Operationalising usage-based learning: A minimal cognitive architecture approach

What cognitive mechanisms enable humans to learn complex languages?

Taking a domain-general approach inspired by artificial intelligence models of cognition, we hypothesise that a minimal cognitive architecture based on sequence memory, chunking and schematizing and driven by reinforcement learning is sufficient to learn complex grammar.

Sequence memory accounts for an accurate perception of order, chunking implements the ability of grouping units together, and schematizing allows decisions to be made based on a symbolic categorization of the units to avoid the combinatorial explosion problem inherent to language acquisition.

Key to this project is that the symbolic categories and the way they combine, i.e, the grammar, will emerge during the learning process and, thus, our model provides an operationalization of usage-based learning and emergent grammar.

Our model will be assessed on both artificial and natural linguistic input and should account for the complete process of grammar learning.

Results can generate ground-breaking progress in our understanding of the human language capacity and will have impact in the fields of language acquisition, cognition, cultural evolution, biological evolution, artificial intelligence and natural language processing.