Modelling vocabulary acquisition in spoken word recognition

Most current models of spoken word recognition relate to the system in its "steady state" (e.g., McClelland and Elman, 1986; Norris, 1993). That is, they apply to the situation where the listener has already acquired a fully fledged list of spoken words. Even in cases where these models incorporate a learning component (e.g., Gaskell & Marslen-Wilson, 1997), the learning phase does not accurately capture the microstructure of learning- it is merely a way of setting the weights in the steady-state model based on a representation of the language environment. Recent research (e.g., Gaskell & Dumay, 2003) has emphasised the timecourse of the lexicalisation of novel words, suggesting a system of dual representation, with both fast and slow learning of different aspects of words. When we encounter a new spoken word (e.g., "cathedruke"), the phonological form is easily stored and made use of immediately, but engagement in lexical competition, as measured by its effect on the recognition of other words (e.g., "cathedral"), emerges over a longer timecourse, and possibly relies on consolidation processes related to sleep. This project will employ computational, behavioural and brain imaging methods to add an extra dimension to models of spoke