Sofia Gavasi

  Faculty of Science and Engineering| Bachelor Data Science and Artificial Intelligence

Towards Explainable AI: Reasoning With Controlled Natural Language

Sofia's elevator pitch
While traditional logic syntax is hard to follow, structured explanations in natural language offer greater clarity and make the underlying reasoning accessible to users who are unfamiliar with formal proof systems. This thesis presents a system that carries out logical reasoning in controlled English, combining a semantic-tableau-based solver with a tailored CNL and a detailed linguistic preprocessing pipeline. Users can enter premises and a conclusion, and the system generates a step-by-step proof tree that shows exactly how each inference is derived or why a branch fails. Experiments on syllogisms, synthetic datasets, and logic puzzles, together with a user study, show that natural-language explanations make the reasoning process more understandable and trustworthy than symbolic proofs.