Interconnected Question Generation with Coreference Alignment and Conversation Flow Modeling

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4853–4862 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4853 Interconnected Question Generation with Coreference Alignment and Conversation Flow Modeling Yifan Gao1∗ Piji Li2 Irwin King1 Michael R. Lyu1 1 Department of Computer Science and Engineering, The Chinese University of Hong Kong, Shatin, N.T., Hong Kong 2 Tencent AI Lab 1{yfgao,king,lyu}@cse.cuhk.edu.hk 2pijili@tencent.com Abstract We study the problem of generating intercon- nected questions in question-answering style conversations. Compared with previous works which generate questions based on a single sentence (or paragraph), this setting is dif- ferent in two major aspects: (1) Questions are highly conversational. Almost half of them refer back to conversation history using coreferences. (2) In a coherent conversation, questions have smooth transitions between turns. We propose an end-to-end neural model with coreference alignment and conversation flow modeling. The coreference alignment modeling explicitly aligns coreferent men- tions in conversation history with correspond- ing pronominal references in generated ques- tions, which makes generated questions inter- connected to conversation history. The con- versation flow modeling builds a coherent con- versation by starting questioning on the first few sentences in a text passage and smoothly shifting the focus to later parts. Extensive ex- periments show that our system outperforms several baselines and can generate highly con- versational questions. The code implementa- tion is released at https://github.com/ Evan-Gao/conversaional-QG. 1 Introduction Question Generation (QG) aims to create human- like questions from a range of inputs, such as natural language text (Heilman and Smith, 2010), knowledge base (Serban et al., 2016) and image (Mostafazadeh et al., 2016). QG is helpful for the knowledge testing in education, i.e., the intel- ligence tutor system, where an instructor can ac- tively ask questions to students given reading com- prehension materials (Heilman and Smith, 2010; Du et al., 2017). Besides, raising good questions ∗ This work was partially done when Yifan Gao was an intern at Tencent AI Lab. Passage: Incumbent Democratic President Bill Clinton was ineligible to serve a third term due to term limitations in the 22nd Amendment of the Constitution, and Vice President Gore was able to secure the Democratic nomination with relative ease. Bush was seen as the early favorite for the Republican nomina- tion and, despite a contentious primary battle with Senator John McCain and other candidates, secured the nomination by Super Tuesday. Bush chose ... Q1: What political party is Clinton a member of? A1: Democratic Q2: What was he ineligible to serve? A2: third term Q3: Why? A3: term limitations Q4: Based on what amendment? A4: 22nd Q5: Of what document? A5: Constitution Q6: Who was his vice president? A6: Gore Q7: Who was the early Republican favorite for A7: Bush the nomination? Q8: Who was the primary battle with? A8: John McCain Q9: What is his title? A9: Senator Q10: When did Bush secure the nomination by? A10: Tuesday Table 1: An example for conversational question gener- ation from a conversational question answering …

Content truncated to 500 words