The UPenn IE Tree

Larry Trask larryt at
Mon Aug 23 15:26:55 UTC 1999

On Thu, 12 Aug 1999 X99Lynx at wrote:

[on the Penn tree for IE]

> I simply must ask some questions about what this means.

> 1. I assume the branching off in this 'Stammbaum' carries the
> inference of being chronological in the sense of earlier or later
> separations.  (Rather than for example the degree of linguistic
> difference between languages.)  This may go without saying, but I'm
> just checking.

Yes; the tree is intended to be a relative -- not absolute --

> So here at this first juncture:

>                                     PIE
>                                    /   \
>                                   /   Anatolian

> Does this mean that PIE co-exists with Anatolian?  It would have to
> wouldn't it?

No.  An ancestral language cannot co-exist with its own descendant.

It is concluded by the Penn group, and widely believed anyway, that
Anatolian was the first branch of IE to split off from the rest. So,
that top node, with its two daughters, represents an initial split of
the single language PIE, with one daughter being the ancestor of
Anatolian, and the other daughter being the single common ancestor of
everything else.  We now often speak of `broad PIE' -- the ancestor of
the whole family -- and `narrow PIE' -- the ancestor of everything
except Anatolian.  Narrow PIE is a sister language of Proto-Anatolian:

			Broad PIE
			  /  \
			 /    \
			/      \
		       /	\
		  Narrow PIE  Proto-Anatolian

> Then where along that left side diagonal does PIE cease to exist?

This is not a question about facts, but only one about terminology. I
don't think anyone wants to apply the term `PIE' to anything below what
I have just called `narrow PIE'.  If you prefer Sturtevant's
`Indo-Hittite' nomenclature, then the last tree gets renamed as follows:

			Proto-Indo-Hittite [sic]
			      /  \
			     /    \
			    /	   \
			   /	    \
			PIE	Proto-Anatolian

> This question of when PIE ends strikes me as an important question,
> for a number of reasons.

Actually, it's only a question of terminology, no more.

> First, reconstruction always seems to proceed as if *PIE were a
> static language - but coexistence could have meant centuries of
> potential change within PIE itself.

Coexistence has nothing to do with it, but otherwise you are dead right.
PIE must have exhibited the same regional and social variation as any
other living language, and it must have changed continuously even during
the centuries when it could still reasonably be regarded as a single

But reconstruction is not good at identifying variation -- which is not
to deny that IEists have often drawn attention to possible instances of
variation within PIE.  As for the change over time within PIE, this has
received a fair amount of attention.  Ideally -- though not always
actually -- reconstruction gives us the latest version of PIE, just
before it began breaking up into distinct daughters.  But it is
perfectly possible in principle to do internal reconstruction *within*
PIE to identify earlier and later stages of it, and precisely this has
been attempted.

For example, the German linguist Specht has argued rather persuasively
that athematic nouns are generally older within PIE than are thematic
nouns.  And several people have devoted attention to the possibility of
reconstructing a version of PIE which is substantially earlier than, and
quite different from, the late version that we commonly see in the
handbooks.  Diakonov (I think it was) coined the name `Pre-PIE' for this
earlier version, while others speak merely of `Early' and `Late' PIE
(and sometimes also of an intervening `Middle' stage).

> Second, it means that languages coexisting with PIE could have been
> influenced by or influenced PIE after splitting off.  And third it
> would mean that PIE could have been influenced by non-PIE influences
> between splittings.

Yes, except that we wouldn't use the name `PIE' for anything descended
from PIE.

> And logically either PIE either coexisted with some of these branch-offs.

No.  It can be misleading to think of one branch as branching off from a
`trunk'.  There is no trunk.  All that really happens is that the single
ancestor (at any given point in time) splits into two (or more) distinct
daughters.  And thereafter it is a matter of chance whether any of those
daughters go on to become ancestral to a large group of languages.

> Or they all branched off at one time and PIE evaporated.

The traditional picture does indeed see PIE as fissioning simultaneously
into ten or twelve daughters.  But it is precisely this picture which is
challenged by the Penn tree.

> OR PIE never disappeared but turned directly into one or a few of
> these languages, which would be direct rather than indirect
> descendents.

Sorry; I don't follow.  All IE languages are direct descendants of PIE,
by definition.

> I think those are all the choices.  There are no others, but each
> one should result in completely different reconstructions of *PIE.

I don't see how.

> 2. You wrote: <<... Indo-Iranian, Greek, and Armenian are in a single
> sub-branch of the IE family together, but Balto-Slavic and Germanic are in
> this branch as well.>>

> If that's the case, then what did that whole subbranch split off from?  At
> the point of the split of Greek Armenian, the left line is still there.
> Above are the Italo-Celtic, Tocharian, Anatolian branches.  Presumably they
> are distinguishable from whatever it is that is out there that might be
> called Proto-BS-Germanic-Indo-Iranian.

> What does this mean for reconstruction of *PIE?  What if it was
> Proto-GrAr-BSGer-IIr that was the branch off and Italo-Celtic was the true
> remainder of the PIE 'trunk'?  (I don't think you can favor one or the other
> branch - why should one be seen as more lineal to PIE than the other?)

Exactly.  It is merely an accident, in the Penn tree, that at almost no
point does a branch split into two branches each of which goes on to
split further into several of the major branches of the family.
According to the Penn results, six of the eight splits recognized
produce two daughters, one of which is the ancestor of only a single
major branch of IE.  The last split yields Baltic and Slavic.  Just two
of the earlier splits produce daughters each of which is the ancestor of
two or more of the major branches we recognize.

> In that case, Italo-Celtic would preserve PIE best and the other
> branch would be the split-off, innovating away from the core.  And
> misleading us as to what PIE was like.

No, no.  First, the Penn tree does not even recognize an `Italo-Celtic'
branch.  It merely concludes that Anatolian was the first to separate
from all the rest, Celtic the first to separate from the remainder, and
Italic the first to separate from the new remainder.  Second, the age of
the split has nothing to do with the degree of conservatism or
innovation in any given branch.  In fact, almost everybody picks
Lithuanian as the most conservative living IE language -- and yet Baltic
split from Slavic only at a very late stage indeed, and Balto-Slavic
split from others only rather late.  The branch leading to Balto-Slavic
has generally been more conservative than the branches leading

> 3. <<The team hypothesize that Germanic started out in life as a
> sister of Balto-Slavic, but that the pre-Germanic speakers came into
> the political orbit of the prehistoric Italo-Celtic peoples and
> absorbed loan words from them at some date prior to Grimm's Law.>>

> How does the team view the new reconstruction of the obstruent
> system (Hopper/Gamkrelidze/etc.) that suggests that Grimm's Law
> actually reflects archaism rather than innovation?

The Penn team didn't make much use of phonological characters, which
they regard as less reliable than lexical and morphological characters.
So this shouldn't make much difference.  It is mainly the divergent
results for lexical and morphological characters which make Germanic

> With that view, would Balto-Slavic become a sister to Germanic that
> came under the influence of IIr, ditching that archaism like IIr?

Dunno, but I doubt it.

> 4. Just hypothetically, if we were to assume that PIE was nothing
> but very early Greek, how would this diagram and the findings behind
> it change?  Would the tree look all that different?  Would it have
> Greek-Armenian at the bottom of the main stem?  Or would it?

Sorry; I don't understand the question.  In a sense, PIE *was* "very
early Greek", since it is the direct ancestor of Greek.  But it was
equally "very early English", or "very early Bengali".

> Does this diagram seem to put IIr in that last position (or
> IIr-BS-Gr) - does that possibly reflect a sampling artifact favoring
> Sanskrit, Germanic and Lithuanian/Slavic - the favored sources in
> many *PIE reconstructions?

Don't know what you mean by "that last position".

Of course, the Penn work is potentially vulnerable to sampling
artefacts.  If they had chosen different characters, or different
languages to represent the major branches, maybe the outcome would have
been somewhat different.  But this sort of thing can be tested for.

Larry Trask
University of Sussex
Brighton BN1 9QH

larryt at

More information about the Indo-european mailing list