A Semantic Space of Color Names

Abstract

Three Russian subjects learned arbitrary pairings between 20 colors and 20 three-letter artificial color names. After different amounts of this training, the subjects rated the difference between the colors associated with every pair of artificial names when these names were presented without the colors. Multidimensional scaling of the ratings after a small amount of training revealed a grouping of the words into four semantic clusters corresponding to the following groups of related colors: the violets, the blues, the greens, and the yellows-through-reds. After more extensive training, multidimensional scaling yielded the full color circle of hues. Further analysis of the data indicated that a spherical model previously proposed by the authors for sensory color space has advantages, also, for the semantic color space obtained when only the names of colors are presented. The results are interpreted in terms of a two-stage process of neuronal analysis of visual inputs in which the activity of four color-opponent channels is followed by differential activation of cells tuned to specific colors.