Re: THEORY: phonemics (was: RE: [CONLANG] Optimum

Mike:
> John Cowan <jcowan@...> wrote:
>
> >And Rosta scripsit:
> >
> >> > >But that still leaves us with
> >> > >allophonic variation that is not conditioned positionally, which
> >> > >is why I gave the example of English /t/ in foot-internal
> >> > >intervocalic position, which can, inter alia, be [t] or tap [D].
> >
> >On reflection, I don't understand this one at all.  For me, [t] vs
> >flap is completely positional, and I thought the dialects for which
> >this is not so don't have flap at all.
> >
> >> I deliberately chose the [t]/[D] allophony because it can't be
> >> defined by underspecification: whereas the final /p/ allophony can be
> >> defined by not specifying relase or aspiration, the intervocalic /t/
> >> allophony can be defined only extensionally, as the list {[t], [D]}.
> >
> >They are both alveolars, though, and the idea of a category that
> >unites stop and flap (which can be thought of as a stop of minimal
> >length) is not absurd.
>
> I also considered this.  In addition, it might be argued that
> the style and register that And mentioned that result in an
> allophony truly defined by extension are actually signifying two
> different languages, and thus it might actually be possible to
> define allophony by underspecification after all.  I didn't
> initially push this because it seems to rely a bit too much on
> hairsplitting.  Moreover, in order to confirm the idea of
> universal underspecified allophony, it seems to me that a broad
> and very detailed survey of languages would need to be undertaken,
> and all that would be required to disprove the whole idea is one
> counterexample.  I did not want to claim that such a counterexample
> does not exist, thus my conservative demeanor.  Also, an arguably
> bigger problem is explaining why it is that a speaker who does
> not use the aforementioned styles and registers might have little
> difficulty *hearing* the allophones.  That leads me to my main
> point.
It is very refreshing to have so conservative an interlocutor in
a debate such as this!

Anyway, in the message where I replied to John, I mentioned vowel
allophony as perhaps less amenable to the underspecificational
approach. E.g. the GOAT, GATE, LOUT, LIGHT diphthongs in British
accents, where there is so much variation in many regions that
it is hard to define a canonical target realization.

> For what it's worth, it just dawned on me that there might well
> be two ways to define a phoneme: productively and perceptually.
> The productive definition of phoneme would entail the normal
> allophones used when uttering the phoneme; the perceptual
> definition would define the allophones that will be heard
> by the same speaker or speech community as belonging to a phoneme.
> The productive set will always be a subset of the perceptual set.
> For example, I basically never will render intervocalic /t/ as [?],
> but that presents no difficulty for me in hearing [?] as /t/ in
> that position.  If we accept the broader perceptual definition
> of phoneme as being basic, then the stance that in some cases
> allophony can be defined only extensionally seems a bit stronger.
I would have thought that both production and perception could
be adequately modelled (in general, not necessarily exceptionlessly)
by defining one or more prototypical allophones, so that the degree
to which a given phone counts as an instance of a given phoneme
is proportional to how near it is to the prototype allophone.

--And.