tmpC814 TMP

Journal of Experimental Psychology:
Learning, Memory, and Cognition

2016, Vol. 42, No. 6, 000
2016 American Psychological Association

0278-7393/16/$12.00 http://dx.doi.org/10.1037/xlm0000283
Memory and Language Improvements Following Cognitive

Control Training
Erika K. Hussey
J. Isaiah Harbison
University of Maryland, College Park and University of Illinois

at Urbana-Champaign
University of Maryland, College Park
Susan E. Teubner-Rhodes
Alan Mishler
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
University of Maryland, College Park and Medical University of

South Carolina
Kayla Velnoskey
Jared M. Novick
University of Maryland, College Park and Yale University
Cognitive control refers to adjusting thoughts and actions when confronted with conflict during information processing. We tested whether this ability is causally linked to performance on certain language
and memory tasks by using cognitive control training to systematically modulate peoples ability to
resolve information-conflict across domains. Different groups of subjects trained on 1 of 3 minimally
different versions of an n-back task: n-back-with-lures (High-Conflict), n-back-without-lures (LowConflict), or 3-back-without-lures (3-Back). Subjects completed a battery of recognition memory and
language processing tasks that comprised both high- and low-conflict conditions before and after training.
We compared the transfer profiles of (a) the High- versus Low-Conflict groups to test how conflict
resolution training contributes to transfer effects, and (b) the 3-Back versus Low-Conflict groups to test
for differences not involving cognitive control. High-Conflict training but not Low-Conflict training
produced discernable benefits on several untrained transfer tasks, but only under selective conditions
requiring cognitive control. This suggests that the conflict-focused intervention influenced functioning on
ostensibly different outcome measures across memory and language domains. 3-Back training resulted
in occasional improvements on the outcome measures, but these were not selective for conditions
involving conflict resolution. We conclude that domain-general cognitive control mechanisms are plastic,
at least temporarily, and may play a causal role in linguistic and nonlinguistic performance.
Keywords: cognitive control, cognitive training, conflict resolution, recognition memory, syntactic
ambiguity resolution
While observing and interacting with the environment, people

sometimes face situations that require them to override the first or
dominant reaction that comes to mind. Such regulation of mental
activity is known as cognitive control, a top-down procedure that

enables the adjustment of thoughts and actions when people encounter and must resolve information-conflict (Badre & Wagner,
Erika K. Hussey, Center for Advanced Study of Language and Program

in Neuroscience and Cognitive Science, University of Maryland, College
Park and Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign; J. Isaiah Harbison, Center for
Advanced Study of Language, University of Maryland, College Park;
Susan E. Teubner-Rhodes, Center for Advanced Study of Language and
Program in Neuroscience and Cognitive Science, University of Maryland,
College Park and Department of OtolaryngologyHead and Neck Surgery, Medical University of South Carolina; Alan Mishler, Center for
Advanced Study of Language, University of Maryland, College Park;
Kayla Velnoskey, Center for Advanced Study of Language, University of
Maryland, College Park and Department of Psychology, Yale University;
Jared M. Novick, Center for Advanced Study of Language, Program in
Neuroscience and Cognitive Science, and Department of Hearing and
Speech Sciences, University of Maryland, College Park.
This material is based on work supported, in whole or in part, with

funding from the United States Government. Any opinions, findings,
and conclusions or recommendations expressed in this material are
those of the author(s) and do not necessarily reflect the views of the
University of Maryland, College Park and/or any agency or entity of the
United States Government. A portion of this work was presented at
the 25th Annual CUNY Conference on Human Sentence Processing
(Columbia, SC), the 53rd Annual Meeting of the Psychonomic
Society (Minneapolis, MN), and the 54th Annual Meeting of the Psychonomic Society (Toronto, Canada). This work originally appeared
in Erika Husseys doctoral thesis (University of Maryland, College
Park).
Correspondence concerning this article should be addressed to Erika
K. Hussey, Beckman Institute for Advanced Science and Technology,
University of Illinois at Urbana-Champaign, 405 North Mathews Avenue, Urbana, IL 61801. E-mail: ehussey@illinois.edu
1
HUSSEY ET AL.
2007; Botvinick, Braver, Barch, Carter, & Cohen, 2001; Braver,

Gray, & Burgess, 2007; Miller & Cohen, 2001; Norman & Shallice, 1986; Shimamura, 2000). For example, finding an alternate
route after discovering an unexpected road closure, or revising the
incorrect interpretation of an ambiguous sentence (consider this
BBC Headline: Vomiting bug closes three wards) are both cases
of when one must override a highly active representation (the
familiar way home; the interpretation that a sick insect is responsible for shuttering hospitals). The ability to dynamically alter
ones processing strategies allows goal-directed behavior to be
consistent with context-specific demands, particularly when such
demands diverge from normal routines. Because cognitive control
is a crucial piece of the cognitive architecture that supports certain
memory and language functions (Jonides & Nee, 2006; Novick,
Trueswell, & Thompson-Schill, 2005), and predicts a range of
linguistic and nonlinguistic impairments in neuropsychological
patients (Hamilton & Martin, 2005; Novick, Kan, Trueswell, &
Thompson-Schill, 2009; Robinson et al., 1998, 2005; Vuong &
Martin, 2011), it is desirable to determine whether cognitive control can be increased, even in the short term. Note that we are not
using the term cognitive control synonymously with executive
function. Throughout this article, we define cognitive control more
narrowly, as the behavioral regulation that follows the detection of
information-processing conflict that serves to resolve such conflict. While this definition resembles Miyake et al.s (2000) inhibition construct, we remain agnostic as to whether cognitive control involves inhibiting task-irrelevant representations, promoting
task-relevant ones, or some combination of both processes (see
also Miyake & Friedman, 2012; Munakata et al., 2011).
Here, we test whether focused practice engaging cognitive control mechanisms modulates memory and language behavior under
high conflict resolution demands. The overall goal is to test
whether temporary training that is expressly designed to target
peoples ability to detect and resolve information-conflict improves domain-general cognitive control functioning. Significant
performance improvements in untrained outcome measures would
(a) add to mounting evidence for a general-purpose cognitive
control system, (b) reveal how nonlinguistic cognitive control
abilities interact with, and perhaps even shape recognition memory
and language processing in both production and comprehension,
and (c) suggest that this domain-general system is, to some extent,
plastic.
Process-Specific Approach to Training

Until recently, high-level cognitive abilities like working memory were assumed to remain static throughout the life span. Recently, there has been a flurry of research testing this assumption,
with the goal of improving overall cognitive ability (Au et al.,
2016; Baniqued et al., 2015; Buschkuehl & Jaeggi, 2010; Sternberg, 2008). Although the evidence for increasing general fluid
intelligence through working memory training remains highly controversial (Harrison et al., 2013; Melby-Lervg & Hulme, 2013;
Rabipour & Raz, 2012; Sprenger et al., 2013), the prospect of
employing more specific training to target particular cognitive
procedures (or mechanisms) remains an active possibility (Oelhafen et al., 2013; von Bastian & Oberauer, 2013). Cognitive
control processes may be subject to positive change across the
developmental spectrum if people frequently engage in everyday
activities demanding cognitive control (Bialystok et al., 2006;

Diamond & Lee, 2011; Gray & Thompson, 2004; Neville et al.,
2013; Stine-Morrow et al., 2014). In addition, lab-based interventions that are carefully designed may also up-regulate cognitive
control that can be generalized to novel tasks relying on shared
procedures (e.g., Dahlin, Neely, Larsson, Bckman, & Nyberg,
2008; Hussey & Novick, 2012; Jaeggi et al., 2010; Kueider, Parisi,
Gross, & Rebok, 2012; Morrison & Chein, 2011; Novick, Hussey,
Teubner-Rhodes, Harbison, & Bunting, 2014; cf. Rapport, Orban,
Kofler, & Friedman, 2013; Waris, Soveri, & Laine, 2015).
The idea that cognitive control functions may be plastic and
enriched through experience has both theoretical and practical
implications. On the theoretical side, researchers debate whether
the cognitive control mechanisms that detect and resolve
information-processing conflict are domain-general (broad in
scope) or domain-specific (narrow in scope). For instance, some
claim that when people are faced with conflicting sources of
informationregardless of differences in the type of input, task
goals, or stimulus characteristics cognitive control processes engage consistently across a variety of domains including recognition memory, language processing, and temporal context retrieval
(Fedorenko, 2014; Kan et al., 2013; Miller & Cohen, 2001; Nee,
Jonides, & Berman, 2007; Novick et al., 2005; Rajah, Ames, &
DEsposito, 2008). Others, by contrast, contend that cognitive
control procedures are domain-specific: various nonoverlapping
systems locally support conflict processing for particular types of
tasks and stimuli that are encountered (Akay & Hazeltine, 2011;
Egner, Delano, & Hirsch, 2007). If cognitive control abilities are
domain-general and plastic, then training-transfer effects should be
observable across tasks with high conflict resolution demands
despite ostensible differences in the task environment in which the
conflict is faced. Alternatively, if conflict resolution is mediated by
domain-specific cognitive control, then transfer effects should be
more localized and observed only on assessment measures that
essentially mimic the conflict environment of the training task
(Egner, 2008). Adjudicating between these theoretical views could
inform translational research: the observation of domain-general
cognitive control plasticity could license investigations into
whether training may be a useful treatment for increasing abilities
in neuropsychological patients whose language and memory impairments appear to stem from a general deficit in conflict resolution abilities (Hamilton & Martin, 2005; Novick et al., 2009,
2010; Robinson et al., 1998, 2005; Thothathiri, Schwartz, &
Thompson-Schill, 2010; Thompson-Schill et al., 2002; Vuong &
Martin, 2011, 2014).
As noted above, the goal of improving intelligence through
working memory training has been met with mixed results and
therefore remains hotly debated (Harrison et al., 2013; Jaeggi et
al., 2008; Karbach, Mang, & Kray, 2010; Melby-Lervg & Hulme,
2013; Redick et al., 2013; Sprenger et al., 2013; Shipstead, Hicks,
& Engle, 2012a, 2012b; Thompson et al., 2013). However, there
may be more promise in process-specific training, as the extent
of transfer from training appears to depend on the degree to which
trained and untrained tasks rely on shared cognitive (and neurobiological) resources regardless of domain (Dahlin et al., 2008;
Oelhafen et al., 2013; Persson et al., 2013; Schneiders, Opitz,
Krick, & Mecklinger, 2011). Under this account, if cognitive
control is targeted and improved throughout an intervention, then
performance should be affected only on transfer tasks with high
COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY
conflict resolution demands. Note that process-specificity differs

from domain-specificity in that the former predicts that training
should transfer to tasks with similar cognitive demands despite
divergences in surface-level properties like task goals or stimulus
characteristics.
In this way, current psycholinguistic theorizing and the neurocognitive literature on recognition memory together inform our
approach to cognitive control training: we use tightly designed and
well-established tasks that permit clear tests of process-specific
predictions related to conflict resolution. In our assessmentoutcome battery, we compare various tasks and conditions with
high conflict resolution demands to those without. We adopt a
process-specific but domain-general framework, and thus predict
that cognitive control training will improve performance only
under conditions when conflict resolution demands exist both
within and outside of the training domain. In what follows, we
briefly review research demonstrating a role for cognitive control
during recognition memory and language processing, outlining
evidence for shared psychological procedures and neurobiological
underpinnings that suggest domain-general cognitive control
mechanisms. We then report our training study, which examines
generalized performance across domains within a process-specific
conflict resolution account, to determine whether these domaingeneral mechanisms are pliable in a way that affects task performance beyond the training context.
When Conflict Arises

On occasion, people receive competing evidence about how best
to characterize a stimulus, owing to unusual instructions, too few
constraints to specify a unique solution, or continuously evolving
input that ultimately mismatches early predictions. Top-down biasing procedures must then signal adjustments in behavior to
distinguish task-relevant from -irrelevant information, filtering
attention for whats important (Botvinick et al., 2001; Milham et
al., 2001; Nelson et al., 2003; Thompson-Schill & Botvinick,
2006; Shimamura, 2000). Both recognition memory and language
processing tasks can create situations that induce two types of
conflict requiring the engagement of cognitive control procedures:
(a) when the input gives rise to a dominant or prepotent representation that people must ultimately override, or (b) when a stimulus
does not induce a dominant response itself but rather elicits multiple underdetermined response candidates that reach equivalent
levels of activation (Botvinick et al., 2001).
During recognition memory, for example, prepotent conflict
arises when subjects encounter recent material that is not part of
the current memory set, thus creating interference from highly
familiar but currently irrelevant memoranda. Such familiarity may
lure item-recognition processes away from the most pertinent
information, that is, the relevant memory set. In such cases, subjects must employ cognitive control to override a prepotent bias to
respond yes to recently seen and therefore extremely recognizable stimuli (Nelson, Reuter-Lorenz, Persson, Sylvester, &
Jonides, 2009; for a review, see Jonides & Nee, 2006). Similarly,
in the domain of sentence processing, readers and listeners consult
multiple linguistic and extralinguistic sources of evidence (e.g.,
lexico-syntactic cues and referential context) that inform interpretation commitments moment-by-moment (MacDonald, Pearlmutter, & Seidenberg, 1994; Tanenhaus, Spivey-Knowlton, Eberhard,
& Sedivy, 1995; Trueswell et al., 1999). Usually, these informational sources conspire to guide readers and listeners toward a
correct interpretation; sometimes, however, there are various, incompatible cues to sentence meaning, resulting in temporary misanalysis and the need to revise (e.g., as in the case of garden-path
recovery; Novick et al., 2005; see also January et al., 2009;
Rabagliati, Pylkknen, & Marcus, 2013; del Ro et al., 2011).
When this occurs, cognitive control may act to override initial
misinterpretations (Hsu & Novick, 2016; January et al., 2009;
Novick et al., 2005; Ye & Zhou, 2009). Finally, the retrieval of
real-world knowledge during language production can also induce
conflict, for example when a speaker must utter a single word from
multiple underdetermined candidates that compete for selection
(e.g., verb generation; Barch, Braver, Sabb, & Noll, 2000; Petersen, Fox, Posner, Mintun, & Raichle, 1988; Persson, Welsh,
Jonides, & Reuter-Lorenz, 2007; Snyder et al., 2010; ThompsonSchill, DEsposito, Aguirre, & Farah, 1997). In cases when a word
has more (vs. fewer) lexical competitors, top-down cognitive control processes may support efforts to select one from several
competing alternatives (Kan & Thompson-Schill, 2004; Nelson et
al., 2009; Thompson-Schill & Botvinick, 2006).
The resolution of such information-processing conflict, broadly
construed, has been associated with a common cognitive control
function across a variety of recognition memory and language
processing tasks (January et al., 2009; Jonides et al., 1998; Kan &
Thompson-Schill, 2004; Nelson et al., 2009; Novick et al., 2005,
2009; Snyder, Banich, & Munakata, 2011; Ye & Zhou, 2009).
Brain-imaging, brain stimulation, and neuropsychological studies
demonstrate consistently that neuroanatomical regions within ventrolateral prefrontal cortex (VLPFC) provide modulatory signals
when conflict arises, and facilitate resolution across different task
environments (Bilenko, Grindrod, Myers, & Blumstein, 2009;
Botvinick et al., 2001; Fedorenko, 2014; Fletcher & Henson, 2001;
Hussey, Ward, Christianson, & Kramer, 2015; Jonides & Nee,
2006; Nelson et al., 2009; Novick et al., 2005, 2009, 2010; Nozari
& Thompson-Schill, 2013; Thompson-Schill, Bedny, & Goldberg,
2005; Vuong & Martin, 2011, 2014; but see Snyder, Banich, &
Munakata, 2014 for evidence that underdetermined and prepotent
response conflict rely on overlapping but partially independent
neural underpinnings). For instance, the same regions within
VLPFC are recruited during various recognition-memory tasks
that require subjects to override a familiarity bias, including the
recent-probes task (DEsposito, Postle, Jonides, & Smith, 1999;
Jonides & Nee, 2006; Jonides, Smith, Marshuetz, Koeppe, &
Reuter-Lorenz, 1998; Nelson et al., 2009; Thompson-Schill et al.,
2002), the local recognition-memory task (Kane, Conway, Miura,
& Colflesh, 2007; Oberauer, 2005; Oberauer & Lange, 2009), and
the n-back task with lures (Chatham et al., 2011; Gray, Chabris, &
Braver, 2003; Owen, McMillan, Laird, & Bullmore, 2005). In a
similar vein, transcranial DC stimulation (tDCS) of LPFCmediated cognitive control regions increases n-back discriminability and recovery from misinterpretation during a self-paced reading paradigm (Hussey et al., 2015). Acute up-regulation of these
ventrolateral PFC regions also reduces language production costs
when generating a single word embedded in a sequence of highly
confusable items that compete for selection (Nozari & ThompsonSchill, 2013). Patients with damage to VLPFC, analogously, demonstrate selective recognition memory deficits under conditions of
high-conflict cognitive control demands (Hamilton & Martin,
HUSSEY ET AL.
2005; Novick, Kan, Trueswell, & Thompson-Schill, 2009;

Thompson-Schill et al., 1998, 2002) that are linked to selective
language processing impairments, including failure to generate a
word from many versus few competitors (Robinson et al., 1998;
Robinson, Shallice, & Cipolotti, 2005; Schnur et al., 2009;
Thompson-Schill et al., 1998) and, during comprehension, failure
to recover from early misanalysis of sentence meaning (Novick et
al., 2009; Novick, Trueswell, & Thompson-Schill, 2010; Vuong &
Martin, 2011, 2014). Finally, behavioral tests of individual differences show that linguistic ambiguity resolution is related to the
ability to detect and resolve information-conflict in other domains,
including recognition memory (Khanna & Boland, 2010; Nilsen &
Graham, 2009; Novick et al., 2014; Teubner-Rhodes et al., 2016).
These correlational findings, in both brain and behavior, suggest a
shared cognitive control system that supports the ability to resolve
information-conflict within memory, revise default interpretations
in language comprehension, and produce a single word from
multiple competing alternatives. But are such general-purpose
cognitive control processes causally linked to memory and language performance under these conditions?
Preliminaries to the Current Study

Here, we examine a range of recognition memory and language
processing conditions that create information-processing conflict.
In particular, we test the extent to which focused practice engaging
cognitive control during a short-term training regimen fine-tunes
processing across domains, reflected in performance improvements specifically under high conflict resolution demands. Importantly, we compare the effects of training on well-studied assessment tasks that include high- and low-conflict conditions, in
addition to assessments that theoretically lack the need for cognitive control. These two design components allow us to compare
across tasks and conditions to address the question of whether
cognitive control mechanisms per se can be sharpened through
process-specific training. Concretely, we ask: do performance improvements during the intervention generalize to novel language
and memory tasks under conditions of information-conflict, but
not to tasks or conditions that do not require cognitive control?
This work is motivated, in part, by some evidence for processspecific effects of cognitive control plasticity: in one study, cognitive control training led to performance benefits on an unpracticed sentence processing task involving garden-path recovery
(Novick et al., 2014). Although subjects in that study completed a
battery of executive function tasks during training, the findings
indicated that performance improvements on only the cognitive
control training task (n-back-with-lures; see also Oelhafen et al.,
2013) significantly predicted readers posttest ability to revise
early misinterpretations in real time following the detection of
syntactic conflict (i.e., discovery of a misinterpretation). In contrast, performance increases on the low-conflict training tasks did
not generalize to garden-path recovery in any way. However, the
training groups transfer effectsthough specific to only the highconflict conditions of the untrained reading task and localized to
the conflict regions of the ambiguous sentences themselveswere
compared with a no-contact control group, which is a suboptimal
contrast (see Boot, Simons, Stothart, & Stutts, 2013; Shipstead,
Redick, & Engle, 2010 for a critique of this practice). Although
these findings are highly suggestive, it remains possible that the
other (low-conflict) training tasks were a necessary component of

a combined suite to confer transfer, and that practice dealing with
conflict per se was not the active ingredient.
In the present study, we are interested in the plasticity of
cognitive control procedures across domains and thus aim to
replicate and extend these earlier findings in crucial ways. First,
we include other transfer tasks in addition to garden-path recovery.
Importantly, some of these tasks have been previously established
to rely on cognitive control to resolve information-conflict under
some conditions but not others, while other tasks are known to be
difficult for reasons unrelated to conflict demands (e.g., understanding sentences containing object relative clauses puts pressure
on working memory resources; see Method). By including multiple measures of the same construct and within-task manipulations
of cognitive control, we can test the extent of transfer across tasks,
as well as the selectivity of these effects to tasks and conditions
that share conflict-processing demands (see Shipstead et al., 2010).
Second, we employ active control groups that undergo a training
regimen involving the same recognition memory task (n-back), but
with a critical feature minimally changedthe presence or absence of lures. Isolating conflict resolution demands by comparing
n-back-with-lures training (High-Conflict) to an otherwise identical regimen without lure items (Low-Conflict) affords the opportunity to understand how practice dealing with conflict in recognition memory contributes to training-related benefits in cognitive
control skills per se (cf. Gray et al., 2003; Kane et al., 2007;
Novick et al., 2014; Oelhafen et al., 2013; Persson et al., 2007;
Szmalec, Verbruggen, Vandierendonck, & Kemps, 2011).
In addition to these two training groups, which will provide a
minimal comparison that tests for domain-general plasticity of
conflict-control procedures, we included a third condition. Some
work suggests that adaptive training tasksthose that adjust task
difficulty based on an individuals real-time performancemay
give rise to larger transfer effects (Brehmer et al., 2011; Holmes,
Gathercole, & Dunning, 2009; Karbach, Strobach, & Schubert,
2015; Klingberg, 2010; Klingberg et al., 2005; Klingberg, Forssberg, & Westerberg, 2002; Leek, 2001; Moreau & Conway, 2014;
but see Shipstead, Redick, & Engle, 2012). This is thought to be
important for making a training task consistently challenging by
keeping subjects at the threshold of their best performance
(Lvden, Bckman, Lindenberger, Schaefer, & Schmiedek, 2010;
but see von Bastian & Eschen, 2016). Thus, we included a group
that practiced a static 3-back task without lure items (3-Back)
against which the Low-Conflict group was compared to test
whether adaptivity provides singular benefits during training.
In sum, our training approach hinges on a key process-specific
assumption: that targeting a specific mental process (conflict detection and resolution) should improve performance on various
tasks that tap that process, but not on other carefully matched tasks
or conditions without such demands even in the face of elevated
task complexity (Dahlin et al., 2008; Hussey & Novick, 2012;
Hussey et al., 2015; Oelhafen et al., 2013; Schneiders et al., 2011).
To this end, we compare training groups that differ in critical
waysthose who practice conflict processing versus those who do
notand predict selective transfer effects to untrained conflict
resolution conditions on a battery of memory and language tasks
for the High-Conflict group compared with the Low-Conflict
group. Specifically, we expect that people receiving High-Conflict
training will demonstrate posttraining improvements on the fol-
lowing task conditions: lure trials during n-back; lure trials on the
local block of a global-local recognition memory task; highassociation/high-competition trials during a verb generation task;
incongruent trials on a Stroop task; and syntactically ambiguous
sentences during real-time language processing (see Method).
These conditions all require resolving information-processing conflict that arises either from proactive interference in recognition
memory (n-back, global-local recognition memory), activation of
multiple, equally valid responses (verb generation), activation of a
prepotent but incorrect response (Stroop), or temporary misinterpretation from syntactic ambiguity (sentence processing). Note
that our task battery probes for different levels of transfer: n-back
tests for task-specific transfer, because it also served as the training
task; the global-local recognition memory task tests for domainspecific transfer, because the stimuli and conflict environment
(lures during recognition memory) are similar between the local
block of this task and the high-conflict n-back training task; and
the remainder of the assessment battery tasks test for domaingeneral transfer, because task goals, stimuli characteristics, and the
environment in which conflict is experienced all differed from the
n-back training task. Importantly, we do not expect to observe
benefits of High-Conflict training on nonconflict control conditions, nor do we expect selective cognitive control gains in the
Low-Conflict treatment group. Of course, the Low-Conflict group
may still benefit from the intervention in some way; however, we
would not expect their performance gains to be strictly localized to
the tasks and conditions that involve conflict resolution demands.
Finally, we also test a widespread belief about cognitive training,
namely regimens that are performance-adaptive yield transferrable
benefits, and so adaptivity by itself produces performance boosts;
thus, task difficulty should adjust to individual trainees performance levels to continually challenge them. We therefore compare
the Low-Conflict (adaptive) trainees to a nonadaptive 3-Back
group, to test whether any transfer effects are conferred by training
adaptivity alone (in the absence of conflict).
Method
one reported taking medications to correct problems related to

neuropsychological or neuropsychiatric impairment. All subjects
had normal or corrected-to-normal vision and normal color vision.
Design
We employed a double-blind pretest/posttest protocol: different
experimenters conducted training and assessment sessions in separate labs without awareness of subjects condition assignments.
All participants visited the training lab for a total of eight hours,
split into 16 30-min sessions in the three-to-six weeks (M ! 4.8
weeks) intervening pretest and posttest visits. To combat attrition
and to promote engagement, each subject was informed of an
incentive program at the halfway point during training (following
the eighth training session) through an email notification that
graphically depicted their individual training performance with
personal high scores clearly marked with a star. The personal high
score was the highest n-back score achieved in a session relative to
all previous sessions for a given subject, calculated as average
accuracy multiplied by average n-level of a session (for the 3-Back
Group, who encountered just one n-level throughout training, this
score was simply a measure of average accuracy; see below for
details). Participants were told that for every new personal best
score, their names would be entered into a lottery to earn a prize
worth up to (an additional) $200.
During pre- and posttest sessions, subjects completed one of two
complementary versions of a recognition memory task, a verb
generation task, a Stroop task, and two sentence reading tasks
while their eye movements were recorded. Each assessment battery was completed in one 2-hr session, with task order counterbalanced and pseudorandomized across assessments. In addition,
at the end of the posttest session, all subjects completed a version
of the n-back task that included blocks of 3- and 6-back trials with
lures. This task was included to examine how well each group
performed relative to the other groups on task conditions experienced during training (e.g., lures for the High-Conflict group;
higher n levels for the adaptive groups).
Subjects
Training Groups
One hundred sixteen healthy native-English-speaking subjects

were recruited from the University of Maryland community to
participate in this experiment. All provided informed consent and
were compensated a total of $200 for 12 complete hours of
participation across 18 lab visits. Each subject was randomly
assigned to one of three training groups: High-Conflict, LowConflict, and 3-Back (see Training Groups below for details).
Thirty-five subjects were excluded from analyses (High-Conflict:
n ! 6; Low-Conflict: n ! 19; 3-Back: n ! 10) for either failing to
complete all study phases (n ! 19) or for allowing at least two
weeks to lapse between any two consecutive lab sessions (n ! 16).
The final participant group comprised 81 people (High-Conflict:
N ! 30, 22 women, Mage ! 19.8 years, Medu ! 14.53 years;
Low-Conflict: N ! 23, 17 women, Mage ! 20.0 years, Medu !
14.09 years; 3-Back: N ! 28, 19 women, Mage ! 20.0 years,
Medu ! 14.36 years). There were no demographic differences
among these groups in terms of average age (p ! .94), education
level (p ! .44), or sex (p ! .87). None of the subjects had a history
of neurological disorders, stroke, or learning disabilities, and no
Subjects were randomly assigned to practice one of three versions of an n-back task during the weeks intervening pretest and
posttest: (a) performance-adaptive n-back with lures (HighConflict), (b) performance-adaptive n-back without lures (LowConflict), and (c) a static 3-back task without lures (3-Back). Lure
presence was manipulated to isolate conflict resolution demands,
and adaptivity was manipulated to test for the singular role of
performance-contingent designs on training outcomes. The LowConflict group therefore served as an internally valid active control
group (Shipstead et al., 2010, 2012b) to test for process-specific
effects of conflict resolution training (High-Conflict vs. LowConflict), and the static 3-back task served as an active control
condition that permitted us to test the importance of performance
adaptivity (Low-Conflict vs. 3-Back). Together, these control conditions allowed us to isolate the extent to which increased conflict
demands and performance adaptivity influence transfer to untrained measures.
High-conflict N-back training. In this task, subjects were
asked to identify (recognize) when a stimulus item appeared n
HUSSEY ET AL.
lures. Fillers were items that did not repeat in the n position or any
highly confusable position. Lures were defined as items that repeated in positions n " 1, n " 2, n # 1, and n # 2 (Gray et al.,
2003; Kane et al., 2007; Novick et al., 2014). For example, in a
4-back condition, the second appearance of K in the sequence k, d,
N, K is considered a lure because it matches the identity of an item
presented 3 instead of 4 trials prior (see top panel of Figure 1).
Thus, lures create information conflict: subjects must override
(inhibit) a familiarity bias to respond Target to any recently
presented item. Importantly, regardless of n-level, all sequences
had the same number of eligible target responses20 a design
trials previously. Single letters were displayed serially for 500ms

following a 500ms fixation cross, with an interstimulus interval of
2 seconds. All letters were drawn from a subset of consonants (b,
c, d, f, h, j, k, l, m, p, q, r, s, t, v, or x) and were displayed in mixed
upper- and lower-case so the task was not purely a visual-matching
exercise. Subjects indicated by button press whether the current
letter, regardless of case, had appeared n items previously by
pressing one of two keys corresponding to Target or NonTarget. All sequences contained 20 " n items, partitioned into 6
targets, 6 lures, and 8 " n fillers. Targets were items that repeated
in the appropriate n-back position. Nontargets were either fillers or
Figure 1. Example sequences from the three variants of n-back training. High-Conflict trainees (upper panel)
practiced sequences containing lures: items that repeated in non-n positions (e.g., the second instance of K is a
lure because it appeared 3 back in a 4-back task). The Low-Conflict group (middle panel) did not encounter lures.
Both High- and Low-Conflict tasks adapted in difficulty: n-level varied from 113 depending on an individuals
performance on the previous sequence. The 3-Back task (bottom panel) was not adaptive, and had five rotating
stimulus sets across sessions (items from symbol set #1 are depicted here; see text).
feature aimed at attenuating response bias inconsistencies across

sequences with variable n-levels. Performance-adaptivity was controlled by adjusting n according to a subjects performance on his
or her previous sequence: if fewer than three errors occurred,
difficulty was increased by one n-level on the next sequence up to
n ! 13; if more than five errors occurred, difficulty was decreased
by one n-level down to n ! 1 (see Jaeggi, Buschkuehl, Jonides, &
Shah, 2011); if 3 to 5 errors occurred, then the current n-level
repeated on the next sequence. All participants began training with
a 2-back task. Feedback in terms of accuracy and average response
time was provided after each sequence, followed by a notification
of the next sequences n-level. All subjects performed two 15-min
blocks of training during each session. Upon returning to the lab
for a new session, subjects started at the most recent n-level that
they had reached during the previous session, thus maintaining
adaptivity across training sessions.
Low-conflict N-back training. The Low-Conflict n-back task
was identical to the High-Conflict version except that lures were
removed from all sequences, resulting in arrangements of 6 targets
and 14"n fillers (see middle panel of Figure 1). Lures were
controlled for all items of a sequence ranging from n " 2 to the
current item. That is, no item repeated within this buffer aside from
targets, minimizing conflict in this task (subjects could largely use
a familiarity bias to complete each trial successfully). Note, however, that items could repeat within a sequence that were not
targets, but they could occur only in n " 3 or later positions,
thereby defining targets as the only repeating recent items in a
sequence.
3-back training. Subjects in this group monitored serially
displayed sequences of 23 items (6 targets and 17 fillers) and were
asked to indicate by button press whether the current item appeared 3 trials previously (see bottom panel of Figure 1). As with
the Low-Conflict condition, there were no lures, specifically
within 5 items of the current letter. Importantly, task difficulty was
not altered as a function of performance, which was the key
difference from the Low-Conflict group. In addition, the 3-Back
task included variable stimulus sets across training sessions to
minimize its repetitive nature and to keep participants engaged.
Stimulus sets included letters, highly imageable single-syllable
words, pronounceable single-syllable nonwords, and two sets of
easily identifiable symbols (taken from webdings; see Figure 1).
Sets were cycled across sessions in the same order for all participants; for example, all subjects saw letters at Session 1, words at
Session 2, one set of symbols at Session 3, nonwords at Session 4,
and a second set of symbols at Session 5 before repeating the same
sequence for Sessions 6 through 10 and 11 through 15. All subjects
finished their final (16th) session with a letter 3-back task.
Pre/Post Transfer Task Assessments and Predictions

Manipulation check: Posttest N-back-with-lures. Subjects
completed an n-back-with-lures task as the final task at their
posttest session. The task began with an 8-min block of 3-back
sequences, followed immediately by an 8-min block of 6-back
sequences, both with lures in positions n " 1, n " 2, n # 1, and
n # 2. Following each sequence, subjects were provided with
accuracy and average response time feedback. N-level was not
varied within a block regardless of performance. Subjects were
explicitly notified when the task transitioned from the 3-back to
the 6-back block. This task was administered to examine if HighConflict trainees outperformed the Low-Conflict trainees on lure
items regardless of n-level, and if the Low-Conflict trainees outperformed the 3-Back trainees generally on the 6-back block.
Global-local recognition memory task. At each pre/post assessment, subjects completed a recognition memory task identical
to the local and global recognition task used in Oberauer (2005,
Experiment 2). Two to five words were sequentially presented for
900 ms (100-ms interstimulus interval) in two to five rectangular
frames arranged horizontally across the screen. Participants were
then presented with recognition probes. Half the lists tested global
recognition, in which one word at a time was presented centrally
below the row of frames, and participants were asked to make a
yes/no judgment about whether the probe word had appeared in the
previous list. For 70 trials, the words were list words (targets); for
the other 70 trials, the words had not appeared on the previous list
(fillers). The local recognition task differed only in the location of
where the probe words appeared. Unlike the global task, local
probes appeared in the frames used during the learning phase and
participants were instructed to respond yes only if the word
appeared in the previous list and in the same frame as during
testing. Of the 140 probes, 70 had appeared in the previous list and
in the same frame (targets). Of the remaining probes, 35 were
words that had not appeared in the previous list (fillers) and 35 had
appeared but in a different frame (lures). Accuracy and response
times in milliseconds were recorded. We were therefore able to test
recognition memory under low conflict (global recognition) and
high conflict (local recognition) demands. Each subject completed
complementary versions of both global and local blocks (with
different words) at each assessment. Version assignment was random and counterbalanced across subjects and assessments.
Verb generation task. In the verb generation task, subjects
see a noun cue (e.g., ball) and must generate the first associated
verb that comes to mind. Longer response times are observed in
conditions where selection demands and memory retrieval demands are high (Botvinick et al., 2001; Martin & Cheng, 2006;
Thompson-Schill et al., 1997; Thompson-Schill & Botvinick,
2006; Wagner et al., 2001). However, as pointed out by Snyder
and colleagues (2008, 2010, 2011), previous verb generation experiments have confounded the effects of competition (i.e., selection demands) and association strength (i.e., memory retrieval
demands). This has resulted in mixed results that support two
alternative accounts of cognitive control. Under the selection account, cognitive control procedures (subserved by VLPFC) resolve
competition when multiple representations are automatically created by a stimulus (e.g., ball kick, catch, bounce, etc. vs.
scissors cut; e.g., Botvinick et al., 2001; Thompson-Schill et al.,
1997; Thompson-Schill & Botvinick, 2006). Under the controlled
retrieval account, cognitive control procedures (subserved by
VLPFC) permit responses from semantic memory to be retrieved,
particularly when it is harder to find one because of low noun-verb
association strength (e.g., valley hike vs. scissors cut; Martin
& Cheng, 2006; Wagner et al., 2001). Because association strength
and selection demands typically covary, it has been difficult to
know exactly when cognitive control should engage (e.g., when
retrieval demands are high and it is hard to find a response, or
when competition demands are high?).
Removing this confound through latent semantic analysis (LSA;
for details, see Snyder & Munakata, 2008; Snyder et al., 2010,
HUSSEY ET AL.
2011; 2014) permits tests of how association strength and competition factors interact:
High Association/Low Competition: scissors cut
High Association/High Competition: phone call, answer,
talk, and so forth
Low Association/Low Competition: leaf fall
Low Association/High Competition: folder open, close,
hold, and so forth
Here, as in previous studies, RTs are higher in low versus high
association conditions, and in high versus low competition conditions (Snyder et al., 2010, 2011). Crucially though, competition
effects are stronger under high association conditions (phone
call, answer, talk) as compared with low association conditions
(folder open, close, hold). This is computed by subtracting
performance on the low competition condition from the high
competition condition for each level of association. As Snyder and
colleagues put it,
When it is easy to retrieve a response, activating multiple competing
responses serves only to increase selection demands, slowing responding. However, when it is difficult to retrieve any response, spreading
activation between multiple weakly associated responses (e.g. between open and close when generating a response for folder) increases
the activation level of all responses, aiding retrieval and partially
offsetting selection costs. (Snyder et al., 2011, p. 3472)
Competition (or conflict) effects, indexed by RTs and VLPFC

activation, increase when retrieval demands are low (high association) compared with when they are high (low association). Thus,
the High Association/High Competition condition requires the
greatest recruitment of cognitive control without confounding the
difficulty encountered with heightened retrieval demands. We
therefore predict that High-Conflict trainees, but not the other
trainees, will show the greatest transfer effects under this verb
generation condition.
Materials and procedure. On each trial, a noun was presented
on a computer screen for up to 3400 ms. Subjects were instructed
to think of and produce the first verb that came to mind (Persson
et al., 2007; Snyder & Munakata, 2008). If no response occurred
during this period, the trial was tagged as a failed retrieval.
Subjects pressed the spacebar to record response times when they
thought of a verb, after which they verbalized their response into
a microphone for later accuracy scoring.
A total of 100 nouns that varied parametrically in terms of
competition and association strength were borrowed from Snyder
et al., 2011, resulting in the four pure conditions outlined above
(see also Snyder & Munakata, 2008 for LSA norming procedure).
Half the items (50 unique nouns) were randomly assigned to a
pretest set, and the other half (50 unique nouns) was assigned to a
posttest set for each subject. This resulted in a total of 13 High
Association/High Competition items, 13 Low Association/Low
Competition items, 12 High Association/Low Competition items,
and 12 Low Association/High Competition in one set and the
remaining items in the second set.
Stroop task. The Stroop task was included as a canonical
measure of cognitive control, where subjects indicate the ink color
of various words that could either be congruent color words (e.g.,
blue in blue font), incongruent color words (e.g., green in blue
font), or neutral noncolor words (e.g., deal in blue font; see
January et al., 2009; Milham et al., 2001). We computed Stroop
interference scores (Incongruent minus Neutral) to assess conflict

resolution and Stroop facilitation effects (Neutral minus Congruent) as a nonconflict baseline.
Procedure. On each trial, a fixation point appeared in the
center of the screen for 750ms, followed by a word written in blue,
yellow, or green font. Subjects were instructed to indicate the font
color by pressing one of three buttons mapped onto responses
corresponding to blue, yellow, or green. Subjects completed two
different blocks of 144 trials each with a 50:25:25 ratio of congruent, incongruent, and neutral trials (see Kane & Engle, 2003).
On congruent trials, the words were color terms that matched their
font color. Incongruent trials were color terms that did not match
their font color. Finally, neutral trials presented noncolor words
that were length and frequency matched to the color words (January et al., 2009; Milham et al., 2001). Participants performed 12
practice trials at the onset of each block prior to starting the actual
trials. Block order was randomized and counterbalanced across
participants. Response time and accuracy were recorded for analysis.
Sentence processing tasks. Subjects read a total of 144 sentences each at pretest and posttest while their eye-movements were
tracked. The task included 24 sentences with an ambiguous versus
unambiguous manipulation to probe conflict resolution procedures
during sentence processing. There were also 24 sentences comparing object versus subject relative clauses to assess processing
difficulty in the absence of conflict. This second sentence processing assessment verified that any effects of High-Conflict training
cannot be ascribed simply to a difficulty account, because highconflict conditions are necessarily harder than low-conflict ones.
At each assessment, these critical items were embedded in 96 filler
sentences, which contained a range of structures to conceal the
syntactic manipulations (see also Novick et al., 2014). After reading each sentence, participants answered a comprehension question aimed at verifying that they processed the sentence meaning.
Yes/no responses were fully balanced in each set.
Syntactically ambiguous versus unambiguous sentences.
Subjects read the same temporarily ambiguous [1] and unambiguous [2] sentences that were used in the training study by Novick
et al. (2014):
1.
While the thief hid the jewelry that was elegant and
expensive sparkled brightly. (Temporarily Ambiguous/
High-Conflict)
2.
The jewelry that was elegant and expensive sparkled

brightly while the thief hid. (Unambiguous/LowConflict)
Sentence [1] is temporarily ambiguous because the verb hid can

be used either transitively (the thief could be hiding something) or
reflexively (the thief could be hiding him/herself). However, the
transitive analysis is strongly supported early onand thus readers
rapidly commit to this interpretation because there is no comma
following hid, which would force the reflexive interpretation, and
the postverbal noun phrase the jewelry is a semantically plausible
direct object for hid particularly in a thieving context (Christianson,
Williams, Zacks, & Ferreira, 2006; Ferreira, Christianson, & Hollingworth, 2001; Novick et al., 2014). However, late-arriving disambiguating evidence (sparkled brightly) renders the preferred object
analysis invalid; readers must therefore revise and capture the alternative reflexive interpretation when they discover a misinterpretation
(syntactic conflict). Such garden-path recovery theoretically involves
the deployment of cognitive control processes (Hsu & Novick, 2016;
January et al., 2009; Novick et al., 2005, 2009, 2014; Ye & Zhou,
2009). In contrast, the reversed clause order in unambiguous cases
like [2] removes the syntactic conflict and the need to revise.
We developed different versions of critical sentences by placing
24 ambiguous verbs into either syntactically ambiguous (highconflict) or unambiguous (low-conflict) sentence frames. We created two counterbalanced lists with 12 sentences of each critical
item type: an ambiguous item in one list became an unambiguous
item in the second list. Comprehension questions probed for the
correct reflexive interpretation (e.g., Did the thief hide himself?); thus, a correct yes response would index successful
garden-path recovery (Christianson et al., 2006; Novick et al.,
2014), and these items all required yes responses.
Better recovery from temporary misanalysis (i.e., conflict resolution) is indexed by higher comprehension accuracy and shorter
reading times on measures of moment-by-moment revision
(second-pass and go-past times; see below) on ambiguous but
not unambiguous sentences at posttest.
Object versus subject relative clauses. Subjects also read sentences like the following, which contained object- and subjectextracted relative clauses (12 of each type):
3.
The farmer who the expert questioned promoted the

product at the fair. (Object- Extracted; HigherDifficulty)
4.
The farmer who questioned the expert promoted the

product at the fair. (Subject- Extracted; LowerDifficulty)
Object relative clauses (italicized in 3) are known to generate

processing difficulty, reflected in increased reading times versus subject relatives (italicized in 4), as a result of working memory demands
associated with storage costs (keeping track of incomplete syntactic
dependencies over longer distances) and integration costs (connecting
new words to the structure built so far; Gibson, 1998). Concretely,
object relative clauses (who the expert questioned) are more difficultand are thus read more slowly than subject relative clauses
(who questioned the expert) because readers must hold the relative clause in mind longer before the verb (promoted) arrives to link
it with the subject noun phrase (Fedorenko, Woodbury, & Gibson,
2013; Gordon & Lowder, 2012; Lewis & Vasishth, 2005; Van Dyke
& Lewis, 2003). Though more difficult, comprehending object relative clauses does not appear to involve conflict-control procedures
(Farmer, Misyak, & Christiansen, 2012). Perhaps the most compelling evidence for this comes from neuropsychological patients with
cognitive control deficits: there is no association between damage to
VLPFC and comprehension of syntactically complex sentences like
object relative clauses (Thothathiri, Kimberg, & Schwartz, 2012); yet,
patients with this lesion profile reliably fail to revise incorrect interpretations of lexical and syntactic ambiguities like those described
earlier (Novick et al., 2009, 2010; Vuong & Martin, 2011). Similarly,
focal stimulation of left lateral prefrontal regions via tDCS affects
interpretations of garden-path sentences, but not real-time processing
of object relative clauses (Hussey et al., 2015).
This (single) dissociation then offers an ideal test bed to compare

the effects of conflict resolution training on a hard sentence processing task involving conflict resolution (garden-path recovery), versus
one that is hard but does not theoretically involve conflict resolution
(object relative clauses). We reasoned that the cognitive control skills
improved during n-back-with-lures training should not influence reading times under all states of complex sentence processing, but rather
only when one interpretation must be revised in favor of another,
requiring conflict resolution. Thus, the relative clause items were
included to (a) test for the specificity of conflict training effects to
only the tasks that share conflict-processing demands, and (b) ensure
that transfer effects from the High-Conflict training group are not
simply attributable to people improving on hard tasks per se, but
rather only on tasks that involve conflict.
Forty-eight relative clause sentences were borrowed from Fedorenko, Gibson, and Rohde (2006) to create 12 object-extracted and
12 subject-extracted relative clauses for each assessment like those in
examples (3) and (4) (no items repeated across assessments). We
created two counterbalanced lists of relative clause sentences: an
object-extracted item in one list became a subject-extracted item in its
counterpart list. These items were also pseudorandomized and counterbalanced across participants and assessments preventing the same
verbs and nouns (e.g., farmers, experts) from appearing within or
across assessments. Comprehension questions verified that participants were processing the sentence for its meaning (e.g., Was the
product promoted on TV?).
Eye-tracking apparatus. To examine the effects of training on
real-time sentence processing, we recorded subjects eye-movements
using an EyeLink 1000 eye-tracker (SR Research), with vertical and
horizontal eye position sampled every millisecond. Stimuli were
presented via EyeTrack Software Version 0.7.10 (www.psych.umass
.edu/eyelab/software). Participants were situated in the Eyelinks forehead and chin rests. Viewing was binocular but the system was set to
monocular recording. The eye-tracker was calibrated to an average
spatial-resolution error of 0.50 or less (with no single point at an error
greater than 1.00) and recalibrated as needed.
Each trial began with a fixation box in the position of where the
leftmost character of the sentence would appear. Once a subject
fixated this box, the sentence appeared automatically, replacing the
box; this procedure served as a trial-by-trial calibration check.
Each sentence was presented in its entirety on a single line.
Subjects were instructed to read each sentence silently at a comfortable pace and press a button when finished to advance to the
comprehension question, to which they responded yes or no via
button press. Before the experiment, subjects completed 10 practice trials to ensure that they understood the procedure. Total task
time averaged 40 min (range ! 20 to 60 min), including recalibration and a scheduled break.
Results
Training Task Performance
As can be seen in Figure 2A, subjects in each training group
improved on their respective n-back training task in terms of
overall accuracy, calculated by last minus first training-session
performance (3-Back: d ! 1.91; Low-Conflict: d ! 1.89; HighConflict: d ! 0.97). Performance was indexed by normalized
n-back accuracy. Specifically, all session values were z scored
10
HUSSEY ET AL.
mixed-effects model evaluating the fixed effects of Training

Group (High-Conflict vs. Low-Conflict) and Session (116) on
standardized n-back score: we observed only a reliable effect of
Session (b ! 0.154, SE ! 0.004, t ! 34.13, p $ .001). The effect
of Training Group and the Group-by-Session interaction did not
reach significance (ps % 0.50). Importantly, this suggests that the
High- and Low-Conflict training groups improved throughout the
intervention to the same degree. Thus, any transfer effects reported
below for one group over another cannot be simply ascribed to
variation in performance and/or greater difficulty of a regimen.
Statistical Analyses on Assessment Measures

We tested the effect of training on transfer-task performance
with multilevel linear regression models using the lme4.0 package
(version 1.17) in R (R Development Core Team, 2015). For all
pre/post assessments, we tested for transfer effects by modeling the
fixed effects of Training Group and Assessment (Pretest, Posttest).
We analyzed the impact of conflict by comparing the High- and
Low-Conflict training conditions, and we analyzed the impact of
adaptivity by comparing the Low-Conflict and 3-Back training
conditions. In all of our models, we report the maximal random
effects structure that converged, including random intercepts for
Subjects and random slopes for Subject-level fixed factors (see
Barr, Levy, Scheepers, & Tily, 2013). Additionally, for the sentence processing measures (comprehension accuracy and eye
movement indices), we also included a random intercept term for
Items and nested random slopes of appropriate Item-level fixed
factors (see Jaeger, 2008). Generalized linear mixed-models were
used for binomial factors, like comprehension accuracy. The significance of model estimates was determined using a KenwardRoger approximation, which models degrees of freedom and the
t-distribution to produce more conservative p values (Mirman,
2014). Group- and condition-level means can be found in Tables
A1 through A3 of the Appendix and the full output of each model
can be found in the Appendix in Tables A4 through A10.
Figure 2. Training performance curves for each group over 16 training
sessions. (A) Mean normalized overall n-back accuracy for each group,
corrected for training Session 1. (B) Mean normalized n-back score (average accuracy multiplied by average n-level) for each adaptive training
group, corrected for training Session 1. Error bars represent & 1 standard
error of the mean.
separately for each training group; then, Session 1 performance

was subtracted from all subsequent sessions (216) on a subjectby-subject basis. Figure 2A illustrates that all training groups
improved performance in overall n-back accuracy, with the highest
mean standardized gains from the first to the final training session
for the 3-Back training group (1.99), followed by the Low- (1.67),
and High-Conflict groups (1.15). Because the adaptive training
groups experienced variable n-levels, we also examined a measure
of n-back performance that captured this important task demand on
each session (n-back score, or average accuracy multiplied by
average n-level on a session). As can be seen in Figure 2B, there
were comparable training gains for both adaptive groups over the
course of the regimens (Low-Conflict: d ! 4.63, first-to-final
session-gain of 3.03; High-Conflict: d ! 3.05, first-to-final
session-gain of 2.50). We confirmed this observation with a linear
Model Interpretation
We implemented dummy contrast coding for the Training
Group factor with the Low-Conflict group as the reference; that is,
the Low-Conflict group served as a baseline against which the
other two groups were compared. This approach was used to
evaluate the minimal contribution of lures and adaptivity during
training on transfer to novel measures (see rationale above, in the
Introduction). For example, the High- and Low-Conflict groups
were contrasted to evaluate the effect of conflict training, and the
3-Back and Low-Conflict groups were compared to assess the
effect of performance adaptivity. Therefore, any significant results
of Conflict Training in our models suggest an effect of n-backwith-lures (conflict) training, while significant Adaptive Training
findings point to an effect of performance-adaptivity (see Linck et
al., 2012). The Assessment factor was always orthogonally coded
to allow us to examine the mean difference across assessments.
We also implemented models of pretest performance for all
pre/post measures by including Training Group and any taskspecific factors as fixed effects. The purpose of this analysis was
twofold: First, it served as a verification that each task-specific
factor was working properly (i.e., we replicated the expected
11
conflict effects) before evaluating any pre/post changes attributable to training. Second, these models allowed us to assess whether
any baseline training group differences were present before the
training interventions.
Results of Posttest N-Back Task: Manipulation Check

Two subjects data were excluded because of poor ($80%)
accuracy on noncritical filler items of both blocks (High-Conflict:
n ! 1; Low-Conflict: n ! 1). Figure 3A shows that the LowConflict (adaptive) trainees were more accurate than the 3-back
(nonadaptive) trainees on the 6- but not the 3-back block. To test
this, we modeled overall accuracy as a function of Adaptive
Training (Low-Conflict vs. 3-Back) and N-Level (3- vs. 6-back).
The model revealed main effects of N-Level, t ! #8.46, p $ .001,
and Adaptive Training, t ! 3.48, p $ .001, and an interaction of
Adaptive Training and N-Level, t ! 3.17, p ! .002. Specifically,
Low-Conflict trainees, who received adaptive training, were reliably more accurate than 3-Back trainees on the 6-back block
(MLC ! 0.88 vs. M3B ! 0.80), but not the 3-back block (MLC !
0.92 vs. M3B ! 0.91), as expected.
Next, we tested whether practice with lure items during training
led to overall better lure and target performance following training
by modeling lure and target accuracy as a function of Conflict
Training (High-Conflict vs. Low-Conflict) and N-Level (3- vs.
6-back). The model of lure accuracy revealed only a main effect of
N-Level, t ! #2.34, p ! .02. As can be seen in Figure 3B,
High-Conflict trainees were numerically more accurate than LowConflict trainees on lure trials for both the 3- (MHC ! 0.95 vs.
MLC ! 0.92) and 6-back blocks (MHC ! 0.90 vs. MLC ! 0.88), but
this observation did not reach significance. The model of target
accuracy, however, revealed a main effect of N-Level, t ! #8.39,
p $ .001 and an interaction of N-Level and Conflict Training, t !
2.36, p ! .02. As can be seen in Figure 3C, Low-Conflict trainees
were more accurate than High-Conflict trainees on target trials on
the 6-back block (MHC ! 0.61 vs. MLC ! 0.73), suggesting that
Low-Conflict trainees were better able to identify targets, which
may have positively influenced these trainees ability to identify
nontargets (including lures), minimizing any effects of Conflict
Training on lure accuracy in this specific task (see Discussion).
Indeed, we modeled signal detection indices of discriminability
and response criterion as a function of Conflict Training and
N-Level. The model of discriminability revealed an interaction of
Conflict Training and N-Level, t ! 1.94, p ! .05, as well as a main
effect of N-Level, t ! #7.15, p $ .001. The Low-Conflict trainees
were better able to discriminate targets from 6-back nontargets
(D ! 2.33) relative to the High-Conflict group (D ! 2.07). In the
model of criterion, we also found an interaction of Conflict Training and N-Level, t ! #2.52, p ! .02 supported by main effects of
Conflict Training, t ! #2.29, p ! .03, and N-Level, t ! 8.33, p $
.001. High-Conflict trainees were more conservative (C ! 0.54)
than Low-Conflict trainees (C ! 0.44). The effect was also exaggerated on 6-back trials (CHC ! 0.72; CLC ! 0.53). This pattern
suggests that the High-Conflict trainees higher accuracy on lure
items compared to the Low-Conflict trainees was due to a more
conservative response pattern. The question is whether taskspecific performance generalizes beyond the training environment.
Figure 3. Posttest n-back task accuracy. (A) Adaptivity manipulation

check by comparing the Low-Conflict and 3-Back groups accuracy on
overall n-back accuracy. (B) Conflict manipulation by comparing the Lowand High-Conflict groups lure accuracy. (C) Conflict manipulation by
comparing the Low- and High-Conflict groups target accuracy. Error bars
represent & 1 standard error of the mean. ! Significant at the p $ .05 level.
12
HUSSEY ET AL.
Results of Transfer Assessment Measures

To preview our results, we observed significant effects of Conflict Training selectively in conflict conditions on three of our four
transfer tasks where Conflict Training was hypothesized to improve performance (and we did not observe transfer effects to a
difficult sentence task without conflict, as expected). This pattern
is summarized in Table 1, which shows that the measures for
which the interaction of Assessment and Conflict Training was
reliable almost exclusively reside among the predicted highconflict conditions of each assessment task (bolded). In what
follows, we report the results for each transfer task in detail.
Global-local recognition memory task. Data from two subjects were excluded from all analyses: One subject in the HighConflict training group had poor filler (nontarget; nonlure) accuracy
($80%); one subject in the Low-Conflict training group had missing
posttest data (Low-Conflict: n ! 1). Following Oberauer (2005), we
focused exclusively on response times to measure recognition performance because prior results have shown that cognitive control differences manifest in response times rather than accuracy.
Pretraining effects. Baseline performance was evaluated in a
model of median response times at pretest that included fixed effects
of Training Group (Conflict Training, Adaptive Training), Block
Type (Local, Global), and Probe Type (Target, Filler, Lure). We
replicated prior findings by observing significant effects of Block
Type (b ! 139.63, SE ! 25.03, t ! 5.58, p $ .001) and Probe Type
(ts % 88.18, ps $ 0.003). Response times to probes in the local block
(M ! 840 ms) were slower than those in the global block (M ! 628
ms), and subjects were slower to respond to lures (M ! 1011 ms)
compared with targets (M ! 705 ms) and fillers (M ! 678 ms).
Importantly, the baseline model did not reveal any effects of Training
Group (ts $ 1.05, ps % 0.29), indicating that all subjects were equally
good at the task prior to the intervention. Next, we assessed recognition memory performance on individual probe types in the local
(lures, targets, fillers) and global (targets, fillers) conditions separately, predicting that conflict training should selectively result in
shorter response times at posttest for lure items on the local block.
Local block training effects. As seen in the left panel of
Figure 4, in the Local (conflict) block, both the High- and LowConflict groups appear to improve from pretest to posttest, with the
High-Conflict group showing a larger improvement. Mixed effects
models that included Assessment (Pretest, Posttest) as an additional fixed factor confirmed this observation: lure trial correct
response times revealed a main effect of Assessment, t ! 2.51, p !
.01 and an interaction of Assessment and Conflict Training (t !
2.30; p ! .02). That is, the High-Conflict group showed a greater
reaction time (RT) improvement from pre- to posttest relative to
the Low-Conflict group (MHC ! 183 ms improvement vs. MLC !
80 ms improvement). Even though the models of the other item
types (targets and fillers) revealed significant main effects of
Assessment (ts % 2.82, ps $ 0.004), neither resulted in a reliable
interaction of Assessment and Conflict Training (ts $ 1.18, ps %
0.24). This indicates that High-Conflict trainees were faster to
respond only to lure items after training relative to Low-Conflict
trainees. There were no effects for the Adaptivity contrast for any
probe type (ts $ 1.17, ps % 0.23), suggesting that performance
adaptivity did not influence global or local recognition memory
procedures.
Global block training effects. Although there were main effects
of Assessment for both target and filler item types of the global block
(ts % 3.11, ps $ 0.002), no effects were observed for either the
conflict or the adaptivity contrasts (ts $ 0.62, ps % 0.54; see Figure
4, right panel). These results are consistent with a process-specific
account of cognitive control training: only High-Conflict training
resulted in faster lure RTs on an untrained recognition memory task
that theoretically involved shared conflict resolution demands; no
effects of Adaptive Training were observed.
Table 1
Summary of Pre/Post Assessment Task Results
Task
Measure and condition
Conflict training
Recognition memory task
Local lures
Local targets
Local fillers
Global targets
Global fillers
High competition/Low association
High competition/High association
Low competition/Low association
Low competition/High association
Interference score
Facilitation score
Ambiguous sentence regression-path time
Unambiguous sentence regression-path time
Ambiguous sentence second-pass time
Unambiguous sentence second-pass time
Object-extracted first-pass time
Subject-extracted first-pass time
Object-extracted second-pass time
Subject-extracted second-pass time
Verb generation task
Stroop task
Garden-path recovery
Relative clause processing
3-back training
!
!
!
!
Note. Bolded rows denote task conditions with heightened conflict resolution demands where we predicted transfer from high-conflict training. For the
garden-path recovery task, see text and figures for details about the sentence regions where effects appear for the reading time measures.
Marginal Training-by-Assessment interaction at the p $ .10 level. ! Significant Training-by-Assessment interaction at the p $ .05 level.
13
Figure 4. Global/local recognition memory task performance. Pre/post change in correct response times for
each of the trial types (targets, fillers, lures), split by local (left panel) and global (right panel) recognition blocks.
Error bars represent & 1 standard error of the mean. !Significant at the p $ .05 level for the Training
Group-by-Assessment interaction term. Note that the respective contrasts for Conflict Training and Adaptive
Training are High-Conflict versus Low-Conflict and Low-Conflict versus 3-Back, thus a line spanning the
High-and Low-Conflict groups corresponds to a reliable interaction of Assessment and Conflict Training.
Verb generation task. Data were excluded from six subjects

because of low naming accuracy ($60%), when subjects produced
either nonverbs or repeated auxiliary verbs (e.g., be, have, do)
across several trials (High-Conflict: n ! 4, Low-Conflict: n ! 1,
3-Back: n ! 1). Button press time was used to index the amount
of time it took for subjects to generate a relevant verb in response
to a noun cue (Persson et al., 2007).
Pretraining effects. We examined initial performance on
verb generation by modeling median response times at pretest
with fixed effects of Training Group (Conflict Training, Adaptive Training), Association (High, Low), and Competition
(High, Low). In the model of condition-level response times, we
observed main effects of Association (b ! 807.57, SE !
120.29, t ! 6.71, p $ .001) and Competition (b ! 412.68, SE !
120.29, t ! 3.43, p ! .006). Consistent with prior work (Snyder
et al., 2010, 2011), low association items resulted in longer RTs
(M ! 2360 ms) compared to high association words (M ! 1558
ms), and subjects generated verbs to high competition nouns
(M ! 2167 ms) more slowly than to low competition nouns
(M ! 1750 ms). We observed no effects of Training Group on
pretest verb generation (ts $ 1.55, ps % 0.12), indicating that
the groups had comparable pretraining performance.
Training effects. We modeled median button press time in
separate mixed-models for each of the four noun conditions. The
models included fixed effects of Assessment (Pretest, Posttest) and
Training (Conflict Training, Adaptive Training), and random effects
of Subjects. Figure 5 illustrates that the High-Conflict trainees, but not
the Low-Conflict trainees, appear to show a selective performance
improvement in the High-Competition/High-Association condition,
as predicted. This was confirmed by a reliable interaction of Assessment and Conflict Training for this task condition (t ! 2.41; p ! .02):
the High-Conflict groups RT improved significantly more than the
Low-Conflict group from pretest to posttest (MHC ! 891-ms improvement vs. MLC ! 195-ms improvement). This effect did not
emerge for High-Competition/Low-Association nouns (t ! 0.09; p !
.93) or Low-Competition/Low-Association nouns (t ! 1.58; p ! .12),
but there was a marginal interaction of Assessment and Conflict
Training for Low-Competition/High-Association items (t ! 1.79; p !
.07). Taken together, the High-Conflict trainees showed significant
improvements over Low-Conflict trainees selectively when a noun is
linked to multiple verb responses (high competition) that are easy to
retrieve (high association, or low retrieval demand), that is, under the
purest conflict resolution pressure without the confound of a heightened retrieval demand (Snyder et al., 2010, 2011). In terms of the role
of adaptivity, we observed no effects of Assessment and Adaptive
Training on any items (ts $ 1.14, ps % 0.25), suggesting that the
3-Back and Low-Conflict groups did not differ in their abilities to
generate verbs.
Stroop task.
Pretraining effects. Baseline Stroop performance (in response
times on correct trials) was examined with a mixed-effects model with
fixed effects of Training Group (Conflict Training, Adaptive Training) and Trial Type (Congruent, Incongruent, Neutral). The model
revealed main effects of Trial Type when contrasting congruent and
neutral trials (b ! #35.39, SE ! 14.14, t ! #2.50, p ! .01) and
incongruent and neutral trials (b ! 78.57, SE ! 14.14, t ! 5.55 p $
.001). These patterns replicate the canonical Stroop effect: Incongruent trials (701ms) elicited slower response times than neutral trials
(609 ms), which were slower than congruent trials (578 ms). Thus, we
computed Stroop interference (incongruent minus neutral RTs) and
Stroop facilitation scores (neutral minus congruent RTs), and used
these values when assessing the effects of training group on Stroop
performance. Finally, the fixed factor of Training was not reliable for
14
HUSSEY ET AL.
Figure 5. Verb generation task performance. Pre/post change in correct response times for noun conditions
parametrically different in terms of competition and association levels. Error bars represent & 1 standard error
of the mean. !Significant at the p $ .05 level and p ! .07 for the Training Group-by-Assessment interaction
term. Note that the respective contrasts for Conflict Training and Adaptive Training are High-Conflict versus
Low-Conflict and Low-Conflict versus 3-Back, thus a line spanning the High-and Low-Conflict groups
corresponds to a reliable interaction of Assessment and Conflict Training.
pretest Stroop performance (ts $ 0.27, ps % 0.787), suggesting that

the groups did not differ before training.
Training effects. We modeled Stroop interference and Stroop
facilitation scores separately as a function of Training (Conflict
Training, Adaptive Training) and Assessment (Pretest, Posttest).
We found only a marginal main effect of Assessment, t ! 1.76,
p ! .08 for Stroop interference scores, and no effects for Stroop
facilitation scores. No effects of training emerged for either measure (Stroop interference: ts $ 0.51, ps $ 0.60; Stroop facilitation:
ts $ 0.75, ps $ 0.45). This suggests that neither training manipulation (Conflict, Adaptivity) had cross-assessment Stroop effects.
This could be attributable to the moderate-to-low reliability of
Stroop performance, indexed by split-half correlations corrected
with the Spearman-Brown Prophecy formula (Pretest Interference ! 0.72, Posttest Interference ! 0.71, Pretest Facilitation !
0.48, Posttest Facilitation ! #0.19). Perhaps related, exit interviews indicated that several subjects across all three training
groups (High-Conflict: n ! 20; Low-Conflict: n ! 15; 3-Back:
n ! 18) had strategically blurred their vision to avoid reading the
words. We will return to this issue in the Discussion.
Sentence processing: Garden-path recovery. Following
Novick et al. (2014), we collected two measurements that are
sensitive to moment-by-moment reinterpretation of garden-path
sentences. First, we assessed changes in accuracy to comprehension questions probing for lingering misanalysis (e.g., Did the thief
hide himself?). Second, we examined changes in eye-movement
patterns on correct trials only, namely, when one would expect
readers to make leftward saccades in search of information to help
them revise an initial misinterpretation and ultimately arrive at the
correct meaning of the sentence. Sentences were divided into four
regions of interest (e.g., While the thief hid/the jewelry/that was
elegant and expensive/sparkled brightly for Ambiguous items and

The jewelry/that was elegant and expensive/sparkled brightly/
while the thief hid for Unambiguous items). Region 4 marks the
point in ambiguous sentences when conflicting evidence first
arrives (e.g., sparkled brightly; see also Novick et al., 2014).
Because region 4 contains different material in ambiguous and
unambiguous sentences, we ran separate multilevel linear regression models for each condition (see Novick et al., 2014).
We assessed real-time sentence reanalysis using measures
of regression-path time and second-pass time reading time.
Regression-path time (also known as go-past time) reflects the
total time it takes a reader to exit a region to the right after first
entering it from the left, including regressions out of the region
first before moving forward (Sturt, 2007). Second-pass reading
time includes all rereadings of a sentence region, a measure that is
commonly believed to reflect reanalysis of information that was
initially misinterpreted (e.g., Clifton et al., 2003; Sturt, 2007;
Rayner, 1998; Trueswell, Tanenhaus, & Kello, 1993). Because
second-pass time includes reading times on sentence regions of
varying lengths, we included region length as a covariate and a
random-slope term for Subjects in our mixed-models (see Ferreira
& Clifton, 1986).
Comprehension accuracy: Pretraining effects. Offline comprehension data from six subjects were excluded because of poor
accuracy on noncritical filler items (High-Conflict: n ! 1; LowConflict: n ! 1; 3-Back: n ! 4). To assess the pretraining performance, we ran a generalized linear mixed model with fixed effects of
Training Group (Conflict Training, Adaptive Training) and Condition
(Ambiguous, Unambiguous) and random effects of Subjects and
Items. We observed the classic lingering garden-path effect through a
robust main effect of Condition (b ! 2.55, SE ! 0.34, z ! 7.52, p $
.001). Accuracy to comprehension questions following ambiguous

items (0.70) was lower than unambiguous items (0.91). Crucially, we
found no baseline differences as a function of Training (zs $ 0.58,
ps % 0.56).
Comprehension accuracy: Training effects. To test the effects
of training across time, we ran comparable generalized linear mixed
model with Assessment (Pretest, Posttest) as an additional fixed
factor. We found a reliable interaction of Condition and Assessment
(b ! #1.49, SE ! 0.42, z ! #3.571, p $ .001) qualified by a main
effect of Condition (b ! 1.75, SE ! 0.21, z ! 8.27, p $ .001).
Unpacking this effect, we found larger cross-assessment accuracy
changes for Ambiguous items (MPre ! 0.70, MPost ! 0.82) compared
to Unambiguous items (MPre ! 0.91, MPost ! 0.90). There were no
effects of Conflict or Adaptive Training, which is consistent with prior
work demonstrating that mere exposure to infrequent constructions
can improve offline comprehension performance (Fine, Jaeger,
Farmer, & Qian, 2013; Wells, Christiansen, Race, Acheson, & MacDonald, 2009). Next we ask whether there are any training-related
differences in readers real-time language processing abilities when
they successfully arrive at the correct interpretation.
Regression-path time: Pretraining effects. Eye-movement
data were not included from eight subjects who could not be calibrated at either assessment (High-Conflict: n ! 2; Low-Conflict: n !
4; 3-Back: n ! 2). We first modeled baseline regression-path time in
the critical region (Region 4; sparkled brightly) as a function of
Training (Conflict Training, Adaptive Training) with random effects
of Subjects and Items separately for each sentence type (ambiguous
vs. unambiguous items), and importantly we found no effects of
Training Group for ambiguous sentences (ts $ 1.21, ps % 0.23). We
did, however, observe baseline differences of Adaptive Training for
unambiguous items (b ! 497, SE ! 249, t ! 2.00, p ! .05): the
3-Back group demonstrated longer regression-path times (2063 ms)
from the final region of unambiguous sentences relative to the LowConflict group (1618 ms; see also Figure 6B).
Regression-path time: Training effects. As can be seen in
Figure 6A, regression-path times stemming from region 4 (the conflict
region) appeared faster at posttest than pretest for the High-Conflict
group but not the Low-Conflict group, whereas the Low-Conflict and
3-Back groups did not differ from each other. No group seems to
show shorter regression-path times in other sentence regions. The
mixed-model for region 4 of ambiguous items yielded a marginal
interaction for Assessment and Conflict Training, t ! 1.79, p ! .07.
The comparable interaction term for Assessment and Adaptive Training did not reach significance, t ! 1.07, p ! .29. That is, regressionpath times stemming from the conflict region improved only for the
subjects practicing an intervention aimed at honing cognitive control
processes (pre/post gain: MHC ! 785ms vs. MLC ! 204ms). This
suggests that, at posttest, High-Conflict trainees had an easier time
revising misinterpretations of ambiguous sentences. Critically, there
was no Assessment-by-Conflict Training interaction for in any other
region of ambiguous sentences (ts $ 0.92; ps % 0.35).
In contrast, High-Conflict training did not appear to affect
regression-path times for unambiguous sentences. Inspection of
Figure 6B reveals that neither the High-Conflict group nor the
Low-Conflict group exhibited any cross-assessment changes in
regression-path time in any region of unambiguous sentences.
Importantly, the interaction term for Assessment and Conflict
Training did not reach significance in region 4, t ! 0.28, p ! .78,
which were comparable with disambiguating region of ambiguous
15
sentences in content and position, respectively. That is, the conflict

manipulation had selective effects, theoretically attributable to the
High-Conflict groups experience with conflict resolution during
training, which is consistent with a process-specific account of the
effects (see also Novick et al., 2014).
Figure 6B shows that the 3-Back group appears to be the only
group to show any cross-assessment change in regression-path time in
region 4 of unambiguous sentences. This is perhaps because 3-Back
trainees started out with reliably longer regression-path times at pretest. The multilevel model of regression-path time in region 4 of
unambiguous items confirmed this improvement through a marginal
interaction of Assessment and Adaptive Training, t ! 1.87, p ! .06.
The 3-Back group improved by 841 ms from pre- to posttest on
unambiguous items, whereas the Low-Conflict group only improved
by 177 ms. The 3-Back groups reading time patterns on unambiguous items suggest possible improvement on an ability distinct from
cognitive control for these trainees: They spend less time returning to
earlier regions following entry into the final sentence region without
the presence of conflict at posttest relative to pretest. On the other
hand, the High-Conflict group, which practiced conflict resolution
throughout training, demonstrated real-time improvements in
regression-path time that stemmed from only the region of conflict
(i.e., in ambiguous items). This effect is consistent with the training
pattern observed by Novick et al. (2014).
To corroborate this selective pattern, namely that High-Conflict
trainees ability to deal with syntactic conflict eases after training, we
also analyzed second-pass reading times, which includes all rereading
times in a sentence region. Second-pass time is a component of
regression-path time that is also widely believed to capture reanalysis
processes in earlier regions following initial misinterpretation (Meseguer, Carreiras, & Clifton, 2002; von der Malsburg & Vasishth,
2013).
Second-pass time: Pretraining effects. We first modeled
baseline second-pass time in all regions as a function of Training (Conflict Training, Adaptive Training) and region string
Length as a covariate with random effects of Subjects and Items
separately for each sentence type (ambiguous vs. unambiguous
items). Importantly, we found no differences among the Training Groups for either sentence type (ts $ 1.58, ps % 0.11), with
the exception of an effect of Adaptive Training in Region 4 of
unambiguous sentences (b ! 227, SE ! 103, t ! 2.19, p ! .03).
This suggests that before training, all training groups were
matched in rereading performance, aside from the Low-Conflict
group demonstrating faster rereading times (length-corrected
rereading time: #273 ms) in the final region of unambiguous
sentences relative to the 3-Back group (length-corrected rereading time: #78 ms).
Second-pass time: Training effects. As can be seen in Figure
7A, High-Conflict trainees appear to show large pre/post changes
in length-corrected rereading times of early regions in ambiguous
sentences. The mixed model of second-pass time yielded an interaction of Assessment and Conflict Training in Regions 1 (t ! 2.38,
p ! .02) and 3 (t ! 1.99, p ! .046) of ambiguous sentences. In
particular, the High-Conflict group was reliably faster to reread
regions 1 and 3 following training (pre/post gains: MR1 ! 216 ms,
MR3 ! 160 ms), compared to the Low-Conflict group (pre/post
gains: MR1 ! #34 ms, MR3 ! #55 ms). We also noted a marginal
interaction of Assessment and Conflict Training in region 4 (b !
165, SE ! 93, t ! 1.77, p ! .08) and no effect in region 2 (t !
HUSSEY ET AL.
16
Figure 6. Regression-path time by sentence region for (A) ambiguous and (B) unambiguous sentences. Marked
in gray is the primary region of comparison, region 4 (the conflict region in ambiguous but not unambiguous
items). Error bars represent & 1 standard error of the mean. p $ .07 for the Training Group-by-Assessment
interaction term. Note that the respective contrasts for Conflict Training and Adaptive Training are High-Conflict
(left panel) versus Low-Conflict (middle panel) and Low-Conflict versus 3-Back (right panel).
1.36, p ! .17). Adaptive Training only interacted with Assessment

for region 4 of ambiguous sentences (b ! 238, SE ! 104, t ! 2.28,
p ! .02; remaining regions: ts $ 1.69; ps % 0.09): the 3-Back
group showed a larger improvement from pre- to posttest compared to the Low-Conflict group. This effect is likely a combined
effect of the 3-Back groups 58-ms improvement (decrease) alongside the Low-Conflict groups cross-assessment 87-ms reading
time increase. Taken together, these results indicate that subjects
were selectively faster to reread earlier sentence information of
ambiguous items only if they underwent training that targeted
17
Figure 7. Length-corrected second-pass reading time for (A) ambiguous and (B) unambiguous sentences. Error
bars represent & 1 standard error of the mean. !Significant at the p $ .05 level for the Training Group-byAssessment interaction term. Note that the respective contrasts for Conflict Training and Adaptive Training are
High-Conflict (left panel) versus Low-Conflict (middle panel) and Low-Conflict versus 3-Back (right panel).
cognitive control. The trainees that did not receive cognitive control practice did not improve in these regions. This effect extends
the regression-path findings reported above and the training pattern observed by Novick et al. (2014).
Importantly, the effect of High-Conflict training on second-pass
reading times did not extend to unambiguous sentences. As shown in
Figure 7B, there were no changes across assessments in the secondpass times of any region for either the High-Conflict or Low-Conflict
group. Indeed, no Assessment-by-Conflict Training interactions
emerged in any region of unambiguous sentences (ts $1.54; ps %
0.12). Crucially, the absence of an effect of Conflict Training for
unambiguous items provides further support for a process-specific
18
HUSSEY ET AL.
account of the pre/post improvements for the High-Conflict group

(relative to the Low-Conflict group), namely that conflict training
only matters under sentence processing conditions when cognitive
control (revising a misinterpretation) is required. However, Adaptive
Training appeared to influence unambiguous sentences. Mixed models revealed an interaction of Assessment and Adaptive Training for
region 1, t ! #2.71, p ! .007: the adaptive Low-Conflict trainees
showed faster rereading times at posttest compared to pretest (108ms
faster) in comparison with the 3-Back group (109 ms slower). However, because the Low-Conflict groups rereading times did not significantly improve across assessments, this result does not seem to
reflect the benefits of Adaptive Training. No other Assessment-byAdaptive Training interactions were observed (ts $1.63; ps % 0.10).
In general, the training activity did not impact second-pass reading
times when parsing sentences that lacked the need to resolve among
competing interpretations.
Sentence processing: Relative clause parsing. We partitioned object- and subject-extracted relative clauses into four analysis
regions (e.g., The farmer/who the expert questioned/promoted the
product/at the fair and The farmer/who questioned the expert/promoted the product/at the fair.), where the second region contains the
critical relative clause. We conducted analyses on correct trials only as
a means of measuring eye-movement patterns during cases when
sentences were read and understood (this resulted in a loss of 13.2%
of the total data at pretest and 15.6% at posttest). First-pass reading
time was used to assess whether training had any discriminate effects
on readers processing difficulty in the relative clause region of
object- and subject-extracted items. We also report second-pass reading time to capture revision efforts following the onset of the reducedrelative portion of the sentence. Because this critical region occurred
early in the sentence (at region 2), minimal regressions were made to
earlier portions of the sentence from this region, preventing us from
having sufficient data to assess regression-path time; thus, we report
only second-pass rereading times (note, however, that of the eligible
regression-path data, we found no effects of Training: ts $ 1.61; ps %
0.11). As detailed above, this analysis was intended to test for the
divergent validity of the conflict training effects by providing a
second sentence processing task that does not routinely recruit cognitive control resources. Moreover, we hoped to rule out interpretations of our results that were bound to the High-Conflict group
improving on hard tasks per se, rather than on tasks that involve
conflict.
First-pass time: Pretraining effects. We evaluated baseline
first-pass reading times at pretest in a mixed model containing
Training (Conflict Training, Adaptive Training) and Sentence
Type (Object-Extracted, Subject-Extracted) as fixed effects, region
string Length as a covariate, and Subjects and Items as random
effects. The model revealed significant main effects of Sentence
Type (b ! 155, SE ! 45, t ! 3.44, p ! .005) and Adaptive
Training (b ! 128, SE ! 54, t ! 2.35, p ! .02). Consistent with
prior work, object-extracted clauses elicited longer first-pass times
in the second sentence region (M ! 203 ms) compared with
subject-extracted items (M ! 72 ms), replicating the expected
difficulty effect. Moreover, the 3-Back group demonstrated longer
first-pass times, regardless of Sentence Type (M ! 177 ms)
compared with the Low-Conflict group (M ! 79 ms). No such
effects emerged for Conflict Training or for any interactions of
Training and Sentence Type at pretest (ts $ 1.33, ps % 0.18).
First-pass time: Training effects. As illustrated in Figure 8,

first-pass reading time appears to pattern differently from pre- to
posttest for the 3-Back and Low-Conflict groups for subject- and
object-extracted items. Indeed, upon evaluating first-pass reading
times across assessments, we found reliable interactions of Assessment and Adaptive Training for both object-relatives, t ! 2.17,
p ! .03 and subject-relatives, t ! 1.96, p ! .05. Following
training, the Low-Conflict group demonstrated slower first-pass
times on the critical region of object-extracted items compared to their
pretest performance on these items (M ! 154 ms slowdown). The
3-Back trainees showed the opposite result with faster first-pass
times at posttest relative to pretest on object-extracted clauses
(M ! 89 ms speed-up). On subject-extracted sentences, the effect
reversed: The Low-Conflict trainees had faster first-pass times at
posttest relative to pretest performance on region 2 (M ! 106 ms
speed-up), whereas the 3-Back trainees had slower first-pass times
at posttest compared to pretest (M ! 118 ms slowdown). Importantly, we found no interactions of Assessment and Conflict Training (ts $ 1.13, ps % 0.25), suggesting that the locus the crossassessment changes in relative clause parsing was due to a
mechanism distinct from cognitive control.
Second-pass time: Pretraining effects. Using the same modeling approach for first-pass reading times, we assessed baseline
second-pass reading times. We found an interaction of Adaptive
Training and Sentence Type (b ! 286, SE ! 139, t ! 2.05, p !
.04): 3-Back trainees spent more time rereading the critical sentence region of object-extracted clauses at pretest compared to the
Low Conflict trainees. But there was no effect of sentence type,
suggesting that object relative clauses do not require more revision
than subject relative clauses, consistent with an account wherein
conflict resolution is not needed.
Second-pass time: Training effects. As can be seen in Figure 9,
second-pass reading times at pretest and posttest were quite similar for
each training group, including the relative clause region of both
object-extracted and subject-extracted relative clause sentences. Indeed, there was no interaction of Assessment and Conflict Training in
the critical relative clause region (region 2) of either sentence type
(ts $ 1.43, ps % 0.15), or in any other region of both sentence types
(ts $ 1.05, ps % 0.30). Models of Adaptive Training, however,
revealed an interaction of Training and Assessment for region 2 of
subject-extracted clauses, t ! #2.09, p ! .04: the adaptive LowConflict trainees showed faster rereading times at posttest compared
to pretest (34 ms faster) in comparison to the 3-Back group (66 ms
slower). Overall, as predicted, cognitive control training did not confer benefits to sentence processing outcomes with greater syntactic
complexity broadly, namely, any sentence type that may be somewhat
difficult to process. Instead, High-Conflict training conferred processing advantages only to a reading task when reinterpretation per se (and
thus the theoretical engagement of cognitive control) was necessary.
Discussion
To date, there is ample evidence that cognitive control helps
resolve competition in recognition memory (Badre & Wagner,
2007; Gray et al., 2003; Jonides & Nee, 2006; Oberauer, 2005;
Nelson et al., 2003) and language processing, during both verb
generation and sentence revision (January et al., 2009; Kan &
Thompson-Schill, 2004; Novick et al., 2005, 2009; ThompsonSchill et al., 2002; Ye & Zhou, 2009). However, much of the
evidence that supports this view has been correlational in nature
19
Figure 8. Length-corrected first-pass time for relative clause items for (A) object-extracted and (B) subjectextracted clauses. Region 2 is highlighted in gray and marks the critical relative clause region. Error bars
represent & 1 standard error of the mean. !Significant at the p $ .05 level for the Training Group-by-Assessment
(but see Hsu & Novick, 2016), drawn for example from VLPFC
patients showing coimpairments across tasks (Hamilton & Martin, 2005; Novick et al., 2010; Robinson, Blair, & Cipolotti,
1998; Vuong & Martin, 2011), brain-imaging work showing
colocalized VLPFC recruitment across tasks that share conflict
resolution demands (January et al., 2009; Ye & Zhou, 2009),
and individual differences studies in both children and adults

showing subject-by-subject covariation in performance
(Khanna & Boland, 2010; Nilsen & Graham, 2009; Novick et
al., 2014; Woodard, Pozzan, & Trueswell, 2016). Though it
seems clear, therefore, that common mechanisms are involved
in linguistic and nonlinguistic cognitive control (Fedorenko,
HUSSEY ET AL.
20
Figure 9. Length-corrected second-pass time for relative clause items for (A) object-extracted and (B)
subject-extracted clauses. Region 2 is highlighted in gray and marks the critical relative clause region. Error bars
represent & 1 standard error of the mean. !Significant at the p $ .05 level for the Training Group-by-Assessment
2014), it has been unclear whether cognitive control causes

differential outcomes on these measures.
Here, we establish that there is a cause-and-effect interplay between
cognitive control and recognition memory and language processing
performance on various tasks. In particular, each of the assessments
we administered involved the need to resolve among active, competing representations and, thus, cognitive control: inhibiting familiar but
irrelevant memoranda during item recognition (global-local recognition memory task), producing a verb under high selection demands by
resolving among competitors (verb generation task), and revising
early misinterpretations of sentence meaning (garden-path recovery

during sentence processing). Critically though, performance on each
of these tasks was reliably influenced by an intervention that targeted
conflict resolution mechanisms, but importantly not by a minimally
different intervention that reduced involvement of such mechanisms.
Concretely, following High-Conflict training, subjects more easily
recognized memory probes that were familiar but extraneous as lures
(shorter response times); more easily generated a verb to a noun cue
that had multiple competing verb associates that were all easy to
retrieve (shorter response times); and more easily reread syntactically
ambiguous sentences (shorter second-pass and go-past reading times
that stemmed from the conflict region). Performance on within-task
control conditions (e.g., item recognition without interference; verb
generation with few competitors; unambiguous sentence reading) did
not improve. We thus ascribe these findings to the selective plasticity
of domain-general cognitive control processes that may actually shape
peoples memory and language abilities under these conditions. In
what follows, we summarize each of the results in turn and then
discuss the overall implications.
Domain-Specific Transfer: Recognition Memory

The local block of the memory assessment largely imitated the
conflict environment of the n-back-with-lures training task because it
occurred in the context of item recognition (lures were recognition
probes that recently appeared but in a different, nontarget location); it
was thus administered to test for domain-specific transfer only. That
is, if domain-specific cognitive control procedures arbitrate conflict
processing, then training-transfer effects should be relatively narrow
in scope; thus, we might expect to observe transfer in the local block,
but not in the global block (which contained no lures), and crucially
not in any other assessment task even under high conflict resolution
demands. Indeed, we observed that High-Conflict trainees response
latencies to lures in the local block diminished significantly, whereas
Low-Conflict trainees response latencies did not (nor did the response latencies of 3-Back trainees). Target- and filler-item performance remained unchanged across assessments in both the local and
global blocks, as expected. Thus, the presence of lures during n-back
practice was likely the critical element that conferred the transfer
effect, perhaps, as the signal detection analyses on the posttest n-back
task suggest. Specifically, on the posttest n-back task, the LowConflict group demonstrated better target/nontarget discriminability
and a less conservative response criterion relative to the High-Conflict
trainees. This may explain the Low-Conflict groups numerically
higher accuracy on global targets. Thus, it is possible that removing
conflict conferred some benefits in low-conflict conditions of this
near-transfer task.
This observation aside, the significant pattern on lure response
times minimally suggests that the same cognitive control system
supports conflict processing across recognition memory tasks that
contain only superficial differences. Moreover, it is to some extent
plastic and trainable: an intervention that involves practice dealing
with interfering representations in memory benefits interference resolution in a similar yet novel setting. To evaluate how general this
effect was, we turn our attention to the remaining pre/post assessment
tasks. If domain-general cognitive control was honed with (HighConflict) n-back lure training, then we predicted selective transfer on
high-conflict conditions in tasks that did not closely match the conflict
21
environment of training (e.g., verb generation and sentence processing).
Domain-General Transfer
Verb generation. The verb generation task is well established
to involve cognitive control when multiple underdetermined response candidates compete (Botvinick et al., 2001; Snyder &
Munakata, 2008; Thompson-Schill et al., 1997). However, prior
debates have argued about what factors trigger cognitive control
during this task to resolve the competition (cf. Martin & Cheng,
2006; Thompson-Schill & Botvinick, 2006): is it selection demands (i.e., competition effects increase as the number of verbs
associated with a noun increases, thus making it harder to select
one among many), or memory retrieval demands (i.e., some nouns
like folder or leaf have weak verb associations, thus increasing
retrieval demands, whereas nouns like phone or scissors have
strong verb associations, thus decreasing retrieval demands)? Following Snyder et al. (2008, 2010, 2011), we assume that both
competition and retrieval elements are at play: namely, competition effects are highest when several verb associates are easily
retrievable (high association, low retrieval demand), thus requiring
the greatest cognitive control because activation spreads among
multiple verbs that are all strongly linked to the same noun (as
call, ring, and answer are all highly associated with
phone, they are excellent candidates that compete for selection,
thereby slowing responding; see Snyder & Munakata, 2008).
Given this theoretical framework, and that conflict resolution
demands are maximized while retrieval demands are minimized
for the High Association/High Competition condition (Snyder et
al., 2011), we predicted thatif cognitive control is domaingeneral cognitive control training via n-back-with-lures should
alleviate the competition pressures in this condition. That is, dealing with competition during n-back would result in shorter response latencies during verb generation in the condition where
conflict demands are the highest and retrieval demands are reduced. Indeed, compared to the Low-Conflict (no lures) training
group, the High-Conflict group demonstrated a significantly larger
reduction in production latencies from pretest to posttest.
Interestingly, a transfer effect was not observed in the three other
conditions with theoretically lower conflict resolution demands, including when there is spreading activation over multiple yet weakly
related responses (Low Association/High Competition). Again, when
retrieving a response is hard, spreading activation among several
weakly connected responses (e.g., among file, close, and open
in response to folder) may boost the activation level of the various
candidates, thus alleviating retrieval demands and neutralizing the
effect of competition (selection). That we do not find transfer of
cognitive control training to such items strongly corroborates Snyder
and colleagues (2011) account that conflict resolution demands may
be lower here (i.e., neutralized) compared with the High Association/
High Competition condition. In addition, we infer that the conflict
resolution process practiced during n-back-with-lures is domaingeneral (transfer from recognition memory to verb generation). Moreover, verb generation, at least when selection/competition pressures
are strongest and retrieval demands are minimized, may be dependent
on the cognitive control process trained during prolonged exposure to
n-back-with-lures.
22
HUSSEY ET AL.
One particularly intriguing aspect of this transfer profile is that the

quality of the conflict experienced during n-back-with-lures and verb
generation is quite differentprepotent in the former case and underdetermined in the latter. That is, correctly recognizing a lure during
n-back requires overriding a dominant (prepotent) yet irrelevant representation, and not responding yes to any recently presented and
therefore familiar stimulus (Oelhafen et al., 2013; Novick et al.,
2014). During verb generation under High Association/High Competition conditions, a noun cue (e.g., phone) does not give rise to a
singular dominant response, but rather activates multiple associated
verbs (e.g., call, ring, answer) that are equally relevant (and are
therefore underdetermined; Snyder & Munakata, 2008). Though the
resolution of competitive interactions has been attributed to common,
domain-general procedures regardless of the source of conflict (Botvinick et al., 2001), others have shown that prepotent conflict and
underdetermined conflict are sometimes handled by overlapping but
partially independent regions within PFC, suggesting that the resolution procedures are not always exactly identical (Snyder et al., 2014).
However, we speculate that our training regimen targeted the overlapping neural substrate in this part-dissociation. That we observe
transfer effects from a task involving prepotent competition to one
involving underdetermined competition strongly supports the idea of
a general cognitive control mechanism that acts to settle the system
into a stable state, irrespective of how the competition originates.
Moreover, the transfer pattern suggests a causal partnership: enhanced
mechanisms that resolve prepotent competition also enhance the resolution of underdetermined response competition. This finding in
particular is important because of its potential for application; it could
inform intervention designs for people with cognitive control deficits
that affect recognition memory and language processing performance
under situations that create both prepotent and underdetermined conflict (Novick et al., 2009, 2010).
Sentence processing: Garden-path recovery versus object
relative clauses. To further test the domain-generality question,
and in particular whether cognitive control also has a causal
influence on a different language processing task, we administered
a reading task involving syntactically ambiguous sentences that
were ripe for misanalysis. A range of patient, brain stimulation,
and brain-imaging studies suggest that one important cognitive
control function is to revise misinterpretations of sentence meaning. That is, the real-time detection of a processing conflictthat
comprehension has gone awryinitiates conflict-control mechanisms that serve to resolve the incompatibility between the two
different interpretations (Hsu & Novick, 2016; January et al.,
2009; Mazuka et al., 2009; Novick et al., 2005, 2009; Vuong &
Martin, 2011; Ye & Zhou, 2009).
The evidence for this has been largely correlational. However, a
recent exception is a training study by Novick and colleagues
(2014) who demonstrated that n-back-with-lures practice positively influenced readers ability to arrive at the intended analysis
following misinterpretation. Specifically, using the same gardenpath materials administered in the current study (e.g., While the
thief hid the jewelry that was elegant and expensive sparkled
brightly), subjects were faster to read past the conflict region
(sparkled brightly) after training. Interestingly, regression-path
times did not change in any other region of ambiguous sentences,
or anywhere in unambiguous sentences at all. This strongly implicated a causal effect of domain-general conflict-control functions
on language comprehension. However, as sketched in the intro-
duction, the comparison group of subjects received no active

contact between pretest and posttest assessments, making it hard to
discern whether cognitive control per se was the causal factor. We
aimed to address this issue in the current experiment.
Here, we carefully isolated the conflict ingredient in an attempt
to replicate the earlier study and identifyas in the recognition
memory and verb generation results summarized thus far
whether this was the crucial feature on which the training effect
depended. Our active Low-Conflict control group trained on an
n-back task with the conflict manipulation removed (i.e., no lures).
Across two eye-movement measures that are sensitive to reanalysis processes, High-Conflict training but not Low-Conflict training
resulted in shorter cross-assessment rereading times on ambiguous
(but not unambiguous) items. For the regression-path measure, the
High-Conflict group demonstrated shorter times stemming from
entry into the disambiguating region (sparkled brightly), which
suggests easier recovery from the moment they encountered evidence that conflicted with an initial incorrect transitive interpretation (the thief was hiding the jewelry). This pattern is consistent
with a process-specific effect of domain-general conflict resolution
training, and replicates Novick et al. (2014).
Moreover, we collected a second eye-movement measurement to
confirm the selective patterns observed thus far. For the second-pass
measure, which is sensitive to reprocessing (Trueswell, Tanenhaus, &
Garnsey, 1994), High-Conflict training but again, not Low-Conflict
trainingyielded reliably shorter rereading times in early sentence
regions of ambiguous but not unambiguous sentences. This suggests
that cognitive control training enabled better reprocessing (i.e., more
efficient integration of information that facilitates revision) following
misanalysis. We argue that this finding cross-validates and reinforces
the above regression-path time results. Across studies, we have consistently observed selective transfer effects of High-Conflict training
to temporarily ambiguous sentences, which suggests process-specific
tuning of cognitive control procedures as a result of training. This
interpretation is further corroborated by recent findings demonstrating
that recovery from misinterpretation is dynamically modulated by the
real-time engagement of cognitive control processes (Hsu & Novick,
2016).
As part of the sentence processing assessment, subjects also read
sentences containing subject- and object-relative clauses. These items
were included to separate a difficulty interpretation from a cognitive
control interpretation of our transfer results, because the other assessments unavoidably confounded these factors (e.g., lures are harder to
correctly recognize than filler items; ambiguous sentences are harder
to understand than unambiguous sentences). Reading object relatives
is more processing intensive compared to subject relatives, indexed by
elevated reading times in the relative clause region. This robust
difficulty effect, which we replicated in the current study in first-pass
reading time before training, has been consistently ascribed to increased working memory demands related to information storage and
integration costs across long distances (Fedorenko, Gibson, & Rohde,
2006, 2007; Gibson, 1998; Gordon, Hendrick, & Levine, 2002; Gordon & Lowder, 2012; but see MacDonald & Christiansen, 2002).
Critically though, no results indicate that such processing difficulty is
related to conflict resolution and cognitive control functions that help
readers and listeners recharacterize interpretations on-the-fly, as in
the case of recovering from misinterpretation of syntactic ambiguities.
For example, patients with VLPFC damage routinely show exaggerated conflict effects on cognitive control tasks like item recognition
with lures (Hamilton & Martin, 2005; Novick et al., 2009);

Thompson-Schill et al., 2002, which predicts their failure to revise
initial misinterpretations of word and sentence ambiguities on a variety of tasks (e.g., Novick et al., 2009; Thothathiri et al., 2012; Vuong
& Martin, 2011). Importantly however, despite their cognitive control
impairment, these patients demonstrate a well-preserved ability to
process and understand difficult object relative clauses, within the
normal range of healthy adults (Thothathiri et al., 2012). Moreover,
following brain stimulation that targets left lateral prefrontal regions,
readers exhibit no change in processing relative clauses despite spending less time processing garden-path sentences (Hussey et al., 2015).
This strongly indicates that reading and understanding such relative
clause items may not draw on conflict-control procedures. These
dissociative findings motivated our prediction that there should be no
effect of cognitive control training on processing object relative
clauses, despite their difficulty as compared to subject relative clauses.
Indeed, no effect of High-Conflict training was observed compared
with Low-Conflict training, a finding that (a) reinforces the notion that
the difficulty involved in processing object relative clauses is attributable to factors other than conflict resolution, perhaps limited working memory capacity; and (b) weakens a difficulty account of our
training-transfer results. Instead, we believe there is a more parsimonious interpretation of our data: the consistent transfer from High- but
not Low-Conflict training under conflict-related conditions in recognition memory, verb generation, and sentence processing is attributable to a cause-and-effect interplay between conflict resolution practice during n-back, and improved conflict resolution in the memory
and language outcome measures (see Table 1).
The Role of Performance-Adaptivity in Training

While the conflict manipulation produced results consistent with
the theory of process-specific and domain-general transfer of cognitive control training, the adaptivity manipulation was less theoretically
motivated and, relatedly, produced less expected results. It is thought
that by adapting the difficulty, participants are kept within a range of
difficulty that is neither too easy nor too difficult, producing the most
effective training. However, the empirical support for adaptivity is not
strong (Shipstead et al., 2012). We, too, did not find consistent
evidence in favor of using adaptivity (see also von Bastian & Eschen,
2016). The rightmost column of Table 1 reveals a collection of
inconsistent task conditions that benefit following 3-back (but not
adaptive) training. Although adaptive training enabled participants to
achieve better performance at higher n-levels at posttest, there was
little evidence that this task-specific training gain conferred benefits
on transfer tasks. For example, in the garden-path task, the 3-Back
group demonstrated greater improvements than the Low-Conflict
group on second-pass and regression-path times of unambiguous
items. Similarly, while parsing relative clauses, the 3-Back group
demonstrated greater improvements than the Low-Conflict group on
first-pass times of subject-extracted items, whereas the opposite pattern was true on object-extracted sentences. However, this could be
because at pretest, the 3-Back groups reading times were often
nominally slower than those of the Low-Conflict group. Thus, their
improvements following training may be attributable to some regression to the mean.
There are several potential explanations for why we did not observe
consistent transfer benefits following practice on performanceadaptive training, compared with static (nonadaptive) training. One
23
possibility, as suggested above, is the contribution of retrieval when

performing the training n-back task. Participants in the Low-Conflict
condition have the least ability and least demand to make use of
retrieval when performing the n-back training task. These participants
are performing at high n-levels, making retrieval more difficult and (in
the absence of lures) retrieval less necessary as they could make use
instead of item familiarity to inform their responses. While 3-back
participants were also working in the absence of lures, decreasing the
need for retrieval, they also only needed to retrieve the third item
back, making retrieval much easier and therefore, more likely.
Alternatively, trainees practicing n-back-without-lures may have
experienced overlearning of the task operations by developing narrow
task-specific strategies that may limit the likelihood of observing
transfer beyond the training task itself (see Shiffrin & Schneider,
1977). This may especially be the case because the 3-Back comparison group was exposed to five different stimulus sets over the course
of the experiment, which may have protected them from developing
a task-specific strategy. Regardless, our results suggest that adaptivity
alone is not sufficient for training gains and, at least in some cases, it
is also not necessary.
Caveats and Limitations

Although our predictions of process-specific and domain-general
transfer of cognitive control training were largely confirmed, there
were two instances where the High-Conflict training did not appear to
transfer to conflict resolution (i.e., the highlighted cells of Table 1 that
do not have a reliable effect): on the training task itself (3-back and
6-back with lures) and on the Stroop task. On the posttest n-backwith-lures task, the High-Conflict group was numerically but not
significantly more accurate than the Low-Conflict group on lure trials.
Taken at face value, this result seems to suggest that conflict training
was simply not effective at boosting conflict resolution. However, we
ascribe this result to task-specific practice effects, wherein extensive
practice on a task enables high-proficiency on tasks with the same
surface-level parameters (Shipstead et al., 2012). Although the LowConflict group did not practice conflict resolution, they did practice
the n-back task extensively, as evidenced by their superior performance on target trials of the posttest 6-back trials. Becoming highly
proficient at n-back-without-lures could facilitate n-back-with-lures
performance because these tasks have identical surface features, requiring participants to track relevant memoranda and identify target
items. However, training should only transfer to untrained tasks with
different surface-level features when they share underlying cognitive
resources. Indeed, only High-Conflict training transferred to measures
of conflict resolution on untrained tasks, suggesting that only HighConflict training selectively tapped and improved cognitive control
resources.
The absence of a transfer effect to Stroop interference resolution
was also surprising. Stroop is a canonical conflict-control task that
activates overlapping neural resources with syntactic ambiguity in
VLPFC (January et al., 2009). However, we believe that our subjects
may have adopted a task-specific strategy to avoid information conflict on the Stroop task. Specifically, during exit interviews, a high
number of subjects in each training group (High-Conflict: n ! 20;
Low-Conflict: n ! 15; 3-Back: n ! 18) reported intentionally blurring their vision during the Stroop task. Such a strategy would prevent
the lexical information of the color-word stimulus from being processed, essentially eliminating competition and obviating the need for
24
HUSSEY ET AL.
cognitive control. Improved conflict resolution following HighConflict training is not expected to impact performance on tasks that
do not require cognitive control, and so it should have little influence
on Stroop performance for the subjects (comprising 65% of our
sample) who adopted this strategy.
Another caveat to our findings is that there was an instance of a
nominal (though not significant) baseline difference between the
training groups. For example, the patterns depicted in Figure 6A
suggest dissimilar regression-path times in the critical region prior to
training between the High- and Low-Conflict trainees, which may
have contributed, in part, to the cross-assessment marginal effects
observed between these groups (and at posttest, regression-path times
appear similar between the groups, raising concerns about regression
to the mean). Critically, however, the regression-path time patterns are
strongly substantiated by the second-pass time results (Figure 7A),
where the High-Conflict group improved significantly more from preto posttest than the Low-Conflict group (and these correct rereading
attempts at posttest were faster than Low-Conflict trainees rereading
attempts, alleviating concerns about regression to the mean). This is
especially relevant given that regression-path time includes firstfixation durations that may not necessarily index revision. Secondpass time, on the other hand, includes only time spent rereading a
region, which by definition excludes nonrevision measures (e.g.,
first-fixation duration). Thus, the second-pass time results more
closely reveal any changes in cognitive control following conflict
training, as it is an index of revision.
Implications and Closing Remarks

We find evidence that cognitive control procedures are amenable to improvement through process-specific training: an
intervention designed to target the process of conflict resolution
yields benefits that selectively transfer to untrained memory
and language conditions with high conflict resolution demands.
This suggests that conflict-control mechanisms are domaingeneral (broad in scope) and play a causal role in linguistic and
nonlinguistic performance. It is critical to reiterate, however,
that we did not endeavor to train working memory in hopes of
shaping intelligence, a pursuit that has been met with both
conviction and great skepticism (Au et al., 2016; Melby-Lervg
& Hulme, 2013, 2016). In fact, we remain agnostic about the
goal of this work and the scope of its benefits (e.g., see
Sprenger et al., 2013).
Yet, mounting evidence suggests that training effects can be
observed if there is sufficient overlap between the trained and
untrained tasks in terms of cognitive and neurobiological substrates, irrespective of domain (Dahlin et al., 2008; Karbach &
Kray, 2009; Lvdn, Bckman, Lindenberger, Schaefer, &
Schmiedek, 2010; Waris et al., 2015; Zelinski, Peters, Hindin,
Petway, & Kennison, 2014). But even training specific executive
functions has been met with some uncertainty (Rabipour & Raz,
2012; Rapport et al., 2013). For example, a recent meta-analysis
questioned the efficacy of different types of executive function
training (Rapport et al., 2013). The authors noted, however, that
there were methodological weaknesses in many of the studies they
reviewed. For example, most studies involved experimental designs that either utilized batteries of training tasks (instead of
single tasks that targeted a particular mental process), or contrasted
training interventions to control conditions with training tasks that
were not minimally different in terms of the trained process of

interest. Both design elements therefore make it difficult to identify and test for process-specificity (Rapport et al., 2013).
Indeed, as pointed out in our introduction, one motivation for
the current study was to pinpoint the locus of the effects we
observed in a prior study that compared transfer effects in participants who practiced a battery of training tasks to a passive control
group (Novick et al., 2014). Our current design addressed this
issue by comparing a range of transfer outcomes between groups
that completed minimally different versions of an n-back training
regimen. Thus, our goal was to test whether performance on
well-studied language and memory tasksall previously established to involve cognitive control across multiple experimental
findingswould improve if subjects trained on a task that shared
conflict-control procedures, which routinely activate portions of
ventrolateral PFC. Our finding that such training transfers to a
number of cognitive control tasks despite surface-level differences
corroborates the notion that there may be promise in processspecific training. This result may open the door to exploring
whether patients with cognitive control deficits (who consistently
fail to perform any of the tasks studied in the current work within
the normal range) may be viable candidates for such training. An
important caution, however, is that our effects could very well be
transientwe do not know whether they last for longer than a
couple of weeks. Future research should explore this in greater
detail.
In sum, our work demonstrates that cognitive control is plastic
and offers the prospect of improving domain-general conflict
resolution abilities through targeted training regimens. Such training could have important implications for a variety of populations
(e.g., young children, neuropsychological patients, elderly adults)
whose limited cognitive control restricts performance across a
range of functions.
References
Akay, ., & Hazeltine, E. (2011). Domain-specific conflict adaptation
without feature repetitions. Psychonomic Bulletin & Review, 18, 505
511. http://dx.doi.org/10.3758/s13423-011-0084-y
Au, J., Buschkuehl, M., Duncan, G. J., & Jaeggi, S. M. (2016). There is no
convincing evidence that working memory training is NOT effective: A
reply to Melby-Lervg and Hulme (2015). Psychonomic Bulletin &
Review, 23, 331337. http://dx.doi.org/10.3758/s13423-015-0967-4
Badre, D., & Wagner, A. D. (2007). Left ventrolateral prefrontal cortex and
the cognitive control of memory. Neuropsychologia, 45, 28832901.
http://dx.doi.org/10.1016/j.neuropsychologia.2007.06.015
Baniqued, P. L., Allen, C. M., Kranz, M. B., Johnson, K., Sipolins, A.,
Dickens, C., . . . Kramer, A. F. (2015). Working memory, reasoning, and
task switching training: Transfer effects, limitations, and great expectations? PLoS ONE, 10, e0142169.
Barch, D. M., Braver, T. S., Sabb, F. W., & Noll, D. C. (2000). Anterior
cingulate and the monitoring of response conflict: Evidence from an
fMRI study of overt verb generation. Journal of Cognitive Neuroscience,
12, 298 309. http://dx.doi.org/10.1162/089892900562110
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects
structure for confirmatory hypothesis testing: Keep it maximal. Journal
of Memory and Language, 68, 255278. http://dx.doi.org/10.1016/j.jml
.2012.11.001
Bialystok, E., Craik, F. I. M., & Ryan, J. (2006). Executive control in a
modified antisaccade task: Effects of aging and bilingualism. Journal of

Experimental Psychology: Learning, Memory, and Cognition, 32, 1341
1354. http://dx.doi.org/10.1037/0278-7393.32.6.1341
Bilenko, N. Y., Grindrod, C. M., Myers, E. B., & Blumstein, S. E. (2009).
Neural correlates of semantic competition during processing of ambiguous words. Journal of Cognitive Neuroscience, 21, 960 975. http://dx
.doi.org/10.1162/jocn.2009.21073
Boot, W. R., Simons, D. J., Stothart, C., & Stutts, C. (2013). The pervasive
problem with placebos in psychology: Why active control groups are not
sufficient to rule out placebo effects. Perspectives on Psychological
Science, 8, 445 454. http://dx.doi.org/10.1177/1745691613491271
Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen, J. D.
(2001). Conflict monitoring and cognitive control. Psychological Review, 108, 624 652. http://dx.doi.org/10.1037/0033-295X.108.3.624
Braver, T. S., Gray, J. R., & Burgess, G. C. (2007). Explaining the many
varieties of working memory variation: Dual mechanisms of cognitive
control. In A. R. A. Conway, C. Jarrold, M. J. Kane, A. Miyake, & J. N.
Towse (Eds.), Variation in working memory (pp. 76 106). New York,
NY: Oxford University Press.
Brehmer, Y., Rieckmann, A., Bellander, M., Westerberg, H., Fischer, H.,
& Bckman, L. (2011). Neural correlates of training-related workingmemory gains in old age. NeuroImage, 58, 1110 1120. http://dx.doi
.org/10.1016/j.neuroimage.2011.06.079
Buschkuehl, M., & Jaeggi, S. M. (2010). Improving intelligence: A literature review. Swiss Medical Weekly, 140, 266 272.
Chatham, C. H., Herd, S. A., Brant, A. M., Hazy, T. E., Miyake, A.,
OReilly, R., & Friedman, N. P. (2011). From an executive network to
executive control: A computational model of the n-back task. Journal of
Cognitive Neuroscience, 23, 3598 3619. http://dx.doi.org/10.1162/
jocn_a_00047
Christianson, K., Williams, C. C., Zacks, R. T., & Ferreira, F. (2006).
Younger and older adults Good-enough interpretations of gardenpath sentences. Discourse Processes, 42, 205238. http://dx.doi.org/10
.1207/s15326950dp4202_6
Clifton, C., Jr., Traxler, M. J., Mohamed, M. T., Williams, R. S., Morris,
R. K., & Rayner, K. (2003). The use of thematic role information in
parsing: Syntactic processing autonomy revisited. Journal of Memory
and Language, 49, 317334. http://dx.doi.org/10.1016/S0749596X(03)00070-6
Dahlin, E., Neely, A. S., Larsson, A., Bckman, L., & Nyberg, L. (2008).
Transfer of learning after updating training mediated by the striatum.
Science, 320, 1510 1512. http://dx.doi.org/10.1126/science.1155466
del Ro, D., Maest, F., Lpez-Higes, R., Moratti, S., Gutirrez, R.,
Maest, C., & del-Pozo, F. (2011). Conflict and cognitive control during
sentence comprehension: Recruitment of a frontal network during the
processing of Spanish object-first sentences. Neuropsychologia, 49,
382391. http://dx.doi.org/10.1016/j.neuropsychologia.2010.12.005
DEsposito, M., Postle, B. R., Jonides, J., & Smith, E. E. (1999). The
neural substrate and temporal dynamics of interference effects in working memory as revealed by event-related functional MRI. PNAS Proceedings of the National Academy of Sciences of the United States of
America, 96, 7514 7519. http://dx.doi.org/10.1073/pnas.96.13.7514
Diamond, A., & Lee, K. (2011). Interventions shown to aid executive
function development in children 4 to 12 years old. Science, 333,
959 964. http://dx.doi.org/10.1126/science.1204529
Egner, T. (2008). Multiple conflict-driven control mechanisms in the
human brain. Trends in Cognitive Sciences, 12, 374 380. http://dx.doi
.org/10.1016/j.tics.2008.07.001
Egner, T., Delano, M., & Hirsch, J. (2007). Separate conflict-specific
cognitive control mechanisms in the human brain. NeuroImage, 35,
940 948. http://dx.doi.org/10.1016/j.neuroimage.2006.11.061
Farmer, T. A., Misyak, J. B., & Christiansen, M. H. (2012). Individual
differences in sentence processing. In M. Spivey, M. Joannisse, & K.
McRae (Eds.), Cambridge handbook of psycholinguistics. Cambridge,
25
UK: Cambridge University Press. http://dx.doi.org/10.1017/

CBO9781139029377.024
Fedorenko, E. (2014). The role of domain-general cognitive control in
language comprehension. Frontiers in Psychology, 5, 335.
Fedorenko, E., Gibson, E., & Rohde, D. (2006). The nature of working
memory capacity in sentence comprehension: Evidence against domainspecific working memory resources. Journal of Memory and Language,
54, 541553. http://dx.doi.org/10.1016/j.jml.2005.12.006
Fedorenko, E., Gibson, E., & Rohde, D. (2007). The nature of working
memory in linguistic, arithmetic and spatial integration processes. Journal of Memory and Language, 56, 246 269. http://dx.doi.org/10.1016/
j.jml.2006.06.007
Fedorenko, E., Woodbury, R., & Gibson, E. (2013). Direct evidence of
memory retrieval as a source of difficulty in non-local dependencies in
language. Cognitive Science, 37, 378 394. http://dx.doi.org/10.1111/
cogs.12021
Ferreira, F., Christianson, K., & Hollingworth, A. (2001). Misinterpretations of garden-path sentences: Implications for models of sentence
processing and reanalysis. Journal of Psycholinguistic Research, 30,
320. http://dx.doi.org/10.1023/A:1005290706460
Ferreira, F., & Clifton, C. J., Jr. (1986). The independence of syntactic
processing. Journal of Memory and Language, 25, 348 368. http://dx
.doi.org/10.1016/0749-596X(86)90006-9
Fine, A. B., Jaeger, T. F., Farmer, T. A., & Qian, T. (2013). Rapid
expectation adaptation during syntactic comprehension. PLoS ONE, 8,
e77661. http://dx.doi.org/10.1371/journal.pone.0077661
Fletcher, P. C., & Henson, R. N. (2001). Frontal lobes and human memory:
Insights from functional neuroimaging. Brain: A Journal of Neurology,
124, 849 881. http://dx.doi.org/10.1093/brain/124.5.849
Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 176. http://dx.doi.org/10.1016/S00100277(98)00034-1
Gordon, P. C., Hendrick, R., & Levine, W. H. (2002). Memory-load
interference in syntactic processing. Psychological Science, 13, 425
430. http://dx.doi.org/10.1111/1467-9280.00475
Gordon, P. C., & Lowder, M. W. (2012). Complex sentence processing: A
review of theoretical perspectives on the comprehension of relative
clauses. Language and Linguistics Compass, 6, 403 415. http://dx.doi
.org/10.1002/lnc3.347
Gray, J. R., Chabris, C. F., & Braver, T. S. (2003). Neural mechanisms of
general fluid intelligence. Nature Neuroscience, 6, 316 322. http://dx
.doi.org/10.1038/nn1014
Gray, J. R., & Thompson, P. M. (2004). Neurobiology of intelligence:
Science and ethics. Nature Reviews Neuroscience, 5, 471 482. http://
dx.doi.org/10.1038/nrn1405
Hamilton, A. C., & Martin, R. C. (2005). Dissociations among tasks
involving inhibition: A single-case study. Cognitive, Affective & Behavioral Neuroscience, 5, 113. http://dx.doi.org/10.3758/CABN.5.1.1
Harrison, T. L., Shipstead, Z., Hicks, K. L., Hambrick, D. Z., Redick, T. S.,
& Engle, R. W. (2013). Working memory training may increase working
memory capacity but not fluid intelligence. Psychological Science, 24,
2409 2419. http://dx.doi.org/10.1177/0956797613492984
Holmes, J., Gathercole, S. E., & Dunning, D. L. (2009). Adaptive training
leads to sustained enhancement of poor working memory in children.
Developmental Science, 12, F9 F15. http://dx.doi.org/10.1111/j.14677687.2009.00848.x
Hsu, N. S., & Novick, J. M. (2016). Dynamic engagement of cognitive
control modulates recovery from misinterpretation during real-time language processing. Psychological Science, 27, 572582. http://dx.doi.org/
10.1177/0956797615625223
Hussey, E. K., & Novick, J. M. (2012). The benefits of executive control
training and the implications for language processing. Frontiers in
Psychology, 3, 158. http://dx.doi.org/10.3389/fpsyg.2012.00158
26
HUSSEY ET AL.
Hussey, E. K., Ward, N., Christianson, K., & Kramer, A. F. (2015).

Language and Memory Improvements following tDCS of Left Lateral
Prefrontal Cortex. PLoS ONE, 10, e0141417. http://dx.doi.org/10.1371/
journal.pone.0141417
Jaeger, T. F. (2008). Categorical data analysis: Away from ANOVAs
(transformation or not) and towards logit mixed models. Journal of
Memory and Language, 59, 434 446. http://dx.doi.org/10.1016/j.jml
.2007.11.007
Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Perrig, W. J. (2008). Improving fluid intelligence with training on working memory. PNAS
Proceedings of the National Academy of Sciences of the United States of
America, 105, 6829 6833. http://dx.doi.org/10.1073/pnas.0801268105
Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Shah, P. (2011). Short- and
long-term benefits of cognitive training. PNAS Proceedings of the National Academy of Sciences of the United States of America, 108,
1008110086. http://dx.doi.org/10.1073/pnas.1103228108
Jaeggi, S. M., Studer-Luethi, B., Buschkuehl, M., Su, Y.-F., Jonides, J., &
Perrig, W. J. (2010). The relationship between n-back performance and
matrix reasoning: Implications for training and transfer. Intelligence, 38,
625 635. http://dx.doi.org/10.1016/j.intell.2010.09.001
January, D., Trueswell, J. C., & Thompson-Schill, S. L. (2009). Colocalization of Stroop and syntactic ambiguity resolution in Brocas
area: Implications for the neural basis of sentence processing. Journal of
Cognitive Neuroscience, 21, 2434 2444. http://dx.doi.org/10.1162/jocn
.2008.21179
Jonides, J., & Nee, D. E. (2006). Brain mechanisms of proactive interference in working memory. Neuroscience, 139, 181193. http://dx.doi
.org/10.1016/j.neuroscience.2005.06.042
Jonides, J., Smith, E. E., Marshuetz, C., Koeppe, R. A., & Reuter-Lorenz,
P. A. (1998). Inhibition in verbal working memory revealed by brain
activation. PNAS Proceedings of the National Academy of Sciences of
the United States of America, 95, 8410 8413. http://dx.doi.org/10.1073/
pnas.95.14.8410
Kan, I. P., Teubner-Rhodes, S., Drummey, A. B., Nutile, L., Krupa, L., &
Novick, J. M. (2013). To adapt or not to adapt: The question of
domain-general cognitive control. Cognition, 129, 637 651. http://dx
.doi.org/10.1016/j.cognition.2013.09.001
Kan, I. P., & Thompson-Schill, S. L. (2004). Selection from perceptual and
conceptual representations. Cognitive, Affective & Behavioral Neuroscience, 4, 466 482. http://dx.doi.org/10.3758/CABN.4.4.466
Kane, M. J., Conway, A. R., Miura, T. K., & Colflesh, G. J. (2007).
Working memory, attention control, and the N-back task: A question of
construct validity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 615 622. http://dx.doi.org/10.1037/0278-7393
.33.3.615
Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the
control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental
Psychology: General, 132, 4770. http://dx.doi.org/10.1037/0096-3445
.132.1.47
Karbach, J., & Kray, J. (2009). How useful is executive control training?
Age differences in near and far transfer of task-switching training.
Developmental Science, 12, 978 990. http://dx.doi.org/10.1111/j.14677687.2009.00846.x
Karbach, J., Mang, S., & Kray, J. (2010). Transfer of task-switching
training in older age: The role of verbal processes. Psychology and
Aging, 25, 677 683. http://dx.doi.org/10.1037/a0019845
Karbach, J., Strobach, T., & Schubert, T. (2015). Adaptive workingmemory training benefits reading, but not mathematics in middle childhood. Child Neuropsychology, 21, 285301. http://dx.doi.org/10.1080/
09297049.2014.899336
Khanna, M. M., & Boland, J. E. (2010). Childrens use of language
context in lexical ambiguity resolution. The Quarterly Journal of
Experimental Psychology, 63, 160 193. http://dx.doi.org/10.1080/

17470210902866664
Klingberg, T. (2010). Training and plasticity of working memory. Trends
in Cognitive Sciences, 14, 317324. http://dx.doi.org/10.1016/j.tics.2010
.05.002
Klingberg, T., Fernell, E., Olesen, P. J., Johnson, M., Gustafsson, P.,
Dahlstrm, K., . . . Westerberg, H. (2005). Computerized training of
working memory in children with ADHDA randomized, controlled
trial. Journal of the American Academy of Child & Adolescent Psychiatry, 44, 177186. http://dx.doi.org/10.1097/00004583-20050200000010
Klingberg, T., Forssberg, H., & Westerberg, H. (2002). Training of working memory in children with ADHD. Journal of Clinical and Experimental Neuropsychology, 24, 781791. http://dx.doi.org/10.1076/jcen
.24.6.781.8395
Kueider, A. M., Parisi, J. M., Gross, A. L., & Rebok, G. W. (2012).
Computerized cognitive training with older adults: A systematic review.
PLoS ONE, 7, e40588. http://dx.doi.org/10.1371/journal.pone.0040588
Leek, M. R. (2001). Adaptive procedures in psychophysical research.
Perception & Psychophysics, 63, 1279 1292. http://dx.doi.org/10.3758/
BF03194543
Lewis, R. L., & Vasishth, S. (2005). An activation-based model of sentence
processing as skilled memory retrieval. Cognitive Science, 29, 375 419.
http://dx.doi.org/10.1207/s15516709cog0000_25
Linck, J. A., Schwieter, J. W., & Sunderman, G. (2012). Inhibitory control
predicts language switching performance in trilingual speech production.
Bilingualism: Language and Cognition, 15, 651 662. http://dx.doi.org/
10.1017/S136672891100054X
Lvdn, M., Bckman, L., Lindenberger, U., Schaefer, S., & Schmiedek, F.
(2010). A theoretical framework for the study of adult cognitive plasticity. Psychological Bulletin, 136, 659 676. http://dx.doi.org/10.1037/
a0020080
MacDonald, M. C., & Christiansen, M. H. (2002). Reassessing working
memory: Comment on Just and Carpenter (1992) and Waters and Caplan
(1996). Psychological Review, 109, 3554. http://dx.doi.org/10.1037/
0033-295X.109.1.35
MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The
lexical nature of syntactic ambiguity resolution. Psychological Review,
101, 676 703. http://dx.doi.org/10.1037/0033-295X.101.4.676
Martin, R. C., & Cheng, Y. (2006). Selection demands versus association
strength in the verb generation task. Psychonomic Bulletin & Review, 13,
396 401. http://dx.doi.org/10.3758/BF03193859
Mazuka, R., Jincho, N., & Oishi, H. (2009). Development of executive
control and language processing. Language and Linguistics Compass, 3,
59 89. http://dx.doi.org/10.1111/j.1749-818X.2008.00102.x
Melby-Lervg, M., & Hulme, C. (2013). Is working memory training
effective? A meta-analytic review. Developmental Psychology, 49, 270
291. http://dx.doi.org/10.1037/a0028228
Melby-Lervg, M., & Hulme, C. (2016). There is no convincing evidence
that working memory training is effective: A reply to Au et al. (2014)
and Karbach and Verhaeghen (2014). Psychonomic Bulletin & Review,
23, 324 330.
Meseguer, E., Carreiras, M., & Clifton, C., Jr. (2002). Overt reanalysis
strategies and eye movements during the reading of mild garden path
sentences. Memory & Cognition, 30, 551561. http://dx.doi.org/10
.3758/BF03194956
Milham, M. P., Banich, M. T., Webb, A., Barad, V., Cohen, N. J., Wszalek,
T., & Kramer, A. F. (2001). The relative involvement of anterior
cingulate and prefrontal cortex in attentional control depends on nature
of conflict. Cognitive Brain Research, 12, 467 473. http://dx.doi.org/
10.1016/S0926-6410(01)00076-3
Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal
cortex function. Annual Review of Neuroscience, 24, 167202. http://dx
.doi.org/10.1146/annurev.neuro.24.1.167

Mirman, D. (2014). Minding the brain. Available from http://
mindingthebrain.blogspot.com/2014/02/three-ways-to-get-parameterspecific-p.html
Miyake, A., & Friedman, N. P. (2012). The nature and organization of
individual differences in executive functions: Four general conclusions.
Current Directions in Psychological Science, 21, 8 14. http://dx.doi
.org/10.1177/0963721411429458
Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A.,
& Wager, T. D. (2000). The unity and diversity of executive functions
and their contributions to complex Frontal Lobe tasks: A latent variable analysis. Cognitive Psychology, 41, 49 100. http://dx.doi.org/10
.1006/cogp.1999.0734
Moreau, D., & Conway, A. R. A. (2014). The case for an ecological
approach to cognitive training. Trends in Cognitive Sciences, 18, 334
336. http://dx.doi.org/10.1016/j.tics.2014.03.009
Morrison, A. B., & Chein, J. M. (2011). Does working memory training
work? The promise and challenges of enhancing cognition by training
working memory. Psychonomic Bulletin & Review, 18, 46 60. http://
dx.doi.org/10.3758/s13423-010-0034-0
Munakata, Y., Herd, S. A., Chatham, C. H., Depue, B. E., Banich, M. T.,
& OReilly, R. C. (2011). A unified framework for inhibitory control.
Trends in Cognitive Sciences, 15, 453 459. http://dx.doi.org/10.1016/j
.tics.2011.07.011
Nee, D. E., Jonides, J., & Berman, M. G. (2007). Neural mechanisms of
proactive interference-resolution. NeuroImage, 38, 740 751. http://dx
.doi.org/10.1016/j.neuroimage.2007.07.066
Nelson, J. K., Reuter-Lorenz, P. A., Persson, J., Sylvester, C. Y., &
Jonides, J. (2009). Mapping interference resolution across task domains:
A shared control process in left inferior frontal gyrus. Brain Research,
1256, 92100. http://dx.doi.org/10.1016/j.brainres.2008.12.001
Nelson, J. K., Reuter-Lorenz, P. A., Sylvester, C. Y., Jonides, J., & Smith,
E. E. (2003). Dissociable neural mechanisms underlying response-based
and familiarity-based conflict in working memory. PNAS Proceedings of
the National Academy of Sciences of the United States of America, 100,
Neville, H. J., Stevens, C., Pakulak, E., Bell, T. A., Fanning, J., Klein, S.,
& Isbell, E. (2013). Family-based training program improves brain
function, cognition, and behavior in lower socioeconomic status preschoolers. PNAS Proceedings of the National Academy of Sciences of the
United States of America, 110, 12138 12143. http://dx.doi.org/10.1073/
pnas.1304437110
Nilsen, E. S., & Graham, S. A. (2009). The relations between childrens
communicative perspective-taking and executive functioning. Cognitive
Psychology, 58, 220 249. http://dx.doi.org/10.1016/j.cogpsych.2008.07
.002
Norman, D., & Shallice, T. (1986). Attention to action: Willed and automatic control of behavior. In R. Davidson, R. G. Schwartz, & D. Shapiro
(Eds.), Consciousness and self-regulation: Advances in research and
theory (pp. 118). New York, NY: Plenum Press. http://dx.doi.org/10
.1007/978-1-4757-0629-1_1
Novick, J. M., Hussey, E. K., Teubner-Rhodes, S. E., Harbison, J. I., &
Bunting, M. (2014). Clearing the garden path: Improving sentence
processing through cognitive control training. Language, Cognition, and
Neuroscience, 29, 186 217. http://dx.doi.org/10.1080/01690965.2012
.758297
Novick, J. M., Kan, I. P., Trueswell, J. C., & Thompson-Schill, S. L.
(2009). A case for conflict across multiple domains: Memory and
language impairments following damage to ventrolateral prefrontal cortex. Cognitive Neuropsychology, 26, 527567. http://dx.doi.org/10
.1080/02643290903519367
Novick, J. M., Trueswell, J. C., & Thompson-Schill, S. L. (2005). Cognitive control and parsing: Reexamining the role of Brocas area in
sentence comprehension. Cognitive, Affective & Behavioral Neuroscience, 5, 263281. http://dx.doi.org/10.3758/CABN.5.3.263
27
Novick, J. M., Trueswell, J. C., & Thompson-Schill, S. L. (2010). Brocas

area and language processing: Evidence for the cognitive control connection. Language and Linguistics Compass, 4, 906 924. http://dx.doi
.org/10.1111/j.1749-818X.2010.00244.x
Nozari, N., & Thompson-Schill, S. L. (2013). More attention when speaking: Does it help or does it hurt? Neuropsychologia, 51, 2770 2780.
http://dx.doi.org/10.1016/j.neuropsychologia.2013.08.019
Oberauer, K. (2005). Binding and inhibition in working memory: Individual and age differences in short-term recognition. Journal of Experimental Psychology: General, 134, 368 387. http://dx.doi.org/10.1037/00963445.134.3.368
Oberauer, K., & Lange, E. B. (2009). Activation and binding in verbal
working memory: A dual-process model for the recognition of nonwords. Cognitive Psychology, 58, 102136. http://dx.doi.org/10.1016/j
.cogpsych.2008.05.003
Oelhafen, S., Nikolaidis, A., Padovani, T., Blaser, D., Koenig, T., & Perrig,
W. J. (2013). Increased parietal activity after training of interference
control. Neuropsychologia, 51, 27812790. http://dx.doi.org/10.1016/j
.neuropsychologia.2013.08.012
Owen, A. M., McMillan, K. M., Laird, A. R., & Bullmore, E. (2005).
N-back working memory paradigm: A meta-analysis of normative functional neuroimaging studies. Human Brain Mapping, 25, 46 59. http://
dx.doi.org/10.1002/hbm.20131
Persson, J., Larsson, A., & Reuter-Lorenz, P. A. (2013). Imaging fatigue of
interference control reveals the neural basis of executive resource depletion. Journal of Cognitive Neuroscience, 25, 338 351. http://dx.doi
.org/10.1162/jocn_a_00321
Persson, J., Welsh, K. M., Jonides, J., & Reuter-Lorenz, P. A. (2007).
Cognitive fatigue of executive processes: Interaction between interference resolution tasks. Neuropsychologia, 45, 15711579. http://dx.doi
.org/10.1016/j.neuropsychologia.2006.12.007
Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M., & Raichle, M. E.
(1988). Positron emission tomographic studies of the cortical anatomy of
single-word processing. Nature, 331, 585589. http://dx.doi.org/10
.1038/331585a0
Rabagliati, H., Pylkknen, L., & Marcus, G. F. (2013). Top-down influence
in young childrens linguistic ambiguity resolution. Developmental Psychology, 49, 1076 1089. http://dx.doi.org/10.1037/a0026918
Rabipour, S., & Raz, A. (2012). Training the brain: Fact and fad in
cognitive and behavioral remediation. Brain and Cognition, 79, 159
179. http://dx.doi.org/10.1016/j.bandc.2012.02.006
Rajah, M. N., Ames, B., & DEsposito, M. (2008). Prefrontal contributions
to domain-general executive control processes during temporal context
retrieval. Neuropsychologia, 46, 1088 1103. http://dx.doi.org/10.1016/
j.neuropsychologia.2007.10.023
Rapport, M. D., Orban, S. A., Kofler, M. J., & Friedman, L. M. (2013). Do
programs designed to train working memory, other executive functions,
and attention benefit children with ADHD? A meta-analytic review of
cognitive, academic, and behavioral outcomes. Clinical Psychology Review, 33, 12371252. http://dx.doi.org/10.1016/j.cpr.2013.08.005
Rayner, K. (1998). Eye movements in reading and information processing:
20 years of research. Psychological Bulletin, 124, 372 422. http://dx
.doi.org/10.1037/0033-2909.124.3.372
R Development Core Team. (2015). R: A language and environment for
statistical computing. R Foundation for Statistical Computing, Vienna,
Austria. Retrieved from http://www.R-project.org
Redick, T. S., Shipstead, Z., Harrison, T. L., Hicks, K. L., Fried, D. E.,
Hambrick, D. Z., . . . Engle, R. W. (2013). No evidence of intelligence
improvement after working memory training: A randomized, placebocontrolled study. Journal of Experimental Psychology: General, 142,
359 379. http://dx.doi.org/10.1037/a0029082
Robinson, G., Blair, J., & Cipolotti, L. (1998). Dynamic aphasia: An
inability to select between competing verbal responses? Brain: A Journal of Neurology, 121, 77 89. http://dx.doi.org/10.1093/brain/121.1.77
28
HUSSEY ET AL.
Robinson, G., Shallice, T., & Cipolotti, L. (2005). A failure of high

level verbal response selection in progressive dynamic aphasia. Cognitive Neuropsychology, 22, 661 694. http://dx.doi.org/10.1080/
02643290442000239
Schneiders, J. A., Opitz, B., Krick, C. M., & Mecklinger, A. (2011).
Separating intra-modal and across-modal training effects in visual working memory: An fMRI investigation. Cerebral Cortex, 21, 25552564.
http://dx.doi.org/10.1093/cercor/bhr037
Schnur, T. T., Schwartz, M. F., Kimberg, D. Y., Hirshorn, E., Coslett,
H. B., & Thompson-Schill, S. L. (2009). Localizing interference during
naming: Convergent neuroimaging and neuropsychological evidence for
the function of Brocas area. PNAS Proceedings of the National Academy of Sciences of the United States of America, 106, 322327. http://
dx.doi.org/10.1073/pnas.0805874106
Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human
information processing: II. perceptual learning, automatic attending and
a general theory. Psychological Review, 84, 127190. http://dx.doi.org/
10.1037/0033-295X.84.2.127
Shimamura, A. P. (2000). The role of the prefrontal cortex in dynamic
filtering. Psychobiology, 28, 207218.
Shipstead, Z., Hicks, K. L., & Engle, R. W. (2012a). Working memory
training remains a work in progress. Journal of Applied Research in
Memory & Cognition, 1, 217219. http://dx.doi.org/10.1016/j.jarmac
.2012.07.009
Shipstead, Z., Hicks, K. L., & Engle, R. W. (2012b). Cogmed working
memory training: Does the evidence support the claims? Journal of
Applied Research in Memory & Cognition, 1, 185193. http://dx.doi
.org/10.1016/j.jarmac.2012.06.003
Shipstead, Z., Redick, T. S., & Engle, R. W. (2010). Does working memory
training generalize? Psychologica Belgica, 50, 245276. http://dx.doi
.org/10.5334/pb-50-3-4-245
Shipstead, Z., Redick, T. S., & Engle, R. W. (2012). Is working memory
training effective? Psychological Bulletin, 138, 628 654. http://dx.doi
.org/10.1037/a0027473
Snyder, H. R., Banich, M. T., & Munakata, Y. (2011). Choosing our words:
Retrieval and selection processes recruit shared neural substrates in left
ventrolateral prefrontal cortex. Journal of Cognitive Neuroscience, 23,
3470 3482.
Snyder, H. R., Banich, M. T., & Munakata, Y. (2014). All competition is
not alike: Neural mechanisms for resolving underdetermined and prepotent competition. Journal of Cognitive Neuroscience, 26, 2608 2623.
http://dx.doi.org/10.1162/jocn_a_00652
Snyder, H. R., Hutchison, N., Nyhus, E., Curran, T., Banich, M. T.,
OReilly, R. C., & Munakata, Y. (2010). Neural inhibition enables
selection during language processing. PNAS Proceedings of the National
Academy of Sciences of the United States of America, 107, 16483
Snyder, H. R., & Munakata, Y. (2008). So many options, so little time: The
roles of association and competition in underdetermined responding.
Psychonomic Bulletin & Review, 15, 10831088. http://dx.doi.org/10
.3758/PBR.15.6.1083
Sprenger, A. M., Atkins, S. M., Bolger, D. J., Harbison, J. I., Novick, J. M.,
Chrabaszcz, J. S., . . . Dougherty, M. R. (2013). Training working
memory: Limits of transfer. Intelligence, 41, 638 663. http://dx.doi.org/
10.1016/j.intell.2013.07.013
Sternberg, R. J. (2008). Increasing fluid intelligence is possible after all.
PNAS Proceedings of the National Academy of Sciences of the United
States of America, 105, 6791 6792. http://dx.doi.org/10.1073/pnas
.0803396105
Stine-Morrow, E. A. L., Payne, B. R., Roberts, B. W., Kramer, A. F.,
Morrow, D. G., Payne, L., . . . Parisi, J. M. (2014). Training versus
engagement as paths to cognitive enrichment with aging. Psychology
and Aging, 29, 891906. http://dx.doi.org/10.1037/a0038244
Sturt, P. (2007). Semantic re-interpretation and garden path recovery.

Cognition, 105, 477 488. http://dx.doi.org/10.1016/j.cognition.2006.10
.009
Szmalec, A., Verbruggen, F., Vandierendonck, A., & Kemps, E. (2011).
Control of interference during working memory updating. Journal of
Experimental Psychology: Human Perception and Performance, 37,
137151. http://dx.doi.org/10.1037/a0020365
Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy,
J. C. (1995). Integration of visual and linguistic information in spoken
language comprehension. Science, 268, 16321634. http://dx.doi.org/10
.1126/science.7777863
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, 213231. http://dx.doi.org/10.1016/j.cognition.2016.02
.011
Thompson, T. W., Waskom, M. L., Garel, K. L., Cardenas-Iniguez, C.,
Reynolds, G. O., Winter, R., . . . Gabrieli, J. D. (2013). Failure of
working memory training to enhance cognition or intelligence. PLoS
ONE, 8, e63614. http://dx.doi.org/10.1371/journal.pone.0063614
Thompson-Schill, S. L., Bedny, M., & Goldberg, R. F. (2005). The frontal
lobes and the regulation of mental activity. Current Opinion in Neurobiology, 15, 219 224. http://dx.doi.org/10.1016/j.conb.2005.03.006
Thompson-Schill, S. L., & Botvinick, M. M. (2006). Resolving conflict: A
response to Martin and Cheng (2006). Psychonomic Bulletin & Review,
13, 402 408. http://dx.doi.org/10.3758/BF03193860
Thompson-Schill, S. L., DEsposito, M., Aguirre, G. K., & Farah, M. J.
(1997). Role of left inferior prefrontal cortex in retrieval of semantic
knowledge: A reevaluation. PNAS Proceedings of the National Academy
of Sciences of the United States of America, 94, 1479214797. http://dx
.doi.org/10.1073/pnas.94.26.14792
Thompson-Schill, S. L., Jonides, J., Marshuetz, C., Smith, E. E.,
DEsposito, M., Kan, I. P., . . . Swick, D. (2002). Effects of frontal lobe
damage on interference effects in working memory. Cognitive, Affective
& Behavioral Neuroscience, 2, 109 120. http://dx.doi.org/10.3758/
CABN.2.2.109
Thompson-Schill, S. L., Swick, D., Farah, M. J., DEsposito, M., Kan, I. P.,
& Knight, R. T. (1998). Verb generation in patients with focal frontal
lesions: A neuropsychological test of neuroimaging findings. PNAS
Proceedings of the National Academy of Sciences of the United States of
America, 95, 1585515860. http://dx.doi.org/10.1073/pnas.95.26.15855
Thothathiri, M., Kimberg, D. Y., & Schwartz, M. F. (2012). The neural
basis of reversible sentence comprehension: Evidence from voxel-based
lesion symptom mapping in aphasia. Journal of Cognitive Neuroscience,
24, 212222. http://dx.doi.org/10.1162/jocn_a_00118
Thothathiri, M., Schwartz, M. F., & Thompson-Schill, S. L. (2010). Selection for position: The role of left ventrolateral prefrontal cortex in
sequencing language. Brain and Language, 113, 28 38. http://dx.doi
.org/10.1016/j.bandl.2010.01.002
Trueswell, J. C., Sekerina, I., Hill, N. M., & Logrip, M. L. (1999). The
kindergarten-path effect: Studying on-line sentence processing in young
children. Cognition, 73, 89 134. http://dx.doi.org/10.1016/S00100277(99)00032-3
Trueswell, J. C., Tanenhaus, M. K., & Garnsey, S. (1994). Semantic
influences on parsing: Use of thematic role information in syntactic
ambiguity resolution. Journal of Memory and Language, 33, 285318.
http://dx.doi.org/10.1006/jmla.1994.1014
Trueswell, J. C., Tanenhaus, M. K., & Kello, C. (1993). Verb-specific
constraints in sentence processing: Separating effects of lexical preference from garden-paths. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 19, 528 553. http://dx.doi.org/10.1037/02787393.19.3.528
Van Dyke, J. A., & Lewis, R. L. (2003). Distinguishing effects of structure
and decay on attachment and repair: A cue-based parsing account of
29

recovery from misanalyzed ambiguities. Journal of Memory and Language, 49, 285316. http://dx.doi.org/10.1016/S0749-596X(03)00081-0
von Bastian, C. C., & Eschen, A. (2016). Does working memory training
have to be adaptive? Psychological Research, 80, 181194. http://dx.doi
.org/10.1007/s00426-015-0655-z
von Bastian, C. C., & Oberauer, K. (2013). Distinct transfer effects of training
different facets of working memory capacity. Journal of Memory and
Language, 69, 36 58. http://dx.doi.org/10.1016/j.jml.2013.02.002
von der Malsburg, T., & Vasishth, S. (2013). Scanpaths reveal syntactic
underspecification and reanalysis strategies. Language and Cognitive Processes, 28, 15451578. http://dx.doi.org/10.1080/01690965.2012.728232
Vuong, L. C., & Martin, R. C. (2011). LIFG-based attentional control and
the resolution of lexical ambiguities in sentence context. Brain and
Language, 116, 2232. http://dx.doi.org/10.1016/j.bandl.2010.09.012
Vuong, L. C., & Martin, R. C. (2014). Domain-specific executive control
and the revision of misinterpretations in sentence comprehension. Language, Cognition and Neuroscience, 29, 312325. http://dx.doi.org/10
.1080/01690965.2013.836231
Wagner, A. D., Par-Blagoev, E. J., Clark, J., & Poldrack, R. A. (2001).
Recovering meaning: Left prefrontal cortex guides controlled semantic
retrieval. Neuron, 31, 329 338. http://dx.doi.org/10.1016/S08966273(01)00359-2

Waris, O., Soveri, A., & Laine, M. (2015). Transfer after working memory
updating training. PLoS ONE, 10, e0138734. http://dx.doi.org/10.1371/
journal.pone.0138734
Wells, J. B., Christiansen, M. H., Race, D. S., Acheson, D. J., & MacDonald, M. C. (2009). Experience and sentence processing: Statistical learning and relative clause comprehension. Cognitive Psychology, 58, 250
271. http://dx.doi.org/10.1016/j.cogpsych.2008.08.002
Woodard, K., Pozzan, L., & Trueswell, J. C. (2016). Taking your own path:
Individual differences in executive function and language processing
skills in child learners. Journal of Experimental Child Psychology, 141,
187209. http://dx.doi.org/10.1016/j.jecp.2015.08.005
Ye, Z., & Zhou, X. (2009). Conflict control during sentence comprehension: FMRI evidence. NeuroImage, 48, 280 290. http://dx.doi.org/10
.1016/j.neuroimage.2009.06.032
Zelinski, E. M., Peters, K. D., Hindin, S., Petway, K. T., II, & Kennison,
R. F. (2014). Evaluating the relationship between change in performance
on training tasks and on untrained outcomes. Frontiers in Human
Neuroscience, 8, 617. http://dx.doi.org/10.3389/fnhum.2014.00617
Appendix
Tables of Means and Model Coefficients
Table A1
Descriptive Statistics for Pre/Post Measures by Task Condition, Training Group, and Assessment
High-conflict group
Pretest
Condition
Recognition task response time (in ms)

Global Filler
624
Global Target
612
Local Filler
723
Local Lure
1024
Local Target
778
Stroop differential response time (in ms)
Facilitation effect
37
Interference effect
75
Verb generation response time (in ms)
High competition/High association
2961
High competition/Low association
4262
Low competition/High association
2583
Low competition/Low association
3712
Garden-path comprehension question accuracy
Ambiguous
.74
Unambiguous
.93
Note.
Low-conflict group
Posttest
Pretest
3-back group
Posttest
Pretest
Posttest
SD
SD
SD
SD
SD
SD
138
123
112
266
129
558
562
639
840
691
117
108
94
208
108
640
641
716
978
781
128
112
112
180
140
571
577
655
893
724
73
72
100
191
127
634
629
730
1023
797
88
89
102
215
136
573
581
682
886
731
94
104
115
196
138
28
70
29
52
31
48
35
79
32
90
35
51
39
75
30
88
38
69
31
58
24
60
3625
4729
4037
4865
2091
3286
1707
2517
2263
4339
1541
1636
2024
3768
1614
2943
1054
5192
614
2806
1849
2545
1401
2531
1113
1765
560
2969
2301
3902
1849
3227
967
2335
821
1813
1931
2782
1457
2276
833
1337
548
1021
.23
.08
.85
.92
.24
.12
.65
.92
.29
.08
.80
.89
.27
.13
.68
.88
.24
.13
.81
.91
.23
.11
M ! mean; SD ! standard deviation; ms ! milliseconds.
(Appendix continues)
30
HUSSEY ET AL.
Table A2
Descriptive Statistics for the Garden-Path Eye Movement Measures by Sentence Condition (Ambiguous vs. Unambiguous), Sentence
Region, Training Group, and Assessment
High-conflict group
Pretest
Condition
Region
Posttest
SD
Garden-path regression-path time (in ms)

Ambiguous
1
655
211
Ambiguous
2
399
172
Ambiguous
3
1032
486
Ambiguous
4
2735
1579
Unambiguous
1
226
82
Unambiguous
2
879
283
Unambiguous
3
597
275
Unambiguous
4
1794
1100
Garden-path second-pass time (length-corrected time
Ambiguous
1
216
348
Ambiguous
2
154
237
Ambiguous
3
179
477
Ambiguous
4
#56
321
Unambiguous
1
#55
264
Unambiguous
2
#181
356
Unambiguous
3
#150
238
Unambiguous
4
#139
350
Note.
Low-conflict group
Pretest
3-back group
Posttest
Pretest
Posttest
SD
SD
SD
SD
SD
619
433
855
1950
251
922
628
1604
in ms)
#13
11
#120
#96
#114
#105
#149
#171
242
191
302
1010
97
367
231
837
655
425
927
2096
249
947
717
1618
224
266
307
840
134
238
329
693
649
393
904
1892
211
864
637
1441
212
161
243
1122
54
225
176
753
664
436
1074
2518
256
963
652
2063
199
153
505
1380
94
192
227
906
626
349
954
1837
246
901
741
1222
181
108
278
902
75
215
253
601
198
109
229
184
192
332
212
266
#7
152
#50
#147
#37
#44
#46
#272
260
182
300
263
138
308
171
248
59
76
10
#60
#67
#156
#162
#151
170
126
289
189
181
415
194
182
75
115
126
#124
#105
#118
#39
#78
385
240
451
185
138
359
244
328
#12
79
#14
#182
#31
#83
#117
#228
231
174
298
168
252
432
217
412
Table A3
Descriptive Statistics for the Relative Clause Eye Movement Measures by Sentence Condition (Object- vs. Subject-Extracted),
Sentence Region, Training Group, and Assessment
High-conflict group
Pretest
Condition
Region
Posttest
SD
Relative clause first-pass time (length-corrected time in ms)

Object-extracted
1
#96
62
#114
Object-extracted
2
181
226
246
Object-extracted
3
#20
125
#6
Object-extracted
4
#91
88
#85
Subject-extracted
1
#114
79
#79
Subject-extracted
2
84
153
117
Subject-extracted
3
43
122
56
Subject-extracted
4
#87
87
#91
Relative clause second-pass time (length-corrected time in ms)
Object-extracted
1
#32
228
#110
Object-extracted
2
157
393
182
Object-extracted
3
113
259
30
Object-extracted
4
#188
206
#234
Subject-extracted
1
#31
201
#98
Subject-extracted
2
#5
339
1
Subject-extracted
3
#29
209
32
Subject-extracted
4
#191
219
#154
Note.
Low-conflict group
Pretest
3-back group
Posttest
Pretest
Posttest
SD
SD
SD
SD
SD
65
228
111
81
82
173
111
83
#95
141
#42
#95
#121
10
2
#71
80
202
139
99
67
113
84
97
#98
245
#4
#74
#101
41
38
#107
64
176
89
68
62
164
147
121
#133
296
15
#82
#129
72
40
#44
81
181
147
93
109
105
198
133
#115
230
#23
#77
#117
#2
63
#67
91
151
128
95
79
131
146
79
172
441
253
195
149
271
245
256
#3
240
#26
#152
#72
16
#5
#170
194
309
139
175
174
238
206
131
#66
108
#3
#151
#117
#18
#61
#173
92
332
233
134
96
336
202
145
#44
280
119
#173
#176
#73
#54
#141
201
675
317
171
137
302
170
167
#80
77
77
#165
#130
#7
#50
#136
100
428
216
201
99
295
191
265
31
Table A4
Estimated Coefficients From Linear Mixed-Effects Models for the Posttest N-Back Task
Fixed effects
Predictor
Total accuracy
(Intercept)
Adaptive training
N-level
Adaptive training " N-level
Lure accuracy
(Intercept)
Conflict training
N-level
Conflict training ' N-level
Target accuracy
(Intercept)
Conflict training
N-level
Conflict training " N-level
Discriminability (D)
(Intercept)
Conflict training
N-level
Criterion (C)
(Intercept)
Conflict training
N-level
Random effects by
subject variance
Coefficient
SE
t value
.88
.06
!.08
.06
.01
.02
.01
.02
105.04!
3.48!
!8.46!
3.17!
.0024
.90
!.06
#.04
#.03
.02
.02
.02
.03
48.25!
!2.34!
#1.46
#.80
.0047
.78
.05
!.21
.12
.02
.04
.03
.05
44.00!
1.45
!8.39!
2.36!
.0074
2.52
.09
!.64
.35
.06
.13
.09
.18
39.25!
.68
!7.15!
1.94!
.1050
.49
!.10
.28
!.17
.02
.04
.03
.07
22.58!
!2.29!
8.33!
!2.52!
0.0095
Note. The conflict training contrast corresponds to the comparison of the high- and low-conflict groups, and the adaptive
training contrast corresponds to the comparison of the low-conflict and 3-back groups. Bold indicates coefficients that are
significant as given by Kenward-Rogers approximations. SE ! standard error.
!
Significant at the p $ .05 level.
32
HUSSEY ET AL.
Table A5
Estimated Coefficients From Linear Mixed-Effects Models for the Recognition Memory Task
Fixed effects
Predictor
Local recognition (Lures)
Intercept
Conflict training
Adaptive training
Assessment
Assessment " Conflict training
Assessment ' Adaptive training
Local recognition (Targets)
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Local recognition (Fillers)
Intercept
Conflict training
Adaptive training
Assessment
Global recognition (Targets)
Intercept
Conflict training
Adaptive training
Assessment
Global recognition (Fillers)
Intercept
Conflict training
Adaptive training
Assessment
Random effects by
subject variance
Coefficient
SE
t value
935.51
#3.93
18.85
84.28
99.79
52.00
43.94
56.72
57.97
33.59
43.36
44.32
21.29!
#.07
.33
2.51!
2.30!
1.17
32966
752.16
#17.88
11.94
56.73
30.51
8.59
26.55
34.28
35.03
20.09
25.93
26.50
28.33!
#.52
.34
2.82!
1.18
.32
12048
685.24
#4.08
20.82
60.88
23.58
#12.10
21.21
27.38
27.98
18.88
24.37
24.91
32.31!
#.15
.74
3.22!
.97
#.49
7215
608.90
#21.74
#3.93
64.00
#14.28
#16.76
20.34
26.26
26.84
20.59
26.59
27.17
29.93!
#.83
#.15
3.11!
#.54
#.62
6157
605.63
#14.60
#2.58
68.35
#2.63
#7.33
21.73
28.05
28.67
21.07
27.20
27.79
27.87!
#.52
#.09
3.24!
#.10
#.26
7224
!
Significant at the p $ .05 level.
33
Table A6
Estimated Coefficients From Linear Mixed-Effects Models for the Verb Generation Task
Fixed effects
Predictor
Coefficient
High competition/Low association response time

Intercept
Conflict training
Adaptive training
Assessment
High competition/High association response time
Intercept
Conflict training
Adaptive training
Assessment
Low competition/Low association response time
Intercept
Conflict training
Adaptive training
Assessment
Low competition/High association response time
Intercept
Conflict training
Adaptive training
Assessment
Random effects by
subject variance
SE
t value
2827
928
347
1011
59
343
469
701
675
457
657
635
6.02!
1.32
.51
2.21!
.09
.54
6890879
1821
595
197
177
694
319
252
377
363
201
288
279
7.23!
1.58
.54
.88
2.41!
1.14
2185968
2549
418
37
323
868
370
351
523
504
383
551
532
7.26!
.80
.07
.84
1.58
.70
3554983
1465
603
109
168
672
175
232
345
332
260
375
363
6.32!
1.75
.33
.65
1.79
.48
1509557
Marginal at the p $ .10 level. ! Significant at the p $ .05 level.
Table A7
Estimated Coefficients From Linear Mixed-Effects Models for the Stroop Task
Fixed effects
Predictor
Interference score
Intercept
Conflict training
Adaptive training
Assessment
Facilitation score
Intercept
Conflict training
Adaptive training
Assessment
Random effects by
subject variable
Coefficient
SE
t value
64.67
#1.60
8.14
27.78
#3.29
2.56
11.69
15.57
15.95
15.81
21.12
21.64
5.53!
#.10
.51
1.76
#.16
.12
1704
34.95
#1.67
#4.86
.89
7.46
#1.87
5.43
7.24
7.41
7.39
9.88
10.12
6.43!
#.23
#.66
.12
.75
#.18
364
34
HUSSEY ET AL.
Table A8
Estimated Coefficients From Generalized Linear Mixed Models of Comprehension Accuracy Following
Garden-Path Sentences
Fixed effects
Predictor
Ambiguous sentence accuracy
Intercept
Conflict training
Adaptive training
Assessment
Unambiguous sentence accuracy
Intercept
Conflict training
Adaptive training
Assessment
Random effects
By subject
variance
By item
variance
4.10!
1.27
.09
4.51!
#.44
#.71
2.3787
.5528
9.05!
.62
#.54
#.11
.21
.81
.8175
1.7273
Coefficient
SE
z value
1.53
.61
.05
1.16
#.15
#.25
.37
.48
.49
.26
.34
.35
3.22
.22
#.20
#.04
.10
.41
.36
.36
.36
.39
.50
.51
.1891
!
significant at the p $ .05 level.
35
Table A9
Estimated Coefficients From Linear Mixed-Effects Models for the Eye Movement Measures for Garden-Path Sentences
Fixed effects
Predictor
Coefficient
Ambiguous sentence regression-path time in region 4

Intercept
2262.56
Conflict training
324.64
Adaptive training
150.03
Assessment
335.58
476.97
322.64
Unambiguous sentence regression-path time in region 4
Intercept
1604.57
Conflict training
157.92
Adaptive training
176.42
Assessment
107.44
59.89
449.01
Ambiguous sentence second-pass time in region 1
Intercept
358.58
Conflict training
62.96
Adaptive training
#5.81
Assessment
77.40
Region length
23.46
284.16
86.19
Unambiguous sentence second-pass time in region 1
Intercept
71.44
Conflict training
34.17
Adaptive training
19.49
Assessment
130.55
Region length
26.70
#72.34
Assessment " Adaptive training
!203.08
Intercept
277.72
Conflict training
52.21
Adaptive training
34.95
Assessment
52.81
Region length
25.20
104.95
27.76
Intercept
217.29
Conflict training
69.72
Adaptive training
43.32
Assessment
119.29
Region length
20.42
#65.77
#64.04
Intercept
306.48
Conflict training
99.47
Adaptive training
45.83
Assessment
41.63
Region length
22.86
251.74
236.45
Intercept
187.46
Conflict training
53.44
Adaptive training
69.61
Assessment
107.18
Region length
17.88
#132.69
#1.63
Random effects
By subject
variance
By item
variance
SE
t value
250.61
306.85
326.58
212.39
266.86
301.94
9.03!
1.06
.46
1.58
1.79
1.07
775823
211518
151.32
181.02
192.70
171.82
214.25
240.32
10.60!
.87
.92
.63
.28
1.87
219444
129381
231.59
120.05
130.35
95.42
9.29
119.53
134.09
1.55
.52
#.04
.81
2.53!
2.38!
.64
23373
34825
106.37
44.72
48.10
53.19
9.51
65.58
74.93
.67
.76
.41
2.45!
2.81!
#1.10
!2.71!
26917
115.19
55.30
59.36
61.50
10.30
77.34
85.27
2.41!
.94
.59
.86
2.45!
1.36
.33
15067
266.39
88.20
94.72
93.50
9.60
119.16
130.93
.82
.79
.46
1.28
2.13!
#.55
#.49
308963
318.09
119.35
127.36
101.92
11.09
126.69
139.59
.96
.83
.36
.41
2.06!
1.99!
1.69
121
101.26
54.85
57.78
68.70
5.38
86.30
92.69
1.85
.97
1.20
1.56
3.32!
#1.54
#.02
79491
6114
8645
422
3192
463
572
3663
4059
4338
524
8874
7012
697
24960
125
Note. The conflict training contrast corresponds to the comparison of the high- and low-conflict groups, and the adaptive training contrast corresponds
to the comparison of the low-conflict and 3-back groups. Bold indicates coefficients that are significant as given by Kenward-Rogers approximations. SE !
standard error.
36
HUSSEY ET AL.
Table A10
Estimated Coefficients From Linear Mixed-Effects Models for the Eye Movement Measures for Relative Clause Sentences
Fixed effects
Predictor
Object-Extracted First-Pass Time in Region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Subject-extracted first-pass time in region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Assessment " Adaptive Training
Object-extracted second-pass time in region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Subject-extracted second-pass time in region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Random effects
t value
By subject
variance
By item
variance
2629
Coefficient
SE
308.88
#12.72
36.24
!99.89
#3.90
69.68
147.04
144.12
35.10
37.68
48.68
5.65
61.80
67.89
2.14!
#.36
.96
!2.05!
#.69
1.13
2.17!
21883
116.88
73.70
19.99
#64.81
#2.93
39.82
112.03
124.23
29.29
31.32
41.70
4.83
52.55
57.13
.94
2.52!
.64
#1.55
#.61
.76
1.96!
104
818.50
#22.83
29.99
56.01
#20.93
#123.75
222.16
364.62
114.58
122.03
103.18
13.69
131.49
146.44
2.24!
#.20
.25
.54
#1.53
#.94
1.52
603024
#340.12
54.00
#5.17
176.21
16.66
#127.52
!265.20
309.49
69.43
73.06
94.79
12.21
118.35
127.09
#1.10
.78
#.07
1.86
1.37
#1.08
!2.09!
432214
65
2634
25816
363
19633
966
Note. The conflict training contrast corresponds to the comparison of the high- and low-conflict groups, and the adaptive training contrast corresponds
to the comparison of the low-conflict and 3-back groups. Bold indicates coefficients that are significant as given by Kenward-Rogers approximations. SE !
standard error.
Received December 4, 2015

Revision received February 11, 2016
Accepted March 10, 2016 !

tmpC814 TMP

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

tmpC814 TMP

Uploaded by

Copyright:

Available Formats

Journal of Experimental Psychology:

Learning, Memory, and Cognition

2016 American Psychological Association

Memory and Language Improvements Following Cognitive

University of Maryland, College Park and University of Illinois

University of Maryland, College Park

University of Maryland, College Park and Medical University of

University of Maryland, College Park

University of Maryland, College Park and Yale University

University of Maryland, College Park

While observing and interacting with the environment, people

activity is known as cognitive control, a top-down procedure that

Erika K. Hussey, Center for Advanced Study of Language and Program

This material is based on work supported, in whole or in part, with

2007; Botvinick, Braver, Barch, Carter, & Cohen, 2001; Braver,

Process-Specific Approach to Training

activities demanding cognitive control (Bialystok et al., 2006;

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

conflict resolution demands. Note that process-specificity differs

When Conflict Arises

2005; Novick, Kan, Trueswell, & Thompson-Schill, 2009;

Preliminaries to the Current Study

other (low-conflict) training tasks were a necessary component of

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

one reported taking medications to correct problems related to

One hundred sixteen healthy native-English-speaking subjects

trials previously. Single letters were displayed serially for 500ms

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

feature aimed at attenuating response bias inconsistencies across

Pre/Post Transfer Task Assessments and Predictions

Competition (or conflict) effects, indexed by RTs and VLPFC

interference scores (Incongruent minus Neutral) to assess conflict

The jewelry that was elegant and expensive sparkled

Sentence [1] is temporarily ambiguous because the verb hid can

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

The farmer who the expert questioned promoted the

The farmer who questioned the expert promoted the

Object relative clauses (italicized in 3) are known to generate

This (single) dissociation then offers an ideal test bed to compare

mixed-effects model evaluating the fixed effects of Training

Statistical Analyses on Assessment Measures

separately for each training group; then, Session 1 performance

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

Results of Posttest N-Back Task: Manipulation Check

Figure 3. Posttest n-back task accuracy. (A) Adaptivity manipulation

Results of Transfer Assessment Measures

Measure and condition

Recognition memory task

Verb generation task

Relative clause processing

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

Verb generation task. Data were excluded from six subjects

pretest Stroop performance (ts $ 0.27, ps % 0.787), suggesting that

elegant and expensive/sparkled brightly for Ambiguous items and

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

.001). Accuracy to comprehension questions following ambiguous

sentences in content and position, respectively. That is, the conflict

1.36, p ! .17). Adaptive Training only interacted with Assessment

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

account of the pre/post improvements for the High-Conflict group

First-pass time: Training effects. As illustrated in Figure 8,

COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY

and individual differences studies in both children and adults