Relative chronology and absolute certainty

[X99Lynx at aol.com]

In a message dated 9/10/1999 11:39:12 PM, Sean Crist wrote:

<<Both analyses work, in the sense that both analyses correctly predict the
attested forms. >>

I just wanted to go back and make a number of things clear about the original
point of this tread.

The issue up till this post was whether the comparative method would yield
the single result given in Sean Crist's original hypothetical ("Suppose that
one of the daughter languages first palatalizes *k before *i, giving /*ci *ke
*ka *ko *ku/, and then merges *e into *i, giving the attested forms /ci ki ka
ko ku/...,"etc.)  My point was that there is a "fundamental uncertainty when
we are trying to reconstruct from two daughters."  And Sean here concedes
that uncertainty, but continues in his post to use some additional approaches
to get to a single result.

Now, my pointing out the duality that happens whenever you reconstruct from
two equal and coeval daughters wasn't aimed at the comparative method.  It
was to point out that there is a degree of uncertainty that sound rules can't
be eliminated.  I need to emphasize that this is not an indictment of the
comparative method.  This is built in to our status as observers.  It applies
to more than historical linguistics.  You'll find the same sort of principle
in the physical sciences.

It simply means that there are degrees of certainty and that conclusions
dealing with time and reconstructing the order of events are never absolute,
but a matter of probabilties.  And the acceptible presence of an alternative
possible explanation that can never be eliminated.

Sean goes on to write:
<<Given that we can't decide between the two analyses
within the Comparative Method, this is one of those cases where it's OK to
fall back on your notions about what's phonetically plausible.>>

Now going to what's 'phonetically plausible" is fine.  It is outside Sean's
original hypothetical, which was originally about internal necessity.  This
is another case of an assumption about phonetic history, which is well and
good, but you can see why it is important that it is stated.   If the
proposed solution to the hypothetical is phonetic plausibility, then that
allows us to examine how valid that assumption is.  And how much uncertainty
it creates.

<<We note that palatalization rules like ki > ci are widely attested in the
languages of the world, but un-palatalization rules like ci > ki are much
rarer.  If we accept this reasoning, then Proto-AB must have had *ka *ke
*ki *ko *ku, >>

Now take a look at the logic here.  ki > ci is widely attested.  ci > ki is
much rarer.  Accepting that, then the proto language MUST HAVE HAD a certain
reconstruction.

But we know that's not true.  The rarity of ci > ki only affects probabilty.
It doesn't make the reconstruction absolutely certain at all.  In fact, it
makes absolute certainty an absolute error.

So why is such absolute terminology being used?  Putting aside that this fact
about the world's languages' habits re 'palatalization' wasn't part of the
original hypothesis, we can still ask a very important question.

Just how widely attested is ci > ki and just how rare is ki > ci?  Especially
in the context of discussions about algorithms and statistics, wouldn't it be
appropriate to talk about what the occurences are here?  How often has ki >
ci happened?  How often has ci > ki happened?  Is there a count of these
events?

If there was ever a place where statistics in linguistics would be helpful,
this is a good one.  Because this is exactly the kind of assertion that
really needs some back-up before it should be credibly accepted as the basis
for accepting this reconstruction.

(I won't go into how selective this external evidence is.  If we are going to
bring the pattern of the world's languages into this, why are we excluding
the possibility that /c/ in the hypothetical might represent loans - like
intervocalic /s/'s in Latin - or other more intricate pathways.  Obviously
the history of the palatal-velar sound changes in this particular language or
its contemporaries would be more relevant than loose generalizations from the
world's languages.  But a hypothetical language can have a dozen different
hypothetical histories, and we should stop somewhere.)

<<If we accept this reasoning, then Proto-AB must have had *ka *ke
*ki *ko *ku, and more importantly here, the palatalization and vowel
merger must have applied in the order I gave them here, and cannot have
applied in the other order.>>

But we should not accept this reasoning.  Not if it is used to justify
another absolute statement.  Because the reasoning only perhaps makes it
likely.  It tells us nothing that justifies such certainty.

Sean continues to add additional hypotheticals to the hypothetical.  One is
particularly interesting:

<<At this point, no phonologist or historical linguist would take seriously
the idea that Language A represents the original situation.  Note that
even if we didn't have any evidence for Language B, we could still
correctly and unambiguously reconstruct Proto-AB purely by performing
Internal Reconstruction on Language A, thanks to the morphologically
conditioned alternations.>>

The new hypothetical Sean is referring to is:

<<...   Language A
        takutu          'I run'
        tacid           'you run'
        takil           'they run'>>

It amazed me to learn that without "any evidence for Language B, we could
still
correctly and unambiguously reconstruct Proto-AB purely by performing
Internal Reconstruction" on the evidence given above.  I presume this has
something to do with some rule about the phonetic formation of second persons
or something gleaned from the world's languages that allows us to recreate
the parent with such absolute certainty.  This is an important rule and I
wish I knew it.

Sean continues:
<<But if this still isn't enough for you, suppose that Proto-AB contrasted
*/k/ and */c/; thus, Proto-AB contrasted the series *ki *ke *ka *ko *ku
against *ci *ce *ca *co *cu....In the case of monomorphemic words, for
example, there'd be no way to undo the merger....>>

Once again, there would be no need to undo a merger.   This is a two step
process, between parent and daughter.  No steps left for unmerging.

<<As I discussed in detail in my last post on phonemic split, we never posit
a sporadic phonemic split if there is an alternative account available
strictly in terms of regular sound change.>>

And as I pointed out, positing a sporadic split in reconstruction is never
the only other alternative.  But more importantly this absolute approach -
assuming 100% certainty when we know that sporadic changes or things that
look like sporadic changes exist - guarantees sooner or later a false picture
of what actually, historically happened.

If a sense of statistical or practical uncertainty can be accepted and even
made good use of in quantum physics, it certainly has some use in historical
linguistics.

Regards,
Steve Long