arXiv cs.AI· Rosario Uceda-Sosa, Maria Chang, Karthikeyan Natesan Ramamurthy, Moninder Singh·· 12 小时前AI 评分36
以范围化语义不变性作为 LLM 可靠性基础
Scoped Semantic Invariance as a Foundation for LLM Reliability
AI 导读
研究将 LLM 对语义等价查询的回答一致性形式化为语义不变性，并证明任意保义变换下的不变性一般不可判定，由此提出限定语义域的范围化语义不变性框架。基于 MITRE D3FEND 本体和八个 Wikidata 衍生领域，不一致性呈结构化分布，集中于围绕 TREE、STAR、CYCLE 模式组织的局部热点子图，且跨模型与领域稳定。
正文
Abstract:LLMs often answer semantically equivalent queries inconsistently, even when the underlying facts are known. We formalize consistency as semantic invariance over equivalence classes of queries and show that invariance under arbitrary meaning-preserving transformations is undecidable in general. This motivates scoped semantic invariance, where model behavior is evaluated relative to bounded semantic domains.
We operationalize this framework using knowledge graphs (KGs), which act as explicit semantic control structures defining local semantic constraints and grounding contexts. Across the MITRE D3FEND ontology and eight Wikidata- derived domains, we show that inconsistencies are structured rather than random, concentrating in localized hotspot subgraphs organized around recurring TREE, STAR, and CYCLE patterns that remain stable across models and domains.
We further show that selective graph grounding improves correctness and cross-model consistency, while indiscriminate grounding introduces regressions. A layered grounding strategy achieves the strongest monotonicity profile, preserving correct knowledge while reducing inconsistency. Finally, we show that hotspot structures discovered from simple consistency probes transfer to more realistic compositional reasoning tasks.
| Comments: | Accepted at EMNLP 2026 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.33267 [cs.AI] |
| (or arXiv:2610.33267v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.33267 arXiv-issued DOI via DataCite |
Submission history
From: Rosario Uceda-Sosa PhD [view email]
[v1]
Fri, 23 Oct 2026 20:26:21 UTC (4,590 KB)
来源：arXiv cs.AI · arxiv.org