AEG: Argumentative Essay Generation via A Dual-Decoder Model with Content Planning

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 5134 - 5148 December 7-11, 2022 ©2022 Association for Computational Linguistics AEG: Argumentative Essay Generation via A Dual-Decoder Model with Content Planning Abstract Argument generation is an important but chal- lenging task in computational argumentation. Existing studies have mainly focused on gen- erating individual short arguments, while re- search on generating long and coherent argu- mentative essays is still under-explored. In this paper, we propose a new task, Argumenta- tive Essay Generation (AEG). Given a writing prompt, the goal of AEG is to automatically generate an argumentative essay with strong persuasiveness. We construct a large-scale dataset, ArgEssay, for this new task and es- tablish a strong model based on a dual-decoder Transformer architecture. Our proposed model contains two decoders, a planning decoder (PD) and a writing decoder (WD), where PD is used to generate a sequence for essay content plan- ning and WD incorporates the planning infor- mation to write an essay. Further, we pre-train this model on a large news dataset to enhance the plan-and-write paradigm. Automatic and human evaluation results show that our model can generate more coherent and persuasive es- says with higher diversity and less repetition compared to several baselines.1 1 Introduction Automatic argument generation, literately the task of generating persuasive arguments on controver- sial issues (Toulmin, 2003; Zukerman et al., 2000), has received many research interests in recent years (Khatib et al., 2021; Schiller et al., 2021). Many works have involved different arguments gen- eration such as the counter-arguments generation (Hua and Wang, 2018; Hua et al., 2019; Hidey and McKeown, 2019; Alshomary et al., 2021b) and the ∗ This work was done during the internship at Huawei Noah’s Ark Lab. †Corresponding Author 1Code and data are available at https://github.com/ HITSZ-HLT/AEG Writing Prompt: Online education is becoming more and more popular. Some people claim that e-learning has so many benefits that it will replace face-to-face education soon. Others say that traditional education is irreplaceable. Discuss both views and give your opinion. Argumentative Essay: Acquiring knowledge virtually has become extremely popular in the present times. While many individuals believe that there are various ad- vantages and might overtake traditional learning in the future, a sizeable group thinks that the traditional method cannot be replaced. :I :::: believe::: that :: the:: use:: of:: the::::::: classroom:::: might ::::: reduce, :: but:it::::: cannot :: be :::::: replaced. This essay will discuss both views and substantiate my view in the course of the essay. : To::::::: commence:::: with, :::: virtual::::: learning::is:::: widely::::::::: implemented ::::: because :it:is::::::: convenient::: and::::::::: cost-effective. It provides us with the opportunity to obtain an education without the hassle of travelling. Students, for in- stance, can attend classes at the comfort of their home, resulting in saving time that might have been spent on commuting in the past. Similarly, learning online can also be considered cost-efficient. Instead of spending an immense amount of college funds, we can attain the same level of qualifications at …

Content truncated to 500 words