Dialo-AP: A Dependency Parsing Based Argument Parser for Dialogues

Proceedings of the 29th International Conference on Computational Linguistics, pages 887–901 October 12–17, 2022. 887 Dialo-AP: A Dependency Parsing Based Argument Parser for Dialogues Sougata Saha, Souvik Das, Rohini Srihari State University of New York at Buffalo Department of Computer Science and Engineering {sougatas, souvikda, rohini}@buffalo.edu Abstract While neural approaches to argument mining (AM) have advanced considerably, most of the recent work has been limited to parsing mono- logues. With an urgent interest in the use of conversational agents for broader societal ap- plications, there is a need to advance the state- of-the-art in argument parsers for dialogues. This enables progress towards more purposeful conversations involving persuasion, debate and deliberation. This paper discusses Dialo-AP, an end-to-end argument parser that constructs argument graphs from dialogues. We formulate AM as dependency parsing of elementary and argumentative discourse units; the system is trained using extensive pre-training and curricu- lum learning comprising nine diverse corpora. Dialo-AP is capable of generating argument graphs from dialogues by performing all sub- tasks of AM. Compared to existing state-of- the-art baselines, Dialo-AP achieves significant improvements across all tasks, which is further validated through rigorous human evaluation. 1 Introduction Argumentation is the process of reasoning sys- tematically in support of an idea, action, or the- ory. It is prevalent in our daily communication and conversations, including online conversations. Since argumentation represents an intrinsic human attribute, the ability of artificial agents (bots) to exhibit this skill can be seen as strong evidence for judging such agents as “human-like”. While computational models of argumentation have been investigated (Bench-Capon and Dunne, 2007; Rah- wan and Simari, 2009; Atkinson et al., 2017), cur- rent progress is impeded by the scarcity of large scale corpora exemplifying use of argumentation and reasoning patterns from discourse. Such cor- pora are necessary if we are to make progress in training argumentative conversational agents. In this paper we experiment with computational ar- gumentation mining (AM) (Mochales and Moens, Figure 1: Dependency representation of dialogical argu- ments, across two turns. 2011; Lippi and Torroni, 2016; Lawrence and Reed, 2019) for automatically analyzing discourse at a pragmatics level, and parsing argumentation struc- tures from dialogues. Furthermore, although con- siderable research can be found in the field of AM, most of the current work has focused on parsing monologues (micro-level models), while neglect- ing dialogues (macro-level models) (Bentahar et al., 2010; Grasso, 2002). Here, we aim to fill the re- search gap in dialogical models for AM by propos- ing Dialo-AP, a novel end-to-end argument parser for dialogues. Arguments primarily comprise claims and premises, with the claim being the central contro- versial statement of an argument, and the premise provides reasoning by supporting or attacking the claims (Stab and Gurevych, 2014b). End-to-end AM for dialogues generally involves performing text segmentation, component classification, and intra/inter-turn relation detection & classification. Formulating AM as dependency parsing (DP) prob- lem operating at a mixture of elementary dis- course unit (EDU) and argumentative discourse unit (ADU) granularity, Dialo-AP is an end-to-end argument parser which takes as …

Content truncated to 500 words