While not immediately having anything at all to do with linguistics, this awesome video demonstrating spontaneous synchronization nonetheless demonstrates a critical idea behind complex emergent phenomenon in linguistics.
The video shows five metronomes lined up on a thin plank, pulsing asynchronously. The plank is then lifted onto two soda cans and gradually the five metronomes begin to sync up and pulse in time with each other, taking just under one minute.
The question is, would they have synced up had they not been lifted onto the cans? I believe the answer is no, they would not have. The energy being transferred between the metronomes was probably too dissipated, too weakly distributed when sitting on the table. When placed atop the cans, the energy was focused in some particularly salient way as to facilitate synchronization (I'm no physicist, this is just my naïve hunch).
I believe a parallel can be drawn with language evolution (and really, language learning in general). There is the notion in language learning theories of matching up internal hypotheses about grammatical structure with evidence from a community of speakers. Eventually, all speakers of "the same language" must form some sort of agreement or synchronization in order to communicate. But that agreement needs proper focal points to be salient. It is not the case that all language patterns get passed along. Some die off. The language patterns that succeed are the ones that have the right focal points to get distributed in the optimal way across a community of speakers. The mechanism of distribution is inter-speaker agreement. This agreement is implicit and emergent. In other words, a language could be defined as the synchronized patterns that speakers have settled on.
But why do speakers agree to adopt pattern A but not pattern B? This is a not entirely well understood, but I think it is clear that some patterns succeed because that were distributed in the right way. good old lucky accident.
I'm thinking of the sort of work that Partha Niyogi has done in has done in his book The Computational Nature of Language Learning and Evolution.
So, what are the soda cans of language evolution?