Discourse parsing for multi-party chat dialogues

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 928–937, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Discourse parsing for multi-party chat dialogues Stergos Afantenos Eric Kow Nicholas Asher Jérémy Perret IRIT, Université Toulouse & CNRS, Univ. Paul Sabatier, 118 Route de Narbonne, 31062 Toulouse {firstname.lastname@irit.fr} Abstract toIn this paper we present the first ever, the best of our knowledge, dis- course parser for multi-party chat dia- logues. Discourse in multi-party dia- logues dramatically differs from mono- logues since threaded conversations are commonplace rendering prediction of the discourse structure compelling. Moreover, the fact that our data come from chats ren- ders the use of syntactic and lexical in- formation useless since people take great liberties in expressing themselves lexically and syntactically. We use the dependency parsing paradigm as has been done in the past (Muller et al., 2012; Li et al., 2014). We learn local probability distri- butions and then use MST for decoding. We achieve 0.680 F1 on unlabelled struc- tures and 0.516 F1 on fully labeled struc- tures which is better than many state of the art systems for monologues, despite the in- herent difficulties that multi-party chat di- alogues have. 1 Introduction Discourse parsing is a difficult, multifaceted prob- lem involving the understanding and modeling of various semantic and pragmatic phenomena as well as understanding the structural properties that a discourse graph can have. Unsurprisingly, most extant theories and computational approaches pos- tulate an extremely simplified version of discourse structure. One of the most widely cited theories, Rhetorical Structure Theory (RST) (Mann and Thompson, 1987; Mann and Thompson, 1988; Taboada and Mann, 2006), requires that only adja- cent discourse units be connected together with a discourse relation. Another widely cited approach, the Penn Discourse Treebank (PDTB) (Prasad et al., 2008), focuses on decisions about the dis- course connectives that label the attachment of po- tentially arbitrary text spans but does not make any claims as to what the overall discourse struc- ture of the resulting annotation looks like. Further, all computational work on the PDTB takes the at- tachments as given in discourse parsing tasks. In both cases, the attachment problem, finding which discourse units are attached to which, is vastly simplified, though this has enabled researchers to explore various approaches for discourse parsing (Marcu, 2000; Sagae, 2009; Hernault et al., 2010; Joty et al., 2012). Our paper’s main contribution is to provide a discourse parsing model for multi-party chat di- alogue (i.e. typed online dialogue), trained on a large corpus we have developed annotated with full discourse structures. We study attachment problem in detail for this genre, without using the simplifying hypotheses mentioned above that we know to be inadequate. In the following section, we describe the Settlers of Catan game and our corpus in more detail and discuss some problem- atic structures for discourse parsing from our cor- pus. We motivate our choice of a particular dis- course theory, the Segmented Discourse Repre- sentation Theory (SDRT), as the …

Content truncated to 500 words