Richer color vocabulary is associated with better color memory but not color perception

Significance

The size of color vocabulary of an individual varies dramatically across the population. We systematically harnessed this variance and showed that the number of color names one possesses almost perfectly predicts her/his color memorization ability; however, the size of color vocabulary is not related to one's basic color perception. In addition, it is known that color names are not evenly distributed across the color wheel. We systematically captured this source of variance and showed that color memorization—and not basic color perception—is more accurate for parts of the color wheel described by higher density of color names.

Abstract

The potential interaction between color naming and psychophysical color recognition has been historically debated. To study this interaction, here we utilized two approaches based on individual differences in color naming and variation of color name density along the color wheel. We tested a pool of Persian speaking subjects with a simple color matching task under two conditions: perceptual and memory-based matching. We also asked subjects to freely name 100 evenly sampled hues along the color wheel. We found that, individuals who possess more names to describe the color wheel have a strong edge in color memorization over those with fewer names. Nevertheless, having more or fewer color names was not related to the subjects’ performance in perceptual color matching. We also calculated the density of color names along the color wheel and observed that parts of the color wheel with higher density of color names are held in memory more accurately. However, similar to the case of individual differences, the density of color names along the wheel did not show any correlation with perceptual color matching performance. Our results demonstrate a strong link between color naming and color memorization both across different individuals and different parts of the color wheel. These results also show that low-level perceptual color matching is not related to color naming, suggesting that the variation in color naming—among the individuals and across the color wheel—is neither the cause nor the effect of variation in low-level color perception.

Humans break the color wheel into smaller segments, each labeled by a color name. Some individuals employ more names to segment the color wheel and some, less sophisticated with naming, use fewer labels to describe it. Do individuals who possess more color names have better color vision? For nearly half a century, considerable debates have been raised over this simple question. On one hand, the “universalist view” suggests that the mental process of color recognition is independent of the linguistic processes that assign names to colors. According to this view, the linguistic color pallet, the vocabulary treasure that discriminates various hues in a language, does not constrain the way humans perceive and remember colors (1, 2). On the other hand, the “relativist view” implies that assignment of color names to various parts of the color wheel interacts with color recognition performance in humans (3–10).

To address the debate between universalist and relativist views, one should experimentally address a “problem of circularity” formalized by Winawer and Witthoft (11). The problem of circularity arises from an inherent problem in the definition of “spacing” in any color space. In order to psychophysically measure the potential effect of linguistic category boundaries on color recognition for different colors (within or between category boundaries), the experimenter first needs to create a physically controlled and evenly spaced color space. Now, if linguistic color categories are taken into account for equal spacing of test colors, then by definition there won’t be any residual linguistic effects left to measure; if not, how can one ensure the test colors were equally spaced to begin with? In the latter case, any observed category effect may reflect the possibility that colors within a linguistic category are psychophysically harder to discriminate and that’s why they have fallen into a natural category in the first place. Winawer and Witthoft have listed three categories of psychophysical approaches to break the circularity problem and investigate the potential relationship between language and color vision: cross-linguistic studies, hemifield specific effects, and verbal interference experiments.

Historically most of the fuel for the debate between universalist and relativist camps come from cross-linguistic studies. In this approach, variation of linguistic color pallet across different languages is used to explain the variation in color recognition performance among the speakers of those languages (e.g., refs. 1, 2, 5, 7–10). In this way the circularity problem is bypassed because the physical spacing of test colors can be kept constant while linguistic category boundaries vary across different languages. Some cross-linguistic studies have revealed interesting effects of language on color memory and learning (e.g., refs. 5, 6, 9), thus supporting the relativist view. Nevertheless, other studies have supported the universalist view (e.g., refs. 1, 2). As informative as cross-linguistic studies are, they come with an inherent problem; cross-cultural confounding factors. Distance from the equator that determines the amount of ultraviolet (UV) damage to the eyes (12), variations in advancement and application of color display technology (13), and cultural variation in exposure to colorful materials (14) are among the many variables that affect color recognition performance and may vary across cultures besides the variation in language. In addition, other sources of variance such as individual differences in color recognition may contaminate cross-linguistic measures of color recognition performance (15). These sources of variance may conceal potential effects of language on color recognition and explain some of the discrepancies in the literature.

Another category of psychophysical studies that concern language and color vision has emerged based on the neurophysiological fact that in most people, language is mostly processed by the left hemisphere (16–18). This leads to the assumption that visual stimuli that are presented to the left hemisphere (right visual field) are influenced more by language. This assumption addresses the circularity problem by keeping the physical spacing of colors constant, yet varying the linguistic effects by varying the visual field. These experiments are done in the context of a single language (19, 20) as well as in combination with cross-linguistic studies (21). The latter case is quite interesting because it controls for potential cross-linguistic confounding factors. In some cases, this category of experiments has revealed effects of linguistic category boundaries on color recognition in the context of speeded visual search task (e.g., refs. 19–21). However, other studies have failed to replicate some of those results and the source of discrepancy is still debated (22, 23).

The third category of evidence comes from verbal interference experiments. The idea behind this class of studies is that if color vision depends on language, it should be affected by the conditions in which language faculty is overloaded by a parallel task (6, 7, 10, 19). The problem of circularity is addressed here by keeping the spacing between colors physically constant, and varying the involvement of the language faculty in the task. These experiments are done both for a single language (e.g., refs. 6, 19) and in the context of cross-linguistic studies (e.g., refs. 3, 7, 10). Verbal interference experiments generally support the relativist view as they reveal various effects of verbal interference on color visual search (19), speeded color matching reaction times (7, 10), color memory (6), and oddball color detection (3). The problem with verbal interference experiments is rather a pragmatic one. Given the difficulty of performing interference experiments for human subjects, the number of verbal interference trials that can be collected from each subject is small and practically limited to the median of 96 trials (ranged from 432 to 16) in the previous studies (e.g., refs. 6, 7, 10, 19). Because of this limitation, the entire body of evidence in this category comes from a carefully selected but limited set of colors around specific linguistic category boundaries such as the blue/green (e.g., refs. 3, 6, 19, 24) blue/dark-blue (e.g., refs. 7, 10) boundaries and multiple (seven) preselected category boundaries (25). The generality of these findings for the entire color wheel is yet to be studied. Building on this rich history, here we introduce two approaches in order to eliminate some of the limitations of previous studies and to investigate the interaction of language with color recognition more directly. In the first approach, in order to avoid cross-cultural confounding factors, we utilized individual differences in color naming and color recognition within only one linguistic domain. There are individual differences in color naming and color recognition even within a language domain (9, 26, 27); we harnessed this natural variance in order to study the covariance of color naming and color recognition in a pool of Persian speaking subjects. Here, to break the problem of circularity, we kept the physical colors constant and varied the observer within a cultural domain.

In the second approach, unlike many previous studies that were limited only to specific color category boundaries, here we measured the subjects’ color naming and color recognition performance for the entire color wheel in a parameterized way. Then, we used heterogeneity of distribution of color names and color recognition performance along the color wheel as a source of variance to study their relationship. In this approach, instead of trying to neutralize the problem of circularity, we directly measured its effects on a physically constant set of colors in the context of two different psychophysical tasks; perceptual and memory-based matching (Discussion). The analytical methods developed here are language independent, thus they can be replicated and compared for any other language domain in the future.

To measure color recognition performance we adopted two separate operational definitions of “color recognition”: a simple low-level color matching task to directly assess concurrent color perception of the subjects, as well as a color memory task. Using these two tasks we aimed at two empirical questions: 1) Is color perception and color memory performance different among individuals with different vocabulary treasure of color names? 2) Independent of individual differences, does the distribution of color names along the color wheel predict color recognition performance for different hues?

To achieve these aims, we tested color recognition abilities of the subjects using a color matching paradigm. In each trial, the subjects were required to change the color of a “test patch” on the screen to match it with that of a “reference patch.” The reference patch was filled randomly with one of the 100 hues evenly sampled from an imaginary circle in the CIEIUV color space (Methods). There were two conditions for this task: the “perceptual matching” condition and the “memory-based matching” condition. In the perceptual matching condition, a test patch and a reference patch were presented simultaneously and the subjects could look at the reference patch as they made their match. This task aimed at documenting the most basic color recognition abilities of the subjects. In the memory-based matching condition, the reference patch was shown for 10 s, then it disappeared; following a delay of 10 s the test patch was presented to the subjects who matched its color to their memory of the reference patch. This task is tailored to measure color memory performance for each sampled hue separately (Fig. 1).

Fig. 1.

(A) Stimuli. One hundred hue samples were selected evenly from the perimeter of an imaginary circle (r = 0.08) centered at gray in CIELUV color space. The color wheel on the Right side of the panel shows the sampled hues, U and V values varied among the 100 sampled colors but the luminance level was kept constant at 1 cd/m². (B) Color matching experiment in two conditions: perceptual matching and memory-based matching. Participants used a computer screen to match the color of 100 hue samples (presented in random order) under two conditions. In perceptual matching condition (Top) two patches of color were presented simultaneously on a black background. In each trial, one of the 100 sample colors, randomly selected, was presented in the top color patch (reference patch). The participants adjusted the color of the bottom patch (test patch) to match it with the reference patch. For each trial, the participants had 60 s to complete the matching procedure. Memory-based matching (Bottom) was similar to perceptual matching except for a 10-s delay introduced between the reference and test patches. In this condition, the participants were asked to withhold the color of the reference patch in memory and adjust the color of the test patch to match the memorized hue (see text for details). (C) Color matching results in a typical subject. The radius of each data point shows the average error of the matches for the 100 color samples. The blue line corresponds to perceptual matching and the red line represents memory-based matching.

To determine the linguistic color pallet of each subject for the sampled color wheel, all of the subjects participated in a separate color naming task (Fig. 2A). The color naming task was presented to subjects after they completed the other two tasks; this was done to avoid potential biases in subjects’ performance in the two other tasks by informing them about our interest in language. In the color naming task the subjects were presented with all 100 sampled hues, one at a time in random order, and asked to type the name of the hue in a dialogue box. At first glance we measured the total number of names that each subject possesses to describe the entire perimeter of the sampled color wheel. The linguistic color pallets of our 20 subjects contained 21 to 50 unique names (mean = 38.80, SD = 7.709) for the tested colors. Female and male subjects possessed an average of 36.14 (median = 38) and 40.23 (median = 41) total color names, respectively. The effect of gender on the number of color names was not significant (t test, t (18) = 1.14, P = 0.2692). Then we investigated the relationship between subjects’ performance in the color matching task and the total name count on their color pallets. The distance between the reference hue and matched hue was defined as “error” for each match. For each subject, this error was averaged across all trials. We noticed a strikingly strong correlation between error and total color name count across subjects for memory-based color matching (r = −0.9428, P < 0.00001); the more color names you possess the better your color memory is (Fig. 2B). However, the clear advantage of people with more color names in the color memory task cannot be the result of their better color perception, simply because there was no correlation between error and total color name count for the perceptual color matching condition (r = −0.1462, P = 0.53).

Fig. 2.

The more names an individual uses to describe colors the better her/his performance is on a color memory matching test. (A) Color naming task. All of the 100 sampled hues were presented to the subjects in random order in separate trials. In each trial, subjects were asked to type the name of the presented color into a text box on the screen (in Persian). (B) The abscissa indicates the number of unique names that each subject has to describe the color wheel. The ordinate represents the error (average of 20 subjects) of the participants on the color matching experiment. Each data point represents one of the 20 participants. While individuals who possess more names to describe the color wheel are similar to others in color perception, they can memorize colors more accurately. Left and Right, respectively, represent perceptual (r = −0.146, P = 0.538) and memory-based (r = −0.942, P < 0.0001) matching conditions. Upper subpanels depict color naming schema of two of the subjects (the ones who had the least and most color names). Each sector on the color wheel corresponds to a unique name that the subject uses to describe that part of the wheel.

Did the subjects explicitly memorize the names of colors to perform the memory-based color matching task? Such an internal naming strategy may have helped the subjects who possess more color names in the memory-based task, thus it can explain the strong correlation observed. In order to assess the potential role of internal color naming in the subjects’ strategy for performing the color memorization task, in an exit interview we asked the participants to describe the strategy they used to remember different colors. Eight of the 20 participants (40%) reported that they memorized the exact test hues and not their names. For this group of subjects correlation between the size of color vocabulary and memory error was high and significant (r = −0.9620, P = 0.0001). The remaining 12 subjects (60%) reported that they matched the colors by memorizing both the hues and the hue labels. A similar near perfect correlation was observed for this group (r = −0.9538, P = 0). Fisher Z test shows no significant difference between these two correlation values (z = −0.2, P = 0.41). This suggests that the link between language and color memory is as strong for those subjects who “consciously and explicitly” think they have not used internal naming to perform the task (SI Appendix, Fig. S1). It is still possible that those subjects used some form of linguistic labeling without being consciously aware of it; we do not intend to reject this theoretical possibility and we leave its further exploration to future studies. In fact, if a form of implicit linguistic labeling mechanism exists, we would classify it as a natural mechanism for color memorization, given that we provided no instructions for performing our task. In any case, here, we only conclude that explicit usage of internal naming for performing memory-based color matching does not affect the correlation between color memory and color vocabulary size.

Next, we aimed at our second empirical question: Is the distribution of color names along the wheel related to color recognition performance for various hues? To create a quantitative and language-independent measure of the distribution of color names we defined “name density.” Name density is the number of unique names that subjects possess to describe a small section of the color wheel. To measure name density for all hues, a sliding window (±5 hues from each sampled hue) was moved along the 100 sampled hues for each subject and the number of unique color names in each window (11 hues wide) was counted. As a result, a name densitogram was created for each subject. The name densitogram indicates the number of unique names that each subject has for each part of the tested color wheel (Fig. 3A).

Fig. 3.

Variation of color name density and color matching performance along the color wheel. (A) Name densitogram. Name density (ND) was defined as the number of names that an individual assigned to each subsection of the color spectrum. The name densitogram shows the density of unique names at each point on the tested color spectrum for each individual (see text for details). (B) Interaction of name density and performance in color matching experiment. Black line indicates ND. Blue and red lines show error (average of 20 subjects) of the matches for the 100 tested hues in perceptual and memory-based matching, respectively. Vertical and horizontal axis labels correspond to error and name density, respectively.

First, we noticed that name density varies significantly in different parts of the color wheel. In general, Persian speaking subjects have fewer names for orange (mean = 2.8) and purple (mean = 3.38) tones and more names for pink (mean = 6.8), and blue (mean: 4.7) and light green (mean = 5.12) shades. Among the pool of subjects one-way ANOVA showed a significant effect of hue on name density for all subjects (F (19,99) = 26.30, P < 0.000001). Analysis of data from individual subjects showed a significant negative correlation between name density and color matching error (across all hues) for the memory-based matching task in 16 out of 20 subjects. The distribution of Pearson r values among subjects was significantly below zero for memory-based matching (mean r = −0.321, SD = 0.189, t test: t (19) = −7.59, P < 0.00001). Name density and color matching error showed significant correlations only in 2 of the 20 subjects for the perceptual matching condition. The distribution of r values was not significantly different from zero for perceptual matching (mean r = −0.0478, SD = 0.138, t test: t (19) = −1.54, P = 0.07).We also averaged the data across all individuals and calculated average error and name density for each hue (Fig. 4). Average name density and average error (both pooled across all subjects) showed a significant negative correlation for memory-based matching (r = −0.555, P < 0.00001) and not for perceptual matching (r = −0.05, P = 0.609). To make sure the effects observed in the memory condition have not originated from any low-level effect potentially present in the results of the perceptual condition, we normalized memory error values to perceptual errors for each hue and correlated them with name density (r = −0.36, P < 0.001). This normalization obviously adds noise to the correlation, but the remaining significant correlation ensures that the memory effect is not shaped by any inhomogeneity in perceptual matching across the color wheel. In sum, this analysis shows that the density of color names varies across the different parts of the color wheel, and that color memory performance is higher in subareas of the wheel that incorporates more color names. However, the density of color names along the wheel does not influence perceptual color matching performance.

Fig. 4.

Distribution of color names over the color space is not even and color matching performance reflects it. The parts of the color wheel that are described by more names (higher name density) are remembered more accurately. The abscissa indicates the name density (ND) around each hue on the color wheel and the ordinate represents the error (average of 20 subjects). Each data point depicts one of the 100 tested hues. Left represents performance in perceptual color matching (r = −0.051, P = 0.609) and Right represents performance in memory-based color matching (r = −0.555, P < 0.0001).

The method of adjustment used in this study allows variation in color matching time and this can potentially affect the results. To study the potential role of matching time in the phenomena observed here, we first explored the potential effect of matching time on color matching accuracy within the time variation range naturally present in the data. We observed no significant correlation between accuracy (inverse of the error) and matching time across the sampled hues (r = −0.013, P = 0.89) and across the individuals (r = 0.08, P = 0.73) for memory-based color matching in our data (also see SI Appendix, Fig. S2). Next, to reassure that matching time does not affect our main findings, we divided the trials into two groups based on trial matching times: fast trials (trials faster than the mean reaction time) and slow trials (trials slower than the mean reaction time). Our main effect was repeated for both trial types with no significant change in the effect size (SI Appendix, Fig. S3).

Discussion

Here we studied the relationship between linguistic color labeling and color recognition performance across different individuals and different parts of the color wheel. This was done in the context of two separate operational definitions of the term color recognition; a simple color matching procedure to measure accuracy of concurrent color perception directly as well as a memory-based matching paradigm which is similar to the color matching task with a 10-s period of retention added.

As for individual differences, our results reveal a near perfect correlation between the total color names that an individual possesses and her/his memory-based color matching performance. This correlation—to the best of our knowledge—is the strongest effect size that has ever been demonstrated in this area of research. The large effect may partly reflect the value of focusing on individual differences within a single culture, which removes cross-cultural confounding factors and allows measurement of the distilled effect of language on color recognition. Reviewing the literature, we found one interesting cross-cultural study (9) that has touched upon this method as a side analysis. Roberson et al. (28) noticed significant yet smaller correlations between the number of color names and color memory performance within pools of Himba and Berinmo speakers. Nevertheless, they did not observe this correlation for English speakers. This discrepancy was explained by the fact that English has more color descriptors compared to Himba and Berinmo, but this assumption is not true about Persian (29). We believe the lack of correlation for English is rooted in the way “memory error” was calculated in Roberson et al. Given the different objective of their study, Roberson et al. chose an all-or-none measure of error; a trial was counted as correct only if the subject picked the exact target color, any other selection was counted as wrong. Here we define a continuous quantitative measure of error (distance from the target color) which can provide a more accurate measurement of error for a potential correlation. To test this hypothesis and to replicate Roberson et al., we reanalyzed our data using the all-or-none definition of error and—similar to Roberson et al.—found no significant correlation between memory error and number of color names (r = −0.3844, P = 0.0943). We are very curious and hopeful to see similar studies by other groups and predict falsifiably that using a quantitative definition of error would reveal similar strong correlations in English and other languages.

The variation of color name density and color recognition performance across the color wheel reveals a similar pattern: color memory performance is better in the parts of the color wheel that are pinned with more color names. This approach is particularly different as many previous studies had to focus on carefully selected hues around very few specific color boundaries. Now we extend the link between color memorization and color naming to the densely sampled continuum of the color wheel. Here, instead of attempting at breaking the circularity problem, we have measured and contrasted it across two psychophysical tasks, using the same set of stimuli. We have shown that color memory performance is higher for the parts of the color wheel that are labeled with more names. Even though we have used a standard color space to determine the spacing of the stimuli, this result may be a reflection of the circularity problem in that high-name-density parts of the color wheel may consist of easier to recognize colors, by measures not captured in LUV color space. But even if that is the case, this pattern does not exist for the same colors tested on a perceptual color matching task. In other words, the effect of linguistic color categories seems to vary for low-level basic color vision and short-term memory for colors. It should be mentioned that we have not tested the entire color space and our results point to an isosaturated color circle at the center of the LUV color space. Future studies can extend this approach to the entire area of the color space.

Our results for the memory-based matching condition are consistent with previous studies on color memorization in the general sense that color naming and color memorization interact with each other (e.g., refs. 2, 6, 9). For the perceptual color matching condition, we found no significant correlation between color naming and perceptual color matching, either in the context of individual differences or for variations of name density across the color wheel. Now, depending on which definition of color recognition to pick, memorization or perceptual matching, our results can be interpreted against or in favor of either of the universalist or relativist positions regarding the relationship between language and color recognition. It appears that for low-level perceptual representation of color, the universalist view takes the upper hand. As cleverly mentioned in Winawer and Witthoft (11) TV companies don’t need to adjust their basic color setting based on their market language. Consistently, color discrimination thresholds don’t seem to be different across speakers of different languages (29). The basic neural mechanisms of color representation is shared among humans, most of the apes, and some monkeys (30); it is unlikely that a recently evolved plug-in like language affects those basic representations. Our results suggest that the perceptual color matching task is performed by neural mechanisms that read out early and basic neural representations of color. However, for the case of color memorization, the relativist view prevails. Both across individuals and different parts of the color wheel, the more color names one possesses the better her or his color memory is. This suggests that memory-based color matching interacts with higher level mechanisms of color processing. These higher level mechanisms may constitute the neural basis for color concept (and name) formation, or alternatively, they may be causally formed by the naming schemas of the language faculty, or both statements may be partially true. This takes us to the question of causality.

Does having more color names cause people to develop better color recognition? Or is it the better color recognition that makes people use more color names? Or perhaps, there is a hidden variable that affects both? The direction of causality between color naming and color recognition performance is one of the oldest unknowns in this domain of research (31). Here, we cannot solve that problem at once, but we address a part of it. Existence of correlation does not imply causality, but the lack of correlation indicates the lack of causality. Given that variation of subjects’ performance in low-level color matching cannot be explained by the variation in their naming performance, we suggest that better color naming cannot be the causal result of having better color perception. Inversely, we can also conclude that better low-level color perception cannot be the cause of having more color names. In this study, our causal inference remains limited to this falsification. The causal relationship between color naming and more complex color tasks such as color memorization is yet to be understood.

In sum, our results suggest that the low-level representation and processing of color, as manifested in direct nonspeeded color matching, is independent of language and is not influenced by it. However, later processing of color, at least for the case of color memorization, seems to be very strongly related to the linguistic representation of color. This is true for variation of linguistic representation of color across individuals and across the color wheel at least for the case of Persian speaking subjects. Given that all of the methods used in the present study are culture independent, we very much hope to see similar investigations in other languages.

Methods

Subjects.

Twenty native Persian speakers, 13 women and 7 men (mean age = 26.35, SD age = 1.95) participated in the color matching and color naming experiments. Participants signed a written consent form and were paid for their time. All of the subjects had normal vision with no history of neurological disorders. Before the experiments, all participants were screened using the Ishihara test for color blindness and they were all normal.

Stimuli.

One hundred hue samples were selected evenly from the perimeter of an imaginary circle (r = 0.08) centered at gray in the CIELUV color space. The stimuli differed only in the u* and the v* chromaticity axes, and the luminance axis (L*) was kept constant (L = 1). The stimuli were generated and presented by MATLAB Psych-Toolbox (R2015b) and controlled by Intel Core 3.50 GHz PC processor, displayed on a photometrically calibrated LG LCD (60 Hz) screen.

Monitor Calibration.

The stimulus display monitor was photometrically calibrated prior to the experiment. Seventeen equidistant color patches were generated for each RGB channel and their luminances were photometrically measured. The luminance response of each color channel was then linearized accordingly. Moreover, the entire battery of 100 hue samples was photometrically measured again and their isoluminance was confirmed. Independent of their hue, all of the hue patches used in this study had the photometric energy of 27.48 cd/m2.

Experimental Design.

Each subject participated in four experiment sessions completed in 4 consecutive days. The first three sessions were for color matching experiments; each consisted of 400 trials (four repetitions of each hue, randomized order). Half of these trials were memory-based matching and the other half were perceptual matching. Perceptual and memory-based trials were presented in random order. There was a mandatory resting period of at least 3 min after each block of 100 trials. A total of 1,200 trials (12 repetition of each hue) was collected from each of the subjects.

Following the completion of the color matching trials, the subjects received a separate color naming task, they were shown the 100 sampled hues (once each) in random order. A blank dialogue box was presented under the color patch. Subjects were asked to type the name of each color in the blank box (in Persian). They were allowed to use up to four words for each “color name.” All subjects received the same instructions in Persian. Testing took place in a quiet, dimly lit room.

Color Matching Experiment.

There were two conditions for each experimental session including perceptual color matching and memory-based color matching. In the perceptual condition, two patches of color were presented simultaneously on a black background. The screen subtended 37.8 by 23.3 degrees of the visual field. Each color patch was a square subtending 8.18 × 8.18 visual degrees, the two squares were arranged vertically. In each trial, one of the 100 reference hues was presented randomly in the top color patch (reference patch). The subjects, freely viewing the stimuli, were instructed to adjust the color of the bottom patch (test patch) to match that of the reference patch. Color adjustment was done by pressing the right or left arrow keys, which allowed the subjects to roll through all 100 reference hues and select the one that looked like the color of the reference patch. The subjects could also use up and down arrow keys to pass through colors in larger steps (jump over 10 hues). For each trial the subject had 60 s to complete the matching procedure (Fig. 1B). In the memory-based color matching condition everything was similar to the perceptual condition except for the timing of the events. The reference patch was presented for 10 s, then it disappeared and the test patch was presented following a 10-s delay. The subjects were instructed to memorize the color of the reference patch and match the subsequently presented test patch to the color of the reference patch. Subjects had 60 s to complete their matching (Fig. 1C).

Data Availability.

Excel data files and Matlab codes data have been deposited in GitHub, https://github.com/Hasantash/Afraz.Richer-color-vocabulary-is-associated-with-better-color-memory/releases/tag/PNAS.