The paper describes a novel, declarative language that enables modelling ubiquitous, contextual and interactive augmented reality environments. The language, called CARL – Contextual Augmented Reality Language, is highly componentised with regards to both the structure of AR scenes as well as the presented AR content. This enables dynamic composition of CARL presentations based on various data sources and depending on the context. CARL separates specification of three categories of entities constituting an AR environment – trackable markers, content objects and interfaces, which makes the language more flexible and particularly well suited to building collective awareness systems based on ubiquitous AR-based information visualisation.