Re: [tied] Abstractness (Was Re: [j] v. [i])

From: Piotr Gasiorowski
Message: 22672
Date: 2003-06-05

----- Original Message -----
From: Jens ElmegÄrd Rasmussen
To: cybalist@yahoogroups.com
Sent: Thursday, June 05, 2003 1:09 AM
Subject: [tied] Abstractness (Was Re: [j] v. [i])


> But /e, i, u/ and /a, i, u/ are equivalent since the phonemes are only
defined by being different from each other.

Phonemes are bundles of distinctive features (or compounds of phonological
"elements", depending on what kind of representation one prefers to work
with), and it _is_ typologically relevant what differentiates them, not only
whether they are different at all. The conventional minimal interpretation
of something transcribed as /e/ is [+front, -high], so the whole system of
contrasts is something like this:

i [+front, +high, - round, -low]
u [-front, +high, +round, -low]
e [+front, -high, -round]

Such an encoding is suboptimal in comparison with the following:

i [+front, +high, -round, -low]
u [-front, +high, +round, -low]
a [-high, +low]

No matter what the values of [front] and [round] for the low vowel (they may
be filled in by rules of allophonic realisation), any pair of phonemes is
differentiated by at least two features. For that reason in a triangular
inventory [a, i, u] are the typical phonetic values of the three vowels. The
vowel symbolised as /a/ may thus have frontish or rounded allophones, but
[+low] alone suffices to differentiate it from the other two vowels (whether
it's fully open or pronounced as a low schwa). Actually, a triangular system
might be encoded as:

i [+front]
u [+round]
a [+low]

The rest falls out from universal constraints and preferences.

> If the choice of the notation /e/ in the former analysis was not meant to
exclude its having a-like allophones, and the choice /a/ of the latter does
not exclude an e-like realization, then the two notations mean exactly the
same thing. And then one cannot exclude one of them and accept the other.

The choice of *e (rather than *a or *o) was motivated first and foremost by
something much more concrete:

Latin /e/ : Greek /e/ : Balto-Slavic /e/, Celtic /e/, Germanic /e ~ i/ :
Indo-Iranian /a/ (palatalising a preceding velar), etc. < *X. (Solution: *X
= /e/.)

The comparative evidence points to [e] as the value of PIE *e. If the
pre-PIE system was triangular, then the choice of *e for the low vowel
expected in such a system is not particularly felicitous. I'd even describe
it as misleading.

Piotr