Research project Modelling usage-based language learning
Foto: Anna Jon-And, DALL-E (AI)
Anna Jon-And and Jérome Michaud (Mälardalens universitet) are investigating the human language ability by creating a minimal AI model and testing its capacity to learn grammar.
What cognitive mechanisms enable humans to learn language? Based on a holistic view of human cognition, our hypothesis is that a minimal artificial intelligence based on sequence memory, chunking, and generalization can learn grammar. Sequence memory is crucial because recent research shows that animals cannot perceive exact word order, while word order is significant in language. Chunking, which is the ability to mentally group several units into one, is important for human information processing, language comprehension, and language acquisition. Generalization is a central requirement for learning in both humans and other animals. Our learning model's task is to find sentences in a written stream of words where cues like punctuation and capitalization have been removed. By exploring and discovering the categories and combinations of words that facilitate the task, we expect the model to autonomously discover and build a functional grammatical system. In this way, we hope to demonstrate how language acquisition can occur with just a few very general mental abilities and that the language learning process does not require any more specific genetic guidance.
Contact:
Anna Jon-AndFunding: Swedish Research Council, VR 2022-02737
Project members:
Anna Jon-And(Fil dr),
Jérome Michaud(Fil dr)
Mälardalens universitet