If we accept the metaphor (or literal fact?) that brains are computers, then it's fairly natural to see the movement of ideas between brains as just data flows in a network. To illustrate this, let's take a simple example. Suppose that as I'm talking to you over lunch in a crowded cafe, I notice that the person behind you has accidentally lit your hair on fire with their cigarette. This idea forms in my brain, and some decision-making processes kick in and determine that I should try to alert you of the situation. A message is constructed, then sent to the vocalization subsystem for transmission, resulting in me shouting "Oh my god, your hair is on fire!" The sound waves comprising this message pass through the air between us and into your ears, which do some slick calculus to weed out my voice from the surrounding din. Language processing turns the sounds into words, and thence into an idea: "Oh [your] god, [my] hair is on fire!" The salient meme, namely that "[your hair] is [on fire]," has been copied from my brain to yours in a rather impressively efficient manner.
Our blossoming understanding of the Universe has taught us that if something in nature is impressively efficient then it's almost certainly not an accident. We've learned that in such cases, we should look for reasons why a shift in the observed direction might have been favored by evolution. Certainly this is a complex inquiry, but it seems to me that developing better communication methods (e.g. language) would very likely have been memetically favored, for the following reasons:
- Any idea that can be had in one brain but not communicated to others is a lost opportunity. Mechanisms which open up the door for replication (such as the emergence of a new word to describe a previously unnamed concept or phenomenon) are likely to be successful.
- The more structured and well-defined a language is, the more precisely it can communicate ideas. A series of grunts and gestures might serve to convey "Your hair is on fire" pretty accurately, but "Joe was glad that Mary and Alex talked to Steve" would be pretty tricky to copy with high fidelity across non-language-capable brains. And fidelity is essential to on-going replication: ideas which are copied erratically are unlikely to re-copy.
- Higher communications bandwidth also favors more replication. Language increases bandwidth by allowing single utterances to stand in for increasingly complex patterns of ideas. For example, "freedom" is a very simple word but it conveys an extremely rich set of ideas whenever we use it. This efficiency means it takes less time to copy each meme, resulting in more memes being copied per given unit of time.
It's interesting to note that while language is a powerful evolved platform for memetic replication, it's not the only one. Any medium that copies patterns of thought from one brain to another is a part of the memetic environment. In particular, we should not overlook the role of art in all its forms (visual art as well as music, performance, and experience). Modern languages are indeed marvelously expressive, but the persistence of these other mediums is strong evidence that there are still many niches in which it's being memetically out-competed as the ideal communications channel. It's no coincidence that "a picture is worth a thousand words" has been such a successful meme.
Another important thing to remember is that the relationship between communications media and memetic evolution runs both ways. I've been discussing how memetics would favor increasingly expressive languages, but I believe we should also expect the reverse effect: if we can develop communications platforms/models with higher bandwidth and higher fidelity, it should increase the rate of memetic evolution of the ideas within those platforms. This feedback is part of what makes the evolutionary algorithm so stunning: the better it gets, the better it gets at getting better.