Examining the Ordering of Rhetorical Strategies in Persuasive Requests

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1299–1306 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1299 Examining the Ordering of Rhetorical Strategies in Persuasive Requests Omar Shaikh, Jiaao Chen, Jon Saad-Falcon, Duen Horng (Polo) Chau, Diyi Yang Georgia Institute of Technology {oshaikh, jchen896, jonsaadfalcon, polo, dyang888}@gatech.edu Abstract Interpreting how persuasive language influ- ences audiences has implications across many domains like advertising, argumentation, and propaganda. Persuasion relies on more than a message’s content. Arranging the order of the message itself (i.e., ordering specific rhetori- cal strategies) also plays an important role. To examine how strategy orderings contribute to persuasiveness, we first utilize a Variational Autoencoder model to disentangle content and rhetorical strategies in textual requests from a large-scale loan request corpus. We then vi- sualize interplay between content and strategy through an attentional LSTM that predicts the success of textual requests. We find that spe- cific (orderings of) strategies interact uniquely with a request’s content to impact success rate, and thus the persuasiveness of a request. 1 Introduction Persuasion has been shown as a powerful tool for catalyzing beneficial social and political changes (Hovland et al., 1953) or enforcing propaganda as a tool of warfare (Finch, 2000). Modeling persua- siveness of text has received much recent attention in the language community (Althoff et al., 2014; Tan et al., 2016; Habernal and Gurevych, 2017; Yang et al., 2019; Srinivasan et al., 2019). Numer- ous qualitative studies have been conducted to un- derstand persuasion, from explorations of rhetoric in presidential campaigns (Bartels, 2006; Popkin and Popkin, 1994) to the impact of a communica- tor’s likability on persuasiveness (Chaiken, 1980). Studies of persuasion and argumentation that have analyzed textual level features (e.g., n-grams, inde- pendent rhetorical strategies) to gauge efficacy have also garnered recent attention (Althoff et al., 2014; Habernal and Gurevych, 2017, 2016b,a; Yang and Kraut, 2017; Yang et al., 2019). Of particular inter- est is Morio et al. (2019), which identified sentence Strategy Definition Concreteness (39%) Use concrete details in request “I need $250 to purchase fishing rods” Reciprocity (18%) Assure user will repay giver “I will pay 5% interest to you” Impact (12%) Highlight the impact of a request “This loan will help teach students” Credibility (8%) Use credentials to establish trust “I have repaid all of my prior loans” Politeness (16%) Use polite language “Highly appreciated.” Other (7%) None of the above Table 1: Sentence level persuasion strategies, and their data distributions (%). Strategy abbreviations are bolded. placements for individual rhetorical strategies in a request. Other research analyzed how different persuasive strategies are more effective on specific stances and personal backgrounds (Durmus and Cardie, 2018, 2019). However, prior work has mainly focused on iden- tifying overall persuasiveness of textual content or analyzing components of persuasion affecting a request. These works largely ignore ordering of specific strategies, a key canon of rhetoric that has a large impact on persuasion effectiveness (Borchers and Hundley, 2018; Cicero, 1862). In the context of online communities, identifying where/how ef- …

Content truncated to 500 words