When speakers produce words, lexical access proceeds through semantic and p
honological levels of processing. If phonological processing begins based o
n partial semantic information, processing is cascaded; otherwise, it is di
screte. In standard models of lexical access, semantically processed words
exert phonological effects only if processing is cascaded. In 3 experiments
, speakers named pictures of objects with homophone names (ball), while aud
itory distracter words were heard beginning 150 ms prior to picture onset.
Distracters speeded picture naming (compared with controls) only when relat
ed to the nondepicted meaning of the picture (e.g., dance), exhibiting an e
arly phonological effect, thereby supporting the cascaded prediction. Distr
acters slowed picture naming when categorically (e.g., frisbee) related to
the depicted picture meaning, but not when associatively (e.g., game) relat
ed to it. An interactive activation model is presented.