Short answer
Not in a fixed order, but not at random either. Berlin and Kay proposed in 1969 that languages gain basic colour terms in seven stages, from black and white through red to a full set of eleven. Tests on the World Color Survey found strong shared tendencies of that kind, while the strict ladder has been abandoned by its own authors. Change is also quicker and messier than a ladder suggests: the cross-language lexicon documents 13 terms whose range is recorded as shifting, including a Japanese light-blue term that became near-universal in about thirty years.
Berlin and Kay were struck by how easily ordinary colour words could be translated between unrelated languages. They defined a basic term as general and salient, then asked native speakers of twenty languages, all living in the San Francisco Bay Area, to mark on a chart of Munsell chips the best example of each basic term and the region it covered, and added a literature search on 78 more languages. Two findings followed. Best examples clustered in a few narrow areas of the chart, while boundaries were drawn unreliably. And ordering languages by their number of basic terms gave a sequence: two terms meant black and white, three added red, four added yellow or green, and so on through seven stages to eleven terms. They read the sequence as evolutionary, and that reading drew most of the criticism.
The sample was the first target: twenty languages, mainly those of industrialised societies and represented by speakers living in California, could hardly stand for the world's. The second was the method, which relativist critics held to be loaded with assumptions that produced the result. In 1997 Saunders and van Brakel argued at length that neither the physiology of vision nor the linguistic evidence supports universal basic categories, and that colour is not an autonomous domain at all. The third objection came from the data. The two Dani terms did not divide into black-plus-dark and white-plus-light as a first-stage language should; they divided cool-and-dark from warm-and-light. Kay and McDaniel responded in 1978 by recasting the stages as successive divisions of broad composite categories instead of successive additions of single ones.
The World Color Survey supplied what the first study lacked: naming of the same chips by speakers of 110 unwritten languages. Kay and Regier ran statistical tests on those data together with the earlier data and reported strong universal tendencies in languages of both industrialised and non-industrialised societies. Lindsey and Brown clustered the naming patterns of individual informants and found that as the number of clusters was allowed to grow, the categories appeared in an order reminiscent of the 1969 sequence. So the tendency is supported. What is not supported is a rule that every language passes through the same stages: the same analyses found several naming systems coexisting among speakers of one language.
Records of actual change are more varied than any sequence. Japanese gained a light-blue category within about thirty years, a split by lightness that the stage theory does not predict. Tsimane' speakers who learned Spanish gave two existing, interchangeable words separate meanings without borrowing anything. Himba took in a word that the Herero dictionary glosses as blue and used it for blue, green and purple. Welsh borrowed its green term from Latin over a thousand years ago and kept the older word for the green of grass. Chinese qīng was frequent in early texts and is uncommon now. Hungarian, which Berlin and Kay thought had twelve basic terms with two reds, has eleven according to later field and corpus studies. Vocabularies split, borrow, narrow and let words lapse.
Recent explanations do not need stages at all. On the efficiency account a colour vocabulary is a trade-off between having few words and being able to say precisely which colour is meant. Zaslavsky and colleagues showed that real naming systems sit close to the best possible trade-off, that the model's categories have soft boundaries and leave some regions inconsistently named, as real ones do, and that as the pressure for precision rises the model passes through a series of restructurings resembling the documented ones. Gibson and colleagues add a reason for the pressure to differ: colour words matter more where objects are distinguished by colour. The Hadza study suggests an earlier step, in which categories are present in how individuals group colours before the community agrees on words for them.
| Stage | Basic terms at this stage in the 1969 scheme | A recorded case, and how well it fits |
|---|---|---|
| I | Black and white | Dani has two terms, but they divide cool-and-dark from warm-and-light, which is what prompted the scheme's revision. |
| II | Black, white and red | Hadza informants agreed fully on these three and on nothing else; their other terms were used without agreement. |
| III | Adds either yellow or green | No term-level record here. |
| IV | Has both yellow and green | Berinmo and Himba each have five basic terms, one of them across green and blue; the Himba study identifies that term with the green-blue term of the staged scheme. |
| V | Adds blue | Tsimane' bilinguals separating green from blue show this division arising from two existing words. |
| VI | Adds brown | One study gives Mandarin nine basic terms including purple, pink and grey but no basic brown, which its authors reconcile with the scheme by a special provision. |
| VII | Adds purple, pink, orange and grey, to eleven | English. Russian, Greek, Italian and perhaps Japanese go beyond it with a second blue, a step the 1969 scheme has no stage for. |
| Language | Term | What the sources record |
|---|---|---|
| Russian | голубой (goluboy) | Paramei (2005) describes the word as symbolically charged and treats its rise to basic status as a cultural development, not a perceptual necessity. |
| Japanese | 青 (ao) | The word once covered green and blue together. Hsieh and colleagues (2020), citing Stanlaw and Kuriki, place the borrowing of the character in a period when ao meant both, probably the fifth century, with blue and green separating in the ninth to tenth centuries. In the eighth-century Man'yōshū it also described greyish things such as clouds and horses (Tobioka 2020). |
| Japanese | 緑 (midori) | Not a basic colour term in the eighth-century Man'yōshū on Tobioka's (2020) analysis, where it fails the criterion of not being an object name; an earlier proposal by Stanlaw had counted it as basic at that stage. |
| Japanese | 水 (mizu) | A change within about thirty years: the 1987 study of ten observers concluded Japanese had eleven basic terms; by 2017 almost every informant used mizu. |
| Mandarin Chinese | 青 (qīng) | Frequent in early Chinese texts and less common now. Hsieh and colleagues describe it as covering greenish to bluish shades in ancient usage and as a single green-blue term from the Tang period. |
| Mandarin Chinese | 藍 (lán) | The same character in Japanese is read ai and names indigo, the dye and its deep blue, not blue in general. |
| Welsh | glas | From a Celtic root shared with Cornish and Breton glas and Old Irish glas, all glossed by the dictionary as blue, green or grey. The compound for grass is built on it. |
| Welsh | gwyrdd | A borrowing from Latin viridis, recorded from the ninth century. It did not displace glas from the green of grass. |
| Welsh | llwyd | Cognate with Old Irish líath, grey, according to the dictionary's etymology. |
| Irish (Old and Middle Irish) | glas | The dictionary also links it to the blue dye from woad. It is cognate with Welsh glas. |
| Himba (a variety of Herero) | burou | Borrowed from Afrikaans by way of Herero, where the dictionary translates it as blue. In Himba the borrowed word took on a wider range than its source. |
| Tsimane' | shandyes | Malik-Moraleda and colleagues (2023) report the word's meaning narrowing among bilinguals without any Spanish word being borrowed. |
| Tsimane' | yụshñus | The study's title calls the bilingual pattern 'the birth of blue': an existing word acquiring a narrower meaning under contact with Spanish. |
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
FactModerate evidence
Berlin and Kay's 1969 field study used native-speaking informants in the San Francisco Bay Area for 20 languages, supplemented by a literature search on 78 more, and proposed a sequence of seven stages leading to eleven basic colour terms.
Caveat: Summarised from a review chapter; the authors later revised the number and nature of the stages.
Source: Basic Color Terms and Basic Color Categories (C. L. Hardin; chapter 11 of Backhaus, Kliegl and Werner, eds, Color Vision: Perspectives from Different Disciplines, 1998); Basic Color Terms: Their Universality and Evolution (Berlin and Kay, 1969)
FactStrong evidence
Statistical tests on colour-naming data from languages of both industrialised and non-industrialised societies found strong universal tendencies in both.
Caveat: Tendencies across languages, not identical categories in every language.
Source: Resolving the question of color naming universals (PNAS 100: 9085–9089, 2003)
FactModerate evidence
A 1997 review concluded that neither psychophysics and neurophysiology nor linguistic evidence supports universal basic colour categories.
Caveat: The conclusion of a critical review published with peer commentary; it predates the public release of the World Color Survey data.
FactModerate evidence
Colour-naming systems across languages achieve near-optimal compression in an information-theoretic model, and the model's optimal systems change through a sequence of structural transitions as the trade-off shifts.
Caveat: A model fitted to present-day naming data. It shows which changes would be efficient, not which changes any language made.
Source: Efficient compression in color naming and its evolution (PNAS 115: 7937–7942, 2018); History constrains the evolution of efficient color naming, enabling historical inference (PNAS 121: e2313603121, 2024)
FactStrong evidence
A field study and a corpus study both conclude that Hungarian vörös is not a basic colour term, leaving Hungarian with eleven basic terms where Berlin and Kay had supposed twelve.
Source: Preliminary study of basic colour terms in modern Hungarian (Linguistica Uralica 43(2): 102–123, 2007); The Hungarian colour terms piros and vörös: a corpus and cognitive linguistic account (Acta Linguistica Hungarica 61: 123–152, 2014)
Colourwise interpretationModerate evidence
The Berlin–Kay sequence is best read as a description of which divisions of colour are most common at each vocabulary size, not as a path each language follows.
Based on: Tests on the World Color Survey support cross-language tendencies; analyses of individual informants find several systems within one language; documented changes in Japanese, Tsimane', Himba and Welsh do not follow the staged order.
Reviewed 1 October 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.