Fluent conversation is a marvel of multi-tasking within the language domain: listeners must simultaneously comprehend the speaker, predict a turn transition point, and plan a response. Experiment 1 used spontaneous conversation to investigate the apparent demands of conversation on working memory by manipulating the difficulty of a secondary task. The experiment found support for Load Theory's (e.g., Lavie et al. 2004) prediction that both conversational fluency and performance on a secondary task would decrease as working memory load increased. However, there was also some support for Pickering and Garrod's (2004, 2013) proposal that dialogue is facilitated by a collection of automatic cognitive operations when interlocutors are well-aligned (i.e., using the same words, phrases, and structures to discuss the same topics). Experiment 2 tested two claims motivated by this account: alignment is necessary for fluent turn transitions, and lexical repetition between speakers is an essential component of the alignment advantage. We found support for the former claim, but not the latter.