For decades, cognitive psychologists and linguists have studied language development by testing theories of learning and development in highly controlled behavioral experiments. Much has been learned from this approach. However, Big Data and computational models allow us to investigate language development in a radically different way: by collecting large datasets of actual speech to children, and simulating what a child can learn from that input with computational models that instantiate specific theories of brain organization and neural processing. When we do this, we learn many exciting facts about the nature of learning and the representation of knowledge, One particularly notable discovery is that the world around us is much more organized and structured than we previously believed. This approach also allows us to compare the way that humans learn, with state-of- the-art machine learning algorithms, providing fruitful insights into both domains.

Prof. Jon Willits