Crucially, elliptical responses that require lexical identity for the slur (one replacement targeting an antecedent noun slur) yield higher prejudice likelihood ratings than those that don't (bare particle & VPE responses).

This calls for a hybrid analysis: the act of uttering a slur produces a performative effect (lost under ellipsis), but slurs also have a non-performative truth-conditional (yet not-at-issue) prejudice component (preserved under ellipsis if the slur itself is recovered). 3/3

So I ran an inference judgement task asking the participants to assess how likely it is that a given speaker is prejudiced against a (fictional) group based on how they respond to an antecedent utterance containing a (fictional) slur targeting this group.

Responses where the speaker repeats the slur yield the highest prejudice likelihood ratings & responses where the speaker indirectly challenges the antecedent use of the slur by replacing it w/a neutral counterpart yield the lowest ones. 2/3