Adapting Serious Game for Fallacious Argumentation to German: Pitfalls, Insights, and Best Practices

Adapting Serious Game for Fallacious Argumentation to German: Pitfalls, Insights, and Best Practices Ivan Habernal†‡, Patrick Pauli†, Iryna Gurevych†‡ †Ubiquitous Knowledge Processing Lab ‡Research Training Group AIPHES Department of Computer Science, Technische Universit¨at Darmstadt, Hochschulstr. 10, 64289 Darmstadt, Germany www.ukp.tu-darmstadt.de www.aiphes.tu-darmstadt.de Abstract As argumentation about controversies is culture- and language-dependent, porting a serious game that deals with daily argumentation to another language requires substantial adaptation. This article presents a study of deploying Argotario (serious game for learning argumentation fallacies) in the German context. We examine all steps that are necessary to end up with a successful serious game platform, such as topic selection, initial data creation, or effective campaigns. Moreover, we analyze users’ behavior and in-game created data in order to assess the dissemination strategies and qualitative aspects of the resulting corpus. We also report on classification experiments based on neural networks and feature-based models. Keywords: argumentation, fallacies, serious games 1. Introduction Computational argumentation and argument mining has been traditionally dealing with understanding argument’s structure (Habernal and Gurevych, 2017; Stab and Gurevych, 2017; Eger et al., 2017; Daxenberger et al., 2017). Re- cently, attention has been paid to pragmatic aspects of ar- guments, such as convincingness (Habernal and Gurevych, 2016b; Habernal and Gurevych, 2016a) or overall quality (Wachsmuth et al., 2017a). A fairly unexplored area of com- putational argumentation is fallacies: arguments that seem to be valid but are not so (Hamblin, 1970). To tackle the nonexistence of corpora for dealing with fallacies compu- tationally, Habernal et al. (2017) published Argotario—a serious game intended to educate players and at the same time collect annotated fallacious arguments. The majority of research on computational argumentation is English-centric (with several exceptions, such as (Peldszus and Stede, 2015; Liebeck et al., 2016; Chow, 2016)). Not only the language itself but the discussed topics and con- troversies are culture-specific. For example, ‘homeschool- ing’ or ‘death penalty’ are almost non-existent in Germany, while being highly controversial subjects of discussion in the United States. As Argotario had been developed within the English context, simply translating the topics and existing arguments and fallacies into another language does not meet the expectations of a serious game user. We thus asked the following research questions. First, what are the best means to tackle language adaptation in serious games that depend on world-knowledge, cultural context, and specific controversies (RQ1)? Second, we were inter- ested in the dynamics and outcomes of deployment of the game, namely whether new lay users understand differences between various fallacies (RQ2), which qualitative aspects are to be expected in user-written fallacies (RQ3), and which advertising channels deliver the best return of investment (RQ4). To answer these questions, we added German language sup- port to Argotario and crowd-sourced initial data to face the ‘cold-start’ problem (RQ1), launched several campaigns (RQ4) and analyzed the obtained data and users’ behavior (RQ2; RQ3). Moreover, we conducted several experiments with feature-based and deep-learning models for classifying fallacies. The main contributions of this article are (1) an extensive study of language adaptation of …

Content truncated to 500 words