Cogn Affect Behav Neurosci (2014) 14:1356–1374 DOI 10.3758/s13415-014-0276-9
Building tasks from verbal instructions: An EEG study on practice trial exposure and task structure complexity during novel sequences of behavior Gareth Roberts & Timothy W. Jones & Elizabeth A. Davis & Trang T. Ly & Mike Anderson
Published online: 6 May 2014 # Psychonomic Society, Inc. 2014
Abstract Configuring the mind to perform a novel task is an effortful process and one that is related to differences in general intelligence. Previous research has suggested that when participants are given instructions for a future task, representations of the rules contained in the instructions can influence subsequent behavior, even when the rules are not necessary to perform the upcoming task. One hypothesis for the continued activation of rule representations suggests that the practice trials participants perform before the experimental trials may instantiate the unnecessary task rules into participants’ mental model of the task (i.e., the task space). To test this hypothesis, EEGs were recorded as participants (N = 66) completed a multirule task designed to contrast the effects of increasing task structure complexity and practice trial exposure. The results showed that, as was predicted, performance is significantly poorer when more task rules are specified in the task instructions. Practice trials with the extra rule did not affect task performance, indicating that an unacted verbal instruction is sufficient to incorporate the rule into participants’ mental model of the task. The EEG results showed that instruction complexity was linked to a phasic increase in frontal theta synchronization but reduced posterior alpha and beta desynchronization. These changes in synchronization occurred during a time period of low intertrial phase coherence and suggest that participants were “checking the task G. Roberts (*) : M. Anderson School of Psychology and Exercise Science, Murdoch University, 90 South St, Murdoch, Perth, Western Australia 6150, Australia e-mail:
[email protected] G. Roberts : M. Anderson Neurocognitive Development Unit, School of Psychology, University of Western Australia, Perth, Western Australia, Australia T. W. Jones : E. A. Davis : T. T. Ly Department of Endocrinology & Diabetes, Princess Margaret Hospital for Children, Perth, Western Australia, Australia
rules” amidst a trial. This transient neural activity may reflect compensatory mechanisms for dealing with increased mindwandering that is more likely to occur in complex tasks. Keywords Frontal midline theta . General intelligence . Goal neglect . Practice trials . Task exposure . Task instructions The ability to use verbal instructions to guide novel sequences of behavior is a uniquely human behavior and one that has led to significant advances in human society and culture (Cole, Bagic, Kass, & Schneider, 2010; Wenke, Gaschler, Nattkemper, & Frensch, 2009). There is an extensive literature on isolated elements of cognitive control, including how humans and nonhuman primates switch between different stimulus–response (S–R) mappings (Rogers & Monsell, 1995) and withhold contextually inappropriate response tendencies (Friedman & Miyake, 2004). These elements of control have had significant implications for clinical, applied, and educational research (Royall et al., 2002); however, in isolation, they do not address how complex goal-directed problems are decomposed and solved (Duncan, 2010). Recently, research has begun to address more central questions of cognitive control, such as how humans prepare their mind for a task they have not encountered before (Power & Petersen, 2013). This process has been selectively referred to as rapid instructed task learning (Cole et al., 2010b), problem space construction (Borst, Taatgen, & van Rijn, 2010), mental program construction (Duncan, 2010), instructional control (Huang, Hazy, Herd, & O’Reilly, 2013), and instruction-based learning (Ruge & Wolfensteller, 2013). In terms of evolutionary theory, the ability to configure the mind to perform an essentially infinite configuration of possible tasks from verbal instruction has a high survival value (Cole, Laurent, & Stocco, 2012). Moreover, besides being a uniquely human behavior that is central to our intelligence as a
Cogn Affect Behav Neurosci (2014) 14:1356–1374
species, the ability to perform novel tasks from instruction also shows high individual differences between humans (Duncan, 2010; Duncan et al., 2008). Therefore, research on the mechanisms behind novel task performance provides an opportunity to bridge broad areas of cognitive neuroscience. In this article, we investigate how neural oscillations support novel task construction and how this process differs between individuals. To successfully perform any given task, mental operations must be selected and organized for an infinite number of possible tasks and corresponding mental processes. We reason that this process of task configuration should be reflected in the ongoing neural oscillations occurring in the brain, and hence, in this study, we use electroencephalography (EEG) to analyze event-related changes in the power and phase of oscillations elicited to a complex multirule experimental paradigm. We investigate individual differences in this process and its relationship to intelligence, because the abstract generality of a task configuration process make it an ideal candidate for a cognitive process that underlies individual differences in general intelligence (Spearman, 1904), which is indicated by covariation in performance among diverse tests of ability, irrespective of their content. Data from functional neuroimaging techniques with high spatial resolution, such as functional magnetic resonance imaging (fMRI) and positron emission topography (PET), have demonstrated the existence of a highly connected cortical network that extends over the prefrontal and parietal cortex (Cole, Pathak, & Schneider, 2010; Duncan, 2010; Woolgar, Hampshire, Thompson, & Duncan, 2011; Woolgar, Thompson, Bor, & Duncan, 2011). The network is selectively referred to as the cognitive control network (Cole & Schneider, 2007), the task-positive network (Fox et al., 2005), the core task-set network (Dosenbach et al., 2006), and the multiple demand system (Duncan, 2010). Relevant to the present study is that the network shows increased activation during diverse tests of cognitive control (Dosenbach et al., 2006), performance on fluid intelligence tests (Bishop, Fossella, Croucher, & Duncan, 2008), and when participants prepare for and perform a novel task (Dumontheil, Thompson, & Duncan, 2010). For simplicity and congruence with the theoretical framework of this article, we adopt Duncan’s (2010) anatomical definition of the network, which consists of the cortex in and around the posterior part of the inferior frontal sulcus, in the anterior insula and adjacent frontal operculum, in the presupplementary motor area (pre-SMA) and dorsal anterior cingulate cortex (dACC), and in the intraparietal sulcus (IPS). Research has shown that these brains regions show an increased bold oxygenation level dependent (BOLD) response when participants receive new task instructions, prepare for the S–R mappings specified in the task instructions, and perform the S–R mappings in behavior (Dumontheil et al., 2010). Despite the plethora of neuroimaging data showing the network’s involvement in broad areas of control, there is
1357
ambiguity with regard to the specific processing roles of regions within the network. Recent voxel-based lesionmapping studies have provided causal evidence for the functional role of these regions in cognitive control yet, in themselves, have not clarified the issue (Glascher et al., 2010; Glascher et al., 2009; Woolgar et al., 2010). For example, Woolgar et al. (2010) showed that cortical damage to any part of the frontoparietal system is associated with a greater loss of IQ, as compared with regions outside the network; however, the differential deficit between regions in the network was relatively minimal. Influential accounts of cognitive control have typically prescribed central roles to the prefrontal cortex (PFC), which serves to bias processing in task-relevant areas and suppress processing of task-irrelevant information (Miller & Cohen, 2001). Since the discovery of and paradigm shift for analyzing networks of the brain (Sporns, 2011), many researchers now view cognitive control as being orchestrated in a distributed manner through the frontoparietal cortex. Further support for a distributed or network-oriented view of cognitive control comes from studies showing that the frontoparietal network is anticorrelated with the default mode network (DMN; Fox et al., 2005). The DMN is thought to reflect internal, retrospective thought and has been typically associated with the brain’s resting state and when participants engage in “mind wandering.” The DMN consists of regions in the medial temporal lobe, medial PFC, posterior cingulate cortex, and ventral precuneus (Raichle & Schneider, 2007). Historically, one of the defining research topics of neurology, neuropsychology, and cognitive neuroscience has been the functional role of the frontal lobes in controlling behavior (Fuster, 2001). Solving this problem has been deceptively difficult due to variability in lesion location, variability in intelligence prior to injury, and variability in compensatory plasticity changes following the brain insult. However, recent methodological advances and theoretical considerations have clarified the role, highlighting how specific regions of the frontal lobes, such as the lateral PFC (lPFC), play a particular important role in cognitive control, task complexity, and the implementation of verbal instructions into S–R mappings (e.g., Hartstra, Kühn, Verguts, & Brass, 2011). Moreover, the lPFC is thought to be central to psychological constructs such as working memory and executive function (Collette et al., 2005; Kane & Engle, 2002). These constructs are widely used in broad areas of cognitive neuroscience and are ultimately concerned with the cognitive processes that guide a person’s behavior toward a desired outcome, such as enhancing task-relevant information and suppressing task-irrelevant (and possibly interfering) information. Importantly, the lPFC is thought to play a central role in metacognitive processes, such as planning and orchestrating a complex multistep sequence of events, an idea that resonates with the historical accounts of Luria (1971, 1980).
1358
In agreement with neuroimaging evidence, case studies of damage to the frontal lobes characterize the behavior exhibited by frontal patients as contextually inappropriate, particularly in situations when the task requirements are novel and goals must be maintained and integrated over time (Duncan, 1986). When engaging in a novel situation, patients with lPFC damage show difficulty in organizing the task requirements into a mental representation that can successfully guide behavior—that is, a task model (Duncan et al., 2008). Behavioral evidence suggests that task models differ from consciously reportable descriptions of task requirements, since patients with frontal lobe damage can often articulate what they should be doing and under what conditions, despite failing to perform the appropriate action in behavior (Duncan, Emslie, Williams, Johnson, & Freer, 1996)—a phenomenon referred to as goal neglect (Duncan, 1995). In a series of experiments, Duncan et al. (2008) found that goal neglect also occurs in the normal population and that a person’s level of general intelligence is closely linked to the occurrence of this behavior–knowledge mismatch. This study was one of the first to highlight the connection between successful goal-directed behavior (i.e., measured by executive functioning tests that assess the control of action) and general intelligence (i.e., measured by fluid intelligence tests that assess a person’s ability to solve novel problems). The connection between these concepts is analogous to the artificial intelligence notion that solving a problem can be conceptualized as a search through problem spaces (Newell & Simon, 1972); similarly, performing a novel task can be conceptualized as a search through task spaces (Cole et al., 2012). A recent investigation of goal neglect (Duncan et al., 2008) reported an informative finding: Goal neglect was not due to demands imposed at real-time execution, such as increased visual search demand or the number of behavioral alternatives to be consideredduring a single trial or trial block; rather, the key factor modulating goal neglect was the number of task rules specified in the task instructions. Importantly, this effect was found in participants in the normal population and was highly correlated with individual differences in intelligence. The experimental task used in the experiment is from Duncan et al. (2008, Experiment 4) and is shown in Fig. 1. On each trial, a pair of digits, surrounded by a pair of colored shapes, was shown centrally on a computer screen. Participants were divided into two groups. One group was given instructions for two tasks (hereafter referred to as the 4-rules condition), one task for digits with surrounding shapes and the other task for digits with no surrounding shapes. Although the participants received instructions and practice for both tasks, the main experimental blocks contained only trials with surrounding shapes, and participants were explicitly told that they would not see any digits without surrounding shapes. The other group of participants (referred to as the 3-rules condition)
Cogn Affect Behav Neurosci (2014) 14:1356–1374
Fig. 1 Diagrammatic representation of the feature match task. Each pair of shapes constitutes a single trial. The shapes were displayed for 1 s, followed by a blank screen for 1 s. Participants responded to the larger number when the shapes matched on one stimulus dimension, but not if the shapes matched on two dimensions or on no dimensions. The trials without shapes surrounding the digits were the extra task rule that was not needed for the experimental trials
was never instructed about the trials with no surrounding shapes. The only difference between the two groups, therefore, was that the 4-rules group had received a more complex set of task rules during instruction and in the practice trials. The task performed by both groups in the experimental trials was exactly the same. However, the results showed that the 4rules participants were much more likely to neglect the rules for the trials with surrounding shapes. Interestingly, the relationship between goal neglect and general intelligence was highest for participants in the 4-rules condition, suggesting a strong relationship between task model construction and general intelligence. A recent behavioral study has extended this result by demonstrating that the relationship between working memory capacity and fluid intelligence is strongest in experimental paradigms that require the maintenance of multiple task rules (Duncan, Schramm, Thompson, & Dumontheil, 2012). By testing the same sample of participants on a diverse battery of working memory tasks, Duncan et al. (2012) showed that performance at tasks requiring the maintenance of multiple rules correlates higher with general intelligence than do paradigms involving the maintenance of items of a specific modality, such as verbal or visual span tasks. These paradigms show modest and reliable correlations with intelligence (see Ackerman, Beier, & Boyle, 2005).
Cogn Affect Behav Neurosci (2014) 14:1356–1374
The effect of instruction complexity adds a new dimension to our conceptual understanding of general intelligence. Although task complexity has often been linked to intelligence (Marshalek, Lohman, & Snow, 1983; Stankov, 2000), real-time execution demands have typically been confounded with the demands of task organization. For example, instructing participants to complete a task with multiple rules and contingencies has usually been followed with participants actually performing the task, not just in its mental preparation. This result is the somewhat unremarkable finding that the more complex a task is, the greater its correlation with intelligence (Marshalek et al., 1983). The result of Duncan et al. (2008) argues, by contrast, that it is the complexity of mental representation that predicts intelligence even when real-time task demands are held constant. The experiments by Duncan et al. (2008), however, did not distinguish whether this “instruction effect” was due to task exposure or, more specifically, initial practice trials with the unnecessary task requirement. For example, in the feature match task, participants in the 4-rules condition were told of two tasks: one for numbers and one for numbers with shapes surrounding them. Participants in this condition received practice trials with both tasks before the real “experimental” trials began. It is a major goal of this study to test whether performing the task consistent with the extra instruction (in an appropriate practice condition) is a necessary precondition for obtaining the extrainstruction effect. The finding by Duncan et al. (2008) that verbal instructions are incorporated into a participant’s task control structure (or into active task sets that influence behavior) is similar to experimental work conducted by Cohen-Kdoshay and Meiran (2009) on Woodworth’s (1938) prepared reflex (PR) metaphor. The PR metaphor suggests that task stimuli can reflexively trigger the corresponding action on the basis of the instructed or planned S–R information, even without any prior practice. In a series of modified flanker paradigms, Cohen-Kdoshay and Meiran demonstrated the existence of the flanker compatibility effect (FCE) on the very first trial following S–R instructions. By carefully controlling the stimuli and distractor items, the authors were able to generate multiple stimulus sets to assess performance on a novel first trial where no long-term memory traces between the stimuli and responses would have occurred. The authors reasoned that the retrieval of long-term memory traces may be the reason for the autonomous response activation seen in the FCE and that, if this account was correct, the FCE would be absent on the first trial but present afterward. However, in contrast to this long-term memory account, the authors found the FCE present on the first trial immediately after the task instructions, thus providing strong support for the PR metaphor. Functional neuroimaging data have demonstrated a clear role for the frontoparietal cortex in the assembly and control of
1359
new mental programs (Duncan, 2010). However, it is clear from the different connectivity characteristics of the regions (Cole et al., 2012; Kelly et al., 2009) that regions of the network may have different functional roles in the construction and maintenance of mental programs. A pressing example is the functional role of the dACC, which has been associated with “general goal control” processes in cognitive architectures (Anderson et al., 2004) but, more influentially, has been viewed as a conflict- or performance-monitoring center of the brain (Botvinick, Braver, Barch, Carter, & Cohen, 2001). Despite these different perspectives, there is evidence that these accounts can be reconciled. For example, large-scale meta-analyses of collated fMRI data sets have shown that the dACC displays start-cue, sustained, and error-related neural activation, a result consistent with both control- and conflict-monitoring perspectives (Dosenbach et al., 2006; Power & Peterson, 2013). EEG is applicable to investigating the functional role of the dACC due to electrophysiological signatures recorded on the scalp that have been localized to the dACC (Carter & van Veen, 2007). Previous research has predominately focused on phase-locked ERP components associated with novelty (Wang, Ulbert, Schomer, Marinkovic, & Halgren, 2005), response conflict (Hämmerer, Li, Müller, & Lindenberger, 2010), semantic conflict (Zurrón, Pouso, Lindín, Galdo, & Díaz, 2009), action mistakes (Larson, Clayson, & Baldwin, 2012), and performance feedback (van de Vijver, Ridderinkhof, & Cohen, 2011). The main ERP components associated with the dACC are the error-related negativity (ERN), feedback-related negativity, frontal-central N200, and frontal-central N400. These components all have similar scalp topography and are associated with performance-monitoring processes locked in time to a specific event (e.g., making a response). For example, the ERN is elicited when participants make a conscious error in speeded-response tasks, and its amplitude has been shown to correlate with the degree of posterror slowing shown on the subsequent trial (Maier, Yeung, & Steinhauser, 2011). These findings have been interpreted as strong support for the conflictmonitoring account of dACC function. However, a recent proposal suggests that frontal midline theta (FMT) oscillations may be the lingua franca to tie these related action-monitoring activities together, due to all these ERP components showing both high phase-locking and increases in power within the theta frequency band near the time they occur (Cavanagh, Zambrano-Vazquez, & Allen, 2011). The term FMT was introduced by Ishihara and Yoshi (1972), who used a set of strenuous mental tests such as continuous arithmetic calculations and found an increased occurrence of EEG theta activity (centered around 6.5 Hz) that was maximal over frontal sites when participants were actively engaged in the tests. Numerous cognitive operations have been shown to modulate FMT power, but it is most readily observable during tasks that involve sustained,
1360
internally directed cognition (Hsieh & Ranaganath, 2013). In particular, FMT has been implicated in cognitive control processes (Sauseng, Griesmayr, Freunberger, & Klimesch, 2010). FMT is modulated by manipulations of cognitive load such as calculation difficulty (Inanaga, 1998), number of items to be maintained in working memory (Onton, Delorme, & Makeig, 2005), and overall “task difficulty” (Mitchell, McNaughton, Flanagan, & Kirk, 2008; Nigbur, Ivanova, & Stürmer, 2011). EEG source imaging studies suggest that FMT originates from the medial PFC—specifically, the dACC and pre-SMA (M. X. Cohen, 2011; Luu, Tucker, & Makeig, 2004; Onton et al., 2005)—a result that has been confirmed by intracranial measurements from the human medial PFC (Wang et al., 2005) and multimodal neuroimaging studies combining EEG and PET (Pizzagalli, Oakes, & Davidson, 2003). In addition, FMT is strongly correlated with the suppression of the DMN and increased activity of the MD system network. For example, activation in the DMN is suppressed with greater working memory demand during go and no-go trials (Barber, Caffo, Pekar, & Mostofsky, 2013). In addition, studies that have recorded simultaneous EEG and fMRI data show that FMT power is negatively correlated with the BOLD signal in the DMN (e.g., Scheeringa et al., 2009).
The present study This study has two main aims. First, we wanted to investigate whether practice with an irrelevant task instruction was necessary for the extra task requirement to negatively affect subsequent performance. Would performance be unaffected when participants were told about an extra task and then told to ignore it, in the absence of exposure to the stimuli? Second, we wanted to investigate the effects of practice and task model complexity on ongoing neural oscillations, especially those showing induced changes in power. We focused on oscillatory activity in theta (4–8 Hz), alpha (8–12 Hz), and beta (13– 30 Hz) bands. These oscillatory bands are associated with broad action control processes (Cavanagh et al., 2011), cortical inhibition (Klimesch, 2012), and maintenance of active task sets (Engel & Fries, 2010), respectively. By analyzing changes in EEG power and coherence across individual trials, we would be able to analyze neural activity that was locked in phase (and would be observable in the ERPs), but also activity that was not locked in phase and, correspondingly, would be obscured in the ERP data. By contrasting these two types of neural activity, we attempted to determine whether task structure complexity reflects a sustained tonic increase in oscillatory activity or whether it reflects a phasic “burst” in oscillatory activity during situations that involve the detection and resolution of conflict or in the recovery from a lapse in attention. One
Cogn Affect Behav Neurosci (2014) 14:1356–1374
hypothesis from the behavioral literature is that goal neglect reflects an attentional lapse due to an inability to inhibit task-unrelated thoughts—a phenomenon analogous to mindwandering in normal participants (McVay & Kane, 2009). Functional neuroimaging studies using fMRI connectivity analyses have shown a decoupling of the frontoparietal network and activation of the DMN, as participants report task-unrelated thoughts (Mason et al., 2007). Therefore, by analyzing EEG signatures associated with increasing task complexity, it may be possible to determine whether task structure complexity reflects a consistent load over the experiment (i.e., a tonic increase) or, rather, increased variability or fluctuations in attentional performance (i.e., a phasic increase). In the present study, we administered an extended version of Duncan’s feature match task while recording EEGs and incorporating an additional manipulation of practice trials. Behaviorally, it was hypothesized that (1) participants given more task rules would show poorer performance and a higher incidence of goal neglect and (2) the number of rules participants received practice with would modulate the effect of instruction on decreased task performance. A final aim of the study was to test an additional hypothesis that task performance would be correlated with IQ (as assessed by the Wechsler Adult Intelligence Scales), particularly in the 4rules condition. Electrophysiologically, it was hypothesized that (1) participants given more task rules would show a greater phasic increase in FMT power, as compared with participants given fewer task rules; (2) increases in FMT power would be modulated by practice with more task rules; and (3) FMT activity would have a suggested source location in or near the dACC and pre-SMA.
Method Participants The study was part of a research project conducted at the Neurocognitive Development Unit at the University of Western Australia, in conjunction with Princess Margaret Hospital, Perth, on the effects of type 1 diabetes and brain development. In total, there were 66 participants, with 49.2 % being a sample of convenience (type 1 diabetics). From the point of view of this study, nondiabetics and diabetic participants were equivalent samples; they did not differ in full-scale IQ (FSIQ), were not differentially affected by the experimental manipulations, and were equally distributed among all conditions, albeit with some subtle differences between the two samples (see Ly, Anderson, McNamara, Davis, & Jones, 2011, for these analyses). The average age was 19.40 years (SD = 2.64), and participants were predominately right-handed (85 %). Color blindness was assessed using the Ishihara color blindness test, and all participants could distinguish between
Cogn Affect Behav Neurosci (2014) 14:1356–1374
the three colors used in the experimental task. The average FSIQ as assessed by the Wechsler Adult Intelligence Scales was 102.21 (SD = 9.60).
Behavioral measures Feature match task The feature match task was adapted from Duncan et al. (2008, Experiment 4). Task presentation was controlled by Presentation software running Windows XP. The task consisted of pairs of shapes being presented that (1) did not match on any stimulus dimensions, (2) matched on one stimulus dimension, or (3) matched on both stimulus dimensions. The rules and correct responses are shown in Table 1. Participants instructed with four rules (including one “irrelevant” rule) were told that they would be performing two tasks and that one task would involve pairs of numbers without any shapes and the other would involve both numbers and shapes. For the numbers task, participants were told that if shapes did not surround the numbers, they were to add the two numbers together and state the sum aloud. Participants instructed with three rules were only presented the rules regarding the numbers surrounded by shapes. There were eight blocks of 30 trials, separated by short breaks controlled by the participant. Each block consisted of 18 no-match trials, 9 one-match trials, and 3 two-match trials. The order of trial types was pseudorandom, as was selection of digits, colors, and shapes for each trial. There were 14 trials in the practice block. To test the influence of practice, practice trials were presented using the same trial ratio as that used in the main experimental blocks and the same practice trial ratio as that used in Duncan et al. (2008). This led to four betweensubjects conditions that varied on the number of rules given to them (either three or four task rules) and the number of trial types that were present in the practice block (two, three, or four). The number of each type of rule for the practice trials is shown in Table 2. In the original feature match task used in Duncan et al. (2008), participants in the 4-rules condition were given 9 no-match trials, 2 one-match trials, 1 two-match trial, and 2 no-surrounds trials, while participants in the 3-rules condition were given 9 no-match trials, 3 one-match trials, and 2 two-match trials.
1361
Calculations of visual angle were based on an approximate distance of 57 cm. Screen background was black. On nosurround trials, the stimuli were a pair of white digits, 0.7° in height and centered 2.3° left and right of a central fixation point. These two digits were always different, drawn from the set of 1–3. For surround trials, the stimuli were similar, except that each digit was contained within a colored outline shape. Shapes were squares, circles, or triangles 2.1° in height and presented in red, green, or blue. For both types of trials, stimuli were presented for 1 s; presentation was followed by a 1-s intertrial interval, during which the screen was blank. Wechsler Adult Intelligence Scales (WAIS–IV; Wechsler, 2008) The WAIS–IV is an IQ test designed to measure general intellectual functioning in adults and older adolescents. The test is composed of 10 core subtests and 5 supplemental subtests, with the 10 core subtests making up the FSIQ. The FSIQ has been shown to correlate with an extracted g factor (defined as the first principal component explaining maximum amount of variance) from the test battery at over .8 in the normal population (Jensen, 1998). EEG acquisition and data analysis EEG activity was recorded using a 38-channel Nuamps amplifier with Easycap (33 scalp sites plus VEOG and mastoid references). Electrodes were placed at Fp1, Fp2, F3, F4, F7, F8, Fz, FC1, FC2, FC5, FC6, FCz, FT9, FT10, C3, C4, Cz, T7, T8, CP1, CP2, CP5, CP6, P3, P4, P7, P8, Pz, PO9, PO10, O1, O2, and Iz. AFz was used as the ground electrode. EEG was recorded with a DC-35-Hz band-pass digitized to 250 Hz. EEGs were referenced online to the right mastoid and rereferenced to linked mastoids offline. Prior to EEG recording, input impedances were brought to 5 kΩ by careful scalp preparation. Two electrodes placed above and below participants’ left-eye recorded eye blinks. The resulting EEG was processed offline using functions from EEGLAB (Delorme & Makeig, 2004) and Fieldtrip (Oostenveld, Fries, Maris, & Schoffelen, 2011) toolboxes in MATLAB 2009b (The Mathworks, Inc.). EEG preprocessing The continuous EEG data sets were band-pass filtered between 1 and 30 Hz using a forward-reverse (zero-phase)
Table 1 Rules used in the feature match task Rule
Criteria
Response
No-Match One-Match Two-Match
If the two shapes presented do not match on color or shape. If the two shapes presented match on color OR on shape. If the two shapes presented match on color AND shape.
Withhold response Press left or right depending on which shape contains the larger number. Withhold response.
1362
Cogn Affect Behav Neurosci (2014) 14:1356–1374
Table 2 Number and type of trials in each instruction condition and practice condition 4-rules, Practice with 4 Rules
4-rules, Practice with 3 Rules
3-rules, Practice with 3 Rules
3-rules, Practice with 2 Rules
7 No-Match Trials 4 One-Match Trials 2 No-Surround Trials 1 Two-Match Trial
9 No-Match Trials 4 One-Match Trials 1 Two-Match Trial
9 No-Match Trials 4 One-Match Trials 1 Two-Match Trial
10 No-Match Trials 4 One-Match Trials
FIR filter (filtfilt) from the MATLAB Signal Processing Toolbox. Epochs were extracted for all stimulus presentations, from 1 s prior to stimulus onset to 2 s after stimulus onset. A semiautomatic artifact rejection procedure was run on the epochs, examining extreme amplitude values, abnormal trends, improbable data, abnormally distributed data, and abnormal spectra (Delorme, Sejnowski, & Makeig, 2007). Epochs were further manually inspected for noisy artifacts and, if detected, removed. Independent components analysis (ICA; Bell & Sejnowski, 1995) was performed on the clean set of data epochs using the runica algorithm implemented in EEGLAB. Using the ICA results, independent components were screened for activity resembling eye blinks, eye movements, muscle movement, and heartbeat on the basis of the independent components’ scalp map topography, power spectrum, and dipole location. Artifactual components were subtracted from the channel data for all analyses. Time-frequency analysis Time-frequency analysis was performed on all scalp electrodes and all independent component clusters using a threecycle Morlet Wavelet family (with a Hanning-tapped window applied) with a 0.5 width and padding ratio of 2. This was computed using the newtimef() function of EEGLAB. On the basis of the number of Wavelet cycles and their frequency width, values of event-related changes in power from stimulus onset to 1,436 ms poststimulus were obtained for the following logarithmically spaced frequency (Hz) bins: 3.00, 3.30, 3.63, 4.00, 4.40, 4.84, 5.33, 5.86, 6.45, 7.09, 7.81, 8.59, 9.45, 10.40, 11.44, 12.59, 13.86, 15.25, 16.78, 18.46, 20.31, 22.35, and 24.60. For all time-frequency analyses, the 400-ms time period preceding stimulus onset was used as the baseline period. The baseline period was averaged across each trial type, but only for trials that were not preceded by an error or a motor response. This was performed separately for each experimental condition. Event-related spectral perturbation (ERSP) plots were computed to show changes in event-related synchronization (ERS) and event-related desynchronization (ERD) on the same scale (logarithmic changes in synchronization expressed in dB). Event-related intertrial-coherence (ITC) plots were also computed. ITC is a complementary measure to ERSP and provides a measure of the phase and amplitude correlation between
individual trials at an individual electrode. ITC values vary between 0 and 1, with 0 representing a random assortment of amplitude and phase across trials and 1 representing a perfect correlation across trials. EEG statistical analysis Statistical analysis of ERPs and time-frequency analyses (ERSP plots) was performed with nonparametric techniques (Maris & Oostenveld, 2007) using functions from Fieldtrip (Oostenveld et al., 2011) and EEGLAB (Delorme & Makeig, 2004). Specifically, for each electrode or independent component analyzed, we computed 2,000 permutations of the data and used False Discovery Rate (FDR) control to correct for the number of statistical comparisons performed. The permutation distribution was obtained by combining the trials of the experimental conditions into a single set, randomly partitioning the trials into subsets, calculating the test statistic on this random partition, and then repeating this process of partitioning and calculating the test statistic a large number of times (N = number of permutations). Cluster analysis of independent components and source localization Independent component clustering was performed on FMT independent components that could be fitted to a single dipole with less than 10 % residual variance. This was performed using a k-means clustering algorithm (implemented in the Statistics toolbox for MATLAB), with dipole location, scalp topography and ERSP activity between 200 and 1,400 ms used as the clustering criteria. DIPFIT2 functions from Fieldtrip (Oostenveld et al., 2011) were used to fit single dipole source models to independent component scalp topographies using a boundary element head model. In the DIPFIT2 toolbox, the BEM model is coregistered with an average brain model (Montreal Neurological Institute) and returns approximate Talairach coordinates for each equivalent dipole source. To further confirm the source localation, we used standardized low-resolution brain electromagnetic tomography (sLORETA; Pascual-Marqui, 2002). The sLORETA method is a standardized discrete, three-dimensional distributed, linear, minimum norm inverse solution. For the present study, we exported single ICA dipole solutions from individual
Cogn Affect Behav Neurosci (2014) 14:1356–1374
participants into the sLORETA algorithm to obtain a distributed source localization of the activity. Procedure Testing was spaced over six 1-day sessions by trained staff at the Neurocognitive Development Unit, University of Western Australia. Participants arrived in the morning and spent the day at the unit, where they participated in different cognitive, social, and neuropsychological tests for this study. Participants were fitted with EEG caps, which took approximately 45 min, and performed two other cognitive measures while EEGs were being recorded. Data from the other tasks were not analyzed for the present study. Instructions for the feature match task were repeated verbatim from Duncan et al. (2008) supplementary materials, except for some minor modifications to the practice trial instructions. Participants were given short breaks between the blocks and between each experimental task while EEGs were being recorded. Knowledge of the feature match task’s rules was assessed by a free recall procedure at the end of the experiment.
1363
feature match task (Fig. 2b). Participants in the 4-rules condition (M rank = 37.44) displayed higher goal neglect than did participants in the 3-rules condition (M rank = 26.39), and this difference was significant, U = 322.00, z = −2.59, p = .009, r = −.33. The difference between experimental conditions remained significant when only the first four blocks were analyzed (to replicate Duncan et al., 2008), U = 355.00, z = −2.14, p = .033, r = −.27, but also when only the final four task blocks were analyzed, U = 376.00, z = −2.41, p = .021, r = −.30. Hence, the hypothesis that participants given more task rules would show increased goal neglect was supported. Moreover, these behavioral results suggest a lasting effect of task complexity on goal neglect, with participants in the 4-rules condition showing higher levels of goal neglect even during the final blocks of the experimental paradigm.
Results Behavioral analysis Three participants were excluded from analysis for excessive button responses during the experiment (e.g., responding on nearly every trial), leaving a sample size of 63 for the behavioral data. Goal neglect was scored using the same scoring procedure as that in Duncan et al. (2008), in which a participant's response for each rule was considered as “passed” (>25 % correct responses) or “failed” (
