Jointly Extracting and Compressing Documents with Summary State Representations

Proceedings of NAACL-HLT 2019, pages 3955–3966 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 3955 Jointly Extracting and Compressing Documents with Summary State Representations Afonso Mendes♠ Shashi Narayan♢∗ Sebasti˜ao Miranda♠ Zita Marinho♡♠ Andr´e F. T. Martins†♣ Shay B. Cohen♢ ♠Priberam Labs, Alameda D. Afonso Henriques, 41, 2o, 1000-123 Lisboa, Portugal ♢School of Informatics, University of Edinburgh, Edinburgh EH8 9AB, UK ♡ Instituto de Sistemas e Rob´otica, Instituto Superior T´ecnico, 1049-001 Lisboa, Portugal †Instituto de Telecomunicac¸˜oes, Instituto Superior T´ecnico, 1049-001 Lisboa, Portugal ♣Unbabel Lda, Rua Visconde de Santar´em, 67-B, 1000-286 Lisboa, Portugal amm@priberam.com, shashi.narayan@gmail.com, ssm@priberam.com, zam@priberam.com, andre.martins@unbabel.com, scohen@inf.ed.ac.uk Abstract We present a new neural model for text sum- marization that first extracts sentences from a document and then compresses them. The pro- posed model offers a balance that sidesteps the difficulties in abstractive methods while gener- ating more concise summaries than extractive methods. In addition, our model dynamically determines the length of the output summary based on the gold summaries it observes dur- ing training, and does not require length con- straints typical to extractive summarization. The model achieves state-of-the-art results on the CNN/DailyMail and Newsroom datasets, improving over current extractive and abstrac- tive methods. Human evaluations demonstrate that our model generates concise and informa- tive summaries. We also make available a new dataset of oracle compressive summaries de- rived automatically from the CNN/DailyMail reference summaries.1 1 Introduction Text summarization is an important NLP problem with a wide range of applications in data-driven industries (e.g., news, health, and defense). Sin- gle document summarization—the task of gener- ating a short summary of a document preserving its informative content (Sp¨arck Jones, 2007)—has been a highly studied research topic in recent years (Nallapati et al., 2016b; See et al., 2017; Fan et al., 2018; Pasunuru and Bansal, 2018). Modern approaches to single document sum- marization using neural network architectures 1Our dataset and code is available at https:// github.com/Priberam/exconsumm. ∗ Now at Google London. (EXCONSUMM Extractive) • (CNN) A top al Qaeda in the Arabian Peninsula leader–who a few years ago was in a U.S. detention facility–was among five killed in an airstrike in Yemen, the terror group said, showing the organization is vulnerable even as Yemen appears close to civil war. • Ibrahim al-Rubaish died Monday night in what AQAP’s media wing, Al-Malahem Media, called a “crusader airstrike.” (EXCONSUMM Compressive) • (CNN) A top al Qaeda in the Arabian Peninsula leader–who a few years ago was in a U.S. detention facility–was among five killed in an airstrike in Yemen , the terror group said, showing the organization is vulnerable even as Yemen appears close to civil war. • Ibrahim al-Rubaish died Monday night in what AQAP’s media wing, Al-Malahem Media, called a “crusader airstrike.” Figure 1: Summaries produced by our model. For illus- tration, the compressive summary shows the removed spans strike-through. have primarily focused on two strategies: extrac- tive and abstractive. The former select a subset of the sentences to assemble a summary (Cheng and Lapata, 2016; Nallapati …

Content truncated to 500 words