Digital Technologies
154
3.3
Modelling complex systems
Keywords
Connectionist, corpus, chunking, neural networks,
statistical modelling
What goes on in a child’s brain when they learn language?
How do they acquire abstract language-specific rules (for
example that English verbs go before their objects, while
Japanese verbs go after their objects)?
The University is engaged in the computational modelling
of language acquisition in children. There is plenty of
evidence which suggests that language acquisition is
similar to statistical learning techniques. We apply
complex statistical ‘learners’ to corpora (a large and
structured set of text used for statistical analysis and
hypothesis testing) of child-directed speech; these
learners allow us to simulate the abilities of children at
various ages. We also apply neural network models to
artificial corpora to examine how language and meaning
interact with each other as language is learned.
Companies such as Google, Facebook, and Amazon
have developed statistical learning techniques that
have an ability to actually understand human
language. These systems are very powerful and
provide the companies with deep insights into the
behaviour of their customers, yet the analytical
engines they have developed still do not encode the
language knowledge of even an average 10 year old.
Our work can help to bridge this gap; by showing how
children acquire their language abilities our research
can also improve the language capabilities of
statistical learning systems.
Relevant centres and groups
•
Child Language Study Centre.
