> Date:         Tue, 19 Jun 2001 05:55:10 +0000
> From: Raymond Brown <[log in to unmask]>
> At 11:42 am +0200 18/6/01, Christophe Grandsire wrote:
> >En réponse à Patrick Dunn <[log in to unmask]>:
> >> When one is offered several choices above, are those variations within a
> >> single word class, or examples of different declensions?

> >Both! it seems to depend on dialect, neighbouring sounds, etc...
> >the picture is very blurred here.

> Of course, it must be.  It was very unthoughtful of the Proto-Indo-European
> speakers not to standardize their language for us; the trouble was they
> were around before anyone had invented writing & that made standardization
> extremely difficult.  PIE is an abstraction; what we have is a set of
> related and continually changing dialects; they had a long history behind
> them and were going to have a long history before them.  The reconstruction
> just pushes beyond the earliest written forms to something that is, we
> hope, vaguely like the set of dialects that gave rise to the family of
> languages we now call Indo-European.

Also, there an amount of minimalism in the traditional reconstruction.
It is quite possible that the ancestral IE speakers did at one stage
utter words like *ph_2tros = /p?\tros/ and *h_3dntos = /?\_w?tntos/
(Greek patrós and odóntos) for the genn.sgg. of father and tooth;
Georgians seem to be happy with that sort of cluster though it's not a
very common type of phonology. (I'm following Beekes here in using
traditional *-notation, but assuming the glottalic version of stops).

But it is also possible that there was all sorts of unstressed vowels
or schwas around that disappeared later, and that the symbols used in
standard reconstructions were actually something quite different that
just happened to develop in a way that's consistent with what we see.

Take zero-grade ablaut, for instance. In traditional *PIE, it causes
massive clusters like the ones above, but in the daughter languages
most of these are resolved by vocalizing laryngeals and semivowels or
dropping consonants. Greek is a bit more cluster-tolerant and has
forms like khthonós < *d^hg'^hmós (note metathesis), but that
could still be consistent with an interpretation of ablaut like this:

                        PIE             //
        short           *Ø i u          /a i u/ > /@ i u/
        stressed        *e ei eu        /"a "i "u/ > /a ai au/
        long            *o oi ou        /a: i: u:/ > /A Ai Au/
                        *e:C            /a:CX/ > /ahC/
                        *o:C            /a:CX/ > /AhC/

where /C/ is any conoid and /X/ is some unknown sound that marked for thematic nominals (since that's the main occurrence of the
lengthened ablaut grade). This would make *PIE into a nice three-vowel
low-cluster-factor language like Arabic, instead of a Caucasian type,
and the forms shown above would become

        *ph_2trós       /paxtar"A:s/
        *h_3dntós       /?\a?tant"A:s/
        *d^hg'^hmós     /t_hak_ham"A:s/

(This is not a serious attempt at explaining all details of IE ablaut,
but ideas like this have been seriously considered by some scholars).

The point is that *PIE is an abstraction. Changing the phonetic model
like this means that the abstract sound laws leading to the earliest
attested stages of the daughter languages will look different --- but
they shouldn't be less plausible. (If they are, the new model is
worse, of course).

Comparativists seem to dislike having symbols in their reconstructions
that do not correspond directly to sets of phonemes across daughter
languages --- which means that will rather leave a zero in an awkward
place than pick a schwa out of thin air. That's understandable, but
it's regrettable if the abstract forms are then taken as evidence that
the languages that actually existed had any specific phonetic

Lars Mathiesen (U of Copenhagen CS Dep) <[log in to unmask]> (Humour NOT marked)