Re: phonemes

[And Rosta <cbmvax!uunet!cuvmb.cc.columbia.edu!ucleaar>]

Lojbab writes:
> And Rosta responds to me:
> >>There is one other facet in this - since Lojban speech is audio-visually
> >>isomorphic, any 'real' sound would also appear in writing.  The buffer
> >>sound, if audible, is not written.  There is no symbol for it - by
> >>definition it is NOT a phoneme of the language.
> >
> >It may not be a lerfu, or the approximate Lojban equivalent of phoneme, but
> >in no phonological theory is there being a graphical symbol for some
> >sound a necessary and sufficient condition for that sound being a phoneme.
> >In traditional phonemic analysis I guess the buffer vowel would tend to
> >be analysed as a phoneme (which happens to be prone to deletion).
>
> My understanding of the definition of a phoneme in a language is that it
> forms a minimal pair with some other phoneme of the language.  The
> buffer is not a Lojban phoneme by definition then, because it ALWAYS has
> an allophone of null, it forms no significant pairs with anything.  Now
> if you want to say that null is inherently a phoneme, because the
> alternative in any position is non-null, go ahead.  But that is the most
> trite of phonological rules.

A minimal pair would be, say, /mlatu/ v. /malatu/.
A phonological account of Lojban must heed the buffer vowel, because (a) it
is not predictable from general laws of phonetics, and (b) it can occur
only at particular points in the phonological string. Mere "alternation
with zero" does not preclude a segment's participation in the phonology:
deletion rules are not uncommon in phonology.
If the buffer vowel really could appear anywhere in the string I would
be prepared to accept your argument.

> I want our phonology description to be something people can understand.
> What linguists come up with as a formula for competence in Lojban is for
> the linguists to decide.  Considering the buffer to be a Lojban vowel
> merely because linguistically it is a sound classified as a vowel is
> counterproductive, because while it makes clusters look like other
> syllables, the Lojban morphology isn't defined by a syllable structure
> per se, but by the patterns of vowels and consonants either separately
> or in clusters as appropriate.  Defining a cluster as /C-C/ where - is
> some phoneme makes it look on paper like it isn't a cluster, and it is
> important to our morphology that clusters be perceived DIFFERENTLY from
> CVC patterns.  You end up adding noise to the description that makes it
> harder to teach the language concealing important patterns under
> unimportant manufactured ones.

There's nothing wrong with simplifying or even distorting the true
picture in an attempt to make learning easier. In this particular case,
it might be preferable to redefine "cluster", possibly using a lojban
word, than have the learner feel that some lojban words are
oppressively difficult to pronounce.

> To say that the buffer is perceived psychologically runs counter to what
> I have read.  I think the more accurate psychological representation is
> that [specific-consonant]+[buffer] is perceived as an allophone of
> [specific-consonant].

The buffer occurs only between consonants. This is not an epiphenomenon of
lingual motion. It is a pattern represented in the mind. If you can
discriminate between buffered and non-buffered dialects, then it's
pretty obvious that you notice the buffer vowel.

> >I argued that hearers try to map any phone onto some phoneme.  So unless
> >the buffer vowel is very very different from the six Lojban vowels it
> >will get mapped onto one of them.  So, if someone speaking English uses
> >[y] (a.k.a.  "u-umlaut"), I will first of all try to map it onto /u:/.
> >So on hearing [byt] I'd guess _boot_ was intended.
>
> There is some level of linguistic noise that doesn't get mapped onto any
> phoneme.  The buffer is an allophone of [noise] %^).

There is a certain degree of noise that plays no part in the phonology.
But the buffer isn't mere noise, else it could surface anywhere.

> From the hearer's
> point of view, the buffer doesn't exist unless consciously looked for
> (which isn't a linguistic function but a metalinguistic or
> paralinguistic one).

Phonology is about more than contrastive segments. It deals with what is
not predictable from the phonetics. For example, you probably pronounce
_letter_ as [leDr] ('D' = tap), or [letr], whereas I might have [let@]
or [le?@]. The [D] and [?] as allos of [t] aren't contrastive, but they're
not accidents of phonetics: the phonology has to explain the allophony.

This is, incidentally, more than linguists playing silly buggers. People
working on speech synthesis have to work with these phonological rules
too, and they have practical ends, not theoretical axes to grind.

> If an English speaker says a word using a [y], I
> will not try to map that sound to any English phoneme because I know it
> isn't an English phoneme.  Unless I have clues from context/familiarity
> with the speaker to know that his/her accent atypically-to-observer-me
> reflects some English phoneme in the form of [y] (as I have come to
> understand in private email with Chris Handley may be true for his [oo]
> of "moon"), I will recognize the sound/word as a foreign borrowing or
> nonce usage.

I don't believe you. You're predicting that you wouldn't understand
either (a) or (b) below. I'm predicting we'd all have much more trouble
with (b).

(a) She went to hear Mozart's Magic [flyt].    (flute)
(b) She believes the Earth is [flyt].          (flat)

> I personally believe that in fluent speech we perceive words that we are
> familiar with as psychological wholes, not as strings of phonemes.  The
> descriptive phonology of Lojban is a teaching tool for non-fluent
> speakers, not a reflection of psychological reality.

Your view is reasonably orthodox. The phoneme is dead. But if words are
perceived as wholes, these wholes are nevertheless analysable and decomposable.

---
And