Volume 63, Issue 1 - Argumentation Technology / Gastherausgeber: Philipp Cimiano, Matthias Hagen, Benno Stein, Host Editor: Paul Molitor

This article summarizes selected aspects of a dissertation project and prior publications related to the DFG-funded RecomRatio research project. As such, it provides an end-to-end overview of a research project that aims at extracting and utilizing Emerging Knowledge represented by two concepts that we define as Emerging Named Entities and Emerging Argument Entities to support medical argumentation retrieval. We use these two concepts to model novelty in general scientific literature and, in particular, in medical argumentation. Therefore, this paper will provide an overview of Emerging Knowledge and definitions of Emerging Named Entities and Emerging Argument Entities. It includes a review of state-of-the-art and related work. A preparatory study shows that Emerging Argument Entities are in use in the medical literature. Based on the state of the art review and the preparatory study, a conceptual system design based on Emerging Named Entity Recognition and a state-of-the-art Argumentation Mining framework (ArgumenText) is introduced to extract Emerging Argument Entities from medical literature and make them available for Argument Retrieval. The conceptual system design supports two Argument Retrieval use cases: 1.) Ranking of result sets based on Emerging Argument Entities, and 2.) Highlighting Emerging Argument Entities within result sets. A case study for the extraction and visualization of Emerging Named Entities and Emerging Argument Entities is implemented based on the conceptual design. This proof-of-concept system is used to conduct technical evaluations regarding the Emerging Named Entity Recognition. Furthermore, prior results of an expert-based evaluation are presented. The article finishes with a conclusion and brief outlook of future work, e. g., supporting the Argument Interchange Format.

Persuasive argumentation depends on multiple aspects, which include not only the content of the individual arguments, but also the way they are presented. The presentation of arguments is crucial – in particular in the context of dialogical argumentation. However, the effects of different discussion styles on the listener are hard to isolate in human dialogues. In order to demonstrate and investigate various styles of argumentation, we propose a multi-agent system in which different aspects of persuasion can be modelled and investigated separately. Our system utilizes argument structures extracted from text-based reviews for which a minimal bias of the user can be assumed. The persuasive dialogue is modelled as a dialogue game for argumentation that was motivated by the objective to enable both natural and flexible interactions between the agents. In order to support a comparison of factual against affective persuasion approaches, we implemented two fundamentally different strategies for both agents: The logical policy utilizes deep Reinforcement Learning in a multi-agent setup to optimize the strategy with respect to the game formalism and the available argument. In contrast, the emotional policy selects the next move in compliance with an agent emotion that is adapted to user feedback to persuade on an emotional level. The resulting interaction is presented to the user via virtual avatars and can be rated through an intuitive interface.

We present an approach to extracting arguments from social media, exemplified by a case study on a large corpus of Twitter messages collected under the #Brexit hashtag during the run-up to the referendum in 2016. Our method is based on constructing dedicated corpus queries that capture predefined argumentation patterns following standard Walton-style argumentation schemes. Query matches are transformed directly into logical patterns, i. e. formulae with placeholders in a general form of modal logic. We prioritize precision over recall, exploiting the fact that the sheer size of the corpus still delivers substantial numbers of matches for all patterns, and with the goal of eventually gaining an overview of widely-used arguments and argumentation schemes. We evaluate our approach in terms of recall on a manually annotated gold standard of 1000 randomly selected tweets for three selected high-frequency patterns. We also estimate precision by manual inspection of query matches in the entire corpus. Both evaluations are accompanied by an analysis of inter-annotator agreement between three independent judges.

In the last decade, the field of argument mining has grown notably. However, only relatively few studies have investigated argumentation in social media and specifically on Twitter. Here, we provide the, to our knowledge, first critical in-depth survey of the state of the art in tweet-based argument mining. We discuss approaches to modelling the structure of arguments in the context of tweet corpus annotation, and we review current progress in the task of detecting argument components and their relations in tweets. We also survey the intersection of argument mining and stance detection, before we conclude with an outlook.

Within the field of argument mining, an important task consists in predicting the frame of an argument, that is, making explicit the aspects of a controversial discussion that the argument emphasizes and which narrative it constructs. Many approaches so far have adopted the framing classification proposed by Boydstun et al. [3], consisting of 15 categories that have been mainly designed to capture frames in media coverage of political articles. In addition to being quite coarse-grained, these categories are limited in terms of their coverage of the breadth of discussion topics that people debate. Other approaches have proposed to rely on issue-specific and subjective (argumentation) frames indicated by users via labels in debating portals. These labels are overly specific and do often not generalize across topics. We present an approach to bridge between coarse-grained and issue-specific inventories for classifying argumentation frames and propose a supervised approach to classifying frames of arguments at a variable level of granularity by clustering issue-specific, user-provided labels into frame clusters and predicting the frame cluster that an argument evokes. We demonstrate how the approach supports the prediction of frames for varying numbers of clusters. We combine the two tasks, frame prediction with respect to media frames categories as well as prediction of clusters of user-provided labels, in a multi-task setting, learning a classifier that performs the two tasks. As main result, we show that this multi-task setting improves the classification on the single tasks, the media frames classification by up to +9.9 % accuracy and the cluster prediction by up to +8 % accuracy.