Robust individual alignment of color qualia structures: toward a structure-based taxonomy of divergent color experiences
Togashi, Y.; Yotsumoto, Y.; Hiramatsu, C.; Tsuchiya, N.; Oizumi, M.
Show abstract
Whether qualitative aspects of consciousness, or qualia in short, are equivalent across individuals is a foundational scientific question. Testing this is challenging because one cannot assume a shared mapping between stimuli and private experience (my "red" may be your "green") [1-3]. Previously, we proposed a structural characterization of qualia [4, 5] and the quantitative assessment of structural correspondences through an unsupervised alignment method [4, 6], which does not presuppose such correspondence. Using this approach, our previous work focused on identifying optimal mappings between relational structures of color qualia at the group level [4]. Given known perceptual diversities [7], however, it remained unknown whether any two individuals structures could be empirically aligned. Here, we resolve this by collecting 4,371 pairwise similarity ratings for 93 colors-from 11 individuals, enabling direct individual-to-individual alignment. We reveal two fundamental, coexisting features. First, we identified two clusters of individuals showing robust within-cluster alignment, corresponding to color-neurotypicals and atypicals. Second, we uncovered a continuous spectrum of diversity: some participants who showed normal color discrimination ability in terms of the Total Error Score (TES) on Farnsworth-Munsell 100 hue test nevertheless failed to align with either cluster, revealing idiosyncratic structures that defy simple categorization. Together, these findings suggest a novel structure-based taxonomy of divergent color qualia that complements conventional performance-based classification. Our method is generalizable to other sensory modalities, and opens a path to the scientific investigation of both shared and idiosyncratic qualitative aspects of consciousness.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.