Search papers, labs, and topics across Lattice.
This paper investigates the phenomenon of completeness-sensitive negative reasoning in large language models (LLMs), focusing on how these models handle queries about absent information. Through the introduction of CROWN-QA, which includes both a controlled synthetic dataset (CROWN-Synth) and a real-document evaluation (CROWN-Real), the authors reveal that LLMs struggle with stable closure judgments, often over-committing to negative answers when evidence is incomplete. The findings highlight a significant asymmetry in the models' reasoning, where they mischaracterize evidence coverage, leading to unreliable distinctions between justified negative answers and cases of insufficient evidence.
LLMs frequently misjudge evidence coverage, leading to a staggering rate of over-closure in negative reasoning tasks.
Large language models (LLMs) are often asked whether something is absent from a record, list, or retrieved context. Yet non-observation licenses a negative answer only when evidence completely covers the query scope; otherwise, the answer should remain unknown. We call this completeness-sensitive negative reasoning. We introduce CROWN-QA, comprising CROWN-Synth, a controlled paired core that fixes the question and observed facts while varying only query-relative coverage, and CROWN-Real, a real-document contrast-set evaluation with controlled coverage variants. Across three LLM families, models show unstable closure judgments and substantial over-closure, failing to reliably distinguish a justified negative answer (Certified-Negative) from insufficient evidence (Unknown). The dominant CROWN-Synth failure is asymmetric: models often recognize implicitly complete evidence yet treat implicitly partial evidence as query-covering. Prompting redistributes errors between over- and under-closure rather than consistently resolving them. Structured certificate elicitation traces many errors to evidence-coverage mischaracterization. CROWN-Real shows that the core partial-coverage asymmetry persists on real-document content, while its strength and the balance between over- and under-closure vary by model, prompt, and source.
