Questionable phonemes

Traditionally, in Watakassí, the phonemic inventory has been considered
to be underlyingly just 15 consonants:
p   t   k
b   d   g
m   n
f   s
v   z
    l
w     y

With the following syllable structure:
(C)(w,y,l)V(F,n,l,*) - F = any fricative, * = gemination
With the following restrictions:
Labial consonants cannot be followed by w
Dental stops and nasals cannot be followed by w
G cannot be followed by y (due to an early change /gj/ -> /j/)
L cannot be followed by l (except when representing a geminate /l/)
Glides cannot be followed by any consonant
Cyi and Cwu are illegal
All other combinations are legal, including things like mla, lwa, lya,
nla

There's another restriction that comes into play when syllables are
combined.  Geminates cannot be followed by another consonant, in other
words, geminates must be intervocalic.  Thus, combinations like *askka
(obvious, since gemination represents the realization of /*/) and *akkla
are not permitted.  However, one may find *orthographic* combinations
like appya, for example, sukklúsi, feet, but that's pronounced as
suklúsi, it's merely an orthographic convention used by some to
represent su*-klús-i, most write it without gemination, i.e., suklúsi, I
generally follow the latter convention

When /t/, /k/, /s/, and /z/ are followed by /j/ they become [tS], [C],
[S], and [Z].  So far, no problem.  [tS] can be analyzed as underlyingly
/tj/.  And, initially, that seems to be a valid analysis.  *[tSla] is
illegal, which would not make sense if /tS/ were a phoneme, but does if
analyzed as */tjla/.

However, there are some problematic cases.  There is a phone, [dZ].
[dZ] is, historically, derived from 3 sources, the most recent of which
is /dj/ (/dj/ and /dZ/ only recently merged), the other two sources are
[da'tS] and [da'Z].  Here's the problem.  Both /di/ and /dZi/ are legal,
thus /dZ/ cannot be analyzed as /dj/.  In addition, /dZ/ may be geminate
(in which case it's realized as [ddZ], this is the cases derived from
[da'tS]).  Due to its origin, it cannot be followed by /l/, /w/, or
/j/.  Yet, it seems to be a phoneme.  It can be geminated (but only
before stressed vowels, due to its origin in [da'tS]), and it contrasts
with /d/ before /i/ (and */dji/ would be impossible), and there's no
other phoneme that could be analyzed to underlie it.  But, if /dZ/ is
admitted as a phoneme, one could state the restrictions on *[tSla] and
the like as "alveopalatal phonemes cannot be followed by glides or /l/",
altho that wouldn't explain the restriction on [C].  Plus, that rule
seems rather ad-hoc, there's no justification for it.

Second: [S] and [Z] *can* occur as geminates, even before /a/ and /u/,
that is, <assya>, [ASSa] is a leagal cluster.  This originates from
asimilatory phenomena, [stS], [sC] -> [S:], [zdZ], [zS], [sdZ] -> [Z:].
However, [tS] cannot.  Thus, [S], [Z], and [dZ] can occur as geminates,
but not [C] or [tS] (except before /i/, when it's analyzed as underlying
/kki/ and /tti/ anyways, legal combinations).  Thus, if [S] and [Z] are
analyzed as underlyingly /sj/ and /zj/, then how can [ASSa] be
explained?  It would have to be analyzed as /assja/, and an explanatin
would have to be found for why /z/ and /s/ can be geminate before /j/,
but not any other consonants.

Third: [C] can occur as a coda.  /i/ is lost after [C] in certain
contexts, for example, in _mánuki_ (I am resting).  It would be tempting
to analyze this simply as an allophonic devoicing rule for /i/, except
that [C] really does act as a coda, mánuki is pronounced ['mAnoC], with
[o], the allophone of /u/ in closed syllables!

It has just occured to me that the second problem could be analyzed as
syllable-final /s/ and /z/ followed by syllable-initial /sj/ and /zj/,
thus no gemination involved on an underlying level (altho on the surface
level appearing to be geminated).  That feels somewhat forced to me, but
I suppose it's possible.  And the first could be viewed as simply a
phoneme /dZ/ with unusual restrictions (has to be, since it can't be
analyzed any other way that I can see).  However, the third problem
remains.  If [C] is underlyingly /kj/, then _mánuki_ would have to be
analyzed as /'manukj/, but only one consonant may exist in the coda
position, and no stop may occupy it.  But if it's a phoneme /C/, then
why can't it be geminate (except before /i/), and why can't it be
followed by /w/ or /l/?  It's palatal, so the restriction on */Cj/ would
be very logical, but why not /Cl/, /Cw/, or /CC/?  And if it's analyzed
as /ki/ (realized as [Ci_0]) in the coda cases, then why the
pronunciations [o] and [e] for /u/ and /o/?

--
Dievas dave dantis; Dievas duos duonos
God gave teeth; God will give bread - Lithuanian proverb
ICQ: 18656696
AIM Screen-Name: NikTailor