Question Answering with Long Multiple-Span Answers

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3840–3849 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 3840 Question Answering with Long Multiple-Span Answers Ming Zhu1, Aman Ahuja1, Da-Cheng Juan2, Wei Wei3, Chandan K. Reddy1 1 Department of Computer Science, Virginia Tech, Arlington, VA 2 Department of Computer Science, National Tsing Hua University, Taiwan 3 Cloud AI Research, Google Inc. {mingzhu, aahuja}@vt.edu, x@dacheng.info, wewei@google.com, reddy@cs.vt.edu Abstract Answering questions in many real-world appli- cations often requires complex and precise in- formation excerpted from texts spanned across a long document. However, currently no such annotated dataset is publicly available, which hinders the development of neural question- answering (QA) systems. To this end, we present MASH-QA1, a Multiple Answer Spans Healthcare Question Answering dataset from the consumer health domain, where answers may need to be excerpted from multiple, non- consecutive parts of text spanned across a long document. We also propose MultiCo, a neural architecture that is able to capture the relevance among multiple answer spans, by using a query-based contextualized sentence selection approach, for forming the answer to the given question. We also demonstrate that conventional QA models are not suitable for this type of task and perform poorly in this setting. Extensive experiments are con- ducted, and the experimental results confirm the proposed model significantly outperforms the state-of-the-art QA models in this multi- span QA setting. 1 Introduction Developing neural networks for question answering (QA) has become an important and fast-growing area of research in the NLP community. Interest in this area is largely driven by the importance and effectiveness of such systems in virtual assistants and search engines. Driven by the development of large-scale datasets such as SQuAD (Rajpurkar et al., 2016, 2018), most of the work in this domain focuses on the task of machine reading compre- hension, where the objective is to find a single short answer span—typically ranging from a few 1Code: https://github.com/mingzhu0527/ MASHQA What are tips for managing my bipolar disor- der? Along with seeing your doctor and therapist and taking your medicines, simple daily habits can make a difference. Start with these strate- gies. (22 words truncated) Pay attention to your sleep. This is especially important for people with bipolar disorder... (178 words truncated) Eat well. There’s no specific diet... (29 words truncated) Focus on the basics: Fa- vor fruits, vegetables, lean protein, and whole grains. And cut down on fat, salt, and sugar. Tame stress. (81 words truncated) You can also listen to music or spend time with posi- tive people who are good company. (73 words truncated) Limit caffeine. It can keep you up at night and possibly affect your mood. (47 words truncated) Avoid alcohol and drugs. They can affect how your medications work. (118 words truncated) Figure 1: An example of a question and its correspond- ing answer (highlighted) from MASH-QA. The answer consists of multiple sentences from the context. All the highlighted sentences will form the comprehensive answer. The context here is 632 words long, so …

Content truncated to 500 words