What is Relatedness?

[X99Lynx at aol.com]

In a message dated 1/22/00 3:22:22 AM, mclasutt at brigham.net wrote:

<<Can vocabulary be borrowed to such an extent that evidence of a relationship
is lost?  Not that I have ever seen.  Even when a language borrows really
extensively from another language (as English has borrowed from Italic),
there is always a base level of commonly-used vocabulary that still retains
its ancestral lexicon.

Can morphology and syntax be borrowed to such an extent that evidence of a
relationship is lost?  Most likely not. >>

My purpose in giving extreme examples was to get at some sense of the basic
mechanics that are going on.  (E.g., you can't detect gravity's democratic
effect on some objects unless you test them in a vacuum.  A feather and a
cannonball will fall at drastically different rates under normal
circumstances.)

My original post was meant to apply to the Upenn tree's methodology.  Without
repeating what I wrote in prior posts, my fundamental questions don't change,
I think, despite the very impressive processes you described in your post.

Relatedness to *PIE is one thing.  The relatedness between the daughter IE
languages is another.  And that difference is because PIE is reconstructed
and its relatedness is entirely based on the evidence given by its daughter
languages.  There therefore should in theory be no "conflicting indicia of
relatedness" in *PIE.

In terms of relatedness between IE daughter languages, you can and do find
conflicting indicia of relatedness.  (The recent posts regarding the Germanic
family tree illustrate how you can read shared attributes both ways.  I think
there is a map in Larry Trask's textbook of shared characteristics in
Germanic (p.186) and I swear you it is positively maddening when you try to
follow the enclosing circle lines as they criss-cross each other.)  The same
thing happens very often in biology, geology and even particle physics.  This
is what the kind of methodology used in the UPenn tree is really meant to
tackle.

There are a number of problems that arise when one decides how to handle
conflicting indicia of relatedness.  For the sake of space, let me give but
one.  What if five lexical attributes appear as shared innovations between IE
languages A and B and therefore evidence a greater relatedness between the
two than a language C - which has none of them.  What if however Language C
shares a totally valid morphological innovation with B not found in Language
A?   How is this reconciled?

Hidden within this problem are the questions of loss (loss in A can reconcile
the dilemma, but loss itself may be "an innovation") and whether retained
characteristics versus pure innovations should be used in a true cladistic
model - i.e., whether shared attributes descended from PIE are proper in this
sort of analysis.  There is also the "count" problem - do five lexicals beat
one morphological?  I've tried to mention these in other posts.

But let me here just make this one point - different solutions shape the tree
differently.  One alternative in the example above is to give each language
an independent branch on the tree - very unsatisfactory as a showcase tree
but accurate in terms of the uncertainty created by the data.  This is a very
common solution in paleobiology.

But the dilemma mentioned above can also be reconciled if the inconsistent
attribute is eliminated.   You can do that accidentially if you limit your
sampling to attributes that cause less problems - e.g., where the "etymology
is certain."    This is far easier to do if you limit yourself to 300-400
attributes across 12 language families over 3000-4000 years - most of them
apparently lexical.

Some cladists will tell you that Deep Time is never a matter of years, but a
matter of unrecoverable information.  Time of course is relative - change
accelerates the effects of time.  In fact, time theory says it is possible
that the conflicting indicia of relatedness between A, B and C may not be
resolvable by human means -  like "deep time" events that happen every day
when viruses reproduce in a petri dish.

And that was one problem I was addressing in my original post - the degree of
uncertainty that varying rates of change in language MUST create.  The
accepted thinking in other fields is that you deal with uncertainty with MORE
data, not less.

If you are going to assure me that all such conflicting indicia of
relatedness are resolvable by internal reconstruction, then perhaps this is a
non-problem.  But if that's not the case - then one problem with the UPenn
may be its way of handling varying rates of change in the IE daughter
languages - precisely because varying rates of change SHOULD produce
different levels of loss and unknowability among related languages.

And that would mean that one may not be measuring filial relatedness with
those 300-400 attributes, but rather an artifact - one may be measuring
nothing more than the varying rates of internal change among those languages.

Regards,
Steve Long