Patrick Ryan wrote:
>
> My vote is for trilled /r/.
>
> Why? The language that has best retained the Nostratic sound-system is
> Arabic; and Arabic has a trilled /r/.
Rhotics are by nature more variable than any other class of consonant.
This is probably because they are normally characterised by _lack_ of
features such as nasality, laterality, height or backness rather than
the presence of a shared trait. It's even arguable that they don't
constitute a natural class, being merely the "default sonorants" that
remain when you subtract all the well-defined classes. (On the other
hand, the class of trills, as opposed to rhotics in general, is easy to
define in phonetic terms, which is why it's possible to have a contrast
between trilled and non-trilled rhotics, as e.g. in Spanish and Albanian).
In any language you can expect a good deal of idiolectal and/or regional
variation in the pronunciation of rhotics, and there are well-known
areal fashions cutting across genetic ties, such as the use of a velar
or uvular approximant or fricative in much of northern Europe. I myself
use a uvular trill, though nobody had it among the people in whose midst
I learnt to speak, so my /r/ a typical case of a spontaneously arising
linguistic mutation. The question of Germanic rhotics has been revisited
by several authots recently, and the rough consensus of those who have
given some attention to the problem seems to be that while some kind of
apico-alveolar articulation predominated in Proto-Germanic, retroflex
and dorso-midpalatal ("bunched") varieties are also very old. There may
have been quite a lot of allophonic variation depending on the position
of the *r (prevocalic or in the syllable coda). My own article on the
pronunciation of OE /r/ is, alas, still unfinished, but I can reveal
that, in my opinion, it varied dialectally as much as its Modern English
descendant.
I think I posted the following link some time ago when we were
discussing the same thing:
http://benjamins.com/jbp/series/DIA/20-1/art/0003a.pdf
As regards the early IE languages, my vote is for the normal rhotic
variability. For example, the RUKI change would be easier to understand
if the Proto-Balto-Slavic and Proto-Indo-Iranian *r had had a dorsal
component (at least preconsonantally).
Piotr