Abstract

People have systematic associations between concepts and colors, and concepts can be represented as distributions of association strengths across color space (Schloss et al., 2018; Rathore et al., 2018). These patterns of color-concept associations are highly structured, which may be critical in explaining how people flexibly draw conceptual inferences about colors across contexts (Mukherjee et al., 2021). Here we consider the structure of the semantic space defined by color-concept associations, evaluating three hypotheses. First, the structure may be arbitrary, so that items evoking similar color associations are conceptually unrelated. Second, it may be redundant with the conceptual structure apparent in natural language, as estimated via word-embeddings computed from large corpora. Third, it may express a representational space important to human cognition but distinct from those latent in natural language. To adjudicate these possibilities, we selected 30 concepts spanning 6 semantic domains (clothes, emotions, scenes, fruits, directions, times-of-day) and collected association ratings between each concept and 58 colors sampled across CIELAB color space. Pairwise dissimilarities between concepts in this color-concept association space could be well captured by embedding them in a reduced 3-dimensional space, where concepts with similar color-concept associations were embedded near each other. A comparison of these embeddings to those extracted from natural-language corpora revealed highly distinct conceptual spaces. To determine whether the color-concept space is reflected in mental representations of word meanings, we used a triadic comparison task to embed the same items based on direct judgments of semantic similarity. These correlated more strongly with color-concept similarities, suggesting that the color-concept associations express aspects of semantic knowledge beyond color that elude word-embeddings derived from language corpora, supporting our third hypothesis. The results provide an initial basis for understanding abstract conceptual relations in human cognition through the lens of visual properties of concepts — in this case, color-concept associations.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call