This study explores the way discourse structure information is used during encoding of linguistic representations, using the distinction between main and subordinate information as a case study. We use the two contrasting constructions: (a) “The singers\textsubscript{MAIN} who admired the violinists\textsubscript{MAIN} invited their mentors to the party”; (b) “The singers\textsubscript{MAIN}, who admired the violinists\textsubscript{SUBORDINATE}, invited their mentors to the party.” While both contain discourse-main information, (b) includes discourse-subordinate information in the clause ``who admired \textit{the violinists}.” Importantly, \textit{the singers} and \textit{the violinists} are both plausible antecedents for \textit{their}, but the overlap in discourse-information between the two NPs differs: (a) overlap ({MAIN, MAIN}); (b) no overlap ({MAIN, SUBORDINATE}). Through two web-based eye-tracking experiments using a visual world paradigm, we find that this overlap leads to competition between the two NPs, evidenced by eye-gaze differences, (a) < (b). We also find that this effect manifests early, even before retrieval, i.e., before pronoun resolution.