ArgU: A Controllable Factual Argument Generator

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 8373–8388 July 9-14, 2023 ©2023 Association for Computational Linguistics ArgU: A Controllable Factual Argument Generator Sougata Saha and Rohini Srihari State University of New York at Buffalo Department of Computer Science and Engineering {sougatas, rohini}@buffalo.edu Abstract Effective argumentation is essential towards a purposeful conversation with a satisfactory out- come. For example, persuading someone to reconsider smoking might involve empathetic, well founded arguments based on facts and ex- pert opinions about its ill-effects and the con- sequences on one’s family. However, the au- tomatic generation of high-quality factual ar- guments can be challenging. Addressing ex- isting controllability issues can make the re- cent advances in computational models for ar- gument generation a potential solution. In this paper, we introduce ArgU: a neural argument generator capable of producing factual argu- ments from input facts and real-world concepts that can be explicitly controlled for stance and argument structure using Walton’s argument scheme-based control codes. Unfortunately, computational argument generation is a rela- tively new field and lacks datasets conducive to training. Hence, we have compiled and re- leased an annotated corpora of 69,428 argu- ments spanning six topics and six argument schemes, making it the largest publicly avail- able corpus for identifying argument schemes; the paper details our annotation and dataset creation framework. We further experiment with an argument generation strategy that es- tablishes an inference strategy by generating an “argument template” before actual argument generation. Our results demonstrate that it is possible to automatically generate diverse ar- guments exhibiting different inference patterns for the same set of facts by using control codes based on argument schemes and stance. 1 Introduction Although arguing is an innate human quality, for- mulating convincing arguments is an art. A success- ful narrative aiming to persuade someone should be rhetorically appealing, trustworthy, factually cor- rect, and logically consistent, which makes formu- lating good arguments challenging. Incorporating neural language models, the relatively new field of computational argument generation has shown promise in assisting with argument synthesis. Argu- ment generators like Project Debater (Slonim et al., 2021) have successfully formulated convincing ar- guments across different domains including legal, politics, education, etc., and can potentially find new argumentative connections. However, lack- ing explicit control mechanisms, neural argument generators often render illogical and inappropri- ate arguments, reducing their trustworthiness and applicability for practical use. Furthermore, train- ing such models requires a considerable amount of quality data, which is hard to collect and an- notate. Hence, we propose ArgU, a controllable neural argument generator trained on a curated and quality-controlled corpus of annotated argument texts from abortion, minimum wage, nuclear en- ergy, gun control, the death penalty and school uniform. Figure 1: Generating stance and argument scheme con- trolled factual arguments using ArgU. ArgU strives to enable effective, scalable and appealing argument generation. As depicted in Figure 1, it takes as input worldly knowledge and concepts as fact variables and coherently combines them to generate an …

Content truncated to 500 words