Language in dialogue: when confederates might be hazardous to your data

Experiments that aim to model language processing in spoken dialogue contexts often use confederates as speakers or addressees. However, the decision of whether to use a confederate, and of precisely how to deploy one, is shaped by researchers’ explicit theories and implicit assumptions about the nature of dialogue. When can a confederate fulfill the role of conversational partner without changing the nature of the dialogue itself? We survey the benefits and risks of using confederates in studies of language in dialogue contexts, identifying four concerns that appear to guide how confederates are deployed. We then discuss several studies that have addressed these concerns differently—and, in some cases, have found different results. We conclude with recommendations for how to weigh the benefits and risks of using experimental confederates in dialogue studies: Confederates are best used when an experimental hypothesis concerns responses to unusual behaviors or low-frequency linguistic forms and when the experimental task calls for the confederate partner to take the initiative as speaker. Confederates can be especially risky in the addressee role, especially if their nonverbal behavior is uncontrolled and if they know more than is warranted by the experimental task.