Print

Print


> [Daniel:] What is chance frequency? Do we just use 50%? That would mean something like, things which happen around half the time require no explanation; things which happen overwhelmingly more or overwhelmingly less than half the time need one. Yet if about half the people I meet are men, about half women, that certainly seems to call for a biological explanation (fortunately, we have one). 

Well, walking in The Scottish Borders and meeting 50 % English and 50 % Scottish people certainly doesn't call for a biological explanation ;)

If we just temporarily accept my theory that a natural grammar is based on logical necessity, and we want to model a prototypical language, the dominant word order provides a problem because the three variables can be organised in six different ways:

SOV, SVO, OSV, OVS, VSO, VOS

By the way this is an example of what I call logical necessity: there is no other possible series for these three variables. 

Anyway, as there are six combinations, the chance frequency of each is 1/6 or ca. 16.7 %. This means we can now make the prediction that each of the types has this percentage among the languages of the world. 

When we then go out and do our field research, if it will turn out that all languages actually have the same dominant word order, e.g. SVO, there's nothing to suggest that this is due to biological factors. I would rather make the assumption that cultural contacts would account for the bias: as people commerce and otherwise interact over language borders, they adopt similar thinking. Or historical: all languages could be based on one SVO protolanguage, and no other types were ever needed.

Well, we already know this is not the case. All languages don't have the same word order; yet the distribution is not the chance frequency.

The search for cultural explanations hasn't yielded many results - societal variables such as marriage type, hierarchies, heritage system etc. do not explain the word order distribution, and geography gives an insufficient explanation.

There's the recent finding that biological factors related to humidity may contribute to the prevalence of tonality, but this is not what we're looking for as we're talking grammar, syntax and typology, not phonology.

So we move on to evolutionary explanations. Chomsky has been compared to Darwin but there's an important difference between the two. While Darwin did extensive research to demonstrate how his theory of natural selection actually works, Chomsky has nothing of the kind. His UG is more of an idea than a fully fledged theory. Anyone can say that the distribution depends on "some neurons, some gene, something in the brain", but this actually demonstrates nothing. 

What's worse, trying to explain the scarcity of OSV with hard-wiring actually suggests that the speakers of e.g. Warao are not people, or disabled at best, being devoid of the "necessary" wiring.

So that's it for nature and nurture. After hundreds of years of research, there's little evidence to support any of them.

> Someone whose idea of a "random" language - that is, their language for describing languages - matches well with observed languages, will decide that the drifting changeability of culture is all the explanation needed. 

Not really, because there's a third possibility, and the language in question doesn't have to be random at all. 

We don't think that physical and mathematical constants are essentially based on the human nature, but rather something external that we're trying to control. Physics and mathematics are tools rather than the hand or the mind itself that uses the tool, and this is also true about language. 

If I'm mistaken here, you'll have to explain the link between grammars and mathematics in a way that allows just one to be based on logical necessity. If you succeed, it will give rise to the notion that methods incorporating the two, such as predicate logic, are not completely logical.
 
> So, I've inadvertently described Bayesianism (well, a Bayesian would go on to describe how to update the description language when it doesn't match reality)

Absolutely, thanks for the term! You can make a Bayesian statistical model with the mere S, V and O, but these are actually not sufficient for a fully expressive language that describes real languages; it's a logical necessity that you'll have to add constituents such as conjunctions and adverbials to make them work.

You'd also have to keep developing the model with simultaneous research on optimality theory as well as statistical research on language databases, and use Bayesian methods. So there will be enough work for many PhD's to come. The good news is that - as I showed in my previous post - we get positive outline results very quickly, and going further will deepen our understanding.

> but my point was just that focusing on those specific questions is what makes it look like more explanation is needed. And keep in mind, those features were specifically designed as things no human language does. We can't expect "chance frequency" when the terms we're working with were specifically selected for deviating from it. Anything can be proven that way! 

I totally agree and I think the original argument was shown to be deficient by contributors on this thread. At any rate the main focus has to be on key features of language rather than assumed anomalies or statistical outliers.