lead to task-
mixing advantages. A third study provides clues with respect
to the origin of the mixing advantage. In this study, Herna´ndez,
Martin, Barcelo, and Costa (2010) also used a color–shape
switching task to test young adult Spanish-Catalan bilinguals
and Spanish-speaking monolinguals. A rule was set at the begin-
ning of a run (e.g., classify by shape), then trials continued for an
unpredictable number without further cues until a second cue
was presented. The second cue was either explicit (e.g.,
classify by color) or implicit (e.g., switch to the other rule or
repeat the previous rule). It was found that switching was slower
than repeating the same criterion but that this effect did not
interact with group. Implicit cues were associated with slower
response times than were explicit cues, and this effect did interact
with language group; bilinguals were faster in the implicit version
but not in the explicit version. The researchers also measured
‘restart costs’’—slower RTs for the first trial than for the second
trial after a repeat cue. Bilinguals had smaller costs than monolin-
guals on this measure too,but again only with implicit cues. These
results suggest that the bilingual participants were better at main-
taining the current set, monitoring the changing situation, and
updating when necessary. Although the task was similar in many
respects to that used by Prior and MacWhinney, the instructions
were presented differently, and the bilinguals’ use of two very
similar languages might account for the differences in results.
If that is the case, one would need to be cautious about generaliz-
ing about differences in local and global task switching between
monolinguals and bilinguals without considering further details
of the participants and task situation.
There are still too few studies to conclude much that is
definitive on the effect of bilingualism on task switching.
Better bilingual performance for mixing costs (Bialystok
et al., 2006) and dealing with implicit cues (Herna´ndez et al.,
2010) suggests that the advantage is in monitoring or set main-
tenance, but the results of the Prior and MacWhinney (2010)
study speak more to the notion of greater mental flexibility
or greater inhibitory control. In addition, bilingual language use
may require different underlying control processes and may
therefore lead to different processing advantages (Prior &
Gollan, 2010). The few current studies involve many differ-
ences in methods and in participants, so the traditional cry of
‘more research is needed!’’ is very much the case before
decisive conclusions can be drawn.
Bilingualism and memory
Since being bilingual necessarily entails the management and
appropriate development of two language systems, it makes
sense that these special skills of mental management should
also apply to aspects of attention, conflict resolution, and
cognitive control. But should bilingualism confer benefits on
other cognitive functions—on memory, for example? The
answer may depend substantially on the type of memory being
investigated. Working memory (the manipulation of small
amounts of material held briefly in mind) is generally considered
to be either part of, or closely related to, executive processes, so
bilingual advantages might be expected with such paradigms.
Bilingual Minds 103
However, performance on semantic memory tasks (tappingstores
of acquired knowledge) is likely to reflect experience with the
type of information tested. Given that we have seen that bilingual
vocabulary levels are typically lower than those of comparable
monolinguals, we might expect that retrieval of verbal informa-
tion would be poorer in bilingual participants, and, as described
in the first section, performance on naming tasks and other tasks
of lexical retrieval do in fact show this pattern. Moreover, perfor-
mance on episodic memory tasks may again depend on the mate-
rial in question.
For both working memory and episodic memory, the
evidence is mixed. In one condition of the Simon task reported
by Bialystok et al. (2004), color patches were presented cen-
trally and so required no cognitive control, and participants
responded to the color by pressing one of two response keys.
In one version, two possible colors mapped to the two keys, and
in the second version, four possible colors mapped to the two
keys, with two colors associated with each key. The four-color
version has greater demands on working memory, so working
memory costs were taken as the difference between the
two-color and the four-color versions. Bilingual participants
aged 30 to 80 years showed smaller costs than did their monolin-
gual counterparts, and were therefore deemed to show a bilin-
gual advantage in working memory. This advantage has
obvious similarities to the bilingual advantage in mixing costs
found in some studies using the task-switching paradigm.
The results of other studies are less clear. Bialystok, Craik,
and Luk (2008a) gave older and younger adult bilinguals and
monolinguals two tests of working memory. The self-ordered
pointing task requires participants to remember which of 12
abstract drawings have been selected previously; no
language-group differences were found. The Corsi Block task
is a test of short-term spatial memory, and in this case there was
a bilingual advantage for younger but not older adults. Feng
(2008) also presented various working memory tasks to mono-
lingual and bilingual children and young adults. In the latter
group, she found no bilingual advantage in either the Corsi
Block task or in alpha span—a word-span task in which parti-
cipants must mentally rearrange a short list of words from a
presented order into alphabetic order. However, Feng did find
a bilingual advantage for both children (Feng, Diamond, &
Bialystok, 2007) and adults (Feng, 2008) in a test of spatial
working memory in which items are presented in a random
order in a 3 3 matrix (for children) or on a 5 5 matrix (for
adults). The task is to recall the positions of the items in
‘matrix order’’—that is, starting at the top left and progressing
through the matrix left to right, line by line.
Whether or not there is a bilingual advantage in working
memory may depend on the type of material used and the way
in which working memory is tested. Working memory tasks
may not be tapping one fixed cognitive mechanism but rather
reflect a family of related functions generally concerned with
holding and manipulating material that is in the focus of atten-
tion (Cowan, 1999) or simply ‘‘held in mind.’’ Tentatively, it
seems to us that a bilingual advantage should be found in work-
ing memory, given the previously reviewed evidence
suggesting that bilinguals have an advantage in set mainte-
nance (e.g., Colzato et al., 2008) and in the related abilities
of monitoring (Costa et al., 2009) and updating (Herna´ndez
et al., 2010).
The effects of bilingualism on episodic memory are also
unclear at present, as only a few studies have been reported.
In the studies described earlier, Fernandes, Craik, Bialystok,
and Kreuger (2007) found poorer word recall by bilinguals, but
Wodniecka et al. (2010) reported that the disadvantage was
overcome when monitoring the list was required, as in the
assessment of recollection. At present, therefore, there is little
clear evidence for a bilingual advantage in episodic memory,
some tentative suggestions of an advantage in working mem-
ory, and a clear disadvantage for bilinguals in the retrieval of
items from semantic memory.
The bilingual advantage across the lifespan
Does the bilingual advantage in cognitive control change
through the lifespan? It is well established that executive con-
trol functions first increase in effectiveness from childhood to
young adulthood and then decline in the course of aging (Craik
& Bialystok, 2006; Dempster, 1992; Diamond, 2002), so it
seems possible that bilingualism might modify such functions
and that the bilingual advantage might also show the same
lifespan trajectory.
If the bilingual advantage in cognitive performance we have
seen in this section is related to the enhancement of the
executive control function, how early might we expect these
differences to emerge given that the executive function system
is late to develop? Similarly, if the cognitive advantage
depends on protracted experience with two languages in which
attention to systems and switching between them becomes
practiced, could such advantages be found in children before
they use language productively? A recent study by Kovacs and
Mehler (2009b) provides dramatic evidence for the very early
appearance of a bilingual advantage in 7-month-old infants.
The infants who participated in the experiments were preverbal
but were classified as bilingual if they had been exposed to two
languages from birth because one parent consistently spoke to
them in one language and the other parent used a different
language. The researchers reported three experiments in which
the infants learned to look for a visually rewarding puppet at
one of two squares on a screen in response to either a speech
stimulus (a trisyllabic nonsense word) or a visual pattern. After
the learning phase, which was performed equally well by
monolingual and bilingual infants, a new cue signaled the
appearance of the visual reward in the alternate square. Thus,
infants had to inhibit their first learned response and switch
to a new response. The finding in all three experiments was that
the bilingual infants learned to switch to the other square but
the monolingual infants did not. The authors suggest that sim-
ply perceiving and processing utterances from the two lan-
guages during the first few months of life serves to accelerate
the development of general executive functions that can then
be applied in a variety of cognitive situations. This interesting
104 Bialystok et al.
result does not negate the notion that some forms of the
bilingual advantage are caused by inhibition of the nonused
language but rather raises the interesting possibility that the
advantage may have more than one causative mechanism.
What happens throughout life once bilingualism has
modified these executive control systems? Does the bilingual
advantage simply increase as the person accumulates
experience dealing with two or more languages? And if
bilingualism offers some protection against age-related cogni-
tive decline (Bialystok, Craik, & Freedman, 2007; Kave´, Eyal,
Shorek, & Cohen-Mansfield, 2008), does an increase in the
bilingual advantage occur simply as a result of monolinguals
showing a steeper decline in cognitive functioning than
bilinguals do?
One problem with assessing these possibilities is that most
studies deal with just one age group, so the opportunity to make
lifespan developmental comparisons is limited. One exception
is an article by Bialystok, Martin, and Viswanathan (2005)
reporting studies on 5-year-olds and young, middle-aged, and
older adults performing the same task, the Simon task. This
series of studies showed a bilingual advantage (faster RTs) that
was substantial in the 5-year-old children, virtually absent in
20-year-old undergraduate participants, but present again in
groups of middle-aged (30–59) and older (60–80) adults. The
authors suggested that the absence of an advantage in young
adults may reflect the fact that cognitive control is most
efficient at that time, so bilingualism provides no further boost.
The two studies involving middle-aged and older adults were
consistent in showing a larger bilingual advantage for the
oldest (60–80) group, because the drop in efficiency from the
middle-aged to older participants was greater for monolinguals
than for bilinguals. This pattern of an especially strong
advantage for the oldest bilingual participants was also found
in three other studies by Bialystok and collaborators (Bialystok
et al., 2004, 2006, 2008a; see Fig. 3b).
In the first two sections of this report, we reviewed
behavioral studies of language and cognition, presenting the
general finding that bilingual children and adults have smaller
vocabularies and slower lexical access times than do their
monolingual peers but that they also show enhanced cognitive
control on a variety of tasks. What are the neural correlates of
these effects? Is it possible to detect these subtle differences
through neuroimaging techniques? In the next section, we
survey the current evidence for structural and functional
changes in the brain that result from bilingual experience.
3. Neural Bases of Language Control in
Whether one speaks just one language or more than one
language, everyday use of language involves cognitive control.
Bilingual speakers do not develop a separate control system;
rather, as we have argued above, the use of two languages
imposes on a single control system additional demands beyond
those experienced by speakers of just one language. Our central
claim is that this control system or network is used by both
monolinguals and bilinguals but that the additional role in
bilingual language processing modifies it, changing its perfor-
mance for all tasks. In Section 2 we examined the cognitive
consequences of such enhanced control. Here we make explicit
the components of the network involved in language control,
demonstrate how they also mediate the cognitive advantages
shown by bilinguals, and explore the neural bases of control
using many of the same tasks discussed in Section 2.
Figure 5 identifies the basic components of the control
network, distinguishing it from the bilingual language system
that it controls. We can think of the bilingual speaker as perform-
ing multiplelanguage tasks such as speaking one language rather
than another. A bilingual must also monitor the language in use
and either maintain it if the circumstance demands (e.g., when
speaking to a monolingual speaker of that language)—and so
avoid inadvertently switching into the other language—or, on
occasion, deliberately switch to the other language if the circum-
stance changes—for example, when a monolingual speaker of
the other language enters the conversation.
The task-switching paradigm described in Section 2 can be
adapted to test language switching in bilinguals, and we use it
here to illustrate the workings of the network for language con-
trol. The task is to name a presented numeral, for instance 4,
in L1 (e.g., French) or in L2 (e.g., English). The participant’s
selection of one task rather than another governs the output
from the bilingual language system; if the task is to ‘‘name in
French,’’ the person says ‘‘quatre.’’ To be successful, the
activation of the selected task (i.e., the mental representation
Cognitive Control Network
Execut ive
Monit or ing
Compet ing t as k
Fig. 5. Basic components of the cognitive control network
for bilinguals, distinguishing it from the bilingual language sys-
tem that it controls. The bilingual language system refers to
a person’s mental representation of their languages; for
present purposes, we leave this undifferentiated and focus
on the components of the control system. A bilingual can
perform different language tasks: He or she can choose to
speak one language rather than another, can switch between
languages, or can translate between them. Task schemas
configure the bilingual language system so as to achieve the
intended task, but these schemas are in competition to con-
trol the bilingual language system. Their activation must be
monitored and, if necessary, adjusted by a higher-order
executive process. For example, a bilingual must either
maintain the current language in use if the circumstance
demands (e.g., when speaking to a monolingual speaker of
that language)—and so avoid inadvertently switching into
the other language—or, on occasion, deliberately switch
to the other language if the circumstance changes—for
example, when a monolingual speaker of the other language
enters the conversation.
Bilingual Minds 105
of the task set, its ‘‘task schema’’) must exceed and continue to
exceed that of the competing task. Therefore, the speaker must
monitor the speech output, and where marked slowing is
detected or an error is noticed (i.e., saying ‘‘four’’ rather than
‘quatre’’) the speaker must make some adjustments. The
speaker might increase the activation of the required task
(‘‘name in French’’) or suppress the activation of the alternative
task (‘‘name in English’’)—as discussed Section 2 when we
examined selection versus inhibition. Executive and monitor-
ing processes are needed to establish new schemas (e.g., in the
case of an experimental task) and invoke ones that are already
part of a person’s repertoire. In this role, these processes work
proactively; in response to performance difficulties, they work
reactively (Green, 1998). A person may be conscious of the
need to make such adjustments when an overt error is made, but
on other occasions control adjustment may occur automati-
cally, as in the way a thermostat adjusts power output in
response to a deviation from the desired temperature (Green,
1998; Paradis, 2009; see Fernandez-Duque & Knight, 2008, for
work suggesting that only conscious control leads to perfor-
mance benefits across tasks).
What produces slower responses or overt naming errors?
Marked slowing in naming in French, for example, may reflect
successful inhibition of a strongly competing name in the other
language (i.e., English), whereas naming in the wrong language
indicates a failure of control. Activation of the English name
may also increase the activation of the task schema for English
and lead to increased competition with the task schema for
French. Resolving such competition requires suppression of the
English task schema. In other words, when a bilingual speaks
two languages regularly, speaking in just one of these
languages requires use of the control network to limit interfer-
ence from the other language and to ensure the continued
dominance of the intended language.
Would there be a difference in the switch cost if the bilingual
were more fluent in French (L1) than in English (L2)? In that
case, French would be the easier task and English the more
difficult task, and the interesting finding is that it takes longer
to switch into the easier task (143 milliseconds, ms) than it does
to switch into the more difficult task (85 ms; Meuter & Allport,
1999). A plausible explanation for this seemingly paradoxical
asymmetry of switch costs is that in order to name in English (the
more difficult task), the easier task (naming in French) must be
strongly inhibited, and it takes more time to reinstate the easier
task, producing an asymmetry in the switching cost. Similar
results were obtained in a study by Misra, Guo, Bobb, and Kroll
(2007). Participants were asked to name pictures in L1 or L2
under either mixed conditions, when either L1 or L2 could be
required, or in blocked conditions, when only one language was
used. Their results showed that naming in L1 was slower under
mixed conditions than it was under blocked conditions and that
L1 naming was slower than L2 naming in the mixed conditions
(an effect of reversed language dominance), supporting the
interpretation that L1 was inhibited to permit the possibility of
L2 naming. No asymmetry of switch costs is found when
bilinguals switch languages voluntarily, yet a complete reversal
of language dominance is found—again suggesting some form
of inhibition of the L1 (Gollan & Ferriera, 2009).
Not all the research is consistent on this point. Finkbeiner,
Almeida, Janssen, and Caramazza (2006) had bilingual
participants name digits in either L1 or L2 and then perform
a picture-naming task in their dominant language. Following
the argument for greater inhibition of the dominant language,
the hypothesis is that it should take longer to name pictures
in L1 if the digit naming had been performed in L2. However,
Finkbeiner et al. found no difference in picture-naming latency
and so concluded that no inhibition of the nonused language
took place. Their conclusion, though, is difficult to reconcile
with evidence of global language inhibition identified in the
later study by Philipp and Koch (2009). A more complete
review of these issues is presented by Kroll et al. (2006).
The experimental research on bilingual task switching
generally uses explicit cues to signal the language required
on the current trial. Deliberate language switching in real life
also requires a speaker to monitor the context for cues as to
which language to speak (e.g., this person speaks L1 but not
L2) and ensure correct language selection and suppression of
any competing responses. Our premise, then, is that the addi-
tional demands on bilingual speakers relative to monolingual
speakers entail greater use of this control network. The partic-
ular tasks that are subject to control are varied (e.g., naming
pictures in one language, describing a scene in a second
language, translating from one language to another). However,
the components involved in monitoring performance and
ensuring correct selection of the intended language task are
applicable to other nonlanguage tasks, and, as we saw in the
previous section, they appear to generalize to nonverbal tasks.
Neural bases of cognitive control
Figure 6 identifies the cortical and subcortical structures that
are components of the cognitive control network in Figure 5.
We follow others in separating the neural structures mediating
control from those that process linguistic or other kinds of sen-
sory or motor data (Posner & Petersen, 1990). The idea is that
these cortical and subcortical structures work together to limit
the effects of interference and to switch between tasks.
For example, they may function as a control loop that continu-
ally monitors attention to the required task (e.g., Botvinick,
Braver, Barch, Carter, & Cohen, 2001; Kerns et al., 2004).
In its monitoring role, the anterior cingulate cortex may detect
and help resolve interference (Lau, Rogers, & Passingham,
2006) and signal the prefrontal cortex, with its widespread
connections to other regions (Dehaene & Changeux, 1991;
Desimone & Duncan, 1995; Miller & Cohen, 2001), to alter the
activation of the task schemas. Another region in the medial
frontal cortex superior to the anterior cingulate cortex, the pre-
supplementary motor area (pre-SMA), is also implicated in the
control of action but seems linked more closely to spontaneously
chosen actions than to response conflict (Lau et al., 2006).
The parietal cortex is involved in representing the task,
through its connection to the prefrontal cortex, and in selecting
106 Bialystok et al.
among competing responses, through its connection to the
basal ganglia (Bunge et al., 2002). The basal ganglia are par-
ticularly important in task switching. Whereas traditional
views (Alexander & Crutcher, 1990; Mink, 1996) emphasize
the role of the basal ganglia in the control of movement,
recent work emphasizes their key role in cognitive control too
(e.g., Graybiel, 2000; Kotz, Schwartze & Schmidt-Kassow,
2009). Both cortical and subcortical structures are therefore
important in understanding how interference is controlled and
task switching achieved, so it is necessary to understand their
role in language control. We shall examine the involvement of
these regions in two broad categories of tasks: those requiring
the control of interference and those based on switching
between tasks and languages.
The control of interference
Using neuroimaging studies, we now consider the neural bases
for controlling interference. These studies mostly rely on func-
tional magnetic resonance imaging (fMRI) to assess the response
of different neural structures when there is an increased demand
to control interference. The basic data are the relative activation
of different neural regions as detected by fMRI. A common
assumption is that an increase in activation reflects an increase
in difficulty. There is more extensive research on the control
of interference in monolingual speakers, so our review makes
use of meta-analyses of data from a number of studies.
Interference control in monolinguals. The argument being
developed here is that bilinguals use the cognitive control
network shown in Figure 6 to control interference from the
competing language. Therefore, it is necessary to establish that
these regions are recruited when monolinguals perform tasks
involving response conflict. We consider work that has looked
at the neural regions involved in controlling interference in
three different tasks that, as described in Section 2, show a
bilingual advantage: a nonverbal flanker task, a Simon task,
and a Stroop task.
Although studies have examined these tasks separately, the
strongest evidence for a common set of regions involved in
cognitive control comes from studies testing two or more of
them in the same individuals (Fan, Flombaum, McCandliss,
Thomas, & Posner, 2003; Liu, Banich, Jacobson & Tanabe,
2004; Peterson et al., 2002). Fan et al. (2003) contrasted perfor-
mance on a flanker task, a Simon task, and a manual version of
the Stroop task in which individuals pressed one of four buttons
corresponding to the font color of a presented word. In all the
tasks, individuals responded faster in congruent trials than in
incongruent trials. Fan et al. identified two regions that showed
a common effect of conflict: one in the anterior cingulate cor-
tex and one in the left prefrontal cortex (see Roberts & Hall,
2008, for a review).
Nee, Wager, and Jonides (2007) examined data from
47 papers using different interference tasks. Their review con-
firmed the importance of the left prefrontal cortex (dorsolateral
Anterior Cingulate Cortex
• Attention
• Conflict monitoring
• Error detection
Inferior Parietal Lobule
• Maintenance of
• Working memory
Basal Ganglia, Caudate
•Language selection
• Set switching
• Language planning
• Lexical selection
• Executive functions
• Decision-making
• Response selection
• Response inhibition
• Working memory
Prefrontal Cortex
Fig. 6. Principal brain structures involved in cognitive control, and their putative functions. From Abutalebi and
Green (2007).
Bilingual Minds 107
region) and the anterior cingulate cortex, along with a region in
the left posterior parietal cortex, in overcoming Stroop conflict.
Neuropsychological data also support the importance of a
frontal region in verbal control. Hamilton and Martin (2005)
found that a patient with damage to a left inferior frontal region
showed a large interference effect in the Stroop task but
interference within the normal range for a spatial-conflict task.
The analyses of Nee et al. also showed that different kinds of
conflict induce slightly different patterns of neural response.
In resolving conflict based on resisting responding to an
infrequent stimulus, frontal and parietal regions in the right
hemisphere, in addition to the left dorsolateral prefrontal cortex
and the anterior cingulate cortex, were activated.
Much research on the control of interference has examined
the role of cortical structures but ignored the role of subcortical
structures, such as the caudate, that, as indicated in Figure 6,
should be involved in selecting among competing responses.
However, there is some relevant research on these subcortical
structures. The left head of the caudate activates when a previ-
ously learned movement has to be inhibited (Shadmeher &
Holcomb, 1999; Parsons, Harrington, & Rao, 2005) and when
a prepotent response has to be blocked (Li, Yan, Sinha, & Lee,
2008). The caudate is also active in controlling interference in
the Stroop task (Ali, Green, Kherif, Devlin, & Price, 2010). A
reasonable supposition, then, is that the caudate is involved in
the inhibition of plans of action and therefore controls both ver-
bal and nonverbal types of interference.
To summarize, neuroimaging research with monolinguals
confirms that a network involving the prefrontal cortex,
anterior cingulate cortex, and caudate is recruited in tasks that
require resolution of conflict from competing responses.
Interference control in bilinguals. We argue that bilinguals
use this same network to control conflict from two languages.
Therefore, if retrieving the name of a picture is effortful for
bilinguals because of the need to overcome interference from
the other language, then we would expect to find evidence for
the involvement of this control network in a picture naming
task. De Bleser et al. (2003) examined covert picture naming
in an L1 and a later-acquired L2. (For technical reasons,
neuroimaging studies sometimes adopt the expedient of asking
participants to mouth picture names or to name pictures only
covertly, so they will not move.) Participants were native
speakers of Flemish/Dutch who had learned French from the
age of 10. In one condition the picture names were cognates
(i.e., the translation equivalents were phonological and
orthographically similar), and in another condition they were
noncognates. For pictures with noncognate names, naming in
the L2 showed more activation in regions responsible for link-
ing conceptual information and word form than did naming in
the L1. The more important result is that activation increased in
two inferior frontal regions associated with more effortful lex-
ical and semantic retrieval. Therefore, data from this study,
along with others (e.g., Abutalebi, Cappa, & Perani, 2001;
Rodriguez-Fornells et al., 2005), suggest that naming in the
L2 is associated with more effortful processing, an idea
consistent with the involvement of cognitive control processes.
Moreover, as proficiency in the L2 increases, the relative
difference in activation between L1 and L2 decreases, again
consistent with the idea that there is a decrease in effort
(Abutalebi & Green, 2007).
Even early and highly proficient bilingual speakers show
evidence of more effortful processing in their L2 and recruit-
ment of control regions, despite demonstrating a processing
profile that is similar to that of native speakers. Kovelman,
Baker, and Petitto (2008) asked Spanish-English bilinguals and
English monolingual speakers to judge whether visually pre-
sented sentences were plausible or not. For the bilinguals, the
sentences were presented in separate experimental blocks for
each language. The English sentences (and their Spanish
translations) varied in their syntactic complexity, being either
subject–object relatives (e.g., ‘‘The child spilled the juice that
stained the carpet’’) or arguably more complex object–subject
relatives (e.g., ‘‘The juice that the child spilled stained the car-
pet’’). As expected, bilingual speakers showed a differential
response to complexity as a function of the presented lan-
guage. Spanish relies more on morphological marking than
word order to signal grammatical relations. Like the English
monolinguals, bilingual speakers showed increased left infer-
ior frontal activation for the more complex English sentences.
In contrast, they showed no differentiation as a function of
complexity when processing the Spanish sentences. However,
the study also showed that bilingual speakers processing
English showed more activation in the left frontal region than
monolingual English speakers did. In other words, processing
even in a language in which they are highly fluent is more
effortful for bilingual speakers and engages regions associ-
ated with cognitive control.
Increased proficiency in the L2 may also alter processing in
the L1 precisely because of increased competition. In reading,
the mappings between letters and sounds differ between lan-
guages, so the same string of letters can give rise to conflicting
pronunciations. For example, what happens when native read-
ers of Italian (which has a regular relationship between letters
and sounds) read in their L1 after learning English, in which the
relationship is irregular? As vocabulary knowledge in English
increases, native Italian readers reading Italian show a linear
increase in activation in a left frontal region associated with
mapping letters to sounds (Nosarti, Mechelli, Green, & Price,
2010). Such an outcome indicates increased competition. More
to the point, there is a linear increase in activation in a left
frontal region used to resolve irregular pronunciations in
monolingual native English readers. Interestingly, this region
is also one that helps resolve lexical competition (e.g., de
Zubicaray, McMahon, Eastburn, & Pringle, 2006). These data
again suggest that bilingual speakers and readers, at least in
contexts where both languages are active, experience increased
verbal conflict and recruit a left frontal region to resolve it.
Other research allows us to see both cortical and subcortical
regions involved in controlling interference. Van Heuven,
Schriefers, Dijkstra, and Hagoort (2008) made use of a special
relationship that exists between words in two languages such as
108 Bialystok et al.
English and Dutch. Their participants were highly proficient
Dutch-English university students who had learned English at
the age of 10 to 12 years. Van Heuven et al. asked participants
to decide whether a presented word was a real English word or
not—an English lexical decision task. Some English words,
termed interlingual homographs, are also real words in Dutch;
for example, room means ‘‘cream’’ in Dutch. In an English lex-
ical decision task, ‘‘room’’ elicits a competing ‘‘No’’ response
because it is a word in Dutch, and in an English lexical decision
task Dutch words should receive a ‘‘No’’ response. Relative to
control words, therefore, correctly deciding that an interlingual
homograph was a real English word elicited increased
activation in three regions displayed in Figure 6: the left infer-
ior prefrontal regions, the anterior cingulate cortex (together
with another region we have noted previously in the medial
frontal cortex, the pre-SMA) and the left caudate. As expected,
van Heuven et al. observed no differential activation for
interlingual homographs in a group of monolingual English
speakers. This experiment left unresolved whether the
activated regions were signaling conflict arising from the
stimulus itself (i.e., ‘‘room’’ elicits two meanings in Dutch-
English bilinguals) or conflict arising from ambiguities asso-
ciated with the response (i.e., is ‘‘room’’ a word in English?).
To determine which regions responded to stimulus-based
rather than response-based conflict, the researchers per-
formed another experiment on a separate group of bilinguals
from the same population. In this case, participants knew that
some of the words might be Dutch words and responded
‘Yes’’ to each real English word regardless of whether it was
also a Dutch word. In this case, interlingual homographs
elicited increased activation only in left prefrontal regions,
suggesting that the left prefrontal regions are sensitive to
stimulus-based conflict. In contrast, the response profile of
the anterior cingulate cortex (and the pre-SMA) and left cau-
date reveals regions that are either sensitive to, or help
resolve, response-based conflict.
The precise impact of the other language might depend on
how active it is. It is reasonable to expect that it will be most
active when it is being used at the same time when bilinguals
are in what Grosjean (1998) termed a bilingual mode and they
are switching between languages. We consider the response of
the control regions in the section on language switching.
Task switching
The second paradigm within which to examine the neural
bases of cognitive control is task switching. Different types
between languages or between other types of tasks. Stroke
damage to a specific structure can lead to difficulties in task
performance and so provide evidence of its causal role in cog-
nitive control that complements the data from neuroimaging
studies. Again, we begin by establishing the neural basis of
task switching in monolinguals and then compare those pat-
terns to data from bilinguals performing task switching and
language switching.
Task switching in monolingual speakers. The occurrence of
a stroke is a tragic and dramatic event that helps to explicate the
role of regions in Figure 6 for task switching. Greater damage
to the left frontal cortex leads to increases in switch costs and so
reflects difficulty in holding the current task in mind or in
selecting the correct response, although inhibition of inap-
propriate tasks or associated responses might be more closely
linked to the right frontal cortex (Aron, Monsell, Sahakian, &
Robbins, 2004). The anterior cingulate cortex is less suscepti-
ble to stroke, but it is sometimes necessary to ablate part of it
surgically. Postoperatively, such patients have difficulty
responding to a cue that requires them to switch the direction
in which they move a joystick (Williams, Bush, Rauch,
Cosgrove, & Eskandar, 2004). Damage to the basal ganglia
also severely impairs a person’s ability to switch between tasks
and to overcome the interference from the prior task. We illus-
trate with a nonverbal task (Yehene, Meiran, & Soroker, 2008).
Yehene et al. asked their patients to press one of two keys in
response to the position of a target schematic face in a 2 2
matrix on the basis of one of two rules. In the top-down task
they had to press Key 1 if the target was in the top half of the
grid and Key 2 if it was in the bottom half. In the left–right task
they pressed Key 1 if it was on the left side of the grid and Key
2 if it was on the right. The rule was cued on each trial. On crit-
ical trials, the correct response depended on the application of
the correct rule, because Key 1 designated a target that was up
or left and Key 2 designated a target on the bottom or right.
Therefore, if a target was in the upper right cell of the grid,
pressing Key 1 was correct for the top-down task but Key 2 was
correct for the left-right task. Basal ganglia patients were
severely impaired when the rule switched in this task, signaling
the importance of that structure in such tasks.
In a meta-analysis of data from neuroimaging studies involv-
ing different types of task switches (e.g., rule switching, changes
in target locations, and different response sets), Wager, Jonides,
and Reading (2004) confirmed that the regions in Figure 6 are
reliably activated on task-switch trials. The prefrontal cortex is
sensitive to changes in demands involved in switching between
tasks (Christoff & Gabrieli, 2000; MacDonald, Cohen, Stenger,
& Carter, 2000) with more complex working memory demands
associated with right frontal activation (Simmonds, Pekar, &
Mostofsky, 2008). The anterior cingulate cortex is sensitive to
changes in tasks and to errors consistent with its rolein monitoring
and in adaptive control in response to errors (e.g., Hyafil,
Summerfield, & Koechlin, 2009). Parietal areas are also involved
in remapping stimuli to response according the new task (e.g.,
Corbetta & Shulman, 2002; Dosenbach et al., 2006). Finally, the
basal ganglia play a role in shifting response. In a study in which
participants tracked a continuous sine wave by controlling a
cursor, activation increased in the left caudate when the current
trial required a movement opposite to that used previously
(Lungu, Binenstock, Pline, Yeaton, & Carey, 2007).
Language switching in bilinguals. In monolingual participants,
the regions identified in Figure 6 were shown to contribute to
task switching. Do they also contribute to language switching
Bilingual Minds 109
and interfere with linguistic behavior in bilingual patients?
Patient reports indicate that damage to the prefrontal cortex,
inferior parietal cortex, or basal ganglia structures affect the
ability of bilingual patients to voluntarily switch from one
language to another. As the anterior cingulate cortex is less
susceptible to stroke, there are fewer reports for this structure,
but all the other regions indicated in this control network
show a clear role in language switching. Damage to either the
left prefrontal lobe (Stengel & Zelmanowitz, 1933; Zatorre,
1989; Fabbro, Skrap, & Aglioti, 2000) or left inferior parietal
lobe (Herschmann & Po¨tzl, 1920; Po¨ tzl, 1925, 1930; Leischner,
1948/1983) can yield pathological switching, that is, unintended
or inappropriate switching between languages. Lesions to the
head of the caudate elicit either selective recovery of the current
language, as if it is no longer possibleto disengage from it (Aglioti
& Fabbro, 1993; Aglioti, Beltramello, Girardi, & Fabbro, 1996),
or pathological switching between languages (Abutalebi,
Miozzo, & Cappa, 2000; Marie¨n, Abutalebi, Engelborghs, &
De Deyn, 2005). In the case reported by Abutalebi et al., A.H.,
a trilingual speaker of Armenian (L1), English (L2), and Italian
(L3), was unable to avoid switching languages when naming
simple pictures. For example, although he named the picture of
a clock correctly in Armenian in an Armenian testing session,
he named it in Italian in the English naming session and in English
in the Italian naming session.
That the circuits underlying language switching are
widespread is also indicated by data from transient cortical and
subcortical electrical stimulation of the brain during surgery for
treating glioma tumors or epileptic foci when the patient is
awake. In the case of bilingual speakers, this stimulation can
lead to involuntary switching from naming pictures in one
language to naming them in another, reflecting the temporary
disruption of control (Moritz-Gasser & Duffau, 2009a, b).
Neuroimaging studies of bilinguals without brain damage
provide complementary data. In a study with early Spanish-
English bilingual speakers, Hernandez et al. (2000) reported
more activation in left dorsolateral prefrontal cortex when
switching between naming pictures in English and Spanish
than when naming pictures in just one language (see also
Hernandez, 2009; Hernandez, Dapretto, & Bookheimer,
2001; Chee, Soon, & Ling Lee, 2003). Price, Green, and von
Studnitz (1999) used single words and found that switching
between languages increased activation in regions associated
with phonological processing (a left inferior frontal region,
Broca’s area, and parietal cortices). Taken together, these data
indicate that language switching or mixing induce increased
frontal and parietal activity consistent with the requirement
to inhibit ongoing activity associated with one task and select
a relevant response in the face of competition.
More recent research provides a fuller picture of the control
regions involved in language switching. Abutalebi and
colleagues (2008) studied German-French bilinguals who
learned French relatively late (around 12 years of age) and were
enrolled in a translation course. The task was to name pictures
in their L1 under one of two conditions. In the single-language
condition, a cue signaled whether they were to name the picture
(e.g., ‘‘cup’’) or generate an associated verb (e.g., ‘‘drink’’). In
the dual-language condition, the cue signaled whether they
were to name the picture in their L1 or in their L2. In this dual
condition, the nontarget language is very active. The key
analysis is the contrast between naming a picture in L1 in the
single-language condition and naming it in L1 in the
dual-language condition. Abutalebi et al. found that naming
pictures in the dual-language condition induced more extensive
activation in the left prefrontal cortex, the anterior cingulate
cortex, and the left caudate nucleus than did naming the
same pictures in the single-language condition. Furthermore,
the study confirmed more extensive activation in these regions
when individuals were using their weaker L2. These results are
strong support for the importance of these regions in selecting
a language in the face of interference.
Other studies have used neuroimaging to examine the neural
basis of the asymmetric cost in switching between a language
in which one is more proficient and a language in which one
is less proficient (Meuter & Allport, 1999). We illustrate this
with a functional imaging study, but there is other work using
evoked reaction potentials that is consistent with the idea that
switching between languages involves a process of actively
inhibiting the other language (Jackson, Swainson, Cunnington,
& Jackson, 2001) even if that does not invariably lead to an
asymmetry in switching cost (e.g., Christoffels, Firk, &
Schiller, 2007; Verhoef, Roelofs, & Chwilla, 2009, 2010).
Wang, Xue, Chen, Xue, and Dong (2007) examined the cost
of switching into L1 (Chinese) versus a newly acquired L2
(English). In line with the view presented here that the same
regions are used for cognitive control and language control,
Wang et al. reported increased activation in the regions associ-
ated with control when subjects switched into L2. The pattern
again is consistent with the idea that bilinguals must inhibit
their L1 to speak in their L2 when they are switching between
the two languages. The persisting suppression delays naming
time when individuals switch back into L1.
The effects of language switching have also been examined
in comprehension, and, surprisingly perhaps, there is also good
evidence for the involvement of control processes. Language
switching elicits a left caudate response in late bilinguals
(German-English/Japanese-English) when they make semantic
decisions about the meanings of words (Crinion et al., 2006).
The left caudate is also activated when bilinguals encounter a
language switch while listening to a narrative and make no
overt response at such a juncture (Abutalebi et al., 2007). The
participants in this study were Italian-French bilinguals who
had acquired French before the age of three and were living
at the time of testing in an Italian community in Geneva, where
French predominates. Switching elicited bilateral inferior fron-
tal activity (along with activation in a language area). Most
interestingly, a switch into the less-exposed language (Italian)
elicited activation of the left caudate and the anterior cingulate
cortex. Such a neural response indicates the need to distinguish
between the processes responsible for implementing control
from processes associated with overcoming the effects of such
control. In the present case, switching into the less dominant
110 Bialystok et al.
language in a comprehension task appears to demand more
neural resources to overcome (suppress) the activation of the
more exposed (dominant) language. In a production task, such
a neural response may give rise, as we have seen, to slower
naming when switching back into the more dominant language
in order to overcome its earlier suppression.
Finally, a special type of language switching occurs when
bilinguals translate from one language to another, and this task
also involves the cortical and subcortical structures depicted in
Figure 6. Price et al. (1999) reported that, in contrast to reading
in different languages, translating activated mainly the anterior
cingulate cortex and bilateral subcortical structures including
the head of caudate. In that study, if participants did not know
the translation equivalent they responded ‘‘No’’ or ‘‘Nein.’
However, in other studies, left inferior frontal activation was
found when that option was not available, both in single-
word tasks (Klein, Milner, Zatorre, Meyer, & Evans, 1995) and
in auditorily presented text translation by simultaneous inter-
preters (Rinne et al., 2000). Further, Rinne et al. (2000)
reported that, since translation into the nonnative language is
the more difficult task, left dorsolateral activation was more
extensive when the interpreters translated into their nonnative
language. The involvement of subcortical structures along with
activity in the left prefrontal cortex is also reported (Lehtonen
et al., 2005). Lehtonen and colleagues studied Finnish-
Norwegian bilinguals who had learned Norwegian as adults
(21–36 years). Participants completed a translation task and a
control task. In the translation task, they silently translated
visually presented Finnish sentences into Norwegian and then
decided whether a presented Norwegian probe sentence was a
correct translation of the Finnish sentence. In the control task,
they silently read a Finnish sentence and determined whether a
Finnish probe sentence was identical to it. The contrast between
the translation and control task yielded substantial activation in
the left (ventrolateral) prefrontal cortex and in a region of the
basal ganglia (globus pallidus) that is activated in suppressing
competing responses (Atallah, Frank, & O’Reilly, 2004; Ali
et al., in press). Taken together, these data provide evidence for
the involvement of the cortical and subcortical regions of the
control network in a task special to bilinguals.
Local switching and mixing costs in bilingual and
monolingual performance. The difference between local
switch costs and mixing costs was discussed in Section 2, with
most studies reporting smaller mixing costs for bilinguals and
with more varied evidence for local switch costs. This distinc-
tion can also be examined using evidence from neuroimaging.
From a control point of view, these two types of cost are
interesting because local switch costs reflect transient control
processes whereas mixing costs reflect the need for sustained
control. In task-mixed blocks, individuals need to keep two
tasks active and monitor the world for cues as to which one
to perform.
Dosenbach et al. (2006) provide a detailed analysis of the
regions involved in initiating a new task, sustaining it over a
sequence of trials, and responding to error. They argue that
the anterior cingulate cortex, together with another
bilateral frontal region (the anterior insula/frontal opercu-
lum) form a core region for implementing and sustaining a
new task. As yet, there are no comparable analyses for lan-
guage switching in bilinguals, so we illustrate with evidence
from two studies that compare local switching and mixing
costs in language tasks.
Braver, Reynolds, and Donaldson (2003) asked participants
to classify words according to whether they referred to objects
that were natural versus created or whether the objects referred
to were large versus small. Participants carried out these tasks
either in separate blocks of trials or mixed in the same block of
trials. The anterior cingulate cortex and prefrontal regions of
the right hemisphere were activated in the mixed blocks but
showed no variation with local switching. In contrast, local
switching was accompanied by activation in left prefrontal and
parietal regions.
Wang, Kuhl, Chen, and Dong (2009) extended these ideas to
language switching. Native speakers of Chinese who started
learning English around 12 years and who rated themselves
as being of low to moderate proficiency in English named
digits silently either in single-language or mixed-language
blocks. Language of response was signaled by a verbal cue
presented 400 ms before the stimulus digit. Consistent with
previous research (e.g., Meuter & Allport, 1999), it took longer
to switch back into Chinese than to switch into English (43 ms
vs. 8 ms.). There was also a mixing cost that was similar for
Chinese and English (but see Christoffels, Firk, & Schiller,
2007; Kroll et al., 2006, for data showing that an L1 can reveal
greater mixing costs). Importantly, however, local switching
and mixing costs were associated with different brain regions.
For mixing costs, there was activation in bilateral prefrontal
and frontal regions. Unlike other studies, Wang and colleagues
reported no differential activation of the anterior cingulate
cortex, a difference they attribute to the more automatic retrie-
val of numeral names. In contrast, and in line with the data of
Braver and colleagues (2003), local switch costs activated left
frontal regions (along with other cortical and subcortical
regions). Based on an analysis of individual data, Wang and
colleagues also proposed that a left parietal region plays a role
in overcoming inhibition or in reactivating the previous
Bilingualism and the neural networks for control
We have summarized research showing the neural regions
involved when individuals control interference in using one
of their languages and the regions involved when they switch
between languages. In both cases, the set of regions depicted
in Figure 6 is activated. These data suggest extensive overlap
with the regions mediating cognitive control when monolin-
gual speakers resolve interference or switch between different
tasks. Such a correspondence supports the proposal that the
bilingual advantage in nonverbal interference tasks and in task
switching arises from their use of neural regions recruited in
language control.
Bilingual Minds 111
We have relied on commonalities in the response of the
control regions in bilingual and monolingual speakers faced
with different tasks, but there may be subtle differences that are
missed in such comparisons. It is important to have studies that
directly compare bilingual and monolingual speakers (matched
on confounding variables such as IQ and socioeconomic class)
performing the same nonverbal conflict or switching task. One
such study has identified differences between bilinguals and
monolinguals (Bialystok, Craik, et al., 2005). The researchers
contrasted two groups of early bilinguals (French-English and
Cantonese-English) with a monolingual English group
performing a Simon task. Bialystok and colleagues used
magnetoencephalography (MEG) to identify the neural basis
of processing differences between the language groups and ana-
lyzed two bands of signals: one associated with attentional con-
trol (theta band) and the other associated with signal processing
(alpha band). The data indicated that there is a common network
used by all participants but with with subtle differences in how
interference is controlled. Faster responding in the bilingual
groups was associated with more activation in the signal-
processing band in two left frontal regions and the left anterior
cingulate cortex, as distinct from the left middle frontal region
associated with faster responding in monolingual speakers. It
will be important to extend such research to other tasks.
Why, then, might bilinguals, at least those who use both
languages on a regular basis and who acquired them early in
life, show an advantage in overcoming interference and in task
switching? The position that we have sought to establish is that
it is due to the need to control linguistic interference with the
corresponding demands to monitor and adapt behavior. Such
control is required when individuals speak two languages.
It may also be required when individuals use two sign lan-
guages but appears not to be important when individuals speak
one language and sign in another. Consistent with this view,
Emmorey, Luk, Pyers, and Bialystok (2008) found that
speech-sign bilinguals responded comparably to monolinguals
and did not show the advantage demonstrated by a group of
speech-speech bilinguals on a flanker task; and Kovelman
et al. (2009) confirmed that bilinguals who spoke one language
and signed another showed no increase in prefrontal activation
when they switched between the two, although they did show
increased activation in language regions associated with
mapping meaning to form.
As noted earlier, whether the source ofthe bilingual advantage
is the voluntary or the involuntary nature of control is an
open question, though it may prove to be the former (cf.
Fernandez-Duque & Knight, 2008). But given that there is such
an advantage, the control network in bilinguals may be more
efficient overall, or bilinguals may adopt a more effective strat-
egy in performing nonverbal tasks. For example, in interfer-
ence tasks they might be better at maintaining the task goal
and so reduce the impact of conflicting information. In task
switching, they may respond more efficiently to a task cue and
retrieve task goals more effectively. If this is the case, switch-
reduced. Longitudinal studies will be important here, because
it is known that older adults shift from a control strategy that is
proactive and maintains task-relevant goals to one that is reac-
tive and retrieves relevant information only when required
(Jimura & Braver, 2010; Paxton, Barch, Racine, & Braver,
2008). The bilingual advantage shown in older adults may
reflect their continued use of a proactive control strategy
supported perhaps by left frontal structures and the anterior
cingulate cortex.
Bilingual experience may also alter the capacity of the
control network by altering the density of grey matter (i.e.,
the nerve cell bodies together with axons and dendrites) in one
or more control regions (e.g., anterior cingulate cortex;
caudate). It may even affect the white matter connections
(i.e., the myelinated axons that connect regions of grey matter).
Prior research indicates that cognitive, linguistic, and motor
abilities can correlate with differences in brain structure,
(e.g., Crinion et al., 2009; Draganski & May, 2008; Gaser &
Schlaug, 2003; Lee et al., 2007; Maguire et al., 2000; Mechelli
et al., 2004). If one or two regions show marked differences
then this would constrain accounts of the neural basis of the
bilingual advantage. Longitudinal studies are important, as
they can rule out preexisting individual differences rather than
bilingual experience as the source of the difference. In this
context, studies of the aging brain (see Section 2) may prove
particularly revealing, because age-related declines can be
related to changes in specific brain structures. Our supposition
is that deteriorating performance found in nonverbal-conflict
tasks will also be found in tasks involving language control.
4. Implications of Bilingualism for Clinical
The behavioral studies reviewed in Sections 1 and 2 reveal a
number of differences between bilinguals and monolinguals
in a variety of cognitive domains. These differences have
proven to be useful for understanding the implications of
bilingualism for cognitive development and cognitive aging.
Moreover, the recent work in neuroimaging and related fields
described in Section 3 is beginning to elucidate the neural
correlates that underlie proficient language use. The question
posed in the present section is whether these findings can be
applied to help practitioners in the areas of neuropsychology,
educational psychology, and speech/language pathology deal
with the problems of bilingual clients and patients.
The challenge to professionals in these applied fields is that
bilingual individuals vary enormously in their language skills.
A few of the many factors that affect the degree of language
proficiency in bilinguals are age and manner of acquisition of
each language, degree of use of each language over a lifetime,
and literacy and level of formal education in each language.
It seems likely that these same factors will also affect the extent
to which bilingualism modifies cognitive processing mechan-
isms. It is difficult to obtain a comprehensive assessment of all
relevant factors in each individual case—yet such assessment is
necessary to interpret test performance accurately. This uncer-
tainty about the details of individual bilingualism combined
112 Bialystok et al.
with the lack of tests developed specifically for use with bilin-
guals, the lack of knowledge about how bilingualism affects
performance on standardized tests that were developed for
monolinguals, and the strong emphasis on language-based
assessment in clinical settings makes it difficult to answer some
of the most common referral questions about bilinguals.
To simplify the following discussion, we assume that the
bilingual individuals had early exposure to two languages and
that English is the dominant language spoken by the majority of
people in the environment. However, much of the discussion
would apply equally well to proficient bilinguals who acquired
one of their languages late in life, to bilinguals who live in
bilingual communities in which one language is not clearly
in the majority, and certainly to situations in which English
is not the majority language.
Three general themes are common when bilingual individu-
als are referred to a clinician for intervention or therapy.
Although the specific questions differ, these same themes are
evident for children, adults, and aging bilinguals. The first
theme is to establish whether there is a cognitive impairment
or language impairment. In children, this question often
takes the form of asking whether the child is learning English
(the second language) as quickly as she or he should be, and
if not, if there is a language impairment or more general
developmental delay. For adults the concern is often linked
to test results. As we saw in Section 1, tests of verbal fluency
and naming generally reveal lower scores for bilinguals than
for monolinguals, and these verbal scores are frequently lower
than indicators of verbal memory or nonverbal functioning for
bilinguals. In a clinical setting, this pattern raises the concern
about the possibility of brain injury or developmental impair-
ment—precisely what those tests were designed to diag-
nose—rather than the history of bilingual language use. For
both children and adults, if language impairment is identified,
there are inevitably questions about the best strategy for
accommodating the impairment and for facilitating communi-
cation and recovery. For example, should treatment be pro-
vided in just one or in both languages? Would it be best to
try to use primarily one language to ease the load on the com-
promised cognitive system by avoiding bilingualism (e.g., by
switching to using only the majority language at home)?
A second theme is the need for advice on the best way to
promote rapid acquisition of English as the person’s second
language. For children the question is frequently framed in
terms of educational options: Is total immersion in English best,
or is it better to encourage parallel development of both
languages by including both as part of the academic curriculum?
In young adults, the concern is centered more on academic
achievement, and questions attempt to determine the role of
bilingual language use in academic outcomes. In middle-aged
and older adults, the focus again shifts to learning the language.
Some individuals are concerned about the length of time it is rea-
sonable to live in a country without learning the environmental
The third theme is more specialized. Clinical intervention is
sometimes sought to assess the adequacy of English
proficiency for a specific purpose, such as functioning in
school or in a professional setting. Adequate proficiency is also
essential for safety and security, as in understanding the con-
versation in a medical interaction or discussing the risks of a
medical procedure. Linguistic levels that may be perfectly ade-
quate for some purposes may fail to support the ability to
understand complex information for which careful thought and
cautious decision making are required. These situations may
also require the services of a clinician.
There are a number of reviews on cognitive and language
assessment of bilinguals that provide useful information on the
challenges that arise, on the kinds of questions to ask in clinical
settings to obtain the necessary information to interpret
bilingual performance on language-based tests, and on how
bilingualism can affect performance on specific tests (e.g.,
Altarriba & Heredia, 2008; Baker, 2000; Cummins, 2000;
Kohnert, 2007; Paradis, 2008; Paradis & Libben, 1987; Pen
& Bedore, 2009; Ponto´n & Leo´n-Carrion, 2001; Rivera-Mindt
et al., 2008; Valde´s & Figueroa, 1994). Here we attempt to
connect questions about assessment of bilinguals more specifi-
cally with the experimental literature reviewed above.
Before considering how bilinguals differ from monolinguals
in their performance on neuropsychological tests, it is helpful
to review what typically happens during a cognitive assess-
ment. Neuropsychologists receive referrals from parents,
schools, and physicians, usually with a very specific question
attached (e.g., Is there a language disability? Is the person
beginning to show signs of early Alzheimer’s disease?). The
neuropsychologist will subsequently review the patient’s
academic record or medical chart and schedule an appointment
to obtain a case history and administer cognitive tests. The gen-
eral questions related to case history are the same for bilinguals
and monolinguals: Were there any complications at birth? Was
a learning disability ever suspected? What was academic per-
formance like through school? What was the highest level of
education attained? What is the employment history? Were
there any losses of consciousness? Is there any history of
substance abuse or other psychiatric conditions? In some cases,
there will also be a detailed language history for bilinguals, to
determine which language is dominant, when and how both
languages were learned, the extent to which both languages are
currently being used, and other factors (e.g., Marian, Blumenfeld,
& Kaushanskaya, 2007).
Subsequently, the neuropsychologist will administer a
series of tests to assess a variety of cognitive domains (e.g.,
mental status, IQ, language, memory, executive functions,
and visuospatial skills), usually with heavier emphasis on
tests that will be useful in answering the specific referral ques-
tion. Often vocabulary tests are used to estimate verbal IQ,
picture naming tests are used to identify the presence of cog-
nitive impairment, and timed verbal fluency tests are given to
look for frontal lobe pathology (Lezak, 1995). Verbal fluency
performance that are associated with certain types of disease
(e.g., deficits in semantic fluency are associated with Alzhei-
mer’s disease whereas deficits in letter fluency are associated
Bilingual Minds 113
with Huntington’s disease; Rohrer, Salmon, Wixted, &
Paulsen, 1999). Assessment of bilinguals is complicated by
the problem that bilingualism itself influences performance
on these measures, and it is often not clear what adjustments
should be made to interpret performance relative to that of
monolinguals on the same tests.
Assessing vocabulary knowledge in bilinguals
A staple of neuropsychological testing is the assessment of
vocabulary, but as we have seen in Section 1, bilinguals,
especially bilingual children, often control a smaller vocabu-
lary in each language than comparable monolinguals do, even
in the absence of other compromising factors. How can clinical
assessment make reliable judgments about the potential for a
disability or disease in contrast to a normal outcome in the
context of bilingual language use?
The approach taken to testing and interpretation often
depends on the nature of the referral question. In some cases,
relatively simple referral questions that can be successfully
addressed without much knowledge about bilingualism arise.
For example, parents may wonder how their child’s English
vocabulary knowledge compares to that of his or her monolin-
gual peers (note that in bilingual societies this question may be
less relevant, particularly if monolinguals are few in number).
In such cases, it is obviously appropriate to administer a test
that was developed for use with monolingual English-
speaking children, and the score obtained will provide a valid
answer to the question being asked. However, the possibility
of interpreting that same test score will not extend beyond the
answer to this one simple question. As a group, bilingual
children who speak a minority language at home (e.g., a non-
English language in an English-speaking environment) will
obtain lower receptive English vocabul ary scores than will mono-
linguals, even if their parents report that they are ‘‘proficient
speakers of English’’ (Bialystok, Luk, et al., 2010). These lower
English vocabulary scores may be found even in children without
much proficiency in the minority home language if the parents are
not native speakers of English, because such children have
reduced exposure to English vocabulary at home, at least com-
pared to children whose parents are native English speakers and
use English exclusively.
The difference in vocabulary size in bilinguals is probably
a better reflection of experience than of ability to learn.
In 6-year-olds, the vocabulary deficit associated with bilingual-
ism seemed to be restricted to test items classified as ‘‘unlikely
to occur in a classroom context’’ (Bialystok, Luk, et al., 2010).
Similar results may be obtained in older bilingual children and
in bilingual adults and, if so, such information could ultimately
be useful for developing vocabulary tests that cater to specific
profiles of bilingual language exposure. In addition, item
analyses may be useful for interpreting individual test scores.
For example, if a bilingual child misses a home-context item
(e.g., ‘‘toaster’’) it may simply mean that there have been no
opportunities to learn this word in English because it is unlikely
to come up in a school context.
Although a group of bilinguals will, on average, score lower
than a group of monolinguals, individual scores will not
necessarily be lower. In the large-scale study of ‘‘fluent
English-speaking’’ bilingual children between the ages of 3 and
10 years (Bialystok, Luk, et al., 2010), the distributions of
bilingual and monolingual scores overlapped much more than
not. This means that although the average bilingual score was
about 10 standard score points (2/3 of a standard deviation)
lower than the average monolingual score, only a small number
of bilinguals scored completely outside the range of
performance for monolinguals. Thus, the majority of bilingual
children described as ‘‘fluent in English’’ will obtain ‘‘normal’
scores on tests developed for monolinguals. However, it is also
probable that these same normal scores will fail to provide an
accurate representation of learning potential.
Vocabulary scores reflect the combined forces of the ability
to learn new vocabulary and the opportunities to learn new
vocabulary. Bilinguals who score within the average range for
monolinguals may have better-than-average ability to learn,
which has allowed them to achieve an average monolingual
score despite having fewer learning opportunities. An impor-
tant consideration in such cases is that comparisons between
monolingual and bilingual children with matched vocabulary
scores may be invalid because bilingual children with
monolingual-like vocabulary scores may be precocious learners.
Conversely, bilinguals whose vocabulary scores fall 2 standard
deviations below the monolingual average could be learning
disabled, or they may simply have had less opportunity to learn
English than their case histories suggest—two conclusions
with very different implications but with equally serious
consequences. Bilinguals who score below average may be
inaccurately diagnosed with impairment when none is present,
or could be diagnosed as ‘‘normal for a bilingual’’ even though
impairment is in fact present and treatment is needed. The less-
frequent cases in which bilinguals obtain scores that are higher
than are typical for monolinguals may indicate exceptional abil-
ity to learn vocabulary or more opportunities to learn English
than the case histories suggest—again, two conclusions with
very different implications. Much of this discussion likely
applies as well to bilingual adults, who also typically obtain
lower vocabulary scores than do monolingual adults (e.g.,
Bialystok et al., 2008a; Portocarrero et al., 2007).
This discussion demonstrates the tremendous challenge in
interpreting individual test scores in bilinguals. Even with the
availability of normative data about bilingual performance on
a given test, several factors continue to complicate interpreta-
tion. Further difficulty arises if one considers a broader range
of bilinguals at different proficiency levels. The previous dis-
cussion applies only to children who are judged by their parents
to be ‘‘fluent in English.’’ Such children can reasonably be
tested in English (and specifically should be tested in English
if English is their dominant language). However, even in such
cases, a more accurate estimation of language skills will
emerge if both languages are tested. Parents may sometimes
overestimate the degree of majority-language fluency that their
children have achieved. Bilinguals who are not dominant in
114 Bialystok et al.
English must be tested in their dominant language, but often
tests for those languages have not been developed, and there
are virtually no tests for different combinations of bilingual
types. One exception that is available in many different lan-
guage combinations is the Bilingual Aphasia Test (the BAT;
Paradis & Libben, 1987). However, the BAT was designed to
assess fluent adult bilinguals for possible language impairment
(i.e., aphasia), and it is not known how bilingual children
should perform on this test or even if the test is useful in asses-
sing bilingual adults who don’t have a high degrees of fluency
in their two languages.
Finally, these simpler cases of ‘‘relatively fluent-in-English
bilinguals’’ are perhaps least likely to present for referral in a
clinic because they have already been successful in attaining
second-language fluency. A more typical presentation will be
someone who seems to be having trouble acquiring second-
language fluency. Parents of young preschool children may
suspect a problem if their child seems to be avoiding English
speakers in the classroom, preferring instead to socialize only
with the small number of other children who happen to speak
the same minority language at home. Parents of older school-
aged children may become concerned about low academic test
scores or large discrepancies between verbal (e.g., reading/
writing) and less-verbal (e.g., math) academic domains.
(Here, ‘‘less verbal’’ is meant to emphasize that all academic
subjects require at least some verbal skills; for example, math
problems sometimes come in paragraph format or require
ability to read instructions.) In such referral cases, it is neces-
sary to assess what the opportunities to learn English have
actually been—sometimes children have actually had less
exposure to English than is assumed—and whether or not nor-
mal amounts of learning have taken place given those oppor-
tunities. Even with adequate assessments of opportunities to
learn, test interpretation is difficult because little to no infor-
mation about exactly how much exposure is needed to per-
form within a particular range on any given test is available
to clinicians.
A creative approach around these problems has been to
provide a learning opportunity during the assessment session
itself and then to determine how much learning takes place, an
approach sometimes called Dynamic Assessment (Gutie´rrez-
Clellen & Pen
˜a, 2001; Pen
˜a, Iglesias, & Lidz, 2001). This
approach is based on interaction between the clinician and the
child. Three types of dynamic assessment are (a) ‘‘testing the
limits,’’ in which feedback is provided and errors pursued
through further questioning; (b) ‘‘graduated prompting,’’ in
which the level of contextual support is manipulated; and (c)
‘test-teach-retest,’’ in which alternative versions of tests of the
same material are repeated after teaching to areas of weakness,
in order to assess learning (Gutie´ rriez-Clellen & Pen
˜a, 2001).
With these methods the amount of exposure is controlled—it
is provided during the testing session itself. Children who fail
to learn (i.e., do not show significant improvement on ‘‘mea-
sures of modifiability’’; Pen
˜a, Resendiz, & Gillam, 2007) are
flagged, with a high rate of accuracy, as probable cases of
developmental delay. Such techniques are extremely useful for
bilinguals and monolinguals alike, and they provide a means
for obtaining accurate assessments with less concern about how
to interpret past opportunities to learn.
In theory, bilingual disadvantages in vocabulary knowledge
should decrease with age as their time to learn words in both
languages increases. Although vocabulary knowledge contin-
ues to increase well into older age (Verhaeghen, 2003), new
words may be learned at a faster rate before knowledge reaches
a particular point (perhaps a typical adult-vocabulary reper-
toire). In other words, bilinguals should ‘‘catch up’’ to monolin-
guals as years of immersion in English accumulate. One way to
test whether this is indeed the case is to ask whether the voca-
bulary deficit associated with bilingualism decreases in chil-
dren as they progress through school and beyond that across
the life-span. Indeed there has been some suggestion that bilin-
gual children achieve monolingual-like vocabulary scores with
increased time in school (Hamers & Blanc, 2000). However,
the ‘‘catching up’’ notion is best tested with a longitudinal
design, and to our knowledge such studies have not been
reported. Moreover, bilinguals may appear to be catching up
only because the test materials are not difficult enough to
reveal persistent differences between bilinguals and monolin-
guals. When tested exclusively for their knowledge of
very-low-frequency words in the relatively dominant language,
for example in studies of the tip-of-the-tongue phenomenon,
adult bilinguals consistently report recognizing fewer of the
targeted vocabulary words than monolinguals do (e.g., Gollan
& Silverberg, 2001; Gollan & Brown, 2006). Tip-of-the-tongue
experiences are retrieval failures in which partial phonological
information is available; they generally occur for
low-frequency words but appear to be more broadly based for
bilinguals. Thus, differences between bilinguals and monolin-
guals in opportunities to learn vocabulary will be less apparent
in settings that only require knowledge of relatively easy,
frequently occurring words than they will be in settings that
require knowledge of difficult, low-frequency words (Gollan
et al., 2008). This may be because, by virtue of using each
language only part of the time, bilinguals will have had
relatively less exposure to words in each language than will
monolinguals (the weaker-links hypothesis described in
Section 1), although they will have had sufficient exposure to
learn frequently encountered words.
Confrontation naming
Confrontation naming is a testing method in which pictures are
presented to participants, who are asked to name them as
rapidly as possible. One of the most commonly used such
neuropsychological tests is the Boston Naming Test (BNT;
Kaplan et al., 1983). This test contains 60 black-and-white line
drawings that show a single object that speakers try to name.
The pictures are easy at the beginning of the test (e.g., a bed)
but become progressively more difficult, ending with uncom-
mon objects encountered in limited contexts. The ability to
name pictures is sensitive to changes in cognitive functioning
and is therefore useful for detecting subtle brain injuries
Bilingual Minds 115
(Lezak, 1995). Unfortunately, this test may have more limited
utility for assessing bilinguals, because cognitively intact bilin-
guals obtain lower scores than monolinguals on the BNT and
other standardized tests of picture naming (e.g., Roberts
et al., 2002) such as the Expressive Vocabulary Test (e.g.,
Portocarrero et al., 2007).
Outside of clinical settings, studies of picture naming
measure both naming success (the number of correct retrievals)
and the time needed to name pictures. Such studies reveal a
very subtle bilingual disadvantage (e.g., it may take bilinguals
60 milliseconds longer than monolinguals to name a picture;
Gollan, Bonanni, & Montoya, 2005). This result applies to
bilinguals immersed in a dominant but second-learned
language (e.g., Gollan et al., 2008) and to bilinguals living in
a bilingual society (Ivanova & Costa, 2008). Picture-naming
deficits in bilinguals could arise for the same reasons as
receptive vocabulary deficits—namely, less frequency of use
of specific words than for monolinguals. Alternatively, it may
be because of dual-language activation—that is, the need to
select one language in the face of competition from the other
one. It is also possible that both factors may be operating. Some
of the burden associated with bilingualism seems to be better
managed with increased age—a result that is consistent with
the notion of a frequency lag for bilinguals. In one picture-
naming study, older bilinguals were relatively faster to produce
low-frequency picture names in a nondominant language than
would be expected based on their otherwise relatively slow
naming times relative to proficiency-matched young bilinguals
(Gollan et al., 2008). Because low-frequency words in the non-
dominant language will be most vulnerable to the frequency-
of-use lag associated with bilingualism, these words are also
most likely to benefit from the increased exposure to language
associated with age.
The age-related advantage for producing low-frequency
words is also evident in studies comparing older to younger
monolingual speakers: Like older bilinguals, older monolinguals
consistently produce names for pictures with very low-frequency
words with greater success than matched young monolinguals
(for review see Gollan & Brown, 2006). It may be that aging
allows for the accumulation of experience to deal with
low-frequency words. The finding that older bilinguals are in
some ways ‘‘better bilinguals’’ than younger bilinguals may seem
unexpected from the perspective of bilingualism as an exercise in
cognitivecontrol. If the frontal lobes (Raz, 2000; West, 1996) and
executive control decline in older age and are needed to suppress
the dominant language during retrieval of the nondominant
language, then older bilinguals should have more difficulty than
young bilinguals in producing low-frequency words in the nondo-
minant language. It might be asked whether older bilinguals
perform better because the low-frequency words are archaic
words more familiar to older than to younger participants.
However, controlled studies select materials that are highly
familiar to both young and old adults, and in the timed picture-
naming study with bilinguals, the low-frequency targets were all
highly familiar and current (e.g., crutches, a whistle, a scarf, a
dustpan; see appendix in Gollan et al., 2008). Most importantly,
the relative age-related advantage appeared only inthe nondomi-
nant language, whereas the same concepts and words did not
demonstrate any age-related advantage in the dominant language
(or in monolinguals). Thus, it seems that accumulated use over a
lifetime has its greatest influence on the very lowest-frequency
words, thereby offsetting some aging-related deficits in retrieval.
A number of factors have been shown to reduce or even
eliminate the bilingual disadvantage in picture naming, and this
raises the question of what would be the best way to adjust tests
of picture naming to accommodate bilingual ability and enable
clinicians to perform reliable assessments. The answer to this
question may vary with the referral question, and the implica-
tions of these findings for diagnosis and treatment of bilinguals
are not yet established. For example, bilinguals name pictures
more quickly (Costa, Caramazza, & Sebastia´n-Galle´s, 2000;
Hoshino & Kroll, 2008) and, in some cases, with no disadvan-
tage relative to monolinguals (Gollan & Acenas, 2004) if the
test consists of pictures with cognate names. Cognates reduce
bilingual disadvantages via joint activation of target phonemes
(sounds) through separate lexical representations in each
language (for a review, see Costa, Santesteban, & Can
2005; for research showing increased activation for cognates,
see Broersma & de Bot, 2006). To illustrate, the lexical repre-
sentations of lemon and its Spanish translation limo
many shared sounds, but grape and its translation uva
activate no shared sounds. A similar reduction in bilingual
disadvantage may be obtained by asking participants to retrieve
names of people (Gollan, Bonanni, & Montoya, 2005).
Bilinguals’ relative ease at producing proper names may have a
different mechanism from cognate effects; bilinguals may
effectively be monolingual for proper-name production because
proper names are generally shared between languages (e.g., Golda
Meir is basically the same in Hebrew, English, Spanish, etc).
The finding that bilinguals are better able to name pictures
with cognate names could be useful clinically. One possibility
is that bilingual picture-naming tests should focus on cognates
(or proper names) for which bilinguals perform much like
monolinguals. However, removing the disadvantage may com-
promise a test as an assessment instrument. For example, the
presence of cognate effects on dominant-language production
implies the presence of dual-language activation even when
bilinguals are tested exclusively in their relatively more domi-
nant language. Thus, a possible problem with using cognates is
that cognates may increase the extent to which both languages
are active, and this may have other undesired effects on test
performance (note that cognate-facilitation effects have also
been found in bilingual children, but this literature has focused
primarily on receptive vocabulary rather than on picture
naming; August, Carlo, Dressler, & Snow, 2005; Mendez
Perez, Pen
˜a, & Bedore, in press).
Similar considerations apply to another way to reduce
bilingual disadvantages in a testing or assessment situation:
to allow bilinguals to use either language to name pictures
(Kohnert, Hernandez, & Bates, 1998; Gollan & Silverberg,
2001). This approach is sometimes called ‘‘composite’’ or
‘conceptual’’ scoring. The scoring method improves
116 Bialystok et al.
bilinguals’ picture-naming scores in young adults (Kohnert
et al., 1998), in elderly bilinguals (Gollan et al., 2007), and
even in bilinguals with Alzheimer’s disease (Gollan, Salmon,
Montoya, & da Pena, 2010). Thus, when naming is untimed,
the composite scoring option is not associated with any
observable processing cost and only facilitates naming perfor-
mance. In timed picture naming, the option to use either lan-
guage produces significant language-switching costs but also
reveals compelling facilitation effects (Gollan & Ferreira,
2009). Specifically, when given the option to use either lan-
guage, unbalanced bilinguals switch languages in a manner
that resembles a more balanced-bilingual profile of language
switching (i.e., no switch-cost asymmetry; Costa & Santeste-
ban, 2004; Costa et al., 2006). In addition, older bilinguals
perform much more like young bilinguals in voluntary lan-
guage switching, whereas they have considerable difficulty
with cued language switching (Hernandez & Kohnert, 1999).
Thus, although language mixing might allow bilinguals to com-
municate better in natural settings, it is not necessarily the case
that allowing language mixing and switching in a clinical setting
will lead to more effective diagnosis and treatment, because the
either-language scoring method may actually obscure differences
between patients and controls (Gollan et al.,2010), which is coun-
terproductive if the goal is to identify impairments in bilinguals.
In bilingual language assessment, the costs associated with lan-
guage switching and mixing can be avoided by testing each lan-
guage in a separate testing block.
The opposite outcome may be found for cognates. It may be,
for example, that language-impaired bilingual children are less
able to benefit from cognate manipulations than typically
developing bilingual children are. If this is so, then the ability
to benefit from cognate status itself could function as a kind of
bilingual-specific litmus test for cognitive impairment. In other
words, failure to demonstrate improved lexical access for cog-
nate words relative to typically developing bilingual children
would signal some type of language impairment. Importantly,
however, it is necessary to consider the relative dominance
of the two languages for the bilingual child and the relation
between that dominance and the language of assessment.
In relatively balanced bilinguals, cognates can reduce bilingual
disadvantages in both the dominant and the nondominant
languages (Gollan & Acenas, 2004; Gollan et al., 2007), but
such reductions are most robust when bilinguals are tested in
their nondominant language (e.g., Costa et al., 2000; Gollan
et al., 2007). A study by van Hell and Dijkstra (2002) using
lexical-decision and word-association tasks showed that high
level of proficiency even in an L3 can influence processing
speed in the dominant language. The clinical significance is
that it is not possible to discount nondominant language
knowledge because even an L3 can have an effect on L1 if the
degree of proficiency in the L3 is high enough. Therefore, it is
possible that cognate effects in the dominant language occur
only in relatively balanced bilinguals who are also cognitively
intact. Alternatively, cognate effects in the nondominant
language might be magnified in cognitively impaired bilinguals.
Additional studies are needed to determine the relations between
cognate effects on the one hand, and language and cognitive
assessment of bilingual children on the other.
Verbal fluency in clinical practice
Research using the verbal fluency test as an experimental tool
was described in Section 1. The results showed consistent bilin-
gual disadvantages on semantic fluency (except when receptive
vocabulary knowledge is matched), with somewhat less severe
or less certain disadvantages on letter fluency. Clinically, the
greater bilingual disadvantage in semantic fluency than in letter
fluency can be misleading, because this is the same pattern of
fluency performance that is found in monolinguals with early
Alzheimer’s disease as compared with normals (Butters,
Granholm, Salmon, Grant, & Wolfe, 1987). This creates a
dilemma for neuropsychologists: Is an individual showing
signs of early Alzheimer’s disease or is she simply showing the
effects of bilingualism on fluency? The verbal fluency test is an
important instrument in the battery to assess patients for cogni-
tive decline, so the ambiguity of the results obtained from bilin-
guals presents a clinical problem. To develop fluency tests for
bilingual speakers, it is necessary to understand why semantic
fluency is more affected by bilingualism than letter fluency is.
As we explained earlier, letter fluency requires greater recruit-
ment of executive control, perhaps offsetting bilinguals’ disad-
vantages in lexical retrieval.
A different approach to assessing older bilinguals is to use a
task related to verbal fluency, one that reflects semantic
processing yet distinguishes the cognitive mechanisms that
underlie the effects of bilingualism from those that are involved
in Alzheimer’s disease. In the semantic-association task (de
Groot, 1989), speakers are given a cue (e.g., ‘‘bride’’) and are
asked to produce the first response that comes to mind in
relation to the cue. The overwhelming majority of responses
in this task are semantically related to the cues, and this is true
for all speakers, whether they are monolingual or bilingual and
whether or not they are cognitively impaired. However,
bilinguals produce slightly but significantly different (or ‘‘less
typical’’) responses than are normally found in monolinguals.
For example, given the cue ‘‘bride,’’ they might say ‘‘pretty’
instead of the more typical ‘‘groom’’ (Anto´ n-Me´ndez &
Gollan, in press). A similar effect was reported in monolinguals
with Alzheimer’s disease as compared to cognitively healthy
controls (Gollan, Salmon, & Paxton, 2006). To this point,
therefore, there is the same interpretation problem as there is
for verbal fluency, because both bilingualism and Alzheimer’s
disease produce the same outcome. However, further experi-
ments with the semantic-association task demonstrated that
only the bilingual effect is modulated by lexical frequency.
Bilinguals produced the same associations as monolinguals
do when the cues were strongly associated to high-frequency
words. In contrast, speakers with Alzheimer’s disease produced
atypical responses regardless of associate frequency (Anto´n-
Me´ndez & Gollan, in press). This evidence is consistent
with the notion that Alzheimer’s disease impairs semantic
representations themselves (Butters, Salmon, & Heindel,
Bilingual Minds 117
1990), whereas in bilinguals, difficulty with lexical access can
sometimes leads them to perform in ways that imply semantic
deficits when none are present.
As with confrontation naming, there is an important role for
cognate status in the performance of verbal-fluency tests, so the
interpretation of results, especially for clinical assessment,
needs to account for this factor. Specifically, in both semantic
and letter fluency, bilinguals who speak languages with many
cognates spontaneously produce as many cognate responses
(e.g., ‘‘lemon’’) as monolinguals do but fewer responses for
words that are not cognates (Sandoval et al., 2010). Put another
way, words that are cognates across the two languages are
generated as often by bilinguals as they are by monolinguals
who only know them in one language, but unique words are
produced less often by bilinguals. In this sense, the greatest
difference in performance is in the lower production of noncog-
nate words by bilinguals, who appear to have easier access to
words that occur in both their languages. These findings suggest
that bilinguals who speak languages with an extremely high pro-
portion of cognates (e.g., Catalan-Spanish bilinguals) may exhibit
no fluency disadvantage, even for semantic fluency.
Another similarity between verbal fluency and confrontation
picture naming is that bilinguals retrieve a greater number of
concept names if they are tested in both languages (Bedore,
˜a, Garcı´a, & Cortez, 2005). However, unlike picture naming
in the BNT, fluency scores do not increase if bilinguals are
allowed to use whichever language comes to mind during a
single trial and so to switch between languages (Gollan et al.,
2002; De Picciotto & Friedland, 2001). The lack of an improve-
ment in fluency scores when both languages are used may reflect
the costs of language switching. The timing allowed to name
each picture in the BNT, about 6 seconds, is too long to detect
the millisecond cost of language switching, so on this task no
switching costs are reported. Presumably, on a more tightly
timed picture-naming task allowing responses in either
language, bilinguals would name fewer pictures than would
monolinguals in a fixed amount of time (e.g., 60 seconds),
because of the additional time needed to carry out the language
switch (Gollan & Ferreira, 2009).
Because bilingualism affects verbal fluency in a number of
interesting ways, there are various possibilities for reducing the
bilingual fluency disadvantage. However, reducing this disad-
vantage may compromise the reliability of the instrument as
an assessment tool for bilinguals, so it is not clear what combi-
nation of fluency tests would be most useful for diagnosis of
cognitive impairment in bilinguals. Minimally, interpretation
of the test scores needs to be modified to accommodate the sys-
tematic differences that accompany bilingual performance, but
ultimately it may be possible to develop fluency tests that are
specifically targeted to a bilingual population.
The assessment of executive functions
Many of the linguistic skills that bilinguals generally perform
more poorly than monolinguals (reviewed in Section 1) are
included in typical assessment batteries, often using the same
instruments as those used in research. Therefore, understanding
how to interpret bilingual performance on those tests is a
crucial concern for neuropsychologists. However, in Section 2
we described a variety of nonverbal cognitive tasks on which
bilinguals generally perform better than monolinguals. These
tasks were measures of executive control and, as we have argued,
the experience of bilingual language use has the beneficial out-
come of enhancing these levels. What are the clinical impli-
cations of this advantage?
The implications of this bilingual advantage for clinical
assessment are more limited than the bilingual disadvantage
in lexical retrieval for several reasons. Perhaps most important
is the great emphasis on verbal skills in clinical assessments,
with a more minor role for nonverbal cognitive performance.
Therefore, the bilingual advantages found in nonlinguistic
tasks will have relatively little effect on the cognitive profiles
generated in clinical settings. Another important point is that
many of the tasks showing bilingual advantages in experimen-
tal studies (e.g., the Simon task and the Attentional Network
Task) are not used in clinical settings.
An important exception is the Stroop color-word-naming
task, which is commonly used to measure attention and is diag-
nostic of a variety of conditions associated with cognitive
impairment (e.g., Lezak, 1995). As we have seen earlier, bilin-
guals generally suffer less Stroop interference and greater
Stroop facilitation than monolinguals do (Bialystok et al.,
2008a; Herna´ ndez et al., 2010). Several considerations make
it difficult to interpret these differences, however. For example,
performance on the Stroop is affected by language proficiency
(Tzelgov, Henik, & Leiser, 1990; Rosselli et al., 2002).
Because of this, it is possible that only highly proficient bilin-
guals will exhibit the advantage in their dominant language and
that disadvantages may be found if bilinguals are tested in a
less dominant language. Equally, it may be that a smaller
Stroop effect would be found for less-proficient bilinguals,
since the meaning of the color word would be less automati-
cally activated and therefore less interfering. However,
Bialystok et al. (2008a) considered that possibility and divided
each of the monolingual and bilingual groups into subgroups
based on the speed with which they read the name of the color
word when it was written in black ink. The idea was that faster
reading times should lead to more interference and therefore a
larger Stroop effect. Therefore, comparing the fast bilingual
readers with the slow monolingual readers should reduce the
size of the Stroop effect, possibly reversing the direction.
Nonetheless, the analysis showed that bilinguals continued to
record a smaller Stroop interference effect than did monolin-
guals, even when considering only the bilinguals for whom
reading the English words was the most automatic.
The facilitation effects found for bilinguals in the Stroop
task might be interpreted as a bilingual disadvantage. Increased
facilitation effects have been found in monolinguals with
Alzheimer’s disease when compared with healthy controls
(Spieler, Balota, & Faust, 1996) and in children when
compared with adults (Wright & Wanley, 2003). The disadvan-
tage view of facilitation is that these effects indicate increased
118 Bialystok et al.
inadvertent focus of attention on the word during color
naming (MacLeod & MacDonald, 2000; Spieler et al., 1996).
Note, however, that the version of the Stroop task used in
experimental research is not exactly the same as the version
used in the clinic. For example, experimental studies typically
use raw interference scores whereas clinic assessment relies on
a speed-adjusted interference score. Similarly, congruent trials
are typically not administered in clinical settings. Therefore,
more information about precisely what types of bilinguals
exhibit a Stroop advantage, the origin of bilingual effects on the
Stroop task, and perhaps most importantly the distribution of
scores is needed.
5. Bilingualism in the World
The constant use of two languages is an experience that leaves
its mark far beyond the immediate and obvious domain of
communication. As we have seen in this review, it modifies the
level to which some features of linguistic systems may be
learned and the way in which they are used; it enhances aspects
of cognitive processing, particularly those involved in the exec-
utive control system; it recruits, and most likely adapts, the
neural networks involved in the control of nonverbal processes
to modify their use for verbal processes; and it intervenes in
clinical assessment by presenting a profile that may not be
accurately captured by monolingual norms. These are signifi-
cant consequences that cover both individual (e.g., cognitive
development and decline) and public (e.g., assessment and
dementia) outcomes. Given this context, the questions posed
in this final section concern the implications of bilingualism for
public policy decisions, especially perhaps in the areas of
education and health care. The current prevalence (and rapid
growth) of bilingualism in today’s highly interconnected world
make these questions relevant and urgent. In light of the
dramatic numbers noted in the Introduction, we conclude by
addressing specific questions about bilingualism that concern
both individual and social issues.
Bilingual education
Not all parents have the opportunity to expose their children to
a second language at home, yet many understand the value of
being able to communicate in another language. One option
in these cases is to find alternatives in formal education.
A popular program in this regard is immersion education.
In these programs, school instruction takes place in a language
that is not the language of the home or the community (e.g.,
French instruction in English Canada, Spanish instruction in
the United States) and children are expected to use this lan-
guage in all their communication with teachers and friends
while at school. Therefore, children develop fairly high compe-
tence in this language, even though they do not typically
achieve the level of a native speaker (for review, see Genesee,
1985; Johnson & Swain, 1997). But does this limited school
exposure make these children ‘‘bilingual’’ by the criteria used
in this review and, therefore, affected by the cognitive and lin-
guistic outcomes we have described?
The question can be cast more broadly as an inquiry regarding
the degree of bilingualism necessary for the outcomes observed
for more fully functioning bilinguals. There is little evidence on
this point, but the available studies suggest that there is a correla-
tion between the degree of bilingualism and the extent of the
impact of bilingualism on cognitive and linguistic processing.
Early studies with children in French immersion programs
showed that both metalinguistic (Bialystok, 1988) and cognitive
(Bialystok & Majumder, 1998) outcomes for these children were
between those found for monolingual children and those found for
bilingual children who were fully fluent in both languages. More
generally, Luk (2008)compared 120 bilingual adults withvarying
degrees of bilingualism to a group of 40 monolinguals on linguis-
tic and cognitive outcomes and again found larger effects to be
associated with greater degree of bilingualism.
Extending this pattern to education, it is reasonable to
assume that there is a cumulative effect of learning language
that, at least in the intense environment of immersion pro-
grams, confers some of the cognitive advantages on children
even if they do not become highly fluent speakers. Importantly,
there are few if any costs of immersion education for most chil-
dren, although individual cases may present special challenges
that need to be considered.
More languages, more benefits?
Bilingualism, as we have explained, leads to specific benefits
in cognitive processing, and even the limited bilingualism that
comes from immersion education produces some minimal form
of this effect. By the same logic, then, does trilingualism lead to
even greater benefits than bilingualism, acting as something
like super-bilingualism? The evidence on this point is scant.
An interesting study by Kave´ et al. (2008) compared general
cognitive level in a large sample of older adults living in Israel
as a function of how many languages they spoke (there were no
monolinguals in the group). They reported significantly higher
maintenance of cognitive status in older age in trilinguals than
in bilinguals, and even greater maintenance by multilinguals
who spoke four or more languages than by trilinguals, although
the measure of cognitive level they used was not very precise.
Similarly, others have reported later age of onset of
Alzheimer’s disease in multilinguals as compared with bi- and
trilinguals, as we will describe (Chertkow et al., 2010). How-
ever, perhaps for bilinguals but almost certainly for
multilinguals, it is possible that people who are able to maintain
knowledge of multiple languages may start out advantaged in
certain ways. It is too early to conclude what the effect of know-
ing more than two languages might be on cognitive outcomes.
A different kind of outcome can be found in language
learning. Monolingual children learning their first language
sometimes use a strategy of disambiguation to rapidly
figure out the meaning of new words by assuming that each
object has one unique name, as discussed in Section 1. However,
Byers-Heinlein and Werker (2009) extended this idea and
Bilingual Minds 119
compared 1½-year-old children who were being raised in
monolingual, bilingual, or trilingual homes. The results
showed a strong reliance on this disambiguation strategy by
monolingual children, a marginal and nonsignificant use of the
strategy by bilingual children, and no evidence at all for this
strategy in trilingual children. Thus, the number of languages
in the environment modified children’s expectations about
words and their meanings, possibly setting the stage for
different paths of language learning.
Bilingual aphasia and its treatment
Aphasia (word-finding difficulties) is the commonest outcome
of stroke, and yet our understanding is largely restricted to
monolingual speakers, whereas a significant portion of stroke
patients are bilingual—a proportion that is set to increase.
Clinical management is hampered because there is no current
basis for predicting speech-production difficulties following
stroke in bilingual speakers. Recovery patterns are diverse
(Green, 2005; Paradis, 2004): For instance, both languages may
recover to the same relative premorbid level (parallel recov-
ery), one may recover better than another, or the progressive
recovery of one language may impair the recovery of the other.
Without an understanding of the causal bases of these recovery
patterns, including the nature of the control processes involved,
there can be no principled basis for treatment and no rational
basis for identifying the resources required for treatment. For
instance, if treatment in one language (e.g., the L1 or current
dominant language) transfers to another, then monolingual
speech therapy could help in the recovery of both languages.
However evidence on this point is equivocal, largely because
there are few well-controlled studies (see Kohnert, 2009, for
a recent review). Even the decision to treat in one language
rather than two reflects an untested assumption that may or may