ABSTRACT
The English past tense is a quasi-regular system, in that many of the irregular verbs share characteristics with regular items. Among high-frequency exceptions, in particular, several have the regular /d/ or /t/ ending but with either a reduction of the vowel (did, said) or a deletion of a stem consonant (had, made). Such forms suggest that many so-called irregular verbs reflect a joint influence of the systematic past-tense pattern captured in fully regular items together with a pressure to be short or simple. We adapt familiar neural network formalisms to show how such forms can arise if the phonological content of word forms are constrained (a) to support accurate communication of the word’s meaning and (b) to be simple.
