Content uploaded by Chie Nakatani
Author content
All content in this area was uploaded by Chie Nakatani on Jan 28, 2015
Content may be subject to copyright.
Available via license: CC BY-NC-ND 3.0
Content may be subject to copyright.
Curbing the attentional blink: Practice keeps the mind’s eye open
Chie Nakatani
a,b
, Shruti Baijal
b,c
, Cees van Leeeuwen
a,b,
n
a
Department of Experimental Psychology, University of Leuven, Tiensestraat 102, Leuven 3000, Belgium
b
Laboratory for Perceptual Dynamics, RIKEN Brain Science Institute, 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan
c
Centre for Behavioural and Cognitive Sciences University of Allahabad, Allahabad 211002, India
article info
Available online 5 January 2012
Keywords:
Lag 1 sparing
RSVP
ERP
N2
P3
Signal detection
abstract
When two targets, T1 and T2, are embedded in a rapid serial visual presentation of distractors,
successful report of T2 depends on its lag from T1: When T2 is separated by a few distracters, it is likely
to be missed; this phenomenon is known as the Attentional Blink (AB). When T2 is presented
consecutively from T1, T2 is likely to be detected despite the temporal proximity of both targets; this
effect is called Lag-1 sparing. We studied how the Lag-1 sparing and AB effects change with practice.
Observers repeated a typical dual-target-report task over separate days, while behavioral indices and
EEG were recorded. Practice increased the Lag-1 sparing and reduced the AB effects, improving the
sensitivity to T2 while leaving the response criterion unchanged. With improving sensitivity, T2-related
amplitude of P3 and negative deflection (ND), an N2 subcomponent, increased. The latter, especially in
the Lag-1 condition, could not fully be explained by changes in the ratio of the T2-hit and miss trials.
ND usually indicates spatial target selectivity but here reflects the selection of temporally proximal
targets. The effect, therefore, suggests common mechanisms for spatial and temporal selectivity.
Relevance of these results for computational models of the AB is discussed.
&2012 Elsevier B.V. All rights reserved.
1. Introduction
For over two decades, the concept of Attentional Blink (AB) has
kept researchers’ attention occupied, no matter what distractors and
potential new research targets have come their way. The AB effect
refers to a class of perceptual phenomena in which an observer
misses a salient target when another target was presented
200–500 ms earlier. The two targets of AB tasks, T1 and T2, are often
embedded in a sequence of non-target items, of which the stimulus
onset asynchrony (SOA) is in the order of 100 ms. This presentation
method is called rapid serial visual presentation (RSVP). Under these
conditions, the observer can report T1, but often misses T2, as if the
mind’s eye blinked after the first target [1,2]
The AB was originally regarded as a product of capacity limita-
tions somewhere along the flow of the perceptual process: either at
an early, target selection stage [3],oratalate,workingmemory
consolidation stage [4,5]. Accordingly, when T2 is presented just
after T1, T2 is often reported successfully despite the proximity
between both targets. This phenomenon is known as Lag-1 sparing
[6–8]. Lag-1 sparing avoids the capacity limitations through integra-
tion of T1 and T2-related information. As a result, observers often
report that the spared T2 was presented before T1 (an effect called
target order error) or, when T1 and T2 are identical, fail to detect
target repetition (known as repetition blindness). Later studies
reported various situations in which these limitations could be
overcome. For example, a contextual cue for T2 can extend Lag-1
sparing to later lags [9,10], or prevent the AB phenomenon from
occurring [11,12]. These studies suggested that, in being sensitive to
similar contextual manipulations, the AB and Lag-1 phenomena are
two sides of the same coin (or, at least, share main component
processes). These processes were addressed in terms of a context-
sensitive target [9,10] or context-based postponement of attentional
engagement to T2 [11,12]. Besides such context, experimental
manipulations, which prevent observers from over-engaging with
the task, turned out to be very effective in preventing the AB
phenomenon [13–17]. For example, insertion of a task-irrelevant
sound during an AB task dramatically improved T2 report [15].
Whereas a top-down resource shift across process components
could improve the AB instantaneously, improvement in the
operation of individual components may develop gradually over
many trials, as a result of practice. The effect of practice is the
focus of our present study. Practice could improve the process
components, such as, target selection and consolidation, and/or
optimizing the control efficiency on resource allocation across
these sub-processes.
Some studies reported that rather than target-related pro-
cesses, practice improved distractor-related ones. Participants in
Contents lists available at SciVerse ScienceDirect
journal homepage: www.elsevier.com/locate/neucom
Neurocomputing
0925-2312/$ - see front matter &2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.neucom.2011.12.022
n
Corresponding author at: Department of Experimental Psychology, University
of Leuven, Tiensestraat 102, Leuven 3000, Belgium.
E-mail address: cees.vanleeuwen@ppw.kuleuven.be (C. van Leeeuwen).
Neurocomputing 84 (2012) 13–22
Maki and Padmanabhan [18] repeated an AB task for several days.
In their task, consistent mapping was used for targets throughout
the practice period: T1 was an uppercase letter (‘T’) and T2 was a
digit (‘9’); distractors were uppercase letters which were chosen
randomly for each trial. After the practice period, the AB effect,
although not completely abolished, was strongly diminished.
Interestingly, when the non-targets were a mixture of digits and
letters, the magnitude of the AB effect recovered instantaneously
to pre-practice level. Moreover, once distractors were restored to
letters-only, the AB effect immediately returned to the practiced
level. The authors, therefore, concluded that practice improved
performance not through the automatization of target-related
processes, but through the suppression of the distractor set
(suppression of distractors was also reported in Ref. [19]). Braun
[20] observed that practice improved distractor-related compo-
nent processes in a center–peripheral AB task that was modified
from Joseph et al. [21]. In this task, T1 is a white-letter, which is to
be identified out of black-letter distractors, all presented at
the center of the display. T2, on the other hand, is presented in
peripheral visual field with a lag from T1 of 0–655 ms. The task
associated with T2 varied in attentional demand; in the low
attentional demand condition, it was a single feature detection
(orientation odd-ball) task, while in the high attentional demand
condition, it was a feature-conjunction detection (T/L identifica-
tion) task. Three groups of observers participated to the experi-
ment; a group of novices and two groups of experts: task-trained
and RSVP-trained groups. When attentional demand for T2 was
high, all three groups showed the AB effect. On the other hand,
when attentional demand was low, the task-trained and RSVP-
trained groups did not show an AB, but the novices did. Based on
these results, the author claimed that practice improved ‘‘aware-
ness to’’ RSVP stimuli independently of resource allocation to T1
and T2-related task components.
In a reply to Braun [21], Joseph et al. [22] emphasized that
detection of T1 and T2 constitutes a typical dual tasking problem
and proposed to consider practice effect in the distracter-related
and target-related processes from the viewpoint of resource
allocation in dual tasks. In general, task performance increases
with each increase in the amount of resource invested up to some
point before it reaches asymptote. Practice makes the slope of this
resource-performance function steeper (Fig. 1 in Ref. [22]). Based
on a capacity limitation account for dual task performance, the
authors assumed a fixed sum of resources allocated to T1 and T2
tasks. Processing of T1 takes freely from the available resources,
but T2 processes need to operate on whichever resource is left
over. If an observer is highly trained for the T2 task (including the
RSVP component), T2 report would still be successful even if
the resource available for T2 was small. If, on the other hand, the
observer was a novice, T2 would be missed. Joseph et al. [22]
claimed that their framework constitutes a ‘‘unitary architecture’’
for practice effects and task demands.
In the dual-task framework, practice makes more resources
available for T2 processing. This may aid component processes such
as target selection, memory consolidation, or distractor suppression.
With practice, each component could, in principle, increase the
signal/noise ratio of its output and/or accelerate processing time.
Some components might show stronger benefits from practice than
others. To study such effects at the level of component processes we
incorporated EEG measurement, which has sufficiently fine time-
resolution for doing so. Also, using EEG measures enables us to
estimate practice effects on the distractor-related activities without
adding further task conditions to the standard AB task.
Novice participants repeated an AB task over two sessions, which
took place on different days. In both sessions, EEG was recorded
from scalp electrodes. From the EEG record, we computed event
related brain potentials (ERPs). In previous attentional blink studies,
ERP amplitude has been related to the level of stimulus information
processing, e.g., [23,24], while its latency has been considered to
reflect timing of component processes, e.g., [25,26]. A number of
computational models of the AB phenomenon predict modulation of
ERP components or ERP like activity [27–33]. ERP measures, there-
fore, provide us a way to discuss our results in relation to these
models. In order to compensate for some methodological short-
comings of the ERP measures, we incorporated into our study the
EEG amplitude of the RSVP-related frequency band.
2. Methods
2.1. Participants
Thirteen university students (2 men and 11 women, 18–28
year-old, 20.85 year-old on average) from the greater Tokyo area
participated in the experiment. All were right-handed and had
normal or corrected-to-normal vision. They received 1000 yen per
hour in reward for participation. The research ethics committee of
RIKEN had given their approval to the experimental procedure.
2.2. Stimuli and task
Similar to our previous study [34], two targets were reported
from a series of alpha-numeric characters. Stimuli were upper-
case letters and digits, excluding G, I, K, X, 0 (zero), and 1 (one).
Each of them fitted within an area of 1.361by 1.361of visual
angle. Their average luminance value was 11.10 cd/m
2
. The
stimuli were shown in RSVP in each trial; SOA was 100 ms and
ISI was 80 ms, i.e., each stimulus was presented for 20 ms and was
followed by an 80 ms blank screen. One of the stimuli was shown
in blue as the first target (T1). Participants judged if T1 was a
letter or a digit; it actually was a letter or a digit in 50/50% of the
trials. T1 was presented either at the fifth or eighth position. The
rest of the stimuli were alphabets presented in white color
(background color was always gray). A letter ‘O’ was the second
target (T2), which was present or absent 50/50% of the times.
Participants responded whether T2 was present or absent. When
present, T2 occurred as either the 1st, 3rd, or 7th stimulus after T1
(Lag 1, Lag 3 and Lag 7 trials).
At least three distractors (non-target white letters) followed
T2 in the RSVP. Afterwards, participants were asked to identify
the category of T1 (letter or digit) and also to report presence/
absence of T2. Participants were instructed to make the best guess
when they were not sure about their perception. Responses were
made by pressing the right or left button in a button box with
right middle or index finger. The category of T1 was reported first
(letter-right button, digit-left button), then the presence/absence
of T2 was reported (present-right button, absent-left button).
Fig. 1 shows the sequence of events in a trial.
2.3. Equipment and procedure
Stimuli were presented using a CRT display (Trinitron Multi-
Scan G520, SONY, Tokyo, Japan) of which the presentation was
synchronized with the display refresh rate (100 Hz). The display
was placed at eye height, at a distance of 105 cm from the
observer. Participants were seated comfortably in a chair with
armrests. A response key box (Etalcia tenkey box, Elecom, Osaka,
Japan) was placed next to their arm rest, in immediate reach of
the participant. The experiment took place in a sound-attenuated
chamber with dim ceiling lights. The experiment consisted of two
experimental sessions, which were held on two different days (in
average 5.69 day apart). On each day, participants completed one
experimental session which consisted of dual task (T1 and T2
C. Nakatani et al. / Neurocomputing 84 (2012) 13–2214
report were required) and single task (T2 report only) blocks. Each
block had 288 trials and consisted of 144 T2-present trials (48
trials 3 lags) and 144 T2-absent trials in random order. In total,
each participant completed 576 dual-task and 576 single-task trials.
Prior to the first experimental session, instruction and 16 practice
trials were given. Afterwards, the electrodes were attached and EEG
was recorded while participants performed in the experiment. The
experiment took about 2.5 h/day, including instruction and EEG
preparation time.
2.4. EEG recording and processing
The experimental tasks were controlled by a software package
(SuperLab Pro version 4.0, Cedrus Corporation, San Pedro, CA)
using a Windows-XP PC. EEG was recorded with a commercial
EEG recording system (EEG1100, Nihon Kohden, Tokyo, Japan)
using a cap with 19 tin electrodes (ElectroCap, Electro-Cap
International, Inc., Eaton, Ohio). The electrodes were placed on
Fp1, Fp2, F3, Fz, F4, Cz, C3, C4, Pz, P3, P4, O1, O2, F7, F8, T3, T4, T5,
and T6. The ground electrode was placed at the mid-sagittal line
between Fz and Cz. Reference electrodes were placed on left and
right ear lobes, which were digitally linked. EOG electrodes were
attached at right and left temples for horizontal EOG (HEOG), and
at above and below of the left eye for vertical EOG (VEOG). Data
were digitized at 500 Hz (.1–100 Hz analog bandwidth).
Ocular and muscular artifacts were semi-automatically removed
from the EEG record using Independent Component Analysis (ICA.
InfoMax, [35]). Among 19 independent components computed,
those which correlated with EOG, those which showed an EMG-
characteristic frequency/scalp distribution pattern, and those which
showed a prominent power at 50 Hz (AC noise) were identified by
experimenters CN and SB and removed. The artifact-removed EEG
was segmented in the range of 500 to þ1500 from T1 onset.
Segments of T1-incorrect trials were removed from further analyses.
This left, for each participant, 120 or more segments each for
T2-present and -absent conditions. Before averaging, the baseline
was adjusted using the mean voltage value between 500 and 0 ms
of each segment.
3. Results
3.1. Behavioral results
3.1.1. T1 correct and T2 hit9T1 correct rates
T1 was reported correctly with a rate higher than 85% in both
sessions. A 2 (Sessions) by 3 (Lags) repeated-measures ANOVA
showed that the proportion of correct T1 report did not change
over Sessions, Fo1 nor across Lags, Fo1. There was no signifi-
cant interaction of Sessions and Lags, Fo1. Thus, T1 performance
was constant across experimental factors. Likewise, in the T2
single task condition, T2 was correctly detected with a rate higher
than 82% in both sessions. In a 2 3 repeated-measures ANOVA,
the proportion of T2 detection did not change over Sessions, Fo1
nor across Lags, Fo1, and there was no interaction of Sessions
and Lags, Fo1. Crucially, the proportion of T2 detection condi-
tional to correct T1 report (T29T1) in a repeated-measures 2 3
ANOVA showed effects of sessions and lags: Sessions, F(1,12)¼
10.00, p¼.008, Lags, F(2,24)¼13.55, po.001, and interaction
between them, F(2,24)¼4.71, p¼.019. Post-hoc t-tests indicated
that the performance improved with the sessions, in the Lags
1 and 3, but not in the Lag 7 conditions, t(12)¼4.22, p¼.001,
t(12)¼3.18, p¼.008, to1, for Lags 1, 3 and 7, respectively (see
Fig. 2a).
Though T2 report improved with sessions, the AB was observed in
both the first and second sessions. The proportion T29T1 correct was
lower in Lag 3 than 7 conditions in the first session t(12)¼6.26,
po.001, as well as in the second session, t(12)¼3.41, p¼.016, (for
multiple comparisons among lag conditions, probabilities reported
Fig. 1. Course of events in the Attentional Blink task. A sequence of 17–20 stimuli,
one light blue and others white were presented on a uniform gray background
after an initial fixation cross SOA from one stimulus to the next was 100 ms.
Participants reported the category (letter or digit) of the blue stimulus and
presence or absence of letter ‘O’ in the sequence. The light blue stimulus (T1)
preceded the letter ‘O’ (T2) with an SOA of 100, 300 or 700 ms, corresponding to a
Lag of 1, 3, or 7 from T1. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
Fig. 2. Performance, sensitivity and response bias measures. (a) Proportion of
correct target report in single (left) and dual (right) task conditions as a function of
the lag between the first and second target: In the single task condition, the
proportion of successful second target detection did not change from Session 1
(solid line) to Session 2 (broken line). However, in the dual task condition, the
proportion successful detection of the second target conditional to correct report
of the first target (T29T1) was better in Session 2 than in Session 1 at Lags 1 and 3
(*represents alphao.05 and vertical bars show SE). (b) Sensitivity index A
0
and
response bias measure B
00
plotted as a function of the lags: Sensitivity improved
between the sessions in Lags 1 and 3, while the bias did not change.
C. Nakatani et al. / Neurocomputing 84 (2012) 13–22 15
were corrected by the Bonferroni method, hereafter). The proportion
correct was about the same in Lags 1 and 7 in the first session,
t(12)¼1.32, p4.1, and in the second, to1. The differences
between Lags 1 and 3 were significant, t(12)¼5.38, po.001 for the
first session, and t(12)¼3.34, p¼.018 for the second. The high
performance in the Lag 1 condition is the phenomenon known as
Lag-1 sparing. We may conclude that the AB effect was reduced and
Lag-1 sparing was increased from the first to the second sessions.
3.1.2. Sensitivity and response bias measures
Sensitivity and response bias to T2 were estimated from T2 hit
and false alarm (FA) rate in both sessions (Fig. 2b). Estimation was
done using the procedures from Macmillan and Creelman [36].
The A
0
statistic was computed as an index of sensitivity to the
presence or absence of T2. This index represents the area under
the receiver operating characteristic curve (ROC), and is therefore
independent of response bias. A
0
¼.5 means chance level, and
larger A
0
, expresses greater sensitivity. Given that A
0
is a measure
of sensitivity based on ROC area, we chose B
00
, which is also ROC
area-based, as our index of bias. The smaller B
00
is, the more lenient
the decision criterion becomes. Results of a 2 3 repeated-measures
ANOVA on A
0
show that sensitivity significantly increased over
Sessions, F(1,12)¼5.51, p¼.037, and depended on Lags, F(2,24)¼
10.66, po.001. The interaction between Sessions and Lags was
marginally significant, F(2,24)¼2.64, p¼.092. Post-hoc t-tests showed
that sensitivity increased over Sessions in Lag 1, t(12)¼2.48, p¼.029,
and Lag 3, t(12)¼2.87, p¼.014, but not in Lag 7, to1. Results of the
ANOVA using B
00
show that the response bias was only marginally
relaxed over the sessions, F(1,12)¼3.52, p¼.085, for the main
effect of Sessions. Post-hoc paired t-tests showed that the marginal
effect was due to the Lag 7 condition, t(12)¼1.87, p¼.086. The main
effect of Lags reached significance F(2,24)¼3.50, p¼.046, but the
interaction did not, Fo1. The analyses, therefore, confirm that the
AB effect decreased and Lag-1 sparing improved over sessions, and
this was mostly due to increasing sensitivity rather than to observers
relaxing the bias.
3.2. Event related potentials
3.2.1. N2/ND
We report on the two ERP components that yielded systematic
target-related effects. The first was a negativity, which appeared
about 200 ms from a target onset, of which the maximum was in
the left occipito-parieto-temporal region. Fig. 3 shows the grand
average ERP waveform of this region in Sessions 1 and 2 for
T2-absent (T1) and -present (T1þT2) conditions. In the T1
condition, a negative spike appeared about 200 ms from the T1
onset. It was followed by a slow positive component. These
T1-time-locked components may be considered as T1-elicited N2
and P3. Both components, however, did not reveal an effect of
Sessions as the following statistical analysis shows. We computed
peak amplitude value applying a peak detection routine (Vision
Analyzer 2.0, Brain Products, Gilching, Germany) to individual
(participant) ERP waveforms. For the T1-elicited N2, the peak
detection routine found the global negative maximum between
150–300 ms from T1 onset. The average value of voltages
725 ms around the maximum was used as peak amplitude to
reduce the effect of high frequency voltage fluctuations. Peak
amplitude was computed for each condition in each session. The
practice effect was evaluated using a multivariate analysis of
variance (MANOVA) using Pillai’s statistic for testing significance.
The CAR package ver. 2.0–8 for R [37] was used for the analysis.
The MANOVA, with factors: Sessions (2), Lags (3), and Electrodes
(O1, P3 and T5), showed no main effect of Sessions, Fo1. (In the
T1 only condition, the Lag factor has no meaning. It was kept in
the design to make the number of trials averaged for ERP
computation, more or less, the same between the T1 only and
T1þT2 conditions.)
Fig. 3. ERPs in the left occipito-temporo-parietal region. Line charts represent ERPs
(grand average over 13 subjects) in Session 1 (black) and 2 (red). The x-axis shows
time from T1 onset and the y-axis indicates amplitude in microvolt. Vertical lines
indicate T1 and T2 onsets. For the sake of comparison, T1þT2 ERPs (in black and red)
were superimposed on the T1 ERPs (in light colors). Electrode locations (O1, T5 and P3)
are shown in the ‘‘head’’ icon. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
C. Nakatani et al. / Neurocomputing 84 (2012) 13–2216
In the T1þT2 condition, another negative component
appeared around 200 ms from T2 onset. The T2-locked compo-
nent seems compatible to the T2-elicited N2 reported by Sergent
et al. [38], of which the amplitude was higher in case of T2 hit
than miss. Kranczioch et al [39] reported a similar negativity,
which, like the current component, showed left laterality. The
authors called the T2-related component ‘negative deflection’
(ND), a naming convention we shall follow. The amplitude of
our ND appeared to increase with the sessions. To evaluate this
effect, the ND peak value was computed as follows. As seen in
Fig. 3, the absolute value of the ND component differed among
lags, due to overlapping T1-related components. To correct this
for following statistical analyses, T1 ERPs were subtracted from
the T1þT2 ERPs. The peak detection routine found the global
negative maximum in the difference ERPs between 150–300 ms
from the T2 onset. The average value of voltages 725 ms around
the maximum was used as peak amplitude to reduce the effect of
high frequency voltage fluctuations. Peak amplitude was com-
puted for each condition in each session. The practice effect was
evaluated using the Sessions Lags Electrodes MANOVA, as we
did for the T1-elicited N2. The MANOVA showed significant
main effect of Sessions; F(1,12)¼23.81, p¼.0003. There was a
marginally significant interaction between Lags and Sessions;
F(2,11)¼3.29, p¼.076. Given the differences in T1–T2 SOA
amongst the lags, we should be cautious in making a comparison
between Lags conditions. Nevertheless, it is notable that the ND
amplitude in the Lags 1 and 3 conditions did not exceed that in
the Lag 7 condition. Therefore, the change in ND amplitude in the
short Lag 1 and 3 conditions may be considered as a recovery to a
non-AB level of amplitude, rather than as facilitation.
Previous studies reported that the amplitude of ND/N2 was
higher in T2-hit than -miss trials [38,39]. The current increase in
T2-hit rate with sessions could therefore explain the rise in ND
amplitude. Alternatively, this could occur independently of T2
detection rates. To settle the issue, we analyzed ND amplitude for
the T2-hit and-miss trials separately. Average ERPs were com-
puted for the T2-hit and -miss trials, respectively. From these,
average ERPs for T2-correct rejection (T1 only) trials were sub-
tracted. In the difference ERPs, the ND components were identi-
fied using the peak detection method. The ND amplitude was
analyzed with a 2(Sessions) 2(T2-report-types: hit or miss)
3(Lags) 3(Electrodes) MANOVA. Two participants who had no
T2-miss trials in the second session (one participant in Lag 3, the
other in Lag 7 conditions) were omitted from the analysis. As
expected, the main effect of T2-report-types was significant;
F(1,10)¼8.40, p¼.016, i.e., the amplitude was larger in the
T2-hit than -miss trials. The effect of Sessions was not significant
F(1,10)¼1.68, p4.1. However, the interaction among Sessions,
Lags and T2-report-types was significant; F(2,9)¼5.92, p¼.023.
The 3-way interaction is shown in Fig. 4. In the T2-hit trials, the
amplitude increased with Sessions in Lag 1 and Lag 3, but not in
Lag 7 conditions. A 2(Sessions) 3(Lags) 3(Electrodes) MANOVA
to the T2-hit amplitudes yielded no main effect of Sessions,
F(1,12)¼2.22, p4.1 but a significant interaction between Sessions
and Lags, F(2,11)¼4.80, p¼.032. Consequently, a 2(Sessions)
3(Electrodes) MANOVA was applied in each lag. In the Lag 1 con-
dition, the main effect of Sessions was significant; F(1,12)¼5.69,
p¼.034. In the Lag 3 condition, however, the main effect was not
significant; F(1,12)¼2,35, p¼.151. In the Lag 7 condition, the
main effect of Sessions was also not significant, Fo1. In the
T2-miss trials, the amplitude increased only in the Lag 7 condition.
A 2(Sessions) 3(Lags) 3(Electrodes) MANOVA showed that the
Sessions Lags interaction was marginally significant; F(2,9) ¼
3.63, p¼.070. This was due to the increased amplitude in the Lag 7
condition; a post-hoc, 2(Sessions) 3(Electrodes) MANOVA showed
that the main effect of Sessions was significant; F(1,11)¼6.82,
p¼.024. Although the amplitude increased, it remained smaller than
that in the T2-hit trials. As seen in Fig. 4, the amplitudes of T2-hit
trials were larger than, approximately, 1.50
m
V, while the ampli-
tudes in the miss trials were smaller than that. These results indicate
that the effect of sessions is a general practice effect that occurs
when T2 is correctly detected, in particular that in the Lag 1 condi-
tion, independently of the proportion of correct detection trials.
3.2.2. P3
The second ERP component observed was a positivity that
appeared with a 300 ms or more latency from target onset; the
maximum of this component was in the midline centro-parietal
region. Fig. 5 shows ERPs in this region. In the T1 condition, ERP
components up to 300 ms from T1 onset did not show an effect of
practice. A positivity peaking around 400 ms from the T1 onset
appeared to decrease in amplitude with sessions. We interpreted
the positivity as the T1-elicited P3. Given that the T1-elicited P3
is a slow activity, the mean amplitude from the wide interval
between 300 and 500 ms was used for statistical analysis. The
mean amplitude was computed using a routine of the Vision
Analyzer 2.0 package. To the means, a 2(Sessions) 2(Electrodes,
Cz and Pz) MANOVA model was applied. The main effect of
Sessions was not significant; F(1,12)¼3.10, p¼.108.
In the T1þT2 condition, another slow positive component
appeared around 400 ms from the T2 onset, which is presumably
a T2-elicited P3. This activity was different among the lags; the
activity in the Lag 7 condition was larger than that in the Lag 1
and 3 conditions. However, the activity did not change between
the sessions. Mean amplitude between 300 and 500 ms from the
T2 onset was computed in the difference waveform between
T1þT2 and T1 ERPs, and a 2(Sessions) 3(Lags) 2(Electrodes)
MANOVA was applied. The main effect of Lags was significant
F(2,11)¼7.02, p¼.010. Post-hoc analyses confirmed that the
amplitude in the Lag 7 condition was larger than that in the
Lag 1 and 3 conditions; the main effect of Lags in a 2(Sessions)
2(Lags 1 vs. 7) 2(Electrodes) MANOVA yielded F(1,12)¼15.29,
p¼.002; the main effect in a 2 2(Lags 3 vs. 7) 2 MANOVA was
F(1,12)¼10.15, p¼.007, and that in a 2 2(Lags 1 vs. 3) 2
Fig. 4. Grand average of ND amplitude in the T2-hit (left) and miss (right) trials.
Error bars represent SE. An asterisk indicates po.05 for the main effect of the
sessions.
C. Nakatani et al. / Neurocomputing 84 (2012) 13–22 17
MANOVA was F(1,12)¼1.10, p4.1 (Bonferroni corrected alpha,
p
Bonf
¼p/3). The main effect of Sessions was not significant, Fo1.
The interaction between Sessions and Lags was significant,
F(2,11)¼4.27, p¼.042. This result we will clarify next.
We tested whether the interaction between Sessions and Lags
was due to the difference between the T2-hit and miss trials;
previous studies reported that the P3 amplitude was larger in
T2-hit than miss trials [23,24,40], thus, the interaction might
merely reflect different ratios of the trials between the sessions.
As we did for the ND, we computed mean amplitude between
300 and 500 ms from the T2 onset in the difference waveform
between the T2-hit and T2-correct rejection (T1 only), and in
the difference waveform between the T2-miss and T2-correct
rejection trials, respectively. A 2(Sessions) 2(T2-report-types:
hit or miss) 3(Lags) 3(Electrodes) MANOVA was applied to the
mean amplitude. The main effect of Lags was again significant
F(1,10)¼6.18, p¼.020. There was no main effect of Sessions, or
interactions: S essions Lags, and Sessions Lags T2-report-type,
all Fo1.01. However, there was a main effect of the T2-report type,
F(1,10)¼6.02, p¼.034. The amplitude was larger in the T2-hit than
miss trials. Thus, the Ses sions Lags interaction in the three-way
MANOVA may be attributed to the increase in T2-hit trials between
the sessions.
3.2.3. Interaction between T1- and T2-related components in the
Lag 1 condition
While the T1-elicited N2 in the T1-only conditions did not
show any practice effect, the one in the Lag 1 condition appeared
to be reduced between sessions (Fig. 3, second panel). We,
therefore, computed the peak amplitude of the T1-elicitted N2
in the T1þT2 ERPs. A MANOVA (2 sessions 3 electrodes)
showed that the main effect of Sessions was not significant,
Fo1. The result confirmed that practice did not shift spatial
attention from a T1- to a T2-selection process. Another possible
interaction between T1- and T2-related components was seen in
P3 peak latency (Fig. 5, the second panel). Given the temporal
proximity of targets in the Lag 1 condition, T1- and T2-elicited P3s
largely overlapped. When we treat them as one complex, its
peak shifted later with practice. We identified peak latency of
the complex between 300 and 600 ms from the T1 onset (this
included the 300–500 ms period from T2 onset). A 2(Sessions)
2(Electrodes) MANOVA was conducted for the latency. The main
effect of Sessions was not significant, Fo1. However, there was
an interaction between Sessions and Electrodes; F(1,12)¼10.86,
p¼.006; in Cz, the peak latency increased from 390 ms to 450 ms
between sessions, while the latency in Pz remained constant at
450 ms. We will consider the implications of these Lag-1 specific
results in the Section 4.
3.2.4. Practice effect on distractor-related activities
Next, we examined distractor-related brain activities. Previous
behavioral studies suggested that practice might suppress dis-
tractor-related activity [18,20]. However, as seen in Figs. 3 and 4,
distractor-related ERPs do not show an apparent difference
between the sessions. The null results might be due to a limita-
tion of ERP measures. The distractors were presented repeatedly
for about 2000 ms. Evoked activities overlap and appeared as a
shift of baseline activity during the RSVP period. Such baseline
activity was canceled by baseline correction, which was applied
prior to averaging over single trial EEG segments for ERP compu-
tation. We therefore used the following procedure to capture
distractor-related activity. Given that RSVP frequency was 10 Hz,
we considered 10-Hz EEG activity as primary distractor-related
activity, and band-passed the EEG for the width of 10.072.0 Hz.
Hilbert transform was applied to the band-passed EEG and
instantaneous amplitude was estimated. The amplitude increased
and decreased slowly (.5 Hz) time locked to the beginning and
ending of an RSVP stimuli. We took the mean amplitude within
the RSVP period, from 500 to 1300 ms from T1 onset in which
the amplitude was high and relatively stationary. The mean
amplitude was further averaged over T2-hit and miss trials. The
mean amplitude was the highest in parietal electrodes. However,
Fig. 5. ERPs in the midline centro-parietal region shown in the same style as
in Fig. 3.
C. Nakatani et al. / Neurocomputing 84 (2012) 13–2218
in these electrodes, there was no effect of Sessions, Lags, or
T2-report type (hit or miss). On the other hand, the mean amplitude
in right temporal-occipital electrodes increased with Sessions. A
2(Sessions) 2(T2-report-type) 3(Lags) 2(Electrodes, T6 and O2)
MANOVA revealed that the main effect of Sessions was significant
F(1,11)¼9.74, p¼.010. However, the interaction between Sessions
and T2-report-type, F(1,11)¼2.12, p4.1, and the main effect of
T2-report-type, F(1,11)¼1.01, p4.1 were not significant. The
results showed that practice increased rather than decreased
distractor-related activity. However the change did not affect to
the T2 report.
In contrast with such a temporally-global distractor effect,
Olivers and Meeter [30] claimed that a distractor after T1 would
determine the success or failure of T2 report. Their Boost-and-
Bounce model assumes that the post-T1 distractor inhibits T2.
The authors suggested that a frontal negativity, which appeared
after 300 ms from T1 would reflect target inhibition due to the
post-T1 distractor. Practice might relax the inhibition. If so, the
negativity would also decrease. We computed the peak amplitude
of the negativity using the peak detection routine. The routine
searched the global negative maximum between 250 and 400 ms
from the T1 onset in the T1þT2 waveform. The peak was
searched for the T2-hit and miss trials, separately. The average
value between 725 ms of the maximum was used for the follow-
ing analysis. A 2(Sessions) 2(T2-report-type; hit vs. miss)
2(Electrodes, F7 and F8) MANOVA was applied in each lag. The
main effect of Sessions, and the Sessions T2-report-type inter-
action were not significant in any lag conditions, Fo1.47. How-
ever, the Sessions Electrodes interaction was significant in the
Lag 1 condition, F(1,12)¼5.37, p¼.039; the amplitude decreased
with sessions in F7, while it increased in F8 electrodes. The
interaction was marginally significant in the Lag 3 condition,
F(1,11)¼3.83, p¼.076; but not in the Lag 7 condition, Fo1. A
paired t-test was conducted at each electrode, but none showed a
significant difference between the sessions, to1.28, p4.1. The
amplitude was larger in T2-miss than hit trials in the Lag 1 condi-
tion; F(1,12)¼9.12, p¼.010, for the main effect of T2-report-type.
However, such pattern was not seen in the other lag conditions;
Fo1.32 for the main effect in Lags 3 and 7. These results suggest
that the post-T1 distractor plays a limited role in the Lag 1 condition
of which T1–T2 SOA is shorter than that in the Lag 3 condition.
3.2.5. Practice effect on task-switching
Finally, we assessed practice effects on task-switching-related
components. The current task involves task switching, of which
the cost is supposed to be additive to the AB phenomenon [41,42].
Practice could reduce the switching cost. Therefore, the current
behavioral practice effect might be due to switching cost reduc-
tion. Previous studies reported that ERP components, such as P2
and N2 of the first target were sensitive to task switching related
factors [43–45]. For example, Astle et al. [43] reported that
amplitudes of frontal P2 and N2 were smaller in no-switching
than in switching conditions. Similarly, Gajewski et al. [44]
reported that the amplitude of fronto-central N2 correlated with
switching cost. If practice reduced the cost, the amplitude of these
components would decrease with sessions. We used frontal N2 to
test this hypothesis, because P2 peak was not reliably detected
across individuals in the current ERP data. The N2 peak amplitude
was computed from the T1 þT2 waveform of the T2-hit and miss
trials. Parameters for N2 peak detection were described earlier.
Following methods in the task switching studies, F3, Fz and F4
electrodes were used for a statistical analysis; a 2(Sessions)
2(T2-report-type; hit vs. miss) 3(Electrodes) MANOVA was applied
to the amplitudes in each lag. The main effect of Sessions was
not significant in any lags. However, the Sessions T2-report-type
interaction was significant in Lags 1 and 3, but not in Lag 7;
F(1,12)¼22.17, po.001, F(1,11) ¼5.74, p¼.035, and F(1,10) ¼1.55,
p4.1, for Lags 1, 3, and 7, respectively. In the Lag 1 and 3 conditions,
apost-hoc2(Sessions)3(Electrodes) MANOVA was applied to the
T2-hit and miss trials, separately. The main effect of Sessions was
not significant in the T2-hit trials. On the other hand, the amplitude
did not decrease but increased with sessions in the T2-miss trials,
F(1,12)¼15.05, p¼.002, F(1,11) ¼4.53, p¼.057 for the main effect of
Sessions in Lags 1 and 3, respectively. The results suggest that task-
switching cost was high, relatively speaking, in the second session
when T2 was missed, while the cost was constant over the sessions
in the T2-hit trials.
4. Discussion
We studied the effect of practice on the Lag-1 sparing and
attentional blink (AB) effects. Participants who received two
sessions separated by several days on average increased Lag-1
sparing while decreased the AB effects in the second session.
Using measures of bias and sensitivity from Signal Detection
Theory, we found that, in both conditions, the sensitivity to the
second target (T2) increased between the sessions. Because T2
occurred in only half of the trials participants could, in principle,
have improved their T2 correct detection scores, merely by
relaxing their response criterion. Participants, however, invariably
used the same response criterion across the two sessions. The
criterion they used, moreover, was very conservative; participants
were more concerned with preventing false-alarms than with
misses. Manipulations suggestive of their ability to increase detec-
tion capacities are likely to make participants more confident; these
may encourage participants to relax their criterion. This alone would
be sufficient to explain the reduction of AB with general detachment
instructions. Our methods, by contrast, assured that no criterion
shift occurred, and so improvements in T2 detection scores could be
attributed to improvements in observer sensitivity.
4.1. Theoretical implications of practice effects in Lag-1 sparing
As T2 sensitivity improved with practice, the amplitude of
T2-elicited Negative Deflection (ND) and P3 increased. Both ND and
P3 show larger amplitude in T2-hit than miss trials [23,24,38–40].
The effect of T2 response type, however, did not account for the
practice effect in the ND amplitude in the Lag 1 condition; in this
condition, practice increased the amplitude of ND in the T2-hit trials
irrespectively of the proportion of these trials. Given the latency and
polarity, ND is considered as a subcomponent of the N2 family.
Different cognitive functions are attributed to different N2 subcom-
ponents (See a review [46]). Whereas two anterior N2 subcompo-
nents can be distinguished, which relate, respectively, to perceptual
template mismatch detection and to cognitive control; the posterior
subcomponents are related to visuospatial selection for targets in
distractors [47–49]. The ND is, given its scalp distribution, related to
the posterior subcomponents.
It might seem odd that processes typically associated with
visuospatial selection are crucial in a task like the current one,
where attentional demands in the temporal domain seem to be
more acute. Previous studies, however, have implicated spatial
selectivity in Lag-1 sparing; for example, the sparing does not
occur when Lag-1 T2 is presented in a location different from T1
[50,51]. These results are consistent with the notion that spatial
attention is relevant to the effect. This, in a task heavy on
temporal attention demand, could be understood if we consider
that the same sensory circuits are involved in spatial and
temporal selectivity, and that there is a trade-off between spatial
and temporal sensitivity [52–54]. Enhanced availability for spatial
C. Nakatani et al. / Neurocomputing 84 (2012) 13–22 19
selectivity in the visual domain, thereby, naturally alleviates the
temporal demands of the task. The broader implication of this
result is that theories and models of visual selection in the
temporal domain should consider spatial and temporal selection
as intrinsically connected.
How could our results be explained from neuro-computational
models of AB and Lag1-sparing? For one, the Simultaneous Type,
Serial Token (ST
2
) model takes into account the visuospatial
constraints for T2 enhancement [55,56]. The ST
2
model assumes
a ‘‘simultaneous’’ (i.e., fast) stage, which is called ‘‘input and
extraction of types.’’ For successful report, however, according to
this model, the extracted type (e.g., target feature) needs to be
bound with episodic information, which is specific to the occur-
rence of an item. This binding process is called ‘‘tokenization.’’
Similar to the Two-stage model [3], the ST
2
model assumes that
tokenization takes place serially, and is facilitated by attention.
Attention is assumed to be deployed in a transient fashion
(approx. 100 ms), in order to allow a precise binding between
item features and its episodic likelihood. The attentional unit is,
thus, called Transient Attentional Enhancement (TAE) unit. Hav-
ing received facilitation from TAE, tokens are further consolidated
in working memory. During the consolidation, TAE unit activity is
inhibited. A T2, which reaches the binding stage during T2
inhibition, is not likely to be tokenized and consolidated in
working memory, therefore likely to be missed. The model
assumes that TAE activity is given to an item which appears at
the same spatial location as T1: visuospatial constraints for TAE
activity. The authors have been developing neural network
implementation of the ST
2
model [33,57,58]. Their newest model
(eST
2
,[33]) produces two simulated ERP components; one reflects
transient TAE activity and the other sustained activity for target
type (target or distractor) information. Following the authors’
terminology, we refer to them as, respectively, Blaster and Type
components. The simulation showed depressed Blaster and Type
components for T2. The pattern of the Blaster and Type compo-
nents corresponds to that of ND and P3 for T2, respectively, in the
current study. We should also note that other models, such as the
COrollary Discharge Of Attention Movement (CODAM) model [28],
Locus Coeruleus–Norepinephrine (LC–NE) model [29], and the
Boost-and-Bounce theory [30] have attentional enhancement
mechanisms for target selection. Our results would require these
neuro-computational models to have an adaptive mechanism as a
function of practice.
After the target selection stage, another practice effect was
observed. Only at Cz, the P3 complex in Lag 1 condition shifted its
peak about 60 ms between the sessions. The peak latency in Session
2 was about the same as that of T2-related P3 in Lags 3 and 7. The
implication of this result is not clear at this point. The pattern could
be interpreted, for example, as a shift in attentional resource from
T1- to T2-related post-selection process [59]; or, alternatively,
T1- and T2-related P3 components might be replaced by a single
P3 component of integrated T1þT2 [60];or,finally,theactivation
level of T2 token information might increase relative to the that of
T1 with practice [33]. At the very least, the current Lag-1 specific
ERP results showed that T1 and T2 processes in the Lag 1 condition
interacted in the post-target-selection stage.
4.2. Practice effect in AB
In the Lag-3 condition of the current study, the ND amplitude
was also larger in T2-hit than in miss trials; however, it did not
significantly increase with practice. The result requires us to
consider another facility for the practice effect on AB, alterna-
tively or in addition to target selectivity. For example, Slagter
et al. [61] claimed that a general attentional detachment from T1
processes could account for such a practice effect. The authors
compared AB task performance before and after three months of
meditation training to increase self-control. Experienced yoga
practitioners and novices took part in the experiment. After
meditation practice, the AB-effect was eliminated in the yoga
practitioners, and correspondingly, T1- elicited P3b (a late sub-
component of P3) decreased without a change in T2-elicited
components. The AB effect was merely reduced in the novice
practitioner group, and no ERP reduction whatsoever took place
here. The results are in line with the overinvestment account,
which claims that correction of overinvestment to T1-related
processes is sufficient to reduce the AB [15,16]. Unfortunately,
their study did not report sensitivity and bias measures, thus it is
not apparent what was changed by practice. (The authors
reported no changes in the frequencies of T2 absence reports.
However, this does not mean there could be no change in
response criterion. Suppose the practitioners had had high sensi-
tivity prior to the practice, i.e., T1 and T1 þT2 distributions are
well separated. If they relaxed response criterion, the T2-miss i.e.,
AB would decrease, while the correct rejection would not.) On the
other hand, our practice method, repetition of the AB task, increased
T2 sensitivity without changing the amplitude of T1-evoked P3. As
we reported, the amplitude appeared to be decreased, but the effect
was not statistically significant. Furthermore, we tested for practice
effects on T1-evoked P3b using the second half of the T1-evoked P3,
i.e., average amplitude between 400 and 500 ms from the T1 onset
in the T1-only condition. The amplitude was tested with the same
MANOVA model used for the T1-evoked P3. The test did not yield a
significant effect of Sessions. Therefore, even though the current
findings do not exclude a role for top-down resource management
strategies [62], general detachment from T1 processes cannot
account for the current practice effect on the AB phenomenon.
One other candidate mechanism for the effect of practice on
AB is distractor suppression [18,20]. Our analyses on the 10-Hz
EEG amplitude indicated that practice did not inhibit but
increased distractor-related activity. The activity could, in princi-
ple, have the function of active distractor suppression; however,
critically, the level of the detractor-related activity did not affect
to a T2 report. The 10-Hz amplitude included activities from
multiple distractors. Rather than any arbitrary distractor, a
specific one could, in principle, determine the fate of T2. The
Boost-and-Bounce theory [30] puts emphasis on T2 inhibition
triggered by a post-T1 distractor. The theory assumes that the
input filter ‘‘boosts’’ a target, but inhibits a distractor. The post-T1
distractor, therefore, evokes inhibition which, as a side effect, also
suppresses T2 processing, i.e., ‘‘bounce’’. Practice might have
relaxed this inhibition. The authors listed a late frontal negativity
(‘post-FSP negativity’) as the primary index of the inhibition.
Although, the negativity in our present study showed a practice
effect, instead of a global decrease in amplitude, it was a decrease
in the lower-left frontal region in combination with an increase in
the lower-right frontal regions, of which the interpretation is not
straightforward. Moreover, the amplitude did not differ between
the T2-hit and miss (i.e., non-AB and AB) trials. From these ERP
results, post-T1 distractors seem to have little relevance to the AB
practice effect.
Finally, we briefly mention practice effects on task-switching.
In the Lag 3 condition, a task-switching related component
increased its amplitude. Given that previous studies predicted
not increase, but decrease in amplitude [43,44], facilitation of
task-switching could not account for the AB practice effect.
4.3. Practice effects in process timing
Some theories predict practice effects on ERP latency, since it
indicates process timing. Martens and colleagues [25,63,64]
reported that individuals who showed no AB (i.e., non-blinkers)
C. Nakatani et al. / Neurocomputing 84 (2012) 13–2220
had shorter latency for T1-elicited P3 than those who showed AB.
Based on their ‘quick mind’ hypothesis, T1 processes may be
accelerated with practice. As seen in Figs. 3 and 4, however, the
latency of T1-time-locked components did not change across
sessions, with a possible exception of P3 complex in the Lag
1 condition, which delayed its peak about 60 ms with practice.
These results suggest that, unlike the group differences reported
by Martens and colleagues, repetition of the AB task did not cause
uniform acceleration of T1 processing.
On the other hand, Nieuwenstein et al. proposed that the
timing of T2 processes determines T2 reportability [11,12].
According to the authors, the preceding process of attentional
selection of items in sensory storage was delayed in AB conditions
(Delayed Attentional Engagement, DAE, account). This, in turn,
delays target consolidation. Sessa et al. [26] reported that the
latency of T2-elicited P3 was longer in Lag 3 than in Lag
7 conditions. In their study, distractors after T2 were omitted in
order to avoid distractor-related activity which attenuates the P3
component. In our study, we could identify a T2-related P3 in
spite of post-T2 distractors in the Lag 3 condition. However, the
peak latency of the T2-related P3 did not change over sessions.
Moreover, the latency of the P3 in Lag 3 was about the same as
that in Lag 7. The difference between Sessa et al.’s [26] and our
results might, in part, be due to the presence/absence of post-T2
distractors. We, therefore, tested the peak latency of the ND,
which was considered as a measure of T2 selection. The latency
did not change with sessions. Thus, the current study did not
provide support for the DAE account.
Taatgen et al. [32] applied the Threaded Cognition model [65]
to the AB phenomenon. The model assumes parallel streams of
visual and memory processes; the visual module encodes stimuli
in a visual buffer, the declarative memory module retrieves
memory content of encoded stimuli from declarative memory,
and the consolidation module creates a reportable representation
in working memory. These modular processes are controlled
(‘‘threaded’’) by a set of production rules. One of the rules,
‘‘Protect consolidation,’’ which blocks the visual process while
target consolidation takes place, is critical for the AB phenom-
enon. According to the model, the AB would be avoided if T2
survived until T1 consolidation is completed, or if the ‘‘Protect
consolidation’’ rule is dropped (the authors, for example, pro-
posed that ‘‘non-blinkers’’ – individuals without an AB – do not
use the rule). Our ERP latency results showed, however, that
practice did not accelerate T1 consolidation. Taatgen et al. [32]
listed no ERP candidate which would reflect the production rule
change. Also, their paper did not discuss how modification of
production rules could take place within individuals. It is, there-
fore, unclear how to apply the theory to our practice effects.
5. Conclusion
Practice in the Attentional Blink task, which involves reporting
two targets in RSVP, increases sensitivity to the second target. No
evidence was found for top-down control: neither shifts in response
criterion nor in resources from the first to the second target
occurred. Our results unambiguously show that brain processes
typically associated with visuospatial selection are a crucial compo-
nent for the practice effect; when the two targets were presented
consecutively in a short time interval, uncorrelated to the atten-
tional allocation to the first target, visuospatial selection of the
second target was enhanced via practice. This, even though the main
constraints in processing appear to be temporal in nature. Spatial
and temporal selection, therefore, must involve a common mechan-
ism. Practice, thus, enhances spatiotemporal selectivity.
References
[1] D.E. Broadbent, M.P. Broadbent, From detection to identification: response to
multiple targets in rapid serial visual presentation, Percept. Psychophys. 42
(1987) 105–113.
[2] J.E. Raymond, K.L. Shapiro, K.M. Arnell, Temporary suppression of visual
processing in an RSVP task: an attentional blink? J. Exp. Psychol. Hum.
Percept. Perform. 18 (1992) 849–860.
[3] M.M. Chun, M.C. Potter, A two-stage model for multiple target detection in
rapid serial visual presentation, J. Exp. Psychol. Hum. Percept. Perform. 21
(1995) 109–127.
[4] P. Jolicoeur, Modulation of the attentional blink by on-line response selec-
tion: evidence from speeded and unspeeded task1 decisions, Mem. Cognit. 26
(1998) 1014–1032.
[5] K.L. Shapiro, J.E. Raymond, K.M. Arnell, Attention to visual pattern informa-
tion produces the attentional blink in rapid serial visual presentation, J. Exp.
Psychol. Hum. Percept. Perform. 20 (1994) 357–371.
[6] N. Kanwisher, Repetition blindness: type recognition without token indivi-
duation, Cognition 27 (1987) 117–143.
[7] M.C. Potter, A. Staub, D.H. O’Connor, The time course of competition for
attention: attention is initially labile, J. Exp. Psychol. Hum. Percept. Perform.
28 (2002) 1149–1162.
[8] T.A. Visser, S.M. Zuvic, W.F. Bischof, V. Di Lollo, The attentional blink with
targets in different spatial locations, Psychon. Bull. Rev. 6 (1999) 432–436.
[9] V. Di Lollo, J. Kawahara, S.M. Ghorashi, J.T. Enns, The attentional blink: resource
depletion or temporary loss of control? Psychol. Res. 69 (2005) 191–200.
[10] J. Kawahara, T. Kumada, V. Di Lollo, The attentional blink is governed by a
temporary loss of control, Psychon. Bull. Rev. 13 (2006) 886–890.
[11] M.R. Nieuwenstein, Top-down controlled, delayed selection in the attentional
blink, J. Exp. Psychol. Hum. Percept. Perform. 32 (2006) 973–985.
[12] M.R. Nieuwenstein, M.M. Chun, R.H. van der Lubbe, I.T. Hooge, Delayed
attentional engagement in the attentional blink, J. Exp. Psychol. Hum.
Percept. Perform. 31 (2005) 1463–1475.
[13] I. Arend, S. Johnston, K. Shapiro, Task-irrelevant visual motion and flicker
attenuate the attentional blink, Psychon. Bull. Rev. 13 (2006) 600–607.
[14] C.N. Olivers, Blink and shrink: the effect of the attentional blink on spatial
processing, J. Exp. Psychol. Hum. Percept. Perform. 30 (2004) 613–631.
[15] C.N. Olivers, S. Nieuwenhuis, The beneficial effect of concurrent task-irrelevant
mental activity on temporal attention, P sychol. Sci. 16 (2005) 265–269.
[16] C.N. Olivers, S. Nieuwenhuis, The beneficial effects of additional task load,
positive affect, and instruction on the attentional blink, J. Exp. Psychol. Hum.
Percept. Perform. 32 (2006) 364–379.
[17] C.N. Olivers, D.G. Watson, Input control processes in rapid serial visual
presentations: target selection and distractor inhibition, J. Exp. Psychol. Hum.
Percept. Perform. 32 (2006) 1083–1092.
[18] W.S. Maki, G. Padmanabhan, Transient suppression of processing during
rapid serial visual presentation: acquired distinctiveness of probes modulates
the attentional blink, Psychon. Bull. Rev. 1 (1994) 499–504.
[19] P.E. Dux, R. Marois, Distractor inhibition predicts individual differences in the
attentional blink, PLoS One 3 (2008) e3330. doi:10.1371/journal.pone.0003330.
[20] J. Braun, Vision and attention: the role of training, Nature 393 (1998)
424–425.
[21] J.S. Joseph, M.M. Chun, K. Nakayama, Attentional requirements in a ’pre-
attentive’ feature search task, Nature 387 (1997) 805–807.
[22] J.S. Joseph, M.M. Chun, K. Nakayama, Reply to J. Braun in: Braun, Vision and
attention: the role of training, Nature 393 (1998) 424–425.
[23] G. McArthur, T. Budd, P. Michie, The attentional blink and P300, Neuroreport
10 (1999) 3691–3695.
[24] E.K. Vogel, S.J. Luck, K.L. Shapiro, Electrophysiological evidence for a post-
perceptual locus of suppression during the attentional blink, J. Exp. Psychol.
Hum. Percept. Perform. 24 (1998) 1656–1674.
[25] S. Martens, J. Munneke, H. Smid, A. Johnson, Quick minds don’t blink:
electrophysiological correlates of individual differences in attentional selec-
tion, J. Cogn. Neurosci. 18 (2006) 1423–1438.
[26] P. Sessa, R. Luria, R. Verleger, R. Dell’Acqua, P3 latency shifts in the attentional
blink: further evidence for second target processing postponement, Brain
Res. 1137 (2007) 131–139.
[27] S. Dehaene, C. Sergent, J.P. Changeux, A neuronal network model linking
subjective reports and objective physiological data during conscious percep-
tion, Proc. Natl. Acad. Sci. USA 100 (2003) 8520–8525.
[28] N. Fragopanagos, S. Kockelkoren, J.G. Taylor, A neurodynamic model of the
attentional blink, Brain Res. Cogn. Brain Res. 24 (2005) 568–586.
[29] S. Nieuwenhuis, M.S. Gilzenrat, B.D. Holmes, J.D. Cohen, The role of the locus
coeruleus in mediating the attentional blink: a neurocomputational theory,
J. Exp. Psychol. Gen. 134 (2005) 291–307.
[30] C.N. Olivers, M. Meeter, A boost and bounce theory of temporal attention,
Psychol. Rev. 115 (2008) 836–863. doi:10.1037/a0013395.
[31] S.I. Shih, The attention cascade model and attentional blink, Cogn. Psychol. 56
(2008) 210–236. doi:10.1016/j.cogpsych.2007.06.001.
[32] N.A. Taatgen, I. Juvina, M. Schipper, J.P. Borst, S. Martens, Too much control
can hurt: a threaded cognition model of the attentional blink, Cogn. Psychol.
59 (2009) 1–29. doi:10.1016/j.cogpsych.2008.12.002.
[33] B. Wyble, H. Bowman, M. Nieuwenstein, The attentional blink provides
episodic distinctiveness: sparing at a cost, J. Exp. Psychol. Hum. Percept.
Perform. 35 (2009) 787–807. doi:10.1037/a0013902.
C. Nakatani et al. / Neurocomputing 84 (2012) 13–22 21
[34] C. Nakatani, J. Ito, A.R. Nikolaev, P. Gong, C. van Leeuwen, Phase synchroniza-
tion analysis of EEG during attentional blink, J. Cogn. Neurosci. 17 (2005)
1969–1979.
[35] S. Makeig, A.J. Bell, T.-P. Jung, T.J. Sejnowski, Independent component
analysis of electroencephalographic data, in: D.S. Touretzky, M.C. Mozer,
M.E. Hasselmo (Eds.), Advances in Information Processing Systems, 8, MIT
Press, Cambridge, MA, 1996.
[36] N.A. Macmillan, C.D. Creelman, Detection Theory: A User’s Guide, 2nd
edition, Lawrence Erlbaum Associates, London, 2005.
[37] J. Fox, S. Weisberg, An {R} Companion to Applied Regression (2.0-8 ed.):
Thousand Oaks CA: Sage, 2011.
[38] C. Sergent, S. Baillet, S. Dehaene, Timing of the brain events underlying access to
consciousness during the attentional blink, Nat. Neurosci. 8 (2005) 1391–1400.
[39] C. Kranczioch, S. Debener, A. Maye, A.K. Engel, Temporal dynamics of access
to consciousness in the attentional blink, Neuroimage 37 (2007) 947–955.
doi:10.1016/j.neuroimage.2007.05.044.
[40] E.K. Vogel, S.J. Luck, Delayed working memory consolidation during the
attentional blink, Psychon. Bull. Rev. 9 (2002) 739–743.
[41] J.F. Juola, J. Botella, A. Palacios, Task- and location-switching effects on visual
attention, Percept. Psychophys. 66 (2004) 1303–1317.
[42] J. Kawahara, S.M. Zuvic, J.T. Enns, V. Di Lollo, Task switching mediates the
attentional blink even without backward masking, Percept. Psychophys. 65
(2003) 339–351.
[43] D.E. Astle, G.M. Jackson, R. Swainson, Fractionating the cognitive control
required to bring about a change in task: a dense-sensor event-related
potential study, J. Cogn. Neurosci. 20 (2008) 255–267.
[44] P.D. Gajewski, T. Kleinsorge, M. Falkenstein, Electrophysiological correlates of
residual switch costs, Cortex 46 (2010) 1138–1148.
[45] S. Hsieh, L.C. Liu, The nature of switch cost: task set configuration or carry-
over effect? Brain Res. Cogn. Brain Res. 22 (2005) 165–175.
[46] J.R. Folstein, C. Van Petten, Influence of cognitive controland mismatchon the N2
component of the ERP: a review, Psychophysiology 45 (2008) 152–170.
[47] M. Kiss, J. Van Velzen, M. Eimer, The N2pc component and its links to
attention shifts and spatially selective visual processing, Psychophysiology
45 (2008) 240–249.
[48] G.F. Woodman, S.J. Luck, Electrophysiological measurement of rapid shifts of
attention during visual search, Nature 400 (1999) 867–869.
[49] G.F. Woodman, S.J. Luck, Serial deployment of attention during visual search,
J. Exp. Psychol. Hum. Percept. Perform. 29 (2003) 121–138.
[50] B.G. Breitmeyer, A. Ehrenstein, K. Pritchard, M. Hiscock, J. Crisan, The roles of
location specificity and masking mechanisms in the attentional blink,
Percept. Psychophys. 61 (1999) 798–809.
[51] A.E. Seiffert, V. Di Lollo, Low-level masking in the attentional blink, J. Exp.
Psychol. Hum. Percept. Perform. 23 (1997) 1061–1073.
[52] S. Gepshtein, I. Tyukin, M. Kubovy, The economics of motion perception and
invariants of visual sensitivity, J. Vis. 7 (2007) 8. doi:10.1167/7.8.8.
[53] R. Marois, M.M. Chun, J.C. Gore, Neural correlates of the attentional blink,
Neuron 28 (2000) 299–308.
[54] A. Raffone, M. Pantani, A global workspace model for phenomenal and access
consciousness, Conscious Cogn. 19 (2010) 580–596.
[55] H. Bowman, B. Wyble, The simultaneous type, serial token model of temporal
attention and working memory, Psychol. Rev. 114 (2007) 38–70.
[56] H. Bowman, B. Wyble, S. Chennu, P. Craston, A reciprocal relationship
between bottom-up trace strength and the attentional blink bottleneck:
relating the LC-NE and ST(2) models, Brain Res. 1202 (2008) 25–42.
[57] S. Chennu, P. Craston, B. Wyble, H. Bowman, Attention increases the temporal
precision of conscious perception: verifying the Neural-ST Model, PLoS
Comput. Biol. 5 (2009) e1000576.
[58] P. Craston, B. Wyble, S. Chennu, H. Bowman, The attentional blink reveals
serial working memory encoding: evidence from virtual and human event-
related potentials, J. Cogn. Neurosci. 21 (2008) 550–566.
[59] K.L. Shapiro, F. Schmitz, S. Martens, B. Hommel, A. Schnitzler, Resource
sharing in the attentional blink, Neuroreport 17 (2006) 163–166.
[60] E.G. Akyurek, B. Hommel, Target integration and the attentional blink, Acta
Psychol. (Amst) 119 (2005) 305–314.
[61] H.A. Slagter, A. Lutz, L.L. Greischar, A.D. Francis, S. Nieuwenhuis, J.M. Davis,
Mental training affects distribution of limited brain resources, PLoS Biol. 5
(2007) e138. doi:10.1371/journal.pbio.0050138.
[62] S.M. Wierda, H. van Rijn, N.A. Taatgen, S. Martens, Distracting the mind
improves performance: an ERP Study, PLoS One 5 (2010) e15024.
doi:10.1371/journal.pone.0015024.
[63] S. Martens, M. Dun, B. Wyble, M. Potter, A quick mind with letters can be a
slow mind with natural scenes: individual differences in attentional selec-
tion, PLoS One 5 (2010) e13562. doi:10.1371/journal.pone.0013562.
[64] S. Martens, O. Korucuoglu, H.G.O.M. Smid, M.R. Nieuwenstein, Quick minds
slowed down: effects of rotation and stimulus category on the attentional
blink, PLoS One 5 (2010) e13509. doi:10.1371/journal.pone.0013509.
[65] D.D. Salvucci, N.A. Taatgen, Threaded cognition: an integrated theory of
concurrent multitasking, Psychol. Rev. 115 (2008) 101–130. doi:10.1037/
0033-295X.115.1.101.
Chie Nakatani received her PhD from Department of
Psychology, University of Massachusetts in USA. She
worked as a Research/Staff Scientist in Laboratory for
Perceptual Dynamics, RIKEN Brain Science Institute
in Japan. Currently she is a Post-doctoral Fellow in
Faculty of Psychology and Educational Sciences, Uni-
versity of Leuven in Belgium. Her research interests are
temporal aspects of visual perception, including atten-
tional blink phenomenon, mental rotation/manipulation
of object representation, inter-saccadic integration of
visual information, and more.
Shruti Baijal holds a PhD in Cognitive Science from
Centre of Behavioural and Cognitive Sciences, Univer-
sity of Allahabad, India. She was a research intern at
Perceptual Dynamics Laboratory, RIKEN BSI, Japan.
Presently, she is a post-doctoral fellow at Department
of Psychology, University of Miami, US. Her research
primarily focuses on understanding how the cognitive
systems of attention and working memory function in
the human brain.
Cees van Leeuwen established the Laboratory for
Perceptual Dynamics at the RIKEN Brain Science Insti-
tute, Japan in 2001. The laboratory combines psycho-
physical methods, recording of electrical brain activity
(EEG) and eye movements in humans, with computa-
tional modeling. Currently, the laboratory is moving
from Japan to the University of Leuven (Belgium),
funded by an Odysseus grant from the Flemish Orga-
nization of Scientific Research FWO. Cees van Leeuwen
is the editor of two scientific journals and has pub-
lished well over 100 articles in peer-reviewed journals
over the last 20 years. During this period, he has
been working to establish a theoretical understanding
of perception and visual awareness based on rigid experimentation and the tools
of complex systems theory.
C. Nakatani et al. / Neurocomputing 84 (2012) 13–2222
