Further on "silent" phonemes

[Hans-Werner Hatting]

On Wed, 26 Apr 2000 11:51:55 GMT Anthony Appleyard wrote:

[ moderator snip ]

>An example of such a "silent phoneme" that some would invent, in a more
>familiar language, is the French "h aspire'" that prevents liaison in some
>French words, e.g. "le haricot" {l#ariko}, "les haricots2 {leariko}, where by
>the above analogy some would write {ariko} as {=ariko}. That this French
>so-called `=' phoneme is derived from a pronounced {h} sound, is merely old
>history (except in Normandy, where this {h} sound persists, or so I read
>once.) Likewise in standard moderm French, final closed {e} as in "je donnai",
>and final open {e} as in "je donnais", are now separate phonemes, whereas they
>were once likely allophones according to whether or not they were followed by
>a now-vanished final consonant.

I would like to suggest another solution to the problem outlined above. I
strongly agree with the opinion of Gabor Sandi, posted in this discussion,
on that for establishing whether two sounds are phonemes or allophones, only
phonetic conditioning of their distribution (as against, e.g., semantic
conditioning - the fact that initial /dh/ is restricted to pronouns in
English is no argument against its phoneme status, and that the /th/-/dh/
distinction serves to distinguish nouns and verbs is, in my eyes, a strong
argument in favour of these being diffrerent phonemes) ought to serve as a
basis. The same also ought to apply to establishing phonemes: we should not
establish a phoneme solely on the basis of one phone triggering the use of
different forms of affixes. Let4s consider the case of the French article
first:

The rule is that various French words and morphemes have different forms
depending on whether they are followed by a word beginning with a vowel or
with a consonant (there are additional syntactical conditions for the choice
between these alloforms, but I think I can leave this aside here for
simplicity4s sake). Now a group of words, synchronically beginning in a
vowel, triggers the alloform used before consonant. (These are mostly
non-Romance loanwords which historically began with an /h/ no longer
pronounced.)To keep the rule neat, one could of course elevate the "h
aspiri" to phoneme status and posit a consonant phoneme /=/ which
phonetically is not distinct to Zero, but triggers the pre- consonant
alloform of preceding words. But this would mean that we have two phonemes
with exactly the same realisation! So wouldn4t it make more sense to say
that the choices between the two alloforms are triggered partly by phonetic
conditions, and partly bv non-phonetic ones (i.e., there is a class A of
words beginning in vowels which triggers alloform 1, and a class B which
triggers alloform 2). This may be not as neat, but if we start to establish
phonemes out of phonetically non-distinct sounds just because they trigger
the choice of different allophones, where to stop? Why not ascribe to French
two phonemes /s1/ and /s2/, one of which triggers a change from "al" to
"aux" in plural formation and the other one doesn4t? One can do so, of
course, but I think this is mixing up description levels and leads to
confusion, not clarity.

I4m not acquainted with Numic, so what I say now is based on John
McLaughlin4s posting in this discussion:

"Here's some very basic data to show that /=/ hasa phonemic status.1) Start
with these noun stems which are representative of the entire bodyof nominal
stems: [waa] 'cedar', [pyjy] 'duck', and [tyhyja] 'deer'2) Now add the
postposition /-pa/ 'on' to each of them: [waahpa] 'on thecedar', [pyjypa]
'on the duck', and [tyhyjava] 'on the deer'. Notice howthe phonetic
realization is different for each of these (remember that eachof these words
represents a class of nouns that operate exactly the sameway).3) Now add the
postposition /-tu/ 'through' to each of them: [waahtu]'through the cedar',
[pyjytu] 'through the duck', and [tyhyjaru] 'throughthe deer'. Notice how
the initial consonants of each of these suffixes changes in the same ways on
the same stems.4) Now incorporate each of these nouns on the verbal stem
/-pa'i/ 'have':[waahpa'i] 'have a cedar', [pyjypa'i] 'have a duck', and
[tyhyjava'i] 'havea deer'5) Now compound each of these nouns with the
nominal /puku/ 'pet':[waahpuku] 'pet cedar' (think bonzai), [pyjypuku] 'pet
duck', and[tyhyjavuku] 'pet deer'By now you should realize that this is not
some feature of the secondelement, but a feature of the stem that causes the
initial consonant of thesecond element to be preaspirated, nonlenited, and
lenited. Unlike thevoicing of /th/ to [dh], it is fully productive in (at
least preobsolescent)Comanche. There is something following each of these
nominal stems which is neutralized in word final position. From Shoshoni
evidence, we know thatthese "final features" are -C (an undifferentiated
consonant that causesgemination in Shoshoni and preaspiration in Comanche),
-n or -= (prenasalization in Shoshoni, a nonlenited stop in Comanche), and
zero(allows lenition in both Shoshoni and Comanche).Now this does bring up
an important point that I'm sure you'll agree with.There is not a clear
boundary line that demarcates when a phoneme has splitor when morphophonemic
distinctions have ceased productivity or when anynumber of changes have
finally and irreversibly taken place. Comanche is avery clear borderline
case. The phonemic status of = (Shoshoni /n/) is notcompletely black or
white."

My impression is that Comanche is similar to the case of French "h aspiri" -
historically distinct phonemes have merged into /0/, and so a rule for
choosing allomorphs based on phonetic conditions has been replaced by a
system of words belonging to three declensional classes, which are
distinguished by which allomorphs they trigger. Of course, if there are
cases where the phonemes -C- and -n- postulated by John McLaughlin are still
visible synchronically in stem-final position(maybe before morphemes
beginning in a vowel?), then one could take these as the basic stem forms
and state the rules for choosing the allomorphs based on phonetical
conditions. But then we probably would not need a phoneme /=/ (historically
< C).

Of course, a "phoneme" /=/ looks like a good shortcut, as it succinctly
conveys information ("triggers the follwing set of allomorphs"), especially
for a lexical entry. It4s comparable to dictionaries of French
distinguishing "h aspiri" from "h muet" by asterisking it. But if we want to
keep the levels of description distinct, and to keep synchronical
descriptions separate from historical explanation, we should not postulate
phonemes in order to formulate rules for the choice of allomorphs. Better to
work with declensional classes, and then add a historical explanation of
their origins.

Best regards,
Hans-Werner Hatting