Evidence Integration for Multi-Hop Reading Comprehension With Graph Neural Networks

• Published in: IEEE transactions on knowledge and data engineering 2022-02, Vol.34 (2), p.631-639
• Main Authors: Song, Linfeng, Wang, Zhiguo, Yu, Mo, Zhang, Yue, Florian, Radu, Gildea, Daniel
• Format: Article
• Language: English
• Subjects: Cognition; Datasets; Encoding; graph neural network; Graph neural networks; Graphs; Knowledge discovery; Logic gates; multi-hop reasoning; Natural language processing; Neural networks; question answering; Questions; Reading comprehension; Task analysis; Urban areas
• ISSN: 1041-4347; 1558-2191
• DOI: 10.1109/TKDE.2020.2982894

Multi-hop reading comprehension focuses on one type of factoid question, where a system needs to properly integrate multiple pieces of evidence to correctly answer a question. Previous work approximates global evidence with local coreference information, encoding coreference chains with DAG-styled GRU layers within a gated-attention reader. However, coreference is limited in providing information for rich inference. We introduce a new method for better connecting global evidence, which forms more complex graphs compared to DAGs. To perform evidence integration on our graphs, we investigate two recent graph neural networks, namely graph convolutional network (GCN) and graph recurrent network (GRN). Experiments on two standard datasets show that richer global information leads to better answers. Our approach shows highly competitive performances on these datasets without deep language models (such as ELMo).