I propose that names be considered more like probability distributions. For example the name “red” has a probability distribution that probably peaks in the nei

I propose that names be considered more like probability distributions. For example the name “red” has a probability distribution that probably peaks in the neighbourhood of #FF0000 but includes a large range of colors; the distribution of the name “crimson” likely has a much narrower distribution. Under this interpretation, and given enough data, one could feasibly extract “the” name for any given colour using a maximum likelihood estimation (what name is most likely to generate the observed colour). And to collect data on this, one could for example ask many random users to choose the range of colours they believe correspond to a given name (“select all red colours”) and merge the results into the distribution. (There are many, many other ways to carry out such a survey; this is but one example). Obviously, you won’t get to 16M names this way, but you could definitely learn quite a lot about where the “boundaries” are between colours from this kind of exercise!