Showing cs.CLShow all
2 papers · 1 filter
cs.CL2026
Regime Boundary Alignment for Evidence-Gated Question Answering
Zeyan Li, Qirong Guo, SIyuan Qiu +3
Retrieval-augmented language models are expected to answer from the retrieved evidence, but in practice they often keep answering when that evidence is missing. We trace this behav…
cs.CL2026
Reasoning Instructions Can Break Answer Decoding in Vision--Language Models
Zeyan Li, Siyuan Qiu, Jianfeng Xu
Chain-of-thought (CoT) instructions can distort multiple-choice VLM evaluation when a scorer appends a reasoning cue but reads answer-label logits before the model generates any ra…