Connectionist networks have been used to model a wide range of cognitive ph
enomena, including developmental, neuropsychological and normal adult behav
iours. They have offered radical alternatives to traditional accounts of we
ll-established facts about cognition. The primary source of the success of
these models is their sensitivity to statistical regularities in their trai
ning environment. This paper provides a brief description of the connection
ist toolbox and how this has developed over the past 2 decades, with partic
ular reference to the problem of reading aloud.