Changing the structure of color categories causally influences color-concept association generalization

People form associations between colors and concepts from experience (Schoenlein & Schloss, 2022), but questions remain concerning how continuous color-concept association distributions are populated from sparse input. The category extrapolation hypothesis proposes that associations for an input color (e.g., a blue) extrapolate to all other colors with the same color category (i.e., all blues) (Rathore et al., 2020). Schoenlein and Schloss (VSS-2022) found that color category boundaries predicted patterns of color-concept association generalization, but this result was correlational. Here, we tested whether category boundaries causally influence association generalization when participants form novel associations between colors and the concepts of two alien species, Slubs and Filks. The category extrapolation hypothesis implies that if participants learn to merge hues from different categories (e.g., blue and purple) into a single overarching category (e.g., blue+purple), they will show greater generalization across category boundaries compared to participants who keep those categories separate. We tested this hypothesis in an experiment with two groups (blue+purple mergers and green+yellow mergers), who completed three tasks. In the first, color naming task, blue+purple mergers learned new color names which merged blues/purples and separated greens/yellows (green+yellow mergers learned the opposite). In the second, color-concept association formation task, participants learned to associate Slubs with a particular blue and Filks a particular green. In the last, color-concept association task, participants indicated (yes/no) whether they associated each species with colors in sequences spanning blue-to-purple and green-to-yellow. These judgments produced psychometric functions over color sequences and the bias term indicated degree of generalization. The results showed that merging hues from different categories resulted in greater generalization (interaction between merging group and color sequence, p=.008). This study provides the first causal evidence that color category boundaries influence color-concept associations, indicating that category extrapolation influences how continuous color-concept association distributions are populated from sparse input.