@inproceedings{uceda-sosa-etal-2026-scoped,
title = "Scoped Semantic Invariance as a Foundation for {LLM} Reliability",
author = "Uceda-Sosa, Rosario and
Chang, Maria and
Natesan Ramamurthy, Karthikeyan and
Singh, Moninder",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2545/",
pages = "57863--57884",
ISBN = "979-8-89176-459-0",
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.\par We operationalize this framework using knowledge graphs (KGs), which act as explicit semantic control structures defining local semantics 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 recur across models and domains.\par We further show that selective graph grounding improves consistency with the KG specification and cross-model agreement, while indiscriminate grounding introduces regressions. A layered grounding strategy achieves the strongest monotonicity profile, preserving KG-consistent knowledge while reducing inconsistency. Finally, we show that hotspot structures discovered from simple consistency probes transfer to more realistic compositional reasoning tasks, positioning them as a practical reliability layer for predicting and targeting failures rather than a canonical decomposition."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="uceda-sosa-etal-2026-scoped">
<titleInfo>
<title>Scoped Semantic Invariance as a Foundation for LLM Reliability</title>
</titleInfo>
<name type="personal">
<namePart type="given">Rosario</namePart>
<namePart type="family">Uceda-Sosa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Karthikeyan</namePart>
<namePart type="family">Natesan Ramamurthy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Moninder</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<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.\par We operationalize this framework using knowledge graphs (KGs), which act as explicit semantic control structures defining local semantics 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 recur across models and domains.\par We further show that selective graph grounding improves consistency with the KG specification and cross-model agreement, while indiscriminate grounding introduces regressions. A layered grounding strategy achieves the strongest monotonicity profile, preserving KG-consistent knowledge while reducing inconsistency. Finally, we show that hotspot structures discovered from simple consistency probes transfer to more realistic compositional reasoning tasks, positioning them as a practical reliability layer for predicting and targeting failures rather than a canonical decomposition.</abstract>
<identifier type="citekey">uceda-sosa-etal-2026-scoped</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2545/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>57863</start>
<end>57884</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Scoped Semantic Invariance as a Foundation for LLM Reliability
%A Uceda-Sosa, Rosario
%A Chang, Maria
%A Natesan Ramamurthy, Karthikeyan
%A Singh, Moninder
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F uceda-sosa-etal-2026-scoped
%X 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.\par We operationalize this framework using knowledge graphs (KGs), which act as explicit semantic control structures defining local semantics 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 recur across models and domains.\par We further show that selective graph grounding improves consistency with the KG specification and cross-model agreement, while indiscriminate grounding introduces regressions. A layered grounding strategy achieves the strongest monotonicity profile, preserving KG-consistent knowledge while reducing inconsistency. Finally, we show that hotspot structures discovered from simple consistency probes transfer to more realistic compositional reasoning tasks, positioning them as a practical reliability layer for predicting and targeting failures rather than a canonical decomposition.
%U https://aclanthology.org/2026.emnlp-main.2545/
%P 57863-57884
Markdown (Informal)
[Scoped Semantic Invariance as a Foundation for LLM Reliability](https://aclanthology.org/2026.emnlp-main.2545/) (Uceda-Sosa et al., EMNLP 2026)
ACL
- Rosario Uceda-Sosa, Maria Chang, Karthikeyan Natesan Ramamurthy, and Moninder Singh. 2026. Scoped Semantic Invariance as a Foundation for LLM Reliability. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 57863–57884, Budapest, Hungary. Association for Computational Linguistics.