The degree to which speakers model their listeners is debatable. This study aims to contribute to the quickly accumulating body of literature assessing speakers' egocentricity and/or collaboration in utterance formation by manipulating the speakers' perspectives in relation to their potential models of their listeners. In 5 experiments, after speakers' beliefs about task difficulty are ascertained, they are systematically manipulated, as are their beliefs about their partners. Based on these beliefs, speakers address a particular cultural audience (in either a monologue or dialogue) designed to help their intended audiences select ambiguous figures from a set. These texts are then given to the speakers' intended or unintended audiences to determine the descriptive efficacy and design specificity of the monologues and dialogues. In some situations, the manipulation of speakers' perspectives should not affect their descriptions if they fully consider their listeners' perspectives. In contrast, if they do not consider their listeners' perspectives at all, the manipulation of beliefs about their partners should not show a significant difference.