Generating Questions for Knowledge Bases via Incorporating Diversified Contexts and Answer-Aware Loss

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 2431–2441, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 2431 Generating Questions for Knowledge Bases via Incorporating Diversified Contexts and Answer-Aware Loss Cao Liu1,2, Kang Liu1,2, Shizhu He1,2, Zaiqing Nie3, 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 3 Alibaba AI Labs, Beijing, 100029, China {cao.liu, kliu, shizhu.he, jzhao}@nlpr.ia.ac.cn zaiqing.nzq@alibaba-inc.com Abstract We tackle the task of question generation over knowledge bases. Conventional methods for this task neglect two crucial research issues: 1) the given predicate needs to be expressed; 2) the answer to the generated question needs to be definitive. In this paper, we strive toward the above two issues via incorporating diversi- fied contexts and answer-aware loss. Specif- ically, we propose a neural encoder-decoder model with multi-level copy mechanisms to generate such questions. Furthermore, the an- swer aware loss is introduced to make generat- ed questions corresponding to more definitive answers. Experiments demonstrate that our model achieves state-of-the-art performance. Meanwhile, such generated question can ex- press the given predicate and correspond to a definitive answer. 1 Introduction Question Generation over Knowledge Bases (K- BQG) aims at generating natural language ques- tions for the corresponding facts on KBs, and it can benefit some real applications. Firstly, K- BQG can automatically annotate question answer- ing (QA) datasets. Secondly, the generated ques- tions and answers will be able to augment the training data for QA systems. More important- ly, KBQG can improve the ability of machines to actively ask questions on human-machine con- versations (Duan et al., 2017; Sun et al., 2018). Therefore, this task has attracted more attention in recent years (Serban et al., 2016; Elsahar et al., 2018). Specifically, KBQG is the task of generating natural language questions according to the in- put facts from a knowledge base with triplet form, like <subject, predicate, object>. For example, as illustrated in Figure 1, KBQG aims at gener- ating a question “Which city is Statue of Liber- ty located in?” (Q3) for the input factual triplet Input <Statue of Liberty, location/containedby, New York City> Output Matching Predicate Definite Answer Question Q1 - Who created the Statue of Liberty? Q2 Where is Statue of Liberty in? Q3 Which city is Statue of Liberty located in? Figure 1: Examples of KBQG. We aims at generating questions like Q3 which expresses (matches) the given predicate and refers to a definitive answer. “<Statue of Liberty, location/containedby1, New York City>”. Here, the generated question is associated to the subject “Statue of Liberty” and the predicate fb:location/containedby) of the input fact, and the answer corresponds to the object “New York City”. As depicted by Serban et al. (2016), KBQG is required to transduce the triplet fact into a ques- tion about the subject and predicate, where the ob- ject is the correct …

Content truncated to 500 words