This paper presents a community of communicating embodied agents which learn an adjacency-based grammar from user interactions. The agents act as intelligent fridge magnets, each printing a word on their respective displays. The user places agents next to other agents on the fridge, removing and replacing them if the word they display is ungrammatical given the current context, thereby indicating grammatical acceptability. We present these agents both as a test bed for exploring research into embodied communicating agents and as a means of investigating how users respond to expressive devices like fridge poetry magnets which learn from user interaction. Categories and Subject Descriptors H.5.2 [Information Interfaces and Presentation]: User Interfaces – natural language, input devices and strategies General Terms Algorithms, Design, Experimentation, Languages. Keywords Embodied communicating agents, interaction-based grammarlearning, expressive user-interfaces, intelligent agents, u...