Abstract
Experimental evidence from picture-naming tasks suggests that lexical access in speech production (lexicalisation) occurs in two non-overlapping stages. Semantic information is used to access an abstract lexical form; only when this stage is complete does phonological realisation begin. Such experimental data are interpreted as evidence against connectionist models of lexicalisation. This paper argues that, counterintuitively, connectionist models are not inconsistent with these data. This proposal is supported by appropriate simulations. Interactive activation models of speech production have the additional advantage of accounting for speech error and other data.