LINGUIST List 2.131

Wednesday, 11 Apr 1991

Disc: (Morpho)phonology

Editor for this issue: <>

Directory

1. John Coleman, (Morpho)phonology
2. Mike Hammond, (Morpho)phonology
3. "Bruce E. Nevin", Halle's classic argument

I said before:
>Whether a phenomenon such as this (the second syllables of `title/titular')
>is morphophonological or just an
>automatic alternation depends on other considerations, especially the
>form of morphophonological representations and the theory of phonetic 
>interpretation which is being assumed. Wojcik's comment seems to me to 
>assume that both of these are settled

Rick Wojcik responds:
>Vowel reduction
>usually refers to a purely phonological phenomenon--what happens when speakers
>try to pronounce vowels in unstressed environments. Purely phonological
>processes such as vowel reduction participate in phonological foreign accent.
>English speakers apply vowel reduction to foreign languages as well as
>their own. The 'title/titular' alternation is of a different nature. It does
>not bear on how we pronounce sounds and plays no concrete role in the
>mispronunciation of foreign words. 

If `title' and `titular' are analysed as, say, /tay.tl/ and /ti.tjUl.a/ there
is an apparently non-physiophonetic alternation between the second syllables:
/l/ vs. /tjUl/. If `title' and `titular' are analysed as /tayt.l/ and
/tic.l./ or /tayt.Vl/ and /tic.Vl.V/, say (c = voiceless alveopalatal 
affricate,  = `schwa' and V = underspecified vowel), then there is no 
phonological distinction between the second syllables (/l/ or /Vl/)
and the quality of the nucleus in both cases might be attributed to 
physiophonetic vowel reduction. If the nucleus of these syllables is 
analysed as combination of [+high], [+back] and [-rnd] then in 
`title' the nucleus and coda occur simultaneously with each other and
with the features [+high, +back, -rnd] in a lump, whereas in `titular' 
the nucleus and coda are not completely simultaneous, and [+high]
spills into the onset. So the distinction between them is again not
psychophonetic but physiophonetic --- a distinction in the timing of 
the phonetic exponents of a single phonological representation.

Here then are three possible analyses of the second syllable opposition in
`title'/`titular'. They each make different claims about whether the
opposition is `merely physiophonetic' e.g. `vowel reduction', or whether
it is cognitively encoded at a more abstract level. Wojcik's claim that
>The 'title/titular' alternation ... does
>not bear on how we pronounce sounds 
rests either on some kind of phonological omnipotence that does not 
require us to dirty our hands with the consideration of actual
analyses of the data, or it is just a theory-internal assertion.

--- John Coleman

It was recently noted by Alexis Manaster-Ramer that "it is [not]
enough to say that orchestra and podagra get initial stress because
the second syllable is open. This is because words with such clusters
between the second and third syllable do not ALWAYS behave this way.
Thus, canasta, Modesto, and so on have second-syllable stress."

I agree completely. Words like Canasta and Modesto show that words
with light penults can exceptionally get penultimate stress. (Cf.
vanilla, Kentucky, etc.) This is in sharp contrast with the behavior
of words with heavy penults. They normally get penultimate stress and
cannot be marked for exceptional antepenultimate stress.
(Incidentally, this follows automatically on the theory of exceptional
stress I pushed in a paper in _Phonology_6.1_.)

Alexis goes on to question the analysis of diabetes as /diabete+s/ to
account for flapping before a nonfinal and nonprevocalic [i]. "Any
theory of English stress which requires diabetes to be analyzed as a
morphological plural strikes me as unacceptable. Are we going to say
the same in the case of Hades and Ulysses?"

If Hades is pronounced [heDiz] then we're committed to a similar
analysis: /hadi+s/. (Frankly, I have no clear intuition about whether
that word is plural or singular.) Since there is no flap in Ulysses,
there is no reason why it can't be treated as having final secondary
stress to account for the full vowel. (Interestingly, I think it'd be
pretty difficult to argue for a plural personal name, hence a
prediction might be made that nothing phonologically like Hades could
be a personal name.)

Alexis also questions the SPE analysis of words like industry: "Is
there any plausibility of the SPE analysis of industry as having a
final [y] rather than a final [I] or [i]? I really doubt that."

I don't really understand what's dubious about it.

mike hammond

I took some time last night after getting the kids to bed to check some
things and try to respond a bit more fully to Rich Wojcik
<rwojcikatc.boeing.com>. I hope this is coherent, as I haven't much
time to edit it this morning.

You and others have referred to "Halle's classic argument against the
phoneme." Since what I had already written was not taken as germane to
this, I assumed that you and others must mean something other than I
remembered. However, all I can turn up is indeed just what I
remembered. Please give me the citation you have in mind if I have this
wrong. I am here today and will be back again on the 24th.

(I depend on Anderson's 1985 rendition, since I don't have access to SPR
at home or in my office and time is short.)

In _The Sound Pattern of Russian_, Halle listed (p. 19) "six formal
conditions which phonological descriptions must satisfy." The third of
these, in two parts, is essentially a restatement of the requirement of
biuniqueness, merged with a strong form of Bloch's (and Halle's)
requirement of phonetic identity.

Anderson in this section identifies biuniqueness with just one part of
Halle's condition 3, reflecting the now conventional understanding of
the term. A brief review of the history of the concept is in order.

Chao (Non-uniqueness, 1934) had distinguished the "reading aspect"
(phonemic to phonetic) from the "writing aspect" (phonetic to phonemic)
of the correspondence between phones and phonemes. The reading aspect
is straightforward, he says; the writing aspect is problematic and not
always attained.

Harris introduced the term "bi-uniqueness" (his hyphen) in the "Long
Components" article. (Anderson 1985 misattributes this to the review of
Newman's Yokuts, also 1944. He apparently took the citation of Harris's
coinage from Hymes & Fought, whose typo, 1944a in place of 1944b, he
seems to have copied without verifying. The page range in Hymes &
Fought is on the money for the occurrence of the term in the long
components paper, and out of range for the Yokuts. I have not verified
whether or not the term occurs there as well. Anderson gives no page
reference.) It means that for each sequence of phones there corresponds
a unique sequence of phonemes, and for each sequence of phonemes there
corresponds a unique sequence of phones. Later, Harris (1951, written
in the second half of the 1940s) drops the term biuniqueness and uses
"one-one correspondence" (required for phonemes) vs. "one-many
correspondence" (characteristic of morphophonemes but disallowed for
phonemes).

Halle's restatement of the "writing aspect" in his condition 3a (p. 21
as quoted by Anderson) is stronger than a 1-1 correspondence. It is a
requirement that one must be able to

 infer . . . the proper phonological representation of
 any speech event, without recourse to any information
 not contained in the physical signal.

This adds to the "writing aspect" of a 1-1 correspondence an additional
empiricist requirement of "discoverability" from the phonetic record
(the "physical signal").

By a curious kind of augmented synecdoche, the "writing aspect" of
biuniqueness, augmented by this discoverability requirement, is what is
now known to every student of generative phonology as "biuniqueness, the
cornerstone of structuralist phonemics." It is a caricature even of what
Bloch claimed in his Postulates or his later article on contrast, for
there at least distributional facts and (albeit uncomfortably for Bloch)
junctural entities were relevant, and he only required that all
allophones share some characteristic phonetic feature(s), something, as
I have said, that Harris did not require.

It is this strange artefact that is the cornerstone of Halle's argument
from neutralization of voicing in Russian on the next two pages. The
affricates [c], [C] (using uppercase for c-hacek) and velar spirant [x]
have no voiced counterpart, but become voiced before a voiced obstruent.
The other consonants do have voiced counterparts.

 In a phonological representation which satisfies both
 condition (3) [the reading aspect] and (3a) [the
 augmented writing aspect], the quoted utterances would
 be symbolized as follows: /m'ok l,i/, /m'og bI/,
 /Z'eC l,i/, /Z'eC bI/. Moreover, a rule would be
 required stating that obstruents lacking voiced
 cognates--i.e., /c/, /C/ and /x/--are voiced in
 position before voiced obstruents. Since this,
 however, is true of all obstruents, the net effect of
 the attempt to meet both condition (3) and (3a) would
 be a splitting up of the obstruents into two classes
 and the addition of a special rule. If condition (3a)
 is dropped, the four utterances would be symbolized as
 follows: {m'ok l,i}, {m'ok bi}, {Z'eC l,i}, {Z'eC bi},
 and the above rule could be generalized to cover all
 obstruents, instead of only {C}, {c} and {x}. It is
 evident that condition (3a) involves a significant
 increase in the complexity of the representation.

(As quoted in Anderson (1985:320), with a comma added. I have
substituted uppercase C for c-hacek, Z for z-hacek, and I for barred i.)

One question that Halle is broaching here in hindsight, it seems to me,
is how abstract may one's underlying representation be? For Bloch, for
Trager & Smith, and for Halle's condition 3a, not very. For Bloomfield
and Harris, the UR could be pretty abstract if that made for a cleaner
description. (Both have unreduced vowels underlying schwa in the
phonetic record, for example, which Bloch and Hockett could not allow.)
Alternatively, how much of the language structure antecedent to the most
recent rounds of merger are still in some sense alive and productive in
the language, and therefore legitimized as abstract entities in UR? For
Sapir, the phonemic representation could be pretty abstract and
etymological if the abstract terms and relations involved were still
lively in the language, that is, a determinant of informants'
perceptions. Needless to say, the abstract/"natural" dispute has not
gone away in the contemporary literature on phonology. (It is
interesting that the familiar writer/rider example is brought to bear
against Hooper [Bybee] in Anderson's "Not `Natural'" paper, suggesting
that NGP may be the heir apparent to the taxonomic tar and feathers.)

In structuralist terms, a conservative ("natural") point of view like
Bloch's would say that, in the present time-slice of Russian, voicing is
allophonic for /c C x/, but morphophonemic for the other phonemes, and
this may be awkward but that's the way languages really truly are and we
just have to live with it. This is what Bloch did with Japanese phonemics,
for example, and it has a special relevance to his interests as a
consummate dialectologist.

As Harris is not bound by Halle's condition 3a, he is not caught in the
dilemma that it sets up. He might concur: "If condition (3a) is
dropped, the four utterances would be symbolized as follows: {m'ok l,i},
{m'ok bi}, {Z'eC l,i}, {Z'eC bi}, and the above rule could be
generalized to cover all obstruents, instead of only {C}, {c} and {x}."
For Harris, a considerable range is available of other options that are
not available to Halle, such as a long component of voicing extending
leftward from voiced obstruents. (Johns, reprinted in Makkai, suggests
a horizontal bundling of features. As Fought points out in Hymes &
Fought, this is really a proposal of a long component solution for
writer/rider which Johns does not recognize as such.)

In the background of Halle's argument is the now tarnished rule-counting
metric for adjudicating alternative grammars, and versions of learning
theory that have lost credibility. (So far as I know, there is no
evidence that language users economize tightly on their neurological
real estate.)

An important metric of the "goodness" of a grammar for Harris is lack of
restriction on combinability of elements. This is because his interest
is in studying language as a mathematical object. One consequence of
this program is that the restrictions that remain have a semantic
(informational) interpretation. But assuredly not all linguists are
interested in a version of linguistics that is a branch of applied
mathematics. (A concise and accessible overview is in the 1989 book
_Language and Information_.) I think Harris would like plurilinear
representations in phonology very much, in part as a representation for
his long components, in part because it neatly partitions different
domains of contrast. For him, it is the contrasts that are the
underlying reality, and all the rest--phonemes, morphophonemes, long
components, and rules--are representations. In this, I believe he
integrated the often counterposed perspectives of his two teachers,
Sapir and Bloomfield. But the psychological validity or non-validity of
anything other than the structure of information in a subfield of a
science (the 1990 book), to which phonology is irrelevant, has yet to be
demonstrated.

Again, please tell me if you had some other "classic argument against
the phoneme" in mind. 

 Bruce
 bnbbn.com