Generating and Resolving Vague Color

We describe a method for distinguishing colors in context using English color terms. Our approach uses linguistic theories of vagueness to build a cognitive model via Bayesian rational analysis. In particular, we formalize the likelihood that a speaker would use a color term to describe one color but not another as a function of the background frequency of the color term, along with the likelihood of selecting standards in context that fit one color and not the other. Our approach exhibits the qualitative flexibility of human color judgments and reaches ceiling performance on a small evaluation corpus.