Text-to-Table: A New Way of Information Extraction

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2518 - 2533 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Text-to-Table: A New Way of Information Extraction Xueqing Wu1, Jiacheng Zhang2, Hang Li2 University of Illinois Urbana-Champaign1 ByteDance AI Lab2 xueqing8@illinois.edu zhangjiacheng.grit,lihang.lh@bytedance.com Abstract We study a new problem setting of informa- tion extraction (IE), referred to as text-to-table. In text-to-table, given a text, one creates a ta- ble or several tables expressing the main con- tent of the text, while the model is learned from text-table pair data. The problem set- ting differs from those of the existing meth- ods for IE. First, the extraction can be carried out from long texts to large tables with com- plex structures. Second, the extraction is en- tirely data-driven, and there is no need to ex- plicitly define the schemas. As far as we know, there has been no previous work that studies the problem. In this work, we formalize text- to-table as a sequence-to-sequence (seq2seq) problem. We first employ a seq2seq model fine- tuned from a pre-trained language model to per- form the task. We also develop a new method within the seq2seq approach, exploiting two additional techniques in table generation: ta- ble constraint and table relation embeddings. We consider text-to-table as an inverse prob- lem of the well-studied table-to-text, and make use of four existing table-to-text datasets in our experiments on text-to-table. Experimen- tal results show that the vanilla seq2seq model can outperform the baseline methods of using relation extraction and named entity extrac- tion. The results also show that our method can further boost the performances of the vanilla seq2seq model. We further discuss the main challenges of the proposed task. The code and data are available at https://github. com/shirley-wu/text_to_table. 1 1 Introduction Information extraction (IE) is a task that aims to extract information of interest from text data and represent the extracted information in a structured form. Traditional IE tasks include named entity recognition which recognizes entities and their 1The work was done when Xueqing Wu was an intern at ByteDance AI Lab. Figure 1: An example of text-to-table from the Rotowire dataset. The text is a report of a basketball game, and the tables are the scores of the teams and players. types (Huang et al., 2015; Ma and Hovy, 2016; Lample et al., 2016; Devlin et al., 2019), relation ex- traction which identifies the relationships between entities (Zheng et al., 2017; Zeng et al., 2018; Luan et al., 2019; Zhong and Chen, 2021), etc. Since the results of IE are structured, they can be easily used by computer systems in different applications such as text mining. In this work, we study IE in a new setting, re- ferred to as text-to-table. First, the system receives a training dataset containing text-table pairs. Each text-table pair contains a text and a table (or ta- bles) representing information needed for the tar- get application extracted from the text. The system learns a …

Content truncated to 500 words