Abstract
Question Generation (QG) concerns the task of automatically generating questions from various inputs such as raw text, database, or semantic representation. People have the ability to ask deep questions about events, evaluation, opinions, synthesis, or reasons, usually in the form of Why, Why-not, How, What-if, which requires an in-depth understanding of the input source and the ability to reason over disjoint relevant contexts. Learning to ask such deep questions has broad application in future intelligent systems, such as dialog systems, online education, intelligent search, among others. This talk will introduce our recent research on generating deep questions that demand high cognitive skills, including questions that require multi-hop reasoning and questions that exhibit certain human-desired properties, such as being answerable by the passage. We will also introduce one practical application of deep QG: how to generate synthetic multi-hop questions in an unsupervised way to improve the performance of multi-hop question answering.
Publications
- Unsupervised Multi-hop Question Answering by Question. NAACL, 2021.
- Exploring Question-Specific Rewards for Generating Deep Questions. COLING, 2020.
- Semantic Graphs for Generating Deep Questions. ACL, 2020.
Resources
- Recent advances in neural question generation. ArXiv, 2019.
- Question Generation Paper List: https://github.com/teacherpeterpan/Question-Generation-Paper-List
Members
- Liangming Pan (PhD Student)
- Yuxi Xie (PhD Student)
- Hengchang Hu (PhD Student)
- Yan Meng (MComp Student)
- Min-Yen Kan (Professor and Advisor)