What is Relatedness?

[Sean Crist]

On Fri, 5 Nov 1999 X99Lynx at aol.com wrote:

> In a message dated 10/26/99 11:09:50 PM, kurisuto at unagi.cis.upenn.edu wrote:

> <<When we have a family of related languages, on what grounds do we decide
> what the internal structure is for the family tree?  In other words, how
> do we decide which of these languages are more closely related than others
> in the family?>>

> But what does "more closely related" mean?   This isn't just a question about
> terminology.  It rather goes to what we are proving and hope to prove by
> establishing "relatedness."

Consider the structure of the Germanic Stammbaum:

             PGmc
            /    \
        PNWGmc    \
        /   \      \
      WGmc  NGmc   EGmc

When I say that WGmc is more closely related to NGmc than to EGmc, I mean
that the lowest node dominating WGmc and NGmc is dominated by the lowest
node dominating WGmc and EGmc.  In other words, WGmc is said to be more
closely related to NGmc than to EGmc because WGmc and NGmc share some
innovations which EGmc did not undergo.

We've previously discussed at great length exactly what this hierarchical
structure represents, so I won't repeat it here.

> If - for example - a language has innovated and borrowed so wildly that it
> retains very little of the ancestor, it may be "more closely related" in some
> chronological sense.

No.  The _amount_ of innovation is not the basis on which we draw
Stammba"ume; we draw these trees on the basis of what is _shared_.

[...]
> A "backwoods" language exposed to new and sophisticated cultural and
> technological input might expand many times over its original form.
> Many new words and concepts might be introduced.

The presence of loan words doesn't alter the genetic affiliation of a
language.

> New tenses might suddenly be needed to indicate matters of time and
> relationships that simply did not matter in the old days.  People who
> calculated time only in terms of the seasons might need to start
> perceiving and discriminating befores and afters, duratives and
> completedness, perfects and aorists.

Languages can certainly develop new tense markings over their history, but
the explanation you've given here is teleological.  There's nothing that
would suggest that languages develop more complex tense systems upon
coming into contact with a technologically more sophisticated culture.

[...]
> This is a side-path off of the discussion in which Larry Trask discussed a
> language changing so much that it might be perceived as a different language.
>  And perhaps it raises the question whether counting the number of apparent
> differences between languages (as e.g. the UPenn tree does to some degree) is
> a valid way to measure genetic distance.

Ringe, Warnow, and Taylor did not just "count" any old differences; what
they were specifically looking for are shared characteristics which cannot
reasonably be attributed to parallel innovation.

> How much must a language borrow, for example, before it starts to owe more to
> the loaning language than it does to a parent or grandparent that has just
> left a few strands of genetic relation?

It might owe a great deal to the loaning language in terms of its lexicon,
but we just don't find cases where languages substantially borrow syntax
or inflectional morphology.  This is one of the reasons why inflectional
morphology is extremely valuable in the determining genetic affiliation.

Loan words are one of those things that we try to factor out when
determining genetic affiliation.  It doesn't matter how many words a
language has borrowed; its genetic affiliation does not and cannot change.
As we've previously discussed, you can often identify loan words, because
they will not have undergone the sound changes which occurred in the
language prior to the borrowing.

[...]
> <<Nearly everybody agrees that we should do so on the basis
> of shared characteristics of the languages which cannot reasonably be
> attributed to parallel innovation or to borrowing. >>

> And of course there is an inverse function that applies here.  If some
> languages had those shared characteristics, but lost them before they became
> documented in writing or otherwise left no evidence - it would be taken as
> evidence of relative unrelatedness.

No, no, no.  It's well known that things such as morphological categories
can be independently lost; this is a very common sort of parallel
innovation.  Ringe, Warnow, and Taylor were well aware of this, and dealt
with this problem by assigning a separate numeric code to each language in
the case of such loss so that such spurious groupings would not occur.

[...]
> The UPenn seems to use a "value" or a neutral count (in terms of shared
> innovations) called "lost."

A _separate_ value is assigned for each branch to mean "lost".

>   This APPEARS to apply only to attributes that
> are presumed to have been present in PIE.  But the fact is that a truly
> innovative language might have lost shared innovations that arose after PIE,
> and those lost shared innovations might have given us a completely different
> picture of relatedness.

> But because the loss happened before documentation, we are mislead into
> thinking they were never there.

> After all, if presumed attributes of PIE can be "lost" in daughter languages,
> then attributes ("shared innovation") of a sub-family - e.g., NW IndoEuropean
> - could have been lost in a particularly innovative member of that subgroup -
> e.g., Greek or Latin - before documentation.  And I don't believe there is
> any methodology that could neceassrily recover them.

See my previous paragraph.

  \/ __ __    _\_     --Sean Crist  (kurisuto at unagi.cis.upenn.edu)
 ---  |  |    \ /     http://www.ling.upenn.edu/~kurisuto/
  _| ,| ,|   -----
  _| ,| ,|    [_]
   |  |  |    [_]