Colour and language

Where Blue Ends

The same palette, divided by 110 languages

The spectrum flows without edges. Words divide it, but the shape of the division changes with the language. Some draw a line. Most change the question.

Draw the line

The band below runs from green to blue without steps. Lightness and saturation are held constant along its whole length, so hue is the only thing that changes. Choose the place where green becomes blue for you.

Tap or click where green stops.

The question fits 35 languages

You drew a line because English asked you for one. The same operation cleanly describes only a third of the World Color Survey. The rest of the palette needs other shapes.

35

Direct transition

35 languages pass once from the dominant term at focal green to the dominant term at focal blue. Only these positions appear on the axis above.

64

Matching endpoints

64 languages use the same dominant term at both focal probes. Two contain a different term between the probes, so this claim is about the endpoints rather than the whole band.

11

Bridge or return

11 languages introduce a third term or return to an earlier one. Their data describes a region rather than a single frontier.

What a whole language looks like

Every speaker named all 330 chips. The grid below shows, for each chip, the word most of them used. Rows run from light to dark, columns travel round the hue circle.

29 speakers · One word from green to blue

  • wɔɔ/wɔ
  • nye/nyie
  • finge

The square beside each word is the mean of the colours that word covers in the grid, so it shows what the word denotes rather than a colour chosen by me. The detached column on the left is the scale without colour, white through black.

Gladstone's wrong question

In 1858, before he became prime minister, William Gladstone published three volumes on Homer. He counted the colour words in the Iliad and the Odyssey and found a thin vocabulary, no clear word for blue, and the sea described seventeen times with the adjective used of wine. He concluded that the Greeks had an undeveloped eye.

The Greek eye was our eye. What was missing was the line, and the line turns out to be the free part. Homer had words for how light behaves on a surface and less use for a shelf of hues. The same observation that led Gladstone to a conclusion about the retina leads, with today's data, to a conclusion about the lexicon.

On those seventeen seas

The order is not arbitrary

In 1969 Brent Berlin and Paul Kay noticed that languages do not add colour words in any order. A language with two words splits them between light and dark. The third is red. Yellow and green follow, then blue, then brown, then the rest. The claim was attacked at once over sample and method, and the World Color Survey was the answer: the same 330 chips, carried to 110 communities, with about 24 speakers each.

The result does not confirm a rigid ladder, but it does show something stable. When all 109,256 best-example choices from both surveys are placed on the hue axis, the density has two clear peaks between yellow and purple: one at 2.5G, one at 2.5PB. Those are the two chips used here as probes. Speakers of every language pointed at them, which is what selected them; English had no part in the choice.

Romanian took its blue from a word for white

Albastru comes from Latin albaster, from albus. The starting sense was grey, washed out, whitish, and Alexandru Ciorănescu's etymological dictionary records it surviving in Moldavian speech and in Maramureș, where a grey sheep is called albastră. The word reached the clear sky by way of the pale one.

The other word, vânăt, is also Latin, from venetus. It covers dark blue tending to violet, the grey-blue of horses, and skin that has been struck or chilled. Neither has a Slavic root, though the second sounds like a loan. Romanian has two Latin words for the blue region, and the one that became the ordinary term set out from white.

The second line runs the other way

The distinction between albastru and vânăt does not live on the band you cut. Hue stays the same; what changes is how dark it is. Russian makes the same cut obligatorily, between goluboy and siniy, and Jonathan Winawer and colleagues showed in 2007 that Russian speakers tell two samples apart faster when they fall on opposite sides of that border, an advantage that disappears when a spoken task occupies their language in parallel.

Where does blue end and dark blue begin?

This line has never been measured for Romanian in any published survey. The band holds hue fixed at 2.5PB and varies only lightness. What you choose stays in your browser.

What is settled and what is still argued

Settled

Languages differ in both the position of their borders and the shape of their categories. In the World Color Survey, 64 of 110 use the same dominant term at focal green and focal blue. Best-example choices nevertheless gather in a few stable regions of the palette.

Open

How much of that gathering comes from the eye and how much from the way the question was put. Debi Roberson, Jules Davidoff and Ian Davies, working with Berinmo speakers in Papua New Guinea and Himba speakers in Namibia, showed that each language's own borders change performance on memory and discrimination tasks, which cuts against a strictly universalist reading. Terry Regier, Paul Kay and Richard Cook replied in 2005 that the focal points hold across the World Color Survey data. Both results stand, because they look at different things: one at the borders, the other at the centres.

What this page cannot show

The original chips are Munsell papers viewed under a standardised illuminant. Your screen is not that illuminant, and almost a third of the chips at full saturation fall outside the sRGB gamut. The band you cut is therefore drawn at a saturation any screen can render honestly, tamer than the real papers. Your position is comparable with the others because it uses the same hue axis, not because your screen is calibrated.

The 20 languages in Berlin and Kay's appendix carry one synthesised speaker each and contain several gaps between the probes. They remain in the reproducible archive and in the history of the idea, but do not enter the three main counts. The World Color Survey averages 24 speakers for each of its 110 languages. The question reached you in a language; its words supplied a shape before you touched the band.

How the figures were computed

For each language the commonest term at each chip was taken, then the dominant term for each hue column. A line enters the comparison only when the sequence between the probes contains one green run followed by one blue run. A third term, a return or a column without observations receives its own class. The table shows how the classification changes when the calculation uses other lightness bands.

Check across row bands
Rows used Direct transitions Matching endpoints Complex shapes Line median
B-I3564117.5BG–10BG
C-G3465117.5BG–10BG
D-F2965165BG–7.5BG
E2366215BG–7.5BG

Sources

The data files are public. The survey's original site no longer responds; the copies used here come from the web archive, and the figures were recomputed from them.