Understanding and Generating Natural Language of Complex Discourse

Join remotely via BlueJeans: https://bluejeans.com/753201752

Abstract: Letting machine understand and generate natural language is inherently difficult because language, as a highly abstract form of human knowledge, conveys not only its literal meaning, but also the intention behind it.

This talk first introduces a modular neural natural language generation framework with end-to-end learning. By separately tackling the challenges of content planning and surface realization, the model is more effective in various tasks: constructing persuasive arguments, writing Wikipedia articles, and generating scientific paper abstracts. It alleviates existing models’ issue of producing bland and incorrect text, a result of lacking global planning. I further argue that knowledge encoding is crucial to system reliability, and demonstrate how to design a knowledge graph representation and a question answering reward that work in tandem to promote high-level content understanding.

The second part of this talk highlights the challenges in understanding complex discourse, such as argumentation. Specifically, in the realm of debates, I will show that the inherent persuasive strengths of different topics and the effects of linguistic styles are intertwined, and both affect debate audience. On Oxford-style debates, our model that captures them jointly predicts audience-adjudicated winners with significantly higher accuracy than models using linguistic features alone. These learnings further motivate new designs of generation systems.

I will conclude this talk by discussing future directions, including user-guided generation, human-robot communication enhancement, and large-scale multilingual media discourse analysis.

Bio: Lu Wang is an Assistant Professor in Khoury College of Computer Sciences at Northeastern University since 2015. She received her Ph.D. in Computer Science from Cornell University and her bachelor degrees in Intelligent Science and Technology and Economics from Peking University. Her research focuses on designing machine learning models for natural language processing tasks, including language generation, abstractive text summarization, argument mining, discourse analysis, and their applications in computational social science (e.g. detecting media bias and polarization). Lu received an outstanding paper award at ACL 2017 and a best paper nomination award at SIGDIAL 2012. Her group’s work is funded by National Science Foundation (NSF), Intelligence Advanced Research Projects Activity (IARPA), and several industry gifts.

11:30-Noon: Round Table with Grad Students (join via BlueJeans seminar link)