How do people generate and decide between the wide array of potential goals available to them at any given moment? We study this question in Minecraft, a game environment that is both open-ended enough to support a diverse array of goals and structured enough to facilitate quantitative evaluation of different goal features that may impact how people respond to different goals. Specifically, we explore the role of goal familiarity, concreteness, and complexity, which we operationalize using both linguistic analyses and by converting human-generated goals into a programmatic domain-specific language. Our results highlight the unique ways in which game environments like Minecraft can facilitate research into how humans engage in open-ended and creative behaviors.