Fixed That for You: Generating Contrastive Claims with Semantic Edits

Proceedings of NAACL-HLT 2019, pages 1756–1767 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 1756 Fixed That for You: Generating Contrastive Claims with Semantic Edits Christopher Hidey Department of Computer Science Columbia University New York, NY 10027 chidey@cs.columbia.edu Kathleen McKeown Department of Computer Science Columbia University New York, NY 10027 kathy@cs.columbia.edu Abstract Understanding contrastive opinions is a key component of argument generation. Central to an argument is the claim, a statement that is in dispute. Generating a counter-argument then requires generating a response in contrast to the main claim of the original argument. To generate contrastive claims, we create a cor- pus of Reddit comment pairs self-labeled by posters using the acronym FTFY (fixed that for you). We then train neural models on these pairs to edit the original claim and pro- duce a new claim with a different view. We demonstrate significant improvement over a sequence-to-sequence baseline in BLEU score and a human evaluation for fluency, coherence, and contrast. 1 Introduction In the Toulmin model (1958), often used in com- putational argumentation research, the center of the argument is the claim, a statement that is in dispute (Govier, 2010). In recent years, there has been increased interest in argument genera- tion (Bilu and Slonim, 2016; Hua and Wang, 2018; Le et al., 2018). Given an argument, a system that generates counter-arguments would need to 1) identify the claims to refute, 2) generate a new claim with a different view, and 3) find supporting evidence for the new claim. We focus on this sec- ond task, which requires an understanding of con- trast. A system that can generate claims with dif- ferent views is a step closer to understanding and generating arguments (Apothloz et al., 1993). We build on previous work in automated claim genera- tion (Bilu et al., 2015) which examined generating opposing claims via explicit negation. However, researchers also noted that not every claim has an exact opposite. Consider a claim from Reddit: Get employers out of the business, pass universal single-payer healthcare. (1) This is an example of a policy claim - a view on what should be done (Schiappa and Nordin, 2013). While negation of this claim is a plausible re- sponse (e.g. asserting there should be no change by stating Do not get employers out of the busi- ness, do not pass universal healthcare), negation limits the diversity of responses that can lead to a productive dialogue. Instead, consider a response that provides an alternative suggestion: Get employers out of the business, deregulate and allow cross-state com- petition. (2) In Example 1, the speaker believes in an in- creased role for government while in Example 2, the speaker believes in a decreased one. As these views are on different sides of the political spec- trum, it is unlikely that a single speaker would utter both claims. In related work, de Marneffe et al. (2008) define two sentences as contradictory when they are extremely unlikely to be true simul- taneously. We …

Content truncated to 500 words