Search Abstracts | Symposia | Slide Sessions | Poster Sessions
When Vision Learns to Speak: Language-Linked Modulation Diverges Between the Visual Cortices of Humans and Macaques
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Haojie Wen1, Kesheng Wang2, Yipeng Li2, Haoyang Chen2, Pinglei Bao2, Yanchao Bi2; 1Beijing Normal University, 2Peking University
The macaque has long served as the canonical model for human vision, reflecting an assumption of evolutionary conservation in both architecture and function. We directly tested this assumption by comparing neural responses to identical naturalistic stimuli in humans (fMRI) and macaques (electrophysiology), using visual models that differ in language supervision. Both species exhibited shared tuning to the core visual features, yet their cortical modulations diverged: human visual responses were better predicted by language-supervised models (CLIP) and by affordances linked to tool use, whereas macaque responses aligned with unsupervised models (MoCo) and low-level visual statistics. The language supervision advantage diminished for animate objects and was absent in the face-selective cortex. These results define boundary conditions on cross-species homology, showing that the visual cortex’s functional modulation in humans has been retuned by language and action systems over the course of evolution.
Topic Areas: Meaning: Lexical Semantics, Multisensory or Sensorimotor Integration