Abstract: As Large Language Models (LLMs) gain prominence in high-stakes domains, the need for transparent and reliable natural language explanations becomes increasingly critical, particularly in sensitive areas like healthcare. While existing Natural Language Generation metrics such as BLEU and ROUGE evaluate syntactic and semantic accuracy, they fail to address essential dimensions like factual accuracy, consistency, and faithfulness. In this talk, I will present a comprehensive framework, Language Explanation Trustworthiness Score (LExT), designed to evaluate the trustworthiness of model-generated natural language explanations by balancing two key aspects—Plausibility and Faithfulness. I will discuss some findings on the application of this domain-agnostic framework in the healthcare domain using public medical datasets, where we evaluated six Language Models, including both general-purpose and domain-specific models. This talk will highlight the importance of tailored evaluation frameworks for assessing natural language explanations in sensitive fields to potentially enhance the trustworthiness and transparency of LLMs in healthcare and beyond. The talk will further give a glimpse of two other works and other activities at the Centre for Responsible AI (CeRAI) at the Wadhwani School of Data Science and AI (WSAI) IIT Madras, India.

Bio: Balaraman Ravindran is the founding head of the Wadhwani School of Data Science and AI, the Robert Bosch Centre for Data Science & AI and the Centre for Responsible AI (CeRAI) at IIT Madras. He has held visiting positions at the Indian Institute of Science, Bangalore, India, the University of Technology, Sydney, Australia and Google Research. He has more than three decades of experience working in reinforcement learning, and his research interest spans responsible AI and deep RL.

He currently serves on the editorial boards of top journals and has helped organize some of the premier conferences in this area, such as AAAI 2021, 2023-24 and KDD 2023. He has published more than 150 papers in premier journals and conferences. His works with students have won multiple best paper awards, the most recent being the best application paper at PAKDD 2021.

He was elected ACM Distinguished Member (2021) for his significant contributions to computing. In 2023, he was elected a Fellow of the Indian National Academy of Engineering, the apex engineering society of India. In 2025. he was elected a Fellow of AAAI (Association for Advancement of AI) for his fundamental contributions to reinforcement learning. 

https://wsai.iitm.ac.in/~ravi/

Light refreshments following the talk.