When Vision Disagrees with Language: Private and Population-Level Color Knowledge in Red–Green Dichromacy
Abstract
Human concepts are shaped by two distinct information sources: private sensorimotor experience and socially transmitted knowledge mainly conveyed through language in an abstract symbolic form. Because information geometry in these sources is largely aligned, their relative contributions to conceptual representation are difficult to disentangle. We exploit a group-level dissociation, in which the alignment between sensory color geometry and population-level language structure diverges systematically (red–green dichromacy vs. trichromacy), and a model-level dissociation (object-color geometry derived from large language models vs. from image-based sensory models), to disentangle vision and language contributions to object knowledge. Sensory color geometry differed behaviorally and neurally between dichromats and trichromats, yet behavioral object-color representations converged across groups and corresponded robustly with language-derived geometry. LadATL consistently expressed language-derived geometry across tasks and groups, whereas LpFG showed a group-dependent representational profile in the chromatic-object task: trichromats showed relatively greater correspondence with sensory-derived geometry, whereas dichromats retained correspondence with both geometries. Dichromats also showed stronger intrinsic connectivity of LadATL with LpFG and EVC. These findings reveal how the brain flexibly draws on sensory evidence and socially transmitted knowledge to maintain population-aligned conceptual representations across heterogeneous sensory experiences.
Related
- Development and psychometric validation of Mpox knowledge and preventive behavior measures in the Turkish population
- Informed forecasting: Leveraging auxiliary knowledge to boost large language model performance on time series forecasting
- Investigating lexical knowledge and lexical processing in order to explore their role in second language utterance fluency
- Relation-aware dynamic knowledge subgraph enhanced large language models for medical multiple-choice question answering
- Relationship among motivation, vocabulary learning strategies and the breadth of vocabulary knowledge: A structural equation modeling study on second language learning
BibTeX
@misc{lu2026when,
title = {{When Vision Disagrees with Language: Private and Population-Level Color Knowledge in Red–Green Dichromacy}},
author = {Jiahui Lu and Ziyi Xiong and Peijun Yuan and Xiaoying Wang and Yanchao Bi},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/ewpv5_v1},
url = {https://osf.io/preprints/psyarxiv/ewpv5_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - When Vision Disagrees with Language: Private and Population-Level Color Knowledge in Red–Green Dichromacy
AU - Jiahui Lu
AU - Ziyi Xiong
AU - Peijun Yuan
AU - Xiaoying Wang
AU - Yanchao Bi
PY - 2026
DA - 2026/09/26
PB - PsyArXiv
DO - 10.31234/osf.io/ewpv5_v1
UR - https://osf.io/preprints/psyarxiv/ewpv5_v1/
AB - Human concepts are shaped by two distinct information sources: private sensorimotor experience and socially transmitted knowledge mainly conveyed through language in an abstract symbolic form. Because information geometry in these sources is largely aligned, their relative contributions to conceptual representation are difficult to disentangle. We exploit a group-level dissociation, in which the alignment between sensory color geometry and population-level language structure diverges systematically (red–green dichromacy vs. trichromacy), and a model-level dissociation (object-color geometry derived from large language models vs. from image-based sensory models), to disentangle vision and language contributions to object knowledge. Sensory color geometry differed behaviorally and neurally between dichromats and trichromats, yet behavioral object-color representations converged across groups and corresponded robustly with language-derived geometry. LadATL consistently expressed language-derived geometry across tasks and groups, whereas LpFG showed a group-dependent representational profile in the chromatic-object task: trichromats showed relatively greater correspondence with sensory-derived geometry, whereas dichromats retained correspondence with both geometries. Dichromats also showed stronger intrinsic connectivity of LadATL with LpFG and EVC. These findings reveal how the brain flexibly draws on sensory evidence and socially transmitted knowledge to maintain population-aligned conceptual representations across heterogeneous sensory experiences.
ER -