| Children with SLI continue to lag behind children with TL in their grammatical development despite apparently equivalent exposures to recasts in conversation (Fey et al., 1999) and an ability to learn from recasts in intervention as quickly as children with TL (Nelson et al., 1996). To test two possible resolutions to this paradox, this experiment manipulated recast density and observed its effect on novel irregular past tense verb accuracy by 13 children (7--8 years) with specific language impairment (SLI) and 13 language similar children (5--6 years) with typical language development (TL). As predicted, children with TL produced the target verbs with greater accuracy than children with SLI when the input included conversational recast densities. Contrary to expectations, the SLI group did not improve their accuracy at higher intervention-like recast densities, and the TL group performances declined significantly. This decline in performance was consistent with the well-established phenomenon of better learning from distributed than massed practice. The SLI group's lack of improvement was attributed to a combination of a reduced working memory capacity and its resultant linguistic processing difficulties, and presentation of a high recast density that was simply too low. This experiment also compared acquisition of the target verbs when the children were exposed to conversational rates of recasts + models or models alone. The children with TL produced the verbs more accurately when recasts were included in the input, while the children with SLI did not differ in their accuracy. Analyses of the children's metalinguistic utterances about their verb learning indicated that recasts were more usable than models. Both groups used recasts to inform their hypotheses about the past tense forms; however, the two groups showed different patterns of this use. Together these results refuted one possible resolution to the paradox: that recasts are no more facilitative than any other type of input. They do not, however, provide definitive support for a second resolution: Children with SLI learn from recasts less efficiently than children with TL. This would require children with SLI to benefit from rates of recasts that were higher than those used in this study. |
