Content uploaded by John W. Schwieter
Author content
All content in this area was uploaded by John W. Schwieter on Apr 11, 2021
Content may be subject to copyright.
is is a contribution from e Mental Lexicon 3:2
© 2008. John Benjamins Publishing Company
is electronic file may not be altered in any way.
e author(s) of this article is/are permitted to use this PDF file to generate printed copies to
be used by way of offprints, for their personal use only.
Permission is granted by the publishers to post this file on a closed server which is accessible
to members (students and staff) only of the author’s/s’ institute.
For any other use of this material prior written permission should be obtained from the
publishers or through the Copyright Clearance Center (for USA: www.copyright.com).
Please contact rights@benjamins.nl or consult our website: www.benjamins.com
Tables of Contents, abstracts and guidelines are available at www.benjamins.com
John Benjamins Publishing Company
e Mental Lexicon : (), –. ./ml...sch
– / - – © John Benjamins Publishing Company
Language switching in bilingual speech
production
In search of the language-specic selection
mechanism
John W. Schwieter and Gretchen Sunderman
Wilfrid Laurier University / Florida State University
Recent research on language production suggests that bilinguals shi from using
inhibitory control mechanisms to a language-specic selective mechanism dur-
ing development (Costa, Santesteban, & Ivanova, 2006). Costa et al. argue that
the robustness of the L2 lexical representations may be critical to the functional-
ity of a language-specic selective mechanism. Accordingly, in the present study
we measured the lexical robustness of a group of 54 English dominant learners
of Spanish using a verbal uency task and investigated its eect on their perfor-
mance in a picture-naming task with language switches. e results suggest that
L2 lexical robustness predicts the shi to a language-specic selective mecha-
nism during speech production. Moreover, we demonstrate a specic threshold
of lexical robustness necessary to engage the mechanism.
Keywords: bilingual speech production, inhibitory control, lexical robustness
Bilinguals have the impressive cognitive skill of being able to speak in one lan-
guage, and in one language alone. Although there are a variety of rst language
(L1) elements (i.e., pronunciation, syntax, etc.) that inuence second language
(L2) speech production (even in highly procient bilinguals), L2 speech rarely
exhibits evidence of L1 lexical intrusions (Poulisse, 1999; Poulisse & Bongaerts,
1994). is phenomenon has led researchers to conclude that since bilinguals are
able to select and produce words exclusively from one of their lexicons, there must
be cognitive mechanisms that are called upon to help them appropriately select
and produce the target word (see Costa, 2005 and La Heij, 2005 for reviews of
work done in this area).
e specic control mechanisms that underpin bilingual speech production
remain uncertain. One possible explanation is Green’s (1986, 1998) Inhibitory
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
Control Model (ICM) which, most simply put, hypothesizes that when a bilingual
speaks in one language, inhibitory control mechanisms are called upon to sup-
press the irrelevant language. Under this assumption, bilinguals are able to restrict
the lexicalization process (i.e., the process by which bilinguals access the appropri-
ate word and map it on to its representative concept) to only one language with
minimal to no lexical intrusions from the irrelevant language.
e ICM has been widely supported empirically by evidence of asymmetri-
cal switching costs during language production tasks (Costa & Santesteban, 2004;
Costa, Santesteban, & Ivanova, 2006; Meuter & Allport, 1999). When less pro-
cient bilinguals switch into their more dominant language (L1), they suer a larger
switching cost than if they switch into their less dominant language (L2). Accord-
ing to the ICM, this is expected because it will take more time to reactive the
larger L1 from suppression. Although the ICM posits that the size of the L1 may
determine the strength of the inhibition eect, it goes without saying that this is
theoretically confounded with the relative strength of L1 representations due to its
higher frequency of usage. Nonetheless, there are instances when highly procient
balanced bilinguals switch between their L1 and L2 and exhibit no costs associ-
ated with the switches given the similar language system. In these cases where
there have been symmetrical switching costs reported, researchers have argued for
alternative mechanisms to the ICM, such as a language-specic selection mecha-
nism (Costa, 2005; Costa & Caramazza, 1999; Costa, Miozzo, & Caramazza, 1999;
Costa et al., 2006; Roelofs, 1998). ese ndings have led researchers (Costa, 2005;
Costa & Santesteban, 2004; Schwieter, in press) to discuss the possibility that less
procient bilinguals rely on dierent control mechanisms than their highly pro-
cient counterparts. ese researchers concluded that the main factor driving the
shi from inhibitory control to a language-specic mechanism was prociency.
is explanation, however, did not hold for long.
Costa et al. (2006) later found that some highly procient bilinguals exhibit
similar speech patterns as less procient bilinguals (i.e., asymmetrical switching
costs). is new nding led Costa et al. to revisit and discuss a number of variables
— language similarity, age of L2 acquisition, and prociency — that do not modu-
late the underlying processes of speech control. As such, they then proceeded to
hypothesize two explanations relating to whether the language-specic selection
mechanism can operate on the lexical representations. e rst possibility was
that the robustness of the lexical representations of the L2 might be crucial for the
functionality of the language-specic selection mechanism. Secondly, they sug-
gested that it might be that the lexical representations are mediated by the rst
language and, thus, when beginning bilinguals need to access the meaning of the
L2 word, they must do so by rst activating the L1. Although the idea of lexical
mediation has been supported by Kroll and Stewart’s (1994) Revised Hierarchical
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
Model, others have taken a response-selection account of performance in picture
naming tasks as an alternative explanation (see Costa et al., 2006 and Finkbeiner,
Gollan, & Caramazza, 2006 for more information). In the end, Costa et al. le
open the two possibilities mentioned above to be explored in future research.
In the present study, we investigate whether L2 lexical robustness explains the
proposed shi away from reliance on inhibitory control to the use of a language-
specic selective mechanism. Following, Costa et al. (2006) we argue that lexical
robustness involves the familiarity with and frequency of access that leads to great-
er automaticity of retrieval of lexical items. Although others have used the term
lexical uency in describing what we and Costa and colleagues call “lexical robust-
ness,” we will use the term lexical robustness exclusively in the present study. In
fact, we do not see lexical uency and lexical robustness as interchangeable terms.
Lexical robustness captures both the strength of the representation and the ability
to access the lexical item. We also argue that lexical robustness is a specic mea-
sure of L2 prociency. Previous studies (Costa & Santesteban, 2004; Costa et al.,
2006) have measured prociency by having participants self-rate their abilities in
speaking, reading, writing and listening. ese ratings may not accurately repre-
sent L2 lexical development but rather a more global picture of prociency. Given
the lack of consistent results using these measures, we posit that a measure that
specically targets lexical robustness may capture the point at which the language-
specic selection mechanism may become functional in a way that previous global
prociency measures have failed to demonstrate.
As such, the present study measures the L2 lexicon size of bilinguals via a
verbal uency measure presented in Gollan, Montoya, and Werner (2002) and
explores how increases in the size of the lexicon aect switching costs. is mea-
sure of lexical robustness is a production measure and captures the quantitative
size of the lexicon. e measure is also continuous in nature. By modeling lexical
robustness as a continuous measure, we are able to describe prociency in more
conceptually-appropriate terms. Prociency should not be viewed as a dichoto-
mous variable; it should capture the developmental continuum that represents
language acquisition. By using a continuous measure of L2 lexicon size, we are able
to identify whether a certain level of lexical robustness coincide with the shi to a
language-specic selective mechanism. In other words, we can investigate whether
a specic threshold of lexical robustness is necessary to engage the mechanism.
We can also investigate the impact of the switching cost along the developmental
continuum using this measure of lexical robustness.
In the sections that follow, we rst discuss Green’s (1986, 1998) ICM and re-
view empirical evidence testing its hypotheses. Following this, we discuss Costa
and Santesteban (2004) and Costa et al. (2006) in detail because they have provided
the primary motivation for the present study. We then introduce the present study
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
(including a discussion of the measure of lexical robustness) and discuss the re-
sults. Finally, we present implications for a bilingual speech production model that
captures the critical role that lexical robustness plays in the lexicalization process.
Inhibitory Control Model
e primary assumption of Green’s (1986, 1998) ICM is that when a bilingual
wants to speak in one language only, that language becomes selected and any fur-
ther production of the nonrelevant language is inhibited. e premise of the ICM
is based on the notion that mental control is a product of inhibition, control sche-
mas, and a supervisory attentional system (SAS). According to this model, there
are multiple levels of control in the bilingual mind with each level corresponding
to a specic schema. Although a wide range of events exist — such as high-level
scripts and low-level articulatory controls — the ICM operates exclusively at the
lemma level. Essentially, Green has proposed that in the regulation of the bilin-
gual lexico-semantic system, a conceptualizer builds conceptual representations,
which are driven by the communicative goal. ese both are mediated by the SAS
together with components of the language system (i.e., language task schemas).
Language task schemas, in turn, compete to control output and thus, accurate se-
lection of any given word requires that the language of production be specied by
the SAS. Once the target language is established, the bilingual mind will turn to
language tags to help determine which non-target words (i.e., those competing for
selection) will need to be inhibited.
e use of language tags is a critical part of Green’s model and also has been
incorporated in other bilingual speech production models (La Heij, 2005; Pou-
lisse & Bongaerts, 1994). Within Green’s framework, the non-target words are
suppressed to a degree that is proportionate to the level of activation based on
language tags (i.e., the more activation that is sent to the non-target language, the
more inhibition will be needed). Because of the asymmetrical system sizes of the
L1 and L2, naturally, more inhibition is applied to the L1 when the L2 is the lan-
guage of production. e opposite is true for the smaller, L2 system: less inhibition
is called upon when the L1 is the language of production.
Figure 1 is an illustration of how the ICM explains the lexicalization process in
bilingual speech production. In this example, upon being presented with a picture
of a chair, an English learner of Spanish is told to name the picture in English.
Upon seeing the picture, the semantic system sends activation to the target lexical
nodes of both languages and to a number of semantically related words. At the
lexical level, because each word contains a language tag, the lexical items belong-
ing to Spanish become inhibited and the bilingual is able to select the word chair
based on its activation level and language tag.
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
Support for the ICM has come from a variety of methodologies including
neurolinguistic studies investigating brain activity (Hernandez, Dapretto, Mazzi-
otta, & Bookheimer, 2001; Price, Green, & von Studnitz, 1999) and natural speech
situations (Grosjean, 1988, 1997; Grosjean & Miller, 1994; Li, 1996, 1998). How-
ever, tasks that require bilinguals to switch back and forth between languages have
proven to be excellent predictors of inhibitory control in bilingual speech produc-
tion. Given the assumption of the ICM that more time is required to reactivate
the larger (L1) system from suppression, it is expected that bilinguals will need
more time (in ms) to switch into their dominant language compared to their less
dominant language. Indeed, strong support for the ICM has come from numeral
naming tasks (Finkbeiner, Almeida, Janssen, & Caramazza, 2006; Meuter, 1994;
Meuter & Allport, 1999) and picture-naming tasks (Costa & Santesteban, 2004;
Costa et al., 2006).
In their seminal article, Meuter and Allport (1999) conducted a battery of
switching experiments with “reasonably procient” bilinguals who were presented
with lists of numerals (1–9). e order of the numerals and length of the lists
were randomized, containing between ve and fourteen numerals. Each target was
placed in either a blue or yellow rectangle indicating that the participant was to
name the numeral in his/her L1 or L2, respectively. Within each list of numerals,
Figure 1. e Inhibitory Control Model (Green, 1986, 1998) (adapted from Finkbeiner,
Gollan, & Caramazza, 2006).
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
Meuter and Allport randomly included anywhere from zero to four L1 and/or L2
language switches. e main ndings from their results can be summarized as
follows: 1) switch trials are slower than nonswitch trials; 2) nonswitch trials have
faster reaction times (RTs) in the L1; 3) L2 switch trials have faster RTs than their
L1 counterparts (which suggests that a switch to L1 is highly dicult); and 4) the
RTs increase with each successive switch.
Meuter and Allport’s (1999) study showed that by looking at the language
switching paradigm, researchers can investigate the extent to which inhibitory
control is utilized in bilingual speech production. eir study also provided some
of the most inuential support for the ICM and its hypotheses were also supported
in other studies (Costa & Santesteban, 2004; Costa et al., 2006; Meuter, 1994). In
sum, they have shown that switching into the L1 is more costly because its lexical
representations have been strongly inhibited during the previous trial to allow the
selection of L2 lexical representation. e switching cost, therefore, is due to the
time required to reactive the lexical node. Switching into the L2 is apparently less
costly because, when naming in L1, its corresponding lexical representation would
not have been inhibited as strongly and therefore the L2 representations would be
more available.
Costa and Santesteban (2004) was the rst study to suggest that the cognitive
control mechanisms involved in lexicalization are modulated by L2 prociency
level. Costa and Santesteban questioned whether or not highly procient bilin-
guals would act dierently than less procient bilinguals in terms of their use of
inhibitory control in speech production. ey hypothesized that inhibitory con-
trol would not be observed in the speech production of highly procient bilinguals
but would be relied upon among less procient bilinguals (i.e., beginning language
learners). Costa and Santesteban conducted a series of ve picture-naming ex-
periments based on the experimental design used in Meuter and Allport (1999).
Costa and Santesteban’s results can be summarized as follows: 1) language switch-
ing costs are present in all bilingual speakers; 2) asymmetrical switching costs are
present for less procient bilinguals but not for highly procient bilinguals; 3) the
switching performance of a highly procient bilingual is independent of the dif-
ference in prociency levels between the two languages involved in the task (but
not in a fourth language (Costa et al., 2006); 4) in a language switching task, highly
procient bilinguals are slower in their dominant than in their non-dominant lan-
guage both for switch and nonswitch trials.
Although Costa and Santesteban (2004) suggested that prociency modulated
the control processes of lexicalization in bilinguals, they le open the possibility
that L2 age of acquisition may be involved as well. Because of this, subsequent
studies have investigated which of these two factors is responsible for a shi in
reliance from inhibitory control to a language-specic selective mechanism. e
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
results in Costa et al. (2006) have suggested that L2 age of acquisition has no eect
on inhibitory control. However, Costa et al. provided somewhat dierent results
than Costa and Santesteban regarding the switching performance of highly pro-
cient bilinguals. Costa and Santesteban reported that highly procient bilinguals
controlled the lexicalization process dierently than less procient bilinguals (by
relying upon a language-specic selective mechanism and inhibitory control, re-
spectively). Costa et al., however, provided evidence suggesting that both inhibi-
tory control and language-specic selection mechanisms are functional in highly
procient bilinguals. In their rst experiment, Costa et al. investigated the impact
of language similarity and L2 age of acquisition on the language-switching perfor-
mance among highly procient Spanish-Basque bilinguals and highly procient
Spanish learners of English. e L1-L2 switching costs were symmetrical for both
groups and neither language similarity nor L2 age of acquisition was reported to
aect the control of lexicalization. eir second experiment explored L2-L3 lan-
guage switches among highly procient Spanish-Catalan bilinguals who spoke
English as their L3. e results revealed symmetrical switching costs between the
strong L2 and weak L3. Costa et al. argued that this nding, coupled with Costa
and Santesteban’s previous ndings, suggest that, “the dierences in the procien-
cy levels between the two languages involved in the task do not lead to asymmetri-
cal switching costs for highly procient bilinguals” (p. 1061).
Costa et al.’s (2006) third and fourth experiments are of special interest to the
present study. In their third experiment, the researchers explored L3-L4 language
switches among highly procient Spanish-Catalan bilinguals who spoke English as
their L3 and French as their L4. e results revealed asymmetrical switching costs
when bilinguals switched between their third and fourth languages. is implied
that when highly procient bilinguals switch between two languages in which they
have relatively low prociency, the magnitude of the switching cost is larger for the
stronger language. ese ndings do not support Costa and Santesteban’s (2004)
claim which hypothesizes that L2 prociency is a determining factor in the shi
toward reliance on a language-specic selection mechanism. Costa et al. hypoth-
esize that these results may be due to the fact that the language-specic selection
mechanism is not functional unless one of the languages involved in the switching
task is one of the bilinguals’ strong languages.
Costa et al.’s (2006) fourth experiment explored language switching between
the L1 and a newly-learned language among highly procient Spanish-Catalan bi-
linguals and Spanish monolinguals (a term Costa et al. used to refer to those with
a very low level of control of the newly learned L2). e results revealed asym-
metrical switching costs for both groups. is implies that the mechanisms used
by both groups for controlling speech production at very early stages of lexical
acquisition were the same. To date, this experiment is the rst to reveal that highly
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
procient bilinguals and monolinguals show the same performance in a language
switching task, suggesting that the mechanisms used during early stages of word
learning are essentially the same. ese ndings reject Costa and Santesteban’s
hypothesis suggesting that with increases in L2 prociency, bilinguals move away
from inhibitory control. If this had been the case, the highly procient bilinguals
in Costa et al.’s fourth experiment should have suered asymmetrical switching
costs when switching between their L1 and a recently learned language. However,
the results of the study suggested that even highly procient bilinguals may revert
back to reliance on inhibitory control, due to possibly limitations of the function-
ality of the language-specic selection mechanism.
Costa et al. (2006) provided evidence showing that in certain instances, pro-
cient bilinguals suer asymmetrical switching costs, thus calling into question
previous claims about the role of prociency in triggering the language-specic
selective mechanism. Given the results of Experiments 3 and 4, Costa et al. modi-
ed their earlier claim that once bilinguals move away from inhibitory control
and make use of a language-specic selection mechanism, they will continue to
rely on it regardless of the relative prociency of the languages involved in the
task. Because highly procient bilinguals suered asymmetrical switching costs
when switching between their L3-L4 and their L1 and a new language, Costa et al.
argued that “whenever the language-switching task involves a language for which
the corresponding words are not well established, highly procient bilinguals will
show asymmetrical switching costs, otherwise symmetrical switching costs will
be obtained” (p. 1067). e researchers identied a theoretical possibility to this
claim: they hypothesized that the robustness of the L2 lexical representations may
be crucial for the functionality of the language-specic selection mechanism. is
implies that a language-specic selection mechanism will only be functional when
lexical representations are integrated in the lexicon. When this is not the case and
lexical representations are weak (e.g., in early stages of L2 acquisition), bilinguals
will rely on mechanisms of inhibitory control. Another possibility they discuss is
that lexical representations are mediated by the L1 and, thus, when bilinguals need
to access the meaning of the L2 word, they must do so by rst activating the L1. If
L2 production operates on L1 mediation, a language-specic selection mechanism
will not be functional because once an L1 representation is selected and translated
into the L2, it must be suppressed to avoid overt production.
To summarize, with respect to how bilinguals control and restrict the lexical-
ization process to one language during speech production, it appears as though
bilinguals make use of control mechanisms to restrict lexicalization to only one
of their languages in order to prevent intrusions from the irrelevant language. In
order to achieve this, however, previous studies have shown that some bilinguals
called up on inhibitory control and others rely on a language-specic selection
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
mechanism. What seems to determine which of the two will be called upon dur-
ing bilingual speech production? Many studies have investigated possible factors
that may shi bilinguals away from inhibitory control towards a language-spe-
cic selection mechanism and have shown that the following do not modulate
this shi: age of acquisition (Costa et al., 2006), language similarity (Costa et al.,
2006), L2 prociency (Costa et al., 2006); and bilingual type (i.e., language learner
vs. heritage speaker) (Schwieter, in press). e present study investigates the iden-
tied possibility outlined in Costa et al., namely, lexical robustness. We opera-
tionalize L2 lexical robustness via a verbal uency measure that taps into the size
and strength of the L2 lexicon. We anticipate that by measuring and exploring the
lexical robustness of the representations of the weaker language instead of rely-
ing on participants’ self-ratings, we will nd that lexical robustness is critical for
the functionality of the language-specic selection mechanism. We hypothesize
that if lexical representations are not robust enough, bilinguals may need to revert
back to reliance on inhibitory control to restrict their lexicalization process to one
language. Moreover, we expect our data to allow us to predict a specic threshold
of lexical robustness that coincides with the shi to a language-specic selective
mechanism during bilingual speech production.
Experiment
In the present study, a verbal uency measure was rst used to measure the L2
lexical robustness for English learners of Spanish. ese participants then named
pictures in their L1 and L2 according to a color cue. Critical variables included tri-
al type (nonswitch and switch) and response language (English and Spanish). By
examining the eects of L1-L2 switches in picture naming, inferences were made
regarding whether or not the cognitive control mechanisms in bilingual speech
production are modulated by lexical robustness.
e experimental procedure in the present study follows Costa and Santeste-
ban (2004). A picture-naming task with language switches was conducted in which
language learners were instructed to name the pictures in either their L1 or L2 as
determined by a color cue to indicate language of production.1 e ICM posits
suppression of the nonrelevant language and therefore assumes that more time is
required to reactivate a larger system (L1) than the smaller system (L2). us, any
asymmetrical switching costs observed between L1 switches and L2 switches will
suggest that speech production entails inhibitory control. is experiment seeks to
investigate the role that the robustness of L2 lexical representations has on the pos-
tulated shi in control mechanisms during bilingual speech production by mea-
suring the L2 lexicon size of bilinguals and exploring how increases in the size of
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
the lexicon aect switching costs. Moreover, in using a continuous measure of the
strength of the L2 lexicon, we will also investigate whether a specic point during
development coincide with the use of a language-specic selective mechanism.
Participants
A total of 54 English-Spanish language learners with varying abilities in Spanish
were recruited from the student and faculty body at a large state university in the
United States. e mean average age for the participants 25 years and their mean
L2 age of acquisition was 18 years. At the time of the present study, all participants
were currently enrolled in at least one course conducted in Spanish. A language
history questionnaire was administered to provide information regarding the par-
ticipants’ language use and experiences. In particular, this questionnaire served
to elicit self-ratings in English and Spanish on their abilities to read, write, speak
and listen. On a ten-point scale, participants’ ratings for their L2 skills ranged as
follows: reading, 4–10; writing, 2–10; speaking, 2–10; and listening, 3–10. Overall,
participants rated their English abilities as 9.5.
Verbal Fluency Measure
e verbal uency measure used in the present study was based on the experi-
mental procedures of Gollan, Montoya, and Werner (2002). Costa et al. (2006)
argue that lexical robustness involves the familiarity with and frequency of access
that leads to greater automaticity of retrieval of lexical items. We believe that a
verbal uency measure in which participants spontaneously and rapidly produce
as many lexical items as possible related to semantic categories is a viable way of
operationalizing the robustness of the L2 lexicon. And, although many neuropsy-
chological studies have used the verbal uency measure as a means to explore a
variety of diseases and disorders, Gollan et al. (2002) argue that it is useful in stud-
ies of bilingualisms as well. ey posit that there is a direct relationship between
word knowledge and verbal uency scores because category size will predict per-
formance on the verbal uency task (see also Borkowski, Benton, & Spreen, 1967;
Rohrer, Wixted, Salmon, & Butters, 1995). Accordingly, a total of ten semantic
categories2 were taken from Gollan et al. (2002) and were individually verbalized
to each participant. Upon hearing the category, each participant was given 60 sec-
onds to produce as many items within that category as s/he could in the L2.3 e
presentation of the categories was randomized for each participant. A total score
was calculated by adding all responses from each of the ten semantic categories.
Each word was only counted once and therefore, words that may have been repeat-
ed during the procedure were not included in the participant’s score. Essentially,
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
the score refers to the total number of words produced in the L2 in ten one-minute
segments. e range of scores varied from 56–220.
Stimuli
In accordance with Costa and Santesteban (2004), the present experiment includes
ten black and white line drawings from the Snodgrass and Vanderwart (1980)
standardized picture list. e pictures were of a pencil, house, car, dog, book, cat,
chair, table, desk and a heart. e materials used for all participants were the same.
As in Costa and Santesteban, all ten pictures were presented individually on a
computer screen and were classied as either a nonswitch trial or a switch trial. A
nonswitch trial is dened as one in which the previous trial is named in the same
language and a switch trial is one in which the previous trial is named in a dier-
ent language.
Design
A range of 5–14 pictures were randomly placed in 100 lists which contained any-
where from 0–4 switch trials. e total number of trials in the experiment was
950 (665 nonswitch trials (70%) and 285 switch trials (30%)). All of the pictures
were individually presented in a colored box as a language cue: blue if the target
was to be produced in English or yellow if the target was to be produced in Span-
ish. ere was equal production of L1 and L2 (i.e., 475 responses in L1 and 475
responses in L2 were elicited). Each of the pictures was presented 95 times during
the entire experiment. For lists with 5–10 pictures in length, no pictures were du-
plicated. However, for lists of 11–14 pictures, the repeated pictures were placed at
least three trials apart from their rst presentation.
Procedure
All participants were individually tested and were given verbal instructions by the
researchers in addition to the instructions they read on the computer screen. Each
participant was placed in front of a computer screen at a distance of about 2.5 feet
and was instructed to name lists of pictures in the language represented by the
colored box. First, each participant went through a training session that included
six lists of pictures similar to the experimental lists. Like the experimental sets,
the practice ones were also randomized for length and amount of switches. e
researchers ensured that sometime during the practice sets, each of the ten pic-
tures appeared at least twice so that the participants were not subjected to any new
pictures during the actual experimental lists.
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
e practice lists and the following 100 identically-structured experimental
lists rst presented either the word “English” inside a blue box or “Spanish” inside
a yellow box for 2000 ms in the center of the computer screen to establish a com-
mon xation point for all pictures. is allowed participants to be able to focus on
the position and only name pictures throughout the experiment without having to
look in other places of the screen for the target. e words “English” and “Spanish”
were presented at the beginning of each list and served as a means to reinforce the
association between color and language of production (i.e., blue represented Eng-
lish and yellow indicated Spanish). In any given list, the rst picture was presented
in the same box as the xation box. is picture remained on the screen for 2000
ms or until the participant responded. Upon response, there was a blank interval
of 800 ms. During this interstimulus interval, the microphone was inactive so that
if a participant took longer to verbalize the preceding target, the computer would
not assume that this was the response for the current trial. e next picture (either
a switch or nonswitch trial) was shown and the cycle was repeated until the end of
the list, at which time an asterisk (*) was presented for 1000 ms to show that the
list had nished and that another one would begin in 1000 ms.
All responses were tape recorded and were coded by the researchers as “cor-
rect” or “incorrect” on the master key. Due to the amount of pictures (950) to be
named, a break was given to the participants every 10 lists (approximately every 7
minutes). During the breaks, participants were allowed to rest for a few minutes
to avoid fatigue.
Results
Only correct responses were included in the analyses of response latencies; cor-
rect and incorrect responses were both included in the error analyses. Accuracy
was coded as either correct (if the participant correctly named the target) or in-
correct (if the participant incorrectly named the target). RTs faster than 300 ms
and slower than 5000 ms were removed from the analyses and treated as outliers
(Ratcli, 1993). Most of these were due to a technical malfunction in which the
microphone did not register the participant’s response. is amount represented
2.0% of the total data collected. In addition, RTs that were 2.5 SDs above or below
the participant’s mean RT for each of the overall conditions were excluded from
the analyses and treated as outliers. is eliminated an additional 1.7% of the total
data collected.
We rst present descriptive statistics. Table 1 shows the mean reaction times
and accuracy percents for L1 and L2 switch and non switch trials.4
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
In English, the L1, participants were slower on the switch trials compared to
the nonswitch trials (a dierence of 75 ms). In Spanish, the L2, the participants
were again slower on the switch trials compared to the nonswitch trials (a dier-
ence of 52 ms). us, the magnitude of the switch cost was greater in the L1, sug-
gesting that switching into the L1 takes longer than into the L2. Regardless of trial
type, participants were always slower in the L1 (944 ms) than in the L2 (875 ms).
We now turn to the results of the regression analyses.
Given that both of our dependent variables (RT and accuracy) as well as
our independent variable (verbal uency) were continuous ratio-level variables
we employed standard Ordinary Least Squares (OLS) multivariate regression.
Because we pooled the data for our 54 subjects across each of the 4 conditions
(English switch trials; English nonswitch trials; Spanish switch trials; Spanish non-
switch trials) our nal unit of analysis is the participant-condition and results in
216 observations. By pooling data in this manner, the results in each participant’s
responses are not statistically independent across each of the combinations of con-
ditions. To account for this statistical dependence, we estimate each of the follow-
ing regression models with robust standard errors clustered within subject (Huber,
1967; White, 1980; Williams, 2000). Finally, to test the threshold hypothesis of
lexical robustness we employ the Brambor, Clark, and Golder (2006) technique.
We plot the estimated eect of switching for a given level of verbal uency with
95% condence levels plotted around this relationship to determine whether there
is a specic point at which verbal uency become a critical factor.
Table 2 presents the results of the multivariate analysis, with Model 1 display-
ing the RT results and Model 2 displaying the accuracy results. In examining the
coecients in Model 1, we rst see that there are statistically signicant main ef-
fects for verbal uency (p < .01). e higher one’s verbal uency score, the faster
they are. We next see that there is a main eect of language (p < .05), indicating
that all participants were slower in English, their native language. ere is also a
signicant main eect of trial type (p < .01), indicating that switch trials resulted
in longer RTs compared to nonswitch trial across all participants. Next we see a
signicant interaction between language and trial type (p < .01), indicating that
switch trials into English were the slowest. Finally, we have a three-way interaction
Table 1. Mean Reaction Times (ms) and Accuracy (%) for L1 (English) and L2 (Spanish)
switch and non-switch trials
L1 L2
RT Acc RT Acc
Switch 981 94.3 901 96.9
Nonswitch 906 96.4 849 98.1
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
between verbal uency, language and trial type (p < .01). For those individuals with
lower levels of verbal uency, switching into English was the most dicult.
e results above are consistent with the general nding (Costa & Santeste-
ban, 2004; Experiments 1 and 2 of Costa et al., 2006) with respect to prociency.
However, in order to understand precisely what these results tell us about a lexical
robustness threshold that we proposed earlier, we must consider two additional
items. First, we must consider at what level of verbal uency the switching cost
phenomenon reduces to zero. Second, we must consider whether the three-way
interaction between verbal uency, language and trial type is statistically signi-
cant across the entire range of the verbal uency score. is second test is impor-
tant because interaction eects with continuous independent variables, while sig-
nicant on average (as indicated by the test of signicance in the regression table),
may not necessarily be signicant across the entire range of a continuous condi-
tioning variable (in this case verbal uency) (Brambor, Clark, & Golder, 2006).
e standard error of the coecient associated with the switching eect changes
over the level of the continuous conditioning variable, verbal uency. erefore,
it is possible that while statistically signicant for some level of verbal uency,
the switch eect may indeed be statistically undistinguishable from zero, at some
other level of verbal uency.
To test for these additional possibilities we conducted two post hoc tests. First,
Figure 2 presents the test of where the switch point appears to reduce to zero. e
Table 2. OLS Regression Models for Switching Task
Reaction Time Accuracy
Model 1 Model 2
Variable B Std Error B Std Error
Verbal Fluency (VF) −1.098*** 0.370 −0.00018 0.00013
Language (L) 51.305** 26.029 −0.02018 0.01246
Trial Type (TT) 61.935*** 20.066 −0.00847 0.00776
VF* L 0.059 0.198 0.00030 0.00012
VF * TT −0.087 0.157 −0.00004 0.00007
L* TT 66.612*** 18.493 −0.01082 0.01250
VF* L * TT −0.413*** 0.143 0.00002 0.00010
Constant 965.511*** 52.707 1.0005*** 0.01259
R20.130 0.229
F-value (d.f.) 19.57 (7, 53)*** 13.47 (7,53)***
Number of Observations 216 216
Number of Subjects 54 54
*** p < 0.01, ** p < 0.05, * p < 0.10
Observations are Participant-Conditions. Standard errors are clustered on participant.
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
Figure presents the eect of switching across English and Spanish over the range
of our verbal uency score. e dierence between the English and Spanish lines
can be understood as the switch cost. e bold section of the lines indicates where
the switch eect is statistically signicant (p < .01). As the gure suggests, for low
verbal uency scores the switch eect is quite large with English taking approxi-
mately 40 ms (100 ms − 60 ms) longer compared to Spanish. As verbal uency
increases, this switch eect attenuates, eventually reaching zero at a verbal u-
ency of approximately 160. e gure suggests that for verbal uency higher than
approximately 160, the switch eect changes in its orientation. One should note
that the slope of the English line is more dramatically aected by changes in ver-
bal uency in Spanish, thus illustrating how rapidly the switch eect is changing
with increases in lexical robustness. For higher verbal uency, it would appear that
the switch eect changes direction, with English now taking longer than Spanish.
However, this eect does not appear to endure long given that the bold shading
quickly trends to insignicance beyond the cross point.
To gain better insight into the possibility of a shi in the switch point we con-
ducted one additional post hoc test. Figure 3 shows the mean eect of switching
across English and Spanish over the range of our verbal uency score. is time
however, we plot the 95% condence bands around these estimated eects. ese
condence bands reveal with high statistical condence (p < .05) the interval with-
in which the eect of switching is likely to lie. e switching eect for English is
0 20 40 60 80 100
Eect of Switching
50 100 150 200 250 300
Verbal Fluency Score (Index of Lexical Robustness)
English Spanish
Signicant at 95% Condence Level
Figure 2. e eect of switching across English and Spanish.
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
shown in bold, while the switching eect for Spanish is shown in regular font. e
condence intervals are plotted around each eect with the relevant dashed lines.
If the condence intervals of the switch eect for Spanish or English overlap then
we can declare that the dierence between the switch eects for that language are
not statistically signicant. However, when the two intervals do not overlap, then
we can be condent that the switching cost for English and Spanish are indeed dif-
ferent from each other (see Appendix 1). As we see from the gure, these intervals
do not overlap for verbal uency scores below approximately 110. Beyond 110,
however, the intervals overlap for the entire range of verbal uency. e results
of the post hoc analyses reveal support for our expectation that there is indeed
a threshold eect of verbal uency (around 110) beyond which the switching ef-
fect collapses to zero. It is important to note that although both Figures 2 and 3
would seem to suggest that the eect goes below zero, the condence intervals
suggest that the eect at that level of verbal uency is statistically no dierent from
zero. erefore the appearance of a below zero eect is really only an artifact of
the graph.
Figure 3. e eect of switching across English and Spanish with condence bands.
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
Error Analyses
As in Costa and Santesteban (2004), no signicant eects were found in the error
analyses.
Discussion
In the experimental procedures of the present study, we conducted a picture-nam-
ing experiment in which bilinguals with varying levels of L2 verbal uency (i.e.,
lexical robustness) switched back and forth between their two languages according
to a color cue that specied the language of production. First, our results answered
the question that Costa et al. (2006) posed regarding the relationship between the
functionality of the language-specic switch mechanism and lexical robustness.
e results indicate that lexical robustness does indeed explain the shi away from
using inhibitory control to using a language-specic mechanism. Bilinguals with
lower levels of L2 lexical robustness exhibited high levels of asymmetry between
the amount of time it takes to switch into the L1 and L2 while those individuals
with higher L2 lexical robustness did not show this asymmetry. Moreover, as previ-
ous studies have reported (Costa & Santesteban, 2004; Costa et al., 2006; Meuter &
Allport, 1999; Segalowitz & de Almeida, 2002), our data revealed that in the context
of a language switching experiment, participants named pictures in their less-dom-
inant language faster than in their more-dominant language. e possible reasons
for this may be explained by the fact that participants compensate the imbalance
of the availability of L1 and L2 lexical representations (see Costa & Santesteban,
2004 for additional discussion). In other words, it is as if the weaker L2 is tempo-
rarily ‘super activated’ and the L1 is more strongly suppressed in the context of
the switching task. Second, our results demonstrate the value of operationalizing
prociency with an objective continuous measure of verbal uency. With this mea-
sure, our results reveal a certain quantitative threshold where a language selective
mechanism may become functional. e verbal uency measure and the statistical
procedures we used to investigate threshold eect allowed us to answer these im-
portant questions that have not been previously addressed in the eld.
Indeed, a series of post-hoc analyses helped to identify the point in L2 de-
velopment of the lexicon at which the language-specic selective mechanism be-
came functional. At approximately a verbal uency score of 110, the switching
cost became zero. Prior to this point, the incremental increases in verbal uency
resulted in decreasing switching costs. is result suggests that when the size of
the L2 lexicon was below this threshold, an individual suered switching costs.
In other words, the L2 lexicon was not suciently developed to allow for the lan-
guage-specic selective mechanism to engage.
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
One might wonder what a verbal uency score of 110 really means. For a sub-
stantive comparison, those participants in our study with approximately 6.7 years
of formal L2 learning (highly advanced language learners) roughly correspond
with participants having verbal uency scores higher than 110. In other words,
the lexical robustness threshold was quite high. Lexical robustness does play a key
role in a developmental shi of the cognitive control processes of bilingual speech
production, but it occurs relatively late in the acquisition process. Needless to say,
future research implementing this lexical robustness measure and these statistical
procedures will be needed to test these predictions with other bilingual popula-
tions with varying levels of skill in their languages and in particular with individu-
als who have an L3 or L4. Moreover, if we recall Figure 2, there appeared to be a
point at which the switching costs reversed. While our post-hoc analyses revealed
that past the 110 threshold, there were no switching costs in our population, fu-
ture research is necessary to investigate a potential reversal of the switching cost.
It could be the case that as a bilingual’s lexical robustness increases in the L2, their
L1 counterparts may take a hit so to speak. By measuring both L1 and L2 lexical
robustness in various bilingual populations, we may nd a dominance eect and a
reversal of switching costs.
Our ndings have important implications for models of bilingual speech
production. It is imperative that current models of bilingual speech production
consider the critical role of L2 lexical robustness in lexical processing and speech
production. Figure 4 shows a bilingual speech production model that incorpo-
rates both mechanisms of inhibitory control (for less procient bilinguals) and a
language-specic selection mechanism (for highly procient bilinguals). We call
this model the Selection by Prociency Model (SbP), a visual representation of
the impact that prociency has on the cognitive processes of lexicalization. We
feel that in light of the results of the present study, our SbP can explain the role
that lexical robustness plays in the shi in cognitive control mechanisms during
speech production and captures both of the theoretical claims put forth by Costa
et al. (2006).
Essentially, the SbP provides an explanation for how an L2 speaker names a
picture in his/her L2. e model species that L2 learners move along a prociency
continuum in a bi-directional manner (i.e., to account for learning and language
attrition). Two separate L2 learners are presented in the model: the one on the le
represents a less procient language learner and the one on the right shows a more
procient L2 learner. e model also posits dierential strengths of the associa-
tion between the conceptual links and lexical links (which are illustrated by the
thickness of the arrows in the model). For example, the darker arrows represent
stronger links and the dashed arrows represent either weaker links or, as in the
right portion of the model, those which may be present but not necessarily called
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
upon. e SbP also demonstrates that for less procient L2 learners, the mecha-
nisms that are called upon during the speech production process entail inhibitory
control (i.e., language task schemas). is is strikingly dierent to what the model
hypothesizes for more procient L2 learners. In the latter case, learners are able to
rely upon higher linguistic cues and to conceptually mediate without mechanisms
of inhibitory control. However, the SbP model assumes that in certain situations
or instances, the highly procient bilingual may need to call upon mechanisms of
inhibitory control again (as shown by the longest diagonal arrow). Finally, as can
be seen at the top of the model, the shi in reliance to language selective mecha-
nisms cannot be achieved for less procient L2 learners until they become much
more procient.
L2
L1
L2
L1
Prociency
development
Concept: Cat
Language Cue: Spanish
Register: Formal
Concept: Cat
Language
Tas k
Schemas
gato
cat
gato
cat
strong links
weak links
path of lexicalization
L1 L2
Figure 4. e Selection by Prociency Model.
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
As mentioned above, there are two explanations of lexicalization depicted
in the SbP depending on the L2 prociency of the bilingual. e le side of the
model represents a less procient bilingual in which there is a stronger associa-
tion between concepts and the L1 words mapped onto them. e links between
L2 words and their concepts, however, are much weaker (see Kroll and Stewart,
1994). erefore, when a less procient bilingual names a picture in his/her L2, s/
he must have to do so by associating the concept to its L1 lexical item rst. e
lexicalization procedure for a less procient bilingual will entail reliance on in-
hibitory control. In the SbP model, the example is given in which a less procient
English-Spanish L2 learner is asked to name a picture of a cat in Spanish. When
this happens, several lexical candidates in both languages receive activation. Be-
cause of low L2 prociency and reliance on L1 to conceptually mediate, competi-
tion will spill over to the lexical level. Words contain language tags that specify
which languages they belong to and help regulate the degree of activation at the
lexical level. More than likely, language tags, in addition to a language task schema,
may ultimately help determine the amount of inhibition required by virtue of the
strength of lexical robustness. Once the words in the non-target language have
been suppressed, it will be easier to select the target language word that matches
the target concept.
e right part of the model represents the speech production of a highly pro-
cient bilingual in which there is a strong association between concepts and words
in both the L1 and L2. When a highly procient bilingual names a picture in his/
her L2, he/she directly accesses the word mapped onto the concept instead of hav-
ing to rely upon the L1 translation. e lexicalization procedure for a highly pro-
cient bilingual will primarily entail reliance on a language-specic selection mech-
anism but, as indicated by the diagonal arrow toward the le side of the model,
may revert to inhibitory control when needed as the linguistic context requires.
e SbP model gives the example in which a highly procient bilingual is asked to
perform the same task. In this case, several lexical candidates in English and Span-
ish receive activation from a language-specic selection mechanism in the form of
a preverbal message. In this cue, bilinguals obtain vital higher linguistic informa-
tion about speech production including but not limited to: 1) the target language;
2) the linguistic register; and 3) the concept to be lexicalized. ese cues work
together to ensure that the target lexical item has a higher activation level than its
competitors’ and therefore, inhibitory control is not necessary because words in
the nonresponse language do not compete for selection.
Although L2 prociency has been previously posited as a modulator for de-
termining which control mechanisms bilingual speech production will entail (see
also Costa & Santesteban, 2004; Experiments 1 and 2 of Costa et al., 2006), the
results of the present study have allowed us to explore a particular element of L2
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
prociency that is responsible, namely, lexical robustness. e SbP is a rst step in
combining issues of inhibitory control and a language-specic selection mecha-
nism in a comprehensive developmental model of bilingual speech production.
Overall, based on the results of the present study, we can entertain the notion
that lexical robustness is one key factor of L2 prociency that determines whether
or not the bilingual will rely on inhibitory control or a language-specic selection
mechanism while speaking. We must be careful to not eliminate prociency as a
construct from future studies investigating control issues in speech production.
Many skills (reading, speaking, vocabulary, more direct concept mediation, etc.)
contribute to what researchers consider to be prociency. e results of the pres-
ent experiment allow us to untangle prociency and nd one particular element
of it, lexical robustness, that may explain why highly procient bilinguals have
performed dierently than less procient bilinguals in past studies. However, one
might still wonder if overall prociency is really the underlying explanation, and
not lexical robustness. If we examine the participants’ overall self-ratings of L2
prociency as elicited by the language history questionnaire and their verbal u-
ency scores, we only nd a moderate (.489) correlation. Of course, future research
will need to assess the relationship between lexical robustness as indexed by verbal
uency and other independent measures, but at a minimum, lexical robustness
seems viable as one of the key components of the elusive construct of prociency.
Conclusion
e present study has provided support for the notion that when strong L2 lexi-
cal representations exist, there will be no need for bilinguals to rely on inhibi-
tory control mechanisms to restrict the lexicalization process to one language. In
other words, our study has demonstrated that the moment of a transition from
inhibitory control to language-specic control coincides with a particular level of
lexical robustness. We also recognize that it may be possible that bilinguals revert
back to inhibitory control in conditions in which the language-specic selection
mechanism cannot function. More research is needed to explore which particular
contexts require a need for inhibitory control and which allow for the bilingual to
rely on the language-specic selection mechanism.
In addition to dening lexical robustness as an important factor in how bilin-
guals control the lexicalization procedure, we modeled prociency as a continuous
variable and used the Brambor, Clark, & Golder (2006) technique to allow us to
better pinpoint a threshold for lexical robustness. We dened the latter as the point
in which bilinguals will no longer have to primarily rely on inhibitory control. We
found evidence for such a developmental shi and a minimum threshold of lexical
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
robustness. We argue that scholars must consider moving away from conceptualiz-
ing prociency as a dichotomous variable and instead utilize continuous measure
that better capture the underlying developmental process. Finally, we presented
our Selection by Prociency Model describing the role that lexical robustness as
well as reliance on the L1 plays in the shi in cognitive control mechanisms during
speech production.
Notes
. Costa et al. (2006) point out that a potential drawback to using the experimental procedures
in the present study is that the large number of repetitions may create a stimulus-response as-
sociation. However, they ultimately maintain that this task is useful for examining bilingual lan-
guage control because it has been shown to be sensitive to the dierential behavior of a variety
of bilingual types.
. For other researchers interested in this task, we list the categories in order from the ones that
elicited the most to the least number of responses: countries, clothing, animals, academic ma-
jors, colors, fruits, vegetables, things with wheels, musical instruments, and sports.
. Although the verbal uency measure was only conducted in the L2 in this study, future re-
search might consider conducting the verbal uency measure in the L1 as well and then remove
eects of the L1 that may be indicative of individual dierences.
. We must note that these numbers are unconditional means. ey do not take into account
the measure of lexical robustness. As such, any discrepancies with the numbers in Table 2 can
be explained by this fact.
References
Borkowski, J. G., Benton, A. L., & Spreen, O. (1967). Word uency and brain damage. Neurop-
sychologia, 5, 135–140.
Brambor, T., Clark, W. R., & Golder, M. (2006). Understanding interaction models: Improving
empirical analyses. Political Analysis, 14, 63–82.
Costa, A. (2005). Lexical access in bilingual production. In J. F. Kroll & A. M. B. de Groot (Eds.),
Handbook of bilingualism: Psycholinguistic approaches (pp. 308–325). New York: Oxford.
Costa, A., & Caramazza, A. (1999). Is lexical selection in bilingual speech production language-
specic? Further evidence from Spanish-English and English-Spanish bilinguals. Bilingual-
ism: Language and Cognition, 2(3), 231–244.
Costa, A., Miozzo, M., & Caramazza, A. (1999). Lexical selection in bilinguals: Do words in
the bilinguals’ two lexicons compete for selection? Journal of Memory and Language, 41,
365–397.
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
Costa, A., & Santesteban, M. (2004). Lexical access in bilingual speech production: Evidence
from language switching in highly procient bilinguals and L2 learners. Journal of Memory
and Language, 50, 491–511.
Costa, A., Santesteban, M., & Ivanova, I. (2006). How do highly procient bilinguals control
their lexicalization process? Inhibitory and language-specic selection mechanisms are
both functional. Journal of Experimental Psychology: Learning, Memory, and Cognition,
32(5), 1057–1074.
Finkbeiner, M., Almeida, J., Janssen, N., & Caramazza, A. (2006). Lexical selection in bilingual
speech production does not involve language suppression. Journal of Experimental Psychol-
ogy: Learning, Memory, and Cognition, 32(5), 1075–1089.
Finkbeiner, M., Gollan, T., & Caramazza, A. (2006). Lexical access in bilingual speakers: What’s
the (hard) problem? Bilingualism: Language and Cognition, 9(2), 153–166.
Gollan, T. H., Montoya, R. I., & Werner, G. A. (2002). Semantic and letter uency in Spanish —
English bilinguals. Neuropsychology, 16(4), 562–576.
Green, D. (1998). Mental control of the bilingual lexico-semantic system. Bilingualism: Lan-
guage and Cognition, 1, 67–81.
Green, D. W. (1986). Control, activation, and resource: A framework and a model for the control
of speech in bilinguals. Brain and Language, 27, 210–223.
Grosjean, F. (1988). Exploring the recognition of guest words in bilingual speech. Language and
Cognitive Processes, 3, 233–274.
Grosjean, F. (1997). Processing mixed languages: Issues, ndings and models. In A. M. B. de
Groot & J. F. Kroll (Eds.), Tutorials in bilingualism: Psycholinguistic perspectives (pp. 225–
254). Mahwah, NJ: Erlbaum.
Grosjean, F., & Miller, J. L. (1994). Going in and out of languages: An example of bilingual ex-
ibility. Psychological Science, 5, 201–206.
Huber, P. J. (1967). e Behavior of Maximum Likelihood Estimates under Nonstandard Con-
ditions, Proceedings of the Fih Berkeley Symposium on Mathematical Statistics and Prob-
ability, 221–233.
Hernandez, A. E., Dapretto, M., Mazziotta, J., & Bookheimer, S. (2001). Language switching
and language representation in Spanish–English bilinguals: An fMRI study. NeuroImage,
14, 510–520.
Huber, P. J. (1967). e Behavior of Maximum Likelihood Estimates under Nonstandard Con-
ditions, Proceedings of the Fih Berkeley Symposium on Mathematical Statistics and Prob-
ability, 221–33.
Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming: Evi-
dence for asymmetric connections between bilingual memory representations. Journal of
Memory and Language, 33, 149–174.
La Heij, W. (2005). Selection processes in monolingual and bilingual lexical access. In J. F. Kroll
& A. M. B. de Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp.
289–307). New York: Oxford.
Li, P. (1996). Spoken word recognition of code-switched words by Chinese-English bilinguals.
Journal of Memory and Language, 35, 757–774.
Li, P. (1998). Mental control, language tags, and language nodes in bilingual lexical processing.
Bilingualism: Language and Cognition, 1, 92–93.
Meuter, R. F. I. (1994). Language switching in naming tasks. Unpublished doctoral dissertation,
Oxford University.
© 2008. John Benjamins Publishing Company
All rights reserved
Language switching
Meuter, R. F. I., & Allport, A. (1999). Bilingual language switching in naming: Asymmetrical
costs of language selection. Journal of Memory and Language, 40, 25–40.
Poulisse, N. (1999). Slips of the tongue: Speech errors in rst and second language production.
Amsterdam, Philadelphia: John Benjamins.
Poulisse, N., & Bongaerts, T. (1994). First language use in second language production. Applied
Linguistics, 15, 36–57.
Price, C. J., Green, D. W., & von Studnitz, R. (1999). A functional imaging study of translation
and language switching. Brain, 122, 2221–2235.
Ratcli, R. (1993). Methods for dealing with reaction time outliers. Psychological Bulletin,
114(3), 510–532.
Roelofs, A. (1998). Lemma selection without inhibition of languages in bilingual speakers. Bilin-
gualism: Language and Cognition, 1, 94–95.
Rohrer, D., Wixted, J. T., Salmon, D. P., & Butters, N. (1995). Retrieval from semantic memory
and its implications for Alzheimer’s disease. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 21, 1127–1139.
Segalowitz, N., & de Almeida, R. G. (2002). Conceptual representation of verbs in bilinguals:
Semantic eld eects and a second language performance paradox. Brain and Language,
81, 517–531.
Schwieter, J. W. (in press). e eects of bilingual type on language selectivity. In J. Watze (Ed.),
Readings in Language Studies, Volume 1: Language Across Disciplinary Boundaries. New
York: e International Society for Language Studies, Inc.
Snodgrass, J. G., & Vanderwart, M. (1980). A standardized set of 260 pictures: Norms for name
agreement, image agreement, familiarity, and visual complexity. Journal of Experimental
Psychology: Human Learning and Memory, 6, 174–215.
White, H. (1980). A Heteroskedasticity-consistent covariance matrix estimator and a direct test
for Heteroskedasticity. Econometrica, 48, 817–838.
Williams, R. L. (2000). A note on robust variance estimation for cluster-correlated data. Biomet-
rics, 56, 645–646.
Corresponding address
John W. Schwieter
Dept. of Languages and Literatures
408 Bricker Academic Building
75 University Avenue
Waterloo, Ontario, N2L 3C5
Phone: 519.884.0710 ext. 2394
Fax: 519.884.7369.
jschwieter@wlu.ca
© 2008. John Benjamins Publishing Company
All rights reserved
John W. Schwieter and Gretchen Sunderman
Appendix 1
To test our proposition that lexical robustness modulates the switch cost eect on production
RTs, we estimate multiplicative regression models in which production RT is a function of the
interaction between lexical robustness (verbal uency) and switching. is hypothesis can be
specied in an econometric regression model of accuracy as follows:
Production RT = β0 + β1 Switching + β2 Verbal Fluency + β3
(Switching)(Verbal Fluency) + ε
where ε is the disturbance term. is suggests that to evaluate our threshold hypothesis in a re-
gression context we must estimate the marginal eect of Switching on RTs. [Note that the term
‘marginal eect’ refers to the standard econometric denition of the change in the dependent
variable that follows a unit change in the independent variable of interest, or, in mathematical
terms, the rst derivative. e word ‘marginal’ in this term in no way connotes ‘unsubstantial’
or ‘unimportant’ but rather the expected change in Y given a 1-unit change in x.] For example,
the marginal eect of Switching on Production RTs, or the expected change in Production RT for
a unit shi in Switching is found by taking the rst derivative of the above regression equation
with respect to Switching. is yields a marginal eect of:
d(Production Rt) = β1 + β3 VerbalFluency
d(Switching)
where, β1 and β3 are the estimated regression coecients. is equation reveals the precise man-
ner in which Verbal Fluency conditions the Switching eect on Production RTs. For a Verbal
Fluency score of zero the eect of Switching on Production RT is simply β1. For increasingly
higher Verbal Fluency scores the eect of Switching on Production RT is β1+β3(Verbal Fluency),
with the nature of the eect being determined by the sign and magnitude of the β3 coecient.
e statistical signicance of this marginal eect is determined by its standard error. However,
unlike additive models, in multiplicative models with continuous variables, the standard error of
the marginal eect varies across the level of conditioning variable, or in our case, verbal uency.
In the above regression equation, if we denote switching as SW and Verbal Fluency as VF, then
the conditional standard error of the marginal eect of SW across values of VF is given as:
conditional S.E. of SW = √ var(B1
) + var(B3
)VF2 + 2 cov(B1
B3
)VF
where var(Bx) represents the variance of the relevant regression coecient and cov(B1B3) rep-
resents the covariance between the SW and the interaction term. To determine the statistical
signicance of the marginal eect of SW we then proceed as we would in a typical regression
context, dividing the estimate by its standard error and seeing if it is exceeds the critical score
of 1.96. Except here we plot the marginal eect of SW on Production with its 95% condence
bands plotted around it. e condence bands are calculated by taking estimated standard error
of the marginal eect of SW for a given level of VF and multiplying it by 1.96. We then take this
resulting width and add and subtract it to the estimated marginal eect. e result is a gure
which displays the 95% condence interval plotted around the marginal eect. erefore, this
Figure represents a valid statistical test of our hypothesis.