LINGUIST List 2.172

Saturday, 27 Apr 1991

Disc: Phonology (Part 2)

Editor for this issue: <>


  1. "Bruce E. Nevin", "American structuralists all agreed"

Message 1: "American structuralists all agreed"

Date: Fri, 26 Apr 91 09:13:00 EDT
From: "Bruce E. Nevin" <>
Subject: "American structuralists all agreed"
> From:
> Ad Wojcik: It is important to remember that Ulaszyn, the Leningrad
> phonologists, and the American structuralists of the 1930's (but not
> earlier!!) all agreed in placing the phonemic level exactly where
> Halle found it--and found it wanting, that is, AFTER all neutralizing
> rules and before all allophonic ones. Halle's argument, of course,
> boils down to saying that automatic neutralization rules and
> allophonic rules are, invariably, one and the same, and that this is
> what is wrong with this level of phonological representation.

Yes, it is true that for American structuralists morphophonemic
alternations are "above" the level of the phoneme and allophonic
alternations "below" it, by definition. However, they did not always
agree as to the status (morphophonemic or allophonic) of a given
alternation. The more usual way of putting this is that they did not
agree as to the definition of the phoneme, and depending on the
definition certain alternations might be allophonic or morphophonemic.
A general aim was to handle an alternation in the phonology (without
lexical or morphological specification) if one could, but some
phonemicists were more constrained by their definition of the phoneme
than others were by theirs.

This is relevant because Halle's argument applies only to some versions
of American structuralist phonemics. On some definitions of the phoneme
found in American structuralism Halle's argument evaporates.

I mentioned previously that for Bloch phonetic likeness (the "invariance
condition") was a requirement in defining the phoneme, and contrast was
a secondary term term, derived from phonetic identity plus distribution.
Harris took contrast as the primitive term (the pair test), and in
consequence was able to deal with both phonetic features and
distribution in a more abstract way than could "natural" phonemicists
like Bloch.

 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.

There is a functional aspect to the distinction between morphophonemic
alternations and allophonic alternations that should be mentioned.
(Ferguson made this point in his review of Halle's SPR.) A Russian
hearing for the first time a neologism or a loan word, say [krog]
(assuming that is a nonsense form), and hearing it only where a voiced
obstruent followed, would have no way of knowing whether it were voiced
[krog] or voiceless [krok] in other environments. Hearing a novel
[krodz] (using [dz] for the voiced counterpart of [c]), she would have
no doubt that it was [kroc] in other environments. This significant
fact about the language is represented only indirectly in the standard
generative (Halle's) analysis. "No form of expression creates new
information: the only question is the availability and organization
which it gives to the information" (Harris "Long components," 1944:132
of reprint in Makkai).

I mentioned the possibility of a componential analysis in the matter of
Harris (1944). A component of voicing might be extracted from the
obstruents, here indicated by a superscript line (and taking p/b as
 [p] = /p/ [b] = /p/

The voicing component marks the distinction. (Harris points out the
relation of this use of a unit-length component to Trubetzkoy's
Merkmal.) If there are five points of articulation for stops, this
reduces the phonemic inventory by four (from 10 stops to 5 voiceless
stops plus component of voicing).

Now define the voicing component to extend leftward over a preceding
obstruent. By extracting this long component, we can treat the voicing
alternation in a parallel way for Halle's two forms.

 ___ ___
 m'ok l,I m'ok pI ZeC l,I ZeC pI

The difference between the morphophonemic alternation and the allophonic
alternation is converted to a distributional restriction on the
component of voicing: the affricates and /x/ cooccur with the voicing
component only in its leftward extension from another obstruent, never
in other environments.

If our proposed loan word [krog] turned out to have a voiced stop (no
alternation with voiceless [k] in other environments), the same sort of
componential analysis would look like this:

 _ ___
 krok l,I krok pI

The component has no additional effect over the already-voiced stop.
Biuniqueness is preserved (no ambiguity or loss of information either
way), but without loss of generality.

Orthographically, the voicing component could be represented as a
segment whose definition (rather than its graphic shape and position)
defines its extent. Thus: /m'ok l,I/, /m'ok ~pI/, /ZeC l,I/, /ZeC ~pI/,
/kro~k l,I/, /kro~k ~pI/. This is entirely parallel to the use of comma
and apostrophe in the Russian transliteration above, and of h, w, y, and
so on in other transcription systems. However, the demonstration of how
the component works is too easy to misinterpret in this form, due to
graphical constraints in "the availability and organization of

This analysis closely parallels several examples of neutralization that
Harris discusses, so it is certainly plausible to suppose that this is
what Harris would do with Halle's example.

Long components differ from standard phonemes in two ways: on the one
hand, complementary distribution is extended from allophones to
sequences of allophones, and on the other hand the notion of allophone
is extended from segments to components of segments. As I mentioned
before, Harris did not *require* allophones to share a characteristic
phonetic feature (Chomsky's (1964) "strong invariance condition"), that
was an optional desideratum of more or less equal status to
distributional considerations.

The relation between Harris' phonology and generative phonology is an
interesting one. Distinctive features are defined in phonetic terms and
are intended to be universal; long components like the other phonemes
are distributional entities, defined in language-specific terms, though
their definition of course draws upon a language-universal theory of
phonetic description. A component may have differing phonetic values
throughout its length, as for example those defined to account for
cluster restrictions; generative theory requires more phonetic
"naturalness" of its features, and defines them only for unit length
(though autosegmental phonology opens this up). For Harris, the
motivation for setting up components is simplification of the grammar of
the particular language being described; for generative theory, the
motivation for distinctive features is to have a universal inventory of
parameters of contrast, and questions of their impact on rule simplicity
apply rarely and in a generalized way. The abstract/natural debate
directly reflects this same tension between descriptive relevance
(expressing relations in a particular language, using abstract terms for
distributional facts) and universality (expressing relations in
"natural" terms that apply to all languages).

These countervailing motivations (for a universal descriptive
terminology and for perspicuous description of the idiosyncrasies of a
given language) can I think be reconciled using the presentation
techniques of autosegmental phonology. Any component extending over
multiple segments can be presented as a feature or features projected
from a separate tier. Some of Harris' long components vary in phonetic
value through their length. Those that are discontinuous (skipping
vowels, applying to all nasals, applying only to the first root
consonant as in the examples of Greek reduplication, etc.) fall out
naturally in an autosegmental treatment. Those designed to describe
limitations on clusters can be replaced by other means developed or
being developed in autosegmental theory.

Conversely, autosegmental phonology vitiates Halle's classic argument
against the phoneme in precisely the same way as does Harris' long
components. It shows that one *can* have an orthography that represents
just what is contrastive in a given descriptive state of a language
without being stuck with the redundancy of identical rules covering
identical morphophonemic and allophonic alternations. These will be
relatively "abstract" phonemes, not so "natural" as Bloch wanted, but
(as Harris might say to him) that's the way languages really, truly are,
and we just have to live with it.

 Bruce Nevin

[End Linguist List, Vol. 2, No. 172]
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue