Shared understanding of color among congenitally
blind and sighted adults
Authors: Judy Sein Kim1*, Brianna Aheimer1, Verónica Montané Manrara1, Marina Bedny
Affiliations: 1Department of Psychological and Brain Sciences, Johns Hopkins University
*Correspondence to: email@example.com
Empiricist philosophers such as Locke famously argued that people born blind could only
acquire shallow, fragmented facts about color. Contrary to this intuition, we report that blind and
sighted people share an in-depth understanding of color, despite disagreeing about arbitrary
color facts. Relative to the sighted, blind individuals are less likely to generate ‘yellow’ for
banana and ‘red’ for stop-sign. However, blind and sighted adults are equally likely to infer that
two bananas (natural kinds) and two stop-signs (artifacts with functional colors) are more likely
to have the same color than two cars (artifacts with non-functional colors), make similar
inferences about novel objects’ colors, and provide similar causal explanations. We argue that
people develop inferentially-rich and intuitive “theories” of color regardless of visual experience.
Linguistic communication is more effective at aligning people’s theories than their knowledge of
Humans acquire knowledge through a variety of means—through direct sensory
experience, communication with others, and by thinking (1,2). The question of where knowledge
comes from is at the heart of many cognitive theories, but disentangling the relative contributions
of these sources is challenging (e.g., 3-8). Color knowledge in blindness provides a wedge into
this puzzle by isolating the contribution of first-person sensory experience (9-11). British
Empiricists like Locke and Hume argued that color ideas are inaccessible to people born blind
since sensory experience is the foundation of knowledge. More recently, Frank Jackson’s
influential thought experiment featuring Mary, a color scientist who lives in a black-and-white
room, inspired many debates about knowledge gained from first-person experience (e.g., 11-15).
Only a handful of empirical studies have examined what people born blind actually know about
color. Landau & Gleitman (16) showed that Kelli, a congenitally blind 4-year-old, applied color
words to concrete objects but not mental entities (e.g., ideas) and understood that color could
only be perceived visually, unlike texture or size. Blind and sighted adults share knowledge of
similarities between colors (e.g. green and blue are similar but different from orange and red) (17-
19). Several studies have also identified differences in blind and sighted people's color
knowledge. A recent study found that unlike sighted adults, adults born blind are unlikely to agree
on the colors of common animals (20). Even for objects for which blind and sighted people
generate the same color labels, blind individuals are less likely to use color as a dimension during
semantic similarity judgments (21). One interpretation of this evidence is that color knowledge
acquired through verbal communication consists largely of associative verbal facts and is thus
inferentially shallow (21-23). An individual born blind might know that the word ‘yellow’ is used to
describe ‘bananas’ without having in-depth understanding of color.
An alternative possibility is that blind individuals share with those who are sighted
inferentially-rich understanding of color. In fact, it is possible that in the absence of first-person
sensory experience of color, inferentially-rich knowledge is more preserved than knowledge of
associative color facts. Sighted children’s inferentially-rich, causal understanding of color enables
them to make inferences about real and novel object colors. Children expect an object’s
relationship with color to differ depending on whether it is a natural kind (e.g. animal, plant, gem)
or an artifact (e.g. machine, tool). When asked, “Could something still be a Glick even if it was a
different color?” 5 year-old children are more likely to say yes for an artifact than for an animal
(24). Such intuitions about color are part of broader frameworks, often referred to as ‘intuitive
theories’, about physical objects (e.g., 25, 26). Children view natural kinds as having intrinsic
essences emanating from nature, but link properties of artifacts to human intentions (e.g., 27-29).
In response to “Why is this object yellow?” children prefer explanations that appeal to biological
mechanisms for natural kinds but human intentions for artifacts (30). Such knowledge about color
can be distinguished from that of other object properties: for example, while different instances of
a natural kind (e.g., bananas) are more likely to have the same color than instances of an artifact
(e.g., cars), natural kinds and artifacts are similarly likely to have consistent shapes (24). The role
of first-person sensory experience in acquiring such causal-explanatory and inferentially-rich color
knowledge is not known.
In the current study, we probed sighted and congenitally blind people’s associative and
causal-explanatory knowledge of color. Experiment 1 first queried associative memory for real
objects’ colors by asking participants to generate “a common color of X” (Figure 1). We next asked
participants to judge how likely two instances of the same object are to have the same color, for
natural kinds (e.g. two bananas) and artifacts (e.g. two cars). We reasoned that if people share
intuitive theories about the relationship between color and object kind, blind and sighted people
would make similar inferences about color consistency, even while disagreeing on associative
facts (i.e., the particular colors of objects). For example, blind individuals might share with the
sighted the intuition that polar bears but not cars have a consistent color, despite showing low
agreement that polar bears are white. Alternatively, the experience of seeing that most polar bears
are white but that cars come in many different colors might be required to learn the different
importance of color for natural kinds and artifacts.
To test nuanced intuitions about the causal mechanisms that lead artifacts to have the
colors they do, we included a third object category: artifacts with functional colors (e.g. stop-sign,
paper, coin). Artifacts vary according to how much and in what way color relates to their function.
For some (e.g., mugs), color is not related to function (holding liquid), while for others (e.g., stop
signs) it is integral (e.g. stop signs are consistently red for visibility and recognizability). If sighted
and blind people appreciate how color relates to artifact function, they should judge stop signs to
have more consistent colors than mugs.
The ability to support generalization to novel instances is a key test of whether knowledge
is inferentially-rich (e.g., 28, 31). In Experiment 2, we thus asked participants to make inferences
about color consistency for novel objects (natural kinds, artifacts with function-relevant color, and
artifacts with function-irrelevant color) in an imaginary island scenario (Fig. 1). If abstract
knowledge about the origins and causes of color is shared, then blind and sighted participants
should be able to make systematic judgments about color consistency on the basis of object
category (e.g., kind of fruit, creature, gem, or household item, gadget, coin) alone. Finally, in
Experiment 3, we elicited open-ended explanations for why objects have their colors (e.g., “Why
is a carrot orange?”). This allowed us to probe the specific nature of blind and sighted people’s
knowledge of the causal mechanisms that give rise to object colors.
Fig1. Experimental conditions and trials for color consistency inference. Participants were asked
about color and usage consistency for real (Experiment 1) and novel (Experiment 2) objects. In both
experiments, color trials asked about natural kinds, artifacts with non-functional colors, and artifacts with
functional colors, while usage trials asked about natural kinds and artifacts. Different items were used in
every trial. For Experiment 1, all items used are listed, and for Experiment 2, one sample trial (an Appendix
with full list of trials can be found in Supplementary Materials).
Knowledge of specific object colors among sighted and blind participants
Blind and sighted participants were asked to name a common color of 54 real objects
(Experiment 1, 30; Experiment 3, 24, collapsed for the current analysis) (Figure 2A). Objects were
chosen from three larger types: natural kinds (NK) (e.g. lemons), artifacts with non-functional
colors (A-NFC) (e.g. cars) and artifacts with functional colors (A-FC) (e.g. stop-signs). Color
naming agreement was quantified for each object using Simpson’s Diversity Index (SDI) (32), and
log-transformed SDIs were modeled using linear mixed effects regression (see Methods for
details). For both sighted and blind groups, color naming agreement was higher for natural kinds
(e.g. lemon) than for artifacts with non-functional colors (e.g. car), but similar to artifacts with
functional colors (e.g. stop signs) (Figure 2B; SDI for sighted NK: M=0.86, SD=0.2; A-NFC:
M=0.49, SD=0.29; A-FC: M=0.74, SD=0.29; blind NK: M=0.5, SD=0.25; A-NFC: M=0.24,
SD=0.09; A-FC: M=0.48, SD=0.26). Naming agreement was substantially higher for sighted
compared to blind participants across all object types, and there was no group-by-object kind
interaction (result of regression, effect of group: , p<0.0001,
; effect of
object type: , p<0.0001,
; object-type-by-group interaction: ,
Fig 2. Object color naming agreement. Blind and sighted participants were asked to name common
colors of real objects (Experiments 1 and 3). (A) Stacked bars show the frequency of the 8 most frequent
colors provided for each object. Frequency for each unique color word is shown as a proportion of all words
provided for an object. (B) Bar graph showing naming agreement (Simpson’s Diversity Index calculated for
individual objects). Mean +/- SEM (across objects). (C) Correlation of ratings from Amazon Mechanical
Turk participants (n=20) for artifact colors’ relevance to function with blind and sighted participants’ color
Color consistency inferences in blind and sighted individuals: real objects
Sighted and blind participants judged the likelihood that two objects (e.g. two lemons),
randomly chosen from the same object category, would have the same color for 10 natural kinds
(NK e.g. lemon), 10 artifacts with non-functional colors (A-NFC e.g. car) and 10 artifacts with
functional colors (A-FC e.g. stop sign) (henceforth color consistency judgment). Participants rated
consistency likelihood on a scale of 1 to 7 (1: not likely, 7: very likely). As a control, participants
judged the likelihood that two people chosen at random would do the same thing with an object
(e.g. a leaf vs. a car) (henceforth usage consistency judgment). Usage consistency was tested
for 10 natural kinds (NK) and 10 artifacts.
Fig 3. Inferences about color and usage consistency across instances of an object. Consistency
judgments for real (Experiment 1) and novel (Experiment 2) objects. Bars are mean +/- SEM.
Sighted participants judged natural kinds (e.g. lemons) to have lower usage consistency
but higher color consistency, relative to artifacts (with non-functional colors, e.g. cars) (Fig. 3;
sighted usage NK: M=3.67, SD=1.59; usage A: M=5.66, SD=1.49, Wilcoxon matched-pairs
signed rank test for usage NK vs. A, two-tailed: z=-3.82, p=0.0001, r=0.88; color NK: M=6.2,
SD=1.12, color A-NFC: M=3.34; SD=1.72; color NK vs. A-NFC, z=3.78, p=0.0002, r=0.87).
Sighted participants’ color consistency ratings for artifacts with functional colors (e.g. stop-signs)
were higher than those for artifacts with non-functional colors and lower than those of natural
kinds (color A-FC: M=5.17, SD=1.81, comparing A-FC vs. A-NFC: z=3.8, p=0.0002, r=0.87; A-FC
vs. NK: z=-3.66, p=0.0003, r=0.84). For all artifacts, we obtained ratings of an object color’s
relevance to the its function from a separate group of sighted Amazon Mechanical Turk
participants. These function relevance judgments for artifacts were positively correlated with
sighted participants’ color consistency judgments (Spearman’s rank correlation: rho=0.61,
p<0.0001; Fig. 2C).
The same effect of object type on color and usage consistency judgments was observed
in the blind group. Blind participants again judged natural kinds to have lower usage consistency
but higher color consistency, compared to artifacts (blind usage NK: M=3.87, SD=1.61; A:
M=6.19, SD=1.15; Wilcoxon matched-pairs test for usage NK vs. A: z=-3.92, p<0.0001, r=0.88;
color NK: M=5.66, SD=1.52; A-NFC: M=3.32, SD=1.47; NK vs. A-NFC: z=3.92, p<0.0001,
r=0.88). Artifacts with functional colors were judged to have higher color consistency than artifacts
with non-functional colors, but lower than natural kinds (color A-FC: M=5.38, SD=1.81; comparing
A-FC vs. A-NFC: z=3.92, p<0.0001, r=0.88; A-FC vs. NK: z=-1.98, p=0.049, r=0.44). Blind
participants’ consistency judgments for artifacts were positively also correlated with MTurk
participants’ ratings of color’s relevance to object function (Spearman’s rho=0.61, p<0.0001; Fig.
When groups were compared directly to each other, object kind and trial type did not
interact with group (mixed ordinal logistic regression, group (blind vs. sighted) x trial type (color
vs. usage) x object kind (NK vs. A-NFC), with sighted group, usage trial, and A-NFC treatment
coded as baselines, no three-way interaction (=-0.24, SE=0.38, z=-0.63, p=0.53). We also
analyzed color judgments separately (group (blind vs. sighted) x object kind (NK vs. A-NFC vs.
A-FC), with sighted and A-NFC as baseline). There was no significant interaction between group
and object kind when comparing artifacts with functional color to artifacts with non-functional color
(=0.45, SE=0.27, z=1.65, p=0.099), although the interaction was significant when comparing
natural kinds to artifacts with non-functional color (=-1.02, SE=0.27, z=-3.78, p=0.0002).
Color consistency inferences in blind and sighted individuals: Novel objects
For real familiar objects, blind and sighted individuals could have made color consistency
judgments based on knowledge of their actual color frequencies (e.g., learned from seeing or
hearing that bananas are often yellow but that cars can be red, blue, black, etc.). Alternatively,
people might have a more general understanding of the relationship between object kind (e.g.
natural kind vs. artifact) and color. To distinguish between these possibilities, we collected color
consistency judgments for novel objects, for which neither blind nor sighted participants could
have directly experienced their color. Participants were presented with “explorer on an island”
scenario and judged the consistency of color and usage for novel natural kinds (e.g. gem, plant)
(5 objects), novel artifacts with non-function-relevant colors (e.g. cleaning-gadget, speaking
device) (5 objects), and novel artifacts with function-relevant colors (e.g. coin, ceremonial
clothing) (5 objects).
As with real objects, both groups judged artifacts to be more likely to have consistent
usage than natural kinds (sighted usage NK: M=2.76, SD=1.42; A: M=5.64, SD=1.38; Wilcoxon
matched-pairs signed rank test for NK vs. A: z=-3.82, p=0.0001, r=0.88; blind usage NK: M=3.25,
SD=1.73; A: M=5.77, SD=1.58; NK vs. A: z=-3.92, p<0.0001, r=0.88). For color trials, consistency
was again judged to be higher for natural kinds than for artifacts with non-functional color by both
groups (sighted color NK: M=5.67, SD=1.12; A-NFC: M=3.73, SD=1.5; NK vs. A-NFC: z=3.81,
p=0.0001, r=0.84; blind color NK: M=5.74, SD=1.42; A-NFC: M=3.95, SD=1.74; NK vs. A-NFC, z
= 3.72, p<0.0001, r=0.85). For both groups, artifacts with functional colors were judged as likely
to have consistent colors as the natural kinds but more likely compared to artifacts with non-
functional colors (for sighted color A-FC: M=5.55, SD=1.43; NK vs. A-FC: z=-0.65, p=0.52; A-
NFC vs. A-FC: z=3.78, p=0.0002, r=0.87; for blind color A-FC: M=5.92, SD=1.50; NK vs. A-FC: z
=1.03, p=0.3; A-NFC vs. A-FC: z=3.9, p<0.0001, r=0.87).
The interaction between group, question type, and object kind was non-significant (mixed
ordinal logistic regression, three-way interaction: =-0.15, SE=0.93, z=-0.16, p=0.88). The group-
by-condition interaction for color trials only were also not significant (for NK vs. A-NFC: =-0.07,
SE=0.36, z=-0.18, p=0.86, for A-FC vs. A-NFC: =0.52, SE=0.39, z=1.39, p=0.17). In sum, blind
and sighted people have similar understanding of the relationship between object kind and color.
Blind and sighted people’s causal explanations of object color (Experiment 3)
In Experiment 3, blind and sighted participants were asked to explain why each object has
its particular color. The explanations were coded according to what type of information they
appealed to: process, depends on…, just is that way, material, social, maker of the object, visibility
and cultural convention (see Supplementary Materials for coding details). Both blind and sighted
participants provided rich and coherent explanations of the cause of object color (Figure 4). Both
groups tended to provide different explanations for natural kinds, artifacts with non-functional
colors, and artifacts with functional colors. For natural kinds, both groups most often said “it just
is that way” (sighted: 33%, blind: 36%) or appealed to a process that give the object its color
(sighted: 32%, blind: 31%). For example, participants often described how the process of
photosynthesis makes plants green. By contrast for artifacts with non-functional colors (e.g. cars)
both blind and sighted participants appealed to people’s social and esthetic preferences (sighted:
64%, blind: 44%), and referred to the material of which the object was made (sighted: 18%, blind:
13%). For example, people frequently stated “personal preference” as a cause for cars, and for
cup, mentioned that they could be different colors depending on whether they are made of plastic,
porcelain, or metal. For artifacts with functional colors, participants most often appealed to cultural
convention (sighted: 57%, blind: 51%) and visibility (sighted: 24%, blind: 23%). For example, for
school bus, participants frequently mentioned tradition and history, and for stop sign, that the color
makes it easy to see.
We examined how similar explanations were across groups by computing Spearman’s
correlation across groups within object kind. The frequencies of explanations by type were highly
correlated across groups for all three kinds of objects (natural kind: rho=0.99, p<0.0001; artifacts with
non-functional color: rho=0.72, p=0.03; artifacts with functional color: rho=0.97, p<0.0001).
Correlations across object kinds within each group were comparatively much lower (within sighted
group: natural vs. A-NFC: rho=-0.31, p=0.4; natural vs. A-FC: rho=-0.27, p=0.5; A-NFC vs. A-FC:
rho=0.78, p=0.01; within blind group: natural vs. A-NFC: rho=-0.02, p=1; natural vs. A-FC: rho=-
0.37, p=0.3; A-NFC vs. A-FC: rho=0.28, p=0.5).
Fig 4. Explanations about object color. Explanation types were coded by 5 different coders
who were blind to group and object. Stacked bar shows the frequency of each explanation type
as a proportion of all explanations provided for an object (within object type) across participants
(within a group). A detailed key of explanation types can be found in Supplementary Materials.
A straightforward idea is that we acquire color knowledge through seeing. Consistent with
this, we find that people who have never seen are less likely to agree with each other and with
sighted people about associative color facts: although 100% of blind participants generate the
label ‘white’ for snow, only 50% say ‘yellow’ for bananas (see also 20, 21). This observation
suggests that when it comes to learning associative color facts, direct visual access is more
effective than linguistic communication.
By contrast, we find that inferentially rich color knowledge is shared among blind and
sighted individuals—blind and sighted participants alike judge that two instances of a natural kind
(e.g. two bananas or two gems) are more likely to have the same color than two instances of an
artifact (e.g. two cars or two mugs). Blind and sighted people also provide similar explanations of
why real objects have the colors that they do, and these explanations vary systematically across
natural kinds and artifacts. For natural kinds, both blind and sighted appeal to an objects’ intrinsic
nature (e.g., “that’s just how it is”, “that’s nature”) or describe processes such as photosynthesis,
growth, or evolution. For artifacts, participants consistently cite individual or groups of people’s
needs and intentions (e.g., culture, aesthetic preference, visibility). Blind individuals produce
coherent explanations for object color even when they do not agree with the sighted about the
typical color of that particular object type. For example, while both groups’ explanations for the
color of polar bears mention their arctic habitat, almost all sighted participants explain that their
white fur allows camouflage in the snow while some blind participants explain that they are black
to absorb heat in the cold. (Interestingly, polar bears indeed have black skin underneath their
transparent fur, and these features are thought to have evolved for both camouflage and heat
absorption) (33). Such cases provide an illustration of causal understanding of color that is
independent of knowing object-color associations.
Blind and sighted people’s intuitions about the relationship between kind and color go
beyond the natural kind/artifact distinction (34). Among artifacts, ratings of how important color is
to an artifact’s function are highly correlated with blind and sighted participants’ ratings of color
consistency. Explanations produced by sighted and blind adults also vary systematically by
artifact type. For household and personal items such as mugs and cars, participants appeal to
aesthetic preferences. For institution-related objects like police uniforms and dollar bills,
participants cite social need for recognition. For stop signs, participants appeal to visibility (e.g.,
“red because red jumps out and warns people to stop”). Across artifacts, sighted and blind alike
appeal to a range of causes such as camouflage, recognizability, cultural convention, symbolism,
history, and aesthetic preference.
Finally, sighted and blind people make similar color consistency inferences for novel
objects with which neither group has visual or linguistic experience. For example, both blind and
sighted participants judge that two instances of a novel gem (natural kind) would be more likely
to have the same color than two instances of a novel household gadget (artifact). Blind and
sighted people also make distinctions within novel artifacts, intuiting which are most likely to have
functionally relevant and therefore consistent colors (e.g. coins, toxic waste containers). Together,
this evidence suggests that people living in the same culture, regardless of their visual experience,
develop similar intuitive theories of color and use these theories to make inferences that go
beyond the data.
While the present evidence suggests that blind and sighted people alike have a coherent
and causal understanding of color, this understanding is likely to differ in substantial ways from
formal scientific color theories (35, 36, 12). Participants’ explanations of object colors did
sometimes cite scientifically studied processes (e.g., photosynthesis), but more commonly
consisted of informal justifications lacking mechanistic detail (e.g., “that’s just how it grows”, “it’s
nature”, “God made it that way”, “manufacturer decided to paint it that way”, “the material it’s made
of”). When more specific causes and processes are mentioned, they are often social and
historical, and unlikely to be taught through formal education (e.g., both blind and sighted
participants mentioned personality of the owner for cars and “the patriarchy” for the color of
wedding dresses). During development, sighted children’s beliefs about color depart
systematically from scientific knowledge. Children mistakenly believe that an object will continue
to have the same color even when the lighting source is changed, that objects emit their own
shadows, and that a green object will have a green shadow (37-39; see also 40). Children's
explanations about such phenomena omit crucial components, such as the source and nature of
light illuminating an object (37). Similar inconsistencies between scientific and intuitive theories
have been observed in numerous other knowledge domains (e.g., physics: 41, 1983; biology: 42;
psychology: 43). Even when educated adults and experts report strong confidence in their own
understanding, their explanations for how things work are coarse and incomplete (44). Future
work is needed to understand the ways in which intuitive theories of color among sighted and
blind people share features with and depart from scientific color theories.
Open questions remain about how blind and sighted people acquire causal intuitions about
color. Linguistic communication likely plays a crucial role. From a young age, children use
testimony as well as more implicit linguistic cues (e.g., labeling) to inform intuitive theories of
physical, biological, and mental phenomena (45, 1, 46). For many previously studied domains of
knowledge, language-induced learning could in principle piggyback on pre-existing structured
knowledge built through sensory observation. For example, learning that the earth is round might
piggyback on learning roundness through vision and touch (47). Even in the case of mental
phenomena, simulation of one’s own feelings and thoughts has been offered as a source of “first-
person” information about others’ minds (48, 49). Analogously, a sighted person might construct
a representation of a novel animal described as blue and large by referencing physical knowledge
previously built up through sensory experience of color and size (20). In the case of color
knowledge among blind individuals, there is no directly pertinent sensory information.
Nevertheless, inferentially rich knowledge is constructed through linguistic communication from
the ground up.
Recent text corpus analyses also find that language is a rich source of semantic
information, including that of physical appearance. Associative algorithms are able to extract
semantic information using word co-occurrence and word neighborhood statistics (e.g., 50-52,
22-23). The available evidence suggests, however, that people’s learning of appearance from
language differs in important respects from the statistical tracking used by current text analysis
algorithms. For example, in the case of animal appearance, people born blind know more about
shape, texture, and size, than what is extracted by co-occurrence tracking algorithms (22, 53).
People born blind appear to use taxonomy and habitat to infer physical characteristics rather than
relying purely on explicit statements about appearance (20). In the case of object color
consistency, it is not clear how tracking word co-occurrence alone would provide the correct
information. Analyses of object and color label co-occurrence suggest that object-color label co-
occurrence consistency does not line up with real-life object-color consistency, nor with sighted
people’s intuitions about the object’s actual color (23). For example, ‘crow’ co-occurs with ‘black’
and ‘white’ with similar frequencies. Nevertheless, we find that blind and sighted people’s
intuitions about color consistency are similar. Moreover, blind participants generate canonical
colors (i.e. black for crow) more often than non-canonical ones (i.e. white for crow), even when
the canonical and non-canonical colors are equally likely to co-occur with the object in text (23,
One source of information blind individuals could use to arrive at color consistency is
generic language (e.g. tomatoes are red). Generics provide evidence that a property is pervasive
to an object, as opposed to specific to a particular instance of that object (e.g., 26, 55-57). For
example, hearing “this car is red,” as opposed to “tomatoes are red” and “stop signs are red” could
provide evidence with respect to which objects tend to have a consistent color. Generic language
could also facilitate learning which of two possible color labels is the canonical one (e.g. this crow
is white vs. crows are black). Generic language alone cannot, however, explain how blind and
sighted people make similar judgments about novel objects, which they have not heard being
described in a generic sentence, or how people born blind generate coherent explanations of
color. Linguistic cues such as generic language, must make contact with early-emerging intuitive
theories and the inferential machinery to transform these theories in light of the linguistic evidence
(57-59). We hypothesize that people born blind use linguistic cues, such as generics, to fill in the
color-specific elements of intuitive theories about objects and their physical properties. For
example, hearing people talk about “favorite colors”, together with evidence that a particular
personal item (e.g. cars) varies in color, might lead one to conclude that personal items are
sometimes colored according to the preferences of the owner. Future work is needed to uncover
precisely how blind and sighted people use language as a source of information when
constructing intuitions about color. One important direction for future research is testing the
acquisition of such knowledge by blind and sighted children. Such studies would reveal exactly
when, how, and with what information such intuitions are constructed.
In summary, we find that blind and sighted individuals alike possess theory-like,
inferentially rich knowledge about the relationship between objects and their colors. These
intuitive theories of color support consistent generalizations in the face of limited information (e.g.,
for novel objects), invoke deep causes (e.g., object function), support the generation of
sophisticated explanations, apply to broad categories (e.g., all plants) as well as to specific
instances (e.g., polar bears), and are specific to color. Interestingly, such structured and
inferentially rich color knowledge appears to be more resilient to the lack of first-person sensory
experience than knowledge of associative color facts. This observation directly contradicts the
common intuition that blind people’s knowledge of color consists of meaningless arbitrary facts.
Language appears to support the updating of causal models much more robustly than it does the
acquisition of arbitrary facts.
Twenty congenitally blind (14F/6M, age: M=30.85, SD=10.59, years of education: M=15.4,
SD=2.23) and nineteen sighted (14F/5M, age: M=31.21, SD=11.21, years of education M=15.79,
SD=1.82) participants took part in the study (participant table can be found in Supplemental
Materials, Table S1). All blind participants reported no experience with color, shape, or motion,
and had at most minimal light perception. All blind participants were tested at the 2018 National
Federation of the Blind Convention in Orlando, Florida. Subtests of the Woodcock Johnson III
Tests of Achievements (Word ID, Word Attack, Synonyms, Antonyms, and Analogies) were
administered to sighted and blind participants, and anyone scoring below two SDs from their own
group’s mean was excluded from further analyses. This resulted in one sighted participant
(participant 20) being excluded. The study consisted of three experiments administered to all
participants within the same session. Experimental procedures were approved by the Johns
Hopkins Homewood Institutional Review Board, and all participants provided informed consent.
Experimental Procedures Overview
Experiment 1 and 3 queried knowledge of and inferences about the colors of real objects
(30 objects in Experiment 1, 24 in Experiment 3). In Experiment 2, participants made color
inferences about 15 novel objects. Experiment 2 was always administered first to prevent the real
object judgments from influencing inferences made about novel objects. Within each experiment,
two different trial orders were used, one for half of the participants within each group.
Experimenters read aloud instructions and trials, and participant answers were audio-recorded
and later transcribed for scoring. The full list of stimuli and instructions can be found in the
Appendix (Supplemental Materials).
Experiment 1: Knowledge of Real Object Colors
In each trial of Experiment 1, participants were asked two questions about an everyday
object (Fig. 1). Three types of questions were asked: color consistency (30 objects), usage
consistency (20 objects), and fillers (20 objects). Objects used for color trials were either natural
(10 objects) or manmade (20 objects), and manmade artifacts could have function-relevant color
(FC, 10 objects) or non-function-relevant color (NFC, 10 objects). Usage trials consisted of 10
natural kinds and 10 artifacts. On filler trials, participants were asked questions about non-color
features (size, shape, and texture). Filler trials consisted of 5 natural kinds and 15 artifacts in
order to balance the overall number of natural kind and manmade trials. The full list of items used
in color and usage trials can be found in main Figure 1.
On color trials, participants were first asked, “What is one common color of (a) [object
name]?”, followed by, “If you picked two [object name]s at random, how likely are they to be the
same color? Rate on a scale of 1 to 7 (1: ‘unlikely’, 3: ‘somewhat likely’, 5: ‘likely’, 7: ‘very likely’).”
For usage trials, the questions were, “What is one common thing you can do with (a/some)
[object name]?” and, “If you picked two people at random and asked them each to do something
with (a/some) [object name], how likely are they to do the same thing, on a scale of 1 to 7?” Usage
trials served as a control condition to ensure blind and sighted participants showed equivalent
performance and were willing to rate artifacts as having some consistent properties.
Experiment 2: Color inferences about novel objects
In Experiment 2, in order to elicit inferences about novel objects parallel to in Experiment
1, participants were first presented an “Explorer on an Island” scenario:
“Imagine that you’re an explorer, and on your travels, you’ve discovered an island in a
remote corner of the world… You learn that the people on this island call themselves Zorkas…
The Zorka people appear to have a highly advanced culture. They have their own language, tools,
machines, buildings, vehicles, foods, customs, and so on. The ecology on this island is also
different from what we’re used to: it has its own plant and animal life, unusual rocks, minerals,
and so on. You’re trying to learn about how things work on this island....”
Participants then heard 35 short vignettes, each of which described an encounter with a
novel object (natural kind, artifacts with functional color, and artifacts with non-functional color;
Fig. 1). In each trial, several appearance features were noted (e.g., “green gem that is spiky and
the size of a hand”). The object was then named (e.g., “The miners tell you that this gem is called
As in Experiment 1, participants were next asked to rate the likelihood that another
instance of the same object would have the same color (e.g., “How likely is it that the next time
you come across another Enly, it is also green?). In usage trials, the question asked the likelihood
that the novel object would be used in the same way if encountered at another time (e.g., “How
likely is it that the next time you come across another Irve, it is also being ripped out of the
ground?”). In addition, there were 10 filler trials (7 natural kind, 3 manmade objects), in which
participants were asked about the likely repeat occurrence of a non-color feature (e.g., shape,
Color trials consisted of 5 natural kinds (plant, algae, gem, liquid from a plant, fruit), 5
artifacts with function-relevant color (coin, road symbol, toxic waste container, ceremonial
clothing, clear substance being used to build a wall), and 5 artifacts with function-irrelevant color
(an odor-emitting gadget, roof cleaning machine, two devices with ambiguous functions).
Usage trials consisted of 5 natural kind (creature, boulder, stone, flower, plant) and 5
artifacts (machine that makes square holes, storage device, toy, machine that turns stones into
goo, and one contraption with ambiguous function).
Filler trials contained 7 natural kind (fruit, two creatures, rock, two plants, gem) and 3
artifacts (game device, type of pool, one contraption with ambiguous function).
Experiment 3: Explanations about the cause of object color
For an additional list of 24 real objects (8 natural kind, 9 manmade with functional color, 7
function-irrelevant color), we asked participants to report their common colors (as in Experiment
1). Common color reports for these 24 objects are collapsed with those from Experiment 1 in main
Figure 2. For these objects, we additionally asked why objects had the particular color (or colors)
that the participant provided: “Why are [object name]s that/those color[s]?” Participants were
instructed to provide whatever explanation felt right to them. Participants were also asked whether
the object has different colored parts, and if an object’s color varies across instances, to report
the other colors. The answer to these questions were not analyzed for the present study.
Quantifying color naming agreement for real objects
Across Experiments 1 and 3, participants named the color of 54 objects (Fig. 1)
(Experiment 1: 30 objects, “What is one common color of…?” and Experiment 3: 24 objects,
“What is the most common color of…?”). For each object, we quantified naming agreement by
using the Simpson’s Diversity Index (SDI) (32, 20). For unique color words (1 to R) provided for
each object across all participants within a group (blind or sighted), a naming agreement score
was calculated according to the equation below. N is the total number of words used across
participants for each object, and n is the number of times each unique word (1 to R) was
provided. The index ranges for 0 to 1, where 0 indicates that the same color word was never
used by two participants (i.e., low color naming agreement), and 1 suggests all participants
provided the same color (i.e., high naming agreement).
Although participants were instructed to provide one color, a few participants provided
multiple colors (at most three, e.g., “red, white, and blue”). All of these colors were included in
the analysis. Further, a small proportion of participants said “I don’t know” or provided words
that were not typical color terms (dark, light, beige, neon). These responses were treated the
same as color terms (“I don’t know” was counted as one word, coded “IDK”). Since SDIs were
not normally distributed, they were log-transformed. To examine differences in color naming
agreement across groups, we then performed linear mixed effects regression on log-
transformed SDIs, using lmer in R (60), with objects as random effects.
Color consistency inference analysis
Consistency likelihood judgments were analyzed using ordinal logistic regression using
the ordinal (61) package in R. Participants and objects were always included as random effects,
and separate models were used in each analysis described (e.g., for real vs. novel objects).
We first compared group differences for natural kinds and artifacts with non-functional
color only, since artifacts with functional color are a special category. This also allowed us to
look at a group (blind vs. sighted) x object kind (natural vs. artifact) x trial type (color
vs. function) three-way interaction. Baselines were coded as sighted group, usage trial, and
artifact. We then compared across groups for color trials only, this time including all three kinds
of objects (natural, artifact with functional color, artifact with non-functional color), with sighted
group and artifact with non-functional color as the baseline.
Correlation with functional relevance of color for artifacts
We obtained ratings from Amazon Mechanical Turk (n=20) for the functional relevance
of color to artifacts. Participants were asked “How important is the color of a [object] to its
function?” and had to rate on a scale of 1 to 7 (not at all to very relevant). Artifacts designated
as ‘artifacts with functional colors’ were those that received an average rating of 4 or above,
and artifacts ‘non-functional colors’ all had ratings below 3 (Table S2). We correlated the
average functional relevance ratings for each object with the average color consistency
judgments, for blind and sighted groups separately (Spearman correlation).
Analysis of explanations
Explanation types were decided by the experimenters based on examining all the
explanations (while blind to group and object). We decided on 9 types of explanations: ‘process’,
‘depends on’, ‘just is’, ‘material’, ‘social/aesthetic’, ‘maker’, ‘visibility’, ‘convention’, and ‘I don’t
know’. A key of explanations can be found in Supplemental Materials (Table S3).
Explanations were coded by four coders who did not know which object or group each
explanation came from. Note, however, that in a small number of instances participants said
the object’s name in their explanations, and at other times, it was fairly easy to discern the
object from the explanation.
There was large variability in how many words participants used in their explanations
(range=1 to 165 words, M=13 words). This meant that each explanation (i.e., what one
participant said for one object) could contain multiple explanation types. For example, a
participants’ answer that the color of a wedding dress is due to “symbolism, or personal style”,
was coded as containing ‘convention’ (for symbolism) and ‘social/aesthetic’ (for personal style)
explanations. However, the same word or phrase (e.g., “personal style”) was never coded for
more than one explanation type.
Some participants gave lengthier explanations than others, without necessarily providing
additional information (e.g., often telling anecdotal stories to make a point). For wedding dress,
for instance, another participant explained: “well, there’s something about tradition, and white
being associated with purity and virginity and all that, but beyond that it’s just a matter of demand,
if you want a baby barf green wedding dress that’s your problem”. This explanation was also
coded with ‘convention’ and ‘social/aesthetic’.
Coding was then filtered according to the criteria that at least three out of four coders
have to agree. The first author (5th coder) made some additional changes, again keeping group
and objects blind, and overruled tagging for <5% explanations. After this process, the number
of explanation types per explanation (again, a single explanation from one participant for one
object) only ranged from 1-3 (mean=1.26).
We compared explanations across groups within each object kind. Within a group and
kind (e.g., sighted group, natural kinds), we calculated how frequently participants (across all
participants within group) used each of the 9 explanation types. The counts were then calculated
as a percentage of all explanations (within group and object kind). We then computed
Spearman’s correlations over the percentages (for 9 types) across groups, as well as across
object kinds within groups.
1. Gelman, S. A. (2009). Learning from others: Children's construction of concepts. Annual review
of psychology, 60, 115-140.
2. Lombrozo, T. (2019). “Learning by Thinking” in Science and in Everyday Life. In The scientific
imagination. Oxford University Press.
3. Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and brain sciences, 22(4), 577-
4. Prinz, J. J. (2005). The return of concept empiricism. In Handbook of categorization in cognitive
science (pp. 679-695). Elsevier Science Ltd.
5. Machery, E. (2006). Two dogmas of neoempiricism. Philosophy Compass, 1(4), 398-412.
6. Spelke, E. S. (1998). Nativism, empiricism, and the origins of knowledge. Infant Behavior and
Development, 21(2), 181-200.
7. Carey, S. (2009). The origin of concepts. Oxford university press.
8. Meltzoff, A. N., & Gopnik, A. (2013). Learning about the mind from evidence: Children’s
development of intuitive theories of perception and personality. Understanding other
minds, 3, 19-34.
9. Locke, J. (1924). 1690. An essay concerning human understanding, 1.
10. Hume, D. (1938). An Abstract of a Treatise of Human Nature, 1740. CUP Archive.
11. Jackson, F. (1982). Epiphenomenal qualia. The Philosophical Quarterly (1950-), 32(127), 127-
12. Jackson, F. (1986). What Mary didn't know. The Journal of Philosophy, 83(5), 291-295.
13. Dennett, D. C. (1993). Consciousness explained. Penguin uk.
14. Tye, M. (2000). Knowing what it is like: The ability hypothesis and the knowledge argument.
Protosociology, Collection of Essays for David Lewis.
15. Chalmers, D. J. (2004). 13 Phenomenal Concepts and the Knowledge Argument. There's
Something About Mary: Essays on phenomenal consciousness and Frank Jackson's
knowledge argument, 269.
16. Gleitman, L., & Landau, B. (1985). Language and experience: Evidence from the blind child.
17. Shepard, R. N., & Cooper, L. A. (1992). Representation of colors in the blind, color-blind, and
normally sighted. Psychological Science, 3(2), 97-104.
18. Marmor, G. S. (1978). Age at onset of blindness and the development of the semantics of
color names. Journal of Experimental Child Psychology, 25(2), 267–278.
19. Saysani, A., Corballis, M. C., & Corballis, P. M. (2018). Colour envisioned: Concepts of colour
in the blind and sighted. Visual Cognition, 26(5), 382–392. https://doi.org/10/ggmt29
20. Kim, J. S., Elli, G. V., & Bedny, M. (2019a). Knowledge of animal appearance among sighted
and blind adults. Proceedings of the National Academy of Sciences, 116(23), 11213-
21. Connolly, A. C., Gleitman, L. R., & Thompson-Schill, S. L. (2007). Effect of congenital
blindness on the semantic representation of some everyday concepts. Proceedings of the
National Academy of Sciences, 104(20), 8241-8246.
22. Lewis, M., Zettersten, M., & Lupyan, G. (2019). Distributional semantics as a source of visual
knowledge. Proceedings of the National Academy of Sciences, 116(39), 19237-19238.
23. Ostarek, M., Van Paridon, J., & Montero-Melis, G. (2019). Sighted people’s language is not
helpful for blind individuals’ acquisition of typical animal colors. Proceedings of the
National Academy of Sciences, 116(44), 21972-21973.
24. Keil, F. C., Smith, W. C., Simons, D. J., & Levin, D. T. (1998). Two dogmas of conceptual
empiricism: Implications for hybrid models of the structure of knowledge. Cognition, 65(2-
25. Keil, F. C. (1992). Concepts, kinds, and cognitive development. mit Press.
26. Gelman, S. A. (2003). The essential child: Origins of essentialism in everyday thought. Oxford
University Press, USA.
27. Greif, M. L., Nelson, D. G., & Kemler, K. FC, & Gutierrez, F.(2006). What do children want to
know about animals and artifacts, 455-459.
28. Gelman, S. A. (1988). The development of induction within natural kind and artifact categories.
Cognitive psychology, 20(1), 65-95.
29. Bloom, P. (1996). Intention, history, and artifact concepts. Cognition, 60(1), 1-29.
30. Springer, K., & Keil, F. C. (1991). Early differentiation of causal mechanisms appropriate to
biological and nonbiological kinds. Child development, 62(4), 767-781.
31. Gerstenberg, T., & Tenenbaum, J. B. (2017). Intuitive theories. Oxford handbook of causal
32. Majid, A., Roberts, S. G., Cilissen, L., Emmorey, K., Nicodemus, B., O’grady, L., ... & Shayan,
S. (2018). Differential coding of perception in the world’s languages. Proceedings of the
National Academy of Sciences, 115(45), 11369-11376.
33. Grojean, R. E., Sousa, J. A., & Henry, M. C. (1980). Utilization of solar radiation by polar
animals: an optical model for pelts. Applied Optics, 19(3), 339-346.
34. Keil, F. C., Greif, M. L., & Kerner, R. S. (2007). A world apart: How concepts of the constructed
world are different in representation and in development. Creations of the mind: Theories
of artifacts and their representation, 231-245.
35. Conceptual differences between children and adults. Mind & Language, 3(3), 167-181.
36. Keil, F. C. (2010). The feasibility of folk science. Cognitive science, 34(5), 826-862.
37. Feher, E., & Meyer, K. R. (1992). Children's conceptions of color. Journal of research in
Science Teaching, 29(5), 505-520.
38. Naranjo Correa, F. L., Martinez Borreguero, G., Perez Rodriguez, A. L., Suero Lopez, M. I.,
& Pardo Fernandez, P. J. (2016). A new online tool to detect color misconceptions. Color
Research & Application, 41(3), 325-329.
39. Anderson, C. W., & Smith, E. L. (1986). Children's Conceptions of Light and Color:
Understanding the Role of Unseen Rays. Research Series No. 166.
40. Cohen, J., & Nichols, S. (2010). Colours, colour relationalism and the deliverances of
introspection. Analysis, 70(2), 218-228.
41. McCloskey, M. (1983). Intuitive physics. Scientific american, 248(4), 122-131.
42. IInagaki, K., & Hatano, G. (2006). Young children's conception of the biological world. Current
Directions in Psychological Science, 15(4), 177-181.
43. Malle, B. F., Knobe, J. M., & Nelson, S. E. (2007). Actor-observer asymmetries in explanations
of behavior: New answers to an old question. Journal of Personality and Social
Psychology, 93(4), 491.
44. Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: An illusion of
explanatory depth. Cognitive science, 26(5), 521-562.
45. Harris, P. L., Koenig, M. A., Corriveau, K. H., & Jaswal, V. K. (2018). Cognitive foundations of
learning from testimony. Annual Review of Psychology, 69, 251-273.
46. Gelman, S. A., & Roberts, S. O. (2017). How language shapes the cultural inheritance of
categories. Proceedings of the National Academy of Sciences, 114(30), 7900-7907.
47. Vosniadou, S., & Brewer, W. F. (1992). Mental models of the earth: A study of conceptual
change in childhood. Cognitive psychology, 24(4), 535-585.
48. Gordon, R. M. (1986). Folk psychology as simulation. Mind & language, 1(2), 158-171.
49. Gallese, V., & Goldman, A. (1998). Mirror neurons and the simulation theory of mind-reading.
Trends in cognitive sciences, 2(12), 493-501.
50. Bruni, E., Boleda, G., Baroni, M., & Tran, N. K. (2012, July). Distributional semantics in
technicolor. In Proceedings of the 50th Annual Meeting of the Association for
Computational Linguistics: Long Papers-Volume 1 (pp. 136-145). Association for
51. Grand, G., Blank, I. A., Pereira, F., & Fedorenko, E. (2018). Semantic projection: recovering
human knowledge of multiple, distinct object features from word embeddings. arXiv
52. Rubinstein, D., Levi, E., Schwartz, R., & Rappoport, A. (2015, July). How well do distributional
models capture different types of semantic knowledge?. In Proceedings of the 53rd Annual
Meeting of the Association for Computational Linguistics and the 7th International Joint
Conference on Natural Language Processing (Volume 2: Short Papers) (pp. 726-730).
53. Kim, J. S., Elli, G. V., & Bedny, M. (2019b). Reply to Lewis et al.: Inference is key to learning
appearance from language, for humans and distributional semantic models alike.
Proceedings of the National Academy of Sciences, 116(39), 19239-19240.
54. Kim, J. S., Elli, G. V., & Bedny, M. (2019c). Reply to Ostarek et al.: Language, but not co-
occurrence statistics, is useful for learning animal appearance. Proceedings of the
National Academy of Sciences, 116(44), 21974-21975.
55. Cimpian, A., Brandone, A. C., & Gelman, S. A. (2010). Generic statements require little
evidence for acceptance but have powerful implications. Cognitive science, 34(8), 1452-
56. Tessler, M. H., & Goodman, N. D. (2019). The language of generalization. Psychological
review, 126(3), 395.
57. Brandone, A. C., & Gelman, S. A. (2009). Differences in preschoolers’ and adults’ use of
generics about novel animals and artifacts: A window onto a conceptual divide. Cognition,
58. Csibra, G., & Gergely, G. (2009). Natural pedagogy. Trends in cognitive sciences, 13(4), 148-
59. Shafto, P., Goodman, N. D., & Frank, M. C. (2012). Learning from others: The consequences
of psychological reasoning for human learning. Perspectives on Psychological Science,
60. Bates D, Mächler M, Bolker B, Walker S (2015). “Fitting Linear Mixed-Effects Models Using
lme4.” Journal of Statistical Software, 67(1).
61. Christensen RHB (2019). “ordinal—Regression Models for Ordinal Data .” R package version
Acknowledgments: We thank the National Federation of the Blind convention, the blind
community, and all of our participants for making this research possible; Funding: This work was
supported by the National Institutes of Health (R01 EY027352 to M.B.) and the Johns Hopkins
University Catalyst Grant (to M.B.); J.S.K was funded by the William Orr Dingwall Dissertation
Fellowship; Author contributions: J.S.K, B.A, V.M, and M.B. designed research; J.S.K, B.A, and
V.M, performed research, J.S.K analyzed data, and J.S.K and M.B. wrote the paper; Competing
interests: Authors declare no competing interests; and Data and materials availability: All data,
code, and materials used in the analysis are available in the repository:
https://github.com/judyseinkim/Intuitive-Theories-of-Color and a detailed description of analyses
can be found in the following document: https://rpubs.com/judyseinkim/color_theory .