Rate of Change: A Closer Look

[X99Lynx at aol.com]

I wrote:
<< So pick a *set* of changes that has occurred in at least five, preferably
ten languages, and the time it took for each of them to occur historically.

In a message dated 6/30/2001 2:54:10 AM, alderson+mail at panix.com writes:
<<Who gets to decide how long the changes took?  Or are you restricting the
sample only to historically attested changes?>>

Oh, yes.  It would have to be that way.  These have to be observable changes
in observable periods of time.  That makes the data "universal" and
reproducible - anyone can measure it.  Of course, you have some statistical
leeway in setting a range for the period the change took.   E.g.,  this set
of changes took 200 years, plus or minus 100 years, and this simply puts a
similar averaged margin of error on your results.  But yes the changes and
the average time they took would have to be measurable by disinterested
observers - just like in calculating half-lives, for example.

You also probably should test a few different sets, just to be sure you had
an adequate control - to show that this one particular set is not an anomaly.

<<In that case, are you prepared to accept the results if they contradict
your view of the prehistoric spread of the Indo-European languages?  Or will
you aver that unattested (that is to say, unwitnessed) changes must have
happened at a different rate? >>

Let me put it this way, this kind of statistical evidence would be very
powerful and would create a prima facie presumption that this statistically
derived "rate" is universal.  The more languages it shows in the more
"certain" it becomes in terms of probability theory.  It would be bad science
indeed to argue against that.

As far as discontinuity behind attested and unattested rates, that cannot be
assumed - as Prof Trask stated - without a clear demonstration that an
outside variable creates a statistical difference between "attested" and
"unattested" languages - e.g., absent something like a distinct iridium
layer, a Yucatan crater and dinosaur extinction at precisely 64 million years
BP.  Obviously, the strong assumption would have to be that this is a
universal (average) rate of change in both attested and unattested languages.
 And because of my training and experience, I'd have every reason to agree.

I should say that I'm not vested in any way in a particular "rate of change."
 I don't have family in Anatolia ready to open up an IE homeland theme park.
I am vested however in the use of proper scientific methodology.  However, on
this list you have some strong personalities with some very strong opinions.

So I find myself taking firm positions about Anatolian origins not because
I'm particularly sure of that idea, but because "strength of conviction"
counts for something in these discussions.  When you are in such a minority,
you simply have to take a devil's advocate position sometimes just to force a
balanced consideration of the evidence on the other side.

On the other hand, the kind of statistical analysis described above is pretty
independent of opinion - my opinion or anyone else's. It speaks for itself
and should go a good way towards settling the matter for everyone.  Or
practically everyone.

<<Suppose that no specific feature ever shows a correlatable average rate of
change?  Suppose that it is instead entire systems that change at a
comparable rate, with no particular change required?  What then?>>

Then the question becomes whether the difference between entire (attested)
systems is measurable.  This would seem to be analogous to the analysis of
differences in complex structural "states."  Again you find some set of
variables - perhaps a much larger set - that predictably matches the
difference you see in the entire system.  If that set (e.g., as an
off-the-cuff example: the twenty most common sounds used in a language) can
predictably identify items from the correct language from a random sample of
items from different languages (let's say 75% of the time) then you have a
pretty good indicator for the entire language system.

Once you have demonstrably shown that kind of correlation, it becomes a good
hypothetical tape measure for any attested language (unless the correlation
can be shown not to apply in other languages.)  More importantly, using that
"representative" set, one should be able to calculate a pretty precise degree
of (attested) change between the same language at two different points in
time.

Of course, one might say that such a representative set or sets cannot be
identified.  But I don't think anyone should be happy with that, since it
would apparently mean that the rate of change in languages is so random and
unpredictable that no scientific statement about it can be valid.

Regards,
Steve Long