Language is foremost an instrument for social interaction. In social interaction, interaction partners have been shown to represent not only their own action but also the action of their partner. Such partner co-representation is typically demonstrated by interference between own and partner action. Yet a partner’s action may also facilitate own action. To investigate this we measured the speed with which speakers can classify objects presented on pictures as “natural” or “manufactured”. In single-subject settings, response times have been shown to decrease with each picture classified within a given semantic category. Here we investigate such cumulative facilitation in a joint task setting in which two task partners take turns classifying pictures. We expect a decrease in response times with each additional member of a given semantic category (cumulative semantic facilitation) and a stronger decrease for semantic categories that are classified jointly by participants and task-partners compared to categories classified solely by participants (partner-elicited semantic facilitation). This would demonstrate that a partner’s actions are represented and in turn can facilitate own action. The shared task was implemented in an online setting in which participants were led to believe they were collaborating with a task partner. To emphasize the joint nature of the task, it was framed as cooperative by informing participants that their own and their partner's performance would be pooled and compared to other participant pairs. Our results will be discussed with regard to the conditions under which partner co-representations occur and with regard to the nature of these representations.