Towards Detecting Counter-considerations in Text

Proceedings of the 2nd Workshop on Argumentation Mining, pages 104–109, Denver, Colorado, June 4, 2015. c⃝2015 Association for Computational Linguistics Towards Detecting Counter-considerations in Text Andreas Peldszus Applied Computational Linguistics FSP Cognitive Science University of Potsdam peldszus@uni-potsdam.de Manfred Stede Applied Computational Linguistics FSP Cognitive Science University of Potsdam stede@uni-potsdam.de Abstract Argumentation mining obviously involves finding support relations between statements, but many interesting instances of argumen- tation also contain counter-considerations, which the author mentions in order to preempt possible objections by the readers. A counter- consideration in monologue text thus involves a switch of perspective toward an imagi- nary opponent. We present a classification ap- proach to classifying counter-considerations and apply it to two different corpora: a se- lection of very short argumentative texts pro- duced in a text generation experiment, and a set of newspaper commentaries. As expected, the latter pose more difficulties, which we in- vestigate in a brief error anaylsis. 1 Introduction The exchange of argument and objection is obvi- ously most typical for dialogue, but to a good ex- tent it is also present in monologue text: Authors do not only provide justifications for their own po- sition – they can also mention potential objections and then refute or outweigh them. In this way they demonstrate to have considered the position of “the other side”, which altogether is designed to rein- force their own position. We use the term ‘counter- consideration’ in a general sense to cover all such moves of an author, no matter whether they are di- rected at the conclusion of the text or at an interme- diate argument, or at some support relation, and irre- spective of whether they are explicitly refuted by the author or merely mentioned and left outweighed by the mass of arguments in favour of the main claim.1 For an author, presenting a counter-consideration involves a switch of perspective by temporarily adopting the opposing viewpoint and then moving back to one’s own. This is a move that generally re- quires some form of explicit linguistic marking so that the reader can follow the line of argumenta- tion. The kinds of marking include explicit belief attribution followed by a contrastive connective sig- naling the return (“Some people think that X. How- ever, this ...”), and there can also be quite com- pact mentions of objections, as in “Even though the project is expensive, we need to pursue it, be- cause...” Detecting counter-considerations is thus a subtask of argumentation mining. It involves identifying the two points of perspective switching, which we henceforth call a move from the proponent role to the opponent role and back. Thus the task can be operationalized as labelling segments of argumenta- tive text in terms of these two roles. Then, counter- considerations are segments labeled as “opponent”. We study this classification problem using two different corpora: a colletion of user-generated short “microtexts”, where we expect the task to be rela- tively easy, and a set of argumentative newspaper pieces that explicitly argue in favour of or …

Content truncated to 500 words