On the Role of Discourse Markers for Discriminating Claims and Premises in Argumentative Discourse

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2236–2242, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. On the Role of Discourse Markers for Discriminating Claims and Premises in Argumentative Discourse Judith Eckle-Kohler1 Roland Kluge3 Iryna Gurevych1,2 1 UKP Lab, Technische Universit¨at Darmstadt 2 UKP Lab, German Institute for Educational Research 3 Real-Time Systems Lab, Technische Universit¨at Darmstadt http://www.ukp.tu-darmstadt.de Abstract This paper presents a study on the role of discourse markers in argumentative dis- course. We annotated a German cor- pus with arguments according to the com- mon claim-premise model of argumen- tation and performed various statistical analyses regarding the discriminative na- ture of discourse markers for claims and premises. Our experiments show that particular semantic groups of discourse markers are indicative of either claims or premises and constitute highly predictive features for discriminating between them. 1 Introduction The growing field of argumentation mining in NLP develops methods to automatically analyze argumentative discourse for enhanced Information Extraction. Terminology: We understand argumentation as a rhetorical arrangement of arguments with the in- tent to convince or persuade the reader of a par- ticular state of affairs. Following previous work in argumentation mining (e.g., (Palau and Moens, 2009; Florou et al., 2013; Peldszus and Stede, 2013a; Stab and Gurevych, 2014)), we define an argument as the pairing of a single claim (an ar- guable text unit) and a (possibly empty) set of premises, which each either support or attack the claim (Besnard and Hunter, 2008). We subsume claims and premises under the term argument unit (AU). Discourse markers in argumentative discourse: Since an argumentation line can only be captured in the context of a coherent text, argumentation mining is closely related to automated discourse analysis (Cabrio et al., 2013), which aims at iden- tifying discourse functions of text segments, and discourse relations (DRs) between adjacent text segments (Webber et al., 2012). Often, so-called discourse markers (DMs) are used to signal DRs. The following example shows that DMs act as lex- ical markers in argumentative discourse as well: the DM however (marking the DR contrast) po- sitions the claim in (1) in the overall argumenta- tion line, while in (2), the DMs as (marking rea- son) and also (marking elaboration) connect the premises with each other and with the claim. (1) However, staying down is pointless from a pedagogic point of view. (2) As the students get frustrated, their performance generally does not im- prove. Also, the point of repeating all courses because of only one or two bad grades is arguable. DMs belong to the word classes of conjunctions and adverbs (also called discourse particles) and are semantically characterized in traditional gram- mar books. The correspondence between DM se- mantics and DR semantics has received consid- erable attention in previous research in linguis- tics, most of which is based on corpora annotated with DRs (Carlson et al., 2003; Wolf and Gibson, 2005; Prasad et al., 2008). In contrast, the role of DMs as potential lexical signals …

Content truncated to 500 words