Abstract

Abstract:

As we continue to push the boundaries of natural language processing, it becomes imperative to better understand how language models interact with, incorporate, and are influenced by knowledge. In this talk, I will navigate the complex interaction between language models and knowledge.

I’ll begin by presenting a forthcoming ACL paper, studying fact verification against knowledge graphs. This demonstrates the ability to ground language model outputs against structured information sources. Following this, I’ll discuss ongoing research in multi-hop multi-set retrieval settings, and then explore how the abundance of knowledge available can potentially undermine retrieve-and-reason architectures. Lastly, I’ll put forward a hypothesis about the role of knowledge in language models and consider the potential advantages of combining generative and retrieval-based NLP approaches.

Bio:



James is Assistant Professor at the KAIST Graduate School of AI, South Korea, working on large-scale and knowledge-intensive natural language understanding. James recently completed his PhD at the University of Cambridge where he developed models and methods for automated fact verification and correction.