The last few years have seen a substantial shift in research focused on Large Language Models (LLMs), with steady advancements in the field. LLMs excel at extracting information from text and images and generating seemingly novel content. However, many researchers nowadays agree that their performance - and that of neural networks (NNs)-based systems in general - could be significantly improved by integrating logical reasoning into the training pipeline. This integration has the potential to address some of the limitations of NNs, such as inconsistency, lack of interpretability, and weak reasoning capabilities.

Logical Reasoning has a long-standing history in Artificial Intelligence (AI) research with various approaches and semantics for deduction, induction, and abduction, and plays a crucial role in the long-term goal of Neuro-Symbolic Artificial Intelligence. While several works exist on the integration of knowledge into general NNs, as of today, few works explicitly target LLMs, presenting a promising path for future research and development.

This Research Topic aims at gathering novel approaches for integrating LLMs and logical reasoning, combining the strength of the two paradigms. Existing logic-based tools can help at overcoming ‘hallucinations’ of LLMs by guiding the reasoning process. This integration will allow more accurate results and confident predictions. This article collection can also serve as a foundation for future research on the rapidly evolving field of Neuro-Symbolic integration.

Topics of interest include, but are not limited to:
• Integration of reasoning systems into LLMs
• Verification of LLMs outputs
• Logic-based approach for knowledge injection
• LLM-guided logical reasoning
• Logical Rules extraction from LLMs
• Counterfactual reasoning with LLMs
• Novel applications and case studies of LLM and Logic

Are you interested in contributing a manuscript to this Research Topic?

Exploring Logic in Large Language Models