View in EDS

Semantic knowledge abstraction: Consistent reasoning in large language models for natural language inference.

Saved in:
Bibliographic Details
Title: Semantic knowledge abstraction: Consistent reasoning in large language models for natural language inference.
Authors: Torres-Moreno, David1 (AUTHOR), Hermosillo-Valadez, Jorge1 (AUTHOR) jhermosillo@uaem.mx
Source: Knowledge-Based Systems. Jan2026, Vol. 332, pN.PAG-N.PAG. 1p.
Subjects: Language models, Semantics methodology, Natural language processing, Discourse analysis
Abstract: • We introduce the Semantic Knowledge Abstraction (SKA) methodological framework to discover and compensate for semantic knowledge gaps of Large Language Models (LLMs) in Natural Language Inference (NLI). • The method guides the LLM's reasoning process, revealing its weaknesses and inconsistencies, which are compensated for by explainable decision strategies that improve NLI accuracy by up to 15 % for some models. • The results show that the framework is useful for improving not-entailment inference and suggest that it could also contribute to develop more robust and reliable models in the field of Natural Language Understanding (NLU). [Display omitted] Despite their strong performance on many NLP tasks, LLMs face significant challenges in semantic abstraction. In this study, we examine how LLMs leverage abstract semantic knowledge in natural language inference (NLI), a task that requires interpreting implicit meanings, contextual conceptual relationships, and semantic connections between words and phrases. To address this, we introduce the Semantic Knowledge Abstraction (SKA) methodological framework, which constructs semantic knowledge at a higher level of abstraction, defined through semantic compatibility and incompatibility for NLI. Within this framework, lexical-semantic relations between the premise and hypothesis are reconfigured to create a more flexible semantic network, inducing alternative reasoning paths in LLMs. These pathways produce consistent response patterns, facilitating consensus on a single inference. Results indicate that SKA effectively identifies and compensates for LLMs' semantic knowledge gaps, yielding substantial accuracy improvements—exceeding 15 % for some models—particularly in the non-entailment class. SKA further enables control of external knowledge at a higher abstraction level, structures semantic relations between entities, supports focused inference through group definitions, and leverages these definitions to guide alignment and reveal prior knowledge. Collectively, these features allow LLMs to reason more flexibly and systematically, demonstrating that semantic abstraction can meaningfully enhance performance in complex NLI tasks. [ABSTRACT FROM AUTHOR]
Copyright of Knowledge-Based Systems is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:• We introduce the Semantic Knowledge Abstraction (SKA) methodological framework to discover and compensate for semantic knowledge gaps of Large Language Models (LLMs) in Natural Language Inference (NLI). • The method guides the LLM's reasoning process, revealing its weaknesses and inconsistencies, which are compensated for by explainable decision strategies that improve NLI accuracy by up to 15 % for some models. • The results show that the framework is useful for improving not-entailment inference and suggest that it could also contribute to develop more robust and reliable models in the field of Natural Language Understanding (NLU). [Display omitted] Despite their strong performance on many NLP tasks, LLMs face significant challenges in semantic abstraction. In this study, we examine how LLMs leverage abstract semantic knowledge in natural language inference (NLI), a task that requires interpreting implicit meanings, contextual conceptual relationships, and semantic connections between words and phrases. To address this, we introduce the Semantic Knowledge Abstraction (SKA) methodological framework, which constructs semantic knowledge at a higher level of abstraction, defined through semantic compatibility and incompatibility for NLI. Within this framework, lexical-semantic relations between the premise and hypothesis are reconfigured to create a more flexible semantic network, inducing alternative reasoning paths in LLMs. These pathways produce consistent response patterns, facilitating consensus on a single inference. Results indicate that SKA effectively identifies and compensates for LLMs' semantic knowledge gaps, yielding substantial accuracy improvements—exceeding 15 % for some models—particularly in the non-entailment class. SKA further enables control of external knowledge at a higher abstraction level, structures semantic relations between entities, supports focused inference through group definitions, and leverages these definitions to guide alignment and reveal prior knowledge. Collectively, these features allow LLMs to reason more flexibly and systematically, demonstrating that semantic abstraction can meaningfully enhance performance in complex NLI tasks. [ABSTRACT FROM AUTHOR]
ISSN:09507051
DOI:10.1016/j.knosys.2025.114825