Accent is an important design cue within human-conversational AI dialogue, especially with speech only conversational agents, where it impacts inferences of conversational AI’s capability as a dialogue partner (termed partner models), along with levels of audience design and adaptation in human-AI language production. Current voice design leads to significant overestimation of the capabilities of such systems, leading to poor partner modelling of the actual capabilities of the system affecting interaction. This PhD project aims to investigate 1) how the use of foreign accented speech by conversational AI agents can be used as a way to influence user partner models; 2) how this impacts informativity in language production during human-conversational AI dialogue and 3) how these effects are influenced by the group structure (e.g., dialogue vs multiparty) and collaborative task.