Korean-Specific Dataset for Table Question Answering

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6114–6120 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6114 Korean-Specific Dataset for Table Question Answering Changwook Jun, Jooyoung Choi, Myoseop Sim, Hyun Kim, Hansol Jang, Kyungkoo Min LG AI Research ISC, 30, Magokjungang 10-ro, Gangseo-gu, Seoul, 07796, Korea {cwjun, jooyoung.choi, myoseop.sim, hyun101.kim, hansol.jang, mingk24}@lgresearch.ai Abstract Existing question answering systems mainly focus on dealing with text data. However, much of the data produced daily is stored in the form of tables that can be found in documents and relational databases, or on the web. To solve the task of question answering over tables, there exist many datasets for table question answering written in English, but few Korean datasets. In this paper, we demonstrate how we construct Korean-specific datasets for table question answering: Korean tabular dataset is a collection of 1.4M tables with corresponding descriptions for unsupervised pre-training language models. Korean table question answering corpus consists of 70k pairs of questions and answers created by crowd-sourced workers. Subsequently, we then build a pre-trained language model based on Transformer and fine-tune the model for table question answering with these datasets. We then report the evaluation results of our model. We make our datasets publicly available via our GitHub repository and hope that those datasets will help further studies for question answering over tables, and for the transformation of table formats. Keywords: Table question-answering, KO-TaBERT, KorWikiTQ 1. Introduction The task of question answering is to correctly an- swer to given questions, which requires a high level of language understanding and machine reading compre- hension abilities. As pre-trained language models on Transformer (Vaswani et al., 2017) have brought sig- nificant improvements in performance in many natu- ral language processing tasks, there have been many studies in machine reading comprehension (MRC) and question answering (QA) tasks (Devlin et al., 2018; Lan et al., 2019; Yang et al., 2019; Yamada et al., 2020; Liu et al., 2019; Clark et al., 2020). There are Stan- ford Question Answering Dataset (SQuAD) bench- marks (Rajpurkar et al., 2016; Rajpurkar et al., 2018), well-known machine reading comprehension bench- marks in the NLP area, that involve reasoning correct answer spans in the evidence document to a given ques- tion. Since SQuAD datasets are composed of pairs of natural language questions and answers, and the cor- responding documents are unstructured textual data, the task is mainly to focus on predicting answers from plain texts. Much of the world’s information produced daily is stored in structured formats such as tables in databases and documents, or tables on the web. Question answer- ing over these structured tables has been generally con- sidered a semantic parsing task in which a natural lan- guage question is translated to a logical form that can be executed to generate the correct answer (Pasupat and Liang, 2015; Zhong et al., 2017; Dasigi et al., 2019; Wang et al., 2019; Rubin and Berant, 2020). There have been many efforts to build semantic …

Content truncated to 500 words