Acceptability ratings of regular and irregular past-tense forms: Evidence for a dual-system model of language from word frequency and phonological neighbourhood effects

What are the computational and representational bases of the mental lexicon of words, and of the rules of grammar which productively combine lexical f orms into larger words, phrases and sentences? "Dual-system" theories posit that lexical forms with non-compositional (arbitrary) sound-meaning pairin gs are stored in memory, whereas compositional structures are subserved by a distinct rule-processing system. "Single-system" theories claim that lexi con and grammar are both subserved by a single associative memory. Investig ations of English past tense may help to resolve this controversy. On the d ual-system view, irregular past-tense forms (e.g. blow-blew) are retrieved from memory, whereas regular past-tense forms (e.g. walk-walked) are produc ed by the application of an -ed-suffixation rule. On the single-system view , both types of past-tense forms are learned and computed in associative me mory. To test these competing theories, acceptability ratings were elicited from native English-speaking adults for regular and irregular past-tense f orms, and their stems, in sentence contexts. Partialling out stem ratings, ratings of irregular past-tense forms (blew) correlated with their frequenc ies and with measures of the number of similar-sounding irregular verbs (th rew, grew), whereas ratings of regular part-tense forms (walked) did not co rrelate with their frequencies or with measures of the number of similar-so unding regular verbs (stalked, balked). The results suggest that irregular past tenses are retrieved from associative memory, whereas regular past ten ses are produced by a suffixation rule.