Just taking some notes on a paper from last year by Helga Andresen, on the ways that role playing by preschool-age children can illuminate language and metacommunication development. I recognized a lot of my own children in the descriptions and examples.
Bateson (1955) postulates the existence of metalinguistic rules which determine how linguistic signs are related to non-linguistic entities like objects, persons, actions and places. These rules themselves, of course, cannot be linguistic in nature.
Bruner  analyzed the development of mother-child interaction within the formats over many months and showed that the child constructs knowledge and anticipation of the interaction sequences and thus successively internalizes the structure of the format. This happens during the second half of the first year. First, the child adjusts his own vocalizations, miming and gestures to this structure until, more and more, he takes over the active part of the ongoing communication. Ritualization and repetition of the interaction make it possible for the child to recognize its structure. Above all, it is the close and fixed relations between verbal utterances and the non-verbal context that give children the chance to realize that the vocal activity of the mother refers to something beyond it and to realize the meanings of the utterances. Therefore the formats may be taken as an instantiation of those metalinguistic, non-verbal rules postulated by Bateson.
These considerations explain why early language use must be sympraxic; otherwise, children would have no chance to grasp the symbolic function of language (Andresen 2005:394).
Andresen has some fascinating examples where she documents interactions between children during roleplay, showing how much of the communication occurs within the roles and how much is metacommunication about the nature of the roleplay (a surprisingly large proportion).
The finding that older children produce less explicit metacommunication is of special interest. At first glance it may be surprising because from a scientific view it would be suggestive to propose that explicit metacommunication demands complex cognitive and communicative abilities which can be developed only on the basis of complex communicative skills which are beyond the scope of 4-year-old children. But metacommunication does not vanish out of play when children grow older; on the contrary, the pretend play of older children is much more complex than in the earlier years and contains a lot of transformations. Qualitative analyses of the older children's play in the Flensburg corpus show that they produce more implicit metacommunication than the younger ones. So, during the preschool period metacommunication changes from explicit to implicit performance (Andresen 2005:401-402).
A passage follows that discusses the complexity of carrying off implicit metacommunication -- which we may take as deliberately structuring communication in a form that itself more or less unambiguously conveys its context. In other words, implicit metacommunciation takes advantage of certain redundancies available in communication -- such as special same-meaning grammatical structures, gestures, tones, etc. -- to context-mark the communication.
Andresen then connects the progression from explicit to implicit in terms of Vygotsky's model of cognitive development:
According to [Vygotsky], internal mental processes arise out of external, interactive and communicative processes in earlier stages. He formulated this phenomenon as the transition from interpsychic to intrapsychic processes and functions during development (Andresen 2005:402).
And the subsequent section considers the development of egocentric speech along a similar timeline. Egocentric speech is inward-directed and regulatory in nature (with respect to actions); Andresen suggests a similar regulatory role for metacommunication. One might mention that egocentric speech has its own metacommunicative elements -- it being hard to mistake someone talking to herself for someone deliberately trying to communicate to others. In any event, this provides an opportunity to argue against simple word-object associations and in favor of the idea that roleplay indicates the ability to create linguistic (i.e., not here-and-now present) objects:
But an analysis of children's role plays clearly shows that already 4- year-olds are indeed able to create objects and meanings by linguistic means: for example, Aunt Maria, in the play of Hilde and Ingrid, who comes into existence through Hilde's utterance on the metacommunicative level and whose existence afterwards can be presupposed within the play. If the children could not create new meanings and communicate them to each other, role play could not take place at all (Andresen 2005:404).
There is a lot of detail in this article, related to the interchange of parent and child joint attention, the development of metacommunication skills in parent-child interactions, and the emergence of role playing as a way for children to take on adult roles they are normally not permitted. It relates well to Tomasello's work and Gregory Bateson's as well as Vygotsky, who is the model for much of the theory.
Andresen H. 2005. Role play and language development in the preschool years. Culture and Psychology 11:387-414. DOI link