Of Trees, nodes, and minimal paths (was Re: Urheimat in Lithuania?)

[Robert Whiting]

On Fri, 24 Mar 2000, Richard M. Alderson III wrote:

> On Wed, 15 Mar 2000, Robert Whiting (whiting at cc.helsinki.fi) wrote
> regarding the conservative nature of Lithuanian:

>> Expressed in terms of a tree model, it sounds very much like a scenario
>> that Steve Long proposed:  At any node on the tree, there is a
>> non-innovating branch and other branches.  If we follow the
>> non-innovating branch from each node, at the bottom of the tree we
>> arrive at a language that is practically identical to PIE (in this
>> case, Lithuanian).  Now everyone said 'no, no, that can't be right
>> because actually all branches innovate, just in different ways.'
>> The example of Lithuanian would seem to argue against this.  Nodes in
>> a tree must be based on some comparative linguistic information.  The
>> only such information that is useful is some innovation that appears
>> in one branch and not in the other.  Shared retentions don't
>> propagate either as waves or trees.  They just stay where they were
>> left, like a well-trained dog.  Of course the fact that Lithuanian
>> was always on the non-innovating branch (or least innovative branch if
>> you prefer) after satemization and RUKI (and only partly on the
>> innovating branch even there; sort of holding on to the end of
>> the branch with one paw) says nothing about where it was located
>> geographically.

> The problem is that there is *no* IE language without significant
> innovations vis-a-vis all the others.  Lithuanian is conservative in
> its *retention* of the nominal case system (for the most part--though
> it loses one and adds a couple based on a Finnic model), but it is
> innovative in the verb; which counts more heavily?

> Further, if we look at the *phonology*, Lithuanian is extremely
> innovative:  It merges the voiced plain and voiced aspirate series of
> stops, it has contrastive palatalized and non-palatalized series of
> obstruents, it merges *o(:) and *a(:) and otherwise disturbs the vowel
> system, and it moves the IE accent from the center of the word to the
> ends.  So now tell me how conservative it is.

> There is *no* branch on the tree that can be labeled "non-innovating",
> and we do ourselves a disservice as linguists, and a greater one to
> the non-linguists looking to us for guidance, by pretending that there
> is.

1) Yes, there is no such thing as a language that has not innovated.

2) Yes, it is likely that some areas of a language will innovate more than
others.

3) Yes, there is no branch of the tree that can be labeled
"non-innovating". (But there are branches at nodes that can be labeled
non-innovating.)

But there is a minimal path through the tree that when followed to the
bottom will find the language that is closest to the parent.  It is not as
simplistic as I originally expressed it (and Steve Long went astray when
he assumed that a non-innovating branch from a node never innovated again
until the next node), but we had a lengthy discussion about this last
August, started by Jon Patrick, under the subject of "Principled
Comparative Method - a new tool".  How the minimal path is calculated was
the essence of the thread and the specific application reported involved
only phonology, but it should be applicable to other areas of language as
well.

Now 2) above says that the minimal path for one area of language may
not (probably won't) be the same as for another.  But then one can just
award prizes by areas or average the path lengths for the areas to find
an overall winner.

But if it is possible to quantify the amount of change along each path,
then unless *every* IE language has innovated *precisely* the same amount
(down to the last decimal point) there must be a minimal path through
the tree.  There *must* be.  1) and 3) above just say that there will be
no path of length 0 in any given area (although there could be for a
specific character).

Personally, I think we do a disservice to linguistics when we say that
linguistic data can't be quantified so there is no point in trying.  It
gives the people who turn to linguists for guidance the idea that
linguists are simply innumerate.  While I feel that there are some
categories of linguistic data that are not readily subject to
quantification (particularly semantic change), there are others that are
(particularly phonological change), and what can be learned from these
quantifications should be pursued for what it may teach us about some of
the conclusions that have been arrived at by intuition.  Our tools aren't
that good yet that we can afford to ignore potential improvements.

Bob Whiting
whiting at cc.helsinki.fi