News Editorials: Towards Summarizing Long Argumentative Texts

Proceedings of the 28th International Conference on Computational Linguistics, pages 5384–5396 Barcelona, Spain (Online), December 8-13, 2020 5384 News Editorials: Towards Summarizing Long Argumentative Texts Shahbaz Syed ∗ Roxanne El Baff † Khalid Al-Khatib ‡ Johannes Kiesel ‡ Benno Stein ‡ Martin Potthast ∗ ∗Leipzig University †German Aerospace Center (DLR), Oberpfaffenhofen ‡Bauhaus-Universität Weimar <shahbaz.syed@uni-leipzig.de> Abstract The automatic summarization of argumentative texts has hardly been explored. This paper takes a further step in this direction, targeting news editorials, i.e., opinionated articles with a well-defined argumentation structure. With Webis-EditorialSum-2020, we present a corpus of 1330 carefully curated summaries for 266 news editorials. We evaluate these summaries based on a tailored annotation scheme, where a high-quality summary is expected to be thesis-indicative, persuasive, reasonable, concise, and self-contained. Our corpus contains at least three high-quality summaries for about 90% of the editorials, rendering it a valuable resource for the development and evaluation of summarization technology for long argumentative texts. We further report details of both, an in-depth corpus analysis, and the evaluation of two extractive summarization models. 1 Introduction News summarization has been, and still is subject of active research to this day (Yao et al., 2017; Shi et al., 2018; Lin and Ng, 2019). However, the inverted pyramid structure of news reports, where the lead paragraphs often have a summarizing quality in and of themselves (PurdueOWL, 2019), induces a bias in many recent news summarization approaches to just copy the opening sentences (Kedzie et al., 2018; Jung et al., 2019). Hence, these approaches fail at argumentative news articles, such as editorials, whose structure differs from that of news reports. Editorials represent the views of an organization (newspaper) on long-standing societal issues and aim to shape public opinion (Firmstone, 2019; Gajevic, 2016). Compared to news reports, which aim to inform objectively about current events, editorials subjectively assess a controversial topic in order to persuade its audience of a specific stance toward it (Van Dijk, 1995; Hynds and Archibald, 1996). This difference in their goals leads to a difference in linguistic choices. An editorial is usually composed of three discourse parts, namely lead, body, and conclusion (Van Dijk, 1992; Rich, 2015). The lead introduces the issue at hand by starting with an anecdote or a question. The body elaborates on arguments and background information, while the conclusion provides an evaluation as well as (possibly) implicit suggestions and calls to action (Bolívar, 2002). The summary of an editorial must hence be constructed with care in order to preserve its argumentative structure and its persuasive means. Research on automatic (news) summarization has so far neglected argumentative texts in general, and the genre of editorials in particular. With this paper we contribute to closing this gap, taking effective steps toward editorial summarization: 1. We define an annotation scheme tailored to editorial summaries, requiring a high-quality summary to be thesis-indicative, persuasive, reasonable, concise, and self-contained. 2. We create a corpus of 1330 summaries for 266 news editorials (five summaries each), manually acquired and evaluated by operationalizing the proposed annotation …

Content truncated to 500 words