Rate of Change: A Closer Look

[X99Lynx at aol.com]

I wrote:
>If we are talking about sound changes that separate two daughter languages,
>we should be able to show that those specific sound changes took a certain
>number of years, days, hours and with historical evidence.  If we are
>claiming that that rate is somehow universal - justifying the application of
>a uniformity assumption - we should be prepared to show that the same rate
>per sound change can be observed in a large enough number of other languages.
>  Or explain why they vary.

In a message dated 6/22/2001 1:56:04 AM, sarima at friesen.net replied:
<<This assumes a constant rate of change,..>>

No it doesn't.  Average rates of change are based on multiple occurrences.
You get an average by finding the same changes in different languages and
averaging the time they took.  Variable rates of change (the opposite of
constant) simply mean that rates will vary in correlation to the change in
some independent variable.

<<What is suggested is a *statistical* median or mean rate, and a
variance in rate.  Given those one can determine a range of reasonable
values for when a *set* of changes took place.  One can certainly determine
a maximum amount of time that could reasonably correlate with a given
amount of change - that is how statistics works.>>

So, you have a historical *set* of changes and some numbers to supply here?
Now you're talking.  I'm ready to go.  My buddy and statistical wiz Seth
Margos has even volunteered to do the crunching.  He's done standard curve
analysis on projects where there were over a half million incidences and over
25 variables.  This should be a piece of cake.

First, what *set* of changes do you want to use?  Remember that we'll need
more than one occurrence of that set.  One occurrence has no statistical
validity, of course.  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.  This will give us your basic mean, mode, medium rates
of change, a rate of variance, standard deviations and some level of
statistical validity.

Now just find the same *set* in a prehistoric language and we'll get a pretty
respectable probable rate of change for that prehistoric language.

And THAT is how statistics works.

Of course, someone may come up with different *sets* of changes and get
different average rates of change, but Seth can handle that, too, if you can
quantify the differences between the *sets.*  (The rate of change might
"vary" predictably according to the *sets*.) But that's jumping ahead.

There's nothing mysterious here.  If Time consistently produces a MEASURABLE
amount of AVERAGE change in some specific feature of a number of historical
languages, that can be validated statistically.  And once it's validated, you
can statistically expect to find the same (plus or minus) "rate of change" in
prehistorical languages that have those same specific features.

Awaiting the data,
Steve Long