This squib aims to show that the acceptability status of sluicing examples with an implicit antecedent in islands varies and discusses what is responsible for this variability. After investigating two representative structural approaches to sluicing that posit unpronounced structure in ellipsis sites, namely, Chung et al.’s Representing language: Essays in honor of Judith Aissen, 2010) LF-recovery analysis and Merchant’s PF-deletion analysis, we demonstrate that the acceptability data presented are challenging for both of them. Acceptable sluicing examples with implicit correlates in (...) islands cast doubt on Chung, Ladusaw, and McCloskey’s strict locality requirement, while unacceptable or degraded sluicing examples necessitate additional constraints for Merchant, who employs E-type anaphora as an escape hatch for island violations in sluicing. The gradient nature of the acceptability status of the examples under discussion calls for a non-structural factor that controls their acceptability. We speculate that it is discourse activation of implicit correlates that plays this crucial role. (shrink)