Question Generation Based on Grammar Knowledge and Fine-grained Classification

Proceedings of the 29th International Conference on Computational Linguistics, pages 6457–6467 October 12–17, 2022. 6457 Question Generation Based on Grammar Knowledge and Fine-grained Classification Yuan Sun1,2,∗ , Sisi Liu1,2 , Zhengcuo Dan1,2 , Xiaobing Zhao1,2,∗ 1Minzu University of China 2National Language Resource Monitoring & Research Center Minority Languages Branch tracy.yuan.sun@gmail.com, liusisi.s@qq.com, nmzxb_cn@163.com *Corresponding Author: Yuan Sun, Xiaobing Zhao Abstract Question generation is the task of automatically generating questions based on given context and answers, and there are problems that the types of questions and answers do not match. In minority languages such as Tibetan, since the grammar rules are complex and the training data is small, the related research on question generation is still in its infancy. To solve the above problems, this paper constructs a ques- tion type classifier and a question generator. We perform fine-grained division of question types and integrate grammatical knowledge into ques- tion type classifiers to improve the accuracy of question types. Then, the types predicted by the question type classifier are fed into the question generator. Our model improves the accuracy of interrogative words in generated questions, and the BLEU-4 on SQuAD reaches 17.52, the BLEU-4 on HotpotQA reaches 19.31, the BLEU-4 on TibetanQA reaches 25.58. 1 Introduction Question generation is an essential research direc- tion in natural language generation, which aims to generate a grammatical question based on the given paragraph and answer, and the question can be answered by the given answer. Recently, ques- tion generation has received extensive attention in academia and industry. It can automatically gen- erate training data for question answering systems (Du and Cardie, 2018) and reading comprehension tasks (Heilman and Smith, 2010). It can also sim- ulate user questions in the field of education (Du et al., 2017), and guide the machine to ask ques- tions actively in the field of dialogue (Shum et al., 2018), and strengthen human-computer interaction (Mostafazadeh et al., 2016). In recent years, the research on question gener- ation has achieved great success, but we analyze the questions generated by an advanced answer- aware s2s model (Zhao et al., 2018) and find that some question types and answers did not match. Paragraph ..., To ensure a space, in 1947, ABC submitted five applications for television station licenses, ... Answer 1947 Gold in what year did abc submit licenses for 5 television stations ? Generated what is the name of the space that was used to ensure a space on television station ? Table 1: English question generation example. As shown in Table 1, the question type of the gold question is to ask questions about time and use "in what year", but the interrogative word of generated question is "what". To solve the problem, some studies have pro- posed to use answer information to improve the interrogative words accuracy of generated ques- tions (Sun et al., 2018; Zhou et al., 2019b; Kang et al., 2019). However, they did not consider the structural and grammatical features of the answers. Some studies classify question types …

Content truncated to 500 words