Discourse &Society, Ahead of Print.
Detecting internet hate speech automatically is an important but difficult task that is recognised as ethically problematic. In comparison to typical computer science approaches, the current study focuses on psychologically meaningful aspects of language, and not on terms pre-defined as hateful. Data consists of the naturally occurring discourse of a gender critical feminist group banned from the Reddit discussion platform for promoting hate based on identity; this is compared with discourse of a feminist group from Reddit, that has not been banned. Notable psychologically meaningful terms of the gender critical group include third-person plural pronouns, and metonymic acronyms that reference the gender critical outgroup, which may represent outgroup derogation, and outgroup homogeneity. It is noted that the banned forum, which is shown to be an online community, may be responding to threats to identity in recognised ways. It is concluded that a socio-cognitive discourse approach to hate speech detection may help address related ethical concerns, including potential social injustice.