DirectQuote: A Dataset for Direct Quotation Extraction and Attribution in News Articles

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6959–6966 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6959 DirectQuote: A Dataset for Direct Quotation Extraction and Attribution in News Articles Yuanchi Zhang1 and Yang Liu1,2,3,4,5 1Department of Computer Science and Technology, Tsinghua University, 2Institute for AI Industry Research, Tsinghua University 3Institute for Artificial Intelligence, Tsinghua University, 4Beijing National Research Center for Information Science and Technology 5Beijing Academy of Artificial Intelligence Beijing China, Beijing China yuanchi-21@mails.tsinghua.edu.cn, liuyang2011@tsinghua.edu.cn Abstract Quotation extraction and attribution are challenging tasks, aiming at determining the spans containing quotations and attributing each quotation to the original speaker. Applying this task to news data is highly related to fact-checking, media monitoring and news tracking. Direct quotations are more traceable and informative, and therefore of great significance among different types of quotations. Therefore, this paper introduces DirectQuote, a corpus containing 19,760 paragraphs and 10,279 direct quotations manually annotated from online news media. To the best of our knowledge, this is the largest and most complete corpus that focuses on direct quotations in news texts. We ensure that each speaker in the annotation can be linked to a specific named entity on Wikidata, benefiting various downstream tasks. In addition, for the first time, we propose several sequence labeling models as baseline methods to extract and attribute quotations simultaneously in an end-to-end manner. Keywords: Direct Quotation Extraction, Direct Quotation Attribution, News Corpus 1. Introduction In linguistics, a quotation is defined as a repetition that covers all kinds of thought, speech and text (Semino and Short, 2004). It is regarded as a crucial method of expressing opinions and perceptions attributed to some speakers (Cappelen and Lepore, 2012). Con- cretely, there are three types of quotations: direct quota- tions, indirect quotations and mixed quotations (Table 1). Among all kinds of quotations, the entire content of the direct quotation (O’Keefe et al., 2013) is in quota- tion marks, which means that what the speaker said is paraphrased verbatim. Direct quotations are of considerable significance among all types of quotations. On the one hand, news writers attribute direct quotations to speakers, making claims credible and authoritative, leading to more trace- able and convincing news (WP et al., 2020). These direct quotations from politicians, public figures, and other celebrities improve the authenticity and fairness of news. On the other hand, the development of so- cial media and advanced language generation models, such as GPT, has led to the proliferation of fake news (Floridi and Chiriatti, 2020). Troubled by rampant me- dia manipulators, people are increasingly doubting the political stance and legitimacy of the news press. There- fore, direct quotations in the news are essential to ensure the transparency and accountability of news (WP et al., 2020). Besides news writing, direct quotations are also actively involved in various NLP tasks. Since direct quotations are highly subjective expressions and assertions, they are used in opinion mining and claim detection task (Balahur et al., 2009) to discover opinions, sentiment analysis …

Content truncated to 500 words