English

DialDefer: A Framework for Detecting and Mitigating LLM Dialogic Deference

Computation and Language 2026-01-19 v1

Abstract

LLMs are increasingly used as third-party judges, yet their reliability when evaluating speakers in dialogue remains poorly understood. We show that LLMs judge identical claims differently depending on framing: the same content elicits different verdicts when presented as a statement to verify ("Is this statement correct?") versus attributed to a speaker ("Is this speaker correct?"). We call this dialogic deference and introduce DialDefer, a framework for detecting and mitigating these framing-induced judgment shifts. Our Dialogic Deference Score (DDS) captures directional shifts that aggregate accuracy obscures. Across nine domains, 3k+ instances, and four models, conversational framing induces large shifts (|DDS| up to 87pp, p < .0001) while accuracy remains stable (<2pp), with effects amplifying 2-4x on naturalistic Reddit conversations. Models can shift toward agreement (deference) or disagreement (skepticism) depending on domain -- the same model ranges from DDS = -53 on graduate-level science to +58 on social judgment. Ablations reveal that human-vs-LLM attribution drives the largest shifts (17.7pp swing), suggesting models treat disagreement with humans as more costly than with AI. Mitigation attempts reduce deference but can over-correct into skepticism, framing this as a calibration problem beyond accuracy optimization.

Keywords

Cite

@article{arxiv.2601.10896,
  title  = {DialDefer: A Framework for Detecting and Mitigating LLM Dialogic Deference},
  author = {Parisa Rabbani and Priyam Sahoo and Ruben Mathew and Aishee Mondal and Harshita Ketharaman and Nimet Beyza Bozdag and Dilek Hakkani-Tür},
  journal= {arXiv preprint arXiv:2601.10896},
  year   = {2026}
}

Comments

10 pages main content, 7 figures, 35 pages total with appendix

R2 v1 2026-07-01T09:06:52.551Z