Using Psycholinguistics to Analyze and Improve Neural Network Language Models

Abstract: Neural networks produce high-performing yet opaque solutions for language processing. Luckily, psycholinguists have long studied high-performing yet opaque language processing systems. In this talk, I will show how psycholinguistic techniques designed to probe human cognition can be applied to neural networks to better understand their learned representations. Further, I will show that psycholinguistic insights can be applied to neural networks to improve their language modeling performance.

Bio: Marten van Schijndel is an assistant professor of linguistics at Cornell University. He studies the incremental representations that humans use to process language. He also studies the linguistic representations of neural networks to identify effective solutions that could be used by humans. Marten has a PhD in linguistics from The Ohio State University and was a cognitive science postdoc at Johns Hopkins University.