LExT: Towards Evaluating Trustworthiness of Natural Language Explanations

ACM Conference on Fairness, Accountability, and Transparency (FAccT 2025), 2025

PDF

Krithi Shailya, Shreya Rajpal, Gokul S. Krishnan, and Balaraman Ravindran.

LExT evaluates the trustworthiness of LLM-generated explanations through faithfulness and plausibility across six medical datasets. The work identifies important shortcomings in the reliability of natural-language explanations.