Automatic Claim Negation: Why, How and When

Proceedings of the 2nd Workshop on Argumentation Mining, pages 84–93, Denver, Colorado, June 4, 2015. c⃝2015 Association for Computational Linguistics Automatic Claim Negation: Why, How and When Yonatan Bilu IBM Haifa Research Lab Mount Carmel Haifa, 31905, Israel yonatanb@il.ibm.com Daniel Hershcovich IBM Haifa Research Lab Mount Carmel Haifa, 31905, Israel danielh@il.ibm.com Noam Slonim IBM Haifa Research Lab Mount Carmel Haifa, 31905, Israel noams@il.ibm.com Abstract The main goal of argumentation mining is to analyze argumentative structures within an argument-rich document, and reason about their composition. Recently, there is also in- terest in the task of simply detecting claims (sometimes called conclusion) in general doc- uments. In this work we ask how this set of detected claims can be augmented further, by adding to it the negation of each detected claim. This presents two NLP problems: how to automatically negate a claim, and when such a negated claim can plausibly be used. We present first steps into solving both these problems, using a rule-based approach for the former and a statistical one towards the latter. 1 Introduction In Monty Python’s famous Argument Clinic Sketch (Chapman and Cleese, 1972), Michael Palin is seek- ing a good argument, and John Cleese, too lazy to provide a real argument, simply contradicts what- ever Mr. Palin is saying. MP: An argument isn’t just contradiction. JC: Well! it can be! MP: No it can’t! An argument is a con- nected series of statements intended to es- tablish a proposition. JC: No it isn’t! MP: Yes it is! It isn’t just contradiction! JC: Look, if I argue with you, I must take up a contrary position! MP: Yes, but it isn’t just saying ’no it isn’t’. In this work we aim to explore this last statement from the perspective of an automatic system, aiming to refute an examined claim. Specifically, given a claim, how should we contradict it? Is it enough to say “No it isn’t”, or is a more complicated algorithm required? And when can we plausibly use an au- tomatically generated contradiction? When would it be considered a valid counter claim, and when would it seem as an even less comprehensible ver- sion of John Cleese? The answers to these questions turn out to be less simple than one might expect at first glance. The main goal of argumentation mining is to an- alyze argumentative structures within a document. Typically, documents in which such structures are abundant, such as from the legal domain (Mochales Palau and Moens, 2011; Bach et al., 2013; Ash- ley and Walker, 2013; Wyner et al., 2010), are an- alyzed, and compound argumentative structures, or argumentation schemes, are sought (Walton, 2012). More recently, there is also interest in automati- cally detecting simple argumentative structures, or the building blocks of such structures, in documents which are not argumentative by nature. For exam- ple, in (Aharoni et al., 2014; Levy et al., 2014) it was shown that context-dependent Claims and Evi- dences (sometimes called Conclusion and Grounds, respectively) are fairly common in Wikipedia arti- …

Content truncated to 500 words