Single Document Summarization as Tree Induction

Proceedings of NAACL-HLT 2019, pages 1745–1755 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 1745 Single Document Summarization as Tree Induction Yang Liu, Ivan Titov and Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB yang.liu2@ed.ac.uk, ititov@inf.ed.ac.uk, mlap@inf.ed.ac.uk Abstract In this paper we conceptualize single- document extractive summarization as a tree induction problem. In contrast to previous approaches (Marcu, 1999; Yoshida et al., 2014) which have relied on linguistically mo- tivated document representations to generate summaries, our model induces a multi-root dependency tree while predicting the output summary. Each root node in the tree is a summary sentence, and the subtrees attached to it are sentences whose content relates to or explains the summary sentence. We design a new iterative refinement algorithm: it induces the trees through repeatedly refining the structures predicted by previous iterations. We demonstrate experimentally on two benchmark datasets that our summarizer1 performs competitively against state-of-the-art methods. 1 Introduction Single-document summarization is the task of au- tomatically generating a shorter version of a doc- ument while retaining its most important informa- tion. The task has received much attention in the natural language processing community due to its potential for various information access applica- tions. Examples include tools which digest textual content (e.g., news, social media, reviews), answer questions, or provide recommendations. Of the many summarization paradigms that have been identified over the years (see Mani 2001 and Nenkova and McKeown 2011 for comprehen- sive overviews), two have consistently attracted attention. In abstractive summarization, various text rewriting operations generate summaries us- ing words or phrases that were not in the original text, while extractive approaches form summaries by copying and concatenating the most important spans (usually sentences) in a document. Recent 1Our code is publicly available at https://github. com/nlpyang/SUMO. approaches to (single-document) extractive sum- marization frame the task as a sequence labeling problem taking advantage of the success of neu- ral network architectures (Bahdanau et al., 2015). The idea is to predict a label for each sentence specifying whether it should be included in the summary. Existing systems mostly rely on recur- rent neural networks (Hochreiter and Schmidhu- ber, 1997) to model the document and obtain a vector representation for each sentence (Nallap- ati et al., 2017; Cheng and Lapata, 2016). Inter- sentential relations are captured in a sequential manner, without taking the structure of the doc- ument into account, although the latter has been shown to correlate with what readers perceive as important in a text (Marcu, 1999). Another prob- lem in neural-based extractive models is the lack of interpretability. While capable of identifying summary sentences, these models are not able to rationalize their predictions (e.g., a sentence is in the summary because it describes important con- tent upon which other related sentences elaborate). The summarization literature offers examples of models which exploit the structure of the un- derlying document, inspired by existing theories of discourse such as Rhetorical Structure The- …

Content truncated to 500 words