Distributional patterns in recast therapy do not systematically align with patterns in conversational data.

Recast therapy is an intervention for morphosyntactic targets but the demanding nature of this child-centered input raises questions about whether the target exemplars differ from their instantiation in naturalistic child-directed speech. We analyzed a corpus of two infrequent structures - object relative clauses (ORC) and passives - produced by 4 clinicians and 9 caregivers (20 dyads) conducting recast therapy to compare performance across groups and utterance types and then compare these findings to distributional patterns reported in the psycholinguistic literature. As compared to caregivers, clinicians overwhelmingly produced more recasts of all kinds and included greater structural variety in their recasts. Differences in the degree, but not direction, were observed for noun phrase types in specific locations in ORCs and passives. Compared to attested patterns in the literature, recasts matched animacy patterns and passive auxiliary choice, but were much more likely to include a relativizer or a by-phrase (i.e., full passives) and place ORCs in sentence-initial positions. Recasting noncanonical frames may lead to input that is neither aligned with conversational patterns nor similar to items from common assessment tools. Consideration of the role that recast rate (dose), variety, and alignment play in learning should drive decisions about intervention approaches and providers.