Grimm's Law and Predictability (ex Re: The Neolithic Hypothesis)

[Robert Whiting]

On Sat, 10 Apr 1999 X99Lynx at aol.com wrote:

> In a message dated 4/10/99 12:02:12 AM, you [JS] wrote:

> <<-- you're confusing description with prediction.  Language change
> _in the > past_ can be described and rules deduced, which can then
> be applied with reasonable confidence to historic languages we don't
> have direct evidence for.>>

> Probability is a tool can be used entirely to evaluate events that
> occured in the past.  Carbon dating for example provides a
> predicatability in dating accompanying non-organic artifacts that has
> been found to be very reliable.  "Predictability" is a factor that
> must be present before sampling validly can be used in analysis.

Exactly backwards.  Statistical analysis is the basis for
probability.  The probability obtained from statistical analysis
is used to predict future events whose outcomes are unknown.  It
is statistical analysis that makes for "predictability," not
"predictability" that makes statistical analysis possible.  The
limitation on this is that the statistical sample has to large
enough to ensure that it is representative.  "Predictability" is
always present but it is not accurate until the sample size is
statistically stabilized.

> Believe me, I'm not confused.

Oh, I believe you.  In general, the less one knows, the less
likely one is to be confused.  If one has only one point of view,
it is almost impossible to be confused.  So I do believe that you
are not confused.  No more were the Church Fathers who forced
Galileo to recant his heretical idea that the earth moved around
the sun confused.  Wrong, yes -- confused, no.  There is a
certainty that comes with single-mindedness that does not allow
for confusion.

> <<None of this makes us able to predict how the language will change
> _in the > future_.>>

> A subject I don't think I've ever addressed.

You address it when you use the terms predictability and
probability, because both of these terms are specific to future
events.  Of course you don't realize that you are doing this
because you don't know what predictability and probability mean.
And this lack of knowledge happily keeps you from being confused,
for as the poet says "... where ignorance is bliss; 'Tis folly to
be wise."  But what you see as a commendable lack of confusion
others see as blissful ignorance.

Technically speaking, probability does not refer to events that
took place in the past.  The term that applies to events that
have already occurred is "likelihood."  I present the definition
of this term from the CRC Concise Encyclopedia of Mathematics
(available on the web at

    http://www.astro.virginia.edu/~eww6n/math/math.html

    Likelihood

    The hypothetical Probability that an event which has already
    occurred would yield a specific outcome. The concept differs
    from that of a probability in that a probability refers to
    the occurrence of future events, while a likelihood refers to
    past events with known outcomes.

And, again, since there are still indications that you are using
a different Webster's than I am, I offer you the following from
Webster's Encyclopedic Unabridged Dictionary (1994):

    predict: v.t. 1. to tell in advance; prophesy. -- v.i. 2.
    to foretell the future; make a prediction. -- Syn. 1, 2.
    presage, divine, augur, project, prognosticate, portend.
    PREDICT, PROPHECY, FORESEE, FORECAST mean to know or tell
    (usually correctly) beforehand what will happen. ...

    predictive: adj. 1. of or pertaining to prediction. 2. used
    or useful for predicting or foretelling the future. 3. being
    an indication of the future or of future conditions.

    (NB: etymologies and examples deleted without indication)

So, again, when you say that you don't think you've ever
addressed the subject of changes in the future through your use
of "predictability," I believe you.  Doubtless you do think this,
but if so, it is just further evidence (if any is needed) that
what you think can't affect reality.

On Wed, 7 Apr 1999 X99Lynx at aol.com wrote

> And they [truisms about the unpredictability of linguistic
> change] are contradicted by the very fact that there is a
> Grimm's law and there is an Indo-European language group and
> there is a way the old sound laws can "predicatably" tell you
> if one word is cognate with another even if they are centuries
> apart.  That's what predictability means.  It means you can look
> at some word on a clay tablet and make a good guess at whether
> they are Greek or not.  Because it follows from prior experience.
> That's predictability.

I have already responded to this once, but your subsequent
postings (with one exception) indicate that you still do not
comprehend the difference between definition and prediction,
so I thought it best to come back here to try once more to make
it clear.

So let's take it step by step (all of you who already know the
difference, please bear with us).  First of all, as I said
elsewhere, a truism can't be contradicted.  It is true by
definition (even if it is trivial).  All one can do with a truism
is point out that a trivial truism (one that simply restates the
premise and adds no new information) is irrelevant as an
argument.  But when the premise is identified as a truism then
the premise is true by definition.  So saying that "a truism is
contradicted" is an oxymoron.

This being the case, it is obvious that Grimm's Law and the IE
language group cannot contradict the truism that linguistic
change is unpredictable.  Now if someone says that "Grimm's Law
predicts that a word that is inherited from PIE that has initial
/p/ in Latin will have initial /f/ in English," this is a loose,
figurative use of the word "predict" not a literal one.  This is
not a prediction based on Grimm's law; this *is* Grimm's Law (in
part).  The statement would be better construed as "Grimm's Law
states ..." or "According to Grimm's Law, ...".

Grimm's Law is a reconstruction of a (series of) sound shift(s)
that took place in the Germanic language.  In fact, far from
being evidence that language change is predictable, it is
confirmation that linguistic change is unpredictable.  For if
Grimm's Law was a predictable linguistic change then all IE
languages should have ("predictably") undergone this change at
some time or another, not just Germanic (and indeed, all
languages everywhere that had these sounds should have shifted
in the same way if this was a "predictable" linguistic change).
So Grimm's Law is not a "predictable" linguistic change because
it cannot be generalized to all languages.  However, because it
is not "predictable" it retains its predictive power in being
able to identify Germanic languages in the overall scheme of IE.

Now if one uses a set of data to create a reconstruction and then
turns around and says that the reconstruction "predicts" the
data, this is simply circularity.  This is how linguists, both
historical and descriptive, get the reputation of operating a
discipline based on circular reasoning.  If a historical linguist
reconstructs a sound shift and then says that the sound shift is
predictable because of the reconstruction, or if a descriptive
linguist writes rules that describe sound changes and then says
that the sounds changed because of the rules, this is circular
reasoning and it is an easy trap to fall into.  This is why
historical reconstructions and descriptive rules do not predict
anything.  They are simply descriptions based on a data set.  They
cannot say anything about the data set without being circular.

Of what value then are these hard-won reconstructions like
Grimm's Law and Verner's Law?  These reconstructions are
descriptions (of events that took place long ago when no one was
around to record them) based on observations, and descriptions
can be used in definitions.  So by reconstructing PIE we are
defining it through description.  By reconstructing the sound
changes that took place in the various branches (e.g., Grimm's
Law) we are describing the various branches in terms of PIE.
But we are not "predicting" the various forms in the present
languages because we already have those.  They are the data
on which the reconstructions are based.  At best we may "predict"
(reconstruct) some intermediate forms that are not attested, but
this is a "prediction of the past" because that's what
reconstruction is.

As an example, if we observe that a horse has four legs (one on
each corner) then we can use this observation as a description
that makes up part of the definition of "horse."  Aha, you will
say based on your idea of predictability, that means that we can
predict that a horse has four legs.  No, I will say, this is not
a prediction, this is simply part of the definition of "horse."
Anything that is a "horse" will have four legs because that is
part of the definition of "horse."  If it doesn't have four legs
it won't be called a "horse."  And the definition stands because
no one has ever seen a horse that didn't have four legs.  This is
a matter of consistency of observation and prior experience.  But
"a horse has four legs" is not a prediction, it is a definition.
If someone finds a type of horse that doesn't have four legs,
then the definition will have to be revised.

Similarly, when we look at the words of a language like
Mycenaean, we are not "predicting" what that language is, we are
seeing how closely it matches our description / definition /
reconstruction of various languages and classifying it according
to which of these it resembles most.  Analyzing these words is
not a matter of "predicting" that they are Greek.  It is a matter
of seeing how closely these words match our definition of Greek
as reconstructed for the appropriate time period.  Matching them
will be a function both of whether they are Greek or not and of
how accurate our description / reconstruction is.  If they match
closely enough to be reasonably sure that they are Greek but not
closely enough to confirm the reconstruction in detail, then the
reconstruction / description / definition will have to be
revised.  So when you say:

     That's what predictability means.  It means you can look at
     some word on a clay tablet and make a good guess at whether
     they are Greek or not.

your use of "predictability" is completely idiosyncratic (and,
based on prior experience with such terms as cognate and stem,
why am I not surprised by this?).  You are saying that
"predictability" is the ability to form a hypothesis while
everybody else is saying that "predictability" is the likelihood
of a hypothesis being correct.  To you, if you can form a
hypothesis about something, that is "predictability"
(particularly if the hypothesis has already been shown to be
correct).  But "predictable" means "capable of being determined
in advance" and "predictability" means "the quality of being
predictable." "Predictable" does not mean "capable of being
hypothesized about (in retrospect)" as you seem to think.
Prediction is something that happens a priori.  So looking at
some word on a clay tablet and deciding whether it is Greek or
not is not a matter of predictability.  It is a matter of how
well one knows the definition of Greek.  It is simply a matter
of using the data set and the reconstruction.  Now if one could
make a good guess at whether the words were Greek or not
*without* looking at the tablet (or without being able to read
the tablet) *that* would be predictability.

In a separate posting you said

     The shock of finding out Linear B was Greek (nobody was
     predicting it, not Evans or even Ventris) ...,

so, obviously, at some point you *couldn't* look at some word on
a clay tablet and guess whether it was Greek or not.  Nobody was
predicting it, therefore, there was a lack of predictability
(although not a lack of hypotheses).  Then as soon as the
language is identified as Greek, you *can* look at some word on a
clay tablet and guess whether it is Greek or not, and that's
supposed to be "predictability."  And this is the evidence upon
which we are to accept your statement that you are not confused?

But don't worry, I still believe that you are not confused.  You
are just using a different language from everybody else and
thereby confusing them because they think you are trying to use
the same language they are.  So if everybody else will just
substitute "describability" whenever you use "predictability"
we will all be talking about the same thing.

I did mention the one exception to your misunderstanding of the
term "predictability," and I would also like to take up the few
shreds of predictability that exist in historical linguistics,
but this is too long already so I will save that for another
posting.

Bob Whiting
whiting at cc.helsinki.fi