It is often assumed that children go through a stage in which they sys
tematically overapply irregular past tense patterns to inappropriate v
erbs, as in wipe-wope, bring-brang, trick-truck, walk-has walken. Such
errors have been interpreted both as reflecting over-use of minor gra
mmatical rules (e.g. 'change i to a'), and as reflecting the operation
of a connectionist pattern associator network that superimposes and b
lends patterns of various degrees of generality. But the actual rate,
time course, and nature of these errors have never been documented. We
analysed 20,000 past tense and participle usages from nine children i
n the CHILDES database, looking for overapplications of irregular vowe
l-change patterns, as in brang, blends, as in branged, productive suff
ixations of -en, as in walken, gross distortions, as in mail-membled,
and double-suffixation, as in walkeded. These errors were collectively
quite rare; children made them in about two tenths of one per cent of
the opportunities, and with few stable patterns: the errors were not
predominantly word-substitutions, did not occur predominantly with irr
egular stems, showed no consistency across verbs or ages, and showed n
o clear age trend. Most (though not all) of the errors were based clos
ely on existing irregular verbs; gross distortions never occurred. We
suggest that both rule-theories and connectionist theories have tended
to overestimate the predominance of such errors. Children master irre
gular forms quite accurately, presumably because irregular forms are j
ust a special case of the arbitrary sound-meaning pairings that define
words, and because children are good at learning words.