Increasing caption length from ~5 words to ~30 words in DCI dataset improves average language-vision alignment scores across all evaluated model pairs

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Increasing caption density (from ~5 to ~30 words) monotonically improves language-vision alignment scores across all vision model familiesfinding0.879
  Supports the claim that information content of modality pairing determines alignment level
• Cross-modal language-vision alignment reaches a maximum of approximately 0.16 on mutual nearest-neighbor metric in Figure 3, well below the theoretical maximum of 1finding0.760
  Quantitative bound on observed alignment; raises the open question of whether this gap reflects noise or real misalignment
• The better an LLM is at language modeling, the more it aligns with vision models, and vice versa — linear relationship between language modeling score and vision-language alignmentfinding0.753
  Core cross-modal empirical result: larger and better language models align better with vision models
• CLIP models exhibit higher language-vision alignment than supervised or self-supervised vision models, but this alignment decreases after fine-tuning on ImageNet classificationfinding0.753
  CLIP training paradigm finding in cross-modal alignment
• Alignment with vision models corresponds to improved performance on downstream language tasks including commonsense reasoning and mathclaim0.750
  Claims that alignment score is a proxy for general capability
• The more descriptive (higher information) a caption is, the better its LLM representation aligns with the visual representation of the corresponding imageclaim0.742
  Preliminary test of the information-level limitation of PRH; denser captions = higher cross-modal alignment
• Color distances learned from language cooccurrence statistics closely mirror those learned from image cooccurrence statistics and human perceptual distances (CIELAB)finding0.739
  Case study confirming that PMI-based learning in different modalities recovers the same perceptual representation
• Training on image data should improve LLM performance, and training on language data should improve vision model performancehypothesis0.731
  Implication of PRH for cross-modal training efficiency