Discourse parsing for multi-party chat dialogues

Abstract

In this paper we present the first ever, to the best of our knowledge, discourse parser for multi-party chat dialogues. Discourse in multi-party dialogues dramatically differs from monologues since threaded conversations are commonplace rendering prediction of the discourse structure compelling. Moreover, the fact that our data come from chats renders the use of syntactic and lexical information 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 distributions and then use MST for decoding. We achieve 0.680 F 1 on unlabelled structures and 0.516 F 1 on fully labeled structures which is better than many state of the art systems for monologues, despite the inherent difficulties that multi-party chat dialogues have.

Integer Linear Programming for Discourse Parsing

“…Predicting full SDRSs (V, E 1 , E 2 , ) with E 2 = ∅ has been to date impossible, because no reliable method has been identified in the literature for calculating edges in E 2 . Instead, most approaches Afantenos et al, 2015, for example) simplify the underlying structures by a head replacement strategy (HR) that removes nodes representing CDUs from the original hypergraphs and replacing any incoming or outgoing edges on these nodes on the heads of those CDUs, forming thus dependency structures and not hypergraphs. A similar approach has also been followed by Hirao et al (2013) and Li et al (2014) in the context of RST to deal with multi-nuclear relations.…”

“…8 For purposes of comparison with the ILP decoder, we tested the Chu-Liu-Edmonds version of the classic Maximum Spanning Tree (MST) algorithm McDonald et al (2005) used for discourse parsing by and Li et al (2014) and by Afantenos et al (2015) on the Settlers corpus. This algorithm requires a specific node to be the root, i.e.…”

“…Our underlying data come from multi-party chat dialogues, which require the prediction of DAGs. We use the dependency parsing paradigm, as has been done in the past Li et al, 2014;Afantenos et al, 2015), but we use the underlying formal framework of SDRT and exploit SDRT's notions of left and right distributive relations. We achieve an Fmeasure of 0.531 for fully labeled structures which beats the previous state of the art.…”

“…Afantenos et al (2015) point out that forcing discourse structures to be trees will perforce miss 9% of the links in their corpus, because a significant number of discourse structures in the corpus are not trees. Although Afantenos et al (2015) is the only prior paper we know of that studies dialogue parsing on multi-party dialogue, and that work relied on methods adapted to treelike structures, we think the area of multi-party dialogue and non-treelike discourse structures is ripe for investigation and potentially important for other genres like the discourse analysis of fora (Wang et al, 2011, for example). This paper proposes a method based on constraints using Integer Linear Programming decoding over local probability distributions to investigate both treelike and non-treelike, full discourse structures for multi-party dialogue.…”

“…But non-projective trees cannot always reflect the structure of a discourse either, as Asher and Lascarides (2003) argue on theoretical grounds. Afantenos et al (2015) provide examples in which a question or a comment by speaker S that is addressed to all the engaged parties in the conversation receives an answer from all the other participants, all of which are then acknowledged by S with a simple OK or No worries, thus creating an intuitive, "lozenge" like structure, in which the acknowledgment has several incoming links representing discourse dependencies.…”