NEUer at SemEval-2021 Task 4: Complete Summary Representation by Filling Answers into Question for Matching Reading Comprehension

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 827–832 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 827 NEUer at SemEval-2021 Task 4: Complete Summary Representation by Filling Answers into Question for Matching Reading Comprehension Zhixiang Chen1†, Yikun Lei1†, Pai Liu1,2†, Guibing Guo1* 1Northeastern University, Shenyang, China 2Westlake University, Hangzhou, China † These three authors contributed equally. {zxchen8830, ethanlei9931, pailiu1998}@gmail.com guogb@swc.neu.edu.cn Abstract SemEval task 4 aims to find a proper option from multiple candidates to resolve the task of machine reading comprehension. Most ex- isting approaches propose to concat question and option together to form a context-aware model. However, we argue that straightfor- ward concatenation can only provide a coarse- grained context for the MRC task, ignoring the specific positions of the option relative to the question. In this paper, we propose a novel MRC model by filling options into the ques- tion to produce a fine-grained context (defined as summary) which can better reveal the re- lationship between option and question. We conduct a series of experiments on the given dataset, and the results show that our approach outperforms other counterparts to a large ex- tent. 1 Introduction In order to make the computer understand, rep- resent and express better, we study the abil- ity of MRC to understand Abstract defini- tions(Spreen and Schulz, 1966; Changizi, 2008) in the Reading Comprehension of Abstract Meaning task(ReCAM). In ReCAM task, there are two kinds of abstract definitions, one is imperceptibility and the other is nonspecificity. To evaluate the model performance comprehensively, three subtasks are designed(Zheng et al., 2021): Subtask 1) ReCAM-Imperceptibility - A pas- sage, a question with a placeholder, and answers used to fill in the placeholder are given. The an- swers are all imperceptibility words so as to evalu- atethe model’s ability to comprehend these imper- ceptibility words. Subtask 2) ReCAM-Nonspecificity - A passage, a question with a placeholder, and answers used to fill in the placeholder are given. The answers are all Nonspecificity words so as to evaluatethe model’s ability to comprehend these Nonspecificity words. Subtask 3) To provide more insights into the relationship between two abstract views, it requires to test the performance of the model trained on one data set and evaluated on the other one. Because question is obtained by hollowing out summary of passage and answer is abstract words to fill in(Zheng et al., 2021), there is a seman- tically complementary relationship between the question and answer. As shown in the Figure 1, in the previous work(Jin et al., 2020; Zhu et al., 2020), the model inputs passage, question and an- swer independently. It can only provide a coarse- grained context for the MRC task, ignoring the specific positions of the option relative to the ques- tion. Quesiton cannot be a complete sentence, and the answer lacks contextual information about the question. To solve this problem, we introduce Complete Summary Representation by Filling An- swers into Question for Maching Reading Com- prehension(ComSR). ComSR convert the original question and …

Content truncated to 500 words