Rui-Chun Tang

Results show mechanism-level evaluation captures architecture-specific sensitivity patterns that behavioral benchmarks cannot; pairing both should become standard NLP practice.

Focusing on visual hallucination, SAFE is introduced, a training-free decoding framework that contrasts visually-grounded and vision-ablated generation paths to produce a token-level contrastive grounding score that identifies when the model favors linguistic priors over visual evidence.

This work proposes ViD, a causally-inspired framework that analyzes attention mechanisms across five distinct patterns, revealing confounding effects from strong language priors and demonstrates that visual-to-language cross-attention effectively suppresses bias while preserving general reasoning capabilities and text generation quality.