Jérôme Michaud: A reinforcement learning approach to grammar induction

Abstract: Grammar induction aims at extracting a grammar from direct exposition to linguistic data. In this talk, I will present a reinforcement learning model of grammar induction that relies on cognitively plausible mechanisms such as sequence memory and chunking. Using a sentence segmentation task, I will show that our model successfully navigates varying degrees of linguistic complexity, exposing efficient adaptation to combinatorial challenges through the reuse of sequential patterns. This is a joint work with Anna Jon-And (SU) funded by the VR project: “Operationalising usage-based learning: A minimal cognitive architecture approach”