the Wheel and Dating PIE

[Larry Trask]

Roz Frank writes:

>  Then does the IE model posit that PIE, understood here as an actual unified
>  linguistic system, was a linguistic isolate? It would seem that the model
>  would have to do this.

No; certainly not.  The recognition and reconstruction of PIE does not, in
itself, carry any implications at all as to whether PIE did or did not have
relatives.  Likewise, the reconstruction of Proto-Celtic or of Proto-Germanic
carries no such implications.

In the case of Celtic and Germanic, it can be demonstrated that these families
have discoverable relatives, including each other.  In the case of PIE,
it has not so far proved possible to identify any secure relatives among
known families, though people keep trying, and maybe one day they'll succeed.

>  Otherwise one would be confronted with a simulation
>  of linguistic prehistory in which PIE could be viewed as merely one member
>  of a language family existing at that point in time.

That is exactly what everyone believes.  But it does not follow that any
discoverable relatives of PIE survived long enough to be recorded.  If they
didn't, we're just out of luck, and our reconstruction can proceed no further
back in time.

>  Stated differently,
>  although I haven't heard this point discussed on the list, a cladistic
>  model requires the end point to coincide with a linguistic system that is
>  viewed as a total linguistic isolate.

No; certainly not.  No one believes Proto-Celtic or Proto-Germanic to be
an isolate.  It's merely that the reconstruction of one of these generally
pays no attention to anything outside the particular family being
investigated.  And the case is no different for PIE.

>  And even if PIE were posited as an
>  isolate, would one not have to propose that, nonetheless, the
>  proto-language, too, would have had the full characteristics of a human
>  language, with the likelihood of suppletions, irregularities and substrata.

Of course, except that there is no particular reason to posit any
significant substrate influences.

>  And I believe that it is this latter point that creates problems. How does
>  the model guarantee that the ultimate origin of the "common vocabulary"
>  should not be traced back, for example, to the substrata that PIE, if
>  understood as a natural language, must have had?

The items assigned to PIE are the items that can be reconstructed for PIE.
How they got into PIE in the first place is another matter.  But,
regardless of origin, if they were present in PIE, then they were part
of PIE, and they were available to be continued into the several daughter
branches.  That's all that counts for the reconstruction.

>  Hence, are we to understand PIE as a convenient shorthand for a set of
>  sharted characteristics or as a term standing for a reified linguistic
>  system spoken in prehistory?

The second.

>  And if it is understood as the second,
>  according to the model, how long did it just tread water? Stated
>  differently, if one chooses the second version, then one must ask how long
>  the unified (undifferentiated) linguistic system, as portrayed by the
>  reconstructions, go unchanged.

It did not remain unchanged for even a single generation.

Our version of the Uniformitarian Principle, a cornerstone of all
scientific investigation, requires that ancient languages should not have
been different from modern ones.  Since all living and recorded languages
are or were constantly changing, so were ancient ones, and so was PIE.

>  Languages do change. Are we to assume that
>  PIE was different?

No; of course not.

>  It seems to me that this is a very slippery aspect of a
>  cladistic modeling of the data.

It is not.  Remember, what we are reconstructing is the proto-language
*at the last moment* at which it was still a more-or-less unified system,
just before it began breaking up.

The earlier changes which had doubtless occurred within the ancestral
stages of PIE cannot be identified by comparative reconstruction, at least
not without some secure relatives of PIE -- which we don't have.  But it is
possible in principle to reconstruct back further within PIE by using
another method: internal reconstruction.  And precisely this has been
attempted by some specialists, perhaps most notably by W. P. Lehmann,
who believes that he can identify at least one, maybe two, significantly
earlier stages ancestral to the PIE that we reconstruct by the comparative
method.

>  On the other hand, if we choose the first alternative, that PIE is a
>  convenient shorthand, it acts like a frame in a moving picture: a
>  convenient way of portraying a stop-action of events that are otherwise
>  inevitably in motion.

That is exactly how we see it, but within the second alternative.

>  And as an aside, are there explict criteria set forth that determine
>  which items are most representative. I'm speaking of crtieria along the
>  lines of those that have been suggested by Larry Trask (and debated by
>  many) concerning the selection of items in Pre-Basque.

But the two cases are very different.  We have lots of IE languages, and so
our main tool is the comparative method here.  But Basque is isolated, and
so the comparative method is of minimal use, and only internal reconstruction
is available.  Hence different criteria are appropriate in the two cases.

>  I would be most
>  interested in knowing if such criteria have been debated and/or laid out
>  explicitly at some point in the past. For example, how many language groups
>  must the item be attested in for it to quality? I assume, for example, that
>  identifying cognates/reflexes of the same item in Sanskrit and Celtic would
>  be sufficient for the item to qualify? Or is the bar set higher for these
>  PIE items, e.g., that the item must be attested in Sanskrit, Germanic and
>  Celtic or Hittite, Slavic and Romance, etc.

There is no unchallengeable answer to this question.  By Meillet's Principle,
we require cognates in at least three branches of IE before we can reconstruct
an etymon for PIE.  But this is only a rule of thumb, and skilful specialists
need not adhere to it slavishly.

>  For example, just glancing over the entries in Buck, it would seem that
>  there isn't as much uniformity for "wheel" across IE languages, as there is
>  for, say, "cart" which shows up most IE languages (obviously with the help
>  of Latin).

But loan words don't count for the purpose of reconstruction.  If an
identifiable Celtic word is borrowed into Latin, from where it descends into
the Romance languages and is borrowed into Germanic and elsewhere, it is still
only the Celtic word which counts.

Larry Trask
COGS
University of Sussex
Brighton BN1 9QH
UK

larryt at cogs.susx.ac.uk