# Neutral shows weak correlation with diversity (ρ=0.05 decTest, ρ=-0.08 conTest)

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

• # Contradiction is positively correlated with diversity (ρ=0.26 decTest, ρ=0.46 conTest)finding0.870
  Ablation result confirming that contradiction predictions indicate diversity
• # Entailment is negatively correlated with diversity (ρ=-0.48 decTest, ρ=-0.65 conTest)finding0.828
  Ablation result confirming that entailment predictions indicate lack of diversity
• Confidence NLI Diversity achieves ρ=0.64 correlation with human diversity judgments on conTestfinding0.828
  Highest human correlation for semantic diversity metric
• Generation length shows weak correlation with ACCatom and RCatom (r = [0.20, -0.12])finding0.794
  Generation length does not strongly affect accuracy or retest consistency at atomic level
• Baseline NLI Diversity – MNLI achieves Spearman ρ=0.59 on conTest diversity parameter correlationfinding0.792
  Comparable to top-performing automatic metric from Tevet and Berant 2021
• The contradiction relation is more useful than the neutral relation for measuring semantic diversityclaim0.781
  Key finding from ablation study on NLI prediction categories
• Strength comparison pair (3,7) with |Δα|=4 outperforms pair (3,5) with |Δα|=2, indicating graded sensitivity to perturbation magnitudefinding0.779
  Shows that introspective accuracy scales with injection strength difference, not binary detection
• AUS_N is a weaker correlate of θ50 than S_max across E3 backbonesfinding0.765
  E3 finding distinguishing the two geometry summaries; breadth less predictive than peak height