Events Calendar

PhD Thesis Proposal (I-Hung Hsu)

Committee Members:
Dr. Prem Natarajan (Chair)
Dr. Nanyun Peng (Co-Chair)
Dr. Dan O'Leary
Dr. Emilio Ferrara 
Dr. Muhao Chen  
Title: Data-efficient event understanding in natural languages.  
 
Abstract:
Many natural languages in the world, such as news or narratives, are composed based on events. By focusing on events, NLP systems can better grasp the plot, infer motivations, consequences, and the dynamics of situations described in text. Despite the rapidly evolving landscape of NLP technology, the challenge of understanding complex events remains significant and usually relies on a large amount of annotated data. In the proposed thesis, we explore and invent algorithms to enhance the efficiency of event understanding in natural languages with minimal data requirements. Many natural languages, such as those found in news or narratives, are structured around events. NLP systems, by concentrating on these events, can more effectively comprehend the narrative, deducing motivations, outcomes, and the dynamics of described situations. Despite rapid advancements in NLP technology nowadays, comprehending complex events remains a formidable challenge, still largely dependent on extensive annotated data. This thesis aims to develop algorithms that enhance the understanding of events in natural languages while minimizing data requirements. We begin by introducing a novel event extraction approach, treating it as a controlled text generation problem. This method leverages indirect supervision from natural language generation to event extraction, facilitating more efficient data learning. We further explore methods to integrate external knowledge into our approach through knowledge-aware prefixes. Finally, we extend our investigation to cross-lingual understanding, broadening the technology's applicability across multiple languages.      

Location: Ronald Tutor Hall of Engineering (RTH) - 306

Audiences: Everyone Is Invited

Contact: CS Events

Event Link: https://usc.zoom.us/j/6139565235