Diversify Question Generation with Continuous Content Selectors and Question Type Modeling

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2134–2143 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2134 Diversify Question Generation with Continuous Content Selectors and Question Type Modeling Zhen Wang, Siwei Rao, Jie Zhang, Zhen Qin, Guangjian Tian, Jun Wang Huawei Noah’s Ark Lab {wangzhen150, raosiwei, clark.zhang}@huawei.com {qin.zhen, tian.guangjian, w.j}@huawei.com Abstract Generating questions based on answers and relevant contexts is a challenging task. Recent work mainly pays attention to the quality of a single generated question. However, ques- tion generation is actually a one-to-many prob- lem, as it is possible to raise questions with dif- ferent focuses on contexts and various means of expression. In this paper, we explore the diversity of question generation and come up with methods from these two aspects. Specif- ically, we relate contextual focuses with con- tent selectors, which are modeled by a continu- ous latent variable with the technique of condi- tional variational auto-encoder (CVAE). In the realization of CVAE, a multimodal prior dis- tribution is adopted to allow for more diverse content selectors. To take into account vari- ous means of expression, question types are ex- plicitly modeled and a diversity-promoting al- gorithm is proposed further. Experimental re- sults on public datasets show that our proposed method can significantly improve the diversity of generated questions, especially from the perspective of using different question types. Overall, our proposed method achieves a better trade-off between generation quality and diver- sity compared with existing approaches. 1 Introduction As a reverse task of question answering (QA), ques- tion generation (QG) aims to generate questions from a given answer and its relevant context. The task holds the potential value of educational pur- pose to generate questions for reading compre- hension materials (Heilman and Smith, 2010). It can also be deployed as chatbot components (Li et al., 2017) for evaluating or improving mental health (Colby, 1975). Moreover, QG can be ap- plied to extend the question-answer pairs (Du and Cardie, 2018) for QA systems. Traditional methods for QG mainly use rigid heuristic rules to transform a sentence into related Source context: the network was engineered and operated by mci telecommunications under a cooperative agreement with the nsf . Target question: who operated the vbsn network? Focus 1: the network was engineered and operated by mci telecommunications under a cooperative agreement with the nsf . (Ours) who operates the network with nsf ? Focus 2: the network was engineered and operated by mci telecommunications under a cooperative agreement with the nsf . (Ours) who operated the network under a cooperative agreement with the nsf ? Focus 3: the network was engineered and operated by mci telecommunications under a cooperative agreement with the nsf . (Ours) in what company was the network engineered and operated by the nsf ? Figure 1: Diversified questions generated by our method for the given passage-answer pair (answer is underlined). Different questions can be raised accord- ing to distinct focuses on the context (colored) and var- ious means of expression (italic). …

Content truncated to 500 words