Speech Science Forum – Marvin Lavechin
Artificial neural networks to analyze and simulate language acquisition in children

Abstract
Lightweight child-worn recorders that collect audio across an entire day allow for a big-data approach to studying language development. By collecting the child's production and linguistic environment, these recordings offer us a uniquely naturalistic view of everyday language uses. However, such recordings quickly accumulate thousands of hours of audio and require the use of automatic speech-processing algorithms, which will be the focus of the first part of my talk. In addition to providing ecologically valid measures of what children hear and say, these recordings can fuel computational models of early language acquisition with what infants truly hear. This opens up new opportunities for running realistic language learning simulations, to which I will dedicate the second part of my talk.
My work is at the intersection of Artificial Intelligence and Cognitive Sciences. In my research program, I develop speech-processing applications for the study of language development. In particular, my research articulates along two major axes: 1) the development and democratization of artificial neural networks to automatically analyze child-centered recordings (supervised learning) and 2) language learning simulations evaluated on tasks inspired by psycholinguistics (unsupervised learning).