Cohort theory in spoken-word recognition assumes that a cohort of word
candidates consistent with incoming sensory information is activated
implicitly as a spoken sound stimulus unfolds over time. Five experime
nts examined implications of this internal-generation-of-words mechani
sm. In Experiments 1 and 2, a ''base'' word was disqualified (the sens
ory information was no longer consistent with the word) either early o
r late in the presentation of a spoken stimulus. On a later recognitio
n-memory test, significantly more false-positive errors occurred to ba
se words following presentations of study items that had late, compare
d to early, disqualification points. Experiments 3-5 tested whether th
is phenomenon could be accounted for in terms of overlapping features
between non-word stimuli and their base words or in terms of a post-id
entification processing mechanism. Experiment 3 replicated Experiments
1 and 2, and demonstrated that differences in early and late disquali
fication points for non-word targets, unlike word targets, were not re
lated to false-positive recognition memory errors. The study inter-ite
m interval in Experiment 4 was reduced to 1 s to minimize the role of
post-identification processing activities, and the results for both wo
rd and non-word targets were consistent with Experiment 3. A word-asso
ciation task in Experiment 5 revealed that the late non-word derivatio
ns used in this research were on the average more effective stimuli th
an the early non-word derivations in eliciting their base words. Howev
er, even when comparisons were restricted to item sets with early and
late non-words that were equally effective in eliciting base words, fa
lse-positive recognition memory errors to target words were higher fol
lowing prior presentations of their late derived non-words than follow
ing prior presentations of their early derived non-words.