Though most documented sign languages make use of space to denote relationships between predicate arguments, studies of emerging sign languages suggest that spatial reference does not emerge fully-formed but takes time to develop. We present an artificial sign language learning experiment that expands the cultural evolutionary framework to investigate complex linguistic constructions. Our results demonstrate the gradual emergence of consistent devices to distinguish between sentence arguments, some of which rely on iconic spatial contrasts. These findings mirror data from emerging sign languages and point to the cultural mechanisms that facilitate the evolution of complex linguistic structures.