Cognitive control and language ability contribute to online reading comprehension: Implications for older adult bilinguals

Abstract

Aims
Previous research has indicated that young adults form predictions for the meaning of upcoming words when contexts are highly constrained. This can lead to processing benefits when expectations are met, but also costs, as indicated by a late, frontally distributed and positive event-related potential (ERP), when an unexpected word is encountered. This effect has been associated with the conflict that arises for prediction errors, as well as attempts to suppress a previously formed prediction. However, individual differences have been found for young adult bilingual and older adult monolingual readers, whereby only those who exhibited better language regulation and executive function skill showed this pattern. The goal of the current study was to investigate how these executive functions influence comprehension skill and behavior for elderly bilinguals.

Approach
We asked whether older adult monolinguals and bilinguals were capable of generating predictions online, and whether cognitive control and language regulation ability were related to the magnitude of prediction costs.

Data and Analysis
Participants ( N = 27) read sentences while their electroencephalogram was recorded, and completed a battery of language and cognitive performance tasks.

Findings
While older adult monolinguals showed some sensitivity to prediction error, older adult bilinguals produced greater prediction costs, an effect that was significantly correlated with both age and control ability.

Originality
This study is the first to show ERP evidence that bilinguals are capable of forming predictions during comprehension in older adulthood.

Significance
These results have important implications for the ways in which bilingualism may influence comprehension across the lifespan. While healthy aging has been proposed to lead to declines in executive function these declines may be mitigated for bilinguals, who have a wealth of experience in negotiating language-related conflict.

Full text

https://doi.org/10.1177/1367006918763143
International Journal of Bilingualism
1 –15
© The Author(s) 2018
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1367006918763143
journals.sagepub.com/home/ijb
Cognitive control and language
ability contribute to online reading
comprehension: Implications for
older adult bilinguals
Megan Zirnstein
Department of Psychology, University of California, Riverside, USA
Janet G van Hell
Department of Psychology, Pennsylvania State University, USA
Judith F Kroll
Department of Psychology, University of California, Riverside, USA; Department of Psychology, Pennsylvania State
University, USA
Abstract
Aims: Previous research has indicated that young adults form predictions for the meaning of
upcoming words when contexts are highly constrained. This can lead to processing benefits when
expectations are met, but also costs, as indicated by a late, frontally distributed and positive
event-related potential (ERP), when an unexpected word is encountered. This effect has been
associated with the conflict that arises for prediction errors, as well as attempts to suppress a
previously formed prediction. However, individual differences have been found for young adult
bilingual and older adult monolingual readers, whereby only those who exhibited better language
regulation and executive function skill showed this pattern. The goal of the current study was to
investigate how these executive functions influence comprehension skill and behavior for elderly
bilinguals.
Approach: We asked whether older adult monolinguals and bilinguals were capable of generating
predictions online, and whether cognitive control and language regulation ability were related to
the magnitude of prediction costs.
Data and Analysis: Participants (N = 27) read sentences while their electroencephalogram was
recorded, and completed a battery of language and cognitive performance tasks.
Findings: While older adult monolinguals showed some sensitivity to prediction error, older
adult bilinguals produced greater prediction costs, an effect that was significantly correlated with
both age and control ability.
Corresponding author:
Megan Zirnstein, Department of Psychology, University of California, 900 University Avenue, Riverside, CA 92521, USA.
Email: megan.zirnstein@ucr.edu
763143IJB0010.1177/1367006918763143International Journal of BilingualismZirnstein et al.
research-article2018
Article

2 International Journal of Bilingualism 00(0)
Originality: This study is the first to show ERP evidence that bilinguals are capable of forming
predictions during comprehension in older adulthood.
Significance: These results have important implications for the ways in which bilingualism may
influence comprehension across the lifespan. While healthy aging has been proposed to lead to
declines in executive function these declines may be mitigated for bilinguals, who have a wealth of
experience in negotiating language-related conflict.
Keywords
Bilingualism, cognitive control, comprehension, aging, event-related potentials
Introduction
A critical question in language science is whether experience in managing the use of multiple lan-
guages has an effect on cognition and neuroplasticity across the lifespan. Some have proposed that
extended bilingual language experience has positive consequences for executive function and even
health outcomes (Alladi et al., 2015; Bak, Vega-Mendoza, & Sorace, 2014; Gold, Kim, Johnson,
Kryscio, & Smith, 2013). These effects are typically found in early development or later in life
when older adults are undergoing cognitive decline (Baum & Titone, 2014; Bialystok, 2017). For
young adults, however, results are often mixed, which some have taken as evidence that no positive
consequences for bilingualism exist (de Bruin, Treccani, & Della Sala, 2015; García-Pentón,
García, Costello, Duñabeitia, & Carreiras, 2016; Valian, 2015; but see Bialystok, Kroll, Green,
MacWhinney, & Craik, 2015), or that they are only likely to occur under special circumstances
(Costa, Hernández, Costa-Faidella, & Sebastián-Gallés, 2009). In addition, the manner in which
these changes in executive function might occur, and the degree to which this pertains to all bilin-
guals, is a topic of active debate.
The most detailed account of the consequences that bilingualism might have for cognition is the
Adaptive Control Hypothesis (Abutalebi & Green, 2007; Green & Abutalebi, 2013). According to
this hypothesis, the ways that bilinguals engage the two languages can impose differential demands
on the executive function system and the neurocircuitry that supports it. This includes attending to
the appropriate language in a conversational setting, monitoring for changes in speaker identity
and potential language switching requirements, and continuously suppressing some or all of the
language not in use. It is this last behavior, the suppression of the non-target language, which has
been highlighted as being uniquely difficult for bilinguals (Van Assche, Duyck, & Gollan, 2013).
However, without this ability, bilingual language production and comprehension would result in
severe cross-language interference and multiple speech production and processing errors, which is
not the case in daily bilingual language use. This suggests that bilinguals are adept at appropriately
regulating the unintended language. Accordingly, language regulation is a potential locus for the
positive consequences of bilingualism, where the ability to exercise control over language has been
hypothesized to transfer to control in other domains.
Executive control in comprehension
Most research on the consequences of bilingualism focuses on the recruitment of executive func-
tions during production, but not comprehension.1 Language comprehension is rife with opportuni-
ties for the employment of these cognitive skills, and multiple studies have shown that executive
function ability, cognitive control in particular, predicts performance in online language comprehen-
sion, for both monolinguals and bilinguals. In a series of studies, Novick, Hussey, Teubner-Rhodes,

Zirnstein et al. 3
Harbison, and Bunting (2014; Teubner-Rhodes et al., 2016) tested whether working memory and
control ability predicted readers’ ability to resolve syntactic ambiguities online. Their results showed
that better inhibitory control predicted better comprehension, that bilinguals were more successful
at performing a version of the control task that involved greater conflict, and that training in this task
resulted in improved comprehension performance.
Whereas this previous work investigated how the use of control during comprehension may be
common across monolinguals and bilinguals, others have turned to the unique nature of bilingual
comprehension. When reading or listening, bilinguals are required to attend to the target language
in question. However, the non-target language is also active (Van Hell & Tanner, 2012), and can
influence processing of the target language at multiple stages. The most notable instance of this is
the case of words that either share meaning and form across both languages (i.e., cognates) or
share their form, but not their meaning (i.e., interlingual homographs). When cross-language
overlap converges, as is the case for cognates, processing benefits occur. When there is cross-
language divergence and conflict, as is the case for homographs, there are processing costs.
However, during bilingual language use, parallel activation is constantly in flux. As a result, while
the suppression of the non-target language is often regarded as the standard for successful produc-
tion and comprehension (i.e., due to a reduction in cross-language interference), this may not
necessarily be true in naturalistic settings. For example, a bilingual reader, who has successfully
down-regulated the activation of the non-target language, may not experience much of a process-
ing benefit when encountering cognates. Thus, while non-target language suppression can sup-
port target language processing in some respects, it can also reduce the cross-language benefits
often available to bilinguals.
Pivneva, Mercier, and Titone (2014) investigated this interplay during a reading comprehension
task that involved sentences with embedded French-English cognates (e.g., piano) and homographs
(e.g., chat). They took into account whether bilinguals were high in proficiency, and therefore
likely to have greater parallel activation, and their ability to suppress conflicting information, as
indexed by their performance on multiple cognitive control measures. Bilinguals with greater
cross-language proficiency experienced the greatest cognate benefits, while those with better con-
trol ability were less susceptible to homograph interference. These findings introduce an added
level of complexity for bilingual comprehension: it is not only a matter of understanding the prob-
lem that cross-language competition presents, but also how bilinguals attempt to resolve that com-
petition online. In particular, it is important to understand how an individual’s language and
cognitive profile may collectively contribute to baseline parallel language activation states and
subsequent ability to regulate those states.
The approaches we have outlined are useful in highlighting what aspects of language experi-
ence are critical for bilingual comprehension, and can help further our understanding of how exec-
utive functions may be impacted by a lifetime of bilingual experience. When considering how
bilingual comprehension might unfold and what repercussions it may have for long-term changes
in cognitive ability, several key issues arise. Firstly, it is important to note the degree of parallel
activation likely to be experienced during testing. This can take into account degree of proficiency
or dominance, but also age of acquisition, immersion context, and, especially for older adults, the
trajectory of language use and loss. Secondly, if cross-language interference is likely to occur,
domain-general executive functions appear to be at play at multiple levels. At one level, the non-
target language itself may be regulated, so that it does not unduly interfere with target language
processing.2 This regulatory ability can, in turn, be influenced by language dominance and the
immersion context itself, as suppression of a less proficient or non-immersed language is likely to
be less effortful. Finally, at another level, other features of the target language may require the
recruitment of cognitive control, in a manner that should be experienced to some degree by any

4 International Journal of Bilingualism 00(0)
individual proficient in that language, regardless of their bilingual status. This could be a conflict
experienced as a result of a syntactic ambiguity or the repercussion of other sentence processes,
such as prediction.
Prediction in comprehension
When reading or listening, comprehenders generate predictions for the meaning of upcoming
words, especially in contexts that are highly semantically constraining. This process can produce
processing benefits when a word with the expected meaning is encountered (Van Berkum, 2008).
However, when expectations are not met, and a less expected word is presented, processing costs
occur. This can be realized in the event-related potentials (ERPs) time-locked to the unexpected
word, including a late, frontally distributed positivity. Peaking at around 500–800 milliseconds,
this ERP effect has been related to the efforts that comprehenders undergo when plausible language
input conflicts with a prediction that has been generated (e.g., DeLong, Groppe, Urbach, & Kutas,
2012; Federmeier, Wlotko, DeOchoa-Dewald, & Kutas, 2007; Federmeier et al., 2010). These
costs could be due to attempts to update expectations for future reading (Kuperberg & Jaeger,
2016) or attempts to revise or suppress a mis-prediction that has come in conflict with a presented
word.
Whether prediction error costs are the result of updating preferences for future prediction pro-
cesses, the result of mediating conflict, or some combination thereof, it appears likely that cogni-
tive control ability may support recovery from prediction error, and thus lead to attenuated ERP
effects. This hypothesis was investigated in a recent study by Zirnstein, van Hell, and Kroll (in
press), in which young adult monolinguals and bilinguals were asked to read sentences while their
electroencephalogram (EEG) was recorded. The critical comparison (see Example 1) involved
sentences of high semantic constraint with embedded target words that were either expected (i.e.,
tip) or unexpected (i.e., ten).
1. They paid for their meal, but forgot to leave a tip/ten for the waitress.
ERPs time-locked to these target words revealed an effect of unexpectedness in frontal electrode
sites, such that less expected words elicited a larger frontally distributed positivity from 500 to 700
ms when compared to expected words. This pattern was found not only for young adult monolin-
gual readers, but also for bilingual readers who had exhibited skill in regulating the non-target
language. Specifically, young adult bilinguals in this study were reading in their less-dominant
second language (L2), but were also immersed in an L2 context that required regulation of the first
language (L1). Their ability to subsequently de-regulate and use the L1 was measured using a ver-
bal fluency task. Previous research (Linck, Kroll, & Sunderman, 2009) has shown that these same
immersion conditions can lead to a reduction in L1 verbal fluency, due to immersion-related dif-
ficulties with language regulation. Likewise, for the young adult bilinguals in the study by Zirnstein
and colleagues (in press), better performance in L1 fluency (i.e., better language regulation) sig-
nificantly predicted the magnitude of prediction error cost in the ERPs. Once such a prediction
error had occurred, better performance on a domain-general executive control task subsequently
mediated the magnitude of this cost, in both monolinguals and bilinguals.
These findings suggest that L2 prediction relies, at least in part, on the ability to free up cogni-
tive resources by regulating the demands that parallel language activation imposes. Previous
research with older adult monolinguals (Federmeier, Kutas, & Schul, 2010) also supports the idea
that executive function ability, and perhaps the availability of cognitive resources, is key to under-
standing whether prediction is likely to occur in populations whose language processing is under

Zirnstein et al. 5
constraint. In their study, only elderly monolinguals with better performance on a semantic cate-
gory fluency task were capable of generating predictions and experiencing subsequent ERP costs
due to unexpectedness. Although the ability to predict the meaning of upcoming words appears
quite prevalent in the processing behavior of young adults, the ability to engage in the same,
resource-demanding processing behaviors in older adulthood appears to rely heavily on executive
functions that are known to decline with age (Brickman et al., 2005).
Current study
The results from this prior work reveal two potential properties of comprehension that are highly
relevant for bilinguals: (1) when processing in one language, regulating the parallel activation of
the non-target language can support comprehension; and (2) once this is in place, aspects of execu-
tive function that support comprehension ability across both monolinguals and bilinguals can be
observed. What is unknown is to what extent these behaviors might work in coordination to sup-
port comprehension for bilingual older adults. While monolingual older adults are less likely to
engage in prediction processes (Wlotko & Federmeier, 2012; Wlotko, Federmeier, & Kutas, 2012),
the persistence of other executive functions (i.e., verbal fluency) may be in place to support predic-
tion generation (DeLong et al., 2012; Federmeier et al., 2010) in a manner that is not typical for the
population at large. These findings are similar to what Zirnstein and colleagues (in press) found
with younger adult bilinguals, suggesting a potential overlap between seemingly disparate lan-
guage processing contexts.
Age-related cognitive decline appears to be an important factor in determining whether older
adults engage in prediction strategies during reading. Some research has suggested that executive
functioning, in particular inhibitory control ability, tends to decline with age (e.g., Hasher, Lustig,
& Zacks, 2007; Hasher & Zacks, 1988), which supports the idea that older adult monolinguals may
have difficulty supporting comprehension strategies that are highly resource demanding. However,
very little is known about the neural changes that occur as a result of age-related cognitive decline
(Burke & Graham, 2012), or how these changes may specifically be delayed as a result of multiple
language use. As such, older adult bilinguals are an ideal group for testing the potential conse-
quences of bilingualism for executive control and comprehension. The goal of the current study is
to investigate how prediction processes might unfold for readers who have benefited from a life-
time of experience in regulating the use of more than one language.
Methods
Participants
Twenty-seven older adults (aged 61–83 years; M = 69.81, SD = 6.25; 16 female), with an average
educational background of 16.89 years (SD = 2.61) participated in the study. Participants were
recruited from the community in the wider Toronto area in Ontario, Canada, and were categorized as
bilingual (N = 12) or monolingual (N = 15) according to their language background. Bilingual and
monolingual groups were matched in age (F(1,25) = .474, p = .498) and education (F(1,25) = .404, p
= .531). All participants were required to be highly proficient in English, right-handed, have corrected
or corrected-to-normal vision, and have no prior history of neurological or reading disorders. They
were compensated 16 Canadian dollars per hour for taking part in the study.
Participants completed the mini-mental state exam (MMSE; Folstein, Folstein, & McHugh,
1975), and scored 27 or higher (30 maximum) in order to be included in the study. Monolinguals
and bilinguals did not differ in their scores (M = 28.96, SD = 1.16; F(1,25) = .171, p = .682).

6 International Journal of Bilingualism 00(0)
Participants also completed Raven’s Progressive matrices (Raven, Raven, & Court, 2000), a meas-
ure of inductive reasoning, in order to estimate general cognitive functioning, which also did not
differ between groups (M = 76.68%, SD = 11.85; F(1,25) = .778, p = .387).
Participants completed a language history questionnaire at the beginning of the study. On a
scale of 1–7 (1 being very poor and 7 being very high), monolinguals rated their ability to read,
write, speak, and comprehend their L1 (English) at an average of 6.73 (SD = 0.59), which was not
different from how bilinguals rated their English ability (M = 6.81, SD = 0.44; F(1,25) = .148, p =
.704). Bilingual participants reported knowing a wide range of languages or dialects other than
English (range: 2–5, M = 3.42, SD = 1.24; including Cantonese, Dutch, French, German, Hokkien,
Hindi/Urdu, Italian, Japanese, Malay, Polish, Sindhi, Sinhalese, Spanish, Ukrainian, and Yiddish).
Self-reported proficiency in their most dominant language other than English (M = 4.85, SD =
2.40) was significantly lower than their English ratings (t(11) = 2.917, p = .014). In addition, while
some reported knowledge of a third, fourth, and even fifth language, none were able to successfully
complete the verbal fluency task for these other languages.
There was, accordingly, a great deal of diversity in the language backgrounds of the bilingual
participants, which included English native speakers with advanced L2 experience (N = 3), bal-
anced bilinguals (in terms of proficiency and use at time of testing; N = 6), and bilinguals whose
dominance had shifted from their original L1 to their later-acquired, yet now more dominant L2
(N = 3). Although it is not within the scope of the current study to compare across these sub-types
of the bilingual experience, it is important to note that all were included in the sample, and that
our measures of language ability (i.e., verbal fluency) were designed to quantify the potential
consequences of this variability.
Materials
Sentence stimuli. Materials consisted of 120 critical items and 40 filler sentences with low semantic
constraint, all in English (see Zirnstein et al., in press, for a detailed description). Of the critical
items, participants saw 40 high constraint sentences with expected target words, 40 high constraint
sentences with unexpected target words, and 40 low constraint sentences with neutral target words.
Sentence contexts were matched in length. Target words were embedded within the sentence and
matched on length and frequency (SUBTLEXus; Brysbaert & New, 2009). Cloze probabilities
(provided by 38 native English speakers who did not take part in the main study) significantly dif-
fered between the high and low constraint contexts (high: .68–1.00; low: .10–.47). Mean associa-
tion strength between targets and all content words in the context (Edinburgh Associative
Thesaurus; Wilson, 1988) was kept at or below .01 for all conditions. All materials were con-
structed to avoid strong collocations between the target and words in prior context (e.g., “The man
proposed to her by getting on bended… knee”). Two words prior to and following the targets were
kept the same across conditions. Items and conditions were counter-balanced across lists.
Cognitive task battery. Participants completed multiple tasks designed to assess individual differ-
ences in cognitive function, including a Semantic Category Verbal Fluency task (Luo, Luk, &
Bialystok, 2010; Rohrer, Wixted, Salmon, & Butters, 1995), an Operation Span task (Turner &
Engle, 1989; Unsworth, Heitz, Schrock, & Engle, 2005), and the distractor version of the AX Con-
tinuous Performance task (AX-CPT; Braver et al., 2001).
In the fluency task, participants viewed four semantic categories in each proficient language
(English for monolinguals; English and one other language for bilinguals), and named as many
tokens as possible within 30 seconds for each category. The total number of correct tokens was
used as the performance score for this task.

Zirnstein et al. 7
In the Operation Span task, participants viewed randomized trials that included sets of paired
simple mathematical equations and letters. Sets could increase from three to seven. For each paired
math question and letter, participants first saw the equation, clicked a mouse, and then saw a num-
ber with a true or false button. They were asked to indicate whether the number matched the solu-
tion to the previous equation. Following this, a letter appeared. The participants’ goal was to
memorize the series of letters in the set, in the correct order, while continuing to solve math prob-
lems, and to recall the series of letters at the end of the set. The total number of correctly recalled
letters, across all sets, was used as the performance measure for this task.
In the AX-CPT, participants viewed 10 practice and 100 critical trials. In each trial, a series of
five letters was presented, one at a time, at the center of the screen, starting with a red-letter cue,
followed by three white-letter distractors, and ending with a red-letter probe. Participants were
asked to press one button in response to every letter, but to change their response at the end of the
sequence if the two red letters (the cue and probe) were A followed by X. AX trials took up 70% of
the total trials, while three other cue-probe conditions took one third of the remaining 30% (AY,
BX, and BY). AY trials were designed to be difficult if participants were highly expecting an X
probe, thus resulting in longer reaction times (RTs) and more errors to the Y probe. BX trials were
intended to tax participants who failed to maintain the goal of the task, and thus took longer to
reject or mistakenly accepted the X probe. Control BY trials reflected performance when neither
the cue nor the probe was manipulated.
Procedure
Participants came in to the lab for two sessions. In the first session, they completed the language
history questionnaire, MMSE, and reading task (during which their EEG was recorded). In the
second session, they completed the cognitive task battery.
Reading task. Participants were tested individually in a room separate from the experimenter. Sen-
tence stimuli were presented on a computer screen approximately 100 cm from the participant,
using E-prime 2.0 software (Psychology Software Tools, Pittsburgh, PA). Participants were
instructed to relax and restrict motor movement and muscle tension while the sentences were being
displayed. At the start of each trial, a blank screen appeared for 1000 ms, followed by a fixation
cross and each word in the sentence, which were presented via rapid serial visual presentation
(RSVP), with a 400 ms presentation duration and 200 ms inter-stimulus interval (ISI) blank screen.
After the final word, a blank screen was presented for 750 ms. After each sentence, a prompt
appeared, either instructing the participant to press a button to continue, or with a true or false state-
ment about the prior sentence.
EEG recording and data analysis. Continuous EEG was recorded from 64 active Ag/AgCl electrodes
(Biosemi Active Two system, Amsterdam, the Netherlands), with a sampling rate of 512 Hz and at
a bandwidth of .01–100 Hz, while referenced to the common mode sense (CMS) electrode, and
with the driven right leg (DRL) electrode acting as the common ground. Six additional electrodes
were placed on the left and right mastoids (for later off-line algebraic re-referencing), as well as on
the outer canthi of each eye and above and below the left eye. Impedances were kept below 20 kΩ.
EEG pre-processing was conducted in EEGLAB (Delorme & Makeig, 2004) and ERPLAB
(erpinfo.org/erplab) toolboxes for MATLAB. Prior to off-line averaging, all single trial waveforms
were screened for amplifier blocking, muscle artifacts, horizontal eye movements, and blinks
across 1200 ms epochs starting 200 ms before the onset of the critical target words. For nine par-
ticipants, blink artifacts were corrected in the continuous EEG using Infomax ICA decomposition

8 International Journal of Bilingualism 00(0)
in EEGLAB. EEG was re-referenced off-line to the average of the left and right mastoids. One
thousand millisecond epochs were extracted at the onset of critical words, with 200 ms pre-stimu-
lus baselines. Grand average ERPs were filtered (0.1–25 Hz), and statistical analyses were con-
ducted on these data.
Repeated measures analyses of variance (ANOVAs) were conducted on a subset of electrodes
(N = 15) to compare within-participant factors of Unexpectedness (Expected, Unexpected),
Anteriority (Frontal, Central, Posterior), and Electrode (with 5 levels), along with a between-par-
ticipants factor of group (monolingual, bilingual). Electrode subsets were determined a priori, with
five electrode sites for frontal (F3, Fz, F4, FC1, and FC2), central (C3, Cz, C4, CP1, and CP2), and
posterior regions (P3, Pz, P4, O1, and O2). Analyses were conducted on the mean amplitude across
three time windows of interest: 300–500, 500–700, and 700–900 ms. Planned contrasts were con-
ducted on the high constraint sentence conditions containing expected or unexpected target words.3
Greenhouse–Geisser corrected p-values were used for all analyses with more than 2 degrees of
freedom in the numerator.
Results and discussion
Below, we report the results for ERPs and the behavioral task battery, with language groups com-
pared to one another, followed by individual difference analyses within each language group sepa-
rately. For group comparisons within task, monolingual and bilingual performance appeared to be
very similar. However, when differences in age, verbal fluency, working memory, and cognitive
control ability are taken into account, we observed distinct effects of bilingualism on
comprehension.
ERPs
No significant interactions with group (monolingual versus bilingual) were found in the ERP anal-
yses. The following results, therefore, focus on effects of Unexpectedness (high constraint expected
versus high constraint unexpected), Anteriority (frontal, central, posterior electrode regions), and
electrode (with five levels within each region). As can be seen in Figure 1, older adults exhibited a
frontally distributed positivity in response to unexpected words. The onset of this effect was earlier
than anticipated (around 350 ms), and the effect lasted throughout the epoch in some channels
(e.g., Fz).
Three hundred to five hundred. There was a significant interaction between Unexpectedness and
Anteriority (F(2,50) = 11.488, p < .001), as well as an interaction between Unexpectedness and
Electrode (F(4,100) = 3.435, p = .021). Follow-up analyses were conducted on each electrode
region separately (frontal, central, and posterior), but no significant effects were found for these
independent regions. Interactions with Unexpectedness were primarily driven by changes in polar-
ity across electrode regions from a small positivity in frontal sites to a negativity in posterior sites
(see Figure 2).
Five hundred to seven hundred. We expected to see an effect of Unexpectedness if older adult
participants had been engaging in prediction processes and subsequently experienced difficulty
with prediction error. There was indeed an interaction between Unexpectedness and Anteriority
in this time window (F(2,50) = 10.113, p = .002). A subsequent main effect of Unexpectedness
was observed for the frontal region (F(1,25) = 4.529, p = .043), while no effects were found
independently for the central or posterior regions. Figure 2 shows the magnitude of this frontally

Zirnstein et al. 9
distributed positivity in response to unexpected words. Although no group interactions were
observed, the magnitude of these prediction error costs may have been driven by bilinguals.
Seven hundred to nine hundred. Previous work on prediction error ERP costs has sometimes indi-
cated a delayed effect in later time windows, especially for bilingual readers (e.g., Zirnstein et al.,
in press). We observed a significant interaction in this late time window between Unexpectedness
and Anteriority (F(2,50) = 5.216, p = .009). A main effect of Unexpectedness was found for the
frontal region (F(1,25) = 4.940, p = .036), while no effects were found in the central or posterior
regions. Prediction error costs in this late time window appear to be a continuation of the frontal
positivity observed from 500 to 700 ms (see Figure 2).
Cognitive tasks
Semantic category fluency. Monolinguals and bilinguals performed similarly on the verbal fluency
task when naming words in English (F(1,25) = .067, p = .798), with monolinguals producing 45.87
tokens on average (SD = 8.30) and bilinguals producing 44.83 (SD = 12.44). This is likely a reflec-
tion of the English-supporting immersion context in which participants had been living for many
years, as well as the impact this context had on bilinguals’ language dominance. Bilingual partici-
pants also completed the task with a separate set of semantic categories in the language they were
most fluent in other than English. Performance in this language was lower (M = 34.42, SD =
22.89), but also more variable than for English (difference between languages: M = 20.42, SD =
18.55), and was therefore not significantly different (t(11) = 1.945, p = .078). English fluency was
Figure 1. Grand average event-related potential waveforms for the effect of Unexpectedness and
Constraint, across nine representative electrodes.

10 International Journal of Bilingualism 00(0)
taken as a measure of proficiency in the immersed language, while fluency in the other language
was taken as a measure of continued bilingual language maintenance and use.
Operation span. Bilingual participants (M = 51.58, SD = 15.36) outperformed monolinguals (M
= 42.13, SD = 16.52) in the total number of letters correctly recalled on the span task. However,
this varied widely among participants, and was not significant in a group comparison (F(1,25) =
2.320, p = .140).
AX-CPT. A repeated measures ANOVA was conducted to assess the degree to which the within-
participants factor of condition (AY, BX, and BY) interacted with group (monolingual, bilingual),
for both RTs to correct trials and proportion of errors. While no significant interactions with group
were found (Fs < 1.00, ps > .05), there was a significant effect of condition for both RTs (F(2,48)
= 50.891, p < .001) and errors (F(2,48) = 5.152, p = .009). Planned contrasts (comparing conditions
AY and BX to the control condition BY) revealed a significant effect for AY in both RTs (F(1,24)
= 53.215, p < .001) and errors (F(1,24) = 15.810, p = .001), as well as a significant effect for BX
RTs (F(1,24) = 5.765, p = .024) and errors (F(1,24) = 9.264, p = .006). While the AY condition was
the most difficult in terms of RT, the BX condition produced more errors (see Figures 3 and 4).
An additional analysis was performed to determine whether monolinguals and bilinguals may have
differed in the strategies they used when approaching task rules. Some participants may have biased
themselves toward maintaining the goal of the task (preparing for X probes, following A cues), result-
ing in longer RTs and more errors to AY trials. Other participants, however, may have attempted to
balance their attention toward the task goal and the possibility of needing to inhibit responses. Previous
researchers have calculated this bias, or Behavioral Shift Index (BSI; Paxton, Barch, Storandt, &
Braver, 2006), using the following formula: (AY – BX) / (AY + BX). We calculated BSI composite
scores (with z-scored RTs and errors). A more positive score on this index implies an over-reliance on
goal maintenance, with subsequent difficulty inhibiting responses in AY trials. Values closer to zero
imply a more balanced approach between maintaining the task goal and correctly inhibiting incorrect
responses. Both monolinguals (M = .21, SD = 2.03) and bilinguals (M = .74, SD = 3.36) exhibited BSI
indices above zero, but did not differ significantly from one another (F(1,25) = .257, p = .617).
Figure 2. Magnitude of the effect of Unexpectedness for monolinguals and bilinguals across all time
windows and electrode regions.

Zirnstein et al. 11
Correlations between ERP effects and cognitive performance. A clear finding from the ERP analyses is
that older adults appear to be quite successful at engaging in prediction processes during online
reading, at least enough to experience subsequent costs when those predictions are not verified.
This finding is somewhat in conflict with results from previous studies (DeLong et al., 2012;
Federmeier et al., 2010), which have found these effects only for older adult monolinguals with
higher executive function ability (i.e., verbal fluency). Group differences in our study, however,
were not observed either in the ERP analyses or in the individual analyses for verbal fluency, work-
ing memory, or performance on the AX-CPT. Despite this, the magnitude of the frontal positivity
seen in the grand average waveforms (Figure 1), and the subtle trend for bilingual older adults to
show a larger effect in this region from 500 to 700 ms (see Figure 2), suggests that a more complex
pattern between bilingual status and reading strategy may be at play. We therefore conducted a
series of correlational analyses to assess the relationship between age and cognitive ability, and the
magnitude of prediction error costs in the ERP results.
For monolinguals, age, English verbal fluency, and performance on the AX-CPT (as indexed by
the BSI) did not significantly correlate with the magnitude of prediction error cost, in any time
Figure 3. Mean reaction time performance in milliseconds on the AX Continuous Performance task (AX-
CPT).
Figure 4. Mean proportion of errors on the AX Continuous Performance task (AX-CPT).

12 International Journal of Bilingualism 00(0)
window or electrode region. Working memory performance did correlate with the effect of
Unexpectedness, but only in the 700–900 ms time window and only in posterior electrodes (r =
.460, p = .042). Working memory ability may have determined the extent to which elderly mono-
linguals were capable of maintaining previously generated predictions over time (i.e., as new
words in the sentence were presented), thus leading to greater prediction error costs.
For bilinguals, however, two prominent effects were revealed. Firstly, age significantly and
negatively correlated with the magnitude of prediction error costs in frontal electrode sites in the
300–500 (r = –.562, p = .029) and 500–700 ms time windows (r = –.543, p = .034). As age
increased, the magnitude of prediction error costs decreased. The presence of this ERP effect,
although technically a cost, is also an indication that readers were able to generate predictions dur-
ing online comprehension, an ability that appears to decline more dramatically with age for elderly
bilinguals, in comparison to monolinguals. Secondly, for bilinguals, a more balanced approach to
the AX-CPT (leading to a BSI closer to zero) was associated with prediction error costs in frontal
electrode sites that appeared earlier (from 300 to 500 ms: r = –.506, p = .047) and remained longer
(from 700 to 900 ms: r = –.508, p = .046) in the ERP record than the typical 500–700 ms effect.
Based on this finding, it is quite likely that elderly bilinguals with balanced cognitive control strat-
egies drove the frontal positivity in the grand average ERPs to appear earlier and last longer than
has been previously demonstrated in any published study of lexical prediction.
Together, these findings suggest that a younger group of older adult bilinguals were the most
likely to engage in prediction during online comprehension. Older adult monolinguals, in contrast,
did not show the same age benefit, and were only likely to have engaged in prediction if their work-
ing memory ability was sufficiently high. Bilinguals also showed evidence for the recruitment of
executive control during comprehension in a way that monolinguals did not. Bilinguals who more
strategically approached the executive control task experienced prediction error costs across a
much longer time course than we expected. It is unclear whether the ability to balance between
goal maintenance and inhibition in the executive control task is what drove older adult bilinguals
to generate predictions in the first place, or whether it instead only determined the time course of
prediction error costs in the ERPs. Regardless, these results indicate that older adult bilinguals are
capable of engaging in prediction processes successfully during reading.
General discussion
The current study provides the first evidence that older adult bilinguals can engage in the highly
resource-demanding task of predicting during online comprehension. The relationship between
executive functions hypothesized to decline due to healthy aging, namely working memory and
cognitive control, and online prediction was not only complex, but also patterned somewhat differ-
ently between language groups. Whereas declines in working memory led to similar declines in
prediction capability for monolinguals, bilinguals instead appeared to be more broadly capable of
generating predictions. However, in the older age range, these qualitative group differences were
not present. More research is needed to identify the exact mechanisms driving these findings, but
this study suggests they are related to the ways that comprehension and executive function change
throughout the aging process, which may be quite different for monolinguals and bilinguals.
Notably, the prediction effects observed in the current study mirror similar anticipatory pro-
cesses found in perception for habitual code-switchers (e.g., Fricke, Kroll, & Dussias, 2016). While
the current findings were isolated in the neurophysiological performance of a highly diverse group
of elderly bilinguals, future work will need to establish a more direct link between the lifespan
trajectory of bilingual language experience, including code-switching behaviors, and how this
impacts the relationship between comprehension and executive function in older adulthood. In
particular, it will be important to investigate the potential relationship between the cross-language

Zirnstein et al. 13
anticipation of language or code-switches and within-language anticipatory processes, such as the
prediction of specific words or meanings.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publi-
cation of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publica-
tion of this article: This research and the writing of this article was supported in part by NSF Grant SMA-
1409973 to M. Zirnstein, J.G. van Hell, and J.F. Kroll, NSF Grant SMA-1657782 to MZ and JK, NSF Grant
OISE-1968369 to JK and JvH, NSF Grants BCS-1535124 to JK and BCS-1349110 and DUE-1561660 to
JvH, and NIH Grant HD082796 to JK.
Notes
1. However, see work by Blumenfeld, Schroeder, Bobb, Freeman and Marian (2016) and Martín, Macizo,
and Bajo (2010) for evidence that bilinguals are adept at utilizing inhibitory control during single word
comprehension.
2. See Green (2017) for recent discussion about the potential limits of bilingual language suppression in the
context of L2 use.
3. Analyses were also conducted to establish the effect of Constraint (high versus low, with expected ver-
sus neutral targets), which indicated significant effects in the 300–500 window (main effect of Constraint
(F(1,26) = 9.916, p = .004); and Constraint by Electrode interaction (F(4,104) = 3.156, p = .025)), with
neutral targets producing a larger and more centro-posterior N400 effect than expected targets (see Figure 1).
References
Abutalebi, J., & Green, D. (2007). Bilingual language production: The neurocognition of language represen-
tation and control. Journal of Neurolinguistics, 20(3), 242–275.
Abutalebi, J., Guidi, L., Borsa, V., Canini, M., Della Rosa, P. A., Parris, B. A., & Weekes, B. S. (2015).
Bilingualism provides a neural reserve for aging populations. Neuropsychologia, 69, 201–210.
Alladi, S., Bak, T. H., Mekala, S., Rajan, A., Chaudhuri, J. R., Mioshi, E., . . . Kaul, S. (2015). Impact of
bilingualism on cognitive outcome after stroke. Stroke, 47(1), 258–261.
Bak, T. H., Vega-Mendoza, M., & Sorace, A. (2014). Never too late? An advantage on tests of auditory atten-
tion extends to late bilinguals. Frontiers in Psychology, 5, 485.
Baum, S., & Titone, D. (2014). Moving toward a neuroplasticity view of bilingualism, executive control, and
aging. Applied Psycholinguistics, 35, 857–894.
Bialystok, E. (2017). The bilingual adaptation: How minds accommodate experience. Psychological Bulletin,
143(3), 233–262.
Bialystok, E., Kroll, J. F., Green, D. W., MacWhinney, B., & Craik, F. I. M. (2015). Publication bias and the
validity of evidence: What’s the connection? Psychological Science, 26(6), 944–946.
Blumenfeld, H. K., Schroeder, S. R., Bobb, S. C., Freeman, M. R., & Marian, V. (2016). Auditory word rec-
ognition across the lifespan: Links between linguistic and nonlinguistic inhibitory control in bilinguals
and monolinguals. Linguistic Approaches to Bilingualism, 6(1–2), 119–146.
Braver, T. S., Barch, D. M., Keys, B. A., Carter, C. S., Cohen, J. D., Kaye, J. A., . . . Reed, B. R. (2001).
Context processing in older adults: Evidence for a theory relating to cognitive control to neurobiology in
healthy aging. Journal of Experimental Psychology: General, 130(4), 746–763.
Brickman, A. M., Paul, R. H., Cohen, R. A., Williams, L. M., MacGregor, K. L., Jefferson, A. L., . . . Gordon,
E. (2005). Category and letter verbal fluency across the adult lifespan: Relationship to EEG theta power.
Archives of Clinical Neuropsychology, 20(5), 561–573.
Brysbaert, M., & New, B. (2009). Moving beyond Kucera and Francis: A critical evaluation of current word
frequency norms and the introduction of a new and improved word frequency measure for American
English. Behavior Research Methods, 41(4), 977–990.

14 International Journal of Bilingualism 00(0)
Burke, D. M., & Graham, E. R. (2012). The neural basis for aging effects on language. In M. Faust (Ed.), The
handbook of the neuropsychology of language (pp. 778–800). Oxford: Blackwell Publishing.
Costa, A., Hernández, M., Costa-Faidella, J., & Sebastián-Gallés, N. (2009). On the bilingual advantage in
conflict processing: Now you see it, now you don’t. Cognition, 113(2), 135–149.
de Bruin, A., Treccani, B., & Della Sala, S. (2015). Cognitive advantage in bilingualism: An example of
publication bias? Psychological Science, 26(1), 99–107.
DeLong, K. A., Groppe, D. M., Urbach, T. P., & Kutas, M. (2012). Thinking ahead or not? Natural aging and
anticipation during reading. Brain and Language, 121(3), 226–239.
Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox for analysis of single-trial EEG
dynamics. Journal of Neuroscience Methods, 134(1), 9–21.
Federmeier, K. D., Kutas, M., & Schul, R. (2010). Age-related and individual differences in the use of predic-
tion during language comprehension. Brain and Language, 115(3), 149–161.
Federmeier, K. D., Wlotko, E., DeOchoa-Dewald, E., & Kutas, M. (2007). Multiple effects of sentential con-
straint on word processing. Brain Research, 1146(1), 75–84.
Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). “Mini-mental state.” A practical method for grading
the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12, 189–198.
Fricke, M., Kroll, J. F., & Dussias, P. E. (2016). Phonetic variation in bilingual speech: A lens for studying the
production-comprehension link. Journal of Memory and Language, 89, 110–137.
García-Pentón, L., García, Y. F., Costello, B., Duñabeitia, J. A., & Carreiras, M. (2016). The neuroanatomy
of bilingualism: How to turn a hazy view into the full picture. Language, Cognition and Neuroscience,
31(3), 303–327.
Gold, B. T., Kim, C., Johnson, N. F., Kryscio, R. J., & Smith, C. D. (2013). Lifelong bilingualism maintains
neural efficiency for cognitive control in aging. The Journal of Neuroscience, 33(2), 387–396.
Green, D. W. (2017). The interactional challenge: L2 learning and use in the third age. In D. Gabryś-Barker
(Ed.), Third age learners of foreign languages. Bristol, UK: Multilingual Matters.
Green, D. W., & Abutalebi, J. (2013). Language control in bilinguals: The adaptive control hypothesis.
Journal of Cognitive Psychology, 25(5), 515–530.
Hasher, L., Lustig, C., & Zacks, R. T. (2007). Inhibitory mechanisms and the control of attention. In A.
Conway, C. Jarrold, M. Kane, A. Miyake, & J. Towse (Eds.), Variation in working memory. (pp. 227–
249). New York: Oxford University Press.
Hasher, L., & Zacks, R. T. (1988). Working memory, comprehension, and aging: A review and a new view.
In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 22, pp. 193–225). New York,
NY: Academic Press.
Kuperberg, G. R., & Jaeger, T. F. (2016). What do we mean by prediction in language comprehension?
Language, Cognition and Neuroscience, 31(1), 32–59.
Linck, J. A., Kroll, J. F., & Sunderman, G. (2009). Losing access to the native language while immersed in a
second language: Evidence for the role of inhibition in second-language learning. Psychological Science,
20(12), 1507–1515.
Luo, L., Luk, G., & Bialystok, E. (2010). Effect of language proficiency and executive control on verbal flu-
ency performance in bilinguals. Cognition, 114(1), 29–41.
Martín, M. C., Macizo, P., & Bajo, T. (2010). Time course of inhibitory processes in bilingual language pro-
cessing. British Journal of Psychology, 101(4), 679–693.
Novick, J. M., Hussey, E., Teubner-Rhodes, S., Harbison, J. I., & Bunting, M. F. (2014). Clearing the gar-
den-path: Improving sentence processing through cognitive control training. Language, Cognition and
Neuroscience, 29(2), 186-217.
Paxton, J. L., Barch, D. M., Storandt, M., & Braver, T. S. (2006). Effects of environmental support and strat-
egy training on older adults’ use of context. Psychology of Aging, 21(3), 499–509.
Pivneva, I., Mercier, J., & Titone, D. (2014). Executive control modulates cross-language lexical activa-
tion during L2 reading: Evidence from eye movements. Journal of Experimental Psychology: Learning,
Memory, & Cognition, 40(3), 787–796.
Raven, J., Raven, J. C., & Court, J. C. (2000). Raven manual: Section 3, standard progressive matrices,
including the parallel and plus versions, 2000 edition. Oxford, UK: Oxford Psychology Press Ltd.

Zirnstein et al. 15
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(5), 1127–1139.
Teubner-Rhodes, S. E., Mishler, A., Corbett, R., Andreu, L., Sanz-Torrent, M., Trueswell, J. C., & Novick,
J. M. (2016). The effects of bilingualism on conflict monitoring, cognitive control, and garden-path
recovery. Cognition, 150, 213–231.
Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journal of Memory and
Language, 28(2), 127–154.
Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span
task. Behavior Research Methods, 37(3), 498–505.
Valian, V. (2015). Bilingualism and cognition. Bilingualism: Language and Cognition, 18, 3–24.
Van Assche, E., Duyck, W., & Gollan, T. H. (2013). Whole-language and item-specific control in bilin-
gual language production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9,
1781–1792.
Van Berkum, J. J. A. (2008). Understanding sentences in context: What brain waves tell us. Current Directions
in Psychological Science, 17(6), 376–380.
Van Hell, J. G., & Tanner, D. (2012). Second language proficiency and cross-language lexical activation.
Language Learning, 62(2), 148–171.
Wilson, M. D. (1988). The MRC psycholinguistic database: Machine readable dictionary, Version 2.
Behavioral Research Methods, Instruments and Computers, 20(1), 6–11.
Wlotko, E. W., & Federmeier, K. D. (2012). Age-related changes in the impact of contextual strength on
multiple aspects of sentence comprehension. Psychophysiology, 49(6), 770–785.
Wlotko, E. W., Federmeier, K. D., & Kutas, M. (2012). To predict or not to predict: Age-related differences
in the use of sentential context. Psychology of Aging, 27(4), 975–988.
Zirnstein, M., van Hell, J. G., & Kroll, J. F. (in press). Cognitive control ability mediates prediction costs in
monolinguals and bilinguals Cognition.
Author biographies
Megan Zirnstein is a research scientist in the Department of Psychology at the University of California,
Riverside, and was previously an NSF SBE and NSF PIRE Postdoctoral Fellow in the Department of
Psychology at the Pennsylvania State University. She received her PhD in Psychology from the University of
California, Davis in 2012, and conducts research using behavioral and neuroscience methods to investigate
how language experience influences comprehension and executive function ability across the lifespan.
Janet van Hell is a professor of Psychology and Linguistics at The Pennsylvania State University. Her
Bilingualism and Language Development Lab studies cognitive and neurocognitive processes related to sec-
ond language learning and the use of two languages in children and adults, as well as language processing in
school-aged children with typical or atypical language development. She combines neuropsychological,
behavioral, and linguistic techniques to study language development and bilingual language processing. For
more information, see http://bild.la.psu.edu/
Judith Kroll is a Distinguished Professor of Psychology at the University of California, Riverside, and
Professor Emeritus of Psychology, Linguistics, and Women’s Studies at The Pennsylvania State University.
The research that she and her students conduct concerns the way that bilinguals juggle the presence of two
languages in one mind and brain. Their work shows that bilingualism provides a tool for revealing the inter-
play between language and cognition that is otherwise obscure in speakers of one language alone. For more
information, see https://bilingualismmindbrain.com/