Towards relation based Argumentation Mining

Proceedings of the 2nd Workshop on Argumentation Mining, pages 29–34, Denver, Colorado, June 4, 2015. c⃝2015 Association for Computational Linguistics Towards relation based Argumentation Mining Lucas Carstens Imperial College London United Kingdom SW2 7AZ, London lc1310@imperial.ac.uk Francesca Toni Imperial College London United Kingdom SW2 7AZ, London ft@imperial.ac.uk Abstract We advocate a relation based approach to Ar- gumentation Mining. Our focus lies on the extraction of argumentative relations instead of the identification of arguments, themselves. By classifying pairs of sentences according to the relation that holds between them we are able to identify sentences that may be factual when considered in isolation, but carry argu- mentative meaning when read in context. We describe scenarios in which this is useful, as well as a corpus of annotated sentence pairs we are developing to provide a testbed for this approach. 1 Introduction Arguments form an integral part of human dis- course. Whether we argue in dialogue with an- other person or advocate the merits of a product in a review, arguments are ubiquitous, in real life as much as in the world wide web. The ever increas- ing amounts of data on the web mean that manual analysis of this content, including debates and ar- guments, seems to become increasingly infeasible. Among other problems Argumentation Mining ad- dresses this issue by developing solutions that au- tomate, or at least facilitate, the process of build- ing Argument Frameworks (AFs) (Amgoud et al., 2008; Dung, 1995) from free text. To build AFs we are generally concerned with two problems, (1) the identification of arguments and (2) the identifi- cation of relations between arguments. With this paper we highlight the intricate link between those two tasks and argue that treating them separately raises a number of issues. On the back of this we propose a relation based way of performing Argu- mentation Mining. Instead of treating the identifica- tion of arguments and their relations to each other as two problems we define it as a single task. We do this by classifying sentences according to whether they stand in an argumentative relation to other sen- tences. We consider any sentence which supports or attacks another sentence to be argumentative. This includes cases such as the one shown in section 3.1, where a sentence contains only parts of an argu- ments (premises or a conclusion) and the remainder of the argument is left implicit for the reader to infer. The remainder of this paper is organised as fol- lows. We discuss related work in section 2. In sec- tion 3 we discuss three issues that arise when trying to decouple the process of identifying arguments and finding relations between them. Following this we describe a relation based approach to perform Argu- mentation Mining for the creation of AFs in section 4. We discuss an application in section 5.1, as well as a corpus design to help us build such applications in section 5.2. We conclude the paper in section 6. 2 Related work Work on Argumentation Mining …

Content truncated to 500 words