Regulation TL;DR: Adversarial Text Summarization of Federal Register Articles

Abstract

Short on time with a reduced attention span, people disengage from reading long text with a "too long, didn't read" justification. While a useful heuristic of managing reading resources, we believe that "tl;dr" is prone to adversarial manipulation. In a seemingly noble effort to produce a bite-sized segments of information fitting social media posts, an adversity could reduce a long text to a short but polarizing summary. In this paper we demonstrate an adversarial text summarization that reduces Federal Register long texts to summaries with obvious liberal or conservative leanings. Contextualizing summaries to a political agenda is hardly new, but a barrage of polarizing "tl;dr" social media posts could derail the public debate about important public policy matters with an unprecedented lack of effort. We show and elaborate on such example "tl;dr" posts to showcase a new and relatively unexplored avenue for information operations on social media.