Re: BBC does it again

Nathan Sanders wrote:
In article <6bcsnhF3akn8nU1@xxxxxxxxxxxxxxxxxx>,
Harlan Messinger <hmessinger.removethis@xxxxxxxxxxx> wrote:

Nathan Sanders wrote:
In article <6bagajF38s8b5U1@xxxxxxxxxxxxxxxxxx>,
Harlan Messinger <hmessinger.removethis@xxxxxxxxxxx> wrote:

Brian M. Scott wrote:
On Wed, 11 Jun 2008 12:54:34 -0400, Harlan Messinger
<hmessinger.removethis@xxxxxxxxxxx> wrote in
<news:6baea9F39n29oU1@xxxxxxxxxxxxxxxxxx> in sci.lang:

While at the same time obscuring something that's
important at the level of the abstract phonology.
<shrug> I don't find that it does so.
You don't see a difference between (a) indicating that the two phonemes are treated as a short/long pair and (b) not so indicating?
I don't think Brian is questioning whether there's a difference. I think he's questioning whether that difference is important.

As a phonologist (am I the only one here?) with functionalist/phonetic leanings, I prefer more surface-y phonemic representations. Using /eI/ and /E/ for English "long e" and "short e" is, to me, more meaningful, intuitive, and directly relatable to the reality of the phonetics. I just don't see the real use in playing mental gymnastics with yourself and your colleagues to jump over arbitrary, self-imposed hurdles by restricting yourself to arbitrarily a-phonetic abstractions like /e: e/ or /e E/.

Underspecification in phonology was a fad that went on far too long, IMO.
I just commented on this in another note, but it's relevant here so I'll reiterate it: One of the advantages of underspecification is that you don't restrict the scope of your phonemic model unnecessarily. If you use /eI/ for English "long E", then you're limiting the utility of your model to the Englishes of speakers who have a diphthong there.

Other Englishes are different languages. Why should different languages be given the same phonemic model?

Why not come up with a different model for each individual speaker? It seems to me that there's a benefit in adopting a representation that covers the largest group possible, and no point in having separate systems that are isomorphic to each other anyway but with different symbols because the phonetic realizations differ from group to group.

You could in theory give French and Spanish the same phonemic model (equivalent to Proto-Romance), and derive their predictable differences by phonological rules that just happen to look like sound changes.

No, Spanish, for example, has nothing to distinguish a /u/ from a /y/.

That's the same thing that you'd be doing by giving different Englishes the same phonemic system with different phonological rules, because the very thing that makes them different English are the different sound changes they have undergone.

Oftentimes the phonetics change *without* breaking the existing phonemic structure. I thought that that's part of what phonemics was about: the existence of an abstract model that remains more stable than the phonetics related to it.

One could argue that dialects to be given the same phonemic model must be mutual intelligible for a single model to be justified, but that isn't a universally accepted definition of "dialect" in linguistics, and there are well-known problems with defining dialects this way (e.g., dialect continua between mutually unintelligible languages).

Underspecification leads to rather bizarre notions. For example, the phonetic place of articulation of /tS/ is fully predictable, since all affricates in English are post-alveolar. All that needs to be specified to distinguish /tS/ from /t/ is affrication (or alternatively, place), leading us to posit /ts/ (or retracted /t_-/) instead, since they don't "needlessly" double-specify the contrast.
I'm not sure I understand that. /ts/ is found in "cats", and we have "cats"/"catch", so how can it be used in place of /tS/?

/ts/ is not a phoneme of English.

I'm don't see what difference that makes. You still need /k&ts/ for "cats" so you can't use it for "catch".

The /tS/ affricate should properly be notated with a tie-bar or superscript S (or /ts/ can be notated with a hyphen /t-s/ to indicate that these are two separate segments, not an affricate).

OK.

Even worse, if you consider radical underspecification (where not only are redundant features not represented, but the default value of a feature is not represented either, since it too is predictable), then you have a lot of abstract wackiness going on.

Consider /t/: it being alveolar is fully predictable, since it isn't /k/ or /p/; it being voiceless is fully predictable, since it isn't /d/; it being oral is fully predictable, since it isn't /n/; etc. So you end up just representing it with some placeholder consonant symbol, since it is the one consonant that gets all of its feature values for free, because they're all algorithmically predictable from the fact that it isn't any of the other consonants.
Well, sure you can play that game for any *one* of the language's phonemes.

If we constrain ourselves with a theory in which predictable information should not be specified in the phonemic model, just because we believe that's a nice thing to do, then this is what we end up with.

Given all the phonemes that a language has EXCEPT one, it isn't predictable what the final one is, so I don't see where you're going with this.

I contend that (a) the fact that this notion leads to apparent absurdity (such as the /tS~ts/ case and the radical underspecification of /t/) is a very good reason to question it, and (b) there is psycholinguistic evidence that shows that the human brain does not in fact elegantly simplify this way, so if we are at all trying to attempt to model reality (rather than playing with abstract mathematical models just for the fun of it), then blind simplification is a bad thing to do.

Are you saying that the brain *doesn't* model these short/long vowel pairs? If true, it's amazing to me that the neat pairing of vowels like this occurs in so many unrelated languages spread throughout the world (Dutch, Hungarian, Arabic, Japanese, etc.).
.

Relevant Pages

  • Re: Transcribing rhotics for ESL
    ... I'm less likely to run into a problem if I teach English to ... Essentially I'm trying to treat multiple languages as a single ... and grouping phonemes across them appropriately. ... English transcriptions can be a source of frustration for students, ...
    (sci.lang)
  • Re: BBC does it again
    ... important at the level of the abstract phonology. ... You don't see a difference between indicating that the two phonemes are treated as a short/long pair and not so indicating? ... Underspecification in phonology was a fad that went on far too long, ... If you use /eI/ for English "long E", then you're limiting the utility of your model to the Englishes of speakers who have a diphthong there. ...
    (sci.lang)
  • Re: A Theory of Light- a unique aspect
    ... languages that many of its vowels are sufficiently indeterminate that they are indistinguishable. ... In any given dialect, there will be sufficient contrasts to distinguish its phonemes But the nature of those contrasts may be different for different dialects, and even from one word to the next. ... It should be obvious that Eagleson has some kind of OCD thing going about English. ...
    (sci.physics)
  • Re: Lost in the mists of time...
    ... >> simply do not introduce concepts like phonemes into our understanding ... >> languages such as English, ... >> phonemes in Hindi or Sanskrit, that are used to teach Hindi or ...
    (sci.lang)
  • Re: Perhaps all of India should adopt one script
    ... suitable for writing all Indian languages? ... of Malayalam, Tamil and English words. ... dialect with a different set of phonemes and a mixture of Malayalam, ... The proportion of such words and the phonemic distances from the script ...
    (sci.lang)