A general consideration of language complexity, pt 2

I wrote this post a couple of years ago. I figure I should complete the trains of thought that started in it.

I listed a bunch of 'models' for what grammar is – granted, in rather exaggerated forms – and presented some of their drawbacks. Finally, I hinted at my own favourite model. So, let's look a bit closer at this particular model.

A model of grammar needs to account for the following facts about language:

• grammar changes over time
• grammar is learned by new speakers (kids, immigrants)
• every speaker has grammar
• even a person who does not know the formal definitions of grammatical terms (and who may not be able to consciously identify something as an object or an indirect object) may very well have a great mastery of the grammar of his or her language.
• grammar needs no central authority to exist, nor any platonic ideal form.

Now, this can be accounted for, in rought terms, by the following:

• a new speaker learns by observing the language and creating hypotheses about how the language works. The speaker uses these hypotheses to parse and create utterances. Hypotheses are sometimes updated, but depending on the individual and on what evidence he is exposed to, some mistaken hypotheses can persist throughout life. Oftentimes, one hypothesis is not enough - there may be several different hypotheses about the same thing (for one example, consider a word that means different things in different registers or regiolects). The hypotheses are really 'patterns', and we could basically state a hypothesis as 'this pattern is used when these particular preconditions hold'
• the speaker's grammar is the set of hypotheses that are in active use in his or her speech and parsing
• the grammar of the language is some kind of 'weighted average' of the grammars of the speakers - speakers who speak much might be more relevant than speakers who speak little. A person who's taken a vow of silence is a member of the speech community, but will have little effect on the grammar. Calculating an actual coefficient for each speaker is impossible – but even worse, the notion of an 'average' of a set of similar rules is also somewhat ill-defined. 

We observe certain things about this:

• human brains are fairly comparable. Certainly there are geniuses and idiots, and certainly any population will have a variation between those extremes. However, human linguistic skill is remarkably similar – we don't find populations that simply cannot learn language.

Since most of language learning is not "directed" by a teacher, we can assume that any two brains in any two situations will find similar amounts of patterns – not the same amount, but comparable amounts, at least as far as order of magnitude goes. Of course, there will be individual variation – but as we look at an increasingly large population, and discard such patterns that only few speakers have, we will probably get even closer to the same average amount of grammar for pretty much all speaker communities of sufficient size.