Fermat's Library | Russian blues reveal effects of language on color descrimination annotated/explained version.

In general, European languages divide the spectrum like pictured be...

There are plenty of interesting differences between languages when ...

There has been a long standing debate surrounding the concept of li...

In the cited studies they use the same verbal interference method u...

Remember, *"within category”* means that the 2 bottom squares (the ...

#### ANOVA ANOVA stands for *Analysis of Variance*. It is a collec...

Russian blues reveal effects of language

on color discrimination

Jonathan Winawer*

†‡

, Nathan Witthoft*

‡

, Michael C. Frank*, Lisa Wu

, Alex R. Wade

, and Lera Boroditsky

‡

*Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139-4307;

Department of Neurology, David Geffen

School of Medicine, University of California, Los Angeles, CA 90095-1769;

Brain Imaging Center, Smith–Kettlewell Eye Research Institute, San Francisco, CA

94115; and

‡

Department of Psychology, Stanford University, Stanford, CA 94305

Communicated by Gordon H. Bower, Stanford University, Stanford, CA, March 7, 2007 (received for review September 22, 2006)

English and Russian color terms divide the color spectrum differ-

ently. Unlike English, Russian makes an obligatory distinction

between lighter blues (‘‘goluboy’’) and darker blues (‘‘siniy’’). We

investigated whether this linguistic difference leads to differences

in color discrimination. We tested English and Russian speakers in

a speeded color discrimination task using blue stimuli that spanned

the siniy/goluboy border. We found that Russian speakers were

faster to discriminate two colors when they fell into different

linguistic categories in Russian (one siniy and the other goluboy)

than when they were from the same linguistic category (both siniy

or both goluboy). Moreover, this category advantage was elimi-

nated by a verbal, but not a spatial, dual task. These effects were

stronger for difﬁcult discriminations (i.e., when the colors were

perceptually close) than for easy discriminations (i.e., when the

colors were further apart). English speakers tested on the identical

stimuli did not show a category advantage in any of the conditions.

These results demonstrate that (i) categories in language affect

performance on simple perceptual color tasks and (ii) the effect of

language is online (and can be disrupted by verbal interference).

categorization 兩 cross-linguistic 兩 Whorf

if ferent languages divide c olor space differently. For exam-

ple, the English ter m ‘‘blue’’ can be used to describe all of

the c olors in Fig. 1. Unlike English, Russian makes an obligatory

distinction between lighter blues (‘‘goluboy’’) and darker blues

(‘‘sin iy’’). Like other basic c olor words, ‘‘sin iy’’ and ‘‘goluboy’’

tend to be learned early by Russian children (1) and share many

of the usage and behavioral properties of other basic color words

(2). There is no single generic word for ‘‘blue’’ in Russian that

can be used to describe all of the colors in Fig. 1 (nor to

adequately translate the title of this work from English to

Russian). Does this difference between languages lead to dif-

ferences in how people discriminate colors?

The question of cross-linguistic differences in color perception

has a long and venerable history (e.g., refs. 3–14) and has been

a cornerstone issue in the debate on whether and how much

language shapes thinking (15). Previous studies have found

cross-linguistic differences in subjective color similarity judg-

ments and c olor confusabilit y in memory (4, 5, 10, 12, 16). For

example, if two c olors are called by the same name in a language,

speakers of that language will judge the two colors to be more

similar and will be more likely to confuse them in memory

c ompared with people whose language assigns different names

to the two c olors. These cross-linguistic differences develop early

in children, and their emergence has been shown to coincide with

the acquisition of color terms (17). Further, cross-linguistic

dif ferences in similarity judgments and recognition memory can

be disrupted by direct verbal interference (13, 18) or by indirectly

preventing subjects from using their normal naming strategies

(10), suggesting that linguistic representations are involved

online in these kinds of color judgments.

However, evidence from memory studies and subjective sim-

ilarit y ratings has left some critics unconvinced (19, 20). Pinker

(19) summarizes the critiques as follows:

Most of the experiments have tested banal ‘‘weak’’

versions of the Whorfian hypothesis, namely that words

can have some effect on memory or categorization. . . .

In a typical experiment, subjects have to commit paint

chips to memory and are tested with a multiple-choice

procedure. In some of these studies, the subjects show

slightly better memory for colors that have readily

available names in their language. . . . A ll [this] shows is

that subjects remembered the chips in two forms, a

non-verbal visual image and a verbal label, presumably

because two types of memory, each one fallible, are

better than one. In another type of experiment subjects

have to say which two of three c olor chips go together;

they often put the ones together that have the same

name in their language. Again, no surprise. I can imagine

the subjects thinking to themselves, ‘‘Now how on earth

does this guy expect me to pick two chips to put

together? He didn’t give me any hints, and they’re all

prett y similar. Well, I’d probably call these t wo ‘green’

and that one ‘blue,’ and that seems as good a reason to

put them together as any.’’

Because previous cross-linguistic comparisons have relied on

memory procedures or subjective judgments, the question of

whether language affects objective c olor discrimination perfor-

mance has remained. Studies testing only c olor memory leave

open the possibility that, when subjects make perceptual dis-

criminations among stimuli that can all be viewed at the same

time, language may have no influence. In studies measuring

subjective similarity, it is possible that any language-congr uent

bias results from a conscious, strategic decision on the part of the

subject (19). Thus, such methods leave open the question of

whether subjects’ normal ability to discriminate colors in an

objective procedure is altered by language.

Here we measure c olor discrimination perfor mance in two

language groups in a simple, objective, perceptual task. Subjects

were simultaneously shown three color squares arranged in a

triad (see Fig. 1) and were asked to say which of the bottom two

c olor squares was perceptually identical to the square on top.

This design combined the advantages of previous tasks in a

way that allowed us to test for the effects of language on color

perception in an objective task, with an implicit measure and

min imal memory demands.

First, the task was objective in that subjects were asked to

provide the correct answer to an unambiguous question, which

they did with high accurac y. This feature of the design addressed

the possibility that subjects rely only on linguistic representations

when faced with an ambiguous task that requires a subjective

Author contributions: J.W., N.W., M.C.F., A.R.W., and L.B. designed research; J.W., N.W.,

M.C.F., and L.W. performed research; J.W., N.W., L.W., and L.B. analyzed data; and J.W.

wrote the paper.

The authors declare no conﬂict of interest.

†

To whom correspondence should be addressed. E-mail: winawer@mit.edu.

7780–7785

兩

PNAS

兩

May 8, 2007

兩

vol. 104

兩

no. 19 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0701644104

judgment. If linguistic representations are only used to make

subjective judgments in ambiguous tasks, then effects of lan-

guage should not show up in an objective unambiguous task with

a clear correct answer.

Sec ond, all stimuli involved in a perceptual decision (in this

case, the three color squares) were present on the screen

simult aneously and remained in full view until the subjects

responded. This allowed subjects to make their decisions in the

presence of the perceptual stimulus and with minimal memory

demands.

Finally, we used the implicit measure of reaction time, a subtle

aspect of behavior that subjects do not generally modulate

ex plicitly. Although subjects may decide to bias their decisions in

choosing between two options in an ambiguous task, it is unlikely

that they explicitly decide to take a little longer in responding in

some trials than in others.

In summary, this design allowed us to test subjects’ discrim-

ination performance of a simple, objective perceptual task.

Further, by asking subjects to perform these perceptual discrim-

inations with and without verbal interference, we are able to ask

whether any cross-linguistic differences in color discrimination

depend on the online involvement of language in the course of

the task.

The questions asked here are as follows. Are there cross-

linguistic differences in color discrimination even for simple,

objective, perceptual discrimination tasks? If so, do these dif-

ferences depend on the online involvement of language? Prev i-

ous studies with English speakers have demonstrated that verbal

interference changes English speakers’ performance in speeded

c olor discrimination (21) and in visual searching (22, 23) across

the English blue/green boundary. If a c olor boundary is present

in one language but not another, will the two language groups

dif fer in their perceptual discrimination performance across that

boundary? Further, will verbal interference affect only the

performance of the language group that makes this linguistic

distinction?

Here we tested English and Russian speakers in an objective

c olor discrimination task across a color boundary that exists in

Russian but not in English. Twenty color stimuli spann ing the

Russian siniy/goluboy range were used (Fig. 1). Subjects were

shown colors arranged in a triad; their task was to indicate as

quickly and ac curately as possible which of the t wo bottom color

squares was identical to the top square. In some trials the

distracter square was from the same Russian category as the

match (i.e., both were goluboy or both were siniy); these were

called ‘‘within-category’’ trials. In other trials the match and the

distracter fell into different Russian categories (i.e., one was

goluboy and one was siniy); these were called ‘‘cross-category’’

trials. For English speakers, all of the colors in all trials fell into

the same basic linguistic category, namely, blue.

If linguistic ef fects on color discrimination are specific to the

categories encoded in a speaker’s language, then Russian

speakers should make faster cross-category discriminations

than within-category discriminations, a category advant age.

For English speakers, it should not matter whether c olors fall

into the same or dif ferent linguistic categories in Russian, so

they should not show any such dif ferences.

Further, if linguistic processes play an active, online role in

perceptual tasks (10), then a verbal dual task, but not a nonlin-

guistic dual task, should diminish the goluboy/sin iy category

advant age found in Russian speakers. To evaluate this possibil-

it y, subjects performed the color discrimination task under three

c onditions: a normal viewing, no-interference condition in which

there was no dual task; a verbal-interference condition, in which

subjects silently rehearsed digit strings while simultaneously

c ompleting the color discrimination trials; and a control, spatial-

interference condition, in which subjects maintained a spatial

pattern in memory while completing color discrimination trials.

The spatial-interference control c ondition was used to examine

whether any differences between the baseline condition and

verbal-interference c ondition were specific to language, or

whether they were due to nonspecific effects of any dual task.

Finally, we had previously found (unpublished work) that lin-

guistic categorie s are more likely to play a role in perceptual tasks

that are more difficult (e.g., ones that involve finer discriminations).

To explore this finding with a new set of color stimuli and speakers

of a different language, we included color discriminations that were

easier (in which the target and distracter color squares were

perceptually dissimilar, the ‘‘far-color comparisons’’) and discrim-

inations that were harder (in which the target and distracter color

squares were perceptually closer, the ‘‘near-color comparisons’’).

Results

Boundaries. To determine each subject’s linguistic color bound-

ary within the range of blues used in this work, we administered

a brief c olor classification task at the end of the experiment

(af ter the main color discrimination blocks). Subjects were asked

to classify each color square used in this work as either goluboy

or siniy (for Russian speakers) or light blue or dark blue (for

English speakers). All subjects classified the lightest stimulus

(stimulus 1 in Fig. 1) as goluboy or light blue and stimulus 20 as

sin iy or dark blue. Each subject’s boundary was identified as the

transition point in these classification responses. If the transition

fell between two stimuli or was ambiguous, the slower reaction

time was used to disambiguate the boundary, because colors

closest to boundaries tend to be categorized more slowly in

simple classification tasks (e.g., ref. 24). The locations of the

goluboy/sin iy boundary (Russian speakers) and the light blue/

dark blue boundary (English speakers) were nearly identical:

8.7 ⫾ 2.2 vs. 8.6 ⫾ 2.5, respectively (mean ⫾ SD).

Analysis. Each subject’s data were analyzed relative to their own

linguistic boundary. Trials were classified as within-category if

the test stimuli fell on the same side of that subject’s boundary

01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20

Fig. 1. The 20 blue colors used in this study are shown at the top of the ﬁgure.

An example triad of color squares used in this study is shown at the bottom of

the ﬁgure. Subjects were instructed to pick which one of the two bottom

squares matched the color of the top square.

Winawer et al. PNAS

兩

May 8, 2007

兩

vol. 104

兩

no. 19

兩

7781

PSYCHOLOGY

(e.g., both goluboy or both light blue) and were classified as

cross-category if they fell on opposite sides of the boundary or

if one of the two stimuli was the boundary. For each subject, the

n ine near-color and the nine far-color comparisons closest to

that subject’s boundary were included in the analysis. This

ensured that the set of stimuli used was centered relative to each

subject’s category boundary.

Additionally, trials were excluded if the response to the

interference stimulus was incorrect during the interference

blocks, if the response to the color task was incorrect, or if the

reaction time for the color discrimination was ⬎3 sec; 12% of

trials were so excluded. Subjects were excluded entirely from

analysis if the above criteria resulted in loss of 25% or more of

the trials, leading to the exclusion of three English and five

Russian speakers.

Summary of Results. Russian speakers showed a category advan-

t age when tested without interference, whereas English speakers

did not (Fig. 2). The category advantage found for Russian

speakers was disrupted by verbal, but not spatial, interference.

English speakers did not show a category advantage in any

c ondition. Further, effects of language were most pronounced

for more difficult discriminations (i.e., the near-color compari-

sons) (Fig. 3).

Detailed Analyses. Subjects were much faster at far-c olor discrim-

inations than near-color discriminations. This effect was re-

flected in separate 2 ⫻ 3 ⫻ 2 repeated-measures ANOVAs

calculated for each language group, with the factors of distance

(near color vs. far color), interference (none vs. spatial vs.

verbal), and category (bet ween vs. within). For each group, there

was a highly significant main effect of distance: in Russian

speakers [926 vs. 1,245 msec, near color vs. far color; F (1, 20) ⫽

267; P ⬍ 0.001] and English speakers [800 vs. 1,078 msec; F (1,

20) ⫽ 144.1; P ⬍ 0.001]. Additionally, a mixed-design ANOVA

using the above three factors as repeated measures and language

as a between-subjects factor showed that Russian speakers were

slower overall than English speakers [1,085 vs. 938 msec; F (1,

40) ⫽ 6.93; P ⫽ 0.012]. This difference might be due to the fact

that the Russian speakers we tested had less experience than the

English speakers in using computers or t aking part in experi-

ments. The mean and SE for each condition are included in

Table 1.

More critical to our hypothesis, the 2 ⫻ 3 ⫻ 2 ANOVA of the

Russian speakers showed that the performance in cross-category

vs. within-category trials was modulated by the interference

c ondition: there was a category advantage under both the no-

interference and the spatial-interference conditions, but not

under the verbal interference condition (Fig. 2) [category ⫻

interference interaction; F (2, 40) ⫽ 5.3; P ⫽ 0.009]. This effect

was completely due to the near-c olor condition (Fig. 3), sup-

ported by a significant three-way interaction among category,

interference, and distance [F (2, 40) ⫽ 3.3; P ⫽ 0.049]. This

finding, that language plays a role only in more difficult tasks

(near-c olor vs. far-color c omparisons, for example), is consistent

with our findings for the blue/green boundary in English in which

a category advantage was observed for harder, but not (unpub-

lished work) easier, discriminations. There were no other sig-

n ificant main effects or interactions in this analysis.

To explore in more det ail the interaction among distance,

category, and interference, several planned t tests were c on-

ducted under each of the separate conditions. In near-color

trials, Russian speakers showed a category advantage without

interference [1,164 vs. 1,288 msec; t (20) ⫽ 2.59; P ⫽ 0.0176] and

with spatial interference [1,162 vs. 1,270 msec; t (20) ⫽ 2.18; P ⫽

0.041] but a trend toward a category disadvantage with verbal

interference [1,325 vs. 1,260 msec; t (20) ⫽ 1.87; P ⫽ 0.076].

储

Moreover, the category advantage was significantly larger in no

interference blocks than in verbal interference blocks [124 vs.

⫺64 msec; t (20) ⫽ 2.93; P ⫽ 0.0082] and in spatial-interference

blocks than in verbal-interference blocks [109 vs. ⫺64 msec; t

(20) ⫽ 3.23; P ⫽ 0.004]. No difference in category advant age was

found between the spatial- and no-interference conditions [t

储

There is in fact a trend toward a reversal of the normal pattern under verbal interference

such that cross-category trials are performed more slowly than within-category trials.

Although this is not a signiﬁcant effect, it is consistent with the reversal in category

advantage under verbal interference reported in another work (23) and may suggest an

obligatory attempt to make a verbal distinction even when a dual task interferes with such

an attempt.

800

900

1000

1100

1200

none spatial verbal

interference condition

cross category

within category

800

900

1000

1100

1200

none spatial verbal

interference condition

cross category

within category

Fig. 2. Russian speakers’ (Left) and English speakers’ (Right) reaction times

(msec) shown for the no-interference, spatial-interference, and verbal-

interference conditions. Both near-color and far-color comparisons are in-

cluded in these graphs. Error bars represent one SE of the estimate of the

two-way interaction between category and interference condition.

-100

-50

100

150

200

no interference

spatial interference

verbal interference

near colors far colors

Russian speakers English speakers

near colors far colors

Fig. 3. Category advantage is plotted for Russian speakers (Left) and English

speakers (Right) as a function of comparison distance (near color vs. far color)

and interference condition (none, spatial, and verbal). Category advantage is

calculated as the difference between the average reaction time for within-

category trials and that for cross-category trials (msec). Error bars represent

one SE of the estimate of the three-way interaction among category, inter-

ference condition, and color distance.

7782

兩

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0701644104 Winawer et al.

(20) ⫽ 0.24; P ⫽ 0.81] nor bet ween any conditions in the far color

trials (P ⱖ 0.78).

Unlike Russian speakers, English speakers did not show any

category advant age [F (1, 20) ⫽ 0.150; P ⫽ 0.703] nor any

category ⫻ interference interaction [F (2, 40) ⫽ 0.422; P ⫽ 0.659]

(Fig. 2), as revealed by the same 2 ⫻ 3 ⫻ 2 ANOVA (category ⫻

interference ⫻ distance) of the English speakers’ data. The only

sign ificant effect in this analysis was a main effect of interfer-

ence, such that English speakers were fastest with no interfer-

ence and slowest with verbal interference [1,113, 1,156, and 1,216

msec for no interference, spatial interference, and verbal inter-

ference, respectively; F (2, 40) ⫽ 5.170; P ⫽ 0.010].

The results of English speakers differed significantly from

those of Russian speakers. In near-c olor trials, the difference in

the category advant age between no interference and verbal

interference was significantly g reater for Russian than English

speakers [189 vs. 15 msec, respectively; t (40) ⫽ 2.17; P ⫽ 0.036].

Likewise, the difference in category advantage between spatial

interference and verbal interference was significantly greater for

Russian speakers than English speakers [173 vs. 14 msec, re-

spectively; t (40) ⫽ 2.142; P ⫽ 0.038]. No differences were

observed for similar comparisons on far color trials (P ⱖ 0.6 for

both comparisons).

Because the performance of Russian speakers on average was

slower than that of English speakers, we considered the possi-

bilit y that the interesting difference between the two language

groups was not due to native language but to overall speed. If

linguistic effects on discrimination were only observed in harder

(or slower) tasks, it is possible that English speakers automati-

cally verbally c oded the light blue/dark blue distinction but were

too quick overall for the linguistic system to be able to influence

the decision process. To test this possibility, we conducted a

un ivariate ANOVA, using language (Russian vs. English) as a

fixed factor and mean reaction time as a covariate. The depen-

dent variable was a composite measure of the linguistic effect of

interest, the categor y advant age under the nonverbal-

interference conditions (the mean of the spatial- and the no-

interference conditions) minus the category advantage under

the verbal-interference condition. For the near-color trials only,

the language group was a significant predictor of the linguistic

ef fect of interest [F (1, 39) ⫽ 4.181; P ⬍ 0.048]. Mean reaction

time was not a significant covariate [F (1, 39) ⫽ 0.349; P ⫽ 0.558].

This analysis confirms that differences in overall speed between

the two language groups were not responsible for the cross-

linguistic differences of interest between the two language

groups.

Accuracy. Because the stimuli were present on the screen until

subjects responded, accuracy was high (96.5 ⫾ 2.1% and 95.7 ⫾

3.2% for English and Russian speakers, respectively). Further

analyses of the accuracy data by language, interference type, and

ef fects of category confir med that the differences of interest

found in reaction times could not be attributed to speed/accuracy

tradeof fs. There was one unex pected result in the accuracy data,

however: For near colors, Russian speakers were more accurate

on within-category, compared with cross-category, trials under

the no-interference condition (93% vs. 87%, or a ⫺6% category

advant age, that is, a category disadvantage), but not under other

interference conditions and not in far-color trials, leading to a

three-way interaction among category, interference, and dis-

t ance [F (2, 40) ⫽ 4.106, P ⫽ 0.024]. To test whether the pattern

of results found in reaction time resulted from a speed/accuracy

tradeof f, we c onducted two further analyses of the near-color

trials. Both analyses suggested that a speed/ac curacy tradeof f

c ould not explain our results. First, the category advantage in

ac curacy showed little difference between the spatial and verbal

interference blocks, and it in fact differed more for the English

speakers (⫺2.4% vs. 1.8%, spatial vs. verbal interference) than

for the Russian speakers (⫺1.9% vs. 0.5%). Second, there was

a significant partial c orrelation between language group (En-

glish vs. Russian, coded as 0 or 1) and a composite measure of

the reaction time effect (see Detailed Analyses above) when

c ontrolling for accuracy (using the same composite measure)

[Pearson’s r (39) ⫽ 0.365; P ⫽ 0.019]. The converse was not true:

there was not a c orrelation between language group and accu-

rac y when controlling for reaction time [r (39) ⫽ 0.096; P ⫽

0.549].

Discussion

We found that Russian speakers were faster to discriminate two

c olors if they fell into different linguistic categories in Russian

(one siniy and the other goluboy) than if the two colors were

f rom the same category (both sin iy or both goluboy). This

category advantage was eliminated by a verbal, but not a spatial,

dual t ask. Further, effects of language were most pronounced on

more difficult, finer discriminations. English speakers tested on

the identical stimuli did not show a category advantage under any

c ondition. These results demonstrate that categories in language

can affect perfor mance of basic perceptual color discrimination

t asks. Further, they show that the effect of language is online,

because it is disrupted by verbal interference. Finally, they show

that color discrimination performance differs across language

groups as a function of what perceptual distinctions are habit-

ually made in a particular language.

The case of the Russian blues suggests that habitual or

obligatory categorical distinctions made in one’s language result

in language-specific categorical distortions in objective percep-

tual t asks.** English speakers, of course, also can subdivide blue

stimuli into light and dark. In fact, English speakers as a g roup

drew nearly the same boundary as did the Russian speakers in

our work. The critical difference in this case is not that English

speakers cannot distinguish between light and dark blues, but

rather that Russian speakers cannot avoid distinguishing them:

they must do so to speak Russian in a c onventional manner. This

c ommunicative requirement appears to cause Russian speakers

to habitually make use of this distinction even when performing

**This may apply to some, but not necessarily all, perceptual tasks. Evidence from other

studies with similar designs suggests that perceptual discriminations that are more

difﬁcult (unpublished work) and ones that are carried out in the right visual ﬁeld (and

therefore more strongly in the left hemisphere of the brain, typically associated with

language) (23) are more likely to be affected by linguistic processes.

Table 1. Mean reaction times in msec (and SEM) for all conditions

Russian speakers English speakers

Near-color Far-color Near-color Far-color

Interference Between Within Between Within Between Within Between Within

None 1,164 (66) 1,288 (77) 998 (55) 999 (55) 900 (51) 914 (52) 758 (36) 735 (32)

Spatial 1,162 (58) 1,270 (56) 1,096 (64) 1,096 (53) 911 (41) 922 (46) 819 (37) 835 (43)

Verbal 1,325 (55) 1,260 (50) 1,146 (62) 1,132 (50) 952 (41) 955 (46) 831 (41) 821 (35)

Winawer et al. PNAS

兩

May 8, 2007

兩

vol. 104

兩

no. 19

兩

7783

PSYCHOLOGY

a perceptual task that does not require language. The fact that

Russian speakers show a category advantage across this color

boundary (both under normal viewing conditions w ithout inter-

ference and despite spatial interference) suggests that language-

specific categorical represent ations are normally brought online

in perceptual decisions.

These results also help to clarify the mechanisms through

which linguistic categories can influence perceptual perfor-

mance. It appears that the influence of linguistic categories on

c olor judgments is not limited to tasks that involve remembering

c olors across a delay. In our task, subjects showed language-

c onsistent distortions in perceptual performance even though all

c olors were in plain view at the time of the perceptual decision.

Further, language-consistent distortions in color judgments were

not limited to ambiguous or subjective judgments where subjects

may explicitly adopt a language-consistent strateg y as a guess at

what the experimenter wants them to do (19). In our task,

subjects showed language-c onsistent distortions in perceptual

performance while making objective judgments in an unambig-

uous perceptual discrimination t ask with a clear, correct answer.

Results from the verbal interference man ipulation provide

further hints about the mechanism through which language

shapes perceptual performance in these tasks. One way that

language-specific distortions in perceptual performance could

arise would be if low-level visual processors tuned to some

particular discriminations showed long-term improvements in

precision, whereas processors tuned to other discriminations

bec ome less precise or remain unchanged (25). Very specific

improvements in perceptual perfor mance are widely observed in

perceptual learning literature and are often thought to reflect

changes in the synaptic connections in early sensory processing

areas (26). Our present results do not offer support for this

possibilit y because a simple task manipulation, ask ing subjects to

remember digit series, eliminated the language-specific distor-

tions in discrimination. If the language-specific distortions in

perceptual discrimination had been a product of a permanent

change in perceptual processors, temporarily disabling access to

linguistic representations with verbal interference should not

have changed the pattern in perceptual perfor mance.

Instead, our results suggest that language-specific distortions in

perceptual performance arise as a function of the interaction of

lower-level perceptual processing and higher-level knowledge sys-

tems (e.g., language) online, in the process of arriving at perceptual

decisions. The exact nature of this interaction cannot be determined

from these data. It could be that information from linguistic systems

directly influences the processing in primary perceptual areas

through feedback connections, or it could be that a later decision

mechanism combines inputs from these two processing streams. In

either case, it appears that language-specific categorical represen-

tations play an online role in simple perceptual tasks that one would

tend to think of as being primarily sensory. Language-specific

representations seem to be brought online spontaneously during

even rather simple perceptual discriminations. The result is that

speakers of different languages show different patterns in percep-

tual discrimination performance when tested under normal viewing

conditions. When normal acce ss to language-specific representa-

tions is disrupted (as under the verbal-interference condition),

language-specific distortions in discrimination performance also

disappear.

These c onclusions are also consistent with three other findings

using similar methodologies. In one study, a verbal dual task was

shown to selectively interfere with blue/green discriminations

among English speakers using the same triad presentations used

here (21). In two studies a v isual field manipulation was used to

test the hypothesis that language effects are more pronounced in

the right visual hemifield (and hence the left, presumably

language-dominant, hemisphere) (22, 23). These studies (22, 23)

found that visual search time was affected more strongly by a

dual verbal task for cross-category searches in the right than the

lef t visual hemifield. In all four studies (the present work and

refs. 21–23), a category advantage was observed in simple

perceptual tasks and the category advantage was selectively

eliminated or reduced by verbal, but not spatial, interference.

Parallel findings using t wo very different manipulations, a

cross-linguistic comparison and a bet ween-hemispheres compar-

ison, converge to make a strong case that language-specific

processes can affect simple, implicit, perceptual decisions.

The Whorfian question is often interpreted as a question of

whether language af fects nonlinguistic processes. Putting the

question in this way presupposes that linguistic and nonlinguistic

processes are highly dissociated in normal human cognition, such

that many tasks are accomplished without the involvement of

language. A different approach to the Whorfian question would

be to ask the extent to which linguistic processes are normally

involved when people engage in all kinds of seemingly nonlin-

guistic tasks (e.g., simple perceptual discriminations that can be

ac complished in the absence of language). Our results suggest

that linguistic representations normally meddle in even surpris-

ingly simple objective perceptual decisions.

Methods

Participants. Twenty-six native Russian speakers (28.9 ⫾ 10.2

years old, mean ⫾ SEM) and 24 native English speakers (26.3 ⫾

9.2 years old) were recruited from the Boston area and tested at

the Massachusetts Institute of Technology (MIT) (Cambridge,

M A). The age of English acquisition for Russian speakers ranged

f rom 7 to 21 years. Participants gave written consent and were

paid for their time. The experimental protocol was approved by

the MIT Human Subjects Committee.

Materials and Design. Each subject completed one block of 136

c olor discrimination trials without any secondary task (‘‘no

interference’’), one block while performing a secondary verbal-

interference task, and one block while performing a control,

spatial-interference task. The order of the blocks was varied

randomly across subjects. After completing the color discrimi-

nation trials, subjects were tested in a separate c olor-naming t ask

to determine their individual linguistic borders. Subjects were

shown the 20 stimuli (twice each) in random order and asked to

classif y each color with a key press, either siniy vs. goluboy (for

Russian speakers) or dark blue vs. light blue (for English

speakers).

Subjects were instructed to make all judgments as quickly and

ac curately as possible. All subjects received the same instruc-

tions in English. Testing took place in a quiet, darkened room.

Color Stimuli. Twenty computer-simulated color chips were cre-

ated for this study, ranging from goluboy or light blue to sin iy or

dark blue (Fig. 1). The Commission Internationale de l’Eclairage

(CIE) Yxy coordinates ranged from 84, 0.214, 0.255 (stimulus 1)

to 5.3, 0.154, 0.09 (stimulus 20). The stimuli differed primarily in

the luminance axis (Y) and the y chromaticit y axis, consistent

with reports on Russian color categorization (e.g., see ref. 1; for

review, see refs. 2 and 27). The color squares were 2.5 cm per

side, and subjects viewed the screen from ⬇60 cm.

Color Discrimination Task. In each c olor discrimination trial, sub-

jects were shown a triad of color squares. One of the colors

presented on the bottom was physically identical to the top color

square (Fig. 1). The task was to indicate which of the bottom

squares matched the top square by pressing a key on the right or

lef t side of the keyboard. The nonmatching/distracter color

square was either very similar to the other two (two steps apart

in our continuum of 20, a near-color c omparison) or more

dif ferent (four steps apart, a far-color comparison).

7784

兩

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0701644104 Winawer et al.

The Interference Conditions. No-interference blocks consisted of

only color discrimination trials as described above. In verbal-

interference blocks, subjects were given an eight-digit number

series to rehearse during the color task. This series was presented

for 3 sec, and subjects were instr ucted to rehearse it silently.

Subjects rehearsed the number series while completing eight

c olor discrimination trials; their recall was then tested by choos-

ing between the original series and a foil which differed by

one digit.

In spatial-interference blocks, subjects v iewed a 4 ⫻ 4 square

grid of which four random squares were shaded black. Subjects

were instructed to remember the grid pattern by maintaining a

picture of it in their mind until tested. As with the verbal-

interference condition, a two-choice test was given after eight

intervening color discrimination trials. The incorrect grid dif-

fered in the location of one shaded square.

The spatial- and verbal-interference tasks were pretested for

dif ficulty in the absence of a primary task and found to result in

equal ac curacy (g rids, 95 ⫾ 1% c orrect; numbers, 96% ⫾ 1%

c orrect; two-tailed t (10) ⫽ 0.94; P ⬎ 0.35). Each of the three

blocks consisted of 136 color trials, w ith 17 interference stimuli

used in each of the two interference blocks. Each color appeared

equally often on the left and right and equally often as the match

and the distracter.

We thank the citizens of Cognation for insightful comments and

discussions. This research was funded by National Science Foundation

CAREER Grant 0608514 (to L.B.).

1. Dav ies IR, Corbett GG, McGurk H, MacDermid C (1998) J Child Lang

25:395–417.

2. Paramei GV (2005) Cross Cult Res 39:10–38.

3. Berlin B, Kay P (1969) Basic Color Terms: Their Universality and Evolution

(Univ of Californ ia Press, Berkeley, CA).

4. Roberson D, Davidoff J, Davies IR, Shapiro LR (2005) Cognit Psychol

50:378–411.

5. Roberson D, Davies I, Davidoff J (2000) J Exp Psychol Gen 129:369–398.

6. Heider ER (1972) J Exp Psychol 93:10–20.

7. Lindsey DT, Brown AM (2002) Psychol Sci 13:506–512.

8. Heider ER, Oliver DC (1972) Cognit Psychol 3:337–354.

9. Lucy JA, Schweder RA (1979) Am Anthropol 81:581–615.

10. Kay P, Kempton W (1984) Am Anthropol 86:65–79.

11. Brown R (1976) Cognition 4:125–153.

12. Pilling M, Davies IR (2004) Br J Psychol 95:429–455.

13. Pilling M, Wiggett A, Ozgen E, Dav ies IR (2003) Mem Cognit 31:538–551.

14. Davies IR, Corbett GG (1997) Br J Psychol 88:493–517.

15. Whorf BL (1956) Language, Thought, and Reality: Selected Writings (Technol

Press of MIT, Cambridge, MA).

16. Dav ies IR, Sowden PT, Jerrett DT, Jerrett T, Corbett GG (1998) Br J Psychol

89:1–15.

17. Roberson D, Davidoff J, Davies IR, Shapiro LR (2004) J Exp Psychol Gen

133:554–571.

18. Roberson D, Davidoff J (2000) Mem Cognit 28:977–986.

19. Pinker S (1994) The Language Instinct (Morrow, New York).

20. Munnich E, Landau B (2003) in Language in Mind: Advances in the Study of

Language and Thought, eds Gentner D, Goldin-Meadow S (MIT Press,

Cambridge, MA), pp 113–157.

21. 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).

22. Drivonikou GV, Kay P, Regier T, Ivry RB, Gilbert AL, Franklin A, Davies IRL

(2007) Proc Natl Acad Sci USA 104:1097–1102.

23. Gilbert A L, Regier T, Kay P, Ivry RB (2006) P roc Natl Acad Sci USA

103:489–494.

24. Bornstein MH, Korda NO (1984) Psychol Res 46:207–222.

25. Goldstone R (1994) J Exp Psychol Gen 123:178–200.

26. Karni A (1996) Brain Res Cognit Brain Res 5:39–48.

27. Moss A, Davies I, Corbett G, Laws G (1990) Lingua 82:313–332.

Winawer et al. PNAS

兩

May 8, 2007

兩

vol. 104

兩

no. 19

兩

7785

PSYCHOLOGY

Discussion

Remember, *"within category”* means that the 2 bottom squares (the match and the distracter) were both dark blue (“siniy”) or light blue (“goluboy”). *"Cross category”* means that one square was a dark blue and the other square was a light blue. You would expect that native Russian speakers would be faster to spot the correct square when the distracter square is a different color than the match one. As the graph indicates, you do see a category advantage (i.e. shorter reaction times) for *cross category* cases. This advantage goes away with verbal interference. #### ANOVA ANOVA stands for *Analysis of Variance*. It is a collection of statistical models used for analyzing the results of an experiment with multiple groups. ANOVA was pioneered by statistician and evolutionary biologist Ronald Fisher. The observed variance in a variable is partitioned into components attributable to the different sources of variation. A Mixed 2X2 ANOVA compares the mean differences between two groups that have been split on two factors. One factor is a within-subjects factor and the other is a between-subjects factor. If you want to learn more about this topic I would suggest [Chapter 11 of "Experimental Design and Analysis" by Howard Seltman](http://www.stat.cmu.edu/~hseltman/309/Book/chapter11.pdf) There has been a long standing debate surrounding the concept of linguistic relativity, which ponders whether language influences thought, and if so, how / how much. The *Sapir-Whorf hypothesis* (also known as *Whorfianism* / *Whorfian Hypothesis*) suggests that language indeed determines (or at least influences) a native speaker’s perception and categorization of experiences. The color debate is an important part of this larger discussion. In the cited studies they use the same verbal interference method used in this paper, namely, silently rehearsing a number while performing another task. Cited studies: - [21] [Effects of language on color discriminability](http://langcog.stanford.edu/papers_new/witthoft-2003-cogsci.pdf) - [22] [Further evidence that Whorfian effects are stronger in the right visual field than the left](http://www.pnas.org/content/pnas/104/3/1097.full.pdf) - [23] [Whorf hypothesis is supported in the right visual field but not the left](http://www.pnas.org/content/103/2/489.short) In general, European languages divide the spectrum like pictured below: ![color spectrum](https://upload.wikimedia.org/wikipedia/commons/thumb/d/d9/Linear_visible_spectrum.svg/605px-Linear_visible_spectrum.svg.png) As stated tho, Russian distinguishes between light blues (“goluboy”) and dark blues (“siniy”). There is no single word that encompasses “goluboy” and “siniy”. These are 2 separate colors in their own right. ![Russian Blues](https://i.imgur.com/XBwxkhI.png) There are plenty of interesting differences between languages when it comes to how they describe colors. For instance, before the modern period, Japanese had just one word, *Ao*, for both blue and green. In fact, in Japan people often refer to the “go” traffic light as being blue in color even tho it is just as green as anywhere else in the world. In Thai *khiaw* refers to green except when you are talking about the sea or the sky in which case it is blue. Sometimes these differences between languages are subtle, but sometimes they can be quite extreme. For instance to the *Dani* people of the highlands of New Guinea objects come in two shades: - *mili* for darker/cooler shades (blue, green, black) - *mola* for lighter/warmer shades (red, yellow, white)

Comments

Products

Project