From: Piotr Gasiorowski
----- Original Message -----
From: Jens Elmegård Rasmussen
Sent: Wednesday, June 04, 2003 2:26 AM
Subject: [tied] Abstractness (Was Re: [j] v. [i])
> I have no idea about preferences, but one compares languages typologically
by using equal standards. And the analysis of PIE that was rejected on
typological grounds was one granting consonantal status to anything that has
a consonantal allophone.
I'm not even sure PIE *j and *w deserve equal status. *w was certainly more
of a consonant than *j was even in phonotactic terms (for one thing, *wr-
and *wl- were permissible initial clusters while *jr- and *jl- were not). If
it were possible to study PIE synchronically, one would perhaps be forced to
regard *w and *u as separate phonemes while treating *j as an allophone of
In living languages, other things being equal, glides that correspond to
high vowels are regarded as non-syllabic allophones of essentially
non-consonantal phonemes, rather than vice versa. It's a natural approach,
since [j], [w], etc. are "marked" with respect to [i], [u], etc., and, to my
knowledge, never occur without the latter (the exact reverse is true of
liquids or nasals, which are _primarily_ consonantal and _secondarily_
vocalic). The fact that French /i/, /u/ and /y/ all have non-syllabic
allophones has not inspired anyone to exclude them from the inventory of
French vowel phonemes. So, if you ask my opinion, if I were to treat PIE
like a living language (that is, using equal standards), I wouldn't hesitate
to grant phonemic status to *i and *u. It's the status of *j and *w that
would remain to be determined.
> You are surely not denying the existence of /y/ and /v/ as consonants in
Sanskrit, are you?
Certainly not. But allophony is a tricky thing. As time goes by, what used
to be consonantal and vocalic allophones go they divergent paths and cease
to be allophones past a certain point, since their phonetic similarity
decreases with time and it no longer occurs to native speakers that they
might be "the same thing". Indo-Aryan /v/ was far advanced on its way
towards becoming an obstruent already in Sanskrit (still later it evolved
into /b/, as in Spanish), and /y/ soon followed suit (ending up as a palatal
affricate, as in Italian). [i] and [j] in Latin were certainly allophones of
the same phoneme (/i/), but they drifted apart in all of Romance.
> Then, if they are in complementary distribution with /i/ and /u/ (which
they indeed are to an astoundingly high degree), what should be the "names"
of the phonemes combining the alternants?
For reasons given above, my preference would be for /i/ and /u/, _if_ the
distribution were fully complementary (see below).
> The point is that, apart from the odd cases of unpredictable
syllabification, there is only one phoneme in this language which *must* be
given the status of a vowel, and that is /a/. Is that not true?
Cases of unpredictable syllabification are telling even if they are rare.
They demonstrate the contrastive potential of segments. A phonemic
distinction is real enough even if its "functional load" is very low.
The analysis of long vowels and diphthongs may be problematic even for
English. I know of several rather different approaches to the English vowel
system, and each of them purports to capture the intuitions of native
speakers. Leaving that aside and concentrating on short vowels, I'd still
argue that the short-vowel inventory of Sanskrit is of a common triangular
type (/a, i, u/). I can't believe that any phonetician would contest that.