Selection and control in bilingual comprehension and production

Full text

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
483
[483--507] 13.4.2015 7:31AM
Part VI
Control

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
484
[483--507] 13.4.2015 7:31AM

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
485
[483--507] 13.4.2015 7:31AM
21
Selection and control in
bilingual comprehension
and production
Judith Kroll, Jason Gullifer, Rhonda McClain, Eleonora Rossi,
and María Cruz Martín
21.1 Introduction
Two observations about bilingualism have dominated the recent litera-
ture. One is the discovery that the bilingual’s two languages are continu-
ously active, even when the task and context require that only one
language be used. The co-activation of the bilingual’s two languages has
been observed in the simplest of tasks, when naming words (e.g.,
Schwartz, Kroll, & Diaz, 2007), naming pictures (e.g., Costa et al., 2000),
and recognizing spoken words (e.g., Marian & Spivey, 2003b). But it has
also been shown to persist when these tasks are placed in sentence context
(e.g., Libben & Titone, 2009) and when features of the two languages are
markedly different, as in the case of two languages that utilize different
written scripts (e.g., Hoshino & Kroll, 2008) or when one language is
written and the other language is signed (e.g., Morford et al., 2011). The
observation that it is not simple for adult bilinguals to exploit cross-
language differences that might serve to separate the two languages
makes the problem of language selection complex. It suggests that there
is a high level of cross-language activation and potential competition when
the intended language is selected. The fact that some groups of bilinguals
code switch, changing language while in the midst of an utterance, pro-
vides additional support for the claim that the parallel activation of the
two languages is not an obstacle, but a feature that can be exploited.
The second observation is that experience in using two or more lan-
guages has consequences not only for language processing itself but also
The writing of this chapter was supported in part by NIH Grant HD053146 and NSF Grants BCS-0955090 and
OISE-0968369 to J. Kroll, by NSF Grant BCS-1226471 to J. Kroll, R. McClain, and E. Rossi, and by NSF Grant
BCS-1251896 to J. Kroll and J. Gullifer.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
486
[483--507] 13.4.2015 7:31AM
for cognition and the brain. Bilinguals, at all points across the lifespan,
have been shown to differ from monolinguals in their ability to negotiate
competition across alternative responses, to switch from one task to
another, and to ignore irrelevant information (e.g., Bialystok, Craik, &
Luk, 2012). Critically, the brain networks that enable cognitive control
are different in bilinguals and monolinguals, with evidence that bilinguals
are able to engage control mechanisms more efficiently than their mono-
lingual peers (e.g., Abutalebi et al., 2012; Gold et al., 2013). What is notable
in these studies of cognitive control is that the tasks that have been used to
assess cognitive performance and brain activation are themselves not
language tasks. Rather, they are non-verbal cognitive tasks that induce
conflict (e.g., Stroop tasks, flanker tasks, the Simon task).
1
The initial interpretation of these two set of observations was that they
are related. Bilinguals must negotiate the stream of competition across
their two languages and experience in doing so creates expertise in the
realm of cognitive control. Indeed, we have used the metaphor of the
juggler to describe the situation that bilinguals face (e.g., Kroll et al.,
2012). Learning to juggle two languages is hypothesized to produce skill
in other cognitive domains in which similar types of competitive processes
are manifest. But what aspect of linguistic athleticism produces these
changes in cognition and the brain? Bilinguals do many different things
with language and it is not at all clear that all or even most of them
necessarily have any sort of direct correspondence with the observed
cognitive and neural consequences.
Brain imaging studies show that the bilingual’s two languages share the
same neural tissue (e.g., Abutalebi, Cappa, & Perani, 2005). When there are
differences in patterns of brain activation for the first (L1) and second (L2)
languages, they are likely to reflect the level of proficiency or skill asso-
ciated with the two languages, with the L2 more likely to require addi-
tional cognitive control (e.g., Abutalebi & Green, 2007). The requirement to
engage control mechanisms in using the L2 and in regulating the activity
of the L1 seems likely to be the source of the observed consequences of
bilingualism. At the same time, there is no accepted model to date of how
language processes differentially recruit cognitive resources to enable
fluent performance and to produce domain-general consequences.
In the recent literature, there are a set of proposals that begin to address
the broader issues of how bilingual experience shapes the neural networks
that support language and its cognitive consequences (e.g., Baum & Titone,
2014; Green & Abutalebi, 2013; Kroll & Bialystok, 2013). In the present
chapter we focus more narrowly on the processes that enable bilinguals to
select the intended language when they read text, listen to spoken
1
There are a few recent papers that attempt to catch the cognitive consequences of language processing as it takes
place online (e.g., Blumenfeld & Marian, 2011; Martín, Macizo, & Bajo, 2010; Wu & Thierry, 2013). We discuss this work
later in the chapter.
486 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
487
[483--507] 13.4.2015 7:31AM
language, or plan speech in one language or the other. We return at the
end of our discussion to consider how mechanisms of selection, which, as
we will show, may differ for comprehension and production, may hold a
clue to better understanding the cognitive consequences of multiple lan-
guage use. Our review and analysis considers evidence primarily from
lexical comprehension and production because that is where there has
been the greatest number of studies and where these processes can be
compared systematically. There is a valid and important question to be
asked about how the claims about lexical processing might be extended to
the comprehension and production of phrases, sentences, and discourse,
but that question is beyond the scope of the present chapter.
As we noted at the onset, there is compelling evidence in all domains
that bilinguals activate both languages in comprehension and in produc-
tion. Curiously, there has been little direct comparison of the evidence in
comprehension and production. Cross-language co-activation may be
resolved quite differently in comprehension than in production because
the intended language may not always be explicitly selected in compre-
hension. In language production, selection is forced by the fact that, at
least for unimodal bilinguals who speak each of the two languages, there is
only one output channel, making it is physically impossible for a bilingual
to produce both languages at the same time. In contrast, in comprehen-
sion, it may not be a requirement to resolve co-activation fully. In this
sense, language selection may not be necessary, or necessary to the same
degree, during language comprehension. The goal of our chapter is to
begin to compare these language processes as means to identify the
scope of the control mechanisms that are engaged when bilinguals use
each language alone or both languages together in mixed language
contexts.
21.2 Bilingual comprehension
The evidence for parallel activation during language comprehension
comes primarily from processing effects involving words with cross-
language overlap, or words that are ambiguous between two languages,
such as cognates and interlingual homographs (for a review, see Dijkstra,
2005). Cognates – translations which share both lexical form and mean-
ing – often exhibit facilitated processing compared to lexically matched
non-cognate control words due to the overlap at all levels (orthography,
phonology, and semantics). In contrast, homographs are words that share
lexical form but not meaning. To illustrate, the word “hotel” is a cognate in
Dutch and English, with virtually the same form and meaning in both
languages. In contrast, the word “room” means cream for your coffee in
Dutch and is pronounced differently in the two languages. Homographs,
also called false friends, often exhibit slower processing compared to
Selection and control in comprehension and production 487

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
488
[483--507] 13.4.2015 7:31AM
lexically matched controls, but depending on the context in which they
are presented, they can exhibit facilitated processing (e.g., Dijkstra, Van
Jaarsveld, & Ten Brinke, 1998; Lemho¨ fer & Dijkstra, 2004). Presumably,
homographs facilitate processing at the levels of lexical form, but the
conflicting semantics create competition that produces interference. The
form of the resulting homograph effect will then depend on the extent to
which task demands require different levels of processing. Theoretically,
the facilitation observed in cognate and homograph processing is pre-
sumed to stem from heightened activation levels, a decreased threshold
for activation, increased functional frequency, common lexical storage, or
a combination of these factors (e.g., Dijkstra, 2005). Regardless of the exact
locus of the facilitation, all accounts attribute the effects of cross-language
interaction to the overlap of lexical codes. Cognate and homograph effects
can occur only for speakers of two or more languages and not for mono-
lingual speakers of any language alone; they are considered the hallmark
of language co-activation or non-selectivity. Models like the Bilingual
Interaction Activation Model (BIA+) proposed by Dijkstra and van
Heuven (2002), assume that bottom-up processes govern the earliest stages
of word recognition, increasing the activation of all form-related codes
regardless of the language that is actually presented.
Perhaps the most straightforward way to investigate language selection
during comprehension is to ask whether bilinguals exploit cues to selec-
tively attend to the language in use. If language cues enable selection to
occur in advance or to reduce cross-language activation that is in process,
then bilinguals should be able to function as monolinguals. That is, the
presence of cross-language ambiguity should have little consequence for
processing. For example, a Chinese–English bilingual reading text in
English, knows immediately that the text is not in Chinese and, in theory,
should be able to shut down the activation of phonology or meaning that is
shared across the two languages. Likewise, the context in which a bilingual
is processing spoken or written information should provide higher level
cues to better enable the selection of the intended language. The evidence,
overall, provides surprisingly little support for the notion that bilinguals
can utilize cues of this sort to select the intended language in advance. The
situation is counterintuitive given the great number of potential language
cues. But there are also some differences in the evidence for the compre-
hension of written text vs. spoken language and we consider below how
the modality of language processing may affect language selection.
21.2.1 The search for cues: modulating non-selectivity in
comprehension
A logical approach in the search for language selection in comprehension
is to determine whether there are cues in the input or in the context of
language use that can modulate markers of language co-activation. If a cue
488 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
489
[483--507] 13.4.2015 7:31AM
is successful in reducing influence of the unintended language or enhan-
cing the activation of the intended language, then the presence of that cue
should eliminate cognate and homograph effects. Language cues can fall
into at least two different categories, although these categories may not be
mutually exclusive. The first category consists of factors that function as a
cue because they exploit cross-language differences. The Chinese–English
reader encountering text in English is an example because of the differ-
ence in written script between Chinese and English. One need not be
bilingual to know that the English text is not Chinese but the question
here is whether the bilingual can use that information to guide processing.
The second category includes cues that capitalize on higher order contex-
tual or linguistic information that may provide a pointer to the upcoming
language. If bilinguals can exploit the presence of contextual cues, then
processing text or listening to speech in one language alone should pro-
vide sufficient information to restrict processing to that language. As we
will show, although there are some examples of language selectivity (e.g.,
Ju & Luce, 2004), more of the evidence demonstrates that bilinguals have
difficulty in using this information to restrict processing (e.g., Van Assche
et al., 2009).
21.2.2 Can bilinguals exploit language-specific features?
Examples of cross-language differences abound during language compre-
hension. One difference that should theoretically influence selection in
word recognition includes the distinct lexical form of words in each
language. While cognates and homographs share lexical form across the
two languages, they also exhibit subtle differences in their degree of form
overlap. The phonology of the bilingual’s two languages is never identical
and the orthography can overlap to varying degrees. In the case of
different-script bilinguals, there may be phonological overlap between
cognates or homographs but no overlap in the orthography. In the case
of bimodal bilingualism, where one language is written or spoken and the
other signed (e.g., American Sign Language and English), the two lan-
guages are structurally distinct. Yet in each case, there is evidence that
readers activate information, including the translation equivalent, in one
language when reading the other (e.g., Morford et al., 2011; Thierry & Wu,
2007). Although the magnitude of cognate and homograph effects in
languages that share orthographic and phonological codes appears to be
modulated by the degree of overlap (e.g., Dijkstra, Grainger, & van Heuven,
1999; Schwartz, Diaz, & Kroll, 2007), in language pairings that do not share
lexical form, distinct cross-language features do not function as a cue to
enable language selective processing, at least when information is pre-
sented in written form.
The evidence on spoken word recognition is more mixed with respect to
cue sensitivity than the evidence on visual word recognition. Some studies
Selection and control in comprehension and production 489

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
490
[483--507] 13.4.2015 7:31AM
have reported a pattern that is consistent with the language non-selectivity
revealed in visual word recognition. Lagrou, Hartsuiker, and Duyck (2011)
compared the performance of Dutch–English bilinguals on an auditory
lexical decision task in which they had to decide whether spoken tokens
were real words in one language alone. The tokens included Dutch–
English homophones, words that sound similar but that have different
meanings in the two languages. The cue that was introduced was the
accentedness of the speaker’s voice, with either a native English or native
Dutch speaker producing the tokens to be judged. Lagrou et al. found that
bilinguals were indeed sensitive to the presence of accentedness overall,
but the accentedness of the spoken words did not affect the magnitude of
the homophone effect. Bilinguals were slower to judge homophones than
control words regardless of whether the accentedness of the spoken voice
provided a cue to the language spoken. The overall sensitivity to accent-
edness suggests that they were not deaf to the cue – they indeed processed
it. But they did not use it selectively to interpret the input as coming from
one language alone.
Other studies of spoken word recognition have used the visual world
paradigm (Allopenna, Magnuson, & Tanenhaus, 1998) to examine cross-
language effects. In the visual world paradigm, participants see a real or
computer-generated grid of objects in front of them while they hear a
spoken word. Their task is to point to or click on the object that is
named by the spoken word. The critical manipulation is whether among
the objects in the visual display, there is another object whose name is
phonologically similar to the target object (e.g., candy–candle). Eye move-
ments are tracked while the task is performed and the basic finding is that
listeners fixate longer on objects whose names are phonological competi-
tors with the target name. Marian and Spivey (2003b) reported that there is
an effect of phonological competitors not only within languages, but also
between languages when bilinguals perform the task in one of their two
languages. Like other word recognition results in the visual domain, these
cross-language effects are typically larger and more robust when perform-
ing the task in the L2, but the fact that they are observed when performing
the task in one language alone reveals the fundamental non-selectivity of
word recognition.
Subsequent studies using the visual world paradigm have shown that
subtle phonetic differences can alter the selectivity of the system. For
example, Ju and Luce (2004) found that language-specific differences in
voice onset time (VOT) changed the degree to which participants fixated on
cross-language homophone distractors. When Spanish words were pro-
nounced with a Spanish VOT, English competitor items were not fixated
on any more than compared to control words. In contrast, when Spanish
words were pronounced with English VOTs, English distractors were then
considered by the participants as evidenced by an increase in the number
of fixations compared to controls. This finding suggests that Spanish
490 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
491
[483--507] 13.4.2015 7:31AM
phonetic differences may have allowed the participants to selectively
access Spanish without influence from English (see also Weber & Cutler,
2004). On the surface, the selectivity imposed by the presence of language-
specific cues to the phonology in the visual world task would seem to be at
odds with the lexical decision results of Lagrou et al. (2011) although the
recognition tasks performed in each case are somewhat different. This
leaves open the possibility that some types of cues may allow for selective
access but it is not yet clear under what circumstances or over what time
course.
In addition to orthographic and phonological ambiguity, language
ambiguous words such as homographs or homophones frequently differ
in their grammatical properties within each language. It appears that
bilinguals may be able to capitalize on these differences in word class to
exploit them as a language cue. Sunderman and Kroll (2006) found that
lexical form interference could be eliminated in a translation recognition
task (i.e., decide whether two words are translations of one another) when
the two words differed in their grammatical class. When two words were
not translation equivalents but the translation of one resembled the other
(e.g., hombre (man)–hambre (hunger) in Spanish) there was interference only
when the grammatical class of the two words matched. Likewise, Baten,
Hofman, and Loeys (2011) showed that word class modulated the degree of
language co-activation in a lexical decision task. They found a facilitatory
homograph effect in lexical decision only when the meaning of the homo-
graph shared grammatical class with its translation. In each of these
examples, it is not clear whether the locus of selection occurs early or
late in processing. According to models such as the BIA+ (Dijkstra & van
Heuven, 2002), the earliest stages of bilingual word recognition should be
data driven and higher-order cues, such as grammatical class, should only
influence selection processes at a relatively late point in processing. In
future research it will be critical to use methods such as eye tracking and
event-related potentials (ERPs) that permit a sensitive analysis of the early
time course of processing to answer this question. However, like the VOT
results in the visual world studies, the report of a pattern of processing that
is influenced by grammatical class also suggests that bilinguals may be
able to exploit cues to language status under some circumstances.
21.2.3 Can bilinguals use context to guide language-specific
lexical access?
Words do not typically appear in isolation, they are embedded within
multiple contexts: in sentences, in text, and in extended discourse. Much
of the research on language selection in bilingual word recognition has
focused on how different contexts of use (particularly list composition and
sentential context) might influence non-selectivity. Cognate facilitation
seems to be relatively robust in the face of different language contexts
Selection and control in comprehension and production 491

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
492
[483--507] 13.4.2015 7:31AM
(e.g., Dijkstra et al., 1998). Dijkstra et al. showed that cognate effects were
present in an isolated lexical decision task regardless of the specific list
composition, whether the lists contained words all in one language, or
whether the list contained other-language distractor words. This was
among the first studies to suggest that top-down contextual information
was not utilized by bilinguals to control activation of the two languages.
With an isolated presentation, it is perhaps not surprising that both
languages will be activated, especially for words with a high degree of
cross-language overlap, such as cognates. In theory, isolated presentation
heightens the ambiguity of language membership of a word. A Spanish–
English bilingual who sees the cognate word “bus” presented alone would
not know whether to activate Spanish or English, unless it were possible to
utilize top-down information about the language membership of other
words in the list. However, everyday language use includes many discourse
and sentential features that go beyond list composition and that might
function as language cues. A question then is whether a richer linguistic
context, such as a sentence context, might provide bilinguals with infor-
mation to predict the language of the upcoming words.
Quite surprisingly, even for words embedded in unilingual sentences,
there is overwhelming evidence that bilinguals activate both language
representations (e.g., Libben & Titone, 2009; Schwartz & Kroll, 2006; van
Hell & de Groot, 2008). Duycket al. (2007) showed that when Dutch–
English bilinguals read English sentences naturalistically cognate words
were fixated on for less time compared to matched control words. This
pattern of fixation differences originated during very early measures of
reading time and persisted until later measures, suggesting that both
languages were activated throughout the entire time course of proces-
sing, despite the presence of a coherent sentence context all in one
language. When bilinguals recognize words within sentences, cognate
and homograph effects are still present. In visual word recognition within
sentence context, eye movements have been used to track lexical co-
activation throughout the entire time course of recognition (e.g.,
Libben & Titone, 2009). Data indicate that both languages are activated
from the initial stages of word recognition that track orthographic encod-
ing and lexical access, and they remain active until later stages that
include semantic integration. This suggests that the presence of a coher-
ent sentence context by itself does not cause lexical access to become
language selective. Indeed, a few studies have shown that these cross-
language effects can be observed even when bilinguals are reading in the
native language alone (e.g., Van Assche et al., 2009). The only exception to
language non-selectivity has been reported in the presence of sentences
that are highly semantically constrained, with highly constrained con-
texts reducing or eliminating language non-selectivity (e.g., Libben &
Titone, 2009; Schwartz & Kroll, 2006; van Hell & de Groot, 2008; but see
Van Assche et al., 2010).
492 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
493
[483--507] 13.4.2015 7:31AM
Another potential cue is in the context of usage. Socio-contextual situa-
tions which require only one language might allow bilinguals to selec-
tively access that language whereas situations in which two languages are
required might show evidence of co-activation. This finding would be
consistent with Grosjean’s (2001) language mode hypothesis, the idea
that the activity of the bilingual’s two languages varies along a continuum
from monolingual mode, with one language used primarily, to bilingual
mode, where both languages are actively used. More recently, Green and
Abutalebi (2013) have extended this hypothesis into the Adaptive Control
Hypothesis stating that the context in which bilinguals acquire and speak
their second language may fundamentally determine how language selec-
tion proceeds. Hence, different language contexts (e.g., unilingual context
vs. code-switching context) are proposed to lead to distinct patterns of
language selection and to place differential demands on the cognitive
and neural processes that support each language. Surprisingly, research
on word recognition in isolation and in sentence context suggests that
context plays less of a role in initial language selection than might be
expected. For example, Gullifer, Kroll, and Dussias (2013) showed that
cognate effects were present and of the same magnitude for words recog-
nized within sentences regardless of whether the language of the sentence
context was consistent within the experiment or whether the language of
the sentence context switched every two sentences. The magnitude of the
cognate effect for cognate translations embedded within the sentence did
not depend on whether the previous sentence was in the same or in a
different language and there was virtually no switch cost to process target
cognates in sentence contexts that had changed language from those that
had not.
In comparison to cognates, homographs appear to be more sensitive to
the language context. After sufficient experience with a unimodal context
in an experiment, bilinguals show effects of “zooming in” to the intended
language. Following sufficient exposure to the language of the sentence
context, homograph effects are diminished or eliminated (e.g., Elston-
Gu¨ ttler, Gunter, & Kotz, 2005). Likewise, in research on isolated word
recognition, the magnitude of homograph effects are more closely related
with aspects of list composition. Homograph effects are greatest, for
example, when other language distractor items are included in the list of
words for a unilingual lexical decision task (Dijkstra et al., 1998). It may be
the case that homographs simply offer a more sensitive measure of rela-
tive activation levels of each language and the control mechanisms that
regulate those activation levels compared to cognates. In a dynamic system
that is under the control of executive function mechanisms and must take
into account the context of language usage the contrast in sensitivity to
context makes sense. Parallel activation of cognates is almost always
beneficial to processing providing a facilitatory effect, where the parallel
activation of homographs imposes a conflict that is only occasionally
Selection and control in comprehension and production 493

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
494
[483--507] 13.4.2015 7:31AM
beneficial to processing in the face of conflict in meaning across languages.
Hence, for homographs, bilinguals who are particularly skilled in control
of both languages may be able to regulate the inhibition of competitors
that are detrimental to fluent processing. By comparison, the resonance
created by the convergence across lexical codes for cognates may be driven
more directly by bottom-up processes that are immune to the influence of
control mechanisms (for additional evidence on access in sentence con-
text, see Titone et al., 2011).
The persistence of language non-selectivity, even in the presence of rich
context, is counterintuitive and puzzling. Indeed, the evidence reviewed
earlier, suggesting that there are no switch costs in lexical access when the
entire sentence context switches back and forth, is surprising. But what
happens when words switch from one language to the other? Language
switching has been used as a means to understand the nature of the
selection mechanism. Lexical switch costs have been documented in a
number of word recognition studies (e.g., Thomas & Allport, 2000).
Typically a switch cost is observed, with longer response times on trials
following a language switch than following a same-language trial. That
result is notable because the persistent activity of the two languages that
we have documented might have led to a prediction of no switch costs if
the bilingual is always in a functionally mixed language context. Critically,
Thomas and Allport found effects of language-specific orthography but no
modulation of those effects in the switch and no-switch conditions, sug-
gesting that language selection affects late stages of processing in word
recognition. As we will see in reviewing the evidence on bilingual produc-
tion, the pattern of switch costs in comprehension stands in contrast to the
one reported for production. While the explicit mechanism of language
selection during comprehension is yet to be unequivocally identified,
there is emerging research elucidating the link between a bilingual’s
capacity for inhibitory control and language selection and co-activation
during comprehension. We turn to that evidence next.
21.2.4 How do bilinguals control selection processes in
comprehension?
The fact that cross-language activation is observed even when bilinguals
are required to only use one of their languages raises the question about
how they select the language they need according to the context. Much of
the recent research on bilingual language processing has focused on
understanding the control mechanisms that allow them to overcome the
negative influence of activating their two languages, and importantly, how
the consequences of this parallel activation impact domain-general cogni-
tive processes.
Two important remarks are relevant for language selection: first, given
that language co-activation occurs, the presence of between-language
494 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
495
[483--507] 13.4.2015 7:31AM
competition introduces the need for a control mechanism that regulates
the activation of the non-target language in order to correctly select the
intended one; second, the non-selective activation of the two languages
may give rise to differences in lexical access and the degree of between-
language competition that depend on several factors that influence bilin-
gual processing (Green & Abutalebi, 2013). For example, lexical selection
may vary depending on the locus of cross-language activation, and so cross-
language interference and the control effects may occur at different levels
of language processing (Kroll, Bobb, & Wodniecka, 2006). Considering this,
it is important to note that the consequences of the cross-language activa-
tion might be not the same for comprehension and production. Production
is a conceptually driven process, in which the concept of a given word is
activated before the word form and phonological properties of possible
target words are available (e.g., Levelt, 1989). Comprehension, as we have
documented in the studies we have reviewed above, is a bottom-up pro-
cess, in which the orthography and phonology drive later conceptual
access (e.g., Dijkstra & van Heuven, 2002). The implication of these funda-
mental differences between production and comprehension is that the
locus and the control mechanisms for lexical selection may differ across
both domains.
While there is no agreement about the nature of the control mechan-
isms that enable language selection, a recent body of evidence suggests the
presence of inhibitory processes (for a review, see Kroll et al., 2008).
Although most of the empirical evidence supporting inhibitory processes
in bilingual processing comes from the language production domain
(Costa & Santesteban, 2004; Meuter & Allport, 1999) recent studies also
demonstrates the involvement of inhibitory processes in bilingual lan-
guage comprehension (Macizo, Martı
´n, & Bajo, 2010).
As reviewed in the earlier section on language non-selectivity, most of
the studies on cross-language interactions have used the general strategy
of examining words that share lexical, orthographic, or phonological
properties in two languages (e.g., false friends or homographs, cognates,
and homophones). So, for example, in visual and spoken word recognition,
competition may arise among lexical neighbors with either similar ortho-
graphy or phonology. Following this strategy, Martı
´n, Macizo, and Bajo
(2010) investigated the role of the control processes involved in language
selection using a paradigm that allows observing the cross-language acti-
vation and how it is resolved. They asked Spanish–English bilinguals to
perform semantic relatedness judgments to English word pairs. On the
critical trials, interlingual homographs were presented among the word
pairs (e.g., the word “pie” means foot in Spanish and cake in English). On
the following trial, the word pair included the English translation of the
Spanish meaning of the homograph. Bilinguals were slower to judge word
pairs that contained a homograph (e.g., pie–toe). Critically, after responding
to homographs, bilinguals slowed their responses when its Spanish
Selection and control in comprehension and production 495

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
496
[483--507] 13.4.2015 7:32AM
meaning became relevant in the subsequent trial as compared to a control
word pair (e.g., foot–hand). The interference effect found in the first trial
was taken as an index of cross-language activation. Given the interference
effect stemming from the parallel activation of the two homograph mean-
ings, the bilinguals appeared to resolve the interference by suppressing
the non-target and competing homograph meaning in order to select the
appropriate one.
Martı
´n et al. (2010) focused on the interval between the time in which
interference was produced and the presentation of the irrelevant homo-
graph meaning as a means to map out the time course of resolving inhibi-
tion. They found that the inhibitory effects were relatively short lived since
they were observed in a time interval between 500 ms and 750 ms. After an
interval of 750 ms, they seem to have recovered from inhibition. These
findings indicate that inhibitory control processes are involved in the
resolution of cross-language competition in comprehension, and this inhi-
bition has transient effects. A critical point is that overcoming inhibition
may take time and it can impose a cost in bilinguals’ performance. The
effects we have described were observed when bilinguals processed words,
out of context, and under artificial experimental conditions. Bilinguals
rarely use their two languages in this manner. In future work it will be
important to ask whether these time-constrained processes are manifest
when bilinguals comprehend language in higher-level context and, if so,
whether context modulates the presence and time course of language
selection.
Converging support for the idea that there is a brief period of inhibitory
control that can be caught on the fly in comprehension has been reported
by Blumenfeld and Marian (2011). They developed a variant of the visual
world paradigm that has been used extensively as a means to investigate
spoken word recognition. On a first trial, native English-speaking bilin-
guals and monolinguals were presented with a display in which they had
to identify a word in English, the L1 for both groups. Their eyes were
tracked while they performed the identification task. One of the pictures
displayed on the first trial had a name that was a phonological competitor
to the target English word that was spoken. Blumenfeld and Marian
reported that both bilinguals and monolinguals experienced within-
language competition on the first trial, revealed in the eye movement
record, with somewhat longer fixations on the phonological competitor
than on the control pictures. Critically, on a subsequent trial, the same grid
was presented but without pictures of objects. Instead, in one of the four
corners of the grid, an asterisk appeared and the participant was asked to
click on the asterisk. They found an inhibitory pattern for the monolin-
guals on the second trial when the asterisk appeared in the position on the
grid that had previously held the phonological competitor. Blumenfeld
and Marian argued that the bilinguals had apparently resolved the inhibi-
tion by the time of the second trial, whereas the monolinguals had not.
496 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
497
[483--507] 13.4.2015 7:32AM
They suggest that bilinguals are more efficient at resolving cross-language
competition online, a result that is congenial with the findings of brain
imaging studies that show that bilinguals produce less activation than
monolinguals in brain areas such as the anterior cingulate cortex (ACC)
that are implicated in the resolution of competition in non-linguistic tasks.
Without a fine-grained analysis of the time course of processing, it is
difficult to compare the Blumenfeld and Marian results directly to those
of Martı
´n et al. (2010), and of course the tasks are quite different, spoken
word recognition in one case and a semantic relatedness judgment in the
other. However, in both studies there is evidence that suggests that bilin-
guals cannot turn off the initial bottom-up processes that produce compe-
tition in comprehension. Rather, they learn to modulate its consequences
in the moments afterwards. Each of these studies also suggests that the
processing of inhibiting competitors is one that extends over a relatively
brief time-span in comprehension. As we will see, this will stand in con-
trast to the findings in production, where inhibitory processes may
include longer-lasting consequences for performance.
21.3 Bilingual production
Like the evidence on comprehension, studies of bilingual speech planning
show that information about both languages is active, at least momenta-
rily, when bilinguals plan to speak even a single word in one language
alone (e.g., Costa, 2005; Hanulova
`, Davidson, & Indefrey, 2011; Kroll et al.,
2006). The parallel activation of two languages occurs for speakers who are
highly proficient in both languages as well as for those who are still
learning the L2. A focus in the most recent research in bilingual language
processing is to understand what mechanisms allow bilinguals to negoti-
ate language activation and, once proficient, to make few language errors
(e.g., Gollan, Sandoval, & Salmon, 2011). Although bilinguals normally
make few errors of language, the importance of a control mechanism
has also been recognized in cases of aphasia where bilingual speakers
who suffered neurological damage cannot properly control language selec-
tion, leading to pathological language mixing (Abutalebi & Green, 2008).
21.3.1 Two views of bilingual speech planning
A number of different mechanisms have been hypothesized to allow
bilinguals to constrain selection to the intended language during speech
planning (for a recent detailed review, see Kroll & Gollan, 2014). Findings
also generally indicate that parallel activation can have multiple conse-
quences, making the locus of language selection in production variable
rather than fixed (e.g., Kroll et al., 2006), and dependent on a range of
factors including the context of the language to be spoken (e.g., language
Selection and control in comprehension and production 497

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
498
[483--507] 13.4.2015 7:32AM
immersion and the language profile of interlocutors), proficiency in the L2,
and the demands of speaking associated with the particular production
task to be performed. One alternative proposes that selection is language
specific (e.g., Costa, Miozzo, & Caramazza, 1999; Finkbeiner et al., 2006;
Finkbeiner, Gollan, & Caramazza, 2006). On this view, there may be activa-
tion of words within the language not in use but the activation of those
words does not make them candidates for selection. This first hypothesis
would require that bilinguals have a mechanism in place that allows them
to use cues to enable them to select the language as efficiently as possible,
a solution that we have called the “mental firewall” (e.g., Kroll et al., 2012).
Like the research on bilingual comprehension, external cues may be
related to linguistic features of the two languages (e.g., phonological or
lexical cues) but also to features of the context in which the two languages
were acquired and are used. However, studies of language production that
have examined the role of language cues, such as cross-language script
differences, have failed to provide evidence that bilinguals can easily
exploit the available cues (Hoshino & Kroll, 2008). Similarly, typological
differences that should provide a clear means to categorize the two lan-
guages do not appear to function effectively as cues. For example, produc-
tion studies in hearing bimodal bilinguals who speak one language and
sign the other, have also shown that there are cross-language influences
even when production engages different articulatory systems (e.g.,
Emmorey et al., 2008). It is possible, however, that script and typological
differences are subtle in the sense that activation of information pertain-
ing to script or typology does not occur in response to a trigger in the
immediate environment. Cues may only become effective as a conse-
quence of the bilingual’s lifetime experience, with collective past experi-
ences likely to be stored in memory (Jared, Poh, & Paivio, 2013), but not
accessible in the moments leading up to speech unless there is an addi-
tional contextual trigger (e.g., Zhang et al., 2013). In this sense, the evi-
dence on production would seem to mirror the results we have reviewed
for comprehension. In each case, there appears to be activation of alter-
natives related to both the target language and the language not in use. For
production, however, this observation is quite counterintuitive, because
unlike comprehension, production is initiated by a top-down process that
first engages ideas and then maps them to their respective linguistic forms.
Logically, it would seem possible to identify the intended language at an
early stage of speech planning.
If the logic of the language selective view is straightforward, providing a
clear explanation of how selectivity is accomplished is not. By a language-
selective account, there would have to be a means for sending greater
activation to the intended language and the studies to date seem to suggest
quite clearly that the intention to plan speech in one language alone is
insufficient to restrict activation to that language (but see La Heij, 2005, for
a defense of the selective position).
498 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
499
[483--507] 13.4.2015 7:32AM
The alternative view is that all activated candidates compete for
selection.
2
Superficially, this view would seem to align with the models
of word recognition like BIA+ that account for non-selectivity in compre-
hension by assuming that bottom-up activation spreads to form-related
words in both languages. But it is important to remember that the top-
down planning of speech means that the codes that are first activated are
related in meaning rather than form with the ideas to be communicated.
Those meanings may subsequently activate word forms, including transla-
tion equivalents and phonological relatives (e.g., Colome
´& Miozzo, 2010;
Hoshino & Kroll, 2008; Hoshino & Thierry, 2011), but the sequence of
processing differs in comprehension and production. Although the initia-
tion of speech planning begins at a conceptual level, the evidence suggests
that cross-language activation reaches all the way to the phonology and
even beyond, to the execution of speech that is manifest in the acoustic
representation of the spoken utterance, to the point where the unintended
word may literally be on the tip of the bilingual’s tongue (e.g., Gollan &
Goldrick, 2012).
21.3.2 Evidence on language switching and language mixing
Studies of language production often adopt a logic similar to the one
applied to word recognition so that bilingual speakers are asked to pro-
duce language ambiguous words, such as cognates. In production, unlike
comprehension, the cognate word is not actually present, so showing that
there is facilitation in naming a picture whose translation is a cognate in
the language not to be spoken provides evidence for cascading activation
all the way to the phonology of the alternative (e.g., Costa et al., 2000;
Hoshino & Kroll, 2008).
In comprehension, the effects of cross-language ambiguity are perhaps
the most compelling in revealing the architecture of the system because
language ambiguous forms affect the earliest stages of recognition. In
production, the evidence for language non-selectivity has relied more on
the consequences of language switching and/or mixing. The logic is sim-
ple. If bilinguals can plan speech in one language alone, as if they were
monolingual speakers, then a cost should be observed when there is
uncertainty about the language to be spoken or when the languages switch
from one utterance to the next. For production, this logic affects the ear-
liest stages of planning, potentially revealing the most basic features of the
architecture that underlie speech, unlike cognate facilitation effects that
necessarily reflect later processing once speech planning has already been
initiated.
2
A third alternative is the frequency lag or weaker links hypothesis that has been proposed by Gollan and colleagues (e.g.,
Gollan et al., 2008) to account for the relatively lower frequency of use of each language in bilinguals relative to
monolinguals. See Kroll and Gollan (2014) for a discussion of how this alternative contrasts with the competition for
selection hypothesis.
Selection and control in comprehension and production 499

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
500
[483--507] 13.4.2015 7:32AM
21.3.3 Costs to language switching and mixing
When bilinguals are forced to switch from one language to the other in a
naming task (e.g., name pictures or digits) there is a cost to processing,
with longer naming latencies following switch than no-switch trials (e.g.,
Meuter & Allport, 1999). In language switching tasks, a picture or number
is presented with a cue signaling in which language the stimulus is to be
named. The finding that there is a cost to naming following a language
switch might be interpreted as support for a language selective model of
bilingual production. However, Meuter and Allport and many subsequent
studies have reported that there is a curious asymmetry in switch costs,
with larger costs when switching from the L2 into the L1 than the reverse.
The greater switch cost into the L1 was initially taken as support for the
claims of the Inhibitory Control (IC) model (Green, 1998) which suggested
that production of the L2 requires inhibition of the L1 and that subsequent
production of L1 then reflects the spillover of that inhibitory process. It is
beyond the scope of the present chapter to provide a comprehensive
review of the switch cost asymmetries and there are other many discus-
sions of how these asymmetries might be interpreted (e.g., Bobb &
Wodniecka, 2013; Gollan & Ferreira, 2009; Schwieter & Sunderman,
2008). For present purposes, the critical observation is that there are
switch costs and that sometimes the more dominant L1 reveals those
costs more clearly than the less dominant L2.
A similar differential pattern has been reported for language mixing
(e.g., Kroll et al., 2000). When the language of production is uncertain, the
L1 suffers a cost and that cost can be observed to the point where even
highly L1-dominant bilinguals are slower to speak the L1 than the L2 and
where the consequences of mixing can be observed not only in behavior
but also in the earliest time course of planning revealed using ERPs and in
brain imaging patterns using fMRI (e.g., Christoffels, Firk, & Schiller, 2007;
Guo et al., 2011;. Kroll et al. (2000) found that Dutch–English speakers were
slower to produce the Dutch names of pictures when faced with the
requirement to produce picture names in English unpredictably relative
to blocked trials in which they produced words in one language only.
Critically, the time for the same speakers to produce English L2 names of
pictures was unaffected by whether picture naming was blocked or mixed,
suggesting that the L1 is active during speech planning in the L2, regard-
less of whether there is a requirement to make it active.
A language switching study that has generated a great deal of discussion
was reported by Costa and Santesteban (2004). They compared the lan-
guage switching performance of balanced and highly proficiency Spanish–
Catalan bilinguals with less proficient speakers for whom there was clear
dominance in the L1. For the L1-dominant speakers, they replicated the
switch cost asymmetry reported by Meuter and Allport (1999) but for the
balanced and high proficiency speakers, the switch costs were symmetric
500 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
501
[483--507] 13.4.2015 7:32AM
for the two languages. The different pattern for the two speaker groups
was initially interpreted as meaning that inhibitory control is required for
those who are less proficient but once high proficiency is achieved, there is
no longer a need for inhibition. Essentially the idea was that once the two
languages are equal sparring partners, they can be engaged without differ-
ential regulation. However, another result in the Costa and Santesteban
study betrays that interpretation. Although there was indeed a difference
in the symmetry of the switch costs for the different types of bilingual
speakers, there was also evidence that even for the high proficiency and
balanced bilinguals, under the conditions of language mixing that are
required by the switching paradigm, the L1 was produced more slowly
than the L2. The fact that these two sources of evidence on switching and
mixing produce different results can be understood if we assume that in
production there may be multiple components of inhibitory control (for a
discussion of these issues in the domain of translation and interpretation,
see de Groot & Christoffels, 2006). The recent research on bilingual produc-
tion has focused on this issue as a means to understand how language
selection might be achieved.
21.3.4 Components of inhibitory control in production
The forced switching that is imposed by the language switching paradigm
requires that bilinguals speak words in each of the two languages in a
sequence that is highly unnatural. Although one might argue that all
decontextualized lexical performance is unnatural to some degree, bilin-
guals often switch between two languages after speaking one language for
a period of time. Another approach to investigating selection in produc-
tion was taken by Misra et al. (2012). The idea was to enable bilinguals to
speak one language alone, the way that they might more naturally, and
then to impose a switch of language across blocks. If the switch costs that
have been documented in forced trial-to-trial switching reflect a process
that is imposed by the artificial nature of the task, then within a few trials,
bilinguals should recover from the switch. Misra et al. asked relatively
proficient Chinese–English bilinguals to name pictures in blocks of trials
in either L1 (Chinese) or L2 (English). The manipulation was only whether
naming was performed first in the L1 and then in the L2 or in the reverse
order. The pictures to be named in each language were identical, with the
prediction of priming on the second presentation of a picture relative to
the first. Misra et al. recorded ERPs while the pictures were named and also
behavioral measures of response time and accuracy. They found that when
pictures were named in the L2 following the L1, the hypothesized priming
was observed. ERPs were less negative on the second naming of a picture
than the first, suggesting that there was facilitation in processing. In
contrast, an inhibitory pattern was observed when the L1 followed the
L2. Not only was there not a pattern consistent with priming, but there was
Selection and control in comprehension and production 501

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
502
[483--507] 13.4.2015 7:32AM
a larger N2 component and greater negativity for the L1 following the L2,
consistent with the presence of inhibition for the L1. Beyond the different
patterns as a function of language naming order, the most surprising
result was that the inhibitory pattern in L1 following L2 was maintained
over the course of two blocks of L1 naming. These bilinguals had ample
opportunity to recover from any momentary inhibition imposed by speak-
ing the L2 but many trials later were still showing the consequences (and
see Phillip, Gade, & Koch, 2007, and Philipp & Koch, 2009, for evidence
from the N-2 repetition paradigm that supports a similar inhibitory
account). The use of the same pictures across languages in the Misra
et al. study did not permit an assessment of the scope of inhibition but
the pattern suggests that there is inhibition that is extended in time. As
noted earlier, there may be multiple components of inhibitory control.
Local inhibition may operate over specific words or conceptual categories
and may extend for brief periods of time. Global inhibition may engage an
entire language and last for a relatively long time.
Guo et al. (2011) used fMRI to track the patterns of brain activation when
bilinguals name in extended blocks as they did in the Misra et al. (2012)
study. The design in the Guo et al. study was identical to the blocked
picture-naming procedure in Misra et al. except that following the blocked
picture-naming trials, Chinese–English bilinguals named pictures in a
mixed language block. Critically, different patterns of brain activation
were found when comparing the blocked and mixed naming trials
(hypothesized to reflect local inhibition) and the spillover effect of naming
in different block orders (hypothesized to reflect global inhibition). The
dorsal anterior cingulate cortex (ACC) and the supplementary motor area
(SMA) appeared to play important roles in local inhibition, while the dorsal
left frontal gyrus and parietal cortex appeared to be important for global
inhibition.
In recent studies, each of these observations has been replicated in
behavioral studies that used different tasks and different language pair-
ings. For example, Van Assche, Duyck, and Gollan (2013) compared per-
formance on a verbal letter fluency task as a function of the order of the
language blocks (L1 or L2 first), the type of speaker (Dutch–English or
Chinese–English), and whether the letter cues were the same or different
across languages. They replicated the block order effect reported by Misra
et al. (2012) in that letter fluency was reduced for the dominant language
when it followed the less dominant language. The groups differed, how-
ever, in that only the Chinese–English bilinguals, but not the Dutch–
English bilinguals, showed these effects globally, regardless of whether
the letter cues were repeated or not. The pattern of results suggests that all
bilinguals, regardless of their proficiency, show evidence of inhibitory
processing, but that proficiency may determine the scope of inhibition.
Martı
´n, Bajo, and Kroll (2013) reported a study that compared the beha-
vioral performance of relatively proficient Chinese–English bilinguals
502 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
503
[483--507] 13.4.2015 7:32AM
when they named pictures in blocked or mixed language trials. The inno-
vation in this study was to add a concurrent updating task that required
speakers to listen to a continuous series of tones and to press a key
whenever three equal tones appear consecutively. The logic here was to
determine whether the addition of a dual task would selectively disrupt
inhibitory processing. The results showed that the updating task elimi-
nated the block order effect. When bilinguals named pictures in L1 follow-
ing L2 while they were performing the dual task, there was no apparent
inhibition of the L1. In contrast, a robust effect of language mixing was
observed regardless of whether bilinguals were performing the updating
task or not. Although bilinguals were slower to name pictures overall
when performing the concurrent updating task than not, the result of
interest was that the disruptive effects were differential for the two
hypothesized components of inhibitory control.
The findings on bilingual production converge on the conclusion that
bilinguals, regardless of their proficiency, inhibit the L1 to enable speech
planning in the L2. The clue first present in the Costa and Santesteban
(2004) language switching data, that there might be more than one
component of inhibitory control, has been supported by the recent stu-
dies that were designed to examine this issue explicitly. It remains to be
determined how not only language proficiency and dominance, but also
the context of language use, may modulate each of these components.
Green and Abutalebi (2013) argue that the neural networks that support
bilingual language processes are necessarily tuned differently in
response to the requirement to engage these processes differentially. So
a bilingual who is a habitual code switcher and living in an environment
in which code switching is prevalent, may engage inhibitory mechan-
isms differently than a bilingual who uses each of his or her two lan-
guages in separate environments. Likewise, individuals who are
immersed in an L2 environment may adjust the need for endogenous
control. Linck, Kroll, and Sunderman (2009) showed that university stu-
dents studying abroad for a semester produced reduced output in their L1
in a category fluency task relative to classroom learners but their L1
performance rebounded upon their return home (and see Baus,
Costa, & Carreiras, 2013, and Levy et al., 2007, for a related laboratory
version of language immersion). Critically, what these new data on lan-
guage production show is that inhibitory control processes are engaged
by the most proficient bilinguals as well as by L2 learners. It remains to be
seen which of these processes are ephemeral, producing short-term
effects that are modulated by context and language usage, and which of
them depend on characteristics of the bilinguals themselves and the
linguistic structure associated with the bilingual language pairings (see
Lev-Ari & Peperkamp, 2013, for a recent discussion of this issue for how
individual differences in inhibitory control may modulate the effects of
theL2onL1phonetics).
Selection and control in comprehension and production 503

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
504
[483--507] 13.4.2015 7:32AM
21.4 Language selection in comprehension vs. production
The research we have reviewed in this chapter demonstrates that language
selection is not a simple nor unitary mechanism. Bilinguals develop the
means to negotiate persistent cross-language activation and to resolve the
resulting competition in ways that reflect the demands that are selective in
comprehension and production. In comprehension, there appear to be
short-lived inhibitory processes whereas in production there may be con-
trol processes that require different types of inhibition and that sometimes
extend over relatively long periods of time. There is a great deal that
remains to be investigated, including the implications of the selection
mechanisms in comprehension and production for domain-general cogni-
tive performance and the way that these language processes, in and of
themselves, may change and adapt in different circumstances.
Green and Abutalebi (2013) argue there may be particular language skills
and contexts that determine how the neural networks that support lan-
guage change with experience. A topic that we haven’t mentioned in our
review but that draws on both comprehension and production in unique
ways is translation and simultaneous interpretation. In a sense, skilled
translation is a special form of extreme bilingualism. For interpreters, in
particular, there are time pressures that render the task of using both
languages quite differently than ordinary bilingualism. Although the
basic research program on these issues is relatively new, the picture that
emerges suggests that these skills sometimes modulate basic mechanisms
of language processing and sometimes do not. For example, in a study
examining lexical processing in sentence context, Iba
´n˜ ez, Macizo, and
Bajo (2010) reported that translators were less likely to engage inhibitory
mechanisms. That finding makes sense, from the perspective that the goal
for translators is to activate the other language continually, but demon-
strating that it affects their performance even when they are not actively
translating suggests that the nature of their language experience has
consequences that extend beyond specific tasks. Likewise, Christoffels,
de Groot, and Kroll (2006) showed that although interpreters have excep-
tional memory skills, a finding that in and of itself is difficult to interpret
with respect to cause and effect, their performance in basic lexical produc-
tion tasks like those we have reviewed in this chapter, is similar to other
high proficiency bilinguals. In this instance, their cognitive abilities did
not appear to modulate language processing.
3
3
In the literature on translation and simultaneous interpretation there is often discussion about the contribution of
possible self-selection factors. Only particular people seek training in these skills and it is not clear whether the observed
cognitive advantages reflect individual differences that led them to seek training or the consequences of the training
itself. Likewise, in a context such as the Netherlands, where virtually all university-educated individuals are multilingual, it
is not clear whether the absence of differences on basic language processing tasks for interpreters and ordinary
bilinguals means that there is no consequence of having these skills for language processing or whether the bilinguals to
504 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
505
[483--507] 13.4.2015 7:32AM
The findings of the studies reported above suggest that the perfor-
mance of professional translators and interpreters might not differ
from other bilinguals while performing different production tasks but,
at the same time, they show differences in the use of cognitive control in
bilingual processing. In line with this source of evidence, a recent study
shows that translators show a similar pattern of performance in compre-
hension tasks as that reported for language production. Martı
´netal.
(under review) tested a group of professional translators using the same
procedure as Macizo et al. (2010). The translators performed a semantic
relatedness task in their L2 (i.e., English) on word pairs including inter-
lingual homographs or control matched words (see the description of the
paradigm in the section about selection processes in comprehension of
this chapter). They found that the translators were similar to the control
bilingual group, showing longer latencies in response to the word pairs
containing homographs. Given that they showed the interference effect
in this kind of trial, it would be expected that after responding to homo-
graphs the translators also performed similar to the control bilinguals
showing an inhibitory effect in the subsequent trial as a sign of the
suppression of the non-target and competing homograph meaning.
However, unlike the bilingual control group, the translators did not
show the inhibitory effect as a result of solving cross-language competi-
tion. This results show nicely how a specific bilingual context can shape
the experience in the usage of control processes involved in language
processing. In the practice of their profession, a translator rarely speaks.
The input for a translator comes mainly from the comprehension domain
and the output is elaborated in written production. Although they do not
have the temporal pressure as interpreters do, they are required to
change from a language to another constantly. It is not strange then,
that when they are required to work only in one language, they perform
as other bilinguals do. However, the particular experience they have
changing between languages shows that the interplay between compre-
hension and production shapes the way in which they use cognitive
control mechanisms and it is reflected in a differential performance
solving cross-language competition (Green & Abutalebi, 2013). A question
for future research will be to determine the scope of the consequences of
particular types of bilingual language experience in both comprehension
and production for altering the manner in which cross-language compe-
tition is negotiated.
One implication of the research that we have discussed is that bilin-
guals and monolinguals might be understood to be processing language
under very different task demands and those differences might be
expected to have consequences not only for language, but also for
whom the interpreters were compared also use their languages in a unique environment that masks the contribution of
other factors.
Selection and control in comprehension and production 505

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
506
[483--507] 13.4.2015 7:32AM
cognition and brain function. Recent imaging studies (e.g., Parker Jones
et al., 2012) have reported differences in patterns of brain activation for
bilinguals and monolinguals even when they perform the same tasks in
their L1 only. Because the neural support for language processing is
shared across the bilingual’s two languages (e.g., Abutalebi et al., 2005)
and because the networks of control that are associated with them change
with language experience (e.g., Abutalebi et al., 2012; Gold et al., 2013), it
may not be surprising that differences between monolinguals and bilin-
guals emerge even when they perform the same tasks only in the native
language.
The characterization we have provided of selection and control in bilin-
gual comprehension and production might allow a reader to come to the
mistaken conclusion that these processes are independent of one another.
To the contrary, we assume that they are tightly linked (e.g., MacDonald,
2013). One context in which it is possible to begin to identify the way in
which performance in production provides a cue to comprehension comes
from studies of code switching behavior. Studies have begun to ask
whether the few markers of disfluency that occur in the context of code
switches are informative about the locus of planning and control in code
switching mode. Hlavac (2011) found that Croatian–English speakers
tended to produce few hesitation and monitoring phenomena (e.g., hesita-
tions, pauses, verbal fillers) overall in code switched speech (around 2
percent of the total utterances contained these phenomena). He also
found that a significantly greater proportion of verbal fillers preceded
than followed a code switch. More importantly, when hesitation and
monitoring phenomena occurred, they were found more often in the
context of code switched words that do not contain the phonology or
morphological patterns of the base language. The Hlavac results suggest
that presence of disfluencies may not be the result of retrieval difficulties
during code switches, but instead may serve to allow the speaker to signal
an upcoming switch to the listener.
The pattern of code switching behavior reported by Hlavac can be
understood within the framework of the Adaptive Control Hypothesis
(Green & Abutalebi, 2013). The description of the highly proficient group
Croatian–English speakers sampled for his study suggests that this group
may have adapted to a language context in which dense code switching is
the norm. The Adaptive Control Hypothesis predicts that these bilinguals
have developed a strategy that allows the speaker to opportunistically
exploit co-activation of two languages, rather than having them in com-
petition. It is our interpretation that the use of integrated word forms in
code switched speech is one sign that cooperative co-activation of L1 and
L2 exists. Code switching may therefore provide a model, and a uniquely
bilingual model, for evaluating the way in which comprehension is tuned
to production.
506 JUDITH KROLL ET AL.

C:/ITOOLS/WMS/CUP-NEW/6143877/WORKINGFOLDER/SWIET/9781107060586C21.3D
507
[483--507] 13.4.2015 7:32AM
21.5 Conclusions
In this chapter we reviewed the basic evidence for language non-selectivity
at the lexical level in bilingual comprehension and production. At a gen-
eral level, the findings in the two domains are similar. There is parallel
activation of the bilingual’s two languages in both comprehension and in
planning for production and it is difficult, if not impossible, for bilinguals
to easily exploit available cues to the language in use to override language
non-selectivity. At a more detailed level, comprehension and production
differ in some important respects, including the nature of the information
that is active and competing across languages, the scope of inhibitory
processes, and the time course over which inhibition occurs. There is a
call now to relate language processing more closely to its cognitive and
neural consequences and to identify the causal basis of these experiential
changes (e.g., Baum & Titone, 2014; Green & Abutalebi, 2013; Kroll &
Bialystok, 2013). The research findings we have discussed provide an
initial basis to pursue this question, paying closer attention to what it is
that bilinguals are doing with language when they speak and understand
one another. Like other proposals, our analysis of differences in compre-
hension and production suggests that the situation is more complex than
we understood it to be. But it also shows that using bilingualism as a tool
demonstrates the way that juggling two languages in one mind and brain
may reveal more about foundational principles than we could ever know
by studying monolingual speakers alone.
Selection and control in comprehension and production 507