Towards open-domain event understanding

Events are ubiquitous in our lives: from our everyday routines to yearly resolutions, from personal affairs to national emergencies, "the only thing constant is change itself''. An event is essentially a change in the state of an object, described by an event type and its arguments. Conventionally, in the natural language processing space, events are grounded in event mentions that are on the word or phrase level. This is in fact, a pigeonholed view of events from three different aspects: (1) we only look at sentences without having the global context in mind; (2) we only care about events that are defined in the ontology; (3) we overlook relationships between event mentions. In this dissertation, we aim to improve the coverage, scope, and structure aspects of event extraction and understanding, by studying how to effectively and efficiently extract events, which events to extract, and how to obtain generalized event structure knowledge. Compared to prior work, we focus on enabling systems that are open-domain and can utilize wider context. Eventually, we showcase how multiple systems can collaborate to enable event forecasting and simulation.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Li, Sha

Contributors dc:contributor

• Ji, Heng
• Han, Jiawei
• Zhai, Chengxiang
• Callison-Burch, Chris

• Event Extraction
• Event Detection
• Event Simulation
• Few-shot Learning
• Semi-supervised Learning

Statement dc:rights

• Copyright 2024 Sha Li

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/127272

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F127272