BLCU_NLP at SemEval-2018 Task 12: An Ensemble Model for Argument Reasoning Based on Hierarchical Attention

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 1104–1108 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics BLCU NLP at SemEval-2018 Task 12: An Ensemble Model for Argument Reasoning Based on Hierarchical Attention Meiqian Zhao, Chunhua Liu, Lu Liu, Yan Zhao, Dong Yu∗ Beijing Language and Culture University {zhaomq195, chunhualiu596, luliu.nlp, zhaoyan.nlp}@gmail.com, yudong@blcu.edu.cn Abstract The argument comprehension reasoning task aims to reconstruct and analyze the argument reasoning. To comprehend an argument and fill the gap between claims and reasons, it is vital to find the implicit supporting warrants behind. In this paper, we propose a hierarchi- cal attention model to identify the right war- rant which explains why the reason stands for the claim. Our model focuses not only on the similar part between warrants and other infor- mation but also on the contradictory part be- tween two opposing warrants. In addition, we use the ensemble method for different mod- els. Our model achieves an accuracy of 61%, ranking second in this task. Experimental re- sults demonstrate that our model is effective to make correct choices. 1 Introduction The argument reasoning comprehension is a cru- cial part for natural language understanding and inference, since argument comprehension requires reconstruction and analysis for its reasoning. Lack of common sense makes it difficult to infer claims from corresponding reasons directly. To fill the gap between claims and reasons, several explo- rations have been performed on argumentative structure of a debate (Hastings, 1963; Walton et al., 2008; Walton, 1990). Argument reason- ing comprehension, a new task in SemEval 2018, sheds some light on the core of reasoning in nat- ural language argumentation: implicit warrants, which are seen as a bridge between claims and rea- sons. Given a reason R, a claim C and two alterna- tive warrants W0 and W1, the goal of this task is to identify the right warrant which can justify the use of R as support for C. The difficulty of the task is the warrants are plausible and lexically close ∗*The corresponding author but lead to contradicting claims (Habernal et al., 2018). To be more specific, the reason R and the claim C are propositions extracted from a natural lan- guage argument. And warrant W is the relation between R and C which is characterized by a rule of inference (Newman and Marshall, 1992). Wal- ton proposed that an argument refers to a claim based on reasons given in the premises (Walton, 1990). The most central part of this task is how to find the warrant for the given R and C. In the argument reasoning comprehension task, the or- ganizer extracts the instances from Room for De- bate section of the New York Times. After a com- plex crowd-sourcing process, 1970 valid instances are provided for the task. Two alternative warrants are provided as candidates, where one can justify the use of R as support for C and the other one can justify R as support for the opposite side of C. …

Content truncated to 500 words