STATISTICS IN LINGUISTICS

[Patrick C. Ryan]

Dear Steve and IEists:

-----Original Message-----
From: X99Lynx at aol.com <X99Lynx at aol.com>
Date: Friday, January 29, 1999 12:49 AM

>In a message dated 1/27/99 11:55:41 AM, proto-language at email.msn.com wrote:

><<Predictive power, IMHO, is based on statistics.>>

>Predictive power is based on an effective understanding of cause and effect
>relationships.

The relationship between cause and effect is statistical. If the same cause
is regularly observed to produce the same effect, the statistic
(probability) is 100% theoretically. Why do you resist this simple truth?

>Statistics are just one way to prove that understanding is correct.

Statistics are the only way to prove that understanding is correct. When a
cause is never observed to produce an effect, the probability is 0%. Why is
that so hard to understand?

>Without a
>proper hypothesis, whether gathered statistically or not,

Hypotheses are not "gathered" statistically. A hypothesis is validated or
not by statistically verified the results of its predictions.

>there is nothing for
>statistics to help prove.

I agree. If you are doing or saying nothing, statistics are minimally
helpful.

>A lingusitic model is a hypothesis, properly used a
>statistical sample can help prove if that model is accurate.

Statistics is the only method that can suggest (not prove) that the model is
accurate.

>However, where there are few occurences to sample and possibly no new
>occurences - as in ancient languages - statistical analyses do not carry
>much "predictative power".

This is ridiculous! Any time you make a prediction, even without formally
applying statistical methods, you are operating statistically. Judgment is
about weighing factors of probability.

>Also recall if you will that I posted on this list the fact that the
>Peresus
>Project showed a higher occuresnce of purus>red in Greek than ereuthem>red.
>This was due to the way Lidell-Scott defined red.  Statistics can only
>yield
>results that fact gathering and prior analysis permit.

Statistics cannot overcome GI.

Pat