The only model of ‘evolution’ that can throw some light on the evolution of language is that of WHEE: history and evolution.com, and one aspect of this is to consider a form of evolution that is a superset of genetic evolution, along with being ‘directional’ (i.e. teleological).
This model has a limited range of data, but the data it does have is remarkable, and it does have some tantalizing glimpses of the ‘macro aspect’ on the issues of language, and the other aspects of the human breakthrough. The problem is larger than that of language and includes ‘mind’, ‘consciousness’, ethical reasoning, art and creativity, and finally the clunker par excellence, ‘soul’, a category we can’t even analyze properly and whose reality, or existence is subject to metaphysical limits.
The value of the model in WHEE is that is can show plausible examples of complex cultural complexes ‘evolving’ in a pattern of what appears to be induction, but with an interactive aspect of man cooperating with ‘evolution’.
The suspected model of this is a simplified version of Kantian (or Schopenhauerian) ‘transcendental idealism, but with no explicit use of its full range of properties, e.g. the construction of space/time. This involves a distinction, crude but roughly supported by the original Kant, of the phenomenal and the noumenal and this can explain the data we see and also how a teleological system might work: a series of finite transitions in the phenomenal realm matches a teleological system in the noumenal.
If we look at the examples of the macro effect in world history we detect to our surprise the correlated temporal match (short of claims of causation) of complex innovation with the relevant finite transitions. Some of the most dramatic examples are complex literatures which spring into existence are remorphings of previous such entities. A look at the Greek Archaic, for example, shows how this form of evolution generates (or generates the prototype that men create) the sudden explosion of linguistic and literary achievements of that transitional ‘Axial Age’ period.
We have thus a comprehensive model of how human language might have evolved: a finite transition sequence of the type seen (a sequence of ten thousand years in this example case, the original would have been much longer) in a system of shaping transformations returning on its population or subsets in (historically) frequency burst of several millennia. A system this complex and intelligent could obviously direct genetic factors, but history doesn’t give us (so far) the genetic aspect. This hierarchy of levels is confusing for genetic evolutionists but in principle there is no problem once the crutch of genetic fundamentalism is overcome. We cannot say how the deeper form factor operates at a level more abstract than the genetic but we have to suspect it is there because it obviously there in the data given in WHEE. It is not in principle any more complex that high and low level programming languages and the relationship between them, e.g. a high level software and its machine code language, etc… That example doesn’t directly answer to our problem, but the analog is obvious.
What is surprising is the super-complexity of the higher form factor ‘software’ and the way it can operate on high level systems, among them language.
This is not a direct answer, but it can show us how in principle an evolutionary system for language can arise.