🚀 go-pugleaf

In article <1993Jan27.183540.15379@midway.uchicago.edu>, goer@kimbark.uchicago.edu (Richard L. Goerwitz) writes:
> connolly@memstvx1.memst.edu writes:
> 
>>> What is a phoneme?
>>
>>I'll go way out on a limb and make most of the readership mad at me
>>with this definition:
>>
>>	A phoneme is the minimal psychological construct representing
>>	certain sounds which occur in a given language.
> 
> This is intuitively correct.  In fact, I don't think that Chomsky him-
> self is really all that concerned about phonemes.  It was Halle that
> got into all the matrices.  I also don't believe that either denied the
> existence of systematic units called phonemes.  They just made them far
> more abstract, and treated them as bundles of features.  The term for
> their kind of phoneme is "systematic phoneme."

I don't have their book at hand, but it is organized into several sections
that, if I remember correctly, are often simply incompatible.

That having been said, I remember that they indeed talk of systematic
phonemes and even introduce a notation: |x| is a "systematic" phoneme
while /x/ is a conventional one.  By pointing out some interesting
problems with the formulation and application of phonological rules,
they show (to their satisfaction) that conventional phonemes are
superfluous, at which point Ockham decided to shave.  The trouble is
that the same arguments apply to systematic phonemes, which in fact
they hardly use.  And the systematic phonemes are *not* tantamount
to feature matrices, since the actual matrices they posit for
words such as _spit_ do not correrspond to *any* sort of phoneme.
Since English words do not begin with consonant + voiced stop,
the matrix for the [p] is unspecified for the feature [voice].  So
the actual matrix would have to correspond to that monstrosity, the
archiphoneme, rather than to any phoneme at any level.  And worse,
since the only consonant allowed ahead of a stop in English is [s],
the matrix for this consonant is simply [+cons].  What sort of phoneme
is that?  Not even one of theirs!  And unless we want to posit an
"arch-archiphoneme" (quod Deus avertat), we have no sort of unit
at all, just a very imperfectly specified matrix.

> The only criticism I would level at your definition is that it is
> vague. 

Conceded.

> Can you offer us a discovery procedure like the descriptivists
> and structuralists had?

The structuralist procedures are pretty good.  It's the constructs
that are wrong because of their hide-bound premise that a given
phone can correspond to only only phoneme, not two or more.  Alterna-
tions such as _relate : relation_ can be blamed on "morphophonemics",
but that's not an explanation: it's an attempt to avoid saying that
some instances of [s^] are /s^/ while others are /ty/.  German _Rad_
'wheel' was analyzed as /ra:t/, while the genitive _Rades_ was
/ra:d@s/.  Heck, the structuralists' procedures could show perfectly
well that /t/ and /d/ were different phonemes; they simply couldn't
bring themselves to analyze _Rad_ as /ra:d/, contrasting with _Rat_
'advice' (genitive _Rates_), which was obviously /ra:t ra:tes/ (no
need for /@/ in German if you play your cards right).

>  Can you offer us a cognitive model and a for-
> malism, like what generative phonologists have?

A cognitive model is asking a bit much, and anything offered by
generative phonologists or anyone else is partial.  I am saying
that phonemes are represented as indivisible units (not bundles
of features), and that these units do not always correspond to units
in speech on a 1:1 basis.  We linguists can and do break these
units down into features and write very plausible phonetic rules
which are difficult even to express any other way.  This may or
may not be a good thing, and I don't know if such rules correspond
to any mental processes.  It might be better merely to note the
correspondces between the abstract units and their varying realization
in speech.

>  Simply calling a phoneme
> a "psychological construct representing certain sounds" doesn't move us
> off of square one.

No, but it brings us there if we've been off chasing ghosts.  Seriously:
setting a plausible starting point is worth something if we haven't been
able to agree where to start.

> Tell you what.  I'll believe that your definition is clear if you can
> take a standard data set, and show how your definition of the phoneme
> offers us a superior way of describing some synchronic phonological pro-
> cess.

I think _spit_ is a least a standard datum.  Does what I said above about
it clarify some of the issues?  Isn't it better than messing with archi-
phonemes or with segments represented merely as [+cons]?  Haven't I
put my phonemes in the right place?  Let's see what others have to say
before I elaborate any further.

--Leo Connolly