Making Sense: Do Language Models Learn to Mean?

In the winter of 2026, Jean-Philippe Magué, a visiting researcher at CRIHN, will organize a lecture series on LLMs with four guests.

The first session, titled "Making Sense: Do Language Models Learn to Mean?", will take place on Monday, January 19, from 3:30 PM to 5:00 PM Montreal time, with the participation of Jian-Yun Nie (Computer Science, University of Montreal).

Here is the session description:

During their training, LLMs perform operations that organize their ability to generalize and produce text. Do these operations build representations that are later mobilized during inference, or is it something else? Can we say they 'make sense,' and by what criteria? In light of various linguistic and cognitive theories, we will discuss the most appropriate level of description for understanding how these models elaborate, manipulate, or simulate meaning. (Zoom link)