YNU-HPCC at SemEval-2018 Task 12: The Argument Reasoning Comprehension Task Using a Bi-directional LSTM with Attention Model

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 1109–1113 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics YUN-HPCC at SemEval-2018 Task 12: The Argument Reasoning Comprehension Task Using a Bi-directional LSTM with Attention Model Quanlei Liao, Xutao Yang, Jin Wang and Xuejie Zhang School of Information Science and Engineering Yunnan University Kunming, P.R. China Contact:xjzhang@ynu.edu.cn Abstract An argument is divided into two parts, the claim and the reason. To obtain a clear- er conclusion, some additional explanation is required. In this task, the explanation- s are called warrants. This paper introduces a bi-directional long short term memory (Bi- LSTM) with an attention model to select a correct warrant from two to explain an argu- ment. We address this question as a question- answering system. For each warrant, the mod- el produces a probability that it is correct. Fi- nally, the system chooses the highest correc- t probability as the answer. Ensemble learn- ing is used to enhance the performance of the model. Among all of the participants, we ranked 15th on the test results. 1 Introduction Reasoning is an important part of human logical thinking. It gives us the ability to draw fresh con- clusions from some of the known points (Judea, 1988). Argument is the basis for reasoning. Ex- cept for the argument’s claim and reason, usually, it needs some additional information. Therefore, what we know is the additional information and arguments reason. The claim also needs warrants 1.for an explanation. An example is shown in Table Obviously, A is a reasonable explanation. The task is to get the reader to find a reasonable ex- planation for the known messages and claims in the two warrants. Due to the small number of al- ternative warrants, this problem can be considered to be a binary classification problem. This idea can be used as the baseline model. However, for system scalability and effectiveness, we treat this problem as the regression problem of probability prediction. The idea calculates the probability for each warrant that it is correct. Because of the di- versity of natural language expression, there are Topic Should It Be Illegal to De- claw Your Cat? Additional Info With legislation pending, New York could become the first state to make re- moving the claws of a cat a crime. Argument Declawing is a crime; in- stead, people should be educated on proper care and training. And since ..., Claim It should be illegal to de- claw your cat . Warrant0 A) owners should not have the right to be in charge of their animals. Warrant1 B) owners should have the right to be in charge of their animals. Table 1: An Example of the Task. many ways in which the same meaning can be ex- pressed. Thus, this approach can be better to ad- dress this situation (Collobert et al., 2011). Another benefit of addressing the problem in this way is to make the problem similar in form to the multi-choice question-answering system. The …

Content truncated to 500 words