Facts, Fabrication, & False Information: linguistic analysis of fake news

2016年，美国爱达荷州双子瀑布镇因涉及四名7至14岁男孩和一名5岁女孩的性行为不端指控而引起全国关注。这一事件——后来被命名为“芬布鲁克案”——迅速演变成充斥着错误信息的媒体闹剧。以传播阴谋论信仰和宣传极端保守议程而闻名的小众媒体开始声称，这四个男孩是叙利亚难民，他们在枪口下轮奸了五岁的女孩，这与最初的故事大相径庭（Bell, 2017）。从那以后，福恩布鲁克案被视为虚假信息传播速度之快的一个标志性例子。我提出的博士研究将研究语言是如何被用来传播错误和虚假信息的。为了正确地解决这个问题，我将使用语料库分析和评价分析相结合的方法。语料库分析提供了一种定量的方法来研究数据，而评价是一种定性的方法，检查跨文本的人际立场。将语料库分析和评价分析结合起来的好处是，我能够分析各种语域和体裁的文本（Martin & White, 2005）。语言学研究人员已经证明，评估与语料库协调程序（如UAM CorpusTool） （O'Donnell, 2016）相结合，是衡量评估意义和变量的统计显著性的有效方法组合（Gales, 2011; Hurt, 2019; Biber, 2014）。评估分析也被用来成功地研究跨文化交流的差异如何影响法律结果（Martin and Zappavigna, 2016）。同样，评价也可以用来研究作者在撰写芬布鲁克案例时的认知立场（承诺标记）和情感立场（情感标记）。为了我的项目的目的，我将使用评估来探索作者如何使用立场标记来传播虚假信息，而不是可信的事实信息。通过这种方式，我提出的研究不仅会为现有（和相关）问题提供新的视角，而且会使用方法上合理的技术来做到这一点。目前，虚假信息的传播是通过使用自动的算法事实核查来打击的；机器被用来识别（可能的）虚假信息。一旦确定，人们就会评估标记的实例。这是由于一个明显的计算限制：计算机还不能解释自然语言的语义和语用方面（Liddy, 2001； Akbik等人，2018）。通过研究语言是如何被用来传播虚假信息的，我将加强现有的关于错误信息和欺骗标记的语言信号的文献（Jiang & Wilson, 2018）。此外，我的研究结果可以为自然语言处理（NLP）检测做出贡献，并为整个社会识别虚假信息提供可靠的数据。

In 2016, the town of Twin Falls, Idaho, USA gained national attention due to allegations of sexual misconduct involving four boys ages seven to fourteen and a five year old girl. The incident -later named the Fawnbrook Case- quickly devolved into a media circus rife with misinformation. Niche media outlets known for spreading conspiratorial beliefs and promoting ultra-conservative agendas began claiming that the four boys in question were Syrian refugees who had gang raped the five year old girl at gunpoint, a polarized departure from the original story (Bell, 2017). Since then, the Fawnbrook Case has been viewed as a hallmark example of how quickly false information can spread.My proposed doctoral research will examine how language is used to promulgate mis- and disinformation. To properly address this topic, I will use a combination of corpus analysis and Appraisal analysis. Corpus analysis offers a quantitative approach to study data while Appraisal is a qualitative approach that examines interpersonal stance across texts. The benefit of combining both corpus analysis and Appraisal analysis is that I am able to analyze texts from a variety ofregisters and genres (Martin & White, 2005). Linguistic researchers have demonstrated that Appraisal -in combination with corpus concordance programs such as the UAM CorpusTool (O'Donnell, 2016)- is a useful combination of methodologies to measure statistical significance of Appraisal meanings and variables (Gales, 2011; Hurt, 2019; Biber, 2014). Appraisal analysis has also been used to successfully study how differences in cross-cultural communication have influenced legal outcomes (Martin and Zappavigna, 2016). Similarly, Appraisal could have been used to study both epistemic stance (markers of commitment) and affective stance (markers of emotion) of authors writing about the Fawnbrook Case. For the purposes of my project, I will use Appraisal to explore how authors use stance markers to circulate false information compared to credible, factual information. In this way, my proposed research will not only offer a new perspective on an existing (and relevant) issue, but use methodologically sound techniques to do so. At present, the spread of false information is combated through the use of automated, algorithmic fact checkers; machines are used to identify instances of (possible) false information. Once identified, humans evaluate the flagged instance. This is due to a glaring computational limitation: computers cannot yet interpret semantic and pragmatic aspects of natural language (Liddy, 2001; Akbik et al., 2018). By studying how language is used to circulate false information, I will strengthen existing literature on the linguistic signaling of misinformation and deception markers (Jiang & Wilson, 2018). Moreover, the results of my research can contribute to Natural Language Processing (NLP) detection and provide sound data for society at large to identify false information.