Useful evidence on negative evidence

J. L. Morgan, K. M. Bonamo, and L. L. Travis (1995)applied synchronous and time-series regression techniques to observational data to detect effects of recasted error correction on children's emerging grammar. Results showed that recasts did not facilitate learning but actually impeded it. In this study, a formal modeling procedure was used to generate similar time series with known, underlying learning relations. The regression procedures used by Morgan et al. could not discriminate between the data generated by models in which recasts (a) totally determined grammatical learning, (b) supplemented other learning, (c) inhibited learning, or (d) had nothing to do with grammatical learning. These modeling results and review of statistical and conceptual problems indicate that Morgan et al.'s analyses (short term as well as long term) are uninterpretable as regards the role of recasts or any form of negative evidence in children's syntactic acquisition. However, their descriptive data confirm many prier reports on when parents use recasts and raise interesting questions about possible bidirectional influences between adult recasts of a syntactic structure and the acquisition of that structure.