In this study, I examine the monologues given by instructors in step aerobics classes, focusing on the way language both arises out of ongoing action and is constitutive of it. Following Levinson (1992), I show how the structure of the activity constrains the interpretations that are made of the utterances that arise throughout the workout. Aerobics participants need specific pragmatic knowledge, a key part of which is the ability to detect and interpret 5 distinct functions of utterance, defined according to their timing and placement in the hierarchical structure of tasks that the class is performing. I demonstrate that it is beat placement, rather than grammatical form or sequential organization, that is the most important cue for this interpretative task. Having presented the utterance functions and the cues to their interpretation in detail, this article goes on to outline how participants achieve the correct assignment of pronoun reference and ellipsis in the instructor's monologue. This is explained by means of an approach to discourse modeling first suggested by Grosz and Sidner (1986), showing how instructors set up ephemeral actions as complex conceptually salient discourse entities, making them accessible for subsequent reference.