Professional Documents
Culture Documents
J. Isaiah Harbison
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.
Kayla Velnoskey
Jared M. Novick
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
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.
HUSSEY ET AL.
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.
& 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,
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.
HUSSEY ET AL.
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.
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
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
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
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.
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).
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.
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,
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.
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)
While the thief hid the jewelry that was elegant and
expensive sparkled brightly. (Temporarily Ambiguous/
High-Conflict)
2.
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.
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.
4.
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.
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.
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.
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.
12
HUSSEY ET AL.
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.
(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
Conflict training
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
Stroop task
Garden-path recovery
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.
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.
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.
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
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.
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.
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.
15
HUSSEY ET AL.
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.
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).
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
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.
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
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.
18
HUSSEY ET AL.
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
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.
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
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).
(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),
HUSSEY ET AL.
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.
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
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).
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
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.
21
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.
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.
22
HUSSEY ET AL.
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.
23
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.
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.
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
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.
25
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.
26
HUSSEY ET AL.
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.
27
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.
28
HUSSEY ET AL.
29
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.
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
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
(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
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.
Condition
Region
Posttest
SD
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
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
(Appendix continues)
31
Table A4
Estimated Coefficients From Linear Mixed-Effects Models for the Posttest N-Back Task
Fixed effects
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.
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
Conflict training " N-level
Criterion (C)
(Intercept)
Conflict training
N-level
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.
(Appendix continues)
32
HUSSEY ET AL.
Table A5
Estimated Coefficients From Linear Mixed-Effects Models for the Recognition Memory Task
Fixed effects
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.
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
Assessment ' Adaptive training
Local recognition (Fillers)
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
Global recognition (Targets)
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
Global recognition (Fillers)
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
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
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.
(Appendix continues)
33
Table A6
Estimated Coefficients From Linear Mixed-Effects Models for the Verb Generation Task
Fixed effects
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.
Predictor
Coefficient
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
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.
Table A7
Estimated Coefficients From Linear Mixed-Effects Models for the Stroop Task
Fixed effects
Predictor
Interference score
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
Facilitation score
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
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
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.
(Appendix continues)
34
HUSSEY ET AL.
Table A8
Estimated Coefficients From Generalized Linear Mixed Models of Comprehension Accuracy Following
Garden-Path Sentences
Fixed effects
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.
Predictor
Ambiguous sentence accuracy
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
Unambiguous sentence accuracy
Intercept
Conflict training
Adaptive training
Assessment
Assessment ' Conflict training
Assessment ' Adaptive training
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
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.
(Appendix continues)
35
Table A9
Estimated Coefficients From Linear Mixed-Effects Models for the Eye Movement Measures for Garden-Path Sentences
Fixed effects
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.
Predictor
Coefficient
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.
(Appendix continues)
36
HUSSEY ET AL.
Table A10
Estimated Coefficients From Linear Mixed-Effects Models for the Eye Movement Measures for Relative Clause Sentences
Fixed effects
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.
Predictor
Object-Extracted First-Pass Time in Region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Assessment ' Conflict training
Assessment " Adaptive training
Subject-extracted first-pass time in region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Assessment ' Conflict training
Assessment " Adaptive Training
Object-extracted second-pass time in region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Assessment ' Conflict training
Assessment ' Adaptive training
Subject-extracted second-pass time in region 2
Intercept
Conflict training
Adaptive training
Assessment
Region length
Assessment ' Conflict training
Assessment " Adaptive training
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.