🚀 go-pugleaf

In article <memo.847737@cix.compulink.co.uk> petex@cix.compulink.co.uk (Peter Christian) writes:

First off;

I don't disagree, in principle, with much that is posted, including
this one. Any kind of a pursuit that results in new knowledge is
a rational. No one knows when it will be put to use. ONe day
someone will find an application for it. This is especially
true of mathematics.

Second; probabilistic and deterministic methods are not
mutually exclusive. Much of the 'scientific laws' have
an implicit or explicit basis in probability theory.

For example, take a simple linear, ordinary, deterministic,
one dimensional and first degree differential equation with constant
coefficients. You can't really find a simpler DE. (actually a
simpler version is with s(t) = 0).

	dY/dt + aY = s(t)


THis results in the famous exponential solution. Simply by
adding another term to the right hand side say w(t) which
is a noise term (i.e. randomness), we can now describe an
essentially deterministic system subjected to random variations.

We can describe the solution(s) in terms of this idea. The
equation describes what would happen to an infinite number of
such systems when subjected to the random innovations, chance
variations etc. The average solution would represent the path
that we would expect most of them to take. Each specific sample
or manifestation could correspond to an individual case which
would vary from another individual development.

> I agree that the search for universal principles is a rational
> pursuit in historical linguistics. The first problem is that there
> are bound to be different principles for different levels of
> structure and that these will interact in complex ways which will
> depend to a great extent on the actual details of the linguistic
> system.

yes. They will interact in complex ways. We don't have to know
the exact nature of the interaction. Probabilistic or statistical
statements reflect our ignorance about processes. These measures
are still better than knowing nothing but of less value if we
have more intimate knowledge of the phenomenon.


> The second problem is that in spite of the impression given by almost
> every book on historical linguistics, no historical change is fully
> explainable (though it will still be describable) without reference
> to the social environment of the change, which of course we have very
> little knowledge of for earlier periods. In fact the work of Labov
> and his followers seems to suggest that we haven't got good enough
> evidence for *any* period where we can't accurately correlate
> linguistic forms with the social categorization of the speaker on a
> statistical basis. 

It's only a matter of degree. Knowing a little bit more is better
than knowing less. There is no need to expect quantum leaps in
knowledge constantly. They occur very infrequently. Knowledge
usually increases in much smaller bits.


So there aren't really two distinct types of
> change (though I admit, it is pragmatically a useful distinction), or
> rather we can never be sure that `environmental' factors did not play
> a part in some historical change. 

Again it's not important. One can divide into two then into two
to get four instead of dividing into four immediately.


I remember collapsing with laughter
> when I read the suggestion that the Second Sound Shift was caused by a
> powerful leader with a speech impediment whose manner of speech was
> imitated by his followers. It's ridiculous because it's pure
> speculation (certainly not because the speech forms of the powerful
> are irrelevant to linguistic change), but just supposing some such
> `social' factor were discovered then all the explanations based on
> `tensions within the system' would be seen to be seriously
> incomplete. 

I've heard a similar story purportedly about why Spanish is
different than South American versions of it. But it doesn't
make any difference why it changed. Mathematical descriptions
don't have to give reasons as to 'why` only have to `adequately
model` the 'how'. And trying to model things that change accross
time immediately brings to mind, obviously the use of a `time`
parameter.



And since by definition we will never have adequate
> sociological evidence for such periods (excluding time machines, of
> course), self sufficient internal principles must be seen as the best
> of a bad job. Which doesn't mean we shouldn't make them as explicit
> and consistent as possible or that we shouldn't have sound criteria
> for distinguishing better and worse accounts, we just shouldn't have
> too much faith in the real historical accuracy of such explanantions.

it doesn't matter. Economic statistics are collected about quarterly
(in some cases monthly). By passing curves thru these points
one can estimate what happened in the middle. Even if the curve
misses them sometimes, it still doesn't matter if we're trying
to find relationships that occur over years, or decades or
centuries.


> Any universal principles derived from such explanations are probably
> best treated as guidelines for what is likely rather than as
> scientific laws, because of this absence of crucial information.

Yes. If the model is somehow 'adequate', you can fiddle with the
model to see what it produces for the unknown periods. But what
really happens is that the model becomes 'reality'--for a while
anyway.  Then someone discovers a new phenomenon --say a new cache
of written materials--if the model cannot adequately account for it,
then it's parameters are twiddled, tweaked, fiddled with. If the
model still cannot account for it, someone produces a new model
which accounts for the old model as a special case and also
accounts for the new findings. And so on, it continues.



 For
> example, a very reasonable `universal principle' would be that complex
> consonant clusters tend to be simplified, but that wouldn't cover the
> explanation of why it sometimes applies and sometimes doesn't, i.e.
> why consonant clusters are not constantly being simplified in every
> language that has them. 

Well, maybe a new Verner or Grimm will come along :-).


What we are left with is a principle in the
> sense of a statistically frequent and plausible type of change
> (clearly better than `consonant clusters tend to become nasalized
> diphthongs', for example) which can be appealed to as `explanations'
> for change, (though they might be regarded as just high level
> classifications of descriptions). This is in fact what historical
> linguists do, but it inevitably falls short of what I take to be the
> implication of `universal principle' in the original posting. 


Dividing up into universal(global) and local is a simple start.
It's binary 0,1. We can easily divide up the range [0,1] so that
we can account for other cases. In your case (consonant clusters)
we might assign a number somewhere between 0.4-0.8,if 1 is
univeral and 0 is strictly local. Again, simplifications have
their use. It makes possible a start in the right direction. If
we intend to solve the most complex phenomena with the simplest
tools, we can only make simple models to start with. This is
historically what happens to almost every science. There's a
saying; "If the only tool you have is a hammer, everything looks
like a nail." There's no other way to do it. More complex
methods come in after some progress has been achieved with simpler
ones.  For example, there seems to be mistaken belief that
Chomsky's representional schemes have no use other than drawing
arrows between symbols. This is clearly not the case since, there
is now a classification of formal languages based on his ideas.
The classification scheme obviously goes in the direction of
which types of languages are more powerful than others. There's
a metric already. More work needs to be done, obviously.


The best
> we can do in my view is have criteria for acceptability of
> explanations, but not criteria for when or why a certain `universal
> principle' operates in a particular language, i.e. not a true
> `universal principle' as I understand the term.

Is this for the present state of the art or forever :-)..

regards
-- 

	mark	

   hubey@amiga.montclair.edu         hubey@apollo.montclair.edu
   hubey@pilot.njin.net               ...!rutgers!pilot!hubey