Attentive Interaction Model: Modeling Changes in View in Argumentation

Proceedings of NAACL-HLT 2018, pages 103–116 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Attentive Interaction Model: Modeling Changes in View in Argumentation Yohan Jo†, Shivani Poddar†, Byungsoo Jeon‡, Qinlan Shen†, Carolyn P. Ros´e†, Graham Neubig† †Language Technologies Institute, ‡Computer Science Department School of Computer Science Carnegie Mellon University {yohanj,spoddar2,byungsoj,qinlans,cprose,gneubig}@cs.cmu.edu Abstract We present a neural architecture for mod- eling argumentative dialogue that explic- itly models the interplay between an Opinion Holder’s (OH’s) reasoning and a challenger’s argument, with the goal of predicting if the argument successfully changes the OH’s view. The model has two components: (1) vulnerable region detection, an attention model that identi- fies parts of the OH’s reasoning that are amenable to change, and (2) interaction encoding, which identifies the relationship between the content of the OH’s reason- ing and that of the challenger’s argument. Based on evaluation on discussions from the Change My View forum on Reddit, the two components work together to pre- dict an OH’s change in view, outperform- ing several baselines. A posthoc analysis suggests that sentences picked out by the attention model are addressed more fre- quently by successful arguments than by unsuccessful ones.1 1 Introduction Through engagement in argumentative dialogue, interlocutors present arguments with the goals of winning the debate or contributing to the joint con- struction of knowledge. Especially modeling the knowledge co-construction process requires un- derstanding of both the substance of viewpoints and how the substance of an argument connects with what it is arguing against. Prior work on argumentation in the NLP community, however, has focused mainly on the first goal and has often reduced the concept of a viewpoint as a discrete 1Our code is available at https://github.com/ yohanjo/aim. side (e.g., pro vs against, or liberal vs conserva- tive), missing more nuanced and complex details of viewpoints. In addition, while the strength of the argument and the side it represents have been addressed relatively often, the dialogical aspects of argumentation have received less attention. To bridge the gap, we present a model that jointly considers an Opinion Holder’s (OH’s) ex- pressed viewpoint with a challenger’s argument in order to predict if the argument succeeded in alter- ing the OH’s view. The first component of the ar- chitecture, vulnerable region detection, aims to identify important parts in the OH’s reasoning that are key to impacting their viewpoint. The intu- ition behind our model is that addressing certain parts of the OH’s reasoning often has little impact in changing the OH’s view, even if the OH realizes the reasoning is flawed. On the other hand, some parts of the OH’s reasoning are more open to de- bate, and thus, it is reasonable for the model to learn and attend to parts that have a better chance to change an OH’s view when addressed. The second component of the architecture, interaction encoding, aims to identify the connection between the OH’s sentences and the challenger’s sentences. Meaningful in- teraction in argumentation may include agree- ment/disagreement, topic relevance, …

Content truncated to 500 words