Detecting Contextomized Quotes in News Headlines by Contrastive Learning

Findings of the Association for Computational Linguistics: EACL 2023, pages 697–704 May 2-6, 2023 ©2023 Association for Computational Linguistics Detecting Contextomized Quotes in News Headlines by Contrastive Learning Seonyeong Song1 Hyeonho Song2,3 Kunwoo Park1 Jiyoung Han2 Meeyoung Cha3,2 1Soongsil University 2KAIST 3Institute of Basic Science kunwoo.park@ssu.ac.kr,jiyoung.han@kaist.ac.kr Abstract Quotes are critical for establishing credibility in news articles. A direct quote enclosed in quotation marks has a strong visual appeal and is a sign of a reliable citation. Unfortu- nately, this journalistic practice is not strictly followed, and a quote in the headline is often “contextomized." Such a quote uses words out of context in a way that alters the speaker’s intention so that there is no semantically matching quote in the body text. We present QuoteCSE, a contrastive learning framework that represents the embedding of news quotes based on domain-driven positive and negative samples to identify such an editorial strategy. The dataset and code are available at https://github.com/ssu-humane/ contextomized-quote-contrastive. 1 Introduction A direct quotation, a verbatim replication of a speaker’s words as opposed to offering news re- porters’ own opinions, manifests news stories’ neu- trality, factuality, and objectivity (Zelizer, 1989). Quoting others also adds color to the news with authentic expressions and conveniently establishes authority based on the speakers’ reputation (The Missouri Group, 2013). Therefore, a direct quota- tion constitutes an integral element of news report- ing (Nylund, 2003). More studies have found a link between the use of direct quotations and fake news. Content anal- yses of news stories document evidence such that deceptive (versus trustworthy) news articles con- tain more direct quotations (Dalecki et al., 2009; Govaert et al., 2020). An equally problematic but less studied concern involving direct quotations is contextomy, quoting words out of context in a way that alters the speaker’s intention. A previous study argued that contextomy is a "common spin tactic" of news reporters promoting their political agenda (McGlone, 2006, p. 332). Figure 1: The central idea of QuoteCSE is based on journalism principles, where quotes from news head- lines and body text should be matched. The proposed contrastive learning framework maximizes the semantic similarity between the headline quote and the matched quote in the body text while minimizing the similarity for other unmatched quotes in the same or other articles. Some news outlets have been notorious for ed- itorializing and sensationalizing their stories with contextomized quotes in news headlines (Han and Lee, 2013). The first example in Table 1 illustrates contextomy. This example has a headline, "A gov- ernment handing out money ... ‘A debt crisis, like Greece, is on the horizon’." The quoted sentence rephrased a financial expert saying in the body text, "If we do not maintain our fiscal health, we may end up like Greece." This is far from word-for- word replication. Instead, the headline reduced the expert’s normative claim about government spend- ing and fiscal distress to a blurb that blasted the national leadership, which was on the opposite side of the political spectrum. As such, a contextomized quote …

Content truncated to 500 words