Journal of Experimental Psychology: Learning, Memory, and Cognition 2017, Vol. 43, No. 1, 23–58 © 2016 American Psychological Association 0278-7393/17/$12.00 http://dx.doi.org/10.1037/xlm0000283 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. 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 University of Maryland, College Park and Medical University of South Carolina University of Maryland, College Park Kayla Velnoskey Jared M. Novick University of Maryland, College Park and Yale University University of Maryland, College Park 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 people’s 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, This article was published Online First July 14, 2016. 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 Otolaryngology—Head 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 Hussey’s 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 23 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. 24 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 one’s 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 people’s 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-Lervåg & 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, Bäckman, & 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 information—regardless 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, & D’Esposito, 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 (Akçay & 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-Lervåg & 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 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. 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 what’s 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, 25 & 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, Pylkkänen, & Marcus, 2013; del Río 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, D’Esposito, 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 (D’Esposito, 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. 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 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 group’s transfer effects—though specific to only the highconflict conditions of the untrained reading task and localized to the conflict regions of the ambiguous sentences themselves—were 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 changed—the 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 tasks—those that adjust task difficulty based on an individual’s real-time performance—may 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 (Lövden, Bäckman, 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 ways—those who practice conflict processing versus those who do not—and 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- 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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 27 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 28 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 responses—20 —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. 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 1–13 depending on an individual’s 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY feature aimed at attenuating response bias inconsistencies across sequences with variable n-levels. Performance-adaptivity was controlled by adjusting n according to a subject’s 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 sequence’s 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 29 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. 30 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 on—and 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 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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 difficult—and 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). 31 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 Eyelink’s 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 32 HUSSEY ET AL. mixed-effects model evaluating the fixed effects of Training Group (High-Conflict vs. Low-Conflict) and Session (1–16) 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. 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. 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 (2–16) 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 COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 33 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. 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. ⴱ Significant at the p ⬍ .05 level. HUSSEY ET AL. 34 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. 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 Recognition memory task Verb generation task Stroop task Garden-path recovery Relative clause processing Measure and condition 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 ⴱ 3-back training ⴱ † † ⴱ † ⴱ ⴱ ⴱ ⴱ 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. ⴱ 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 35 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. ⴱSignificant 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 group’s 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 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. 36 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. ⴱSignificant 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 ⬍ 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY .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 ⫽ 785 ms vs. MLC ⫽ 204 ms). 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 37 sentences in content and position, respectively. That is, the conflict manipulation had selective effects, theoretically attributable to the High-Conflict group’s 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 group’s 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. 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. 38 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 group’s 58-ms improvement (decrease) alongside the Low-Conflict group’s 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 39 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY Figure 7. Length-corrected second-pass reading time for (A) ambiguous and (B) unambiguous sentences. Error bars represent ⫾ 1 standard error of the mean. ⴱSignificant 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. 40 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 group’s 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 41 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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. ⴱSignificant 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), 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. 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. 42 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. ⴱSignificant 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). 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 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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 people’s 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 group’s 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 43 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 that—if 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. 44 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 different—prepotent 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 conflict—that comprehension has gone awry—initiates 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 identify—as 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 training—yielded 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 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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 group’s 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 45 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 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. 46 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-Lervåg & 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; Lövdén, Bäckman, 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 tasks—all previously established to involve cognitive control across multiple experimental findings—would 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 transient—we 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 Akçay, Ç., & 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-Lervåg and Hulme (2015). Psychonomic Bulletin & Review, 23, 331–337. 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, 2883–2901. 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, 255–278. 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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., & Bäckman, 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., O’Reilly, 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, 205–238. 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, 317–334. http://dx.doi.org/10.1016/S0749596X(03)00070-6 Dahlin, E., Neely, A. S., Larsson, A., Bäckman, 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 Río, D., Maestú, F., López-Higes, R., Moratti, S., Gutiérrez, R., Maestú, 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, 382–391. http://dx.doi.org/10.1016/j.neuropsychologia.2010.12.005 D’Esposito, 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, 47 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, 541–553. 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, 3–20. 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, 1–76. 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, 1–13. 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, 572–582. 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 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. 48 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, 10081–10086. 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 Broca’s 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, 181–193. 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, 47–70. 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, 285–301. http://dx.doi.org/10.1080/ 09297049.2014.899336 Khanna, M. M., & Boland, J. E. (2010). Children’s 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, 317–324. http://dx.doi.org/10.1016/j.tics.2010 .05.002 Klingberg, T., Fernell, E., Olesen, P. J., Johnson, M., Gustafsson, P., Dahlström, K., . . . Westerberg, H. (2005). Computerized training of working memory in children with ADHD—A randomized, controlled trial. Journal of the American Academy of Child & Adolescent Psychiatry, 44, 177–186. 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, 781–791. 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 Lövdén, M., Bäckman, 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, 35–54. 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-Lervåg, 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-Lervåg, 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, 551–561. 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, 167–202. http://dx .doi.org/10.1146/annurev.neuro.24.1.167 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 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., & O’Reilly, 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, 92–100. 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, 11171–11175. http://dx.doi.org/10.1073/pnas.1334125100 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 children’s 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. 1–18). 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, 527–567. 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 Broca’s area in sentence comprehension. Cognitive, Affective & Behavioral Neuroscience, 5, 263–281. http://dx.doi.org/10.3758/CABN.5.3.263 49 Novick, J. M., Trueswell, J. C., & Thompson-Schill, S. L. (2010). Broca’s 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, 102–136. 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, 2781–2790. 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, 1571–1579. 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, 585–589. http://dx.doi.org/10 .1038/331585a0 Rabagliati, H., Pylkkänen, L., & Marcus, G. F. (2013). Top-down influence in young children’s 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., & D’Esposito, 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, 1237–1252. 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 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. 50 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, 2555–2564. 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 Broca’s area. PNAS Proceedings of the National Academy of Sciences of the United States of America, 106, 322–327. 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, 127–190. 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, 207–218. 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, 217–219. 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, 185–193. 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, 245–276. 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., O’Reilly, 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– 16488. http://dx.doi.org/10.1073/pnas.1002291107 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, 1083–1088. 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, 891–906. 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, 137–151. 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, 1632–1634. 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, 213–231. 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., D’Esposito, 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, 14792–14797. http://dx .doi.org/10.1073/pnas.94.26.14792 Thompson-Schill, S. L., Jonides, J., Marshuetz, C., Smith, E. E., D’Esposito, 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., D’Esposito, 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, 15855–15860. 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, 212–222. 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, 285–318. 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 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. COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY recovery from misanalyzed ambiguities. Journal of Memory and Language, 49, 285–316. 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, 181–194. 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, 1545–1578. 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, 22–32. 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, 312–325. http://dx.doi.org/10 .1080/01690965.2013.836231 Wagner, A. D., Paré-Blagoev, E. J., Clark, J., & Poldrack, R. A. (2001). Recovering meaning: Left prefrontal cortex guides controlled semantic 51 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, 187–209. 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 M 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 M SD M SD M SD M SD M 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) HUSSEY ET AL. 52 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 M 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 M SD M SD M SD M SD M 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 M ⫽ mean; SD ⫽ standard deviation; ms ⫽ milliseconds. 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 M Posttest SD M 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 M SD M SD M SD M 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 M ⫽ mean; SD ⫽ standard deviation; ms ⫽ milliseconds. (Appendix continues) COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 53 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 ⴛ N-level Lure accuracy (Intercept) Conflict training N-level Conflict training ⫻ N-level Target accuracy (Intercept) Conflict training N-level Conflict training ⴛ N-level Discriminability (D’) (Intercept) Conflict training N-level Conflict training ⴛ N-level Criterion (C) (Intercept) Conflict training N-level Conflict training ⴛ N-level Random effects by subject variance Coefficient SE t value .88 .06 ⴚ.08 .06 .01 .02 .01 .02 105.04ⴱ 3.48ⴱ ⴚ8.46ⴱ 3.17ⴱ .0024 .90 ⴚ.06 ⫺.04 ⫺.03 .02 .02 .02 .03 48.25ⴱ ⴚ2.34ⴱ ⫺1.46 ⫺.80 .0047 .78 .05 ⴚ.21 .12 .02 .04 .03 .05 44.00ⴱ 1.45 ⴚ8.39ⴱ 2.36ⴱ .0074 2.52 .09 ⴚ.64 .35 .06 .13 .09 .18 39.25ⴱ .68 ⴚ7.15ⴱ 1.94ⴱ .1050 .49 ⴚ.10 .28 ⴚ.17 .02 .04 .03 .07 22.58ⴱ ⴚ2.29ⴱ 8.33ⴱ ⴚ2.52ⴱ 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. ⴱ Significant at the p ⬍ .05 level. (Appendix continues) HUSSEY ET AL. 54 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 ⴛ 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 21.29ⴱ ⫺.07 .33 2.51ⴱ 2.30ⴱ 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 28.33ⴱ ⫺.52 .34 2.82ⴱ 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 32.31ⴱ ⫺.15 .74 3.22ⴱ .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 29.93ⴱ ⫺.83 ⫺.15 3.11ⴱ ⫺.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 27.87ⴱ ⫺.52 ⫺.09 3.24ⴱ ⫺.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. ⴱ Significant at the p ⬍ .05 level. (Appendix continues) COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 55 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 High competition/Low association response time Intercept Conflict training Adaptive training Assessment Assessment ⫻ Conflict training Assessment ⫻ Adaptive training High competition/High association response time Intercept Conflict training Adaptive training Assessment Assessment ⴛ Conflict training Assessment ⫻ Adaptive training Low competition/Low association response time Intercept Conflict training Adaptive training Assessment Assessment ⫻ Conflict training Assessment ⫻ Adaptive training Low competition/High association response time Intercept Conflict training Adaptive training Assessment Assessment ⫻ Conflict training Assessment ⫻ Adaptive training Random effects by subject variance SE t value 2827 928 347 1011 59 343 469 701 675 457 657 635 6.02ⴱ 1.32 .51 2.21ⴱ .09 .54 6890879 1821 595 197 177 694 319 252 377 363 201 288 279 7.23ⴱ 1.58 .54 .88 2.41ⴱ 1.14 2185968 2549 418 37 323 868 370 351 523 504 383 551 532 7.26ⴱ .80 .07 .84 1.58 .70 3554983 1465 603 109 168 672 175 232 345 332 260 375 363 6.32ⴱ 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. † Marginal at the p ⬍ .10 level. ⴱ 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 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 5.53ⴱ ⫺.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 6.43ⴱ ⫺.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. † Marginal at the p ⬍ .10 level. ⴱ Significant at the p ⬍ .05 level. (Appendix continues) HUSSEY ET AL. 56 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 4.10ⴱ 1.27 .09 4.51ⴱ ⫺.44 ⫺.71 2.3787 .5528 9.05ⴱ .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. ⴱ significant at the p ⬍ .05 level. (Appendix continues) COGNITIVE CONTROL TRAINING FOR LANGUAGE AND MEMORY 57 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 Ambiguous sentence regression-path time in region 4 Intercept 2262.56 Conflict training 324.64 Adaptive training 150.03 Assessment 335.58 Assessment ⫻ Conflict training 476.97 Assessment ⫻ Adaptive training 322.64 Unambiguous sentence regression-path time in region 4 Intercept 1604.57 Conflict training 157.92 Adaptive training 176.42 Assessment 107.44 Assessment ⫻ Conflict training 59.89 Assessment ⫻ Adaptive training 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 Assessment ⴛ Conflict training 284.16 Assessment ⫻ Adaptive training 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 Assessment ⫻ Conflict training ⫺72.34 Assessment ⴛ Adaptive training ⴚ203.08 Ambiguous sentence second-pass time in region 2 Intercept 277.72 Conflict training 52.21 Adaptive training 34.95 Assessment 52.81 Region length 25.20 Assessment ⫻ Conflict training 104.95 Assessment ⫻ Adaptive training 27.76 Unambiguous sentence second-pass time in region 2 Intercept 217.29 Conflict training 69.72 Adaptive training 43.32 Assessment 119.29 Region length 20.42 Assessment ⫻ Conflict training ⫺65.77 Assessment ⫻ Adaptive training ⫺64.04 Ambiguous sentence second-pass time in region 3 Intercept 306.48 Conflict training 99.47 Adaptive training 45.83 Assessment 41.63 Region length 22.86 Assessment ⴛ Conflict training 251.74 Assessment ⫻ Adaptive training 236.45 Unambiguous sentence second-pass time in region 3 Intercept 187.46 Conflict training 53.44 Adaptive training 69.61 Assessment 107.18 Region length 17.88 Assessment ⫻ Conflict training ⫺132.69 Assessment ⫻ Adaptive training ⫺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 9.03ⴱ 1.06 .46 1.58 1.79† 1.07 775823 211518 151.32 181.02 192.70 171.82 214.25 240.32 10.60ⴱ .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 2.53ⴱ 2.38ⴱ .64 23373 34825 106.37 44.72 48.10 53.19 9.51 65.58 74.93 .67 .76 .41 2.45ⴱ 2.81ⴱ ⫺1.10 ⴚ2.71ⴱ 26917 115.19 55.30 59.36 61.50 10.30 77.34 85.27 2.41ⴱ .94 .59 .86 2.45ⴱ 1.36 .33 15067 266.39 88.20 94.72 93.50 9.60 119.16 130.93 .82 .79 .46 1.28 2.13ⴱ ⫺.55 ⫺.49 308963 318.09 119.35 127.36 101.92 11.09 126.69 139.59 .96 .83 .36 .41 2.06ⴱ 1.99ⴱ 1.69 121 101.26 54.85 57.78 68.70 5.38 86.30 92.69 1.85† .97 1.20 1.56 3.32ⴱ ⫺1.54 ⫺.02 79491 6114 8645 422 3192 463 3663 4059 572 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. † Marginal at the p ⬍ .10 level. ⴱ Significant at the p ⬍ .05 level. (Appendix continues) HUSSEY ET AL. 58 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 ⴛ Adaptive training Subject-extracted first-pass time in region 2 Intercept Conflict training Adaptive training Assessment Region length Assessment ⫻ Conflict training Assessment ⴛ 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 ⴛ Adaptive training Random effects t value By subject variance By item variance 2629 Coefficient SE 308.88 ⫺12.72 36.24 ⴚ99.89 ⫺3.90 69.68 147.04 144.12 35.10 37.68 48.68 5.65 61.80 67.89 2.14ⴱ ⫺.36 .96 ⴚ2.05ⴱ ⫺.69 1.13 2.17ⴱ 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 2.52ⴱ .64 ⫺1.55 ⫺.61 .76 1.96ⴱ 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 2.24ⴱ ⫺.20 .25 .54 ⫺1.53 ⫺.94 1.52 603024 ⫺340.12 54.00 ⫺5.17 176.21 16.66 ⫺127.52 ⴚ265.20 309.49 69.43 73.06 94.79 12.21 118.35 127.09 ⫺1.10 .78 ⫺.07 1.86† 1.37 ⫺1.08 ⴚ2.09ⴱ 432214 65 2634 2 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. † Marginal at the p ⬍ .10 level. ⴱ Significant at the p ⬍ .05 level. Received December 4, 2015 Revision received February 11, 2016 Accepted March 10, 2016 䡲