Claire Bowern writes in The Conversation about her research into color terms:
My colleague Hannah Haynie and I were interested in how color terms might change over time, and in particular, in how color terms might change as a system. That is, do the words change independently, or does change in one word trigger a change in others? In our research, recently published in the journal PNAS, we used a computer modeling technique more common in biology than linguistics to investigate typical patterns and rates of color term change. Contrary to previous assumptions, what we found suggests that color words aren’t unique in how they evolve in language. […]
That there’s something unique about the stability of color concepts is an assumption we wanted to investigate. We were also interested in patterns of color naming and where color terms come from. And we wanted to look at the rates of change – that is, if color terms are added, do speakers tend to add lots of them? Or are the additions more independent, with color terms added one at a time? […]
In order to answer these questions, we used techniques originally developed in biology. Phylogenetic methods use computers to study the remote past. In brief, we use probability theory, combined with a family tree of languages, to make a model of what the history of the color words might have been. […]
Our results supported some of the previous findings, but questioned others. In general, our findings backed up Berlin and Kay’s ideas about the sequential adding of terms, in the order they proposed. For the most part, our color data showed that Australian languages also show the patterns of color term naming that have been proposed elsewhere in the world; if there are three named colors, they will be black, white and red (not, for example, black, white and purple). But we show that it is most likely that Australian languages have lost color terms, as well as gained them. This contradicts 40 years of assumptions of how color terms change – and makes color words look a lot more like other words.
I freely admit I don’t understand the method works (“we estimate likely reconstructions, evaluate that model for how well it fits our hypotheses, tweak the model parameters a bit to produce a different set of results, score that model, and so on”), and I’m curious what my readers think of the method and the results.