Argument Mining with Structured SVMs and RNNs

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 985–995 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1091 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 985–995 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1091 Argument Mining with Structured SVMs and RNNs Vlad Niculae Cornell University vlad@cs.cornell.edu Joonsuk Park Williams College jpark@cs.williams.edu Claire Cardie Cornell University cardie@cs.cornell.edu Abstract We propose a novel factor graph model for argument mining, designed for settings in which the argumentative relations in a document do not necessarily form a tree structure. (This is the case in over 20% of the web comments dataset we release.) Our model jointly learns elementary unit type classification and argumentative re- lation prediction. Moreover, our model supports SVM and RNN parametrizations, can enforce structure constraints (e.g., transitivity), and can express dependencies between adjacent relations and proposi- tions. Our approaches outperform unstruc- tured baselines in both web comments and argumentative essay datasets. 1 Introduction Argument mining consists of the automatic identi- fication of argumentative structures in documents, a valuable task with applications in policy mak- ing, summarization, and education, among others. The argument mining task includes the tightly-knit subproblems of classifying propositions into ele- mentary unit types and detecting argumentative re- lations between the elementary units. The desired output is a document argumentation graph struc- ture, such as the one in Figure 1, where proposi- tions are denoted by letter subscripts, and the asso- ciated argumentation graph shows their types and support relations between them. Most annotation and prediction efforts in ar- gument mining have focused on tree or forest structures (Peldszus and Stede, 2015; Stab and Gurevych, 2016), constraining argument struc- tures to form one or more trees. This makes the problem computationally easier by enabling the use of maximum spanning tree–style parsing ap- [ Calling a debtor at work is counter-intuitive; ]a [ if collectors are continuously calling someone at work, other employees may report it to the debtor’s supervisor. ]b [ Most companies have es- tablished rules about receiving or making per- sonal calls during working hours. ]c [ If a col- lector or creditor calls a debtor on his/her cell phone and is informed that the debtor is at work, the call should be terminated. ]d [ No calls to em- ployers should be allowed, ]e [ as this jeopardizes the debtor’s job. ]f b (VALUE) a (VALUE) d (POLICY) c (FACT) f (VALUE) e (POLICY) Figure 1: Example annotated CDCP comment.1 proaches. However, argumentation in the wild can be less well-formed. The argument put forth in Figure 1, for instance, consists of two compo- nents: a simple tree structure and a more com- plex graph structure (c jointly supports b and d). In this work, we design a flexible and highly expressive structured prediction model for argu- ment mining, jointly learning to classify elemen- tary units (henceforth propositions) and to identify the argumentative relations between …

Content truncated to 500 words