B RA IN RE S EAR CH 1 08 0 (2 0 0 6) 2–8

a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m

w w w. e l s e v i e r. c o m / l o c a t e / b r a i n r e s

Research Report

Selection for action and selection for awareness:
Evidence from hemispatial neglect
Robert Rafal a,⁎, Robert Ward a , Shai Danziger b
a

School of Psychology, University of Wales, Bangor, Brigantia Building, Penrallt Road, Bangor, Gwynedd LL 57 2AS, UK
School of Management, Ben-Gurion University of the Negev, Israel

b

A R T I C LE I N FO

AB S T R A C T

Article history:

In bedside testing of patients with hemispatial neglect, we have found that extinction for

Accepted 11 January 2005

contralesional stimuli is less when the contralesional and ipsilesional items are different on

Available online 6 March 2006

the dimension to be reported relative to when they are the same. Importantly, a study that
investigated this observation found that similarity on visual features that are not necessary

Keywords:

for response does not impact the amount of extinction. These findings suggest that response

Attention

requirements may determine what stimuli will and what stimuli will not gain access to

Hemispatial neglect

awareness. In a related study, we found that extinction of contralesional stimuli was not

Consciousness

determined by perceptual similarity of the ipsilesional and contralesional items but by

Extinction

whether they shared the same semantics (e.g., ONE + 1 ) or the same response (e.g., ONE =

Response selection

WON). Here, we report a single case study in which extinction was determined by whether
the competing items shared the same response, regardless of whether they shared or differed
in their visual features or semantics. When asked to read the item in each field, there was
equivalent extinction in the conditions (ONE + ONE) and (ONE + 1) but less extinction in the
condition (ONE + TWO). By contrast, when asked to count the number of characters in each
field, there was more extinction in the condition (ONE + TWO) than (ONE + 1). When asked to
categorize each item as either a word or digit, the degree of extinction was determined both
by whether the items shared the same semantics and by whether they required the same
response. The results are consistent with a biased competition model in which competition
for selection is resolved flexibly depending on response requirements. Furthermore, the data
provide evidence that unattended stimuli are processed to the level of response.
© 2006 Elsevier B.V. All rights reserved.

1.

Introduction

The patient shown in Fig. 1 had suffered a stroke involving the
right parietal lobe 10 days before this video (with thanks to
Sharon Posner and Perpetua Productions) was made. When
the examiner's finger was wiggled in his left (i.e., contralesional) visual field he detected it—so he is not blind in that
field. However, when a finger was wiggled simultaneously in

his right visual field, this competing stimulus captured his
attention, and he failed to detect the contralesional stimulus
(Fig. 1). That is, the competing stimulus in the ipsilesional field
causes the contralesional stimulus to be extinguished from his
awareness.
Under conditions like this, in which a simple visual feature
must be detected and the competing stimuli are in opposite
visual fields (i.e., not within the same receptive fields of

⁎ Corresponding author.
E-mail address: r.rafal@bangor.ac.uk (R. Rafal).
0006-8993/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.brainres.2005.01.108

B RA IN RE S EA R CH 1 08 0 (2 0 0 6 ) 2 –8

3

function of attention is selection for action. We proceeded
from the perspective that attentional selection is needed, not
only because of limitations in perceptual processing capacity,
but also because we are limited in the number of actions that
we can initiate simultaneously. We provide evidence that
visual information is selected for, or excluded from, awareness and guidance of goal directed responses at the level of
response selection, immediately prior to action.

1.2.

Fig. 1 – Conventional bedside test demonstrating visual
extinction. (See text).

neurons in the early visual pathway), attenuation of processing for the unattended stimulus occurs very early in visual
pathway. Such evidence is often interpreted to support ‘early
selection’ accounts of attention.

1.1.

Early versus late selection

Does ‘early selection’ determine whether a visual signal will
gain access to consciousness? Vuilleumier and Rafal, (2000)
measured vocal reaction times in four patients in a simple visual
extinction task like that shown in Fig. 1. Stimuli were presented
in the left visual field, the right visual field, or both simultaneously; and the patients were asked to report ‘left’, ‘right’, or
‘both’. Exposure duration of the stimuli were adjusted such that,
for each patient, extinction occurred (i.e., ‘right’ reported) on
about 50% of bilateral trials (where the correct response should
have been ‘both’). Not surprisingly, on unilateral trials, vocal RTs
were longer for left than for right stimuli. On bilateral trials
where the patients reported ‘both’, vocal RT was, again
unsurprisingly, about the same as on unilateral left trials; i.e.,
response latency was determined by the stimulus requiring the
most time to process. The interesting finding in this experiment
was the vocal RTs on extinction trials, i.e., bilateral trials where
the patient reported ‘right’. If inattention on these trials had
caused gating from awareness at the earliest levels of the visual
pathway, such that no further processing of the left stimulus
had occurred before a response was made, then one might
expect RT s to be the same as for a ‘right’ response on unilateral
right trials (which were phenomenally the same as extinction
trials). This was not the case, however: RTs on these trials was
actually the same as for bilateral trials where both targets were
detected. It seems, then, that extinguished left stimuli took as
much time to process as detected left stimuli, before being gated
from awareness. In this sense, extinction might be construed as
something more like denial than as an attenuation of perceptual
processing due to early attentional selection.
The current communication reports a single case study of a
patient with visual extinction. It tests the hypothesis that one

Extinction and stimulus repetition

In bedside tests of extinction, we have found that when
patients are asked to identify visually presented objects, there
is less extinction when two competing objects are different,
than when are the same (Rafal, 1994). In Fig. 2, for example,
when the patient is shown two different objects (top), he
reports the identity of both of them. By contrast, when shown
two forks (bottom), he only reports the identity of the one on
the right. One way of thinking about this effect of stimulus
repetition is that when there are two different items present,
both are ‘singletons’; and this may be viewed as a special case
of singleton ‘popout’ that occurs in visual search tasks in
healthy observers (Folk et al., 1992). Note, however, that in Fig.
2, each fork is of a different color. Why didn't this difference
attract attention and prevent extinction? The reason, as it
turns out, is that, in this task, the patient was asked to report
the identity of the stimulus and not its color. Would the
patient have been more likely to detect both forks if the task
had been to report the colors of the stimuli?
An experimental investigation (Baylis et al., 1993) examined
this question in five patients (including the patient shown in
Fig. 1), and confirmed that only the similarity of response
relevant features determined whether the contralesional
object was extinguished. Since this paradigm is the same as
that used in the current case study, it will be described in some
detail here. All the patients had recently demonstrated visual
extinction at the bedside. On each trial, colored letters
appeared either unilaterally or bilaterally and the patients
were asked to report what they saw in each field. Exposure
duration of stimuli was adjusted such that extinction occurred
on approximately 1/3–1/2 of trials with bilateral items. Patients
reported either the identity or color of letters in different
blocks. Whereas there was a similar amount of extinction
whether the stimuli were the same or different color in the
letter naming block, in the color report blocks extinction was
greater for the same color stimuli (when compared to the same
stimuli in the letter report blocks or different colored stimuli in
the color report block). That is, extinction–failure to report the
presence of the contralesional item on bilateral presentations–
occurred more frequently when the bilateral stimuli were the
same on the attribute to be reported than when they were
different on this attribute, and more frequently than when the
stimuli were the same on the irrelevant attribute. These results
demonstrate not only that information about shape and color
of unattended objects is encoded, but also that awareness of
the presence of this information is contingent upon task goals.
Notice, that in that study, the patients' responses were
determined solely on the basis of simple visual features (color
or shape). Competition for awareness could therefore have
been due either to competition among objects sharing the

4

B RA IN RE S EAR CH 1 08 0 (2 0 0 6) 2 –8

Fig. 2 – Extinction and the effect of stimulus similarity. (See
text).

relevant visual feature, or between objects sharing the same
implications for response. The former could be implemented
on the basis of competitive feature selection early in the visual
pathway and/or response, while the latter implies biased
selection at a later level of semantics and/or response.
Another study, which leads to the current work, attempted
to determine the level at which selection takes place (Rafal et

al., 2002). In the first experiment of this study, responses were
based on the semantics of the items, while visual similarity
between concurrently presented items was manipulated. The
stimuli were the words “ONE” and “TWO” and the numerals
“1” and “2” that appeared, randomly, in the left field (one-sixth
of trials), the right field (one-sixth), or both fields (two-thirds of
trials). The patients were asked to read the item(s) and to
report what was present in each visual field by responding
“one” (indicating either the word “ONE” or the digit “1”), “two”
(the word “TWO” or the digit “2”), or “nothing”. Thus,
responses were contingent upon the semantic meaning of
the stimulus (i.e., what it denotes in terms of numerosity), and
independent of its visual features (i.e., whether a word or
digit). Patients verbally reported the location and identity of
the stimuli (e.g., ‘a one on the right and a two on the left'; ‘a
one on the right and nothing on the left’), frequently pointing
to the items in each field while naming them. On the critical
bilateral trials, the items could: be identical (e.g., ONE + ONE; or
2 + 2) and, thus, share the same visual features, meaning and
response; differ visually but share the same meaning and
require the same response (e.g., 1 + ONE; or TWO + 2); or differ
both with regard to visual features, meaning, and required
response. This last condition included both trials in which the
items came from the same category (e.g., 1 + 2) or from
different categories (e.g., 1 + TWO). If attention resolved
competition at the level of perceptual features, then perceptual similarity should modulate the degree of extinction. If
attention resolved competition at the level of the semantic
meaning required for response selection, then visual similarity between the competing items should not influence
extinction.
The results of this experiment demonstrated that, for all
three patients, competing items that were identical (e.g., ONE +
ONE, or 2 + 2) resulted in higher rates of extinction than did
competing items that had different meanings and required
different responses (e.g., ONE + TWO, 1 + 2, or ONE + 2).
However, there was just as much extinction in the condition
(ONE + 1) as in the condition (ONE + ONE), even though in the
case of the first pair, the competing items were not physically
similar. The data from that experiment for Patient JP,
investigated in the current report, are shown in Table 1 and
Fig. 4A. Thus, in this task that required semantic processing to
produce a correct response, there was no evidence that
differences in visual similarity influenced awareness. Rather,
selection in this experiment was dependent upon whether the
items shared the same meaning and response.

Table 1 – Percent of targets missed, and percent of those detected that were incorrectly identified for Patient JP in each
condition of Experiment 1 reported by Rafal et al., (2002) and Experiment 1 and 2 of the current report
Example

ONE 1
ONE 2
ONE ONE
ONE TWO
ONE –
– ONE

Read items (“One” or “Two”)
(Rafal et al., 2002)

Categorize (“Word” or “Digit”)
Experiment 1

Count (“One” or “Three”)
Experiment 2

Misses

Errors

Misses

Errors

Misses

Errors

57% (23/40)
41% (16/39)
60% (25/42)
37% (15/41)
37% (15/40)
3% (1/39)

0% (0/17)
0% (0/23)
12% (2/17)
0% (0/26)
4% (1/25)
3% (1/38)

42% (16/38)
26% (10/38)
69% (27/39)
47% (18/38)
0% (0/32)
5% (2/38)

0% (0/22)
11% (3/28)
17% (2/12)
0% (0/20)
6% (2/32)
8% (3/36)

20% (21/107)
27% (26/95)
35% (37/107)
35% (34/96)
1% (1/106)
0% (105/105)

26% (22/86)
29% (20/69)
37% (26/70)
34% (21/62)
12% (13/105)
11% (12/105)

B RA IN RE S EA R CH 1 08 0 (2 0 0 6 ) 2 –8

A second experiment reported by Rafal et al., (2002)
examined whether selection for action could determine
awareness independent of semantics. A fourth patient was
tested in an experiment that examined competition between
homophones that shared the same response (based on
phonology) but that were different in terms of both visual
features and semantics. The results showed that there was
just as much extinction when the competing items required
the same response (e.g., ONE + WON, or TOO + TWO), even
though they differed in semantics and visual features, as
when they were identical (e.g., ONE + ONE, WON + WON,
TOO + TOO, or TWO + TWO).
Here, we report two further experiments in a single case
study of Patient JP, who was one of the participants in the
study reported by Rafal et al., (2002). Our purpose was twofold.
First, we sought further evidence that the effect of stimulus
similarity on extinction was determined at the level of
response selection. Further, we wanted to specifically test
whether attentional selection could be biased at several levels
of processing under conditions where more than one level of
processing could be relevant to selecting a correct response. In
the first experiment, JP was shown the same stimuli she had
been tested with in the study by Rafal et al., (2002). However,
instead of being asked to read each item, she was asked to
classify it and report whether it was a “Word” or Digit”. In the
second experiment, she was asked only to report the number
of characters presented on each side (i.e., to respond ‘three’ if
there was a word, or ‘one’ if there was a digit).

1.3.

Case report

JP was 81 years old when she suffered a stroke in September
2000. She had previously been in good health and living
independently. The stroke involved most of the cortex in right
middle cerebral artery territory: parts of lateral frontal cortex
(including the frontal eye field), temporal lobe (including the
temporo-parietal junction), and inferior and superior parietal
lobes (Fig. 3). Very fortunately, it spared the primary motor
cortex and subcortical region of the internal capsule. Right
arm weakness recovered completely, but she was left with left
hemispatial neglect, mild sensory impairment in the left hand

Fig. 3 – T1-weighted MRI scan of patient JP.

5

and arm and dressing apraxia. She remained fully lucid and
socially engaged. She was able to return home to live alone,
with supervision by her family, and was independent in
everyday activities at the time of testing, 5 months after her
stroke.

2.

Experiment 1

2.1.

Procedure

The apparatus and stimuli were identical to those used by
Rafal et al., (2002). The task, instead of reading the items, was
to report for each visual field, whether there was a “word” a
“digit” or “nothing”. A laptop computer was used on which
black stimuli were briefly presented on a white background.
The stimuli were the words “ONE” and “TWO” and the
numerals “1” and “2” (Helvetica 36) that appeared, randomly,
in the left field (one-sixth of trials), the right field (one-sixth),
or both fields simultaneously (two-thirds of trials). The stimuli
were centered on the vertical meridian at approximately 3°
eccentricity from a + sign at fixation which offset when the
stimuli appeared.
Training on the task was initiated by presenting the stimuli
for several seconds and encouraging the patient to inspect
each visual field and to report what item, if any, was present in
each. When JP was reliably able to report both items on most
trials with bilateral stimuli, a brief break was given and
training resumed using a briefer exposure duration. This
process continued until she began to manifest extinction on
some trials. At this stage, she was encouraged to ‘guess’̦ if she
thought she detected an item but was unsure of its identify.
After training, JP was tested in 5 sessions comprising 240 trials.
Exposure durations across blocks ranged from 2250 to 2500 ms
in order to maintain a rate of extinction between 1/3 and 1/2 of
bilateral trials.

2.2.

Analysis

Because our main interest is in the conditions determining the
patient's detection of visual stimuli in contralesional space
under conditions of competition, the primary dependent
measure was the degree of extinction: i.e., errors in detecting
the presence of any stimulus (i.e., reporting “nothing”) when a
stimulus was, in fact, present. Misidentifications of an item
(e.g., reporting ‘word’ when the correct response was ‘digit’)
were scored as detections of the stimulus, because the
patients had in fact detected the stimulus. This approach
also obviated a potential confound that might arise from a
response bias in which a strategy was adopted to either use
the same response used for the ipsilesional items or,
conversely, to systematically select a different response
under conditions of uncertainty. That is if, for example, on
bilateral trials the patient detected a contralesional stimulus
but was unsure of its identity, the patient could have adopted
the strategy of reporting an identity different to that reported
for the ipsilesional item. Such a guessing strategy could, if
correct identification were the dependent measure, result in
an artificial inflation of ‘correct’ responses when the stimuli
were different than when they were the same. Since, in our

6

B RA IN RE S EAR CH 1 08 0 (2 0 0 6) 2–8

analyses, such incorrect identification of contralesional items
on bilateral trials were scored as non-extinction trials, a
strategy of this type would not bias the results. Similarly, if JP
adopted a strategy, when she did not detect a contralesional
item, of reporting that an item was present and guessing its
identity, this strategy would affect scores on bilateral trials
equally whether the items were same or different. Given task
instructions and training, it was likely that JP would, if a
contralesional item had been detected but its identity
uncertain, guess one of the two possible stimulus identities
rather than guess that ‘nothing’ had occurred. Our measure of
extinction, therefore, may reasonably be assumed to accurately reflect a failure to detect the presence of any item,
rather than the improbable alternative that patient systematically adopted a strategy to guess that ‘nothing’ had been
presented contralesionally when she had, in fact, detected an
item but were uncertain of its identity.

2.3.

Results

Table 1 reports the data for this experiment, and includes the
number and percentage of misses (i.e., reporting ‘nothing’
when a target had in fact been present), and the number and
percentage of errors in identification on trials where the
contralesional target was detected, in each condition. For
comparison purposes, Table 1 also shows the data for Patient
JP reported by Rafal et al., (2002) when her task was to read the
item on each side.
Fig. 4B shows that, in this experiment, there was an effect
of category [χ2(1) = 9.03, P = 0.0027]. That is, pairs of items
that required the same response showed more extinction
than those requiring different responses, regardless of
perceptual or semantic similarity. Note, however, that there
was also an independent effect of semantics [χ2(1) = 5.55, P <
0.02]. Thus, attentional selection gated access to awareness
at both the level of semantics and the level of response
selection. However, on these findings, we cannot conclude
that attentional selection at the level of semantics is
automatic and obligatory since, in this experiment, semantics could be used in determining the correct response, e.g.,
patients might read the stimuli as a step in determining their
category.

3.

Experiment 2

This experiment examined whether selection on the basis of
semantics is obligatory and occurs when semantic access is
not only unnecessary, but detrimental to response selection.
JP was asked to report the number of characters on each side:
the correct response for words was “3” since each word
contained three letters and for digits “1”. Exposure durations
ranged from 600 to 2750 ms over 16 blocks comprising 768
trials. The word ONE was replaced, in this experiment, by the
word SIX and the numeral 1 by 6. With this stimulus set, not
only is the semantic meaning of the items irrelevant to
response selection, but simple reading of the items would
compete with generation of a correct response, since neither
“Two” nor “Six” was ever a correct response. Thus, the task not
only made semantic analysis of the stimuli irrelevant, but

required that reading of the items be inhibited in order to
prevent them from competing with the correct response.
The results (Table 1 and Fig. 4C) showed that extinction
occurred more frequently [χ2(1) = 7.82, P = 0.0052] when the
competing items required the same response (73/203, 36%)
than when they required different responses (47/202, 23%).
There was no effect [χ2(1) = .910, P > 0.3] of semantic (word vs.
digit) similarity: same = 27%, different = 31%.

4.

Discussion

Our awareness of the visual environment is limited to only
those parts of a scene to which we attend. An emerging view is
that stimulus representations compete for processing (Duncan
et al., 1997), and for access to awareness. Neurobiological
evidence indicates that attention biases this competition via
selective enhancement and attenuation that can begin early in
the visual pathway (Heinze et al., 1994; Moran and Desimone,
1985), including primary visual cortex (Vuilleumier et al., 2001),
and as early as the lateral geniculate (O'Connor et al., 2002).
Nevertheless, there is abundant evidence that unattended
information is transmitted to later stages of processing, and
that the identity of unattended items is registered outside of
our awareness. Masked priming studies in normal individuals
(Marcel, 1980) have shown that this unattended information
activates semantic representations; and indeed that it also
activates motor responses relevant to the task at hand (Cohen
et al., 1995; Dehaene et al., 1998). Priming experiments in
neurological patients with the syndrome of hemispatial
neglect (Berti and Rizzolatti, 1992; McGlinchey-Berroth et al.,
1993) offer some of the most compelling evidence that
information outside of awareness undergoes processing at
least to a semantic level of representation. Thus, limitations in
awareness are not attributable merely to a limitation in
processing resources in sensory pathways.
The current observations converge with these studies of
semantic priming, and extend earlier work in our laboratory
showing an effect of stimulus repetition on visual extinction
(Baylis et al., 1993; Rafal et al., 2002). They extend the earlier
work by showing that, given the same stimuli, the degree of
extinction is determined by the task at hand; and, specifically,
that the degree of extinction can be modulated by whether the
competing stimuli require the same response. Thus, the
reading task, in which response was determined by semantics
(ONE + 1), resulted in more extinction than (ONE + TWO);
whereas the opposite pattern of performance was observed
when the task was counting the number of characters or
categorizing each stimulus as a word or digit.
These experiments examined the level of processing at
which attentional set influences detection. They should not be
interpreted to mean that attentional set is the only factor
determining the detectability of signals, or that sensory
factors, such as the size or brightness of stimuli do not have
their own independent affects on target detectability. Regardless of task, it would be expected, for example, that (ONE + 1)
would result in less extinction than (1 + ONE). We conducted
these analyses and, though the effect did not reach statistical
significance in any of the three experiments, when averaged
across all three experiments, the difference in extinction

B RA IN RE S EA R CH 1 08 0 (2 0 0 6 ) 2 –8

between Digit-Word versus Word-Digit pairs was statistically
reliable [χ2(1) = 5.3, P < 0.025]. This raises the question of
whether the advantage for different items might have derived
only from those trials in which the larger item was in the
contralesional field. This could have been the case in the
semantic (reading) task in which, for example, there was a
trend for more extinction between (1 + ONE) than (ONE + ONE).
However, this could not have accounted for the opposite result
with these same pairs in the categorizing or counting tasks. It

7

also cannot account for the observations that homophones
(ONE + WON) result in as much extinction as items that are
also semantically and physically identical (ONE + ONE) (Rafal
et al., 2002).
If attention gated access to consciousness entirely by
attenuating the processing of unattended stimuli early in the
visual pathway, then visual extinction could not be influenced
by the semantic meaning of the competing stimuli or, as
demonstrated here and in our previous work (Rafal et al.,
2002), by the task used to test for it. Extinction is, however,
determined both by what the competing stimuli are, and by
task demands. Whether or not a contralesional object is
extinguished is dependent upon: (1) whether the competing
objects are grouped on the basis of Gestalt principles
(Mattingley et al., 1997; Ward et al., 1994); (2) whether the
competing item is the same, or different, on the dimension to
be reported (Baylis et al., 1993; Rafal et al., 2002); (3) the task
used to probe for extinction—counting, identification, or
localization (Vuilleumier and Rafal, 2000); and (4) potential
relevance—that is, real objects suffer less extinction than
meaningless stimuli (Ward and Goodrich, 1996), and the
degree of extinction is modulated by the social relevance
and emotional valence of the stimuli (Vuilleumier, 2000;
Vuilleumier and Schwartz, 2001a,b,c).
Previous research in both normal individuals, and in
patients with hemispatial neglect, have shown that visual
signals that do not gain access to consciousness may ‘capture’
attention in the sense that they confer a subsequent
advantage in target detection at that location (Danziger et
al., 1998; McCormick, 1997), or summon a reflexive eye
movement toward it (Ladavas et al., 1997; Theeuwes et al.,
1998). The observations in JP reveal the mechanisms underlying attentional capture that not only bias processing at a
location, but that also lead to disengagement of attention from
its current focus so that signals at that location are explicitly
detected and accessible to processing by limited capacity
systems for voluntary action. They provide further evidence
that attentional capture is contingent upon attentional control
settings and suggest that attention can grant or deny access to
the gates of consciousness at a stage of processing at which

Fig. 4 – (A) Mean percent of bilateral trials with extinction, for
Patient JP, in a task in which she was asked to read the item
in each field and, therefore, in which response was
determined by the semantic meaning of the stimuli (from
Rafal et al., 2002). Above each column is an example of a
bilateral trial in which, from left to right, the competing items:
(a) share the same meaning and response, but are from
different categories and are visually dissimilar; (b) are
identical in meaning, category, visual features, and response;
(c) have different meaning and responses and are from
different categories; (d) are different in meaning, visual
features but are from the same category. (B) Mean percent of
bilateral trials with extinction, for Patient JP, in a task
(Experiment 1) in which she was asked to categorize the item
on each side as a ‘word’ or ‘digit’. (C) Percent of bilateral trials
with extinction for patient JP in a task (Experiment 2) in
which required response was to ‘count’ the number of
characters

8

B RA IN RE S EAR CH 1 08 0 (2 0 0 6) 2–8

the meaning of the visual stimulus is utilized in selecting it for
action.
Our results show that the exclusion from awareness
observed in visual extinction is based on similarity of the
attributes relevant for response. When patients can respond
based on a simple visual feature, for example, report the color
or shape (Baylis et al., 1993), or size (Experiment 2 here), then
unselected items may be excluded from awareness early in
processing. When the response must be determined by
semantics, as in reading each item, there is just as much
extinction between (ONE + 1) or (1 + ONE) as between (ONE +
ONE) or (1 + 1). Thus, in this situation where physical features
cannot be used to choose a response, attentional selection
does not appear to occur as early in visual processing (Rafal et
al., 2002). Indeed, since in a reading task there is as much
extinction between homophones (ONE + WON) as physically
and semantically related items (ONE + ONE), attentional
selection may be implemented even after semantic processing, at a level of representation immediately prior to response
(Rafal et al., 2002). Nevertheless, when semantic processing
was beneficial for choosing the correct response in the
categorization task (Experiment 1 here), attentional selection
occurred at both the level of semantics and the level of
response. Our findings are consistent with a framework in
which competition for selection and awareness occurs at
many levels of processing from sensation to action. Exclusion
from awareness can arise at any level (or levels) of processing
relevant for response selection.

Acknowledgment
This work was supported by the Wellcome Trust.

REFERENCES

Baylis, G., Driver, J., Rafal, R., 1993. Extinction and stimulus
repetition. J. Cogn. Neurosci. 5, 453–466.
Berti, A., Rizzolatti, G., 1992. Visual processing without awareness:
evidence from unilateral neglect. J. Cogn. Neurosci. 4, 345–351.
Cohen, A., Ivry, R., Rafal, R., Kohn, C., 1995. Response code
activation by stimuli in the neglected visual field.
Neuropsychology 9, 165–173.
Danziger, S., Kingstone, A., Rafal, R., 1998. Reflexive orienting to
signals in the neglected visual field. Psychol. Sci. 9, 119–123.
Dehaene, S., Naccache, L., Le Clec, H.G., Koechlin, E., Mueller, M.,
Dehaene-Lambertz, G., van de Moortele, P.F., Le Bihan, D.,
1998. Imaging unconscious semantic priming. Nature 395,
597–600.
Duncan, J., Humphreys, G., Ward, R., 1997. Competitive brain
activity in visual attention. Curr. Opin. Neurobiol. 7, 255–261.

Folk, C.L., Remington, R.W., Johnston, J.C., 1992. Involuntary covert
orienting is contingent on attentional control settings. J. Exp.
Psychol. Hum. Percept. Perform. 18, 1030–1044.
Heinze, H.J., Mangun, G.R., Burchert, W., Hinrichs, H., Scholz, M.,
Munte, T.F., Gos, A., Scherg, M., Johannes, S., Hundeshagen, H.,
et al., 1994. Combined spatial and temporal imaging of brain
activity during visual selective attention in humans. Nature
372, 543–546.
Ladavas, E., Zeloni, G., Zaccara, G., Gangemi, P., 1997. Eye
movements and orienting of attention in patients with visual
neglect. J. Cogn. Neurosci. 9, 67–74.
Marcel, A., 1980. Conscious and unconscious perception:
experiments on visual masking and word recognition. Cognit.
Psychol. 15, 197–237.
Mattingley, J.B., Davis, G., Driver, J., 1997. Preattentive filling-in of
visual surfaces in parietal extinction. Science 275, 671–673.
McCormick, P.A., 1997. Orienting attention without awareness.
J. Exp. Psychol. Hum. Percept. Perform. 23, 168–180.
McGlinchey-Berroth, R., Milberg, W.P., Verfaellie, M., Alexander,
M., Kilduff, P.T., 1993. Semantic processing in the neglected
visual field: evidence from a lexical decision task. Cognit.
Neuropsychol. 10, 79–108.
Moran, J., Desimone, R., 1985. Selective attention gates visual
processing in the extrastriate cortex. Science 229, 782–784.
O'Connor, D.H., Fukui, M.M., Pinsk, M.A., Kastner, S., 2002.
Attention modulates responses in the human lateral
geniculate nucleus. Nat. Neurosci. 5, 1203–1209.
Rafal, R.D., 1994. Neglect. Curr. Opin. Neurobiol. 4, 2312–2316.
Rafal, R., Danziger, S., Grossi, G., Machado, L., Ward, R., 2002. Visual
detection is gated by attending for action: evidence from
hemispatial neglect. Proc. Natl. Acad. Sci. U.S.A. 26, 26.
Theeuwes, J., Kramer, A.F., Hahn, S., Irwin, D.E., 1998. Our eyes do
not always go where we want them to go capture of the eyes
by new objects. Psychol. Sci. 9, 379–385.
Vuilleumier, P., 2000. Faces call for attention: evidence from
patients with visual extinction. Neuropsychologia 38 (5),
693–700.
Vuilleumier, P., Rafal, R., 2000. A systematic study of extinction:
between and within field deficits of attention in hemispatial
neglect. Brain 123, 1263–1279.
Vuilleumier, P., Sagiv, N., Hazeltine, E., Poldrack, R., Swick, D.,
Rafal, R., Gabrieli, J., 2001. Neural fate of seen and unseen faces
in visuospatial neglect: a combined event-related fMRI and ERP
study. Proc. Natl. Acad. Sci. 98, 3495–3500.
Vuilleumier, P., Schwartz, S., 2001a. Beware and be aware: capture
of spatial attention by fear-related stimuli in neglect.
NeuroReport 12, 1119–1122.
Vuilleumier, P., Schwartz, S., 2001b. Emotional facial expressions
capture attention. Neurology 56, 153–158.
Vuilleumier, P., Schwartz, S., 2001c. Modulation of visual
perception by eye gaze direction in patients with spatial
neglect and extinction. NeuroReport 12 (10), 2101–2104.
Ward, R., Goodrich, S., 1996. Differences between objects and
nonobjects in visual extinction: a competition for attention.
Psychol. Sci. 7, 177–180.
Ward, R., Goodrich, S., Driver, J., 1994. Grouping reduces visual
extinction: neuropsychological evidence for weight-linkage in
visual selection. Visual Cognition 1, 101–129.