In this work package, we constructed a new computational model that explains how children discover new words and become faster at recognising these words across development. When listening to a conversation in an unfamiliar language, we find it difficult to pinpoint the individual words as there are no reliable acoustic boundaries that separate them in fluent speech. Infants learning their first language face a similar problem. The speech they hear is mainly produced as a continuous speech stream without any pauses. Yet, regardless of the specific languages they are learning, infants still manage to build a lexicon and gradually become productive language users. We demonstrated that children could discover words using a learning mechanism called chunking, where sequences of linguistic elements are grouped into units. We also suggest that these chunks have different strengths that determine how long it takes children to recognise words. We implemented this theory as a computational architecture called CIPAL and simulated how children learn words in 15 different languages. We have also used this model to simulate the behavioural patterns observed in language development experiments with artificial languages. Consequently, we have a deeper understanding of children’s vocabulary development and a new computational model for exploring language learning

Project Team

Fernand Gobet (Lead), Andrew Jessop, Julian Pine, Evan Kidd, Padraic Monaghan and Caroline Rowland

Duration

September 2020 – Feb 2024

Project Number

3.2

Key Outputs

Jessop, A., Pine, J., & Gobet, F. (2025).Chunk-based incremental processing and learning: An integrated theory of word discovery, implicit statistical learning, and speed of lexical processing. Psychological Review. Advance online publication.

Related