Multilevel Annotation of Agreement and Disagreement in Italian News Blogs

Multilevel Annotation of Agreement and Disagreement in Italian News Blogs Fabio Celli, Giuseppe Riccardi, Firoj Alam Dept. of Computer Science, University of Trento, Italy {fabio.celli, giuseppe.riccardi, firoj.alam}@unitn.it Abstract In this paper, we present a corpus of news blog conversations in Italian annotated with gold standard agreement/disagreement relations at message and sentence levels. This is the first resource of this kind in Italian. From the analysis of ADRs at the two levels emerged that agreement annotated at message level is consistent and generally reflected at sentence level, and that the structure of disagreement is more complex. The manual error analysis revealed that this resource is useful not only for the analysis of argumentation, but also for the detection of irony/sarcasm in online debates. The corpus and annotation tool are available for research purposes on request. Keywords: Corpora, Agreement/Disagreement, Social Media, News Blogs 1. Introduction and related Work On-line social conversations concur to the formation of opinions and shared knowledge which influence decision makers. A large amount of multiparty conversations take place online every day in social forums and news blogs (Ruiz et al., 2011), and participants express agreement and disagreement with respect to each others’ positions and statements. From a communication analysis perspective, conversation in social media are asynchronous and partic- ipants can reply to any other, using text messages or pre- coded actions (e.g. like buttons). Previous work on Agree- ment/Disagreement Relations (henceforth ADRs) in asyn- chronous online debates, focused either on messages and overall positions of participants (Murakami and Raymond, 2010) (Somasundaran and Wiebe, 2009) (Abu-Jbara et al., 2012), or on the detection of ADRs in pairs of candidate sentences or parts of messages (Andreas et al., 2012). Moti- vated by the interest in the analysis of argumentation struc- tures in asynchronous conversations, we produced a corpus annotated at message and sentence level in Italian. To do so, we developed a specific annotation tool. Both the cor- pus and the tool are available for research purposes1 under a LGPL license. To the best of our knowledge, this is the first resource of ADRs in Italian. We believe that the two lev- els of annotations maybe useful for argumentation analysis (Schneider et al., 2013) as well as summarization (Di Fab- brizio et al., 2014), irony/sarcasm detection (Reyes et al., 2013) and other kind of parasemantic analyses in the social media domain (Basile and Nissim, 2013), (Celli and Polo- nio, 2013). 2. Definitions of ADRs ADRs in conversations can be defined in general terms as shared public commitments, that ground the speech acts performed by the bloggers within the conversations (Las- carides and Asher, 2008). Figure 1 reports an exmple of messages in agreement and disagreement to a news arti- cle. From an operational point of view, previous works in asynchronous conversations defined ADRs in different 1The language resources can requested at http://sisl.disi.unitn.it/ ways. Bender considered ADRs as relationships among bloggers to a multiparty conversation, expressed at message level, with a post or turn text unit (Bender et al., 2011); Walker …

Content truncated to 500 words