OPERA: Operation-Pivoted Discrete Reasoning over Text

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1655 - 1666 July 10-15, 2022 ©2022 Association for Computational Linguistics OPERA: Operation-Pivoted Discrete Reasoning over Text Yongwei Zhou1, Junwei Bao2∗, Chaoqun Duan2, Haipeng Sun2, Jiahui Liang2, Yifan Wang2, Jing Zhao2, Youzheng Wu2, Xiaodong He2, Tiejun Zhao1† 1Harbin Institute of Technology 2JD AI Research ywzhou@hit-mtlab.net baojunwei@jd.com tjzhao@hit.edu.cn Abstract Machine reading comprehension (MRC) that requires discrete reasoning involving symbolic operations, e.g., addition, sorting, and count- ing, is a challenging task. According to this nature, semantic parsing-based methods pre- dict interpretable but complex logical forms. However, logical form generation is nontriv- ial and even a little perturbation in a logical form will lead to wrong answers. To alleviate this issue, multi-predictor -based methods are proposed to directly predict different types of answers and achieve improvements. However, they ignore the utilization of symbolic opera- tions and encounter a lack of reasoning ability and interpretability. To inherit the advantages of these two types of methods, we propose OPERA, an operation-pivoted discrete reason- ing framework, where lightweight symbolic operations (compared with logical forms) as neural modules are utilized to facilitate the rea- soning ability and interpretability. Specifically, operations are first selected and then softly exe- cuted to simulate the answer reasoning proce- dure. Extensive experiments on both DROP1 and RACENum datasets show the reasoning ability of OPERA. Moreover, further analysis verifies its interpretability. 2 1 Introduction Machine reading comprehension (MRC) that re- quires discrete reasoning is a valuable and chal- lenging task (Dua et al., 2019), especially involv- ing symbolic operations such as addition, sorting, and counting. The examples in Table 1 illustrate the task. To answer the question “Who threw the longest pass”, it requires a model to choose the per- son with the longest pass from all the people and ∗Work was done during the first author’s internship at JD AI mentored by Junwei Bao: baojunwei001@gmail.com. †Corresponding author. 1https://leaderboard.allenai.org/drop/ submissions/public 2Codes is released at https://github.com/ JD-AI-Research-NLP/OPERA. Passage: Houston would tie the game in the second quarter with kicker Kris Brown getting a 53-yard and a 24-yard field goal. Oakland would take the lead in the third quarter with wide receiver Johnnie Lee Higgins catching a 29-yard touchdown pass from Russell, followed up by an 80-yard punt return... Question: Who threw the longest pass? Answer: Russell Table 1: An example of question-answer pair along with a passage from DROP dataset (Dua et al., 2019). Question words in color indicate the potential operations for reasoning, i.e., ARGMAX and KEY_VALUE. corresponding distance pairs based on the given passage. This task has various application scenar- ios in the real world, such as analyzing financial reports and sports news. Existing approaches for this task can be roughly divided into two categories: the semantic-parsing- based (Chen et al., 2020b; Gupta et al., 2020) and the multi-predictor-based methods (Dua et al., 2019; Ran et al., 2019; Hu et al., 2019; Chen et al., 2020a; Zhou et al., 2021). The former …

Content truncated to 500 words