The Role of Discourse Units in Near-Extractive Summarization

By Junyi Jessy Li, Kapil Thadani, Amanda Stent. 2016

64 citations recorded by the source corpus.

Although human-written summaries of documents tend to involve significant edits to the source text, most automated summa-rizers are extractive and select sentences verbatim. In this work we examine how elementary discourse units (EDUs) from Rhetorical Structure Theory can be used to extend extractive summarizers to produce a wider range of human-like summaries. Our analysis demonstrates that EDU segmentation is effective in preserving human-labeled summarization concepts within sentences and also aligns with near-extractive summaries constructed by news editors. Finally, we show that us-ing EDUs as units of content selection instead of sentences leads to stronger summarization performance in near-extractive scenarios, especially under tight budgets.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Mohit Bansal
2. Jianfeng Gao
3. Yejin Choi
4. Caiming Xiong
5. Richard Socher
6. Zhe Gan
7. Luke S. Zettlemoyer
8. Xipeng Qiu

In Natural Language Processing

1. Mohit Bansal
2. Jianfeng Gao
3. Yejin Choi
4. Caiming Xiong
5. Richard Socher
6. Xipeng Qiu
7. Graham Neubig
8. Zhe Gan

Citations in the corpus

• A Deep Reinforced Model for Abstractive Summarization (2017, methodology): “Even though this dataset has been used to train extractive summarization systems (Hong and Nenkova, 2014; Li et al., 2016) or closely-related models for predicting the importance of a phrase in an article (Yang and Nenkova, 2014; Nye and Nenkova, 2015; Hong et al., 2015), we are the first group to…”
• Neural Text Summarization: A Critical Evaluation (2019, methodology): “It has been primarily used for extractive summarization (Hong and Nenkova, 2014; Li et al., 2016) and phrase-importance prediction (Yang and Nenkova, 2014; Nye and Nenkova, 2015).”
• Deep Communicating Agents for Abstractive Summarization (2018, methodology): “We use the scripts provided in Li et al. (2016) to extract and pre-process the NYT dataset with some modifications in order to replicate the pre-processing steps presented in Paulus et al. (2018).”
• Exploring Content Selection in Summarization of Novel Chapters (2020, methodology): “Previous work has summarized documents using Rhetorical Structure Theory (RST) (Mann and Thompson, 1988) to extract elementary discourse units (EDUs) for compression and more contentpacked summaries (Daumé III and Marcu, 2002; Li et al., 2016; Arumae et al., 2019).”
• Intrinsic Evaluation of Summarization Datasets (2020, background): “Conversely, NYT has been used to build extractive systems (Hong and Nenkova, 2014; Li et al., 2016).”
• Discourse-Aware Neural Extractive Text Summarization (2019, background): “Li et al. (2016) demonstrated that using EDUs as units of content selection leads to stronger summarization performance.”
• Joint Parsing and Generation for Abstractive Summarization (2019, background): “…and Klein 2011; Wang et al. 2013; Durrett, Berg-Kirkpatrick, and Klein 2016), rhetorical structure (Christensen et al. 2013; Yoshida et al. 2014; Li, Thadani, and Stent 2016), and abstract meaning representation (Liu et al. 2015; Liao, Lebanoff, and Liu 2018; Hardy and Vlachos 2018) to…”
• Recent advances in document summarization (2017, background): “recent study [108] demonstrates that segmenting EDUs (elementary discourse units, cf.”