Softening Ontological Reasoning with Large Language Models

Abstract

Logic-based Knowledge Graphs (KGs) and Knowledge Representation and Reasoning (KRR) have emerged as fundamental methodologies in many data-intensive areas, fostering trust and accountability for effective decision-making. However, the knowledge captured by such approaches is often restricted by the rigidity of their structured rule-based formalisms. More recently, the rising adoption of Large Language Models (LLMs) has introduced a new layer of semantic understanding and flexibility in human-data interaction. Yet, these models are inherently limited in reasoning capabilities and lack systematic and explainable outcomes due to their opaque nature. To address today’s challenge of combining the strengths of both technologies, we propose a novel neurosymbolic solution that leverages the power of LLMs to “soften” rule activations, enhancing adaptability in ontological reasoning while preserving robustness and transparency of KRR systems.