Is Graph Structure Necessary for Multi-hop Question Answering?

Abstract

Recently, attempting to model texts as graph structure and introducing graph neural networks to deal with it has become a trend in many NLP research areas. In this paper, we investigate whether the graph structure is necessary for multi-hop question answering. Our analysis is centered on HotpotQA. We construct a strong baseline model to establish that, with the proper use of pre-trained models, graph structure may not be necessary for multi-hop question answering. We point out that both graph structure and adjacency matrix are task-related prior knowledge, and graph-attention can be considered as a special case of self-attention. Experiments and visualized analysis demonstrate that graph-attention or the entire graph structure can be replaced by self-attention or Transformers.

arXiv comment6 pages, to appear at EMNLP 2020

References in corpus (5)

• Bidirectional Attention Flow for Machine Comprehension
• Introduction to the CoNLL-2002 Shared Task: Language-Independent Named Entity Recognition
• Exploring Graph-structured Passage Representation for Multi-hop Reading Comprehension with Graph Neural Networks
• Hierarchical Graph Network for Multi-hop Question Answering
• Select, Answer and Explain: Interpretable Multi-hop Reading Comprehension over Multiple Documents

Cited by in corpus (8)

• Machine Reading Comprehension: The Role of Contextualized Language Models and Beyond
• A Survey on Explainability in Machine Reading Comprehension
• Hierarchical Graph Network for Multi-hop Question Answering
• Graph-free Multi-hop Reading Comprehension: A Select-to-Guide Strategy
• Adversarial Training for Machine Reading Comprehension with Virtual Embeddings
• Towards Interpretable and Reliable Reading Comprehension: A Pipeline Model with Unanswerability Prediction
• Memory Augmented Sequential Paragraph Retrieval for Multi-hop Question Answering
• Decomposing Complex Questions Makes Multi-Hop QA Easier and More Interpretable