New words for colour arise as if through a kind of game: the speaker might talk about a particular colour using a word that the hearer is not familiar with. Will the hearer figure out what the speaker is referring to, and if so, will the hearers then adopt the same word themselves, jettisoning their own word for that colour, or recognizing a new sub-category? It is out of many interactions of this sort, which may or may not help to spread a word – be it a colour or an object – that the population’s shared language arises.
For colour words, though, this linguistic negotiation is also biased by how we perceive them. We do not see all parts of the visible spectrum equally: it is easier for us to see small changes in hue in some parts than in others. Loreto and colleagues propose that this so-called “just noticeable difference function” of colour perception makes it more likely that agents in their model will reach consensus faster on a word for some hues than others.
The speed at which agents reached a consensus on a name fell into a distinct hierarchy, and the order in which this happened – red first, then violet, green/yellow, blue, orange and then cyan – is very close to that identified by Berlin and Kay. (Black and white, which are not themselves spectral colours, must be assumed at the outset as the crude distinction between dark and light.) Crucially, this sequence cannot be predicted purely from the “just noticeable difference function” – that is, from the physiology of colour vision – but arises only when it is fed into the naming game.
The match is far from perfect, however. For one thing, violet does not appear in Berlin and Kay’s hierarchy. Loreto and colleagues explain its emergence in their sequence as an artificial consequence of the way reddish hues crop up at both ends of the visible spectrum. And Berlin and Kay listed brown after blue. But brown isn’t a spectral colour – it is a kind of dark yellow/orange, and so can be considered a variant shade of orange.
Whether or not you accept those explanations for the discrepancies, this model of language evolution looks set to offer a good basis for exploring factors such as cultural differences and contingencies, like those Jarman discovered. It also offers a good basis for how language gets transmitted between cultures, often mutating in the process.