Event Transition in Commonsense Reasoning

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Event Transition in Commonsense Reasoning"

By

WANG Weiqi

Abstract:

A long-standing area of interest in commonsense reasoning focuses on the 
transitions between stative and dynamic events. For example, "PersonX 
feels hungry, as a result, PersonX eats a hamburger" is a transition from 
a stative event to a dynamic event. While current commonsense knowledge 
bases focus on the relationship between two events, this project aims to 
explore whether existing commonsense reasoning models can identify the 
effect of a dynamic event between the transition of two stative events or 
vice versa. To provide commonsense reasoning models with the data to 
reason upon, we annotate a high-quality dataset that focuses on the 
transition between events based on a popular commonsense knowledge base, 
ATOMIC. Then, commonsense reasoning models are asked to predict whether 
the transition between events was correct or not. Results show that 
current state-of-the-art models cannot achieve satisfactory accuracy in 
such tasks, indicating that event transition prediction is still a 
critical task to be solved by machines.


Date            : 7 May 2022 (Saturday)

Time            : 10:40-11:20

Zoom Link       : https://hkust.zoom.us/j/6761083097

Meeting ID      : 676 108 3097

Advisor         : Dr. SONG Yangqiu

2nd Reader      : Prof. ZHANG Nevin Lianwen