Controlling Text Edition by Changing Answers of Specific Questions

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 1288–1299 August 1–6, 2021. ©2021 Association for Computational Linguistics 1288 Controlling Text Edition by Changing Answers of Specific Questions Lei Sha, Patrick Hohenecker, Thomas Lukasiewicz Department of Computer Science University of Oxford, United Kingdom {lei.sha, thomas.lukasiewicz}@cs.ox.ac.uk patrick@serein.ai Abstract In this paper, we introduce the new task of con- trollable text edition, in which we take as input a long text, a question, and a target answer, and the output is a minimally modified text, so that it fits the target answer. This task is very impor- tant in many situations, such as changing some conditions, consequences, or properties in a le- gal document, or changing some key informa- tion of an event in a news text. This is very challenging, as it is hard to obtain a parallel corpus for training, and we need to first find all text positions that should be changed and then decide how to change them. We constructed the new dataset WIKIBIOCTE for this task based on the existing dataset WIKIBIO (origi- nally created for table-to-text generation). We use WIKIBIOCTE for training, and manually labeled a test set for testing. We also propose novel evaluation metrics and a novel method for solving the new task. Experimental results on the test set show that our proposed method is a good fit for this novel NLP task. 1 Introduction In many cases, we need to change some specific content in a document. For example, in the legal domain, the items and conditions in contract doc- uments often need to be revised many times. We would like to use artificial intelligence to conduct this process for human editors. A major difficulty of this process is that the machine learning model should decide where to edit and how to edit. Usually, the place of specific content (“where to edit”) can be located by a question, and the content updating (“how to edit”) can be determined by the answer of the question. Therefore, in this paper, we propose the new task of controllable text edition (CTE). In this task, we would like to achieve the fol- lowing goal: adjust some content of a document D, to make the answer A of a document-related ques- tion Q changed to a new answer A′. For example, Lang Ping (born 10 December 1960) is a former Chinese volleyball player and the current head coach of China women's national volleyball team. She was the former head coach of the United States women's national volleyball team, herself being the MVP of women volleyball in 1984 Olympics. Question (Q): What is Lang PingNs current job? Original Answer (A): the Lang Ping (born 10 December 1960) is a former Chinese volleyball player and the current director of the National Sports Administration. She was the former head coach of the United States women's national volleyball team, herself being the MVP of women volleyball in 1984 Olympics. New Answer (AN): the director of the National Sports Administration Origin document …

Content truncated to 500 words