Modelling Performance in Tests of Spoken Language