Stronger Transformers for Neural Multi-Hop Question Generation

Prior work on automated question generation has almost exclusively focused on generating simple questions whose answers can be extracted from a single document. However, there is an increasing interest in developing systems that are capable of more complex multi-hop question generation, where answer...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2020-10
Main Authors: Devendra Singh Sachan, Wu, Lingfei, Sachan, Mrinmaya, Hamilton, William
Format: Article
Language:English
Subjects:
Foundation construction
Questions
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Prior work on automated question generation has almost exclusively focused on generating simple questions whose answers can be extracted from a single document. However, there is an increasing interest in developing systems that are capable of more complex multi-hop question generation, where answering the questions requires reasoning over multiple documents. In this work, we introduce a series of strong transformer models for multi-hop question generation, including a graph-augmented transformer that leverages relations between entities in the text. While prior work has emphasized the importance of graph-based models, we show that we can substantially outperform the state-of-the-art by 5 BLEU points using a standard transformer architecture. We further demonstrate that graph-based augmentations can provide complimentary improvements on top of this foundation. Interestingly, we find that several important factors--such as the inclusion of an auxiliary contrastive objective and data filtering could have larger impacts on performance. We hope that our stronger baselines and analysis provide a constructive foundation for future work in this area.
ISSN:2331-8422