Understanding Understanding — in General, and in Large Language Models

David Schlangen  : Professor, “Foundations of Computational Linguistics”, University of Potsdam

Abstract:
In the first part of the talk I will present an overview of the activities of my group (Foundations of CompLing at Uni Potsdam), organised around the quest to understand how linguistic communication can yield shared understanding. I will present an analytical framework in which similarities and differences in method and approach between text NLU (natural language understanding) and SLU (situated language understanding) can be described.
In the second part, and motivated by this framework, I will present a new computational framework for the evaluation of “chat-optimized LLMs” (cLLMs) like ChatGPT for their capabilities in task-oriented interaction.