Adult language knowledge is organized around different levels of linguistic structure: sound categories, words, sub-word meaningful units, and grammatical rules for combining words into utterances. While it is widely acknowledged that learning at each level of linguistic representation depends on learning at other levels, the precise nature of these interactions, as well as the learning mechanisms, remain unknown. These interactions may support ?virtuous cycles? where children's inferences build upon one another iteratively; the interactions may also introduce vectors for cascading failures, where delays in learning at one level of representation have wide-reaching effects on learning. As such, characterizing how these learning processes interact is a critical step towards understanding early language acquisi- tion, itself a determining factor in early literacy, and in turn early educational attainment. The proposed project focuses on the acquisition of the English regular plural, an abstract aspect of language structure which implicates learning at all of these levels, and to which children exhibit an early sensitivity. Focusing on a carefully selected test case allows for exhaustive experimental characteriza- tion (Speci?c Aim 1) and the development of models that capture the interactions between children's growing knowledge of sounds, meaning, and the grammar (Speci?c Aim 2). As such, this research constitutes a ?rst step towards a broad-coverage mechanistic model of the interactions between learn- ing processes in early language acquisition. Early language processing abilities are known to be key contributors to literacy and educational attainment, with subsequent effects for health and well-being. By providing an enhanced causal model of children's language early language attainment, this research will help researchers develop better diagnostics and design improved interventions. The fellowship training plan presented here focuses on 1) deepening the applicant's knowledge of psycholinguistic modeling techniques and 2) teaching the applicant to run behavioral experiments with toddlers. The research will be performed at two site: the trainee will work on computational modeling at MIT (primary host institution, approximately 2/3 of the Fellowship duration), and conduct behavioral experiments at Duke (approximately 1/3 of the Fellowship duration). This multi-institution setup will allow the trainee to work with a mentorship team with the specialized knowledge, equipment, and facilities needed to achieve the project goals, especially sophisticated modeling techniques for language learning (Dr. Levy, MIT) and experimental techniques for evaluating the language knowledge of toddlers (Drs. Bergelson and Tomasello, Duke). This combination of mentors, labs, and institutions is perfectly suited to provide the training necessary to prepare the applicant for a tenure-track professorship, while answering critical outstanding research questions.
The goal of this project is to learn how young children begin to combine information from different aspects of language in both their comprehension and production, using experiments and computational models. Focusing on a key test case, this work will help us build theories that capture the complex interactions between children's growing knowledge of sounds, words, meaning, and grammar. The results of this work will support the development of diagnostics that evaluate children at risk for language delays, and inform the design of effective interventions .