Text Analysis in Adversarial Settings: Does Deception Leave a Stylistic\n Trace?

Abstract

Textual deception constitutes a major problem for online security. Many\nstudies have argued that deceptiveness leaves traces in writing style, which\ncould be detected using text classification techniques. By conducting an\nextensive literature review of existing empirical work, we demonstrate that\nwhile certain linguistic features have been indicative of deception in certain\ncorpora, they fail to generalize across divergent semantic domains. We suggest\nthat deceptiveness as such leaves no content-invariant stylistic trace, and\ntextual similarity measures provide superior means of classifying texts as\npotentially deceptive. Additionally, we discuss forms of deception beyond\nsemantic content, focusing on hiding author identity by writing style\nobfuscation. Surveying the literature on both author identification and\nobfuscation techniques, we conclude that current style transformation methods\nfail to achieve reliable obfuscation while simultaneously ensuring semantic\nfaithfulness to the original text. We propose that future work in style\ntransformation should pay particular attention to disallowing semantically\ndrastic changes.\n