Abstract Children with language disorders are at high risk for academic failure, especially with respect to literacy. Virtually all treatment programs focus on systematically manipulating language input from parents or clinicians. The research presented in this proposal creates a theoretical framework for making detailed predictions regarding what input will be most helpful for children. We build on models of pedagogical sampling, developed in the cognitive development literature, in order to design input that is optimized for teaching language to children. In this framework, teachers optimize data for statistical learners by selecting data that maximize learners' belief in the correct hypothesis, relative to alternative hypotheses. We apply this pedagogical sampling framework to interventions for children with specific language impairment (SLI) that are designed to teach the productive use of grammatical morphemes. Aim 1 develops models of morphological and syntactic learning with memory constraints; Aim 2 validates the learning models' ability to predict intervention outcomes; and Aim 3 develops algorithms for predicting what new types of input data (e.g., specific sets of verbs) will be most effective for teaching a child that a particular grammatical morpheme is productive. In the long term, models derived in this project will facilitate the future development of systems that can give detailed recommendations for input that would be most useful to children with a variety of language profiles.
|Effective start/end date||8/5/18 → 7/31/21|
- National Institute on Deafness and Other Communication Disorders: $242,437.00
- National Institute on Deafness and Other Communication Disorders: $192,311.00