Remove or keep up?

A bottom-up assessment of online hate speech

Abstract

Multi-component methods of analyzing hate speech seem to open new horizons in training human annotators and algorithms to facilitate online hate detection and moderation. One of the limitations of these endeavors lies on the fact that the components proposed are either treated as if they were of equal significance or hierarchized within scales that are based on scholars’ understandings and assessments – informed and relevant as these might be – with no feedback from common laypeople. They constitute “top-down” understandings of the performative power of hate speech. The present contribution seeks to address this gap by analyzing how ordinary people, especially young people in our case, assess different forms of hate speech. Our aim is to identify what they estimate to be acceptable or tolerable, and the actions they would choose to undertake, if they were to monitor online posts (i. e., take them down or keep them up). We consider hate speech as an “ordinary concept” reflecting “ordinary people’s norms”. The research reveals an overall tendency to content removal, especially in the case of calls to lethal aggressions and swearwords which seem to operate as offense intensifiers. Incitements to non-lethal actions and emotional expressions may be tolerated or even accepted by some, being perceived as non-hateful”. A distinction between hate action, hate speech, and hate expression revealed to be significant. At the same time, punctuation seems to have no significant effect on the overall judgments. Finally, combinations of hate features increase the perceived hatefulness and, even more substantially, the willingness to take down the content in question.