LLM-Guided Semantic Bootstrapping for Interpretable Text Classification with Tsetlin Machines

Abstract

This work proposes an LLM-guided semantic bootstrapping framework that distills semantic knowledge from large language models into interpretable feature spaces for Tsetlin Machines. The approach enables transparent, logic-based text classification that retains competitive accuracy while providing human-readable explanations, bridging the gap between the capability of LLMs and the interpretability of propositional-logic learners.

Publication
Findings of the Association for Computational Linguistics: ACL 2026
Yuangang Li
Yuangang Li
PhD Student at UCI