Re: Subtitutes for English /T/ and /D/

In article <ifu6a313pakf0p0coak2s0t19bhrmnaha3@xxxxxxx>,
Ruud Harmsen <realemailonsite@xxxxxxxxxxxxxxxxxx> wrote:

Sun, 22 Jul 2007 10:24:23 -0400: Nathan Sanders
<nsanders@xxxxxxxxxxxx>: in sci.lang:

Sometimes Bob will simply never produce that vowel, in which case, the
syncope rule has applied perfectly.

Any case in which the vowel did not delete is a case in which the rule
failed to apply perfectly.

There's a slippery line, or a thin line, or a minefield, I don't know
what exactly, between "never produce" and "did not delete".

I can quite confidently state that I "never" (see below) produce the
original vowel in the English plural of words like "wives".

For
examples, listen to Portuguese Portugese for elided vowels that are
nevertheless perhaps all fully there. Or hardly, hardly audible, or
not ar all, or only perveicable for native speaker or others who have
learnt to understand the language.

They are there, or they are not. MRIs and spectrograms do not lie.
If the speakers are not articulating a vowel, then it isn't there.
(If they are articulating it, but it is inaudible, then clearly for
them, the vowel is there, but for listeners---and most importantly, L1
learners acquiring the language from this speaker---it is not.)

I repeat: the basic misunderstanding here is to think that this is a
binary matter, a yes or no, logical zero or logical one kind of issue.
In reality, it simply isn't. Denying that makes phonology a university
model that has no bearing on real languages.

Of course there is random variation in all aspects of speech, because
humans are not perfectly tuned instruments (indeed, there is
essentially no physical phenomenon in the universe that is truly
perfectly behaved). This random variation is noise in the data, and
when some phenomenon of interest is indistinguishable from the noise,
the phenomenon cannot be claimed to exist.

For example, it is natural for the amount of voicing in a sound to
vary. If you were to measure my pronunciations of words ending in
voiced stops, you would likely find some instances that in fact have
no voicing at all. But as long as this is within the range of
expected variation, it would be pointless to claim that I have an
active rule of final devoicing (otherwise, we would have to claim that
all speakers of all languages have such a rule, which is a rather
extreme version of Universal Grammar!).

Where we draw the line between noise and not-noise is a matter of
debate, of course, but that's best left to statisticians to work out;
linguists don't need to invent their own pet theory of statistical
significance, and we shouldn't have to define it every time we say
"never" or "always".

Sure. There is no one true phonemic analysis of any given language.

But a phonemic distinction is phonemic or it is not, with nothing in
between. I detect a contradiction here.

There is no contradiction. [i] and [I] are in contrastive
distribution in English, so they clearly belong to different phonemes.
Whether those phonemes are /iy i/ or /i I/ or /i: i/ is not absolute.
Any of those phonemecizations are valid.

Phonemes are part of a linguist's model of language, and according to
how phonemes are standardly used, something either is or is not a
phoneme. All this other stuff with variability and marginality is
modeled somewhere else (usually in the phonological rules).

Difficult. Academic. Not practical. Hard to accept.

If you find the concept of phonological rules difficult, then I
suggest taking an introductory phonology course.

If your issue is instead with variability in grammar, I can point you
to some work that's been done on stochastic grammars (but it's all
within the framework of Optimality Theory, which you may not be
familiar with). See also work in variationist sociolinguistics for a
different take.

Yes. And because there is no fixed point in time when it reaches
exactly 100%, there will be an intermediate stage in which both the
"two phonemes" and the "one phoneme" description has some validity.

As long as the speaker shows variability in their speech, /

That is: always.

Only in the uninteresting sense in which all physical phenomena show
variability.

In the interesting sense of statistically significant variability, it
most certainly is not "always".

an analysis
which hard-encodes the sound change into the phoneme inventory is not
correct.

OK, so any proposed phoneme inventory is incorrect.

I don't see how that follows. If a sound change is in progress, it
makes no sense to hard-code it into the phoneme system, since speakers
are obviously still working with the original, pre-sound change system.

Once the sound change has run its course (as the original intervocalic
voicing and syncope rules of English have done), then hard-coding is
possible (depending on the actual facts, of course).

Phonology unmasked. Nice for making models for the sake of it, but
unusable for real-life languages?

Simplifications have to made in every science. Just look at classical
mechanics. They have simplifications of all sorts, from approximating
complex shapes as simpler ones, to assuming massless springs and
frictionless surfaces to derive certain results. And everything from
temperature to minor air currents to quantum-level fluctuations in
mass are routinely ignored for ordinary purposes in classical
mechanics.

I wouldn't go that far myself, but
it seems to be the logical consequence of what you wrote.

It's a necessary fact about science. You simply cannot (and should
not try to) model every single minute variation that can possibly
arise in your data. Some things are just statistically insignificant
noise (and some things are too complex and too minor to worry about
until larger, more basic questions have been answered).

(assuming of course that
this was a full, unconditioned merger, leaving no telltale remnants of
the former pre-merge contrast).

Right, and in real life, that never happens.

Nonsense! There are plenty of mergers that leave no trace.

OK, but there is always that pesky long period in which the traces are
still there, and when exactly that period ends is undeterminable.

A given speaker either has those traces or they do not. If they do
not, then they did not acquire them, and thus, they mark the starting
point of the true merger (for their direct line of acquisition, at
least). Of course, determining this is difficult, since we can't
monitor someone's speech continuously for their entire life. We have
to carefully extract representative samples and extrapolate.

And for a given population, this is even more difficult. Not only are
we looking at sample utterances from individuals, we're also looking
at sample individuals from the population. The statistical
computation grows more complex, the data are necessarily more noisy,
and the extrapolation will necessarily be more distant from individual
occurrences in reality. That doesn't mean we can't create the model,
or that the model isn't valid.

And certainly, just because we say that our data are statistically
consistent with a model in which a particular phenomenon "never
occurs", that doesn't mean that we couldn't ever actually find some
speakers somewhere who do in fact exhibit this phenomena. But it's
generally understood by those familiar with the scientific study of
real life phenomena that such variation is just a part of the way the
universe works and doesn't need to be mentioned, accounted for, or
explained every single time we talk about some particular issue,
especially in informal settings like Usenet.

Nathan

--
Nathan Sanders
Linguistics Program
Williams College
http://wso.williams.edu/~nsanders/
.

Relevant Pages

  • Re: Phonemes
    ... >> the same phoneme as that used in the word by other speakers who ... >> produce it as a back vowel, and which I perceive as the same phoneme. ... then you are not perceiving the two vowels to be the ...
    (alt.usage.english)
  • Re: why cant the BBC
    ... and must not appear in a phonemic transcription. ... short vowels, normal practice is for the long vowel phonemes are ... rejects the level and entity of the phoneme entirely (and, IMHO, ... Possibly relevant distinctive features for them are long/short, ...
    (sci.lang)
  • Re: The "u" and "v" in older written English is confvsing
    ... systematic replacement of all realizations of the phoneme. ... explain that speakers who produce the uvular r don't produce the alveolar and vice versa, at least not without a conscious switch of register or dialect. ... I've heard German speakers of both types, in abundance, and French speakers of both types, not in the same abundance, but I've never heard any speaker of neither French nor German mix the two articulations. ... the fact that the r is uvular word-initially and alveolar/apical elsewhere would rather suggest that the uvular r is the new allophone. ...
    (sci.lang)
  • Re: Literary phonetic alphabet
    ... Furry has the NURSE vowel, which we've so far avoided. ... Are you claiming that they have the Mary phoneme (/eh / in T-S), and not the marry phoneme, either in your dialect or more generally in non-mergerers? ...
    (sci.lang)
  • Re: country names
    ... You say "Furthermore, if a phoneme has multiple different allophones, ... which one is actually recorded in UPSID?". ... in UPSID either) is vowel frequency. ...
    (sci.lang)