We'll be watching a recording of Dr. Potts' LSA 2025 plenary address.
 

In his plenary address, he investigates how large language models (LLMs) exhibit sophisticated linguistic behaviors and the abstract mechanisms that enable these capabilities.

Dr. Potts also delved into the powerful framework of causal abstraction to explore how LLMs represent and process linguistic structures, such as morphological agreement, negation, and long-distance dependencies. Using the English Preposing in PP (PiPP) construction as a case study, he will demonstrate how LLMs capture intricate linguistic patterns, even in rare constructions. This analysis will provide insights into how LLMs can contribute to linguistic theory and deepen our understanding of the inductive biases that enable models to learn complex language phenomena.

Zoom: https://cuboulder.zoom.us/j/97014876908