Keep It Simple: Unsupervised Simplification of Multi-Paragraph Text

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 6365–6378 August 1–6, 2021. ©2021 Association for Computational Linguistics 6365 Keep it Simple: Unsupervised Simplification of Multi-Paragraph Text Philippe Laban UC Berkeley Tobias Schnabel Microsoft Paul N. Bennett Microsoft Marti A. Hearst UC Berkeley∗ Abstract This work presents Keep it Simple (KiS), a new approach to unsupervised text simplifica- tion which learns to balance a reward across three properties: fluency, salience and simplic- ity. We train the model with a novel algorithm to optimize the reward (k-SCST), in which the model proposes several candidate simpli- fications, computes each candidate’s reward, and encourages candidates that outperform the mean reward. Finally, we propose a realis- tic text comprehension task as an evaluation method for text simplification. When tested on the English news domain, the KiS model out- performs strong supervised baselines by more than 4 SARI points, and can help people com- plete a comprehension task an average of 18% faster while retaining accuracy, when com- pared to the original text. 1 Introduction The main objective of text simplification is to make a complex text accessible to a wide audience by increasing its readability. In contrast with text sum- marization – in which key content is selected to remain in the summary and other content is elided – in text simplification, ideally all relevant content is preserved. We propose that text simplification algorithms need to balance three properties: (1) fluency: the simplified text should use well-formed English sen- tences, (2) salience: the simplified text should relay the same information as the original, and (3) sim- plicity: the simplified text should be syntactically and lexically simpler than the original. Figure 1 provides intuition for the necessity of each of the three properties. It shows the origi- nal text and the output of the full proposed model compared to three reduced versions: ∗ Author emails: {phillab,hearst}@berkeley.edu, {Tobias.Schnabel,Paul.N.Bennett}@microsoft.com Original: NASA's Curiosity rover just celebrated a major milestone — 3,000 days on the surface of Mars. To mark the occasion, the space agency has released a stunning new panorama of the red planet, captured by the rover. Model Full: NASA's Curiosity rover has now passed 3,000 days of travel on the surface of Mars. To mark the milestone, the space agency released a huge panorama of Mars, as seen by the rover. Model No Fluency: NASA's Curiosity rover. celebrated. A major milestone — 3,000 days on. The of.. To mark. The space agency has. a stunning new panorama.. red planet. captured by. The rover. However Model No Salience: NASA's Curiosity rover just celebrated a major milestone. The space agency has released a stunning new panoramic of the red planet, captured by the team. It was by the rover's panoramic camera. Model No Simplicity: NASA's Curiosity rover has celebrated a major milestone, 3,000 days on the ground of Mars. To mark the occasion, the space agency has unveiled a stunning new panoramic view of …

Content truncated to 500 words