The structure of a complex data object such as a multimedia presentation is stored in a persistent representation comprising a directly executable program. The instructions of the program are created by translating recorded method calls on a function library when the object is created. The structure of the object is recreated during playback by executing the program instructions in a stack-based virtual machine in a user's computer. The persistent representation thus stores the structure as a series of operations for building the object, rather than as a static plan of the structure itself.