Generating Natural Answers by Incorporating Copying and Retrieving Mechanisms in Sequence-to-Sequence Learning

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 199–208 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1019 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 199–208 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1019 Generating Natural Answers by Incorporating Copying and Retrieving Mechanisms in Sequence-to-Sequence Learning Shizhu He1, Cao Liu1,2, Kang Liu1 and Jun Zhao1,2 1 National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2 University of Chinese Academy of Sciences, Beijing, 100049, China {shizhu.he, cao.liu, kliu, jzhao}@nlpr.ia.ac.cn Abstract Generating answer with natural language sentence is very important in real-world question answering systems, which need- s to obtain a right answer as well as a coherent natural response. In this paper, we propose an end-to-end ques- tion answering system called COREQA in sequence-to-sequence learning, which in- corporates copying and retrieving mech- anisms to generate natural answers with- in an encoder-decoder framework. Specif- ically, in COREQA, the semantic units (words, phrases and entities) in a natural answer are dynamically predicted from the vocabulary, copied from the given ques- tion and/or retrieved from the correspond- ing knowledge base jointly. Our em- pirical study on both synthetic and real- world datasets demonstrates the efficien- cy of COREQA, which is able to generate correct, coherent and natural answers for knowledge inquired questions. 1 Introduction Question answering (QA) systems devote to pro- viding exact answers, often in the form of phrases and entities for natural language questions (Wood- s, 1977; Ferrucci et al., 2010; Lopez et al., 2011; Yih et al., 2015), which mainly focus on analyzing questions, retrieving related facts from text snip- pets or knowledge bases (KBs), and finally pre- dicting the answering semantic units-SU (word- s, phrases and entities) through ranking (Yao and Van Durme, 2014) and reasoning (Kwok et al., 2001). However, in real-world environments, most people prefer the correct answer replied with a more natural way. For example, most existing <ÀîÁ¬½Ü£¬³öÉúµØµã£¬±±¾©> <ÀîÁ¬½Ü£¬¹ú¼®£¬Ð¼ÓÆÂ> <ÀîÁ¬½Ü£¬³öÉúÄêÔ£¬1963Äê4ÔÂ26ÈÕ> ... ÀîÁ¬½ÜÊÇÄÄÀïÈË£¿ ÀîÁ¬½Ü³öÉúÓÚ ±±¾© £¬Ëû <ÀîÁ¬½Ü£¬ÐÔ±ð£¬ÄÐ> ÏÖÔÚÊÇ Ð¼ÓÆ ¹ú¼®¡£ Copy Reasoning From Question From KB Question Response where wasJet Li Jet Li was born in Beijing. He is now a Singaporean citizen. Copying and Retrieving Predicting Copying from Question Retrieving from KB Question Natural Answer Do you know from ? Knowledge Base Figure 1: Incorporating copying and retrieving mechanisms in generating a natural answer. commercial products such as Siri1 will reply a nat- ural answer “Jet Li is 1.64m in height.” for the question “How tall is Jet Li?”, rather than only answering one entity “1.64m”. Basic on this ob- servation, we define the “natural answer” as the natural response in our daily communication for replying factual questions, which is usually ex- pressed in a complete/partial natural language sen- tence rather than a single entity/phrase. In this case, the system needs to not only parse question, retrieve relevant facts from KB but also generate a proper reply. To …

Content truncated to 500 words