Identification and Characterization of Newsworthy Verbs in World News

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1440–1445, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Identification and Characterization of Newsworthy Verbs in World News Benjamin Nye University of Pennsylvania bepnye@gmail.com Ani Nenkova University of Pennsylvania nenkova@seas.upenn.edu Abstract We present a data-driven technique for acquir- ing domain-level importance of verbs from the analysis of abstract/article pairs of world news articles. We show that existing lexical re- sources capture some the semantic character- istics for important words in the domain. We develop a novel characterization of the associ- ation between verbs and personal story narra- tives, which is descriptive of verbs avoided in summaries for this domain. 1 Introduction Summarization, either by people or machine, calls for the ability to identify important content. Com- putational approaches to identifying important con- tent fall into the two extremes of a possible spec- trum. On one end, the types of important infor- mation for a given domain and topic are prede- fined as information extraction templates defined by experts, as in the earliest approaches to multi- document summarization (Radev and McKeown, 1998) and the recently introduced guided summa- rization (Owczarzak and Dang, 2011). On the other extreme, traditional systems work only with indica- tors of importance coming solely from the input to be summarized, or possibly also from the context of the input, i.e . analyzing the anchor text of links to a webpage, or comments on a blog post or citations to a scientific article (Nenkova and McKeown, 2012). Here we explore the feasibility of data-driven identification of important information in the world news domain. We specifically focus on the analy- sis of verbs, which is the first step of identifying event types of special interest. The goal is to col- lect evidence of verb importance globally, without regard to a particular input or its context. Such ideas have been explored in the past as subcompo- nents of extractive summarizers (Schiffman et al., 2002; Hong and Nenkova, 2014) or as features de- rived from small datasets for sentence compression (Woodsend and Lapata, 2012). In contrast, in our work we rely on large corpora and exclusively focus on the task of acquiring input independent indica- tors of importance. We also constrain our analysis to a single domain, which allows us to examine the semantic aspects of the verbs that may contribute to their perceived importance. We leverage a dataset of human-written sum- maries of news articles to objectively ground the def- inition of word importance. Summaries are intended to convey important information while omitting the less important pieces, so words that are important in a newsworthy sense will occur more frequently in summaries. The same data and intuition was used recently to develop a large corpus for determining entity salience (Dunietz and Gillick, 2014). We derive a list of over one thousand verbs that have statistically significant bias to appear in the summaries (important verbs) and verbs with higher rate …

Content truncated to 500 words