A dense corpus study of past tense and plural overregularization in English

In the "blocking-and-retrieval-failure" account of overregularization (OR; G. F. Marcus, 1995; G. F. Marcus et al., 1992), the claim that a symbolic rule generates regular inflection is founded on pervasively low past tense OR rates and the lack of a substantive difference between past tense and plural OR rates. Evidence of extended periods of OR in the face of substantial correct input (M. Maratsos, 2000) and of an initial period in which nouns are more likely to be overregularized than verbs (V. A. Marchman, K. Plunkett, & J. Goodman, 1997) casts doubt on the blocking account and suggests instead an interplay between type and token frequency effects that is more consistent with usage-based approaches (e.g., J. Bybee, 1995; K. Köpcke, 1998; K. Plunkett & V. Marchman, 1993). However, previous naturalistic studies have been limited by data that account for only 1–2% of child speech. The current study reports analyses of verb and noun ORs in a dense naturalistic corpus (1 child, 2;00.12–3;11.06 [years;months.days]) that captures 8–10% of child speech and input. The data show (a) a marked difference in verb and noun OR rates; (b) evidence of a relationship between relative regular/irregular type frequencies and the onset and rate of past tense and plural ORs; (c) substantial OR periods for some verbs and nouns despite hundreds of correct tokens in child speech and input; and (d) a strong negative correlation between input token frequencies and OR rates for verbs and nouns. The implications of these findings for blocking and other accounts of OR are discussed