ArticlePDF Available

The Development of Color Categories in Two Languages: A Longitudinal Study.

Authors:

Abstract and Figures

This study unites investigations into the linguistic relativity of color categories with research on children's category acquisition. Naming, comprehension, and memory for colors were tracked in 2 populations over a 3-year period. Children from a seminomadic equatorial African culture, whose language contains 5 color terms, were compared with a group of English children. Despite differences in visual environment, language, and education, they showed similar patterns of term acquisition. Both groups acquired color vocabulary slowly and with great individual variation. Those knowing no color terms made recognition errors based on perceptual distance, and the influence of naming on memory increased with age. An initial perceptually driven color continuum appears to be progressively organized into sets appropriate to each culture and language.
Content may be subject to copyright.
The Development of Color Categories in Two Languages:
A Longitudinal Study
Debi Roberson
University of Essex
Jules Davidoff
Goldsmiths College
Ian R. L. Davies
University of Surrey
Laura R. Shapiro
Goldsmiths College
This study unites investigations into the linguistic relativity of color categories with research on
children’s category acquisition. Naming, comprehension, and memory for colors were tracked in 2
populations over a 3-year period. Children from a seminomadic equatorial African culture, whose
language contains 5 color terms, were compared with a group of English children. Despite differences in
visual environment, language, and education, they showed similar patterns of term acquisition. Both
groups acquired color vocabulary slowly and with great individual variation. Those knowing no color
terms made recognition errors based on perceptual distance, and the influence of naming on memory
increased with age. An initial perceptually driven color continuum appears to be progressively organized
into sets appropriate to each culture and language.
Although the physiological basis of color vision is essentially
the same for all humans with normal trichromatic color vision
(Mollon, 1999), there is considerable diversity in the way that
different languages segment the continuum of visible colors. Some
languages are reported to use as few as two terms to describe all
colors (Heider, 1972); others use many more (Kay, Berlin, &
Merrifield, 1991; MacLaury, 1987). The variability exists even for
those terms deemed by Berlin and Kay (1969) to be basic (mono-
lexemic, present in the idiolect of all observers, and not subsumed
within the meaning of other terms). Once one considers secondary
terms, there is far greater diversity. Nevertheless, despite these
diverse naming systems, there are empirical data that support the
proposal for panhuman cognitive universals in color categorization
transcending terminological differences (Heider & Olivier, 1972;
Heider, 1972). Rosch (1973) further argued that these universal
categories are each based on the same focal colors regardless of the
number of terms in the speaker’s language.
The view that there is between-cultures agreement as to the
fundamental names, or basic color terms (BCTs), and that these
derive from the structure of the visual system, which is shared
among all humans, remains prevalent. Indeed, some recent studies
of languages that have 11 basic categories (Guest & Van Laar,
2000; Lin, Luo, MacDonald, & Tarrant, 2001; Sturges & Whit-
field, 1997) have provided some support for the preeminence of
category centers, or foci. However, there is no correspondence
between BCTs and any processes yet found in the visual system
(Boynton, 1997; Valberg, 2001; Webster, Miyahara, Malkoc, &
Raker, 2000); nameability is an important feature of color sets,
independent of any perceptual qualities of focality (Guest & Van
Laar, 2002), and cross-cultural studies have found no increased
salience for the proposed universal focal colors (Davidoff, Davies,
& Roberson, 1999; Jameson & Alvarado, 2003a). The present
study, therefore, reopens the debate on innate, panhuman cognitive
universals in color categorization (the universalist hypothesis) by
turning to a new source of evidence. It examines the acquisition of
color terms.
The studies reported here bring together two well-researched
fields of investigation into color categorization. The first field
concerns a broad range of recent studies that have reported effects
of language on a variety of cognitive tasks (number systems
[Gumperz & Levinson, 1997], spatial relations [Bowerman &
Choi, 2001; Levinson, 1996], artifact categories [Malt & Johnson,
1998; Malt, Sloman, & Gennari, 2003], modes of motion [Gennari,
Sloman, Malt, & Fitch, 2000], time [Boroditsky, 2001], shape
[Lucy, 1992; Roberson, Davidoff, & Shapiro, 2002], and gram-
matical gender [Sera, Berge, & del Castillo Pintado, 1994; Sera et
al., 2002]). In particular, a number of studies with adults have
shown effects of language on color categorization (Kay & Kemp-
ton, 1984; Pilling, Wiggett, O
¨
zgen, & Davies, 2003; Roberson &
Debi Roberson, Department of Psychology, University of Essex,
Wivenhoe Park, Colchester, United Kingdom; Jules Davidoff and Laura R.
Shapiro, Department of Psychology, Goldsmiths College, London, United
Kingdom; Ian R. L. Davies, Department of Psychology, University of
Surrey, Guildford, Surrey, United Kingdom.
Laura R. Shapiro is now at the Department of Psychology, University of
Warwick, Coventry, United Kingdom.
This research was supported by Economic and Social Research Council
Grant R0002383-10. We are grateful to Kemuu Jakarama, who acted as
interpreter, and to the Himba participants in these studies. The stimuli and
tasks were adapted from Boyles (2001).
Correspondence concerning this article should be addressed to Debi
Roberson, Department of Psychology, University of Essex, Wivenhoe
Park, Colchester CO3 4SQ, United Kingdom. E-mail: robedd@essex.ac.uk
Journal of Experimental Psychology: General Copyright 2004 by the American Psychological Association
2004, Vol. 133, No. 4, 554 –571 0096-3445/04/$12.00 DOI: 10.1037/0096-3445.133.4.554
554
Davidoff, 2000; Roberson, Davies, & Davidoff, 2000). For exam-
ple, Roberson et al. (2000) reported studies with adults in a remote
hunter– gatherer tribe in Papua New Guinea. Those studies found
substantial differences in perceptual judgments and memory per-
formance between a language with 11 BCTs and one with only 5
(Berinmo). Contrary to the influential view that language and
cognitive organization can be independent (Heider, 1972; Rosch,
1973), these differences suggest that language not only facilitates
memory performance but also affects the perceived similarity of
stimuli (the relativist hypothesis), a result also found by Kay and
Kempton (1984).
A second, largely independent field of study has examined the
tardy acquisition of color terms by children (Andrick & Tager-
Flusberg, 1986; Mervis, Bertrand, & Pani, 1995; Heider, 1971;
Sandhofer & Smith, 1999; Soja, 1994). These studies considered
the acquisition of color terms in English-speaking children for
whom the set of basic terms to be acquired consists of just those
that are presumed to be universally present before the correct
acquisition of the linguistic terms. Thus, given the evidence that
primary categories appear to be in place at 4 months of age
(Bornstein, Kessen, & Weiskopf, 1976), the relatively late acqui-
sition of color terms has puzzled many researchers. Bornstein
(1985) found that 3-year-old children were slow to learn paired
associates to colors relative to abstract shapes, and Sandhofer and
Smith (1999) found that young children were slower both to
comprehend color terms than size terms and to match items on the
basis of color (but see Pitchford & Mullen, 2001).
Estimates of the age at which children acquire a minimum color
vocabulary (four basic terms) have consistently dropped from the
7– 8 years of age estimated by Binet and Simon (1908, cited in
Bornstein, 1985) to 2–3 years (Andrick & Tager-Flusberg, 1986;
Shatz, Behrend, Gelman, & Ebeling, 1996). Indeed, several studies
have argued that, with constant intensive training, children as
young as 1.5 years can produce and use some color terms accu-
rately (Cruse, 1977; Mervis et al., 1995). However, Rice (1980)
reported that hundreds of trials were needed to achieve this out-
come, compared with the single-presentation learning demon-
strated for object terms (Carey, 1978). In addition, without such
intensive input, estimates of the age at which children acquire
color terms fall between 2 and 6 years, depending on the number
of terms examined and the measures of knowledge taken. These
studies used a variety of methodologies and examined either only
a single measure (naming or comprehension) or the acquisition of
only a small subset of terms over a very short period of time. We
therefore introduce strict controls on our response measures, ex-
amining both naming and comprehension systematically over a
3-year period to establish a more reliable measure of children’s
color term acquisition.
A related and unanswered question, raised initially by Bornstein
(1985), concerns how color term acquisition differs in speakers of
different languages. In the framework of a presumed innate, uni-
versal, fixed set of color categories, Bornstein (1985) predicted
that acquiring color terms would be even more difficult for chil-
dren learning a language in which the universal set must be
overwritten by a new set, even if there were fewer terms to be
learned. Children might have to assimilate their existing, hue-
based universal categories into a new and orthogonal set of lin-
guistic categories based on another dimension, such as lightness in
the case of the Dani reported by Heider (1972). Similarly, Bower-
man and Choi (2003) suggested that the more robust and prepotent
the prelinguistic organization of the perceived world is, the greater
is the resistance that language acquisition has to overcome to
restructure mental life. Thus, whereas Davidoff et al. (1999) pre-
sented compelling evidence for linguistic influence on color cat-
egories in adults, the acquisition of a set of name categories that
are different from the presumed set of innate, universal categories
might show a different developmental pattern than that of English-
speaking children.
We set out to address these questions in a study that included
two groups speaking different languages. One group of young
English children was tested initially before entering preschool and
subsequently through 3 years of formal education. The other group
of children was also monitored longitudinally and belonged to the
Himba, a seminomadic cattle-herding tribe in northern Namibia.
Himba people speak a dialect of the Herero language, but cultural
isolation over the last 100 years has resulted in a variety of cultural
and linguistic differences. Although villages communities have a
permanent base to which they return in the short rainy season, they
move around a series of temporary bush camps during the dry
season to find grass and water for their cattle. The area has few
natural resources, and Himba people’s homes, clothes, tools, and
artifacts are made from cattle products. Contact with the outside
world is sporadic, and, for a number of the youngest children
tested, the experimenter was the first White person they had
encountered. Few of the children received any formal education
during the period of the study. For those who did, school atten-
dance was restricted to brief periods, and color terms were not
included in the curriculum. Himba has five BCTs, according to the
criteria of Kay et al. (1991), although one, burou, is a recently
borrowed term from Herero. The BCTs are monolexemic, not
subsumed under the meaning of other terms, not restricted to a
narrow class of objects, and understood by all observers. The five
terms are serandu (broadly, red with orange and pink), dumbu
(broadly, beige with yellow and some light green), zoozu (broadly,
all dark colors and black), vapa (broadly, all light colors and
white), and burou (broadly, green with blue and purple).
The present study seeks to answer questions concerning color
category acquisition, primarily from analyses of the child’s color
term knowledge and memory errors. There are different predic-
tions for errors made during early color term acquisition, depend-
ing on their origin. If there is a predetermined, universal set of 11
prelinguistic cognitive categories, then children who have yet to
learn any of the appropriate labels for their language should show
similar patterns of memory errors, inasmuch as these reflect an
inevitable organization of color. Within-category confusions
should be more common than across-category confusions, even
with equal-difference steps between stimuli (measured in a per-
ceptually uniform metric, such as the L*a*b* dimensions of Com-
mission Internationale d’Eclairage [CIE] space; Wyszecki &
Stiles, 1982). A crucial aspect of these patterns is that they resem-
ble the organization of color by adult English speakers into 11
basic categories. Thus, if children start with a predetermined set of
perceptual color categories, English children learning color terms
need only to learn to map a set of labels onto their existing color
categories. As they also receive extensive training in a cultural
environment where these are highly salient, English children
should learn the universal set of color categories before Himba
children learn their linguistic set. Himba children should not only
555
COLOR CATEGORY DEVELOPMENT
lack explicit instruction but also be disadvantaged by having to
override the universal set to learn the set appropriate to their own
language.
An alternative account suggests that errors derive from the color
terms in the speaker’s language. Children might acquire some
familiarity with the set of categories appropriate to their language
and culture before they learn to apply any terms correctly. In this
case, one might expect to see a reflection of the different adult
patterns of naming in the cognitive organization of color even for
those children who have yet to succeed in naming color stimuli. At
later stages, the pattern of errors should be clearly dependent on
the color terms in the speaker’s language. In addition, young
children’s error-prone performance in color naming may itself
influence their cognitive representation of color. If, as Slobin
(1996, 2003) suggested, “thinking for speaking” (Slobin, 2003, p.
158) maintains some patterns of association at a high level of
activation, this may contribute to the difficulty children have in
overcoming initial naming errors or overextensions. However, this
account requires a different explanation for memory errors made
by children who definitely know no color terms. Without a pre-
determined organization, initial errors must depend on perceptual
similarity. Confusions should be made with the nearest perceptual
neighbors regardless of category. Hence, both groups of children
who know no color terms in their own language would be expected
to make similar errors. As children acquire color terms, the effects
of categorization (i.e., greater within- than between-categories
similarity) should become more important than perceptual
similarity.
Previous studies with English-speaking children (Andrick &
Tager-Flusberg, 1986; Sandhofer & Smith, 1999) found consider-
able discrepancy between children’s productive naming of color
stimuli and their comprehension of color terms. It could be that
productive naming is a more difficult task, as the number of
available responses is potentially infinite. Moreover, Zelazo and
colleagues (Zelazo, Frye, & Rapus, 1996; Zelazo & Reznick,
1991) found an asynchronous pattern of knowledge development
in young children, depending on the modality of response required.
Where verbal responding and manual responding were contrasted,
children often pointed to a correct choice but said the incorrect
one. Such behavior might be particularly evident in a population of
children, like the Himba, if they are learning to overwrite universal
cognitive categories with a new set of linguistic ones. The present
study, therefore, in addition to eliciting the lexical color terms that
children knew, examines both productive and comprehension mea-
sures of children’s color term knowledge. We took a composite
measure of term attainment first proposed by Soja (1994) that
includes both correct naming of stimuli and correct indication of
appropriate stimuli when children are asked to point to stimuli that
should be called by a particular name (e.g., red).
In summary, the study has two principal aims. The first aim is
to compare universal and relativist hypotheses of color categori-
zation by comparing children’s acquisition of color categories over
time in two different languages and cultures to establish whether
the pattern of acquisition relates to the appropriate set of adult
categories. Important data should come, in particular, from chil-
dren’s memory confusions at the first time of testing, before they
appropriately label colors. The second aim, in light of the widely
varying reports of children’s acquisition of color concepts, is to
systematically compare the similarities and differences of the
learning trajectories of children from two different cultures as they
acquire their appropriate set of color categories, by tracking nam-
ing and comprehension of color terms and cognitive organization
of color longitudinally over a 3-year period.
Method
Participants
The English children were 32 three-year-olds (12 boys, 20 girls) with a
mean age of 37.9 months and 36 four-year-olds (14 boys, 22 girls) with a
mean age of 47.5 months. The Himba children were 42 three-year-olds (25
boys, 17 girls) with an estimated
1
mean age of 40 months and 27 four-year-
olds (13 boys, 14 girls) with an estimated mean age of 50 months. Of the
69 Himba children, 63 became the longitudinal sample, completing all six
tests (30 boys, 33 girls). These children were tested on each occasion in
their home village in northern Namibia. None had received any schooling
at the first time of testing. At the sixth time of testing, 34 children had
attended a newly introduced mobile school for a period of between 6
months and 1 year. The remaining children had never attended school. Of
the 32 three-year-old English children, 28 became the longitudinal sample,
completing all six tests (11 boys, 17 girls). These children were tested in
their own home on the first occasion and subsequently in preschool and
then in primary school in Witham, England. Six Himba children and 8
English children moved with their family to a new home during the course
of the study and became unavailable for further testing. All children in both
populations had normal color vision, as assessed by the City Color Vision
Test (Fletcher, 1980). In addition, 24 adult English volunteers and 24 adult
Himba speakers (paid for their participation) with normal color vision
named the 22 stimuli.
All Himba children were rewarded for their participation with gifts of
maize flour for their family and small toys. English children received
stickers.
Materials
The stimulus set consisted of two identical sets of 22 Color Aid matte-
surface colored squares (Color Aid Corporation, New York, NY) measur-
ing 2 in. (5.08 cm) square and backed with stiff card. Eleven of the stimuli
were the best examples (focal colors) of the English basic categories black,
white, gray, red, green, blue, yellow, pink, orange, purple, and brown. The
other 11 were chosen to be intermediate between each of the chromatic
categories (e.g., halfway between pink and orange), given the arrangement
of color in a spherical three-dimensional space. The Appendix gives the
designations and CIE L*a*b* coordinates for each color.
Figure 1 shows the distribution of Himba color terms used by adults to
name a range of 160 Munsell stimuli varying in hue (horizontal axis) and
lightness (vertical axis) compared with the terms used by English adults.
Together these terms were used to name 86.2% of the stimuli.
In addition, a number of secondary terms, particular to the Himba dialect
and normally used specifically to describe the color of animal hides (cattle,
goats, etc.), were also used by a number of observers (vinde, vahe, kuze,
and honi). These represented 8.6% of the total names given.
1
We estimated age using a combination of measures. Taking family
histories within each village established which children had been born
around the time of independence (1991), and we also used rough estimates
of the order of births of other children after that time and the number of
rainy seasons between the birth of siblings. Additional measures included
checking that children could not touch the contralateral ear with their arm
raised over their head (normally attained at around 5 years), had all their
milk teeth, and were less than 120 cm tall.
556
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
For Himba speakers, those tiles most central to each of the five categories
were calculated on the basis of adult naming agreement. For three categories
(serandu, zoozu, and dumbu), more than one tile was considered to be a best
example on the basis of adult naming agreement (three for serandu, two each
for zoozu and burou). The nine tiles designated as best examples are hereafter
referred to as focals, as this terminology has been consistently used in past
research (e.g., Heider, 1972; Rosch, 1973). However, the criterion used here
(consensus of adult naming agreement) does not necessarily imply that these
are the only good examples of adult Himba categories. Adult speakers of
Himba, like the Berinmo participants tested by Roberson et al. (2000) and the
Dugum Dani participants tested by Heider (1972), showed little agreement
when asked to select the best examples of their basic categories from a much
larger set of colors (160). The categories of these languages may not have a
single focal point for all speakers. Indeed, even for English speakers there is
variability in choices of best examples.
Procedure
English children were tested by the experimenter in their own home in
the first instance and subsequently in a quiet corner of their preschool or
Figure 1. Distribution of Himba naming and choices of best examples for the 160-chip saturated array (for 31
observers) compared with those of English speakers for the same array. Numbers represent the number of
individuals choosing an exemplar as the best example of the category. Munsell lightness values are indicated on
the vertical axis; hue values are indicated on the horizontal axis. R red; Y yellow; G green; B blue;
P purple.
557
COLOR CATEGORY DEVELOPMENT
primary school. Children were tested seated at a table near a window in
conditions of natural daylight. Himba children were tested by the experi-
menter with the help of an interpreter, in their home village, seated at a
table in shaded natural daylight. All responses were recorded by the
experimenter in the order in which they were given.
The children were asked to complete four tasks on all test sessions. The
first was a listing task in which children were asked to “tell me all the
colors that you know.” Children who did not respond were encouraged,
“Do you know the names of any colors?” Color naming and comprehen-
sion were tested with two tasks, one productive, in which children were
shown each tile, one at a time, and asked, “What color is this?” followed
by a comprehension task in which all 22 tiles were laid out, in random
order, in front of the child and the child was instructed, “Show me a red
[serandu] one.” When the child had made a selection, this was noted by the
experimenter and the child was asked, “Is there another red [serandu]
one?” This procedure continued until the child responded negatively.
Children were classified as knowing a term when they both correctly
named the focal and pointed correctly to it when asked in the comprehen-
sion task, with the proviso that they not point to it in response to more than
one other term (on the basis of Soja, 1994, and Pitchford & Mullen, 2001).
We chose the lenient criterion of accepting one wrong selection because
the continued questioning by an adult might encourage children to choose
tiles more than once. All children also completed a memory task in which
the complete array of 22 tiles was laid out, in random order, in front of the
child and then covered up. The child was shown a tile (from an identical
set) for 5 s and told that the same colored tile was hiding under the cloth.
The tile was then removed, and the array of 22 tiles was exposed without
delay. The child was asked to find the color that he or she had just seen.
Two practice trials were given, with feedback, using tiles not included in
the set of 22, to establish that the children understood the task. Reaction
times were not formally measured on the recognition memory task, as
stimuli were hand presented. Informal measurement confirmed, however,
that responses were typically made within 15 s of when the test array was
uncovered by children in both populations. The list task was always
completed first, and the naming task was always completed before the
comprehension task. Order of tasks for the memory and naming/compre-
hension tasks was counterbalanced across participants and across test
sessions.
Results
First Testing
Children Who Know No Color Terms
Children were considered as knowing a color term if they passed
both the naming and the comprehension tests. Table 1 shows the
distribution of term knowledge at the first test. Our first analyses
are restricted to those children who knew no color terms, as it is in
this group of children that the predictions of universal BCTs can be
most clearly tested. We consider children who knew one or more
color terms in the following section. Nine English children (6 girls,
3 boys) and 27 Himba children (11 girls, 15 boys) failed to pass the
combined naming and comprehension criteria for any of their color
terms at first testing. Of the 9 English children, only 1 failed both
tasks for all 11 focal items. The remaining 8 passed either the
naming or the comprehension task for a mean of 1.78 focal items
(range 1–7). Similarly, for the Himba children, only 1 failed
both tasks for all 9 focal items. The remaining 26 children passed
either the naming or the comprehension task for a mean of 1.89
focal items (range 1– 8) and may have had some partial knowl-
edge of the appropriate set of categories for their language and
culture. Thus, our strict criteria for allocating children as knowing
no color terms are conservative for the prediction that their mem-
ory error patterns would be based on perceptual distance.
Memory confusions. The strong proposal that color categories
are universal, innate, and independent of language predicts that
both Himba and English children who knew no color terms would
share the same set of cognitive categories (those that correspond to
the English basic categories). Such organization of color categories
should be evidenced by similar confusions in memory between the
two populations, because colors belonging to the same category
should appear more alike than those from different categories.
However, it is also the case that items within the same category
(two red tiles) are closer to each other in perceptual terms than, for
instance, a red tile and a green tile. A pattern of memory confu-
sions based solely on perceptual distance (rather than categoriza-
tion) would also find more memory errors for perceptually closer
items, but there should be differences. For example, naming of
poor examples of categories (e.g., dark orange, dark blue) over-
rides the greater perceptual similarity to an exemplar of another
name category (red, black) in favor of the categorical name (or-
ange, blue). Therefore, we examined memory confusions with
respect to both perceptual distance and proposed predetermined
categories to see how far each factor would explain the data.
Separate 22 22 dissimilarity matrices were constructed for the
memory confusions of English and Himba children. SPSS repli-
cated multidimensional scaling (RMDS; Young & Harris, 1994)
was applied to the matrices. In a multidimensional solution of a
matrix, if two stimuli were always confused with each other, they
would occupy the same point in space. If they were never con-
fused, they would be placed as far apart as possible (given their
relationships to all the other stimuli). In comparing two solutions,
RMDS measures the departure from goodness of fit using S stress
(squared stress) values that represent the distance that each point in
the solution needs to be moved for the two arrays to fit perfectly
over each other. Kruskal’s (1964) primary approach was used for
tied ranks.
Table 1
Number of English and Himba Children at Each Level of
Knowledge for the Focal Items of Each Basic Color Term at the
First Test
Children and
color
Fail both
tasks
Name
correctly Comprehend
Pass both
tasks
English
Red 12 6 12 38
Blue 7 18 4 39
Green 8 13 6 41
Yellow 10 11 4 43
Pink 17 11 8 32
Purple 25 16 4 23
Orange 20 17 8 23
Brown 23 8 23 14
Black 12 8 10 38
White 14 6 10 38
Gray 51 15 0 2
Himba
Serandu 15 11 12 31
Zoozu 33 7 15 14
Dumbu 21 14 13 21
Vapa 14 9 7 39
Burov 47 7 10 5
558
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
The memory confusion matrices for those children who knew no
color terms in each population were compared with the confusion
matrix that results from judgments based on log perceptual dis-
tance (E, calculated as the Euclidean distance between each pair
of stimuli in the perceptually uniform CIE L*a*b* space;
Wyszecki & Stiles, 1982) and with the matrix derived from name
similarity in English (on the basis of adult naming of the 22
stimuli). For the full set of 22 stimuli, S stress values for each
comparison are given in Figure 2.
The Mantel (1967) test was used to compare the relative
strengths of the relationships between pairs of matrices (Legrende,
2000; Mantel, 1967). There were significant relationships between
Himba memory and log perceptual distance (r .343, p .001),
English memory and log perceptual distance (r .494, p .001),
and English and Himba memory (r .551, p .001). A Fisher’s
r-to-z test revealed no significant difference among the strengths of
these relationships (z 1). Thus, both groups of children who
knew no color terms made very similar errors, and these appear to
be closely associated with perceptual distance. In contrast, a com-
parison of the pattern that would be produced by English naming
revealed no significant relationship to Himba memory (r .079,
p .2), English memory (r .061, p .2), or log perceptual
distance (r .108, p .1).
In summary, memory errors for children who knew no color
terms were strongly influenced by perceptual distance. It is im-
portant to note that the error patterns of both groups of children
were clearly not influenced by the 11 basic color categories of
English. Errors appeared to be based on perceptual distance rather
than a particular set of predetermined categories. Navy blue, for
example, was more often confused with the more perceptually
similar purple or black than with focal blue. This point is ad-
dressed further in the longitudinal analyses.
Memory for focal colors. Another prediction of a universalist
hypothesis comes from the fact that focal items are presumed to
have a natural salience regardless of a particular language’s ter-
minology (Rosch, 1973). According to this view, the same focal
points exist for languages with fewer terms, even if those terms
reflect conjunctive categories—for example, blue with green or
yellow with green (Berlin & Kay, 1969). Therefore, stimuli central
to the 11 English categories (focal stimuli) should be remembered
better than other items, even by individuals who cannot name
them. Our stimuli were chosen to contrast naming and memory for
the focal points of English color categories to stimuli that lie
between these categories. Although some of the 11 focal stimuli
are also central to Himba categories, some are not; indeed, some
colors that are focal for Himba are peripheral to English categories.
There were thus four different types of focal status among the
stimuli: tiles that were focal for English only (gray, brown, orange,
purple, green, and pink), tiles that were focal for Himba only (dark
navy blue, dark orange, orangey pinky red, and greeny blue), tiles
that were focal for both languages (black, white, red, yellow, and
blue), and tiles that were focal for neither language (the 7 remain-
ing stimuli).
Table 2 shows the mean proportions of memory accuracy for
both groups for each type of stimulus. A 2 (language: English vs.
Himba) 4 (target type: English focal vs. Himba focal vs. focal
in both languages focal in neither language) analysis of variance
(ANOVA) examined memory accuracy across the four different
types of stimulus. There were a significant main effect of language
(English children were more accurate than Himba children), F(1,
34) 14.47,
2
.43, p .001, and a significant effect of target
type, F(3, 102) 5.05,
2
.15, p .003, but no significant
interaction, F(3, 102) 1. To further investigate the effect of
target type, we combined data from both language groups and
conducted a further one-way (target type: English focal vs. Himba
focal vs. focal in both languages focal in neither language)
ANOVA. There was a significant effect of target type, F(3, 105)
4.21,
2
.12, p .008. Post hoc Newman–Keuls analysis
revealed that only those targets that were focal in neither language
were recognized significantly less often than those that were focal
in both languages ( p .01). There were no other significant
differences.
List task. Although children only offered color terms, there
was no consistency about the terms listed. Of the 9 English
Figure 2. Stress values for multidimensional scaling comparisons of memory confusions by Himba children
knowing no color terms and English children knowing no color terms compared with stress values found for
perceptual distances and for naming by adult English speakers.
559
COLOR CATEGORY DEVELOPMENT
children who knew no color terms, 4 also listed none. Of the other
5, 2 listed three, 1 listed five, and 2 listed six terms. Of the 27
Himba children who knew no color terms, 24 listed none. One
child listed two terms, 1 listed three, and 1 listed four.
Both English and Himba children with no color term knowledge
made similar memory errors, and it is crucial to note that neither
error pattern resembles that derived from the 11 basic terms of
English. Furthermore, for these children, there was no specific
advantage for colors that were focal to either language. However,
colors not focal in either language were recognized significantly
less successfully than those that were focal in either English,
Himba, or both languages. Closer examination of the four sets,
however, reveals that perceptual distances were not equal across
the sets. In particular, the distance of each item from its nearest
neighbors (E) in the set that were focal in neither language was
significantly smaller than those of the other three sets (all ps
.05). Within the set that was not focal in either language, 67.5% of
errors were for the items with the smallest (E
2
) differences. Thus,
the high rate of errors for this set derives from perceptual confu-
sion and is unrelated to focality. For both groups of children, there
was no significant difference in memory accuracy for the other
three groups of stimuli, despite the fact that some Himba focals are
nonfocal for English speakers and vice versa.
Thus, before a child had acquired any color terms, there was no
evidence of any innate categorical organization. Memory confu-
sions appeared to be based on perceptual distance rather than a
particular set of predetermined categories. It seems that the innate
neurophysiological basis for perceptual discrimination is not itself
sufficient to provide color categories (see also Gellatly, 1995).
Indeed, the argument is stronger in light of the conservative criteria
for defining color term knowledge. The majority of children who
knew no color terms showed partial knowledge of their own set of
color terms and volunteered only color terms in the list task,
although these bore no relation to their knowledge. Such partial
knowledge might have been expected to drive apart the cognitive
organization of color for the two groups and make the memory
patterns more closely resemble that of their community’s adult
naming. Such partial knowledge, however, does not appear to be
enough to override perceptual similarity when memory confusions are
made. We return to this issue in considering the longitudinal data.
Children Who Know One or More Color Terms
Order of term acquisition. Universalist theories (e.g., Kay &
McDaniel, 1978) predicted, on the basis of fuzzy set theory, that
English speakers would learn primary terms (red, yellow, green,
blue) before secondary terms (orange, brown, purple, pink), be-
cause the primary terms correspond to neural opponent process
responses, whereas the secondary terms correspond to combina-
tions of responses. Although the physiological basis for this argu-
ment has been shown to be false, many researchers continue to
infer the case for heightened salience of primary over secondary
terms on these grounds (see Jameson & Alvarado, 2003b, and
Saunders & van Brakel, 1997, for discussions). Indeed, previous
studies of English-speaking children have not upheld the predic-
tion concerning primary terms (Mervis, Catlin, & Rosch, 1975;
Shatz et al., 1996).
The English data reported here show a pattern generally similar
to that found in Pitchford and Mullen (2001), and, overall, children
knew more primary (M 2.37) than secondary (M 1.35) colors,
t(58) 7.45, p .001. However, when the data were considered
for individuals, there was little consistency in which terms were
learned first.
For the English sample, of the 4 children who knew one term, 2
knew blue, 1 knew green, and 1 knew pink. Of the 2 children who
knew two terms, 1 knew yellow and red, and the other knew yellow
and green. Three-term combinations included brown, pink, black;
brown, green, black; brown, blue, red; and yellow, white, green.
None of the 8 children who knew four terms knew the same four,
and only 2 of the 12 children who knew five terms knew the same
five. To estimate the degree to which these patterns match the
probability that primary terms are generally learned before sec-
ondary ones, we subjected the data to a Guttman analysis (Ham-
mond, 1990). Only 51 of the 2,048 possible random combinations
of known terms were observed. When we matched our findings to
a model in which primary terms appear before secondary ones,
only 3 children’s profiles matched the model, whereas 48 did not,
producing a coefficient of reproducibility of only .415 (upward of
.700 is generally considered useful).
Despite the smaller potential variation, a similar pattern was
observed among Himba children. Of the 16 children who knew
only one term, 7 knew vapa, 4 knew serandu, 2 knew dumbu,2
knew zoozu, and 1 knew burou. Of the 10 children who knew two
terms, 5 knew serandu and vapa, 2 knew zoozu and vapa, 2 knew
serandu and dumbu, and 1 knew dumbu and vapa. Every possible
combination was observed for three- and four-term knowledge,
and no child knew all five terms.
For children who knew at least one color term, there was little
evidence, in either language, of a predictable order of acquisition.
Item dependence was previously noted by Mervis et al. (1975),
who reported that more children knew red and orange than knew
purple and brown. In the present study, fewer English children
knew brown and gray than other terms, supporting the findings of
Pitchford and Mullen (2001), but 3 of the English children who
knew just three color terms did know brown. Pitchford and Mullen
(2001) argued that, in addition to restricted lightness levels, these
colors have low perceptual salience relative to other basic colors.
Gray and brown, they suggested, are such common colors that they
give little predictive information for object identification and so
are harder to learn. It might be that salience could be boosted for
some individuals, for instance, by owning a brown dog or a gray
cat. However, there is no evidence that children learn color terms
from their reliable associations to particular objects. Indeed, it is
unlikely given the neuropsychological evidence for a double dis-
Table 2
Proportion Mean Accuracy for 9 English and 27 Himba
Children Knowing No Color Terms for Items That Are Focal for
English Only (n 6), for Himba Only (n 4), for Both
Languages (n 5), or for Neither Language (n 7)
Focal
English Himba
MSEMSE
English only .42 .06 .18 .03
Himba only .43 .08 .19 .04
Both languages .50 .12 .23 .04
Neither language .25 .08 .13 .02
560
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
sociation in performance on the two tasks (Davidoff & Roberson,
2004; Luzzatti & Davidoff, 1994; Miceli et al., 2001). This issue
is considered again in the examination of the longitudinal data.
Another similarity in performance between the two populations
concerned the equivalence of naming and comprehension. In the
children’s attainment of color terms, we did not observe the
predicted asynchrony of knowledge development (comprehension
better than naming) found in other areas by Zelazo and Reznick
(1991). In other tasks, comprehension may be easier because the
name is given to the child and the range of items that can be
matched to it is limited to the stimulus set. The naming task may
be harder because the child must generate a label from a poten-
tially infinite set. However, there are two possible reasons why an
asynchronous pattern of knowledge development might not occur
here. First, children from both cultures only volunteered color
words in response to the question, “What color is this?” Therefore,
knowledge of the appropriate lexical category restricted the poten-
tially available set of responses. Second, children may have over-
extended their choices in color comprehension tasks as well as in
naming tasks (a result found by Backsheider & Shatz, 1993). The
consistent lack of asynchrony in both populations is an indication
that children’s difficulties in establishing color categories are at a
conceptual level (i.e., where to place boundaries along a continu-
ous dimension) rather than at a lexical level (i.e., selection of the
appropriate label for a known group of stimuli).
Memory for focal colors. We tested memory for focal colors
using the same procedures as with children who knew no color
terms. Table 3 shows the mean proportions of memory accuracy
for English and Himba children for targets focal only in English,
focal only in Himba, focal in both languages, and focal in neither
language. Accuracy for each different type of target was examined
with a 2 (language: English vs. Himba) 4 (target type: English
focal vs. Himba focal vs. focal in both languages vs. focal in
neither language) ANOVA. There were significant main effects of
language, F(1, 99) 22.65,
2
.26, p .001, and of target type,
F(3, 297) 34.79,
2
.35, p .001, as well as a significant
interaction between language and target type, F(3, 297) 5.52,
2
.06, p .001. Newman–Keuls pairwise comparisons of the
interaction revealed that English children recognized significantly
more targets than Himba children of those that were focal in both
languages, focal in English, and focal in neither language (all ps
.05). English children recognized significantly more targets that
were focal in both languages than any other type (all ps .01) and
more targets focal in English than those focal in Himba or in
neither language ( p .05 or p .01, respectively). Himba
children recognized significantly more of those targets that were
focal in both languages than those that were focal in English or in
neither language (both ps .01) and more targets focal in Himba
than those focal in English or in neither language ( p .05 or p
.01, respectively). Thus, both groups of children recognized more
of the stimuli that were focal than those that were not focal in their
own language.
The previous analysis takes no account of the effects of color
term knowledge on memory for focal colors. For a child who knew
only 1 term, there was only 1 chance to correctly identify the focal
tile of the term that he or she knew but 10 chances to correctly
identify the focal tiles of the terms that he or she did not know. The
opposite was true for a child who knew 10 terms. To examine the
effect of term knowledge on memory, we therefore calculated the
proportion of correct memory choices for terms known and not
known for each child. Figure 3 illustrates memory accuracy for
both groups by terms known.
For English children who knew at least one but not all terms, the
proportion of correct identifications for the focal chips was com-
pared in a 2 (age: 3 vs. 4) 2 (knowledge: known vs. not known)
analysis of covariance (with number of terms known as covariant).
There was only one reliable effect: a significant effect of knowl-
edge, F(1, 55) 18.45,
2
.34, p .001. There were no
significant effects of age, F(1, 55) 1.94,
2
.04, p .169, or
number of terms known, F(1, 55) 1.66,
2
.03, p .133, nor
any interactions (all Fs 1). For Himba children who knew at
least one and not all terms, the identical analysis again revealed a
very similar pattern of results. There was a significant effect of
knowledge, F(1, 48) 36.41,
2
.76, p .001, but no
significant effect of age, F(1, 48) 2.39,
2
.05, p .129, no
significant effect of number of terms known, F(1, 48) 1, and no
significant interactions (all Fs 1). Thus, the analyses showed, for
both populations, that there was a stable relationship between
memory accuracy and term knowledge that was unrelated to age.
Children in both populations and at both age groups, at all levels
of color knowledge, correctly remembered more stimuli for terms
that they knew than for those that they did not know.
A calculation of the mean perceptual distance of memory errors
also showed an effect of knowledge. Mean error distances are
shown in Table 4.
An analysis by item for children who made errors both on tiles
for which they did and on tiles for which they did not know the
term revealed that English children made errors to more distant
distractors when the terms were not known than when they were
known, t(10) 3.89, p .01. The same was true for Himba
children, t(8) 2.42, p .05.
We further investigated the data to establish whether passing
both tests (our criterion for knowing a term) was more beneficial
than passing either test alone. These analyses considered sepa-
rately, for each group, only the items that were focal in the group’s
own language, as it was not possible to judge children’s naming as
either correct or incorrect for items on which adult naming agree-
ment was very low. Table 1 shows the number of children in each
group at each stage of knowledge for each color term. Figure 4
illustrates the effect of knowledge status on memory. For the
English focal items, a one-way (level: fail both tests vs. pass
naming vs. pass comprehension vs. pass both tests) ANOVA on
recognition accuracy showed a significant effect of knowledge
Table 3
Proportion Mean Accuracy for 59 English and 42 Himba
Children Knowing at Least One Color Term for Items That Are
Focal for English Only (n 6), for Himba Only (n 4), for
Both Languages (n 5), or for Neither Language (n 7)
Focal
English Himba
MSEMSE
English only .48 .32 .27 .03
Himba only .39 .04 .36 .04
Both languages .66 .04 .41 .04
Neither language .35 .03 .15 .02
561
COLOR CATEGORY DEVELOPMENT
status, F(3, 30) 7.70,
2
.69, p .001. Newman–Keuls
pairwise comparisons showed that those targets for which both
tests were failed were recognized significantly less often than
those in all other conditions (all ps .05). There were no other
significant differences. For the Himba focal items, a similar one-
way (level) ANOVA again showed a significant effect of knowl-
edge status, F(3, 27) 3.08,
2
.35, p .05. Newman–Keuls
pairwise comparisons showed that those targets for which both
tests were failed were recognized significantly less often than
those for which both tests were passed ( p .05). There were no
other significant differences.
It is not surprising that English children performed significantly
better than Himba children on the recognition memory task; this
was the case even for children at the youngest age group who had
not yet started preschool. However, and more important, effects of
focality only appear as the child acquires color terms. English
children remembered significantly more of the items that were
focal in both languages and more of the items that were focal only
in English. Himba children remembered significantly more of the
items focal in both languages and those focal only in Himba. Thus,
focality is language dependent rather than universal. Knowledge of
the appropriate term also reduces the distance of errors when they
are made, and even partial knowledge of a term increases the
probability of correct identification in the memory task.
List task. No child from either culture offered anything other
than color terms (rather than other words) when asked to list the
color words they knew. However, whereas English children listed
only BCTs, 4 Himba children included at least one secondary term.
English 3-year-old children listed a mean of 1.26 terms (range
0 6). English 4-year-olds listed a mean of 2.73 terms (range
0 –7). Himba 3-year-old children listed a mean of 0.85 terms
(range 0 4), whereas 4-year-olds listed a mean of 1.40 terms
(range 0 –5). Children from both cultures often offered color
terms that they could not identify. Of those terms listed for which
a child did not know the term, 30% were to targets for which both
tests were failed, 42% were to targets that were only named
correctly, and 28% were to targets that were only comprehended.
The list task was the only one for which color knowledge did not
appear important. Like the children who knew no color terms,
children frequently listed a term they did not know. Thus, children
from both cultures seem to acquire the concept of color and a set
of terms that describe colors before they can correctly apply those
terms. A similar result was found previously for English-speaking
children (Andrick & Tager-Flusberg, 1986; Davidoff & Mitchell,
1993; Gottfried & Tonks, 1996; Mervis et al., 1995). The lack of
relationship between the listing of color terms and category knowl-
edge is further evidence against the idea that learning color terms
requires only the mapping of linguistic terms onto existing cate-
gories. The argument is stronger for English-speaking children
because they listed only basic terms.
Longitudinal Data
By the sixth time of testing, all English children had completed
2.5 years of education. At the sixth test, 29 Himba children had
never attended school, whereas 34 children had attended one of the
newly established mobile schools for between 6 months and 1
year. Attendance at school was determined solely by geographical
location and not by ability. Where schooling was available, all
children in a village attended. Instruction at mobile schools was
given in Herero, and color terms were not included in the curric-
ulum. However, as Herero has additional color terms ( pinge [pink]
and grine [green]) that are not used in Himba, in the subsequent
analyses, findings are considered with and without school
attenders.
Order of Term Acquisition
The numbers of the 28 English children in the longitudinal study
who passed the knowledge criteria for each term on each occasion
are shown in Table 5. For the 63 Himba children in the longitu-
dinal study, the numbers of children who passed knowledge cri-
teria for each term on each occasion are shown in Table 6. Passing
the criterion for three of the five Himba terms involved correctly
naming and comprehending more than one tile. As this might be a
harder criterion to reach than that required for English children
(one tile per term), numbers in parentheses in Table 6 show the
number of Himba children who passed the criterion for each term
when just one tile was considered for each color term. For the 29
Himba children who had never attended school, the mean number
of terms known at Test 6 was 3.21. For the 34 children who had
attended mobile schools, the mean number of terms known was
2.74. There was no significant difference in term knowledge
between these groups, t(61) 1.62, p .05.
Figure 3. Memory performance on focal stimuli for 3- and 4-year-old
Himba and English children at first test, grouped according to the number
of terms known. yrs years.
Table 4
Mean Error Distance (Measured in Commission Internationale
d’Eclairage L*a*b* Space) for Memory Confusions Made by
Participants to Focal Colors for Which They Knew the Term
and Those for Which They Did Not
Children Known Not known
English 38.70 112.64
Himba 43.91 74.18
562
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
At the sixth test for English children, 1 child knew 5 terms, 1
knew 7, 2 knew 8, 6 knew 9, 7 knew 10, and 12 knew all 11 terms.
At the sixth test for Himba children, 2 knew no color terms, 7
knew 1, 14 knew 2, 16 knew 3, 23 knew 4, and only 1 knew all 5
terms. In the interval between each testing time (6 months), most
English children and some Himba children acquired more than 1
new color term. It thus was not possible to determine the order in
which terms were acquired in a systematic way. However, two
additional measures were used in an attempt to address the ques-
tion of the order of term acquisition. In the first analysis, each BCT
was assigned an ordinal position for each child (allowing for ties
and based on the first time that the child passed the knowledge
criteria for that term). The rank order of terms for English children
across all six tests was yellow, red, orange, blue, green, black,
pink, white, purple, brown, gray. In an alternative analysis, the
average age at which both tests were passed was calculated for
each color term. According to this criterion, the order of terms for
English children was yellow, red, black, blue, orange, white, pink,
green, purple, brown, gray (the average age of acquisition ranged
from 44.13 months for yellow to 52.9 months for gray). Both
measures are imprecise, given the time elapsed between tests and
the fact that some children acquired up to five new terms between
two testing sessions, but neither reveals a division strictly based on
primary and secondary terms. Previous investigations of children’s
color term acquisition assessed the age at which children know
color terms either on a single test (Heider, 1971; Macario, 1991;
Pitchford & Mullen, 2001) or over a short acquisition period
(Andrick & Tager-Flusberg, 1986; Mervis et al., 1995; Sandhofer
& Smith, 1999). A longitudinal study allowed us to verify that
such estimates reflect permanent category acquisition rather than a
transitory effect of intensive training.
For English children, early failures on the naming task exclu-
sively involved incorrect use of a BCT (e.g., white called pink,
orange called yellow, brown called black, red called orange); later
failures largely involved incorrect use of secondary terms (e.g., red
called lilac, pink called skin colored, purple called multicolored).
Only one error at Test 6 involved the wrong application of a basic
term (pink called purple). For Himba children, both early and late
Figure 4. Mean proportion of correct memory identifications of focal stimuli in their own language by English
and Himba children at each stage of color term knowledge.
Table 5
Number of Children out of 28 Passing Both Productive and Comprehension Tasks for the Focal
Items of Each of the 11 Basic English Color Terms on Each of Six Tests
Time
of test Red Green Yellow Blue Pink Purple Orange Brown Black White Gray
11511 171210 4 10 4 1311 2
2 15 17 17 18 13 14 12 8 18 15 9
3 20 20 19 19 14 14 12 5 15 18 9
42225 27241917 15 15 242511
52224 24222323 20 18 262618
62828 26252227 22 22 282820
563
COLOR CATEGORY DEVELOPMENT
failures on the naming task frequently involved incorrect use of
both basic and secondary terms (e.g., zoozu called dumbu or vinde,
serandu called dumbu or honi, vapa called dumbu or vambi).
For both populations, some children who passed the criteria for
knowing a term on one occasion failed it on another. For the
English children, the four terms considered primary receive intense
adult input and feedback in the first year of preschool. Sixty-six
percent of English children knew all of red, blue, green, and yellow
by the end of their first year of preschool (Test 3), compared with
42% who knew all the other basic terms, but this figure rises to
55% with the exclusion of brown and gray. The difference in
number of subsequent errors for terms that had been passed on a
previous occasion for primary terms (M 1.71) and secondary
terms (M 2.18) was not significant, t(27) 1.61, p .1. Across
all six tests, the average proportion of times on which children
passed the criterion for a term on one occasion but subsequently
failed was 8% (range 0%–18%). For the Himba children, across
all six tests, the average proportion of times on which they passed
the criterion for a term on one occasion but subsequently failed on
a following occasion was also 8% (range 0%–24%).
In summary, the rate of subsequent failure of one or another
criterion after having passed both on a previous test was small but
very similar across the two populations, despite the English chil-
dren receiving more intensive reinforcement in the intervening 6
months between tests. The present data suggest that previous
studies using only a single test give slight underestimates for the
age of color term acquisition.
Memory for Focal Colors
The effect of focal status on memory over time was examined in
each language, in analyses similar to those used for the first
testing, with a 4 (target type: English focal vs. Himba focal vs.
focal in both languages vs. focal in neither language) 6 (time of
test: 1– 6) ANOVA. These results are illustrated in Figure 5. For
English children, there were significant main effects of target type,
F(3, 486) 62.42,
2
.39, p .001, and time of test, F(5,
162) 18.38,
2
.57, p .001, but no significant interaction,
F(15, 486) 1.60,
2
.05, p .069. Performance on all types
of target improved significantly over time. To examine the effect
of target type, we collapsed data across the six tests and conducted
a one-way ANOVA comparison (target type: English focal vs.
Himba focal vs. focal in both languages vs. focal in neither
language). The main effect was significant, F(3, 333) 30.60,
2
.27, p .001. Examination of the simple main effects
revealed that targets that were focal in both languages were rec-
ognized significantly better than any other target type (all ps
.01) and targets that were focal in English were recognized sig-
nificantly more often than those that were focal in Himba or in
neither language (both ps .01).
For Himba children, in the initial analysis of Target Type
Time of Test, there were significant main effects of time of test,
F(5, 372) 22.51,
2
.54, p .001, and of target type, F(2,
744) 29.21,
2
.39, p .001, but no significant interaction,
F(10, 744) 1. Performance on all types of target improved
significantly over time. To examine the effect of target type, we
Table 6
Number of Children out of 63 Passing Both Productive and
Comprehension Tasks for the Focal Items of Each of the Five
Himba Color Terms on Each of Six Tests
Time of test Serandu Vapa Zoozu Dumbu Burou
1 25 (31) 30 11 (13) 17 4 (10)
2 31 (42) 43 34 (38) 21 0 (13)
3 29 (47) 46 30 (32) 23 3 (18)
4 37 (55) 49 32 (34) 30 8 (14)
5 37 (55) 57 23 (27) 35 7 (21)
6 46 (58) 60 29 (29) 43 8 (15)
Note. Numbers in parentheses represent the number of children passing
the tests when only one focal item was considered for each term.
Figure 5. Mean correct memory identifications for English focal items,
Himba focal items, focal items for both languages, and focal items for
neither language by English and Himba children across all tests.
564
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
collapsed data across the six tests and conducted a similar one-way
ANOVA (target type: English focal vs. Himba focal vs. focal in
both languages vs. focal in neither language). The main effect of
target type was significant, F(3, 753) 34.38,
2
.21, p .001.
Examination of the simple main effects revealed that targets that
were focal in both languages and focal in Himba were recognized
significantly more often than those that were focal in English or in
neither language (all ps .01).
The difference between the two languages and the superior
recognition of those items focal to the speaker’s language persisted
over time in spite of the overall improvement in recognition
memory.
Overall, English children’s recognition performance was better
across all tests. One possible explanation is that color terms are
taught in English schools. However, in part, the increased recog-
nition accuracy may be accounted for by the acquisition of more
smaller categories because the broader categories of the Himba
language allow more within-category errors. By the final test, the
number of tiles selected for each term by English children was
close to one. Thus, knowledge of the appropriate term was a better
memory aid for English children, for whom there is only one best
example of each basic term in the set, than for the Himba children,
for whom there are several equally good examples of each category.
Memory performance across all six tests for focal items from
categories for which children knew terms was compared with
memory for focal items from categories for which they did not.
Table 7 shows the mean proportion of terms known across tests.
For both groups, known items were identified significantly more
often than unknown items in the first tests (all ts 2.09, all ps
.07). For both groups, by Test 6, there was no significant differ-
ence in recognition accuracy between known and unknown items.
These results are illustrated in Figure 6.
Perceptual Versus Categorical Errors
We compared the distance (measured in CIE space) of errors to
the focal items of known and not known terms over time. Figure
7 illustrates these results. In addition to a reduction in the number
of errors over time, for both groups the mean perceptual distance
of erroneous choices also decreased across tests. For English
children, across all tests, there was a trend for smaller error
distances when a term was known than when it was not, t(5)
2.337, p .06. For Himba children, across all tests, error distances
were significantly smaller when a term was known, t(5) 4.332,
p .01.
In most cases, making more within-category errors would also
reduce the perceptual distance of confusions. We therefore exam-
ined two cases in which perceptual and categorical errors could be
directly contrasted. Of particular interest was children’s perfor-
mance on the navy blue tile, which lies perceptually between
English focal blue and black. For English speakers, this tile is in
the same category as the focal blue tile. For Himba speakers,
however, it is in the same category as the black tile (and both are
equally focal). Within the set of 22 tiles, there is a closer percep-
tual alternative than either of these; this is the English focal purple
tile. If choices were only influenced by perceptual similarity, the
purple tile should be a more frequent erroneous choice than either
the lighter blue or the black tile for both populations.
We examined, across tests, the numbers of times that those
children who made a recognition error for the navy blue tile chose
the focal blue tile, the black tile, or the focal purple tile. Figure 8
illustrates these results. A 2 (language: English vs. Himba) 3
(error type: perceptual vs. English category vs. Himba category)
chi-square comparison of English and Himba children’s errors at
the first two tests (counting only the first error per child) revealed
Table 7
Mean Proportion of Terms Known Across All Six Tests
Children
Test
123456
English .354 .506 .581 .727 .831 .896
Himba .276 .409 .416 .473 .505 .590
Figure 6. Mean number of memory errors to focal items known and not
known by all children in both groups across tests. *p .05. **p .01.
565
COLOR CATEGORY DEVELOPMENT
no significant difference between groups,
2
(2, N 82) 0.68,
p .7. However, a 2 (language: English vs. Himba) 3 (error
type: perceptual vs. English category vs. Himba category) chi-
square comparison of English and Himba children’s errors at the
final two tests (counting only the first error per child) revealed a
significant difference between groups,
2
(2, N 50) 9.81, p
.01. An examination of the residuals revealed significant differ-
ences between the two groups in numbers of choices of the blue
tile (z 2.2, p .05) and the black tile (z 2.0, p .05) but no
significant difference in choices of the purple tile (z 0.2, p .4).
Thus, initially both groups made errors of all types, but by the final
two tests, significantly more children from each group made cat-
egorical errors appropriate to their own language. To eliminate the
possibility that this result was skewed by the attendance at school
of some of the Himba children during the final year of testing, we
reanalyzed the data, removing the errors of all children who had
attended school. The pattern was not affected; a chi-square com-
parison of the remaining data still revealed a significant difference
between groups,
2
(2, N 32) 12.27, p .01.
A similar comparison was carried out for children’s perfor-
mance on the dark orange tile, which lies perceptually between
English focal red and focal orange. Figure 9 illustrates these
results. For English speakers, this tile is in the orange category. For
Himba speakers, however, it is in the same category as the red tile
(and also focal). The red tile is also the closest perceptual alter-
native within the set. If choices were only influenced by perceptual
similarity, the most frequent erroneous choice for both populations
would be the red tile.
We examined, across tests, the number of times that those
children who made a recognition error for the dark orange tile
chose either the focal red or the focal orange tile. A 2 (language:
English vs. Himba) 2 (error type: English category vs. Himba
category) chi-square comparison of English and Himba children’s
errors at the first two tests (counting only the first error per child)
revealed no significant difference between groups,
2
(1, N
53) 1.16, p .2. A 2 (language: English vs. Himba) 2 (error
type: English category vs. Himba category) chi-square comparison
of English and Himba children’s errors at the fifth and sixth tests
revealed a significant difference between groups,
2
(1, N 42)
11.98, p .001. Again, by the final test, the two groups made
categorical errors consonant with their own language. To eliminate
the possibility that this result was skewed by the attendance at
school of some of the Himba children during the final year of
testing, we reanalyzed the data, removing the errors of all children
who had attended school. Again, an analysis of the remaining data
revealed a significant difference between groups,
2
(1, N 35)
12.16, p .001.
List Task
The probability of a child listing a term that he or she knew
rather than one that he or she did not know was investigated over
time. Figure 10 illustrates these results. For English children, a
one-way ANOVA (time of test: 1– 6) showed a significant effect of
time, F(5, 145) 11.02,
2
.28, p .001, with a significant
linear trend, F(1, 150) 53.44,
2
.27, p .001. Over time,
children were increasingly likely to list terms that they knew rather
than those that they did not know. For Himba children, a similar
one-way ANOVA (time of test: 1– 6) showed a significant effect of
time, F(5, 372) 12.72,
2
.15, p .001, with a significant
linear trend, F(1, 377) 55.43,
2
.13, p .001. Again, across
tests, children were increasingly likely to list terms that they knew
rather than those that they did not know.
The longitudinal data show that the pattern of term acquisition
is slow for both languages. Several of the English children could
not correctly name and point to named examples of all 11 BCTs at
age 6, even after 3 years of instruction. In the context of a
gradually improving memory performance, the longitudinal data
reinforce the first testing data, showing that children gained a
particular advantage in memory for the colors focal to their own
language.
Our data point strongly to effects based on color term knowl-
edge. Across all tests, children from both groups recognized a
greater proportion of target chips for terms they knew than for
Figure 7. Mean perceptual distance of errors to known and not known focal targets by English and Himba
children across tests. T time.
566
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
terms they did not know; however, this is an effect that necessarily
decreased over time. For a child who knows no color terms, it may
be necessary to consider all 22 alternatives in the recognition
phase. For a child who knows 1 term (e.g., pink) and who is shown
a blue tile, the probability that the child will be accurate should
increase, because none of the four alternatives that could be called
pink need be considered in the recognition phase. In the extreme
case, once a child knows 10 terms, the range of possible alterna-
tives at recognition is dramatically reduced, as none of the known
alternatives need be considered for a tile the child does not know.
Thus, recognition for tiles for which the term is unknown should
approach a ceiling as a child acquires a full set of color terms.
Our longitudinal sample also provides data for the contrast
between perceptual and categorical mechanisms underlying color
term acquisition. As the number of errors in the memory task
decreased steadily over time, so did the perceptual distance of
those errors to the target. For Himba children, in particular, the
perceptual distance of errors to targets for which the term was
known was significantly less than the perceptual distance of errors
to targets for which the term was not known. However, these data
need to be considered in light of the increasing influence of
naming and categorically derived performance.
The influence is highlighted by the case of a nonfocal tile that
falls into different categories for adult speakers of the two lan-
guages; there is a widening difference between the types of errors
that children make over time. Speakers of each language showed
an increasing tendency to make within-category rather than cross-
category errors. The name-based errors eventually overcome the
tendency to make errors to the closest perceptual stimulus.
Figure 8. Mean proportion of erroneous identifications for the navy blue
target tile to the black, the best example blue, or the closest perceptual
match (purple). T time.
Figure 9. Mean proportion of erroneous identifications for the dark
orange target tile to the best example red (closest perceptual match) or the
best example orange tile. T time.
567
COLOR CATEGORY DEVELOPMENT
General Discussion
In this study we set out to investigate the origin and develop-
ment of color term knowledge and cognitive color categories in
two very different languages and cultures. Himba children live in
a sparsely populated, arid environment; their homes, clothes, tools,
and artifacts are made from cattle products, and their contact with
the outside world is sporadic. The English children tested live in a
town in a temperate climate, attend school, watch television, and
are surrounded by brightly colored, man-made objects. Despite
these differences and the difference in the number of terms even-
tually acquired, there were remarkable similarities between the two
populations.
The first aim of the study was to compare the universalist and
relativist explanations for the development of color categories.
Most Himba children receive no formal education, and for those
who do attend school, knowledge of color terms is not a curricu-
lum requirement. Despite such considerable environmental, lin-
guistic, and educational differences, universalist theories of color
categorization predict that both groups of children will have the
same initial cognitive organization of the color space (11 catego-
ries) and that this organization will remain in place for both groups
of children, despite the superimposition of differing sets of lin-
guistic categories (Bornstein, 1985; Rosch, 1973). We have pre-
sented three main arguments against the universalist position from
our new data.
The first argument concerns the errors made in recognition.
Previous research with adults also used memory errors as the main
data to argue for linguistic relativity (Roberson et al., 2000);
however, that procedure was criticized recently by Munnich and
Landau (2003). They argued that language-specific errors might
merely derive from the participant using a verbal code to remem-
ber colors. In other tasks, children as young as 16 months attend to
labels as well as to perceptual properties when encountering novel
objects and often overlook perceptual similarity in favor of judging
same-label items as most similar (Welder & Graham, 2001).
Therefore, language-based memory errors might not necessarily
contradict the universal account of color categories. However,
these arguments cannot apply to children who know no color
terms.
For children with no color term knowledge, the patterns of both
English and Himba memory errors appeared very similar, and it is
crucial that neither pattern resembled that derived from the 11
basic categories of English. Both appeared to be based on percep-
tual distance rather than a particular set of predetermined catego-
ries. Thus, the present data argue against an innate origin for the 11
BCTs in English.
However, a large proportion of the world’s major languages
have the same number of color categories, and one may ask why.
It is possible that the 11-color organization yields the optimal
combination of discriminability and cognitive economy for recog-
nition and representation of large numbers of colors. If so, lan-
guages with fewer terms would gain by introducing or borrowing
new terms when increasing technological advances or contact with
other cultures introduces a greater need to communicate more
precisely about color. Nevertheless, even if the 11-term organiza-
tion were found to be optimal and eventually adopted by all
cultures, it need not be innate. The present results suggest that
children do not have a universal set of predetermined categories
but rather gradually acquire the organization of categories that are
appropriate to their own language and culture.
The second argument from the present data against the univer-
salist position concerns the question of whether there is a predict-
able order in which color terms are acquired. Although the order of
acquisition observed over time differed according to the measure
Figure 10. Probability (Prob.) of listing known rather than unknown terms for both populations across tests.
T time.
568
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
used, no measure showed the pattern, predicted by universalist
theory, in which primary colors (red, blue, green, and yellow) were
learned first. Over the course of the longitudinal study, neither
population showed a predictable order of acquisition, although
English children generally acquired the terms brown and gray later
than other terms, which supports similar results found in previous
studies of English children (e.g., Macario, 1991; Mervis et al.,
1975; Pitchford & Mullen, 2001; Shatz et al., 1996). The present
study supports previous findings of the lack of a predictable order
of term acquisition in both languages.
The third argument against the universalist position concerns the
role of a particular set of focal colors. For both Himba and English
children who knew no color terms, there was no advantage in
memory for terms that were focal in either language. However, an
advantage for focal colors became evident as soon as children
acquired color terms. Of those children who knew at least one
color term at the first time of testing, English children showed
superior memory performance for the items that are focal in
English and in both sets of categories. Himba children showed
superior recognition for those items that are focal in Himba and in
both Himba and English categories. Such rapid divergence in the
cognitive organization of color for the two groups from the time
that the first terms are learned suggests that cognitive color cate-
gories are learned rather than innate.
The advantage present for focal colors at first testing increased
throughout the longitudinal study. Thus, the importance that Rosch
(1973) gave to focality in establishing categories seems justified
from the present data; nevertheless, it is important to stress that the
focality is not universal but, as shown both at first testing and
longitudinally, language dependent. For English children, this ef-
fect is unsurprising, as these are just the colors that are taught from
the earliest age and most readily available in their play things. For
Himba children, focality was determined on the basis of adult
naming agreement. Those targets deemed focal were those for
which over 90% of adults agreed on the name. Other targets
received little adult naming agreement. Himba children do not
encounter constant presentation, through printing, dyeing, and
film, of best example, highly saturated colors. In their environ-
ment, only muted, natural colors are encountered, for which adult
naming may often disagree. Children should then be faster to learn
those colors that adults reliably call by the same name— hence the
superiority for focal colors.
The second principal aim of the study was to examine the
trajectory of color term acquisition in both cultures. Previous
studies found conflicting evidence about the age at which children
reliably produce and use color terms appropriately (Andrick &
Tager-Flusberg, 1986; Backscheider & Shatz, 1993; Macario,
1991; Rice, 1980; Sandhofer & Smith, 1999), with estimates of the
age of acquisition varying between 2 and 6 years. The present
results suggest that children continue to refine their conceptual
color categories for some years after they first show evidence of
term knowledge for focal colors. Children know that a set of terms
refers to color and can select color as a property on which to match
objects as early as 2 years of age (Soja, 1994). In the present study,
3-year-olds in both cultures listed only color terms when asked.
However, some children from both language groups in this study
could not correctly apply all their BCTs (even though the English
children had had 3 years of specific instruction). Although the
expected discrepancy between naming and comprehension (Zelazo
et al., 1996; Zelazo & Reznick, 1991) was not observed in either
group, the acquisition of color terms did appear to be genuinely
slow, effortful, and error prone in both cultures.
It is not surprising that English children acquired their first color
words earlier than the Himba children. Greater exposure to colored
objects and the increased cultural salience of color in Western
society may contribute to an earlier conceptual understanding of
color as a separable dimension. However, from then on, the dif-
ferences between the groups are less marked than the similarities,
which are clearly seen in the children’s performance on the rec-
ognition memory tasks. We have already noted the similar impor-
tance that language-dependent focal colors have for both groups;
even more important are the similar effects of color term knowl-
edge. For both populations, once color terms are acquired, memory
performance is determined by the number of terms known. Chil-
dren make more correct identifications of focal items for terms that
they know than for terms that they do not know, regardless of the
absolute number of terms known. Thus, the effect of term knowl-
edge on memory cannot be an artifact of superior memory and
language skills of children with higher general intelligence; chil-
dren who know more terms get the same proportion of the items
they know correct as those who know few. Knowledge of even one
color term appears to change the cognitive organization of color,
and from this point on there are language-dependent differences
between the two groups. Once knowledge is acquired, it appears to
restructure the cognitive organization of color in a reliable way,
and this restructuring relates to term acquisition per se, not to
maturation or educational input.
Acquisition of term knowledge causes a reduction of memory
errors, and these change in nature over time. The effects of naming
are particularly evident in the case of two items that are called by
the same name in one language and by different names in another,
such as navy blue or dark orange. By the time children are 6 years
old, the few errors that are made to these tiles are to within- rather
than cross-category items, regardless of perceptual distance. It is
not simply the case that improving memory allows children to
make fewer and less distant errors.
In summary, the present study set out to examine when and how
children acquire a set of color categories appropriate to their own
language and culture (Roberson et al., 2000). The present results
suggest that what is universal about the acquisition of color vo-
cabularies is a gradual progression from an uncategorized organi-
zation of color based on perceptual similarity (where dimensions
are viewed as continua) to a structured organization of categories
that varies across languages and cultures. The increase in the
influence of linguistic categorization on memory for colors is
progressive and cumulative in both groups. Moreover, without
intensive adult input, color category acquisition is universally slow
and effortful.
References
Andrick, G. R., & Tager-Flusberg, H. (1986). The acquisition of colour
terms. Journal of Child Language, 13, 119 –134.
Backsheider, A. G., & Shatz, M. (1993). Children’s acquisition of the
lexical domain of colour. Proceedings of the Chicago Linguistic Society,
29, 11–22.
Berlin, B., & Kay, P. (1969). Basic color terms: Their universality and
evolution. Berkeley: University of California Press.
569
COLOR CATEGORY DEVELOPMENT
Bornstein, M. H. (1985). On the development of color naming in young
children: Data and theory. Brain and Language, 26, 72–93.
Bornstein, M. H., Kessen, W., & Weiskopf, S. (1976). Color vision and hue
categorization in young human infants. Journal of Experimental Psy-
chology: Human Perception and Performance, 2, 115–129.
Boroditsky, L. (2001). Does language shape thought? Mandarin and En-
glish speakers’ conceptions of time. Cognitive Psychology, 43, 1–22.
Bowerman, M., & Choi, S. (2001). Shaping meanings for language: Uni-
versal and language-specific in the acquisition of spatial semantic cate-
gories. In M. Bowerman & S. Levinson (Eds.), Language acquisition
and conceptual development (pp. 475–511). Cambridge, England: Cam-
bridge University Press.
Bowerman, M., & Choi, S. (2003). Space under construction: Language-
specific spatial categorization in first language acquisition. In D. Gent-
ner & S. Goldin-Meadow (Eds.), Language in mind: Advances in the
study of language and thought (pp. 387– 428). Cambridge, MA: MIT
Press.
Boyles, S. (2001). Children’s colour naming and a test of the linguistic
relativity hypothesis. Unpublished master’s thesis, University of Surrey,
Guildford, England.
Boynton, R. M. (1997). Insights gained from naming the OSA colors. In
C. L. Hardin (Ed.), Color categories in thought and language (pp.
135–150). Cambridge, England: Cambridge University Press.
Carey, S. (1978). The child as word learner. In J. Bresnan, G. Miller, & M.
Halle (Eds.), Linguistic theory and psychological reality (pp. 264 –293).
Cambridge, MA: MIT Press.
Cruse, D. A. (1977). A note on the learning of colour names. Journal of
Child Language, 4, 305–311.
Davidoff, J., Davies, I., & Roberson, D. (1999). Colour categories of a
stone-age tribe. Nature, 398, 203–204, 402, 604.
Davidoff, J., & Mitchell, P. (1993). The colour cognition of children.
Cognition, 48, 121–137.
Davidoff, J., & Roberson, D. (2004). Preserved thematic and impaired
taxonomic categorisation: A case study. Language & Cognitive Pro-
cesses, 19, 137–174.
Fletcher, R. (1980). The City University Colour Vision Test. Windsor,
England: Keeler.
Gellatly, A. (1995). Colourful Whorfian ideas: Linguistic and cultural
influences on the perception and cognition of colour, and on the inves-
tigation of them. Mind & Language, 10, 119 –225.
Gennari, S. P., Sloman, S., Malt, B., & Fitch, W. T. (2000). Motion events
in language and cognition. Cognition, 83, 49 –79.
Gottfried, G. M., & Tonks, J. M. (1996). Specifying the relation between
novel and known: Input affects the acquisition of novel color terms.
Child Development, 67, 850 866.
Guest, S., & Van Laar, D. L. (2000). The structure of colour naming space.
Vision Research, 40, 723–734.
Guest, S., & Van Laar, D. L. (2002). The effect of name category and
discriminability on the search characteristics of colour sets. Perception,
31, 445– 461.
Gumperz, J. J., & Levinson, S. C. (1997). Rethinking linguistic relativity.
Current Anthropology, 32, 613– 623.
Hammond, B. J. (1990). Testing time: Advances in logic analysis. IEE
Review, 36, 103–106.
Heider, E. R. (1971). “Focal” color areas and the development of color
names. Developmental Psychology, 4, 447– 455.
Heider, E. R. (1972). Universals in color naming and memory. Journal of
Experimental Psychology, 93, 10 –20.
Heider, E. R., & Olivier, D. C. (1972). The structure of the color space in
naming and memory for two languages. Cognitive Psychology, 3, 337–
354.
Jameson, K. A., & Alvarado, N. (2003a). Differences in color naming and
color salience in Vietnamese and English. Color Research & Applica-
tion, 28, 113–138.
Jameson, K. A., & Alvarado, N. (2003b). The relational correspondence
between category exemplars and naming. Philosophical Psychology, 16,
26 49.
Kay, P., Berlin, B., & Merrifield, W. (1991). Biocultural implications of
systems of color naming: Journal of Linguistic Anthropology, 1, 12–25.
Kay, P., & Kempton, W. (1984). What is the Sapir-Whorf hypothesis?
American Anthropologist, 86, 65–78.
Kay, P., & McDaniel, C. K. (1978). The linguistic significance of the
meanings of basic colour terms. Language, 54, 610 646.
Kruskal, J. B. (1964). Multidimensional scaling by optimizing goodness of
fit to a non-metric hypothesis. Psychometrika, 29, 1–27.
Legrende, P. (2000). Comparison of permutation methods for the partial
correlation and partial Mantel tests. Journal of Statistics and Computer
Simulation, 67, 37–73.
Levinson, S. C. (1996). Language and space. Annual Review of Anthro-
pology, 25, 353–382.
Lin, H., Luo, M. R., MacDonald, L. W., & Tarrant, A. W. S. (2001). A
cross-cultural colour-naming study. Part I: Using an unconstrained
method. Color Research and Application, 26, 40 60.
Lucy, J. A. (1992). Language and diversity of thought: A reformulation of
the linguistic relativity hypothesis. Cambridge, England: Cambridge
University Press.
Luzzatti, C., & Davidoff, J. (1994). Impaired retrieval of object-colour
knowledge with preserved colour naming. Neuropsychologia, 32, 933–
950.
Macario, J. F. (1991). Young children’s use of colour in classification:
Foods and canonically coloured objects. Cognitive Development, 6,
17– 46.
MacLaury, R. E. (1987). Color and cognition in Mesoamerica: Construct-
ing categories as vantages. Austin: University of Texas Press.
Malt, B. C., & Johnson, E. C. (1998). Artifact category membership and
the intentional-historical theory. Cognition, 66, 79 85.
Malt, B. C., Sloman, S. A., & Gennari, S. (2003). Speaking versus thinking
about objects and actions. In D. Gentner & S. Goldin-Meadow (Eds.),
Language in mind: Advances in the study of language and thought (pp.
81–112). Cambridge, MA: MIT Press.
Mantel, B. (1967). The detection of disease clustering and a generalized
regression approach. Cancer Research, 27, 209 –220.
Mervis, C. B., Bertrand, J., & Pani, J. R. (1995). Transaction of cognitive-
linguistic abilities and adult input: A case-study of the acquisition of
color terms and color-based subordinate object categories. British Jour-
nal of Developmental Psychology, 13, 285–302.
Mervis, C. B., Catlin, J., & Rosch, E. (1975). Development of the structure
of color categories. Developmental Psychology, 11, 54 60.
Miceli, G., Fouch, E., Capasso, R., Shelton, J. R., Tomaiuolo, F., &
Caramazza, A. (2001). The dissociation of color from form and function
knowledge. Nature Neuroscience, 4, 662– 667.
Mollon, J. D. (1999). Color vision: Opsins and options. Proceedings of the
National Academy of Sciences, USA, 96, 4743– 4745.
Munnich, E., & Landau, B. (2003). The effects of spatial language on
spatial representation: Setting some boundaries. In D. Gentner & S.
Goldin-Meadow (Eds.), Language in mind: Advances in the study of
language and thought (pp. 113–156). Cambridge: MA: MIT Press.
Pilling, M., Wiggett, A., O
¨
zgen, E., & Davies, I. R. L. (2003). Is color
categorical perception really perceptual? Memory & Cognition, 31,
538 –551.
Pitchford, N. J., & Mullen, K. T. (2001). Conceptualization of perceptual
attributes: A special case for color? Journal of Experimental Child
Psychology, 80, 289 –314.
Rice, M. (1980). Cognition to language: Categories, word meanings, and
training. Baltimore: University Park Press.
Roberson, D., & Davidoff, J. (2000). The categorical perception of colours
and facial expressions: The effect of verbal interference. Memory &
Cognition, 28, 977–986.
Roberson, D., Davidoff, J., & Shapiro, L. (2002). Squaring the circle: The
570
ROBERSON, DAVIDOFF, DAVIES, AND SHAPIRO
cultural relativity of good shape. Journal of Cognition and Culture, 2,
29 –53.
Roberson, D., Davies, I., & Davidoff, J. (2000). Colour categories are not
universal: Replications and new evidence from a Stone-Age culture.
Journal of Experimental Psychology: General, 129, 369 –398.
Rosch, E. H. (1973). Natural categories. Cognitive Psychology, 4, 328
350.
Sandhofer, C. M., & Smith, L. B. (1999). Learning color words involves a
system of mappings. Developmental Psychology, 35, 668 679.
Saunders, B. A. C., & van Brakel, J. (1997). Are there non-trivial con-
straints on colour categorization? Behavioral & Brain Sciences, 20,
167–178.
Sera, M. D., Berge, C., & del Castillo Pintado, J. (1994). Grammatical and
conceptual forces in the attribution of gender by English and Spanish
speakers. Cognitive Development, 9, 261–292.
Sera, M. D., Elieff, C., Forbes, J., Burch, M. C., Rodriguez, W., & Dubois,
D. P. (2002). When language affects cognition and when it does not: An
analysis of grammatical gender and classification. Journal of Experi-
mental Psychology: General, 131, 377–397.
Shatz, M., Behrend, D., Gelman, S. A., & Ebeling, K. S. (1996). Colour
term knowledge in two-year-olds: Evidence for early competence. Jour-
nal of Child Language, 23, 177–199.
Slobin, D. L. (1996). From “thought and language” to “thinking for
speaking.” In J. J. Gumperz & S. C. Levinson (Eds.), Rethinking lin-
guistic relativity (pp. 70 –96). Cambridge. England: Cambridge Univer-
sity Press.
Slobin, D. L. (2003). Language and thought online: Cognitive conse-
quences of linguistic relativity. In D. Gentner & S. Goldin-Meadow
(Eds.), Language in mind: Advances in the study of language and
thought (pp. 157–192). Cambridge, MA: MIT Press.
Soja, N. N. (1994). Young children’s concept of color and its relation to the
acquisition of color words. Child Development, 65, 918 –937.
Sturges, J., & Whitfield, T. W. A. (1997). Salient features of Munsell
colour space as a function of monolexemic naming and response laten-
cies. Vision Research, 37, 307–313.
Valberg, A. (2001). Unique hues: An old problem for a new generation.
Vision Research, 41, 1645–1657.
Webster, M. A., Miyahara, E., Malkoc, G., & Raker, V. E. (2000).
Variations in normal color vision. II. Unique hues. Journal of the Optical
Society of America A: Optics, Image Science and Vision, 17, 1545–1555.
Welder, A. N., & Graham, S. A. (2001). The influence of shape similarity
and shared labels on infants’ inductive inferences about nonobvious
object properties. Child Development, 72, 1653–1673.
Wyszecki, G., & Stiles, W. S. (1982). Color science— concepts and meth-
ods, quantitative data and formulae. New York: Wiley.
Young, F. W., & Harris, D. F. (1994). Multidimensional scaling. In M. J.
Norusis (Ed.), SPSS Professional Statistics 6.1 (pp. 155–222). Chicago:
SPSS.
Zelazo, P. D., Frye, D., & Rapus, T. (1996). An age-related dissociation
between knowing rules and using them. Cognitive Development, 11,
37– 63.
Zelazo, P. D., & Reznick, J. S. (1991). Age-related asynchrony of knowl-
edge and action. Child Development, 62, 719 –735.
Received November 3, 2003
Revision received August 3, 2004
Accepted August 3, 2004
Appendix
Designations of the 22 Color Aid Tiles in the Commission Internationale d’Eclairage
L*a*b* Metric
Tile L* a* b*
Focal
in
English
Focal in
Himba
Most common
English name
for nonfocal items
1 71.11 0.75 0.61 Gray x
410000White Vapa
6 44.12 6.06 7.77 Brown X
23 81.63 36.51 72.4 Orange X
28 39.6 25.81 21.39 Purple X
29 36.44 6.62 17.36 X Zoozu Navy blue
33 63.42 49.8 1.72 X X Dark pink
42 34.71 2.68 2.27 Black Zoozu
45 90.42 8.97 83.5 Yellow Dumbu
46 91.57 23.78 85.43 X X Yellow
47 75.21 61.19 59 X Serandu Dark orange
49 58.82 61.92 43.04 Red Serandu
50 49.31 0.92 48.11 Blue Burou
53 50.97 31.61 13.57 X Burou Turquoise
54 57.32 42.37 1.97 Green X
59 88.65 35.45 15.16 Pink X
77 69.6 51.72 35.63 X Serandu Orange/red
78 74.21 32.03 48.23 X X Yellow/green
79 70.05 53.54 22.62 X X Peach/beige
80 87.12 26.95 57.93 X X Pink/peach
81 83.23 39.45 53.08 X X Peach/mauve
82 56.7 60.68 12.55 X X Tan
Note. Stimuli were measured and viewed under 6,500 K illumination (simulating northern daylight), under
which they are calibrated. Although naturalistic viewing conditions vary slightly over time, this is likely to have
added noise to the data rather than any systematic confound.
571
COLOR CATEGORY DEVELOPMENT
... On the other hand, English speakers could not do so as their language only has one word for both shades 4 . Similarly, comparing English and Berinmo speaking subjects, English speakers distinguished shades of blue and green stimuli as different, whereas Berinmo language speakers, whose language has no distinct words for these two colours, failed to do so 5 . These studies show that linguistic features informed the formation of categories in the continuous spectrum of sensory information. ...
... The effects of verbal labels on perception have been well studied in the visual and auditory modalities 4,5,10,[24][25][26][27] . Information on how the perceptual systems are affected by linguistic cues in other modalities are not yet well studied. ...
Preprint
Full-text available
The influence of language on perceptual processes, referred to as the Whorfian hypothesis, has been a contentious issue. Cross-linguistic research and lab-based experiments have shown that verbal labels can facilitate perceptual and discriminatory processes, mostly in visual and auditory modalities. Here, we investigated whether verbal labels improve performance in a tactile texture discrimination task using natural textures. We also explored whether the grammatical category of these verbal labels plays a role in discrimination ability. In our experiments, we asked the participants to discriminate between pairs of textures presented to the fingertip after a five-day training phase. During the training phase, the tactile textures and English pseudowords were co-presented consistently in the congruent (experimental) condition and inconsistently in the incongruent (control) condition, allowing them to form implicit associations only in the former condition. The pseudoword verbal labels belonged to two grammatical categories, verb-like and noun-like. We found an improvement in the texture discrimination ability only for the congruent condition, irrespective of the grammatical category.
... On the other hand, English speakers could not do so as their language only has one word for both shades 4 . Similarly, comparing English and Berinmo speaking subjects, English speakers distinguished shades of blue and green stimuli as different, whereas Berinmo language speakers, whose language has no distinct words for these two colours, failed to do so 5 . These studies show that linguistic features informed the formation of categories in the continuous spectrum of sensory information. ...
... The effects of verbal labels on perception have been well studied in the visual and auditory modalities 4,5,10,[24][25][26][27] . Information on how the perceptual systems are affected by linguistic cues in other modalities are not yet well studied. ...
Preprint
Full-text available
The influence of language on perceptual processes, referred to as the Whorfian hypothesis, has been a contentious issue. Cross-linguistic research and lab-based experiments have shown that verbal labels can facilitate perceptual and discriminatory processes, mostly in visual and auditory modalities. Here, we investigated whether verbal labels improve performance in a tactile texture discrimination task using natural textures. We also explored whether the grammatical category of these verbal labels plays a role in discrimination ability. In our experiments, we asked the participants to discriminate between pairs of textures presented to the fingertip after a five-day training phase. During the training phase, the tactile textures and English pseudowords were co-presented consistently in the congruent (experimental) condition and inconsistently in the incongruent (control) condition, allowing them to form implicit associations only in the former condition. The pseudoword verbal labels belonged to two grammatical categories, verb-like and noun-like. We found an improvement in the texture discrimination ability only for the congruent condition, irrespective of the grammatical category.
... Rather than examining color emotion from a psychological perspective, these studies were more focused on the functions of light in tasks. However, color emotion are also important [35]. For example, yellow light is mainly used in the bedroom to evoke a feeling of warmth, whereas white light is primarily used in offices to provide a feeling of cleanness and efficiency. ...
Article
Full-text available
The social robotics field has been growing in prominence. Many developed countries have begun to apply such robots in various fields, such as shopping reception, education, home companion, medical care, and security. However, expectations of social robots have also changed. In addition to performing their specific tasks, social robots are also expected to consider users’ emotional needs. Therefore, scholars have begun to explore robots’ social cues in an effort to enhance their role in society and enable them to provide higher-quality services. This study used the robot's voice type, head-light color, and application field as independent variables to discuss the optimal robot social cues for different applications and to provide a reference for future production and research. Education, shopping reception, and home companion applications were selected as the most common areas employing social robots. According to the vocal fundamental frequencies used in related studies, three voice types were included: male, female, and child. The three values for head-light color, namely warm, neutral, and cold colors, were set according to color temperature theory. The robot social attribute scale, with the addition of an acceptance component, was used to evaluate respondent perceptions. Results revealed that male voices provide users with the highest impression of competence, whereas children's voices have the lowest competence. However, results for the warmth component were completely different: in this aspect, children's voices had the highest evaluations and male voices the lowest. For head-light colors, neutral colors had the highest overall acceptance, the highest competence evaluation, and the lowest discomfort. The neutral color's warmth judgment was the same as that of the warm color, and its competence rating was the same as that of the cold color. For home companion robots, we recommend a child’s voice with neutral colors as the first choice, and a child’s voice with warm color as the second option. Because females rejected male voices, male voices are not recommended in home companion. For education and business, a male voice and neutral colors are the first choice. A female voice is also the second option. In cases with less focus on warmth impression but greater emphasis on competence impression, cold colors could be employed.
... Alike, the location of the chips with the highest confidence for grine and for burou in our new data were boundary colours in our earlier data. So, there is no evidence that latent categories were responsible for augmenting the colour lexicon in the adult Himba nor indeed that they drive colour naming in the development of colour naming in Himba children (Roberson et al., 2004). ...
Article
Full-text available
Languages differ markedly in the number of colour terms in their lexicons. The Himba, for example, a remote culture in Namibia, were reported in 2005 to have only a 5-colour term language. We re-examined their colour naming using a novel computer-based method drawing colours from across the gamut rather than only from the saturated shell of colour space that is the norm in cross-cultural colour research. Measuring confidence in communication, the Himba now have seven terms, or more properly categories, that are independent of other colour terms. Thus, we report the first augmentation of major terms, namely green and brown, to a colour lexicon in any language. A critical examination of supervised and unsupervised machine-learning approaches across the two datasets collected at different periods shows that perceptual mechanisms can, at most, only to some extent explain colour category formation and that cultural factors, such as linguistic similarity are the critical driving force for augmenting colour terms and effective colour communication.
... Colour concept may have determined the emergence of human-specific colour categories, stemming from the generalisation around artificial colours. This idea could explain the cross cultural differences in colour (Roberson, Davidoff, Davies, & Shapiro, 2004;Roberson, Davies, & Davidoff, 2000), and the fact that the number of colour categories in industrialised societies is changing with time, and seem to be increasing (e.g. Lindsey & Brown, 2014). ...
Thesis
We examined the effects of brain lesions in humans on the interdependences between three modules of cortical colour processing, namely colour perception, naming and object-colour knowledge. We first focused on colour categorisation - a case-in-point of the interplay between perception and language. Reviewed evidence from cognitive development, comparative psychology and cognitive neuroscience hints that colour categorisation originates from neither perception nor language, as assumed by the Nature-Nurture debate. Instead, colour categories may reflect relevant objects in the environment. To assess the causal link between categorization and naming, we investigated a stroke patient, RDS. Despite severe difficulties in naming chromatic colours, due to a left occipito-temporal lesion, RDS’s colour categorisation was relatively spared. Multimodal MRI experiments revealed that the language-perception connectivity is essential for efficient colour naming but not for categorisation. Investigation of object-colour knowledge in the context of RDS’s colour-naming impairment showed that RDS could not link colour perception to neither language nor semantic knowledge. He could not associate a visual colour to a colour name or to the shape of its typical object. Overall, we demonstrated three functional segregations in colour processing: between (1) colour categorisation and colour naming, (2) naming of chromatic and achromatic colours and (3) knowing about coloured objects and knowing about abstract colours. The main purpose of high-level cortical colour mechanisms could be providing sensory and semantic information to guide object-related behaviour, by achieving (1) stable colour perception, (2) relevant colour categories, and (3) joint mental representations of shapes and colours. These neural computations may have been recycled in cultural evolution to isolate colours from objects and label them with names.
Thesis
Les expériences de notre nuit sont souvent décrites comme des îlots d'activité mentale, internement générées dans un océan d'inconscience. En sous-texte de cette vision se cachent deux pré-supposés que le sommeil lent est un modèle d'inconscience et que le traitement sensoriel du monde extérieur en sommeil paradoxal ne peut être qu'inconscient. Dans cette thèse, nous avons voulu tester ces pré-supposés avec une approche empruntant à trois littératures complémentaires : celle de la conscience, celle du sommeil sain et pathologique et celle de la philosophie de l'esprit. Dans une première étude nous avons mis en évidence l'existence de "blackout' de nuit : une absence total de rappel de contenu du couche au lever dans l'hypersomnie Idiopathique. Nous pensons que notre démonstration de l'existence du phénomène de blackout est intéressante car elle permet, par contraste, de mettre en évidence l'existence d'une expérience minimale de la nuit, comme les philosophes l'avaient suggéré. Dans deux autres études nous avons montré la capacité de patients narcoleptiques (lucides ou non) à traiter l'extérieur pendant des siestes en utilisant comme réponses les muscles de leurs visages. Cela suggère qu'un traitement conscient dans le sommeil peut avoir lieu en sommeil paradoxal chez ces patients. L'ensemble de ce travail de thèse invite à penser que l'idée selon laquelle on perd conscience pendant que l'on dort serait à réévaluer. En effet, une réelle perte de conscience dans le sommeil, si elle existe, pourrait être plutôt transitoire et négligeable face à la fabuleuse pluralité des processus qui se déroulent en son sein.
This study presents an analysis of the incommensurability about the representations or models elaborated by children from an Indigenous community within three areas or cultural domains, namely, the ethnic, daily (domestic), and school domains and their implications in relation to science education. The children belong to an Indigenous Nahuatl community which is located in the Sierra Norte of Puebla, México. The analysis is based on the children’s model constructions concerning two scientific topics: (1) the mixture of colours and (2) the Earth’s shape and day-night cycle. Both topics belong to the school program. The children’s models are the result of previous investigations that are analysed within this work from an epistemological approach where non-translatability criteria are used to determine the incommensurability of the corresponding models to the different cultural areas. The results show that the models in each area are effectively incommensurable and are therefore coexistent without any integration or displacement process or the presence of conflict among the three cultural areas on a daily life basis. A brief discussion is showed on the implications of results on the difficulties of crossing border processes and the potential conceptual conflicts in young students.
Article
Статья продолжает серию исследований, рассматривающих проблему категориального цветовосприятия на различных участках зрительного поля для пограничных и фокальных цветов. Известно, что центральные поля зрения преимущественно принимают участие в процессах идентификации и категоризации объектов, а зрительная периферия отвечает за процесс детекции и пространственную локализацию поступающей зрительной информации. В работе основное внимание уделено возрастным закономерностям развития категориальных эффектов цветовосприятия.
Book
An overview is given of cross-cultural psychology and cultural psychology, focusing on theory and methodology. In Section 1 historical developments in research are traced; it is found that initially extensive psychological differences tend to shrink when more carefully designed studies are conducted. Section 2 addresses the conceptualization of “culture” and of “a culture”. For psychological research the notion “culture” is considered too vague; more focal explanatory concepts are required. Section 3 describes methodological issues, taking the notion of the empirical cycle as a lead for both qualitative and quantitative research. Pitfalls in research design and data analysis of behavior-comparative studies, and the need for replication are discussed. Section 4 suggests to move beyond research on causal relationships and to incorporate additional questions, addressing the function and the development of behavior patterns in ontogenetic, phylogenetic and historical time. Section 5 emphasizes the need for applied research serving the global village.
Article
Vocal signalling systems, as used by humans and various non-human animals, exhibit discrete and continuous properties that can naturally be used to express discrete and continuous information, such as distinct words to denote objects in the world and prosodic features to convey the emotions of the speaker. However, continuous aspects are not always expressed with the continuous properties of an utterance but are frequently categorised into discrete symbols. While the existence of symbols in communication is self-evident, the emergence of discretisation from a continuous space is not well understood. In this paper, we investigate the emergence of discrete symbols in regions with a continuous semantics by simulating the learning process of two agents that acquire a shared signalling system. The task is formalised as a reinforcement learning problem with a continuous form and meaning space. We identify two causes for the emergence of discretisation that do not originate in discrete semantics: 1) premature convergence to sub-optimal signalling conventions and 2) topological mismatch between the continuous form space and the continuous semantic space. The insights presented in this paper shed light on the origins of discrete symbols, whose existence is assumed by a large body of research concerned with the emergence of syntactic structures and meaning in language.
Article
This review describes some recent, unexpected findings concerning variation in spatial language across cultures, and places them in the context of the general anthropology of space on the one hand, and theories of spatial cognition in the cognitive sciences on the other. There has been much concern with the symbolism of space in anthropological writings, but little on concepts of space in practical activities. This neglect of everyday spatial notions may be due to unwitting ethnocentrism, the assumption in Western thinking generally that notions of space are universally of a single kind. Recent work shows that systems of spatial reckoning and description can in fact be quite divergent across cultures, linguistic differences correlating with distinct cognitive tendencies. This unexpected cultural variation raises interesting questions concerning the relation between cultural and linguistic concepts and the biological foundations of cognition. It argues for more sophisticated models relating culture and cognition than we currently have available.
Article
This paperback reprint of a classic book deals with all phases of light, color, and color vision, providing comprehensive data, formulas, concepts, and procedures needed in basic and applied research in color vision, colorimetry, and photometry.
Article
Rethinking Linguistic Relativity. John J. Gumperz and Stephen C. Levinson. eds. Studies in the Social and Cultural Foundations of Language, 17. Cambridge: Cambridge University Press, 1996. 488 pp.