1 Introduction
What are the consequences of training eye movements? Although there are several
models of visual attention (eg Itti and Koch 2000; Logan 1996) and oculomotor control
(eg Findlay and Walker 1999), these models underspecify the influences of top ^ down
factors, such as training or experience in a specific context. The research described
here incorporates eye-movement training into Findlay and Walker's (1999) dual-process
model of saccadic planning to suggest that domain-specific experts should be more
aware of established models of eye-movement control when developing visual training
protocols for use in applied settings.
Findlay and Walker's framework specifies a system where two metrics are programmed
for every saccade: `when' the saccade should occur, and `where' it should be
directed. The model is hierarchical, and although it describes the `where' and `when'
pathways primarily in functional terms, its lower levels in particular are considerably
influenced by work on oculomotor physiology. Accordingly, in processing models of
eye movements (Morrison 1984; Reichle et al 1998) `when' is determined by the processing
of the fixated stimulus, and Findlay and Walker link this to activation in the
rostral pole region of the superior colliculus, which carries a neural representation of
the fovea (Munoz and Wurtz 1993). When processing of the current object has reached
a certain threshold, programming of the next saccade occurs. The location of the next
saccade is determined by the greatest salience peak represented on a topographic map
(cf Munoz and Wurtz 1995a, 1995b), though the intrinsic salience of any object (contours,
contrast, etc) may be influenced by top ^ down factors, such as inhibition of areas of the
salience map (Godijn and Theeuwes 2002).
The need to fixate and the urge to make a saccade (represented as `when' activation
in the `Fixate' centre, and `where' activation in the `Move' centre, respectively, within
Findlay and Walker's model) are in competition, and are bound together by reciprocal
inhibitory links. This has important ramifications for attempts to train eye movements,
Training eye movements: Can training people where to look
hinder the processing of fixated objects?
Perception, 2008, volume 37, pages 1729 ^ 1744
Richard Dewhurst, David Crundall
School of Psychology, University of Nottingham, University Park, Nottingham NG7 2RD, UK;
e-mail: lwxrcd@psychology.nottingham.ac.uk; dec@psychology.nottingham.ac.uk
Received 5 November 2007, in revised form 21 April 2008; published online 27 October 2008
Abstract. An experiment designed to test the effects of different forms of training in a visualsearch-like
task is reported. Observers were presented with a series of displays containing a
central letter and a ring of peripheral characters, one of which was a digit. Odd digits (catch
trials) required a space-bar response; even digits required a different response contingent on the
identity of the central letter. Two forms of training provided information either about the location
of the peripheral digit, or about a quick way to classify the central letter. The aim was to relate
training to Findlay and Walker's (1999, Behavioral and Brain Sciences 22 661 ^ 721) model of
saccadic eye-movement control by affecting the hypothesised Move and Fixate centres respectively.
The results showed that training benefited search, but training of the Move centre alone
generated significantly longer re-inspections of the central region (in a feedback condition). This
highlights that the emphasis often placed upon broadening the range of visual search when training
eye movements may be misplaced. More specifically, special attention should be given, not
only to advising people how to move their eyes, but also to improving the ability to effectively
process important visual stimuli when fixating.
doi:10.1068/p5944
as any attempt to focus upon one aspect of eye movements may degrade behaviour that
is dependent upon the other.
A good example of negative consequences which may be related to this competitive
interaction is available in the applied literature on driving. Drivers often fail to see
motorbikes at junctions, even though they often report looking towards their location
(Clark et al 2004). This phenomenon has been coined the `look but fail to see' accident
causation factor (Brown 2002), and is supported by considerable research (see
also Herslund and Jorgensen 2003; Martens and Fox 2007). Here, increased activity in
the Move centre associated with vigilantly scanning the visual scene may decrease
activity in the Fixate centre when fixated objects are being processed, in this case
motorbikes, and as a result they may not be processed adequately despite being fixated.
Following this logic, it is unlikely that certain visual-search strategies recommended
to drivers, such as repeating a left ^ right scanpath at junctions (UK Department
for Transport 2007) or maintaining active and board visual search (Mills 2005),
will be successful in their objectives because they focus upon maximising eye movements,
which again may decrease activity in the Fixate centre owing to Move-centre
competition.
The competitive interaction between the Move and Fixate centres provides a viable
explanation why some previous attempts to train eye movements may not have had the
desired outcomes. Donovan et al (2005) gave feedback to novice radiographers when
they were performing a fracture-detection task. Following this feedback (which showed
them where they initially looked on the X-ray and for how long) their eye movements
more closely resembled those of expert radiographers. Interestingly, however, the use
of an expert scanpath alone was not sufficient to give rise to an improvement in the
fracture-detection task. This suggests that advice which changes eye movements alone
does not necessarily lead to improvements in performance, possibly because of the type
of conflict between the Move and Fixate centres described.
The present research employs a novel experimental paradigm to assess the consequences
of directing training specifically at the Move centre of Findlay and Walker's
model. A visual-search task was used in which a central letter was presented within a
circular array of potential targets (see figure 1a). Participants were required to start
each trial by processing the central letter, before searching the peripheral array for a
digit amongst non-digits. This design was specifically chosen to require the initial
inspection of the central letter (involving the Fixate centre) followed by a series of eye
movements (involving the Move centre).
When an even-number peripheral target was located, the required response depended
on the central letter: if the central letter was A, T, M, or V, the correct response was
to press the `1' key on the computer's number-pad; alternatively, if the central letter
was L, F, K, or N, the correct response was to press the `3' key on the computer's
number-pad. In contrast, the correct response to odd digits was unrelated to the central
letter, a space-bar press being required irrespective of the central letter displayed
(these trial types served as catch trials to ensure participants searched the peripheral
array and located the peripheral target). Eye movements were recorded while participants
completed this task.
The paradigm used provides the opportunity to pre-inform participants about the
sequence of locations the target will appear in from one trial to the next; this forms
the basis of training directed at the Move centre (see figure 1b), which will be referred
to as `Move Training'. The intention was to create a task where the demands placed
on the Fixate centre by the initial letter-discrimination judgment competed with
Move centre activation when participants were instructed in `Move Training'. It is for
this reason that correct answers were contingent upon both the central letter and the
peripheral number; we sought a task where two items in the display competed for
1730 R Dewhurst, D Crundall
attention within a single trial. If, for instance, there was no central letter and participants
only responded to the peripheral number (with separate responses for odd and even
numbers for example), then the `Move Training' sequence would induce far less competition
between the Move and Fixate centres because the upcoming target location it
predicted would be in the next trial. Indeed, this would be the case in any task where
Move ^ Fixate competition was sought inter-, rather than intra-, trial. If the basic task
is to be comparable for an untrained group, and `Move Training' is to be beneficial in
improving search performance yet detrimental in reducing information processing at
fixation, a task such as the one used here is necessary.
It is predicted that the hypothesised competition induced by `Move Training' in
our task will encourage disengagement from the central letter before it is fully processed.
The most obvious way this effect could be revealed therefore is by reduced
central gaze durations at the start of a trial, accompanied by less overall fixations
and faster search times to locate the peripheral target. However, it is also conceivable
on the basis of considerable previous research (eg Engel 1971; Eriksen and Hoffman
1972; Koivisto et al 2004; Motter and Holsapple 2007; Posner et al 1980) that the
absolute time spent on the centre upon trial commencement will not change, and
the competitive interaction will be expressed as a premature shift of covert attention.
Either way, insufficient initial processing of the central letter could lead to the detrimental
performance of Move-trained participants in a number of ways: a decrease in
accuracy, or participants could need to re-inspect the central letter more regularly after
they have identified the peripheral target in order to respond correctly. These possible
outcomes would advance our understanding of the eye-movement statistics associated
with looking but failing to see.
Given the hypothesis that `Move Training' may be detrimental to performance because
of its effects on the Fixate centre, we also ask whether providing training directed at the
Fixate centre, in concert with `Move Training', can counteract any negative consequences
(a) (b)
Figure 1. (a) An example of one stimulus as shown in a single trial display. The target (the
only stimulus in the peripheral array that is a number, in this case the number 2) is located
458 anticlockwise from the 12 o'clock position. (b) The training directed at the Move centre.
The numbers indicate place-holder positions for the peripheral stimuli. From trial to trial the
peripheral target number appeared in the following sequence: in trial one it was presented
in position 1, followed by position 4 in trial two, then positions 7, 2, 5, 8, 3, 6 from one trial
to the next (this is indicated by the arrows in the figure). This sequence is repeated throughout
all experimental blocks.
Training eye movements 1731
that arise from `Move Training' alone. Training was directed at the Fixate centre by
informing participants that the first set of letters (A, T, M, and V) are all symmetrical
around the vertical meridian, whereas the second set (L, F, K, N) are not. Therefore,
when initially processing the central letter at the beginning of a trial, the discrimination
judgment is reduced in difficulty. This manipulation is introduced on the basis
of previous research, cited by Findlay and Walker (1999) as evidence for top ^ down
mediation of the Fixate centre, which shows that, when a stimulus is unchanged but
its informational load is decreased, fixation durations decrease also (eg Gould 1973).
It is predicted that when directed in both `Move Training' and training aimed towards
the Fixate centre (this combination of training strategies will hereafter be referred to
as `Both Training') that participants will no longer disengage from the central letter
before it is fully processed, and as a result they will not need to re-inspect it as
regularly in order to maintain accuracy. `Both Training' was chosen in preference to
`Fixate Training' in isolation because we wished to demonstrate that any deficits on the
task associated with `Move Training' alone could be ameliorated. In so doing we hope
to show that the predicted visual-search benefits of `Move Training' (in terms of faster
search times and fewer fixations per trial) are attainable without any look-but-failto-see-like
errors. As such, a Fixate-trained-only group would provide little evidence
of how visual-training protocols could be improved, because, without explicit direction
where to move the eyes, comparable benefits in visual search would not be expected.
Three groups of participants were tested in two blocks of trials. The first block served
as a baseline to establish initial performance on the task. For the second block, one
group of participants was told the `Move Training' strategy, another group was told `Both
Training' strategies, and a third and final group was used as a control comparison, being
given `No Training' throughout.
Although it is not central to the main hypothesis, it is conceivable that participants
may guess some of the experimental contingencies when they are not advised of the
training strategies. Therefore participants were asked upon debrief whether they noticed
any of the rules which governed stimulus presentation. This, coupled with a brief
examination of typical scanning patterns within a trial, will address questions regarding
explicit or implicit learning of the experiment's contingencies. Moreover, a simple
examination of the typical scanpaths adopted will also highlight regularities in visualsearch
behaviour. For example, one may predict, on the basis of previous observations
of scanpath uniformity (Foulsham and Underwood 2008), that participants will search
the peripheral array in a clockwise or counterclockwise manner (see Findlay and Brown
2006) prior to awareness of the predictable sequence in which the number target
occurs.
One final manipulation was introduced into the design. As the study aims to raise
awareness amongst researchers in applied domains about the need to refer to established
models of eye-movement control when devising eye-movement training regimes,
it was thought prudent to assess the effect of feedback upon task performance. The
ability to interact with the environment and evaluate the consequences of actions is
particularly important in applied settings such as driving, sport, human ^ computer
interfaces, etc, and the source of error monitoring can be internal or external. Therefore,
the availability of feedback was varied in the current study, with the second block being
equally divided into two counterbalanced sections, one in which visual feedback was
always available, and one in which no feedback was provided. Although no specific
predictions were made about the effects of feedback, it was anticipated that the performance
of untrained participants would suffer most from its withdrawal, simply because
this group had no strategy to rely on.
1732 R Dewhurst, D Crundall
2 Method
2.1 Participants
Thirty-eight paid participants (with a mean age of 23 years; twenty-six female) were
recruited from the student population of the University of Nottingham. All participants
reported normal or corrected-to-normal vision.
2.2 Stimuli and apparatus
There were 512 individual stimuli in total; the central letter could be one of 8 possibilities:
A, T, M, V, L, F, K, or N; the peripheral number target could appear in one
of 8 locations around the centre, and was one of 8 numbers in `calculator-style' font:
0, 2, 3, 4, 5, 6, 7, and 9 (1 and 8 were not used because they contained too few, and
too many component lines, respectively, which could unfairly assist in their detection).
Zero was classified as an even number and participants were made aware of this. All
of the factors differentiating the stimuli mentioned above were represented with equal
probability.
24 non-digit distractors were created by removing one or more component lines
from the number 8. Each distractor had approximately the same visual footprint, and
different distractors occupied the remaining 7 peripheral locations of the stimulus that
did not contain the target. Which distractor occurred in which position was determined
at random.
All letters and digits subtended 0.8 deg visual angle horizontally. The peripheral
shapes subtended 1.6 deg vertically, while the central letters subtended 1.2 deg vertically.
The centre of the central letter was coincident with the centre of the display.
The peripheral shapes were located on the perimeter of an imaginary circle, with a
radius of 10 deg from the centre, at 458 increments.
A PC operating with a Pentium 3 processor was used to run the experiment via
E-prime software. Stimuli were presented on a 19 inch Samsung SyncMaster LCD
screen, running at 10246768 pixel resolution. A desk-mounted SMI Red-Eye III eye
tracker was used to record eye movements of the right eye at 50 Hz. Fixations were
defined as periods of ocular stability in-between saccades with a minimum duration
of 80 ms.
2.3 Design
The experiment was divided into two blocks of 128 trials. Baseline performance was
assessed in block 1, with no eye-movement training being given. In-between blocks 1
and 2 training was provided for the Move Trained (MT) and Both Trained (BT) groups;
the No Training group (NT) acted as a control.
The MT group (N ˆ 14) was informed that the target would appear in the predictable
sequence shown in figure 1b from one trial to the next. The target was presented
in this order with 100% probability throughout both blocks of the experiment for each
group of participants. The BT group (N ˆ 10) was given an additional strategy to that
of the MT group; the participants were made aware that the letters A, T, M, and V
(to which the same response is required) are all vertically symmetrical; whereas the
letters L, F, K, and N (which also share the same required response) are vertically
asymmetrical. The NT group (N ˆ 14) was given neither of these strategies.
Feedback was also manipulated for several reasons. First, on the basis of previous
pilot work in our laboratory, there was reason to believe that the predicted detriments
associated with MT would be most apparent when feedback was provided. Second,
because the ability to monitor errors when performing motor tasks is controversial
(Hajcak et al 2005; van Veen et al 2004), we thought it prudent to include this measure.
The final reason this measure was incorporated into the design was because in applied
settings, such as driving, the ability to interact with the environment is vital, and the
source of error monitoring can be internal or external.
Training eye movements 1733
Feedback was a within-groups factor, with half of block 2 containing feedback (FB)
and half not (NFB). The order of this manipulation was counterbalanced across
subjects. Throughout block 1 feedback was always given, by way of a red (incorrect) or
blue (correct) transient screen after each trial (`time-outs' were considered incorrect).
For NFB trials the blue or red mask was replaced by a grey mask in all instances,
such that feedback was no longer available.
With the exception of the sequence of peripheral target location being pre-determined,
stimuli were selected quasi-randomly. Catch trials (where the peripheral target was an
odd number) occurred with 33.3% probability, while trials with an even-number target
occurred with 66.6% probability. As the number of trials per block is smaller than the
number of stimuli, not all stimuli occurred within a block of trials.
Several dependent variables contributed to the measures taken from the design.
Behavioural data in terms of manual reaction times and accuracy were recorded.
Several measures of eye movements were also taken. The stimulus array was divided
into regions of interest for eye-tracking analysis. A circle, with a radius of 2.9 deg
from the centre of the display, was created within which any fixation made was classified
as a fixation with the purpose of central-letter processing. Initial central gaze
duration, the cumulation of fixation durations from the first fixation on the central
letter to the last fixation before this region is left, was calculated. After this initial
period of inspection, any fixation made within this central region was classified as a
re-fixation. The mean re-inspection duration (ie the average of re-fixation durations
for all trials in an entire block, including those in which no re-fixation occurred) was
calculated as well as the average number of re-fixations per trial.
Finally, overall visual-search performance was assessed, first by analysing the average
number of fixations per trial, and second by analysing the time taken to fixate
the peripheral target from the start of the trial. The latter of these measures was
calculated by creating regions of interest around the peripheral array items. If a fixation
was made outside the area of the imaginary circle designated for the central letter,
but within the area of another eccentric circle with a radius 5.8 deg from the centre
of the display, it was classified as being unrelated to the processing of any of the
displayed items in the stimulus array. Beyond the second eccentric circle, fixations were
divided into those on the peripheral number target and those on the distractors. This
was done by splitting the stimulus array into regions 22.58 (of a total 3608 for the
entire stimulus array) either side of centre of the peripheral shapes. If a fixation landed
within the target's region it was flagged, therefore allowing the time taken to fixate
the target to be calculated. To complement these parametric measures of visual-search
performance, representative scanpaths of visual-search behaviour within a trial will be
shown for example participants.
For purposes of analysis the data collected were treated in terms of changes in
block 2 relative to baseline (although overall means will be given also). For each participant,
by subtracting the mean value of a dependent variable at baseline from its mean
value in both the FB and NFB portions of block 2, it is possible to quantify relative
changes in performance. This controls for idiosyncrasies in ability between participants
as the difference score obtained reflects individual change on a given measure as a
function of training or, in the case of the NT group, exposure alone.
2.4 Procedure
On arrival, informed consent was obtained from the participants. Detailed instruction was
available on the computer screen explaining the task, and the experimenter was present
throughout to ensure participants understood what they had to do. After the participants
agreed to take part and confirmed that they understood the task, the experimenter
calibrated the eye tracker (a chin-rest was used to maintain a stable and fixed viewing
1734 R Dewhurst, D Crundall
position of 70 cm from the screen throughout the experiment). Following calibration,
participants were initially given a practice block of 30 trials, which were identical to
the experimental blocks described below. Once participants had completed the initial
practice block, and the experimenter was satisfied that they understood the task, they
could proceed.
A fixation cross was presented only at the start of a trial block. A stimulus was then
selected according to the stipulations outlined in the design section above. The stimulus
remained on the screen until a response was made for a maximum duration of
10 s, following which a feedback mask was presented for 200 ms. Stimulus presentation
followed by feedback continuously cycled until the end of a block.
Participants were required to begin each trial fixating the central letter before searching
the peripheral array for the target number. When A, T, M, or V were presented in
the centre, and the peripheral target was an even number, the correct response was to
press the `1' key on the keyboard number-pad; when L, F, K, or N were presented, and
peripheral target was an even number, the `3' key on the number-pad was the correct
response. When the peripheral target was an odd number the trial was a catch trial, and
the correct response was to press the space-bar irrespective of which letter was displayed.
Participants were asked to respond as quickly and accurately as possible throughout.
Participants completed the baseline block first, after completion of which the MT
and BT groups received their respective training via on-screen instructions. The experimenter
also provided explanation if required. Participants were then given a further
30 trials of practice to become familiar with using the training suggested to them. The
NT group was given an additional 30 trials practice also, to ensure equal exposure to
the task. Following this practice, the experimenter re-calibrated the eye tracker as before.
The second block proceeded next. As feedback was counterbalanced, the order of
feedback availability in block 2 varied between participants; they were always informed,
however, whether the trials they were about to complete contained feedback or not by
on-screen instructions.
On completion of the experiment, participants were debriefed as to the aim of the
experiment and its hypotheses. A short questionnaire was also completed in which
generic demographic information was collected, along with information about whether
the experiment's contingencies were guessed.
3 Results
3.1 Reaction times (RTs)
Practice trials, catch trials, and anticipatory responses (RT 5 200 ms) were removed
before analysis for all analyses described. The average RT to correct trials across
groups and conditions was 2859 ms (SD ˆ 605X6 ms). Although the behavioural data
collected were analysed in terms of difference scores, summary statistics showing the
mean in each condition are given in table 1.
The relative effect of training (or practice for the NT group) was calculated by
subtracting the participants' mean baseline (block 1) RTs from their mean RTs in both
parts of block 2 (FB and NFB). A negative value therefore reflects a decrease, whereas
a positive value reflects an increase. These RTs were analysed by a 362 mixed factorial
ANOVA, with three levels of training and two levels of feedback (this analysis was
repeated for all dependent measures taken).
A significant main effect of training was observed (F2 35 ˆ 4X5, MSE ˆ 341062,
p ˆ 0X018), reflecting the greater improvement in RT for the trained groups (MT:
xblock 2 À xblock 1 ˆ À799 ms, SEM ˆ 79.7 ms; BT: xblock 2 À xblock 1 ˆ À861 ms, SEM ˆ
129X1 ms) compared to the NT group (xblock 2 À xblock 1 ˆ À412 ms, SEM ˆ 62X0 ms).
A-posteriori comparisons with Tukey's HSD confirmed that, while the trained groups
differed significantly from the NT group ( p 5 0X05), the MT and BT groups did
,
Training eye movements 1735
not differ from each other ( p ˆ 0X931). There was no main effect of feedback, and
feedback did not interact with training.
To check that no group had a quasi-advantage in the ability to perform the task from
the outset, baseline (block 1) RTs were compared by a one-way ANOVA with training
as the grouping variable. No baseline differences were observed between groups for RTs,
or any of the measures reported henceforth.
3.2 Accuracy
No significant main effects were found in the accuracy analysis, nor did the factors
interact. Accuracy was high, however (x ˆ 93% correct overall, SD ˆ 5X3%).
3.3 Eye-movement recordings
To properly understand the behavioural data obtained in the context of eye-movement
control it is necessary to examine certain statistics of visual search and central-letter
processing. In addition to the removal of practice, catch, anticipatory, and incorrect
trials, trials in which the peripheral target was not fixated, the duration of the initial
central gaze on the letter was 550 ms, and the central letter was not fixated within
4 fixations, were also excluded prior to analysis of the eye-movement data. Respectively,
these criteria reduce the possibilities that participants were locating the target in
peripheral vision, spending an insufficient time foveating the centre at the start of
the trial, or searching the peripheral array first (even though they were specifically told
not to do this). After these exclusions, an average of 36 trials contributed to each
participant's cell means for the eye-movement measures taken. If a participant had less
than 10 trials contributing to their cell mean for the baseline, feedback, or no-feedback
blocks, he/she was removed from the subsequent eye-movement analysis. Loss of calibration
among certain participants means that seven were excluded according to this
criterion, three from the NT group and four from the MT group. Additionally, one
participant was excluded (from the BT group) because of not completing the task
properly. After removal of these participants, an average of 40 trials now contributed
to the cell means. Although the eye-movement data were also analysed in terms of
difference scores, the means in each condition are given in table 2.
3.4 Overall visual-search performance
The average number of fixations per trial (x ˆ 8X31 overall; SD ˆ 1X82) gives an indication
of search efficiency. Analysis of the relative changes in these data revealed a significant
main effect of training only (F2 27 ˆ 8X4, MSE ˆ 3X6, p ˆ 0X001), highlighting that the
trained groups (MT: xblock 2 À xblock 1 ˆ À2X5, SEM ˆ 0X3; BT: xblock 2 À xblock 1 ˆ À3X16,
SEM ˆ 0X4) make less fixations following training than the reduction afforded by practice
alone (NT: xblock 2 À xblock 1 ˆ À0X8, SEM ˆ 0X23). Tukey HSD a posteriori comparisons
confirmed these differences (NT versus MT, BT: ps 5 0X05; MT versus BT: p ˆ 0X54).
,
Table 1. Mean reaction times and accuracy for each group in each condition. Standard errors
are in parentheses.
Training Block 1 (baseline) Block 2
group
feedback no feedback
NT reaction time/ms 3258 (143) 2890 (148) 2802 (150)
accuracya% 92 (1.5) 92 (1.2) 95 (1.2)
MT reaction timeams 3279 (113) 2433 (128) 2524 (135)
accuracya% 92 (1.2) 92 (1.1) 93 (1.2)
BT reaction timeams 3416 (188) 2481 (142) 2628 (145)
accuracya% 93 (1.5) 95 (1.2) 94 (1.8)
1736 R Dewhurst, D Crundall
Visual-search performance can also be assessed by looking at the time from commencement
of the trial to fixation of the peripheral target. These data were extracted from the
eye-movement recordings also, and, overall, participants fixated the target within an
average of 1542 ms (SD ˆ 593X7 ms). The relative-change analysis revealed a significant
main effect of training only (F2 27 ˆ 6X7, MSE ˆ 234948, p ˆ 0X004), with the trained groups
(MT: xblock 2 À xblock 1 ˆ À625 ms, SEM ˆ 84X9 ms; BT: xblock 2 À xblock 1 ˆ À838 ms,
SEM ˆ 110X7 ms) improving more than the NT group (xblock 2 À xblock 1 ˆ À286 ms,
SEM ˆ 39X0 ms. A-posteriori comparisons with Tukey's HSD revealed differences between
the NT and BT groups only ( p ˆ 0X004); the comparison between the NT and MT
groups narrowly failed to reach conventional statistical significance ( p ˆ 0X078), and the
MT and BT groups did not differ ( p ˆ 0X38). This pattern of results complements the RT
analysis, demonstrating the improved search efficiency of the trained groups.
3.5 Visual scanning behaviour
To relate the measures of visual-search performance outlined above to eye-movement
patterns, typical scanpaths within a trial are plotted for two example participants
(figure 2). Although detailed statistical analysis of these scanpaths is outside the scope
of this article (see Foulsham and Underwood 2008 for a more comprehensive approach),
they do reveal a number of interesting findings pertinent to the hypothesis. First, given
the circular arrangement of stimuli in the periphery, participants commonly favoured a
clockwise or counterclockwise eye-movement sequence (when untrained), as predicted.
Such visual-search behaviour is similar to what has been previously described as a
`convex-hull' scanpath, where participants predominantly search the perimeter of an
array of items making frequent forays into the centre (Findlay and Brown 2006). Given
the layout of our stimuli, this is not so surprising; however, there were certain noteworthy
characteristics of our participants' fixation sequences en route. Figure 2a (left
panel) shows that, in trial 52 of the baseline block, subject 13 (from the NT group)
commenced searching the periphery at target position 6, continuing up to position 8
before backtracking to position 7 to re-check the distractor there. After doing so, the
original search did not resume from where it left off; instead, this participant moved
over to the opposite side of the display and began to search in a clockwise order once
more. An interesting observation of the search behaviour adopted in this trial is that,
,
Table 2. Summary statistics for the measures of eye movements taken. Means are shown for
each group in each condition with standard errors in parentheses.
Training Eye-movement statistic Block 1 Block 2
group (baseline)
feedback no feedback
NT number of fixations per trial 9.6 (0.4) 8.9 (0.3) 8.8 (0.3)
time to first target fixationams 1963 (137) 1702 (138) 1651 (134)
initial central gaze durationams 473 (52) 397 (36) 455 (58)
re-inspection durationams 186 (33) 143 (28) 185 (39)
number of re-fixations per trial 1.1 (0.2) 0.7 (0.1) 0.9 (0.2)
MT number of fixations per trial 9.4 (0.3) 6.9 (0.3) 6.9 (0.4)
time to first target fixationams 1828 (226) 1197 (181) 1207 (180)
initial central gaze durationams 643 (102) 608 (96) 664 (118)
re-inspection durationams 146 (31) 187 (48) 133 (39)
number of re-fixations per trial 0.7 (0.2) 0.7 (0.2) 0.6 (0.2)
BT number of fixations per trial 10.2 (0.8) 6.8 (0.6) 7.2 (0.5)
time to first target fixationams 1967 (173) 1132 (121) 1125 (135)
initial central gaze durationams 399 (38) 539 (140) 472 (98)
re-inspection durationams 178 (51) 163 (51) 203 (59)
number of re-fixations per trial 1.0 (0.3) 0.7 (0.2) 0.9 (0.2)
Training eye movements 1737
once this subject reached the original location at which the search began, he/she remembered
that the left hand side of the display had already been searched and continued
up to the remaining segment that was yet to be inspected. This is an excellent example
of an entirely exhaustive search where all possible regions were inspected before the
target was found. As was frequently the case, the target fixation was the penultimate
fixation of the trial, the final fixation being a re-fixation on the central letter.
Despite the minor practice effects of the NT group reported above, it can also be seen
that, even towards the end of block 2, the same untrained participant failed to notice the
predictable sequence of target presentation. This is nicely shown in figure 2a (right panel)
where the participant began a counterclockwise search on the opposite side of the screen
to the target and still passed over the target after it was eventually fixated, only backtracking
to its location two fixations later. Indeed, surprisingly, when completing the
debrief questionnaire, no subject explicitly reported guessing either the predictable
order of target presentation or the symmetry rule which applied to the central letter.
(Subsequent versions of this experiment have since been carried out with similar subject
numbers, and only very occasionally did participants work out either of the strategies.)
The visual-search behaviour of the MT group at baseline was similar to that of the NT
group. The example trial from subject 21 illustrates this (figure 2b, left panel). During
block 2, however, the benefit of MT is apparent: Subject 21 commenced searching the
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
Verticalscreendimensionapixels
0 200 400 600 800 1000 0 200 400 600 800 1000
0 200 400 600 800 1000 0 200 400 600 800 1000
Horizontal screen dimensionapixels
Figure 2a. The sequence of eye fixations made by an untrained participant (subject 13) in block 1
(left: trial 52) and block 2 (right: trial 187). Arrows indicate the order of fixations in the
sequence not saccades per se. Each panel should be read from top to bottom, the first fixation
in the bottom panels showing how the search continued from the last fixation in the top panels
(in reality the sequence was continuous but is subdivided here to avoid clutter and overlap).
Where two fixations are consecutive, overlapping and not easily discernible from subsequent
return fixations `62' is indicated on the figure. The x and y axes of each quadrant represent
screen dimensions in pixels (10246768, respectively).
1738 R Dewhurst, D Crundall
periphery by returning to the target location of the previous trial, then promptly moved to
the target location of the trial in hand (figure 2b, right panel). This strategy was quite
common and is interesting because it indicates that subjects sometimes used the target
location from the previous trial as a placeholder for the upcoming trial. This preference
may suggest that, in our task, executing saccades to the exact location dictated by the MT
strategy is more demanding than returning to the previous target position for guidance.
The final point of interest in the visual-scanning behaviour of the MT participants
is that they often made consecutive re-fixations at the end of a trial, typically alternating
between the target and the central letter several times before responding. This is
important as it is in-line with the predictions of the main hypothesis. Eye movements
of this type are exemplified in the bottom right panel of figure 2b.
3.6 Central-letter processing: Initial central gaze duration
The summation of fixation durations for consecutive fixations made within the region of
interest for the central letter, from the first fixation within this area until a saccade is made
outside of it, gives an indication of how efficiently the central letter is initially processed.
It also allows one to infer whether the MT strategy employed exerts any influence
over this initial processing. These data were extracted from the eye-movement recordings,
and, overall, the mean central gaze duration was 515 ms (SD ˆ 227X8 ms). However, the
relative change analysis revealed no significant differences, and is not reported further.
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
Verticalscreendimensionapixels
0 200 400 600 800 1000 0 200 400 600 800 1000
Horizontal screen dimensionapixels
0 200 400 600 800 1000 0 200 400 600 800 1000
Figure 2b. The sequence of eye fixations made by a participant (subject 21) from the MT group
in block 1 (left: trial 23) and block 2 (right: trial 152). Arrows indicate the order of fixations in
the sequence not saccades per se. Each panel should be read from top to bottom, the first fixation
in the bottom panels showing how the search continued from the last fixation in the top panels
(in reality the sequence was continuous but is subdivided here to avoid clutter and overlap).
Where two fixations are consecutive, overlapping and not easily discernible from subsequent
return fixations `62' is indicated on the figure. The x and y axes of each quadrant represent screen
dimensions in pixels (10246768, respectively).
Training eye movements 1739
In an attempt to see whether other measures of initial letter processing yielded
any significant results, both the mean number of fixations of which the central gaze
duration consisted and the mean fixation duration for fixations contained within the
central gaze duration were extracted and analysed. Again, however, the relative change
analysis on these data revealed no significant effects. Therefore, one must conclude
that neither training, feedback, nor a combination of these factors interacting affects
the duration or amount of fixations made on the central letter.
3.7 Re-inspections of the central letter
Apart from fixations classified as comprising the initial central gaze duration, any
other fixation made on the central letter was identified as a re-fixation. For continuity,
and to avoid any further data loss, we chose to calculate mean re-inspection durations
(the average of re-fixation durations for all trials in an entire block, including those in
which no re-fixation occurred). This approach was considered more logical than calculating
the mean duration of re-fixations when they occurred, since we did not expect
re-fixations for the NT and BT groups to be as numerous. Therefore, this measure
keeps the number of trials contributing to each cell mean constant, favouring fair
comparisons. However, it should be noted that the effects on re-inspection durations
reported below remain for the more conventional measure of mean re-fixation. These
data are not presented, however, because there were indeed too few trials contributing
to certain participants cell means.
Overall the mean duration of re-inspections was 169 ms (SD ˆ 128X5 ms). There was
no main effect of training in the relative change analysis for this measure, nor was there
a main effect of feedback. Crucially, however, these factors did interact (F2 27 ˆ 3X8,
MSE ˆ 4029, p ˆ 0X036). This interaction is charted in figure 3a and reflects the fact
that the MT group has increased mean re-inspection durations, but only when external
visual feedback was presented. In contrast, the provision of feedback did not increase
the mean re-inspection durations of the other two groups; on the contrary, it seems to
have reduced them. This suggests that feedback is beneficial unless trained on the Move
centre in isolation. For the NT group, and to some extent the BT group, when feedback
was available, the initial time spent processing the central letter seemed sufficient;
therefore these groups showed a reduced need to return to it. When the MT group
received feedback, however, the initial time spent processing the central letter seemed
less than optimal; therefore these participants needed to re-inspect it for longer durations
in order to respond correctly.
,
75.0
50.0
25.0
0.0
À25.0
À50.0
À75.0
Changeinre-inspectiondurationams
0.2
0.1
0.0
À0.1
À0.2
À0.3
À0.4
À0.5
À0.6
No training
Move training
Both training
Changeinnumberofre-fixationspertrial
Feedback No feedback Feedback No feedback
(a)
(b)
Figure 3. (a) Relative change from baseline in mean re-inspection durations for each group, across
feedback conditions. (b) Reduction from baseline in the mean number of re-fixations per trial for
each group, across feedback conditions. Error bars show Æ1 SEM.
1740 R Dewhurst, D Crundall
When feedback is not provided, however, this pattern changes: all groups now show
more of a trend towards re-inspecting for comparable durations to those observed at
baseline (ie 0 on the y axis of figure 3a). In fact, when feedback is not provided, the
trend described above appears to reverse somewhat.
To assess these results, a posteriori tests were carried out. Two one-way ANOVAs
were conducted, one comparing the differences between groups in the FB condition,
and one comparing the differences between groups in the NFB condition. Only the
first of these yielded a significant between-groups effect (F2 27 ˆ 4X7, MSE ˆ 4081,
p ˆ 0X018), and Tukey HSD a posteriori comparisons showed that the difference in
this analysis lies with the increase in mean re-inspection durations for the MT group
and the decrease in mean re-inspection durations for the NT group ( p ˆ 0X015); the
NT group did not differ from the BT group ( p ˆ 0X609), and the MT group did not
differ from the BT group ( p ˆ 0X150). These analyses confirm that the MT group
re-inspects the centre for longer than the NT group, but only in the FB condition.
The same analysis was conducted on the average number of re-fixations per trial
(overall x ˆ 0X83, SD ˆ 0X58). There were no main effects, nor was the interaction
significant. However, because we had reason to believe that the MT group would
re-fixate the centre more regularly than the NT group, two independent-samples t-tests
were carried out: the first on the difference between the NT and MT groups in the FB
condition was significant (t19 ˆ À2X53, p ˆ 0X021); the second on the same groups in
the NFB condition was not significant. This mirrors the re-inspection duration analysis
(see figure 3a). Moreover, although all groups showed a slight decrease in re-fixations
(figure 3b), a finding which is consistent with practice at the task, the MT group showed
the smallest reduction. This, therefore, demonstrates that the re-inspection duration
results described above are not the artifact of a trade-off where the longer re-inspections
of the MT group are a consequence of this group making fewer re-fixations numerically.
4 Discussion
Several predictions were made from the outset regarding the consequences of training
eye movements. It was hypothesised that isolated eye-movement training directed towards
the Move centre of Findlay and Walker's model would in some way degrade the ability
to process the centrally located letter, owing to competition between the need to fixate
this item (or Fixate centre activity) and the urge to saccade towards the known location
of the peripheral target (or Move centre activity). It was predicted that this effect would
be exposed in one of two ways. First, it could appear as an effect on accuracy, with the
MT group disengaging from the central letter before having fully processed it, therefore
being uncertain of the correct response to make. Alternatively, MT participants might
re-inspect the central letter more regularly for accuracy to be preserved. Evidence for the
latter case was provided, not in the number of re-fixations per se, but in the re-inspection
durations. The data confirm that, although all groups showed comparable processing
time on the central letter in the first instance (reflected in the initial central gaze duration
measure), the MT group alone failed to process the item adequately within this time,
needing to re-inspect it for longer in order to respond correctly.
On the surface, it appears that directing eye-movement training towards the Move
centre of Findlay and Walker's model is beneficial: it improves visual-search performance
in terms of reaction times and overall search efficiency, desirable goals for any
eye-movement training strategy; yet conversely, more subtle detrimental effects on the
processing of individually fixated items can be missed. It is necessary, however, in light
of there being no differential effect of training on the initial processing of the central
letter, to relate the suggested competition between the Move and Fixate centres to
some form of covert attention shift towards the saccade target before the eyes move.
,
Training eye movements 1741
This idea can be supported by considerable evidence about covert attention (eg Engel
1971; Eriksen and Hoffman 1972; Koivisto et al 2004; Motter and Holsapple 2007;
Posner et al 1980). Koivisto et al (2004) used an inattentional blindness paradigm to
show that, when participants fixate the central region of a display and orient covertly
to peripheral targets, they fail to notice an unexpected stimulus, even when it is always
presented centrally at the point of fixation. A similar process of `looking but failing
to see' may well account for the present results. However, because in our task participants
were not explicitly instructed to keep their eyes fixed at the centre, an interesting
conclusion emerges: the delay between a covert shift of attention and a proceeding
saccade can vary according to task demands, even when subjects are free to move their
eyes. Whether the time course of covert shifts can vary is somewhat controversial in
the literature; nevertheless spatial selection in Findlay and Walker's model provides a
component which could potentially account for the effects observed here (this feature
of Findlay and Walker's framework operates much like the traditional spotlight or
zoom lens models of attention, selectively enhancing a spatially restricted region before
the eyes move). Further research with the present paradigm, using variable stimulus
onset asynchronies for the peripheral array, or a gaze-contingent window outside of
which peripheral information is obscured, would help address these issues. At present,
these remain avenues for future research.
The finding of increased re-inspection durations for the MT group provides an
explanation why some previous attempts to train eye movements may not have had
the desired outcomes (Chapman et al 2002; Donovan et al 2005; Quevedo et al 1999).
Moreover, it should raise awareness amongst researchers when developing eye-movement
training regimes for use in applied settings, because it suggests that the emphasis often
placed upon moving the eyes (eg Mills 2005; UK Department for Transport 2007) may
be misplaced. Indeed, we have shown that the most effective way to employ visual training
is to concentrate efforts towards the Fixate centre as well as the Move centre. When
this is done (as demonstrated with the BT group in the present experiment) the deficit
associated with MT alone is abolished. The training directed at the Fixate centre,
employed by the BT group, seems to reduce the processing demands of the central letter
despite its appearance remaining unchanged. This is in line with similar reports of higher
influences on the Fixate centre when stimulus characteristics are kept constant (Gould 1973;
Zingale and Kowler 1987).
However, the finding of longer re-inspection durations for the MT group was only
the case when visual feedback was presented in-between trials informing participants
of the accuracy of their responses. When an uninformative transient grey mask was
presented after each trial, irrespective of the response, participants given MT no longer
showed any evidence of competition between the Move and Fixate centres detracting
from the ability to process the central letter. One potential reason for this is that, when
executing the highly coordinated sequence of eye movements and stimulus processing
required, the MT group is better able to carry out the task when they can internally
regulate error monitoring. External feedback therefore might lead to a bottleneck in a
capacity-limited system (see Pashler 1994) where the competing demands to remember
the sequence of eye movements required, the response to be made to the central letter,
and to process the feedback mask, overload cognitive resources. As a result of this
conflict, it is the initial processing of the central letter which is hindered for the MT
participants, a problem which is remedied by looking back towards the letter for longer
durations. This does not occur for the NT and BT groups, however, because for these
participant groups one of the factors causing the bottleneck is removed: the NT group
does not have a sequence of eye movements to remember, and the difficulty of processing
the central letter is reduced for the BT group. As a result, the feedback mask has
an interfering effect on the MT group only, while it assists the NT and BT groups.
1742 R Dewhurst, D Crundall
Interestingly, previous research has shown that, when a voluntary saccade is being
planned, oculomotor capture from an abrupt onset is likely (Theeuwes et al 1999),
whereas this capture does not occur when the task is to remain fixated (Tse et al
2002). Although the feedback masks we used were global transients, it may be that they
cause a similar process of interference for the MT group when voluntarily executing
saccades dictated by the training scanpath. This may not occur for the other two groups,
however, because competition between the Move centre and Fixate centre is weaker,
and there are fewer cognitive demands causing a bottleneck.
The argument presented above suggests that the MT group prefers to internalise
error monitoring. However, how might this process be carried out? It is feasible that,
when feedback is not provided, MT participants have a higher likelihood of generating
feedback Error Related Negativity (fERNöHolroyd et al 2006), a frontal lobe ERP component
associated with evaluating the outcomes of actions in the absence of external
feedback. If this is the case, the higher and less specified levels of Findlay and Walker's
model should be revised to take account of this possibility when training eye movements.
The pattern of results observed suggests that training directed at both the Move
and Fixate centres of Findlay and Walker's model is advisable. Although MT in isolation
can be beneficial, this was only the case when visual feedback was not pertinent
to the task at hand. Therefore, because, particularly in applied settings, interaction
with the environment and awareness of visual reinforcement is vital, it seems less likely
that eye-movement training directed solely at the Move centre will be helpful. Clearly,
however, there are big differences between the experimental task we used and `real
life' scenarios; we controlled exactly the characteristics of the stimuli, therefore knew
precisely what training strategies to advise. Nevertheless, despite the differences between
this study and applied contexts, it is evident that additional focus from domain-specific
experts should be given to the processing of fixated items, in concert with assisting in
the ability to visually locate them which has been the predominant drive in recent years.
To refer back to the example of the `look but fail to see' accident causation factor
(Brown 2002) in driving, implementing visual training that simultaneously tackles the
problems of locating and processing important objects may be as simple as advising
drivers to fixate for longer. For example, ``look right CHECK, look left CHECK, then right
again'' (UK Department for Transport 2007).
Acknowledgments.We would like to thank Peter Chapman,Thomas Foulsham for useful discussions,
the Economic Social Research Council for funding this work, and two anonymous reviewers for
their helpful comments on an earlier draft of this manuscript.
References
Brown I, 2002 ``A review of the `look but fail to see' accident causation factor'' Behavioural Research
in Road Safety 11 116 ^ 124
Chapman P R, Underwood G, Roberts K, 2002 ``Visual search patterns in trained and untrained
novice drivers'' Transportation Research Part FöTraffic Psychology and Behaviour 5 157 ^ 167
Clark D, Ward P, Truman W, Bartle C, 2004 ``An in-depth case study of motorcycle accidents
using police road accident files'' Proceedings of the 14th Seminar on Behavioural Research in
Road Safety (London: Department for Transport) pp 5 ^ 20
Donovan T, Manning D J, Phillips P W, Highman S, Crawford T, 2005 ``The effect of feedback
on performance in a fracture detection task'' Progress in Biomedical Optics and Imaging
(SPIE) 6 79 ^ 85
Engel F L, 1971 ``Visual conspicuity, directed attention and retinal locus'' Vision Research 11 563 ^ 575
Eriksen C W, Hoffman J E, 1972 ``Temporal and spatial characteristics of selective encoding from
visual displays'' Perception & Psychophysics 12 201 ^ 204
Findlay J M, Brown V, 2006 ``Eye scanning of multi-element displays: I. Scanpath planning''
Vision Research 46 179 ^ 195
Findlay J M, Walker R, 1999 ``A model of saccade generation based on parallel processing and
competitive inhibition'' Behavioral and Brain Sciences 22 661 ^ 721
Training eye movements 1743
Foulsham T, Underwood G, 2008 ``What can saliency models predict about eye movements?
Spatial and sequential aspects of fixations during encoding and recognition'' Journal of Vision
8(2:6) 1 ^ 17
Godijn R, Theeuwes J, 2002 ``Programming of endogenous and exogenous saccades: Evidence for
a competitive integration model'' Journal of Experimental Psychology: Human Perception and
Performance 28 1039 ^ 1054
Gould J D, 1973 ``Eye movements during visual search and memory search'' Journal of Experimental
Psychology 98 184 ^ 195
Hajcak G, Holroyd C B, Moser J S, Simons R F, 2005 ``Brain potentials associated with expected
and unexpected good and bad outcomes'' Psychophysiology 42 161 ^ 170
Herslund M B, Jorgensen N O, 2003 ``Looked-but-failed-to-see-errors in traffic'' Accident Analysis
and Prevention 35 885 ^ 891
Holroyd C B, Hajcak G, Larsen J T, 2006 ``The good, the bad and the neutral: Electrophysiological
responses to feedback stimuli'' Brain Research 1105 93 ^ 101
Itti L, Koch C, 2000 ``A saliency-based search mechanism for overt and covert shifts of visual
attention'' Vision Research 40 1489 ^ 1506
Koivisto M, Hyona J, Revonsuo A, 2004 ``The effects of eye movements, spatial attention, and
stimulus features on inattentional blindness'' Vision Research 44 3211 ^ 3221
Logan G D, 1996 ``The CODE theory of visual attention: An integration of space-based and
object-based attention'' Psychological Review 103 603 ^ 649
Martens M H, Fox M R J, 2007 ``Do familiarity and expectations change perception? Drivers'
glances and response to changes'' Transportation Research Part FöTraffic Psychology and
Behaviour 10 476 ^ 492
Mills K, 2005 Disciplined Attention: How to ImproveYour Visual Attention WhenYou Drive (London:
Profile Books)
Morrison R E, 1984 ``Manipulation of stimulus onset delay in readingöevidence for parallel
programming of saccades'' Journal of Experimental Psychology: Human Perception and Performance
10 667 ^ 682
Motter B C, Holsapple J, 2007 ``Saccades and covert shifts of attention during active visual
search: Spatial distributions, memory, and items per fixation'' Vision Research 47 1261 ^ 1281
Munoz D P,Wurtz R H, 1993 ``Fixation cells in monkey superior colliculus. 1. Characteristics of cell
discharge'' Journal of Neurophysiology 70 559 ^ 575
Munoz D P,Wurtz R H, 1995a ``Saccade-related activity in monkey superior colliculus. 1. Characteristics
of burst and buildup cells'' Journal of Neurophysiology 73 2313 ^ 2333
Munoz D P, Wurtz R H, 1995b ``Saccade-related activity in monkey superior colliculus. 2. Spread
of activity during saccades'' Journal of Neurophysiology 73 2334 ^ 2348
Pashler H, 1994 ``Dual-task inference in simple tasksödata and theory'' Psychological Bulletin
116 220 ^ 244
Posner M I, Snyder C R R, Davidson B J, 1980 ``Attention and the detection of signals'' Journal
of Experimental Psychology: General 109 160 ^ 174
Quevedo L, Sole J, Palmi J, Planas A, Soana C, 1999 ``Experimental study of visual training effects
in shooting initiation'' Clinical Experimental Optometry 82 23 ^ 28
Reichle E D, Pollatsek A, Fisher D L, Rayner K, 1998 ``Toward a model of eye movement control
in reading'' Psychological Review 105 125 ^ 157
Theeuwes J, Kramer A F, Hahn S, Irwin D E, Zelinsky G J, 1999 ``Influence of attentional
capture on oculomotor control'' Journal of Experimental Psychology: Human Perception and
Performance 25 1595 ^ 1608
Tse P U, Shienberg D L, Logothetis N K, 2002 ``Fixational eye movements are not affected by
abrupt onsets that capture attention'' Vision Research 42 1663 ^ 1669
UK Department for Transport, 2007 ``Think Bike'' Campaign (London: UK Department for Transport)
Veen V van, Holroyd C B, Cohen J D, Stenger V A, Carter C S, 2004 ``Errors without conflict:
Implications for performance monitoring theories of anterior cingulate cortex'' Brain and
Cognition 56 267 ^ 276
Zingale C M, Kowler E, 1987 ``Planning sequences of saccades'' Vision Research 27 1327 ^ 1341
ß 2008 a Pion publication
1744 R Dewhurst, D Crundall
ISSN 0301-0066 (print)
Conditions of use. This article may be downloaded from the Perception website for personal research
by members of subscribing organisations. Authors are entitled to distribute their own article (in printed
form or by e-mail) to up to 50 people. This PDF may not be placed on any website (or other online
distribution system) without permission of the publisher.
www.perceptionweb.com
ISSN 1468-4233 (electronic)