Yingming Wang

Extensive experiments across multiple datasets and models demonstrate the effectiveness of SR-NLE in improving explanation faithfulness.

SR-NLE leverages two feedback strategies — natural language self-feedback and attribution-based important-word feedback — to guide the refinement process.

🚀 Thrilled to share that our paper "Self-Critique and Refinement for Faithful Natural Language Explanations" has been accepted to EMNLP 2025 Main! Big thanks to my amazing advisor, @atanasovapepa, for her guidance and support during this work! arxiv.org/abs/2505.22823