A fundamental unsolved question is what cognitive mechanisms enable humans to learn grammar. Recent research in comparative cognition points out sequence memory as central for human mental abilities. In this talk Anna Jon-And, CEK, will present joint work with Jérôme Michaud, where they build a cognitive architecture aiming at accounting for language acquisition from a domain-general perspective. They hypothesize that a minimal model based on sequence memory, chunking and schematizing, driven by reinforcement learning, is sufficient to learn complex grammar from exposure to natural language input. She will present the results of a pilot model containing only sequence memory and chunking. In simulations, the model turns out to be successful at the task of identifying sentences in small artificial languages, indicating that it is a good starting point for an extended architecture. She will discuss the potential of the model to account for the emergence of linguistic structure as a self-organizing solution to the combinatorial explosion problem inherent to language acquisition.
The webinar will be presented at the Uppsala Working Group on Empirical linguistics.