The present paper investigates the use of the propositional anaphor so in embedded responses in English (e.g., A: Is John coming to the party? B: I think so.). I make two claims concerning the use of so that I couch in the discourse framework of Farkas and Bruce (2009) and the framework on that-clauses and attitude verbs by Kratzer (2006) and Moulton (2009) respectively. First, I argue that the use of so is only defined if the referent of so is the top item on the Table T_c in the context c at the speech time, i.e., if the referent is still under discussion. Second, I argue that so is a predicate modifier that supplies an anaphoric argument for the predicate it modifies.