Propagate-Selector: Detecting Supporting Sentences for Question Answering via Graph Neural Networks

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5400–5407 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5400 Propagate-Selector: Detecting Supporting Sentences for Question Answering via Graph Neural Networks Seunghyun Yoon1,2, Franck Dernoncourt2, Doo Soon Kim2, Trung Bui2, Kyomin Jung1 1Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea 2Adobe Research, San Jose, CA, USA {mysmilesh, kjung}@snu.ac.kr, {franck.dernoncourt, dkim, bui}@adobe.com Abstract In this study, we propose a novel graph neural network called propagate-selector (PS), which propagates information over sentences to understand information that cannot be inferred when considering sentences in isolation. First, we design a graph structure in which each node represents an individual sentence, and some pairs of nodes are selectively connected based on the text structure. Then, we develop an iterative attentive aggregation and a skip-combine method in which a node interacts with its neighborhood nodes to accumulate the necessary information. To evaluate the performance of the proposed approaches, we conduct experiments with the standard HotpotQA dataset. The empirical results demonstrate the superiority of our proposed approach, which obtains the best performances, compared to the widely used answer-selection models that do not consider the intersentential relationship. Keywords: question answering, supporting sentence, graph neural network 1. Introduction Understanding texts and being able to answer a question posed by a human is a long-standing goal in the artificial intelligence field. With the rapid advancement of neural network-based models and the availability of large-scale datasets, such as SQuAD (Rajpurkar et al., 2016) and Trivi- aQA (Joshi et al., 2017), researchers have begun to concen- trate on building automatic question-answering (QA) sys- tems. One example of such a system is the machine-reading question-answering (MRQA) model, which provides an- swers to questions from given passages (Seo et al., 2016; Xiong et al., 2016; Wang et al., 2017; Shen et al., 2017b). Recently, research has revealed that most questions in ex- isting MRQA datasets do not require reasoning across sen- tences in the given context (passage); instead, they can be answered by looking at only a single sentence (Weis- senborn et al., 2017). Using this characteristic, a simple model can achieve performance competitive with that of a sophisticated model. However, in most real scenarios of QA applications, more than one sentence should be utilized to extract a correct answer. To alleviate this limitation of previous datasets, another type of dataset was developed in which answering the ques- tion requires reasoning over multiple sentences in the given passages (Yang et al., 2018; Welbl et al., 2018). Figure 1 shows an example of a recently released dataset, the Hot- potQA. This dataset consists of not only question-answer pairs with context passages but also supporting sentence information for answering the question annotated by a hu- man. In this study, we build a model that exploits the relational information among sentences in passages to classify the supporting sentences that contain the essential informa- tion for answering the question. To this end, we propose a novel graph …

Content truncated to 500 words