Introduction
The principle question asked throughout this research is whether upright faces
capture visual-spatial attention. Empirical support that it does, will provide an
explanation for the Face-Detection Effect {Purcell, 1986 #27;Purcell, 1988
#24;Shelley-Tremblay, 1999 #121}. The Face-Detection Effect refers to the
phenomenon whereby upright faces were faster to be localised than its inverted and
scrambled counterparts. If spatial attention involuntarily orients towards upright faces,
it would explain why participants are quicker to be consciously aware of their
presence and their spatial position. The 4 experiments reported in this chapter test the
hypothesis that a face in its canonical orientation is more easily detectable than its
inverted counterpart because it captures spatial attention. This is achieved by pitting
this putative property of upright faces against the attentional demands of a classic
visual search task. If faces capture attention, then their mere presence in the periphery
of the visual search array should distract and hence, impair visual search performance;
as measured by target-detection timings and accuracy.
The introductory sections of this chapter will discuss 2 main theoretical
themes – namely, attentional capture and the attentional “spotlight” – as researched
using the classical experimental design of visual search tasks. First though, the visual
search paradigm will be described alongside the two different patterns of searches that
can occur depending on the elements used in a search array; that is, parallel and serial
visual search. Attentional capture is said to explain parallel searches and serial search
patterns can be described with the metaphor of an attentional spotlight.
Hence, it will be useful to discuss attentional capture within the context of
parallel search performances and the search conditions that induces it. A coverage of
stimulus properties known to capture spatial attention will be discussed within the
context of the visual search design. Here, parallels will also be made between these
stimulus properties and how they might relate to faces. Following this, the commonlyused metaphor of an “attentional spotlight” will be discussed, given its usefulness as a
theoretical construct within the attentional literature. As before, its theoretical
implications to our experiments will be addressed.
Visual search design
The visual search paradigm was introduced in the previous chapter as a means
of studying spatial attention. It is thus a useful paradigm that has been employed in all
the experiments reported in this chapter that set out to study whether the upright faceconfiguration possesses attentional-capture properties. A typical visual search
experiment involves requiring participants to search for a pre-specified target among a
varying number (set-size) of non-target distractors (e.g. {Treisman, 1985
#162;Treisman, 1988 #122}. As mentioned previously, task-performance on in visual
search experiments are measured in terms of reaction-time and accuracy, as a function
of set-size. Properties of visual search can be inferred from the slope-function of these
performance variables in relation to varying set-size. In this section, visual-search
experiments conducted in the past and their consistent main findings will be presented
in greater detail. This is to grant the reader a better appreciation of the general
experimental design in this chapter.
Commonly, search is labelled as highly efficient when the target can be
distinguished from all non-targets by the minimum of a single feature. For example,
searching for a red target embedded within an array of green distractors. In such cases,
target detection times are hardly affected by set-size at all {Wolfe, 1998 #124}. Such
efficient searches are sometimes described as parallel searches, whereby the target
‘pops out” by virtue of its salient visual difference to its neighbours. In contrast,
search time for targets that can only be distinguished from its distractors by a
combination of features, increases linearly with incrementing set-size {Wolfe, 1998
#124}. For example, searching for a red X within an array of red Os and green Xs.
Searches that vary in accordance to set-size are termed, serial searches. These terms
“parallel” and “serial” describe the theoretical search process for targets; that is, in
‘parallel searches’ every element in the array can be analysed for a single
distinguishing feature simultaneously, whereas serial search for a target defined by a
conjunction of features can only be achieved by analysing every element in the search
array one by one {Treisman, 1988 #122}.
There are several explanatory accounts for the qualitative difference between
parallel and serial search. The feature-integration theory (FIT) is one such account. In
the FIT, visual search is based on the use of topological featural maps, each
responding retinotopically to the presence of a specific visual attribute e.g colour,
line-orientation, etc. {Treisman, 1982 #68;Treisman, 1988 #81;Treisman, 1998 #65}.
A target that is distinguishable from its surrounding on the basis of only one feature,
requires the use of only one featural map. Parallel search occurs because by using just
one featural map, all the items in the visual scene can be simultaneously assessed for
this given target feature and the target containing the critical attribute simply ‘popsout’ on the featural map. In contrast, serial searches occur when a critical target is
different from its neighbouring distractors, not by a single feature, but by possessing a
unique combination of featural attributes. According to the FIT, the spatial location of
the critical target cannot be solved by the use of only one featural map in such
instances. The critical target is the item that possesses all the required features and
hence, its position has to be topologically congruent across all the corresponding
feature maps that make up the target criterion. Treisman and colleagues
(1982;1988;1998) claim that to do so, it is necessary to examine each item
individually and thus, search can only progress serially. There are several alternative
theories to the FIT, that can also account for these qualitative differences in visual
search processes (see for example: Guided search model {Wolfe, 1994 #134};
similarity theory {Duncan, 1989 #112}). Nonetheless, they are all in agreement that
target-distractor similarity is an critical determinant of task difficulty in visual
searches. Parallel searches occur when the target is highly distinguishable from its
surrounding distractors and serial searches take place when the critical target shares
visual features with its neighbours. In Experiment 1, this factor is used to control for
the experimental variable of task-difficulty.
Another consistent finding is that the slope of the reaction-time to set-size
function of target-absent trials are often twice as steep as target-present trials
{Sternberg, 1966 #114;Wolfe, 1994 #134;Wolfe, 1998 #124}. This is particularly true
with serial searches and indicates that serial searches are self-terminating. With serial
searches, participants search the array by by examining each element individually and
in serial progression. With target-present trials, the search ends when the target is
detected. This can happen at any point in the course of the serial search depending on
the search ‘route’.With target-absent trials, however, the entire visual array has to be
fully examined before target absence can be accurately determined. Therefore,
responses latencies are expected to be longer on target-absent trials, in comparison to
target-present trials. This finding is taken up and used as a controlled variable for
task-difficulty in Experiment 2,3 and 4.
Though a modest set-up, findings from visual search experiments have over
the past 20 years made far-reaching contributions to our understanding of the
processes that underlie visual cognition. We intend to use the same paradigm to
investigate whether faces, presented in their normal upright orientation, capture
attention.
Attentional capture
An alternative way of describing parallel searches is through attentional
capture. The term attentional capture is associated to known visual attributes that
induce reflexive and involuntary orienting of attention towards their location visual
scene. For example, it is well-established that sudden changes in the visual scene i.e.
abrupt visual onsets, have a special propensity to attract attention {Yantis, 1990
#115;Jonides, 1988 #116;Yantis, 1984 #117}. An instance of such a cue might be the
sudden dimming or brightening of a particular location in the visual scene. One of the
key properties associated with this exogenous form of attentional control is that it is
involuntary. That is to say, that attention capture depends entirely on the visual
stimulus’ attributes, not on the goals of the observers. Thus, visual search times
decreases with abrupt-onset targets and increase with abrupt-onset distractors even
when, the location of the abrupt-onset distractor is completely uncorrelated with the
location of the target (e.g. {Yantis, 1990 #115}). Attentional orienting of such a
nature can be described as stimulus-driven.
Relevant to the thread of the discussion so far, has been a recent attribution of
attentional capture to feature “singletons” (e.g. {Joseph, 1996 #120;Theeuwes, 1991
#118;Theeuwes, 1992 #119}. Feature “singletons” are described as items that possess
at least one featural difference that renders it highly discriminable from its neighbours
- such as target items in parallel searches. For instance, Neisser (1964) noted that
curved letters e.g. C are more easily detected as a single member of a letter-string of
angular letters e.g. K, and vice versa. The measure of target-distractor similarity are
often based on the primitive dimensions of colour, line-orientation, motion direction –
visual properties that have neurological correlates in the early visual pathway{see
{Bravo, 1992 #140;Cheal, 1992 #222} for a list of “features”}.
The obvious relation between attentional capture and parallel visual search
performance is this common statement that highly distinctive items stand-out in the
visual scene such that the amount of visual clutter of irrelevant items don’t matter.
Nonetheless, the ultimate test in proving that feature ‘singletons’ can capture attention,
lies in the successful demonstration that ‘singletons’can draw attention towards itself,
even in the instances when they are completely task irrelevant or, when doing so is
highly detrimental to primary task performance.
One way of doing so is by having an irrelevant distractor as a feature singleton
and to measure its attention-capturing property as an index of how much it prolongs
the target detection time. For example, the time it takes to detect an oblique line (/) in
an array of many Os (or an O amongsts oblique lines) is lengthened simply by having
a couple of uniquely coloured distractors {Pashler, 1988 #123}: Experiment 7).
Participants are explicitly instructed to ignore the irrelevant distractors at all times.
This simple experiment illustrates spatial attention as a a finite resource that can be
automatically recruited, against the participants intentions, by task-irrelevant albeit
visually unique and salient distractors.
This line of research was furthered by Theeuwes (1991, 1992) when his
participants were asked to perform the easy search of a unique shape (e.g . diamond)
within an array of distractors (eg. circles). A line was enscribed within each shape,
target and distractors, that was either horizontal or vertical and successful detection
was noted if participants were able to accurately state the orientation of the line,
within the target. On half the trials, all the elements in the array were of the same
colour and in the remaining trials, just one distractor was of a different colour.
Participants were informed that colour was task-irrelevant. Nonetheless, the presence
of the uniquely coloured distractor significantly slowed down target detection time.
In a similar experiment, the test display was segmented into 4 distinct spatial
quadrants and in the test phase, participants were required to search for the letter L
which was embedded in an array of distractor Ts. {Joseph, 1996 #120}. Participants
responded by specifying the quadrant that contained the target. Preceding the testphase, however, was a non-informative cue array that contained a “singleton” in either
the same quadrant as the one containing the target in the upcoming test-array, or any
one of the other non-target quadrants. Participants were told that this preceding cue
array was non-informative and did not predict the target position in the test array.
Nonetheless, detection times were fastest when the “singleton” in the cue array fell
within the same quadrant as the test array.
These experiments show that a feature singleton can reflexively orient
attention to its location in space, even when this orienting does not benefit task
demands. This is seems to be especially true when search in the test trial is easy, such
as with parallel searches. The reason for this is because attentional capture is believed
to work best when there is no top-down control in place {Theeuwes, 1991 #118}. In
the first chapter, it has been suggested that upright-faces could capture attention.
Against the backdrop of research concerning attentional capture by featural
‘singletons’, it seems that upright faces could be doing so in the same fashion. In
Chapter 1, it has been suggested that the configural pattern of an upright face could
have equivalent status as certain primary features e.g. colour, orientation. Like these
primary visual features, there exists neural substrates that respond specifically to its
presence {Rolls, 1984 #166;Perrett, 1982 #164;Perrett, 1984 #165}. If this is true,
observers could be particularly vulnerable to attentional capture by upright faceconfigurations when conducting parallel searches. This hypothesis is put to the test in
Experiment 1.
Visual Attentional Spotlight
The “attentional spotlight” is a convenient metaphor used to describe visual
spatial attention {Treisman, 1988 #81}. The spatial extent of this “spotlight” is
believed to range from 1° to 20° of visual angle (Eriksen, & Hoffman, 1973; Hughes
& Zimba, 1985), depending on the task. Importantly, information processing is
quicker and more efficient for items within the “spotlight” and inhibited if found,
outside the “spotlight’s” perimeter {Brefczynski, 1999 #74;Posner, 1990
#110;Slotnick, 2002 #80}. In fact, retinotopic mapping can be established between the
foci of attended stimuli and the topography of cortical activations, highlighting the
spatial aspect of visual attention and justifying the use of this metaphor {Slotnick,
2002 #80;Brefczynski, 1999 #74}.
The “spotlight” metaphor is best used to describe the guided control of visual
attention in serial search tasks (e.g. {Treisman, 1988 #122}. As mentioned previously,
serial search occurs when the target is embedded in a visual array of similar
distractors. The more similar the target is to its surrounding distractors, the smaller the
area of the “spotlight” and hence, more time will be required for target detection
{Neisser, 1964 #113;Duncan, 1989 #112}. Therefore, the serial search process can
described as a process in which a small attentional “spotlight” has to examine every
single element in the scene one after another in order to find the specified target. This
is used to account for the linear relationship between target-detection latencies and the
set-size of distractors in serial searches {Wolfe, 1998 #124}.
A narrow attentional spotlight, induced by high target-distractor similarity can
be postulated to be more vulnerable to attentional capture by irrelevant distractors.
Because the spotlight is retinotopic {Brefczynski, 1999 #74}, being attracted away
from its task-relevant visual field space would severely impair task performance. The
experiments reported in this chapter are designed to operate on this principle. If
upright faces do capture spatial attention, it should strongly impair serial search
performance of a critical task that is situated in a different visual space location to
itself.
Attentional capture by upright faces: serial or parallel search
mechanism?
The face-detection effect has only been tested in an uncluttered scene {Purcell,
1986 #27;Purcell, 1988 #24;Shelley-Tremblay, 1999 #121}. To recap, participants
had to detect the presence of an image which could have been an upright, inverted or
scrambled face. It was found that performance was best for upright faces in that it
required the least amount of presentation time in order to be successfully detected.
An offered explanation for this effect, is that upright faces capture spatial
attention. However, it is not clear if it might capture spatial attention as a feature
singleton might, in a visual array undergoing parallel search (cf. {Theeuwes, 1991
#118;Joseph, 1996 #120;Pashler, 1988 #123}; or by reflexively orienting a very
narrow attentional spotlight - induced by target-distractor similarity – away from the
critical target. Experiment 1 was designed to clarify this issue by utilising faces as a
peripheral distractor with visual search arrays that either induced parallel or serial
search.
More importantly, the claim that upright faces could capture attention was
tested by measuring their influence on visual search performance, on a central
primary task. In the following experiments reported, the visual search consisted
entirely of of non-face objects i.e. letters of the alphabet. This was contrived such as
to avoid the conscious use of face-identification processes that could have
inadvertently nullify the ability of upright face-configurations to draw attention to
themselves (cf. {Kuehn, 1994 #135;Nothdurft, 1993 #136}. At all times, participants
were focused on detecting a pre-specified letter of the alphabet i.e. ‘K’ and were
explicitly instructed to ignore the appearance of any peripheral task-irrelevant
distractors.
The hypothesis predicts that visual search performance would be significantly
impaired by the presence of upright faces on both parallel and serial search trials.
Upright faces will have a larger impact on parallel searches than serial searches if they
behave like feature singletons. The reverse will be true if upright faces orient spatial
attention in a general fashion.
Experiment 1:
Attentional capture by Upright-face distractors on Parallel vs. Serial searches
de Gelder and Rouw (2001) argues that the face-detection effect arises from an
innate visual predisposition to upright face-configurations {de Gelder, 2001 #33}.
This experiment sets out to operationalize this hypothesis within the context of welldocumented search processes (e.g. {Wolfe, 1994 #134}.
In this experiment, participants were instructed to search for a pre-specified
target in a centrally located visual array. Meanwhile, a peripheral task-irrelevant
distractor appeared on half of the trials. The current hypothesis predicted that the taskirrelevant distractor should impair visual search performance by distracting attention
away from the central search display, especially if it was an upright faceconfiguration. Thus, the measure to which upright faces capture spatial attention was
considered to be the extent to which an accompanying task-irrelevant upright faceconfiguration impaired search performance by distracting spatial attention away from
the task-relevant search array. Half of the task-irrelevant distractors that appeared
were inverted face-images which acted as controls. The face-detection system that
brings about involuntary orienting of attention is believed to be respond specifically to
coarse-grained image information found in a face i.e. configural pattern of a face
{Purcell, 1988 #24;de Gelder, 2001 #33}. Thus, the inverted-face images were not
expected to draw attention away from the critical search array in the same way as an
upright face image was predicted to.
In addition, the influence of upright face-configurations on two different types
of search processes (parallel and serial) were considered. This is to provide a
qualitative description of the manner by which upright face-configurations capture
attention. As explained in the preceding section, it is yet unclear whether an uprightface configuration capture attention by visual-salience, and “popping-out” of a
cluttered visual scene as does a feature-singleton when parallel search processes are in
action; or whether it does so by drawing a narrow attentional ‘spotlight’, that is
deployed during serial searches, towards itself. Parallel searches are deemed easy and
induced by low visual similarity between the target and its non-targets in the visual
array. The converse is true for serial searches that are believed to be brought about by
high visual similarity between targets and non-targets {Duncan, 1989 #112;Wolfe,
1994 #134}. Pitting the presence of an upright face-configuration against either of
these search processes will clarify this issue.
Design
On each trial, participants were required to search a central visual search array
for a pre-specified target. There were two levels of task-difficulty that was controlled
by target-similarity between target and non-target items in the search array (see
Materials section for details). The experiment consisted of 6 blocks of trials, presented
in random order. Each block contained 32 trials that were equally divided for taskdifficulty. Altogether, there were 192 experimental trials on which half were
presented with an accompanying task-irrelevant distractor and half without. On the
trials with a task-irrelevant distractor, half the distractors were upright faces and the
remaining half were inverted faces.The experiment can thus be described as a 2 (task
difficulty: easy vs. difficult) x 3 (distractor-type: no distractor vs. face-distractor vs.
nonface-distractor) repeated-measures design. Two dependent measures were
recorded; namely, reaction time and performance accuracy on the primary visual
search task.
Materials
Each trial was made up of a visual search display that half of the time, was
accompanied by a peripheral task-irrelevant distractor. This section covers how both
elements of the trial was created and paired together. To avoid confusion, the terms
target and non-target will be used exclusively to refer to items of the visual search
array, while the term distractor refers to task-irrelevant distractors that are either
upright or inverted faces.
Visual search display:
Two sets of four alphabetic letters each made up the visual search array for the
primary task. The two sets of letters differed from each other with regards to their
visual-similarity to the target letter; that is, the letter ‘K’ which can be described as an
angular letter {Neisser, 1964 #113}. Thus, one set of letters consisted solely of
angular letters (target-similar: X, Y, N, V), and the other set of highly dissimilar
circular letters (target-dissimilar: O, C, D, G).Each visual search array was a 2 x 4
arrangement; or 2 rows of letter-strings, each comprising 4 letters.
24 unique letter-strings were rendered for each set by applying a full
permutation to the corresponding items. Following this, letter-strings containing the
target item were derived by a process of simply replacing an element of a letter-string
with the target letter ‘K’, By repeating this process for each non-target letter, across
all the 4 possible positions, 96 unique letter-strings containing the target letter ‘K’
were derivable for each set. Each of these target letter-strings were randomly paired
with a non-target letter-string made up of items from the same set, resulting in a total
of 192 visual search arrays. That is to say, the entire search array would have
consisted entirely of either target-similar or target-dissimilar non-targets.
The visual search arrays were counter-balanced for position of the target-letter
string about a central fixation cross. In addition, the position of the target letter-string
within the array was also counter-balanced across the target’s position within the
letter string, as well as it’s position relative to the distractor – if present.
Task-Irrelevant Distractors (Upright and Inverted Faces):
Still images of 24 different faces were selected from a bank of images taken of
police trainees supplied by the Home Office [see Bruce et al. (1999) for more details
on this set of materials]. Each image was edited using Adobe Photoshop ver. 6.0, to
only reveal the face, as shown in Fig. 2.1a. The faces are looking straight ahead, with
a neutral face expression.
Corresponding inverted face-distractors were created by a 180° rotation of
each of the 24 images, using Adobe Photoshop ver. 6.0. An example is given in Fig.
2.1b. Care is taken that each upright distractor and its inverted counterpart was
equivalent with regards to brightness, contrast, size and featural detail.
Fig 2.1: Examples of the task-irrelevant distractors created from photorealistic face-images: a) upright
face-type distractor; b) inverted face-type distractor.
Each distractor image was used twice in the experiment, a total of 96
distractors. Each image appeared in 1 of 6 specified positions, these 6 positions were
fixed evenly-spaced points on an imaginary circle around the visual search array (see
Fig. 2.2c). The image position was specified by a list and a latin-square was
conducted such that each image was presented in all of the given 6 positions, across
all the participants.
Apparatus:
The experiment was controlled by an experimental software package Psyscope
v.1.12 on a Macintosh G4. The display was presented on a 17’ flat-screen monitor
that was distanced 57cm from the head rest. A fixed head-rest was aligned to the
fixation cross of the experiment and ensured that a consistent visual display size was
maintained throughout the experiment. At 57cm, 1cm on the screen subtends
approximately 1º visual angle on the retina.
Participants
30 first-year Psychology undergraduates (age range: 18-19 years) participated
in this study for course credits. All participants reported normal or corrected-tonormal vision.
Procedure:
On each experimental trial, participants performed a two-alternative forcedchoice (2AFC) letter detection task that involved searching a 2 x 4 array - containing
8 letters of the alphabet - for a target letter i.e. ‘K’. Each trial began with a fixation
cross that lasted for 1000 msec. Following this, the critical search array was presented.
The array was in the form of two letter-strings, one situated above the fixation cross
and the other, below it. Only one letter-string contained the target letter ‘K’. This
critical display lasted for only 200msecs and the participants’ task was to identify the
letter-string containing the target with key-press response: ‘Y’ - letter-string above the
fixation-cross; ‘B’ - below fixation-cross. The timings of each experimental trial are
illustrated in Fig. 1 and the inter-trial interval was 500ms. Participants were required
to keep both index fingers on the response keys throughout the experiment and
handedness-dominance was counter-balanced across participants.
Participants’ were encouraged to respond as quickly as possible, without
making errors. Corrective feedback was also provided, to discourage accuracy-speed
trade off. The mean reaction time (RT) and accuracy of their responses were treated
as measures of their primary task performance.
On half of the trials, this array was accompanied by a distractor (upright or
inverted face) that was positioned in any of 6 possible positions - equidistant from and
evenly spaced around the letter array. Figure 2.2 provides an example of a trial timeline. The inter-trial interval was 500ms.
Fig. 2.2: a) Example of a typical visual search trial in Experimental 1 and 2. In this example, critical
search array consists of target letter ‘K’ amongst visually-similar (angular) letters. Inset b) provides
an example of an easy search (Experiment 1); c) possible positions of the task-irrelevant distractors.
The task-irrelevant distractors subtended a circle of 4º radius and were
presented in any of six positions, evenly spaced along the circumference of a circle
centred 9º about the fixation cross. Letter strings were positioned 0.5º above and
below the fixation cross. Each unit in the search array subtended 0.2º x 0.2º and the
dimensions of the entire search array were 1.3º x 1.1º.
Results:
Two measures were taken for task performance on the letter-detection task;
namely, accuracy and reaction time.
A 2 (Task Difficulty) x 3 (Distractor-type) ANOVA was conducted on each
set of dependent variable. For measures of accuracy, there were main effects of taskdifficulty (F(1,29)=184.4, p<0.001) but not of distractor-type (F(2,58)=0.096, p=0.91).
Graph 2.1 gives a summary of the results across conditions. The manipulated variable
of task-difficulty has been proven to be effective. Searches for the target found
amongst visually-similar non-targets were less accurate than when visually dissimilar
non-targets made up the array instead. In addition, there is no significant interaction
was between task-difficulty and distractor-type (F(2,58)=0.21, p=0.815). This further
indicates that the distractor-type has no differential effect on either type of search
process.
Performance accuracy (%)
100
95
90
85
80
75
easy
difficult
Task-difficulty levels
absent
inverted face
upright face
Graph 2.1: Comparison of accuracy scores in task-difficulty and serial order conditions
across distractor-types. Error bars represent standard errors.
The same pattern of results was repeated for the measure of reaction-time.
Again, main effect of task-difficulty was demonstrated, but not for distractor-type.
Response latency is lower for searches of the target in an array of visually-dissimilar
non-targets compared to visually-similar non-targets (F(1,29)=47.5, p<0.001). As
with the measure of accuracy, the type of distractors do not influence performance on
the primary as measured by reaction time (F(2,58)=0.266, p=0.77). Again, there is no
interaction between the variable of task-difficulty and distractor-type, indicating that
the distractors do not affect either type of searches any differently from each other. A
summary of the RT means are presented in Table 2.1.
Easy-task
Mean Reaction
Time (msec)
Difficult-task
No
distractor
Inverted-face Upright-face No
Inverted-face
distractor
distractor
distractor distractor
Upright-face
distractor
567.3
557.4
564.2
650.0
659.9
670.9
18.7
17.9
22.2
26.4
25.7
Standard Error. 27.9
Table 2.1: Summary of mean RTs (msecs) and standard errors across experimental conditions
To analyse these results in greater detail, a post-hoc analysis was conducted
that collapsed the trials that contained a distractor, regardless of whether it was an
upright or inverted face. Thus a 2 (task difficulty: easy vs. difficult) x 2(distractortype: absent vs. present) repeated measures Anova was conducted for this purpose.
The analysis reiterated the above findings, with significant main effects of taskdifficulty for both accuracy measures (F(1,29)=182.6, p<0.001) and reaction times
(F(1,29)=32.8, p<0.001). However, the mere presence of distractors failed to have
any significant influence on the primary task performance at all, across both measures
of accuracy (F(1,29)=0.190, p=0.67) and reaction time (F(1,29)=0.12, p=0.73).
Discussion:
This experiment was designed to investigate the cognitive mechanism that
underlie the face-detection effect reported in Purcell and Stewart (1988). The claim is
that participants are quicker to be conscious of the position of an upright facial
configural pattern than control images containing the same visual elements, but in an
uncommon orientation i.e. inverted faces. This is even when the images were
presented so briefly that image-identification was kept at chance-level.
To explain these results, some authors have suggested that visual-spatial
attention might be selective for familiar visual configural patterns, in particular faces
(Purcell & Stewart, 1988; de Gelder & Rouw, 2000). When processing a complex
visual scene, an upright facial configuration could play the role of a highly salient cue,
one that guides the “attentional spotlight” such that extra processing is involuntarily
dedicated to its visual proximity. If this is so, and if visual-spatial attention can be
assumed to be a limited resource, the simultaneous presentation of upright faces
should effectively distract and hence, impair performance on the letter-search task in
this experiment. However, our results fail to conform to this given prediction.
In Experiment 1, irrespective of whether the search array was accompanied by
either an upright or inverted face-distractor, performance on the primary task was the
same, on both measures of accuracy and RT. In fact, task performance was
completely unaffected by the presence of a distractor. The complete inefficacy of the
distractors in impairing primary task performance, might provide insight into possible
task confounds that might be inherent in the experimental design.
The pattern of our results confirm that the letter-detection task worked as
expected and employed the same attentional mechanisms as that reported in classic
visual search experiments. Targets were consistently harder to detect in an array that
consisted of visually similar non-targets and were more noticeable when placed
amongst dissimilar non-targets {Duncan, 1989 #112;Neisser, 1964 #113}.
Furthermore, these results suggest that the presence of face-distractors – both upright
and inverted - do not influence the search-process at all; that is, by capturing visualspatial attention.
Nonetheless, these results are arguably inconclusive as they could have been
confounded by a task-solving strategy brought about by the given task-response. In
this experiment, participants were asked to make a 2-force choice decision to indicate
the target position; that is, above/below the fixation cross. Post-experimental
discussions with the participants revealed that on occasion, the task was solved by
inference. When failing to detect a target in a letter-string, participants sometimes
inferred that it was to be found in the other. Thus some participants adopted the
strategy of monitoring only one letter-string. Such a strategy could have diminished
the influence of distractors on the visual search task, by reducing the search area to
such an extent that it precluded the distractors entirely. In addition, the results suggest
that the presence of distractors, failed to impair search performance completely. This
is highly unlikely and raises the possibility that even in the instance that distractors
were effective in impairing search performance, participants could have returned an
accurate response by making an inferential decision.
Another matter to note, is the high performance on searches for the target in a
visually-dissimilar array i.e. easy-task. The range of accuracy scores were 94.1% to
94.9%. The results in the task-easy condition could be inconclusive due to ceiling
effects. Also, it could also mean that the task-easy condition is so effortless that it is
able to offset any detrimental influence that might be exerted upon it by an upright
face.
The next experiment is designed to take these likely confounds into account.
In the following experiment, modifications were adopted to: i) prevent inferential
responses, ii) reduce the overall visual display size to fall within foveal region and, iii)
to minimise the possibility of a ceiling effect in primary task-performance.
Experiment 2: Presence Detection task
Experiment 1 failed to confirm the primary hypothesis that upright faces
“capture” visual-spatial attention. Nonetheless, there existed task confounds in the
previous experiment that prevented us from making any conclusive findings.
Experiment 2 is a modification of the previous one, that attempts to resolve these
confounding issues.
The basic design of the experimental trial and timings remained the same as
Experiment 1, with the following modifications. Firstly, the primary task has been
modified to prevent participants from making inferential responses. For Experiment 2,
participants were required to decide whether or not the target was present or absent on
each trial, rather than indicate the letter-string that contained it. Therefore, unlike
Experiment 1, only half of the trials in this experiment contained a target. The benefit
of this modification was that participants could no longer deliver an accurate response
based on inference. The only way to be truly confident about target-presence was to
actually find the target.
As a second modification, dissimilar non-targets are not used at all. Instead,
the all the visual search arrays used in this experiment consisted only of visually
similar non-targets e.g. ‘N’; that is, angular letters similar to the target letter ‘K’. As
noted in the discussion section of Experiment 1, performance on the search arrays
using visually dissimilar non-targets e.g. ‘C’, were exceedingly good. This resulted in
high performance accuracy that hinted of ceiling effects. To avoid this, only the visual
search arrays consisting of visually similar non-targets are used in Experiment 2. This
also served the advantage of reducing the number of trials to a manageable level. This
is because whereas previously, all trials contained a target-present search array,
changing the task-response required doubling the number of trials to accomodate both
target-present and target-absent trials.
With the removal of trials that contained search arrays of visually dissimilar
non-targets, the independent variable of task-difficulty had to be replaced by targetpresence instead. Trials that contain a target are assumed to be easier than trials that
do not because target-absent trials are believed to require twice the amount of time
needed by target-present trials for successful performance {Sternberg, 1966
#114;Wolfe, 1994 #134;Wolfe, 1998 #124}. This can only be explained by
considering that one can only be confident of a target’s absence by examining every
element in the search array. This is not true of target-present trials. Search for targets
embedded amongst visual similar non-targets is believed to be serial and selfterminating on target-present trials. Hence, the participant is, on average, likely to
detect the target halfway through a thorough scanning of the search array, on a targetpresent trial. With limited viewing times imposed upon the critical search array, this
consistent visual search finding should be reflected by lower accuracy scores and
slower reaction timings on target-absent trials.
The removal of trials with dissimilar non-target means that we are now unable
to test the influence of upright faces on parallel searches for a singleton. However,
recent evidence indicates that serial search performance starts to resemble parallel
searches with practice {Sireteanu, 1995 #138}. That is to say, that search performance
is efficient and no longer influenced by an increase in the aray-size . This might be
due to an enlargement of the “spotlight’s” area; or a more effective distribution of
attention across the test display. In fact, certain authors suggest that serial and parallel
searches are not distinct processes, but instead fall at two ends of a continuum that is
modulated by learning {Bravo, 1992 #140;Townsend, 1990 #139}. To look into this,
an additional variable of serial order is considered in this experiment, by comparing
performance on the first 3 blocks of trials against the last 3 blocks. Given that
inference is no longer a valid strategy in this experiment, any serial-order effect
might be taken to support this view of a mutable transition from serial to parallel
search performance that takes place with practice. If so, the second half of the
experiment could be treated as the equivalent of our parallel search trials and we
should expect an interaction between serial-order and distractor-type, if upright faces
influence search performance of parallel searches but not serial search (or vice versa).
Finally, to ensure that the presence of task-irrelevant distractors were noticed,
the retinal size of the visual display was reduced in Experiment 2. This was effected
by increasing the distance between the display and headrest to 96cm. By doing so, the
search array and distractor now fell within the foveal region, about the fixation cross;
that is, approximately 5.2° by 5.2°. Hence, the viewing conditions for the critical
search array and task-irrelevant distractors were equivalent.
As before, the primary hypothesis is that an upright face-image is a more
effective distractor than its inverted counterpart. Therefore, the main prediction is that
trials containing the former - and not the latter - will capture visual-spatial attention
and hamper its application to the primary task. This should be reflected by lower
accuracy rates and slower RTs in primary task performance. In addition, the first 3
blocks of experimental trials are expected to reflect serial search processes and the
subsequent 3 blocks as parallel processes (if practice effects can be demonstrated). By
examining the interaction between serial-order and distractor-type, we might be better
able to pin-point the type of visual search that is more vulnerable to an upright face’s
distractability.
Design:
This experiment has a 2 (serial order: first 3 blocks vs. last 3 blocks) x 2
(target presence: present vs. absent) x 3 (distractor-type: no distractor vs. facedistractor vs. nonface-distractor) repeated-measures design. Similar to Experiment 1,
there are 6 blocks of 32 trials each, presented in random order and equally
representative of target present and absent trials within each block; there is a total of
192 experimental trials. Half are presented with an accompanying task-irrelevant
distractor and half without. On the trials with a distractor, half the distractors are
upright faces and half are inverted faces. The main dependent variables are measures
of primary task performance; namely, reaction timings and response accuracy.
Participants:
35 first-year undergraduates with normal or corrected-to-normal vision took
part in this experiment for course credits. The age-range is 18 to 20 years old.
Materials:
The materials in this experiment were the same as Experiment 1, with the only
changes being the complete replacement of search-arrays that contained target
dissimilar non-targets e.g. ‘C’, with target-absent search arrays comprising target
similar non-targets. This was to facilitate for the condition of target-presence. Targetabsent visual search arrays were created by the following procedure.
The search-arrays comprising non-targets similar to the target letter ‘K’ were
duplicated. For this duplicate set, the target-letter was replaced by a non-target letter from the set of visually-similar letters - that was not already present in the letter-string.
Thus, there were 192 search-arrays in total, half containing a target-letter ‘K’ and half
without the target.
Apparatus:
The apparatus set-up was the same as in Experiment 1, with the exception of
increasing the distance between the video display unit and the fixed chin-rest to 96cm.
This modification meant that 1cm on the screen subtended approximately 0.59° on the
retina.
Procedure:
The trial proceedings and timings remained unchanged from Experiment 1.
However, instead of determining the target’s spatial position, participants were now
instructed to decide if the target is present or absent. Participants were tested on a
2AFC procedure as before, and to respond with by pressing one of 2 pre-designated
keys on a standard keyboard.
In each trial, participants began by fixating on the central cross that was
followed by the presentation of the critical search array and a possible distractor in
one of 6 positions. Participants then scanned the search array for target presence and
made an appropriate keypress as soon as possible. As before, participants were
encouraged to make accurate and fast responses The test stimuli was briefly presented
(200ms) and was followed by a central fixation cross that stayed on until a keypress
response is made. Again, corrective feedback was provided.
The distractors subtended a circle of 2.375º radius and were presented in any
of six positions, evenly spaced along the circumference of a circle centred 5.3º about
the fixation cross. The schematic arrangement was the same as that used in
Experiment 1 (see Fig 2.2c). Letter strings were positioned 0.3º above and below the
fixation cross. Each unit in the search array subtended 0.12º x 0.12º and the
dimensions of the entire search array were 0.77º x 0.65º. The whole display fell within
a circular area of diameter 5.3°, approximately the size of the foveal region.
Results
The dependent variables were i) response accuracy and ii) reaction timings of
correct responses on the primary task.
A 2 (Serial order) x 2 (Target presence) x 3 (Distractor-type) ANOVA was
conducted on each set of dependent variable – accuracy and detection RT. The
analysis of accuracy scores revealed statistically significant main effects for the
independent variables of serial order (F(1,34)=4.58, p=0.04) and target-presence
(F(1,34)=57.4, p<0.001). Performance improved with practice, resulting in more
accurate responses in the latter three blocks of the experiment compared to the first
three. Also, under limited viewing conditions, participants performed better with
target-present rather than target-absent search arrays. Contrary to the hypothesis, there
was no main effect for the variable of distractor-type (F(2,68)=2.14, p=0.13).
Performance on the primary task was not substantially influenced by the condition of
distractor-type. In addition, there were no significant interactions, particularly
between serial-order and distractor-type (F(2,68)=1.03, p=0.36), that could have
indicated a differential influence of the orientation of distractor-type on varying levels
of search expertise.
A repeat analysis with reaction-time measures, revealed a similar pattern of
results. Statistically significant main effects were reflected for serial order
(F(1,34)=39.15, p<0.001) and target-presence (F(1,34)=98.62, p<0.001), but not for
distractor-type (F(2,68)=0.49, p=0.61). Again, this indicated faster responses on
target-present trials and improved performance with practice. Contrary to
experimental predictions, the orientation of face-distractors failed to influence
primary task performance. There was a significant interaction between the
independent variables of serial-order and target-presence (F(1,34)=4.618, p=0.39).
Close examination of the mean reaction times (see Appendix A: Experiment 2)
reveals that while response latencies were generally shorter for target-present than
target-absent trials, this difference was larger for the first three blocks of trials
compared to the later three blocks. This interaction demonstrates the effect that
practice might exert on search performance, the transition of serial-like to parallel-like
search efficiency.
The summary of mean accuracy scores and RTs of the main effects are shown
in Table 2.2. To summarise, responses made on the trials contained within the final 3
blocks were faster and more accurate, compared to the first 3. This was also true for
trials that contained the target within their search-arrays, as opposed to those that did
not. Both accuracy and RTs do not appear to vary across trials that are distinguished
by the condition of distractor-type that was predicted to impair primary task
performance.
Serial-order
Mean acc
(%)
Std error
Mean RT
(msec)
Std error
Target
Distractor-type
First
Second
Present
Absent
Absent Inverted face Upright face
65.0
1.7
68.2
1.9
77.3
1.6
55.9
2.6
68.2
1.6
66.4
1.7
65.3
2.1
827
30.0
730
21.8
688
18.9
869
30.8
776
23.2
776
24.8
784
25.0
Table. 2.1: Summary of mean accuracy and reaction times of responses and respective standard
errors of visual search task, across independent variables of serial-order, target-presence and
distractor type.
Following this, the data for distractor-type was collapsed across upright and
inverted faces distractor-types. This was to investigate if the presence of a distractor
had an influence on the primary task performance regardless of type.
A 2 (serial order 1 vs. 2) x 2 (target absent vs present) x 2 (distractor absent vs
present) repeated measures ANOVA was conducted for the measure of response
accuracy. This analysis showed significant main effects for serial-order (F(1,34)=6.41,
p=0.016), target presence (F(1,34)=59.87, p<0.001) and finally, distractor presence
(F(1,34)=4.9, p=0.034). No significant interactions were revealed.
The same analysis conducted for the measure of RTs only showed main
effects for serial-order (F(1,34)=43.36, p<0.001) and target presence (F(1,34)=97.97,
p<0.001). There was a significant interaction between serial-order and target-presence
(F(1,34)=4.31, p=0.046).
A summary of the mean accuracy scores and RTs is presented in table 2.2.
Altogether, primary task performance is faster and more accurate in the first 3 blocks
compared to the final 3. This is also true for when the target is present in the search
array, compared to when it is absent. In addition, the interaction of serial-order and
target-presence for RT reveals practise effects, that participants find it easier to cope
with target-absent trials with practise. The presence of a distractor, regardless of
whether it is an upright or inverted face image, significantly reduces response
accuracy but does not slow down performance on the primary task.
Serial-order
Mean acc (%)
std. error
Mean RT
(msec)
std. error
Target
Distractor-type
First
65.2
1.7
Second
68.9
1.9
Present
77.6
1.6
Absent
56.5
2.5
Absent
68.3
1.6
Present
65.8
1.8
824.4
27.2
731.5
21.8
685.9
18.9
870.1
30.5
776.1
23.2
779.8
24.4
Table 2.2: Summary table of mean response accuracy and RTs, and respective standard errors, for the
main independent variables of i)target-presence and ii)distractor-type.
Discussion:
The experimental design of Experiment 1 contained several confounds that
were addressed in this second experiment. Firstly, the responses made in this
experiment could only be accurately derived from conducting a visual search of the
letter array and not by inference. This eliminates the use of an internal strategy to
compensate for difficulties caused by the experimental limitations imposed in visual
search; such as, limited visual exposure or distractor presence. In addition, the whole
trial display is reduced in size, specifically to allow it to fall within the foveal region.
Results from experiment 1 failed to confirm the hypothesis that upright faces
involuntarily captures selective attention, hence severely limiting its successful
application to the primary task; that is, a purposeful visual search of a letter array. The
purpose of Experiment 2 was to resolve the aforesaid confounds and thus, clarify the
results of Experiment 1. The results obtained here are similar to those of Experiment 1.
Like Experiment 1, performance on the visual search task was not influenced
by the type of distractor used. Upright faces were not more effective than inverted
faces in distracting spatial attention away from the primary task. Nonetheless, it can
be seen from this experiment that both distractors can exert a detrimental effect on
visual search accuracy when they fall in the same foveal region. Thus, the noneffectiveness of distractors in the previous experiment might be attributed to the
greater distance between themselves and the search array.
The pattern of performance on the visual search task – independent of
distractor-type - is in accordance with our expectations. With letter arrays that do not
contain the target, responses are less accurate and slower compared to those that do
contain a target. The analogy of a “attentional spotlight” is useful in explaining this.
To solve the visual search task, each element of the array has to be scanned in a serial
fashion, especially when the target is amongst like elements {Duncan, 1989 #112}.
Search is expected to be faster for arrays that contain the target compared to those that
do not, because the only way to return an accurate response for the latter is by
assessing every element in the array. Searches on the former are terminated upon
target detection and there is no imperative to analyse the whole array in such instances.
It was originally thought that if upright faces captured spatial attention and hence,
skew the path of this attentional spotlight, performance on a serial search task would
be severely impaired on trials that had a simultaneous presentation of upright faces.
This hypothesis cannot be confirmed by the results of Experiment 2. Thus, the results
of this experiment confirm that a serial search is taking place.
In addition, the interaction between serial-order and target-presence further
suggests that practice might have improve search efficiency, such that serial search
performance in the first 3 blocks assumes the efficiency of parallel searches with
practice, in the last 3 blocks. Hence, our results strongly suggests that upright faces do
not capture attention used in serial searches any more than inverted face controls; and
that it is unlikely to capture attention used in parallel searches either. Therefore, the
face-detection effect cannot be explained by a visual-spatial attentional system that is
selective for upright faces.
For further confirmation of this finding, two pilot experiments were
subsequently conducted that allowed for slight modifications to the experimental
design of Experiment 2. Namely: Experiment 3 introduces a backward visual mask
that immediately follows the critical stimuli and Experiment 4 reduces the distance
between distractor and letter array.
The original investigations of the face-detection effect used a backward visual
mask to disrupt any visual-processing of the stimuli after it was removed from the
visual display {Purcell, 1986 #27;Purcell, 1988 #24;Rolls, 1994 #7}. The experiments
reported here do not do so. Experiment 3 tests the possibility of face-detection effect
as a construct of this particular paradigm.
Unlike the previous experiment, distractors in Experiment 2 are significantly
effective in impairing target search performance. Experiment 4 is conducted to assess
the measure of an upright-face’s ability to capture attention, as a function of its
proximity to the critical search array. In a similar task, Jenkins and Burton (2001)
demonstrated covert face-learning effects when the upright face occupied the same
visual space as the letter array. Attentional capture properties of an upright face could
be bounded by foveal proximity to the critical array and only exhibit such properties
within a certain radius of the relevant stimuli.
Experiment 3 and 4
Two experiments are reported here, with only slight modifications on the
design of Experiment 2. These two experiments served the purpose of replicating the
results obtained in Experiment 2 The slight modifications are to the display stimuli
and were introduced to increase the visual salience of the task-irrelevant distractor
stimuli. The condition of serial-order has been removed from further analyses because
the main focus is to replicate the finding that the orientation of face-distractors do not
influence serial search performance and by inference, do not capture spatial attention.
In Experiment 3, backward visual masking was introduced (cf. Purcell & Stewart,
1986;1988) and Experiment 4 brought the distractors even closer to the critical search
array.
Experiment 3 followed the same design as Experiment 2, except that a visual
mask follows the critical search array instead of the standard central fixation cross
used in both Experiment 1 and 2 (see Fig. 2). This visual-pattern of random noise
replaces the test display and stays on until a keypress response is made. This
rectangular pattern measures 12.77° x 13.3°.
Fig. 2: Time-line of experimental trial in Expt 3 with backward visual masking
In Experiment 4 the linear distance between distractors and fixation cross was
reduced from 5.3° to 3°. This is the only difference between Experiment 3 and 4. All
the experimental trials and test conditions were maintained.
Participants
There were 8 undergraduates with normal or corrected-to-normal vision for
each experiment. The age-range was 19-22 years. Participants were awarded course
credits for their participation.
Materials and Apparatus
All the materials and apparatus used for these 2 experiments were the same as
Experiment 2. The pattern of random noise used as a backward-visual mask was
created by applying an effects filter to a blank template, using Adobe Photoshop v.
6.5.
Procedures
Participants were provided with the same task instructions as Experiment 2.
The experimental script was identical with the exceptions of the aforementioned
differences.
Experiment 3: Backward-visual Masking
Results
A 2 (target-presence: present vs. absent) x 3 (distractor-type: absent vs.
inverted face vs. upright face) repeated-measures ANOVA was conducted on the
measures of response accuracy and RTs.
Again, the analysis for response accuracy reflected a significant main effect of
target-presence (F(1,7)=34.57, p=0.001) but not for distractor-type (F(2,14)=0.38,
p=0.691). The same was shown for reaction-time measures, with a significant main
effect of target-presence (F(1,7)=57.6, p<0.001), though not for distractor-type
(F(2,14)=0.44, p=0.66). Finally, there were no significant 2-way interactions to report.
Table 2.3 summarises the mean accuracy and RTs of the participants responses.
No distractor
Present
73.7
Mean acc (%)
3.6
std. errors
595.3
Mean RT (msec)
55.1
std. errors
Absent
44.5
4.5
812.8
96.5
Inverted
Upright
face-distractor
face-distractor
Present Absent Present Absent
73.4
44.3
74.5
45.8
4.7
6.4
4.2
5.5
643.4 735.8
613.4
831.9
75.6
106.6
66.5
133.8
Table 2.3: Summary table of mean response accuracy and RTs as well as corresponding
standard errors, for the main independent variables of i) target-presence and ii) distractortype.
Performance was faster and more accurate on search trials that contained a
target. Despite the introduced modifications, the orientation of face distractor-type
continued to effect no influence on primary task performance.
The only statistically significant interaction between target-presence and
distractor-type was for RTs (F(2,14)=4.39, p=0.03). This interaction is illustrated in
the line-graph 2.2. However, this interaction did not address the issues presented in
our hypothesis. Furthermore, this interaction was not replicated in Experiment 4 and
thus, was not be subject to further analysis.
target-present
850
target-absent
800
750
700
650
600
550
500
absent
inverted-face
upright-face
Graph 2.2: Comparison of mean RTs across distractor-types for search performance on arrays
with target-present and target-absent.
Experiment 4: Reduced distance between distractors and the critical array
Results:
A 2 (target-presence: present vs. absent) x 3 (distractor-type: absent vs.
inverted face vs. upright face) ANOVA was conducted on measures of response
accuracy and RTs.
On the measure of response accuracy, the independent variable of targetpresence was the only significant main effect (F(1,7)=31.57, p=0.001). The main
effect of distractor-type was not statistically significant (F(2,14)=0.059, p=0.943).
The same analysis of RTs also showed a significant main effect for target-presence
(F(1,7)=10.3, p=0.015) but not for distractor-type (F(2,14)=0.68, p=0.52). Table 2.4
presents a summary of the mean response accuracy and reaction-times.
No distractor
Present
72.9
Mean acc (%)
2.8
std. errors
Mean RT (msec) 487.5
39.7
std. errors
Absent
47.7
4.7
636.6
46.5
Inverted
Upright
face-distractor
face-distractor
Present Absent Present Absent
77.1
46.9
73.4
49.5
3.5
5.7
2.9
5.2
513.1 637.3
484.3
630.9
38.4
37.7
34.8
55.8
Fig. 2.4: Summary table of mean response accuracy and RTs, and respective standard errors,
for the main independent variables of i) target-presence and ii) distractor-type.
Therefore, primary task performance was faster and more accurate on search
arrays that contained a target. Search times and accuracy were independent of
distractor-type and did not vary across this condition.
Discussion:
The results obtained in these two reported pilots were similar to those already
obtained in Experiment 2. As with Experiment 2, there was a significant main effect
of target-presence. Despite the modifications introduced, it cannot be shown that the
orientation of the face-distractors was an influential factor on the attentional
mechanisms that underpinned serial visual search. Even by increasing the prominence
of these distractors, by introducing a backward visual mask and drawing the
distractors closer to the critical search array failed to induce the expected results. We
conclude that upright faces were not more effective than inverted faces in distracting
spatial attention away from a letter-search task.
Experiments 3 and 4 demonstrated that the effect of target-presence is
extremely robust, even with small samples of participants. This supports the
assumption that spatial attention is a limited resource as well as the analogy of an
attentional spotlight that scans each individual element of an array until the critical
target is found {Wolfe, 1998 #124}. Nonetheles, the presence of task-irrelevant
upright faces in a proximal area did not appear to selectively attract this attentional
“spotlight” away from the critical stimuli and hamper with this conscious search
process.
Overall Discussion
The main motivation behind the experiments reported here has been to
understand the face-detection effect in terms of early visual attention. In their dualroute model of face-processing, de Gelder and Rouw (2001) suggested that the facedetection effect could be better described as an epiphenomenon of a primitive ability
to automatically orient visual attention towards upright face-configurations. In this
chapter, visual attention has been defined teleologically; that is, in terms of how it
might be utilised in visual searches. Based on de Gelder and Rouw’s (2001) model,
the general prediction has been that - depending on the upright face-configuration’s
spatial proximity to the critical search target – the search process could be facilitated
or impeded by the upright face’s tendency to draw visual attention towards its
location in the visual scene.
Although past studies have generally been unsuccessful in using the visual
search paradigm to support this hypothesis, it has been argued earlier in the chapter
that this failure could have resulted from their use of the upright face-configuration
itself as the search target {Nothdurft, 1993 #136;Brown, 1997 #132;Kuehn, 1994
#135}. In doing so, their participants were inevitably recruiting face-identification
processes in their visual searches, that could have overshadowed the more primitive
and involuntary processes of face-detection. So far, the only experiment based on the
visual search paradigm that has provided support for this hypothesis, required
participants to search for a non-face target i.e. line-orientation {Gorea, 1990 #21}.
The aims in running the experiments here has been three-fold. Firstly, there is
a strong motivation to establish a relevance between the face-detection effect and
goal-directed behaviour. Given the claims about face-detection as an automatic and
involuntary process, these experiments investigate whether the presence of upright
faces could have an influence on goal-directed behaviour that has been designed to be
as irrelevant to faces as possible; namely, visual search for a particular letter of the
alphabet. Next, face-detection is theorised to be a primitive process that is distinctive
from face-identification. Thus, efforts have been taken to remove any conscious
involvement of face-identification processes in the experimental tasks and to see
whether the face-detection effect might continue to exhibit itself nonetheless, by
capturing visual attention. Finally, these experiments seek to define the face-detection
process in terms of visual attention, which is a better-understood and well-defined
cognitive process. This will allow better integration of the face-detection effect into
our current understanding of the visual system and face-specific processes.
Under certain conditions, the search process can be metaphorically likened to
the use of an attentional ‘spotlight’ that sweeps through a visual scene, inspecting
each item in turn, it reaches the spatial location of the critical target. Such a search can
be described as serial and self-terminating {Wolfe, 1994 #134;Wolfe, 1998 #124}. As
already mentioned, certain feature singletons e.g. colour are known to capture spatial
attention (e.g.{Theeuwes, 1991 #118} and it is on this basis that the same claim has
been put forward for upright face-configurations. If indeed, the upright faceconfiguration captures visual attention and if attention is like a ‘spotlight’ that covers
a finite spatial area, we ought have expected the accompanying presence of an upright
face-configuration to impair search performance on the critical display of letterstimuli. In any case, the presence of a distractor, whether upright or inverted, was
expected to detract from optimum visual search performance, simply by being an
extra non-target element in the visual scene. However, the general findings of these
experiments did not support this prediction. Whilst the presence of a task-irrelevant
distractor did reduce search accuracy on the primary task (Experiment 2,3,4), upright
face-configuration did not impair performance any more than an inverted faceconfiguration. Despite this, the pattern of visual search performance in all the reported
experiments here have been consistent with that found other visual-search studies.
In Experiment 1, our a priori expectations of task-difficulty was validated. In
replication of previous findings, target-distractor similarity was a consistent
determinant of search performance. Thus, the target-letter ‘K’ was more easily found
when it was embedded in an array of curved letter-distractors e.g. ‘C’, than when
amongst similar, angular letter-distractors e.g. ‘X' (cf. Neisser, 1964; Duncan &
Humphreys, 1989). Judging by the findings of experiments 2,3 and 4, the search
process for the target-letter ‘K’ when amongst visually-similar distractors is also
comparable to that of a serial and self-terminating , judging by the findings of
experiment 2,3 and 4 (cf. Wolfe, 1994; 1998).
Participants of experiments 2, 3 and 4 had a different task demand from those
of the first experiment. Instead of reporting the spatial location of the target, the only
requirement was to indicate their awareness of target-presence. Therefore,
experimental trials were either target-absent or target-present. The serial-search model
predicted worse performance for target-absent trials and this was a highly consistent
finding across experiment 2, 3 and 4 {Sternberg, 1966 #114;Wolfe, 1994 #134;Wolfe,
1998 #124}. This prediction is based on the supposition that if serial-search (as
opposed to parallel search) is used, one can be absolutely certain that a target is absent
only by scanning the entire display, item-by-item. In contrast, serial searches on
target-present trials self-terminate the instant the target is found, resulting in shorter
response latency; and more accurate performance for limited presentation times. In
other words, we could be certain that visual spatial attention was deployed in these
experiments; the same form of attention that is sometimes described as an attentional
“spotlight” {Treisman, 1988 #122}.
The pattern of these visual search results provide the assurance that the
experimental design has fulfilled its purpose of invoking the use of well-defined
visual search processes on the primary task. Experiment 1 allowed a comparison of
the influence of an upright face-configuration on both parallel and serial search
processes. With experiments 2, 3 and 4, an assessment was made on how the presence
of an upright face-configuration could guide serial search. The results obtained
indicate that the serial search process is not influenced by an upright faceconfiguration at all.
Contrary to our predictions, the presence of an upright face-distractor did not
reliably detract from normal visual search performance; measured in reaction-time
and accuracy. In fact, the presence of face-type distractors did not influence visual
search in Experiment 1 at all. This prompted several changes in the original
experimental design. The improvements suggested included:

changing the task to prevent alternative task-solving strategies e.g.
inference.

only testing the influence of upright faces on search performance for
arrays with target-similar elements.

drawing the face-type distractors closer to the visual search array so
that it fell in the foveal region.
The most significant change introduced has been that of the primary taskresponse. In Experiment 1, participants were required to indicate if the target letter
was to be found in the letter-string above or below the fixation cross. This
inadvertently introduced a confound in that participants were plausibly able to return
an accurate response by only monitoring one of the two letter-strings. Since all the
trials were target-present, participants were able to accurately infer that the target was
to be found in the complementary letter-string, if it was absent in the letter-string
actively monitored. By changing the task to a target-presence (yes/no) task in
Experiment 2, participants were now compelled to search the whole array in order to
come up with an accurate answer. The results in Experiment 1 also indicated strong
practise effects for searches for the critical letter-target amongst target-dissimilar
distractors. To reduce performance variability across trials, only the difficult search
trials were used in all subsequent experiments. Finally, the task-irrelevant face
distractors were drawn closer to the critical search array for Experiments 2,3 and 4, so
that it fell within the foveal region and difficult to ignore. The final recommendation
was made because it was noted that the task-irrelevant distractors failed to influence
search performances at all, regardless of their type. This suggested that the taskirrelevant distractors were completely ignored even if they were images of upright
faces.
In Expt 2, the presence of task-irrelevant distractors impaired detection
accuracy on the primary task. This implied that the modifications worked and that at
the very least, the presence of distractors was exerting a perceptual load on the
primary task, if not an attentional one. Nonetheless, the severity of impairment on the
primary visual-search performance was completely independent of the type of taskirrelevant distractors. Upright faces did not reduce search accuracy and response
latency on the letter-detection task any more than inverted faces did.
Finally, further modifications were systematically introduced in Experiment 3
and 4 to render the task-irrelevant distractors comparable to those used in the facedetection experiments {Purcell, 1986 #27;Purcell, 1988 #24} as well as increase their
visual salience. In Experiment 3, a backward visual mask was introduced after the
critical display was presented just like with the face-detection experimental paradigm
(cf. {Purcell, 1986 #27;Purcell, 1988 #24}. Results did not differ from those obtained
in Experiment 2. Subsequently, the upright/inverted face-distractors were drawn even
closer to the primary search display for Experiment 4. Again, this failed to bring about
findings that supported the hypothesis. Namely, that an upright face-configuration
leads to the involuntary capture of visual attention.
Given the current findings, it is appropriate to conclude that an upright faceconfiguration is not capable of attracting visual attention away from the spatial locus
of a critical search array. Still, it is important to note that in requiring participants to
conduct a visual search for the letter ‘K’, these experiments have explicitly
encouraged the establishment of an endogenous search strategy. Therefore, it could
still be that the face-detection effect is based on an automatic process that orients
exogenous visual attention towards upright face-configurations; but, only in the
complete absence of an endogenous attentional strategy. The purpose of using nonface stimuli in the critical search display was to completely remove any possible
involvement of the face-identification system in the task-solving process. The
findings of the studies reported here indicate that simply in having a well-specified
target to search for eliminates any influence that an upright face-configuration might
be able to exert on visual orienting.
In the dual-route model by de Gelder and Rouw (2001) (see Fig.1.3), the
general object-identification system can be tuned to a continuum of whole-based or
part-based perceptual encoding strategy; the face-identification system is considered
to be a specialised sub-system of this, and is preferentially biased to whole-based
encoding. By having a specific visual item to search for, one is pre-disposed to
employing the object-identification system, be it a face or letter of the alphabet. Farah
and colleagues {Farah, 1992 #191;Farah, 1998 #192} might even go so far as to argue
that just as face-identification is biased to whole-based perceptual encoding, letteridentification is ideally placed to benefit from part-based encoding. This is because
letters are the building blocks of whole words, but are of significant importance by
themselves. Essentially, the findings in this study concur with those conducted by
Nothdurft (1993) and Kuehn and Jolicuer (1994). That is, upright faces are not
capable of capturing visual attention when one is searching for a well-defined target;
whether the target should be an upright face or a known letter of the alphabet.
It is not entirely clear how this account might explain the results of Gorea and
Julesz (1991) whereby, they did find a benefit for target search performances if the
target was a constituent of the upright face pattern. In this particular experiment, the
target was merely any non-oblique line in an array of oblique lines of varying angular
difference from the targets. Thus, we could argue that since orientation is, arguably,
not a member of any well-defined object category (cf. faces, letters of the alphabet),
participants were able to circumvent the object-identification system in their searches.
In doing so, it allowed the the search process to be completely sensitized to the
influence of the face-detection system that effectively oriented observers to the
upright face-configuration in the display.
When introducing this chapter, abrupt visual onset was raised as a stimuli that
lays strong claim to the ability to capture attention {Yantis, 1990 #115;Jonides, 1988
#116;Yantis, 1984 #117}. However, there have been failed attempts in inducing
attentional capture, even with abrupt visual onsets. Reconsidering the supporting
literature on attentional capture, it is noted that a priori knowledge of target position
can allow the observer to demonstrate resilience to performance decrements induced
by attentional capture {Yantis, 1990 #115}. Behavioural evidence has revealed that
regions of attentional inhibition can be found, surrounding a region of attentional
facilitation {Mounts, 2000 #79;Mounts, 2000 #78}. It could have been that always
presenting the critical search array in the same region resulted in creating a
circumscribed region of attentional facilitation where the critical array was expected
to be, and a surrounding region of attentional inhibition that completely suppressed
the attentional-capture properties of an upright face could have exerted. However, this
is not to say that an upright face cannot capture attention under conditions when the
observer has no a priori expectations.
In the next chapter, a study will be presented in which participants were not
predisposed towards either the identification of any category of visual stimuli or any
particular spatial region in the visual field. Having established that upright faceconfigurations do not capture attention when an endogenous search strategy is in
place, the next chapter sets out to investigate whether upright faces can orient visual
attention towards itself in the complete absence of any endogenous search strategies.