Much of the second language acquisition (SLA) scholarship suggests that conversational skills are best acquired through communication in the target language. Although in recent decades communicative approaches to language teaching have seen widespread adoption in the classroom, it remains exceedingly difficult to assign conversational homework with the tools currently available. This reality has created a gap between the way in which foreign language courses are often implemented and the manner in which the SLA theory community might recommend. It is our belief that automatic speech recognition technology in general and spoken dialogue systems in particular have the potential to bridge this gap. In this paper, we lay out some principles behind dialogue system design in this context, and introduce a prototype language learning dialogue system in Mandarin Chinese.