Machine Learning Seminar Series Spring 2026 | Discourse Centric Models and Evaluation
All Seminars Held on Wednesdays 12pm - 1pm
Abstract: Large Language Models are powerful models of language even for long stretches of generation. This has shifted the research frontier in natural language processing to where global coherence and consistency, strategic content planning and selection, and high-fidelity evaluation play increasingly important roles. These belong to discourse-level capabilities. I will first present a model of discourse in the framework of Questions Under Discussion (QUD), which provides the necessary data structure to move beyond a local view of text generation and informs us how to strategically decide which questions are worth answering in a given context. Next, I will share how this model can be used to measure and evaluate novelty, revealing the structural homogeneity in LLM outputs even when they differ in content. Finally, I will tackle the evaluation challenge for open-ended generation, focusing on fine-grained evaluation systems that capture implicit aspects absent from user specification. In all, this talk aims to address the need for discourse-level understanding and control in today’s generative models.
Bio: Jessy Li is an associate professor in the Linguistics Department and (by courtesy) the Computer Science Department at the University of Texas at Austin. She earned her Ph.D. from the Department of Computer and Information Science at the University of Pennsylvania. Her research interests are in computational linguistics and natural language processing, specifically discourse and pragmatics, natural language generation, evaluation, and language & code. She is a recipient of an NSF CAREER Award, ACL and EMNLP Outstanding Paper Awards, an ACM SIGSOFT Distinguished Paper Award, among others. Jessy is an AI research lead at the NSF-Simons CosmicAI Institute, and an affiliated faculty member at the Oden Institute for Computational Engineering & Sciences. She is the current Secretary of NAACL and a program chair of EMNLP 2026.