Abstract
A growing body of research suggests that dual-task interference in sensory consolidation (e.g., the attentional blink, AB) and response selection (e.g., the psychological refractory period, PRP) stems from a common central bottleneck of information processing. With regard to response selection, it is well known that training reduces dual-task interference. We tested whether training that is known to be effective for response selection can also reduce dual-task interference in sensory consolidation. Over two experiments, performance on a PRP paradigm (Exp. 1) and on AB paradigms (differing in their stimuli and task demands, Exps. 1 and 2) was examined after participants had completed a relevant training regimen (T1 practice for both paradigms), an irrelevant training regimen (comparable sensorimotor training, not related to T1 for both tasks), a visual-search training regimen (Exp. 2 only), or after participants had been allocated to a no-training control group. Training that had shown to be effective for reducing dual-task interference in response selection was also found to be effective for reducing interference in sensory consolidation. In addition, we found some evidence that training benefits transferred to the sensory consolidation of untrained stimuli. Collectively, these findings show that training benefits can transfer across cognitive operations that draw on the central bottleneck in information processing. These findings have implications for theories of the AB and for the design of cognitive-training regimens that aim to produce transferable training benefits.
Similar content being viewed by others
Despite the staggering complexity and immense processing power of the brain, humans are subject to striking limitations when attempting to do more than one thing at a time (Marois & Ivanoff, 2005). For example, large response time (RT) costs are incurred when individuals attempt to perform two simple speeded sensorimotor tasks in close succession, as is the case in the psychological refractory period (PRP) paradigm (Pashler, 1994). In addition, accuracy costs are observed when individuals attempt to detect/identify more than one target from a rapidly presented stream of visual information, if the second target succeeds the first by 200–500 ms, a phenomenon known as the attentional blink (AB; Raymond, Shapiro, & Arnell, 1992). Fortunately, some evidence suggests that with training we can overcome these dual-task limitations (Choi, Chang, Shibata, Sasaki, & Watanabe, 2012; Kamienkowski, Pashler, Dehaene, & Sigman, 2011; Ruthruff, Johnston, & Van Selst, 2001). It remains to be explored how generalizable training benefits can be—for example; is it possible to transfer training benefits from one dual-task context to another? Furthermore, do training benefits have to be stimulus-specific? Or can we generalize performance benefits across stimulus classes? Here, we investigate these questions over the course of two experiments. We show that a training regimen that reduces dual-task interference in the PRP paradigm can also reduce the size of the AB. Secondly; we demonstrate some evidence that training benefits can be generalized across stimulus classes.
The PRP and AB
In the PRP paradigm, participants are presented with two stimuli separated by a variable stimulus onset asynchrony (SOA, the temporal gap between the stimulus onsets) each requiring a speeded response. The PRP effect is the observation that as the SOA is decreased, RTs to the second task increase dramatically, even if the stimuli are unmasked, and thus data-unlimited (Norman & Bobrow, 1975). As Sternberg (1969) showed, the analysis of sensory information up to a response can be divided into distinct stages of processing. Borrowing from this work, most accounts of the PRP effect divide task processing into three stages: early/perceptual, central/response section and late/response execution. The early and late stages are assumed to operate in parallel, meaning that during early or late processing in one task concurrent processing in another task is possible. In contrast, the central stage constitutes a bottleneck (although see Brisson & Jolicœur, 2007; Ruiz Fernández & Ulrich, 2010; Ulrich et al., 2006, for evidence of bottleneck processes at early/perceptual and late/response output stages). The nature of the bottleneck itself remains in debate. According to the structural bottleneck model (Pashler, 1984, 1994), central processes must operate in a serial manner; therefore Task 2 central processing may not begin until Task 1 central processing is complete (causing Task 2 delays). Capacity sharing accounts (Navon & Miller, 2002; Tombu & Jolicœur, 2003) share this structural view, but instead hypothesize that the bottleneck is a limited capacity parallel processor. Alternately, it has been argued that dual-task costs reflect a strategic consequence of task demands (EPIC framework; Meyer & Kieras, 1997a, b; Meyer et al., 1995), or are a result of a strategy invoked to minimize task crosstalk (ECTVA; Logan & Gordon, 2001).
In the AB paradigm, participants are presented with a rapid serial visual presentation (RSVP; usually at a rate of approximately 10/s; Potter & Levy, 1969) of stimuli at fixation, with two targets (T1 and T2) embedded amongst distractors. The relative position of T2 with respect to T1 (the T1–T2 lag) is varied and target detection rates are assessed. As the lag is decreased, detection rates for T2 decrease sharply. Specifically, given successful detection of T1, T2 performance is impaired when it follows T1 by approximately 200–500 ms (Raymond et al., 1992). Although a range of theories have been proposed to account for the information processing constraints reflected by the AB, theoretical accounts generally fall into two camps (Dux & Marois, 2009). According to bottleneck/resource theories (e.g., Chun & Potter, 1995; Jolicœur & Dell’Acqua, 1998), T1 processing postpones T2 processing at short lags, and because of the data-limited nature of the stimuli in the AB paradigm, results in T2 being overwritten before it can be selected for extended processing. In contrast, selection models argue that inhibitory processes are activated by (1) post-T1 distractors (Olivers & Meeter, 2008; Olivers, Hulleman, Spalek, Kawahara, & Di Lollo, 2011), (2) a competitive mechanism that enhances the episodic distinctiveness of T1 (Wyble, Bowman, & Nieuwenstein, 2009), or, (3) an overly conservative (yet implicit) strategy to protect T1 consolidation (Taatgen, Juvina, Schipper, Borst, & Martens, 2009). As a consequence, T2 processing is delayed or suppressed at short lags. Alternatively, other selection accounts such as the temporary loss of control theory (Di Lollo, Kawahara, Ghorashi, & Enns, 2005) claim that the AB reflects distractors disrupting input filters during T1 encoding. Dux and Marois (2009) have proposed that both T1 encoding limitations as well as cognitive control/selection operations give rise to the AB.
A common mechanism underlying the PRP and AB
Unified bottleneck theories predict that the sensory-consolidation limitations evident in the AB, and the response-selection limitations reflected by the PRP, result, at least partly, from a common central capacity limitation (Arnell & Duncan, 2002; Jolicœur, 1999; Jolicœur, Dell’Acqua, & Crebolder, 2001; Marti, Sigman, & Dehaene, 2011; Ruthruff & Pashler, 2001; Tombu et al., 2011; Zylberberg, Fernández Slezak, Roelfsema, Dehaene, & Sigman, 2010). In the short-term consolidation (STC) framework (Jolicœur & Dell’Acqua, 1998), encoding visual information into durable storage (sensory consolidation) requires the same central mechanisms upon which response selection operations draw. According to this theory, the central processing mechanism can only support a single information processing stage at a time; that is, it can support short-term consolidation, or response selection, but it cannot support both operations concurrently. Consequently this limitation results in the postponement of Stimulus 2 processing at short T1–T2 lags/SOAs, thus increasing the likelihood in an AB paradigm that subsequent distractors will overwrite T2, which will not enter awareness, and that the Task 2 response will be prolonged in the PRP paradigm. According to the global workspace model (Dehaene & Changeux, 2011; Dehaene, Kerszberg, & Changeux, 2001; Dehaene & Naccache, 2001), sensory-consolidation and response-selection processes have access to a neuronal workspace that makes information globally available to multiple brain systems, forming a conscious percept of the task relevant stimulus. Processing in the global workspace is strictly serial and thus, when two percepts require access simultaneously, conscious access/the response to the second percept is delayed. Although the STC framework and the global workspace model both attribute the dual-task interference in the AB and PRP paradigms to a unified cause, it is worth pointing out that the two theories do so in a different manner. Whereas in the STC framework the AB and PRP effects are caused by limitations in distinct, yet mutually interfering stages of processing (STC, response selection), in the global workspace model a common limited capacity stage of processing is responsible for both effects (the global workspace).
Behavioral and electrophysiological evidence indicating a common limitation has stemmed from hybrid tasks that incorporate elements of both the AB and PRP paradigms. For example, adding online response selection demands to T1 in an AB paradigm further impairs T2 detection at shorter lags (Jolicœur, 1999). Likewise, encoding a visual stimulus for later report increases RTs to a temporally overlapping speeded sensorimotor task (Jolicœur & Dell’Acqua, 1998), and delays Task 2 related ERP components (Arnell, Helion, Hurdelbrink, & Pasieka, 2004) that have also been shown to be absent when T2 is missed in an AB paradigm (Kranczioch, Debener, & Engel, 2003; Sergent, Baillet, & Dehaene, 2005; Vogel, Luck, & Shapiro, 1998). In addition, imaging evidence indicates overlap in frontoparietal networks involved in the encoding and response-selection bottlenecks (see Marois & Ivanoff, 2005, for a review). Increased activity in frontoparietal areas has been observed when target displays are subjected to manipulations known to give rise to the AB (Marois, Chun, & Gore, 2000), and when T2 is detected under RSVP conditions (Kranczioch, Debener, Schwarzbach, Goebel, & Engel, 2005; Marois, Yi, & Chun, 2004). Similarly, frontoparietal areas have been found to display patterns of delayed activity that track delays in both perceptual encoding and response selection when demands are high, relative to when demands are low (Dux, Ivanoff, Asplund, & Marois, 2006; Hesselmann, Flandin, & Dehaene, 2011; Jiang, 2004; Marti et al., 2011; Tombu et al., 2011)—as would be predicted by unified bottleneck theories.
Training to overcome dual-task limitations
Although dual-task settings can lead to considerable task impairments, such costs can be drastically reduced with practice. As we described above, it has been well documented that training improves performance in the PRP paradigm (see Pashler, Johnston, & Ruthruff, 2001, for a review). In fact, training on the first task alone is sufficient to reduce the PRP effect (Ruthruff et al., 2001), presumably because decision stages of the task, which draw on central resources and involve the mapping of a stimulus to an appropriate response, have been shortened (Dux et al., 2009; Kamienkowski et al., 2011; Van Selst & Jolicœur, 1997) or automatized (Maquestiaux, Laguë-Beauvais, Bherer, & Ruthruff, 2008; Ruthruff, Hazeltine, & Remington, 2006; Ruthruff, Van Selst, Johnston, & Remington, 2006). As a result, Task 2 central stages can be initiated sooner. Far less is known about whether training reduces the AB. Recent findings indicate that repeating the dual-target RSVP task over time may improve T2 performance (Klein, Arend, Beauducel, & Shapiro, 2011; Maki & Padmanabhan, 1994; Nakatani, Baijal, & van Leeuwen, 2009), however, it has yet to be addressed whether training that improves PRP performance can influence performance in the AB paradigm. This was the question we addressed in the first experiment of the study.
Experiment 1
As we reviewed above, training can drastically reduce the size of the PRP effect. However, it remains to be explored whether these training benefits can generalize to other dual-task settings that presumably share a causal locus with the PRP effect. Given the hypothesis that training reduces the PRP effect by shortening the duration of central processing, and the hypothesis that the AB also relies on central processing, it is conceivable that training that is beneficial in the PRP context might extend to the AB context.
At one extreme, training might increase the efficiency of the central bottleneck for any task that relies upon it. For example, in the case of the STC framework (Jolicœur & Dell’Acqua, 1998), both sensory-consolidation and response-selection processes rely on the same capacity limited central mechanisms. If training improves the efficiency of these central mechanisms, performance should improve for both sensory consolidation and response selection. In the case of the global workspace model (Dehaene & Changeux, 2011; Marti et al., 2011), in which it is assumed that limited access to a serial neuronal workspace gives rise to both the PRP effect and the AB, we may expect that training will improve the access of information to this workspace, thus enhancing performance on both paradigms. At the other extreme, training might be task-specific. In this case training that reduces the PRP effect would not be expected to affect the AB.
In Experiment 1, participants took part in two sessions (pre- and posttraining) during which they completed a PRP paradigm and a “categorical”-AB paradigm. In the PRP paradigm, participants first performed a data-unlimited (not masked) speeded letter-identification task requiring a manual response (the visual–manual task), followed at either a short or long SOA by a speeded sound discrimination task requiring a vocal response (the auditory–vocal task). In the categorical-AB paradigm, participants were presented with an RSVP stream of digits in which two target letters were embedded at various lags relative to each other. Both letters were to be encoded for recall at the end of the trial. The first letter was drawn from the same set of letters that had been used in Task 1 of the PRP paradigm, whereas the second letter was drawn from a different set of letters. So, Task 1 in the categorical-AB paradigm was essentially a data-limited version of the task employed as Task 1 in the PRP paradigm, with unspeeded response demands.
Between these two sessions, participants either completed two weeks of training or were assigned to a control group (no training, retested after a two-week interval). If a training regimen improves performance for both the PRP and the categorical-AB paradigms, it would be important to know whether the underlying mechanisms for performance improvements are the same for both paradigms. Previous research has indicated that training-related reductions in the PRP effect occur because the duration of central bottleneck stages have been reduced, most likely because of the strengthened representation for the stimulus–response mappings employed at test (e.g., Dux et al., 2009). We therefore included two training groups: a relevant-training group, which trained on the task employed as Task 1 in the PRP paradigm used in the pre- and posttraining sessions, and an irrelevant-training group, which trained on a similar but unrelated speeded visual–manual task. In fact, the only difference between the relevant-training group and the irrelevant-training group was that, whereas relevant training involved rapidly responding to letters, irrelevant training involved rapidly responding to colors. If training-related reductions in dual-task interference were to be only found for the relevant-training group, this would suggest that the improvements were largely due to strengthened representations of the stimulus–response mappings. If irrelevant training were also to reduce dual-task interference, this would suggest a more general improvement in processing efficiency that went beyond merely strengthening the representations of specific stimulus–response mappings.
Method
Participants
The participants were included if they were between 17 and 49 years of age, had normal or corrected-to-normal vision, had no history of psychiatric or neurological disorder, and scored <90 % accuracy for T2 | T1 at lag 2 on the AB tasks. The groups were matched for age, gender, handedness, and years of education (see Table 1 for the participants’ characteristics). We found no significant differences between the groups on any of these measures (ps > .2).
All participants received AUD10 per hour for participation. Participants assigned to training groups were also able to earn bonus dollars for accuracy and for beating RT deadlines (~AUD15 per participant). All procedures were cleared in accordance with the ethical review processes of the School of Psychology at The University of Queensland and the guidelines of the National Statement on Ethical Conduct in Human Research.
Training-related improvements in performance on the PRP paradigm are generally large (RTs typically decrease by several hundred milliseconds; e.g., Dux et al., 2009; Maquestiaux et al., 2008; Schumacher et al., 2001; Van Selst, Ruthruff, & Johnston, 1999); therefore, we estimated that training benefits on dual-task limitations should be represented by a medium to large population effect size (f = .3; Cohen, 1988). A power calculation (Cohen, 1988; Faul, Erdfelder, Lang, & Buchner, 2007) revealed that to achieve 80 % power to detect a significant 3 (group) × 2 (session) × 2 (SOA) interaction in the PRP data, or a significant 3 (group) × 2 (session) × 4 (lag) interaction in the categorical-AB data, a total of 30 participants (ten per group) would be required. A total of 32 participants were recruited for the study. Of these, two failed to show poorer T2 | T1 accuracy at earlier lags (2 and 3) relative to later lags (5 and 6) and were excluded from further participation. The remaining 30 participants were randomly allocated to the relevant-training, irrelevant-training, and control groups (see Table 1 for further information).
Materials
All procedures were carried out using a 21-in. Sony Trinitron CRT monitor and a Macintosh 2.5-GHz Mini computer. All tasks were programmed in MATLAB R2009b (The MathWorks, Natick, MA) using the Psychophysics Toolbox 3.0.9 extension (Brainard, 1997; Pelli, 1997). Participants sat at an approximate viewing distance of 57 cm from the computer screen for all the tasks used in the two experiments.
Procedure
Irrespective of group allocation participants initially completed a PRP paradigm and a categorical-AB paradigm.Footnote 1 This initial session will be referred to from here onward as the pretraining session. Subsequently, the participants assigned to the relevant- and irrelevant-training conditions attended eight training sessions held over two weeks, attending a maximum of one session per day. Following the training sessions, or after two weeks for the control group, participants again completed the PRP and categorical-AB paradigms (posttraining session). The pre- and posttraining sessions occurred at the same time of day. The paradigms were presented in counterbalanced order within and across sessions (pre- and posttraining). For a schematic representation of all the tasks, refer to Fig. 1. The timeline of the tasks completed by each group is presented in Table 2.
Schematic representations of trials for the psychological refractory period (PRP) and categorical attentional blink (AB) paradigms and the training tasks of Experiment 1. All participants completed the PRP and AB paradigms at pre- and posttraining. Participants assigned to the relevant- and irrelevant-training groups completed 4,080 training trials between pre- and posttraining. Control participants had a no-training two-week interval between pre- and posttraining. (T1 = Target 1, T2 = Target 2, ms = milliseconds)
PRP paradigm
For each trial, participants were instructed to complete two four-alternative choice tasks as quickly and accurately as possible; Task 1 required a manual response to one of four letters (H, S, A, or B) using one of the four corresponding keys (V, B, N, or M on a standard QWERTY keyboard; the keys were covered with white stickers, and participants were required to respond with the four fingers of their dominant hand). Task 2 required participants to make one of four nonword vocal responses (“Pag,” “Mab,” “Dat,” or “Taf”) to one of four complex tones. The tones were selected from those previously employed by Dux et al. (2006). Three practice blocks were performed immediately prior to (i.e., the same day as) the main PRP experiment. The first practice block was administered in order to familiarize participants with the stimulus–response mapping used for the auditory–vocal task. On each trial, one of the four complex tones used in the main experiment was presented, to which participants made the corresponding vocal response. Each tone was presented five times over the course of the practice block, for a total of 20 trials. A similar 20-trial practice block was also administered for the visual–manual task. On each of these trials, participants were presented with one of the four letters used in Task 1 of the main experiment, to which they made the corresponding manual response. Each letter was presented five times over the course of the practice block. Finally, a block of 20 dual-task trials were administered in order to familiarize participants with the procedure used in the main experiment. These dual-task trials were identical to the dual-task trials employed in the main experiment. The auditory–manual, visual–manual, and dual-task practice sessions were repeated until a criterion of 90 % correct was achieved.
Each dual-task trial began with a black fixation cross presented on a white background. After 500 ms, the fixation cross was removed, initiating a blank interval with a random duration lasting between 200 and 1,000 ms. The Task 1 letter stimulus followed, presented centrally in black Courier New font for 200 ms and subtending a visual angle of approximately 1.0° × 1.0°. The Task 1 stimulus was unmasked, and thus data-unlimited. After either a short (200-ms) or a long (1,000-ms) SOA, the Task 2 auditory stimulus was presented for 200 ms. A blank screen was presented for an additional 2,800 ms and was followed by a message informing participants to press the space bar when they were ready to begin the next trial. The experimental trials consisted of 50 repetitions each of the two SOA trial types. The presentation of the experimental trials was randomized, as was the selection of the Task 1 and Task 2 stimuli for each trial. RTs for each task were taken from the task-specific stimulus onset.
Categorical-AB paradigm
For the categorical-AB paradigm, two letter targets appeared in an RSVP stream of digit distractors. T1 was drawn from the same stimulus set used in Task 1 of the PRP (H, S, A, and B), using the same stimulus–response mappings (once again, participants were instructed to respond with the same four fingers of the dominant hand). T2 was one of four prespecified letters (J, D, E, or K), and participants were instructed to identify the target using one of four keys (F, G, H, or J). Participants had to wait until the end of the RSVP stream to respond to both T1 and T2, and they were informed to take as long as they wished to make their responses. So, in the categorical-AB paradigm, Task 1 was the same as Task 1 in the PRP paradigm, except that the stimulus was data-limited and the response was delayed and unspeeded. Task 2 also required the identification of one of four letters and the unspeeded selection of one of four responses; however, the stimulus identities and response options differed from those of T1, and consequently, from those presented in the PRP paradigm.
Each trial began with a central fixation cross, presented for 500 ms. At fixation offset, an RSVP stream of eight digit distractors and two target letters was displayed. All stimuli were presented centrally in black Courier New font on a white screen, for 100 ms each, and subtended 1.0° of visual angle at a viewing distance of 57 cm. The distractors were randomly selected from the digits 2 to 9 (inclusive). On all trials, T1 was presented at Serial Position 3, with T2 following at a lag of 2, 3, 5, or 6. Participants completed 20 practice trials just prior to commencing the main experiment. The experimental trials consisted of a randomized presentation of 50 repetitions of the four lag trial types, resulting in 200 experimental trials. The T1 and T2 identities were selected randomly from trial to trial.
Training (relevant vs. irrelevant)
The goal of the training procedures was to improve RTs on either Task 1 of the PRP paradigm (relevant training) or a comparable sensorimotor task (irrelevant training). Participants completed 4,080 practice trials over 17 training blocks. Sixteen blocks were completed over eight training sessions; the seventeenth was completed just prior to the second testing session. The first session began with an overview of the training program, whereas the remaining sessions began with visual feedback (in the form of a line graph) of a participant’s median RTs achieved over the previous training blocks. Median RTs were chosen so as to reduce the influence of outliers. For both the relevant- and irrelevant-training groups, each trial began with a fixation cross, presented for 750 ms. Subsequently, a blank screen was presented for one of four randomly selected intervals (100, 200, 300, or 400 ms), followed by the target stimulus for 200 ms. The target stimulus was unmasked, and thus data-unlimited, as had been the case in the PRP paradigm. For the relevant-training group, the target stimulus was randomly selected from the four letter stimuli used as Task 1 in the PRP and AB paradigms, whereas for the irrelevant group, the target stimulus was randomly selected from one of four colored discs (red, blue, green, or yellow). A blank screen followed until the participant responded, and a tone was played if the participant made an error. Each of the four stimuli was presented 120 times over the course of each practice session. This resulted in 480 trials per session, presented over two blocks of 240 trials each. The only additional difference between the relevant and irrelevant groups was that the response keys for the irrelevant-training group were shifted four keys rightward on the keyboard (“,”–“.”–“/”–“shift”; again, these keys were covered with blank stickers, and participants were instructed to respond with the first through fourth fingers of their dominant hand) and were assigned to the colors red, blue, green, and yellow, respectively.
In order to encourage quick and accurate responses, participants were provided with performance feedback at the end of each training block. If participants scored over 95 % correct and met their RT target on over 70 % of the trials, a bonus dollar was awarded. When participants maintained that accuracy and met their RT target on over 75 % of the trials, a new RT target was calculated, and a further two bonus dollars were awarded ($3 in total). RT targets were derived using the mean and standard deviation of the previous block’s RTs. The 75th percentile was calculated and employed as the new RT target. Along with the performance feedback presented at the end of each block, the total number of dollars awarded for that block, the total number of dollars earned overall, and the RT target for the next block were also presented.
Results
For trials that required a speeded response (training and PRP paradigm), outlier screening was performed for each participant in each session for each task separately. Trials were excluded if a given RT was either <100 ms or >3 standard deviations (SDs) above the mean for that participant in that condition and task. Excluded RT trials accounted for <1.5 % of the data.
In order to allow for the application of additive factors logic to the categorical-AB accuracy data, the percentage correct accuracy data were corrected for guessing and then log transformed to the base 10 (see Schweickert, 1985). Correction for guessing ensures that the accuracy data reflect performance on the cognitive process being targeted and not guessing. We assume that when a target is not identified, participants guess the target identity. Since the target identities could be one of four possible targets, we assume that if participants guess, they have a one-in-four chance of being correct. Therefore, to obtain guess corrected accuracy data, we used the formula 1 – [(error rate/3) × 4]. The log transformation rescales the data so that a manipulation that has the same proportional effect at different accuracy levels will have the same absolute effect on the scale. This facilitates comparison between pre- and posttraining, and helps protect against possible ceiling effects when detecting interactions. For clarity, untransformed accuracy values are reported in brackets next to transformed values.
Training
We examined training effects by subjecting the mean RTs from the training sessions to a 2 (group) × 17 (block) mixed analysis of variance (ANOVA). The results revealed significant main effects of block [F(3.65, 65.72) = 33.18, MSE = .003, p < .001, η p 2 = .65] and group [F(1, 18) = 5.47, MSE = .05, p = .03, η p 2 = .23], but no interaction between the two (p = .32, η p 2 = .06). As can be seen in Fig. 2, participants responded significantly faster as training progressed, with mean RTs dropping from 620 ms in Block 1 to 488 ms in Block 17. As is also evident from Fig. 2, participants in the relevant-training group were faster than those in the irrelevant-training group (mean RTs of 498 and 555 ms, respectively). This likely reflects the automaticity of letter reading relative to color identification (e.g., MacLeod, 1991; Stroop, 1935).
Accuracy (left, as proportions correct) and response times (RTs, right) for the relevant- and irrelevant-training groups over the 17 training blocks of Experiment 1. Both groups showed equivalent decreases in RTs over training, while retaining accuracy. Error bars represent 95 % confidence intervals of the means for the relevant and irrelevant groups, based on the within-group Subject × Block error term (Masson & Loftus, 2003)
An identical mixed ANOVA was also conducted on the accuracy data. Neither of the main effects nor their interaction approached significance (ps > .18). Indeed, accuracy was consistently high (mean = 96 %, SD 2 %) across all training blocks.
PRP paradigm
Although we analyzed RTs to both Tasks 1 and 2, of primary interest was the effect of training on the PRP effect, as reflected in Task 2 RT improvements. More specifically, if training improves performance in the PRP paradigm, we would expect this to be reflected in a reduced PRP effect (or an attenuated difference between Task 2 RTs for the two SOA conditions).
Task 1 RT
The Task 1 RT data were submitted to a 3 (group) × 2 (session) × 2 (SOA) mixed ANOVA. The means for each condition are presented in the top portion of the left panel in Fig. 3. Overall, we observed a main effect of SOA, F(1, 27) = 26.88, MSE = .004, p < .001, η p 2 = .50. RTs were longer at the 200-ms SOA (mean = 757 ms, SD = 182 ms) than at the 1,000-ms SOA (mean = 697 ms, SD = 157 ms), possibly reflecting capacity sharing between Tasks 1 and 2 (Navon & Miller, 2002; Tombu & Jolicœur, 2002, 2003). A significant main effect of session was also observed, F(1, 27) = 76.80, MSE = .014, p < .001, η p 2 = .74. Overall, participants were faster at posttraining (mean = 632 ms, SD = 159 ms) than at pretraining (mean = 823 ms, SD = 200 ms). A significant Group × Session interaction was also found [F(2, 27) = 5.26, MSE = .014, p = .01, η p 2 = .28]. To further investigate the interaction, difference scores were calculated for each participant (pretraining mean RT – posttraining mean RT). The relevant-training group showed a significantly greater decrease in RTs relative to the control group [F(1, 18) = 9.45, MSE = .016, p = .01, η p 2 = .33]. No other follow-up comparisons achieved significance (ps > .09, η p 2s < .14).
Response times (RTs) to Task 1 and Task 2 in the psychological refractory period (PRP) paradigm as a function of SOA (left) and the PRP effect (Task 2 RT 200 ms SOA – Task 2 RT 1,000 ms SOA) at pre- and posttraining as a function of group (right) in Experiment 1. Only the relevant-training group showed a change in the PRP effect from pre- to posttraining. Error bars represent 95 % confidence intervals of the means (calculated with the Subject × Session × SOA error term; Masson & Loftus, 2003)
Task 2 RT
The Task 2 RT data were subjected to an identical 3 (group) × 2 (session) × 2 (SOA) mixed ANOVA. The means for each condition are presented in the bottom portion of the left panel in Fig. 3. Crucially, a significant Group × Session × SOA interaction [F(2, 27) = 5.71, MSE = .006, p = .009, η p 2 = .30] was observed, indicating that the nature of training determined its impact on the PRP effect. To identify the groups that differed from one another, each pair of groups was entered into a Group × Session × SOA ANOVA. The results revealed that training had a differential impact on the PRP effect for the relevant group relative to either the control group [F(1, 18) = 12.11, MSE = .004, p = .003, η p 2 = .40] or the irrelevant group [F(1, 18) = 7.07, MSE = .006, p = .016, η p 2 = .28], whereas the irrelevant and control groups did not differ from each other [F(1, 18) = 0.06, MSE = .006, p = .81, η p 2 = .003].
Further follow-up analyses examined the specific impact of training on the PRP effect for each group. To do so, a PRP effect value was calculated for each participant at pre- and posttraining by subtracting the mean RT in the 1,000-ms condition from that in the 200-ms condition. The results are plotted in the right panel in Fig. 3. The relevant-training group showed a significant reduction in the PRP effect with training [F(1, 9) = 18.95, MSE = .008, p = .002, η p 2 = .68]. No reduction in the PRP effect was found for the irrelevant-training group [F(1, 9) = 0.03, MSE = .016, p = .89, η p 2 = .003] or the control group [F(1, 9) = 0.41, MSE = .009, p = .54, η p 2 = .04]. Thus, the present results suggest that only task-specific training reduces the PRP effect. Given that the relevant training targeted Task 1 performance, and that only the early and central stages of Task 1 processing contribute to the PRP effect, relevant training must have improved performance by reducing the duration of early and/or central stages of processing in Task 1. Thus, although Task 1 RTs were also reduced for both the irrelevant and control groups at test, the absence of a reduction in the PRP effect indicates that any changes to Task 1 processing must have occurred at a locus later than that which gives rise to the PRP effect.
T1 and T2 accuracy
The accuracy data for Task 1 and Task 2 are presented in Table 3. Importantly, 3 (group) × 2 (session) × 2 (SOA) mixed ANOVAs performed on the Task 1 and Task 2 accuracy data showed no evidence for speed–accuracy trade-offs (see Table 3). A significant Group × Session interaction was observed in the Task 1 accuracy data [F(2, 27) = 3.89, MSE = .002, p = .03, η p 2 = .22]. To investigate this interaction, each pair of groups was entered into a Group × Session × SOA ANOVA. The relevant group showed a greater increase in accuracy from pre- to posttraining relative to the irrelevant group [F(1, 18) = 5.0, MSE = .002, p = .04, η p 2 = .22], and also relative to the control group [F(1, 18) = 7.25, MSE = .001, p = .02, η p 2 = .29]. A significant Group × Session interaction was not observed when the irrelevant group was compared to the control group [F(1, 18) = 0.37, MSE = .001, p = .55, η p 2 = .02]. A significant Session × SOA interaction was also observed for the Task 2 accuracy data [F(1, 27) = 4.11, MSE = .001, p = .053, η p 2 = .13]: A significant increase in Task 2 accuracy was observed from pre- to posttraining at the short SOA [F(1, 29) = 8.18, MSE = .01, p = .01, η p 2 = .22]. This effect was not present at the long SOA [F(1, 29) = 3.29, MSE = .011, p = .08, η p 2 = .10]. No other interactions achieved significance (all ps > .3).
Categorical-AB paradigm
The results of the PRP analysis showed that, in the context of PRP interference, dual-task training benefits were task-specific. Training a data-unlimited speeded visual–manual task, using different stimuli than those used at test, did not reduce dual-task interference in the PRP paradigm. The primary goal of the present analysis was to investigate whether the training regimen that proved successful in the context of the PRP paradigm (relevant training) would transfer to the context of the categorical-AB paradigm. Here, although the task being performed at test and during training was the same (letter identification), the context was different (training on a data-unlimited speeded version of the task, testing on a data-limited, unspeeded version of the task). In order to determine whether performance changes were brought about by the same training regimen that improved PRP performance, we assessed the impacts of both relevant (speeded letter identification) and irrelevant (speeded color identification) training on performance on the categorical-AB paradigm.
If training benefits are specific to the task conditions that were trained upon, training on a data-unlimited speeded version of the task should not result in training benefits when testing for dual-task interference using the data-limited, unspeeded categorical-AB paradigm. On the other hand, if training can benefit more than one processing operation that relies on the central bottleneck, then relevant training should also result in a reduced categorical AB. Furthermore, if the locus of improvement is the same for the PRP and the categorical AB, then irrelevant training should not reduce dual-task interference in the categorical-AB paradigm.
Again, although we analyzed accuracy changes for both T1 and T2, our crucial measure of training benefits pertained to attenuated differences in T2 given T1 (T2 | T1) performance between early and late lag conditions with training. More specifically, we looked for reductions in the magnitude of the AB with training.
T1
A 3 (group) × 2 (session) × 4 (lag) mixed ANOVA revealed a general increase in accuracy from pre- [mean = 85 % (1.90), SD = 8.57 (0.07)] to posttraining [mean = 90 % (1.94), SD = 6.80 (0.05); F(1, 27) = 13.08, MSE = .003, p = .001, η p 2 = .33; see Fig. 4, top left panel]. We also found a significant main effect of lag [F(3, 81) = 4.21, MSE < .001, p = .008, η p 2 = .14]. Post-hoc least significant difference comparisons revealed that accuracy was significantly lower at lag 2 [mean = 89 % (1.93), SD = 9.39 % (0.05)] than at lag 3 [mean = 90 % (1.95), SD = 7.37 % (0.03), p < .01]. None of the other comparisons achieved significance (all ps > .06).
Pre- and posttraining Target 1 (T1) and Target 2 given Target 1 (T2 | T1) accuracy (guess-corrected [GC] and log-transformed) in the attentional blink (AB) paradigm, plotted as a function of lag (left) in Experiment 1. Accuracy difference scores (plotted by groups, right) were calculated by subtracting the pretraining T2 | T1 performance at each lag from the posttraining T2 | T1 performance using the GC, log-transformed data. Both the relevant- and irrelevant-training group showed a reduced AB magnitude posttraining relative to the control group, evidenced by greater accuracy changes at earlier than at later lags. This effect was not present in the control group. Error bars represent 95 % confidence intervals of the means (calculated with the Subject × Session × Lag error term; Masson & Loftus, 2003). ** p < .005
T2 | T1
MacLean and Arnell (2012) argued that two key criteria must be fulfilled in order to demonstrate that the AB has been modulated by an experimental manipulation. First, it must be demonstrated that the lag-dependent effect on T2 | T1 accuracy (the key characteristic of the AB) has been influenced by the experimental conditions. In the present design, this would be indicated by a significant Session × Lag interaction. Second, it would be important that the interaction be driven by accuracy changes at early lags (i.e., those occurring within the AB temporal window) rather than accuracy changes occurring at later lags (at lags that lie outside of the AB temporal window).
A significant Group × Session × Lag interaction [F(6, 81) = 3.23, MSE < .001, p = .007, η p 2 = .19; see Fig. 4, left panel, bottom] indicated that the effect of training (session) on the size of the AB (lag) was influenced by the group to which participants were assigned. Subsequent analysis showed a significant Group × Session × Lag interaction when the relevant-training group was compared to the control group [F(3, 54) = 3.06, MSE = .003, p = .04, η p 2 = .15], and when the irrelevant-training group was compared to the control group [F(3, 54) = 3.64, MSE = .003, p = .02, η p 2 = .17]. When the relevant group was compared to the irrelevant group, the Group × Session × Lag interaction failed to reach significance [F(3, 54) = 2.69, MSE = .001, p = .06, η p 2 = .13]. Post-hoc comparisons did not reveal any significant differences between the two groups when the change in accuracy from pre- to posttest was compared at each lag (all ps > .10). However, given that the three-way interaction test was on the margins of significance, caution must be taken when interpreting differences between the relevant and irrelevant training regimens.
To further investigate group differences, and to establish whether the observed performance changes met the key criteria for evidence for AB attenuation (MacLean & Arnell, 2012)—that is, whether the performance changes were due to greater improvements at early lags than at later lags—2 (session) × 4 (lag) repeated measures ANOVAs were conducted for each group. The relevant-training group showed a significant Session × Lag interaction [F(3, 27) = 7.90, MSE = .004, p = .001, η p 2 = .47], as did the irrelevant-training group [F(3, 27) = 8.06, MSE = .002, p = .001, η p 2 = .47]. However, a significant interaction was not present in the control group [F(3, 27) = 0.773, MSE = .003, p = .52, η p 2 = .08]. This indicates that both the relevant- and irrelevant-training groups showed modulation of the AB, whereas the control group did not.
Although all participants performed significantly below ceiling for T2 | T1 at later lags pretraining (using one-sample t tests against ceiling for each group, all ps < .014), we wanted to ensure that the significant interactions that were observed in the relevant and irrelevant groups were driven by changes in T2 | T1 accuracy performance at earlier lags, rather than possible ceiling effects for performance at the later lags. In order to achieve this, difference scores were calculated at each lag for each participant. T2 | T1 accuracy scores obtained at the pretraining session were subtracted from the posttraining accuracy scores (see Fig. 4, rightmost panel). As can be seen, both the relevant-training and irrelevant-training groups showed large changes in performance at earlier lags.
Post-hoc LSD comparisons were performed on these difference-score data. The relevant-training group showed significantly greater improvements at lag 2 [mean = 17.28 % (0.16), SD = 11.98 % (0.13)] than at lags 5 [mean = 10.10 % (0.08), SD = 10.53 % (0.08), p = .02], and 6 (mean = 4.22 % (0.03), SD = 10.25 % (0.08), p = .007]. Improvement at lag 3 [mean = 21.16 % (0.20), SD = 14.17 % (0.17)] was significantly greater than at lag 5 (p = .02) and lag 6 (p = .009). Similarly, the irrelevant group showed significantly greater improvements at lag 2 [mean = 16.33 % (0.16), SD = 11.42 % (0.14)] than at lag 5 [mean = 11.23 % (0.09), SD = 9.20 % (0.08), p = .04] and lag 6 [mean = 3.93 % (0.03), SD = 7.36 % (0.05), p = .006]. Their performance improvement at lag 3 [mean = 11.53 % (0.09), SD = 10.31 % (0.10)] was significantly greater than at lag 6 (p = .03). No significant differences were observed for the control group (all ps > .2).
To recap, the observed significant interaction between the three groups showed that both relevant and irrelevant training were effective at reducing the size of the AB, whereas no such reduction was shown for the control group. The post-hoc analysis showed that the performance changes observed for the relevant- and irrelevant-training groups occurred for early lags, rather than later lags, confirming attenuation of the AB in these two groups. These findings stand in stark contrast to those for the PRP paradigm, in which only relevant training was effective in reducing dual-task interference.
Discussion
In Experiment 1, we explored two key issues associated with training-related reductions in dual-task interference. The first related to the generalizability of training across dual-task paradigms, whereas the second related to the specificity of training required for reducing dual-task interference. Three groups of participants performed both a PRP and an AB paradigm before and after completing a training regimen. The relevant-training group received training on Task 1 from the PRP paradigm, whereas the irrelevant-training group received training on an analogous, but unrelated, speeded alternative forced choice task. The control group received no specific training, but was retested after a similar amount of time as the other groups. Clear differences for the influence of training on the AB and PRP were observed. Specifically, whereas only relevant training reduced the PRP effect, both relevant and irrelevant training attenuated the AB. This suggests that whereas reduction of the PRP effect requires training the specific stimulus–response mappings employed at test (when single-task training is employed for the training regimen), more general training is effective in the context of the categorical-AB paradigm. Despite clear differences between the training task and the tasks that make up the AB paradigm (data-unlimited speeded letter/color-discrimination task during training, data-limited unspeeded letter-discrimination tasks in the AB paradigm), training (relevant and irrelevant) nonetheless reduced dual-task interference.
These results suggest that although a training regimen that reduces the PRP effect can also attenuate AB magnitude, it is not necessarily for the same reason, because the irrelevant training regimen reduced dual-task interference in the categorical-AB paradigm, but not the PRP paradigm. Given that both relevant and irrelevant training regimens improve performance on the AB, something other than strengthened specific stimulus–response mappings must be driving the observed performance benefits. Furthermore, the results of Experiment 1 indicated that training benefits were not stimulus-specific and improved AB performance in general. It may be that this benefit transferred across AB tasks that had different stimuli and task demands. One goal of Experiment 2 was to examine the extent of this generalization.
It is also crucial to include an active control group when testing the effects of training interventions (Redick et al., 2013). This ensures that any detected performance changes could be attributed to the effect of the training regimen employed, rather than to motivational differences that might arise when comparing participants who have invested time and effort toward completing a challenging training regimen to a no-training control group. Given that AB performance has previously been shown to be modulated by factors such as attentional investment (Olivers & Nieuwenhuis, 2006), it was important to ascertain that the attenuated ABs observed in Experiment 1 were due to the specific features of the relevant and irrelevant training regimens, rather than factors such as investment or effort. This was a further goal of Experiment 2.
Experiment 2
In Experiment 1, we showed that training a speeded, data-unlimited task reduces the size of the AB, even when the stimuli and task employed during training differ from those used at test. However, it is not clear from Experiment 1 whether this performance benefit was due to the processing mechanisms tapped during the training task, or to motivational factors such as attentional investment. Experiment 2 assessed the influence of an extra training condition on AB performance. In addition to the relevant, irrelevant and control groups, a fourth group received visual search training. Visual search performance is known to not correlate with AB performance (Arnell, Howe, Joanisse, & Klein, 2006), therefore any reduction in the size of the AB observed in this condition can be attributed to reward and motivational factors alone. If reductions in the size of the AB are due to a specific processing mechanism tapped by the relevant- and irrelevant-training conditions, and are not due to reward or motivational factors, then these training regimens should attenuate the AB to a greater extent than the visual-search training regimen. In contrast, if reward or motivational factors are driving the reduction in dual-task interference observed in Experiment 1, then visual search training should result in reductions in the size of the AB that are akin to those seen in the relevant- and irrelevant-training conditions.
We were also interested in examining the extent to which training-related reductions in dual-task interference generalized to AB paradigms employing different classes of stimuli. To do so, participants completed three AB paradigms during the pre- and posttraining sessions. The first AB paradigm was the categorical-AB paradigm used in Experiment 1. The second paradigm also used the same digit distractors and T1 demands, but required the detection of the presence or absence of the digit 2 for T2 (detection AB). The third AB paradigm required the identification of one of four possible targets for T1 and T2, but involved indoor and outdoor scene images for the targets and scrambled scenes for the distractors (scene AB). If training reduces the size of the AB regardless of the stimuli employed during training, then greater improvements should be observed for early lags relative to later lags, and this should be consistent across AB paradigms. In contrast, if training benefits are stronger for particular stimuli or task demands, then training benefits should differ across paradigms.
Method
Participants
The participants were recruited according to the criteria detailed in Experiment 1 (see Table 4). A total of 61 participants were included in the study. Participants were randomly assigned to the relevant, irrelevant, visual-search, and no-training control groups. Of these, eight participants (N = 2 from each group) failed to achieve a score above floor (0) on at least one of the T2 | T1 measures after guess correction and were subsequently removed from the analysis.
Materials and procedure
During the pretraining session, all participants completed three AB paradigms (categorical-AB, detect-AB, and scene-AB; see Table 5 for a summary), with the order in which these paradigms were performed counterbalanced across participants. Given that participants had reached asymptotic performance at around 2,500 training trials in Experiment 1, the participants in the relevant, irrelevant, and visual-search training groups completed 2,970 training trials over three consecutive days in Experiment 2. Following the training sessions, or after three days for the control group, participants were retested on the three AB paradigms (posttraining session).
AB paradigm
The structure of the AB paradigms was the same as that described for Experiment 1, except that no white stickers were placed on the response keys. Since all participants had previously interacted with the keyboards, and therefore already held some motor associations with the keys, it was deemed to make little difference whether or not the keys were covered. Furthermore, T2 could be presented at lags of 2, 3, 4, 5, and 7. For each paradigm, the experimental trials consisted of a randomized presentation of 40 repetitions of the five lag trial types, resulting in 200 experimental trials.
The categorical-AB paradigm was the same as had been presented in Experiment 1. The detection-AB paradigm was identical to the categorical-AB paradigm, except that for T2, participants were required to detect whether or not the digit 2 had been present in the RSVP stream. For the third AB paradigm (scene AB), T1 was drawn from four possible indoor scenes, and T2 was drawn from four possible outdoor scenes (taken from Marois et al., 2004). The target scenes were presented among scrambled scene distractors. All of the scene images subtended 11° of visual angle at a viewing distance of 57 cm. For a schematic representation of the detection- and scene-AB paradigms, refer to Fig. 5.
Schematic representations of trials for the detect-AB and scene-AB paradigms in Experiment 2. All participants completed the three AB paradigms (categorical-AB, detect-AB, and scene-AB) at both pretraining and posttraining. Participants assigned to the relevant, irrelevant, and visual-search training groups completed 2,970 training trials between pre- and posttraining. Control participants had a no-training three-day interval between pre- and posttraining. (AB = attentional blink; T1 = Target 1, T2 = Target 2)
Training
The structure of the sessions was identical to that in Experiment 1, except that the participants in the training conditions completed 2,970 training trials over three sessions. Each session consisted of five blocks of 198 trials. Instead of the presentation of a single letter or colored disc, visual-search training involved the presentation of a target (the letter T) among 8, 12, or 16 randomly placed distractors (Ls). The distractor set sizes were presented equally often, and the presentation order was randomized within the training session. The target stimulus could be oriented to 90° or 270°. Participants were required to identify the orientation of the target using the “A” and “Z” keys on a standard keyboard. The search array was displayed for 3 s or until the participant’s response. Participants were instructed to respond with the index and middle fingers of their dominant hand.
In order to encourage quick and accurate responses, participants were provided with performance feedback (as described in Exp. 1) at the end of every block of training trials (five feedbacks per session).
Results
Training
For the relevant and irrelevant training regimens, RT data were entered into a 2 (group) × 15 (block) mixed ANOVA. The results replicated those of Experiment 1, with a main effect of session [F(14, 350) = 10.75, MSE = .007, p < .001, η p 2 = .30]. RTs were slower on the first training block (mean = 745 ms, SD = 281 ms) than on the last training block (mean = 562, SD = 93). We found a significant main effect of group [F(1, 25) = 6.8, MSE = .093, p = .02, η p 2 = .21], with the relevant-training group responding faster (mean = 561 ms, SD = 78 ms) than the irrelevant-training group (mean = 636 ms, SD 15 ms). Crucially, no significant Group × Session interaction emerged (p = .14), suggesting that comparable training benefits were observed for both groups. For the accuracy data, we observed a significant main effect of block [F(14, 350) = 1.94, MSE = .009, p = .02, η p 2 = .07], which most likely reflected minor speed–accuracy trade-offs for some participants in the final blocks of training (first block mean = 96 %, SD = 3 %; final block mean = 90 %, SD = 18 %).
The RT data from the visual-search training regimen were entered into a 3 (set size) × 15 (block) repeated measures ANOVA. We found a significant Set Size × Block interaction [F(28, 336) = 4.32, MSE = .001, p < .001, η p 2 = .26]. Visual inspection of the data (see Fig. 6) revealed that this was due to a smaller RT difference for the search arrays containing 12 and 16 distractors at the beginning of training, relative to the remaining training blocks. Importantly, a main effect of session [F(14, 168) = 36.52, MSE = .011, p < .001, η p 2 = .75] indicated that responses were faster in the final block (mean = 749 ms, SD = 92 ms) than in the first block (mean = 1,126 ms, SD = 276 ms). The accuracy data were calculated across set sizes for each block and were subject to a repeated measures ANOVA with Block as the within-subjects factor. The effect of block failed to reach significance (p = .11), indicating that accuracy was consistent across blocks. Accuracy was consistently high (mean = 96 %, SD = 5 %) across the 15 training blocks.
Response times (RTs) for the relevant, irrelevant (top panel), and visual-search (bottom panel) training groups over the 15 training blocks of Experiment 2. Error bars represent 95 % confidence intervals of the means (based on the within-group Subject × Block error term for the relevant- and irrelevant-training groups, and on the within-group Subject × Block × Set Size error term for the visual-search group; Masson & Loftus, 2003). Accuracy was consistently high across the training blocks. RTs for the relevant- and irrelevant-training groups replicated those from Experiment 1
AB paradigms
Again, although we analyzed accuracy for both T1 and T2, our crucial measure of training benefits pertained to reductions in the magnitude of the AB. The guess-correction and log-transformation procedures were the same as for Experiment 1, with the exception of the accuracy data for T2 | T1 in the detection-AB task. For these data, participants’ overall false alarm (FA) rate was subtracted from the T2 | T1 accuracy at each lag (Dale, Dux, & Arnell, 2013) [overall mean pretraining FA = 8 % (SD = 7 %), overall mean posttraining FA = 6 % (SD = 6 %)] prior to log transformation). We found no significant differences between the groups in FAs for the pre- and posttraining sessions (both ps > .2).
T1
A 4 (group) × 3 (AB paradigm) × 2 (session) × 5 (lag) mixed ANOVA revealed a significant four-way interaction [F(24, 392) = 1.83, MSE = .001, p = .01, η p 2 = .1]. We did not attempt to interpret this interaction, since our focus was on T2 performance. However, visual inspection of the data showed that, although accuracy was generally high across the conditions, accuracy at lag 7 was conspicuously higher for the irrelevant group than for the other groups, which most likely contributed to the interaction. Accuracy across the three paradigms improved from the pretraining session [mean = 79.5 % (1.89), SD = 15.2 % (0.07)] to the posttraining session [mean = 89.1 % (1.95), SD = 9.5 % (0.07); F(1, 49) = 47.78, MSE = .025, p < .001, η p 2 = .49]. We observed a main effect of lag [F(4, 196) = 4.25, MSE = .002, p = .003, η p 2 = .08], with T1 accuracy being lower when T2 was presented at lag 2 [mean = 83.2 % (1.913), SD = 13 % (0.05)] than at lag 4 [mean = 84.6 % (1.92), SD = 13.6 % (0.06), p = .04], lag 5 [mean = 85.3 % (1.924), SD = 13.1 % (0.06), p = < .001], and lag 7 [mean = 85.2 % (1.924), SD = 13.2 % (0.05), p = .003]. T1 accuracy was also lower when T2 was presented at lag 3 [mean = 84.2 % (1.917), SD = 13.9 % (0.06)] relative to lag 5 (p = .02). Finally, a significant main effect of paradigm was apparent [F(2, 98) = 4.72, MSE = .016, p = .011]: Accuracy for T1 was higher in the detect-AB paradigm [mean = 86.6 % (1.93), SD = 11.6 % (0.04)] than in the categorical-AB paradigm [mean = 84.1 % (1.92), SD = 13.6 % (0.06), p = .006] or the scene-AB paradigm [mean = 82.8 % (1.91), SD = 14.5 % (0.07), p = .015].
T2 | T1
Given the results of Experiment 1, the goal of the present T2 | T1 analysis was to determine which groups demonstrated a significant Session × Lag interaction (driven by significant changes at early lags rather than at later lags (MacLean & Arnell, 2012)). If general motivational factors drove the attenuation of the AB observed in Experiment 1, then all three training groups (relevant, irrelevant, and visual search) should show a significant Session × Lag interaction. If a specific processing mechanism that is tapped by the relevant and irrelevant training regimens is responsible for attenuating the AB, then only these two groups should show a significant Session × Lag interaction. Furthermore, if training benefits are not specific to the stimuli that were trained, then the Session × Lag interactions should be consistent across paradigms.
Although the omnibus 4 (group) × 3 (paradigm) × 2 (session) × 5 (lag) mixed ANOVA did not reveal a significant four-way interaction [F(8, 32) = 1.03, MSE = .007, p = .4], we had strong a-priori hypotheses regarding the expected interactions for the training and control groups. Given our a-priori predictions, a 3 (paradigm) × 2 (session) × 5 (lag) repeated measures ANOVA was conducted for each group. Consistent with Experiment 1, both the relevant- and irrelevant-training groups showed a significant Session × Lag interaction [relevant training, F(4, 48) = 9.33, MSE = .006, p < .001, η p 2 = .44; irrelevant training, F(4, 52) = 4.63, MSE = .006, p = .003, η p 2 = .26]. Importantly, this effect did not interact with paradigm for either group [relevant training, F(8, 96) = 1.6, MSE = .008, p = .14; irrelevant training, F(8, 96) = 0.928, MSE = .007, p = .5], suggesting that the AB was consistently attenuated across all three AB paradigms. Simple-effects comparisons of sessions at each level of lag (using a Bonferroni-adjusted alpha of .01 to correct for comparisons across five lags) revealed significant increases in accuracy for the relevant-training group at lag 2 [pretraining mean = 49.88 % (1.62), SD = 22.3 % (0.31); posttraining mean = 67.4 % (1.8), SD = 22.2 % (0.17); η p 2 = .59], lag 3 [pretraining mean = 61.72 % (1.73), SD = 26.3 % (0.26); posttraining mean = 81.2 % (1.9), SD = 16.4 % (0.11); η p 2 = .62], and lag 4 [pretraining mean = 72.08 % (1.82), SD = 24.3 % (0.22); posttraining mean = 84.84 % (1.92), SD = 14.7 % (0.09); η p 2 = .59, all ps = .001]. The same analysis for the irrelevant-training group revealed significant increases in T2 | T1 accuracy at lag 3 [pretraining mean = 57.59 % (1.7), SD = 24.3 % (0.26); posttraining mean = 75.41 % (1.84), SD = 22.4 % (0.22); η p 2 = .57], and at lag 4 [pretraining mean = 66.48 % (1.79), SD = 20.7 % (0.2); posttraining mean = 81.44 % (1.89), SD = 16.75 % (0.16); η p 2 = .64, both ps = .001]. These results suggest that AB magnitude was attenuated across the three AB paradigms for both the relevant-training and irrelevant-training groups (see Fig. 7). In contrast, no significant Session × Lag interaction was found for either the visual-search training group [F(4, 48) = 1.32, MSE = .005, p = .28, η p 2 = .10] or the control group [F(4, 48) = 2.23, MSE = .011, p = .08, η p 2 = .15]; therefore, the evidence was insufficient to conclude that AB magnitude was attenuated for either of these groups.
Pre- and posttraining Target 2 given Target 1 (T2 | T1) accuracy (guess-corrected and log-transformed) in the categorical-AB, detect-AB, and scene-AB paradigms, as well as for the overall Session × Lag interaction (across columns) for the relevant, irrelevant, visual-search, and no-training control groups (rows) in Experiment 2. T2 | T1 accuracy is plotted across lags. Both the relevant- and irrelevant-training groups showed a reduced AB magnitude posttraining, evidenced by greater accuracy changes at earlier than at later lags. This improvement was consistent across the three AB tasks. This effect was not present in the control group or the visual-search training groups. Error bars represent 95 % confidence intervals of the means (calculated with the Subject × Session × Lag error term; Masson & Loftus, 2003). AB = attentional blink. ** p < .005
Since we did not find a significant interaction between the groups in the omnibus Group × Task × Session × Lag ANOVA, evidence was not sufficient to conclude that the groups differed from one another. Given the suggestive pattern of results above found for the relevant- and irrelevant-training groups, we plotted difference scores (T2 | T1 performance at each lag, post – pre) for each paradigm separately for each group (see Fig. 8). The performance changes for the relevant-training group are indicative of a general decrease in AB magnitude across the three tasks (evidenced by a greater change at early relative to later lags), with the largest improvements occurring for the detect-AB paradigm. Visual inspection of the data suggests that this pattern of results may also have occurred for the irrelevant-training group, and potentially for the no-training control group. The performance changes for the visual-search group suggest a more mixed pattern of change.
Guess-corrected (GC), log-transformed accuracy (Acc) difference scores at each lag for each attentional blink (AB) task of Experiment 2. Each panel represents one group. Error bars represent 95 % confidence intervals of the means (calculated with the Subject × Task × Session × Lag error term; Masson & Loftus, 2003). ** p < .005, referring to the Task × Session × Lag interaction
Given that the patterns of results depicted in Fig. 8 are suggestive of performance changes that may have differed across the groups, and that statistical power is an essential element for determining the efficacy of cognitive-training regimens (see, e.g., Redick et al., 2013), we decided to combine the categorical-AB data from Experiments 1 and 2 in a follow-up analysis. The data from the relevant, irrelevant, and no-training control groups were entered into a 3 (group) × 2 (session) × 4 (lag [2, 3, 5, and 6 from Exp. 1, and 2, 3, 5, and 7 from Exp. 2]) repeated measures ANOVA. A significant Session × Lag × Group interaction emerged [F(6, 201) = 2.25, MSE = .003, p = .04, η p 2 = .06]. Subsequent analyses revealed a significant Group × Session × Lag interaction when the relevant group were compared to the control group [F(3, 132) = 2.76, MSE = .003, p = .047, η p 2 = .06]. This was driven by a significant Session × Lag interaction for the relevant-training group [F(3, 66) = 9.67, MSE = .004, p < .001, η p 2 = .31] that was not present in the control group [F(3, 66) = 2.54, MSE = .003, p = .06, η p 2 = .10]. Although some evidence supported attenuation of the AB across sessions in the control condition, significantly greater attenuation was observed in the relevant-training condition (see Fig. 9). This result shows unequivocally that relevant training reduces dual-task interference in the AB paradigm to a greater extent than does no training (see Fig. 9).
Guess-corrected (GC), log-transformed accuracy (Acc) at pre- (top left) and post- (top right) training for the categorical-AB paradigm, combined across Experiments 1 and 2. Bottom panel: Difference scores at each lag for each group. Error bars represent 95 % confidence intervals of the means (calculated with the Subject × Session × Lag error term; Masson & Loftus, 2003). AB = attentional blink
In contrast, any conclusions regarding irrelevant training were less clear-cut. A follow-up analysis did not reveal a significant Group × Session × Lag interaction when the irrelevant-training group was compared to the control group [F(3, 135) = 2.28, MSE = .003, p = .08, η p 2 = .05], nor when the irrelevant-training group was compared to the relevant group [F(3, 135) = 1.74, MSE = .003, p = .16, η p 2 = .04], although when it was considered in isolation, the irrelevant-training group did show a significant Session × Lag interaction [F(3, 69) = 5.40, MSE = .003, p = .002, η p 2 = .19]. Therefore, with the present data, we do not have the evidence to conclude whether or not the reductions in the size of the AB observed in the irrelevant group were different from those found for either the relevant-training or the no-training control group.
Discussion
Although our a-priori analysis indicated that relevant and irrelevant training resulted in a significant attenuation of the AB, our follow-up analysis combining the data across both experiments only confirmed significant differences between the relevant-training group and the no-training control group. We can therefore conclude that training on a relevant speeded, unmasked task does reduce dual-task interference in the AB paradigm. This has implications for our understanding of bottlenecks in information processing and for the design of cognitive-training regimens. Both of these issues will be revisited in the General discussion.
Furthermore, although our a-priori analysis indicated that the irrelevant-training group showed significant attenuation of the AB, our follow-up analysis did not confirm this to be significantly different from that observed in either the no-training control group or the relevant-training group. Therefore, strong conclusions about the effect of irrelevant training on dual-task performance in the AB paradigm cannot be drawn at this time. Further work assessing training transfer across stimuli at the level of sensory consolidation should utilize more sensitive measures in order to test whether training on task-irrelevant stimuli can result in transferable sensory-consolidation benefits.
General discussion
Over the course of two experiments, we tested the generalizability of training benefits for both the PRP and AB paradigms. In Experiment 1, participants performed a PRP paradigm involving a data-unlimited, speeded letter discrimination task followed by, at either a short or a long SOA, a speeded auditory–vocal task. In the AB paradigm, participants were presented with an RSVP stream of digits, in which two target letters were embedded at various lags. Both letters had to be reported at the end of the trial, each with an unspeeded response. The first letter was drawn from the same letter set as that used in Task 1 of the PRP paradigm and used identical responses, whereas the second was drawn from a different set of letters, using different response keys. Therefore, Task 1 in the AB paradigm was the same as that used in the PRP paradigm, except that the stimulus was data-limited, and the response was delayed and unspeeded. Three training regimens were employed: relevant training, irrelevant training, and control. In the relevant-training group participants practiced the data-unlimited speeded letter discrimination task employed as Task 1 in the PRP paradigm. In the irrelevant-training condition an unrelated data-unlimited speeded visual–manual task was practiced. Finally, in the control group no practice was provided. The results showed that a training regimen that improves performance on the PRP paradigm also improves AB performance, as the group that practiced the letter discrimination task showed attenuated PRPs and ABs, relative to a control group that did not receive any training. Furthermore, the finding that the AB (and not the PRP) was attenuated for the group who trained on stimuli unrelated to those used at test, relative to a no-training control group, suggested that training benefits at the level of sensory consolidation may be transferable across stimulus–response mappings.
A second experiment explored the generalizability of training to attenuate the AB, both across stimuli and tasks. A second goal of this experiment was to constrain the locus of improvement that gave rise to AB performance changes with training. Participants completed three AB paradigms, each involving different stimuli and task demands, and were assigned to relevant, irrelevant, visual-search training, or to a no-training control group. In line with the findings of Experiment 1, a-priori comparisons suggested that both the relevant- and irrelevant-training groups showed significant attenuation of the AB, whereas the visual-search training and control groups did not. Furthermore, the AB attenuation did not interact with task, suggesting that these two groups had obtained a generalized sensory-consolidation benefit through training. However, AB performance differences were not found between the groups; therefore, the evidence was insufficient to draw conclusions regarding the impact of training on AB performance.
A follow-up analysis that combined data from the two experiments indicated that whereas the relevant-training group showed a larger attenuation of the AB than the control group, the irrelevant-training group did not show an attenuation of the AB that was significantly different from either the relevant-training group or the no-training control group [interestingly, the Session × Lag effect sizes (η p 2) from both experiments were largest for the relevant-training group (Exp. 1, η p 2 = .47; Exp. 2, η p 2 = .44), equal or second-largest for the irrelevant-training group (Exp. 1, η p 2 = .47; Exp. 2, η p 2 = .26), followed by the no-training control (Exp. 1, η p 2 = .08; Exp. 2, η p 2 = .15) and visual-search training groups (Exp. 2, η p 2 = .10).] Therefore, although there is evidence that the relevant training regimen modulated AB performance, this evidence is currently insufficient to draw conclusions regarding performance changes caused by the irrelevant training regimen.
We found consistent evidence that improving the efficiency of stimulus–response mappings on a speeded, data-unlimited task does result in improved performance on the same stimuli in an unspeeded, data-limited task. The evidence regarding transfer of training benefits across stimulus–response mappings under data-limited conditions was mixed. It may be that the reduction of the total number of training trials from Experiment 1 to Experiment 2 caused the absence of group differences in the second experiment. In short, it is possible that improvements to sensory consolidation are greater when the training regimen is longer. Secondly, it is possible that with shorter training regimens, the transferable benefit is subtle, and incremental to the benefit that can be gained from repeating an AB task over two sessions. Alternately, it may be that no training benefit transfers across stimulus–response mappings. Future research should employ more sensitive measures of sensory consolidation in order to disentangle these possibilities.
Collectively, these results have implications for our understanding of the central bottleneck, the AB, and for the design of cognitive training regimens. These implications are discussed in turn below.
The central bottleneck
The findings from Experiments 1 and 2 have interesting implications for conceptualizations of the central bottleneck thought to be responsible for interference in both the PRP and the AB. The present data demonstrate that a training regimen that has been shown to improve response selection can also result in sensory-consolidation improvements. Becoming more efficient at executing specific stimulus–response mappings benefits limitations in both sensory consolidation and response selection, which, in line with other findings (Jolicœur, 1998, 1999; Marti et al., 2011; Tombu et al., 2011), demonstrates that the two processes are closely related. Furthermore, the data show that training benefits can transfer across these processing operations.
Theoretical accounts posit different potential sources of improvement. Firstly, the global workspace model posits that the PRP and AB reflect limitations in the same mental process (which is delayed in the case of the PRP, and is absent for the AB; Dehaene, Sergent, & Changeux, 2003; Marti, Sackur, Sigman, & Dehaene, 2010; Marti et al., 2011). Therefore, training may have tapped a unitary process that underpins the PRP and AB. However, the data presented here are suggestive that training benefits may transfer across stimulus–response mappings in the case of the AB, and not for the PRP. Further experiments are required to investigate this possibility. This finding would be problematic to an account that posits that the two limitations share a unitary cause. Therefore, disentangling whether training benefits transfer across stimulus–response mappings to benefit AB performance would provide an interesting test of the global workspace model’s underlying assumptions regarding the relationship between the PRP and the AB.
In contrast, Jolicœur and Dell’Acqua’s (1998) short-term consolidation theory postulates that the consolidation of visual information (short-term consolidation, STC) into short-term memory, and response selection are separate mental processes, however both operations require access to the same central resource in order to be performed. Capacity limitations in STC are thought to underlie the AB, whereas limitations in response selection give rise to the PRP effect. Further to this, when STC processes are engaged; response selection processes are either delayed or slowed, which gives rise to increased AB and PRP effects in hybrid paradigms (e.g., Jolicœur, 1999). Training an unmasked, speeded task may drive improvements in both STC and response selection, or the efficiency of the common resource that engages both processes. Future research should aim to tease apart these possibilities.
Locus of AB improvements
The present findings are also informative as to theoretical accounts of the AB. We found that increasing the efficiency of responses to unmasked stimuli attenuated AB magnitude. The boost-and-bounce (Olivers & Meeter, 2008) account posits that T2 impairments are the result of transitions between attention boosts (or excitatory processes) for targets and bounces (or inhibitory processes) for post-T1 distractors rather than capacity limitations in target processing. It is not readily identifiable how increased efficiency under unmasked conditions would affect this transitory period unless the degree of bounce depends upon the speed with which targets can be processed. This would suggest that the transition depends in part upon the capacities required for target processing, which is problematic to the boost and bounce claim that the AB is not a capacity problem. Overall, the results appear more reconcilable with accounts that posit a role for target processing efficiency in the AB.
Implications for cognitive-training regimens
Overall, the results indicate that when training data-unlimited speeded sensorimotor tasks, it is possible to generalize training benefits to data-limited, unspeeded conditions in order to overcome dual-task limitations. Whether sensory-consolidation benefits generalize across stimuli has not been fully demonstrated here, and the idea needs to be tested further. It has previously been demonstrated that training benefits can generalize across stimuli at the locus of response selection, when trained stimulus–response mappings are categorically related to untrained stimulus–response mappings (Pashler & Bayliss, 1991). Previous investigations into the impact of cognitive-training regimens have yielded intensive debate as to whether generalization of training benefits can be achieved (Bergman Nutley et al., 2011; Diamond & Lee, 2011; Jaeggi, Buschkuehl, Jonides, & Perrig, 2008; Johnstone, Roodenrys, Phillips, Watt, & Mantz, 2010; Klingberg, 2010; Lustig, Shah, Seidler, & Reuter-Lorenz, 2009; Morrison & Chein, 2011; Owen et al., 2010; Redick et al., 2013; Tang & Posner, 2009; Tang, Yang, Leve, & Harold, 2012). The present findings, when considered in conjunction with those reported by Pashler and Bayliss (1991), demonstrate that the generalizability of training benefits can transfer both across and within processing operations (i.e., within response selection, and across to sensory consolidation), depending on the training regimens and transfer tests that are employed. Therefore, the present findings highlight the importance of identifying target-processing operations, and the range of cognitive skills that are served by those processing operations, when designing cognitive programs that aim for generalizability of training benefits.
Conclusions
Over the course of two experiments, we have explored the generalizability of training to overcome limitations caused by a central bottleneck in information processing. We found that a training regimen known to improve PRP performance can also improve AB performance. We also found some evidence that training unmasked stimuli that are different from those employed at test may also benefit AB performance; however, further research will be required before a stronger conclusion can be reached.
We observed that this training benefit extends across AB tasks using varying task demands. This suggests that AB attenuation can result from improvements to a target categorization mechanism that operates under both data-limited and data-unlimited conditions. This finding is problematic for theories of the AB that do not posit a causal role for the efficiency of target processing. The present findings also suggest that training benefits can generalize across different levels of information processing, which is informative for the design of cognitive-training regimens.
Notes
A hybrid AB–PRP paradigm (speeded Task 1, delayed Task 2) was also administered during the pre- and posttraining sessions. The administration of the three tasks (AB, PRP, hybrid AB–PRP) was fully counterbalanced across testing sessions and participants. Two participants responded at chance levels on the hybrid AB–PRP paradigm. Since this resulted in the loss of 1/5th of a data set from the irrelevant group, and consequently reduced our power to detect a significant 3 (group) × 2 (session) × 4 (lag) interaction to 55 %, we do not present the analysis of this task here.
References
Arnell, K. M., & Duncan, J. (2002). Separate and shared sources of dual-task cost in stimulus identification and response selection. Cognitive Psychology, 44, 105–147. doi:10.1006/cogp.2001.0762
Arnell, K. M., Helion, A. M., Hurdelbrink, J. A., & Pasieka, B. (2004). Dissociating sources of dual-task interference using human electrophysiology. Psychonomic Bulletin & Review, 11, 77–83. doi:10.3758/bf03206464
Arnell, K. M., Howe, A. E., Joanisse, M. F., & Klein, R. M. (2006). Relationships between attentional blink magnitude, RSVP target accuracy, and performance on other cognitive tasks. Memory & Cognition, 34, 1472–1483. doi:10.3758/bf03195912
Bergman Nutley, S., Söderqvist, S., Bryde, S., Thorell, L. B., Humphreys, K., & Klingberg, T. (2011). Gains in fluid intelligence after training non-verbal reasoning in 4-year-old children: A controlled, randomized study. Developmental Science, 14, 591–601. doi:10.1111/j.1467-7687.2010.01022.x
Brainard, D. H. (1997). The psychophysics toolbox. Spatial Vision, 10, 433–436. doi:10.1163/156856897x00357
Brisson, B., & Jolicœur, P. (2007). Cross-modal multitasking processing deficits prior to the central bottleneck revealed by event-related potentials. Neuropsychologia, 45, 3038–3053. doi:10.1016/j.neuropsychologia.2007.05.022
Choi, H., Chang, L.-H., Shibata, K., Sasaki, Y., & Watanabe, T. (2012). Resetting capacity limitations revealed by long-lasting elimination of attentional blink through training. Proceedings of the National Academy of Sciences, 109, 12242–12247. doi:10.1073/pnas.1203972109
Chun, M. M., & Potter, M. C. (1995). A two-stage model for multiple target detection in rapid serial visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 21, 109–127. doi:10.1037/0096-1523.21.1.109
Cohen, J. (1988). Statistical power analysis for the behavioural sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
Dale, G., Dux, P. E., & Arnell, K. M. (2013). Individual differences within and across attentional blink tasks revisited. Attention, Perception, & Psychophysics, 75, 456–467. doi:10.3758/s13414-012-0415-8
Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70, 200–227. doi:10.1016/j.neuron.2011.03.018
Dehaene, S., Kerszberg, M., & Changeux, J. P. (2001). A neuronal model of a global workspace in effortful cognitive tasks. In P. C. Marijuan (Ed.), Cajal and consciousness: Scientific approaches to consciousness on the centennial of Ramon y Cajal’s Textura (Vol. 929) (pp. 152–165). New York, NY: New York Academy of Sciences.
Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79, 1–37. doi:10.1016/s0010-0277(00)00123-2
Dehaene, S., Sergent, C., & Changeux, J. (2003). A neuronal network model linking subjective reports and objective physiological data during conscious perception. Proceedings of the National Academy of Sciences, 100, 8520–8525. doi:10.1073/pnas.1332574100
Di Lollo, V., Kawahara, J., Ghorashi, S. M., & Enns, J. T. (2005). The attentional blink: Resource depletion or temporary loss of control? Psychological Research, 69, 191–200. doi:10.1007/s00426-004-0173-x
Diamond, A., & Lee, K. (2011). Interventions shown to aid executive function development in children 4 to 12 years old. Science, 333, 959–964. doi:10.1126/science.1204529
Dux, P. E., Ivanoff, J., Asplund, C. L., & Marois, R. (2006). Isolation of a central bottleneck of information processing with time-resolved fMRI. Neuron, 52, 1109–1120. doi:10.1016/j.neuron.2006.11.009
Dux, P. E., & Marois, R. (2009). The attentional blink: A review of data and theory. Attention, Perception, & Psychophysics, 71, 1683–1700. doi:10.3758/app. 71.8.1683
Dux, P. E., Tombu, M. N., Harrison, S., Rogers, B. P., Tong, F., & Marois, R. (2009). Training improves multitasking performance by increasing the speed of information processing in human prefrontal cortex. Neuron, 63, 127–138. doi:10.1016/j.neuron.2009.06.005
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175–191. doi:10.3758/bf03193146
Hesselmann, G., Flandin, G., & Dehaene, S. (2011). Probing the cortical network underlying the psychological refractory period: A combined EEG-fMRI study. NeuroImage, 56, 1608–1621. doi:10.1016/j.neuroimage.2011.03.017
Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Perrig, W. J. (2008). Improving fluid intelligence with training on working memory. Proceedings of the National Academy of Sciences, 105, 6829–6833. doi:10.1073/pnas.0801268105
Jiang, Y. (2004). Resolving dual-task interference: An fMRI study. NeuroImage, 22, 748–754. doi:10.1016/j.neuroimage.2004.01.043
Johnstone, S. J., Roodenrys, S., Phillips, E., Watt, A. J., & Mantz, S. (2010). A pilot study of combined working memory and inhibition training for children with AD/HD. Attention Deficit and Hyperactivity Disorders, 2, 31–42.
Jolicœur, P. (1998). Modulation of the attentional blink by on-line response selection: Evidence from speeded and unspeeded Task1 decisions. Memory & Cognition, 26, 1014–1032. doi:10.3758/bf03201180
Jolicœur, P. (1999). Concurrent response-selection demands modulate the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 25, 1097–1113. doi:10.1037/0096-1523.25.4.1097
Jolicœur, P., & Dell’Acqua, R. (1998). The demonstration of short-term consolidation. Cognitive Psychology, 36, 138–202. doi:10.1006/cogp.1998.0684
Jolicœur, P., Dell’Acqua, R., & Crebolder, J. M. (2001). The attentional blink bottleneck. In K. Shapiro (Ed.), The limits of attention: Temporal constraints in human information processing (pp. 82–99). New York: Oxford University Press.
Kamienkowski, J. E., Pashler, H., Dehaene, S., & Sigman, M. (2011). Effects of practice on task architecture: Combined evidence from interference experiments and random-walk models of decision making. Cognition, 119, 81–95. doi:10.1016/j.cognition.2010.12.010
Klein, C., Arend, I. C., Beauducel, A., & Shapiro, K. L. (2011). Individuals differ in the attentional blink: Mental speed and intra-subject stability matter. Intelligence, 39, 27–35. doi:10.1016/j.intell.2010.11.004
Klingberg, T. (2010). Training and plasticity of working memory. Trends in Cognitive Sciences, 14, 317–324. doi:10.1016/j.tics.2010.05.002
Kranczioch, C., Debener, S., & Engel, A. K. (2003). Event-related potential correlates of the attentional blink phenomenon. Cognitive Brain Research, 17, 177–187. doi:10.1016/s0926-6410(03)00092-2
Kranczioch, C., Debener, S., Schwarzbach, J., Goebel, R., & Engel, A. K. (2005). Neural correlates of conscious perception in the attentional blink. NeuroImage, 24, 704–714. doi:10.1016/j.neuroimage.2004.09.024
Logan, G. D., & Gordon, R. D. (2001). Executive control of visual attention in dual-task situations. Psychological Review, 108, 393–434. doi:10.1037/0033-295x.108.2.393
Lustig, C., Shah, P., Seidler, R., & Reuter-Lorenz, P. A. (2009). Aging, training, and the brain: A review and future directions. Neuropsychology Review, 19, 504–522. doi:10.1007/s11065-009-9119-9
MacLean, M. H., & Arnell, K. M. (2012). A conceptual and methodological framework for measuring and modulating the attentional blink. Attention, Perception, & Psychophysics, 74, 1080–1097. doi:10.3758/s13414-012-0338-4
MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An integrative review. Psychological Bulletin, 109, 163–203. doi:10.1037/0033-2909.109.2.163
Maki, W., & Padmanabhan, G. (1994). Transient suppression of processing during rapid serial visual presentation: Acquired distinctiveness of probes modulates the attentional blink. Psychonomic Bulletin & Review, 1, 499–504. doi:10.3758/bf03210954
Maquestiaux, F., Laguë-Beauvais, M., Bherer, L., & Ruthruff, E. (2008). Bypassing the central bottleneck after single-task practice in the psychological refractory period paradigm: Evidence for task automatization and greedy resource recruitment. Memory & Cognition, 36, 1262–1282. doi:10.3758/mc.36.7.1262
Marois, R., Chun, M. M., & Gore, J. C. (2000). Neural correlates of the attentional blink. Neuron, 28, 299–308.
Marois, R., & Ivanoff, J. (2005). Capacity limits of information processing in the brain. Trends in Cognitive Sciences, 9, 296–305. doi:10.1016/j.tics.2005.04.010
Marois, R., Yi, D. J., & Chun, M. M. (2004). The neural fate of consciously perceived and missed events in the attentional blink. Neuron, 41, 465–472.
Marti, S., Sackur, J., Sigman, M., & Dehaene, S. (2010). Mapping introspection’s blind spot: Reconstruction of dual-task phenomenology using quantified introspection. Cognition, 115, 303–313. doi:10.1016/j.cognition.2010.01.003
Marti, S., Sigman, M., & Dehaene, S. (2011). A shared cortical bottleneck underlying attentional blink and psychological refractory period. NeuroImage, 59, 2883–2898. doi:10.1016/j.neuroimage.2011.09.063
Masson, M. E., & Loftus, G. R. (2003). Using confidence intervals for graphically based data interpretation. Canadian Journal of Experimental Psychology, 57, 203–220. doi:10.1037/h0087426
Meyer, D. E., & Kieras, D. E. (1997a). A computational theory of executive cognitive processes and multiple-task performance: Part 2. Accounts of psychological refractory-period phenomena. Psychological Review, 104, 749–791. doi:10.1037/0033-295x.104.4.749
Meyer, D. E., & Kieras, D. E. (1997b). A computational theory of executive cognitive processes and multiple-task performance: Part I. Basic mechanisms. Psychological Review, 104, 3–65. doi:10.1037/0033-295x.104.1.3
Meyer, D. E., Kieras, D. E., Lauber, E. J., Schumacher, E. H., Glass, J. M., Zurbriggen, E. L., & Apfelblat, D. (1995). Adaptive executive control: Flexible multiple-task performance without pervasive immutable response-selection bottlenecks. Acta Psychologica, 90, 163–190. doi:10.1016/0001-6918(95)00026-q
Morrison, A., & Chein, J. (2011). Does working memory training work? The promise and challenges of enhancing cognition by training working memory. Psychonomic Bulletin & Review, 18, 46–60. doi:10.3758/s13423-010-0034-0
Nakatani, C., Baijal, S., & van Leeuwen, C. (2009). Practice begets the second target: Task repetition and the attentional blink effect. In S. Narayanan (Ed.), Progress in brain research (Vol. 176) (pp. 123–134). Amsterdam, The Netherlands: Elsevier.
Navon, D., & Miller, J. (2002). Queuing or sharing? A critical evaluation of the single-bottleneck notion. Cognitive Psychology, 44, 193–251. doi:10.1006/cogp.2001.0767
Norman, D. A., & Bobrow, D. G. (1975). On data-limited and resource-limited processes. Cognitive Psychology, 7, 44–64. doi:10.1016/0010-0285(75)90004-3
Olivers, C. N. L., Hulleman, J., Spalek, T., Kawahara, J., & Di Lollo, V. (2011). The sparing is far from spurious: Reevaluating within-trial contingency effects in the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 37, 396–408. doi:10.1037/a0020379
Olivers, C. N. L., & Meeter, M. (2008). A boost and bounce theory of temporal attention. Psychological Review, 115, 836–863. doi:10.1037/a0013395
Olivers, C. N. L., & Nieuwenhuis, S. (2006). The beneficial effects of additional task load, positive affect, and instruction on the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 32, 364–379. doi:10.1037/0096-1523.32.2.364
Owen, A. M., Hampshire, A., Grahn, J. A., Stenton, R., Dajani, S., Burns, A. S., & Ballard, C. G. (2010). Putting brain training to the test. Nature, 465, 775–778. doi:10.1038/nature09042
Pashler, H. (1984). Processing stages in overlapping tasks: Evidence for a central bottleneck. Journal of Experimental Psychology: Human Perception and Performance, 10, 358–377. doi:10.1037/0096-1523.10.3.358
Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116, 220–244. doi:10.1037/0033-2909.116.2.220
Pashler, H., & Bayliss, G. (1991). Procedural learning: 1. Locus of practice effects in speeded choice tasks. Journal of Experimental Psychology: Learning, Memory, and Cognition, 1, 20–32. doi:10.1037/0278-7393.17.1.20
Pashler, H., Johnston, J. C., & Ruthruff, E. (2001). Attention and performance. Annual Review of Psychology, 52, 629–651. doi:10.1146/annurev.psych.52.1.629
Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: transforming numbers into movies. Spatial Vision, 10, 437–442. doi:10.1163/156856897X00366
Potter, M. C., & Levy, E. I. (1969). Recognition memory for a rapid sequence of pictures. Journal of Experimental Psychology, 81, 10–15. doi:10.1037/h0027470
Raymond, J. E., Shapiro, K. L., & Arnell, K. M. (1992). Temporary suppression of visual processing in an RSVP task: An attentional blink? Journal of Experimental Psychology: Human Perception and Performance, 18, 849–860. doi:10.1037/0096-1523.18.3.849
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, placebo-controlled study. Journal of Experimental Psychology: General, 142, 359–379. doi:10.1037/a0029082
Ruiz Fernández, S., & Ulrich, R. (2010). Late backward effects in the refractory period paradigm: Effects of Task 2 execution on Task 1 performance. Psychological Research, 74, 378–387. doi:10.1007/s00426-009-0260-0
Ruthruff, E., Hazeltine, E., & Remington, R. W. (2006a). What causes residual dual-task interference after practice? Psychological Research, 70, 494–503. doi:10.1007/s00426-005-0012-8
Ruthruff, E., Johnston, J. C., & Van Selst, M. (2001). Why practice reduces dual-task interference. Journal of Experimental Psychology: Human Perception and Performance, 27, 3–21. doi:10.1037/0096-1523.27.1.3
Ruthruff, E., & Pashler, H. (2001). Perceptual and central interference in dual-task performance. In K. Shapiro (Ed.), The limits of attention: Temporal constraints in human information processing (pp. 100–123). New York, NY: Oxford University Press.
Ruthruff, E., Van Selst, M., Johnston, J. C., & Remington, R. W. (2006b). How does practice reduce dual-task interference: Integration, automatization, or just stage-shortening? Psychological Research, 70, 125–142. doi:10.1007/s00426-004-0192-7
Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D. E., Lauber, E. J., Kieras, D. E., & Meyer, D. E. (2001). Virtually perfect time sharing in dual-task performance: Uncorking the central cognitive bottleneck. Psychological Science, 12, 101–108. doi:10.1111/1467-9280.00318
Schweickert, R. (1985). Separable effects of factors on speed and accuracy: Memory scanning, lexical decision, and choice tasks. Psychological Bulletin, 97, 530–546. doi:10.1037/0033-2909.97.3.530
Sergent, C., Baillet, S., & Dehaene, S. (2005). Timing of the brain events underlying access to consciousness during the attentional blink. Nature Neuroscience, 8, 1391–1400. doi:10.1038/nn1549
Sternberg, S. (1969). The discovery of processing stages: Extensions of Donders’ method. Acta Psychologica, 30, 276–315. doi:10.1016/0001-6918(69)90055-9
Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643–662. doi:10.1037/h0054651
Taatgen, N. A., Juvina, I., Schipper, M., Borst, J. P., & Martens, S. (2009). Too much control can hurt: A threaded cognition model of the attentional blink. Cognitive Psychology, 59, 1–29. doi:10.1016/j.cogpsych.2008.12.002
Tang, Y.-Y., & Posner, M. I. (2009). Attention training and attention state training. Trends in Cognitive Sciences, 13, 222–227. doi:10.1016/j.tics.2009.01.009
Tang, Y.-Y., Yang, L., Leve, L. D., & Harold, G. T. (2012). Improving executive function and its neurobiological mechanisms through a mindfulness-based intervention: Advances within the field of developmental neuroscience. Child Development Perspectives, 6, 361–366. doi:10.1111/j.1750-8606.2012.00250.x
Tombu, M. N., Asplund, C. L., Dux, P. E., Godwin, D., Martin, J. W., & Marois, R. (2011). A unified attentional bottleneck in the human brain. Proceedings of the National Academy of Sciences, 108, 13426–13431. doi:10.1073/pnas.1103583108
Tombu, M., & Jolicœur, P. (2002). All-or-none bottleneck versus capacity sharing accounts of the psychological refractory period phenomenon. Psychological Research, 66, 274–286. doi:10.1007/s00426-002-0101-x
Tombu, M., & Jolicœur, P. (2003). A central capacity sharing model of dual-task performance. Journal of Experimental Psychology: Human Perception and Performance, 29, 3–18. doi:10.1037/0096-1523.29.1.3
Ulrich, R., Fernandez, S. R., Jentzsch, I., Rolke, B., Schroter, H., & Leuthold, H. (2006). Motor limitation in dual-task processing under ballistic movement conditions. Psychological Science, 17, 788–793. doi:10.1111/j.1467-9280.2006.01783.x
Van Selst, M., & Jolicœur, P. (1997). Decision and response in dual-task interference. Cognitive Psychology, 33, 266–307. doi:10.1006/cogp.1997.0662
Van Selst, M., Ruthruff, E., & Johnston, J. C. (1999). Can practice eliminate the psychological refractory period effect? Journal of Experimental Psychology: Human Perception and Performance, 25, 1268–1283. doi:10.1037/0096-1523.25.5.1268
Vogel, E. K., Luck, S. J., & Shapiro, K. L. (1998). Electrophysiological evidence for a postperceptual locus of suppression during the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 24, 1656–1674. doi:10.1037/0096-1523.24.6.1656
Wyble, B., Bowman, H., & Nieuwenstein, M. (2009). The attentional blink provides episodic distinctiveness: Sparing at a cost. Journal of Experimental Psychology: Human Perception and Performance, 35, 787–807. doi:10.1037/a0013902
Zylberberg, A., Fernández Slezak, D., Roelfsema, P. R., Dehaene, S., & Sigman, M. (2010). The brain’s router: A cortical network model of serial processing in the primate brain. PLoS Computational Biology, 6, e1000765. doi:10.1371/journal.pcbi.1000765
Author note
An Australian Research Council Discovery Grant and APD Fellowship (No. DP0986387, to P.E.D.), an Australian Research Council Future Fellowship (No. FT120100033, to P.E.D.), and the ARC-SRI Science of Learning Research Centre (Grant No. SR120300015 to P.E.D.) supported this work. We thank Roger Remington for helpful discussions on this work, and Amy Taylor and Luke Hearne for help with the data collection.
Author information
Authors and Affiliations
Corresponding authors
Rights and permissions
About this article
Cite this article
Garner, K.G., Tombu, M.N. & Dux, P.E. The influence of training on the attentional blink and psychological refractory period. Atten Percept Psychophys 76, 979–999 (2014). https://doi.org/10.3758/s13414-014-0638-y
Published:
Issue Date:
DOI: https://doi.org/10.3758/s13414-014-0638-y