Processing Speaker-Specific Information in Two Stages During the Interpretation of Referential Precedents