Reproduce “Probing the Geometry of Color Representations in Multilingual Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pablo Loyola, Edison Marrese-Taylor, Andrés Hoyos-Idrobo
The authors examine how language models represent color relationships across layers, comparing model similarity patterns with CIE–LAB and across languages. They find weak alignment with CIE–LAB that often declines toward the final layer, while cross-language similarity is strongest near the input and often falls with depth. These results show distinct color-language and perceptual structure in tested models and datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.