Commonsense Reasoning in Natural Language Processing

PhD Qualifying Examination


Title: "Commonsense Reasoning in Natural Language Processing"

by

Mr. Tianqing FANG


Abstract:

Understanding commonsense knowledge in human language is one of the ultimate 
goals of artificial intelligence. To achieve this goal, commonsense reasoning 
tasks of different formats are proposed to conduct empirical studies on 
commonsense in certain domains, for example, daily events, social interactions, 
and simple physics. The mainstream commonsense reasoning tasks can be 
categorized into three types, Commonsense Question Answering (QA), Commonsense 
Knowledge Base Completion (CKBC), and Commonsense Knowledge Generation. In 
recent years, a surge of reasoning models has been proposed to tackle these 
challenges. The most popular way is to design large-scale pretrained language 
models (PTLM), where single models of PTLMs have shown great success on several 
QA benchmarks, surpassing human performance. Next is the models incorporating 
external commonsense or background knowledge resources on top of PTLMs, which 
help understand commonsense scenarios in a more explainable way. The third line 
of research is multi-task learning, where several commonsense benchmarks are 
trained in a multi-task setting to achieve better performance on single tasks 
individually. In this survey, we first introduce the important milestones in 
the field of commonsense resources including CommonSense Knowledge Bases (CSKB) 
and Knowledge Graphs (CSKG). Then we introduce important benchmarks of 
reasoning tasks categorized as Commonsense QA, CKBC, and Commonsense 
Generation. Last but not least, we will introduce state-of-the-art models that 
tackle commonsense reasoning.


Date:			Thursday, 27 May 2021

Time:                  	2:00pm - 4:00pm

Zoom meeting: 
https://hkust.zoom.us/j/94654173775?pwd=VGo5RTNVaHlIZElWVDBNSHNudkhpZz09

Committee Members:	Dr. Yangqiu Song (Supervisor)
 			Dr. Xiaojuan Ma (Chairperson)
 			Prof. Nevin Zhang
 			Prof. Xiaofang Zhou


**** ALL are Welcome ****