Color naming across languages reflects color use

Abstract

What determines how languages categorize colors? We analyzed results of the World Color Survey (WCS) of 110 languages to show that despite gross differences across languages, communication of chromatic chips is always better for warm colors (yellows/reds) than cool colors (blues/greens). We present an analysis of color statistics in a large databank of natural images curated by human observers for salient objects and show that objects tend to have warm rather than cool colors. These results suggest that the cross-linguistic similarity in color-naming efficiency reflects colors of universal usefulness and provide an account of a principle (color use) that governs how color categories come about. We show that potential methodological issues with the WCS do not corrupt information-theoretic analyses, by collecting original data using two extreme versions of the color-naming task, in three groups: the Tsimane', a remote Amazonian hunter-gatherer isolate; Bolivian-Spanish speakers; and English speakers. These data also enabled us to test another prediction of the color-usefulness hypothesis: that differences in color categorization between languages are caused by differences in overall usefulness of color to a culture. In support, we found that color naming among Tsimane' had relatively low communicative efficiency, and the Tsimane' were less likely to use color terms when describing familiar objects. Color-naming among Tsimane' was boosted when naming artificially colored objects compared with natural objects, suggesting that industrialization promotes color usefulness.

Keywords: Whorfian hypothesis; basic color terms; color categorization; color cognition; information theory.

Fig. 1.

The Amazonian Tsimane' people show large individual differences in color naming, but at the population level, similar color categories to those observed among Bolivian-Spanish and English speakers. Color naming of 80 chips evenly sampling the Munsell array, presented singly in random sequence under controlled lighting, in Tsimane', Spanish-speakers in neighboring San Borja (Bolivia), and English-speaking students near Boston (see SI Appendix, Table S2 for the key relating the axes to Munsell chip designations). Color of each diamond corresponds to the modal color for the chip (see SI Appendix, Table S3 for the key matching the color with the terms in each language). Diamond size shows the fraction of people who gave the modal response. All participants showed 100% consistency for black and white chips: negro, blanco (Bolivian-Spanish); tsincus, jaibas (Tsimane'). The location of the numbers overlying the plot indicate the color chips in the 160-chip Munsell array that were most frequently selected as the best example of the subset of modal color terms queried (SI Appendix, Table S4). The numbers are the percentage of respondents who made the given selection. Note that two modal color terms in Tsimane', yụshñus and shandyes, correspond to the same chip (E8). English speakers were asked about red, green, yellow, blue, orange, brown, purple, and pink. Bolivian-Spanish speakers were asked about rojo, verde, amarillo, azul, celeste, anaranjado, morado, cafe, and rosa. Tsimane' were asked about jäinäs (∼red), yụshñus (∼blue), shandyes (∼green), itsidyeisi (∼purple), cafedyeisi (∼brown), and chames (∼yellow). Data are from the free-choice version of the task (n = 58 Tsimane', 20 Bolivian-Spanish, 31 English); data from the fixed-choice version of the task, conducted in separate participants (n = 41 Tsimane', 25 Bolivian-Spanish, 29 English), yielded similar results (SI Appendix, Fig. S6).

Fig. 2.

Variability of color labels (entropy, Eq. 3) for familiar objects, ordered by Tsimane' results. On average, Tsimane' has higher entropy over color words for a particular object (1.06 bits, compared with English, 0.33 bits, and Bolivian-Spanish, 0.30 bits).

Fig. 3.

Communication efficiency of color naming, across languages and among color chips. (A) Communication efficiency for each language of the WCS (open symbols), Tsimane’ (black symbols), Bolivian-Spanish (dark gray symbols), and English (light gray symbols), as a function of number of unique color words used by the population of participants tested in each language. The two data sets collected in Tsimane’, Bolivian-Spanish, and Tsimane’ show that variability in experimental methods have little impact on assessments of communicative efficiency of color naming, licensing the use of the WCS data for further analysis. Circles show data from experiments in which participants were constrained to use a fixed vocabulary of basic color terms; squares show data where participants were free to use any term. Number of participants stated as (N=fixed choice, free choice). Communicative efficiency for each language was computed using Eq. 2. (B) Color chips rank-ordered by their average surprisal (computed using Eq. 1) for Tsimane' and Bolivian-Spanish (pattern for English overlaps Spanish, omitted for clarity). SI Appendix, Table S5 provides the chip identity in rank order. The asterisks represent focal colors determined as described in Fig. 2. The sequences of colors in each population are highly correlated (Spearman rank correlation between Bolivian-Spanish and English, ρ = 0.87; between Bolivian-Spanish and Tsimane', ρ = 0.51; and between English and Tsimane', ρ = 0.53).

Fig. 4.

Color chips rank-ordered by their average surprisal (computed using Eq. 1) for all languages in the WCS, and the three languages tested here. Each row shows data for a given language, and the languages are ordered according to their overall communication efficiency (Eq. 2).

Fig. 5.

The color statistics of objects predict the average surprisal of colors. Objects in the Microsoft Research Asia database of 20,000 photographs were identified by human observers who were blind to the purpose of our study (31). The colors of the pixels in the images were binned into the 80 colors defined by the Munsell chips used in the behavioral experiments (across the images there were 9.2 × 10⁸ object pixels and 1.54 × 10⁹ background pixels). The y axis shows the [(number of pixels of given color in objects)/(number of pixels of given color in objects + number of pixels of given color in backgrounds)]; the color chip ranking is that obtained for the Tsimane'. Error bars are SE. The three languages were not significantly different from each other (English: slope = −0.0064, ρ = −0.57, P value = 3 × 10⁻⁸; Bolivian-Spanish: slope = −0.0049, ρ = −0.44, P value = 5 × 10⁻⁴; Tsimane': slope = −0.0054, ρ = −0.49, P value = 5 × 10⁻⁶).

Fig. 6.

The Tsimane' use color terms less frequently than English speakers. (A) Contrastive-labeling paradigm, adapted to assess use of color terms in normal communication. (B) Percent of trials in which participants used a color word to describe objects presented in sequential pairs. Members of each pair were identical except for color (e.g., green banana/yellow banana). Tsimane' speakers were less likely to use a color word (mixed effect logistic regression, β = −5.23, z = −5.48, P < 0.0001). Among Tsimane', a mixed-effect logistic regression shows a main effect of artificiality (β = 3.59, z = 4.00, P < 0.0001) and presentation order (β = 1.57, z = 3.09, P < 0.01) with no interaction (β = 0.91, z = 1.19, P = 0.23). Among English, we find a main effect of presentation order (β = 1.53, z = 4.00, P < 0.001).