Eye movements were monitored as French participants followed spoken instruc
tions to use a computer mouse to click on one of four displayed pictures. E
xperiment 1 demonstrated that, in the absence of grammatical gender in the
context preceding the referent name [e.g., cliquez sur les boutons (click o
n the((plural neut.)) buttons((masc.)))], participants fixated pictures wit
h names sharing initial sounds with the target [e.g., bouteilles (bottles((
fem.)))] more than on pictures with phonologically unrelated names, replica
ting "cohort" effects previously found with this paradigm. When a gender ma
rked article immediately preceded the noun [e.g., cliquez sur le bouton (cl
ick on the((masc.)) button)], the early activation of the gender-inconsiste
nt cohort was completely eliminated (Experiment 2). This demonstrates that
the set of candidates initially considered for recognition of the noun is c
onstrained by the gender-marked article. Two alternative accounts of these
results, one based on grammatical level of processing and the other based o
n form-based statistics, are discussed. (C) 2000 Academic Press.