Text Editing by Command

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5259–5274 June 6–11, 2021. ©2021 Association for Computational Linguistics 5259 Text Editing by Command Felix Faltings♦,∗ Michel Galley♠ Gerold Hintz♠ Chris Brockett♠ Chris Quirk♠ Jianfeng Gao♠ Bill Dolan♠ ♦Department of Computer Science, ETH Z¨urich ♠Microsoft Research fafelix@student.ethz.ch {mgalley,gehint,chrisbkt,chrisq,jfgao,billdol}@microsoft.com Abstract A prevailing paradigm in neural text genera- tion is one-shot generation, where text is pro- duced in a single step. The one-shot setting is inadequate, however, when the constraints the user wishes to impose on the generated text are dynamic, especially when authoring longer documents. We address this limitation with an interactive text generation setting in which the user interacts with the system by is- suing commands to edit existing text. To this end, we propose a novel text editing task, and introduce WikiDocEdits, a dataset of single- sentence edits extracted from Wikipedia revi- sion histories. We show that our Interactive Editor, a transformer-based model trained on this dataset, outperforms baselines and obtains positive results in both automatic and human evaluations. We present empirical and qualita- tive analyses of this model’s performance.1 1 Introduction A long-standing goal of natural language process- ing research has been to generate long-form text (Lebowitz, 1985; Fan et al., 2018; Rashkin et al., 2020). Recent large generative language models such as GPT-2 (Radford et al., 2019), and GPT- 3 (Brown et al., 2020), demonstrate an impres- sive ability to generate fluent text, but their outputs are difficult to control beyond a prompt, and they manifest a tendency to hallucinate facts (Wiseman et al., 2017). Much recent work has thus focused on making such models more controllable (Keskar et al., 2019; Hu et al., 2017; Zhang et al., 2020; Dathathri et al., 2019), and factually grounded (Guu et al., 2020; Liu et al., 2018b). * Work done at Microsoft Research. and1All our code (including code to recreate our data) pre-trained models will be made available at: http://microsoft.com/research/project/ interactive-document-generation Barack Obama was the 44th President of the United States. Barack Obama was the 44th President of the United States and the first African-American to hold the office. Barack Obama was the 44th President of the United States from 2009 to 2017 and the first African-American to hold the office. expand add years in office Figure 1: An illustration of our interactive text gen- eration setting. This is an example generated by our model. The blue panels represent the text being edited, taken from the document shown on the right. The orange panels represent user edit commands. The model grounds edits in query results from a commer- cial search engine. Most such work only considers a one-shot gen- eration setting. Given a set of inputs, which may be a prompt, a control code (Keskar et al., 2019), or a table of data (Liu et al., 2018b) for example, the system generates text in a single step. Humans, though, often produce text through an evolutionary process involving multiple draft-edit cycles. …

Content truncated to 500 words