Gerald Penn is a Professor of Computer Science at University of Toronto, and a Fellow of Computer Science at St. Michael's College. His research interests are spoken language processing and computational linguistics.  He is a senior member of IEEE and AAAI, and a past recipient of the Ontario Early Researcher Award.  His joint work with Geoffrey Hinton and Hui Jiang on signal processing with neural networks revolutionized acoustic modelling for speech recognition systems, and received the IEEE Signal Processing Society's Best Paper Award.  He has led numerous research projects, including those funded by Avaya, Bell Canada, CAE, the Communications Security Establishment, the Connaught Fund, Microsoft, Naver Corporation, NSERC, the German Ministry for Training and Research, SMART Technologies, the U.S. Army and the U.S. Office of the Director of National Intelligence.

Do Language Models Know Language? Triumphalist portraits of large language models (LLMs) boast that language models have mastered a level of language understanding that natural language processing (NLP) researchers have laboured for years to attain using complex architectures consisting of diverse component models that each require large amounts of training data. Do they? How do we know? And, if so, how do they manage this? In this talk, we will examine some recent results in relation to LLMs that cast doubt upon these claims, while affirming the utility of LLMs in present-day NLP.