Two experiments are reported in which people resolve references to sets of entities (e.g., lies) that have previously been introduced either explicitly into a text ("the lies") or implicitly via a cognate verb (a form of the verb "to lie"). Previous work has show that pronominal references to such entities were judged as relatively unacceptable and required longer judgement times when judgements were positive compared with cases in which the antecedent was explicit. This finding suggests that the inference from the activity of lying to a set of lies is made in the backward direction. New results show a different pattern, with no penalty in either times or acceptability judgements for the implicit case. The results are discussed in terms of hypotheses about reference processing and the notion of the centrality of an antecedent in a scenario.