Towards Explainable AI: Reasoning With Controlled Natural Language

Sofia Gavasi, Nico Roos

Track: Type A (Regular Papers)

Keywords: Explainable AI · Reasoning · Controlled Natural Language · Natural Language Inference · Semantic Tableau.

Abstract: This paper investigates how formal logical reasoning can be made accessible and explainable through natural language. It introduces a system that integrates the semantic tableau method with a Controlled Natural Language, enabling inference over structured English inputs while preserving logical rigor. The system performs syntactic and semantic preprocessing, classifies sentence structures, and applies adapted tableau rules to construct visual proof trees. It supports extended reasoning patterns, including the handling of explicit exceptions. Experimental evaluation on logic tasks, synthetic datasets, and user studies demonstrates that the system can reason accurately within its constraints and improve user comprehension and trust through trans- parent explanations. These results highlight the potential of logic-based, linguistically grounded systems for advancing explainable AI.

Serve As Reviewer: ~Annette_Ten_Teije2, ~Leon_Van_der_Torre2, ~Frank_Van_Harmelen2, ~Cees_Witteveen1, ~Bart_Verheij1

Submission Number: 5