Whorf hypothesis is supported in the right visual
field but not the left
Aubrey L. Gilbert*†‡
, Terry Regier§
, Paul Kay¶ʈ
**, and Richard B. Ivry*
*Helen Wills Neuroscience Institute and Departments of †Psychology and ʈLinguistics, University of California, Berkeley, CA 94720; §Department of
Psychology, University of Chicago, Chicago, IL 60637; and ¶International Computer Science Institute, 1947 Center Street, Berkeley, CA 94704
Contributed by Paul Kay, November 16, 2005
The question of whether language affects perception has been
debated largely on the basis of cross-language data, without
considering the functional organization of the brain. The nature of
this neural organization predicts that, if language affects perception,
it should do so more in the right visual field than in the left
visual field, an idea unexamined in the debate. Here, we find
support for this proposal in lateralized color discrimination tasks.
Reaction times to targets in the right visual field were faster when
the target and distractor colors had different names; in contrast,
reaction times to targets in the left visual field were not affected
by the names of the target and distractor colors. Moreover, this
pattern was disrupted when participants performed a secondary
task that engaged verbal working memory but not a task making
comparable demands on spatial working memory. It appears that
people view the right (but not the left) half of their visual world
through the lens of their native language, providing an unexpected
resolution to the language-and-thought debate.
categorical perception ͉ color ͉ hemispheric laterality ͉ linguistic relativity
Does the language we speak affect our perception of the
world (1, 2)?††
This classic question continues to provoke
controversy (3–5). Even in the thoroughly investigated domain of
color, some studies suggest that language may affect color
perception (6–8), although others suggest that it does not
(9–12). However, this question has not been addressed in terms
of the functional organization of the brain, which suggests a
possible resolution. Specifically, consideration of brain organization
predicts that language may affect perception in the right
half of the visual field but not in the left half, consistent with both
views at once.
Speakers of English judge colors that straddle the English
category boundary between ‘‘green’’ and ‘‘blue’’ to be less similar
than speakers of Tarahumara, a Uto-Aztecan language of Mexico
that, like the majority of the world’s languages, uses a single
word for these colors (6). Such categorical perception of color is
eliminated under conditions designed to interfere with verbal
processing (6–8). These results suggest that language may affect
perceptual discrimination through the spontaneous but unspoken
use of lexical codes. If so, these lexical codes are likely to be
more strongly represented in the left hemisphere (LH) of the
brain because a preferential involvement of the LH is observed
for almost all language tasks (13, 14), including those requiring
lexical access (15).
Given the contralateral nature of visual projections to the
cortex, information from the right visual field (RVF) would, at
least initially, have preferential access to, and be more susceptible
to modulation by, lexical representations in the LH. This
fact suggests that an effect of language on perceptual discrimination
would be seen primarily for stimuli in the RVF. Although
some earlier findings (16–19) are compatible with this proposal,
to our knowledge it has not been directly tested. We do so here.
We test three predictions based on the hypothesis that language
will disproportionately influence color discrimination in
the RVF as compared with the left visual field (LVF). First,
discrimination between colors from different lexical categories
(i.e., that have different names) should be faster when stimuli are
displayed in the RVF than when they are displayed in the LVF
because the lexical distinction will enhance the perceptual
difference. Second, discrimination between colors from the
same lexical category should be slower in the RVF than in the
LVF, because the assignment of the same name to two colors will
diminish the perceptual difference. Third, these laterality effects
should be disrupted when language resources are taxed by the
demands of an interference task.
Experiment 1, Results, and Discussion
We used a visual search task to test the predictions outlined
above. The search involved colors drawn from a set of four,
which we designated A, B, C, and D. These four colors form a
graded series from green to blue, the green–blue boundary
falling between B and C (Fig. 1a). In the visual search task, each
stimulus display consisted of a ring of colored squares surrounding
a central fixation marker. All of the squares were of the same
color except for one, the target (Fig. 1b). The target and
distractor colors were either from within the same lexical
category (e.g., different shades of blue) or from different lexical
categories (e.g., a green and a blue). On each trial, participants
were asked to indicate whether the target was in the left or right
half of the circle, by making speeded keyboard responses with the
corresponding hand. In this manner, we manipulated two variables:
the visual field of the target and the categorical relationship
between the target and distractor colors. This visual search
task was performed with or without a concurrent verbalinterference
task, which consisted of silently rehearsing an
eight-digit number throughout a block of visual search trials and
recalling it at the end of the block. There were four types of
target–distractor pairs: 1-step within-category (AB and CD),
1-step between-category (BC), 2-step between-category (AC
and BD), and 3-step between-category (AD).
Trials in which the participant pressed the wrong key or any
nondesignated key or in which the reaction time (RT) was Ͼ2 SD
from the participant’s mean were not included in the analysis of
the visual search data. About 8% of all trials were excluded by
the criteria just mentioned, 75% of these because of erroneous
responses. There was an approximately equal distribution of
excluded trials between the two visual fields, and error rates were
similar for within- and between-category trials. On Ϸ89% of the
blocks with verbal interference, participants entered the correct
eight-digit number; on the remaining blocks, errors were typiConflict
of interest statement: No conflicts declared.
Abbreviations: LH, left hemisphere; LVF, left visual field; RT, reaction time; RVF, right visual
field.
‡To whom correspondence may be addressed at: Helen Wills Neuroscience Institute and
Department of Psychology, 3210 Tolman Hall #1650, University of California, Berkeley,
CA 94720-1650. E-mail: mayfly@berkeley.edu.
**To whom correspondence may be addressed. E-mail: kay@icsi.berkeley.edu.
††We use the term ‘‘perception’’ throughout this paper in the broad sense in which it is
commonly employed in the language-and-thought literature; however, a more restricted
concept of perception is also examined in General Discussion.
© 2005 by The National Academy of Sciences of the USA
www.pnas.org͞cgi͞doi͞10.1073͞pnas.0509868103 PNAS ͉ January 10, 2006 ͉ vol. 103 ͉ no. 2 ͉ 489–494
NEUROSCIENCEPSYCHOLOGY
cally of the form that participants either failed to report one digit
or inverted the order of two digits.
Not surprisingly, there was a significant effect of step size:
targets were detected more quickly when they were less similar
to distractors [F(2,10) ϭ 12.36, P Ͻ 0.01]. The mean RTs for the
3-step, 2-step, and 1-step pairs, respectively, were 402 ms, 407 ms,
and 422 ms in the no-interference condition and 428 ms, 433
ms, and 444 ms in the interference condition (collapsed over
within- and between-category pairs). This observation confirms
that RT in the visual search task is sensitive to the discriminability
of the stimuli.
We focus on the 1-step pairs, because it is only at this step
size that both within- and between-category pairs exist. The
RT data from the no-interference condition were analyzed by
using a 2 (visual field: left vs. right) ϫ 2 (pair type: within- vs.
between-category) ANOVA (Fig. 1c). There was a reliable
main effect of pair type, with between-category pairs faster
than within-category pairs [F(1,10) ϭ 8.43, P Ͻ 0.02]. There was
no effect of visual field [F(1,10) ϭ 1.57, P ϭ 0.24]. The interaction
was significant [F(1,10) ϭ 16.11, P Ͻ 0.01], indicating faster
between-category discrimination in the RVF.
For the between-category pair, RTs were faster by 14 ms when
the target appeared in the RVF than when it appeared in the
LVF (t ϭ 2.23, P Ͻ 0.05). The opposite pattern approached
significance for the within-category pairs: RTs were faster by 8
ms when the target appeared in the LVF than when it appeared
in the RVF (t ϭ 1.84, P ϭ 0.09).
We also compared RTs for within-category vs. betweencategory
pairs within each visual field. For RVF targets, participants’
responses to the between-category pair were 24 ms faster
than to within-category pairs (t ϭ 2.78, P Ͻ 0.02). For LVF
targets, RTs were similar for within- and between-category pairs
(t ϭ 0.19, P ϭ 0.85). Thus, color names influenced RVF
responses but not LVF responses.
We performed a similar ANOVA on the data from the
verbal-interference condition (Fig. 1d). The effect of pair type
was significant [F(1,10) ϭ 6.75, P Ͻ 0.05], but the effect of visual
field was not [F(1,10) ϭ 2.89, P ϭ 0.12]. The interaction was again
significant [F(1,10) ϭ 26.27, P Ͻ 0.01], but the pattern of results
was quite different from that observed without interference.
Although we had expected that the RVF advantage for the
between-category pair would be attenuated when the search task
was performed concurrently with the verbal-interference task,
we found that RTs on these trials were actually 22 ms faster when
the target appeared in the LVF (t ϭ 2.79, P Ͻ 0.02). Similarly,
RTs for within-category pairs were now 17 ms faster when the
target appeared in the RVF (t ϭ 2.31, P Ͻ 0.05).
Considering the data by visual field for RVF targets, RTs were
now 26 ms slower for the between-category pair than for
within-category pairs (t ϭ 2.33, P Ͻ 0.05). The opposite pattern
approached significance for the LVF: RTs were faster by 13 ms
for the between-category pair than for the within-category pairs
(t ϭ 2.05, P ϭ 0.07). Thus, the addition of the verbal-interference
task reversed the visual field asymmetries observed when the
visual search task was performed alone.
Finally, we performed a three-way ANOVA (visual field ϫ
pair type ϫ secondary task: no interference vs. verbal interference)
directly to compare the no-interference and interference
conditions. As expected, there was a main effect of
secondary task, with slower RTs when the participants were
concurrently engaged in the verbal-interference task [F(1,10) ϭ
6.12, P Ͻ 0.05]. Most importantly, the three-way interaction
(secondary task: none vs. verbal ϫ visual field ϫ pair type) was
reliable [F(2,10) ϭ 16.37, P Ͻ 0.01]. Consistent with the results
reported above, the secondary task ϫ pair type interaction was
only reliable when targets appeared in the RVF [F(1,10) ϭ
27.21, P Ͻ 0.01] and not when they appeared in the LVF
[F(1,10) ϭ 2.74, P ϭ 0.13].
The results of experiment 1 are consistent with the hypothesis
that linguistic categories selectively influence color discrimination
in the RVF. Color names modulated color discrimination,
enhancing between-category distinctions and perhaps reducing
Fig. 1. Lexical categories influence perception in the RVF. (a) Print-rendered versions of the four colors used. (b) Sample display for the visual search task.
Participants were required to press one of two response keys, indicating the side containing the target color. (c) In the no-interference condition, RTs were faster
for the between-category pair and slower for the within-category pairs when targets appeared in the RVF compared with when they appeared in the LVF.
(d) Effects were reversed with verbal interference. *, P Ͻ 0.05, two-tailed t test, df ϭ 10; ns, nonsignificant. Values are mean Ϯ SEM.
490 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0509868103 Gilbert et al.
within-category distinctions, but only when the target appeared
in the RVF. These effects were disrupted by verbal interference.
Experiment 2, Results, and Discussion
We have assumed that the disruption caused by the interference
task was related to its specifically verbal processing requirements.
However, it is important to verify this point (7, 8). To this
end, we conducted a second experiment, similar to the first, but
employing both verbal and nonverbal concurrent interference
tasks. In the verbal-interference task, participants were shown a
word for a color (e.g., ‘‘red’’) before each visual search trial. They
had to remember the word during the visual search; upon seeing
the color term at the beginning of the next trial, they were to
respond if it matched the one they had seen previously (one-back
match). In the nonverbal-interference task, participants had to
remember a spatial grid of squares instead (Fig. 2a). We
expected that the influence of language on color discrimination
in the RVF would be disrupted by the verbal-interference task
but not by the nonverbal-interference task.
Trials were excluded from the RT analysis by using the same
criteria as in experiment 1. The mean number of excluded trials
was 6%, 8%, and 9% for no-interference, verbal-interference,
and nonverbal-interference conditions, respectively, with almost
all of these exclusions due to trials in which the wrong key was
pressed. Errors were approximately equal for RVF targets and
LVF targets.
In the verbal-interference task, participants responded correctly
on 91% of the trials for which the interference display
matched the previous one; this proportion was 89% in the
nonverbal-interference condition. In addition, there were a small
number of false alarms. In analyses of the visual search data, we
excluded trials occurring after any responses to these secondary
tasks.
As in experiment 1, RT was inversely related to step size [F(1,10) ϭ
10.07, P Ͻ 0.01]. The mean RTs for the 2-step and 1-step pairs,
respectively, were 415 ms and 436 ms in the no-interference
condition, 430 ms and 456 ms in the verbal-interference condition,
and 433 ms and 456 ms in the nonverbal-interference condition
(collapsed over within- and between-category pairs; the 3-step pair
AD was not tested in experiment 2).
Focusing again on the 1-step pairs, the results for the nointerference
and verbal-interference conditions replicated the
results for the corresponding conditions in experiment 1 (Fig. 2
b and c). Initial analyses with 2 (visual field) ϫ 2 (pair type)
ANOVAs for the no-interference and verbal-interference conditions
again demonstrated reliable main effects of pair type [no
interference, between-category faster: F(1,10) ϭ 9.95, P Ͻ 0.01;
verbal interference, within-category faster: F(1,10) ϭ 5.87, P Ͻ
0.05]. The effect of visual field approached significance for both
conditions [no interference: F(1,10) ϭ 3.38, P ϭ 0.10; verbal
interference: F(1,10) ϭ 3.72, P ϭ 0.08].
The visual field by pair type interaction was significant in both
of these conditions, [no interference: F(1,10) ϭ 12.60, P Ͻ 0.01;
verbal interference: F(1,10) ϭ 23.46, P Ͻ 0.01]. In the nointerference
condition, between-category RTs were faster by 16
ms when the target appeared in the RVF than when it appeared
in the LVF (t ϭ 2.46, P Ͻ 0.05). The opposite pattern was
observed for the within-category pairs: RTs were faster by 11 ms
when the target appeared in the LVF than when it appeared in
the RVF (t ϭ 2.27, P Ͻ 0.05). Similar to our analysis for
experiment 1, we also compared RTs within each visual field
separately. When targets appeared in the RVF, participants’
responses were faster by 30 ms for the between-category pair
than for the within-category pairs (t ϭ 2.93, P Ͻ 0.02). In
contrast, when targets appeared in the LVF, RTs for the
within-category pairs were not significantly different from those
for the between-category pair (t ϭ 0.05, P ϭ 0.96).
As in experiment 1, this pattern was reversed when the
participants were required to perform a concurrent verbalinterference
task. For the between-category pair, RTs were 20
ms faster when the target appeared in the LVF than in the RVF
(t ϭ 2.68, P Ͻ 0.05). The RTs for within-category pairs were 13
Fig. 2. Modulation of color-category effects in the RVF is specific to linguistic demands of the interference task. (a) Trial events. Within a block of trials, the
visual search task was interleaved with blank displays, displays containing a color word, or displays containing a spatial grid. (b and c) No-interference and
verbal-interference results replicate those obtained in the first experiment. (d) For the nonverbal-interference condition, performance followed a pattern similar
to that observed in the no-interference condition. *, P Ͻ 0.05, two-tailed t test, df ϭ 10; ns, nonsignificant.
Gilbert et al. PNAS ͉ January 10, 2006 ͉ vol. 103 ͉ no. 2 ͉ 491
NEUROSCIENCEPSYCHOLOGY
ms faster when the target appeared in the RVF than when it
appeared in the LVF (t ϭ 2.56, P Ͻ 0.05). For the within-visual
field analyses for RVF targets, RTs were 23 ms slower for the
between-category pair than for within-category pairs (t ϭ 2.75,
P Ͻ 0.05). The opposite pattern approached significance in the
LVF: RTs were 10 ms slower for within-category pairs than for
between-category pairs (t ϭ 1.88, P ϭ 0.09).
As expected, the results for the nonverbal-interference condition
were similar to those observed in the no-interference
condition but with generally slower RTs. (Fig. 2d). The main
effect of pair type was significant [F(1,10) ϭ 5.66, P Ͻ 0.05], but
the effect of visual field was not [F(1,10) ϭ 2.55, P ϭ 0.14], and
these factors interacted [F(1,10) ϭ 7.50, P Ͻ 0.05]. These results
did not resemble those of the verbal-interference condition. For
the between-category pair, RTs were 13 ms faster when targets
appeared in the RVF than when they appeared in the LVF (t ϭ
2.32, P Ͻ 0.05). For within-category pairs, RTs did not differ
significantly for targets appearing in the RVF and LVF (t ϭ 0.26,
P ϭ 0.80). In the within-visual-field analyses, RVF RTs were 22
ms faster for the between-category pair than for within-category
pairs (t ϭ 2.75, P Ͻ 0.05). This pattern did not reach significance
in the LVF (t ϭ 1.65, P ϭ 0.13).
To compare the verbal- and nonverbal-interference conditions
directly, we performed a three-way ANOVA. This analysis
revealed a significant three-way interaction between visual field,
pair type, and secondary task (by using just two factors, verbal
and nonverbal) [F(2,20) ϭ 12.53, P Ͻ 0.01]. As can be seen in Fig.
2, the two interference tasks had different effects for targets
presented in the RVF, with only the verbal-interference task
reversing the between-category advantage. In contrast, the two
interference tasks had similar effects for targets in the LVF.
Taken together, these results are consistent with the hypothesis
that language modulates color discrimination in the RVF by
means of lexical categories, whereas it has little, if any, effect on
the discrimination of color in the LVF. Moreover, this pattern is
disrupted by verbal, but not by nonverbal, interference.
Experiment 3, Results, and Discussion
The logic of experiments 1 and 2 rests on the assumption that
targets in the RVF will be primarily represented in the LH.
Although a large body of studies with neurologically intact
individuals supports this assumption (14), lateralized inputs can
readily become available to both hemispheres, primarily through
interhemispheric communication across the corpus callosum.
Callosotomy patients provide a less inferential method to bias
processing to a single hemisphere and thus can provide converging
evidence that language selectively modulates representations
in the LH͞RVF. We tested one such individual, JW, on
a version of the visual search task. JW underwent a two-stage
callosotomy operation in 1979–1980 for intractable epilepsy
(20). The linguistic competence of each hemisphere in this
individual has been extensively documented, and, despite slight
improvement in right hemisphere linguistic competence when
tested 15 years after surgery, JW has remained consistently left
hemisphere language dominant (21). JW has suffered two
strokes over the past two years, resulting in LH lesions primarily
in the parietal lobe. Chronic deficits related to the stroke include
a right-sided somatosensory deficit and reduced dexterity of the
right hand. There is no clinically observable neglect or aphasia,
and JW had little difficulty understanding and performing the
tasks.
Only 5% of the trials were excluded by using the criteria
described in experiment 1. Of these excluded trials, 45% involved
erroneous responses, 20% involved trials in which a nondesignated
key was pressed, and the remainder had RTs beyond 2 SD
from the mean. Slightly more exclusions due to a nondesignated
key being pressed occurred with the right hand (62%), likely
because of JW’s coordination problems related to previous
strokes, and more exclusions based on the RT criteria were
imposed for left-hand responses (68%) than right-hand responses.
The erroneous responses were distributed equally for
RVF and LVF targets. Similar to the results of the first two
experiments, JW was faster at responding on trials with 2-step
pairs than on those with 1-step pairs [F(1, 10) ϭ 10.2, P Ͻ 0.01].
The mean RTs for the 2- and 1-step pairs were 434 ms and 487
ms, respectively. The 3-step pair, AD, was not tested in this
experiment.
For the 1-step pairs (Fig. 3), we observed reliable main effects of
visual field [F(1,7) ϭ 22.46, P Ͻ 0.01] and pair type [F(1,7) ϭ 7.98, P Ͻ
0.05], as well as an interaction of these variables [F(1, 10) ϭ 17.91, P Ͻ
0.01]. JW was considerably slower in responding to targets in the
RVF, consistent with his right-hand apraxia and somatosensory
deficits related to the strokes. Nonetheless, RVF responses for the
between-category pair were still 67 ms faster than for the withincategory
pairs (t ϭ 14.3, P Ͻ 0.05). For targets in the LVF, RTs did
not differ for the between- and within-category pairs (t ϭ 1.01, P ϭ
0.50). Another way to consider the interaction is to observe that the
RVF RT cost was 204 ms for the within-category pairs but only 130
ms for the between-category pair. Thus, although the callosotomy
patient is much slower to respond to RVF targets in general, the
cost is significantly reduced when the target and distractor can be
lexically differentiated.
General Discussion
The results of these three experiments establish that, for the
visual search task used here, the Whorf hypothesis is supported
most strongly, even exclusively, in the RVF. Experiment 1
showed that color names influence color discrimination in the
RVF but not the LVF. Discrimination of colors with different
names is faster in the RVF than in the LVF. This experiment also
established that this laterality effect was attenuated and even
reversed when verbal working memory was taxed by a concurrent
interference task. Experiment 2 replicated these results, showing
that discrimination of colors in the same category was slower in
the RVF than in the LVF, and that a concurrent interference
task that taxed spatial working memory did not disrupt the
laterality effect. Converging evidence that the laterality effect
was related to hemispheric asymmetries was obtained in experiment
3; here, a split-brain patient was shown to have an RVF
advantage for between-category discrimination over withincategory
discrimination.
Whorf referred to the effects of language in terms of how it
is used to ‘‘cut nature up’’ (ref. 1, p. 213) without clearly
delineating the distinction between perception and other aspects
of cognition, such as decision making. When we use the term
‘‘perception,’’ we intend it in the same broad sense. We employ
the expressions ‘‘perceptual discrimination’’ and ‘‘discriminaFig.
3. Lexical categories influence perception in the RVF of a callosotomy
patient. *, P Ͻ 0.05, two-tailed t test, df ϭ 1; ns, nonsignificant.
492 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0509868103 Gilbert et al.
tion’’ specifically to describe our participants’ behavior on the
visual search task.
There are at least two reasons that color discrimination might
be affected by language in our findings, as in the literature more
generally. One possibility is that language affects perception
directly; by this view, language alters the nature of the early
visual response to colors, with this effect especially pronounced
in the left hemisphere. Another possibility is that language
facilitates postperceptual processes (22, 23). On this view, the
effect of language on the decision process is much stronger in the
left hemisphere, leading to the categorical effects observed in
performance. It is, of course, also possible that there is some
truth in both hypotheses. Our results do not discriminate among
these possibilities.‡‡
However, whether language affects perception,
postperceptual processes, or both, any influence of language
on perceptual discrimination clearly falls within the broad
formulation of Whorf’s hypothesis.
Previous studies addressing the possible influence of language
on perception have tended to look for a simple yes or no answer
to the question. Our findings suggest a more complex picture,
based on the functional organization of the brain. The LH
appears to sharpen visual distinctions between lexically defined
categories and to blur visual distinctions within these categories,
whereas the right hemisphere does so much less, if at all. To the
degree that these results can be generalized to everyday perception,
our representation of the visual world may be, at one
and the same time, filtered and not filtered through the categories
of language.
Materials and Methods
Experiment 1. Participants. Thirteen participants were recruited
from the University of California, Berkeley, community. The
participants (mean age ϭ 20 years) were right-handed, native
English speakers, and, based on self-reports, had normal color
vision.
Equipment and stimuli. The stimuli A, B, C, and D (Munsell 7.5G,
2.5BG, 7.5BG, and 2.5B) were used in a previous study (6); we
used computer-generated versions of them. The color calibration
wizard at www.easyrgb.com was used to standardize the
computer monitor. The red–green–blue (RGB) values were
identified by using a searchable database, also at www.easyrgb.
com. The brightness and saturation values were adjusted to make
them equal, based on the independent judgments of four observers.
The 8-bit RGB values were as follows: A ϭ 0, 171, and
129; B ϭ 0, 170, and 149; C ϭ 0, 170, and 170; and D ϭ 0, 149,
and 170.§§
The RGB values for the background were 178, 178,
and 178.
Procedure. First, each participant’s blue–green lexical boundary
was determined. In each trial of this preliminary task, a square
stimulus (one of the colors, A, B, C, or D) was presented
centrally on a neutral gray screen for 200 ms. Participants labeled
the stimulus either green or blue by pressing the ‘‘G’’ or ‘‘B’’ keys
on the computer keyboard. Each stimulus was presented 10 times
in a total of 40 randomized trials. The lexical boundary was
defined as the point of subjective equality (i.e., the estimated
value where green would be reported half of the time). Only the
11 participants who placed the blue–green boundary between
stimuli B and C were included in analysis of the visual search
data.
The visual search itself was then conducted. Each trial began
with the onset of a central fixation marker. After 1,000 ms, the
stimulus display appeared, consisting of a ring of 12 squares
surrounding the fixation marker. Responses were made by
pressing one of two horizontally aligned keys on the keyboard,
by using either the left (‘‘Q’’ key) or right (‘‘P’’ key) index finger.
The visual search display remained visible until a response was
made. After the response, no feedback was provided, and the
screen went blank for 250 ms before the fixation marker appeared
to indicate the start of the next trial. Although participants
were instructed to maintain fixation, we did not monitor
eye movements.
There were 12 target–distractor pairs, formed by using all
pairwise combinations of the four colors (six pairs) and having
each member of a pair serve once as target and once as
distractor. Because the target could occupy any of 12 positions,
there were 144 possible stimulus configurations. Each participant
completed six 144-trial blocks, with each stimulus configuration
appearing at random, once per block. For three of the
blocks, the visual search task was performed alone; for the other
three, the search task was performed with the concurrent
verbal-interference task. For the verbal-interference task, the
eight digits were randomly selected without replacement from
the set of single-digit numbers and were presented before the
start of the block of visual search trials. The numbers were
presented sequentially, with each digit appearing for 1,000 ms
followed by an interdigit interval of 200 ms. After the visual
search block, the participant recalled the list, typing in the
numbers on the keyboard number pad.
Participants completed three blocks of one condition (visual
search alone or concurrent with verbal interference) and then
three blocks of the other condition. The order of tasks was
counterbalanced across subjects. Before the first block of each
task, a short 10-trial practice block was administered.
Experiment 2. Participants. Thirteen participants were recruited
from the University of California, Berkeley, community, using
the same criteria as in experiment 1. The mean age of the
participants was again 20 years.
Procedure. The lexical boundary determination task was performed
as in experiment 1, and again, 11 of the 13 original
participants placed the green–blue boundary between stimuli B
and C. We again excluded the data of the other two participants
from the analysis of the visual search data.
The visual search task was similar to that of experiment 1, with
three changes. First, the relatively uninformative 3-step pair
(AD) was excluded. Second, for each of the remaining five pairs,
only one member, randomly selected, served as the target. These
two changes reduced the total number of stimulus displays to 60
(five pairs, with the target appearing in each of the 12 possible
locations), a number that allowed us to complete testing with the
search task alone as well as with the two interference tasks.
Third, the exposure time of the visual search displays was limited
to 200 ms, thus rendering eye movements, if present, ineffective
in identifying the target.
For the verbal-interference task, the interference display
consisted of a single color word drawn from the set: ‘‘beige,’’
‘‘black,’’ ‘‘brown,’’ ‘‘gray,’’ ‘‘orange,’’ ‘‘pink,’’ ‘‘purple,’’ ‘‘red,’’
‘‘violet,’’ ‘‘white,’’ and ‘‘yellow.’’ For the nonverbal-interference
task, the displays consisted of a 5-cm ϫ 5-cm grid in which 12 of
the 25 squares were black and 13 were white. A set of 15 displays
‡‡Behavioral methods may not prove sufficient to distinguish between these alternatives.
Physiological procedures might identify both the location and timing at which the two
hemispheres diverge in their response to stimuli such as those used in our color discrimination
task (24, 25).
§§After the experiment, we determined the coordinates of the color stimuli in the
CIEL*a*b* and CIEL*u*v* color spaces by using the conversion software at www.easyrgb.
com, with parameters set to illuminant ‘‘daylight’’ and observer ‘‘2 degrees.’’ For the
CIEL*a*b* space, the interstimulus distances were: AB, 12.3; BC,12.5; and CD, 15.7. For the
CIEL*u*v* space, the interstimulus distances were: AB, 16.3; BC, 17.5; and CD, 19.5. In both
spaces, the within-green (AB) distance is slightly less than the between-category (BC)
distance, and the within-blue differences is larger. In both cases, the average of the
within-distances (AB and CD) exceeds the between-distance (BC). Whereas we collapse
across the within-category pairs in the analyses reported in the main text, these differences
led us to perform secondary analyses in which we treated the data for the
within-green and within-blue conditions separately, rather than pooled. In no case did
the outcome of these secondary analyses deviate from that reported in the main text.
Gilbert et al. PNAS ͉ January 10, 2006 ͉ vol. 103 ͉ no. 2 ͉ 493
NEUROSCIENCEPSYCHOLOGY
was created. On no-interference blocks, a blank screen was
presented in lieu of an interference display.
The visual search and interference displays were interleaved
(Fig. 2a). A fixation marker was presented for 1,250 ms. It was
then replaced by a blank screen (no interference), a color name
(verbal interference), or one of the spatial grids (nonverbal
interference) for 1,250 ms. The fixation screen then reappeared
for another 1,250 ms, followed by the visual search display for
200 ms. The instructions emphasized that participants should
respond as quickly as possible to the visual search displays, by
using their left or right index finger to press the ‘‘Q’’ or ‘‘P’’ key
on the keyboard to indicate the side of the target. During
interference blocks, they were also required to press the space
bar with both thumbs whenever they detected that the secondary
task stimulus was the same as that shown in the previous display
(one-back match). No response was required when the secondary
task stimulus changed. Matches occurred in Ϸ10% of trials.
This low percentage was selected because RTs on the visual
search task are likely to be inflated when these responses are
made shortly after another overt response (26). For this reason,
we excluded trials occurring after overt secondary task responses
in our analysis of the visual search task data. Extensive pilot
testing was conducted to select displays in which accuracy on the
nonverbal one-back task was similar to that on the verbal
one-back task, when each was performed concurrently with the
visual search task.
Each participant completed four 60-trial blocks for each of the
three tasks: visual search under no-interference, verbalinterference,
and nonverbal-interference conditions. The order
of the three tasks was counterbalanced across subjects. A 10-trial
practice block preceded the first test block for each condition.
Experiment 3. Equipment. Testing was conducted at the patient’s
home, necessitating the use of a laptop computer. As in the
previous two experiments, the color calibration wizard at www.
easyrgb.com was used to standardize the colors presented on the
laptop monitor.
Procedure. JW performed the lexical boundary determination
task and visual search tasks (without an interference task). On
a first block of 20 trials for the lexical boundary task, the stimuli
appeared in the center of the screen for 200 ms. The patient
made vocal responses, and these responses were recorded by the
experimenter. After this portion of the experiment, a 40-trial
block was conducted in which the stimuli were presented for 200
ms in either the LVF or RVF. Again, vocal responses were
obtained. For all three presentation locations, JW placed the
lexical boundary between stimuli B and C. It is of interest is that
the patient was able to vocally label the colors when presented
in the LVF, indicating at least some access to lexical codes.
However, when the boundary task was repeated before a second
session the next day, the patient failed to name stimuli presented
in the LVF, always reporting, ‘‘I didn’t see anything’’ (although
this report almost always came immediately after the stimulus).
The change in performance is not uncommon with split-brain
patients and likely reflects different states of cortical function
due to arousal.
As in experiment 2, the 3-step pair was excluded, resulting in
five total pairs. Each member of the pair served both as target
and distractor. The 10 pair configurations were each presented
six times per block of 60 trials, with target location selected
randomly. JW completed eight such 60-trial blocks of the visual
search task. Responses were made with the index finger of the
left or right hand, to indicate that the target was on the
corresponding side of the display. Eye fixation was monitored by
the experimenter, who prompted the patient to maintain fixation
on the few trials in which movements were detected. Four blocks
were tested on each of 2 days and the first block of each session
was preceded by 10 practice trials.
We thank Lera Boroditsky for providing counsel; Paul Aparicio for aid
with programming pilot studies; Michael Posner, Christian Fiebach,
David Presti, and Michael Webster for their comments on the manuscript;
and Steven Shevell for suggesting the title. This work was
supported by National Science Foundation Grants 0418404 and 0418283
(to P.K. and T.R.) and National Institutes of Health Grant NS40813 (to
R.B.I.).
1. Carroll, J. B., ed. (1956) Language, Thought, and Reality: Selected Writings of
Benjamin Lee Whorf (MIT Press, Cambridge, MA).
2. Gumperz, J. & Levinson, S. C. (1996) Rethinking Linguistic Relativity (Cambridge
Univ. Press, Cambridge, U.K.).
3. Gordon, P. (2004) Science 306, 496–499.
4. Casasanto, D. (2005) Science 307, 1721–1722.
5. Gordon, P. (2005) Science 307, 1722.
6. Kay, P. & Kempton, W. (1984) Am. Anthropol. 86, 65–79.
7. Roberson, D. & Davidoff, J. (2000) Mem. Cognit. 28, 977–986.
8. Witthoft, N., Winawer, J., Wu, L., Frank, M., Wade, A. & Boroditsky, L. (2003)
in Proceedings of the 25th Annual Meeting of the Cognitive Science Society, eds.
Alterman, R. & Kirsh, D. (Lawrence Erlbaum, Mahwah, NJ), pp. 1247–1252.
9. Heider, E. R. (1972) J. Exp. Psychol. 93, 10–20.
10. Heider, E. R. & Olivier, D. C. (1972) Cognit. Psychol. 3, 337–354.
11. Lindsey, D. T. & Brown, A. M. (2002) Psychol. Sci. 13, 506–512.
12. Franklin, A., Clifford, A., Williamson, E. & Davies, I. (2005) J. Exp. Child
Psychol. 90, 114–141.
13. Corballis, M. C. (1991) The Lopsided Ape: Evolution of the Generative Mind
(Oxford Univ. Press, New York).
14. Hellige, J. B. (1993) Hemispheric Asymmetry: What’s Right and What’s Left
(Harvard Univ. Press, Cambridge, MA).
15. Damasio, H., Grabowski, T. J., Tranel, D., Hichwa, R. D. & Damasio, A. R.
(1996) Nature 380, 499–505.
16. Davidoff, J. (1976) Q. J. Exp. Psychol. 28, 387–394.
17. Hannay, H. J. (1979) Brain Lang. 8, 191–201.
18. Malone, D. R. & Hannay, H. J. (1978) Neuropsychologia 16, 51–59.
19. Kosslyn, S. M., Koenig, O., Barrett, A., Cave, C. B., Tang, J. &
Gabrieli, J. D. E. (1989) J. Exp. Psychol. Hum. Percept. Perform. 15, 723–
735.
20. Sidtis, J. J., Volpe, B. T., Wilson, D. H., Rayport, M. & Gazzaniga, M. S. (1981)
J. Neurosci. 1, 323–331.
21. Baynes, K., Wessinger, C. M., Fendrich, R. & Gazzaniga, M. S. (1995)
Neuropsychologia 33, 1225–1242.
22. MacLeod, C. M. (1991) Psychol. Bull. 109, 163–203.
23. Pilling, M., Wiggett, A., O¨ zgen, E. & Davies, I. R. L. (2003) Mem. Cognit. 31,
538–551.
24. Raz, A., Fan, J. & Posner, M. I. (2005) Proc. Natl. Acad. Sci. USA 102,
9978–9983.
25. Kosslyn, S. M., Thompson, W. L., Costantini-Ferrando, M. F., Alpert, N. M.
& Spiegel, D. (2000) Am. J. Psychiatry 157, 1279–1284.
26. Pashler, H. (1994) Psychol. Bull. 116, 220–244.
494 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0509868103 Gilbert et al.